diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-19 00:47:55 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-19 00:47:55 +0000 |
commit | 26a029d407be480d791972afb5975cf62c9360a6 (patch) | |
tree | f435a8308119effd964b339f76abb83a57c29483 /js/src/jit/riscv64 | |
parent | Initial commit. (diff) | |
download | firefox-26a029d407be480d791972afb5975cf62c9360a6.tar.xz firefox-26a029d407be480d791972afb5975cf62c9360a6.zip |
Adding upstream version 124.0.1.upstream/124.0.1
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'js/src/jit/riscv64')
57 files changed, 33913 insertions, 0 deletions
diff --git a/js/src/jit/riscv64/Architecture-riscv64.cpp b/js/src/jit/riscv64/Architecture-riscv64.cpp new file mode 100644 index 0000000000..ea4a364b92 --- /dev/null +++ b/js/src/jit/riscv64/Architecture-riscv64.cpp @@ -0,0 +1,100 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "jit/riscv64/Architecture-riscv64.h" + +#include "jit/FlushICache.h" // js::jit::FlushICache +#include "jit/RegisterSets.h" +#include "jit/Simulator.h" +namespace js { +namespace jit { +Registers::Code Registers::FromName(const char* name) { + for (size_t i = 0; i < Total; i++) { + if (strcmp(GetName(i), name) == 0) { + return Code(i); + } + } + + return Invalid; +} + +FloatRegisters::Code FloatRegisters::FromName(const char* name) { + for (size_t i = 0; i < Total; i++) { + if (strcmp(GetName(i), name) == 0) { + return Code(i); + } + } + + return Invalid; +} + +FloatRegisterSet FloatRegister::ReduceSetForPush(const FloatRegisterSet& s) { +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + LiveFloatRegisterSet mod; + for (FloatRegisterIterator iter(s); iter.more(); ++iter) { + if ((*iter).isSingle()) { + // Even for single size registers save complete double register. + mod.addUnchecked((*iter).doubleOverlay()); + } else { + mod.addUnchecked(*iter); + } + } + return mod.set(); +} + +FloatRegister FloatRegister::singleOverlay() const { + MOZ_ASSERT(!isInvalid()); + if (kind_ == Codes::Double) { + return FloatRegister(encoding_, Codes::Single); + } + return *this; +} + +FloatRegister FloatRegister::doubleOverlay() const { + MOZ_ASSERT(!isInvalid()); + if (kind_ != Codes::Double) { + return FloatRegister(encoding_, Codes::Double); + } + return *this; +} + +uint32_t FloatRegister::GetPushSizeInBytes( + const TypedRegisterSet<FloatRegister>& s) { +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + return s.size() * sizeof(double); +} +void FlushICache(void* code, size_t size) { +#if defined(JS_SIMULATOR) + js::jit::SimulatorProcess::FlushICache(code, size); + +#elif defined(__linux__) +# if defined(__GNUC__) + intptr_t end = reinterpret_cast<intptr_t>(code) + size; + __builtin___clear_cache(reinterpret_cast<char*>(code), + reinterpret_cast<char*>(end)); + +# else + _flush_cache(reinterpret_cast<char*>(code), size, BCACHE); +# endif +#else +# error "Unsupported platform" +#endif +} + +bool CPUFlagsHaveBeenComputed() { + // TODO Add CPU flags support + // Flags were computed above. + return true; +} + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/Architecture-riscv64.h b/js/src/jit/riscv64/Architecture-riscv64.h new file mode 100644 index 0000000000..c75bd05ff1 --- /dev/null +++ b/js/src/jit/riscv64/Architecture-riscv64.h @@ -0,0 +1,514 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_Architecture_riscv64_h +#define jit_riscv64_Architecture_riscv64_h + +// JitSpewer.h is included through MacroAssembler implementations for other +// platforms, so include it here to avoid inadvertent build bustage. +#include "mozilla/MathAlgorithms.h" + +#include <algorithm> +#include <iterator> + +#include "jit/JitSpewer.h" +#include "jit/shared/Architecture-shared.h" +#include "js/Utility.h" + +namespace js { +namespace jit { + +static const uint32_t SimdMemoryAlignment = + 16; // Make it 4 to avoid a bunch of div-by-zero warnings + +// RISCV64 has 32 64-bit integer registers, x0 though x31. +// The program counter is not accessible as a register. + +// RISCV INT Register Convention: +// Name Alias Usage +// x0 zero hardwired to 0, ignores writes +// x1 ra return address for calls +// x2 sp stack pointer +// x3 gp global pointer +// x4 tp thread pointer +// x5-x7 t0-t2 temporary register 0 +// x8 fp/s0 Callee-saved register 0 or frame pointer +// x9 s1 Callee-saved register 1 +// x10-x11 a0-a1 return value or function argument +// x12-x17 a2-a7 function argument 2 +// x18-x27 s2-s11 Callee-saved register +// x28-x31 t3-t6 temporary register 3 + +// RISCV-64 FP Register Convention: +// Name Alias Usage +// $f0-$f7 $ft0-$ft7 Temporary registers +// $f8-$f9 $fs0-$fs1 Callee-saved registers +// $f10-$f11 $fa0-$fa1 Return values +// $f12-$f17 $fa2-$fa7 Args values +// $f18-$f27 $fs2-$fs11 Callee-saved registers +// $f28-$f31 $ft8-$ft11 Temporary registers +class Registers { + public: + enum RegisterID { + x0 = 0, + x1, + x2, + x3, + x4, + x5, + x6, + x7, + x8, + x9, + x10, + x11, + x12, + x13, + x14, + x15, + x16, + x17, + x18, + x19, + x20, + x21, + x22, + x23, + x24, + x25, + x26, + x27, + x28, + x29, + x30, + x31, + zero = x0, + ra = x1, + sp = x2, + gp = x3, + tp = x4, + t0 = x5, + t1 = x6, + t2 = x7, + fp = x8, + s1 = x9, + a0 = x10, + a1 = x11, + a2 = x12, + a3 = x13, + a4 = x14, + a5 = x15, + a6 = x16, + a7 = x17, + s2 = x18, + s3 = x19, + s4 = x20, + s5 = x21, + s6 = x22, + s7 = x23, + s8 = x24, + s9 = x25, + s10 = x26, + s11 = x27, + t3 = x28, + t4 = x29, + t5 = x30, + t6 = x31, + invalid_reg, + }; + typedef uint8_t Code; + typedef RegisterID Encoding; + union RegisterContent { + uintptr_t r; + }; + + typedef uint32_t SetType; + + static uint32_t SetSize(SetType x) { + static_assert(sizeof(SetType) == 4, "SetType must be 32 bits"); + return mozilla::CountPopulation32(x); + } + static uint32_t FirstBit(SetType x) { + return mozilla::CountTrailingZeroes32(x); + } + static uint32_t LastBit(SetType x) { + return 31 - mozilla::CountLeadingZeroes32(x); + } + static const char* GetName(uint32_t code) { + static const char* const Names[] = { + "zero", "ra", "sp", "gp", "tp", "t0", "t1", "t2", "fp", "s1", "a0", + "a1", "a2", "a3", "a4", "a5", "a6", "a7", "s2", "s3", "s4", "s5", + "s6", "s7", "s8", "s9", "s10", "s11", "t3", "t4", "t5", "t6"}; + static_assert(Total == std::size(Names), "Table is the correct size"); + if (code >= Total) { + return "invalid"; + } + return Names[code]; + } + + static Code FromName(const char*); + + static const Encoding StackPointer = sp; + static const Encoding Invalid = invalid_reg; + static const uint32_t Total = 32; + static const uint32_t TotalPhys = 32; + static const uint32_t Allocatable = 24; + static const SetType NoneMask = 0x0; + static const SetType AllMask = 0xFFFFFFFF; + static const SetType ArgRegMask = + (1 << Registers::a0) | (1 << Registers::a1) | (1 << Registers::a2) | + (1 << Registers::a3) | (1 << Registers::a4) | (1 << Registers::a5) | + (1 << Registers::a6) | (1 << Registers::a7); + + static const SetType VolatileMask = + ArgRegMask | (1 << Registers::t0) | (1 << Registers::t1) | + (1 << Registers::t2) | (1 << Registers::t3) | (1 << Registers::t4) | + (1 << Registers::t5) | (1 << Registers::t6); + + // We use this constant to save registers when entering functions. This + // is why $ra is added here even though it is not "Non Volatile". + static const SetType NonVolatileMask = + (1 << Registers::ra) | (1 << Registers::fp) | (1 << Registers::s1) | + (1 << Registers::s2) | (1 << Registers::s3) | (1 << Registers::s4) | + (1 << Registers::s5) | (1 << Registers::s6) | (1 << Registers::s7) | + (1 << Registers::s8) | (1 << Registers::s9) | (1 << Registers::s10) | + (1 << Registers::s11); + + static const SetType NonAllocatableMask = + (1 << Registers::zero) | // Always be zero. + (1 << Registers::t4) | // Scratch reg + (1 << Registers::t5) | // Scratch reg + (1 << Registers::t6) | // Scratch reg or call reg + (1 << Registers::s11) | // Scratch reg + (1 << Registers::ra) | (1 << Registers::tp) | (1 << Registers::sp) | + (1 << Registers::fp) | (1 << Registers::gp); + + static const SetType AllocatableMask = AllMask & ~NonAllocatableMask; + + // Registers returned from a JS -> JS call. + static const SetType JSCallMask = (1 << Registers::a2); + + // Registers returned from a JS -> C call. + static const SetType CallMask = (1 << Registers::a0); + + static const SetType WrapperMask = VolatileMask; +}; + +// Smallest integer type that can hold a register bitmask. +typedef uint32_t PackedRegisterMask; + +class FloatRegisters { + public: + enum FPRegisterID { + f0 = 0, + f1, + f2, + f3, + f4, + f5, + f6, + f7, + f8, + f9, + f10, + f11, + f12, + f13, + f14, + f15, + f16, + f17, + f18, + f19, + f20, + f21, + f22, + f23, + f24, + f25, + f26, + f27, + f28, + f29, + f30, + f31, + invalid_reg, + ft0 = f0, + ft1 = f1, + ft2 = f2, + ft3 = f3, + ft4 = f4, + ft5 = f5, + ft6 = f6, + ft7 = f7, + fs0 = f8, + fs1 = f9, + fa0 = f10, + fa1 = f11, + fa2 = f12, + fa3 = f13, + fa4 = f14, + fa5 = f15, + fa6 = f16, + fa7 = f17, + fs2 = f18, + fs3 = f19, + fs4 = f20, + fs5 = f21, + fs6 = f22, + fs7 = f23, + fs8 = f24, + fs9 = f25, + fs10 = f26, + fs11 = f27, // Scratch register + ft8 = f28, + ft9 = f29, + ft10 = f30, // Scratch register + ft11 = f31 + }; + + enum Kind : uint8_t { Double, NumTypes, Single }; + + typedef FPRegisterID Code; + typedef FPRegisterID Encoding; + union RegisterContent { + float s; + double d; + }; + + static const char* GetName(uint32_t code) { + static const char* const Names[] = { + "ft0", "ft1", "ft2", "ft3", "ft4", "ft5", "ft6", "ft7", + "fs0", "fs2", "fa0", "fa1", "fa2", "fa3", "fa4", "fa5", + "fa6", "fa7", "fs2", "fs3", "fs4", "fs5", "fs6", "fs7", + "fs8", "fs9", "fs10", "fs11", "ft8", "ft9", "ft10", "ft11"}; + static_assert(TotalPhys == std::size(Names), "Table is the correct size"); + if (code >= Total) { + return "invalid"; + } + return Names[code]; + } + + static Code FromName(const char* name); + + typedef uint32_t SetType; + + static const Code Invalid = invalid_reg; + static const uint32_t Total = 32; + static const uint32_t TotalPhys = 32; + static const uint32_t Allocatable = 23; + static const SetType AllPhysMask = 0xFFFFFFFF; + static const SetType AllMask = 0xFFFFFFFF; + static const SetType AllDoubleMask = AllMask; + // Single values are stored as 64 bits values (NaN-boxed) when pushing them to + // the stack, we do not require making distinctions between the 2 types, and + // therefore the masks are overlapping.See The RISC-V Instruction Set Manual + // for 14.2 NaN Boxing of Narrower Values. + static const SetType AllSingleMask = AllMask; + static const SetType NonVolatileMask = + SetType((1 << FloatRegisters::fs0) | (1 << FloatRegisters::fs1) | + (1 << FloatRegisters::fs2) | (1 << FloatRegisters::fs3) | + (1 << FloatRegisters::fs4) | (1 << FloatRegisters::fs5) | + (1 << FloatRegisters::fs6) | (1 << FloatRegisters::fs7) | + (1 << FloatRegisters::fs8) | (1 << FloatRegisters::fs9) | + (1 << FloatRegisters::fs10) | (1 << FloatRegisters::fs11)); + static const SetType VolatileMask = AllMask & ~NonVolatileMask; + + // fs11/ft10 is the scratch register. + static const SetType NonAllocatableMask = + SetType((1 << FloatRegisters::fs11) | (1 << FloatRegisters::ft10)); + + static const SetType AllocatableMask = AllMask & ~NonAllocatableMask; +}; + +template <typename T> +class TypedRegisterSet; + +struct FloatRegister { + public: + typedef FloatRegisters Codes; + typedef Codes::Code Code; + typedef Codes::Encoding Encoding; + typedef Codes::SetType SetType; + + static uint32_t SetSize(SetType x) { + static_assert(sizeof(SetType) == 4, "SetType must be 32 bits"); + x &= FloatRegisters::AllPhysMask; + return mozilla::CountPopulation32(x); + } + + static uint32_t FirstBit(SetType x) { + static_assert(sizeof(SetType) == 4, "SetType"); + return mozilla::CountTrailingZeroes64(x); + } + static uint32_t LastBit(SetType x) { + static_assert(sizeof(SetType) == 4, "SetType"); + return 31 - mozilla::CountLeadingZeroes64(x); + } + + static FloatRegister FromCode(uint32_t i) { + uint32_t code = i & 0x1f; + return FloatRegister(Code(code)); + } + bool isSimd128() const { return false; } + bool isInvalid() const { return invalid_; } + FloatRegister asSingle() const { + MOZ_ASSERT(!invalid_); + return FloatRegister(Encoding(encoding_), FloatRegisters::Single); + } + FloatRegister asDouble() const { + MOZ_ASSERT(!invalid_); + return FloatRegister(Encoding(encoding_), FloatRegisters::Double); + } + FloatRegister asSimd128() const { MOZ_CRASH(); } + constexpr Code code() const { + MOZ_ASSERT(!invalid_); + return encoding_; + } + Encoding encoding() const { return encoding_; } + const char* name() const { return FloatRegisters::GetName(code()); } + bool volatile_() const { + MOZ_ASSERT(!invalid_); + return !!((SetType(1) << code()) & FloatRegisters::VolatileMask); + } + bool operator!=(FloatRegister other) const { return code() != other.code(); } + bool operator==(FloatRegister other) const { return code() == other.code(); } + bool aliases(FloatRegister other) const { + return other.encoding_ == encoding_; + } + uint32_t numAliased() const { return 1; } + FloatRegister aliased(uint32_t aliasIdx) const { + MOZ_ASSERT(aliasIdx == 0); + return *this; + } + // Ensure that two floating point registers' types are equivalent. + bool equiv(FloatRegister other) const { + MOZ_ASSERT(!invalid_); + return kind_ == other.kind_; + } + constexpr uint32_t size() const { + MOZ_ASSERT(!invalid_); + if (kind_ == FloatRegisters::Double) { + return sizeof(double); + } + MOZ_ASSERT(kind_ == FloatRegisters::Single); + return sizeof(float); + } + uint32_t numAlignedAliased() { return numAliased(); } + FloatRegister alignedAliased(uint32_t aliasIdx) { + MOZ_ASSERT(aliasIdx < numAliased()); + return aliased(aliasIdx); + } + SetType alignedOrDominatedAliasedSet() const { return SetType(1) << code(); } + static constexpr RegTypeName DefaultType = RegTypeName::Float64; + + template <RegTypeName Name = DefaultType> + static SetType LiveAsIndexableSet(SetType s) { + return SetType(0); + } + + template <RegTypeName Name = DefaultType> + static SetType AllocatableAsIndexableSet(SetType s) { + static_assert(Name != RegTypeName::Any, "Allocatable set are not iterable"); + return LiveAsIndexableSet<Name>(s); + } + + FloatRegister singleOverlay() const; + FloatRegister doubleOverlay() const; + + static TypedRegisterSet<FloatRegister> ReduceSetForPush( + const TypedRegisterSet<FloatRegister>& s); + + uint32_t getRegisterDumpOffsetInBytes() { +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + return code() * sizeof(double); + } + static Code FromName(const char* name); + + // This is used in static initializers, so produce a bogus value instead of + // crashing. + static uint32_t GetPushSizeInBytes(const TypedRegisterSet<FloatRegister>& s); + + private: + typedef Codes::Kind Kind; + // These fields only hold valid values: an invalid register is always + // represented as a valid encoding and kind with the invalid_ bit set. + Encoding encoding_; // 32 encodings + Kind kind_; // Double, Single; more later + bool invalid_; + + public: + constexpr FloatRegister(Encoding encoding, Kind kind) + : encoding_(encoding), kind_(kind), invalid_(false) { + MOZ_ASSERT(uint32_t(encoding) < Codes::Total); + } + + constexpr FloatRegister(Encoding encoding) + : encoding_(encoding), kind_(FloatRegisters::Double), invalid_(false) { + MOZ_ASSERT(uint32_t(encoding) < Codes::Total); + } + + constexpr FloatRegister() + : encoding_(FloatRegisters::invalid_reg), + kind_(FloatRegisters::Double), + invalid_(true) {} + + bool isSingle() const { + MOZ_ASSERT(!invalid_); + // On riscv64 arch, float register and double register using the same + // register file. + return kind_ == FloatRegisters::Single || kind_ == FloatRegisters::Double; + } + bool isDouble() const { + MOZ_ASSERT(!invalid_); + return kind_ == FloatRegisters::Double; + } + + Encoding code() { return encoding_; } +}; + +template <> +inline FloatRegister::SetType +FloatRegister::LiveAsIndexableSet<RegTypeName::Float32>(SetType set) { + return set & FloatRegisters::AllSingleMask; +} + +template <> +inline FloatRegister::SetType +FloatRegister::LiveAsIndexableSet<RegTypeName::Float64>(SetType set) { + return set & FloatRegisters::AllDoubleMask; +} + +template <> +inline FloatRegister::SetType +FloatRegister::LiveAsIndexableSet<RegTypeName::Any>(SetType set) { + return set; +} + +inline bool hasUnaliasedDouble() { return false; } +inline bool hasMultiAlias() { return false; } + +static const uint32_t ShadowStackSpace = 0; +static const uint32_t JumpImmediateRange = INT32_MAX; + +#ifdef JS_NUNBOX32 +static const int32_t NUNBOX32_TYPE_OFFSET = 4; +static const int32_t NUNBOX32_PAYLOAD_OFFSET = 0; +#endif + +static const uint32_t SpillSlotSize = + std::max(sizeof(Registers::RegisterContent), + sizeof(FloatRegisters::RegisterContent)); + +inline uint32_t GetRISCV64Flags() { return 0; } + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_Architecture_riscv64_h */ diff --git a/js/src/jit/riscv64/Assembler-riscv64.cpp b/js/src/jit/riscv64/Assembler-riscv64.cpp new file mode 100644 index 0000000000..5370192987 --- /dev/null +++ b/js/src/jit/riscv64/Assembler-riscv64.cpp @@ -0,0 +1,1548 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +// Copyright (c) 1994-2006 Sun Microsystems Inc. +// All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// - Redistributions of source code must retain the above copyright notice, +// this list of conditions and the following disclaimer. +// +// - Redistribution in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution. +// +// - Neither the name of Sun Microsystems or the names of contributors may +// be used to endorse or promote products derived from this software without +// specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS +// IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, +// THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR +// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +// The original source code covered by the above license above has been +// modified significantly by Google Inc. +// Copyright 2021 the V8 project authors. All rights reserved. +#include "jit/riscv64/Assembler-riscv64.h" + +#include "mozilla/DebugOnly.h" +#include "mozilla/Maybe.h" + +#include "gc/Marking.h" +#include "jit/AutoWritableJitCode.h" +#include "jit/ExecutableAllocator.h" +#include "jit/riscv64/disasm/Disasm-riscv64.h" +#include "vm/Realm.h" + +using mozilla::DebugOnly; +namespace js { +namespace jit { + +#define UNIMPLEMENTED_RISCV() MOZ_CRASH("RISC_V not implemented"); + +bool Assembler::FLAG_riscv_debug = false; + +void Assembler::nop() { addi(ToRegister(0), ToRegister(0), 0); } + +// Size of the instruction stream, in bytes. +size_t Assembler::size() const { return m_buffer.size(); } + +bool Assembler::swapBuffer(wasm::Bytes& bytes) { + // For now, specialize to the one use case. As long as wasm::Bytes is a + // Vector, not a linked-list of chunks, there's not much we can do other + // than copy. + MOZ_ASSERT(bytes.empty()); + if (!bytes.resize(bytesNeeded())) { + return false; + } + m_buffer.executableCopy(bytes.begin()); + return true; +} + +// Size of the relocation table, in bytes. +size_t Assembler::jumpRelocationTableBytes() const { + return jumpRelocations_.length(); +} + +size_t Assembler::dataRelocationTableBytes() const { + return dataRelocations_.length(); +} +// Size of the data table, in bytes. +size_t Assembler::bytesNeeded() const { + return size() + jumpRelocationTableBytes() + dataRelocationTableBytes(); +} + +void Assembler::executableCopy(uint8_t* buffer) { + MOZ_ASSERT(isFinished); + m_buffer.executableCopy(buffer); +} + +uint32_t Assembler::AsmPoolMaxOffset = 1024; + +uint32_t Assembler::GetPoolMaxOffset() { + static bool isSet = false; + if (!isSet) { + char* poolMaxOffsetStr = getenv("ASM_POOL_MAX_OFFSET"); + uint32_t poolMaxOffset; + if (poolMaxOffsetStr && + sscanf(poolMaxOffsetStr, "%u", &poolMaxOffset) == 1) { + AsmPoolMaxOffset = poolMaxOffset; + } + isSet = true; + } + return AsmPoolMaxOffset; +} + +// Pool callbacks stuff: +void Assembler::InsertIndexIntoTag(uint8_t* load_, uint32_t index) { + MOZ_CRASH("Unimplement"); +} + +void Assembler::PatchConstantPoolLoad(void* loadAddr, void* constPoolAddr) { + MOZ_CRASH("Unimplement"); +} + +void Assembler::processCodeLabels(uint8_t* rawCode) { + for (const CodeLabel& label : codeLabels_) { + Bind(rawCode, label); + } +} + +void Assembler::WritePoolGuard(BufferOffset branch, Instruction* dest, + BufferOffset afterPool) { + DEBUG_PRINTF("\tWritePoolGuard\n"); + int32_t off = afterPool.getOffset() - branch.getOffset(); + if (!is_int21(off) || !((off & 0x1) == 0)) { + printf("%d\n", off); + MOZ_CRASH("imm invalid"); + } + // JAL encode is + // 31 | 30 21 | 20 | 19 12 | 11 7 | 6 0 | + // imm[20] | imm[10:1] | imm[11] | imm[19:12] | rd | opcode| + // 1 10 1 8 5 7 + // offset[20:1] dest JAL + int32_t imm20 = (off & 0xff000) | // bits 19-12 + ((off & 0x800) << 9) | // bit 11 + ((off & 0x7fe) << 20) | // bits 10-1 + ((off & 0x100000) << 11); // bit 20 + Instr instr = JAL | (imm20 & kImm20Mask); + dest->SetInstructionBits(instr); + DEBUG_PRINTF("%p(%x): ", dest, branch.getOffset()); + disassembleInstr(dest->InstructionBits(), JitSpew_Codegen); +} + +void Assembler::WritePoolHeader(uint8_t* start, Pool* p, bool isNatural) { + static_assert(sizeof(PoolHeader) == 4); + + // Get the total size of the pool. + const uintptr_t totalPoolSize = sizeof(PoolHeader) + p->getPoolSize(); + const uintptr_t totalPoolInstructions = totalPoolSize / kInstrSize; + + MOZ_ASSERT((totalPoolSize & 0x3) == 0); + MOZ_ASSERT(totalPoolInstructions < (1 << 15)); + + PoolHeader header(totalPoolInstructions, isNatural); + *(PoolHeader*)start = header; +} + +void Assembler::copyJumpRelocationTable(uint8_t* dest) { + if (jumpRelocations_.length()) { + memcpy(dest, jumpRelocations_.buffer(), jumpRelocations_.length()); + } +} + +void Assembler::copyDataRelocationTable(uint8_t* dest) { + if (dataRelocations_.length()) { + memcpy(dest, dataRelocations_.buffer(), dataRelocations_.length()); + } +} + +void Assembler::RV_li(Register rd, int64_t imm) { + UseScratchRegisterScope temps(this); + if (RecursiveLiCount(imm) > GeneralLiCount(imm, temps.hasAvailable())) { + GeneralLi(rd, imm); + } else { + RecursiveLi(rd, imm); + } +} + +int Assembler::RV_li_count(int64_t imm, bool is_get_temp_reg) { + if (RecursiveLiCount(imm) > GeneralLiCount(imm, is_get_temp_reg)) { + return GeneralLiCount(imm, is_get_temp_reg); + } else { + return RecursiveLiCount(imm); + } +} + +void Assembler::GeneralLi(Register rd, int64_t imm) { + // 64-bit imm is put in the register rd. + // In most cases the imm is 32 bit and 2 instructions are generated. If a + // temporary register is available, in the worst case, 6 instructions are + // generated for a full 64-bit immediate. If temporay register is not + // available the maximum will be 8 instructions. If imm is more than 32 bits + // and a temp register is available, imm is divided into two 32-bit parts, + // low_32 and up_32. Each part is built in a separate register. low_32 is + // built before up_32. If low_32 is negative (upper 32 bits are 1), 0xffffffff + // is subtracted from up_32 before up_32 is built. This compensates for 32 + // bits of 1's in the lower when the two registers are added. If no temp is + // available, the upper 32 bit is built in rd, and the lower 32 bits are + // devided to 3 parts (11, 11, and 10 bits). The parts are shifted and added + // to the upper part built in rd. + if (is_int32(imm + 0x800)) { + // 32-bit case. Maximum of 2 instructions generated + int64_t high_20 = ((imm + 0x800) >> 12); + int64_t low_12 = imm << 52 >> 52; + if (high_20) { + lui(rd, (int32_t)high_20); + if (low_12) { + addi(rd, rd, low_12); + } + } else { + addi(rd, zero_reg, low_12); + } + return; + } else { + UseScratchRegisterScope temps(this); + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + // 64-bit case: divide imm into two 32-bit parts, upper and lower + int64_t up_32 = imm >> 32; + int64_t low_32 = imm & 0xffffffffull; + Register temp_reg = rd; + // Check if a temporary register is available + if (up_32 == 0 || low_32 == 0) { + // No temp register is needed + } else { + temp_reg = temps.hasAvailable() ? temps.Acquire() : InvalidReg; + } + if (temp_reg != InvalidReg) { + // keep track of hardware behavior for lower part in sim_low + int64_t sim_low = 0; + // Build lower part + if (low_32 != 0) { + int64_t high_20 = ((low_32 + 0x800) >> 12); + int64_t low_12 = low_32 & 0xfff; + if (high_20) { + // Adjust to 20 bits for the case of overflow + high_20 &= 0xfffff; + sim_low = ((high_20 << 12) << 32) >> 32; + lui(rd, (int32_t)high_20); + if (low_12) { + sim_low += (low_12 << 52 >> 52) | low_12; + addi(rd, rd, low_12); + } + } else { + sim_low = low_12; + ori(rd, zero_reg, low_12); + } + } + if (sim_low & 0x100000000) { + // Bit 31 is 1. Either an overflow or a negative 64 bit + if (up_32 == 0) { + // Positive number, but overflow because of the add 0x800 + slli(rd, rd, 32); + srli(rd, rd, 32); + return; + } + // low_32 is a negative 64 bit after the build + up_32 = (up_32 - 0xffffffff) & 0xffffffff; + } + if (up_32 == 0) { + return; + } + // Build upper part in a temporary register + if (low_32 == 0) { + // Build upper part in rd + temp_reg = rd; + } + int64_t high_20 = (up_32 + 0x800) >> 12; + int64_t low_12 = up_32 & 0xfff; + if (high_20) { + // Adjust to 20 bits for the case of overflow + high_20 &= 0xfffff; + lui(temp_reg, (int32_t)high_20); + if (low_12) { + addi(temp_reg, temp_reg, low_12); + } + } else { + ori(temp_reg, zero_reg, low_12); + } + // Put it at the bgining of register + slli(temp_reg, temp_reg, 32); + if (low_32 != 0) { + add(rd, rd, temp_reg); + } + return; + } + // No temp register. Build imm in rd. + // Build upper 32 bits first in rd. Divide lower 32 bits parts and add + // parts to the upper part by doing shift and add. + // First build upper part in rd. + int64_t high_20 = (up_32 + 0x800) >> 12; + int64_t low_12 = up_32 & 0xfff; + if (high_20) { + // Adjust to 20 bits for the case of overflow + high_20 &= 0xfffff; + lui(rd, (int32_t)high_20); + if (low_12) { + addi(rd, rd, low_12); + } + } else { + ori(rd, zero_reg, low_12); + } + // upper part already in rd. Each part to be added to rd, has maximum of 11 + // bits, and always starts with a 1. rd is shifted by the size of the part + // plus the number of zeros between the parts. Each part is added after the + // left shift. + uint32_t mask = 0x80000000; + int32_t shift_val = 0; + int32_t i; + for (i = 0; i < 32; i++) { + if ((low_32 & mask) == 0) { + mask >>= 1; + shift_val++; + if (i == 31) { + // rest is zero + slli(rd, rd, shift_val); + } + continue; + } + // The first 1 seen + int32_t part; + if ((i + 11) < 32) { + // Pick 11 bits + part = ((uint32_t)(low_32 << i) >> i) >> (32 - (i + 11)); + slli(rd, rd, shift_val + 11); + ori(rd, rd, part); + i += 10; + mask >>= 11; + } else { + part = (uint32_t)(low_32 << i) >> i; + slli(rd, rd, shift_val + (32 - i)); + ori(rd, rd, part); + break; + } + shift_val = 0; + } + } +} + +int Assembler::GeneralLiCount(int64_t imm, bool is_get_temp_reg) { + int count = 0; + // imitate Assembler::RV_li + if (is_int32(imm + 0x800)) { + // 32-bit case. Maximum of 2 instructions generated + int64_t high_20 = ((imm + 0x800) >> 12); + int64_t low_12 = imm << 52 >> 52; + if (high_20) { + count++; + if (low_12) { + count++; + } + } else { + count++; + } + return count; + } else { + // 64-bit case: divide imm into two 32-bit parts, upper and lower + int64_t up_32 = imm >> 32; + int64_t low_32 = imm & 0xffffffffull; + // Check if a temporary register is available + if (is_get_temp_reg) { + // keep track of hardware behavior for lower part in sim_low + int64_t sim_low = 0; + // Build lower part + if (low_32 != 0) { + int64_t high_20 = ((low_32 + 0x800) >> 12); + int64_t low_12 = low_32 & 0xfff; + if (high_20) { + // Adjust to 20 bits for the case of overflow + high_20 &= 0xfffff; + sim_low = ((high_20 << 12) << 32) >> 32; + count++; + if (low_12) { + sim_low += (low_12 << 52 >> 52) | low_12; + count++; + } + } else { + sim_low = low_12; + count++; + } + } + if (sim_low & 0x100000000) { + // Bit 31 is 1. Either an overflow or a negative 64 bit + if (up_32 == 0) { + // Positive number, but overflow because of the add 0x800 + count++; + count++; + return count; + } + // low_32 is a negative 64 bit after the build + up_32 = (up_32 - 0xffffffff) & 0xffffffff; + } + if (up_32 == 0) { + return count; + } + int64_t high_20 = (up_32 + 0x800) >> 12; + int64_t low_12 = up_32 & 0xfff; + if (high_20) { + // Adjust to 20 bits for the case of overflow + high_20 &= 0xfffff; + count++; + if (low_12) { + count++; + } + } else { + count++; + } + // Put it at the bgining of register + count++; + if (low_32 != 0) { + count++; + } + return count; + } + // No temp register. Build imm in rd. + // Build upper 32 bits first in rd. Divide lower 32 bits parts and add + // parts to the upper part by doing shift and add. + // First build upper part in rd. + int64_t high_20 = (up_32 + 0x800) >> 12; + int64_t low_12 = up_32 & 0xfff; + if (high_20) { + // Adjust to 20 bits for the case of overflow + high_20 &= 0xfffff; + count++; + if (low_12) { + count++; + } + } else { + count++; + } + // upper part already in rd. Each part to be added to rd, has maximum of 11 + // bits, and always starts with a 1. rd is shifted by the size of the part + // plus the number of zeros between the parts. Each part is added after the + // left shift. + uint32_t mask = 0x80000000; + int32_t i; + for (i = 0; i < 32; i++) { + if ((low_32 & mask) == 0) { + mask >>= 1; + if (i == 31) { + // rest is zero + count++; + } + continue; + } + // The first 1 seen + if ((i + 11) < 32) { + // Pick 11 bits + count++; + count++; + i += 10; + mask >>= 11; + } else { + count++; + count++; + break; + } + } + } + return count; +} + +void Assembler::li_ptr(Register rd, int64_t imm) { + m_buffer.enterNoNops(); + m_buffer.assertNoPoolAndNoNops(); + // Initialize rd with an address + // Pointers are 48 bits + // 6 fixed instructions are generated + DEBUG_PRINTF("li_ptr(%d, %lx <%ld>)\n", ToNumber(rd), imm, imm); + MOZ_ASSERT((imm & 0xfff0000000000000ll) == 0); + int64_t a6 = imm & 0x3f; // bits 0:5. 6 bits + int64_t b11 = (imm >> 6) & 0x7ff; // bits 6:11. 11 bits + int64_t high_31 = (imm >> 17) & 0x7fffffff; // 31 bits + int64_t high_20 = ((high_31 + 0x800) >> 12); // 19 bits + int64_t low_12 = high_31 & 0xfff; // 12 bits + lui(rd, (int32_t)high_20); + addi(rd, rd, low_12); // 31 bits in rd. + slli(rd, rd, 11); // Space for next 11 bis + ori(rd, rd, b11); // 11 bits are put in. 42 bit in rd + slli(rd, rd, 6); // Space for next 6 bits + ori(rd, rd, a6); // 6 bits are put in. 48 bis in rd + m_buffer.leaveNoNops(); +} + +void Assembler::li_constant(Register rd, int64_t imm) { + m_buffer.enterNoNops(); + m_buffer.assertNoPoolAndNoNops(); + DEBUG_PRINTF("li_constant(%d, %lx <%ld>)\n", ToNumber(rd), imm, imm); + lui(rd, (imm + (1LL << 47) + (1LL << 35) + (1LL << 23) + (1LL << 11)) >> + 48); // Bits 63:48 + addiw(rd, rd, + (imm + (1LL << 35) + (1LL << 23) + (1LL << 11)) << 16 >> + 52); // Bits 47:36 + slli(rd, rd, 12); + addi(rd, rd, (imm + (1LL << 23) + (1LL << 11)) << 28 >> 52); // Bits 35:24 + slli(rd, rd, 12); + addi(rd, rd, (imm + (1LL << 11)) << 40 >> 52); // Bits 23:12 + slli(rd, rd, 12); + addi(rd, rd, imm << 52 >> 52); // Bits 11:0 + m_buffer.leaveNoNops(); +} + +ABIArg ABIArgGenerator::next(MIRType type) { + switch (type) { + case MIRType::Int32: + case MIRType::Int64: + case MIRType::Pointer: + case MIRType::WasmAnyRef: + case MIRType::StackResults: { + if (intRegIndex_ == NumIntArgRegs) { + current_ = ABIArg(stackOffset_); + stackOffset_ += sizeof(uintptr_t); + break; + } + current_ = ABIArg(Register::FromCode(intRegIndex_ + a0.encoding())); + intRegIndex_++; + break; + } + case MIRType::Float32: + case MIRType::Double: { + if (floatRegIndex_ == NumFloatArgRegs) { + current_ = ABIArg(stackOffset_); + stackOffset_ += sizeof(double); + break; + } + current_ = ABIArg(FloatRegister( + FloatRegisters::Encoding(floatRegIndex_ + fa0.encoding()), + type == MIRType::Double ? FloatRegisters::Double + : FloatRegisters::Single)); + floatRegIndex_++; + break; + } + case MIRType::Simd128: { + MOZ_CRASH("RISCV64 does not support simd yet."); + break; + } + default: + MOZ_CRASH("Unexpected argument type"); + } + return current_; +} + +bool Assembler::oom() const { + return AssemblerShared::oom() || m_buffer.oom() || jumpRelocations_.oom() || + dataRelocations_.oom() || !enoughLabelCache_; +} + +int Assembler::disassembleInstr(Instr instr, bool enable_spew) { + if (!FLAG_riscv_debug && !enable_spew) return -1; + disasm::NameConverter converter; + disasm::Disassembler disasm(converter); + EmbeddedVector<char, 128> disasm_buffer; + + int size = + disasm.InstructionDecode(disasm_buffer, reinterpret_cast<byte*>(&instr)); + DEBUG_PRINTF("%s\n", disasm_buffer.start()); + if (enable_spew) { + JitSpew(JitSpew_Codegen, "%s", disasm_buffer.start()); + } + return size; +} + +uintptr_t Assembler::target_address_at(Instruction* pc) { + Instruction* instr0 = pc; + DEBUG_PRINTF("target_address_at: pc: 0x%p\t", instr0); + Instruction* instr1 = pc + 1 * kInstrSize; + Instruction* instr2 = pc + 2 * kInstrSize; + Instruction* instr3 = pc + 3 * kInstrSize; + Instruction* instr4 = pc + 4 * kInstrSize; + Instruction* instr5 = pc + 5 * kInstrSize; + + // Interpret instructions for address generated by li: See listing in + // Assembler::set_target_address_at() just below. + if (IsLui(*reinterpret_cast<Instr*>(instr0)) && + IsAddi(*reinterpret_cast<Instr*>(instr1)) && + IsSlli(*reinterpret_cast<Instr*>(instr2)) && + IsOri(*reinterpret_cast<Instr*>(instr3)) && + IsSlli(*reinterpret_cast<Instr*>(instr4)) && + IsOri(*reinterpret_cast<Instr*>(instr5))) { + // Assemble the 64 bit value. + int64_t addr = (int64_t)(instr0->Imm20UValue() << kImm20Shift) + + (int64_t)instr1->Imm12Value(); + MOZ_ASSERT(instr2->Imm12Value() == 11); + addr <<= 11; + addr |= (int64_t)instr3->Imm12Value(); + MOZ_ASSERT(instr4->Imm12Value() == 6); + addr <<= 6; + addr |= (int64_t)instr5->Imm12Value(); + + DEBUG_PRINTF("addr: %lx\n", addr); + return static_cast<uintptr_t>(addr); + } + // We should never get here, force a bad address if we do. + MOZ_CRASH("RISC-V UNREACHABLE"); +} + +void Assembler::PatchDataWithValueCheck(CodeLocationLabel label, + ImmPtr newValue, ImmPtr expectedValue) { + PatchDataWithValueCheck(label, PatchedImmPtr(newValue.value), + PatchedImmPtr(expectedValue.value)); +} + +void Assembler::PatchDataWithValueCheck(CodeLocationLabel label, + PatchedImmPtr newValue, + PatchedImmPtr expectedValue) { + Instruction* inst = (Instruction*)label.raw(); + + // Extract old Value + DebugOnly<uint64_t> value = Assembler::ExtractLoad64Value(inst); + MOZ_ASSERT(value == uint64_t(expectedValue.value)); + + // Replace with new value + Assembler::UpdateLoad64Value(inst, uint64_t(newValue.value)); +} + +uint64_t Assembler::ExtractLoad64Value(Instruction* inst0) { + DEBUG_PRINTF("\tExtractLoad64Value: \tpc:%p ", inst0); + if (IsJal(*reinterpret_cast<Instr*>(inst0))) { + int offset = inst0->Imm20JValue(); + inst0 = inst0 + offset; + } + Instruction* instr1 = inst0 + 1 * kInstrSize; + if (IsAddiw(*reinterpret_cast<Instr*>(instr1))) { + // Li64 + Instruction* instr2 = inst0 + 2 * kInstrSize; + Instruction* instr3 = inst0 + 3 * kInstrSize; + Instruction* instr4 = inst0 + 4 * kInstrSize; + Instruction* instr5 = inst0 + 5 * kInstrSize; + Instruction* instr6 = inst0 + 6 * kInstrSize; + Instruction* instr7 = inst0 + 7 * kInstrSize; + if (IsLui(*reinterpret_cast<Instr*>(inst0)) && + IsAddiw(*reinterpret_cast<Instr*>(instr1)) && + IsSlli(*reinterpret_cast<Instr*>(instr2)) && + IsAddi(*reinterpret_cast<Instr*>(instr3)) && + IsSlli(*reinterpret_cast<Instr*>(instr4)) && + IsAddi(*reinterpret_cast<Instr*>(instr5)) && + IsSlli(*reinterpret_cast<Instr*>(instr6)) && + IsAddi(*reinterpret_cast<Instr*>(instr7))) { + int64_t imm = (int64_t)(inst0->Imm20UValue() << kImm20Shift) + + (int64_t)instr1->Imm12Value(); + MOZ_ASSERT(instr2->Imm12Value() == 12); + imm <<= 12; + imm += (int64_t)instr3->Imm12Value(); + MOZ_ASSERT(instr4->Imm12Value() == 12); + imm <<= 12; + imm += (int64_t)instr5->Imm12Value(); + MOZ_ASSERT(instr6->Imm12Value() == 12); + imm <<= 12; + imm += (int64_t)instr7->Imm12Value(); + DEBUG_PRINTF("imm:%lx\n", imm); + return imm; + } else { + FLAG_riscv_debug = true; + disassembleInstr(inst0->InstructionBits()); + disassembleInstr(instr1->InstructionBits()); + disassembleInstr(instr2->InstructionBits()); + disassembleInstr(instr3->InstructionBits()); + disassembleInstr(instr4->InstructionBits()); + disassembleInstr(instr5->InstructionBits()); + disassembleInstr(instr6->InstructionBits()); + disassembleInstr(instr7->InstructionBits()); + MOZ_CRASH(); + } + } else { + DEBUG_PRINTF("\n"); + Instruction* instrf1 = (inst0 - 1 * kInstrSize); + Instruction* instr2 = inst0 + 2 * kInstrSize; + Instruction* instr3 = inst0 + 3 * kInstrSize; + Instruction* instr4 = inst0 + 4 * kInstrSize; + Instruction* instr5 = inst0 + 5 * kInstrSize; + Instruction* instr6 = inst0 + 6 * kInstrSize; + Instruction* instr7 = inst0 + 7 * kInstrSize; + disassembleInstr(instrf1->InstructionBits()); + disassembleInstr(inst0->InstructionBits()); + disassembleInstr(instr1->InstructionBits()); + disassembleInstr(instr2->InstructionBits()); + disassembleInstr(instr3->InstructionBits()); + disassembleInstr(instr4->InstructionBits()); + disassembleInstr(instr5->InstructionBits()); + disassembleInstr(instr6->InstructionBits()); + disassembleInstr(instr7->InstructionBits()); + MOZ_ASSERT(IsAddi(*reinterpret_cast<Instr*>(instr1))); + // Li48 + return target_address_at(inst0); + } +} + +void Assembler::UpdateLoad64Value(Instruction* pc, uint64_t value) { + DEBUG_PRINTF("\tUpdateLoad64Value: pc: %p\tvalue: %lx\n", pc, value); + Instruction* instr1 = pc + 1 * kInstrSize; + if (IsJal(*reinterpret_cast<Instr*>(pc))) { + pc = pc + pc->Imm20JValue(); + instr1 = pc + 1 * kInstrSize; + } + if (IsAddiw(*reinterpret_cast<Instr*>(instr1))) { + Instruction* instr0 = pc; + Instruction* instr2 = pc + 2 * kInstrSize; + Instruction* instr3 = pc + 3 * kInstrSize; + Instruction* instr4 = pc + 4 * kInstrSize; + Instruction* instr5 = pc + 5 * kInstrSize; + Instruction* instr6 = pc + 6 * kInstrSize; + Instruction* instr7 = pc + 7 * kInstrSize; + MOZ_ASSERT(IsLui(*reinterpret_cast<Instr*>(pc)) && + IsAddiw(*reinterpret_cast<Instr*>(instr1)) && + IsSlli(*reinterpret_cast<Instr*>(instr2)) && + IsAddi(*reinterpret_cast<Instr*>(instr3)) && + IsSlli(*reinterpret_cast<Instr*>(instr4)) && + IsAddi(*reinterpret_cast<Instr*>(instr5)) && + IsSlli(*reinterpret_cast<Instr*>(instr6)) && + IsAddi(*reinterpret_cast<Instr*>(instr7))); + // lui(rd, (imm + (1LL << 47) + (1LL << 35) + (1LL << 23) + (1LL << 11)) >> + // 48); // Bits 63:48 + // addiw(rd, rd, + // (imm + (1LL << 35) + (1LL << 23) + (1LL << 11)) << 16 >> + // 52); // Bits 47:36 + // slli(rd, rd, 12); + // addi(rd, rd, (imm + (1LL << 23) + (1LL << 11)) << 28 >> 52); // Bits + // 35:24 slli(rd, rd, 12); addi(rd, rd, (imm + (1LL << 11)) << 40 >> 52); // + // Bits 23:12 slli(rd, rd, 12); addi(rd, rd, imm << 52 >> 52); // Bits 11:0 + *reinterpret_cast<Instr*>(instr0) &= 0xfff; + *reinterpret_cast<Instr*>(instr0) |= + (((value + (1LL << 47) + (1LL << 35) + (1LL << 23) + (1LL << 11)) >> 48) + << 12); + *reinterpret_cast<Instr*>(instr1) &= 0xfffff; + *reinterpret_cast<Instr*>(instr1) |= + (((value + (1LL << 35) + (1LL << 23) + (1LL << 11)) << 16 >> 52) << 20); + *reinterpret_cast<Instr*>(instr3) &= 0xfffff; + *reinterpret_cast<Instr*>(instr3) |= + (((value + (1LL << 23) + (1LL << 11)) << 28 >> 52) << 20); + *reinterpret_cast<Instr*>(instr5) &= 0xfffff; + *reinterpret_cast<Instr*>(instr5) |= + (((value + (1LL << 11)) << 40 >> 52) << 20); + *reinterpret_cast<Instr*>(instr7) &= 0xfffff; + *reinterpret_cast<Instr*>(instr7) |= ((value << 52 >> 52) << 20); + disassembleInstr(instr0->InstructionBits()); + disassembleInstr(instr1->InstructionBits()); + disassembleInstr(instr2->InstructionBits()); + disassembleInstr(instr3->InstructionBits()); + disassembleInstr(instr4->InstructionBits()); + disassembleInstr(instr5->InstructionBits()); + disassembleInstr(instr6->InstructionBits()); + disassembleInstr(instr7->InstructionBits()); + MOZ_ASSERT(ExtractLoad64Value(pc) == value); + } else { + Instruction* instr0 = pc; + Instruction* instr2 = pc + 2 * kInstrSize; + Instruction* instr3 = pc + 3 * kInstrSize; + Instruction* instr4 = pc + 4 * kInstrSize; + Instruction* instr5 = pc + 5 * kInstrSize; + Instruction* instr6 = pc + 6 * kInstrSize; + Instruction* instr7 = pc + 7 * kInstrSize; + disassembleInstr(instr0->InstructionBits()); + disassembleInstr(instr1->InstructionBits()); + disassembleInstr(instr2->InstructionBits()); + disassembleInstr(instr3->InstructionBits()); + disassembleInstr(instr4->InstructionBits()); + disassembleInstr(instr5->InstructionBits()); + disassembleInstr(instr6->InstructionBits()); + disassembleInstr(instr7->InstructionBits()); + MOZ_ASSERT(IsAddi(*reinterpret_cast<Instr*>(instr1))); + set_target_value_at(pc, value); + } +} + +void Assembler::set_target_value_at(Instruction* pc, uint64_t target) { + DEBUG_PRINTF("\tset_target_value_at: pc: %p\ttarget: %lx\n", pc, target); + uint32_t* p = reinterpret_cast<uint32_t*>(pc); + MOZ_ASSERT((target & 0xffff000000000000ll) == 0); +#ifdef DEBUG + // Check we have the result from a li macro-instruction. + Instruction* instr0 = pc; + Instruction* instr1 = pc + 1 * kInstrSize; + Instruction* instr3 = pc + 3 * kInstrSize; + Instruction* instr5 = pc + 5 * kInstrSize; + MOZ_ASSERT(IsLui(*reinterpret_cast<Instr*>(instr0)) && + IsAddi(*reinterpret_cast<Instr*>(instr1)) && + IsOri(*reinterpret_cast<Instr*>(instr3)) && + IsOri(*reinterpret_cast<Instr*>(instr5))); +#endif + int64_t a6 = target & 0x3f; // bits 0:6. 6 bits + int64_t b11 = (target >> 6) & 0x7ff; // bits 6:11. 11 bits + int64_t high_31 = (target >> 17) & 0x7fffffff; // 31 bits + int64_t high_20 = ((high_31 + 0x800) >> 12); // 19 bits + int64_t low_12 = high_31 & 0xfff; // 12 bits + *p = *p & 0xfff; + *p = *p | ((int32_t)high_20 << 12); + *(p + 1) = *(p + 1) & 0xfffff; + *(p + 1) = *(p + 1) | ((int32_t)low_12 << 20); + *(p + 2) = *(p + 2) & 0xfffff; + *(p + 2) = *(p + 2) | (11 << 20); + *(p + 3) = *(p + 3) & 0xfffff; + *(p + 3) = *(p + 3) | ((int32_t)b11 << 20); + *(p + 4) = *(p + 4) & 0xfffff; + *(p + 4) = *(p + 4) | (6 << 20); + *(p + 5) = *(p + 5) & 0xfffff; + *(p + 5) = *(p + 5) | ((int32_t)a6 << 20); + MOZ_ASSERT(target_address_at(pc) == target); +} + +void Assembler::WriteLoad64Instructions(Instruction* inst0, Register reg, + uint64_t value) { + DEBUG_PRINTF("\tWriteLoad64Instructions\n"); + // Initialize rd with an address + // Pointers are 48 bits + // 6 fixed instructions are generated + MOZ_ASSERT((value & 0xfff0000000000000ll) == 0); + int64_t a6 = value & 0x3f; // bits 0:5. 6 bits + int64_t b11 = (value >> 6) & 0x7ff; // bits 6:11. 11 bits + int64_t high_31 = (value >> 17) & 0x7fffffff; // 31 bits + int64_t high_20 = ((high_31 + 0x800) >> 12); // 19 bits + int64_t low_12 = high_31 & 0xfff; // 12 bits + Instr lui_ = LUI | (reg.code() << kRdShift) | + ((int32_t)high_20 << kImm20Shift); // lui(rd, (int32_t)high_20); + *reinterpret_cast<Instr*>(inst0) = lui_; + + Instr addi_ = + OP_IMM | (reg.code() << kRdShift) | (0b000 << kFunct3Shift) | + (reg.code() << kRs1Shift) | + (low_12 << kImm12Shift); // addi(rd, rd, low_12); // 31 bits in rd. + *reinterpret_cast<Instr*>(inst0 + 1 * kInstrSize) = addi_; + + Instr slli_ = + OP_IMM | (reg.code() << kRdShift) | (0b001 << kFunct3Shift) | + (reg.code() << kRs1Shift) | + (11 << kImm12Shift); // slli(rd, rd, 11); // Space for next 11 bis + *reinterpret_cast<Instr*>(inst0 + 2 * kInstrSize) = slli_; + + Instr ori_b11 = OP_IMM | (reg.code() << kRdShift) | (0b110 << kFunct3Shift) | + (reg.code() << kRs1Shift) | + (b11 << kImm12Shift); // ori(rd, rd, b11); // 11 bits + // are put in. 42 bit in rd + *reinterpret_cast<Instr*>(inst0 + 3 * kInstrSize) = ori_b11; + + slli_ = OP_IMM | (reg.code() << kRdShift) | (0b001 << kFunct3Shift) | + (reg.code() << kRs1Shift) | + (6 << kImm12Shift); // slli(rd, rd, 6); // Space for next 11 bis + *reinterpret_cast<Instr*>(inst0 + 4 * kInstrSize) = + slli_; // slli(rd, rd, 6); // Space for next 6 bits + + Instr ori_a6 = OP_IMM | (reg.code() << kRdShift) | (0b110 << kFunct3Shift) | + (reg.code() << kRs1Shift) | + (a6 << kImm12Shift); // ori(rd, rd, a6); // 6 bits are + // put in. 48 bis in rd + *reinterpret_cast<Instr*>(inst0 + 5 * kInstrSize) = ori_a6; + disassembleInstr((inst0 + 0 * kInstrSize)->InstructionBits()); + disassembleInstr((inst0 + 1 * kInstrSize)->InstructionBits()); + disassembleInstr((inst0 + 2 * kInstrSize)->InstructionBits()); + disassembleInstr((inst0 + 3 * kInstrSize)->InstructionBits()); + disassembleInstr((inst0 + 4 * kInstrSize)->InstructionBits()); + disassembleInstr((inst0 + 5 * kInstrSize)->InstructionBits()); + disassembleInstr((inst0 + 6 * kInstrSize)->InstructionBits()); + MOZ_ASSERT(ExtractLoad64Value(inst0) == value); +} + +// This just stomps over memory with 32 bits of raw data. Its purpose is to +// overwrite the call of JITed code with 32 bits worth of an offset. This will +// is only meant to function on code that has been invalidated, so it should +// be totally safe. Since that instruction will never be executed again, a +// ICache flush should not be necessary +void Assembler::PatchWrite_Imm32(CodeLocationLabel label, Imm32 imm) { + // Raw is going to be the return address. + uint32_t* raw = (uint32_t*)label.raw(); + // Overwrite the 4 bytes before the return address, which will + // end up being the call instruction. + *(raw - 1) = imm.value; +} + +void Assembler::target_at_put(BufferOffset pos, BufferOffset target_pos, + bool trampoline) { + if (m_buffer.oom()) { + return; + } + DEBUG_PRINTF("\ttarget_at_put: %p (%d) to %p (%d)\n", + reinterpret_cast<Instr*>(editSrc(pos)), pos.getOffset(), + reinterpret_cast<Instr*>(editSrc(pos)) + target_pos.getOffset() - + pos.getOffset(), + target_pos.getOffset()); + Instruction* instruction = editSrc(pos); + Instr instr = instruction->InstructionBits(); + switch (instruction->InstructionOpcodeType()) { + case BRANCH: { + instr = SetBranchOffset(pos.getOffset(), target_pos.getOffset(), instr); + instr_at_put(pos, instr); + } break; + case JAL: { + MOZ_ASSERT(IsJal(instr)); + instr = SetJalOffset(pos.getOffset(), target_pos.getOffset(), instr); + instr_at_put(pos, instr); + } break; + case LUI: { + set_target_value_at(instruction, + reinterpret_cast<uintptr_t>(editSrc(target_pos))); + } break; + case AUIPC: { + Instr instr_auipc = instr; + Instr instr_I = + editSrc(BufferOffset(pos.getOffset() + 4))->InstructionBits(); + MOZ_ASSERT(IsJalr(instr_I) || IsAddi(instr_I)); + + intptr_t offset = target_pos.getOffset() - pos.getOffset(); + if (is_int21(offset) && IsJalr(instr_I) && trampoline) { + MOZ_ASSERT(is_int21(offset) && ((offset & 1) == 0)); + Instr instr = JAL; + instr = SetJalOffset(pos.getOffset(), target_pos.getOffset(), instr); + MOZ_ASSERT(IsJal(instr)); + MOZ_ASSERT(JumpOffset(instr) == offset); + instr_at_put(pos, instr); + instr_at_put(BufferOffset(pos.getOffset() + 4), kNopByte); + } else { + MOZ_RELEASE_ASSERT(is_int32(offset + 0x800)); + MOZ_ASSERT(instruction->RdValue() == + editSrc(BufferOffset(pos.getOffset() + 4))->Rs1Value()); + int32_t Hi20 = (((int32_t)offset + 0x800) >> 12); + int32_t Lo12 = (int32_t)offset << 20 >> 20; + + instr_auipc = + (instr_auipc & ~kImm31_12Mask) | ((Hi20 & kImm19_0Mask) << 12); + instr_at_put(pos, instr_auipc); + + const int kImm31_20Mask = ((1 << 12) - 1) << 20; + const int kImm11_0Mask = ((1 << 12) - 1); + instr_I = (instr_I & ~kImm31_20Mask) | ((Lo12 & kImm11_0Mask) << 20); + instr_at_put(BufferOffset(pos.getOffset() + 4), instr_I); + } + } break; + default: + UNIMPLEMENTED_RISCV(); + break; + } +} + +const int kEndOfChain = -1; +const int32_t kEndOfJumpChain = 0; + +int Assembler::target_at(BufferOffset pos, bool is_internal) { + if (oom()) { + return kEndOfChain; + } + Instruction* instruction = editSrc(pos); + Instruction* instruction2 = nullptr; + if (IsAuipc(instruction->InstructionBits())) { + instruction2 = editSrc(BufferOffset(pos.getOffset() + kInstrSize)); + } + return target_at(instruction, pos, is_internal, instruction2); +} + +int Assembler::target_at(Instruction* instruction, BufferOffset pos, + bool is_internal, Instruction* instruction2) { + DEBUG_PRINTF("\t target_at: %p(%x)\n\t", + reinterpret_cast<Instr*>(instruction), pos.getOffset()); + disassembleInstr(instruction->InstructionBits()); + Instr instr = instruction->InstructionBits(); + switch (instruction->InstructionOpcodeType()) { + case BRANCH: { + int32_t imm13 = BranchOffset(instr); + if (imm13 == kEndOfJumpChain) { + // EndOfChain sentinel is returned directly, not relative to pc or pos. + return kEndOfChain; + } else { + DEBUG_PRINTF("\t target_at: %d %d\n", imm13, pos.getOffset() + imm13); + return pos.getOffset() + imm13; + } + } + case JAL: { + int32_t imm21 = JumpOffset(instr); + if (imm21 == kEndOfJumpChain) { + // EndOfChain sentinel is returned directly, not relative to pc or pos. + return kEndOfChain; + } else { + DEBUG_PRINTF("\t target_at: %d %d\n", imm21, pos.getOffset() + imm21); + return pos.getOffset() + imm21; + } + } + case JALR: { + int32_t imm12 = instr >> 20; + if (imm12 == kEndOfJumpChain) { + // EndOfChain sentinel is returned directly, not relative to pc or pos. + return kEndOfChain; + } else { + DEBUG_PRINTF("\t target_at: %d %d\n", imm12, pos.getOffset() + imm12); + return pos.getOffset() + imm12; + } + } + case LUI: { + uintptr_t imm = target_address_at(instruction); + uintptr_t instr_address = reinterpret_cast<uintptr_t>(instruction); + if (imm == kEndOfJumpChain) { + return kEndOfChain; + } else { + MOZ_ASSERT(instr_address - imm < INT_MAX); + int32_t delta = static_cast<int32_t>(instr_address - imm); + MOZ_ASSERT(pos.getOffset() > delta); + return pos.getOffset() - delta; + } + } + case AUIPC: { + MOZ_ASSERT(instruction2 != nullptr); + Instr instr_auipc = instr; + Instr instr_I = instruction2->InstructionBits(); + MOZ_ASSERT(IsJalr(instr_I) || IsAddi(instr_I)); + int32_t offset = BrachlongOffset(instr_auipc, instr_I); + if (offset == kEndOfJumpChain) return kEndOfChain; + DEBUG_PRINTF("\t target_at: %d %d\n", offset, pos.getOffset() + offset); + return offset + pos.getOffset(); + } + default: { + UNIMPLEMENTED_RISCV(); + } + } +} + +uint32_t Assembler::next_link(Label* L, bool is_internal) { + MOZ_ASSERT(L->used()); + BufferOffset pos(L); + int link = target_at(pos, is_internal); + if (link == kEndOfChain) { + L->reset(); + return LabelBase::INVALID_OFFSET; + } else { + MOZ_ASSERT(link >= 0); + DEBUG_PRINTF("next: %p to offset %d\n", L, link); + L->use(link); + return link; + } +} + +void Assembler::bind(Label* label, BufferOffset boff) { + JitSpew(JitSpew_Codegen, ".set Llabel %p %d", label, currentOffset()); + DEBUG_PRINTF(".set Llabel %p\n", label); + // If our caller didn't give us an explicit target to bind to + // then we want to bind to the location of the next instruction + BufferOffset dest = boff.assigned() ? boff : nextOffset(); + if (label->used()) { + uint32_t next; + + // A used label holds a link to branch that uses it. + do { + BufferOffset b(label); + DEBUG_PRINTF("\tbind next:%d\n", b.getOffset()); + // Even a 0 offset may be invalid if we're out of memory. + if (oom()) { + return; + } + int fixup_pos = b.getOffset(); + int dist = dest.getOffset() - fixup_pos; + next = next_link(label, false); + DEBUG_PRINTF("\t%p fixup: %d next: %d\n", label, fixup_pos, next); + DEBUG_PRINTF("\t fixup: %d dest: %d dist: %d %d %d\n", fixup_pos, + dest.getOffset(), dist, nextOffset().getOffset(), + currentOffset()); + Instruction* instruction = editSrc(b); + Instr instr = instruction->InstructionBits(); + if (IsBranch(instr)) { + if (dist > kMaxBranchOffset) { + MOZ_ASSERT(next != LabelBase::INVALID_OFFSET); + MOZ_RELEASE_ASSERT((next - fixup_pos) <= kMaxBranchOffset); + MOZ_ASSERT(IsAuipc(editSrc(BufferOffset(next))->InstructionBits())); + MOZ_ASSERT( + IsJalr(editSrc(BufferOffset(next + 4))->InstructionBits())); + DEBUG_PRINTF("\t\ttrampolining: %d\n", next); + } else { + target_at_put(b, dest); + BufferOffset deadline(b.getOffset() + + ImmBranchMaxForwardOffset(CondBranchRangeType)); + m_buffer.unregisterBranchDeadline(CondBranchRangeType, deadline); + } + } else if (IsJal(instr)) { + if (dist > kMaxJumpOffset) { + MOZ_ASSERT(next != LabelBase::INVALID_OFFSET); + MOZ_RELEASE_ASSERT((next - fixup_pos) <= kMaxJumpOffset); + MOZ_ASSERT(IsAuipc(editSrc(BufferOffset(next))->InstructionBits())); + MOZ_ASSERT( + IsJalr(editSrc(BufferOffset(next + 4))->InstructionBits())); + DEBUG_PRINTF("\t\ttrampolining: %d\n", next); + } else { + target_at_put(b, dest); + BufferOffset deadline( + b.getOffset() + ImmBranchMaxForwardOffset(UncondBranchRangeType)); + m_buffer.unregisterBranchDeadline(UncondBranchRangeType, deadline); + } + } else { + MOZ_ASSERT(IsAuipc(instr)); + target_at_put(b, dest); + } + } while (next != LabelBase::INVALID_OFFSET); + } + label->bind(dest.getOffset()); +} + +void Assembler::Bind(uint8_t* rawCode, const CodeLabel& label) { + if (label.patchAt().bound()) { + auto mode = label.linkMode(); + intptr_t offset = label.patchAt().offset(); + intptr_t target = label.target().offset(); + + if (mode == CodeLabel::RawPointer) { + *reinterpret_cast<const void**>(rawCode + offset) = rawCode + target; + } else { + MOZ_ASSERT(mode == CodeLabel::MoveImmediate || + mode == CodeLabel::JumpImmediate); + Instruction* inst = (Instruction*)(rawCode + offset); + Assembler::UpdateLoad64Value(inst, (uint64_t)(rawCode + target)); + } + } +} + +bool Assembler::is_near(Label* L) { + MOZ_ASSERT(L->bound()); + return is_intn((currentOffset() - L->offset()), kJumpOffsetBits); +} + +bool Assembler::is_near(Label* L, OffsetSize bits) { + if (L == nullptr || !L->bound()) return true; + return is_intn((currentOffset() - L->offset()), bits); +} + +bool Assembler::is_near_branch(Label* L) { + MOZ_ASSERT(L->bound()); + return is_intn((currentOffset() - L->offset()), kBranchOffsetBits); +} + +int32_t Assembler::branch_long_offset(Label* L) { + if (oom()) { + return kEndOfJumpChain; + } + intptr_t target_pos; + BufferOffset next_instr_offset = nextInstrOffset(2); + DEBUG_PRINTF("\tbranch_long_offset: %p to (%d)\n", L, + next_instr_offset.getOffset()); + if (L->bound()) { + JitSpew(JitSpew_Codegen, ".use Llabel %p on %d", L, + next_instr_offset.getOffset()); + target_pos = L->offset(); + } else { + if (L->used()) { + LabelCahe::Ptr p = label_cache_.lookup(L->offset()); + MOZ_ASSERT(p); + MOZ_ASSERT(p->key() == L->offset()); + target_pos = p->value().getOffset(); + target_at_put(BufferOffset(target_pos), next_instr_offset); + DEBUG_PRINTF("\tLabel %p added to link: %d\n", L, + next_instr_offset.getOffset()); + bool ok = label_cache_.put(L->offset(), next_instr_offset); + if (!ok) { + NoEnoughLabelCache(); + } + return kEndOfJumpChain; + } else { + JitSpew(JitSpew_Codegen, ".use Llabel %p on %d", L, + next_instr_offset.getOffset()); + L->use(next_instr_offset.getOffset()); + DEBUG_PRINTF("\tLabel %p added to link: %d\n", L, + next_instr_offset.getOffset()); + bool ok = label_cache_.putNew(L->offset(), next_instr_offset); + if (!ok) { + NoEnoughLabelCache(); + } + return kEndOfJumpChain; + } + } + intptr_t offset = target_pos - next_instr_offset.getOffset(); + MOZ_ASSERT((offset & 3) == 0); + MOZ_ASSERT(is_int32(offset)); + return static_cast<int32_t>(offset); +} + +int32_t Assembler::branch_offset_helper(Label* L, OffsetSize bits) { + if (oom()) { + return kEndOfJumpChain; + } + int32_t target_pos; + BufferOffset next_instr_offset = nextInstrOffset(); + DEBUG_PRINTF("\tbranch_offset_helper: %p to %d\n", L, + next_instr_offset.getOffset()); + // This is the last possible branch target. + if (L->bound()) { + JitSpew(JitSpew_Codegen, ".use Llabel %p on %d", L, + next_instr_offset.getOffset()); + target_pos = L->offset(); + } else { + BufferOffset deadline(next_instr_offset.getOffset() + + ImmBranchMaxForwardOffset(bits)); + DEBUG_PRINTF("\tregisterBranchDeadline %d type %d\n", deadline.getOffset(), + OffsetSizeToImmBranchRangeType(bits)); + m_buffer.registerBranchDeadline(OffsetSizeToImmBranchRangeType(bits), + deadline); + if (L->used()) { + LabelCahe::Ptr p = label_cache_.lookup(L->offset()); + MOZ_ASSERT(p); + MOZ_ASSERT(p->key() == L->offset()); + target_pos = p->value().getOffset(); + target_at_put(BufferOffset(target_pos), next_instr_offset); + DEBUG_PRINTF("\tLabel %p added to link: %d\n", L, + next_instr_offset.getOffset()); + bool ok = label_cache_.put(L->offset(), next_instr_offset); + if (!ok) { + NoEnoughLabelCache(); + } + return kEndOfJumpChain; + } else { + JitSpew(JitSpew_Codegen, ".use Llabel %p on %d", L, + next_instr_offset.getOffset()); + L->use(next_instr_offset.getOffset()); + bool ok = label_cache_.putNew(L->offset(), next_instr_offset); + if (!ok) { + NoEnoughLabelCache(); + } + DEBUG_PRINTF("\tLabel %p added to link: %d\n", L, + next_instr_offset.getOffset()); + return kEndOfJumpChain; + } + } + + int32_t offset = target_pos - next_instr_offset.getOffset(); + DEBUG_PRINTF("\toffset = %d\n", offset); + MOZ_ASSERT(is_intn(offset, bits)); + MOZ_ASSERT((offset & 1) == 0); + return offset; +} + +Assembler::Condition Assembler::InvertCondition(Condition cond) { + switch (cond) { + case Equal: + return NotEqual; + case NotEqual: + return Equal; + case Zero: + return NonZero; + case NonZero: + return Zero; + case LessThan: + return GreaterThanOrEqual; + case LessThanOrEqual: + return GreaterThan; + case GreaterThan: + return LessThanOrEqual; + case GreaterThanOrEqual: + return LessThan; + case Above: + return BelowOrEqual; + case AboveOrEqual: + return Below; + case Below: + return AboveOrEqual; + case BelowOrEqual: + return Above; + case Signed: + return NotSigned; + case NotSigned: + return Signed; + default: + MOZ_CRASH("unexpected condition"); + } +} + +Assembler::DoubleCondition Assembler::InvertCondition(DoubleCondition cond) { + switch (cond) { + case DoubleOrdered: + return DoubleUnordered; + case DoubleEqual: + return DoubleNotEqualOrUnordered; + case DoubleNotEqual: + return DoubleEqualOrUnordered; + case DoubleGreaterThan: + return DoubleLessThanOrEqualOrUnordered; + case DoubleGreaterThanOrEqual: + return DoubleLessThanOrUnordered; + case DoubleLessThan: + return DoubleGreaterThanOrEqualOrUnordered; + case DoubleLessThanOrEqual: + return DoubleGreaterThanOrUnordered; + case DoubleUnordered: + return DoubleOrdered; + case DoubleEqualOrUnordered: + return DoubleNotEqual; + case DoubleNotEqualOrUnordered: + return DoubleEqual; + case DoubleGreaterThanOrUnordered: + return DoubleLessThanOrEqual; + case DoubleGreaterThanOrEqualOrUnordered: + return DoubleLessThan; + case DoubleLessThanOrUnordered: + return DoubleGreaterThanOrEqual; + case DoubleLessThanOrEqualOrUnordered: + return DoubleGreaterThan; + default: + MOZ_CRASH("unexpected condition"); + } +} + +// Break / Trap instructions. +void Assembler::break_(uint32_t code, bool break_as_stop) { + // We need to invalidate breaks that could be stops as well because the + // simulator expects a char pointer after the stop instruction. + // See constants-mips.h for explanation. + MOZ_ASSERT( + (break_as_stop && code <= kMaxStopCode && code > kMaxTracepointCode) || + (!break_as_stop && (code > kMaxStopCode || code <= kMaxTracepointCode))); + + // since ebreak does not allow additional immediate field, we use the + // immediate field of lui instruction immediately following the ebreak to + // encode the "code" info + ebreak(); + MOZ_ASSERT(is_uint20(code)); + lui(zero_reg, code); +} + +void Assembler::ToggleToJmp(CodeLocationLabel inst_) { + Instruction* inst = (Instruction*)inst_.raw(); + MOZ_ASSERT(IsAddi(inst->InstructionBits())); + int32_t offset = inst->Imm12Value(); + MOZ_ASSERT(is_int12(offset)); + Instr jal_ = JAL | (0b000 << kFunct3Shift) | + (offset & 0xff000) | // bits 19-12 + ((offset & 0x800) << 9) | // bit 11 + ((offset & 0x7fe) << 20) | // bits 10-1 + ((offset & 0x100000) << 11); // bit 20 + // jal(zero, offset); + *reinterpret_cast<Instr*>(inst) = jal_; +} + +void Assembler::ToggleToCmp(CodeLocationLabel inst_) { + Instruction* inst = (Instruction*)inst_.raw(); + + // toggledJump is allways used for short jumps. + MOZ_ASSERT(IsJal(inst->InstructionBits())); + // Replace "jal zero_reg, offset" with "addi $zero, $zero, offset" + int32_t offset = inst->Imm20JValue(); + MOZ_ASSERT(is_int12(offset)); + Instr addi_ = OP_IMM | (0b000 << kFunct3Shift) | + (offset << kImm12Shift); // addi(zero, zero, low_12); + *reinterpret_cast<Instr*>(inst) = addi_; +} + +bool Assembler::reserve(size_t size) { + // This buffer uses fixed-size chunks so there's no point in reserving + // now vs. on-demand. + return !oom(); +} + +static JitCode* CodeFromJump(Instruction* jump) { + uint8_t* target = (uint8_t*)Assembler::ExtractLoad64Value(jump); + return JitCode::FromExecutable(target); +} + +void Assembler::TraceJumpRelocations(JSTracer* trc, JitCode* code, + CompactBufferReader& reader) { + while (reader.more()) { + JitCode* child = + CodeFromJump((Instruction*)(code->raw() + reader.readUnsigned())); + TraceManuallyBarrieredEdge(trc, &child, "rel32"); + } +} + +static void TraceOneDataRelocation(JSTracer* trc, + mozilla::Maybe<AutoWritableJitCode>& awjc, + JitCode* code, Instruction* inst) { + void* ptr = (void*)Assembler::ExtractLoad64Value(inst); + void* prior = ptr; + + // Data relocations can be for Values or for raw pointers. If a Value is + // zero-tagged, we can trace it as if it were a raw pointer. If a Value + // is not zero-tagged, we have to interpret it as a Value to ensure that the + // tag bits are masked off to recover the actual pointer. + uintptr_t word = reinterpret_cast<uintptr_t>(ptr); + if (word >> JSVAL_TAG_SHIFT) { + // This relocation is a Value with a non-zero tag. + Value v = Value::fromRawBits(word); + TraceManuallyBarrieredEdge(trc, &v, "jit-masm-value"); + ptr = (void*)v.bitsAsPunboxPointer(); + } else { + // This relocation is a raw pointer or a Value with a zero tag. + // No barrier needed since these are constants. + TraceManuallyBarrieredGenericPointerEdge( + trc, reinterpret_cast<gc::Cell**>(&ptr), "jit-masm-ptr"); + } + + if (ptr != prior) { + if (awjc.isNothing()) { + awjc.emplace(code); + } + Assembler::UpdateLoad64Value(inst, uint64_t(ptr)); + } +} + +/* static */ +void Assembler::TraceDataRelocations(JSTracer* trc, JitCode* code, + CompactBufferReader& reader) { + mozilla::Maybe<AutoWritableJitCode> awjc; + while (reader.more()) { + size_t offset = reader.readUnsigned(); + Instruction* inst = (Instruction*)(code->raw() + offset); + TraceOneDataRelocation(trc, awjc, code, inst); + } +} + +UseScratchRegisterScope::UseScratchRegisterScope(Assembler* assembler) + : available_(assembler->GetScratchRegisterList()), + old_available_(*available_) {} + +UseScratchRegisterScope::~UseScratchRegisterScope() { + *available_ = old_available_; +} + +Register UseScratchRegisterScope::Acquire() { + MOZ_ASSERT(available_ != nullptr); + MOZ_ASSERT(!available_->empty()); + Register index = GeneralRegisterSet::FirstRegister(available_->bits()); + available_->takeRegisterIndex(index); + return index; +} + +bool UseScratchRegisterScope::hasAvailable() const { + return (available_->size()) != 0; +} + +void Assembler::retarget(Label* label, Label* target) { + spew("retarget %p -> %p", label, target); + if (label->used() && !oom()) { + if (target->bound()) { + bind(label, BufferOffset(target)); + } else if (target->used()) { + // The target is not bound but used. Prepend label's branch list + // onto target's. + int32_t next; + BufferOffset labelBranchOffset(label); + + // Find the head of the use chain for label. + do { + next = next_link(label, false); + labelBranchOffset = BufferOffset(next); + } while (next != LabelBase::INVALID_OFFSET); + + // Then patch the head of label's use chain to the tail of + // target's use chain, prepending the entire use chain of target. + target->use(label->offset()); + target_at_put(labelBranchOffset, BufferOffset(target)); + MOZ_CRASH("check"); + } else { + // The target is unbound and unused. We can just take the head of + // the list hanging off of label, and dump that into target. + target->use(label->offset()); + } + } + label->reset(); +} + +bool Assembler::appendRawCode(const uint8_t* code, size_t numBytes) { + if (m_buffer.oom()) { + return false; + } + while (numBytes > SliceSize) { + m_buffer.putBytes(SliceSize, code); + numBytes -= SliceSize; + code += SliceSize; + } + m_buffer.putBytes(numBytes, code); + return !m_buffer.oom(); +} + +void Assembler::ToggleCall(CodeLocationLabel inst_, bool enabled) { + Instruction* i0 = (Instruction*)inst_.raw(); + Instruction* i1 = (Instruction*)(inst_.raw() + 1 * kInstrSize); + Instruction* i2 = (Instruction*)(inst_.raw() + 2 * kInstrSize); + Instruction* i3 = (Instruction*)(inst_.raw() + 3 * kInstrSize); + Instruction* i4 = (Instruction*)(inst_.raw() + 4 * kInstrSize); + Instruction* i5 = (Instruction*)(inst_.raw() + 5 * kInstrSize); + Instruction* i6 = (Instruction*)(inst_.raw() + 6 * kInstrSize); + + MOZ_ASSERT(IsLui(i0->InstructionBits())); + MOZ_ASSERT(IsAddi(i1->InstructionBits())); + MOZ_ASSERT(IsSlli(i2->InstructionBits())); + MOZ_ASSERT(IsOri(i3->InstructionBits())); + MOZ_ASSERT(IsSlli(i4->InstructionBits())); + MOZ_ASSERT(IsOri(i5->InstructionBits())); + if (enabled) { + Instr jalr_ = JALR | (ra.code() << kRdShift) | (0x0 << kFunct3Shift) | + (i5->RdValue() << kRs1Shift) | (0x0 << kImm12Shift); + *((Instr*)i6) = jalr_; + } else { + *((Instr*)i6) = kNopByte; + } +} + +void Assembler::PatchShortRangeBranchToVeneer(Buffer* buffer, unsigned rangeIdx, + BufferOffset deadline, + BufferOffset veneer) { + if (buffer->oom()) { + return; + } + DEBUG_PRINTF("\tPatchShortRangeBranchToVeneer\n"); + // Reconstruct the position of the branch from (rangeIdx, deadline). + ImmBranchRangeType branchRange = static_cast<ImmBranchRangeType>(rangeIdx); + BufferOffset branch(deadline.getOffset() - + ImmBranchMaxForwardOffset(branchRange)); + Instruction* branchInst = buffer->getInst(branch); + Instruction* veneerInst_1 = buffer->getInst(veneer); + Instruction* veneerInst_2 = + buffer->getInst(BufferOffset(veneer.getOffset() + 4)); + // Verify that the branch range matches what's encoded. + DEBUG_PRINTF("\t%p(%x): ", branchInst, branch.getOffset()); + disassembleInstr(branchInst->InstructionBits(), JitSpew_Codegen); + DEBUG_PRINTF("\t instert veneer %x, branch:%x deadline: %x\n", + veneer.getOffset(), branch.getOffset(), deadline.getOffset()); + MOZ_ASSERT(branchRange <= UncondBranchRangeType); + MOZ_ASSERT(branchInst->GetImmBranchRangeType() == branchRange); + // emit a long jump slot + Instr auipc = AUIPC | (t6.code() << kRdShift) | (0x0 << kImm20Shift); + Instr jalr = JALR | (zero_reg.code() << kRdShift) | (0x0 << kFunct3Shift) | + (t6.code() << kRs1Shift) | (0x0 << kImm12Shift); + + // We want to insert veneer after branch in the linked list of instructions + // that use the same unbound label. + // The veneer should be an unconditional branch. + int32_t nextElemOffset = target_at(buffer->getInst(branch), branch, false); + int32_t dist; + // If offset is 0, this is the end of the linked list. + if (nextElemOffset != kEndOfChain) { + // Make the offset relative to veneer so it targets the same instruction + // as branchInst. + dist = nextElemOffset - veneer.getOffset(); + } else { + dist = 0; + } + int32_t Hi20 = (((int32_t)dist + 0x800) >> 12); + int32_t Lo12 = (int32_t)dist << 20 >> 20; + auipc = SetAuipcOffset(Hi20, auipc); + jalr = SetJalrOffset(Lo12, jalr); + // insert veneer + veneerInst_1->SetInstructionBits(auipc); + veneerInst_2->SetInstructionBits(jalr); + // Now link branchInst to veneer. + if (IsBranch(branchInst->InstructionBits())) { + branchInst->SetInstructionBits(SetBranchOffset( + branch.getOffset(), veneer.getOffset(), branchInst->InstructionBits())); + } else { + MOZ_ASSERT(IsJal(branchInst->InstructionBits())); + branchInst->SetInstructionBits(SetJalOffset( + branch.getOffset(), veneer.getOffset(), branchInst->InstructionBits())); + } + DEBUG_PRINTF("\tfix to veneer:"); + disassembleInstr(branchInst->InstructionBits()); +} +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/Assembler-riscv64.h b/js/src/jit/riscv64/Assembler-riscv64.h new file mode 100644 index 0000000000..e8462626a7 --- /dev/null +++ b/js/src/jit/riscv64/Assembler-riscv64.h @@ -0,0 +1,690 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +// Copyright (c) 1994-2006 Sun Microsystems Inc. +// All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// - Redistributions of source code must retain the above copyright notice, +// this list of conditions and the following disclaimer. +// +// - Redistribution in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution. +// +// - Neither the name of Sun Microsystems or the names of contributors may +// be used to endorse or promote products derived from this software without +// specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS +// IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, +// THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR +// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +// The original source code covered by the above license above has been +// modified significantly by Google Inc. +// Copyright 2021 the V8 project authors. All rights reserved. + +#ifndef jit_riscv64_Assembler_riscv64_h +#define jit_riscv64_Assembler_riscv64_h + +#include "mozilla/Assertions.h" +#include "mozilla/Sprintf.h" + +#include <stdint.h> + +#include "jit/CompactBuffer.h" +#include "jit/JitCode.h" +#include "jit/JitSpewer.h" +#include "jit/Registers.h" +#include "jit/RegisterSets.h" +#include "jit/riscv64/Architecture-riscv64.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/extension/base-riscv-i.h" +#include "jit/riscv64/extension/extension-riscv-a.h" +#include "jit/riscv64/extension/extension-riscv-c.h" +#include "jit/riscv64/extension/extension-riscv-d.h" +#include "jit/riscv64/extension/extension-riscv-f.h" +#include "jit/riscv64/extension/extension-riscv-m.h" +#include "jit/riscv64/extension/extension-riscv-v.h" +#include "jit/riscv64/extension/extension-riscv-zicsr.h" +#include "jit/riscv64/extension/extension-riscv-zifencei.h" +#include "jit/riscv64/Register-riscv64.h" +#include "jit/shared/Assembler-shared.h" +#include "jit/shared/Disassembler-shared.h" +#include "jit/shared/IonAssemblerBufferWithConstantPools.h" +#include "js/HashTable.h" +#include "wasm/WasmTypeDecls.h" +namespace js { +namespace jit { + +struct ScratchFloat32Scope : public AutoFloatRegisterScope { + explicit ScratchFloat32Scope(MacroAssembler& masm) + : AutoFloatRegisterScope(masm, ScratchFloat32Reg) {} +}; + +struct ScratchDoubleScope : public AutoFloatRegisterScope { + explicit ScratchDoubleScope(MacroAssembler& masm) + : AutoFloatRegisterScope(masm, ScratchDoubleReg) {} +}; + +struct ScratchRegisterScope : public AutoRegisterScope { + explicit ScratchRegisterScope(MacroAssembler& masm) + : AutoRegisterScope(masm, ScratchRegister) {} +}; + +class MacroAssembler; + +inline Imm32 Imm64::secondHalf() const { return hi(); } +inline Imm32 Imm64::firstHalf() const { return low(); } + +static constexpr uint32_t ABIStackAlignment = 8; +static constexpr uint32_t CodeAlignment = 16; +static constexpr uint32_t JitStackAlignment = 8; +static constexpr uint32_t JitStackValueAlignment = + JitStackAlignment / sizeof(Value); +static const uint32_t WasmStackAlignment = 16; +static const uint32_t WasmTrapInstructionLength = 2 * sizeof(uint32_t); +// See comments in wasm::GenerateFunctionPrologue. The difference between these +// is the size of the largest callable prologue on the platform. +static constexpr uint32_t WasmCheckedCallEntryOffset = 0u; +static constexpr uint32_t WasmCheckedTailEntryOffset = 20u; + +static const Scale ScalePointer = TimesEight; + +class Assembler; + +static constexpr int32_t SliceSize = 1024; + +typedef js::jit::AssemblerBufferWithConstantPools< + SliceSize, 4, Instruction, Assembler, NumShortBranchRangeTypes> + Buffer; + +class Assembler : public AssemblerShared, + public AssemblerRISCVI, + public AssemblerRISCVA, + public AssemblerRISCVF, + public AssemblerRISCVD, + public AssemblerRISCVM, + public AssemblerRISCVC, + public AssemblerRISCVZicsr, + public AssemblerRISCVZifencei { + GeneralRegisterSet scratch_register_list_; + + static constexpr int kInvalidSlotPos = -1; + +#ifdef JS_JITSPEW + Sprinter* printer; +#endif + bool enoughLabelCache_ = true; + + protected: + using LabelOffset = int32_t; + using LabelCahe = + HashMap<LabelOffset, BufferOffset, js::DefaultHasher<LabelOffset>, + js::SystemAllocPolicy>; + LabelCahe label_cache_; + void NoEnoughLabelCache() { enoughLabelCache_ = false; } + CompactBufferWriter jumpRelocations_; + CompactBufferWriter dataRelocations_; + Buffer m_buffer; + bool isFinished = false; + Instruction* editSrc(BufferOffset bo) { return m_buffer.getInst(bo); } + + struct RelativePatch { + // the offset within the code buffer where the value is loaded that + // we want to fix-up + BufferOffset offset; + void* target; + RelocationKind kind; + + RelativePatch(BufferOffset offset, void* target, RelocationKind kind) + : offset(offset), target(target), kind(kind) {} + }; + + js::Vector<RelativePatch, 8, SystemAllocPolicy> jumps_; + + void addPendingJump(BufferOffset src, ImmPtr target, RelocationKind kind) { + enoughMemory_ &= jumps_.append(RelativePatch(src, target.value, kind)); + if (kind == RelocationKind::JITCODE) { + jumpRelocations_.writeUnsigned(src.getOffset()); + } + } + + void addLongJump(BufferOffset src, BufferOffset dst) { + CodeLabel cl; + cl.patchAt()->bind(src.getOffset()); + cl.target()->bind(dst.getOffset()); + cl.setLinkMode(CodeLabel::JumpImmediate); + addCodeLabel(std::move(cl)); + } + + public: + static bool FLAG_riscv_debug; + + Assembler() + : scratch_register_list_((1 << t5.code()) | (1 << t4.code()) | + (1 << t6.code())), +#ifdef JS_JITSPEW + printer(nullptr), +#endif + m_buffer(/*guardSize*/ 2, /*headerSize*/ 2, /*instBufferAlign*/ 8, + /*poolMaxOffset*/ GetPoolMaxOffset(), /*pcBias*/ 8, + /*alignFillInst*/ kNopByte, /*nopFillInst*/ kNopByte), + isFinished(false) { + } + static uint32_t NopFill; + static uint32_t AsmPoolMaxOffset; + static uint32_t GetPoolMaxOffset(); + bool reserve(size_t size); + bool oom() const; + void setPrinter(Sprinter* sp) { +#ifdef JS_JITSPEW + printer = sp; +#endif + } + void finish() { + MOZ_ASSERT(!isFinished); + isFinished = true; + } + void enterNoPool(size_t maxInst) { m_buffer.enterNoPool(maxInst); } + void leaveNoPool() { m_buffer.leaveNoPool(); } + bool swapBuffer(wasm::Bytes& bytes); + // Size of the instruction stream, in bytes. + size_t size() const; + // Size of the data table, in bytes. + size_t bytesNeeded() const; + // Size of the jump relocation table, in bytes. + size_t jumpRelocationTableBytes() const; + size_t dataRelocationTableBytes() const; + void copyJumpRelocationTable(uint8_t* dest); + void copyDataRelocationTable(uint8_t* dest); + // Copy the assembly code to the given buffer, and perform any pending + // relocations relying on the target address. + void executableCopy(uint8_t* buffer); + // API for speaking with the IonAssemblerBufferWithConstantPools generate an + // initial placeholder instruction that we want to later fix up. + static void InsertIndexIntoTag(uint8_t* load, uint32_t index); + static void PatchConstantPoolLoad(void* loadAddr, void* constPoolAddr); + // We're not tracking short-range branches for ARM for now. + static void PatchShortRangeBranchToVeneer(Buffer*, unsigned rangeIdx, + BufferOffset deadline, + BufferOffset veneer); + struct PoolHeader { + uint32_t data; + + struct Header { + // The size should take into account the pool header. + // The size is in units of Instruction (4bytes), not byte. + union { + struct { + uint32_t size : 15; + + // "Natural" guards are part of the normal instruction stream, + // while "non-natural" guards are inserted for the sole purpose + // of skipping around a pool. + uint32_t isNatural : 1; + uint32_t ONES : 16; + }; + uint32_t data; + }; + + Header(int size_, bool isNatural_) + : size(size_), isNatural(isNatural_), ONES(0xffff) {} + + Header(uint32_t data) : data(data) { + static_assert(sizeof(Header) == sizeof(uint32_t)); + MOZ_ASSERT(ONES == 0xffff); + } + + uint32_t raw() const { + static_assert(sizeof(Header) == sizeof(uint32_t)); + return data; + } + }; + + PoolHeader(int size_, bool isNatural_) + : data(Header(size_, isNatural_).raw()) {} + + uint32_t size() const { + Header tmp(data); + return tmp.size; + } + + uint32_t isNatural() const { + Header tmp(data); + return tmp.isNatural; + } + }; + + static void WritePoolHeader(uint8_t* start, Pool* p, bool isNatural); + static void WritePoolGuard(BufferOffset branch, Instruction* inst, + BufferOffset dest); + void processCodeLabels(uint8_t* rawCode); + BufferOffset nextOffset() { return m_buffer.nextOffset(); } + // Get the buffer offset of the next inserted instruction. This may flush + // constant pools. + BufferOffset nextInstrOffset(int numInstr = 1) { + return m_buffer.nextInstrOffset(numInstr); + } + void comment(const char* msg) { spew("; %s", msg); } + +#ifdef JS_JITSPEW + inline void spew(const char* fmt, ...) MOZ_FORMAT_PRINTF(2, 3) { + if (MOZ_UNLIKELY(printer || JitSpewEnabled(JitSpew_Codegen))) { + va_list va; + va_start(va, fmt); + spew(fmt, va); + va_end(va); + } + } + +#else + MOZ_ALWAYS_INLINE void spew(const char* fmt, ...) MOZ_FORMAT_PRINTF(2, 3) {} +#endif + +#ifdef JS_JITSPEW + MOZ_COLD void spew(const char* fmt, va_list va) MOZ_FORMAT_PRINTF(2, 0) { + // Buffer to hold the formatted string. Note that this may contain + // '%' characters, so do not pass it directly to printf functions. + char buf[200]; + + int i = VsprintfLiteral(buf, fmt, va); + if (i > -1) { + if (printer) { + printer->printf("%s\n", buf); + } + js::jit::JitSpew(js::jit::JitSpew_Codegen, "%s", buf); + } + } +#endif + + enum Condition { + Overflow = overflow, + Below = Uless, + BelowOrEqual = Uless_equal, + Above = Ugreater, + AboveOrEqual = Ugreater_equal, + Equal = equal, + NotEqual = not_equal, + GreaterThan = greater, + GreaterThanOrEqual = greater_equal, + LessThan = less, + LessThanOrEqual = less_equal, + Always = cc_always, + CarrySet, + CarryClear, + Signed, + NotSigned, + Zero, + NonZero, + }; + + enum DoubleCondition { + // These conditions will only evaluate to true if the comparison is ordered + // - i.e. neither operand is NaN. + DoubleOrdered, + DoubleEqual, + DoubleNotEqual, + DoubleGreaterThan, + DoubleGreaterThanOrEqual, + DoubleLessThan, + DoubleLessThanOrEqual, + // If either operand is NaN, these conditions always evaluate to true. + DoubleUnordered, + DoubleEqualOrUnordered, + DoubleNotEqualOrUnordered, + DoubleGreaterThanOrUnordered, + DoubleGreaterThanOrEqualOrUnordered, + DoubleLessThanOrUnordered, + DoubleLessThanOrEqualOrUnordered, + FIRST_UNORDERED = DoubleUnordered, + LAST_UNORDERED = DoubleLessThanOrEqualOrUnordered + }; + + Register getStackPointer() const { return StackPointer; } + void flushBuffer() {} + static int disassembleInstr(Instr instr, bool enable_spew = false); + int target_at(BufferOffset pos, bool is_internal); + static int target_at(Instruction* instruction, BufferOffset pos, + bool is_internal, Instruction* instruction2 = nullptr); + uint32_t next_link(Label* label, bool is_internal); + static uintptr_t target_address_at(Instruction* pos); + static void set_target_value_at(Instruction* pc, uint64_t target); + void target_at_put(BufferOffset pos, BufferOffset target_pos, + bool trampoline = false); + virtual int32_t branch_offset_helper(Label* L, OffsetSize bits); + int32_t branch_long_offset(Label* L); + + // Determines if Label is bound and near enough so that branch instruction + // can be used to reach it, instead of jump instruction. + bool is_near(Label* L); + bool is_near(Label* L, OffsetSize bits); + bool is_near_branch(Label* L); + + void nopAlign(int m) { + MOZ_ASSERT(m >= 4 && (m & (m - 1)) == 0); + while ((currentOffset() & (m - 1)) != 0) { + nop(); + } + } + virtual void emit(Instr x) { + MOZ_ASSERT(hasCreator()); + m_buffer.putInt(x); +#ifdef DEBUG + if (!oom()) { + DEBUG_PRINTF( + "0x%lx(%lx):", + (uint64_t)editSrc(BufferOffset(currentOffset() - sizeof(Instr))), + currentOffset() - sizeof(Instr)); + disassembleInstr(x, JitSpewEnabled(JitSpew_Codegen)); + } +#endif + } + virtual void emit(ShortInstr x) { MOZ_CRASH(); } + virtual void emit(uint64_t x) { MOZ_CRASH(); } + virtual void emit(uint32_t x) { + DEBUG_PRINTF( + "0x%lx(%lx): uint32_t: %d\n", + (uint64_t)editSrc(BufferOffset(currentOffset() - sizeof(Instr))), + currentOffset() - sizeof(Instr), x); + m_buffer.putInt(x); + } + + void instr_at_put(BufferOffset offset, Instr instr) { + DEBUG_PRINTF("\t[instr_at_put\n"); + DEBUG_PRINTF("\t%p %d \n\t", editSrc(offset), offset.getOffset()); + disassembleInstr(editSrc(offset)->InstructionBits()); + DEBUG_PRINTF("\t"); + *reinterpret_cast<Instr*>(editSrc(offset)) = instr; + disassembleInstr(editSrc(offset)->InstructionBits()); + DEBUG_PRINTF("\t]\n"); + } + + static Condition InvertCondition(Condition); + + static DoubleCondition InvertCondition(DoubleCondition); + + static uint64_t ExtractLoad64Value(Instruction* inst0); + static void UpdateLoad64Value(Instruction* inst0, uint64_t value); + static void PatchDataWithValueCheck(CodeLocationLabel label, ImmPtr newValue, + ImmPtr expectedValue); + static void PatchDataWithValueCheck(CodeLocationLabel label, + PatchedImmPtr newValue, + PatchedImmPtr expectedValue); + static void PatchWrite_Imm32(CodeLocationLabel label, Imm32 imm); + + static void PatchWrite_NearCall(CodeLocationLabel start, + CodeLocationLabel toCall) { + Instruction* inst = (Instruction*)start.raw(); + uint8_t* dest = toCall.raw(); + + // Overwrite whatever instruction used to be here with a call. + // Always use long jump for two reasons: + // - Jump has to be the same size because of PatchWrite_NearCallSize. + // - Return address has to be at the end of replaced block. + // Short jump wouldn't be more efficient. + // WriteLoad64Instructions will emit 6 instrs to load a addr. + Assembler::WriteLoad64Instructions(inst, ScratchRegister, (uint64_t)dest); + Instr jalr_ = JALR | (ra.code() << kRdShift) | (0x0 << kFunct3Shift) | + (ScratchRegister.code() << kRs1Shift) | (0x0 << kImm12Shift); + *reinterpret_cast<Instr*>(inst + 6 * kInstrSize) = jalr_; + } + static void WriteLoad64Instructions(Instruction* inst0, Register reg, + uint64_t value); + + static uint32_t PatchWrite_NearCallSize() { return 7 * sizeof(uint32_t); } + + static void TraceJumpRelocations(JSTracer* trc, JitCode* code, + CompactBufferReader& reader); + static void TraceDataRelocations(JSTracer* trc, JitCode* code, + CompactBufferReader& reader); + + static void ToggleToJmp(CodeLocationLabel inst_); + static void ToggleToCmp(CodeLocationLabel inst_); + static void ToggleCall(CodeLocationLabel inst_, bool enable); + + static void Bind(uint8_t* rawCode, const CodeLabel& label); + // label operations + void bind(Label* label, BufferOffset boff = BufferOffset()); + void bind(CodeLabel* label) { label->target()->bind(currentOffset()); } + uint32_t currentOffset() { return nextOffset().getOffset(); } + void retarget(Label* label, Label* target); + static uint32_t NopSize() { return 4; } + + static uintptr_t GetPointer(uint8_t* instPtr) { + Instruction* inst = (Instruction*)instPtr; + return Assembler::ExtractLoad64Value(inst); + } + + static bool HasRoundInstruction(RoundingMode) { return false; } + + void verifyHeapAccessDisassembly(uint32_t begin, uint32_t end, + const Disassembler::HeapAccess& heapAccess) { + MOZ_CRASH(); + } + + void setUnlimitedBuffer() { m_buffer.setUnlimited(); } + + GeneralRegisterSet* GetScratchRegisterList() { + return &scratch_register_list_; + } + + void EmitConstPoolWithJumpIfNeeded(size_t margin = 0) {} + + // As opposed to x86/x64 version, the data relocation has to be executed + // before to recover the pointer, and not after. + void writeDataRelocation(ImmGCPtr ptr) { + // Raw GC pointer relocations and Value relocations both end up in + // TraceOneDataRelocation. + if (ptr.value) { + if (gc::IsInsideNursery(ptr.value)) { + embedsNurseryPointers_ = true; + } + dataRelocations_.writeUnsigned(nextOffset().getOffset()); + } + } + + bool appendRawCode(const uint8_t* code, size_t numBytes); + + void assertNoGCThings() const { +#ifdef DEBUG + MOZ_ASSERT(dataRelocations_.length() == 0); + for (auto& j : jumps_) { + MOZ_ASSERT(j.kind == RelocationKind::HARDCODED); + } +#endif + } + + // Assembler Pseudo Instructions (Tables 25.2, 25.3, RISC-V Unprivileged ISA) + void break_(uint32_t code, bool break_as_stop = false); + void nop(); + void RV_li(Register rd, intptr_t imm); + static int RV_li_count(int64_t imm, bool is_get_temp_reg = false); + void GeneralLi(Register rd, int64_t imm); + static int GeneralLiCount(intptr_t imm, bool is_get_temp_reg = false); + void RecursiveLiImpl(Register rd, intptr_t imm); + void RecursiveLi(Register rd, intptr_t imm); + static int RecursiveLiCount(intptr_t imm); + static int RecursiveLiImplCount(intptr_t imm); + // Returns the number of instructions required to load the immediate + static int li_estimate(intptr_t imm, bool is_get_temp_reg = false); + // Loads an immediate, always using 8 instructions, regardless of the value, + // so that it can be modified later. + void li_constant(Register rd, intptr_t imm); + void li_ptr(Register rd, intptr_t imm); +}; + +class ABIArgGenerator { + public: + ABIArgGenerator() + : intRegIndex_(0), floatRegIndex_(0), stackOffset_(0), current_() {} + ABIArg next(MIRType); + ABIArg& current() { return current_; } + uint32_t stackBytesConsumedSoFar() const { return stackOffset_; } + void increaseStackOffset(uint32_t bytes) { stackOffset_ += bytes; } + + protected: + unsigned intRegIndex_; + unsigned floatRegIndex_; + uint32_t stackOffset_; + ABIArg current_; +}; + +// Note that nested uses of these are allowed, but the inner calls must imply +// an area of code which exists only inside the area of code implied by the +// outermost call. Otherwise AssemblerBufferWithConstantPools::enterNoPool +// will assert. +class BlockTrampolinePoolScope { + public: + explicit BlockTrampolinePoolScope(Assembler* assem, int margin) + : assem_(assem) { + assem_->enterNoPool(margin); + } + ~BlockTrampolinePoolScope() { assem_->leaveNoPool(); } + + private: + Assembler* assem_; + BlockTrampolinePoolScope() = delete; + BlockTrampolinePoolScope(const BlockTrampolinePoolScope&) = delete; + BlockTrampolinePoolScope& operator=(const BlockTrampolinePoolScope&) = delete; +}; + +class UseScratchRegisterScope { + public: + explicit UseScratchRegisterScope(Assembler* assembler); + ~UseScratchRegisterScope(); + + Register Acquire(); + bool hasAvailable() const; + void Include(const GeneralRegisterSet& list) { + *available_ = GeneralRegisterSet::Intersect(*available_, list); + } + void Exclude(const GeneralRegisterSet& list) { + *available_ = GeneralRegisterSet::Subtract(*available_, list); + } + + private: + GeneralRegisterSet* available_; + GeneralRegisterSet old_available_; +}; + +// Class Operand represents a shifter operand in data processing instructions. +class Operand { + public: + enum Tag { REG, FREG, MEM, IMM }; + Operand(FloatRegister freg) : tag(FREG), rm_(freg.code()) {} + + explicit Operand(Register base, Imm32 off) + : tag(MEM), rm_(base.code()), offset_(off.value) {} + + explicit Operand(Register base, int32_t off) + : tag(MEM), rm_(base.code()), offset_(off) {} + + explicit Operand(const Address& addr) + : tag(MEM), rm_(addr.base.code()), offset_(addr.offset) {} + + explicit Operand(intptr_t immediate) : tag(IMM), rm_() { value_ = immediate; } + // Register. + Operand(const Register rm) : tag(REG), rm_(rm.code()) {} + // Return true if this is a register operand. + bool is_reg() const { return tag == REG; } + bool is_freg() const { return tag == FREG; } + bool is_mem() const { return tag == MEM; } + bool is_imm() const { return tag == IMM; } + inline intptr_t immediate() const { + MOZ_ASSERT(is_imm()); + return value_; + } + bool IsImmediate() const { return !is_reg(); } + Register rm() const { return Register::FromCode(rm_); } + int32_t offset() const { + MOZ_ASSERT(is_mem()); + return offset_; + } + + FloatRegister toFReg() const { + MOZ_ASSERT(tag == FREG); + return FloatRegister::FromCode(rm_); + } + + Register toReg() const { + MOZ_ASSERT(tag == REG); + return Register::FromCode(rm_); + } + + Address toAddress() const { + MOZ_ASSERT(tag == MEM); + return Address(Register::FromCode(rm_), offset()); + } + + private: + Tag tag; + uint32_t rm_; + int32_t offset_; + intptr_t value_; // valid if rm_ == no_reg + + friend class Assembler; + friend class MacroAssembler; +}; + +static const uint32_t NumIntArgRegs = 8; +static const uint32_t NumFloatArgRegs = 8; +static inline bool GetIntArgReg(uint32_t usedIntArgs, Register* out) { + if (usedIntArgs < NumIntArgRegs) { + *out = Register::FromCode(a0.code() + usedIntArgs); + return true; + } + return false; +} + +static inline bool GetFloatArgReg(uint32_t usedFloatArgs, FloatRegister* out) { + if (usedFloatArgs < NumFloatArgRegs) { + *out = FloatRegister::FromCode(fa0.code() + usedFloatArgs); + return true; + } + return false; +} + +// Get a register in which we plan to put a quantity that will be used as an +// integer argument. This differs from GetIntArgReg in that if we have no more +// actual argument registers to use we will fall back on using whatever +// CallTempReg* don't overlap the argument registers, and only fail once those +// run out too. +static inline bool GetTempRegForIntArg(uint32_t usedIntArgs, + uint32_t usedFloatArgs, Register* out) { + // NOTE: We can't properly determine which regs are used if there are + // float arguments. If this is needed, we will have to guess. + MOZ_ASSERT(usedFloatArgs == 0); + + if (GetIntArgReg(usedIntArgs, out)) { + return true; + } + // Unfortunately, we have to assume things about the point at which + // GetIntArgReg returns false, because we need to know how many registers it + // can allocate. + usedIntArgs -= NumIntArgRegs; + if (usedIntArgs >= NumCallTempNonArgRegs) { + return false; + } + *out = CallTempNonArgRegs[usedIntArgs]; + return true; +} + +} // namespace jit +} // namespace js +#endif /* jit_riscv64_Assembler_riscv64_h */ diff --git a/js/src/jit/riscv64/AssemblerMatInt.cpp b/js/src/jit/riscv64/AssemblerMatInt.cpp new file mode 100644 index 0000000000..81c7fa7c40 --- /dev/null +++ b/js/src/jit/riscv64/AssemblerMatInt.cpp @@ -0,0 +1,217 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ +//===- RISCVMatInt.cpp - Immediate materialisation -------------*- C++ +//-*--===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM +// Exceptions. See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// +#include "mozilla/MathAlgorithms.h" +#include "mozilla/Maybe.h" + +#include "gc/Marking.h" +#include "jit/AutoWritableJitCode.h" +#include "jit/ExecutableAllocator.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/disasm/Disasm-riscv64.h" +#include "vm/Realm.h" +namespace js { +namespace jit { +void Assembler::RecursiveLi(Register rd, int64_t val) { + if (val > 0 && RecursiveLiImplCount(val) > 2) { + unsigned LeadingZeros = mozilla::CountLeadingZeroes64((uint64_t)val); + uint64_t ShiftedVal = (uint64_t)val << LeadingZeros; + int countFillZero = RecursiveLiImplCount(ShiftedVal) + 1; + if (countFillZero < RecursiveLiImplCount(val)) { + RecursiveLiImpl(rd, ShiftedVal); + srli(rd, rd, LeadingZeros); + return; + } + } + RecursiveLiImpl(rd, val); +} + +int Assembler::RecursiveLiCount(int64_t val) { + if (val > 0 && RecursiveLiImplCount(val) > 2) { + unsigned LeadingZeros = mozilla::CountLeadingZeroes64((uint64_t)val); + uint64_t ShiftedVal = (uint64_t)val << LeadingZeros; + // Fill in the bits that will be shifted out with 1s. An example where + // this helps is trailing one masks with 32 or more ones. This will + // generate ADDI -1 and an SRLI. + int countFillZero = RecursiveLiImplCount(ShiftedVal) + 1; + if (countFillZero < RecursiveLiImplCount(val)) { + return countFillZero; + } + } + return RecursiveLiImplCount(val); +} + +inline int64_t signExtend(uint64_t V, int N) { + return int64_t(V << (64 - N)) >> (64 - N); +} + +void Assembler::RecursiveLiImpl(Register rd, int64_t Val) { + if (is_int32(Val)) { + // Depending on the active bits in the immediate Value v, the following + // instruction sequences are emitted: + // + // v == 0 : ADDI + // v[0,12) != 0 && v[12,32) == 0 : ADDI + // v[0,12) == 0 && v[12,32) != 0 : LUI + // v[0,32) != 0 : LUI+ADDI(W) + int64_t Hi20 = ((Val + 0x800) >> 12) & 0xFFFFF; + int64_t Lo12 = Val << 52 >> 52; + + if (Hi20) { + lui(rd, (int32_t)Hi20); + } + + if (Lo12 || Hi20 == 0) { + if (Hi20) { + addiw(rd, rd, Lo12); + } else { + addi(rd, zero_reg, Lo12); + } + } + return; + } + + // In the worst case, for a full 64-bit constant, a sequence of 8 + // instructions (i.e., LUI+ADDIW+SLLI+ADDI+SLLI+ADDI+SLLI+ADDI) has to be + // emitted. Note that the first two instructions (LUI+ADDIW) can contribute + // up to 32 bits while the following ADDI instructions contribute up to 12 + // bits each. + // + // On the first glance, implementing this seems to be possible by simply + // emitting the most significant 32 bits (LUI+ADDIW) followed by as many + // left shift (SLLI) and immediate additions (ADDI) as needed. However, due + // to the fact that ADDI performs a sign extended addition, doing it like + // that would only be possible when at most 11 bits of the ADDI instructions + // are used. Using all 12 bits of the ADDI instructions, like done by GAS, + // actually requires that the constant is processed starting with the least + // significant bit. + // + // In the following, constants are processed from LSB to MSB but instruction + // emission is performed from MSB to LSB by recursively calling + // generateInstSeq. In each recursion, first the lowest 12 bits are removed + // from the constant and the optimal shift amount, which can be greater than + // 12 bits if the constant is sparse, is determined. Then, the shifted + // remaining constant is processed recursively and gets emitted as soon as + // it fits into 32 bits. The emission of the shifts and additions is + // subsequently performed when the recursion returns. + + int64_t Lo12 = Val << 52 >> 52; + int64_t Hi52 = ((uint64_t)Val + 0x800ull) >> 12; + int ShiftAmount = 12 + mozilla::CountTrailingZeroes64((uint64_t)Hi52); + Hi52 = signExtend(Hi52 >> (ShiftAmount - 12), 64 - ShiftAmount); + + // If the remaining bits don't fit in 12 bits, we might be able to reduce + // the shift amount in order to use LUI which will zero the lower 12 bits. + bool Unsigned = false; + if (ShiftAmount > 12 && !is_int12(Hi52)) { + if (is_int32((uint64_t)Hi52 << 12)) { + // Reduce the shift amount and add zeros to the LSBs so it will match + // LUI. + ShiftAmount -= 12; + Hi52 = (uint64_t)Hi52 << 12; + } + } + RecursiveLi(rd, Hi52); + + if (Unsigned) { + } else { + slli(rd, rd, ShiftAmount); + } + if (Lo12) { + addi(rd, rd, Lo12); + } +} + +int Assembler::RecursiveLiImplCount(int64_t Val) { + int count = 0; + if (is_int32(Val)) { + // Depending on the active bits in the immediate Value v, the following + // instruction sequences are emitted: + // + // v == 0 : ADDI + // v[0,12) != 0 && v[12,32) == 0 : ADDI + // v[0,12) == 0 && v[12,32) != 0 : LUI + // v[0,32) != 0 : LUI+ADDI(W) + int64_t Hi20 = ((Val + 0x800) >> 12) & 0xFFFFF; + int64_t Lo12 = Val << 52 >> 52; + + if (Hi20) { + // lui(rd, (int32_t)Hi20); + count++; + } + + if (Lo12 || Hi20 == 0) { + // unsigned AddiOpc = (IsRV64 && Hi20) ? RISCV::ADDIW : RISCV::ADDI; + // Res.push_back(RISCVMatInt::Inst(AddiOpc, Lo12)); + count++; + } + return count; + } + + // In the worst case, for a full 64-bit constant, a sequence of 8 + // instructions (i.e., LUI+ADDIW+SLLI+ADDI+SLLI+ADDI+SLLI+ADDI) has to be + // emitted. Note that the first two instructions (LUI+ADDIW) can contribute + // up to 32 bits while the following ADDI instructions contribute up to 12 + // bits each. + // + // On the first glance, implementing this seems to be possible by simply + // emitting the most significant 32 bits (LUI+ADDIW) followed by as many + // left shift (SLLI) and immediate additions (ADDI) as needed. However, due + // to the fact that ADDI performs a sign extended addition, doing it like + // that would only be possible when at most 11 bits of the ADDI instructions + // are used. Using all 12 bits of the ADDI instructions, like done by GAS, + // actually requires that the constant is processed starting with the least + // significant bit. + // + // In the following, constants are processed from LSB to MSB but instruction + // emission is performed from MSB to LSB by recursively calling + // generateInstSeq. In each recursion, first the lowest 12 bits are removed + // from the constant and the optimal shift amount, which can be greater than + // 12 bits if the constant is sparse, is determined. Then, the shifted + // remaining constant is processed recursively and gets emitted as soon as + // it fits into 32 bits. The emission of the shifts and additions is + // subsequently performed when the recursion returns. + + int64_t Lo12 = Val << 52 >> 52; + int64_t Hi52 = ((uint64_t)Val + 0x800ull) >> 12; + int ShiftAmount = 12 + mozilla::CountTrailingZeroes64((uint64_t)Hi52); + Hi52 = signExtend(Hi52 >> (ShiftAmount - 12), 64 - ShiftAmount); + + // If the remaining bits don't fit in 12 bits, we might be able to reduce + // the shift amount in order to use LUI which will zero the lower 12 bits. + bool Unsigned = false; + if (ShiftAmount > 12 && !is_int12(Hi52)) { + if (is_int32((uint64_t)Hi52 << 12)) { + // Reduce the shift amount and add zeros to the LSBs so it will match + // LUI. + ShiftAmount -= 12; + Hi52 = (uint64_t)Hi52 << 12; + } + } + + count += RecursiveLiImplCount(Hi52); + + if (Unsigned) { + } else { + // slli(rd, rd, ShiftAmount); + count++; + } + if (Lo12) { + // addi(rd, rd, Lo12); + count++; + } + return count; +} + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/CodeGenerator-riscv64.cpp b/js/src/jit/riscv64/CodeGenerator-riscv64.cpp new file mode 100644 index 0000000000..1c890799ed --- /dev/null +++ b/js/src/jit/riscv64/CodeGenerator-riscv64.cpp @@ -0,0 +1,2896 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "jit/riscv64/CodeGenerator-riscv64.h" + +#include "mozilla/MathAlgorithms.h" + +#include "jsnum.h" + +#include "jit/CodeGenerator.h" +#include "jit/InlineScriptTree.h" +#include "jit/JitRuntime.h" +#include "jit/MIR.h" +#include "jit/MIRGraph.h" +#include "vm/JSContext.h" +#include "vm/Realm.h" +#include "vm/Shape.h" + +#include "jit/shared/CodeGenerator-shared-inl.h" +#include "vm/JSScript-inl.h" + +using namespace js; +using namespace js::jit; + +using JS::GenericNaN; +using mozilla::FloorLog2; +using mozilla::NegativeInfinity; + +// shared +CodeGeneratorRiscv64::CodeGeneratorRiscv64(MIRGenerator* gen, LIRGraph* graph, + MacroAssembler* masm) + : CodeGeneratorShared(gen, graph, masm) {} + +Operand CodeGeneratorRiscv64::ToOperand(const LAllocation& a) { + if (a.isGeneralReg()) { + return Operand(a.toGeneralReg()->reg()); + } + if (a.isFloatReg()) { + return Operand(a.toFloatReg()->reg()); + } + return Operand(ToAddress(a)); +} + +Operand CodeGeneratorRiscv64::ToOperand(const LAllocation* a) { + return ToOperand(*a); +} + +Operand CodeGeneratorRiscv64::ToOperand(const LDefinition* def) { + return ToOperand(def->output()); +} + +#ifdef JS_PUNBOX64 +Operand CodeGeneratorRiscv64::ToOperandOrRegister64( + const LInt64Allocation input) { + return ToOperand(input.value()); +} +#else +Register64 CodeGeneratorRiscv64::ToOperandOrRegister64( + const LInt64Allocation input) { + return ToRegister64(input); +} +#endif + +void CodeGeneratorRiscv64::branchToBlock(FloatFormat fmt, FloatRegister lhs, + FloatRegister rhs, MBasicBlock* mir, + Assembler::DoubleCondition cond) { + // Skip past trivial blocks. + Label* label = skipTrivialBlocks(mir)->lir()->label(); + if (fmt == DoubleFloat) { + masm.branchDouble(cond, lhs, rhs, label); + } else { + masm.branchFloat(cond, lhs, rhs, label); + } +} + +void OutOfLineBailout::accept(CodeGeneratorRiscv64* codegen) { + codegen->visitOutOfLineBailout(this); +} + +MoveOperand CodeGeneratorRiscv64::toMoveOperand(LAllocation a) const { + if (a.isGeneralReg()) { + return MoveOperand(ToRegister(a)); + } + if (a.isFloatReg()) { + return MoveOperand(ToFloatRegister(a)); + } + MoveOperand::Kind kind = a.isStackArea() ? MoveOperand::Kind::EffectiveAddress + : MoveOperand::Kind::Memory; + Address address = ToAddress(a); + MOZ_ASSERT((address.offset & 3) == 0); + + return MoveOperand(address, kind); +} + +void CodeGeneratorRiscv64::bailoutFrom(Label* label, LSnapshot* snapshot) { + MOZ_ASSERT_IF(!masm.oom(), label->used()); + MOZ_ASSERT_IF(!masm.oom(), !label->bound()); + + encode(snapshot); + + InlineScriptTree* tree = snapshot->mir()->block()->trackedTree(); + OutOfLineBailout* ool = new (alloc()) OutOfLineBailout(snapshot); + addOutOfLineCode(ool, + new (alloc()) BytecodeSite(tree, tree->script()->code())); + + masm.retarget(label, ool->entry()); +} + +void CodeGeneratorRiscv64::bailout(LSnapshot* snapshot) { + Label label; + masm.jump(&label); + bailoutFrom(&label, snapshot); +} + +bool CodeGeneratorRiscv64::generateOutOfLineCode() { + if (!CodeGeneratorShared::generateOutOfLineCode()) { + return false; + } + + if (deoptLabel_.used()) { + // All non-table-based bailouts will go here. + masm.bind(&deoptLabel_); + + // Push the frame size, so the handler can recover the IonScript. + // Frame size is stored in 'ra' and pushed by GenerateBailoutThunk + // We have to use 'ra' because generateBailoutTable will implicitly do + // the same. + masm.move32(Imm32(frameSize()), ra); + + TrampolinePtr handler = gen->jitRuntime()->getGenericBailoutHandler(); + masm.jump(handler); + } + + return !masm.oom(); +} + +class js::jit::OutOfLineTableSwitch + : public OutOfLineCodeBase<CodeGeneratorRiscv64> { + MTableSwitch* mir_; + CodeLabel jumpLabel_; + + void accept(CodeGeneratorRiscv64* codegen) { + codegen->visitOutOfLineTableSwitch(this); + } + + public: + OutOfLineTableSwitch(MTableSwitch* mir) : mir_(mir) {} + + MTableSwitch* mir() const { return mir_; } + + CodeLabel* jumpLabel() { return &jumpLabel_; } +}; + +void CodeGeneratorRiscv64::emitTableSwitchDispatch(MTableSwitch* mir, + Register index, + Register base) { + Label* defaultcase = skipTrivialBlocks(mir->getDefault())->lir()->label(); + + // Lower value with low value + if (mir->low() != 0) { + masm.subPtr(Imm32(mir->low()), index); + } + + // Jump to default case if input is out of range + int32_t cases = mir->numCases(); + masm.branchPtr(Assembler::AboveOrEqual, index, ImmWord(cases), defaultcase); + + // To fill in the CodeLabels for the case entries, we need to first + // generate the case entries (we don't yet know their offsets in the + // instruction stream). + OutOfLineTableSwitch* ool = new (alloc()) OutOfLineTableSwitch(mir); + addOutOfLineCode(ool, mir); + + // Compute the position where a pointer to the right case stands. + masm.ma_li(base, ool->jumpLabel()); + + BaseIndex pointer(base, index, ScalePointer); + + // Jump to the right case + masm.branchToComputedAddress(pointer); +} + +template <typename T> +void CodeGeneratorRiscv64::emitWasmLoad(T* lir) { + const MWasmLoad* mir = lir->mir(); + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + + Register memoryBase = ToRegister(lir->memoryBase()); + Register ptr = ToRegister(lir->ptr()); + Register ptrScratch = InvalidReg; + if (!lir->ptrCopy()->isBogusTemp()) { + ptrScratch = ToRegister(lir->ptrCopy()); + } + + if (mir->base()->type() == MIRType::Int32) { + masm.move32To64ZeroExtend(ptr, Register64(scratch2)); + ptr = scratch2; + ptrScratch = ptrScratch != InvalidReg ? scratch2 : InvalidReg; + } + + // ptr is a GPR and is either a 32-bit value zero-extended to 64-bit, or a + // true 64-bit value. + masm.wasmLoad(mir->access(), memoryBase, ptr, ptrScratch, + ToAnyRegister(lir->output())); +} + +template <typename T> +void CodeGeneratorRiscv64::emitWasmStore(T* lir) { + const MWasmStore* mir = lir->mir(); + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + + Register memoryBase = ToRegister(lir->memoryBase()); + Register ptr = ToRegister(lir->ptr()); + Register ptrScratch = InvalidReg; + if (!lir->ptrCopy()->isBogusTemp()) { + ptrScratch = ToRegister(lir->ptrCopy()); + } + + if (mir->base()->type() == MIRType::Int32) { + masm.move32To64ZeroExtend(ptr, Register64(scratch2)); + ptr = scratch2; + ptrScratch = ptrScratch != InvalidReg ? scratch2 : InvalidReg; + } + + // ptr is a GPR and is either a 32-bit value zero-extended to 64-bit, or a + // true 64-bit value. + masm.wasmStore(mir->access(), ToAnyRegister(lir->value()), memoryBase, ptr, + ptrScratch); +} + +void CodeGeneratorRiscv64::generateInvalidateEpilogue() { + // Ensure that there is enough space in the buffer for the OsiPoint + // patching to occur. Otherwise, we could overwrite the invalidation + // epilogue + for (size_t i = 0; i < sizeof(void*); i += Assembler::NopSize()) { + masm.nop(); + } + + masm.bind(&invalidate_); + + // Push the return address of the point that we bailed out at to the stack + masm.Push(ra); + + // Push the Ion script onto the stack (when we determine what that + // pointer is). + invalidateEpilogueData_ = masm.pushWithPatch(ImmWord(uintptr_t(-1))); + + // Jump to the invalidator which will replace the current frame. + TrampolinePtr thunk = gen->jitRuntime()->getInvalidationThunk(); + + masm.jump(thunk); +} + +void CodeGeneratorRiscv64::visitOutOfLineBailout(OutOfLineBailout* ool) { + // Push snapshotOffset and make sure stack is aligned. + masm.subPtr(Imm32(sizeof(Value)), StackPointer); + masm.storePtr(ImmWord(ool->snapshot()->snapshotOffset()), + Address(StackPointer, 0)); + + masm.jump(&deoptLabel_); +} + +void CodeGeneratorRiscv64::visitOutOfLineTableSwitch( + OutOfLineTableSwitch* ool) { + MTableSwitch* mir = ool->mir(); + masm.nop(); + masm.haltingAlign(sizeof(void*)); + masm.bind(ool->jumpLabel()); + masm.addCodeLabel(*ool->jumpLabel()); + BlockTrampolinePoolScope block_trampoline_pool( + &masm, mir->numCases() * sizeof(uint64_t)); + for (size_t i = 0; i < mir->numCases(); i++) { + LBlock* caseblock = skipTrivialBlocks(mir->getCase(i))->lir(); + Label* caseheader = caseblock->label(); + uint32_t caseoffset = caseheader->offset(); + + // The entries of the jump table need to be absolute addresses and thus + // must be patched after codegen is finished. + CodeLabel cl; + masm.writeCodePointer(&cl); + cl.target()->bind(caseoffset); + masm.addCodeLabel(cl); + } +} + +void CodeGeneratorRiscv64::visitOutOfLineWasmTruncateCheck( + OutOfLineWasmTruncateCheck* ool) { + FloatRegister input = ool->input(); + Register output = ool->output(); + Register64 output64 = ool->output64(); + MIRType fromType = ool->fromType(); + MIRType toType = ool->toType(); + Label* oolRejoin = ool->rejoin(); + TruncFlags flags = ool->flags(); + wasm::BytecodeOffset off = ool->bytecodeOffset(); + + if (fromType == MIRType::Float32) { + if (toType == MIRType::Int32) { + masm.oolWasmTruncateCheckF32ToI32(input, output, flags, off, oolRejoin); + } else if (toType == MIRType::Int64) { + masm.oolWasmTruncateCheckF32ToI64(input, output64, flags, off, oolRejoin); + } else { + MOZ_CRASH("unexpected type"); + } + } else if (fromType == MIRType::Double) { + if (toType == MIRType::Int32) { + masm.oolWasmTruncateCheckF64ToI32(input, output, flags, off, oolRejoin); + } else if (toType == MIRType::Int64) { + masm.oolWasmTruncateCheckF64ToI64(input, output64, flags, off, oolRejoin); + } else { + MOZ_CRASH("unexpected type"); + } + } else { + MOZ_CRASH("unexpected type"); + } +} + +ValueOperand CodeGeneratorRiscv64::ToValue(LInstruction* ins, size_t pos) { + return ValueOperand(ToRegister(ins->getOperand(pos))); +} + +ValueOperand CodeGeneratorRiscv64::ToTempValue(LInstruction* ins, size_t pos) { + return ValueOperand(ToRegister(ins->getTemp(pos))); +} + +void CodeGenerator::visitBox(LBox* box) { + const LAllocation* in = box->getOperand(0); + ValueOperand result = ToOutValue(box); + + masm.moveValue(TypedOrValueRegister(box->type(), ToAnyRegister(in)), result); +} + +void CodeGenerator::visitUnbox(LUnbox* unbox) { + MUnbox* mir = unbox->mir(); + + Register result = ToRegister(unbox->output()); + + if (mir->fallible()) { + const ValueOperand value = ToValue(unbox, LUnbox::Input); + Label bail; + switch (mir->type()) { + case MIRType::Int32: + masm.fallibleUnboxInt32(value, result, &bail); + break; + case MIRType::Boolean: + masm.fallibleUnboxBoolean(value, result, &bail); + break; + case MIRType::Object: + masm.fallibleUnboxObject(value, result, &bail); + break; + case MIRType::String: + masm.fallibleUnboxString(value, result, &bail); + break; + case MIRType::Symbol: + masm.fallibleUnboxSymbol(value, result, &bail); + break; + case MIRType::BigInt: + masm.fallibleUnboxBigInt(value, result, &bail); + break; + default: + MOZ_CRASH("Given MIRType cannot be unboxed."); + } + bailoutFrom(&bail, unbox->snapshot()); + return; + } + + LAllocation* input = unbox->getOperand(LUnbox::Input); + if (input->isRegister()) { + Register inputReg = ToRegister(input); + switch (mir->type()) { + case MIRType::Int32: + masm.unboxInt32(inputReg, result); + break; + case MIRType::Boolean: + masm.unboxBoolean(inputReg, result); + break; + case MIRType::Object: + masm.unboxObject(inputReg, result); + break; + case MIRType::String: + masm.unboxString(inputReg, result); + break; + case MIRType::Symbol: + masm.unboxSymbol(inputReg, result); + break; + case MIRType::BigInt: + masm.unboxBigInt(inputReg, result); + break; + default: + MOZ_CRASH("Given MIRType cannot be unboxed."); + } + return; + } + + Address inputAddr = ToAddress(input); + switch (mir->type()) { + case MIRType::Int32: + masm.unboxInt32(inputAddr, result); + break; + case MIRType::Boolean: + masm.unboxBoolean(inputAddr, result); + break; + case MIRType::Object: + masm.unboxObject(inputAddr, result); + break; + case MIRType::String: + masm.unboxString(inputAddr, result); + break; + case MIRType::Symbol: + masm.unboxSymbol(inputAddr, result); + break; + case MIRType::BigInt: + masm.unboxBigInt(inputAddr, result); + break; + default: + MOZ_CRASH("Given MIRType cannot be unboxed."); + } +} + +void CodeGeneratorRiscv64::splitTagForTest(const ValueOperand& value, + ScratchTagScope& tag) { + masm.splitTag(value.valueReg(), tag); +} + +void CodeGenerator::visitCompareI64(LCompareI64* lir) { + MCompare* mir = lir->mir(); + const mozilla::DebugOnly<MCompare::CompareType> type = mir->compareType(); + MOZ_ASSERT(type == MCompare::Compare_Int64 || + type == MCompare::Compare_UInt64); + + const LInt64Allocation lhs = lir->getInt64Operand(LCompareI64::Lhs); + const LInt64Allocation rhs = lir->getInt64Operand(LCompareI64::Rhs); + Register lhsReg = ToRegister64(lhs).reg; + Register output = ToRegister(lir->output()); + bool isSigned = mir->compareType() == MCompare::Compare_Int64; + Assembler::Condition cond = JSOpToCondition(lir->jsop(), isSigned); + + if (IsConstant(rhs)) { + masm.cmpPtrSet(cond, lhsReg, ImmWord(ToInt64(rhs)), output); + } else if (rhs.value().isGeneralReg()) { + masm.cmpPtrSet(cond, lhsReg, ToRegister64(rhs).reg, output); + } else { + masm.cmpPtrSet(cond, lhsReg, ToAddress(rhs.value()), output); + } +} + +void CodeGenerator::visitCompareI64AndBranch(LCompareI64AndBranch* lir) { + MCompare* mir = lir->cmpMir(); + const mozilla::DebugOnly<MCompare::CompareType> type = mir->compareType(); + MOZ_ASSERT(type == MCompare::Compare_Int64 || + type == MCompare::Compare_UInt64); + + const LInt64Allocation lhs = lir->getInt64Operand(LCompareI64::Lhs); + const LInt64Allocation rhs = lir->getInt64Operand(LCompareI64::Rhs); + Register lhsReg = ToRegister64(lhs).reg; + bool isSigned = mir->compareType() == MCompare::Compare_Int64; + Assembler::Condition cond = JSOpToCondition(lir->jsop(), isSigned); + + if (IsConstant(rhs)) { + emitBranch(lhsReg, ImmWord(ToInt64(rhs)), cond, lir->ifTrue(), + lir->ifFalse()); + } else if (rhs.value().isGeneralReg()) { + emitBranch(lhsReg, ToRegister64(rhs).reg, cond, lir->ifTrue(), + lir->ifFalse()); + } else { + emitBranch(lhsReg, ToAddress(rhs.value()), cond, lir->ifTrue(), + lir->ifFalse()); + } +} + +void CodeGenerator::visitCompare(LCompare* comp) { + MCompare* mir = comp->mir(); + Assembler::Condition cond = JSOpToCondition(mir->compareType(), comp->jsop()); + const LAllocation* left = comp->getOperand(0); + const LAllocation* right = comp->getOperand(1); + const LDefinition* def = comp->getDef(0); + + if (mir->compareType() == MCompare::Compare_Object || + mir->compareType() == MCompare::Compare_Symbol || + mir->compareType() == MCompare::Compare_UIntPtr || + mir->compareType() == MCompare::Compare_WasmAnyRef) { + if (right->isConstant()) { + MOZ_ASSERT(mir->compareType() == MCompare::Compare_UIntPtr); + masm.cmpPtrSet(cond, ToRegister(left), Imm32(ToInt32(right)), + ToRegister(def)); + } else if (right->isGeneralReg()) { + masm.cmpPtrSet(cond, ToRegister(left), ToRegister(right), + ToRegister(def)); + } else { + masm.cmpPtrSet(cond, ToRegister(left), ToAddress(right), ToRegister(def)); + } + return; + } + + if (right->isConstant()) { + masm.cmp32Set(cond, ToRegister(left), Imm32(ToInt32(right)), + ToRegister(def)); + } else if (right->isGeneralReg()) { + masm.cmp32Set(cond, ToRegister(left), ToRegister(right), ToRegister(def)); + } else { + masm.cmp32Set(cond, ToRegister(left), ToAddress(right), ToRegister(def)); + } +} + +void CodeGenerator::visitCompareAndBranch(LCompareAndBranch* comp) { + const MCompare* mir = comp->cmpMir(); + const MCompare::CompareType type = mir->compareType(); + const LAllocation* lhs = comp->left(); + const LAllocation* rhs = comp->right(); + MBasicBlock* ifTrue = comp->ifTrue(); + MBasicBlock* ifFalse = comp->ifFalse(); + Register lhsReg = ToRegister(lhs); + const Assembler::Condition cond = JSOpToCondition(type, comp->jsop()); + + if (type == MCompare::Compare_Object || type == MCompare::Compare_Symbol || + type == MCompare::Compare_UIntPtr || + type == MCompare::Compare_WasmAnyRef) { + if (rhs->isConstant()) { + emitBranch(ToRegister(lhs), Imm32(ToInt32(rhs)), cond, ifTrue, ifFalse); + } else if (rhs->isGeneralReg()) { + emitBranch(lhsReg, ToRegister(rhs), cond, ifTrue, ifFalse); + } else { + MOZ_CRASH("NYI"); + } + return; + } + + if (rhs->isConstant()) { + emitBranch(lhsReg, Imm32(ToInt32(comp->right())), cond, ifTrue, ifFalse); + } else if (comp->right()->isGeneralReg()) { + emitBranch(lhsReg, ToRegister(rhs), cond, ifTrue, ifFalse); + } else { + // TODO(riscv): emitBranch with 32-bit comparision + ScratchRegisterScope scratch(masm); + masm.load32(ToAddress(rhs), scratch); + emitBranch(lhsReg, Register(scratch), cond, ifTrue, ifFalse); + } +} + +void CodeGenerator::visitDivOrModI64(LDivOrModI64* lir) { + Register lhs = ToRegister(lir->lhs()); + Register rhs = ToRegister(lir->rhs()); + Register output = ToRegister(lir->output()); + + Label done; + + // Handle divide by zero. + if (lir->canBeDivideByZero()) { + Label nonZero; + masm.ma_b(rhs, rhs, &nonZero, Assembler::NonZero); + masm.wasmTrap(wasm::Trap::IntegerDivideByZero, lir->bytecodeOffset()); + masm.bind(&nonZero); + } + + // Handle an integer overflow exception from INT64_MIN / -1. + if (lir->canBeNegativeOverflow()) { + Label notOverflow; + masm.branchPtr(Assembler::NotEqual, lhs, ImmWord(INT64_MIN), ¬Overflow); + masm.branchPtr(Assembler::NotEqual, rhs, ImmWord(-1), ¬Overflow); + if (lir->mir()->isMod()) { + masm.ma_xor(output, output, Operand(output)); + } else { + masm.wasmTrap(wasm::Trap::IntegerOverflow, lir->bytecodeOffset()); + } + masm.jump(&done); + masm.bind(¬Overflow); + } + + if (lir->mir()->isMod()) { + masm.ma_mod64(output, lhs, rhs); + } else { + masm.ma_div64(output, lhs, rhs); + } + + masm.bind(&done); +} + +void CodeGenerator::visitUDivOrModI64(LUDivOrModI64* lir) { + Register lhs = ToRegister(lir->lhs()); + Register rhs = ToRegister(lir->rhs()); + Register output = ToRegister(lir->output()); + + Label done; + + // Prevent divide by zero. + if (lir->canBeDivideByZero()) { + Label nonZero; + masm.ma_b(rhs, rhs, &nonZero, Assembler::NonZero); + masm.wasmTrap(wasm::Trap::IntegerDivideByZero, lir->bytecodeOffset()); + masm.bind(&nonZero); + } + + if (lir->mir()->isMod()) { + masm.ma_modu64(output, lhs, rhs); + } else { + masm.ma_divu64(output, lhs, rhs); + } + + masm.bind(&done); +} + +void CodeGeneratorRiscv64::emitBigIntDiv(LBigIntDiv* ins, Register dividend, + Register divisor, Register output, + Label* fail) { + // Callers handle division by zero and integer overflow. + masm.ma_div64(/* result= */ dividend, dividend, divisor); + + // Create and return the result. + masm.newGCBigInt(output, divisor, initialBigIntHeap(), fail); + masm.initializeBigInt(output, dividend); +} + +void CodeGeneratorRiscv64::emitBigIntMod(LBigIntMod* ins, Register dividend, + Register divisor, Register output, + Label* fail) { + // Callers handle division by zero and integer overflow. + masm.ma_mod64(/* result= */ dividend, dividend, divisor); + + // Create and return the result. + masm.newGCBigInt(output, divisor, initialBigIntHeap(), fail); + masm.initializeBigInt(output, dividend); +} + +void CodeGenerator::visitWasmLoadI64(LWasmLoadI64* lir) { + const MWasmLoad* mir = lir->mir(); + + Register memoryBase = ToRegister(lir->memoryBase()); + Register ptrScratch = InvalidReg; + if (!lir->ptrCopy()->isBogusTemp()) { + ptrScratch = ToRegister(lir->ptrCopy()); + } + + Register ptrReg = ToRegister(lir->ptr()); + if (mir->base()->type() == MIRType::Int32) { + // See comment in visitWasmLoad re the type of 'base'. + masm.move32ZeroExtendToPtr(ptrReg, ptrReg); + } + + masm.wasmLoadI64(mir->access(), memoryBase, ptrReg, ptrScratch, + ToOutRegister64(lir)); +} + +void CodeGenerator::visitWasmStoreI64(LWasmStoreI64* lir) { + const MWasmStore* mir = lir->mir(); + + Register memoryBase = ToRegister(lir->memoryBase()); + Register ptrScratch = InvalidReg; + if (!lir->ptrCopy()->isBogusTemp()) { + ptrScratch = ToRegister(lir->ptrCopy()); + } + + Register ptrReg = ToRegister(lir->ptr()); + if (mir->base()->type() == MIRType::Int32) { + // See comment in visitWasmLoad re the type of 'base'. + masm.move32ZeroExtendToPtr(ptrReg, ptrReg); + } + + masm.wasmStoreI64(mir->access(), ToRegister64(lir->value()), memoryBase, + ptrReg, ptrScratch); +} + +void CodeGenerator::visitWasmSelectI64(LWasmSelectI64* lir) { + MOZ_ASSERT(lir->mir()->type() == MIRType::Int64); + + Register cond = ToRegister(lir->condExpr()); + const LInt64Allocation falseExpr = lir->falseExpr(); + + Register64 out = ToOutRegister64(lir); + MOZ_ASSERT(ToRegister64(lir->trueExpr()) == out, + "true expr is reused for input"); + + if (falseExpr.value().isRegister()) { + masm.moveIfZero(out.reg, ToRegister(falseExpr.value()), cond); + } else { + Label done; + masm.ma_b(cond, cond, &done, Assembler::NonZero, ShortJump); + masm.loadPtr(ToAddress(falseExpr.value()), out.reg); + masm.bind(&done); + } +} + +void CodeGenerator::visitWasmReinterpretFromI64(LWasmReinterpretFromI64* lir) { + MOZ_ASSERT(lir->mir()->type() == MIRType::Double); + MOZ_ASSERT(lir->mir()->input()->type() == MIRType::Int64); + masm.fmv_d_x(ToFloatRegister(lir->output()), ToRegister(lir->input())); +} + +void CodeGenerator::visitWasmReinterpretToI64(LWasmReinterpretToI64* lir) { + MOZ_ASSERT(lir->mir()->type() == MIRType::Int64); + MOZ_ASSERT(lir->mir()->input()->type() == MIRType::Double); + masm.fmv_x_d(ToRegister(lir->output()), ToFloatRegister(lir->input())); +} + +void CodeGenerator::visitExtendInt32ToInt64(LExtendInt32ToInt64* lir) { + const LAllocation* input = lir->getOperand(0); + Register output = ToRegister(lir->output()); + + if (lir->mir()->isUnsigned()) { + masm.move32To64ZeroExtend(ToRegister(input), Register64(output)); + } else { + masm.slliw(output, ToRegister(input), 0); + } +} + +void CodeGenerator::visitWrapInt64ToInt32(LWrapInt64ToInt32* lir) { + const LAllocation* input = lir->getOperand(0); + Register output = ToRegister(lir->output()); + + if (lir->mir()->bottomHalf()) { + if (input->isMemory()) { + masm.load32(ToAddress(input), output); + } else { + masm.slliw(output, ToRegister(input), 0); + } + } else { + MOZ_CRASH("Not implemented."); + } +} + +void CodeGenerator::visitSignExtendInt64(LSignExtendInt64* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + Register64 output = ToOutRegister64(lir); + switch (lir->mode()) { + case MSignExtendInt64::Byte: + masm.move32To64SignExtend(input.reg, output); + masm.move8SignExtend(output.reg, output.reg); + break; + case MSignExtendInt64::Half: + masm.move32To64SignExtend(input.reg, output); + masm.move16SignExtend(output.reg, output.reg); + break; + case MSignExtendInt64::Word: + masm.move32To64SignExtend(input.reg, output); + break; + } +} + +void CodeGenerator::visitWasmExtendU32Index(LWasmExtendU32Index* lir) { + Register input = ToRegister(lir->input()); + Register output = ToRegister(lir->output()); + MOZ_ASSERT(input == output); + masm.move32To64ZeroExtend(input, Register64(output)); +} + +void CodeGenerator::visitWasmWrapU32Index(LWasmWrapU32Index* lir) { + Register input = ToRegister(lir->input()); + Register output = ToRegister(lir->output()); + MOZ_ASSERT(input == output); + masm.move64To32(Register64(input), output); +} + +void CodeGenerator::visitClzI64(LClzI64* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + Register64 output = ToOutRegister64(lir); + masm.clz64(input, output.reg); +} + +void CodeGenerator::visitCtzI64(LCtzI64* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + Register64 output = ToOutRegister64(lir); + masm.ctz64(input, output.reg); +} + +void CodeGenerator::visitNotI64(LNotI64* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + Register output = ToRegister(lir->output()); + + masm.ma_cmp_set(output, input.reg, zero, Assembler::Equal); +} + +void CodeGenerator::visitWasmTruncateToInt64(LWasmTruncateToInt64* lir) { + FloatRegister input = ToFloatRegister(lir->input()); + Register64 output = ToOutRegister64(lir); + + MWasmTruncateToInt64* mir = lir->mir(); + MIRType fromType = mir->input()->type(); + + MOZ_ASSERT(fromType == MIRType::Double || fromType == MIRType::Float32); + + auto* ool = new (alloc()) OutOfLineWasmTruncateCheck(mir, input, output); + addOutOfLineCode(ool, mir); + + Label* oolEntry = ool->entry(); + Label* oolRejoin = ool->rejoin(); + bool isSaturating = mir->isSaturating(); + + if (fromType == MIRType::Double) { + if (mir->isUnsigned()) { + masm.wasmTruncateDoubleToUInt64(input, output, isSaturating, oolEntry, + oolRejoin, InvalidFloatReg); + } else { + masm.wasmTruncateDoubleToInt64(input, output, isSaturating, oolEntry, + oolRejoin, InvalidFloatReg); + } + } else { + if (mir->isUnsigned()) { + masm.wasmTruncateFloat32ToUInt64(input, output, isSaturating, oolEntry, + oolRejoin, InvalidFloatReg); + } else { + masm.wasmTruncateFloat32ToInt64(input, output, isSaturating, oolEntry, + oolRejoin, InvalidFloatReg); + } + } +} + +void CodeGenerator::visitInt64ToFloatingPoint(LInt64ToFloatingPoint* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + FloatRegister output = ToFloatRegister(lir->output()); + + MIRType outputType = lir->mir()->type(); + MOZ_ASSERT(outputType == MIRType::Double || outputType == MIRType::Float32); + + if (outputType == MIRType::Double) { + if (lir->mir()->isUnsigned()) { + masm.convertUInt64ToDouble(input, output, Register::Invalid()); + } else { + masm.convertInt64ToDouble(input, output); + } + } else { + if (lir->mir()->isUnsigned()) { + masm.convertUInt64ToFloat32(input, output, Register::Invalid()); + } else { + masm.convertInt64ToFloat32(input, output); + } + } +} + +void CodeGenerator::visitTestI64AndBranch(LTestI64AndBranch* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + MBasicBlock* ifTrue = lir->ifTrue(); + MBasicBlock* ifFalse = lir->ifFalse(); + + emitBranch(input.reg, Imm32(0), Assembler::NonZero, ifTrue, ifFalse); +} + +void CodeGenerator::visitTestIAndBranch(LTestIAndBranch* test) { + const LAllocation* opd = test->getOperand(0); + MBasicBlock* ifTrue = test->ifTrue(); + MBasicBlock* ifFalse = test->ifFalse(); + + emitBranch(ToRegister(opd), Imm32(0), Assembler::NonZero, ifTrue, ifFalse); +} + +void CodeGenerator::visitMinMaxD(LMinMaxD* ins) { + FloatRegister first = ToFloatRegister(ins->first()); + FloatRegister second = ToFloatRegister(ins->second()); + + MOZ_ASSERT(first == ToFloatRegister(ins->output())); + + if (ins->mir()->isMax()) { + masm.maxDouble(second, first, true); + } else { + masm.minDouble(second, first, true); + } +} + +void CodeGenerator::visitMinMaxF(LMinMaxF* ins) { + FloatRegister first = ToFloatRegister(ins->first()); + FloatRegister second = ToFloatRegister(ins->second()); + + MOZ_ASSERT(first == ToFloatRegister(ins->output())); + + if (ins->mir()->isMax()) { + masm.maxFloat32(second, first, true); + } else { + masm.minFloat32(second, first, true); + } +} + +void CodeGenerator::visitAddI(LAddI* ins) { + const LAllocation* lhs = ins->getOperand(0); + const LAllocation* rhs = ins->getOperand(1); + const LDefinition* dest = ins->getDef(0); + + MOZ_ASSERT(rhs->isConstant() || rhs->isGeneralReg()); + + // If there is no snapshot, we don't need to check for overflow + if (!ins->snapshot()) { + if (rhs->isConstant()) { + masm.ma_add32(ToRegister(dest), ToRegister(lhs), Imm32(ToInt32(rhs))); + } else { + masm.addw(ToRegister(dest), ToRegister(lhs), ToRegister(rhs)); + } + return; + } + + Label overflow; + if (rhs->isConstant()) { + masm.ma_add32TestOverflow(ToRegister(dest), ToRegister(lhs), + Imm32(ToInt32(rhs)), &overflow); + } else { + masm.ma_add32TestOverflow(ToRegister(dest), ToRegister(lhs), + ToRegister(rhs), &overflow); + } + + bailoutFrom(&overflow, ins->snapshot()); +} + +void CodeGenerator::visitAddI64(LAddI64* lir) { + const LInt64Allocation lhs = lir->getInt64Operand(LAddI64::Lhs); + const LInt64Allocation rhs = lir->getInt64Operand(LAddI64::Rhs); + + MOZ_ASSERT(ToOutRegister64(lir) == ToRegister64(lhs)); + + if (IsConstant(rhs)) { + masm.add64(Imm64(ToInt64(rhs)), ToRegister64(lhs)); + return; + } + + masm.add64(ToOperandOrRegister64(rhs), ToRegister64(lhs)); +} + +void CodeGenerator::visitSubI(LSubI* ins) { + const LAllocation* lhs = ins->getOperand(0); + const LAllocation* rhs = ins->getOperand(1); + const LDefinition* dest = ins->getDef(0); + + MOZ_ASSERT(rhs->isConstant() || rhs->isGeneralReg()); + + // If there is no snapshot, we don't need to check for overflow + + if (!ins->snapshot()) { + if (rhs->isConstant()) { + masm.ma_sub32(ToRegister(dest), ToRegister(lhs), Imm32(ToInt32(rhs))); + } else { + masm.ma_sub32(ToRegister(dest), ToRegister(lhs), ToRegister(rhs)); + } + return; + } + + Label overflow; + if (rhs->isConstant()) { + masm.ma_sub32TestOverflow(ToRegister(dest), ToRegister(lhs), + Imm32(ToInt32(rhs)), &overflow); + } else { + masm.ma_sub32TestOverflow(ToRegister(dest), ToRegister(lhs), + ToRegister(rhs), &overflow); + } + + bailoutFrom(&overflow, ins->snapshot()); +} + +void CodeGenerator::visitSubI64(LSubI64* lir) { + const LInt64Allocation lhs = lir->getInt64Operand(LSubI64::Lhs); + const LInt64Allocation rhs = lir->getInt64Operand(LSubI64::Rhs); + + MOZ_ASSERT(ToOutRegister64(lir) == ToRegister64(lhs)); + + if (IsConstant(rhs)) { + masm.sub64(Imm64(ToInt64(rhs)), ToRegister64(lhs)); + return; + } + + masm.sub64(ToOperandOrRegister64(rhs), ToRegister64(lhs)); +} + +void CodeGenerator::visitMulI(LMulI* ins) { + const LAllocation* lhs = ins->lhs(); + const LAllocation* rhs = ins->rhs(); + Register dest = ToRegister(ins->output()); + MMul* mul = ins->mir(); + + MOZ_ASSERT_IF(mul->mode() == MMul::Integer, + !mul->canBeNegativeZero() && !mul->canOverflow()); + + if (rhs->isConstant()) { + int32_t constant = ToInt32(rhs); + Register src = ToRegister(lhs); + + // Bailout on -0.0 + if (mul->canBeNegativeZero() && constant <= 0) { + Assembler::Condition cond = + (constant == 0) ? Assembler::LessThan : Assembler::Equal; + bailoutCmp32(cond, src, Imm32(0), ins->snapshot()); + } + + switch (constant) { + case -1: + if (mul->canOverflow()) { + bailoutCmp32(Assembler::Equal, src, Imm32(INT32_MIN), + ins->snapshot()); + } + + masm.ma_sub32(dest, zero, src); + break; + case 0: + masm.move32(zero, dest); + break; + case 1: + masm.move32(src, dest); + break; + case 2: + if (mul->canOverflow()) { + Label mulTwoOverflow; + masm.ma_add32TestOverflow(dest, src, src, &mulTwoOverflow); + + bailoutFrom(&mulTwoOverflow, ins->snapshot()); + } else { + masm.addw(dest, src, src); + } + break; + default: + uint32_t shift = FloorLog2(constant); + + if (!mul->canOverflow() && (constant > 0)) { + // If it cannot overflow, we can do lots of optimizations. + uint32_t rest = constant - (1 << shift); + + // See if the constant has one bit set, meaning it can be + // encoded as a bitshift. + if ((1 << shift) == constant) { + masm.slliw(dest, src, shift % 32); + return; + } + + // If the constant cannot be encoded as (1<<C1), see if it can + // be encoded as (1<<C1) | (1<<C2), which can be computed + // using an add and a shift. + uint32_t shift_rest = FloorLog2(rest); + if (src != dest && (1u << shift_rest) == rest) { + masm.slliw(dest, src, (shift - shift_rest) % 32); + masm.add32(src, dest); + if (shift_rest != 0) { + masm.slliw(dest, dest, shift_rest % 32); + } + return; + } + } + + if (mul->canOverflow() && (constant > 0) && (src != dest)) { + // To stay on the safe side, only optimize things that are a + // power of 2. + + if ((1 << shift) == constant) { + ScratchRegisterScope scratch(masm); + // dest = lhs * pow(2, shift) + masm.slliw(dest, src, shift % 32); + // At runtime, check (lhs == dest >> shift), if this does + // not hold, some bits were lost due to overflow, and the + // computation should be resumed as a double. + masm.sraiw(scratch, dest, shift % 32); + bailoutCmp32(Assembler::NotEqual, src, Register(scratch), + ins->snapshot()); + return; + } + } + + if (mul->canOverflow()) { + Label mulConstOverflow; + masm.ma_mul32TestOverflow(dest, ToRegister(lhs), Imm32(ToInt32(rhs)), + &mulConstOverflow); + + bailoutFrom(&mulConstOverflow, ins->snapshot()); + } else { + masm.ma_mul32(dest, src, Imm32(ToInt32(rhs))); + } + break; + } + } else { + Label multRegOverflow; + + if (mul->canOverflow()) { + masm.ma_mul32TestOverflow(dest, ToRegister(lhs), ToRegister(rhs), + &multRegOverflow); + bailoutFrom(&multRegOverflow, ins->snapshot()); + } else { + masm.mulw(dest, ToRegister(lhs), ToRegister(rhs)); + } + + if (mul->canBeNegativeZero()) { + Label done; + masm.ma_b(dest, dest, &done, Assembler::NonZero, ShortJump); + + // Result is -0 if lhs or rhs is negative. + // In that case result must be double value so bailout + UseScratchRegisterScope temps(&masm); + Register scratch = temps.Acquire(); + masm.or_(scratch, ToRegister(lhs), ToRegister(rhs)); + bailoutCmp32(Assembler::Signed, scratch, scratch, ins->snapshot()); + + masm.bind(&done); + } + } +} + +void CodeGenerator::visitMulI64(LMulI64* lir) { + const LInt64Allocation lhs = lir->getInt64Operand(LMulI64::Lhs); + const LInt64Allocation rhs = lir->getInt64Operand(LMulI64::Rhs); + const Register64 output = ToOutRegister64(lir); + + if (IsConstant(rhs)) { + int64_t constant = ToInt64(rhs); + switch (constant) { + case -1: + masm.neg64(ToRegister64(lhs)); + return; + case 0: + masm.xor64(ToRegister64(lhs), ToRegister64(lhs)); + return; + case 1: + // nop + return; + case 2: + masm.add(output.reg, ToRegister64(lhs).reg, ToRegister64(lhs).reg); + return; + default: + if (constant > 0) { + if (mozilla::IsPowerOfTwo(static_cast<uint64_t>(constant + 1))) { + ScratchRegisterScope scratch(masm); + masm.movePtr(ToRegister64(lhs).reg, scratch); + masm.slli(output.reg, ToRegister64(lhs).reg, + FloorLog2(constant + 1)); + masm.sub64(scratch, output); + return; + } else if (mozilla::IsPowerOfTwo( + static_cast<uint64_t>(constant - 1))) { + int32_t shift = mozilla::FloorLog2(constant - 1); + ScratchRegisterScope scratch(masm); + masm.movePtr(ToRegister64(lhs).reg, scratch); + masm.slli(output.reg, ToRegister64(lhs).reg, shift); + masm.add64(scratch, output); + return; + } + // Use shift if constant is power of 2. + int32_t shift = mozilla::FloorLog2(constant); + if (int64_t(1) << shift == constant) { + masm.lshift64(Imm32(shift), ToRegister64(lhs)); + return; + } + } + Register temp = ToTempRegisterOrInvalid(lir->temp()); + masm.mul64(Imm64(constant), ToRegister64(lhs), temp); + } + } else { + Register temp = ToTempRegisterOrInvalid(lir->temp()); + masm.mul64(ToOperandOrRegister64(rhs), ToRegister64(lhs), temp); + } +} + +void CodeGenerator::visitDivI(LDivI* ins) { + // Extract the registers from this instruction + Register lhs = ToRegister(ins->lhs()); + Register rhs = ToRegister(ins->rhs()); + Register dest = ToRegister(ins->output()); + Register temp = ToRegister(ins->getTemp(0)); + MDiv* mir = ins->mir(); + + Label done; + + // Handle divide by zero. + if (mir->canBeDivideByZero()) { + if (mir->trapOnError()) { + Label nonZero; + masm.ma_b(rhs, rhs, &nonZero, Assembler::NonZero); + masm.wasmTrap(wasm::Trap::IntegerDivideByZero, mir->bytecodeOffset()); + masm.bind(&nonZero); + } else if (mir->canTruncateInfinities()) { + // Truncated division by zero is zero (Infinity|0 == 0) + Label notzero; + masm.ma_b(rhs, rhs, ¬zero, Assembler::NonZero, ShortJump); + masm.move32(Imm32(0), dest); + masm.ma_branch(&done, ShortJump); + masm.bind(¬zero); + } else { + MOZ_ASSERT(mir->fallible()); + bailoutCmp32(Assembler::Zero, rhs, rhs, ins->snapshot()); + } + } + + // Handle an integer overflow exception from -2147483648 / -1. + if (mir->canBeNegativeOverflow()) { + Label notMinInt; + masm.move32(Imm32(INT32_MIN), temp); + masm.ma_b(lhs, temp, ¬MinInt, Assembler::NotEqual, ShortJump); + + masm.move32(Imm32(-1), temp); + if (mir->trapOnError()) { + Label ok; + masm.ma_b(rhs, temp, &ok, Assembler::NotEqual); + masm.wasmTrap(wasm::Trap::IntegerOverflow, mir->bytecodeOffset()); + masm.bind(&ok); + } else if (mir->canTruncateOverflow()) { + // (-INT32_MIN)|0 == INT32_MIN + Label skip; + masm.ma_b(rhs, temp, &skip, Assembler::NotEqual, ShortJump); + masm.move32(Imm32(INT32_MIN), dest); + masm.ma_branch(&done, ShortJump); + masm.bind(&skip); + } else { + MOZ_ASSERT(mir->fallible()); + bailoutCmp32(Assembler::Equal, rhs, temp, ins->snapshot()); + } + masm.bind(¬MinInt); + } + + // Handle negative 0. (0/-Y) + if (!mir->canTruncateNegativeZero() && mir->canBeNegativeZero()) { + Label nonzero; + masm.ma_b(lhs, lhs, &nonzero, Assembler::NonZero, ShortJump); + bailoutCmp32(Assembler::LessThan, rhs, Imm32(0), ins->snapshot()); + masm.bind(&nonzero); + } + // Note: above safety checks could not be verified as Ion seems to be + // smarter and requires double arithmetic in such cases. + + // All regular. Lets call div. + if (mir->canTruncateRemainder()) { + masm.ma_div32(dest, lhs, rhs); + } else { + MOZ_ASSERT(mir->fallible()); + + Label remainderNonZero; + masm.ma_div_branch_overflow(dest, lhs, rhs, &remainderNonZero); + bailoutFrom(&remainderNonZero, ins->snapshot()); + } + + masm.bind(&done); +} + +void CodeGenerator::visitDivPowTwoI(LDivPowTwoI* ins) { + Register lhs = ToRegister(ins->numerator()); + Register dest = ToRegister(ins->output()); + Register tmp = ToRegister(ins->getTemp(0)); + int32_t shift = ins->shift(); + + if (shift != 0) { + MDiv* mir = ins->mir(); + if (!mir->isTruncated()) { + // If the remainder is going to be != 0, bailout since this must + // be a double. + masm.slliw(tmp, lhs, (32 - shift) % 32); + bailoutCmp32(Assembler::NonZero, tmp, tmp, ins->snapshot()); + } + + if (!mir->canBeNegativeDividend()) { + // Numerator is unsigned, so needs no adjusting. Do the shift. + masm.sraiw(dest, lhs, shift % 32); + return; + } + + // Adjust the value so that shifting produces a correctly rounded result + // when the numerator is negative. See 10-1 "Signed Division by a Known + // Power of 2" in Henry S. Warren, Jr.'s Hacker's Delight. + if (shift > 1) { + masm.sraiw(tmp, lhs, 31); + masm.srliw(tmp, tmp, (32 - shift) % 32); + masm.add32(lhs, tmp); + } else { + masm.srliw(tmp, lhs, (32 - shift) % 32); + masm.add32(lhs, tmp); + } + + // Do the shift. + masm.sraiw(dest, tmp, shift % 32); + } else { + masm.move32(lhs, dest); + } +} + +void CodeGenerator::visitModI(LModI* ins) { + // Extract the registers from this instruction + Register lhs = ToRegister(ins->lhs()); + Register rhs = ToRegister(ins->rhs()); + Register dest = ToRegister(ins->output()); + Register callTemp = ToRegister(ins->callTemp()); + MMod* mir = ins->mir(); + Label done, prevent; + + masm.move32(lhs, callTemp); + + // Prevent INT_MIN % -1; + // The integer division will give INT_MIN, but we want -(double)INT_MIN. + if (mir->canBeNegativeDividend()) { + masm.ma_b(lhs, Imm32(INT_MIN), &prevent, Assembler::NotEqual, ShortJump); + if (mir->isTruncated()) { + // (INT_MIN % -1)|0 == 0 + Label skip; + masm.ma_b(rhs, Imm32(-1), &skip, Assembler::NotEqual, ShortJump); + masm.move32(Imm32(0), dest); + masm.ma_branch(&done, ShortJump); + masm.bind(&skip); + } else { + MOZ_ASSERT(mir->fallible()); + bailoutCmp32(Assembler::Equal, rhs, Imm32(-1), ins->snapshot()); + } + masm.bind(&prevent); + } + + // 0/X (with X < 0) is bad because both of these values *should* be + // doubles, and the result should be -0.0, which cannot be represented in + // integers. X/0 is bad because it will give garbage (or abort), when it + // should give either \infty, -\infty or NAN. + + // Prevent 0 / X (with X < 0) and X / 0 + // testing X / Y. Compare Y with 0. + // There are three cases: (Y < 0), (Y == 0) and (Y > 0) + // If (Y < 0), then we compare X with 0, and bail if X == 0 + // If (Y == 0), then we simply want to bail. + // if (Y > 0), we don't bail. + + if (mir->canBeDivideByZero()) { + if (mir->isTruncated()) { + if (mir->trapOnError()) { + Label nonZero; + masm.ma_b(rhs, rhs, &nonZero, Assembler::NonZero); + masm.wasmTrap(wasm::Trap::IntegerDivideByZero, mir->bytecodeOffset()); + masm.bind(&nonZero); + } else { + Label skip; + masm.ma_b(rhs, Imm32(0), &skip, Assembler::NotEqual, ShortJump); + masm.move32(Imm32(0), dest); + masm.ma_branch(&done, ShortJump); + masm.bind(&skip); + } + } else { + MOZ_ASSERT(mir->fallible()); + bailoutCmp32(Assembler::Equal, rhs, Imm32(0), ins->snapshot()); + } + } + + if (mir->canBeNegativeDividend()) { + Label notNegative; + masm.ma_b(rhs, Imm32(0), ¬Negative, Assembler::GreaterThan, ShortJump); + if (mir->isTruncated()) { + // NaN|0 == 0 and (0 % -X)|0 == 0 + Label skip; + masm.ma_b(lhs, Imm32(0), &skip, Assembler::NotEqual, ShortJump); + masm.move32(Imm32(0), dest); + masm.ma_branch(&done, ShortJump); + masm.bind(&skip); + } else { + MOZ_ASSERT(mir->fallible()); + bailoutCmp32(Assembler::Equal, lhs, Imm32(0), ins->snapshot()); + } + masm.bind(¬Negative); + } + + masm.ma_mod32(dest, lhs, rhs); + + // If X%Y == 0 and X < 0, then we *actually* wanted to return -0.0 + if (mir->canBeNegativeDividend()) { + if (mir->isTruncated()) { + // -0.0|0 == 0 + } else { + MOZ_ASSERT(mir->fallible()); + // See if X < 0 + masm.ma_b(dest, Imm32(0), &done, Assembler::NotEqual, ShortJump); + bailoutCmp32(Assembler::Signed, callTemp, Imm32(0), ins->snapshot()); + } + } + masm.bind(&done); +} + +void CodeGenerator::visitModPowTwoI(LModPowTwoI* ins) { + Register in = ToRegister(ins->getOperand(0)); + Register out = ToRegister(ins->getDef(0)); + MMod* mir = ins->mir(); + Label negative, done; + + masm.move32(in, out); + masm.ma_b(in, in, &done, Assembler::Zero, ShortJump); + // Switch based on sign of the lhs. + // Positive numbers are just a bitmask + masm.ma_b(in, in, &negative, Assembler::Signed, ShortJump); + { + masm.and32(Imm32((1 << ins->shift()) - 1), out); + masm.ma_branch(&done, ShortJump); + } + + // Negative numbers need a negate, bitmask, negate + { + masm.bind(&negative); + masm.neg32(out); + masm.and32(Imm32((1 << ins->shift()) - 1), out); + masm.neg32(out); + } + if (mir->canBeNegativeDividend()) { + if (!mir->isTruncated()) { + MOZ_ASSERT(mir->fallible()); + bailoutCmp32(Assembler::Equal, out, zero, ins->snapshot()); + } else { + // -0|0 == 0 + } + } + masm.bind(&done); +} + +void CodeGenerator::visitModMaskI(LModMaskI* ins) { + Register src = ToRegister(ins->getOperand(0)); + Register dest = ToRegister(ins->getDef(0)); + Register tmp0 = ToRegister(ins->getTemp(0)); + Register tmp1 = ToRegister(ins->getTemp(1)); + MMod* mir = ins->mir(); + + if (!mir->isTruncated() && mir->canBeNegativeDividend()) { + MOZ_ASSERT(mir->fallible()); + + Label bail; + masm.ma_mod_mask(src, dest, tmp0, tmp1, ins->shift(), &bail); + bailoutFrom(&bail, ins->snapshot()); + } else { + masm.ma_mod_mask(src, dest, tmp0, tmp1, ins->shift(), nullptr); + } +} + +void CodeGenerator::visitBitNotI(LBitNotI* ins) { + const LAllocation* input = ins->getOperand(0); + const LDefinition* dest = ins->getDef(0); + MOZ_ASSERT(!input->isConstant()); + + masm.nor(ToRegister(dest), ToRegister(input), zero); +} + +void CodeGenerator::visitBitNotI64(LBitNotI64* ins) { + const LAllocation* input = ins->getOperand(0); + MOZ_ASSERT(!input->isConstant()); + Register inputReg = ToRegister(input); + MOZ_ASSERT(inputReg == ToRegister(ins->output())); + masm.nor(inputReg, inputReg, zero); +} + +void CodeGenerator::visitBitOpI(LBitOpI* ins) { + const LAllocation* lhs = ins->getOperand(0); + const LAllocation* rhs = ins->getOperand(1); + const LDefinition* dest = ins->getDef(0); + // all of these bitops should be either imm32's, or integer registers. + switch (ins->bitop()) { + case JSOp::BitOr: + if (rhs->isConstant()) { + masm.ma_or(ToRegister(dest), ToRegister(lhs), Imm32(ToInt32(rhs))); + } else { + masm.or_(ToRegister(dest), ToRegister(lhs), ToRegister(rhs)); + masm.slliw(ToRegister(dest), ToRegister(dest), 0); + } + break; + case JSOp::BitXor: + if (rhs->isConstant()) { + masm.ma_xor(ToRegister(dest), ToRegister(lhs), Imm32(ToInt32(rhs))); + } else { + masm.ma_xor(ToRegister(dest), ToRegister(lhs), + Operand(ToRegister(rhs))); + masm.slliw(ToRegister(dest), ToRegister(dest), 0); + } + break; + case JSOp::BitAnd: + if (rhs->isConstant()) { + masm.ma_and(ToRegister(dest), ToRegister(lhs), Imm32(ToInt32(rhs))); + } else { + masm.and_(ToRegister(dest), ToRegister(lhs), ToRegister(rhs)); + masm.slliw(ToRegister(dest), ToRegister(dest), 0); + } + break; + default: + MOZ_CRASH("unexpected binary opcode"); + } +} + +void CodeGenerator::visitBitOpI64(LBitOpI64* lir) { + const LInt64Allocation lhs = lir->getInt64Operand(LBitOpI64::Lhs); + const LInt64Allocation rhs = lir->getInt64Operand(LBitOpI64::Rhs); + + MOZ_ASSERT(ToOutRegister64(lir) == ToRegister64(lhs)); + + switch (lir->bitop()) { + case JSOp::BitOr: + if (IsConstant(rhs)) { + masm.or64(Imm64(ToInt64(rhs)), ToRegister64(lhs)); + } else { + masm.or64(ToOperandOrRegister64(rhs), ToRegister64(lhs)); + } + break; + case JSOp::BitXor: + if (IsConstant(rhs)) { + masm.xor64(Imm64(ToInt64(rhs)), ToRegister64(lhs)); + } else { + masm.xor64(ToOperandOrRegister64(rhs), ToRegister64(lhs)); + } + break; + case JSOp::BitAnd: + if (IsConstant(rhs)) { + masm.and64(Imm64(ToInt64(rhs)), ToRegister64(lhs)); + } else { + masm.and64(ToOperandOrRegister64(rhs), ToRegister64(lhs)); + } + break; + default: + MOZ_CRASH("unexpected binary opcode"); + } +} + +void CodeGenerator::visitShiftI(LShiftI* ins) { + Register lhs = ToRegister(ins->lhs()); + const LAllocation* rhs = ins->rhs(); + Register dest = ToRegister(ins->output()); + + if (rhs->isConstant()) { + int32_t shift = ToInt32(rhs) & 0x1F; + switch (ins->bitop()) { + case JSOp::Lsh: + if (shift) { + masm.slliw(dest, lhs, shift % 32); + } else { + masm.move32(lhs, dest); + } + break; + case JSOp::Rsh: + if (shift) { + masm.sraiw(dest, lhs, shift % 32); + } else { + masm.move32(lhs, dest); + } + break; + case JSOp::Ursh: + if (shift) { + masm.srliw(dest, lhs, shift % 32); + } else { + // x >>> 0 can overflow. + if (ins->mir()->toUrsh()->fallible()) { + bailoutCmp32(Assembler::LessThan, lhs, Imm32(0), ins->snapshot()); + } + masm.move32(lhs, dest); + } + break; + default: + MOZ_CRASH("Unexpected shift op"); + } + } else { + // The shift amounts should be AND'ed into the 0-31 range + masm.ma_and(dest, ToRegister(rhs), Imm32(0x1F)); + + switch (ins->bitop()) { + case JSOp::Lsh: + masm.sllw(dest, lhs, dest); + break; + case JSOp::Rsh: + masm.sraw(dest, lhs, dest); + break; + case JSOp::Ursh: + masm.srlw(dest, lhs, dest); + if (ins->mir()->toUrsh()->fallible()) { + // x >>> 0 can overflow. + bailoutCmp32(Assembler::LessThan, dest, Imm32(0), ins->snapshot()); + } + break; + default: + MOZ_CRASH("Unexpected shift op"); + } + } +} + +void CodeGenerator::visitShiftI64(LShiftI64* lir) { + const LInt64Allocation lhs = lir->getInt64Operand(LShiftI64::Lhs); + LAllocation* rhs = lir->getOperand(LShiftI64::Rhs); + + MOZ_ASSERT(ToOutRegister64(lir) == ToRegister64(lhs)); + + if (rhs->isConstant()) { + int32_t shift = int32_t(rhs->toConstant()->toInt64() & 0x3F); + switch (lir->bitop()) { + case JSOp::Lsh: + if (shift) { + masm.lshift64(Imm32(shift), ToRegister64(lhs)); + } + break; + case JSOp::Rsh: + if (shift) { + masm.rshift64Arithmetic(Imm32(shift), ToRegister64(lhs)); + } + break; + case JSOp::Ursh: + if (shift) { + masm.rshift64(Imm32(shift), ToRegister64(lhs)); + } + break; + default: + MOZ_CRASH("Unexpected shift op"); + } + return; + } + + switch (lir->bitop()) { + case JSOp::Lsh: + masm.lshift64(ToRegister(rhs), ToRegister64(lhs)); + break; + case JSOp::Rsh: + masm.rshift64Arithmetic(ToRegister(rhs), ToRegister64(lhs)); + break; + case JSOp::Ursh: + masm.rshift64(ToRegister(rhs), ToRegister64(lhs)); + break; + default: + MOZ_CRASH("Unexpected shift op"); + } +} + +void CodeGenerator::visitRotateI64(LRotateI64* lir) { + MRotate* mir = lir->mir(); + LAllocation* count = lir->count(); + + Register64 input = ToRegister64(lir->input()); + Register64 output = ToOutRegister64(lir); + Register temp = ToTempRegisterOrInvalid(lir->temp()); + + MOZ_ASSERT(input == output); + + if (count->isConstant()) { + int32_t c = int32_t(count->toConstant()->toInt64() & 0x3F); + if (!c) { + return; + } + if (mir->isLeftRotate()) { + masm.rotateLeft64(Imm32(c), input, output, temp); + } else { + masm.rotateRight64(Imm32(c), input, output, temp); + } + } else { + if (mir->isLeftRotate()) { + masm.rotateLeft64(ToRegister(count), input, output, temp); + } else { + masm.rotateRight64(ToRegister(count), input, output, temp); + } + } +} + +void CodeGenerator::visitUrshD(LUrshD* ins) { + Register lhs = ToRegister(ins->lhs()); + Register temp = ToRegister(ins->temp()); + + const LAllocation* rhs = ins->rhs(); + FloatRegister out = ToFloatRegister(ins->output()); + + if (rhs->isConstant()) { + masm.srliw(temp, lhs, ToInt32(rhs) % 32); + } else { + masm.srlw(temp, lhs, ToRegister(rhs)); + } + + masm.convertUInt32ToDouble(temp, out); +} + +void CodeGenerator::visitClzI(LClzI* ins) { + Register input = ToRegister(ins->input()); + Register output = ToRegister(ins->output()); + + masm.Clz32(output, input); +} + +void CodeGenerator::visitCtzI(LCtzI* ins) { + Register input = ToRegister(ins->input()); + Register output = ToRegister(ins->output()); + + masm.Ctz32(output, input); +} + +void CodeGenerator::visitPopcntI(LPopcntI* ins) { + Register input = ToRegister(ins->input()); + Register output = ToRegister(ins->output()); + Register tmp = ToRegister(ins->temp0()); + + masm.Popcnt32(input, output, tmp); +} + +void CodeGenerator::visitPopcntI64(LPopcntI64* ins) { + Register64 input = ToRegister64(ins->getInt64Operand(0)); + Register64 output = ToOutRegister64(ins); + Register tmp = ToRegister(ins->getTemp(0)); + + masm.Popcnt64(input.scratchReg(), output.scratchReg(), tmp); +} + +void CodeGenerator::visitPowHalfD(LPowHalfD* ins) { + FloatRegister input = ToFloatRegister(ins->input()); + FloatRegister output = ToFloatRegister(ins->output()); + ScratchDoubleScope fpscratch(masm); + + Label done, skip; + + // Masm.pow(-Infinity, 0.5) == Infinity. + masm.loadConstantDouble(NegativeInfinity<double>(), fpscratch); + UseScratchRegisterScope temps(&masm); + Register scratch = temps.Acquire(); + + masm.ma_compareF64(scratch, Assembler::DoubleNotEqualOrUnordered, input, + fpscratch); + masm.ma_branch(&skip, Assembler::Equal, scratch, Operand(1)); + // masm.ma_bc_d(input, fpscratch, &skip, Assembler::DoubleNotEqualOrUnordered, + // ShortJump); + masm.fneg_d(output, fpscratch); + masm.ma_branch(&done, ShortJump); + + masm.bind(&skip); + // Math.pow(-0, 0.5) == 0 == Math.pow(0, 0.5). + // Adding 0 converts any -0 to 0. + masm.loadConstantDouble(0.0, fpscratch); + masm.fadd_d(output, input, fpscratch); + masm.fsqrt_d(output, output); + + masm.bind(&done); +} + +void CodeGenerator::visitMathD(LMathD* math) { + FloatRegister src1 = ToFloatRegister(math->getOperand(0)); + FloatRegister src2 = ToFloatRegister(math->getOperand(1)); + FloatRegister output = ToFloatRegister(math->getDef(0)); + + switch (math->jsop()) { + case JSOp::Add: + masm.fadd_d(output, src1, src2); + break; + case JSOp::Sub: + masm.fsub_d(output, src1, src2); + break; + case JSOp::Mul: + masm.fmul_d(output, src1, src2); + break; + case JSOp::Div: + masm.fdiv_d(output, src1, src2); + break; + default: + MOZ_CRASH("unexpected opcode"); + } +} + +void CodeGenerator::visitMathF(LMathF* math) { + FloatRegister src1 = ToFloatRegister(math->getOperand(0)); + FloatRegister src2 = ToFloatRegister(math->getOperand(1)); + FloatRegister output = ToFloatRegister(math->getDef(0)); + + switch (math->jsop()) { + case JSOp::Add: + masm.fadd_s(output, src1, src2); + break; + case JSOp::Sub: + masm.fsub_s(output, src1, src2); + break; + case JSOp::Mul: + masm.fmul_s(output, src1, src2); + break; + case JSOp::Div: + masm.fdiv_s(output, src1, src2); + break; + default: + MOZ_CRASH("unexpected opcode"); + } +} + +void CodeGenerator::visitTruncateDToInt32(LTruncateDToInt32* ins) { + emitTruncateDouble(ToFloatRegister(ins->input()), ToRegister(ins->output()), + ins->mir()); +} + +void CodeGenerator::visitTruncateFToInt32(LTruncateFToInt32* ins) { + emitTruncateFloat32(ToFloatRegister(ins->input()), ToRegister(ins->output()), + ins->mir()); +} + +void CodeGenerator::visitWasmBuiltinTruncateDToInt32( + LWasmBuiltinTruncateDToInt32* lir) { + emitTruncateDouble(ToFloatRegister(lir->getOperand(0)), + ToRegister(lir->getDef(0)), lir->mir()); +} + +void CodeGenerator::visitWasmBuiltinTruncateFToInt32( + LWasmBuiltinTruncateFToInt32* lir) { + emitTruncateFloat32(ToFloatRegister(lir->getOperand(0)), + ToRegister(lir->getDef(0)), lir->mir()); +} + +void CodeGenerator::visitWasmTruncateToInt32(LWasmTruncateToInt32* lir) { + auto input = ToFloatRegister(lir->input()); + auto output = ToRegister(lir->output()); + + MWasmTruncateToInt32* mir = lir->mir(); + MIRType fromType = mir->input()->type(); + + MOZ_ASSERT(fromType == MIRType::Double || fromType == MIRType::Float32); + + auto* ool = new (alloc()) OutOfLineWasmTruncateCheck(mir, input, output); + addOutOfLineCode(ool, mir); + + Label* oolEntry = ool->entry(); + if (mir->isUnsigned()) { + if (fromType == MIRType::Double) { + masm.wasmTruncateDoubleToUInt32(input, output, mir->isSaturating(), + oolEntry); + } else if (fromType == MIRType::Float32) { + masm.wasmTruncateFloat32ToUInt32(input, output, mir->isSaturating(), + oolEntry); + } else { + MOZ_CRASH("unexpected type"); + } + + masm.bind(ool->rejoin()); + return; + } + + if (fromType == MIRType::Double) { + masm.wasmTruncateDoubleToInt32(input, output, mir->isSaturating(), + oolEntry); + } else if (fromType == MIRType::Float32) { + masm.wasmTruncateFloat32ToInt32(input, output, mir->isSaturating(), + oolEntry); + } else { + MOZ_CRASH("unexpected type"); + } + + masm.bind(ool->rejoin()); +} + +void CodeGenerator::visitCopySignF(LCopySignF* ins) { + FloatRegister lhs = ToFloatRegister(ins->getOperand(0)); + FloatRegister rhs = ToFloatRegister(ins->getOperand(1)); + FloatRegister output = ToFloatRegister(ins->getDef(0)); + + masm.fsgnj_s(output, lhs, rhs); +} + +void CodeGenerator::visitCopySignD(LCopySignD* ins) { + FloatRegister lhs = ToFloatRegister(ins->getOperand(0)); + FloatRegister rhs = ToFloatRegister(ins->getOperand(1)); + FloatRegister output = ToFloatRegister(ins->getDef(0)); + + masm.fsgnj_d(output, lhs, rhs); +} + +void CodeGenerator::visitValue(LValue* value) { + const ValueOperand out = ToOutValue(value); + + masm.moveValue(value->value(), out); +} + +void CodeGenerator::visitDouble(LDouble* ins) { + const LDefinition* out = ins->getDef(0); + + masm.loadConstantDouble(ins->value(), ToFloatRegister(out)); +} + +void CodeGenerator::visitFloat32(LFloat32* ins) { + const LDefinition* out = ins->getDef(0); + masm.loadConstantFloat32(ins->value(), ToFloatRegister(out)); +} + +void CodeGenerator::visitTestDAndBranch(LTestDAndBranch* test) { + FloatRegister input = ToFloatRegister(test->input()); + ScratchDoubleScope fpscratch(masm); + + MBasicBlock* ifTrue = test->ifTrue(); + MBasicBlock* ifFalse = test->ifFalse(); + + masm.loadConstantDouble(0.0, fpscratch); + // If 0, or NaN, the result is false. + if (isNextBlock(ifFalse->lir())) { + branchToBlock(DoubleFloat, input, fpscratch, ifTrue, + Assembler::DoubleNotEqual); + } else { + branchToBlock(DoubleFloat, input, fpscratch, ifFalse, + Assembler::DoubleEqualOrUnordered); + jumpToBlock(ifTrue); + } +} + +void CodeGenerator::visitTestFAndBranch(LTestFAndBranch* test) { + FloatRegister input = ToFloatRegister(test->input()); + ScratchFloat32Scope fpscratch(masm); + + MBasicBlock* ifTrue = test->ifTrue(); + MBasicBlock* ifFalse = test->ifFalse(); + + masm.loadConstantFloat32(0.0f, fpscratch); + // If 0, or NaN, the result is false. + + if (isNextBlock(ifFalse->lir())) { + branchToBlock(SingleFloat, input, fpscratch, ifTrue, + Assembler::DoubleNotEqual); + } else { + branchToBlock(SingleFloat, input, fpscratch, ifFalse, + Assembler::DoubleEqualOrUnordered); + jumpToBlock(ifTrue); + } +} + +void CodeGenerator::visitCompareD(LCompareD* comp) { + FloatRegister lhs = ToFloatRegister(comp->left()); + FloatRegister rhs = ToFloatRegister(comp->right()); + Register dest = ToRegister(comp->output()); + + Assembler::DoubleCondition cond = JSOpToDoubleCondition(comp->mir()->jsop()); + masm.ma_compareF64(dest, cond, lhs, rhs); +} + +void CodeGenerator::visitCompareF(LCompareF* comp) { + FloatRegister lhs = ToFloatRegister(comp->left()); + FloatRegister rhs = ToFloatRegister(comp->right()); + Register dest = ToRegister(comp->output()); + + Assembler::DoubleCondition cond = JSOpToDoubleCondition(comp->mir()->jsop()); + masm.ma_compareF32(dest, cond, lhs, rhs); +} + +void CodeGenerator::visitCompareDAndBranch(LCompareDAndBranch* comp) { + FloatRegister lhs = ToFloatRegister(comp->left()); + FloatRegister rhs = ToFloatRegister(comp->right()); + + Assembler::DoubleCondition cond = + JSOpToDoubleCondition(comp->cmpMir()->jsop()); + MBasicBlock* ifTrue = comp->ifTrue(); + MBasicBlock* ifFalse = comp->ifFalse(); + + if (isNextBlock(ifFalse->lir())) { + branchToBlock(DoubleFloat, lhs, rhs, ifTrue, cond); + } else { + branchToBlock(DoubleFloat, lhs, rhs, ifFalse, + Assembler::InvertCondition(cond)); + jumpToBlock(ifTrue); + } +} + +void CodeGenerator::visitCompareFAndBranch(LCompareFAndBranch* comp) { + FloatRegister lhs = ToFloatRegister(comp->left()); + FloatRegister rhs = ToFloatRegister(comp->right()); + + Assembler::DoubleCondition cond = + JSOpToDoubleCondition(comp->cmpMir()->jsop()); + MBasicBlock* ifTrue = comp->ifTrue(); + MBasicBlock* ifFalse = comp->ifFalse(); + + if (isNextBlock(ifFalse->lir())) { + branchToBlock(SingleFloat, lhs, rhs, ifTrue, cond); + } else { + branchToBlock(SingleFloat, lhs, rhs, ifFalse, + Assembler::InvertCondition(cond)); + jumpToBlock(ifTrue); + } +} + +void CodeGenerator::visitBitAndAndBranch(LBitAndAndBranch* lir) { + ScratchRegisterScope scratch(masm); + if (lir->right()->isConstant()) { + masm.ma_and(scratch, ToRegister(lir->left()), Imm32(ToInt32(lir->right()))); + } else { + masm.ma_and(scratch, ToRegister(lir->left()), ToRegister(lir->right())); + } + emitBranch(scratch, Register(scratch), lir->cond(), lir->ifTrue(), + lir->ifFalse()); +} + +void CodeGenerator::visitWasmUint32ToDouble(LWasmUint32ToDouble* lir) { + masm.convertUInt32ToDouble(ToRegister(lir->input()), + ToFloatRegister(lir->output())); +} + +void CodeGenerator::visitWasmUint32ToFloat32(LWasmUint32ToFloat32* lir) { + masm.convertUInt32ToFloat32(ToRegister(lir->input()), + ToFloatRegister(lir->output())); +} + +void CodeGenerator::visitNotI(LNotI* ins) { + masm.cmp32Set(Assembler::Equal, ToRegister(ins->input()), Imm32(0), + ToRegister(ins->output())); +} + +void CodeGenerator::visitNotD(LNotD* ins) { + // Since this operation is not, we want to set a bit if + // the double is falsey, which means 0.0, -0.0 or NaN. + FloatRegister in = ToFloatRegister(ins->input()); + Register dest = ToRegister(ins->output()); + ScratchDoubleScope fpscratch(masm); + + masm.loadConstantDouble(0.0, fpscratch); + masm.ma_compareF64(dest, Assembler::DoubleEqualOrUnordered, in, fpscratch); +} + +void CodeGenerator::visitNotF(LNotF* ins) { + // Since this operation is not, we want to set a bit if + // the float32 is falsey, which means 0.0, -0.0 or NaN. + FloatRegister in = ToFloatRegister(ins->input()); + Register dest = ToRegister(ins->output()); + ScratchFloat32Scope fpscratch(masm); + + masm.loadConstantFloat32(0.0f, fpscratch); + masm.ma_compareF32(dest, Assembler::DoubleEqualOrUnordered, in, fpscratch); +} + +void CodeGenerator::visitMemoryBarrier(LMemoryBarrier* ins) { + masm.memoryBarrier(ins->type()); +} + +void CodeGenerator::visitWasmLoad(LWasmLoad* lir) { emitWasmLoad(lir); } + +void CodeGenerator::visitWasmStore(LWasmStore* lir) { emitWasmStore(lir); } + +void CodeGenerator::visitAsmJSLoadHeap(LAsmJSLoadHeap* ins) { + const MAsmJSLoadHeap* mir = ins->mir(); + const LAllocation* ptr = ins->ptr(); + const LDefinition* out = ins->output(); + const LAllocation* boundsCheckLimit = ins->boundsCheckLimit(); + + bool isSigned; + int size; + bool isFloat = false; + switch (mir->access().type()) { + case Scalar::Int8: + isSigned = true; + size = 8; + break; + case Scalar::Uint8: + isSigned = false; + size = 8; + break; + case Scalar::Int16: + isSigned = true; + size = 16; + break; + case Scalar::Uint16: + isSigned = false; + size = 16; + break; + case Scalar::Int32: + isSigned = true; + size = 32; + break; + case Scalar::Uint32: + isSigned = false; + size = 32; + break; + case Scalar::Float64: + isFloat = true; + size = 64; + break; + case Scalar::Float32: + isFloat = true; + size = 32; + break; + default: + MOZ_CRASH("unexpected array type"); + } + + if (ptr->isConstant()) { + MOZ_ASSERT(!mir->needsBoundsCheck()); + int32_t ptrImm = ptr->toConstant()->toInt32(); + MOZ_ASSERT(ptrImm >= 0); + if (isFloat) { + if (size == 32) { + masm.loadFloat32(Address(HeapReg, ptrImm), ToFloatRegister(out)); + } else { + masm.loadDouble(Address(HeapReg, ptrImm), ToFloatRegister(out)); + } + } else { + masm.ma_load(ToRegister(out), Address(HeapReg, ptrImm), + static_cast<LoadStoreSize>(size), + isSigned ? SignExtend : ZeroExtend); + } + return; + } + + Register ptrReg = ToRegister(ptr); + + if (!mir->needsBoundsCheck()) { + if (isFloat) { + if (size == 32) { + masm.loadFloat32(BaseIndex(HeapReg, ptrReg, TimesOne), + ToFloatRegister(out)); + } else { + masm.loadDouble(BaseIndex(HeapReg, ptrReg, TimesOne), + ToFloatRegister(out)); + } + } else { + masm.ma_load(ToRegister(out), BaseIndex(HeapReg, ptrReg, TimesOne), + static_cast<LoadStoreSize>(size), + isSigned ? SignExtend : ZeroExtend); + } + return; + } + + Label done, outOfRange; + masm.wasmBoundsCheck32(Assembler::AboveOrEqual, ptrReg, + ToRegister(boundsCheckLimit), &outOfRange); + // Offset is ok, let's load value. + if (isFloat) { + if (size == 32) { + masm.loadFloat32(BaseIndex(HeapReg, ptrReg, TimesOne), + ToFloatRegister(out)); + } else { + masm.loadDouble(BaseIndex(HeapReg, ptrReg, TimesOne), + ToFloatRegister(out)); + } + } else { + masm.ma_load(ToRegister(out), BaseIndex(HeapReg, ptrReg, TimesOne), + static_cast<LoadStoreSize>(size), + isSigned ? SignExtend : ZeroExtend); + } + masm.ma_branch(&done, ShortJump); + masm.bind(&outOfRange); + // Offset is out of range. Load default values. + if (isFloat) { + if (size == 32) { + masm.loadConstantFloat32(float(GenericNaN()), ToFloatRegister(out)); + } else { + masm.loadConstantDouble(GenericNaN(), ToFloatRegister(out)); + } + } else { + masm.move32(Imm32(0), ToRegister(out)); + } + masm.bind(&done); +} + +void CodeGenerator::visitAsmJSStoreHeap(LAsmJSStoreHeap* ins) { + const MAsmJSStoreHeap* mir = ins->mir(); + const LAllocation* value = ins->value(); + const LAllocation* ptr = ins->ptr(); + const LAllocation* boundsCheckLimit = ins->boundsCheckLimit(); + + bool isSigned; + int size; + bool isFloat = false; + switch (mir->access().type()) { + case Scalar::Int8: + isSigned = true; + size = 8; + break; + case Scalar::Uint8: + isSigned = false; + size = 8; + break; + case Scalar::Int16: + isSigned = true; + size = 16; + break; + case Scalar::Uint16: + isSigned = false; + size = 16; + break; + case Scalar::Int32: + isSigned = true; + size = 32; + break; + case Scalar::Uint32: + isSigned = false; + size = 32; + break; + case Scalar::Float64: + isFloat = true; + size = 64; + break; + case Scalar::Float32: + isFloat = true; + size = 32; + break; + default: + MOZ_CRASH("unexpected array type"); + } + + if (ptr->isConstant()) { + MOZ_ASSERT(!mir->needsBoundsCheck()); + int32_t ptrImm = ptr->toConstant()->toInt32(); + MOZ_ASSERT(ptrImm >= 0); + + if (isFloat) { + FloatRegister freg = ToFloatRegister(value); + Address addr(HeapReg, ptrImm); + if (size == 32) { + masm.storeFloat32(freg, addr); + } else { + masm.storeDouble(freg, addr); + } + } else { + masm.ma_store(ToRegister(value), Address(HeapReg, ptrImm), + static_cast<LoadStoreSize>(size), + isSigned ? SignExtend : ZeroExtend); + } + return; + } + + Register ptrReg = ToRegister(ptr); + Address dstAddr(ptrReg, 0); + + if (!mir->needsBoundsCheck()) { + if (isFloat) { + FloatRegister freg = ToFloatRegister(value); + BaseIndex bi(HeapReg, ptrReg, TimesOne); + if (size == 32) { + masm.storeFloat32(freg, bi); + } else { + masm.storeDouble(freg, bi); + } + } else { + masm.ma_store(ToRegister(value), BaseIndex(HeapReg, ptrReg, TimesOne), + static_cast<LoadStoreSize>(size), + isSigned ? SignExtend : ZeroExtend); + } + return; + } + + Label outOfRange; + masm.wasmBoundsCheck32(Assembler::AboveOrEqual, ptrReg, + ToRegister(boundsCheckLimit), &outOfRange); + + // Offset is ok, let's store value. + if (isFloat) { + if (size == 32) { + masm.storeFloat32(ToFloatRegister(value), + BaseIndex(HeapReg, ptrReg, TimesOne)); + } else + masm.storeDouble(ToFloatRegister(value), + BaseIndex(HeapReg, ptrReg, TimesOne)); + } else { + masm.ma_store(ToRegister(value), BaseIndex(HeapReg, ptrReg, TimesOne), + static_cast<LoadStoreSize>(size), + isSigned ? SignExtend : ZeroExtend); + } + + masm.bind(&outOfRange); +} + +void CodeGenerator::visitWasmCompareExchangeHeap( + LWasmCompareExchangeHeap* ins) { + MWasmCompareExchangeHeap* mir = ins->mir(); + Register memoryBase = ToRegister(ins->memoryBase()); + Register ptrReg = ToRegister(ins->ptr()); + BaseIndex srcAddr(memoryBase, ptrReg, TimesOne, mir->access().offset()); + MOZ_ASSERT(ins->addrTemp()->isBogusTemp()); + + Register oldval = ToRegister(ins->oldValue()); + Register newval = ToRegister(ins->newValue()); + Register valueTemp = ToTempRegisterOrInvalid(ins->valueTemp()); + Register offsetTemp = ToTempRegisterOrInvalid(ins->offsetTemp()); + Register maskTemp = ToTempRegisterOrInvalid(ins->maskTemp()); + + masm.wasmCompareExchange(mir->access(), srcAddr, oldval, newval, valueTemp, + offsetTemp, maskTemp, ToRegister(ins->output())); +} + +void CodeGenerator::visitWasmAtomicExchangeHeap(LWasmAtomicExchangeHeap* ins) { + MWasmAtomicExchangeHeap* mir = ins->mir(); + Register memoryBase = ToRegister(ins->memoryBase()); + Register ptrReg = ToRegister(ins->ptr()); + Register value = ToRegister(ins->value()); + BaseIndex srcAddr(memoryBase, ptrReg, TimesOne, mir->access().offset()); + MOZ_ASSERT(ins->addrTemp()->isBogusTemp()); + + Register valueTemp = ToTempRegisterOrInvalid(ins->valueTemp()); + Register offsetTemp = ToTempRegisterOrInvalid(ins->offsetTemp()); + Register maskTemp = ToTempRegisterOrInvalid(ins->maskTemp()); + + masm.wasmAtomicExchange(mir->access(), srcAddr, value, valueTemp, offsetTemp, + maskTemp, ToRegister(ins->output())); +} + +void CodeGenerator::visitWasmAtomicBinopHeap(LWasmAtomicBinopHeap* ins) { + MOZ_ASSERT(ins->mir()->hasUses()); + MOZ_ASSERT(ins->addrTemp()->isBogusTemp()); + + MWasmAtomicBinopHeap* mir = ins->mir(); + Register memoryBase = ToRegister(ins->memoryBase()); + Register ptrReg = ToRegister(ins->ptr()); + Register valueTemp = ToTempRegisterOrInvalid(ins->valueTemp()); + Register offsetTemp = ToTempRegisterOrInvalid(ins->offsetTemp()); + Register maskTemp = ToTempRegisterOrInvalid(ins->maskTemp()); + + BaseIndex srcAddr(memoryBase, ptrReg, TimesOne, mir->access().offset()); + + masm.wasmAtomicFetchOp(mir->access(), mir->operation(), + ToRegister(ins->value()), srcAddr, valueTemp, + offsetTemp, maskTemp, ToRegister(ins->output())); +} + +void CodeGenerator::visitWasmAtomicBinopHeapForEffect( + LWasmAtomicBinopHeapForEffect* ins) { + MOZ_ASSERT(!ins->mir()->hasUses()); + MOZ_ASSERT(ins->addrTemp()->isBogusTemp()); + + MWasmAtomicBinopHeap* mir = ins->mir(); + Register memoryBase = ToRegister(ins->memoryBase()); + Register ptrReg = ToRegister(ins->ptr()); + Register valueTemp = ToTempRegisterOrInvalid(ins->valueTemp()); + Register offsetTemp = ToTempRegisterOrInvalid(ins->offsetTemp()); + Register maskTemp = ToTempRegisterOrInvalid(ins->maskTemp()); + + BaseIndex srcAddr(memoryBase, ptrReg, TimesOne, mir->access().offset()); + masm.wasmAtomicEffectOp(mir->access(), mir->operation(), + ToRegister(ins->value()), srcAddr, valueTemp, + offsetTemp, maskTemp); +} + +void CodeGenerator::visitWasmStackArg(LWasmStackArg* ins) { + const MWasmStackArg* mir = ins->mir(); + if (ins->arg()->isConstant()) { + masm.storePtr(ImmWord(ToInt32(ins->arg())), + Address(StackPointer, mir->spOffset())); + } else { + if (ins->arg()->isGeneralReg()) { + masm.storePtr(ToRegister(ins->arg()), + Address(StackPointer, mir->spOffset())); + } else if (mir->input()->type() == MIRType::Double) { + masm.storeDouble(ToFloatRegister(ins->arg()), + Address(StackPointer, mir->spOffset())); + } else { + masm.storeFloat32(ToFloatRegister(ins->arg()), + Address(StackPointer, mir->spOffset())); + } + } +} + +void CodeGenerator::visitWasmStackArgI64(LWasmStackArgI64* ins) { + const MWasmStackArg* mir = ins->mir(); + Address dst(StackPointer, mir->spOffset()); + if (IsConstant(ins->arg())) { + masm.store64(Imm64(ToInt64(ins->arg())), dst); + } else { + masm.store64(ToRegister64(ins->arg()), dst); + } +} + +void CodeGenerator::visitWasmSelect(LWasmSelect* ins) { + MIRType mirType = ins->mir()->type(); + + Register cond = ToRegister(ins->condExpr()); + const LAllocation* falseExpr = ins->falseExpr(); + + if (mirType == MIRType::Int32 || mirType == MIRType::WasmAnyRef) { + Register out = ToRegister(ins->output()); + MOZ_ASSERT(ToRegister(ins->trueExpr()) == out, + "true expr input is reused for output"); + if (falseExpr->isRegister()) { + masm.moveIfZero(out, ToRegister(falseExpr), cond); + } else { + masm.cmp32Load32(Assembler::Zero, cond, cond, ToAddress(falseExpr), out); + } + return; + } + + FloatRegister out = ToFloatRegister(ins->output()); + MOZ_ASSERT(ToFloatRegister(ins->trueExpr()) == out, + "true expr input is reused for output"); + + if (falseExpr->isFloatReg()) { + if (mirType == MIRType::Float32) { + masm.ma_fmovz(SingleFloat, out, ToFloatRegister(falseExpr), cond); + } else if (mirType == MIRType::Double) { + masm.ma_fmovz(DoubleFloat, out, ToFloatRegister(falseExpr), cond); + } else { + MOZ_CRASH("unhandled type in visitWasmSelect!"); + } + } else { + Label done; + masm.ma_b(cond, cond, &done, Assembler::NonZero, ShortJump); + + if (mirType == MIRType::Float32) { + masm.loadFloat32(ToAddress(falseExpr), out); + } else if (mirType == MIRType::Double) { + masm.loadDouble(ToAddress(falseExpr), out); + } else { + MOZ_CRASH("unhandled type in visitWasmSelect!"); + } + + masm.bind(&done); + } +} + +// We expect to handle only the case where compare is {U,}Int32 and select is +// {U,}Int32, and the "true" input is reused for the output. +void CodeGenerator::visitWasmCompareAndSelect(LWasmCompareAndSelect* ins) { + bool cmpIs32bit = ins->compareType() == MCompare::Compare_Int32 || + ins->compareType() == MCompare::Compare_UInt32; + bool selIs32bit = ins->mir()->type() == MIRType::Int32; + + MOZ_RELEASE_ASSERT( + cmpIs32bit && selIs32bit, + "CodeGenerator::visitWasmCompareAndSelect: unexpected types"); + + Register trueExprAndDest = ToRegister(ins->output()); + MOZ_ASSERT(ToRegister(ins->ifTrueExpr()) == trueExprAndDest, + "true expr input is reused for output"); + + Assembler::Condition cond = Assembler::InvertCondition( + JSOpToCondition(ins->compareType(), ins->jsop())); + const LAllocation* rhs = ins->rightExpr(); + const LAllocation* falseExpr = ins->ifFalseExpr(); + Register lhs = ToRegister(ins->leftExpr()); + + masm.cmp32Move32(cond, lhs, ToRegister(rhs), ToRegister(falseExpr), + trueExprAndDest); +} + +void CodeGenerator::visitWasmReinterpret(LWasmReinterpret* lir) { + MOZ_ASSERT(gen->compilingWasm()); + MWasmReinterpret* ins = lir->mir(); + + MIRType to = ins->type(); + mozilla::DebugOnly<MIRType> from = ins->input()->type(); + + switch (to) { + case MIRType::Int32: + MOZ_ASSERT(from == MIRType::Float32); + masm.fmv_x_w(ToRegister(lir->output()), ToFloatRegister(lir->input())); + break; + case MIRType::Float32: + MOZ_ASSERT(from == MIRType::Int32); + masm.fmv_w_x(ToFloatRegister(lir->output()), ToRegister(lir->input())); + break; + case MIRType::Double: + case MIRType::Int64: + MOZ_CRASH("not handled by this LIR opcode"); + default: + MOZ_CRASH("unexpected WasmReinterpret"); + } +} + +void CodeGenerator::visitUDivOrMod(LUDivOrMod* ins) { + Register lhs = ToRegister(ins->lhs()); + Register rhs = ToRegister(ins->rhs()); + Register output = ToRegister(ins->output()); + Label done; + + // Prevent divide by zero. + if (ins->canBeDivideByZero()) { + if (ins->mir()->isTruncated()) { + if (ins->trapOnError()) { + Label nonZero; + masm.ma_b(rhs, rhs, &nonZero, Assembler::NonZero); + masm.wasmTrap(wasm::Trap::IntegerDivideByZero, ins->bytecodeOffset()); + masm.bind(&nonZero); + } else { + // Infinity|0 == 0 + Label notzero; + masm.ma_b(rhs, rhs, ¬zero, Assembler::NonZero, ShortJump); + masm.move32(Imm32(0), output); + masm.ma_branch(&done, ShortJump); + masm.bind(¬zero); + } + } else { + bailoutCmp32(Assembler::Equal, rhs, Imm32(0), ins->snapshot()); + } + } + + masm.ma_modu32(output, lhs, rhs); + + // If the remainder is > 0, bailout since this must be a double. + if (ins->mir()->isDiv()) { + if (!ins->mir()->toDiv()->canTruncateRemainder()) { + bailoutCmp32(Assembler::NonZero, output, output, ins->snapshot()); + } + // Get quotient + masm.ma_divu32(output, lhs, rhs); + } + + if (!ins->mir()->isTruncated()) { + bailoutCmp32(Assembler::LessThan, output, Imm32(0), ins->snapshot()); + } + + masm.bind(&done); +} + +void CodeGenerator::visitEffectiveAddress(LEffectiveAddress* ins) { + const MEffectiveAddress* mir = ins->mir(); + Register base = ToRegister(ins->base()); + Register index = ToRegister(ins->index()); + Register output = ToRegister(ins->output()); + + BaseIndex address(base, index, mir->scale(), mir->displacement()); + masm.computeEffectiveAddress(address, output); +} + +void CodeGenerator::visitNegI(LNegI* ins) { + Register input = ToRegister(ins->input()); + Register output = ToRegister(ins->output()); + + masm.ma_sub32(output, zero, input); +} + +void CodeGenerator::visitNegI64(LNegI64* ins) { + Register64 input = ToRegister64(ins->getInt64Operand(0)); + MOZ_ASSERT(input == ToOutRegister64(ins)); + masm.neg64(input); +} + +void CodeGenerator::visitNegD(LNegD* ins) { + FloatRegister input = ToFloatRegister(ins->input()); + FloatRegister output = ToFloatRegister(ins->output()); + + masm.fneg_d(output, input); +} + +void CodeGenerator::visitNegF(LNegF* ins) { + FloatRegister input = ToFloatRegister(ins->input()); + FloatRegister output = ToFloatRegister(ins->output()); + + masm.fneg_s(output, input); +} + +void CodeGenerator::visitWasmAddOffset(LWasmAddOffset* lir) { + MWasmAddOffset* mir = lir->mir(); + Register base = ToRegister(lir->base()); + Register out = ToRegister(lir->output()); + + Label ok; + masm.ma_add32TestCarry(Assembler::CarryClear, out, base, Imm32(mir->offset()), + &ok); + masm.wasmTrap(wasm::Trap::OutOfBounds, mir->bytecodeOffset()); + masm.bind(&ok); +} + +void CodeGenerator::visitWasmAddOffset64(LWasmAddOffset64* lir) { + MWasmAddOffset* mir = lir->mir(); + Register64 base = ToRegister64(lir->base()); + Register64 out = ToOutRegister64(lir); + + Label ok; + masm.ma_addPtrTestCarry(Assembler::CarryClear, out.reg, base.reg, + ImmWord(mir->offset()), &ok); + masm.wasmTrap(wasm::Trap::OutOfBounds, mir->bytecodeOffset()); + masm.bind(&ok); +} + +void CodeGenerator::visitAtomicTypedArrayElementBinop( + LAtomicTypedArrayElementBinop* lir) { + MOZ_ASSERT(!lir->mir()->isForEffect()); + + AnyRegister output = ToAnyRegister(lir->output()); + Register elements = ToRegister(lir->elements()); + Register outTemp = ToTempRegisterOrInvalid(lir->temp2()); + Register valueTemp = ToTempRegisterOrInvalid(lir->valueTemp()); + Register offsetTemp = ToTempRegisterOrInvalid(lir->offsetTemp()); + Register maskTemp = ToTempRegisterOrInvalid(lir->maskTemp()); + Register value = ToRegister(lir->value()); + Scalar::Type arrayType = lir->mir()->arrayType(); + + if (lir->index()->isConstant()) { + Address mem = ToAddress(elements, lir->index(), arrayType); + masm.atomicFetchOpJS(arrayType, Synchronization::Full(), + lir->mir()->operation(), value, mem, valueTemp, + offsetTemp, maskTemp, outTemp, output); + } else { + BaseIndex mem(elements, ToRegister(lir->index()), + ScaleFromScalarType(arrayType)); + masm.atomicFetchOpJS(arrayType, Synchronization::Full(), + lir->mir()->operation(), value, mem, valueTemp, + offsetTemp, maskTemp, outTemp, output); + } +} + +void CodeGenerator::visitAtomicTypedArrayElementBinopForEffect( + LAtomicTypedArrayElementBinopForEffect* lir) { + MOZ_ASSERT(lir->mir()->isForEffect()); + + Register elements = ToRegister(lir->elements()); + Register valueTemp = ToTempRegisterOrInvalid(lir->valueTemp()); + Register offsetTemp = ToTempRegisterOrInvalid(lir->offsetTemp()); + Register maskTemp = ToTempRegisterOrInvalid(lir->maskTemp()); + Register value = ToRegister(lir->value()); + Scalar::Type arrayType = lir->mir()->arrayType(); + + if (lir->index()->isConstant()) { + Address mem = ToAddress(elements, lir->index(), arrayType); + masm.atomicEffectOpJS(arrayType, Synchronization::Full(), + lir->mir()->operation(), value, mem, valueTemp, + offsetTemp, maskTemp); + } else { + BaseIndex mem(elements, ToRegister(lir->index()), + ScaleFromScalarType(arrayType)); + masm.atomicEffectOpJS(arrayType, Synchronization::Full(), + lir->mir()->operation(), value, mem, valueTemp, + offsetTemp, maskTemp); + } +} + +void CodeGenerator::visitCompareExchangeTypedArrayElement( + LCompareExchangeTypedArrayElement* lir) { + Register elements = ToRegister(lir->elements()); + AnyRegister output = ToAnyRegister(lir->output()); + Register outTemp = ToTempRegisterOrInvalid(lir->temp()); + + Register oldval = ToRegister(lir->oldval()); + Register newval = ToRegister(lir->newval()); + Register valueTemp = ToTempRegisterOrInvalid(lir->valueTemp()); + Register offsetTemp = ToTempRegisterOrInvalid(lir->offsetTemp()); + Register maskTemp = ToTempRegisterOrInvalid(lir->maskTemp()); + Scalar::Type arrayType = lir->mir()->arrayType(); + + if (lir->index()->isConstant()) { + Address dest = ToAddress(elements, lir->index(), arrayType); + masm.compareExchangeJS(arrayType, Synchronization::Full(), dest, oldval, + newval, valueTemp, offsetTemp, maskTemp, outTemp, + output); + } else { + BaseIndex dest(elements, ToRegister(lir->index()), + ScaleFromScalarType(arrayType)); + masm.compareExchangeJS(arrayType, Synchronization::Full(), dest, oldval, + newval, valueTemp, offsetTemp, maskTemp, outTemp, + output); + } +} + +void CodeGenerator::visitAtomicExchangeTypedArrayElement( + LAtomicExchangeTypedArrayElement* lir) { + Register elements = ToRegister(lir->elements()); + AnyRegister output = ToAnyRegister(lir->output()); + Register outTemp = ToTempRegisterOrInvalid(lir->temp()); + + Register value = ToRegister(lir->value()); + Register valueTemp = ToTempRegisterOrInvalid(lir->valueTemp()); + Register offsetTemp = ToTempRegisterOrInvalid(lir->offsetTemp()); + Register maskTemp = ToTempRegisterOrInvalid(lir->maskTemp()); + Scalar::Type arrayType = lir->mir()->arrayType(); + + if (lir->index()->isConstant()) { + Address dest = ToAddress(elements, lir->index(), arrayType); + masm.atomicExchangeJS(arrayType, Synchronization::Full(), dest, value, + valueTemp, offsetTemp, maskTemp, outTemp, output); + } else { + BaseIndex dest(elements, ToRegister(lir->index()), + ScaleFromScalarType(arrayType)); + masm.atomicExchangeJS(arrayType, Synchronization::Full(), dest, value, + valueTemp, offsetTemp, maskTemp, outTemp, output); + } +} + +void CodeGenerator::visitCompareExchangeTypedArrayElement64( + LCompareExchangeTypedArrayElement64* lir) { + Register elements = ToRegister(lir->elements()); + Register oldval = ToRegister(lir->oldval()); + Register newval = ToRegister(lir->newval()); + Register64 temp1 = ToRegister64(lir->temp1()); + Register64 temp2 = ToRegister64(lir->temp2()); + Register out = ToRegister(lir->output()); + Register64 tempOut(out); + Scalar::Type arrayType = lir->mir()->arrayType(); + + masm.loadBigInt64(oldval, temp1); + masm.loadBigInt64(newval, tempOut); + + if (lir->index()->isConstant()) { + Address dest = ToAddress(elements, lir->index(), arrayType); + masm.compareExchange64(Synchronization::Full(), dest, temp1, tempOut, + temp2); + } else { + BaseIndex dest(elements, ToRegister(lir->index()), + ScaleFromScalarType(arrayType)); + masm.compareExchange64(Synchronization::Full(), dest, temp1, tempOut, + temp2); + } + + emitCreateBigInt(lir, arrayType, temp2, out, temp1.scratchReg()); +} + +void CodeGenerator::visitAtomicExchangeTypedArrayElement64( + LAtomicExchangeTypedArrayElement64* lir) { + Register elements = ToRegister(lir->elements()); + Register value = ToRegister(lir->value()); + Register64 temp1 = ToRegister64(lir->temp1()); + Register64 temp2 = Register64(ToRegister(lir->temp2())); + Register out = ToRegister(lir->output()); + Scalar::Type arrayType = lir->mir()->arrayType(); + + masm.loadBigInt64(value, temp1); + + if (lir->index()->isConstant()) { + Address dest = ToAddress(elements, lir->index(), arrayType); + masm.atomicExchange64(Synchronization::Full(), dest, temp1, temp2); + } else { + BaseIndex dest(elements, ToRegister(lir->index()), + ScaleFromScalarType(arrayType)); + masm.atomicExchange64(Synchronization::Full(), dest, temp1, temp2); + } + + emitCreateBigInt(lir, arrayType, temp2, out, temp1.scratchReg()); +} + +void CodeGenerator::visitAtomicTypedArrayElementBinop64( + LAtomicTypedArrayElementBinop64* lir) { + MOZ_ASSERT(lir->mir()->hasUses()); + + Register elements = ToRegister(lir->elements()); + Register value = ToRegister(lir->value()); + Register64 temp1 = ToRegister64(lir->temp1()); + Register64 temp2 = ToRegister64(lir->temp2()); + Register out = ToRegister(lir->output()); + Register64 tempOut = Register64(out); + + Scalar::Type arrayType = lir->mir()->arrayType(); + AtomicOp atomicOp = lir->mir()->operation(); + + masm.loadBigInt64(value, temp1); + + if (lir->index()->isConstant()) { + Address dest = ToAddress(elements, lir->index(), arrayType); + masm.atomicFetchOp64(Synchronization::Full(), atomicOp, temp1, dest, + tempOut, temp2); + } else { + BaseIndex dest(elements, ToRegister(lir->index()), + ScaleFromScalarType(arrayType)); + masm.atomicFetchOp64(Synchronization::Full(), atomicOp, temp1, dest, + tempOut, temp2); + } + + emitCreateBigInt(lir, arrayType, temp2, out, temp1.scratchReg()); +} + +void CodeGenerator::visitAtomicTypedArrayElementBinopForEffect64( + LAtomicTypedArrayElementBinopForEffect64* lir) { + MOZ_ASSERT(!lir->mir()->hasUses()); + + Register elements = ToRegister(lir->elements()); + Register value = ToRegister(lir->value()); + Register64 temp1 = ToRegister64(lir->temp1()); + Register64 temp2 = ToRegister64(lir->temp2()); + + Scalar::Type arrayType = lir->mir()->arrayType(); + AtomicOp atomicOp = lir->mir()->operation(); + + masm.loadBigInt64(value, temp1); + + if (lir->index()->isConstant()) { + Address dest = ToAddress(elements, lir->index(), arrayType); + masm.atomicEffectOp64(Synchronization::Full(), atomicOp, temp1, dest, + temp2); + } else { + BaseIndex dest(elements, ToRegister(lir->index()), + ScaleFromScalarType(arrayType)); + masm.atomicEffectOp64(Synchronization::Full(), atomicOp, temp1, dest, + temp2); + } +} + +void CodeGenerator::visitAtomicLoad64(LAtomicLoad64* lir) { + Register elements = ToRegister(lir->elements()); + Register temp = ToRegister(lir->temp()); + Register64 temp64 = ToRegister64(lir->temp64()); + Register out = ToRegister(lir->output()); + const MLoadUnboxedScalar* mir = lir->mir(); + + Scalar::Type storageType = mir->storageType(); + + auto sync = Synchronization::Load(); + masm.memoryBarrierBefore(sync); + if (lir->index()->isConstant()) { + Address source = + ToAddress(elements, lir->index(), storageType, mir->offsetAdjustment()); + masm.load64(source, temp64); + } else { + BaseIndex source(elements, ToRegister(lir->index()), + ScaleFromScalarType(storageType), mir->offsetAdjustment()); + masm.load64(source, temp64); + } + masm.memoryBarrierAfter(sync); + emitCreateBigInt(lir, storageType, temp64, out, temp); +} + +void CodeGenerator::visitAtomicStore64(LAtomicStore64* lir) { + Register elements = ToRegister(lir->elements()); + Register value = ToRegister(lir->value()); + Register64 temp1 = ToRegister64(lir->temp1()); + + Scalar::Type writeType = lir->mir()->writeType(); + + masm.loadBigInt64(value, temp1); + auto sync = Synchronization::Store(); + masm.memoryBarrierBefore(sync); + if (lir->index()->isConstant()) { + Address dest = ToAddress(elements, lir->index(), writeType); + masm.store64(temp1, dest); + } else { + BaseIndex dest(elements, ToRegister(lir->index()), + ScaleFromScalarType(writeType)); + masm.store64(temp1, dest); + } + masm.memoryBarrierAfter(sync); +} + +void CodeGenerator::visitWasmCompareExchangeI64(LWasmCompareExchangeI64* lir) { + Register memoryBase = ToRegister(lir->memoryBase()); + Register ptr = ToRegister(lir->ptr()); + Register64 oldValue = ToRegister64(lir->oldValue()); + Register64 newValue = ToRegister64(lir->newValue()); + Register64 output = ToOutRegister64(lir); + uint32_t offset = lir->mir()->access().offset(); + + BaseIndex addr(memoryBase, ptr, TimesOne, offset); + masm.wasmCompareExchange64(lir->mir()->access(), addr, oldValue, newValue, + output); +} + +void CodeGenerator::visitWasmAtomicExchangeI64(LWasmAtomicExchangeI64* lir) { + Register memoryBase = ToRegister(lir->memoryBase()); + Register ptr = ToRegister(lir->ptr()); + Register64 value = ToRegister64(lir->value()); + Register64 output = ToOutRegister64(lir); + uint32_t offset = lir->mir()->access().offset(); + + BaseIndex addr(memoryBase, ptr, TimesOne, offset); + masm.wasmAtomicExchange64(lir->mir()->access(), addr, value, output); +} + +void CodeGenerator::visitWasmAtomicBinopI64(LWasmAtomicBinopI64* lir) { + Register memoryBase = ToRegister(lir->memoryBase()); + Register ptr = ToRegister(lir->ptr()); + Register64 value = ToRegister64(lir->value()); + Register64 output = ToOutRegister64(lir); + Register64 temp(ToRegister(lir->getTemp(0))); + uint32_t offset = lir->mir()->access().offset(); + + BaseIndex addr(memoryBase, ptr, TimesOne, offset); + + masm.wasmAtomicFetchOp64(lir->mir()->access(), lir->mir()->operation(), value, + addr, temp, output); +} + +void CodeGenerator::visitNearbyInt(LNearbyInt*) { MOZ_CRASH("NYI"); } + +void CodeGenerator::visitNearbyIntF(LNearbyIntF*) { MOZ_CRASH("NYI"); } + +void CodeGenerator::visitSimd128(LSimd128* ins) { MOZ_CRASH("No SIMD"); } + +void CodeGenerator::visitWasmTernarySimd128(LWasmTernarySimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmBinarySimd128(LWasmBinarySimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmBinarySimd128WithConstant( + LWasmBinarySimd128WithConstant* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmVariableShiftSimd128( + LWasmVariableShiftSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmConstantShiftSimd128( + LWasmConstantShiftSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmSignReplicationSimd128( + LWasmSignReplicationSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmShuffleSimd128(LWasmShuffleSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmPermuteSimd128(LWasmPermuteSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmReplaceLaneSimd128(LWasmReplaceLaneSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmReplaceInt64LaneSimd128( + LWasmReplaceInt64LaneSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmScalarToSimd128(LWasmScalarToSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmInt64ToSimd128(LWasmInt64ToSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmUnarySimd128(LWasmUnarySimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmReduceSimd128(LWasmReduceSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmReduceAndBranchSimd128( + LWasmReduceAndBranchSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmReduceSimd128ToInt64( + LWasmReduceSimd128ToInt64* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmLoadLaneSimd128(LWasmLoadLaneSimd128* ins) { + MOZ_CRASH("No SIMD"); +} + +void CodeGenerator::visitWasmStoreLaneSimd128(LWasmStoreLaneSimd128* ins) { + MOZ_CRASH("No SIMD"); +} diff --git a/js/src/jit/riscv64/CodeGenerator-riscv64.h b/js/src/jit/riscv64/CodeGenerator-riscv64.h new file mode 100644 index 0000000000..793c834085 --- /dev/null +++ b/js/src/jit/riscv64/CodeGenerator-riscv64.h @@ -0,0 +1,210 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_CodeGenerator_riscv64_h +#define jit_riscv64_CodeGenerator_riscv64_h + +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/MacroAssembler-riscv64.h" +#include "jit/shared/CodeGenerator-shared.h" + +namespace js { +namespace jit { + +class CodeGeneratorRiscv64; +class OutOfLineBailout; +class OutOfLineTableSwitch; + +using OutOfLineWasmTruncateCheck = + OutOfLineWasmTruncateCheckBase<CodeGeneratorRiscv64>; + +class CodeGeneratorRiscv64 : public CodeGeneratorShared { + friend class MoveResolverLA; + + protected: + CodeGeneratorRiscv64(MIRGenerator* gen, LIRGraph* graph, + MacroAssembler* masm); + + NonAssertingLabel deoptLabel_; + + Operand ToOperand(const LAllocation& a); + Operand ToOperand(const LAllocation* a); + Operand ToOperand(const LDefinition* def); + +#ifdef JS_PUNBOX64 + Operand ToOperandOrRegister64(const LInt64Allocation input); +#else + Register64 ToOperandOrRegister64(const LInt64Allocation input); +#endif + + MoveOperand toMoveOperand(LAllocation a) const; + + template <typename T1, typename T2> + void bailoutCmp32(Assembler::Condition c, T1 lhs, T2 rhs, + LSnapshot* snapshot) { + Label bail; + masm.branch32(c, lhs, rhs, &bail); + bailoutFrom(&bail, snapshot); + } + template <typename T1, typename T2> + void bailoutTest32(Assembler::Condition c, T1 lhs, T2 rhs, + LSnapshot* snapshot) { + Label bail; + masm.branchTest32(c, lhs, rhs, &bail); + bailoutFrom(&bail, snapshot); + } + template <typename T1, typename T2> + void bailoutCmpPtr(Assembler::Condition c, T1 lhs, T2 rhs, + LSnapshot* snapshot) { + Label bail; + masm.branchPtr(c, lhs, rhs, &bail); + bailoutFrom(&bail, snapshot); + } + void bailoutTestPtr(Assembler::Condition c, Register lhs, Register rhs, + LSnapshot* snapshot) { + // TODO(riscv64) Didn't use branchTestPtr due to '-Wundefined-inline'. + MOZ_ASSERT(c == Assembler::Zero || c == Assembler::NonZero || + c == Assembler::Signed || c == Assembler::NotSigned); + Label bail; + if (lhs == rhs) { + masm.ma_b(lhs, rhs, &bail, c); + } else { + ScratchRegisterScope scratch(masm); + masm.and_(scratch, lhs, rhs); + masm.ma_b(scratch, scratch, &bail, c); + } + bailoutFrom(&bail, snapshot); + } + void bailoutIfFalseBool(Register reg, LSnapshot* snapshot) { + Label bail; + ScratchRegisterScope scratch(masm); + masm.ma_and(scratch, reg, Imm32(0xFF)); + masm.ma_b(scratch, scratch, &bail, Assembler::Zero); + bailoutFrom(&bail, snapshot); + } + + void bailoutFrom(Label* label, LSnapshot* snapshot); + void bailout(LSnapshot* snapshot); + + bool generateOutOfLineCode(); + + template <typename T> + void branchToBlock(Register lhs, T rhs, MBasicBlock* mir, + Assembler::Condition cond) { + masm.ma_b(lhs, rhs, skipTrivialBlocks(mir)->lir()->label(), cond); + } + void branchToBlock(FloatFormat fmt, FloatRegister lhs, FloatRegister rhs, + MBasicBlock* mir, Assembler::DoubleCondition cond); + + // Emits a branch that directs control flow to the true block if |cond| is + // true, and the false block if |cond| is false. + template <typename T> + void emitBranch(Register lhs, T rhs, Assembler::Condition cond, + MBasicBlock* mirTrue, MBasicBlock* mirFalse) { + if (isNextBlock(mirFalse->lir())) { + branchToBlock(lhs, rhs, mirTrue, cond); + } else { + branchToBlock(lhs, rhs, mirFalse, Assembler::InvertCondition(cond)); + jumpToBlock(mirTrue); + } + } + void testZeroEmitBranch(Assembler::Condition cond, Register reg, + MBasicBlock* ifTrue, MBasicBlock* ifFalse) { + emitBranch(reg, Imm32(0), cond, ifTrue, ifFalse); + } + + void emitTableSwitchDispatch(MTableSwitch* mir, Register index, + Register base); + + template <typename T> + void emitWasmLoad(T* ins); + template <typename T> + void emitWasmStore(T* ins); + + void generateInvalidateEpilogue(); + + // Generating a result. + template <typename S, typename T> + void atomicBinopToTypedIntArray(AtomicOp op, Scalar::Type arrayType, + const S& value, const T& mem, + Register flagTemp, Register outTemp, + Register valueTemp, Register offsetTemp, + Register maskTemp, AnyRegister output); + + // Generating no result. + template <typename S, typename T> + void atomicBinopToTypedIntArray(AtomicOp op, Scalar::Type arrayType, + const S& value, const T& mem, + Register flagTemp, Register valueTemp, + Register offsetTemp, Register maskTemp); + + public: + // Out of line visitors. + void visitOutOfLineBailout(OutOfLineBailout* ool); + void visitOutOfLineTableSwitch(OutOfLineTableSwitch* ool); + void visitOutOfLineWasmTruncateCheck(OutOfLineWasmTruncateCheck* ool); + + protected: + void testNullEmitBranch(Assembler::Condition cond, const ValueOperand& value, + MBasicBlock* ifTrue, MBasicBlock* ifFalse) { + UseScratchRegisterScope temps(&masm); + Register scratch = temps.Acquire(); + masm.splitTag(value.valueReg(), scratch); + emitBranch(scratch, ImmTag(JSVAL_TAG_NULL), cond, ifTrue, ifFalse); + } + void testUndefinedEmitBranch(Assembler::Condition cond, + const ValueOperand& value, MBasicBlock* ifTrue, + MBasicBlock* ifFalse) { + UseScratchRegisterScope temps(&masm); + Register scratch = temps.Acquire(); + masm.splitTag(value.valueReg(), scratch); + emitBranch(scratch, ImmTag(JSVAL_TAG_UNDEFINED), cond, ifTrue, ifFalse); + } + void testObjectEmitBranch(Assembler::Condition cond, + const ValueOperand& value, MBasicBlock* ifTrue, + MBasicBlock* ifFalse) { + UseScratchRegisterScope temps(&masm); + Register scratch = temps.Acquire(); + masm.splitTag(value.valueReg(), scratch); + emitBranch(scratch, ImmTag(JSVAL_TAG_OBJECT), cond, ifTrue, ifFalse); + } + + void emitBigIntDiv(LBigIntDiv* ins, Register dividend, Register divisor, + Register output, Label* fail); + void emitBigIntMod(LBigIntMod* ins, Register dividend, Register divisor, + Register output, Label* fail); + + template <typename T> + void emitWasmLoadI64(T* ins); + template <typename T> + void emitWasmStoreI64(T* ins); + + ValueOperand ToValue(LInstruction* ins, size_t pos); + ValueOperand ToTempValue(LInstruction* ins, size_t pos); + + // Functions for LTestVAndBranch. + void splitTagForTest(const ValueOperand& value, ScratchTagScope& tag); +}; + +typedef CodeGeneratorRiscv64 CodeGeneratorSpecific; + +// An out-of-line bailout thunk. +class OutOfLineBailout : public OutOfLineCodeBase<CodeGeneratorRiscv64> { + protected: + LSnapshot* snapshot_; + + public: + OutOfLineBailout(LSnapshot* snapshot) : snapshot_(snapshot) {} + + void accept(CodeGeneratorRiscv64* codegen) override; + + LSnapshot* snapshot() const { return snapshot_; } +}; + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_CodeGenerator_riscv64_h */ diff --git a/js/src/jit/riscv64/LIR-riscv64.h b/js/src/jit/riscv64/LIR-riscv64.h new file mode 100644 index 0000000000..eaf02ab7dd --- /dev/null +++ b/js/src/jit/riscv64/LIR-riscv64.h @@ -0,0 +1,408 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_LIR_riscv64_h +#define jit_riscv64_LIR_riscv64_h + +namespace js { +namespace jit { + +class LUnbox : public LInstructionHelper<1, 1, 0> { + protected: + LUnbox(LNode::Opcode opcode, const LAllocation& input) + : LInstructionHelper(opcode) { + setOperand(0, input); + } + + public: + LIR_HEADER(Unbox); + + explicit LUnbox(const LAllocation& input) : LInstructionHelper(classOpcode) { + setOperand(0, input); + } + + static const size_t Input = 0; + + MUnbox* mir() const { return mir_->toUnbox(); } + const char* extraName() const { return StringFromMIRType(mir()->type()); } +}; + +class LUnboxFloatingPoint : public LUnbox { + MIRType type_; + + public: + LIR_HEADER(UnboxFloatingPoint); + + LUnboxFloatingPoint(const LAllocation& input, MIRType type) + : LUnbox(classOpcode, input), type_(type) {} + + MIRType type() const { return type_; } +}; + +// Convert a 32-bit unsigned integer to a double. +class LWasmUint32ToDouble : public LInstructionHelper<1, 1, 0> { + public: + LIR_HEADER(WasmUint32ToDouble) + + explicit LWasmUint32ToDouble(const LAllocation& input) + : LInstructionHelper(classOpcode) { + setOperand(0, input); + } +}; + +// Convert a 32-bit unsigned integer to a float32. +class LWasmUint32ToFloat32 : public LInstructionHelper<1, 1, 0> { + public: + LIR_HEADER(WasmUint32ToFloat32) + + explicit LWasmUint32ToFloat32(const LAllocation& input) + : LInstructionHelper(classOpcode) { + setOperand(0, input); + } +}; + +class LDivI : public LBinaryMath<1> { + public: + LIR_HEADER(DivI); + + LDivI(const LAllocation& lhs, const LAllocation& rhs, const LDefinition& temp) + : LBinaryMath(classOpcode) { + setOperand(0, lhs); + setOperand(1, rhs); + setTemp(0, temp); + } + + MDiv* mir() const { return mir_->toDiv(); } +}; + +class LDivPowTwoI : public LInstructionHelper<1, 1, 1> { + const int32_t shift_; + + public: + LIR_HEADER(DivPowTwoI) + + LDivPowTwoI(const LAllocation& lhs, int32_t shift, const LDefinition& temp) + : LInstructionHelper(classOpcode), shift_(shift) { + setOperand(0, lhs); + setTemp(0, temp); + } + + const LAllocation* numerator() { return getOperand(0); } + int32_t shift() const { return shift_; } + MDiv* mir() const { return mir_->toDiv(); } +}; + +class LModI : public LBinaryMath<1> { + public: + LIR_HEADER(ModI); + + LModI(const LAllocation& lhs, const LAllocation& rhs, + const LDefinition& callTemp) + : LBinaryMath(classOpcode) { + setOperand(0, lhs); + setOperand(1, rhs); + setTemp(0, callTemp); + } + + const LDefinition* callTemp() { return getTemp(0); } + MMod* mir() const { return mir_->toMod(); } +}; + +class LModPowTwoI : public LInstructionHelper<1, 1, 0> { + const int32_t shift_; + + public: + LIR_HEADER(ModPowTwoI); + + LModPowTwoI(const LAllocation& lhs, int32_t shift) + : LInstructionHelper(classOpcode), shift_(shift) { + setOperand(0, lhs); + } + + int32_t shift() const { return shift_; } + MMod* mir() const { return mir_->toMod(); } +}; + +class LModMaskI : public LInstructionHelper<1, 1, 2> { + const int32_t shift_; + + public: + LIR_HEADER(ModMaskI); + + LModMaskI(const LAllocation& lhs, const LDefinition& temp0, + const LDefinition& temp1, int32_t shift) + : LInstructionHelper(classOpcode), shift_(shift) { + setOperand(0, lhs); + setTemp(0, temp0); + setTemp(1, temp1); + } + + int32_t shift() const { return shift_; } + MMod* mir() const { return mir_->toMod(); } +}; + +// Takes a tableswitch with an integer to decide +class LTableSwitch : public LInstructionHelper<0, 1, 2> { + public: + LIR_HEADER(TableSwitch); + + LTableSwitch(const LAllocation& in, const LDefinition& inputCopy, + const LDefinition& jumpTablePointer, MTableSwitch* ins) + : LInstructionHelper(classOpcode) { + setOperand(0, in); + setTemp(0, inputCopy); + setTemp(1, jumpTablePointer); + setMir(ins); + } + + MTableSwitch* mir() const { return mir_->toTableSwitch(); } + const LAllocation* index() { return getOperand(0); } + const LDefinition* tempInt() { return getTemp(0); } + // This is added to share the same CodeGenerator prefixes. + const LDefinition* tempPointer() { return getTemp(1); } +}; + +// Takes a tableswitch with an integer to decide +class LTableSwitchV : public LInstructionHelper<0, BOX_PIECES, 3> { + public: + LIR_HEADER(TableSwitchV); + + LTableSwitchV(const LBoxAllocation& input, const LDefinition& inputCopy, + const LDefinition& floatCopy, + const LDefinition& jumpTablePointer, MTableSwitch* ins) + : LInstructionHelper(classOpcode) { + setBoxOperand(InputValue, input); + setTemp(0, inputCopy); + setTemp(1, floatCopy); + setTemp(2, jumpTablePointer); + setMir(ins); + } + + MTableSwitch* mir() const { return mir_->toTableSwitch(); } + + static const size_t InputValue = 0; + + const LDefinition* tempInt() { return getTemp(0); } + const LDefinition* tempFloat() { return getTemp(1); } + const LDefinition* tempPointer() { return getTemp(2); } +}; + +class LMulI : public LBinaryMath<0> { + public: + LIR_HEADER(MulI); + + LMulI() : LBinaryMath(classOpcode) {} + + MMul* mir() { return mir_->toMul(); } +}; + +class LUDivOrMod : public LBinaryMath<0> { + public: + LIR_HEADER(UDivOrMod); + + LUDivOrMod() : LBinaryMath(classOpcode) {} + + MBinaryArithInstruction* mir() const { + MOZ_ASSERT(mir_->isDiv() || mir_->isMod()); + return static_cast<MBinaryArithInstruction*>(mir_); + } + + bool canBeDivideByZero() const { + if (mir_->isMod()) { + return mir_->toMod()->canBeDivideByZero(); + } + return mir_->toDiv()->canBeDivideByZero(); + } + + bool trapOnError() const { + if (mir_->isMod()) { + return mir_->toMod()->trapOnError(); + } + return mir_->toDiv()->trapOnError(); + } + + wasm::BytecodeOffset bytecodeOffset() const { + MOZ_ASSERT(mir_->isDiv() || mir_->isMod()); + if (mir_->isMod()) { + return mir_->toMod()->bytecodeOffset(); + } + return mir_->toDiv()->bytecodeOffset(); + } +}; + +class LWasmCompareExchangeI64 + : public LInstructionHelper<INT64_PIECES, 2 + INT64_PIECES + INT64_PIECES, + 0> { + public: + LIR_HEADER(WasmCompareExchangeI64); + + LWasmCompareExchangeI64(const LAllocation& ptr, + const LInt64Allocation& oldValue, + const LInt64Allocation& newValue, + const LAllocation& memoryBase) + : LInstructionHelper(classOpcode) { + setOperand(0, ptr); + setInt64Operand(1, oldValue); + setInt64Operand(1 + INT64_PIECES, newValue); + setOperand(1 + 2 * INT64_PIECES, memoryBase); + } + + const LAllocation* ptr() { return getOperand(0); } + const LInt64Allocation oldValue() { return getInt64Operand(1); } + const LInt64Allocation newValue() { + return getInt64Operand(1 + INT64_PIECES); + } + const LAllocation* memoryBase() { return getOperand(1 + 2 * INT64_PIECES); } + const MWasmCompareExchangeHeap* mir() const { + return mir_->toWasmCompareExchangeHeap(); + } +}; + +class LWasmAtomicExchangeI64 + : public LInstructionHelper<INT64_PIECES, 2 + INT64_PIECES, 0> { + public: + LIR_HEADER(WasmAtomicExchangeI64); + + LWasmAtomicExchangeI64(const LAllocation& ptr, const LInt64Allocation& value, + const LAllocation& memoryBase) + : LInstructionHelper(classOpcode) { + setOperand(0, ptr); + setInt64Operand(1, value); + setOperand(1 + INT64_PIECES, memoryBase); + } + + const LAllocation* ptr() { return getOperand(0); } + const LInt64Allocation value() { return getInt64Operand(1); } + const LAllocation* memoryBase() { return getOperand(1 + INT64_PIECES); } + const MWasmAtomicExchangeHeap* mir() const { + return mir_->toWasmAtomicExchangeHeap(); + } +}; + +class LWasmAtomicBinopI64 + : public LInstructionHelper<INT64_PIECES, 2 + INT64_PIECES, 2> { + public: + LIR_HEADER(WasmAtomicBinopI64); + + LWasmAtomicBinopI64(const LAllocation& ptr, const LInt64Allocation& value, + const LAllocation& memoryBase) + : LInstructionHelper(classOpcode) { + setOperand(0, ptr); + setInt64Operand(1, value); + setOperand(1 + INT64_PIECES, memoryBase); + } + + const LAllocation* ptr() { return getOperand(0); } + const LInt64Allocation value() { return getInt64Operand(1); } + const LAllocation* memoryBase() { return getOperand(1 + INT64_PIECES); } + const MWasmAtomicBinopHeap* mir() const { + return mir_->toWasmAtomicBinopHeap(); + } +}; + +class LDivOrModI64 : public LBinaryMath<1> { + public: + LIR_HEADER(DivOrModI64) + + LDivOrModI64(const LAllocation& lhs, const LAllocation& rhs, + const LDefinition& temp) + : LBinaryMath(classOpcode) { + setOperand(0, lhs); + setOperand(1, rhs); + setTemp(0, temp); + } + + const LDefinition* remainder() { return getTemp(0); } + MBinaryArithInstruction* mir() const { + MOZ_ASSERT(mir_->isDiv() || mir_->isMod()); + return static_cast<MBinaryArithInstruction*>(mir_); + } + + bool canBeDivideByZero() const { + if (mir_->isMod()) { + return mir_->toMod()->canBeDivideByZero(); + } + return mir_->toDiv()->canBeDivideByZero(); + } + bool canBeNegativeOverflow() const { + if (mir_->isMod()) { + return mir_->toMod()->canBeNegativeDividend(); + } + return mir_->toDiv()->canBeNegativeOverflow(); + } + wasm::BytecodeOffset bytecodeOffset() const { + MOZ_ASSERT(mir_->isDiv() || mir_->isMod()); + if (mir_->isMod()) { + return mir_->toMod()->bytecodeOffset(); + } + return mir_->toDiv()->bytecodeOffset(); + } +}; + +class LUDivOrModI64 : public LBinaryMath<1> { + public: + LIR_HEADER(UDivOrModI64); + + LUDivOrModI64(const LAllocation& lhs, const LAllocation& rhs, + const LDefinition& temp) + : LBinaryMath(classOpcode) { + setOperand(0, lhs); + setOperand(1, rhs); + setTemp(0, temp); + } + + const LDefinition* remainder() { return getTemp(0); } + const char* extraName() const { + return mir()->isTruncated() ? "Truncated" : nullptr; + } + + MBinaryArithInstruction* mir() const { + MOZ_ASSERT(mir_->isDiv() || mir_->isMod()); + return static_cast<MBinaryArithInstruction*>(mir_); + } + bool canBeDivideByZero() const { + if (mir_->isMod()) { + return mir_->toMod()->canBeDivideByZero(); + } + return mir_->toDiv()->canBeDivideByZero(); + } + wasm::BytecodeOffset bytecodeOffset() const { + MOZ_ASSERT(mir_->isDiv() || mir_->isMod()); + if (mir_->isMod()) { + return mir_->toMod()->bytecodeOffset(); + } + return mir_->toDiv()->bytecodeOffset(); + } +}; + +class LWasmTruncateToInt64 : public LInstructionHelper<1, 1, 0> { + public: + LIR_HEADER(WasmTruncateToInt64); + + explicit LWasmTruncateToInt64(const LAllocation& in) + : LInstructionHelper(classOpcode) { + setOperand(0, in); + } + + MWasmTruncateToInt64* mir() const { return mir_->toWasmTruncateToInt64(); } +}; + +class LInt64ToFloatingPoint : public LInstructionHelper<1, 1, 0> { + public: + LIR_HEADER(Int64ToFloatingPoint); + + explicit LInt64ToFloatingPoint(const LInt64Allocation& in) + : LInstructionHelper(classOpcode) { + setInt64Operand(0, in); + } + + MInt64ToFloatingPoint* mir() const { return mir_->toInt64ToFloatingPoint(); } +}; + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_LIR_riscv64_h */ diff --git a/js/src/jit/riscv64/Lowering-riscv64.cpp b/js/src/jit/riscv64/Lowering-riscv64.cpp new file mode 100644 index 0000000000..d8c1a49e3d --- /dev/null +++ b/js/src/jit/riscv64/Lowering-riscv64.cpp @@ -0,0 +1,1101 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "jit/riscv64/Lowering-riscv64.h" + +#include "mozilla/MathAlgorithms.h" + +#include "jit/Lowering.h" +#include "jit/MIR.h" +#include "jit/riscv64/Assembler-riscv64.h" + +#include "jit/shared/Lowering-shared-inl.h" + +using namespace js; +using namespace js::jit; + +using mozilla::FloorLog2; + +LTableSwitch* LIRGeneratorRiscv64::newLTableSwitch(const LAllocation& in, + const LDefinition& inputCopy, + MTableSwitch* tableswitch) { + return new (alloc()) LTableSwitch(in, inputCopy, temp(), tableswitch); +} + +LTableSwitchV* LIRGeneratorRiscv64::newLTableSwitchV( + MTableSwitch* tableswitch) { + return new (alloc()) LTableSwitchV(useBox(tableswitch->getOperand(0)), temp(), + tempDouble(), temp(), tableswitch); +} + +void LIRGeneratorRiscv64::lowerForShift(LInstructionHelper<1, 2, 0>* ins, + MDefinition* mir, MDefinition* lhs, + MDefinition* rhs) { + ins->setOperand(0, useRegister(lhs)); + ins->setOperand(1, useRegisterOrConstant(rhs)); + define(ins, mir); +} + +template <size_t Temps> +void LIRGeneratorRiscv64::lowerForShiftInt64( + LInstructionHelper<INT64_PIECES, INT64_PIECES + 1, Temps>* ins, + MDefinition* mir, MDefinition* lhs, MDefinition* rhs) { + ins->setInt64Operand(0, useInt64RegisterAtStart(lhs)); + + static_assert(LShiftI64::Rhs == INT64_PIECES, + "Assume Rhs is located at INT64_PIECES."); + static_assert(LRotateI64::Count == INT64_PIECES, + "Assume Count is located at INT64_PIECES."); + + ins->setOperand(INT64_PIECES, useRegisterOrConstant(rhs)); + + defineInt64ReuseInput(ins, mir, 0); +} + +template void LIRGeneratorRiscv64::lowerForShiftInt64( + LInstructionHelper<INT64_PIECES, INT64_PIECES + 1, 0>* ins, + MDefinition* mir, MDefinition* lhs, MDefinition* rhs); +template void LIRGeneratorRiscv64::lowerForShiftInt64( + LInstructionHelper<INT64_PIECES, INT64_PIECES + 1, 1>* ins, + MDefinition* mir, MDefinition* lhs, MDefinition* rhs); + +// x = !y +void LIRGeneratorRiscv64::lowerForALU(LInstructionHelper<1, 1, 0>* ins, + MDefinition* mir, MDefinition* input) { + ins->setOperand(0, useRegister(input)); + define( + ins, mir, + LDefinition(LDefinition::TypeFrom(mir->type()), LDefinition::REGISTER)); +} + +// z = x + y +void LIRGeneratorRiscv64::lowerForALU(LInstructionHelper<1, 2, 0>* ins, + MDefinition* mir, MDefinition* lhs, + MDefinition* rhs) { + ins->setOperand(0, useRegister(lhs)); + ins->setOperand(1, useRegisterOrConstant(rhs)); + define( + ins, mir, + LDefinition(LDefinition::TypeFrom(mir->type()), LDefinition::REGISTER)); +} + +void LIRGeneratorRiscv64::lowerForALUInt64( + LInstructionHelper<INT64_PIECES, INT64_PIECES, 0>* ins, MDefinition* mir, + MDefinition* input) { + ins->setInt64Operand(0, useInt64RegisterAtStart(input)); + defineInt64ReuseInput(ins, mir, 0); +} + +void LIRGeneratorRiscv64::lowerForALUInt64( + LInstructionHelper<INT64_PIECES, 2 * INT64_PIECES, 0>* ins, + MDefinition* mir, MDefinition* lhs, MDefinition* rhs) { + ins->setInt64Operand(0, useInt64RegisterAtStart(lhs)); + ins->setInt64Operand(INT64_PIECES, willHaveDifferentLIRNodes(lhs, rhs) + ? useInt64OrConstant(rhs) + : useInt64OrConstantAtStart(rhs)); + defineInt64ReuseInput(ins, mir, 0); +} + +void LIRGeneratorRiscv64::lowerForMulInt64(LMulI64* ins, MMul* mir, + MDefinition* lhs, MDefinition* rhs) { + bool needsTemp = false; + bool cannotAliasRhs = false; + bool reuseInput = true; + + ins->setInt64Operand(0, useInt64RegisterAtStart(lhs)); + ins->setInt64Operand(INT64_PIECES, + (willHaveDifferentLIRNodes(lhs, rhs) || cannotAliasRhs) + ? useInt64OrConstant(rhs) + : useInt64OrConstantAtStart(rhs)); + + if (needsTemp) { + ins->setTemp(0, temp()); + } + if (reuseInput) { + defineInt64ReuseInput(ins, mir, 0); + } else { + defineInt64(ins, mir); + } +} + +void LIRGeneratorRiscv64::lowerForFPU(LInstructionHelper<1, 1, 0>* ins, + MDefinition* mir, MDefinition* input) { + ins->setOperand(0, useRegister(input)); + define( + ins, mir, + LDefinition(LDefinition::TypeFrom(mir->type()), LDefinition::REGISTER)); +} + +template <size_t Temps> +void LIRGeneratorRiscv64::lowerForFPU(LInstructionHelper<1, 2, Temps>* ins, + MDefinition* mir, MDefinition* lhs, + MDefinition* rhs) { + ins->setOperand(0, useRegister(lhs)); + ins->setOperand(1, useRegister(rhs)); + define( + ins, mir, + LDefinition(LDefinition::TypeFrom(mir->type()), LDefinition::REGISTER)); +} + +template void LIRGeneratorRiscv64::lowerForFPU(LInstructionHelper<1, 2, 0>* ins, + MDefinition* mir, + MDefinition* lhs, + MDefinition* rhs); +template void LIRGeneratorRiscv64::lowerForFPU(LInstructionHelper<1, 2, 1>* ins, + MDefinition* mir, + MDefinition* lhs, + MDefinition* rhs); + +void LIRGeneratorRiscv64::lowerForCompareI64AndBranch( + MTest* mir, MCompare* comp, JSOp op, MDefinition* left, MDefinition* right, + MBasicBlock* ifTrue, MBasicBlock* ifFalse) { + LCompareI64AndBranch* lir = new (alloc()) + LCompareI64AndBranch(comp, op, useInt64Register(left), + useInt64OrConstant(right), ifTrue, ifFalse); + add(lir, mir); +} + +void LIRGeneratorRiscv64::lowerForBitAndAndBranch(LBitAndAndBranch* baab, + MInstruction* mir, + MDefinition* lhs, + MDefinition* rhs) { + baab->setOperand(0, useRegisterAtStart(lhs)); + baab->setOperand(1, useRegisterOrConstantAtStart(rhs)); + add(baab, mir); +} + +LBoxAllocation LIRGeneratorRiscv64::useBoxFixed(MDefinition* mir, Register reg1, + Register reg2, + bool useAtStart) { + MOZ_ASSERT(mir->type() == MIRType::Value); + + ensureDefined(mir); + return LBoxAllocation(LUse(reg1, mir->virtualRegister(), useAtStart)); +} + +LAllocation LIRGeneratorRiscv64::useByteOpRegister(MDefinition* mir) { + return useRegister(mir); +} + +LAllocation LIRGeneratorRiscv64::useByteOpRegisterAtStart(MDefinition* mir) { + return useRegisterAtStart(mir); +} + +LAllocation LIRGeneratorRiscv64::useByteOpRegisterOrNonDoubleConstant( + MDefinition* mir) { + return useRegisterOrNonDoubleConstant(mir); +} + +LDefinition LIRGeneratorRiscv64::tempByteOpRegister() { return temp(); } +LDefinition LIRGeneratorRiscv64::tempToUnbox() { return temp(); } + +void LIRGeneratorRiscv64::lowerUntypedPhiInput(MPhi* phi, + uint32_t inputPosition, + LBlock* block, size_t lirIndex) { + lowerTypedPhiInput(phi, inputPosition, block, lirIndex); +} +void LIRGeneratorRiscv64::lowerInt64PhiInput(MPhi* phi, uint32_t inputPosition, + LBlock* block, size_t lirIndex) { + lowerTypedPhiInput(phi, inputPosition, block, lirIndex); +} +void LIRGeneratorRiscv64::defineInt64Phi(MPhi* phi, size_t lirIndex) { + defineTypedPhi(phi, lirIndex); +} + +void LIRGeneratorRiscv64::lowerNegI(MInstruction* ins, MDefinition* input) { + define(new (alloc()) LNegI(useRegisterAtStart(input)), ins); +} +void LIRGeneratorRiscv64::lowerNegI64(MInstruction* ins, MDefinition* input) { + defineInt64ReuseInput(new (alloc()) LNegI64(useInt64RegisterAtStart(input)), + ins, 0); +} + +void LIRGeneratorRiscv64::lowerMulI(MMul* mul, MDefinition* lhs, + MDefinition* rhs) { + LMulI* lir = new (alloc()) LMulI; + if (mul->fallible()) { + assignSnapshot(lir, mul->bailoutKind()); + } + + lowerForALU(lir, mul, lhs, rhs); +} + +void LIRGeneratorRiscv64::lowerDivI(MDiv* div) { + if (div->isUnsigned()) { + lowerUDiv(div); + return; + } + + // Division instructions are slow. Division by constant denominators can be + // rewritten to use other instructions. + if (div->rhs()->isConstant()) { + int32_t rhs = div->rhs()->toConstant()->toInt32(); + // Check for division by a positive power of two, which is an easy and + // important case to optimize. Note that other optimizations are also + // possible; division by negative powers of two can be optimized in a + // similar manner as positive powers of two, and division by other + // constants can be optimized by a reciprocal multiplication technique. + int32_t shift = FloorLog2(rhs); + if (rhs > 0 && 1 << shift == rhs) { + LDivPowTwoI* lir = + new (alloc()) LDivPowTwoI(useRegister(div->lhs()), shift, temp()); + if (div->fallible()) { + assignSnapshot(lir, div->bailoutKind()); + } + define(lir, div); + return; + } + } + + LDivI* lir = new (alloc()) + LDivI(useRegister(div->lhs()), useRegister(div->rhs()), temp()); + if (div->fallible()) { + assignSnapshot(lir, div->bailoutKind()); + } + define(lir, div); +} + +void LIRGeneratorRiscv64::lowerDivI64(MDiv* div) { + if (div->isUnsigned()) { + lowerUDivI64(div); + return; + } + + LDivOrModI64* lir = new (alloc()) + LDivOrModI64(useRegister(div->lhs()), useRegister(div->rhs()), temp()); + defineInt64(lir, div); +} + +void LIRGeneratorRiscv64::lowerModI(MMod* mod) { + if (mod->isUnsigned()) { + lowerUMod(mod); + return; + } + + if (mod->rhs()->isConstant()) { + int32_t rhs = mod->rhs()->toConstant()->toInt32(); + int32_t shift = FloorLog2(rhs); + if (rhs > 0 && 1 << shift == rhs) { + LModPowTwoI* lir = + new (alloc()) LModPowTwoI(useRegister(mod->lhs()), shift); + if (mod->fallible()) { + assignSnapshot(lir, mod->bailoutKind()); + } + define(lir, mod); + return; + } else if (shift < 31 && (1 << (shift + 1)) - 1 == rhs) { + LModMaskI* lir = new (alloc()) + LModMaskI(useRegister(mod->lhs()), temp(LDefinition::GENERAL), + temp(LDefinition::GENERAL), shift + 1); + if (mod->fallible()) { + assignSnapshot(lir, mod->bailoutKind()); + } + define(lir, mod); + return; + } + } + LModI* lir = + new (alloc()) LModI(useRegister(mod->lhs()), useRegister(mod->rhs()), + temp(LDefinition::GENERAL)); + + if (mod->fallible()) { + assignSnapshot(lir, mod->bailoutKind()); + } + define(lir, mod); +} + +void LIRGeneratorRiscv64::lowerModI64(MMod* mod) { + if (mod->isUnsigned()) { + lowerUModI64(mod); + return; + } + + LDivOrModI64* lir = new (alloc()) + LDivOrModI64(useRegister(mod->lhs()), useRegister(mod->rhs()), temp()); + defineInt64(lir, mod); +} + +void LIRGeneratorRiscv64::lowerUDiv(MDiv* div) { + MDefinition* lhs = div->getOperand(0); + MDefinition* rhs = div->getOperand(1); + + LUDivOrMod* lir = new (alloc()) LUDivOrMod; + lir->setOperand(0, useRegister(lhs)); + lir->setOperand(1, useRegister(rhs)); + if (div->fallible()) { + assignSnapshot(lir, div->bailoutKind()); + } + + define(lir, div); +} + +void LIRGeneratorRiscv64::lowerUDivI64(MDiv* div) { + LUDivOrModI64* lir = new (alloc()) + LUDivOrModI64(useRegister(div->lhs()), useRegister(div->rhs()), temp()); + defineInt64(lir, div); +} + +void LIRGeneratorRiscv64::lowerUMod(MMod* mod) { + MDefinition* lhs = mod->getOperand(0); + MDefinition* rhs = mod->getOperand(1); + + LUDivOrMod* lir = new (alloc()) LUDivOrMod; + lir->setOperand(0, useRegister(lhs)); + lir->setOperand(1, useRegister(rhs)); + if (mod->fallible()) { + assignSnapshot(lir, mod->bailoutKind()); + } + + define(lir, mod); +} + +void LIRGeneratorRiscv64::lowerUModI64(MMod* mod) { + LUDivOrModI64* lir = new (alloc()) + LUDivOrModI64(useRegister(mod->lhs()), useRegister(mod->rhs()), temp()); + defineInt64(lir, mod); +} + +void LIRGeneratorRiscv64::lowerUrshD(MUrsh* mir) { + MDefinition* lhs = mir->lhs(); + MDefinition* rhs = mir->rhs(); + + MOZ_ASSERT(lhs->type() == MIRType::Int32); + MOZ_ASSERT(rhs->type() == MIRType::Int32); + + LUrshD* lir = new (alloc()) + LUrshD(useRegister(lhs), useRegisterOrConstant(rhs), temp()); + define(lir, mir); +} + +void LIRGeneratorRiscv64::lowerPowOfTwoI(MPow* mir) { + int32_t base = mir->input()->toConstant()->toInt32(); + MDefinition* power = mir->power(); + + auto* lir = new (alloc()) LPowOfTwoI(useRegister(power), base); + assignSnapshot(lir, mir->bailoutKind()); + define(lir, mir); +} + +void LIRGeneratorRiscv64::lowerBigIntDiv(MBigIntDiv* ins) { + auto* lir = new (alloc()) LBigIntDiv(useRegister(ins->lhs()), + useRegister(ins->rhs()), temp(), temp()); + define(lir, ins); + assignSafepoint(lir, ins); +} + +void LIRGeneratorRiscv64::lowerBigIntMod(MBigIntMod* ins) { + auto* lir = new (alloc()) LBigIntMod(useRegister(ins->lhs()), + useRegister(ins->rhs()), temp(), temp()); + define(lir, ins); + assignSafepoint(lir, ins); +} + +void LIRGeneratorRiscv64::lowerBigIntLsh(MBigIntLsh* ins) { + auto* lir = new (alloc()) LBigIntLsh( + useRegister(ins->lhs()), useRegister(ins->rhs()), temp(), temp(), temp()); + define(lir, ins); + assignSafepoint(lir, ins); +} + +void LIRGeneratorRiscv64::lowerBigIntRsh(MBigIntRsh* ins) { + auto* lir = new (alloc()) LBigIntRsh( + useRegister(ins->lhs()), useRegister(ins->rhs()), temp(), temp(), temp()); + define(lir, ins); + assignSafepoint(lir, ins); +} + +void LIRGeneratorRiscv64::lowerTruncateDToInt32(MTruncateToInt32* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Double); + + define(new (alloc()) LTruncateDToInt32(useRegister(opd), tempDouble()), ins); +} + +void LIRGeneratorRiscv64::lowerTruncateFToInt32(MTruncateToInt32* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Float32); + + define(new (alloc()) LTruncateFToInt32(useRegister(opd), tempFloat32()), ins); +} + +void LIRGeneratorRiscv64::lowerBuiltinInt64ToFloatingPoint( + MBuiltinInt64ToFloatingPoint* ins) { + MOZ_CRASH("We don't use it for this architecture"); +} + +void LIRGeneratorRiscv64::lowerWasmSelectI(MWasmSelect* select) { + auto* lir = new (alloc()) + LWasmSelect(useRegisterAtStart(select->trueExpr()), + useAny(select->falseExpr()), useRegister(select->condExpr())); + defineReuseInput(lir, select, LWasmSelect::TrueExprIndex); +} + +void LIRGeneratorRiscv64::lowerWasmSelectI64(MWasmSelect* select) { + auto* lir = new (alloc()) LWasmSelectI64( + useInt64RegisterAtStart(select->trueExpr()), + useInt64(select->falseExpr()), useRegister(select->condExpr())); + defineInt64ReuseInput(lir, select, LWasmSelectI64::TrueExprIndex); +} + +// On riscv we specialize the only cases where compare is {U,}Int32 and select +// is {U,}Int32. +bool LIRGeneratorShared::canSpecializeWasmCompareAndSelect( + MCompare::CompareType compTy, MIRType insTy) { + return insTy == MIRType::Int32 && (compTy == MCompare::Compare_Int32 || + compTy == MCompare::Compare_UInt32); +} + +void LIRGeneratorShared::lowerWasmCompareAndSelect(MWasmSelect* ins, + MDefinition* lhs, + MDefinition* rhs, + MCompare::CompareType compTy, + JSOp jsop) { + MOZ_ASSERT(canSpecializeWasmCompareAndSelect(compTy, ins->type())); + auto* lir = new (alloc()) LWasmCompareAndSelect( + useRegister(lhs), useRegister(rhs), compTy, jsop, + useRegisterAtStart(ins->trueExpr()), useRegister(ins->falseExpr())); + defineReuseInput(lir, ins, LWasmCompareAndSelect::IfTrueExprIndex); +} + +void LIRGeneratorRiscv64::lowerWasmBuiltinTruncateToInt32( + MWasmBuiltinTruncateToInt32* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Double || opd->type() == MIRType::Float32); + + if (opd->type() == MIRType::Double) { + define(new (alloc()) LWasmBuiltinTruncateDToInt32( + useRegister(opd), useFixed(ins->instance(), InstanceReg), + LDefinition::BogusTemp()), + ins); + return; + } + + define(new (alloc()) LWasmBuiltinTruncateFToInt32( + useRegister(opd), useFixed(ins->instance(), InstanceReg), + LDefinition::BogusTemp()), + ins); +} + +void LIRGeneratorRiscv64::lowerWasmBuiltinTruncateToInt64( + MWasmBuiltinTruncateToInt64* ins) { + MOZ_CRASH("We don't use it for this architecture"); +} + +void LIRGeneratorRiscv64::lowerWasmBuiltinDivI64(MWasmBuiltinDivI64* div) { + MOZ_CRASH("We don't use runtime div for this architecture"); +} + +void LIRGeneratorRiscv64::lowerWasmBuiltinModI64(MWasmBuiltinModI64* mod) { + MOZ_CRASH("We don't use runtime mod for this architecture"); +} + +void LIRGeneratorRiscv64::lowerAtomicLoad64(MLoadUnboxedScalar* ins) { + const LUse elements = useRegister(ins->elements()); + const LAllocation index = + useRegisterOrIndexConstant(ins->index(), ins->storageType()); + + auto* lir = new (alloc()) LAtomicLoad64(elements, index, temp(), tempInt64()); + define(lir, ins); + assignSafepoint(lir, ins); +} + +void LIRGeneratorRiscv64::lowerAtomicStore64(MStoreUnboxedScalar* ins) { + LUse elements = useRegister(ins->elements()); + LAllocation index = + useRegisterOrIndexConstant(ins->index(), ins->writeType()); + LAllocation value = useRegister(ins->value()); + + add(new (alloc()) LAtomicStore64(elements, index, value, tempInt64()), ins); +} + +void LIRGenerator::visitBox(MBox* box) { + MDefinition* opd = box->getOperand(0); + + // If the operand is a constant, emit near its uses. + if (opd->isConstant() && box->canEmitAtUses()) { + emitAtUses(box); + return; + } + + if (opd->isConstant()) { + define(new (alloc()) LValue(opd->toConstant()->toJSValue()), box, + LDefinition(LDefinition::BOX)); + } else { + LBox* ins = new (alloc()) LBox(useRegister(opd), opd->type()); + define(ins, box, LDefinition(LDefinition::BOX)); + } +} + +void LIRGenerator::visitUnbox(MUnbox* unbox) { + MDefinition* box = unbox->getOperand(0); + MOZ_ASSERT(box->type() == MIRType::Value); + + LUnbox* lir; + if (IsFloatingPointType(unbox->type())) { + lir = new (alloc()) + LUnboxFloatingPoint(useRegisterAtStart(box), unbox->type()); + } else if (unbox->fallible()) { + // If the unbox is fallible, load the Value in a register first to + // avoid multiple loads. + lir = new (alloc()) LUnbox(useRegisterAtStart(box)); + } else { + lir = new (alloc()) LUnbox(useAtStart(box)); + } + + if (unbox->fallible()) { + assignSnapshot(lir, unbox->bailoutKind()); + } + + define(lir, unbox); +} + +void LIRGenerator::visitAbs(MAbs* ins) { + define(allocateAbs(ins, useRegisterAtStart(ins->input())), ins); +} + +void LIRGenerator::visitCopySign(MCopySign* ins) { + MDefinition* lhs = ins->lhs(); + MDefinition* rhs = ins->rhs(); + + MOZ_ASSERT(IsFloatingPointType(lhs->type())); + MOZ_ASSERT(lhs->type() == rhs->type()); + MOZ_ASSERT(lhs->type() == ins->type()); + + LInstructionHelper<1, 2, 2>* lir; + if (lhs->type() == MIRType::Double) { + lir = new (alloc()) LCopySignD(); + } else { + lir = new (alloc()) LCopySignF(); + } + + lir->setTemp(0, temp()); + lir->setTemp(1, temp()); + + lir->setOperand(0, useRegisterAtStart(lhs)); + lir->setOperand(1, useRegister(rhs)); + defineReuseInput(lir, ins, 0); +} + +void LIRGenerator::visitPowHalf(MPowHalf* ins) { + MDefinition* input = ins->input(); + MOZ_ASSERT(input->type() == MIRType::Double); + LPowHalfD* lir = new (alloc()) LPowHalfD(useRegisterAtStart(input)); + defineReuseInput(lir, ins, 0); +} + +void LIRGenerator::visitExtendInt32ToInt64(MExtendInt32ToInt64* ins) { + defineInt64( + new (alloc()) LExtendInt32ToInt64(useRegisterAtStart(ins->input())), ins); +} + +void LIRGenerator::visitSignExtendInt64(MSignExtendInt64* ins) { + defineInt64(new (alloc()) + LSignExtendInt64(useInt64RegisterAtStart(ins->input())), + ins); +} + +void LIRGenerator::visitInt64ToFloatingPoint(MInt64ToFloatingPoint* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Int64); + MOZ_ASSERT(IsFloatingPointType(ins->type())); + + define(new (alloc()) LInt64ToFloatingPoint(useInt64Register(opd)), ins); +} + +void LIRGenerator::visitSubstr(MSubstr* ins) { + LSubstr* lir = new (alloc()) + LSubstr(useRegister(ins->string()), useRegister(ins->begin()), + useRegister(ins->length()), temp(), temp(), temp()); + define(lir, ins); + assignSafepoint(lir, ins); +} + +void LIRGenerator::visitCompareExchangeTypedArrayElement( + MCompareExchangeTypedArrayElement* ins) { + MOZ_ASSERT(ins->arrayType() != Scalar::Float32); + MOZ_ASSERT(ins->arrayType() != Scalar::Float64); + + MOZ_ASSERT(ins->elements()->type() == MIRType::Elements); + MOZ_ASSERT(ins->index()->type() == MIRType::IntPtr); + + const LUse elements = useRegister(ins->elements()); + const LAllocation index = + useRegisterOrIndexConstant(ins->index(), ins->arrayType()); + + const LAllocation newval = useRegister(ins->newval()); + const LAllocation oldval = useRegister(ins->oldval()); + + if (Scalar::isBigIntType(ins->arrayType())) { + LInt64Definition temp1 = tempInt64(); + LInt64Definition temp2 = tempInt64(); + + auto* lir = new (alloc()) LCompareExchangeTypedArrayElement64( + elements, index, oldval, newval, temp1, temp2); + define(lir, ins); + assignSafepoint(lir, ins); + return; + } + + // If the target is a floating register then we need a temp at the + // CodeGenerator level for creating the result. + + LDefinition outTemp = LDefinition::BogusTemp(); + LDefinition valueTemp = LDefinition::BogusTemp(); + LDefinition offsetTemp = LDefinition::BogusTemp(); + LDefinition maskTemp = LDefinition::BogusTemp(); + + if (ins->arrayType() == Scalar::Uint32 && IsFloatingPointType(ins->type())) { + outTemp = temp(); + } + + if (Scalar::byteSize(ins->arrayType()) < 4) { + valueTemp = temp(); + offsetTemp = temp(); + maskTemp = temp(); + } + + LCompareExchangeTypedArrayElement* lir = new (alloc()) + LCompareExchangeTypedArrayElement(elements, index, oldval, newval, + outTemp, valueTemp, offsetTemp, + maskTemp); + + define(lir, ins); +} + +void LIRGenerator::visitAtomicExchangeTypedArrayElement( + MAtomicExchangeTypedArrayElement* ins) { + MOZ_ASSERT(ins->elements()->type() == MIRType::Elements); + MOZ_ASSERT(ins->index()->type() == MIRType::IntPtr); + + const LUse elements = useRegister(ins->elements()); + const LAllocation index = + useRegisterOrIndexConstant(ins->index(), ins->arrayType()); + + const LAllocation value = useRegister(ins->value()); + + if (Scalar::isBigIntType(ins->arrayType())) { + LInt64Definition temp1 = tempInt64(); + LDefinition temp2 = temp(); + + auto* lir = new (alloc()) LAtomicExchangeTypedArrayElement64( + elements, index, value, temp1, temp2); + define(lir, ins); + assignSafepoint(lir, ins); + return; + } + + // If the target is a floating register then we need a temp at the + // CodeGenerator level for creating the result. + + MOZ_ASSERT(ins->arrayType() <= Scalar::Uint32); + + LDefinition outTemp = LDefinition::BogusTemp(); + LDefinition valueTemp = LDefinition::BogusTemp(); + LDefinition offsetTemp = LDefinition::BogusTemp(); + LDefinition maskTemp = LDefinition::BogusTemp(); + + if (ins->arrayType() == Scalar::Uint32) { + MOZ_ASSERT(ins->type() == MIRType::Double); + outTemp = temp(); + } + + if (Scalar::byteSize(ins->arrayType()) < 4) { + valueTemp = temp(); + offsetTemp = temp(); + maskTemp = temp(); + } + + LAtomicExchangeTypedArrayElement* lir = + new (alloc()) LAtomicExchangeTypedArrayElement( + elements, index, value, outTemp, valueTemp, offsetTemp, maskTemp); + + define(lir, ins); +} + +void LIRGenerator::visitAtomicTypedArrayElementBinop( + MAtomicTypedArrayElementBinop* ins) { + MOZ_ASSERT(ins->arrayType() != Scalar::Uint8Clamped); + MOZ_ASSERT(ins->arrayType() != Scalar::Float32); + MOZ_ASSERT(ins->arrayType() != Scalar::Float64); + + MOZ_ASSERT(ins->elements()->type() == MIRType::Elements); + MOZ_ASSERT(ins->index()->type() == MIRType::IntPtr); + + const LUse elements = useRegister(ins->elements()); + const LAllocation index = + useRegisterOrIndexConstant(ins->index(), ins->arrayType()); + const LAllocation value = useRegister(ins->value()); + + if (Scalar::isBigIntType(ins->arrayType())) { + LInt64Definition temp1 = tempInt64(); + LInt64Definition temp2 = tempInt64(); + + // Case 1: the result of the operation is not used. + // + // We can omit allocating the result BigInt. + + if (ins->isForEffect()) { + auto* lir = new (alloc()) LAtomicTypedArrayElementBinopForEffect64( + elements, index, value, temp1, temp2); + add(lir, ins); + return; + } + + // Case 2: the result of the operation is used. + + auto* lir = new (alloc()) + LAtomicTypedArrayElementBinop64(elements, index, value, temp1, temp2); + define(lir, ins); + assignSafepoint(lir, ins); + return; + } + + LDefinition valueTemp = LDefinition::BogusTemp(); + LDefinition offsetTemp = LDefinition::BogusTemp(); + LDefinition maskTemp = LDefinition::BogusTemp(); + + if (Scalar::byteSize(ins->arrayType()) < 4) { + valueTemp = temp(); + offsetTemp = temp(); + maskTemp = temp(); + } + + if (ins->isForEffect()) { + LAtomicTypedArrayElementBinopForEffect* lir = + new (alloc()) LAtomicTypedArrayElementBinopForEffect( + elements, index, value, valueTemp, offsetTemp, maskTemp); + add(lir, ins); + return; + } + + // For a Uint32Array with a known double result we need a temp for + // the intermediate output. + + LDefinition outTemp = LDefinition::BogusTemp(); + + if (ins->arrayType() == Scalar::Uint32 && IsFloatingPointType(ins->type())) { + outTemp = temp(); + } + + LAtomicTypedArrayElementBinop* lir = + new (alloc()) LAtomicTypedArrayElementBinop( + elements, index, value, outTemp, valueTemp, offsetTemp, maskTemp); + define(lir, ins); +} + +void LIRGenerator::visitReturnImpl(MDefinition* opd, bool isGenerator) { + MOZ_ASSERT(opd->type() == MIRType::Value); + + LReturn* ins = new (alloc()) LReturn(isGenerator); + ins->setOperand(0, useFixed(opd, JSReturnReg)); + add(ins); +} + +void LIRGenerator::visitAsmJSLoadHeap(MAsmJSLoadHeap* ins) { + MDefinition* base = ins->base(); + MOZ_ASSERT(base->type() == MIRType::Int32); + + MDefinition* boundsCheckLimit = ins->boundsCheckLimit(); + MOZ_ASSERT_IF(ins->needsBoundsCheck(), + boundsCheckLimit->type() == MIRType::Int32); + + LAllocation baseAlloc = useRegisterAtStart(base); + + LAllocation limitAlloc = ins->needsBoundsCheck() + ? useRegisterAtStart(boundsCheckLimit) + : LAllocation(); + + // We have no memory-base value, meaning that HeapReg is to be used as the + // memory base. This follows from the definition of + // FunctionCompiler::maybeLoadMemoryBase() in WasmIonCompile.cpp. + MOZ_ASSERT(!ins->hasMemoryBase()); + auto* lir = + new (alloc()) LAsmJSLoadHeap(baseAlloc, limitAlloc, LAllocation()); + define(lir, ins); +} + +void LIRGenerator::visitAsmJSStoreHeap(MAsmJSStoreHeap* ins) { + MDefinition* base = ins->base(); + MOZ_ASSERT(base->type() == MIRType::Int32); + + MDefinition* boundsCheckLimit = ins->boundsCheckLimit(); + MOZ_ASSERT_IF(ins->needsBoundsCheck(), + boundsCheckLimit->type() == MIRType::Int32); + + LAllocation baseAlloc = useRegisterAtStart(base); + + LAllocation limitAlloc = ins->needsBoundsCheck() + ? useRegisterAtStart(boundsCheckLimit) + : LAllocation(); + + // See comment in LIRGenerator::visitAsmJSStoreHeap just above. + MOZ_ASSERT(!ins->hasMemoryBase()); + add(new (alloc()) LAsmJSStoreHeap(baseAlloc, useRegisterAtStart(ins->value()), + limitAlloc, LAllocation()), + ins); +} + +void LIRGenerator::visitWasmLoad(MWasmLoad* ins) { + MDefinition* base = ins->base(); + // 'base' is a GPR but may be of either type. If it is 32-bit, it is + // sign-extended on riscv64 platform and we should explicitly promote it + // to 64-bit by zero-extension when use it as an index register in memory + // accesses. + MOZ_ASSERT(base->type() == MIRType::Int32 || base->type() == MIRType::Int64); + + LAllocation memoryBase = + ins->hasMemoryBase() ? LAllocation(useRegisterAtStart(ins->memoryBase())) + : LGeneralReg(HeapReg); + LAllocation ptr; + ptr = useRegisterAtStart(base); + + if (ins->type() == MIRType::Int64) { + auto* lir = new (alloc()) LWasmLoadI64(ptr, memoryBase); + if (ins->access().offset()) { + lir->setTemp(0, tempCopy(base, 0)); + } + + defineInt64(lir, ins); + return; + } + + auto* lir = new (alloc()) LWasmLoad(ptr, memoryBase); + if (ins->access().offset()) { + lir->setTemp(0, tempCopy(base, 0)); + } + + define(lir, ins); +} + +void LIRGenerator::visitWasmStore(MWasmStore* ins) { + MDefinition* base = ins->base(); + // See comment in visitWasmLoad re the type of 'base'. + MOZ_ASSERT(base->type() == MIRType::Int32 || base->type() == MIRType::Int64); + + MDefinition* value = ins->value(); + + LAllocation memoryBase = + ins->hasMemoryBase() ? LAllocation(useRegisterAtStart(ins->memoryBase())) + : LGeneralReg(HeapReg); + + if (ins->access().type() == Scalar::Int64) { + LAllocation baseAlloc = useRegisterAtStart(base); + LInt64Allocation valueAlloc = useInt64RegisterAtStart(value); + auto* lir = new (alloc()) LWasmStoreI64(baseAlloc, valueAlloc, memoryBase); + if (ins->access().offset()) { + lir->setTemp(0, tempCopy(base, 0)); + } + + add(lir, ins); + return; + } + + LAllocation baseAlloc = useRegisterAtStart(base); + LAllocation valueAlloc = useRegisterAtStart(value); + auto* lir = new (alloc()) LWasmStore(baseAlloc, valueAlloc, memoryBase); + if (ins->access().offset()) { + lir->setTemp(0, tempCopy(base, 0)); + } + + add(lir, ins); +} + +void LIRGenerator::visitWasmNeg(MWasmNeg* ins) { + if (ins->type() == MIRType::Int32) { + define(new (alloc()) LNegI(useRegisterAtStart(ins->input())), ins); + } else if (ins->type() == MIRType::Float32) { + define(new (alloc()) LNegF(useRegisterAtStart(ins->input())), ins); + } else { + MOZ_ASSERT(ins->type() == MIRType::Double); + define(new (alloc()) LNegD(useRegisterAtStart(ins->input())), ins); + } +} + +void LIRGenerator::visitWasmTruncateToInt64(MWasmTruncateToInt64* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Double || opd->type() == MIRType::Float32); + + defineInt64(new (alloc()) LWasmTruncateToInt64(useRegister(opd)), ins); +} + +void LIRGenerator::visitWasmUnsignedToDouble(MWasmUnsignedToDouble* ins) { + MOZ_ASSERT(ins->input()->type() == MIRType::Int32); + LWasmUint32ToDouble* lir = + new (alloc()) LWasmUint32ToDouble(useRegisterAtStart(ins->input())); + define(lir, ins); +} + +void LIRGenerator::visitWasmUnsignedToFloat32(MWasmUnsignedToFloat32* ins) { + MOZ_ASSERT(ins->input()->type() == MIRType::Int32); + LWasmUint32ToFloat32* lir = + new (alloc()) LWasmUint32ToFloat32(useRegisterAtStart(ins->input())); + define(lir, ins); +} + +void LIRGenerator::visitWasmCompareExchangeHeap(MWasmCompareExchangeHeap* ins) { + MDefinition* base = ins->base(); + // See comment in visitWasmLoad re the type of 'base'. + MOZ_ASSERT(base->type() == MIRType::Int32 || base->type() == MIRType::Int64); + + LAllocation memoryBase = + ins->hasMemoryBase() ? LAllocation(useRegisterAtStart(ins->memoryBase())) + : LGeneralReg(HeapReg); + if (ins->access().type() == Scalar::Int64) { + auto* lir = new (alloc()) LWasmCompareExchangeI64( + useRegister(base), useInt64Register(ins->oldValue()), + useInt64Register(ins->newValue()), memoryBase); + defineInt64(lir, ins); + return; + } + + LDefinition valueTemp = LDefinition::BogusTemp(); + LDefinition offsetTemp = LDefinition::BogusTemp(); + LDefinition maskTemp = LDefinition::BogusTemp(); + + if (ins->access().byteSize() < 4) { + valueTemp = temp(); + offsetTemp = temp(); + maskTemp = temp(); + } + + LWasmCompareExchangeHeap* lir = new (alloc()) + LWasmCompareExchangeHeap(useRegister(base), useRegister(ins->oldValue()), + useRegister(ins->newValue()), valueTemp, + offsetTemp, maskTemp, memoryBase); + + define(lir, ins); +} + +void LIRGenerator::visitWasmAtomicExchangeHeap(MWasmAtomicExchangeHeap* ins) { + MDefinition* base = ins->base(); + // See comment in visitWasmLoad re the type of 'base'. + MOZ_ASSERT(base->type() == MIRType::Int32 || base->type() == MIRType::Int64); + + LAllocation memoryBase = + ins->hasMemoryBase() ? LAllocation(useRegisterAtStart(ins->memoryBase())) + : LGeneralReg(HeapReg); + + if (ins->access().type() == Scalar::Int64) { + auto* lir = new (alloc()) LWasmAtomicExchangeI64( + useRegister(base), useInt64Register(ins->value()), memoryBase); + defineInt64(lir, ins); + return; + } + + LDefinition valueTemp = LDefinition::BogusTemp(); + LDefinition offsetTemp = LDefinition::BogusTemp(); + LDefinition maskTemp = LDefinition::BogusTemp(); + + if (ins->access().byteSize() < 4) { + valueTemp = temp(); + offsetTemp = temp(); + maskTemp = temp(); + } + + LWasmAtomicExchangeHeap* lir = new (alloc()) + LWasmAtomicExchangeHeap(useRegister(base), useRegister(ins->value()), + valueTemp, offsetTemp, maskTemp, memoryBase); + define(lir, ins); +} + +void LIRGenerator::visitWasmAtomicBinopHeap(MWasmAtomicBinopHeap* ins) { + MDefinition* base = ins->base(); + // See comment in visitWasmLoad re the type of 'base'. + MOZ_ASSERT(base->type() == MIRType::Int32 || base->type() == MIRType::Int64); + LAllocation memoryBase = + ins->hasMemoryBase() ? LAllocation(useRegisterAtStart(ins->memoryBase())) + : LGeneralReg(HeapReg); + + if (ins->access().type() == Scalar::Int64) { + auto* lir = new (alloc()) LWasmAtomicBinopI64( + useRegister(base), useInt64Register(ins->value()), memoryBase); + lir->setTemp(0, temp()); + defineInt64(lir, ins); + return; + } + + LDefinition valueTemp = LDefinition::BogusTemp(); + LDefinition offsetTemp = LDefinition::BogusTemp(); + LDefinition maskTemp = LDefinition::BogusTemp(); + + if (ins->access().byteSize() < 4) { + valueTemp = temp(); + offsetTemp = temp(); + maskTemp = temp(); + } + + if (!ins->hasUses()) { + LWasmAtomicBinopHeapForEffect* lir = new (alloc()) + LWasmAtomicBinopHeapForEffect(useRegister(base), + useRegister(ins->value()), valueTemp, + offsetTemp, maskTemp, memoryBase); + add(lir, ins); + return; + } + + LWasmAtomicBinopHeap* lir = new (alloc()) + LWasmAtomicBinopHeap(useRegister(base), useRegister(ins->value()), + valueTemp, offsetTemp, maskTemp, memoryBase); + + define(lir, ins); +} + +void LIRGenerator::visitWasmTernarySimd128(MWasmTernarySimd128* ins) { + MOZ_CRASH("ternary SIMD NYI"); +} + +void LIRGenerator::visitWasmBinarySimd128(MWasmBinarySimd128* ins) { + MOZ_CRASH("binary SIMD NYI"); +} + +#ifdef ENABLE_WASM_SIMD +bool MWasmTernarySimd128::specializeBitselectConstantMaskAsShuffle( + int8_t shuffle[16]) { + return false; +} +#endif + +bool MWasmBinarySimd128::specializeForConstantRhs() { + // Probably many we want to do here + return false; +} + +void LIRGenerator::visitWasmBinarySimd128WithConstant( + MWasmBinarySimd128WithConstant* ins) { + MOZ_CRASH("binary SIMD with constant NYI"); +} + +void LIRGenerator::visitWasmShiftSimd128(MWasmShiftSimd128* ins) { + MOZ_CRASH("shift SIMD NYI"); +} + +void LIRGenerator::visitWasmShuffleSimd128(MWasmShuffleSimd128* ins) { + MOZ_CRASH("shuffle SIMD NYI"); +} + +void LIRGenerator::visitWasmReplaceLaneSimd128(MWasmReplaceLaneSimd128* ins) { + MOZ_CRASH("replace-lane SIMD NYI"); +} + +void LIRGenerator::visitWasmScalarToSimd128(MWasmScalarToSimd128* ins) { + MOZ_CRASH("scalar-to-SIMD NYI"); +} + +void LIRGenerator::visitWasmUnarySimd128(MWasmUnarySimd128* ins) { + MOZ_CRASH("unary SIMD NYI"); +} + +void LIRGenerator::visitWasmReduceSimd128(MWasmReduceSimd128* ins) { + MOZ_CRASH("reduce-SIMD NYI"); +} + +void LIRGenerator::visitWasmLoadLaneSimd128(MWasmLoadLaneSimd128* ins) { + MOZ_CRASH("load-lane SIMD NYI"); +} + +void LIRGenerator::visitWasmStoreLaneSimd128(MWasmStoreLaneSimd128* ins) { + MOZ_CRASH("store-lane SIMD NYI"); +} diff --git a/js/src/jit/riscv64/Lowering-riscv64.h b/js/src/jit/riscv64/Lowering-riscv64.h new file mode 100644 index 0000000000..03ccb3ac8f --- /dev/null +++ b/js/src/jit/riscv64/Lowering-riscv64.h @@ -0,0 +1,110 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_Lowering_riscv64_h +#define jit_riscv64_Lowering_riscv64_h + +#include "jit/shared/Lowering-shared.h" + +namespace js { +namespace jit { + +class LIRGeneratorRiscv64 : public LIRGeneratorShared { + protected: + LIRGeneratorRiscv64(MIRGenerator* gen, MIRGraph& graph, LIRGraph& lirGraph) + : LIRGeneratorShared(gen, graph, lirGraph) {} + + LTableSwitch* newLTableSwitch(const LAllocation& in, + const LDefinition& inputCopy, + MTableSwitch* ins); + LTableSwitchV* newLTableSwitchV(MTableSwitch* ins); + + void lowerForShift(LInstructionHelper<1, 2, 0>* ins, MDefinition* mir, + MDefinition* lhs, MDefinition* rhs); + template <size_t Temps> + void lowerForShiftInt64( + LInstructionHelper<INT64_PIECES, INT64_PIECES + 1, Temps>* ins, + MDefinition* mir, MDefinition* lhs, MDefinition* rhs); + + void lowerForALU(LInstructionHelper<1, 1, 0>* ins, MDefinition* mir, + MDefinition* input); + void lowerForALU(LInstructionHelper<1, 2, 0>* ins, MDefinition* mir, + MDefinition* lhs, MDefinition* rhs); + void lowerForALUInt64(LInstructionHelper<INT64_PIECES, INT64_PIECES, 0>* ins, + MDefinition* mir, MDefinition* input); + void lowerForALUInt64( + LInstructionHelper<INT64_PIECES, 2 * INT64_PIECES, 0>* ins, + MDefinition* mir, MDefinition* lhs, MDefinition* rhs); + void lowerForMulInt64(LMulI64* ins, MMul* mir, MDefinition* lhs, + MDefinition* rhs); + + void lowerForFPU(LInstructionHelper<1, 1, 0>* ins, MDefinition* mir, + MDefinition* src); + template <size_t Temps> + void lowerForFPU(LInstructionHelper<1, 2, Temps>* ins, MDefinition* mir, + MDefinition* lhs, MDefinition* rhs); + + void lowerForCompareI64AndBranch(MTest* mir, MCompare* comp, JSOp op, + MDefinition* left, MDefinition* right, + MBasicBlock* ifTrue, MBasicBlock* ifFalse); + void lowerForBitAndAndBranch(LBitAndAndBranch* baab, MInstruction* mir, + MDefinition* lhs, MDefinition* rhs); + + // Returns a box allocation. reg2 is ignored on 64-bit platforms. + LBoxAllocation useBoxFixed(MDefinition* mir, Register reg1, Register reg2, + bool useAtStart = false); + + LAllocation useByteOpRegister(MDefinition* mir); + LAllocation useByteOpRegisterAtStart(MDefinition* mir); + LAllocation useByteOpRegisterOrNonDoubleConstant(MDefinition* mir); + + LDefinition tempByteOpRegister(); + LDefinition tempToUnbox(); + + bool needTempForPostBarrier() { return true; } + + void lowerUntypedPhiInput(MPhi* phi, uint32_t inputPosition, LBlock* block, + size_t lirIndex); + void lowerInt64PhiInput(MPhi*, uint32_t, LBlock*, size_t); + void defineInt64Phi(MPhi*, size_t); + + void lowerNegI(MInstruction* ins, MDefinition* input); + void lowerNegI64(MInstruction* ins, MDefinition* input); + void lowerMulI(MMul* mul, MDefinition* lhs, MDefinition* rhs); + void lowerDivI(MDiv* div); + void lowerDivI64(MDiv* div); + void lowerModI(MMod* mod); + void lowerModI64(MMod* mod); + void lowerUDiv(MDiv* div); + void lowerUDivI64(MDiv* div); + void lowerUMod(MMod* mod); + void lowerUModI64(MMod* mod); + void lowerUrshD(MUrsh* mir); + void lowerPowOfTwoI(MPow* mir); + void lowerBigIntDiv(MBigIntDiv* ins); + void lowerBigIntMod(MBigIntMod* ins); + void lowerBigIntLsh(MBigIntLsh* ins); + void lowerBigIntRsh(MBigIntRsh* ins); + void lowerTruncateDToInt32(MTruncateToInt32* ins); + void lowerTruncateFToInt32(MTruncateToInt32* ins); + void lowerBuiltinInt64ToFloatingPoint(MBuiltinInt64ToFloatingPoint* ins); + void lowerWasmSelectI(MWasmSelect* select); + void lowerWasmSelectI64(MWasmSelect* select); + void lowerWasmBuiltinTruncateToInt64(MWasmBuiltinTruncateToInt64* ins); + void lowerWasmBuiltinTruncateToInt32(MWasmBuiltinTruncateToInt32* ins); + void lowerWasmBuiltinDivI64(MWasmBuiltinDivI64* div); + void lowerWasmBuiltinModI64(MWasmBuiltinModI64* mod); + + void lowerAtomicLoad64(MLoadUnboxedScalar* ins); + void lowerAtomicStore64(MStoreUnboxedScalar* ins); +}; + +typedef LIRGeneratorRiscv64 LIRGeneratorSpecific; + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_Lowering_riscv64_h */ diff --git a/js/src/jit/riscv64/MacroAssembler-riscv64-inl.h b/js/src/jit/riscv64/MacroAssembler-riscv64-inl.h new file mode 100644 index 0000000000..096d00b384 --- /dev/null +++ b/js/src/jit/riscv64/MacroAssembler-riscv64-inl.h @@ -0,0 +1,2079 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_MacroAssembler_riscv64_inl_h +#define jit_riscv64_MacroAssembler_riscv64_inl_h + +#include "jit/riscv64/MacroAssembler-riscv64.h" + +namespace js { +namespace jit { + +template <> +inline void MacroAssembler::cmpPtrSet(Assembler::Condition cond, Address lhs, + ImmPtr rhs, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + cmpPtrSet(cond, Register(scratch2), rhs, dest); +} + +template <> +inline void MacroAssembler::cmpPtrSet(Assembler::Condition cond, Register lhs, + Address rhs, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(lhs != scratch); + loadPtr(rhs, scratch); + cmpPtrSet(cond, lhs, Register(scratch), dest); +} + +template <> +inline void MacroAssembler::cmpPtrSet(Assembler::Condition cond, Address lhs, + Register rhs, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(rhs != scratch2); + loadPtr(lhs, scratch2); + cmpPtrSet(cond, Register(scratch2), rhs, dest); +} + +template <> +inline void MacroAssembler::cmp32Set(Assembler::Condition cond, Register lhs, + Address rhs, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(lhs != scratch); + load32(rhs, scratch); + cmp32Set(cond, lhs, Register(scratch), dest); +} + +template <> +inline void MacroAssembler::cmp32Set(Assembler::Condition cond, Address lhs, + Register rhs, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(rhs != scratch2); + load32(lhs, scratch2); + cmp32Set(cond, Register(scratch2), rhs, dest); +} + +//{{{ check_macroassembler_style +CodeOffset MacroAssembler::sub32FromStackPtrWithPatch(Register dest) { + CodeOffset offset = CodeOffset(currentOffset()); + MacroAssemblerRiscv64::ma_liPatchable(dest, Imm32(0)); + sub(dest, StackPointer, dest); + return offset; +} + +template <class L> +void MacroAssembler::branchTest32(Condition cond, Register lhs, Imm32 rhs, + L label) { + MOZ_ASSERT(cond == Zero || cond == NonZero || cond == Signed || + cond == NotSigned); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_and(scratch, lhs, rhs); + ma_b(scratch, scratch, label, cond); +} +template <class L> +void MacroAssembler::branchTest32(Condition cond, Register lhs, Register rhs, + L label) { + MOZ_ASSERT(cond == Zero || cond == NonZero || cond == Signed || + cond == NotSigned); + if (lhs == rhs) { + ma_b(lhs, rhs, label, cond); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + and_(scratch, lhs, rhs); + ma_b(scratch, scratch, label, cond); + } +} +template <class L> +void MacroAssembler::branchTest64(Condition cond, Register64 lhs, + Register64 rhs, Register temp, L label) { + branchTestPtr(cond, lhs.reg, rhs.reg, label); +} + +template <typename T> +void MacroAssembler::branchAdd32(Condition cond, T src, Register dest, + Label* overflow) { + switch (cond) { + case Overflow: + ma_add32TestOverflow(dest, dest, src, overflow); + break; + case CarryClear: + case CarrySet: + ma_add32TestCarry(cond, dest, dest, src, overflow); + break; + default: + MOZ_CRASH("NYI"); + } +} +template <typename T> +void MacroAssembler::branchAddPtr(Condition cond, T src, Register dest, + Label* label) { + switch (cond) { + case Overflow: + ma_addPtrTestOverflow(dest, dest, src, label); + break; + case CarryClear: + case CarrySet: + ma_addPtrTestCarry(cond, dest, dest, src, label); + break; + default: + MOZ_CRASH("NYI"); + } +} +template <typename T> +void MacroAssembler::branchMul32(Condition cond, T src, Register dest, + Label* overflow) { + MOZ_ASSERT(cond == Assembler::Overflow); + ma_mul32TestOverflow(dest, dest, src, overflow); +} +template <typename T> +void MacroAssembler::branchRshift32(Condition cond, T src, Register dest, + Label* label) { + MOZ_ASSERT(cond == Zero || cond == NonZero); + rshift32(src, dest); + branch32(cond == Zero ? Equal : NotEqual, dest, Imm32(0), label); +} +// the type of 'T src' maybe a Register, maybe a Imm32,depends on who call it. +template <typename T> +void MacroAssembler::branchSub32(Condition cond, T src, Register dest, + Label* label) { + switch (cond) { + case Overflow: + ma_sub32TestOverflow(dest, dest, src, label); + break; + case NonZero: + case Zero: + case Signed: + case NotSigned: + ma_sub32(dest, dest, src); + ma_b(dest, dest, label, cond); + break; + default: + MOZ_CRASH("NYI"); + } +} +template <typename T> +void MacroAssembler::branchSubPtr(Condition cond, T src, Register dest, + Label* label) { + switch (cond) { + case Overflow: + ma_subPtrTestOverflow(dest, dest, src, label); + break; + case NonZero: + case Zero: + case Signed: + case NotSigned: + subPtr(src, dest); + ma_b(dest, dest, label, cond); + break; + default: + MOZ_CRASH("NYI"); + } +} +template <typename T> +void MacroAssembler::branchTestGCThingImpl(Condition cond, const T& address, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + ma_b(tag, ImmTag(JS::detail::ValueLowerInclGCThingTag), label, + (cond == Equal) ? AboveOrEqual : Below); +} +template <typename T> +void MacroAssembler::testBigIntSet(Condition cond, const T& src, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(src, scratch2); + ma_cmp_set(dest, tag, ImmTag(JSVAL_TAG_BIGINT), cond); +} + +template <typename T> +void MacroAssembler::testBooleanSet(Condition cond, const T& src, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(src, scratch2); + ma_cmp_set(dest, tag, ImmTag(JSVAL_TAG_BOOLEAN), cond); +} + +template <typename T> +void MacroAssembler::testNumberSet(Condition cond, const T& src, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(src, scratch2); + ma_cmp_set(dest, tag, ImmTag(JS::detail::ValueUpperInclNumberTag), + cond == Equal ? BelowOrEqual : Above); +} + +template <typename T> +void MacroAssembler::testStringSet(Condition cond, const T& src, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(src, scratch2); + ma_cmp_set(dest, tag, ImmTag(JSVAL_TAG_STRING), cond); +} + +template <typename T> +void MacroAssembler::testSymbolSet(Condition cond, const T& src, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(src, scratch2); + ma_cmp_set(dest, tag, ImmTag(JSVAL_TAG_SYMBOL), cond); +} + +// Also see below for specializations of cmpPtrSet. +template <typename T1, typename T2> +void MacroAssembler::cmp32Set(Condition cond, T1 lhs, T2 rhs, Register dest) { + ma_cmp_set(dest, lhs, rhs, cond); +} +template <typename T1, typename T2> +void MacroAssembler::cmpPtrSet(Condition cond, T1 lhs, T2 rhs, Register dest) { + ma_cmp_set(dest, lhs, rhs, cond); +} +void MacroAssembler::abs32(Register src, Register dest) { + ScratchRegisterScope scratch(asMasm()); + sraiw(scratch, src, 31); + xor_(dest, src, scratch); + subw(dest, dest, scratch); +} +void MacroAssembler::absFloat32(FloatRegister src, FloatRegister dest) { + fabs_s(dest, src); +} + +void MacroAssembler::absDouble(FloatRegister src, FloatRegister dest) { + fabs_d(dest, src); +} +void MacroAssembler::add32(Register src, Register dest) { + ma_add32(dest, dest, src); +} + +void MacroAssembler::add32(Imm32 imm, Register dest) { + ma_add32(dest, dest, imm); +} + +void MacroAssembler::add32(Imm32 imm, Register src, Register dest) { + ma_add32(dest, src, imm); +} + +void MacroAssembler::add32(Imm32 imm, const Address& dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(dest, scratch2); + ma_add32(scratch2, scratch2, imm); + store32(scratch2, dest); +} +void MacroAssembler::add64(Register64 src, Register64 dest) { + addPtr(src.reg, dest.reg); +} + +void MacroAssembler::add64(const Operand& src, Register64 dest) { + if (src.is_mem()) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register64 scratch64(scratch); + + load64(src.toAddress(), scratch64); + add64(scratch64, dest); + } else { + add64(Register64(src.toReg()), dest); + } +} + +void MacroAssembler::add64(Imm32 imm, Register64 dest) { + ma_add64(dest.reg, dest.reg, imm); +} + +void MacroAssembler::add64(Imm64 imm, Register64 dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(dest.reg != scratch); + mov(ImmWord(imm.value), scratch); + add(dest.reg, dest.reg, scratch); +} +void MacroAssembler::addDouble(FloatRegister src, FloatRegister dest) { + fadd_d(dest, dest, src); +} + +void MacroAssembler::addFloat32(FloatRegister src, FloatRegister dest) { + fadd_s(dest, dest, src); +} +void MacroAssembler::addPtr(Register src, Register dest) { + ma_add64(dest, dest, Operand(src)); +} + +void MacroAssembler::addPtr(Imm32 imm, Register dest) { + ma_add64(dest, dest, imm); +} + +void MacroAssembler::addPtr(ImmWord imm, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + movePtr(imm, scratch); + addPtr(scratch, dest); +} +void MacroAssembler::addPtr(Imm32 imm, const Address& dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + loadPtr(dest, scratch); + addPtr(imm, scratch); + storePtr(scratch, dest); +} + +void MacroAssembler::addPtr(const Address& src, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + loadPtr(src, scratch); + addPtr(scratch, dest); +} +void MacroAssembler::and32(Register src, Register dest) { + ma_and(dest, dest, src); +} + +void MacroAssembler::and32(Imm32 imm, Register dest) { + ma_and(dest, dest, imm); +} + +void MacroAssembler::and32(Imm32 imm, const Address& dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(dest, scratch2); + ma_and(scratch2, imm); + store32(scratch2, dest); +} + +void MacroAssembler::and32(const Address& src, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(src, scratch2); + ma_and(dest, dest, scratch2); +} +void MacroAssembler::and64(Imm64 imm, Register64 dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, ImmWord(imm.value)); + ma_and(dest.reg, dest.reg, scratch); +} + +void MacroAssembler::and64(Register64 src, Register64 dest) { + ma_and(dest.reg, dest.reg, src.reg); +} + +void MacroAssembler::and64(const Operand& src, Register64 dest) { + if (src.is_mem()) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register64 scratch64(scratch); + + load64(src.toAddress(), scratch64); + ma_and(dest.scratchReg(), scratch64.scratchReg()); + } else { + ma_and(dest.scratchReg(), src.toReg()); + } +} + +void MacroAssembler::andPtr(Register src, Register dest) { + ma_and(dest, dest, src); +} + +void MacroAssembler::andPtr(Imm32 imm, Register dest) { + ma_and(dest, dest, imm); +} + +void MacroAssembler::branch8(Condition cond, const Address& lhs, Imm32 rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(scratch2 != lhs.base); + + switch (cond) { + case Assembler::Equal: + case Assembler::NotEqual: + case Assembler::Above: + case Assembler::AboveOrEqual: + case Assembler::Below: + case Assembler::BelowOrEqual: + load8ZeroExtend(lhs, scratch2); + branch32(cond, scratch2, Imm32(uint8_t(rhs.value)), label); + break; + + case Assembler::GreaterThan: + case Assembler::GreaterThanOrEqual: + case Assembler::LessThan: + case Assembler::LessThanOrEqual: + load8SignExtend(lhs, scratch2); + branch32(cond, scratch2, Imm32(int8_t(rhs.value)), label); + break; + + default: + MOZ_CRASH("unexpected condition"); + } +} + +void MacroAssembler::branch8(Condition cond, const BaseIndex& lhs, Register rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(scratch2 != lhs.base); + + computeScaledAddress(lhs, scratch2); + + switch (cond) { + case Assembler::Equal: + case Assembler::NotEqual: + case Assembler::Above: + case Assembler::AboveOrEqual: + case Assembler::Below: + case Assembler::BelowOrEqual: + load8ZeroExtend(Address(scratch2, lhs.offset), scratch2); + branch32(cond, scratch2, rhs, label); + break; + + case Assembler::GreaterThan: + case Assembler::GreaterThanOrEqual: + case Assembler::LessThan: + case Assembler::LessThanOrEqual: + load8SignExtend(Address(scratch2, lhs.offset), scratch2); + branch32(cond, scratch2, rhs, label); + break; + + default: + MOZ_CRASH("unexpected condition"); + } +} + +void MacroAssembler::branch16(Condition cond, const Address& lhs, Imm32 rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(scratch2 != lhs.base); + + switch (cond) { + case Assembler::Equal: + case Assembler::NotEqual: + case Assembler::Above: + case Assembler::AboveOrEqual: + case Assembler::Below: + case Assembler::BelowOrEqual: + load16ZeroExtend(lhs, scratch2); + branch32(cond, scratch2, Imm32(uint16_t(rhs.value)), label); + break; + + case Assembler::GreaterThan: + case Assembler::GreaterThanOrEqual: + case Assembler::LessThan: + case Assembler::LessThanOrEqual: + load16SignExtend(lhs, scratch2); + branch32(cond, scratch2, Imm32(int16_t(rhs.value)), label); + break; + + default: + MOZ_CRASH("unexpected condition"); + } +} +template <class L> +void MacroAssembler::branch32(Condition cond, Register lhs, Register rhs, + L label) { + ma_b(lhs, rhs, label, cond); +} + +template <class L> +void MacroAssembler::branch32(Condition cond, Register lhs, Imm32 imm, + L label) { + ma_b(lhs, imm, label, cond); +} + +void MacroAssembler::branch32(Condition cond, const Address& lhs, Register rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(lhs, scratch2); + ma_b(scratch2, rhs, label, cond); +} + +void MacroAssembler::branch32(Condition cond, const Address& lhs, Imm32 rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(lhs, scratch2); + ma_b(scratch2, rhs, label, cond); +} + +void MacroAssembler::branch32(Condition cond, const AbsoluteAddress& lhs, + Register rhs, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(lhs, scratch2); + ma_b(scratch2, rhs, label, cond); +} + +void MacroAssembler::branch32(Condition cond, const AbsoluteAddress& lhs, + Imm32 rhs, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(lhs, scratch2); + ma_b(scratch2, rhs, label, cond); +} + +void MacroAssembler::branch32(Condition cond, const BaseIndex& lhs, Imm32 rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(lhs, scratch2); + ma_b(scratch2, rhs, label, cond); +} + +void MacroAssembler::branch32(Condition cond, wasm::SymbolicAddress addr, + Imm32 imm, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(addr, scratch2); + ma_b(scratch2, imm, label, cond); +} +void MacroAssembler::branch64(Condition cond, Register64 lhs, Imm64 val, + Label* success, Label* fail) { + MOZ_ASSERT(cond == Assembler::NotEqual || cond == Assembler::Equal || + cond == Assembler::LessThan || + cond == Assembler::LessThanOrEqual || + cond == Assembler::GreaterThan || + cond == Assembler::GreaterThanOrEqual || + cond == Assembler::Below || cond == Assembler::BelowOrEqual || + cond == Assembler::Above || cond == Assembler::AboveOrEqual, + "other condition codes not supported"); + + branchPtr(cond, lhs.reg, ImmWord(val.value), success); + if (fail) { + jump(fail); + } +} + +void MacroAssembler::branch64(Condition cond, Register64 lhs, Register64 rhs, + Label* success, Label* fail) { + MOZ_ASSERT(cond == Assembler::NotEqual || cond == Assembler::Equal || + cond == Assembler::LessThan || + cond == Assembler::LessThanOrEqual || + cond == Assembler::GreaterThan || + cond == Assembler::GreaterThanOrEqual || + cond == Assembler::Below || cond == Assembler::BelowOrEqual || + cond == Assembler::Above || cond == Assembler::AboveOrEqual, + "other condition codes not supported"); + + branchPtr(cond, lhs.reg, rhs.reg, success); + if (fail) { + jump(fail); + } +} + +void MacroAssembler::branch64(Condition cond, const Address& lhs, Imm64 val, + Label* label) { + MOZ_ASSERT(cond == Assembler::NotEqual || cond == Assembler::Equal, + "other condition codes not supported"); + + branchPtr(cond, lhs, ImmWord(val.value), label); +} + +void MacroAssembler::branch64(Condition cond, const Address& lhs, + Register64 rhs, Label* label) { + MOZ_ASSERT(cond == Assembler::NotEqual || cond == Assembler::Equal, + "other condition codes not supported"); + + branchPtr(cond, lhs, rhs.reg, label); +} + +void MacroAssembler::branch64(Condition cond, const Address& lhs, + const Address& rhs, Register scratch, + Label* label) { + MOZ_ASSERT(cond == Assembler::NotEqual || cond == Assembler::Equal, + "other condition codes not supported"); + MOZ_ASSERT(lhs.base != scratch); + MOZ_ASSERT(rhs.base != scratch); + + loadPtr(rhs, scratch); + branchPtr(cond, lhs, scratch, label); +} + +void MacroAssembler::branchDouble(DoubleCondition cc, FloatRegister frs1, + FloatRegister frs2, Label* L) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_compareF64(scratch, cc, frs1, frs2); + ma_b(scratch, Imm32(1), L, Equal); +} +void MacroAssembler::branchFloat(DoubleCondition cc, FloatRegister frs1, + FloatRegister frs2, Label* L) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_compareF32(scratch, cc, frs1, frs2); + ma_b(scratch, Imm32(1), L, Equal); +} +void MacroAssembler::branchMulPtr(Condition cond, Register src, Register dest, + Label* label) { + MOZ_ASSERT(cond == Assembler::Overflow); + ma_mulPtrTestOverflow(dest, dest, src, label); +} +void MacroAssembler::branchNeg32(Condition cond, Register reg, Label* label) { + MOZ_ASSERT(cond == Overflow); + neg32(reg); + branch32(Assembler::Equal, reg, Imm32(INT32_MIN), label); +} + +void MacroAssembler::branchPrivatePtr(Condition cond, const Address& lhs, + Register rhs, Label* label) { + branchPtr(cond, lhs, rhs, label); +} + +template <class L> +void MacroAssembler::branchPtr(Condition cond, Register lhs, Register rhs, + L label) { + ma_b(lhs, rhs, label, cond); +} + +void MacroAssembler::branchPtr(Condition cond, Register lhs, Imm32 rhs, + Label* label) { + ma_b(lhs, rhs, label, cond); +} + +void MacroAssembler::branchPtr(Condition cond, Register lhs, ImmPtr rhs, + Label* label) { + if (rhs.value == nullptr && (cond == Zero || cond == NonZero)) { + ma_b(lhs, lhs, label, cond); + } else { + ma_b(lhs, rhs, label, cond); + } +} + +void MacroAssembler::branchPtr(Condition cond, Register lhs, ImmGCPtr rhs, + Label* label) { + ma_b(lhs, rhs, label, cond); +} + +void MacroAssembler::branchPtr(Condition cond, Register lhs, ImmWord rhs, + Label* label) { + ma_b(lhs, rhs, label, cond); +} + +template <class L> +void MacroAssembler::branchPtr(Condition cond, const Address& lhs, Register rhs, + L label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchPtr(Condition cond, const Address& lhs, ImmPtr rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchPtr(Condition cond, const Address& lhs, ImmGCPtr rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchPtr(Condition cond, const Address& lhs, ImmWord rhs, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchPtr(Condition cond, const AbsoluteAddress& lhs, + Register rhs, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchPtr(Condition cond, const AbsoluteAddress& lhs, + ImmWord rhs, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchPtr(Condition cond, wasm::SymbolicAddress lhs, + Register rhs, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchPtr(Condition cond, const BaseIndex& lhs, + Register rhs, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchPtr(Condition cond, const BaseIndex& lhs, + ImmWord rhs, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchPtr(cond, scratch2, rhs, label); +} + +void MacroAssembler::branchTest32(Condition cond, const AbsoluteAddress& lhs, + Imm32 rhs, Label* label) { + MOZ_ASSERT(cond == Zero || cond == NonZero || cond == Signed || + cond == NotSigned); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(lhs, scratch2); + and32(rhs, scratch2); + ma_b(scratch2, scratch2, label, cond); +} + +void MacroAssembler::branchTest32(Condition cond, const Address& lhs, Imm32 rhs, + Label* label) { + MOZ_ASSERT(cond == Zero || cond == NonZero || cond == Signed || + cond == NotSigned); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(lhs, scratch2); + and32(rhs, scratch2); + ma_b(scratch2, scratch2, label, cond); +} +void MacroAssembler::branchTestBigInt(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_BIGINT), label, cond); +} + +void MacroAssembler::branchTestBigInt(Condition cond, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestBigInt(cond, scratch2, label); +} + +void MacroAssembler::branchTestBigInt(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestBigInt(cond, tag, label); +} + +void MacroAssembler::branchTestBigInt(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + computeEffectiveAddress(address, scratch2); + splitTag(scratch2, scratch2); + branchTestBigInt(cond, scratch2, label); +} +void MacroAssembler::branchTestBigIntTruthy(bool b, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + unboxBigInt(value, scratch2); + load32(Address(scratch2, BigInt::offsetOfDigitLength()), scratch2); + ma_b(scratch2, Imm32(0), label, b ? NotEqual : Equal); +} +void MacroAssembler::branchTestBoolean(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_BOOLEAN), label, cond); +} + +void MacroAssembler::branchTestBoolean(Condition cond, + const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestBoolean(cond, scratch2, label); +} + +void MacroAssembler::branchTestBoolean(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestBoolean(cond, tag, label); +} + +void MacroAssembler::branchTestBoolean(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestBoolean(cond, tag, label); +} +void MacroAssembler::branchTestBooleanTruthy(bool b, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + unboxBoolean(value, scratch2); + ma_b(scratch2, scratch2, label, b ? NonZero : Zero); +} +void MacroAssembler::branchTestDouble(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + Condition actual = (cond == Equal) ? BelowOrEqual : Above; + ma_b(tag, ImmTag(JSVAL_TAG_MAX_DOUBLE), label, actual); +} + +void MacroAssembler::branchTestDouble(Condition cond, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestDouble(cond, scratch2, label); +} + +void MacroAssembler::branchTestDouble(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestDouble(cond, tag, label); +} + +void MacroAssembler::branchTestDouble(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestDouble(cond, tag, label); +} + +void MacroAssembler::branchTestDoubleTruthy(bool b, FloatRegister value, + Label* label) { + ScratchDoubleScope fpscratch(*this); + loadConstantDouble(0.0, fpscratch); + DoubleCondition cond = b ? DoubleNotEqual : DoubleEqualOrUnordered; + branchDouble(cond, value, fpscratch, label); +} +void MacroAssembler::branchTestGCThing(Condition cond, const Address& address, + Label* label) { + branchTestGCThingImpl(cond, address, label); +} + +void MacroAssembler::branchTestGCThing(Condition cond, const BaseIndex& address, + Label* label) { + branchTestGCThingImpl(cond, address, label); +} + +void MacroAssembler::branchTestGCThing(Condition cond, + const ValueOperand& address, + Label* label) { + branchTestGCThingImpl(cond, address, label); +} +void MacroAssembler::branchTestInt32(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_INT32), label, cond); +} + +void MacroAssembler::branchTestInt32(Condition cond, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestInt32(cond, scratch2, label); +} + +void MacroAssembler::branchTestInt32(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestInt32(cond, tag, label); +} + +void MacroAssembler::branchTestInt32(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestInt32(cond, tag, label); +} +void MacroAssembler::branchTestInt32Truthy(bool b, const ValueOperand& value, + Label* label) { + ScratchRegisterScope scratch(*this); + ExtractBits(scratch, value.valueReg(), 0, 32); + ma_b(scratch, scratch, label, b ? NonZero : Zero); +} +void MacroAssembler::branchTestMagic(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_MAGIC), label, cond); +} + +void MacroAssembler::branchTestMagic(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestMagic(cond, tag, label); +} + +void MacroAssembler::branchTestMagic(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestMagic(cond, tag, label); +} + +template <class L> +void MacroAssembler::branchTestMagic(Condition cond, const ValueOperand& value, + L label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + ma_b(scratch2, ImmTag(JSVAL_TAG_MAGIC), label, cond); +} + +void MacroAssembler::branchTestMagic(Condition cond, const Address& valaddr, + JSWhyMagic why, Label* label) { + uint64_t magic = MagicValue(why).asRawBits(); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + loadPtr(valaddr, scratch); + ma_b(scratch, ImmWord(magic), label, cond); +} +void MacroAssembler::branchTestNull(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_NULL), label, cond); +} + +void MacroAssembler::branchTestNull(Condition cond, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestNull(cond, scratch2, label); +} + +void MacroAssembler::branchTestNull(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestNull(cond, tag, label); +} + +void MacroAssembler::branchTestNull(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestNull(cond, tag, label); +} +void MacroAssembler::branchTestNumber(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + Condition actual = cond == Equal ? BelowOrEqual : Above; + ma_b(tag, ImmTag(JS::detail::ValueUpperInclNumberTag), label, actual); +} + +void MacroAssembler::branchTestNumber(Condition cond, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestNumber(cond, scratch2, label); +} +void MacroAssembler::branchTestObject(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_OBJECT), label, cond); +} + +void MacroAssembler::branchTestObject(Condition cond, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestObject(cond, scratch2, label); +} + +void MacroAssembler::branchTestObject(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestObject(cond, tag, label); +} + +void MacroAssembler::branchTestObject(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestObject(cond, tag, label); +} +void MacroAssembler::branchTestPrimitive(Condition cond, + const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestPrimitive(cond, scratch2, label); +} +void MacroAssembler::branchTestPrimitive(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JS::detail::ValueUpperExclPrimitiveTag), label, + (cond == Equal) ? Below : AboveOrEqual); +} +template <class L> +void MacroAssembler::branchTestPtr(Condition cond, Register lhs, Register rhs, + L label) { + MOZ_ASSERT(cond == Zero || cond == NonZero || cond == Signed || + cond == NotSigned); + if (lhs == rhs) { + ma_b(lhs, rhs, label, cond); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_and(scratch, lhs, Operand(rhs)); + ma_b(scratch, scratch, label, cond); + } +} + +void MacroAssembler::branchTestPtr(Condition cond, Register lhs, Imm32 rhs, + Label* label) { + MOZ_ASSERT(cond == Zero || cond == NonZero || cond == Signed || + cond == NotSigned); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_and(scratch, lhs, rhs); + ma_b(scratch, scratch, label, cond); +} + +void MacroAssembler::branchTestPtr(Condition cond, const Address& lhs, + Imm32 rhs, Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(lhs, scratch2); + branchTestPtr(cond, scratch2, rhs, label); +} +void MacroAssembler::branchTestString(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_STRING), label, cond); +} + +void MacroAssembler::branchTestString(Condition cond, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestString(cond, scratch2, label); +} + +void MacroAssembler::branchTestString(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestString(cond, tag, label); +} + +void MacroAssembler::branchTestString(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestString(cond, tag, label); +} +void MacroAssembler::branchTestStringTruthy(bool b, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + unboxString(value, scratch2); + load32(Address(scratch2, JSString::offsetOfLength()), scratch2); + ma_b(scratch2, Imm32(0), label, b ? NotEqual : Equal); +} +void MacroAssembler::branchTestSymbol(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_SYMBOL), label, cond); +} + +void MacroAssembler::branchTestSymbol(Condition cond, const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestSymbol(cond, scratch2, label); +} + +void MacroAssembler::branchTestSymbol(Condition cond, const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestSymbol(cond, tag, label); +} + +void MacroAssembler::branchTestSymbol(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestSymbol(cond, tag, label); +} +void MacroAssembler::branchTestUndefined(Condition cond, + const ValueOperand& value, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + splitTag(value, scratch2); + branchTestUndefined(cond, scratch2, label); +} + +void MacroAssembler::branchTestUndefined(Condition cond, const Address& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestUndefined(cond, tag, label); +} + +void MacroAssembler::branchTestUndefined(Condition cond, + const BaseIndex& address, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register tag = extractTag(address, scratch2); + branchTestUndefined(cond, tag, label); +} +void MacroAssembler::branchTestUndefined(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(tag, ImmTag(JSVAL_TAG_UNDEFINED), label, cond); +} +void MacroAssembler::branchTestValue(Condition cond, const BaseIndex& lhs, + const ValueOperand& rhs, Label* label) { + MOZ_ASSERT(cond == Assembler::Equal || cond == Assembler::NotEqual); + branchPtr(cond, lhs, rhs.valueReg(), label); +} +void MacroAssembler::branchToComputedAddress(const BaseIndex& addr) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(addr, scratch2); + branch(scratch2); +} +void MacroAssembler::branchTruncateDoubleMaybeModUint32(FloatRegister src, + Register dest, + Label* fail) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Trunc_w_d(dest, src, scratch); + ma_b(scratch, Imm32(0), fail, Assembler::Equal); +} + +void MacroAssembler::branchTruncateDoubleToInt32(FloatRegister src, + Register dest, Label* fail) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Trunc_w_d(dest, src, scratch); + ma_b(scratch, Imm32(0), fail, Assembler::Equal); +} +void MacroAssembler::branchTruncateFloat32MaybeModUint32(FloatRegister src, + Register dest, + Label* fail) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Trunc_w_s(dest, src, scratch); + ma_b(scratch, Imm32(0), fail, Assembler::Equal); +} + +void MacroAssembler::branchTruncateFloat32ToInt32(FloatRegister src, + Register dest, Label* fail) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Trunc_w_s(dest, src, scratch); + ma_b(scratch, Imm32(0), fail, Assembler::Equal); +} + +void MacroAssembler::byteSwap16SignExtend(Register src) { + JitSpew(JitSpew_Codegen, "[ %s\n", __FUNCTION__); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register scratch2 = temps.Acquire(); + // src 0xFFFFFFFFFFFF8000 + andi(scratch, src, 0xFF); // + slli(scratch, scratch, 8); // scratch 0x00 + ma_li(scratch2, 0xFF00); // scratch2 0xFF00 + and_(src, src, scratch2); // src 0x8000 + srli(src, src, 8); // src 0x0080 + or_(src, src, scratch); // src 0x0080 + slliw(src, src, 16); + sraiw(src, src, 16); + JitSpew(JitSpew_Codegen, "]"); +} + +void MacroAssembler::byteSwap16ZeroExtend(Register src) { + JitSpew(JitSpew_Codegen, "[ %s\n", __FUNCTION__); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register scratch2 = temps.Acquire(); + andi(scratch, src, 0xFF); + slli(scratch, scratch, 8); + ma_li(scratch2, 0xFF00); + and_(src, src, scratch2); + srli(src, src, 8); + or_(src, src, scratch); + slliw(src, src, 16); + srliw(src, src, 16); + JitSpew(JitSpew_Codegen, "]"); +} + +void MacroAssembler::byteSwap32(Register src) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ByteSwap(src, src, 4, scratch); +} +void MacroAssembler::byteSwap64(Register64 src) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ByteSwap(src.reg, src.reg, 8, scratch); +} +void MacroAssembler::clampIntToUint8(Register reg) { + // If reg is < 0, then we want to clamp to 0. + Label skip, skip2; + slti(ScratchRegister, reg, 0); + ma_branch(&skip, NotEqual, ScratchRegister, Operand(1)); + ma_li(reg, Imm32(0)); + jump(&skip2); + bind(&skip); + // If reg is >= 255, then we want to clamp to 255. + ma_branch(&skip2, LessThanOrEqual, reg, Operand(255)); + ma_li(reg, Imm32(255)); + bind(&skip2); +} + +void MacroAssembler::clz32(Register src, Register dest, bool knownNotZero) { + Clz32(dest, src); +} +void MacroAssembler::clz64(Register64 src, Register dest) { + Clz64(dest, src.reg); +} + +void MacroAssembler::ctz64(Register64 src, Register dest) { + Ctz64(dest, src.reg); +} + +void MacroAssembler::cmp16Set(Condition cond, Address lhs, Imm32 rhs, + Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(scratch2 != lhs.base); + + switch (cond) { + case Assembler::Equal: + case Assembler::NotEqual: + case Assembler::Above: + case Assembler::AboveOrEqual: + case Assembler::Below: + case Assembler::BelowOrEqual: + load16ZeroExtend(lhs, scratch2); + ma_cmp_set(dest, scratch2, Imm32(uint16_t(rhs.value)), cond); + break; + + case Assembler::GreaterThan: + case Assembler::GreaterThanOrEqual: + case Assembler::LessThan: + case Assembler::LessThanOrEqual: + load16SignExtend(lhs, scratch2); + ma_cmp_set(dest, scratch2, Imm32(int16_t(rhs.value)), cond); + break; + + default: + MOZ_CRASH("unexpected condition"); + } +} + +void MacroAssembler::cmp32Load32(Condition cond, Register lhs, + const Address& rhs, const Address& src, + Register dest) { + ScratchRegisterScope scratch(*this); + MOZ_ASSERT(lhs != scratch && dest != scratch); + load32(rhs, scratch); + cmp32Load32(cond, lhs, scratch, src, dest); +} + +void MacroAssembler::cmp32Load32(Condition cond, Register lhs, Register rhs, + const Address& src, Register dest) { + Label skip; + branch32(Assembler::InvertCondition(cond), lhs, rhs, &skip); + load32(src, dest); + bind(&skip); +} + +void MacroAssembler::cmp32Load32(Condition cond, Register lhs, Imm32 rhs, + const Address& src, Register dest) { + Label skip; + branch32(Assembler::InvertCondition(cond), lhs, rhs, &skip); + load32(src, dest); + bind(&skip); +} + +void MacroAssembler::cmp32LoadPtr(Condition cond, const Address& lhs, Imm32 rhs, + const Address& src, Register dest) { + Label skip; + branch32(Assembler::InvertCondition(cond), lhs, rhs, &skip); + loadPtr(src, dest); + bind(&skip); +} + +void MacroAssembler::cmp32Move32(Condition cond, Register lhs, Imm32 rhs, + Register src, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + cmp32Set(cond, lhs, rhs, scratch2); + moveIfNotZero(dest, src, scratch2); +} + +void MacroAssembler::cmp32Move32(Condition cond, Register lhs, Register rhs, + Register src, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + cmp32Set(cond, lhs, rhs, scratch2); + moveIfNotZero(dest, src, scratch2); +} + +void MacroAssembler::cmp32Move32(Condition cond, Register lhs, + const Address& rhs, Register src, + Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(lhs != scratch2 && src != scratch2 && dest != scratch2); + load32(rhs, scratch2); + cmp32Move32(cond, lhs, scratch2, src, dest); +} +void MacroAssembler::cmp32MovePtr(Condition cond, Register lhs, Imm32 rhs, + Register src, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + cmp32Set(cond, lhs, rhs, scratch2); + moveIfNotZero(dest, src, scratch2); +} +void MacroAssembler::cmp64Set(Condition cond, Address lhs, Imm64 rhs, + Register dest) { + ma_cmp_set(dest, lhs, ImmWord(uint64_t(rhs.value)), cond); +} +void MacroAssembler::cmp8Set(Condition cond, Address lhs, Imm32 rhs, + Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(scratch2 != lhs.base); + + switch (cond) { + case Assembler::Equal: + case Assembler::NotEqual: + case Assembler::Above: + case Assembler::AboveOrEqual: + case Assembler::Below: + case Assembler::BelowOrEqual: + load8ZeroExtend(lhs, scratch2); + ma_cmp_set(dest, scratch2, Imm32(uint8_t(rhs.value)), cond); + break; + + case Assembler::GreaterThan: + case Assembler::GreaterThanOrEqual: + case Assembler::LessThan: + case Assembler::LessThanOrEqual: + load8SignExtend(lhs, scratch2); + ma_cmp_set(dest, scratch2, Imm32(int8_t(rhs.value)), cond); + break; + + default: + MOZ_CRASH("unexpected condition"); + } +} +void MacroAssembler::cmpPtrMovePtr(Condition cond, Register lhs, Register rhs, + Register src, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + cmpPtrSet(cond, lhs, rhs, scratch2); + moveIfNotZero(dest, src, scratch2); +} + +void MacroAssembler::cmpPtrMovePtr(Condition cond, Register lhs, + const Address& rhs, Register src, + Register dest) { + MOZ_CRASH("NYI"); +} + +void MacroAssembler::ctz32(Register rd, Register rs, bool knownNotZero) { + Ctz32(rd, rs); +} + +void MacroAssembler::decBranchPtr(Condition cond, Register lhs, Imm32 rhs, + Label* label) { + subPtr(rhs, lhs); + branchPtr(cond, lhs, Imm32(0), label); +} +void MacroAssembler::divFloat32(FloatRegister src, FloatRegister dest) { + fdiv_s(dest, dest, src); +} + +void MacroAssembler::divDouble(FloatRegister src, FloatRegister dest) { + fdiv_d(dest, dest, src); +} +void MacroAssembler::fallibleUnboxPtr(const ValueOperand& src, Register dest, + JSValueType type, Label* fail) { + MOZ_ASSERT(type == JSVAL_TYPE_OBJECT || type == JSVAL_TYPE_STRING || + type == JSVAL_TYPE_SYMBOL || type == JSVAL_TYPE_BIGINT); + // dest := src XOR mask + // scratch := dest >> JSVAL_TAG_SHIFT + // fail if scratch != 0 + // + // Note: src and dest can be the same register + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(src.valueReg() != scratch); + mov(ImmWord(JSVAL_TYPE_TO_SHIFTED_TAG(type)), scratch); + xor_(dest, src.valueReg(), scratch); + srli(scratch, dest, JSVAL_TAG_SHIFT); + ma_b(scratch, Imm32(0), fail, Assembler::NotEqual); +} + +void MacroAssembler::fallibleUnboxPtr(const Address& src, Register dest, + JSValueType type, Label* fail) { + loadValue(src, ValueOperand(dest)); + fallibleUnboxPtr(ValueOperand(dest), dest, type, fail); +} + +void MacroAssembler::fallibleUnboxPtr(const BaseIndex& src, Register dest, + JSValueType type, Label* fail) { + loadValue(src, ValueOperand(dest)); + fallibleUnboxPtr(ValueOperand(dest), dest, type, fail); +} +void MacroAssembler::flexibleLshift32(Register src, Register dest) { + lshift32(src, dest); +} +void MacroAssembler::flexibleRshift32Arithmetic(Register src, Register dest) { + rshift32Arithmetic(src, dest); +} +void MacroAssembler::flexibleRshift32(Register src, Register dest) { + rshift32(src, dest); +} +void MacroAssembler::inc64(AbsoluteAddress dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register scratch2 = temps.Acquire(); + ma_li(scratch, ImmWord(uintptr_t(dest.addr))); + ld(scratch2, scratch, 0); + addi(scratch2, scratch2, 1); + sd(scratch2, scratch, 0); +} + +void MacroAssembler::load32SignExtendToPtr(const Address& src, Register dest) { + load32(src, dest); +} +void MacroAssembler::loadAbiReturnAddress(Register dest) { movePtr(ra, dest); } + +void MacroAssembler::lshift32(Register src, Register dest) { + sllw(dest, dest, src); +} + +void MacroAssembler::lshift32(Imm32 imm, Register dest) { + slliw(dest, dest, imm.value % 32); +} +void MacroAssembler::lshift64(Register shift, Register64 dest) { + sll(dest.reg, dest.reg, shift); +} + +void MacroAssembler::lshift64(Imm32 imm, Register64 dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + slli(dest.reg, dest.reg, imm.value); +} +void MacroAssembler::lshiftPtr(Register shift, Register dest) { + sll(dest, dest, shift); +} + +void MacroAssembler::lshiftPtr(Imm32 imm, Register dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + slli(dest, dest, imm.value); +} +void MacroAssembler::maxDouble(FloatRegister other, FloatRegister srcDest, + bool handleNaN) { + Float64Max(srcDest, srcDest, other); +} +void MacroAssembler::maxFloat32(FloatRegister other, FloatRegister srcDest, + bool handleNaN) { + Float32Max(srcDest, srcDest, other); +} +void MacroAssembler::memoryBarrier(MemoryBarrierBits barrier) { + if (barrier) { + sync(); + } +} +void MacroAssembler::minDouble(FloatRegister other, FloatRegister srcDest, + bool handleNaN) { + Float64Min(srcDest, srcDest, other); +} +void MacroAssembler::minFloat32(FloatRegister other, FloatRegister srcDest, + bool handleNaN) { + Float32Min(srcDest, srcDest, other); +} +void MacroAssembler::move16SignExtend(Register src, Register dest) { + slli(dest, src, xlen - 16); + srai(dest, dest, xlen - 16); +} +void MacroAssembler::move16To64SignExtend(Register src, Register64 dest) { + move32To64SignExtend(src, dest); + move16SignExtend(dest.reg, dest.reg); +} +void MacroAssembler::move32SignExtendToPtr(Register src, Register dest) { + slliw(dest, src, 0); +} +void MacroAssembler::move32To64SignExtend(Register src, Register64 dest) { + slliw(dest.reg, src, 0); +} +void MacroAssembler::move32To64ZeroExtend(Register src, Register64 dest) { + slli(dest.reg, src, 32); + srli(dest.reg, dest.reg, 32); +} +void MacroAssembler::move32ZeroExtendToPtr(Register src, Register dest) { + slli(dest, src, 32); + srli(dest, dest, 32); +} +void MacroAssembler::move64(Register64 src, Register64 dest) { + movePtr(src.reg, dest.reg); +} + +void MacroAssembler::move64(Imm64 imm, Register64 dest) { + movePtr(ImmWord(imm.value), dest.reg); +} + +void MacroAssembler::move64To32(Register64 src, Register dest) { + slliw(dest, src.reg, 0); +} + +void MacroAssembler::move8ZeroExtend(Register src, Register dest) { + MOZ_CRASH("NYI"); +} + +void MacroAssembler::move8SignExtend(Register src, Register dest) { + slli(dest, src, xlen - 8); + srai(dest, dest, xlen - 8); +} +void MacroAssembler::move8To64SignExtend(Register src, Register64 dest) { + move32To64SignExtend(src, dest); + move8SignExtend(dest.reg, dest.reg); +} +void MacroAssembler::moveDoubleToGPR64(FloatRegister src, Register64 dest) { + fmv_x_d(dest.reg, src); +} + +void MacroAssembler::moveGPR64ToDouble(Register64 src, FloatRegister dest) { + fmv_d_x(dest, src.reg); +} +void MacroAssembler::moveFloat32ToGPR(FloatRegister src, Register dest) { + fmv_x_w(dest, src); +} +void MacroAssembler::moveGPRToFloat32(Register src, FloatRegister dest) { + fmv_w_x(dest, src); +} +void MacroAssembler::mul32(Register rhs, Register srcDest) { + mulw(srcDest, srcDest, rhs); +} + +void MacroAssembler::mul32(Imm32 imm, Register srcDest) { + ScratchRegisterScope scratch(asMasm()); + move32(imm, scratch); + mul32(scratch, srcDest); +} + +void MacroAssembler::mulHighUnsigned32(Imm32 imm, Register src, Register dest) { + ScratchRegisterScope scratch(asMasm()); + ma_li(scratch, uint32_t(imm.value)); + mul(dest, src, scratch); + srli(dest, dest, 32); +} + +void MacroAssembler::mul64(Imm64 imm, const Register64& dest) { + ScratchRegisterScope scratch(asMasm()); + MOZ_ASSERT(dest.reg != scratch); + mov(ImmWord(imm.value), scratch); + mul(dest.reg, dest.reg, scratch); +} + +void MacroAssembler::mul64(Imm64 imm, const Register64& dest, + const Register temp) { + MOZ_ASSERT(temp == Register::Invalid()); + mul64(imm, dest); +} + +void MacroAssembler::mul64(const Register64& src, const Register64& dest, + const Register temp) { + MOZ_ASSERT(temp == Register::Invalid()); + mul(dest.reg, dest.reg, src.reg); +} + +void MacroAssembler::mul64(const Operand& src, const Register64& dest, + const Register temp) { + if (src.is_mem()) { + ScratchRegisterScope scratch(asMasm()); + Register64 scratch64(scratch); + + load64(src.toAddress(), scratch64); + mul64(scratch64, dest, temp); + } else { + mul64(Register64(src.toReg()), dest, temp); + } +} +void MacroAssembler::mulBy3(Register src, Register dest) { + ScratchRegisterScope scratch(asMasm()); + MOZ_ASSERT(src != scratch); + add(scratch, src, src); + add(dest, scratch, src); +} +void MacroAssembler::mulDouble(FloatRegister src, FloatRegister dest) { + fmul_d(dest, dest, src); +} +void MacroAssembler::mulDoublePtr(ImmPtr imm, Register temp, + FloatRegister dest) { + ScratchRegisterScope scratch(asMasm()); + ScratchDoubleScope fpscratch(asMasm()); + movePtr(imm, scratch); + loadDouble(Address(scratch, 0), fpscratch); + mulDouble(fpscratch, dest); +} +void MacroAssembler::mulFloat32(FloatRegister src, FloatRegister dest) { + fmul_s(dest, dest, src); +} +void MacroAssembler::mulPtr(Register rhs, Register srcDest) { + mul(srcDest, srcDest, rhs); +} + +void MacroAssembler::negateDouble(FloatRegister reg) { fneg_d(reg, reg); } + +void MacroAssembler::negateFloat(FloatRegister reg) { fneg_s(reg, reg); } + +void MacroAssembler::neg64(Register64 reg) { sub(reg.reg, zero, reg.reg); } + +void MacroAssembler::negPtr(Register reg) { sub(reg, zero, reg); } + +void MacroAssembler::neg32(Register reg) { subw(reg, zero, reg); } +void MacroAssembler::not32(Register reg) { nor(reg, reg, zero); } + +void MacroAssembler::notPtr(Register reg) { nor(reg, reg, zero); } + +void MacroAssembler::or32(Register src, Register dest) { + ma_or(dest, dest, src); +} + +void MacroAssembler::or32(Imm32 imm, Register dest) { ma_or(dest, dest, imm); } + +void MacroAssembler::or32(Imm32 imm, const Address& dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(dest, scratch2); + ma_or(scratch2, imm); + store32(scratch2, dest); +} + +void MacroAssembler::or64(Register64 src, Register64 dest) { + ma_or(dest.reg, dest.reg, src.reg); +} + +void MacroAssembler::or64(const Operand& src, Register64 dest) { + if (src.is_mem()) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register64 scratch64(scratch); + + load64(src.toAddress(), scratch64); + or64(scratch64, dest); + } else { + or64(Register64(src.toReg()), dest); + } +} +void MacroAssembler::or64(Imm64 imm, Register64 dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, ImmWord(imm.value)); + ma_or(dest.reg, dest.reg, scratch); +} + +void MacroAssembler::orPtr(Register src, Register dest) { + ma_or(dest, dest, src); +} + +void MacroAssembler::orPtr(Imm32 imm, Register dest) { ma_or(dest, dest, imm); } + +void MacroAssembler::patchSub32FromStackPtr(CodeOffset offset, Imm32 imm) { + DEBUG_PRINTF("patchSub32FromStackPtr at offset %lu with immediate %d\n", + offset.offset(), imm.value); + Instruction* inst0 = + (Instruction*)m_buffer.getInst(BufferOffset(offset.offset())); + Instruction* inst1 = (Instruction*)(inst0 + 4); + MOZ_ASSERT(IsLui(*reinterpret_cast<Instr*>(inst0))); + MOZ_ASSERT(IsAddi(*reinterpret_cast<Instr*>(inst1))); + + int64_t value = imm.value; + int64_t high_20 = ((value + 0x800) >> 12); + int64_t low_12 = value << 52 >> 52; + + uint32_t* p = reinterpret_cast<uint32_t*>(inst0); + + (*p) = (*p) & 0xfff; + (*p) = (*p) | ((int32_t)high_20 << 12); + + *(p + 1) = *(p + 1) & 0xfffff; + *(p + 1) = *(p + 1) | ((int32_t)low_12 << 20); + disassembleInstr(inst0->InstructionBits()); + disassembleInstr(inst1->InstructionBits()); + MOZ_ASSERT((int32_t)(inst0->Imm20UValue() << kImm20Shift) + + (int32_t)(inst1->Imm12Value()) == + imm.value); +} + +void MacroAssembler::popcnt32(Register input, Register output, Register tmp) { + Popcnt32(output, input, tmp); +} +void MacroAssembler::popcnt64(Register64 input, Register64 output, + Register tmp) { + Popcnt64(output.reg, input.reg, tmp); +} +void MacroAssembler::quotient32(Register rhs, Register srcDest, + bool isUnsigned) { + if (isUnsigned) { + ma_divu32(srcDest, srcDest, rhs); + } else { + ma_div32(srcDest, srcDest, rhs); + } +} + +void MacroAssembler::remainder32(Register rhs, Register srcDest, + bool isUnsigned) { + if (isUnsigned) { + ma_modu32(srcDest, srcDest, rhs); + } else { + ma_mod32(srcDest, srcDest, rhs); + } +} +void MacroAssembler::rotateLeft64(Imm32 count, Register64 src, Register64 dest, + Register temp) { + Dror(dest.reg, src.reg, Operand(64 - (count.value % 64))); +} +void MacroAssembler::rotateLeft64(Register count, Register64 src, + Register64 dest, Register temp) { + ScratchRegisterScope scratch(asMasm()); + ma_mod32(scratch, count, Operand(64)); + negw(scratch, scratch); + addi(scratch, scratch, 64); + Dror(dest.reg, src.reg, Operand(scratch)); +} + +void MacroAssembler::rotateLeft(Imm32 count, Register input, Register dest) { + JitSpew(JitSpew_Codegen, "[ rotateLeft\n"); + Ror(dest, input, Operand(32 - (count.value % 32))); + JitSpew(JitSpew_Codegen, "]\n"); +} +void MacroAssembler::rotateLeft(Register count, Register input, Register dest) { + JitSpew(JitSpew_Codegen, "[ rotateLeft\n"); + ScratchRegisterScope scratch(asMasm()); + ma_mod32(scratch, count, Operand(32)); + negw(scratch, scratch); + addi(scratch, scratch, 32); + Ror(dest, input, Operand(scratch)); + JitSpew(JitSpew_Codegen, "]\n"); +} +void MacroAssembler::rotateRight64(Register count, Register64 src, + Register64 dest, Register temp) { + Dror(dest.reg, src.reg, Operand(count)); +} +void MacroAssembler::rotateRight64(Imm32 count, Register64 src, Register64 dest, + Register temp) { + Dror(dest.reg, src.reg, Operand(count.value)); +} +void MacroAssembler::rotateRight(Imm32 count, Register input, Register dest) { + Ror(dest, input, Operand(count.value)); +} +void MacroAssembler::rotateRight(Register count, Register input, + Register dest) { + Ror(dest, input, Operand(count)); +} +void MacroAssembler::rshift32Arithmetic(Register src, Register dest) { + sraw(dest, dest, src); +} + +void MacroAssembler::rshift32Arithmetic(Imm32 imm, Register dest) { + sraiw(dest, dest, imm.value % 32); +} +void MacroAssembler::rshift32(Register src, Register dest) { + srlw(dest, dest, src); +} + +void MacroAssembler::rshift32(Imm32 imm, Register dest) { + srliw(dest, dest, imm.value % 32); +} + +void MacroAssembler::rshift64Arithmetic(Imm32 imm, Register64 dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + srai(dest.reg, dest.reg, imm.value); +} + +void MacroAssembler::rshift64Arithmetic(Register shift, Register64 dest) { + sra(dest.reg, dest.reg, shift); +} + +void MacroAssembler::rshift64(Register shift, Register64 dest) { + srl(dest.reg, dest.reg, shift); +} + +void MacroAssembler::rshift64(Imm32 imm, Register64 dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + srli(dest.reg, dest.reg, imm.value); +} + +void MacroAssembler::rshiftPtrArithmetic(Imm32 imm, Register dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + srai(dest, dest, imm.value); +} +void MacroAssembler::rshiftPtr(Register shift, Register dest) { + srl(dest, dest, shift); +} + +void MacroAssembler::rshiftPtr(Imm32 imm, Register dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + srli(dest, dest, imm.value); +} +void MacroAssembler::spectreBoundsCheck32(Register index, Register length, + Register maybeScratch, + Label* failure) { + MOZ_RELEASE_ASSERT(!JitOptions.spectreIndexMasking); + branch32(Assembler::BelowOrEqual, length, index, failure); +} + +void MacroAssembler::spectreBoundsCheck32(Register index, const Address& length, + Register maybeScratch, + Label* failure) { + MOZ_RELEASE_ASSERT(!JitOptions.spectreIndexMasking); + branch32(Assembler::BelowOrEqual, length, index, failure); +} +void MacroAssembler::spectreBoundsCheckPtr(Register index, Register length, + Register maybeScratch, + Label* failure) { + MOZ_RELEASE_ASSERT(!JitOptions.spectreIndexMasking); + branchPtr(Assembler::BelowOrEqual, length, index, failure); +} + +void MacroAssembler::spectreBoundsCheckPtr(Register index, + const Address& length, + Register maybeScratch, + Label* failure) { + MOZ_RELEASE_ASSERT(!JitOptions.spectreIndexMasking); + branchPtr(Assembler::BelowOrEqual, length, index, failure); +} +void MacroAssembler::spectreMovePtr(Condition, Register, Register) { + MOZ_CRASH("spectreMovePtr"); +} +void MacroAssembler::spectreZeroRegister(Condition cond, Register scratch, + Register dest) { + MOZ_CRASH("spectreZeroRegister"); +} +void MacroAssembler::sqrtDouble(FloatRegister src, FloatRegister dest) { + fsqrt_d(dest, src); +} +void MacroAssembler::sqrtFloat32(FloatRegister src, FloatRegister dest) { + fsqrt_s(dest, src); +} +FaultingCodeOffset MacroAssembler::storeUncanonicalizedFloat32( + FloatRegister src, const Address& addr) { + return ma_fst_s(src, addr); +} +FaultingCodeOffset MacroAssembler::storeUncanonicalizedFloat32( + FloatRegister src, const BaseIndex& addr) { + return ma_fst_s(src, addr); +} + +FaultingCodeOffset MacroAssembler::storeUncanonicalizedDouble( + FloatRegister src, const Address& addr) { + return ma_fst_d(src, addr); +} +FaultingCodeOffset MacroAssembler::storeUncanonicalizedDouble( + FloatRegister src, const BaseIndex& addr) { + return ma_fst_d(src, addr); +} +void MacroAssembler::sub32(Register src, Register dest) { + subw(dest, dest, src); +} + +void MacroAssembler::sub32(Imm32 imm, Register dest) { + ma_sub32(dest, dest, imm); +} + +void MacroAssembler::sub32(const Address& src, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + load32(src, scratch); + subw(dest, dest, scratch); +} + +void MacroAssembler::sub64(Register64 src, Register64 dest) { + sub(dest.reg, dest.reg, src.reg); +} + +void MacroAssembler::sub64(const Operand& src, Register64 dest) { + if (src.is_mem()) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register64 scratch64(scratch); + + load64(src.toAddress(), scratch64); + sub64(scratch64, dest); + } else { + sub64(Register64(src.toReg()), dest); + } +} + +void MacroAssembler::sub64(Imm64 imm, Register64 dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(dest.reg != scratch); + ma_li(scratch, ImmWord(imm.value)); + sub(dest.reg, dest.reg, scratch); +} + +void MacroAssembler::subDouble(FloatRegister src, FloatRegister dest) { + fsub_d(dest, dest, src); +} + +void MacroAssembler::subFloat32(FloatRegister src, FloatRegister dest) { + fsub_s(dest, dest, src); +} + +void MacroAssembler::subPtr(Register src, const Address& dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + loadPtr(dest, scratch); + subPtr(src, scratch); + storePtr(scratch, dest); +} + +void MacroAssembler::subPtr(const Address& addr, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + loadPtr(addr, scratch); + subPtr(scratch, dest); +} +void MacroAssembler::subPtr(Imm32 imm, Register dest) { + ma_sub64(dest, dest, imm); +} +void MacroAssembler::subPtr(Register src, Register dest) { + sub(dest, dest, src); +} +void MacroAssembler::test32LoadPtr(Condition cond, const Address& addr, + Imm32 mask, const Address& src, + Register dest) { + MOZ_RELEASE_ASSERT(!JitOptions.spectreStringMitigations); + Label skip; + branchTest32(Assembler::InvertCondition(cond), addr, mask, &skip); + loadPtr(src, dest); + bind(&skip); +} +void MacroAssembler::test32MovePtr(Condition, const Address&, Imm32, Register, + Register) { + MOZ_CRASH(); +} +void MacroAssembler::xor32(Register src, Register dest) { + ma_xor(dest, dest, src); +} + +void MacroAssembler::xor32(Imm32 imm, Register dest) { + ma_xor(dest, dest, imm); +} + +void MacroAssembler::xor32(Imm32 imm, const Address& dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(dest, scratch2); + xor32(imm, scratch2); + store32(scratch2, dest); +} + +void MacroAssembler::xor32(const Address& src, Register dest) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(src, scratch2); + xor32(scratch2, dest); +} +void MacroAssembler::xor64(Register64 src, Register64 dest) { + ma_xor(dest.reg, dest.reg, src.reg); +} + +void MacroAssembler::xor64(const Operand& src, Register64 dest) { + if (src.is_mem()) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register64 scratch64(scratch); + + load64(src.toAddress(), scratch64); + xor64(scratch64, dest); + } else { + xor64(Register64(src.toReg()), dest); + } +} +void MacroAssembler::xor64(Imm64 imm, Register64 dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, ImmWord(imm.value)); + ma_xor(dest.reg, dest.reg, scratch); +} +void MacroAssembler::xorPtr(Register src, Register dest) { + ma_xor(dest, dest, src); +} + +void MacroAssembler::xorPtr(Imm32 imm, Register dest) { + ma_xor(dest, dest, imm); +} +//}}} check_macroassembler_style + +void MacroAssemblerRiscv64Compat::incrementInt32Value(const Address& addr) { + asMasm().add32(Imm32(1), addr); +} + +void MacroAssemblerRiscv64Compat::retn(Imm32 n) { + // pc <- [sp]; sp += n + loadPtr(Address(StackPointer, 0), ra); + asMasm().addPtr(n, StackPointer); + jr(ra, 0); +} +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_MacroAssembler_riscv64_inl_h */ diff --git a/js/src/jit/riscv64/MacroAssembler-riscv64.cpp b/js/src/jit/riscv64/MacroAssembler-riscv64.cpp new file mode 100644 index 0000000000..93ccf1cc27 --- /dev/null +++ b/js/src/jit/riscv64/MacroAssembler-riscv64.cpp @@ -0,0 +1,6631 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +// Copyright 2021 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/MacroAssembler-riscv64.h" + +#include "jsmath.h" +#include "jit/Bailouts.h" +#include "jit/BaselineFrame.h" +#include "jit/JitFrames.h" +#include "jit/JitRuntime.h" +#include "jit/MacroAssembler.h" +#include "jit/MoveEmitter.h" +#include "jit/riscv64/SharedICRegisters-riscv64.h" +#include "util/Memory.h" +#include "vm/JitActivation.h" // jit::JitActivation +#include "vm/JSContext.h" + +#include "jit/MacroAssembler-inl.h" + +namespace js { +namespace jit { + +MacroAssembler& MacroAssemblerRiscv64::asMasm() { + return *static_cast<MacroAssembler*>(this); +} + +const MacroAssembler& MacroAssemblerRiscv64::asMasm() const { + return *static_cast<const MacroAssembler*>(this); +} + +void MacroAssemblerRiscv64::ma_cmp_set(Register rd, Register rj, ImmWord imm, + Condition c) { + if (imm.value <= INT32_MAX) { + ma_cmp_set(rd, rj, Imm32(uint32_t(imm.value)), c); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + ma_cmp_set(rd, rj, scratch, c); + } +} + +void MacroAssemblerRiscv64::ma_cmp_set(Register rd, Register rj, ImmPtr imm, + Condition c) { + ma_cmp_set(rd, rj, ImmWord(uintptr_t(imm.value)), c); +} + +void MacroAssemblerRiscv64::ma_cmp_set(Register rd, Address address, Imm32 imm, + Condition c) { + // TODO(riscv): 32-bit ma_cmp_set? + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + ma_load(scratch2, address, SizeWord); + ma_cmp_set(rd, Register(scratch2), imm, c); +} + +void MacroAssemblerRiscv64::ma_cmp_set(Register rd, Address address, + ImmWord imm, Condition c) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + ma_load(scratch2, address, SizeDouble); + ma_cmp_set(rd, Register(scratch2), imm, c); +} + +void MacroAssemblerRiscv64::ma_cmp_set(Register rd, Register rj, Imm32 imm, + Condition c) { + if (imm.value == 0) { + switch (c) { + case Equal: + case BelowOrEqual: + ma_sltu(rd, rj, Operand(1)); + break; + case NotEqual: + case Above: + sltu(rd, zero, rj); + break; + case AboveOrEqual: + case Below: + ori(rd, zero, c == AboveOrEqual ? 1 : 0); + break; + case GreaterThan: + case LessThanOrEqual: + slt(rd, zero, rj); + if (c == LessThanOrEqual) { + xori(rd, rd, 1); + } + break; + case LessThan: + case GreaterThanOrEqual: + slt(rd, rj, zero); + if (c == GreaterThanOrEqual) { + xori(rd, rd, 1); + } + break; + case Zero: + ma_sltu(rd, rj, Operand(1)); + break; + case NonZero: + sltu(rd, zero, rj); + break; + case Signed: + slt(rd, rj, zero); + break; + case NotSigned: + slt(rd, rj, zero); + xori(rd, rd, 1); + break; + default: + MOZ_CRASH("Invalid condition."); + } + return; + } + + switch (c) { + case Equal: + case NotEqual: + ma_xor(rd, rj, imm); + if (c == Equal) { + ma_sltu(rd, rd, Operand(1)); + } else { + sltu(rd, zero, rd); + } + break; + case Zero: + case NonZero: + case Signed: + case NotSigned: + MOZ_CRASH("Invalid condition."); + default: + Condition cond = ma_cmp(rd, rj, imm, c); + MOZ_ASSERT(cond == Equal || cond == NotEqual); + + if (cond == Equal) xori(rd, rd, 1); + } +} + +Assembler::Condition MacroAssemblerRiscv64::ma_cmp(Register dest, Register lhs, + Register rhs, Condition c) { + switch (c) { + case Above: + // bgtu s,t,label => + // sltu at,t,s + // bne at,$zero,offs + sltu(dest, rhs, lhs); + return NotEqual; + case AboveOrEqual: + // bgeu s,t,label => + // sltu at,s,t + // beq at,$zero,offs + sltu(dest, lhs, rhs); + return Equal; + case Below: + // bltu s,t,label => + // sltu at,s,t + // bne at,$zero,offs + sltu(dest, lhs, rhs); + return NotEqual; + case BelowOrEqual: + // bleu s,t,label => + // sltu at,t,s + // beq at,$zero,offs + sltu(dest, rhs, lhs); + return Equal; + case GreaterThan: + // bgt s,t,label => + // slt at,t,s + // bne at,$zero,offs + slt(dest, rhs, lhs); + return NotEqual; + case GreaterThanOrEqual: + // bge s,t,label => + // slt at,s,t + // beq at,$zero,offs + slt(dest, lhs, rhs); + return Equal; + case LessThan: + // blt s,t,label => + // slt at,s,t + // bne at,$zero,offs + slt(dest, lhs, rhs); + return NotEqual; + case LessThanOrEqual: + // ble s,t,label => + // slt at,t,s + // beq at,$zero,offs + slt(dest, rhs, lhs); + return Equal; + default: + MOZ_CRASH("Invalid condition."); + } + return Always; +} + +Assembler::Condition MacroAssemblerRiscv64::ma_cmp(Register dest, Register lhs, + Imm32 imm, Condition c) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_RELEASE_ASSERT(lhs != scratch); + + switch (c) { + case Above: + case BelowOrEqual: + if (imm.value != 0x7fffffff && is_intn(imm.value + 1, 12) && + imm.value != -1) { + // lhs <= rhs via lhs < rhs + 1 if rhs + 1 does not overflow + ma_sltu(dest, lhs, Operand(imm.value + 1)); + + return (c == BelowOrEqual ? NotEqual : Equal); + } else { + ma_li(scratch, imm); + sltu(dest, scratch, lhs); + return (c == BelowOrEqual ? Equal : NotEqual); + } + case AboveOrEqual: + case Below: + if (is_intn(imm.value, 12)) { + ma_sltu(dest, lhs, Operand(imm.value)); + } else { + ma_li(scratch, imm); + sltu(dest, lhs, scratch); + } + return (c == AboveOrEqual ? Equal : NotEqual); + case GreaterThan: + case LessThanOrEqual: + if (imm.value != 0x7fffffff && is_intn(imm.value + 1, 12)) { + // lhs <= rhs via lhs < rhs + 1. + ma_slt(dest, lhs, Operand(imm.value + 1)); + return (c == LessThanOrEqual ? NotEqual : Equal); + } else { + ma_li(scratch, imm); + slt(dest, scratch, lhs); + return (c == LessThanOrEqual ? Equal : NotEqual); + } + case GreaterThanOrEqual: + case LessThan: + if (is_intn(imm.value, 12)) { + ma_slt(dest, lhs, imm); + } else { + ma_li(scratch, imm); + slt(dest, lhs, scratch); + } + return (c == GreaterThanOrEqual ? Equal : NotEqual); + default: + MOZ_CRASH("Invalid condition."); + } + return Always; +} + +void MacroAssemblerRiscv64::ma_cmp_set(Register rd, Register rj, Register rk, + Condition c) { + switch (c) { + case Equal: + // seq d,s,t => + // xor d,s,t + // sltiu d,d,1 + xor_(rd, rj, rk); + ma_sltu(rd, rd, Operand(1)); + break; + case NotEqual: + // sne d,s,t => + // xor d,s,t + // sltu d,$zero,d + xor_(rd, rj, rk); + sltu(rd, zero, rd); + break; + case Above: + // sgtu d,s,t => + // sltu d,t,s + sltu(rd, rk, rj); + break; + case AboveOrEqual: + // sgeu d,s,t => + // sltu d,s,t + // xori d,d,1 + sltu(rd, rj, rk); + xori(rd, rd, 1); + break; + case Below: + // sltu d,s,t + sltu(rd, rj, rk); + break; + case BelowOrEqual: + // sleu d,s,t => + // sltu d,t,s + // xori d,d,1 + sltu(rd, rk, rj); + xori(rd, rd, 1); + break; + case GreaterThan: + // sgt d,s,t => + // slt d,t,s + slt(rd, rk, rj); + break; + case GreaterThanOrEqual: + // sge d,s,t => + // slt d,s,t + // xori d,d,1 + slt(rd, rj, rk); + xori(rd, rd, 1); + break; + case LessThan: + // slt d,s,t + slt(rd, rj, rk); + break; + case LessThanOrEqual: + // sle d,s,t => + // slt d,t,s + // xori d,d,1 + slt(rd, rk, rj); + xori(rd, rd, 1); + break; + case Zero: + MOZ_ASSERT(rj == rk); + // seq d,s,$zero => + // sltiu d,s,1 + ma_sltu(rd, rj, Operand(1)); + break; + case NonZero: + MOZ_ASSERT(rj == rk); + // sne d,s,$zero => + // sltu d,$zero,s + sltu(rd, zero, rj); + break; + case Signed: + MOZ_ASSERT(rj == rk); + slt(rd, rj, zero); + break; + case NotSigned: + MOZ_ASSERT(rj == rk); + // sge d,s,$zero => + // slt d,s,$zero + // xori d,d,1 + slt(rd, rj, zero); + xori(rd, rd, 1); + break; + default: + MOZ_CRASH("Invalid condition."); + } +} + +void MacroAssemblerRiscv64::ma_compareF32(Register rd, DoubleCondition cc, + FloatRegister cmp1, + FloatRegister cmp2) { + switch (cc) { + case DoubleEqualOrUnordered: + case DoubleEqual: + feq_s(rd, cmp1, cmp2); + break; + case DoubleNotEqualOrUnordered: + case DoubleNotEqual: { + Label done; + CompareIsNanF32(rd, cmp1, cmp2); + ma_branch(&done, Equal, rd, Operand(1)); + feq_s(rd, cmp1, cmp2); + bind(&done); + NegateBool(rd, rd); + break; + } + case DoubleLessThanOrUnordered: + case DoubleLessThan: + flt_s(rd, cmp1, cmp2); + break; + case DoubleGreaterThanOrEqualOrUnordered: + case DoubleGreaterThanOrEqual: + fle_s(rd, cmp2, cmp1); + break; + case DoubleLessThanOrEqualOrUnordered: + case DoubleLessThanOrEqual: + fle_s(rd, cmp1, cmp2); + break; + case DoubleGreaterThanOrUnordered: + case DoubleGreaterThan: + flt_s(rd, cmp2, cmp1); + break; + case DoubleOrdered: + CompareIsNotNanF32(rd, cmp1, cmp2); + return; + case DoubleUnordered: + CompareIsNanF32(rd, cmp1, cmp2); + return; + } + if (cc >= FIRST_UNORDERED && cc <= LAST_UNORDERED) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + CompareIsNanF32(scratch, cmp1, cmp2); + or_(rd, rd, scratch); + } +} + +void MacroAssemblerRiscv64::ma_compareF64(Register rd, DoubleCondition cc, + FloatRegister cmp1, + FloatRegister cmp2) { + switch (cc) { + case DoubleEqualOrUnordered: + case DoubleEqual: + feq_d(rd, cmp1, cmp2); + break; + case DoubleNotEqualOrUnordered: + case DoubleNotEqual: { + Label done; + CompareIsNanF64(rd, cmp1, cmp2); + ma_branch(&done, Equal, rd, Operand(1)); + feq_d(rd, cmp1, cmp2); + bind(&done); + NegateBool(rd, rd); + } break; + case DoubleLessThanOrUnordered: + case DoubleLessThan: + flt_d(rd, cmp1, cmp2); + break; + case DoubleGreaterThanOrEqualOrUnordered: + case DoubleGreaterThanOrEqual: + fle_d(rd, cmp2, cmp1); + break; + case DoubleLessThanOrEqualOrUnordered: + case DoubleLessThanOrEqual: + fle_d(rd, cmp1, cmp2); + break; + case DoubleGreaterThanOrUnordered: + case DoubleGreaterThan: + flt_d(rd, cmp2, cmp1); + break; + case DoubleOrdered: + CompareIsNotNanF64(rd, cmp1, cmp2); + return; + case DoubleUnordered: + CompareIsNanF64(rd, cmp1, cmp2); + return; + } + + if (cc >= FIRST_UNORDERED && cc <= LAST_UNORDERED) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + CompareIsNanF64(scratch, cmp1, cmp2); + or_(rd, rd, scratch); + } +} + +void MacroAssemblerRiscv64Compat::movePtr(Register src, Register dest) { + mv(dest, src); +} +void MacroAssemblerRiscv64Compat::movePtr(ImmWord imm, Register dest) { + ma_li(dest, imm); +} + +void MacroAssemblerRiscv64Compat::movePtr(ImmGCPtr imm, Register dest) { + ma_li(dest, imm); +} + +void MacroAssemblerRiscv64Compat::movePtr(ImmPtr imm, Register dest) { + movePtr(ImmWord(uintptr_t(imm.value)), dest); +} +void MacroAssemblerRiscv64Compat::movePtr(wasm::SymbolicAddress imm, + Register dest) { + DEBUG_PRINTF("[ %s\n", __FUNCTION__); + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + append(wasm::SymbolicAccess(CodeOffset(nextOffset().getOffset()), imm)); + ma_liPatchable(dest, ImmWord(-1), Li64); + DEBUG_PRINTF("]\n"); +} + +bool MacroAssemblerRiscv64Compat::buildOOLFakeExitFrame(void* fakeReturnAddr) { + asMasm().PushFrameDescriptor(FrameType::IonJS); // descriptor_ + asMasm().Push(ImmPtr(fakeReturnAddr)); + asMasm().Push(FramePointer); + return true; +} + +void MacroAssemblerRiscv64Compat::convertUInt32ToDouble(Register src, + FloatRegister dest) { + fcvt_d_wu(dest, src); +} + +void MacroAssemblerRiscv64Compat::convertUInt64ToDouble(Register src, + FloatRegister dest) { + fcvt_d_lu(dest, src); +} + +void MacroAssemblerRiscv64Compat::convertUInt32ToFloat32(Register src, + FloatRegister dest) { + fcvt_s_wu(dest, src); +} + +void MacroAssemblerRiscv64Compat::convertDoubleToFloat32(FloatRegister src, + FloatRegister dest) { + fcvt_s_d(dest, src); +} + +template <typename F> +void MacroAssemblerRiscv64::RoundHelper(FPURegister dst, FPURegister src, + FPURegister fpu_scratch, + FPURoundingMode frm) { + BlockTrampolinePoolScope block_trampoline_pool(this, 20); + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + + MOZ_ASSERT((std::is_same<float, F>::value) || + (std::is_same<double, F>::value)); + // Need at least two FPRs, so check against dst == src == fpu_scratch + MOZ_ASSERT(!(dst == src && dst == fpu_scratch)); + + const int kFloatMantissaBits = + sizeof(F) == 4 ? kFloat32MantissaBits : kFloat64MantissaBits; + const int kFloatExponentBits = + sizeof(F) == 4 ? kFloat32ExponentBits : kFloat64ExponentBits; + const int kFloatExponentBias = + sizeof(F) == 4 ? kFloat32ExponentBias : kFloat64ExponentBias; + Label done; + + { + UseScratchRegisterScope temps2(this); + Register scratch = temps2.Acquire(); + // extract exponent value of the source floating-point to scratch + if (std::is_same<F, double>::value) { + fmv_x_d(scratch, src); + } else { + fmv_x_w(scratch, src); + } + ExtractBits(scratch2, scratch, kFloatMantissaBits, kFloatExponentBits); + } + + // if src is NaN/+-Infinity/+-Zero or if the exponent is larger than # of bits + // in mantissa, the result is the same as src, so move src to dest (to avoid + // generating another branch) + if (dst != src) { + if (std::is_same<F, double>::value) { + fmv_d(dst, src); + } else { + fmv_s(dst, src); + } + } + { + Label not_NaN; + UseScratchRegisterScope temps2(this); + Register scratch = temps2.Acquire(); + // According to the wasm spec + // (https://webassembly.github.io/spec/core/exec/numerics.html#aux-nans) + // if input is canonical NaN, then output is canonical NaN, and if input is + // any other NaN, then output is any NaN with most significant bit of + // payload is 1. In RISC-V, feq_d will set scratch to 0 if src is a NaN. If + // src is not a NaN, branch to the label and do nothing, but if it is, + // fmin_d will set dst to the canonical NaN. + if (std::is_same<F, double>::value) { + feq_d(scratch, src, src); + bnez(scratch, ¬_NaN); + fmin_d(dst, src, src); + } else { + feq_s(scratch, src, src); + bnez(scratch, ¬_NaN); + fmin_s(dst, src, src); + } + bind(¬_NaN); + } + + // If real exponent (i.e., scratch2 - kFloatExponentBias) is greater than + // kFloat32MantissaBits, it means the floating-point value has no fractional + // part, thus the input is already rounded, jump to done. Note that, NaN and + // Infinity in floating-point representation sets maximal exponent value, so + // they also satisfy (scratch2 - kFloatExponentBias >= kFloatMantissaBits), + // and JS round semantics specify that rounding of NaN (Infinity) returns NaN + // (Infinity), so NaN and Infinity are considered rounded value too. + ma_branch(&done, GreaterThanOrEqual, scratch2, + Operand(kFloatExponentBias + kFloatMantissaBits)); + + // Actual rounding is needed along this path + + // old_src holds the original input, needed for the case of src == dst + FPURegister old_src = src; + if (src == dst) { + MOZ_ASSERT(fpu_scratch != dst); + fmv_d(fpu_scratch, src); + old_src = fpu_scratch; + } + + // Since only input whose real exponent value is less than kMantissaBits + // (i.e., 23 or 52-bits) falls into this path, the value range of the input + // falls into that of 23- or 53-bit integers. So we round the input to integer + // values, then convert them back to floating-point. + { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + if (std::is_same<F, double>::value) { + fcvt_l_d(scratch, src, frm); + fcvt_d_l(dst, scratch, frm); + } else { + fcvt_w_s(scratch, src, frm); + fcvt_s_w(dst, scratch, frm); + } + } + // A special handling is needed if the input is a very small positive/negative + // number that rounds to zero. JS semantics requires that the rounded result + // retains the sign of the input, so a very small positive (negative) + // floating-point number should be rounded to positive (negative) 0. + // Therefore, we use sign-bit injection to produce +/-0 correctly. Instead of + // testing for zero w/ a branch, we just insert sign-bit for everyone on this + // path (this is where old_src is needed) + if (std::is_same<F, double>::value) { + fsgnj_d(dst, dst, old_src); + } else { + fsgnj_s(dst, dst, old_src); + } + + bind(&done); +} + +template <typename CvtFunc> +void MacroAssemblerRiscv64::RoundFloatingPointToInteger(Register rd, + FPURegister fs, + Register result, + CvtFunc fcvt_generator, + bool Inexact) { + // Save csr_fflags to scratch & clear exception flags + if (result != Register::Invalid()) { + BlockTrampolinePoolScope block_trampoline_pool(this, 6); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + + int exception_flags = kInvalidOperation; + if (Inexact) exception_flags |= kInexact; + csrrci(scratch, csr_fflags, exception_flags); + + // actual conversion instruction + fcvt_generator(this, rd, fs); + + // check kInvalidOperation flag (out-of-range, NaN) + // set result to 1 if normal, otherwise set result to 0 for abnormal + frflags(result); + andi(result, result, exception_flags); + seqz(result, result); // result <-- 1 (normal), result <-- 0 (abnormal) + + // restore csr_fflags + csrw(csr_fflags, scratch); + } else { + // actual conversion instruction + fcvt_generator(this, rd, fs); + } +} + +void MacroAssemblerRiscv64::Trunc_uw_d(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_wu_d(dst, src, RTZ); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Trunc_w_d(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_d(dst, src, RTZ); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Trunc_uw_s(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_wu_s(dst, src, RTZ); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Trunc_w_s(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_s(dst, src, RTZ); + }, + Inexact); +} +void MacroAssemblerRiscv64::Trunc_ul_d(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_lu_d(dst, src, RTZ); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Trunc_l_d(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_l_d(dst, src, RTZ); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Trunc_ul_s(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_lu_s(dst, src, RTZ); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Trunc_l_s(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_l_s(dst, src, RTZ); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Floor_d_d(FPURegister dst, FPURegister src, + FPURegister fpu_scratch) { + RoundHelper<double>(dst, src, fpu_scratch, RDN); +} + +void MacroAssemblerRiscv64::Ceil_d_d(FPURegister dst, FPURegister src, + FPURegister fpu_scratch) { + RoundHelper<double>(dst, src, fpu_scratch, RUP); +} + +void MacroAssemblerRiscv64::Trunc_d_d(FPURegister dst, FPURegister src, + FPURegister fpu_scratch) { + RoundHelper<double>(dst, src, fpu_scratch, RTZ); +} + +void MacroAssemblerRiscv64::Round_d_d(FPURegister dst, FPURegister src, + FPURegister fpu_scratch) { + RoundHelper<double>(dst, src, fpu_scratch, RNE); +} + +void MacroAssemblerRiscv64::Floor_s_s(FPURegister dst, FPURegister src, + FPURegister fpu_scratch) { + RoundHelper<float>(dst, src, fpu_scratch, RDN); +} + +void MacroAssemblerRiscv64::Ceil_s_s(FPURegister dst, FPURegister src, + FPURegister fpu_scratch) { + RoundHelper<float>(dst, src, fpu_scratch, RUP); +} + +void MacroAssemblerRiscv64::Trunc_s_s(FPURegister dst, FPURegister src, + FPURegister fpu_scratch) { + RoundHelper<float>(dst, src, fpu_scratch, RTZ); +} + +void MacroAssemblerRiscv64::Round_s_s(FPURegister dst, FPURegister src, + FPURegister fpu_scratch) { + RoundHelper<float>(dst, src, fpu_scratch, RNE); +} + +void MacroAssemblerRiscv64::Round_w_s(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_s(dst, src, RNE); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Round_w_d(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_d(dst, src, RNE); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Ceil_w_s(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_s(dst, src, RUP); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Ceil_w_d(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_d(dst, src, RUP); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Floor_w_s(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_s(dst, src, RDN); + }, + Inexact); +} + +void MacroAssemblerRiscv64::Floor_w_d(Register rd, FPURegister fs, + Register result, bool Inexact) { + RoundFloatingPointToInteger( + rd, fs, result, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_d(dst, src, RDN); + }, + Inexact); +} + +// Checks whether a double is representable as a 32-bit integer. If so, the +// integer is written to the output register. Otherwise, a bailout is taken to +// the given snapshot. This function overwrites the scratch float register. +void MacroAssemblerRiscv64Compat::convertDoubleToInt32(FloatRegister src, + Register dest, + Label* fail, + bool negativeZeroCheck) { + if (negativeZeroCheck) { + fclass_d(dest, src); + ma_b(dest, Imm32(kNegativeZero), fail, Equal); + } + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Trunc_w_d(dest, src, scratch, true); + ma_b(scratch, Imm32(0), fail, Equal); +} + +void MacroAssemblerRiscv64Compat::convertDoubleToPtr(FloatRegister src, + Register dest, Label* fail, + bool negativeZeroCheck) { + if (negativeZeroCheck) { + fclass_d(dest, src); + ma_b(dest, Imm32(kNegativeZero), fail, Equal); + } + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Trunc_l_d(dest, src, scratch, true); + ma_b(scratch, Imm32(0), fail, Equal); +} + +// Checks whether a float32 is representable as a 32-bit integer. If so, the +// integer is written to the output register. Otherwise, a bailout is taken to +// the given snapshot. This function overwrites the scratch float register. +void MacroAssemblerRiscv64Compat::convertFloat32ToInt32( + FloatRegister src, Register dest, Label* fail, bool negativeZeroCheck) { + if (negativeZeroCheck) { + fclass_d(dest, src); + ma_b(dest, Imm32(kNegativeZero), fail, Equal); + } + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Trunc_w_s(dest, src, scratch, true); + ma_b(scratch, Imm32(0), fail, Equal); +} + +void MacroAssemblerRiscv64Compat::convertFloat32ToDouble(FloatRegister src, + FloatRegister dest) { + fcvt_d_s(dest, src); +} + +void MacroAssemblerRiscv64Compat::convertInt32ToFloat32(Register src, + FloatRegister dest) { + fcvt_s_w(dest, src); +} + +void MacroAssemblerRiscv64Compat::convertInt32ToFloat32(const Address& src, + FloatRegister dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + load32(src, scratch); + fcvt_s_w(dest, scratch); +} + +void MacroAssemblerRiscv64Compat::movq(Register rj, Register rd) { mv(rd, rj); } + +// Memory. +FaultingCodeOffset MacroAssemblerRiscv64::ma_loadDouble(FloatRegister dest, + Address address) { + int16_t encodedOffset; + Register base; + + if (!is_int12(address.offset)) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, Imm32(address.offset)); + add(ScratchRegister, address.base, ScratchRegister); + base = ScratchRegister; + encodedOffset = 0; + } else { + encodedOffset = address.offset; + base = address.base; + } + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + fld(dest, base, encodedOffset); + return fco; +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_loadFloat(FloatRegister dest, + Address address) { + int16_t encodedOffset; + Register base; + + if (!is_int12(address.offset)) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, Imm32(address.offset)); + add(ScratchRegister, address.base, ScratchRegister); + base = ScratchRegister; + encodedOffset = 0; + } else { + encodedOffset = address.offset; + base = address.base; + } + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + flw(dest, base, encodedOffset); + return fco; +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_load( + Register dest, Address address, LoadStoreSize size, + LoadStoreExtension extension) { + int16_t encodedOffset; + Register base; + + if (!is_int12(address.offset)) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, Imm32(address.offset)); + add(ScratchRegister, address.base, ScratchRegister); + base = ScratchRegister; + encodedOffset = 0; + } else { + encodedOffset = address.offset; + base = address.base; + } + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + switch (size) { + case SizeByte: + if (ZeroExtend == extension) { + lbu(dest, base, encodedOffset); + } else { + lb(dest, base, encodedOffset); + } + break; + case SizeHalfWord: + if (ZeroExtend == extension) { + lhu(dest, base, encodedOffset); + } else { + lh(dest, base, encodedOffset); + } + break; + case SizeWord: + if (ZeroExtend == extension) { + lwu(dest, base, encodedOffset); + } else { + lw(dest, base, encodedOffset); + } + break; + case SizeDouble: + ld(dest, base, encodedOffset); + break; + default: + MOZ_CRASH("Invalid argument for ma_load"); + } + return fco; +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_store( + Register data, const BaseIndex& dest, LoadStoreSize size, + LoadStoreExtension extension) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + asMasm().computeScaledAddress(dest, scratch2); + return asMasm().ma_store(data, Address(scratch2, dest.offset), size, + extension); +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_store( + Imm32 imm, const BaseIndex& dest, LoadStoreSize size, + LoadStoreExtension extension) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register address = temps.Acquire(); + // Make sure that scratch contains absolute address so that + // offset is 0. + computeScaledAddress(dest, address); + + // Scrach register is free now, use it for loading imm value + ma_li(scratch, imm); + + // with offset=0 ScratchRegister will not be used in ma_store() + // so we can use it as a parameter here + return ma_store(scratch, Address(address, 0), size, extension); +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_store( + Imm32 imm, Address address, LoadStoreSize size, + LoadStoreExtension extension) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + return ma_store(scratch, address, size, extension); +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_store( + Register data, Address address, LoadStoreSize size, + LoadStoreExtension extension) { + int16_t encodedOffset; + Register base; + + if (!is_int12(address.offset)) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, Imm32(address.offset)); + add(ScratchRegister, address.base, ScratchRegister); + base = ScratchRegister; + encodedOffset = 0; + } else { + encodedOffset = address.offset; + base = address.base; + } + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + switch (size) { + case SizeByte: + sb(data, base, encodedOffset); + break; + case SizeHalfWord: + sh(data, base, encodedOffset); + break; + case SizeWord: + sw(data, base, encodedOffset); + break; + case SizeDouble: + sd(data, base, encodedOffset); + break; + default: + MOZ_CRASH("Invalid argument for ma_store"); + } + return fco; +} + +// Memory. +void MacroAssemblerRiscv64::ma_storeDouble(FloatRegister dest, + Address address) { + int16_t encodedOffset; + Register base; + + if (!is_int12(address.offset)) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, Imm32(address.offset)); + add(ScratchRegister, address.base, ScratchRegister); + base = ScratchRegister; + encodedOffset = 0; + } else { + encodedOffset = address.offset; + base = address.base; + } + fsd(dest, base, encodedOffset); +} + +void MacroAssemblerRiscv64::ma_storeFloat(FloatRegister dest, Address address) { + int16_t encodedOffset; + Register base; + + if (!is_int12(address.offset)) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, Imm32(address.offset)); + add(ScratchRegister, address.base, ScratchRegister); + base = ScratchRegister; + encodedOffset = 0; + } else { + encodedOffset = address.offset; + base = address.base; + } + fsw(dest, base, encodedOffset); +} + +void MacroAssemblerRiscv64::computeScaledAddress(const BaseIndex& address, + Register dest) { + Register base = address.base; + Register index = address.index; + int32_t shift = Imm32::ShiftOf(address.scale).value; + UseScratchRegisterScope temps(this); + Register tmp = dest == base ? temps.Acquire() : dest; + if (shift) { + MOZ_ASSERT(shift <= 4); + slli(tmp, index, shift); + add(dest, base, tmp); + } else { + add(dest, base, index); + } +} + +void MacroAssemblerRiscv64Compat::wasmLoadI64Impl( + const wasm::MemoryAccessDesc& access, Register memoryBase, Register ptr, + Register ptrScratch, Register64 output, Register tmp) { + access.assertOffsetInGuardPages(); + uint32_t offset = access.offset(); + MOZ_ASSERT_IF(offset, ptrScratch != InvalidReg); + + // Maybe add the offset. + if (offset) { + asMasm().addPtr(ImmWord(offset), ptrScratch); + ptr = ptrScratch; + } + + asMasm().memoryBarrierBefore(access.sync()); + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + FaultingCodeOffset fco; + switch (access.type()) { + case Scalar::Int8: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lb(output.reg, ScratchRegister, 0); + break; + case Scalar::Uint8: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lbu(output.reg, ScratchRegister, 0); + break; + case Scalar::Int16: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lh(output.reg, ScratchRegister, 0); + break; + case Scalar::Uint16: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lhu(output.reg, ScratchRegister, 0); + break; + case Scalar::Int32: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lw(output.reg, ScratchRegister, 0); + break; + case Scalar::Uint32: + // TODO(riscv): Why need zero-extension here? + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lwu(output.reg, ScratchRegister, 0); + break; + case Scalar::Int64: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + ld(output.reg, ScratchRegister, 0); + break; + default: + MOZ_CRASH("unexpected array type"); + } + + asMasm().append(access, js::wasm::TrapMachineInsnForLoad(access.byteSize()), + fco); + asMasm().memoryBarrierAfter(access.sync()); +} + +void MacroAssemblerRiscv64Compat::wasmStoreI64Impl( + const wasm::MemoryAccessDesc& access, Register64 value, Register memoryBase, + Register ptr, Register ptrScratch, Register tmp) { + access.assertOffsetInGuardPages(); + uint32_t offset = access.offset(); + MOZ_ASSERT_IF(offset, ptrScratch != InvalidReg); + + // Maybe add the offset. + if (offset) { + asMasm().addPtr(ImmWord(offset), ptrScratch); + ptr = ptrScratch; + } + + asMasm().memoryBarrierBefore(access.sync()); + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + FaultingCodeOffset fco; + switch (access.type()) { + case Scalar::Int8: + case Scalar::Uint8: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + sb(value.reg, ScratchRegister, 0); + break; + case Scalar::Int16: + case Scalar::Uint16: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + sh(value.reg, ScratchRegister, 0); + break; + case Scalar::Int32: + case Scalar::Uint32: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + sw(value.reg, ScratchRegister, 0); + break; + case Scalar::Int64: + add(ScratchRegister, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + sd(value.reg, ScratchRegister, 0); + break; + default: + MOZ_CRASH("unexpected array type"); + } + + asMasm().append(access, js::wasm::TrapMachineInsnForLoad(access.byteSize()), + fco); + asMasm().memoryBarrierAfter(access.sync()); +} + +void MacroAssemblerRiscv64Compat::profilerEnterFrame(Register framePtr, + Register scratch) { + asMasm().loadJSContext(scratch); + loadPtr(Address(scratch, offsetof(JSContext, profilingActivation_)), scratch); + storePtr(framePtr, + Address(scratch, JitActivation::offsetOfLastProfilingFrame())); + storePtr(ImmPtr(nullptr), + Address(scratch, JitActivation::offsetOfLastProfilingCallSite())); +} + +void MacroAssemblerRiscv64Compat::profilerExitFrame() { + jump(asMasm().runtime()->jitRuntime()->getProfilerExitFrameTail()); +} + +void MacroAssemblerRiscv64Compat::move32(Imm32 imm, Register dest) { + ma_li(dest, imm); +} + +void MacroAssemblerRiscv64Compat::move32(Register src, Register dest) { + slliw(dest, src, 0); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load8ZeroExtend( + const Address& address, Register dest) { + return ma_load(dest, address, SizeByte, ZeroExtend); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load8ZeroExtend( + const BaseIndex& src, Register dest) { + return ma_load(dest, src, SizeByte, ZeroExtend); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load8SignExtend( + const Address& address, Register dest) { + return ma_load(dest, address, SizeByte, SignExtend); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load8SignExtend( + const BaseIndex& src, Register dest) { + return ma_load(dest, src, SizeByte, SignExtend); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load16ZeroExtend( + const Address& address, Register dest) { + return ma_load(dest, address, SizeHalfWord, ZeroExtend); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load16ZeroExtend( + const BaseIndex& src, Register dest) { + return ma_load(dest, src, SizeHalfWord, ZeroExtend); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load16SignExtend( + const Address& address, Register dest) { + return ma_load(dest, address, SizeHalfWord, SignExtend); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load16SignExtend( + const BaseIndex& src, Register dest) { + return ma_load(dest, src, SizeHalfWord, SignExtend); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load32(const Address& address, + Register dest) { + return ma_load(dest, address, SizeWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load32(const BaseIndex& address, + Register dest) { + return ma_load(dest, address, SizeWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load32(AbsoluteAddress address, + Register dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + movePtr(ImmPtr(address.addr), ScratchRegister); + return load32(Address(ScratchRegister, 0), dest); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::load32( + wasm::SymbolicAddress address, Register dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + movePtr(address, ScratchRegister); + return load32(Address(ScratchRegister, 0), dest); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::loadPtr(const Address& address, + Register dest) { + return ma_load(dest, address, SizeDouble); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::loadPtr(const BaseIndex& src, + Register dest) { + return ma_load(dest, src, SizeDouble); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::loadPtr(AbsoluteAddress address, + Register dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + movePtr(ImmPtr(address.addr), ScratchRegister); + return loadPtr(Address(ScratchRegister, 0), dest); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::loadPtr( + wasm::SymbolicAddress address, Register dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + movePtr(address, ScratchRegister); + return loadPtr(Address(ScratchRegister, 0), dest); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::loadPrivate( + const Address& address, Register dest) { + return loadPtr(address, dest); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store8(Imm32 imm, + const Address& address) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, imm); + return ma_store(ScratchRegister, address, SizeByte); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store8(Register src, + const Address& address) { + return ma_store(src, address, SizeByte); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store8(Imm32 imm, + const BaseIndex& dest) { + return ma_store(imm, dest, SizeByte); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store8(Register src, + const BaseIndex& dest) { + return ma_store(src, dest, SizeByte); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store16( + Imm32 imm, const Address& address) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, imm); + return ma_store(ScratchRegister, address, SizeHalfWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store16( + Register src, const Address& address) { + return ma_store(src, address, SizeHalfWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store16(Imm32 imm, + const BaseIndex& dest) { + return ma_store(imm, dest, SizeHalfWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store16( + Register src, const BaseIndex& address) { + return ma_store(src, address, SizeHalfWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store32( + Register src, AbsoluteAddress address) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + movePtr(ImmPtr(address.addr), ScratchRegister); + return store32(src, Address(ScratchRegister, 0)); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store32( + Register src, const Address& address) { + return ma_store(src, address, SizeWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store32( + Imm32 src, const Address& address) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + move32(src, ScratchRegister); + return ma_store(ScratchRegister, address, SizeWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store32(Imm32 imm, + const BaseIndex& dest) { + return ma_store(imm, dest, SizeWord); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::store32(Register src, + const BaseIndex& dest) { + return ma_store(src, dest, SizeWord); +} + +template <typename T> +FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr(ImmWord imm, + T address) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, imm); + return ma_store(ScratchRegister, address, SizeDouble); +} + +template FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr<Address>( + ImmWord imm, Address address); +template FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr<BaseIndex>( + ImmWord imm, BaseIndex address); + +template <typename T> +FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr(ImmPtr imm, + T address) { + return storePtr(ImmWord(uintptr_t(imm.value)), address); +} + +template FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr<Address>( + ImmPtr imm, Address address); +template FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr<BaseIndex>( + ImmPtr imm, BaseIndex address); + +template <typename T> +FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr(ImmGCPtr imm, + T address) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + movePtr(imm, ScratchRegister); + return storePtr(ScratchRegister, address); +} + +template FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr<Address>( + ImmGCPtr imm, Address address); +template FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr<BaseIndex>( + ImmGCPtr imm, BaseIndex address); + +FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr( + Register src, const Address& address) { + return ma_store(src, address, SizeDouble); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr( + Register src, const BaseIndex& address) { + return ma_store(src, address, SizeDouble); +} + +FaultingCodeOffset MacroAssemblerRiscv64Compat::storePtr(Register src, + AbsoluteAddress dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + movePtr(ImmPtr(dest.addr), ScratchRegister); + return storePtr(src, Address(ScratchRegister, 0)); +} + +void MacroAssemblerRiscv64Compat::testNullSet(Condition cond, + const ValueOperand& value, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + splitTag(value, ScratchRegister); + ma_cmp_set(dest, ScratchRegister, ImmTag(JSVAL_TAG_NULL), cond); +} + +void MacroAssemblerRiscv64Compat::testObjectSet(Condition cond, + const ValueOperand& value, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + splitTag(value, ScratchRegister); + ma_cmp_set(dest, ScratchRegister, ImmTag(JSVAL_TAG_OBJECT), cond); +} + +void MacroAssemblerRiscv64Compat::testUndefinedSet(Condition cond, + const ValueOperand& value, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + splitTag(value, ScratchRegister); + ma_cmp_set(dest, ScratchRegister, ImmTag(JSVAL_TAG_UNDEFINED), cond); +} + +void MacroAssemblerRiscv64Compat::unboxInt32(const ValueOperand& operand, + Register dest) { + slliw(dest, operand.valueReg(), 0); +} + +void MacroAssemblerRiscv64Compat::unboxInt32(Register src, Register dest) { + slliw(dest, src, 0); +} + +void MacroAssemblerRiscv64Compat::unboxInt32(const Address& src, + Register dest) { + load32(Address(src.base, src.offset), dest); +} + +void MacroAssemblerRiscv64Compat::unboxInt32(const BaseIndex& src, + Register dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + computeScaledAddress(src, ScratchRegister); + load32(Address(ScratchRegister, src.offset), dest); +} + +void MacroAssemblerRiscv64Compat::unboxBoolean(const ValueOperand& operand, + Register dest) { + ExtractBits(dest, operand.valueReg(), 0, 32); +} + +void MacroAssemblerRiscv64Compat::unboxBoolean(Register src, Register dest) { + ExtractBits(dest, src, 0, 32); +} + +void MacroAssemblerRiscv64Compat::unboxBoolean(const Address& src, + Register dest) { + ma_load(dest, Address(src.base, src.offset), SizeWord, ZeroExtend); +} + +void MacroAssemblerRiscv64Compat::unboxBoolean(const BaseIndex& src, + Register dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + computeScaledAddress(src, ScratchRegister); + ma_load(dest, Address(ScratchRegister, src.offset), SizeWord, ZeroExtend); +} + +void MacroAssemblerRiscv64Compat::unboxDouble(const ValueOperand& operand, + FloatRegister dest) { + fmv_d_x(dest, operand.valueReg()); +} + +void MacroAssemblerRiscv64Compat::unboxDouble(const Address& src, + FloatRegister dest) { + ma_loadDouble(dest, Address(src.base, src.offset)); +} + +void MacroAssemblerRiscv64Compat::unboxDouble(const BaseIndex& src, + FloatRegister dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + loadPtr(src, scratch); + unboxDouble(ValueOperand(scratch), dest); +} + +void MacroAssemblerRiscv64Compat::unboxString(const ValueOperand& operand, + Register dest) { + unboxNonDouble(operand, dest, JSVAL_TYPE_STRING); +} + +void MacroAssemblerRiscv64Compat::unboxString(Register src, Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_STRING); +} + +void MacroAssemblerRiscv64Compat::unboxString(const Address& src, + Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_STRING); +} + +void MacroAssemblerRiscv64Compat::unboxSymbol(const ValueOperand& operand, + Register dest) { + unboxNonDouble(operand, dest, JSVAL_TYPE_SYMBOL); +} + +void MacroAssemblerRiscv64Compat::unboxSymbol(Register src, Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_SYMBOL); +} + +void MacroAssemblerRiscv64Compat::unboxSymbol(const Address& src, + Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_SYMBOL); +} + +void MacroAssemblerRiscv64Compat::unboxBigInt(const ValueOperand& operand, + Register dest) { + unboxNonDouble(operand, dest, JSVAL_TYPE_BIGINT); +} + +void MacroAssemblerRiscv64Compat::unboxBigInt(Register src, Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_BIGINT); +} + +void MacroAssemblerRiscv64Compat::unboxBigInt(const Address& src, + Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_BIGINT); +} + +void MacroAssemblerRiscv64Compat::unboxObject(const ValueOperand& src, + Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_OBJECT); +} + +void MacroAssemblerRiscv64Compat::unboxObject(Register src, Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_OBJECT); +} + +void MacroAssemblerRiscv64Compat::unboxObject(const Address& src, + Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_OBJECT); +} + +void MacroAssemblerRiscv64Compat::unboxValue(const ValueOperand& src, + AnyRegister dest, + JSValueType type) { + if (dest.isFloat()) { + Label notInt32, end; + asMasm().branchTestInt32(Assembler::NotEqual, src, ¬Int32); + convertInt32ToDouble(src.valueReg(), dest.fpu()); + ma_branch(&end); + bind(¬Int32); + unboxDouble(src, dest.fpu()); + bind(&end); + } else { + unboxNonDouble(src, dest.gpr(), type); + } +} + +void MacroAssemblerRiscv64Compat::boxDouble(FloatRegister src, + const ValueOperand& dest, + FloatRegister) { + fmv_x_d(dest.valueReg(), src); +} + +void MacroAssemblerRiscv64Compat::boxNonDouble(JSValueType type, Register src, + const ValueOperand& dest) { + MOZ_ASSERT(src != dest.valueReg()); + boxValue(type, src, dest.valueReg()); +} + +void MacroAssemblerRiscv64Compat::boolValueToDouble(const ValueOperand& operand, + FloatRegister dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + convertBoolToInt32(operand.valueReg(), ScratchRegister); + convertInt32ToDouble(ScratchRegister, dest); +} + +void MacroAssemblerRiscv64Compat::int32ValueToDouble( + const ValueOperand& operand, FloatRegister dest) { + convertInt32ToDouble(operand.valueReg(), dest); +} + +void MacroAssemblerRiscv64Compat::boolValueToFloat32( + const ValueOperand& operand, FloatRegister dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + convertBoolToInt32(operand.valueReg(), ScratchRegister); + convertInt32ToFloat32(ScratchRegister, dest); +} + +void MacroAssemblerRiscv64Compat::int32ValueToFloat32( + const ValueOperand& operand, FloatRegister dest) { + convertInt32ToFloat32(operand.valueReg(), dest); +} + +void MacroAssemblerRiscv64Compat::loadConstantFloat32(float f, + FloatRegister dest) { + ma_lis(dest, f); +} + +void MacroAssemblerRiscv64Compat::loadInt32OrDouble(const Address& src, + FloatRegister dest) { + Label notInt32, end; + // If it's an int, convert it to double. + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Register SecondScratchReg = temps.Acquire(); + loadPtr(Address(src.base, src.offset), ScratchRegister); + srli(SecondScratchReg, ScratchRegister, JSVAL_TAG_SHIFT); + asMasm().branchTestInt32(Assembler::NotEqual, SecondScratchReg, ¬Int32); + loadPtr(Address(src.base, src.offset), SecondScratchReg); + convertInt32ToDouble(SecondScratchReg, dest); + ma_branch(&end); + + // Not an int, just load as double. + bind(¬Int32); + unboxDouble(src, dest); + bind(&end); +} + +void MacroAssemblerRiscv64Compat::loadInt32OrDouble(const BaseIndex& addr, + FloatRegister dest) { + Label notInt32, end; + + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Register SecondScratchReg = temps.Acquire(); + // If it's an int, convert it to double. + computeScaledAddress(addr, SecondScratchReg); + // Since we only have one scratch, we need to stomp over it with the tag. + loadPtr(Address(SecondScratchReg, 0), ScratchRegister); + srli(SecondScratchReg, ScratchRegister, JSVAL_TAG_SHIFT); + asMasm().branchTestInt32(Assembler::NotEqual, SecondScratchReg, ¬Int32); + + computeScaledAddress(addr, SecondScratchReg); + loadPtr(Address(SecondScratchReg, 0), SecondScratchReg); + convertInt32ToDouble(SecondScratchReg, dest); + ma_branch(&end); + + // Not an int, just load as double. + bind(¬Int32); + // First, recompute the offset that had been stored in the scratch register + // since the scratch register was overwritten loading in the type. + computeScaledAddress(addr, SecondScratchReg); + unboxDouble(Address(SecondScratchReg, 0), dest); + bind(&end); +} + +void MacroAssemblerRiscv64Compat::loadConstantDouble(double dp, + FloatRegister dest) { + ma_lid(dest, dp); +} + +Register MacroAssemblerRiscv64Compat::extractObject(const Address& address, + Register scratch) { + loadPtr(Address(address.base, address.offset), scratch); + ExtractBits(scratch, scratch, 0, JSVAL_TAG_SHIFT); + return scratch; +} + +Register MacroAssemblerRiscv64Compat::extractTag(const Address& address, + Register scratch) { + loadPtr(Address(address.base, address.offset), scratch); + ExtractBits(scratch, scratch, JSVAL_TAG_SHIFT, 64 - JSVAL_TAG_SHIFT); + return scratch; +} + +Register MacroAssemblerRiscv64Compat::extractTag(const BaseIndex& address, + Register scratch) { + computeScaledAddress(address, scratch); + return extractTag(Address(scratch, address.offset), scratch); +} + +///////////////////////////////////////////////////////////////// +// X86/X64-common/ARM/LoongArch interface. +///////////////////////////////////////////////////////////////// +///////////////////////////////////////////////////////////////// +// X86/X64-common/ARM/MIPS interface. +///////////////////////////////////////////////////////////////// +void MacroAssemblerRiscv64Compat::storeValue(ValueOperand val, + const BaseIndex& dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + computeScaledAddress(dest, ScratchRegister); + storeValue(val, Address(ScratchRegister, dest.offset)); +} + +void MacroAssemblerRiscv64Compat::storeValue(JSValueType type, Register reg, + BaseIndex dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + + computeScaledAddress(dest, ScratchRegister); + + int32_t offset = dest.offset; + if (!is_int12(offset)) { + UseScratchRegisterScope temps(this); + Register SecondScratchReg = temps.Acquire(); + ma_li(SecondScratchReg, Imm32(offset)); + add(ScratchRegister, ScratchRegister, SecondScratchReg); + offset = 0; + } + + storeValue(type, reg, Address(ScratchRegister, offset)); +} + +void MacroAssemblerRiscv64Compat::storeValue(ValueOperand val, + const Address& dest) { + storePtr(val.valueReg(), Address(dest.base, dest.offset)); +} + +void MacroAssemblerRiscv64Compat::storeValue(JSValueType type, Register reg, + Address dest) { + if (type == JSVAL_TYPE_INT32 || type == JSVAL_TYPE_BOOLEAN) { + store32(reg, dest); + JSValueShiftedTag tag = (JSValueShiftedTag)JSVAL_TYPE_TO_SHIFTED_TAG(type); + store32(((Imm64(tag)).secondHalf()), Address(dest.base, dest.offset + 4)); + } else { + ScratchRegisterScope SecondScratchReg(asMasm()); + MOZ_ASSERT(dest.base != SecondScratchReg); + ma_li(SecondScratchReg, ImmTag(JSVAL_TYPE_TO_TAG(type))); + slli(SecondScratchReg, SecondScratchReg, JSVAL_TAG_SHIFT); + InsertBits(SecondScratchReg, reg, 0, JSVAL_TAG_SHIFT); + storePtr(SecondScratchReg, Address(dest.base, dest.offset)); + } +} + +void MacroAssemblerRiscv64Compat::storeValue(const Value& val, Address dest) { + UseScratchRegisterScope temps(this); + Register SecondScratchReg = temps.Acquire(); + if (val.isGCThing()) { + writeDataRelocation(val); + movWithPatch(ImmWord(val.asRawBits()), SecondScratchReg); + } else { + ma_li(SecondScratchReg, ImmWord(val.asRawBits())); + } + storePtr(SecondScratchReg, Address(dest.base, dest.offset)); +} + +void MacroAssemblerRiscv64Compat::storeValue(const Value& val, BaseIndex dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Register SecondScratchReg = temps.Acquire(); + computeScaledAddress(dest, ScratchRegister); + + int32_t offset = dest.offset; + if (!is_int12(offset)) { + ma_li(SecondScratchReg, Imm32(offset)); + add(ScratchRegister, ScratchRegister, SecondScratchReg); + offset = 0; + } + storeValue(val, Address(ScratchRegister, offset)); +} + +void MacroAssemblerRiscv64Compat::loadValue(const BaseIndex& addr, + ValueOperand val) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + computeScaledAddress(addr, ScratchRegister); + loadValue(Address(ScratchRegister, addr.offset), val); +} + +void MacroAssemblerRiscv64Compat::loadValue(Address src, ValueOperand val) { + loadPtr(Address(src.base, src.offset), val.valueReg()); +} + +void MacroAssemblerRiscv64Compat::tagValue(JSValueType type, Register payload, + ValueOperand dest) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + MOZ_ASSERT(dest.valueReg() != ScratchRegister); + JitSpew(JitSpew_Codegen, "[ tagValue"); + if (payload != dest.valueReg()) { + mv(dest.valueReg(), payload); + } + ma_li(ScratchRegister, ImmTag(JSVAL_TYPE_TO_TAG(type))); + InsertBits(dest.valueReg(), ScratchRegister, JSVAL_TAG_SHIFT, + 64 - JSVAL_TAG_SHIFT); + if (type == JSVAL_TYPE_INT32 || type == JSVAL_TYPE_BOOLEAN) { + InsertBits(dest.valueReg(), zero, 32, JSVAL_TAG_SHIFT - 32); + } + JitSpew(JitSpew_Codegen, "]"); +} + +void MacroAssemblerRiscv64Compat::pushValue(ValueOperand val) { + // Allocate stack slots for Value. One for each. + asMasm().subPtr(Imm32(sizeof(Value)), StackPointer); + // Store Value + storeValue(val, Address(StackPointer, 0)); +} + +void MacroAssemblerRiscv64Compat::pushValue(const Address& addr) { + // Load value before allocate stack, addr.base may be is sp. + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + loadPtr(Address(addr.base, addr.offset), ScratchRegister); + ma_sub64(StackPointer, StackPointer, Imm32(sizeof(Value))); + storePtr(ScratchRegister, Address(StackPointer, 0)); +} + +void MacroAssemblerRiscv64Compat::popValue(ValueOperand val) { + ld(val.valueReg(), StackPointer, 0); + ma_add64(StackPointer, StackPointer, Imm32(sizeof(Value))); +} + +void MacroAssemblerRiscv64Compat::breakpoint(uint32_t value) { break_(value); } + +void MacroAssemblerRiscv64Compat::ensureDouble(const ValueOperand& source, + FloatRegister dest, + Label* failure) { + Label isDouble, done; + { + ScratchTagScope tag(asMasm(), source); + splitTagForTest(source, tag); + asMasm().branchTestDouble(Assembler::Equal, tag, &isDouble); + asMasm().branchTestInt32(Assembler::NotEqual, tag, failure); + } + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + unboxInt32(source, ScratchRegister); + convertInt32ToDouble(ScratchRegister, dest); + jump(&done); + + bind(&isDouble); + unboxDouble(source, dest); + + bind(&done); +} + +void MacroAssemblerRiscv64Compat::handleFailureWithHandlerTail( + Label* profilerExitTail, Label* bailoutTail) { + // Reserve space for exception information. + int size = (sizeof(ResumeFromException) + ABIStackAlignment) & + ~(ABIStackAlignment - 1); + asMasm().subPtr(Imm32(size), StackPointer); + mv(a0, StackPointer); // Use a0 since it is a first function argument + + // Call the handler. + using Fn = void (*)(ResumeFromException* rfe); + asMasm().setupUnalignedABICall(a1); + asMasm().passABIArg(a0); + asMasm().callWithABI<Fn, HandleException>( + ABIType::General, CheckUnsafeCallWithABI::DontCheckHasExitFrame); + + Label entryFrame; + Label catch_; + Label finally; + Label returnBaseline; + Label returnIon; + Label bailout; + Label wasm; + Label wasmCatch; + + // Already clobbered a0, so use it... + load32(Address(StackPointer, ResumeFromException::offsetOfKind()), a0); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ExceptionResumeKind::EntryFrame), &entryFrame); + asMasm().branch32(Assembler::Equal, a0, Imm32(ExceptionResumeKind::Catch), + &catch_); + asMasm().branch32(Assembler::Equal, a0, Imm32(ExceptionResumeKind::Finally), + &finally); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ExceptionResumeKind::ForcedReturnBaseline), + &returnBaseline); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ExceptionResumeKind::ForcedReturnIon), &returnIon); + asMasm().branch32(Assembler::Equal, a0, Imm32(ExceptionResumeKind::Bailout), + &bailout); + asMasm().branch32(Assembler::Equal, a0, Imm32(ExceptionResumeKind::Wasm), + &wasm); + asMasm().branch32(Assembler::Equal, a0, Imm32(ExceptionResumeKind::WasmCatch), + &wasmCatch); + + breakpoint(); // Invalid kind. + + // No exception handler. Load the error value, restore state and return from + // the entry frame. + bind(&entryFrame); + asMasm().moveValue(MagicValue(JS_ION_ERROR), JSReturnOperand); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfFramePointer()), + FramePointer); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfStackPointer()), + StackPointer); + + // We're going to be returning by the ion calling convention + ma_pop(ra); + jump(ra); + nop(); + + // If we found a catch handler, this must be a baseline frame. Restore + // state and jump to the catch block. + bind(&catch_); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfTarget()), a0); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfFramePointer()), + FramePointer); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfStackPointer()), + StackPointer); + jump(a0); + + // If we found a finally block, this must be a baseline frame. Push three + // values expected by the finally block: the exception, the exception stack, + // and BooleanValue(true). + bind(&finally); + ValueOperand exception = ValueOperand(a1); + loadValue(Address(sp, ResumeFromException::offsetOfException()), exception); + + ValueOperand exceptionStack = ValueOperand(a2); + loadValue(Address(sp, ResumeFromException::offsetOfExceptionStack()), + exceptionStack); + + loadPtr(Address(sp, ResumeFromException::offsetOfTarget()), a0); + loadPtr(Address(sp, ResumeFromException::offsetOfFramePointer()), + FramePointer); + loadPtr(Address(sp, ResumeFromException::offsetOfStackPointer()), sp); + + pushValue(exception); + pushValue(exceptionStack); + pushValue(BooleanValue(true)); + jump(a0); + + // Return BaselineFrame->returnValue() to the caller. + // Used in debug mode and for GeneratorReturn. + Label profilingInstrumentation; + bind(&returnBaseline); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfFramePointer()), + FramePointer); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfStackPointer()), + StackPointer); + loadValue(Address(FramePointer, BaselineFrame::reverseOffsetOfReturnValue()), + JSReturnOperand); + jump(&profilingInstrumentation); + + // Return the given value to the caller. + bind(&returnIon); + loadValue(Address(StackPointer, ResumeFromException::offsetOfException()), + JSReturnOperand); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfFramePointer()), + FramePointer); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfStackPointer()), + StackPointer); + + // If profiling is enabled, then update the lastProfilingFrame to refer to + // caller frame before returning. This code is shared by ForcedReturnIon + // and ForcedReturnBaseline. + bind(&profilingInstrumentation); + { + Label skipProfilingInstrumentation; + // Test if profiler enabled. + AbsoluteAddress addressOfEnabled( + asMasm().runtime()->geckoProfiler().addressOfEnabled()); + asMasm().branch32(Assembler::Equal, addressOfEnabled, Imm32(0), + &skipProfilingInstrumentation); + jump(profilerExitTail); + bind(&skipProfilingInstrumentation); + } + + mv(StackPointer, FramePointer); + pop(FramePointer); + ret(); + + // If we are bailing out to baseline to handle an exception, jump to + // the bailout tail stub. Load 1 (true) in ReturnReg to indicate success. + bind(&bailout); + loadPtr(Address(sp, ResumeFromException::offsetOfBailoutInfo()), a2); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfStackPointer()), + StackPointer); + ma_li(ReturnReg, Imm32(1)); + jump(bailoutTail); + + // If we are throwing and the innermost frame was a wasm frame, reset SP and + // FP; SP is pointing to the unwound return address to the wasm entry, so + // we can just ret(). + bind(&wasm); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfFramePointer()), + FramePointer); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfStackPointer()), + StackPointer); + ma_li(InstanceReg, ImmWord(wasm::FailInstanceReg)); + ret(); + + // Found a wasm catch handler, restore state and jump to it. + bind(&wasmCatch); + loadPtr(Address(sp, ResumeFromException::offsetOfTarget()), a1); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfFramePointer()), + FramePointer); + loadPtr(Address(StackPointer, ResumeFromException::offsetOfStackPointer()), + StackPointer); + jump(a1); +} + +CodeOffset MacroAssemblerRiscv64Compat::toggledJump(Label* label) { + CodeOffset ret(nextOffset().getOffset()); + BranchShort(label); + return ret; +} + +CodeOffset MacroAssemblerRiscv64Compat::toggledCall(JitCode* target, + bool enabled) { + DEBUG_PRINTF("\ttoggledCall\n"); + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + BufferOffset bo = nextOffset(); + CodeOffset offset(bo.getOffset()); + addPendingJump(bo, ImmPtr(target->raw()), RelocationKind::JITCODE); + ma_liPatchable(ScratchRegister, ImmPtr(target->raw())); + if (enabled) { + jalr(ScratchRegister); + } else { + nop(); + } + MOZ_ASSERT_IF(!oom(), nextOffset().getOffset() - offset.offset() == + ToggledCallSize(nullptr)); + return offset; +} + +void MacroAssembler::subFromStackPtr(Imm32 imm32) { + if (imm32.value) { + asMasm().subPtr(imm32, StackPointer); + } +} + +void MacroAssembler::clampDoubleToUint8(FloatRegister input, Register output) { + JitSpew(JitSpew_Codegen, "[ clampDoubleToUint8"); + Label nan, done; + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + feq_d(scratch, input, input); + beqz(scratch, &nan); + addi(zero, scratch, 0x11); + Round_w_d(output, input); + clampIntToUint8(output); + ma_branch(&done); + // Input is nan + bind(&nan); + mv(output, zero_reg); + bind(&done); + JitSpew(JitSpew_Codegen, "]"); +} + +//{{{ check_macroassembler_style +// =============================================================== +// MacroAssembler high-level usage. +bool MacroAssembler::convertUInt64ToDoubleNeedsTemp() { return false; } +CodeOffset MacroAssembler::call(Label* label) { + BranchAndLink(label); + return CodeOffset(currentOffset()); +} +CodeOffset MacroAssembler::call(Register reg) { + jalr(reg, 0); + return CodeOffset(currentOffset()); +} +CodeOffset MacroAssembler::call(wasm::SymbolicAddress target) { + UseScratchRegisterScope temps(this); + temps.Exclude(GeneralRegisterSet(1 << CallReg.code())); + movePtr(target, CallReg); + return call(CallReg); +} +CodeOffset MacroAssembler::farJumpWithPatch() { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register scratch2 = temps.Acquire(); + // Allocate space which will be patched by patchFarJump(). + CodeOffset farJump(nextInstrOffset(5).getOffset()); + auipc(scratch, 0); + lw(scratch2, scratch, 4 * sizeof(Instr)); + add(scratch, scratch, scratch2); + jr(scratch, 0); + spew(".space 32bit initValue 0xffff ffff"); + emit(UINT32_MAX); + return farJump; +} +CodeOffset MacroAssembler::moveNearAddressWithPatch(Register dest) { + return movWithPatch(ImmPtr(nullptr), dest); +} +CodeOffset MacroAssembler::nopPatchableToCall() { + BlockTrampolinePoolScope block_trampoline_pool(this, 7); + // riscv64 + nop(); // lui(rd, (int32_t)high_20); + nop(); // addi(rd, rd, low_12); // 31 bits in rd. + nop(); // slli(rd, rd, 11); // Space for next 11 bis + nop(); // ori(rd, rd, b11); // 11 bits are put in. 42 bit in rd + nop(); // slli(rd, rd, 6); // Space for next 6 bits + nop(); // ori(rd, rd, a6); // 6 bits are put in. 48 bis in rd + nop(); // jirl + return CodeOffset(currentOffset()); +} +FaultingCodeOffset MacroAssembler::wasmTrapInstruction() { + BlockTrampolinePoolScope block_trampoline_pool(this, 2); + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + illegal_trap(kWasmTrapCode); + ebreak(); + return fco; +} +size_t MacroAssembler::PushRegsInMaskSizeInBytes(LiveRegisterSet set) { + return set.gprs().size() * sizeof(intptr_t) + set.fpus().getPushSizeInBytes(); +} + +template <typename T> +void MacroAssembler::branchValueIsNurseryCellImpl(Condition cond, + const T& value, Register temp, + Label* label) { + MOZ_ASSERT(cond == Assembler::Equal || cond == Assembler::NotEqual); + Label done; + branchTestGCThing(Assembler::NotEqual, value, + cond == Assembler::Equal ? &done : label); + + // temp may be InvalidReg, use scratch2 instead. + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + + getGCThingValueChunk(value, scratch2); + loadPtr(Address(scratch2, gc::ChunkStoreBufferOffset), scratch2); + branchPtr(InvertCondition(cond), scratch2, ImmWord(0), label); + + bind(&done); +} + +template <typename T> +void MacroAssembler::storeUnboxedValue(const ConstantOrRegister& value, + MIRType valueType, const T& dest) { + MOZ_ASSERT(valueType < MIRType::Value); + + if (valueType == MIRType::Double) { + boxDouble(value.reg().typedReg().fpu(), dest); + return; + } + + if (value.constant()) { + storeValue(value.value(), dest); + } else { + storeValue(ValueTypeFromMIRType(valueType), value.reg().typedReg().gpr(), + dest); + } +} + +template void MacroAssembler::storeUnboxedValue(const ConstantOrRegister& value, + MIRType valueType, + const Address& dest); +template void MacroAssembler::storeUnboxedValue( + const ConstantOrRegister& value, MIRType valueType, + const BaseObjectElementIndex& dest); + +// =============================================================== +// Jit Frames. + +uint32_t MacroAssembler::pushFakeReturnAddress(Register scratch) { + CodeLabel cl; + + ma_li(scratch, &cl); + Push(scratch); + bind(&cl); + uint32_t retAddr = currentOffset(); + + addCodeLabel(cl); + return retAddr; +} + +//=============================== +// AtomicOp + +template <typename T> +static void AtomicExchange(MacroAssembler& masm, + const wasm::MemoryAccessDesc* access, + Scalar::Type type, const Synchronization& sync, + const T& mem, Register value, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register output) { + ScratchRegisterScope scratch(masm); + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + bool signExtend = Scalar::isSignedIntType(type); + unsigned nbytes = Scalar::byteSize(type); + + switch (nbytes) { + case 1: + case 2: + break; + case 4: + MOZ_ASSERT(valueTemp == InvalidReg); + MOZ_ASSERT(offsetTemp == InvalidReg); + MOZ_ASSERT(maskTemp == InvalidReg); + break; + default: + MOZ_CRASH(); + } + + Label again; + + masm.computeEffectiveAddress(mem, scratch); + + if (nbytes == 4) { + masm.memoryBarrierBefore(sync); + masm.bind(&again); + BlockTrampolinePoolScope block_trampoline_pool(&masm, 5); + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_w(true, true, output, scratch); + masm.or_(scratch2, value, zero); + masm.sc_w(true, true, scratch2, scratch, scratch2); + masm.ma_b(scratch2, Register(scratch2), &again, Assembler::NonZero, + ShortJump); + + masm.memoryBarrierAfter(sync); + + return; + } + + masm.andi(offsetTemp, scratch, 3); + masm.subPtr(offsetTemp, scratch); + masm.slliw(offsetTemp, offsetTemp, 3); + masm.ma_li(maskTemp, Imm32(UINT32_MAX >> ((4 - nbytes) * 8))); + masm.sllw(maskTemp, maskTemp, offsetTemp); + masm.nor(maskTemp, zero, maskTemp); + switch (nbytes) { + case 1: + masm.andi(valueTemp, value, 0xff); + break; + case 2: + masm.ma_and(valueTemp, value, Imm32(0xffff)); + break; + } + masm.sllw(valueTemp, valueTemp, offsetTemp); + + masm.memoryBarrierBefore(sync); + + masm.bind(&again); + + BlockTrampolinePoolScope block_trampoline_pool(&masm, 10); + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_w(true, true, output, scratch); + masm.and_(scratch2, output, maskTemp); + masm.or_(scratch2, scratch2, valueTemp); + + masm.sc_w(true, true, scratch2, scratch, scratch2); + + masm.ma_b(scratch2, Register(scratch2), &again, Assembler::NonZero, + ShortJump); + + masm.srlw(output, output, offsetTemp); + + switch (nbytes) { + case 1: + if (signExtend) { + masm.slliw(output, output, 32 - 8); + masm.sraiw(output, output, 32 - 8); + } else { + masm.andi(valueTemp, value, 0xff); + } + break; + case 2: + if (signExtend) { + masm.slliw(output, output, 32 - 16); + masm.sraiw(output, output, 32 - 16); + } else { + masm.ma_and(valueTemp, value, Imm32(0xffff)); + } + break; + } + + masm.memoryBarrierAfter(sync); +} + +template <typename T> +static void AtomicExchange64(MacroAssembler& masm, + const wasm::MemoryAccessDesc* access, + const Synchronization& sync, const T& mem, + Register64 value, Register64 output) { + MOZ_ASSERT(value != output); + UseScratchRegisterScope temps(&masm); + Register SecondScratchReg = temps.Acquire(); + masm.computeEffectiveAddress(mem, SecondScratchReg); + + Label tryAgain; + + masm.memoryBarrierBefore(sync); + + masm.bind(&tryAgain); + BlockTrampolinePoolScope block_trampoline_pool(&masm, 5); + if (access) { + masm.append(*access, js::wasm::TrapMachineInsn::Load64, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_d(true, true, output.reg, SecondScratchReg); + masm.movePtr(value.reg, ScratchRegister); + masm.sc_d(true, true, ScratchRegister, SecondScratchReg, ScratchRegister); + masm.ma_b(ScratchRegister, ScratchRegister, &tryAgain, Assembler::NonZero, + ShortJump); + + masm.memoryBarrierAfter(sync); +} + +template <typename T> +static void AtomicFetchOp64(MacroAssembler& masm, + const wasm::MemoryAccessDesc* access, + const Synchronization& sync, AtomicOp op, + Register64 value, const T& mem, Register64 temp, + Register64 output) { + MOZ_ASSERT(value != output); + MOZ_ASSERT(value != temp); + UseScratchRegisterScope temps(&masm); + Register SecondScratchReg = temps.Acquire(); + masm.computeEffectiveAddress(mem, SecondScratchReg); + + Label tryAgain; + + masm.memoryBarrierBefore(sync); + + masm.bind(&tryAgain); + BlockTrampolinePoolScope block_trampoline_pool(&masm, 5); + if (access) { + masm.append(*access, js::wasm::TrapMachineInsn::Load64, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_d(true, true, output.reg, SecondScratchReg); + + switch (op) { + case AtomicFetchAddOp: + masm.add(temp.reg, output.reg, value.reg); + break; + case AtomicFetchSubOp: + masm.sub(temp.reg, output.reg, value.reg); + break; + case AtomicFetchAndOp: + masm.and_(temp.reg, output.reg, value.reg); + break; + case AtomicFetchOrOp: + masm.or_(temp.reg, output.reg, value.reg); + break; + case AtomicFetchXorOp: + masm.xor_(temp.reg, output.reg, value.reg); + break; + default: + MOZ_CRASH(); + } + + masm.sc_d(true, true, temp.reg, SecondScratchReg, temp.reg); + masm.ma_b(temp.reg, temp.reg, &tryAgain, Assembler::NonZero, ShortJump); + + masm.memoryBarrierAfter(sync); +} + +template <typename T> +static void AtomicEffectOp(MacroAssembler& masm, + const wasm::MemoryAccessDesc* access, + Scalar::Type type, const Synchronization& sync, + AtomicOp op, const T& mem, Register value, + Register valueTemp, Register offsetTemp, + Register maskTemp) { + ScratchRegisterScope scratch(masm); + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + unsigned nbytes = Scalar::byteSize(type); + + switch (nbytes) { + case 1: + case 2: + break; + case 4: + MOZ_ASSERT(valueTemp == InvalidReg); + MOZ_ASSERT(offsetTemp == InvalidReg); + MOZ_ASSERT(maskTemp == InvalidReg); + break; + default: + MOZ_CRASH(); + } + + Label again; + + masm.computeEffectiveAddress(mem, scratch); + + if (nbytes == 4) { + masm.memoryBarrierBefore(sync); + masm.bind(&again); + + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_w(true, true, scratch2, scratch); + + switch (op) { + case AtomicFetchAddOp: + masm.addw(scratch2, scratch2, value); + break; + case AtomicFetchSubOp: + masm.subw(scratch2, scratch2, value); + break; + case AtomicFetchAndOp: + masm.and_(scratch2, scratch2, value); + break; + case AtomicFetchOrOp: + masm.or_(scratch2, scratch2, value); + break; + case AtomicFetchXorOp: + masm.xor_(scratch2, scratch2, value); + break; + default: + MOZ_CRASH(); + } + + masm.sc_w(true, true, scratch2, scratch, scratch2); + masm.ma_b(scratch2, Register(scratch2), &again, Assembler::NonZero, + ShortJump); + + masm.memoryBarrierAfter(sync); + + return; + } + + masm.andi(offsetTemp, scratch, 3); + masm.subPtr(offsetTemp, scratch); + masm.slliw(offsetTemp, offsetTemp, 3); + masm.ma_li(maskTemp, Imm32(UINT32_MAX >> ((4 - nbytes) * 8))); + masm.sllw(maskTemp, maskTemp, offsetTemp); + masm.nor(maskTemp, zero, maskTemp); + + masm.memoryBarrierBefore(sync); + + masm.bind(&again); + + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_w(true, true, scratch2, scratch); + masm.srlw(valueTemp, scratch2, offsetTemp); + + switch (op) { + case AtomicFetchAddOp: + masm.addw(valueTemp, valueTemp, value); + break; + case AtomicFetchSubOp: + masm.subw(valueTemp, valueTemp, value); + break; + case AtomicFetchAndOp: + masm.and_(valueTemp, valueTemp, value); + break; + case AtomicFetchOrOp: + masm.or_(valueTemp, valueTemp, value); + break; + case AtomicFetchXorOp: + masm.xor_(valueTemp, valueTemp, value); + break; + default: + MOZ_CRASH(); + } + + switch (nbytes) { + case 1: + masm.andi(valueTemp, valueTemp, 0xff); + break; + case 2: + masm.ma_and(valueTemp, valueTemp, Imm32(0xffff)); + break; + } + + masm.sllw(valueTemp, valueTemp, offsetTemp); + + masm.and_(scratch2, scratch2, maskTemp); + masm.or_(scratch2, scratch2, valueTemp); + + masm.sc_w(true, true, scratch2, scratch, scratch2); + + masm.ma_b(scratch2, Register(scratch2), &again, Assembler::NonZero, + ShortJump); + + masm.memoryBarrierAfter(sync); +} + +template <typename T> +static void AtomicFetchOp(MacroAssembler& masm, + const wasm::MemoryAccessDesc* access, + Scalar::Type type, const Synchronization& sync, + AtomicOp op, const T& mem, Register value, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register output) { + ScratchRegisterScope scratch(masm); + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + bool signExtend = Scalar::isSignedIntType(type); + unsigned nbytes = Scalar::byteSize(type); + + switch (nbytes) { + case 1: + case 2: + break; + case 4: + MOZ_ASSERT(valueTemp == InvalidReg); + MOZ_ASSERT(offsetTemp == InvalidReg); + MOZ_ASSERT(maskTemp == InvalidReg); + break; + default: + MOZ_CRASH(); + } + + Label again; + + masm.computeEffectiveAddress(mem, scratch); + + if (nbytes == 4) { + masm.memoryBarrierBefore(sync); + masm.bind(&again); + + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_w(true, true, output, scratch); + + switch (op) { + case AtomicFetchAddOp: + masm.addw(scratch2, output, value); + break; + case AtomicFetchSubOp: + masm.subw(scratch2, output, value); + break; + case AtomicFetchAndOp: + masm.and_(scratch2, output, value); + break; + case AtomicFetchOrOp: + masm.or_(scratch2, output, value); + break; + case AtomicFetchXorOp: + masm.xor_(scratch2, output, value); + break; + default: + MOZ_CRASH(); + } + + masm.sc_w(true, true, scratch2, scratch, scratch2); + masm.ma_b(scratch2, Register(scratch2), &again, Assembler::NonZero, + ShortJump); + + masm.memoryBarrierAfter(sync); + + return; + } + + masm.andi(offsetTemp, scratch, 3); + masm.subPtr(offsetTemp, scratch); + masm.slliw(offsetTemp, offsetTemp, 3); + masm.ma_li(maskTemp, Imm32(UINT32_MAX >> ((4 - nbytes) * 8))); + masm.sllw(maskTemp, maskTemp, offsetTemp); + masm.nor(maskTemp, zero, maskTemp); + + masm.memoryBarrierBefore(sync); + + masm.bind(&again); + + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_w(true, true, scratch2, scratch); + masm.srlw(output, scratch2, offsetTemp); + + switch (op) { + case AtomicFetchAddOp: + masm.addw(valueTemp, output, value); + break; + case AtomicFetchSubOp: + masm.subw(valueTemp, output, value); + break; + case AtomicFetchAndOp: + masm.and_(valueTemp, output, value); + break; + case AtomicFetchOrOp: + masm.or_(valueTemp, output, value); + break; + case AtomicFetchXorOp: + masm.xor_(valueTemp, output, value); + break; + default: + MOZ_CRASH(); + } + + switch (nbytes) { + case 1: + masm.andi(valueTemp, valueTemp, 0xff); + break; + case 2: + masm.andi(valueTemp, valueTemp, 0xffff); + break; + } + + masm.sllw(valueTemp, valueTemp, offsetTemp); + + masm.and_(scratch2, scratch2, maskTemp); + masm.or_(scratch2, scratch2, valueTemp); + + masm.sc_w(true, true, scratch2, scratch, scratch2); + + masm.ma_b(scratch2, Register(scratch2), &again, Assembler::NonZero, + ShortJump); + + switch (nbytes) { + case 1: + if (signExtend) { + masm.slliw(output, output, 32 - 8); + masm.sraiw(output, output, 32 - 8); + } else { + masm.andi(output, output, 0xff); + } + break; + case 2: + if (signExtend) { + masm.slliw(output, output, 32 - 16); + masm.sraiw(output, output, 32 - 16); + } else { + masm.andi(output, output, 0xffff); + } + break; + } + + masm.memoryBarrierAfter(sync); +} + +// ======================================================================== +// JS atomic operations. + +template <typename T> +static void CompareExchangeJS(MacroAssembler& masm, Scalar::Type arrayType, + const Synchronization& sync, const T& mem, + Register oldval, Register newval, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register temp, + AnyRegister output) { + if (arrayType == Scalar::Uint32) { + masm.compareExchange(arrayType, sync, mem, oldval, newval, valueTemp, + offsetTemp, maskTemp, temp); + masm.convertUInt32ToDouble(temp, output.fpu()); + } else { + masm.compareExchange(arrayType, sync, mem, oldval, newval, valueTemp, + offsetTemp, maskTemp, output.gpr()); + } +} + +template <typename T> +static void AtomicExchangeJS(MacroAssembler& masm, Scalar::Type arrayType, + const Synchronization& sync, const T& mem, + Register value, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register temp, AnyRegister output) { + if (arrayType == Scalar::Uint32) { + masm.atomicExchange(arrayType, sync, mem, value, valueTemp, offsetTemp, + maskTemp, temp); + masm.convertUInt32ToDouble(temp, output.fpu()); + } else { + masm.atomicExchange(arrayType, sync, mem, value, valueTemp, offsetTemp, + maskTemp, output.gpr()); + } +} + +template <typename T> +static void AtomicFetchOpJS(MacroAssembler& masm, Scalar::Type arrayType, + const Synchronization& sync, AtomicOp op, + Register value, const T& mem, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register temp, AnyRegister output) { + if (arrayType == Scalar::Uint32) { + masm.atomicFetchOp(arrayType, sync, op, value, mem, valueTemp, offsetTemp, + maskTemp, temp); + masm.convertUInt32ToDouble(temp, output.fpu()); + } else { + masm.atomicFetchOp(arrayType, sync, op, value, mem, valueTemp, offsetTemp, + maskTemp, output.gpr()); + } +} + +void MacroAssembler::atomicEffectOpJS(Scalar::Type arrayType, + const Synchronization& sync, AtomicOp op, + Register value, const BaseIndex& mem, + Register valueTemp, Register offsetTemp, + Register maskTemp) { + AtomicEffectOp(*this, nullptr, arrayType, sync, op, mem, value, valueTemp, + offsetTemp, maskTemp); +} + +void MacroAssembler::atomicEffectOpJS(Scalar::Type arrayType, + const Synchronization& sync, AtomicOp op, + Register value, const Address& mem, + Register valueTemp, Register offsetTemp, + Register maskTemp) { + AtomicEffectOp(*this, nullptr, arrayType, sync, op, mem, value, valueTemp, + offsetTemp, maskTemp); +} +void MacroAssembler::atomicExchange64(const Synchronization& sync, + const Address& mem, Register64 value, + Register64 output) { + AtomicExchange64(*this, nullptr, sync, mem, value, output); +} + +void MacroAssembler::atomicExchange64(const Synchronization& sync, + const BaseIndex& mem, Register64 value, + Register64 output) { + AtomicExchange64(*this, nullptr, sync, mem, value, output); +} + +void MacroAssembler::atomicExchangeJS(Scalar::Type arrayType, + const Synchronization& sync, + const Address& mem, Register value, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register temp, + AnyRegister output) { + AtomicExchangeJS(*this, arrayType, sync, mem, value, valueTemp, offsetTemp, + maskTemp, temp, output); +} + +void MacroAssembler::atomicExchangeJS(Scalar::Type arrayType, + const Synchronization& sync, + const BaseIndex& mem, Register value, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register temp, + AnyRegister output) { + AtomicExchangeJS(*this, arrayType, sync, mem, value, valueTemp, offsetTemp, + maskTemp, temp, output); +} + +void MacroAssembler::atomicExchange(Scalar::Type type, + const Synchronization& sync, + const Address& mem, Register value, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register output) { + AtomicExchange(*this, nullptr, type, sync, mem, value, valueTemp, offsetTemp, + maskTemp, output); +} + +void MacroAssembler::atomicExchange(Scalar::Type type, + const Synchronization& sync, + const BaseIndex& mem, Register value, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register output) { + AtomicExchange(*this, nullptr, type, sync, mem, value, valueTemp, offsetTemp, + maskTemp, output); +} + +void MacroAssembler::atomicFetchOpJS(Scalar::Type arrayType, + const Synchronization& sync, AtomicOp op, + Register value, const Address& mem, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register temp, + AnyRegister output) { + AtomicFetchOpJS(*this, arrayType, sync, op, value, mem, valueTemp, offsetTemp, + maskTemp, temp, output); +} + +void MacroAssembler::atomicFetchOpJS(Scalar::Type arrayType, + const Synchronization& sync, AtomicOp op, + Register value, const BaseIndex& mem, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register temp, + AnyRegister output) { + AtomicFetchOpJS(*this, arrayType, sync, op, value, mem, valueTemp, offsetTemp, + maskTemp, temp, output); +} + +void MacroAssembler::atomicFetchOp(Scalar::Type type, + const Synchronization& sync, AtomicOp op, + Register value, const Address& mem, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register output) { + AtomicFetchOp(*this, nullptr, type, sync, op, mem, value, valueTemp, + offsetTemp, maskTemp, output); +} + +void MacroAssembler::atomicFetchOp(Scalar::Type type, + const Synchronization& sync, AtomicOp op, + Register value, const BaseIndex& mem, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register output) { + AtomicFetchOp(*this, nullptr, type, sync, op, mem, value, valueTemp, + offsetTemp, maskTemp, output); +} +void MacroAssembler::branchPtrInNurseryChunk(Condition cond, Register ptr, + Register temp, Label* label) { + MOZ_ASSERT(cond == Assembler::Equal || cond == Assembler::NotEqual); + MOZ_ASSERT(ptr != temp); + MOZ_ASSERT(ptr != ScratchRegister); // Both may be used internally. + MOZ_ASSERT(temp != ScratchRegister); + MOZ_ASSERT(temp != InvalidReg); + + ma_and(temp, ptr, Imm32(int32_t(~gc::ChunkMask))); + branchPtr(InvertCondition(cond), Address(temp, gc::ChunkStoreBufferOffset), + zero, label); +} +void MacroAssembler::branchTestValue(Condition cond, const ValueOperand& lhs, + const Value& rhs, Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(lhs.valueReg() != scratch); + moveValue(rhs, ValueOperand(scratch)); + ma_b(lhs.valueReg(), scratch, label, cond); +} +void MacroAssembler::branchValueIsNurseryCell(Condition cond, + const Address& address, + Register temp, Label* label) { + branchValueIsNurseryCellImpl(cond, address, temp, label); +} + +void MacroAssembler::branchValueIsNurseryCell(Condition cond, + ValueOperand value, Register temp, + Label* label) { + branchValueIsNurseryCellImpl(cond, value, temp, label); +} +void MacroAssembler::call(const Address& addr) { + UseScratchRegisterScope temps(this); + temps.Exclude(GeneralRegisterSet(1 << CallReg.code())); + loadPtr(addr, CallReg); + call(CallReg); +} +void MacroAssembler::call(ImmPtr target) { + BufferOffset bo = m_buffer.nextOffset(); + addPendingJump(bo, target, RelocationKind::HARDCODED); + ma_call(target); +} +void MacroAssembler::call(ImmWord target) { call(ImmPtr((void*)target.value)); } + +void MacroAssembler::call(JitCode* c) { + DEBUG_PRINTF("[ %s\n", __FUNCTION__); + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BufferOffset bo = m_buffer.nextOffset(); + addPendingJump(bo, ImmPtr(c->raw()), RelocationKind::JITCODE); + ma_liPatchable(scratch, ImmPtr(c->raw())); + callJitNoProfiler(scratch); + DEBUG_PRINTF("]\n"); +} + +void MacroAssembler::callWithABIPre(uint32_t* stackAdjust, bool callFromWasm) { + MOZ_ASSERT(inCall_); + uint32_t stackForCall = abiArgs_.stackBytesConsumedSoFar(); + + // Reserve place for $ra. + stackForCall += sizeof(intptr_t); + + if (dynamicAlignment_) { + stackForCall += ComputeByteAlignment(stackForCall, ABIStackAlignment); + } else { + uint32_t alignmentAtPrologue = callFromWasm ? sizeof(wasm::Frame) : 0; + stackForCall += ComputeByteAlignment( + stackForCall + framePushed() + alignmentAtPrologue, ABIStackAlignment); + } + + *stackAdjust = stackForCall; + reserveStack(stackForCall); + + // Save $ra because call is going to clobber it. Restore it in + // callWithABIPost. NOTE: This is needed for calls from SharedIC. + // Maybe we can do this differently. + storePtr(ra, Address(StackPointer, stackForCall - sizeof(intptr_t))); + + // Position all arguments. + { + enoughMemory_ &= moveResolver_.resolve(); + if (!enoughMemory_) { + return; + } + + MoveEmitter emitter(asMasm()); + emitter.emit(moveResolver_); + emitter.finish(); + } + + assertStackAlignment(ABIStackAlignment); +} + +void MacroAssembler::callWithABIPost(uint32_t stackAdjust, ABIType result, + bool callFromWasm) { + // Restore ra value (as stored in callWithABIPre()). + loadPtr(Address(StackPointer, stackAdjust - sizeof(intptr_t)), ra); + + if (dynamicAlignment_) { + // Restore sp value from stack (as stored in setupUnalignedABICall()). + loadPtr(Address(StackPointer, stackAdjust), StackPointer); + // Use adjustFrame instead of freeStack because we already restored sp. + adjustFrame(-stackAdjust); + } else { + freeStack(stackAdjust); + } + +#ifdef DEBUG + MOZ_ASSERT(inCall_); + inCall_ = false; +#endif +} + +void MacroAssembler::callWithABINoProfiler(Register fun, ABIType result) { + // Load the callee in scratch2, no instruction between the movePtr and + // call should clobber it. Note that we can't use fun because it may be + // one of the IntArg registers clobbered before the call. + UseScratchRegisterScope temps(this); + temps.Exclude(GeneralRegisterSet(1 << CallReg.code())); + movePtr(fun, CallReg); + + uint32_t stackAdjust; + callWithABIPre(&stackAdjust); + call(CallReg); + callWithABIPost(stackAdjust, result); +} + +void MacroAssembler::callWithABINoProfiler(const Address& fun, ABIType result) { + // Load the callee in scratch2, as above. + UseScratchRegisterScope temps(this); + temps.Exclude(GeneralRegisterSet(1 << CallReg.code())); + loadPtr(fun, CallReg); + + uint32_t stackAdjust; + callWithABIPre(&stackAdjust); + call(CallReg); + callWithABIPost(stackAdjust, result); +} + +void MacroAssembler::ceilDoubleToInt32(FloatRegister src, Register dest, + Label* fail) { + UseScratchRegisterScope temps(this); + ScratchDoubleScope fscratch(*this); + Label performCeil, done; + // If x < -1 or x > 0 then perform ceil. + loadConstantDouble(0, fscratch); + branchDouble(Assembler::DoubleGreaterThan, src, fscratch, &performCeil); + loadConstantDouble(-1.0, fscratch); + branchDouble(Assembler::DoubleLessThanOrEqual, src, fscratch, &performCeil); + + Register scratch = temps.Acquire(); + // If binary value is not zero, the input was not 0, so we bail. + { + moveFromDoubleHi(src, scratch); + branch32(Assembler::NotEqual, scratch, zero, fail); + } + + bind(&performCeil); + Ceil_w_d(dest, src, scratch); + ma_b(scratch, Imm32(1), fail, NotEqual); + bind(&done); +} + +void MacroAssembler::ceilFloat32ToInt32(FloatRegister src, Register dest, + Label* fail) { + UseScratchRegisterScope temps(this); + ScratchDoubleScope fscratch(*this); + Label performCeil, done; + // If x < -1 or x > 0 then perform ceil. + loadConstantFloat32(0, fscratch); + branchFloat(Assembler::DoubleGreaterThan, src, fscratch, &performCeil); + loadConstantFloat32(-1.0, fscratch); + branchFloat(Assembler::DoubleLessThanOrEqual, src, fscratch, &performCeil); + + Register scratch = temps.Acquire(); + // If binary value is not zero, the input was not 0, so we bail. + { + fmv_x_w(scratch, src); + branch32(Assembler::NotEqual, scratch, zero, fail); + } + bind(&performCeil); + Ceil_w_s(dest, src, scratch); + ma_b(scratch, Imm32(1), fail, NotEqual); + bind(&done); +} +void MacroAssembler::comment(const char* msg) { Assembler::comment(msg); } + +template <typename T> +static void CompareExchange64(MacroAssembler& masm, + const wasm::MemoryAccessDesc* access, + const Synchronization& sync, const T& mem, + Register64 expect, Register64 replace, + Register64 output) { + MOZ_ASSERT(expect != output && replace != output); + ScratchRegisterScope scratch(masm); + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.computeEffectiveAddress(mem, scratch); + + Label tryAgain; + Label exit; + + masm.memoryBarrierBefore(sync); + + masm.bind(&tryAgain); + + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_d(true, true, output.reg, scratch); + + masm.ma_b(output.reg, expect.reg, &exit, Assembler::NotEqual, ShortJump); + masm.movePtr(replace.reg, scratch2); + masm.sc_d(true, true, scratch2, scratch, scratch2); + masm.ma_b(scratch2, Register(scratch2), &tryAgain, Assembler::NonZero, + ShortJump); + + masm.memoryBarrierAfter(sync); + + masm.bind(&exit); +} + +void MacroAssembler::compareExchange64(const Synchronization& sync, + const Address& mem, Register64 expect, + Register64 replace, Register64 output) { + CompareExchange64(*this, nullptr, sync, mem, expect, replace, output); +} + +void MacroAssembler::compareExchange64(const Synchronization& sync, + const BaseIndex& mem, Register64 expect, + Register64 replace, Register64 output) { + CompareExchange64(*this, nullptr, sync, mem, expect, replace, output); +} + +void MacroAssembler::compareExchangeJS(Scalar::Type arrayType, + const Synchronization& sync, + const Address& mem, Register oldval, + Register newval, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register temp, AnyRegister output) { + CompareExchangeJS(*this, arrayType, sync, mem, oldval, newval, valueTemp, + offsetTemp, maskTemp, temp, output); +} + +void MacroAssembler::compareExchangeJS(Scalar::Type arrayType, + const Synchronization& sync, + const BaseIndex& mem, Register oldval, + Register newval, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register temp, AnyRegister output) { + CompareExchangeJS(*this, arrayType, sync, mem, oldval, newval, valueTemp, + offsetTemp, maskTemp, temp, output); +} + +void MacroAssembler::convertInt64ToDouble(Register64 src, FloatRegister dest) { + fcvt_d_l(dest, src.scratchReg()); +} +void MacroAssembler::convertInt64ToFloat32(Register64 src, FloatRegister dest) { + fcvt_s_l(dest, src.scratchReg()); +} +void MacroAssembler::convertIntPtrToDouble(Register src, FloatRegister dest) { + fcvt_d_l(dest, src); +} +void MacroAssembler::convertUInt64ToDouble(Register64 src, FloatRegister dest, + Register tmp) { + fcvt_d_lu(dest, src.scratchReg()); +} +void MacroAssembler::convertUInt64ToFloat32(Register64 src, FloatRegister dest, + Register tmp) { + fcvt_s_lu(dest, src.scratchReg()); +} +void MacroAssembler::copySignDouble(FloatRegister lhs, FloatRegister rhs, + FloatRegister output) { + fsgnj_d(output, lhs, rhs); +} +void MacroAssembler::enterFakeExitFrameForWasm(Register cxreg, Register scratch, + ExitFrameType type) { + enterFakeExitFrame(cxreg, scratch, type); +} +void MacroAssembler::flexibleDivMod32(Register rhs, Register srcDest, + Register remOutput, bool isUnsigned, + const LiveRegisterSet&) { + if (isUnsigned) { + ma_modu32(remOutput, srcDest, rhs); + ma_divu32(srcDest, srcDest, rhs); + } else { + ma_mod32(remOutput, srcDest, rhs); + ma_div32(srcDest, srcDest, rhs); + } +} +void MacroAssembler::flexibleQuotient32(Register rhs, Register srcDest, + bool isUnsigned, + const LiveRegisterSet&) { + quotient32(rhs, srcDest, isUnsigned); +} + +void MacroAssembler::flexibleRemainder32(Register rhs, Register srcDest, + bool isUnsigned, + const LiveRegisterSet&) { + remainder32(rhs, srcDest, isUnsigned); +} + +void MacroAssembler::floorDoubleToInt32(FloatRegister src, Register dest, + Label* fail) { + JitSpew(JitSpew_Codegen, "[ %s", __FUNCTION__); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Floor_w_d(dest, src, scratch); + ma_b(scratch, Imm32(1), fail, NotEqual); + fmv_x_d(scratch, src); + ma_branch(fail, Equal, scratch, Operand(0x8000000000000000)); + JitSpew(JitSpew_Codegen, "]"); +} +void MacroAssembler::floorFloat32ToInt32(FloatRegister src, Register dest, + Label* fail) { + JitSpew(JitSpew_Codegen, "[ %s", __FUNCTION__); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Floor_w_s(dest, src, scratch); + ma_b(scratch, Imm32(1), fail, NotEqual); + fmv_x_w(scratch, src); + ma_branch(fail, Equal, scratch, Operand(int32_t(0x80000000))); + JitSpew(JitSpew_Codegen, "]"); +} +void MacroAssembler::flush() {} +void MacroAssembler::loadStoreBuffer(Register ptr, Register buffer) { + ma_and(buffer, ptr, Imm32(int32_t(~gc::ChunkMask))); + loadPtr(Address(buffer, gc::ChunkStoreBufferOffset), buffer); +} + +void MacroAssembler::moveValue(const TypedOrValueRegister& src, + const ValueOperand& dest) { + if (src.hasValue()) { + moveValue(src.valueReg(), dest); + return; + } + + MIRType type = src.type(); + AnyRegister reg = src.typedReg(); + + if (!IsFloatingPointType(type)) { + boxNonDouble(ValueTypeFromMIRType(type), reg.gpr(), dest); + return; + } + + ScratchDoubleScope fpscratch(asMasm()); + FloatRegister scratch = fpscratch; + FloatRegister freg = reg.fpu(); + if (type == MIRType::Float32) { + convertFloat32ToDouble(freg, scratch); + freg = scratch; + } + boxDouble(freg, dest, scratch); +} +void MacroAssembler::moveValue(const ValueOperand& src, + const ValueOperand& dest) { + if (src == dest) { + return; + } + movePtr(src.valueReg(), dest.valueReg()); +} + +void MacroAssembler::moveValue(const Value& src, const ValueOperand& dest) { + if (!src.isGCThing()) { + ma_li(dest.valueReg(), ImmWord(src.asRawBits())); + return; + } + + writeDataRelocation(src); + movWithPatch(ImmWord(src.asRawBits()), dest.valueReg()); +} +void MacroAssembler::nearbyIntDouble(RoundingMode, FloatRegister, + FloatRegister) { + MOZ_CRASH("not supported on this platform"); +} +void MacroAssembler::nearbyIntFloat32(RoundingMode, FloatRegister, + FloatRegister) { + MOZ_CRASH("not supported on this platform"); +} + +void MacroAssembler::oolWasmTruncateCheckF32ToI32(FloatRegister input, + Register output, + TruncFlags flags, + wasm::BytecodeOffset off, + Label* rejoin) { + Label notNaN; + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + CompareIsNotNanF32(scratch, input, input); + ma_branch(¬NaN, Equal, scratch, Operand(1)); + wasmTrap(wasm::Trap::InvalidConversionToInteger, off); + bind(¬NaN); + + Label isOverflow; + const float two_31 = -float(INT32_MIN); + ScratchFloat32Scope fpscratch(*this); + if (flags & TRUNC_UNSIGNED) { + loadConstantFloat32(two_31 * 2, fpscratch); + ma_compareF32(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_branch(&isOverflow, Equal, scratch, Operand(1)); + loadConstantFloat32(-1.0f, fpscratch); + ma_compareF32(scratch, Assembler::DoubleGreaterThan, input, fpscratch); + ma_b(scratch, Imm32(1), rejoin, Equal); + } else { + loadConstantFloat32(two_31, fpscratch); + ma_compareF32(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_branch(&isOverflow, Equal, scratch, Operand(1)); + loadConstantFloat32(-two_31, fpscratch); + ma_compareF32(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_b(scratch, Imm32(1), rejoin, Equal); + } + bind(&isOverflow); + wasmTrap(wasm::Trap::IntegerOverflow, off); +} + +void MacroAssembler::oolWasmTruncateCheckF64ToI32(FloatRegister input, + Register output, + TruncFlags flags, + wasm::BytecodeOffset off, + Label* rejoin) { + Label notNaN; + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + CompareIsNotNanF64(scratch, input, input); + ma_branch(¬NaN, Equal, scratch, Operand(1)); + wasmTrap(wasm::Trap::InvalidConversionToInteger, off); + bind(¬NaN); + + Label isOverflow; + const double two_31 = -double(INT32_MIN); + ScratchDoubleScope fpscratch(*this); + if (flags & TRUNC_UNSIGNED) { + loadConstantDouble(two_31 * 2, fpscratch); + ma_compareF64(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_branch(&isOverflow, Equal, scratch, Operand(1)); + loadConstantDouble(-1.0, fpscratch); + ma_compareF64(scratch, Assembler::DoubleGreaterThan, input, fpscratch); + ma_b(scratch, Imm32(1), rejoin, Equal); + } else { + loadConstantDouble(two_31, fpscratch); + ma_compareF64(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_branch(&isOverflow, Equal, scratch, Operand(1)); + loadConstantDouble(-two_31 - 1, fpscratch); + ma_compareF64(scratch, Assembler::DoubleGreaterThan, input, fpscratch); + ma_b(scratch, Imm32(1), rejoin, Equal); + } + bind(&isOverflow); + wasmTrap(wasm::Trap::IntegerOverflow, off); +} + +void MacroAssembler::oolWasmTruncateCheckF32ToI64(FloatRegister input, + Register64 output, + TruncFlags flags, + wasm::BytecodeOffset off, + Label* rejoin) { + Label notNaN; + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + CompareIsNotNanF32(scratch, input, input); + ma_branch(¬NaN, Equal, scratch, Operand(1)); + wasmTrap(wasm::Trap::InvalidConversionToInteger, off); + bind(¬NaN); + + Label isOverflow; + const float two_63 = -float(INT64_MIN); + ScratchFloat32Scope fpscratch(*this); + if (flags & TRUNC_UNSIGNED) { + loadConstantFloat32(two_63 * 2, fpscratch); + ma_compareF32(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_branch(&isOverflow, Equal, scratch, Operand(1)); + loadConstantFloat32(-1.0f, fpscratch); + ma_compareF32(scratch, Assembler::DoubleGreaterThan, input, fpscratch); + ma_b(scratch, Imm32(1), rejoin, Equal); + } else { + loadConstantFloat32(two_63, fpscratch); + ma_compareF32(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_branch(&isOverflow, Equal, scratch, Operand(1)); + loadConstantFloat32(-two_63, fpscratch); + ma_compareF32(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_b(scratch, Imm32(1), rejoin, Equal); + } + bind(&isOverflow); + wasmTrap(wasm::Trap::IntegerOverflow, off); +} + +void MacroAssembler::oolWasmTruncateCheckF64ToI64(FloatRegister input, + Register64 output, + TruncFlags flags, + wasm::BytecodeOffset off, + Label* rejoin) { + Label notNaN; + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + CompareIsNotNanF64(scratch, input, input); + ma_branch(¬NaN, Equal, scratch, Operand(1)); + wasmTrap(wasm::Trap::InvalidConversionToInteger, off); + bind(¬NaN); + + Label isOverflow; + const double two_63 = -double(INT64_MIN); + ScratchDoubleScope fpscratch(*this); + if (flags & TRUNC_UNSIGNED) { + loadConstantDouble(two_63 * 2, fpscratch); + ma_compareF64(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_branch(&isOverflow, Equal, scratch, Operand(1)); + loadConstantDouble(-1.0, fpscratch); + ma_compareF64(scratch, Assembler::DoubleGreaterThan, input, fpscratch); + ma_b(scratch, Imm32(1), rejoin, Equal); + } else { + loadConstantDouble(two_63, fpscratch); + ma_compareF64(scratch, Assembler::DoubleGreaterThanOrEqual, input, + fpscratch); + ma_branch(&isOverflow, Equal, scratch, Operand(1)); + loadConstantDouble(-two_63, fpscratch); + ma_compareF64(scratch, Assembler::DoubleGreaterThan, input, fpscratch); + ma_b(scratch, Imm32(1), rejoin, Equal); + } + bind(&isOverflow); + wasmTrap(wasm::Trap::IntegerOverflow, off); +} +void MacroAssembler::patchCallToNop(uint8_t* call) { + uint32_t* p = reinterpret_cast<uint32_t*>(call) - 7; + *reinterpret_cast<Instr*>(p) = kNopByte; + *reinterpret_cast<Instr*>(p + 1) = kNopByte; + *reinterpret_cast<Instr*>(p + 2) = kNopByte; + *reinterpret_cast<Instr*>(p + 3) = kNopByte; + *reinterpret_cast<Instr*>(p + 4) = kNopByte; + *reinterpret_cast<Instr*>(p + 5) = kNopByte; + *reinterpret_cast<Instr*>(p + 6) = kNopByte; +} + +CodeOffset MacroAssembler::callWithPatch() { + BlockTrampolinePoolScope block_trampoline_pool(this, 2); + DEBUG_PRINTF("\tcallWithPatch\n"); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + int32_t imm32 = 1 * sizeof(uint32_t); + int32_t Hi20 = ((imm32 + 0x800) >> 12); + int32_t Lo12 = imm32 << 20 >> 20; + auipc(scratch, Hi20); // Read PC + Hi20 into scratch. + jalr(scratch, Lo12); // jump PC + Hi20 + Lo12 + DEBUG_PRINTF("\tret %d\n", currentOffset()); + return CodeOffset(currentOffset()); +} + +void MacroAssembler::patchCall(uint32_t callerOffset, uint32_t calleeOffset) { + DEBUG_PRINTF("\tpatchCall\n"); + BufferOffset call(callerOffset - 2 * sizeof(uint32_t)); + DEBUG_PRINTF("\tcallerOffset %d\n", callerOffset); + int32_t offset = BufferOffset(calleeOffset).getOffset() - call.getOffset(); + if (is_int32(offset)) { + Instruction* auipc_ = (Instruction*)editSrc(call); + Instruction* jalr_ = (Instruction*)editSrc( + BufferOffset(callerOffset - 1 * sizeof(uint32_t))); + DEBUG_PRINTF("\t%p %lu\n\t", auipc_, callerOffset - 2 * sizeof(uint32_t)); + disassembleInstr(auipc_->InstructionBits()); + DEBUG_PRINTF("\t%p %lu\n\t", jalr_, callerOffset - 1 * sizeof(uint32_t)); + disassembleInstr(jalr_->InstructionBits()); + DEBUG_PRINTF("\t\n"); + MOZ_ASSERT(IsJalr(jalr_->InstructionBits()) && + IsAuipc(auipc_->InstructionBits())); + MOZ_ASSERT(auipc_->RdValue() == jalr_->Rs1Value()); + int32_t Hi20 = (((int32_t)offset + 0x800) >> 12); + int32_t Lo12 = (int32_t)offset << 20 >> 20; + instr_at_put(call, SetAuipcOffset(Hi20, auipc_->InstructionBits())); + instr_at_put(BufferOffset(callerOffset - 1 * sizeof(uint32_t)), + SetJalrOffset(Lo12, jalr_->InstructionBits())); + } else { + MOZ_CRASH(); + } +} + +void MacroAssembler::patchFarJump(CodeOffset farJump, uint32_t targetOffset) { + uint32_t* u32 = reinterpret_cast<uint32_t*>( + editSrc(BufferOffset(farJump.offset() + 4 * kInstrSize))); + MOZ_ASSERT(*u32 == UINT32_MAX); + *u32 = targetOffset - farJump.offset(); +} + +void MacroAssembler::patchNearAddressMove(CodeLocationLabel loc, + CodeLocationLabel target) { + PatchDataWithValueCheck(loc, ImmPtr(target.raw()), ImmPtr(nullptr)); +} + +void MacroAssembler::patchNopToCall(uint8_t* call, uint8_t* target) { + uint32_t* p = reinterpret_cast<uint32_t*>(call) - 7; + Assembler::WriteLoad64Instructions((Instruction*)p, ScratchRegister, + (uint64_t)target); + DEBUG_PRINTF("\tpatchNopToCall %lu %lu\n", (uint64_t)target, + ExtractLoad64Value((Instruction*)p)); + MOZ_ASSERT(ExtractLoad64Value((Instruction*)p) == (uint64_t)target); + Instr jalr_ = JALR | (ra.code() << kRdShift) | (0x0 << kFunct3Shift) | + (ScratchRegister.code() << kRs1Shift) | (0x0 << kImm12Shift); + *reinterpret_cast<Instr*>(p + 6) = jalr_; +} +void MacroAssembler::Pop(Register reg) { + ma_pop(reg); + adjustFrame(-int32_t(sizeof(intptr_t))); +} + +void MacroAssembler::Pop(FloatRegister f) { + ma_pop(f); + adjustFrame(-int32_t(sizeof(double))); +} + +void MacroAssembler::Pop(const ValueOperand& val) { + popValue(val); + adjustFrame(-int32_t(sizeof(Value))); +} + +void MacroAssembler::PopRegsInMaskIgnore(LiveRegisterSet set, + LiveRegisterSet ignore) { + int32_t diff = + set.gprs().size() * sizeof(intptr_t) + set.fpus().getPushSizeInBytes(); + const int32_t reserved = diff; + + for (GeneralRegisterBackwardIterator iter(set.gprs()); iter.more(); ++iter) { + diff -= sizeof(intptr_t); + if (!ignore.has(*iter)) { + loadPtr(Address(StackPointer, diff), *iter); + } + } + +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + for (FloatRegisterBackwardIterator iter(set.fpus().reduceSetForPush()); + iter.more(); ++iter) { + diff -= sizeof(double); + if (!ignore.has(*iter)) { + loadDouble(Address(StackPointer, diff), *iter); + } + } + MOZ_ASSERT(diff == 0); + freeStack(reserved); +} + +void MacroAssembler::pushReturnAddress() { push(ra); } + +void MacroAssembler::popReturnAddress() { pop(ra); } +void MacroAssembler::PopStackPtr() { + loadPtr(Address(StackPointer, 0), StackPointer); + adjustFrame(-int32_t(sizeof(intptr_t))); +} +void MacroAssembler::freeStackTo(uint32_t framePushed) { + MOZ_ASSERT(framePushed <= framePushed_); + ma_sub64(StackPointer, FramePointer, Imm32(framePushed)); + framePushed_ = framePushed; +} +void MacroAssembler::PushBoxed(FloatRegister reg) { + subFromStackPtr(Imm32(sizeof(double))); + boxDouble(reg, Address(getStackPointer(), 0)); + adjustFrame(sizeof(double)); +} + +void MacroAssembler::Push(Register reg) { + ma_push(reg); + adjustFrame(int32_t(sizeof(intptr_t))); +} + +void MacroAssembler::Push(const Imm32 imm) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + ma_push(scratch); + adjustFrame(int32_t(sizeof(intptr_t))); +} + +void MacroAssembler::Push(const ImmWord imm) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + ma_push(scratch); + adjustFrame(int32_t(sizeof(intptr_t))); +} + +void MacroAssembler::Push(const ImmPtr imm) { + Push(ImmWord(uintptr_t(imm.value))); +} + +void MacroAssembler::Push(const ImmGCPtr ptr) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, ptr); + ma_push(scratch); + adjustFrame(int32_t(sizeof(intptr_t))); +} + +void MacroAssembler::Push(FloatRegister f) { + ma_push(f); + adjustFrame(int32_t(sizeof(double))); +} + +void MacroAssembler::PushRegsInMask(LiveRegisterSet set) { + int32_t diff = + set.gprs().size() * sizeof(intptr_t) + set.fpus().getPushSizeInBytes(); + const int32_t reserved = diff; + + reserveStack(reserved); + for (GeneralRegisterBackwardIterator iter(set.gprs()); iter.more(); ++iter) { + diff -= sizeof(intptr_t); + storePtr(*iter, Address(StackPointer, diff)); + } + +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + for (FloatRegisterBackwardIterator iter(set.fpus().reduceSetForPush()); + iter.more(); ++iter) { + diff -= sizeof(double); + storeDouble(*iter, Address(StackPointer, diff)); + } + MOZ_ASSERT(diff == 0); +} + +void MacroAssembler::roundFloat32ToInt32(FloatRegister src, Register dest, + FloatRegister temp, Label* fail) { + JitSpew(JitSpew_Codegen, "[ %s", __FUNCTION__); + ScratchDoubleScope fscratch(*this); + Label negative, done; + // Branch to a slow path if input < 0.0 due to complicated rounding rules. + // Note that Fcmp with NaN unsets the negative flag. + { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + fmv_x_w(scratch, src); + ma_branch(fail, Equal, scratch, Operand(int32_t(0x80000000))); + fmv_w_x(temp, zero); + ma_compareF32(scratch, DoubleLessThan, src, temp); + ma_branch(&negative, Equal, scratch, Operand(1)); + } + // Handle the simple case of a positive input, and also -0 and NaN. + // Rounding proceeds with consideration of the fractional part of the input: + // 1. If > 0.5, round to integer with higher absolute value (so, up). + // 2. If < 0.5, round to integer with lower absolute value (so, down). + // 3. If = 0.5, round to +Infinity (so, up). + { + // Convert to signed 32-bit integer, rounding halfway cases away from zero. + // In the case of overflow, the output is saturated. + // In the case of NaN and -0, the output is zero. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + RoundFloatingPointToInteger( + dest, src, scratch, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_s(dst, src, RMM); + }, + false); + ma_b(scratch, Imm32(1), fail, NotEqual); + jump(&done); + } + + // Handle the complicated case of a negative input. + // Rounding proceeds with consideration of the fractional part of the input: + // 1. If > 0.5, round to integer with higher absolute value (so, down). + // 2. If < 0.5, round to integer with lower absolute value (so, up). + // 3. If = 0.5, round to +Infinity (so, up). + bind(&negative); + { + // Inputs in [-0.5, 0) need 0.5 added; other negative inputs need + // the biggest double less than 0.5. + Label join; + loadConstantFloat32(GetBiggestNumberLessThan(0.5), temp); + loadConstantFloat32(-0.5, fscratch); + branchFloat(Assembler::DoubleLessThan, src, fscratch, &join); + loadConstantFloat32(0.5, temp); + bind(&join); + addFloat32(src, temp); + // Round all values toward -Infinity. + // In the case of overflow, the output is saturated. + // NaN and -0 are already handled by the "positive number" path above. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + RoundFloatingPointToInteger( + dest, temp, scratch, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_s(dst, src, RDN); + }, + false); + ma_b(scratch, Imm32(1), fail, NotEqual); + // If output is zero, then the actual result is -0. Fail. + branchTest32(Assembler::Zero, dest, dest, fail); + } + bind(&done); + JitSpew(JitSpew_Codegen, "]"); +} + +void MacroAssembler::roundDoubleToInt32(FloatRegister src, Register dest, + FloatRegister temp, Label* fail) { + JitSpew(JitSpew_Codegen, "[ %s", __FUNCTION__); + + ScratchDoubleScope fscratch(*this); + Label negative, done; + // Branch to a slow path if input < 0.0 due to complicated rounding rules. + // Note that Fcmp with NaN unsets the negative flag. + { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + fmv_x_d(scratch, src); + ma_branch(fail, Equal, scratch, Operand(0x8000000000000000)); + fmv_d_x(temp, zero); + ma_compareF64(scratch, DoubleLessThan, src, temp); + ma_branch(&negative, Equal, scratch, Operand(1)); + } + // Handle the simple case of a positive input, and also -0 and NaN. + // Rounding proceeds with consideration of the fractional part of the input: + // 1. If > 0.5, round to integer with higher absolute value (so, up). + // 2. If < 0.5, round to integer with lower absolute value (so, down). + // 3. If = 0.5, round to +Infinity (so, up). + { + // Convert to signed 32-bit integer, rounding halfway cases away from zero. + // In the case of overflow, the output is saturated. + // In the case of NaN and -0, the output is zero. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + RoundFloatingPointToInteger( + dest, src, scratch, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_d(dst, src, RMM); + }, + false); + ma_b(scratch, Imm32(1), fail, NotEqual); + jump(&done); + } + + // Handle the complicated case of a negative input. + // Rounding proceeds with consideration of the fractional part of the input: + // 1. If > 0.5, round to integer with higher absolute value (so, down). + // 2. If < 0.5, round to integer with lower absolute value (so, up). + // 3. If = 0.5, round to +Infinity (so, up). + bind(&negative); + { + // Inputs in [-0.5, 0) need 0.5 added; other negative inputs need + // the biggest double less than 0.5. + Label join; + loadConstantDouble(GetBiggestNumberLessThan(0.5), temp); + loadConstantDouble(-0.5, fscratch); + branchDouble(Assembler::DoubleLessThan, src, fscratch, &join); + loadConstantDouble(0.5, temp); + bind(&join); + addDouble(src, temp); + // Round all values toward -Infinity. + // In the case of overflow, the output is saturated. + // NaN and -0 are already handled by the "positive number" path above. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + RoundFloatingPointToInteger( + dest, temp, scratch, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_d(dst, src, RDN); + }, + false); + ma_b(scratch, Imm32(1), fail, NotEqual); + // If output is zero, then the actual result is -0. Fail. + branchTest32(Assembler::Zero, dest, dest, fail); + } + bind(&done); + JitSpew(JitSpew_Codegen, "]"); +} + +void MacroAssembler::setupUnalignedABICall(Register scratch) { + MOZ_ASSERT(!IsCompilingWasm(), "wasm should only use aligned ABI calls"); + setupNativeABICall(); + dynamicAlignment_ = true; + + or_(scratch, StackPointer, zero); + + // Force sp to be aligned + asMasm().subPtr(Imm32(sizeof(uintptr_t)), StackPointer); + ma_and(StackPointer, StackPointer, Imm32(~(ABIStackAlignment - 1))); + storePtr(scratch, Address(StackPointer, 0)); +} +void MacroAssembler::shiftIndex32AndAdd(Register indexTemp32, int shift, + Register pointer) { + if (IsShiftInScaleRange(shift)) { + computeEffectiveAddress( + BaseIndex(pointer, indexTemp32, ShiftToScale(shift)), pointer); + return; + } + lshift32(Imm32(shift), indexTemp32); + addPtr(indexTemp32, pointer); +} +void MacroAssembler::speculationBarrier() { MOZ_CRASH(); } +void MacroAssembler::storeRegsInMask(LiveRegisterSet set, Address dest, + Register) { + FloatRegisterSet fpuSet(set.fpus().reduceSetForPush()); + unsigned numFpu = fpuSet.size(); + int32_t diffF = fpuSet.getPushSizeInBytes(); + int32_t diffG = set.gprs().size() * sizeof(intptr_t); + + MOZ_ASSERT(dest.offset >= diffG + diffF); + + for (GeneralRegisterBackwardIterator iter(set.gprs()); iter.more(); ++iter) { + diffG -= sizeof(intptr_t); + dest.offset -= sizeof(intptr_t); + storePtr(*iter, dest); + } + MOZ_ASSERT(diffG == 0); + +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + for (FloatRegisterBackwardIterator iter(fpuSet); iter.more(); ++iter) { + FloatRegister reg = *iter; + diffF -= reg.size(); + numFpu -= 1; + dest.offset -= reg.size(); + if (reg.isDouble()) { + storeDouble(reg, dest); + } else if (reg.isSingle()) { + storeFloat32(reg, dest); + } else { + MOZ_CRASH("Unknown register type."); + } + } + MOZ_ASSERT(numFpu == 0); + diffF -= diffF % sizeof(uintptr_t); + MOZ_ASSERT(diffF == 0); +} +void MacroAssembler::truncDoubleToInt32(FloatRegister src, Register dest, + Label* fail) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Label zeroCase, done; + // Convert scalar to signed 32-bit fixed-point, rounding toward zero. + // In the case of overflow, the output is saturated. + // In the case of NaN and -0, the output is zero. + RoundFloatingPointToInteger( + dest, src, scratch, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_d(dst, src, RTZ); + }, + false); + ma_b(scratch, Imm32(1), fail, NotEqual); + // If the output was zero, worry about special cases. + branch32(Assembler::Equal, dest, Imm32(0), &zeroCase); + jump(&done); + // Handle the case of a zero output: + // 1. The input may have been NaN, requiring a failure. + // 2. The input may have been in (-1,-0], requiring a failure. + // 3. +0, return 0. + { + bind(&zeroCase); + + // If input is a negative number that truncated to zero, the real + // output should be the non-integer -0. + // The use of "lt" instead of "lo" also catches unordered NaN input. + ScratchDoubleScope fscratch(*this); + fmv_d_x(fscratch, zero); + ma_compareF64(scratch, DoubleLessThan, src, fscratch); + ma_b(scratch, Imm32(1), fail, Equal); + + // Check explicitly for -0, bitwise. + fmv_x_d(dest, src); + branchTestPtr(Assembler::Signed, dest, dest, fail); + movePtr(ImmWord(0), dest); + } + + bind(&done); +} +void MacroAssembler::truncFloat32ToInt32(FloatRegister src, Register dest, + Label* fail) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Label zeroCase, done; + // Convert scalar to signed 32-bit fixed-point, rounding toward zero. + // In the case of overflow, the output is saturated. + // In the case of NaN and -0, the output is zero. + RoundFloatingPointToInteger( + dest, src, scratch, + [](MacroAssemblerRiscv64* masm, Register dst, FPURegister src) { + masm->fcvt_w_s(dst, src, RTZ); + }, + false); + ma_b(scratch, Imm32(1), fail, NotEqual); + // If the output was zero, worry about special cases. + branch32(Assembler::Equal, dest, Imm32(0), &zeroCase); + jump(&done); + // Handle the case of a zero output: + // 1. The input may have been NaN, requiring a failure. + // 2. The input may have been in (-1,-0], requiring a failure. + // 3. +0, return 0. + { + bind(&zeroCase); + + // If input is a negative number that truncated to zero, the real + // output should be the non-integer -0. + // The use of "lt" instead of "lo" also catches unordered NaN input. + ScratchDoubleScope fscratch(*this); + fmv_w_x(fscratch, zero); + ma_compareF32(scratch, DoubleLessThan, src, fscratch); + ma_b(scratch, Imm32(1), fail, Equal); + + // Check explicitly for -0, bitwise. + fmv_x_w(dest, src); + branchTestPtr(Assembler::Signed, dest, dest, fail); + movePtr(ImmWord(0), dest); + } + + bind(&done); +} +void MacroAssembler::wasmAtomicEffectOp(const wasm::MemoryAccessDesc& access, + AtomicOp op, Register value, + const Address& mem, Register valueTemp, + Register offsetTemp, + Register maskTemp) { + AtomicEffectOp(*this, &access, access.type(), access.sync(), op, mem, value, + valueTemp, offsetTemp, maskTemp); +} + +void MacroAssembler::wasmAtomicEffectOp(const wasm::MemoryAccessDesc& access, + AtomicOp op, Register value, + const BaseIndex& mem, + Register valueTemp, Register offsetTemp, + Register maskTemp) { + AtomicEffectOp(*this, &access, access.type(), access.sync(), op, mem, value, + valueTemp, offsetTemp, maskTemp); +} +template <typename T> +static void WasmAtomicExchange64(MacroAssembler& masm, + const wasm::MemoryAccessDesc& access, + const T& mem, Register64 value, + Register64 output) { + AtomicExchange64(masm, &access, access.sync(), mem, value, output); +} + +void MacroAssembler::wasmAtomicExchange64(const wasm::MemoryAccessDesc& access, + const Address& mem, Register64 src, + Register64 output) { + WasmAtomicExchange64(*this, access, mem, src, output); +} + +void MacroAssembler::wasmAtomicExchange64(const wasm::MemoryAccessDesc& access, + const BaseIndex& mem, Register64 src, + Register64 output) { + WasmAtomicExchange64(*this, access, mem, src, output); +} +void MacroAssembler::wasmAtomicExchange(const wasm::MemoryAccessDesc& access, + const Address& mem, Register value, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register output) { + AtomicExchange(*this, &access, access.type(), access.sync(), mem, value, + valueTemp, offsetTemp, maskTemp, output); +} + +void MacroAssembler::wasmAtomicExchange(const wasm::MemoryAccessDesc& access, + const BaseIndex& mem, Register value, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register output) { + AtomicExchange(*this, &access, access.type(), access.sync(), mem, value, + valueTemp, offsetTemp, maskTemp, output); +} +void MacroAssembler::wasmAtomicFetchOp64(const wasm::MemoryAccessDesc& access, + AtomicOp op, Register64 value, + const Address& mem, Register64 temp, + Register64 output) { + AtomicFetchOp64(*this, &access, access.sync(), op, value, mem, temp, output); +} +void MacroAssembler::wasmAtomicFetchOp64(const wasm::MemoryAccessDesc& access, + AtomicOp op, Register64 value, + const BaseIndex& mem, Register64 temp, + Register64 output) { + AtomicFetchOp64(*this, &access, access.sync(), op, value, mem, temp, output); +} + +void MacroAssembler::atomicFetchOp64(const Synchronization& sync, AtomicOp op, + Register64 value, const Address& mem, + Register64 temp, Register64 output) { + AtomicFetchOp64(*this, nullptr, sync, op, value, mem, temp, output); +} + +void MacroAssembler::atomicFetchOp64(const Synchronization& sync, AtomicOp op, + Register64 value, const BaseIndex& mem, + Register64 temp, Register64 output) { + AtomicFetchOp64(*this, nullptr, sync, op, value, mem, temp, output); +} + +void MacroAssembler::atomicEffectOp64(const Synchronization& sync, AtomicOp op, + Register64 value, const Address& mem, + Register64 temp) { + AtomicFetchOp64(*this, nullptr, sync, op, value, mem, temp, temp); +} + +void MacroAssembler::atomicEffectOp64(const Synchronization& sync, AtomicOp op, + Register64 value, const BaseIndex& mem, + Register64 temp) { + AtomicFetchOp64(*this, nullptr, sync, op, value, mem, temp, temp); +} +void MacroAssembler::wasmAtomicFetchOp(const wasm::MemoryAccessDesc& access, + AtomicOp op, Register value, + const Address& mem, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register output) { + AtomicFetchOp(*this, &access, access.type(), access.sync(), op, mem, value, + valueTemp, offsetTemp, maskTemp, output); +} + +void MacroAssembler::wasmAtomicFetchOp(const wasm::MemoryAccessDesc& access, + AtomicOp op, Register value, + const BaseIndex& mem, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register output) { + AtomicFetchOp(*this, &access, access.type(), access.sync(), op, mem, value, + valueTemp, offsetTemp, maskTemp, output); +} +void MacroAssembler::wasmBoundsCheck32(Condition cond, Register index, + Register boundsCheckLimit, Label* ok) { + ma_b(index, boundsCheckLimit, ok, cond); +} + +void MacroAssembler::wasmBoundsCheck32(Condition cond, Register index, + Address boundsCheckLimit, Label* ok) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + load32(boundsCheckLimit, scratch2); + ma_b(index, Register(scratch2), ok, cond); +} + +void MacroAssembler::wasmBoundsCheck64(Condition cond, Register64 index, + Register64 boundsCheckLimit, Label* ok) { + ma_b(index.reg, boundsCheckLimit.reg, ok, cond); +} + +void MacroAssembler::wasmBoundsCheck64(Condition cond, Register64 index, + Address boundsCheckLimit, Label* ok) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + loadPtr(boundsCheckLimit, scratch2); + ma_b(index.reg, scratch2, ok, cond); +} + +void MacroAssembler::wasmCompareExchange64(const wasm::MemoryAccessDesc& access, + const Address& mem, + Register64 expect, + Register64 replace, + Register64 output) { + CompareExchange64(*this, &access, access.sync(), mem, expect, replace, + output); +} + +void MacroAssembler::wasmCompareExchange64(const wasm::MemoryAccessDesc& access, + const BaseIndex& mem, + Register64 expect, + Register64 replace, + Register64 output) { + CompareExchange64(*this, &access, access.sync(), mem, expect, replace, + output); +} + +template <typename T> +static void CompareExchange(MacroAssembler& masm, + const wasm::MemoryAccessDesc* access, + Scalar::Type type, const Synchronization& sync, + const T& mem, Register oldval, Register newval, + Register valueTemp, Register offsetTemp, + Register maskTemp, Register output) { + bool signExtend = Scalar::isSignedIntType(type); + unsigned nbytes = Scalar::byteSize(type); + + switch (nbytes) { + case 1: + case 2: + break; + case 4: + MOZ_ASSERT(valueTemp == InvalidReg); + MOZ_ASSERT(offsetTemp == InvalidReg); + MOZ_ASSERT(maskTemp == InvalidReg); + break; + default: + MOZ_CRASH(); + } + + Label again, end; + UseScratchRegisterScope temps(&masm); + Register SecondScratchReg = temps.Acquire(); + masm.computeEffectiveAddress(mem, SecondScratchReg); + + if (nbytes == 4) { + masm.memoryBarrierBefore(sync); + masm.bind(&again); + + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_w(true, true, output, SecondScratchReg); + masm.ma_b(output, oldval, &end, Assembler::NotEqual, ShortJump); + masm.mv(ScratchRegister, newval); + masm.sc_w(true, true, ScratchRegister, SecondScratchReg, ScratchRegister); + masm.ma_b(ScratchRegister, ScratchRegister, &again, Assembler::NonZero, + ShortJump); + + masm.memoryBarrierAfter(sync); + masm.bind(&end); + + return; + } + + masm.andi(offsetTemp, SecondScratchReg, 3); + masm.subPtr(offsetTemp, SecondScratchReg); +#if !MOZ_LITTLE_ENDIAN() + masm.as_xori(offsetTemp, offsetTemp, 3); +#endif + masm.slli(offsetTemp, offsetTemp, 3); + masm.ma_li(maskTemp, Imm32(UINT32_MAX >> ((4 - nbytes) * 8))); + masm.sll(maskTemp, maskTemp, offsetTemp); + masm.nor(maskTemp, zero, maskTemp); + + masm.memoryBarrierBefore(sync); + + masm.bind(&again); + + if (access) { + masm.append(*access, wasm::TrapMachineInsn::Atomic, + FaultingCodeOffset(masm.currentOffset())); + } + + masm.lr_w(true, true, ScratchRegister, SecondScratchReg); + + masm.srl(output, ScratchRegister, offsetTemp); + + switch (nbytes) { + case 1: + if (signExtend) { + masm.SignExtendByte(valueTemp, oldval); + masm.SignExtendByte(output, output); + } else { + masm.andi(valueTemp, oldval, 0xff); + masm.andi(output, output, 0xff); + } + break; + case 2: + if (signExtend) { + masm.SignExtendShort(valueTemp, oldval); + masm.SignExtendShort(output, output); + } else { + masm.andi(valueTemp, oldval, 0xffff); + masm.andi(output, output, 0xffff); + } + break; + } + + masm.ma_b(output, valueTemp, &end, Assembler::NotEqual, ShortJump); + + masm.sll(valueTemp, newval, offsetTemp); + masm.and_(ScratchRegister, ScratchRegister, maskTemp); + masm.or_(ScratchRegister, ScratchRegister, valueTemp); + masm.sc_w(true, true, ScratchRegister, SecondScratchReg, ScratchRegister); + + masm.ma_b(ScratchRegister, ScratchRegister, &again, Assembler::NonZero, + ShortJump); + + masm.memoryBarrierAfter(sync); + + masm.bind(&end); +} + +void MacroAssembler::compareExchange(Scalar::Type type, + const Synchronization& sync, + const Address& mem, Register oldval, + Register newval, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register output) { + CompareExchange(*this, nullptr, type, sync, mem, oldval, newval, valueTemp, + offsetTemp, maskTemp, output); +} + +void MacroAssembler::compareExchange(Scalar::Type type, + const Synchronization& sync, + const BaseIndex& mem, Register oldval, + Register newval, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register output) { + CompareExchange(*this, nullptr, type, sync, mem, oldval, newval, valueTemp, + offsetTemp, maskTemp, output); +} + +void MacroAssembler::wasmCompareExchange(const wasm::MemoryAccessDesc& access, + const Address& mem, Register oldval, + Register newval, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register output) { + CompareExchange(*this, &access, access.type(), access.sync(), mem, oldval, + newval, valueTemp, offsetTemp, maskTemp, output); +} + +void MacroAssembler::wasmCompareExchange(const wasm::MemoryAccessDesc& access, + const BaseIndex& mem, Register oldval, + Register newval, Register valueTemp, + Register offsetTemp, Register maskTemp, + Register output) { + CompareExchange(*this, &access, access.type(), access.sync(), mem, oldval, + newval, valueTemp, offsetTemp, maskTemp, output); +} + +void MacroAssembler::wasmLoad(const wasm::MemoryAccessDesc& access, + Register memoryBase, Register ptr, + Register ptrScratch, AnyRegister output) { + wasmLoadImpl(access, memoryBase, ptr, ptrScratch, output, InvalidReg); +} + +void MacroAssembler::wasmLoadI64(const wasm::MemoryAccessDesc& access, + Register memoryBase, Register ptr, + Register ptrScratch, Register64 output) { + wasmLoadI64Impl(access, memoryBase, ptr, ptrScratch, output, InvalidReg); +} + +void MacroAssembler::wasmStore(const wasm::MemoryAccessDesc& access, + AnyRegister value, Register memoryBase, + Register ptr, Register ptrScratch) { + wasmStoreImpl(access, value, memoryBase, ptr, ptrScratch, InvalidReg); +} + +void MacroAssembler::wasmStoreI64(const wasm::MemoryAccessDesc& access, + Register64 value, Register memoryBase, + Register ptr, Register ptrScratch) { + wasmStoreI64Impl(access, value, memoryBase, ptr, ptrScratch, InvalidReg); +} + +void MacroAssemblerRiscv64::Clear_if_nan_d(Register rd, FPURegister fs) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Label no_nan; + feq_d(ScratchRegister, fs, fs); + bnez(ScratchRegister, &no_nan); + mv(rd, zero_reg); + bind(&no_nan); +} + +void MacroAssemblerRiscv64::Clear_if_nan_s(Register rd, FPURegister fs) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Label no_nan; + feq_s(ScratchRegister, fs, fs); + bnez(ScratchRegister, &no_nan); + mv(rd, zero_reg); + bind(&no_nan); +} + +void MacroAssembler::wasmTruncateDoubleToInt32(FloatRegister input, + Register output, + bool isSaturating, + Label* oolEntry) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Trunc_w_d(output, input, ScratchRegister); + if (isSaturating) { + Clear_if_nan_d(output, input); + } else { + ma_b(ScratchRegister, Imm32(1), oolEntry, Assembler::NotEqual); + } +} + +void MacroAssembler::wasmTruncateDoubleToInt64( + FloatRegister input, Register64 output, bool isSaturating, Label* oolEntry, + Label* oolRejoin, FloatRegister tempDouble) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Trunc_l_d(output.reg, input, ScratchRegister); + if (isSaturating) { + bind(oolRejoin); + Clear_if_nan_d(output.reg, input); + } else { + ma_b(ScratchRegister, Imm32(1), oolEntry, Assembler::NotEqual); + } +} + +void MacroAssembler::wasmTruncateDoubleToUInt32(FloatRegister input, + Register output, + bool isSaturating, + Label* oolEntry) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Trunc_uw_d(output, input, ScratchRegister); + if (isSaturating) { + Clear_if_nan_d(output, input); + } else { + ma_b(ScratchRegister, Imm32(1), oolEntry, Assembler::NotEqual); + } +} + +void MacroAssembler::wasmTruncateDoubleToUInt64( + FloatRegister input, Register64 output, bool isSaturating, Label* oolEntry, + Label* oolRejoin, FloatRegister tempDouble) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Trunc_ul_d(output.reg, input, ScratchRegister); + if (isSaturating) { + bind(oolRejoin); + Clear_if_nan_d(output.reg, input); + } else { + ma_b(ScratchRegister, Imm32(1), oolEntry, Assembler::NotEqual); + } +} + +void MacroAssembler::wasmTruncateFloat32ToInt32(FloatRegister input, + Register output, + bool isSaturating, + Label* oolEntry) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Trunc_w_s(output, input, ScratchRegister); + if (isSaturating) { + Clear_if_nan_s(output, input); + } else { + ma_b(ScratchRegister, Imm32(1), oolEntry, Assembler::NotEqual); + } +} + +void MacroAssembler::wasmTruncateFloat32ToInt64( + FloatRegister input, Register64 output, bool isSaturating, Label* oolEntry, + Label* oolRejoin, FloatRegister tempFloat) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Trunc_l_s(output.reg, input, ScratchRegister); + + if (isSaturating) { + bind(oolRejoin); + Clear_if_nan_s(output.reg, input); + } else { + ma_b(ScratchRegister, Imm32(1), oolEntry, Assembler::NotEqual); + } +} + +void MacroAssembler::wasmTruncateFloat32ToUInt32(FloatRegister input, + Register output, + bool isSaturating, + Label* oolEntry) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Trunc_uw_s(output, input, ScratchRegister); + if (isSaturating) { + Clear_if_nan_s(output, input); + } else { + ma_b(ScratchRegister, Imm32(1), oolEntry, Assembler::NotEqual); + } +} + +void MacroAssembler::wasmTruncateFloat32ToUInt64( + FloatRegister input, Register64 output, bool isSaturating, Label* oolEntry, + Label* oolRejoin, FloatRegister tempFloat) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + Trunc_ul_s(output.reg, input, ScratchRegister); + + if (isSaturating) { + bind(oolRejoin); + Clear_if_nan_s(output.reg, input); + } else { + ma_b(ScratchRegister, Imm32(1), oolEntry, Assembler::NotEqual); + } +} + +// TODO(riscv64): widenInt32 should be nop? +void MacroAssembler::widenInt32(Register r) { + move32To64SignExtend(r, Register64(r)); +} + +#ifdef ENABLE_WASM_TAIL_CALLS +void MacroAssembler::wasmMarkSlowCall() { mv(ra, ra); } + +const int32_t SlowCallMarker = 0x8093; // addi ra, ra, 0 + +void MacroAssembler::wasmCheckSlowCallsite(Register ra_, Label* notSlow, + Register temp1, Register temp2) { + MOZ_ASSERT(ra_ != temp2); + load32(Address(ra_, 0), temp2); + branch32(Assembler::NotEqual, temp2, Imm32(SlowCallMarker), notSlow); +} +#endif // ENABLE_WASM_TAIL_CALLS +//}}} check_macroassembler_style + +// This method generates lui, dsll and ori instruction block that can be +// modified by UpdateLoad64Value, either during compilation (eg. +// Assembler::bind), or during execution (eg. jit::PatchJump). +void MacroAssemblerRiscv64::ma_liPatchable(Register dest, Imm32 imm) { + m_buffer.ensureSpace(2 * sizeof(uint32_t)); + int64_t value = imm.value; + int64_t high_20 = ((value + 0x800) >> 12); + int64_t low_12 = value << 52 >> 52; + lui(dest, high_20); + addi(dest, dest, low_12); +} + +void MacroAssemblerRiscv64::ma_liPatchable(Register dest, ImmPtr imm) { + return ma_liPatchable(dest, ImmWord(uintptr_t(imm.value))); +} + +void MacroAssemblerRiscv64::ma_liPatchable(Register dest, ImmWord imm, + LiFlags flags) { + DEBUG_PRINTF("\tma_liPatchable\n"); + if (Li64 == flags) { + li_constant(dest, imm.value); + } else { + li_ptr(dest, imm.value); + } +} + +void MacroAssemblerRiscv64::ma_li(Register dest, ImmGCPtr ptr) { + BlockTrampolinePoolScope block_trampoline_pool(this, 6); + writeDataRelocation(ptr); + ma_liPatchable(dest, ImmPtr(ptr.value)); +} +void MacroAssemblerRiscv64::ma_li(Register dest, Imm32 imm) { + RV_li(dest, imm.value); +} +void MacroAssemblerRiscv64::ma_li(Register dest, Imm64 imm) { + RV_li(dest, imm.value); +} +void MacroAssemblerRiscv64::ma_li(Register dest, CodeLabel* label) { + DEBUG_PRINTF("[ %s\n", __FUNCTION__); + BlockTrampolinePoolScope block_trampoline_pool(this, 7); + BufferOffset bo = m_buffer.nextOffset(); + JitSpew(JitSpew_Codegen, ".load CodeLabel %p", label); + ma_liPatchable(dest, ImmWord(/* placeholder */ 0)); + label->patchAt()->bind(bo.getOffset()); + label->setLinkMode(CodeLabel::MoveImmediate); + DEBUG_PRINTF("]\n"); +} +void MacroAssemblerRiscv64::ma_li(Register dest, ImmWord imm) { + RV_li(dest, imm.value); +} + +// Shortcut for when we know we're transferring 32 bits of data. +void MacroAssemblerRiscv64::ma_pop(Register r) { + ld(r, StackPointer, 0); + addi(StackPointer, StackPointer, sizeof(intptr_t)); +} + +void MacroAssemblerRiscv64::ma_push(Register r) { + if (r == sp) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + // Pushing sp requires one more instruction. + mv(ScratchRegister, sp); + r = ScratchRegister; + } + + addi(StackPointer, StackPointer, (int32_t) - sizeof(intptr_t)); + sd(r, StackPointer, 0); +} + +// multiplies. For now, there are only few that we care about. +void MacroAssemblerRiscv64::ma_mul32TestOverflow(Register rd, Register rj, + Register rk, Label* overflow) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + MulOverflow32(rd, rj, rk, ScratchRegister); + ma_b(ScratchRegister, Register(zero), overflow, Assembler::NotEqual); +} +void MacroAssemblerRiscv64::ma_mul32TestOverflow(Register rd, Register rj, + Imm32 imm, Label* overflow) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + MulOverflow32(rd, rj, Operand(imm.value), ScratchRegister); + ma_b(ScratchRegister, Register(zero), overflow, Assembler::NotEqual); +} + +void MacroAssemblerRiscv64::ma_mulPtrTestOverflow(Register rd, Register rj, + Register rk, + Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(rd != scratch); + + if (rd == rj) { + or_(scratch, rj, zero); + rj = scratch; + rk = (rd == rk) ? rj : rk; + } else if (rd == rk) { + or_(scratch, rk, zero); + rk = scratch; + } + + mul(rd, rj, rk); + mulh(scratch, rj, rk); + srai(scratch2, rd, 63); + ma_b(scratch, Register(scratch2), overflow, Assembler::NotEqual); +} + +// MulOverflow32 sets overflow register to zero if no overflow occured +void MacroAssemblerRiscv64::MulOverflow32(Register dst, Register left, + const Operand& right, + Register overflow) { + UseScratchRegisterScope temps(this); + BlockTrampolinePoolScope block_trampoline_pool(this, 11); + Register right_reg; + Register scratch = temps.Acquire(); + Register scratch2 = temps.Acquire(); + if (right.is_imm()) { + ma_li(scratch, right.immediate()); + right_reg = scratch; + } else { + MOZ_ASSERT(right.is_reg()); + right_reg = right.rm(); + } + + MOZ_ASSERT(left != scratch2 && right_reg != scratch2 && dst != scratch2 && + overflow != scratch2); + MOZ_ASSERT(overflow != left && overflow != right_reg); + sext_w(overflow, left); + sext_w(scratch2, right_reg); + + mul(overflow, overflow, scratch2); + sext_w(dst, overflow); + xor_(overflow, overflow, dst); +} + +int32_t MacroAssemblerRiscv64::GetOffset(int32_t offset, Label* L, + OffsetSize bits) { + if (L) { + offset = branch_offset_helper(L, bits); + } else { + MOZ_ASSERT(is_intn(offset, bits)); + } + return offset; +} + +bool MacroAssemblerRiscv64::CalculateOffset(Label* L, int32_t* offset, + OffsetSize bits) { + if (!is_near(L, bits)) return false; + *offset = GetOffset(*offset, L, bits); + return true; +} + +void MacroAssemblerRiscv64::BranchShortHelper(int32_t offset, Label* L) { + MOZ_ASSERT(L == nullptr || offset == 0); + offset = GetOffset(offset, L, OffsetSize::kOffset21); + Assembler::j(offset); +} + +bool MacroAssemblerRiscv64::BranchShortHelper(int32_t offset, Label* L, + Condition cond, Register rs, + const Operand& rt) { + MOZ_ASSERT(L == nullptr || offset == 0); + MOZ_ASSERT(rt.is_reg() || rt.is_imm()); + UseScratchRegisterScope temps(this); + Register scratch = Register(); + if (rt.is_imm()) { + scratch = temps.Acquire(); + ma_li(scratch, Imm64(rt.immediate())); + } else { + MOZ_ASSERT(rt.is_reg()); + scratch = rt.rm(); + } + BlockTrampolinePoolScope block_trampoline_pool(this, 2); + { + switch (cond) { + case Always: + if (!CalculateOffset(L, &offset, OffsetSize::kOffset21)) return false; + Assembler::j(offset); + EmitConstPoolWithJumpIfNeeded(); + break; + case Equal: + // rs == rt + if (rt.is_reg() && rs == rt.rm()) { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset21)) return false; + Assembler::j(offset); + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::beq(rs, scratch, offset); + } + break; + case NotEqual: + // rs != rt + if (rt.is_reg() && rs == rt.rm()) { + break; // No code needs to be emitted + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::bne(rs, scratch, offset); + } + break; + + // Signed comparison. + case GreaterThan: + // rs > rt + if (rt.is_reg() && rs == rt.rm()) { + break; // No code needs to be emitted. + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::bgt(rs, scratch, offset); + } + break; + case GreaterThanOrEqual: + // rs >= rt + if (rt.is_reg() && rs == rt.rm()) { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset21)) return false; + Assembler::j(offset); + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::bge(rs, scratch, offset); + } + break; + case LessThan: + // rs < rt + if (rt.is_reg() && rs == rt.rm()) { + break; // No code needs to be emitted. + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::blt(rs, scratch, offset); + } + break; + case LessThanOrEqual: + // rs <= rt + if (rt.is_reg() && rs == rt.rm()) { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset21)) return false; + Assembler::j(offset); + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::ble(rs, scratch, offset); + } + break; + + // Unsigned comparison. + case Above: + // rs > rt + if (rt.is_reg() && rs == rt.rm()) { + break; // No code needs to be emitted. + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::bgtu(rs, scratch, offset); + } + break; + case AboveOrEqual: + // rs >= rt + if (rt.is_reg() && rs == rt.rm()) { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset21)) return false; + Assembler::j(offset); + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::bgeu(rs, scratch, offset); + } + break; + case Below: + // rs < rt + if (rt.is_reg() && rs == rt.rm()) { + break; // No code needs to be emitted. + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + bltu(rs, scratch, offset); + } + break; + case BelowOrEqual: + // rs <= rt + if (rt.is_reg() && rs == rt.rm()) { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset21)) return false; + Assembler::j(offset); + } else { + if (!CalculateOffset(L, &offset, OffsetSize::kOffset13)) return false; + Assembler::bleu(rs, scratch, offset); + } + break; + default: + MOZ_CRASH("UNREACHABLE"); + } + } + return true; +} + +// BRANCH_ARGS_CHECK checks that conditional jump arguments are correct. +#define BRANCH_ARGS_CHECK(cond, rs, rt) \ + MOZ_ASSERT((cond == Always && rs == zero && rt.rm() == zero) || \ + (cond != Always && (rs != zero || rt.rm() != zero))) + +bool MacroAssemblerRiscv64::BranchShortCheck(int32_t offset, Label* L, + Condition cond, Register rs, + const Operand& rt) { + BRANCH_ARGS_CHECK(cond, rs, rt); + + if (!L) { + MOZ_ASSERT(is_int13(offset)); + return BranchShortHelper(offset, nullptr, cond, rs, rt); + } else { + MOZ_ASSERT(offset == 0); + return BranchShortHelper(0, L, cond, rs, rt); + } +} + +void MacroAssemblerRiscv64::BranchShort(Label* L) { BranchShortHelper(0, L); } + +void MacroAssemblerRiscv64::BranchShort(int32_t offset, Condition cond, + Register rs, const Operand& rt) { + BranchShortCheck(offset, nullptr, cond, rs, rt); +} + +void MacroAssemblerRiscv64::BranchShort(Label* L, Condition cond, Register rs, + const Operand& rt) { + BranchShortCheck(0, L, cond, rs, rt); +} + +void MacroAssemblerRiscv64::BranchLong(Label* L) { + // Generate position independent long branch. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + int32_t imm; + imm = branch_long_offset(L); + GenPCRelativeJump(scratch, imm); +} + +void MacroAssemblerRiscv64::BranchAndLinkLong(Label* L) { + // Generate position independent long branch and link. + int32_t imm; + imm = branch_long_offset(L); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + GenPCRelativeJumpAndLink(scratch, imm); +} + +void MacroAssemblerRiscv64::ma_branch(Label* L, Condition cond, Register rs, + const Operand& rt, JumpKind jumpKind) { + if (L->used()) { + if (jumpKind == ShortJump && BranchShortCheck(0, L, cond, rs, rt)) { + return; + } + if (cond != Always) { + Label skip; + Condition neg_cond = InvertCondition(cond); + BranchShort(&skip, neg_cond, rs, rt); + BranchLong(L); + bind(&skip); + } else { + BranchLong(L); + EmitConstPoolWithJumpIfNeeded(); + } + } else { + if (jumpKind == LongJump) { + if (cond != Always) { + Label skip; + Condition neg_cond = InvertCondition(cond); + BranchShort(&skip, neg_cond, rs, rt); + BranchLong(L); + bind(&skip); + } else { + BranchLong(L); + EmitConstPoolWithJumpIfNeeded(); + } + } else { + BranchShort(L, cond, rs, rt); + } + } +} + +// Branches when done from within riscv code. +void MacroAssemblerRiscv64::ma_b(Register lhs, Address addr, Label* label, + Condition c, JumpKind jumpKind) { + ScratchRegisterScope scratch(asMasm()); + MOZ_ASSERT(lhs != scratch); + ma_load(scratch, addr, SizeDouble); + ma_b(lhs, Register(scratch), label, c, jumpKind); +} + +void MacroAssemblerRiscv64::ma_b(Register lhs, ImmPtr imm, Label* l, + Condition c, JumpKind jumpKind) { + asMasm().ma_b(lhs, ImmWord(uintptr_t(imm.value)), l, c, jumpKind); +} + +// Branches when done from within loongarch-specific code. +void MacroAssemblerRiscv64::ma_b(Register lhs, ImmWord imm, Label* label, + Condition c, JumpKind jumpKind) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(lhs != scratch); + ma_li(scratch, imm); + ma_b(lhs, Register(scratch), label, c, jumpKind); +} + +void MacroAssemblerRiscv64::ma_b(Register lhs, Imm32 imm, Label* label, + Condition c, JumpKind jumpKind) { + if ((c == NonZero || c == Zero || c == Signed || c == NotSigned) && + imm.value == 0) { + ma_b(lhs, lhs, label, c, jumpKind); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(lhs != scratch); + ma_li(scratch, imm); + ma_b(lhs, Register(scratch), label, c, jumpKind); + } +} + +void MacroAssemblerRiscv64::ma_b(Address addr, Imm32 imm, Label* label, + Condition c, JumpKind jumpKind) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + ma_load(scratch2, addr); + ma_b(Register(scratch2), imm, label, c, jumpKind); +} + +void MacroAssemblerRiscv64::ma_b(Register lhs, Register rhs, Label* label, + Condition c, JumpKind jumpKind) { + switch (c) { + case Equal: + case NotEqual: + ma_branch(label, c, lhs, rhs, jumpKind); + break; + case Always: + ma_branch(label, c, zero, Operand(zero), jumpKind); + break; + case Zero: + MOZ_ASSERT(lhs == rhs); + ma_branch(label, Equal, lhs, Operand(zero), jumpKind); + break; + case NonZero: + MOZ_ASSERT(lhs == rhs); + ma_branch(label, NotEqual, lhs, Operand(zero), jumpKind); + break; + case Signed: + MOZ_ASSERT(lhs == rhs); + ma_branch(label, LessThan, lhs, Operand(zero), jumpKind); + break; + case NotSigned: + MOZ_ASSERT(lhs == rhs); + ma_branch(label, GreaterThanOrEqual, lhs, Operand(zero), jumpKind); + break; + default: { + ma_branch(label, c, lhs, rhs, jumpKind); + break; + } + } +} + +void MacroAssemblerRiscv64::ExtractBits(Register rt, Register rs, uint16_t pos, + uint16_t size, bool sign_extend) { +#if JS_CODEGEN_RISCV64 + MOZ_ASSERT(pos < 64 && 0 < size && size <= 64 && 0 < pos + size && + pos + size <= 64); + slli(rt, rs, 64 - (pos + size)); + if (sign_extend) { + srai(rt, rt, 64 - size); + } else { + srli(rt, rt, 64 - size); + } +#elif JS_CODEGEN_RISCV32 + MOZ_ASSERT(pos < 32); + MOZ_ASSERT(size > 0); + MOZ_ASSERT(size <= 32); + MOZ_ASSERT((pos + size) > 0); + MOZ_ASSERT((pos + size) <= 32); + slli(rt, rs, 32 - (pos + size)); + if (sign_extend) { + srai(rt, rt, 32 - size); + } else { + srli(rt, rt, 32 - size); + } +#endif +} + +void MacroAssemblerRiscv64::InsertBits(Register dest, Register source, int pos, + int size) { +#if JS_CODEGEN_RISCV64 + MOZ_ASSERT(size < 64); +#elif JS_CODEGEN_RISCV32 + MOZ_ASSERT(size < 32); +#endif + UseScratchRegisterScope temps(this); + Register mask = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + Register source_ = temps.Acquire(); + // Create a mask of the length=size. + ma_li(mask, Imm32(1)); + slli(mask, mask, size); + addi(mask, mask, -1); + and_(source_, mask, source); + slli(source_, source_, pos); + // Make a mask containing 0's. 0's start at "pos" with length=size. + slli(mask, mask, pos); + not_(mask, mask); + // cut area for insertion of source. + and_(dest, mask, dest); + // insert source + or_(dest, dest, source_); +} + +void MacroAssemblerRiscv64::InsertBits(Register dest, Register source, + Register pos, int size) { +#if JS_CODEGEN_RISCV64 + MOZ_ASSERT(size < 64); +#elif JS_CODEGEN_RISCV32 + MOZ_ASSERT(size < 32); +#endif + UseScratchRegisterScope temps(this); + Register mask = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + Register source_ = temps.Acquire(); + // Create a mask of the length=size. + ma_li(mask, Imm32(1)); + slli(mask, mask, size); + addi(mask, mask, -1); + and_(source_, mask, source); + sll(source_, source_, pos); + // Make a mask containing 0's. 0's start at "pos" with length=size. + sll(mask, mask, pos); + not_(mask, mask); + // cut area for insertion of source. + and_(dest, mask, dest); + // insert source + or_(dest, dest, source_); +} + +void MacroAssemblerRiscv64::ma_add32(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + if (is_int12(rt.immediate())) { + addiw(rd, rs, static_cast<int32_t>(rt.immediate())); + } else if ((-4096 <= rt.immediate() && rt.immediate() <= -2049) || + (2048 <= rt.immediate() && rt.immediate() <= 4094)) { + addiw(rd, rs, rt.immediate() / 2); + addiw(rd, rd, rt.immediate() - (rt.immediate() / 2)); + } else { + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + ma_li(scratch, rt.immediate()); + addw(rd, rs, scratch); + } + } else { + MOZ_ASSERT(rt.is_reg()); + addw(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_add64(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + if (is_int12(rt.immediate())) { + addi(rd, rs, static_cast<int32_t>(rt.immediate())); + } else if ((-4096 <= rt.immediate() && rt.immediate() <= -2049) || + (2048 <= rt.immediate() && rt.immediate() <= 4094)) { + addi(rd, rs, rt.immediate() / 2); + addi(rd, rd, rt.immediate() - (rt.immediate() / 2)); + } else { + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + ma_li(scratch, rt.immediate()); + add(rd, rs, scratch); + } + } else { + MOZ_ASSERT(rt.is_reg()); + add(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_sub32(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + if (is_int12(-rt.immediate())) { + addiw(rd, rs, + static_cast<int32_t>( + -rt.immediate())); // No subi instr, use addi(x, y, -imm). + } else if ((-4096 <= -rt.immediate() && -rt.immediate() <= -2049) || + (2048 <= -rt.immediate() && -rt.immediate() <= 4094)) { + addiw(rd, rs, -rt.immediate() / 2); + addiw(rd, rd, -rt.immediate() - (-rt.immediate() / 2)); + } else { + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, rt.immediate()); + subw(rd, rs, scratch); + } + } else { + MOZ_ASSERT(rt.is_reg()); + subw(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_sub64(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + if (is_int12(-rt.immediate())) { + addi(rd, rs, + static_cast<int32_t>( + -rt.immediate())); // No subi instr, use addi(x, y, -imm). + } else if ((-4096 <= -rt.immediate() && -rt.immediate() <= -2049) || + (2048 <= -rt.immediate() && -rt.immediate() <= 4094)) { + addi(rd, rs, -rt.immediate() / 2); + addi(rd, rd, -rt.immediate() - (-rt.immediate() / 2)); + } else { + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, rt.immediate()); + sub(rd, rs, scratch); + } + } else { + MOZ_ASSERT(rt.is_reg()); + sub(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_and(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + if (is_int12(rt.immediate())) { + andi(rd, rs, rt.immediate()); + } else { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + and_(rd, rs, ScratchRegister); + } + } else { + MOZ_ASSERT(rt.is_reg()); + and_(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_or(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + if (is_int12(rt.immediate())) { + ori(rd, rs, rt.immediate()); + } else { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + or_(rd, rs, ScratchRegister); + } + } else { + MOZ_ASSERT(rt.is_reg()); + or_(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_xor(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + if (is_int12(rt.immediate())) { + xori(rd, rs, rt.immediate()); + } else { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + xor_(rd, rs, ScratchRegister); + } + } else { + MOZ_ASSERT(rt.is_reg()); + xor_(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_nor(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + nor(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + nor(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_div32(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + divw(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + divw(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_divu32(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + divuw(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + divuw(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_div64(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + div(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + div(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_divu64(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + divu(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + divu(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_mod32(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + remw(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + remw(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_modu32(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + remuw(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + remuw(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_mod64(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + rem(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + rem(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_modu64(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + remu(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + remu(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_mul32(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + mulw(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + mulw(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_mulh32(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + mul(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + mul(rd, rs, rt.rm()); + } + srai(rd, rd, 32); +} + +void MacroAssemblerRiscv64::ma_mul64(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + mul(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + mul(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_mulh64(Register rd, Register rs, Operand rt) { + if (rt.is_imm()) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, rt.immediate()); + mulh(rd, rs, ScratchRegister); + } else { + MOZ_ASSERT(rt.is_reg()); + mulh(rd, rs, rt.rm()); + } +} + +void MacroAssemblerRiscv64::ma_sll64(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + sll(rd, rs, rt.rm()); + } else { + MOZ_ASSERT(rt.is_imm()); + uint8_t shamt = static_cast<uint8_t>(rt.immediate()); + slli(rd, rs, shamt); + } +} + +void MacroAssemblerRiscv64::ma_sll32(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + sllw(rd, rs, rt.rm()); + } else { + MOZ_ASSERT(rt.is_imm()); + uint8_t shamt = static_cast<uint8_t>(rt.immediate()); + slliw(rd, rs, shamt); + } +} + +void MacroAssemblerRiscv64::ma_sra64(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + sra(rd, rs, rt.rm()); + } else { + MOZ_ASSERT(rt.is_imm()); + uint8_t shamt = static_cast<uint8_t>(rt.immediate()); + srai(rd, rs, shamt); + } +} + +void MacroAssemblerRiscv64::ma_sra32(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + sraw(rd, rs, rt.rm()); + } else { + MOZ_ASSERT(rt.is_imm()); + uint8_t shamt = static_cast<uint8_t>(rt.immediate()); + sraiw(rd, rs, shamt); + } +} + +void MacroAssemblerRiscv64::ma_srl64(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + srl(rd, rs, rt.rm()); + } else { + MOZ_ASSERT(rt.is_imm()); + uint8_t shamt = static_cast<uint8_t>(rt.immediate()); + srli(rd, rs, shamt); + } +} + +void MacroAssemblerRiscv64::ma_srl32(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + srlw(rd, rs, rt.rm()); + } else { + MOZ_ASSERT(rt.is_imm()); + uint8_t shamt = static_cast<uint8_t>(rt.immediate()); + srliw(rd, rs, shamt); + } +} + +void MacroAssemblerRiscv64::ma_slt(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + slt(rd, rs, rt.rm()); + } else { + MOZ_ASSERT(rt.is_imm()); + if (is_int12(rt.immediate())) { + slti(rd, rs, static_cast<int32_t>(rt.immediate())); + } else { + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + ma_li(scratch, rt.immediate()); + slt(rd, rs, scratch); + } + } +} + +void MacroAssemblerRiscv64::ma_sltu(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + sltu(rd, rs, rt.rm()); + } else { + MOZ_ASSERT(rt.is_imm()); + if (is_int12(rt.immediate())) { + sltiu(rd, rs, static_cast<int32_t>(rt.immediate())); + } else { + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + ma_li(scratch, rt.immediate()); + sltu(rd, rs, scratch); + } + } +} + +void MacroAssemblerRiscv64::ma_sle(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + slt(rd, rt.rm(), rs); + } else { + MOZ_ASSERT(rt.is_imm()); + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + ma_li(scratch, rt.immediate()); + slt(rd, scratch, rs); + } + xori(rd, rd, 1); +} + +void MacroAssemblerRiscv64::ma_sleu(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + sltu(rd, rt.rm(), rs); + } else { + MOZ_ASSERT(rt.is_imm()); + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + ma_li(scratch, rt.immediate()); + sltu(rd, scratch, rs); + } + xori(rd, rd, 1); +} + +void MacroAssemblerRiscv64::ma_sgt(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + slt(rd, rt.rm(), rs); + } else { + MOZ_ASSERT(rt.is_imm()); + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + ma_li(scratch, rt.immediate()); + slt(rd, scratch, rs); + } +} + +void MacroAssemblerRiscv64::ma_sgtu(Register rd, Register rs, Operand rt) { + if (rt.is_reg()) { + sltu(rd, rt.rm(), rs); + } else { + MOZ_ASSERT(rt.is_imm()); + // li handles the relocation. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 9); + ma_li(scratch, rt.immediate()); + sltu(rd, scratch, rs); + } +} + +void MacroAssemblerRiscv64::ma_sge(Register rd, Register rs, Operand rt) { + ma_slt(rd, rs, rt); + xori(rd, rd, 1); +} + +void MacroAssemblerRiscv64::ma_sgeu(Register rd, Register rs, Operand rt) { + ma_sltu(rd, rs, rt); + xori(rd, rd, 1); +} + +static inline bool IsZero(const Operand& rt) { + if (rt.is_reg()) { + return rt.rm() == zero_reg; + } else { + MOZ_ASSERT(rt.is_imm()); + return rt.immediate() == 0; + } +} + +void MacroAssemblerRiscv64::ma_seq(Register rd, Register rs, Operand rt) { + if (rs == zero_reg) { + ma_seqz(rd, rt); + } else if (IsZero(rt)) { + seqz(rd, rs); + } else { + ma_sub64(rd, rs, rt); + seqz(rd, rd); + } +} + +void MacroAssemblerRiscv64::ma_sne(Register rd, Register rs, Operand rt) { + if (rs == zero_reg) { + ma_snez(rd, rt); + } else if (IsZero(rt)) { + snez(rd, rs); + } else { + ma_sub64(rd, rs, rt); + snez(rd, rd); + } +} + +void MacroAssemblerRiscv64::ma_seqz(Register rd, const Operand& rt) { + if (rt.is_reg()) { + seqz(rd, rt.rm()); + } else { + ma_li(rd, rt.immediate() == 0); + } +} + +void MacroAssemblerRiscv64::ma_snez(Register rd, const Operand& rt) { + if (rt.is_reg()) { + snez(rd, rt.rm()); + } else { + ma_li(rd, rt.immediate() != 0); + } +} + +void MacroAssemblerRiscv64::ma_neg(Register rd, const Operand& rt) { + MOZ_ASSERT(rt.is_reg()); + neg(rd, rt.rm()); +} + +void MacroAssemblerRiscv64::ma_jump(ImmPtr dest) { + DEBUG_PRINTF("[ %s\n", __FUNCTION__); + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + asMasm().ma_liPatchable(scratch, dest); + jr(scratch, 0); + DEBUG_PRINTF("]\n"); +} +// fp instructions +void MacroAssemblerRiscv64::ma_lid(FloatRegister dest, double value) { + ImmWord imm(mozilla::BitwiseCast<uint64_t>(value)); + + if (imm.value != 0) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + fmv_d_x(dest, scratch); + } else { + fmv_d_x(dest, zero); + } +} +// fp instructions +void MacroAssemblerRiscv64::ma_lis(FloatRegister dest, float value) { + Imm32 imm(mozilla::BitwiseCast<uint32_t>(value)); + + if (imm.value != 0) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + fmv_w_x(dest, scratch); + } else { + fmv_w_x(dest, zero); + } +} + +void MacroAssemblerRiscv64::ma_sub32TestOverflow(Register rd, Register rj, + Register rk, Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + sub(scratch, rj, rk); + subw(rd, rj, rk); + ma_b(rd, Register(scratch), overflow, Assembler::NotEqual); +} + +void MacroAssemblerRiscv64::ma_sub32TestOverflow(Register rd, Register rj, + Imm32 imm, Label* overflow) { + if (imm.value != INT32_MIN) { + asMasm().ma_add32TestOverflow(rd, rj, Imm32(-imm.value), overflow); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(rj != scratch); + ma_li(scratch, Imm32(imm.value)); + asMasm().ma_sub32TestOverflow(rd, rj, scratch, overflow); + } +} + +void MacroAssemblerRiscv64::ma_add32TestOverflow(Register rd, Register rj, + Register rk, Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + add(scratch, rj, rk); + addw(rd, rj, rk); + ma_b(rd, Register(scratch), overflow, Assembler::NotEqual); +} + +void MacroAssemblerRiscv64::ma_add32TestOverflow(Register rd, Register rj, + Imm32 imm, Label* overflow) { + // Check for signed range because of addi + if (is_intn(imm.value, 12)) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + addi(scratch, rj, imm.value); + addiw(rd, rj, imm.value); + ma_b(rd, scratch, overflow, Assembler::NotEqual); + } else { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + ma_li(scratch2, imm); + ma_add32TestOverflow(rd, rj, scratch2, overflow); + } +} + +void MacroAssemblerRiscv64::ma_subPtrTestOverflow(Register rd, Register rj, + Register rk, + Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT_IF(rj == rd, rj != rk); + MOZ_ASSERT(rj != scratch2); + MOZ_ASSERT(rk != scratch2); + MOZ_ASSERT(rd != scratch2); + + Register rj_copy = rj; + + if (rj == rd) { + ma_or(scratch2, rj, zero); + rj_copy = scratch2; + } + + { + Register scratch = temps.Acquire(); + MOZ_ASSERT(rd != scratch); + + sub(rd, rj, rk); + // If the sign of rj and rk are the same, no overflow + ma_xor(scratch, rj_copy, rk); + // Check if the sign of rd and rj are the same + ma_xor(scratch2, rd, rj_copy); + ma_and(scratch2, scratch2, scratch); + } + + ma_b(scratch2, zero, overflow, Assembler::LessThan); +} + +void MacroAssemblerRiscv64::ma_addPtrTestOverflow(Register rd, Register rj, + Register rk, + Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(rd != scratch); + + if (rj == rk) { + if (rj == rd) { + ma_or(scratch, rj, zero); + rj = scratch; + } + + add(rd, rj, rj); + ma_xor(scratch, rj, rd); + ma_b(scratch, zero, overflow, Assembler::LessThan); + } else { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(rj != scratch); + MOZ_ASSERT(rd != scratch2); + + if (rj == rd) { + ma_or(scratch2, rj, zero); + rj = scratch2; + } + + add(rd, rj, rk); + slti(scratch, rj, 0); + slt(scratch2, rd, rj); + ma_b(scratch, Register(scratch2), overflow, Assembler::NotEqual); + } +} + +void MacroAssemblerRiscv64::ma_addPtrTestOverflow(Register rd, Register rj, + Imm32 imm, Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + + if (imm.value == 0) { + ori(rd, rj, 0); + return; + } + + if (rj == rd) { + ori(scratch2, rj, 0); + rj = scratch2; + } + + ma_add64(rd, rj, imm); + + if (imm.value > 0) { + ma_b(rd, rj, overflow, Assembler::LessThan); + } else { + MOZ_ASSERT(imm.value < 0); + ma_b(rd, rj, overflow, Assembler::GreaterThan); + } +} + +void MacroAssemblerRiscv64::ma_addPtrTestOverflow(Register rd, Register rj, + ImmWord imm, + Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + + if (imm.value == 0) { + ori(rd, rj, 0); + return; + } + + if (rj == rd) { + MOZ_ASSERT(rj != scratch2); + ori(scratch2, rj, 0); + rj = scratch2; + } + + ma_li(rd, imm); + add(rd, rj, rd); + + if (imm.value > 0) { + ma_b(rd, rj, overflow, Assembler::LessThan); + } else { + MOZ_ASSERT(imm.value < 0); + ma_b(rd, rj, overflow, Assembler::GreaterThan); + } +} + +void MacroAssemblerRiscv64::ma_add32TestCarry(Condition cond, Register rd, + Register rj, Register rk, + Label* overflow) { + MOZ_ASSERT(cond == Assembler::CarrySet || cond == Assembler::CarryClear); + MOZ_ASSERT_IF(rd == rj, rk != rd); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + addw(rd, rj, rk); + sltu(scratch, rd, rd == rj ? rk : rj); + ma_b(Register(scratch), Register(scratch), overflow, + cond == Assembler::CarrySet ? Assembler::NonZero : Assembler::Zero); +} + +void MacroAssemblerRiscv64::ma_add32TestCarry(Condition cond, Register rd, + Register rj, Imm32 imm, + Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + MOZ_ASSERT(rj != scratch2); + ma_li(scratch2, imm); + ma_add32TestCarry(cond, rd, rj, scratch2, overflow); +} + +void MacroAssemblerRiscv64::ma_subPtrTestOverflow(Register rd, Register rj, + Imm32 imm, Label* overflow) { + // TODO(riscv): Check subPtrTestOverflow + MOZ_ASSERT(imm.value != INT32_MIN); + ma_addPtrTestOverflow(rd, rj, Imm32(-imm.value), overflow); +} + +void MacroAssemblerRiscv64::ma_addPtrTestCarry(Condition cond, Register rd, + Register rj, Register rk, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(rd != rk); + MOZ_ASSERT(rd != scratch); + add(rd, rj, rk); + sltu(scratch, rd, rk); + ma_b(scratch, Register(scratch), label, + cond == Assembler::CarrySet ? Assembler::NonZero : Assembler::Zero); +} + +void MacroAssemblerRiscv64::ma_addPtrTestCarry(Condition cond, Register rd, + Register rj, Imm32 imm, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + + // Check for signed range because of addi + if (is_intn(imm.value, 12)) { + addi(rd, rj, imm.value); + sltiu(scratch2, rd, imm.value); + ma_b(scratch2, scratch2, label, + cond == Assembler::CarrySet ? Assembler::NonZero : Assembler::Zero); + } else { + ma_li(scratch2, imm); + ma_addPtrTestCarry(cond, rd, rj, scratch2, label); + } +} + +void MacroAssemblerRiscv64::ma_addPtrTestCarry(Condition cond, Register rd, + Register rj, ImmWord imm, + Label* label) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + + // Check for signed range because of addi_d + if (is_intn(imm.value, 12)) { + uint32_t value = imm.value; + addi(rd, rj, value); + ma_sltu(scratch2, rd, Operand(value)); + ma_b(scratch2, scratch2, label, + cond == Assembler::CarrySet ? Assembler::NonZero : Assembler::Zero); + } else { + ma_li(scratch2, imm); + ma_addPtrTestCarry(cond, rd, rj, scratch2, label); + } +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_load( + Register dest, const BaseIndex& src, LoadStoreSize size, + LoadStoreExtension extension) { + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + asMasm().computeScaledAddress(src, scratch2); + return asMasm().ma_load(dest, Address(scratch2, src.offset), size, extension); +} +void MacroAssemblerRiscv64::ma_pop(FloatRegister f) { + fld(f, StackPointer, 0); + addi(StackPointer, StackPointer, sizeof(double)); +} + +void MacroAssemblerRiscv64::ma_push(FloatRegister f) { + addi(StackPointer, StackPointer, (int32_t) - sizeof(double)); + fsd(f, StackPointer, 0); +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_fld_s(FloatRegister ft, + Address address) { + int32_t offset = address.offset; + Register base = address.base; + + FaultingCodeOffset fco; + if (is_intn(offset, 12)) { + fco = FaultingCodeOffset(currentOffset()); + flw(ft, base, offset); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(base != scratch); + ma_li(scratch, Imm32(offset)); + ma_add64(scratch, base, scratch); + fco = FaultingCodeOffset(currentOffset()); + flw(ft, scratch, 0); + } + return fco; +} +FaultingCodeOffset MacroAssemblerRiscv64::ma_fld_d(FloatRegister ft, + Address address) { + int32_t offset = address.offset; + Register base = address.base; + + FaultingCodeOffset fco; + if (is_intn(offset, 12)) { + fco = FaultingCodeOffset(currentOffset()); + fld(ft, base, offset); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(base != scratch); + ma_li(scratch, Imm32(offset)); + ma_add64(scratch, base, scratch); + fco = FaultingCodeOffset(currentOffset()); + fld(ft, scratch, 0); + } + return fco; +} +FaultingCodeOffset MacroAssemblerRiscv64::ma_fst_d(FloatRegister ft, + Address address) { + int32_t offset = address.offset; + Register base = address.base; + + FaultingCodeOffset fco; + if (is_intn(offset, 12)) { + fco = FaultingCodeOffset(currentOffset()); + fsd(ft, base, offset); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(base != scratch); + ma_li(scratch, Imm32(offset)); + ma_add64(scratch, base, scratch); + fco = FaultingCodeOffset(currentOffset()); + fsd(ft, scratch, 0); + } + return fco; +} +FaultingCodeOffset MacroAssemblerRiscv64::ma_fst_s(FloatRegister ft, + Address address) { + int32_t offset = address.offset; + Register base = address.base; + FaultingCodeOffset fco; + if (is_intn(offset, 12)) { + fco = FaultingCodeOffset(currentOffset()); + fsw(ft, base, offset); + } else { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(base != scratch); + ma_li(scratch, Imm32(offset)); + ma_add64(scratch, base, scratch); + fco = FaultingCodeOffset(currentOffset()); + fsw(ft, scratch, 0); + } + return fco; +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_fst_d(FloatRegister ft, + BaseIndex address) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + asMasm().computeScaledAddress(address, scratch); + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + asMasm().ma_fst_d(ft, Address(scratch, address.offset)); + return fco; +} + +FaultingCodeOffset MacroAssemblerRiscv64::ma_fst_s(FloatRegister ft, + BaseIndex address) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + asMasm().computeScaledAddress(address, scratch); + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + asMasm().ma_fst_s(ft, Address(scratch, address.offset)); + return fco; +} + +void MacroAssemblerRiscv64::ma_fld_d(FloatRegister ft, const BaseIndex& src) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + asMasm().computeScaledAddress(src, scratch); + asMasm().ma_fld_d(ft, Address(scratch, src.offset)); +} + +void MacroAssemblerRiscv64::ma_fld_s(FloatRegister ft, const BaseIndex& src) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + asMasm().computeScaledAddress(src, scratch); + asMasm().ma_fld_s(ft, Address(scratch, src.offset)); +} + +void MacroAssemblerRiscv64::ma_call(ImmPtr dest) { + DEBUG_PRINTF("[ %s\n", __FUNCTION__); + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + UseScratchRegisterScope temps(this); + temps.Exclude(GeneralRegisterSet(1 << CallReg.code())); + asMasm().ma_liPatchable(CallReg, dest); + jalr(CallReg, 0); + DEBUG_PRINTF("]\n"); +} + +void MacroAssemblerRiscv64::CompareIsNotNanF32(Register rd, FPURegister cmp1, + FPURegister cmp2) { + UseScratchRegisterScope temps(this); + BlockTrampolinePoolScope block_trampoline_pool(this, 3); + Register scratch = temps.Acquire(); + + feq_s(rd, cmp1, cmp1); // rd <- !isNan(cmp1) + feq_s(scratch, cmp2, cmp2); // scratch <- !isNaN(cmp2) + ma_and(rd, rd, scratch); // rd <- !isNan(cmp1) && !isNan(cmp2) +} + +void MacroAssemblerRiscv64::CompareIsNotNanF64(Register rd, FPURegister cmp1, + FPURegister cmp2) { + UseScratchRegisterScope temps(this); + BlockTrampolinePoolScope block_trampoline_pool(this, 3); + Register scratch = temps.Acquire(); + + feq_d(rd, cmp1, cmp1); // rd <- !isNan(cmp1) + feq_d(scratch, cmp2, cmp2); // scratch <- !isNaN(cmp2) + ma_and(rd, rd, scratch); // rd <- !isNan(cmp1) && !isNan(cmp2) +} + +void MacroAssemblerRiscv64::CompareIsNanF32(Register rd, FPURegister cmp1, + FPURegister cmp2) { + CompareIsNotNanF32(rd, cmp1, cmp2); // rd <- !isNan(cmp1) && !isNan(cmp2) + ma_xor(rd, rd, Operand(1)); // rd <- isNan(cmp1) || isNan(cmp2) +} + +void MacroAssemblerRiscv64::CompareIsNanF64(Register rd, FPURegister cmp1, + FPURegister cmp2) { + CompareIsNotNanF64(rd, cmp1, cmp2); // rd <- !isNan(cmp1) && !isNan(cmp2) + ma_xor(rd, rd, Operand(1)); // rd <- isNan(cmp1) || isNan(cmp2) +} + +void MacroAssemblerRiscv64::Clz32(Register rd, Register xx) { + // 32 bit unsigned in lower word: count number of leading zeros. + // int n = 32; + // unsigned y; + + // y = x >>16; if (y != 0) { n = n -16; x = y; } + // y = x >> 8; if (y != 0) { n = n - 8; x = y; } + // y = x >> 4; if (y != 0) { n = n - 4; x = y; } + // y = x >> 2; if (y != 0) { n = n - 2; x = y; } + // y = x >> 1; if (y != 0) {rd = n - 2; return;} + // rd = n - x; + + Label L0, L1, L2, L3, L4; + UseScratchRegisterScope temps(this); + Register x = rd; + Register y = temps.Acquire(); + Register n = temps.Acquire(); + MOZ_ASSERT(xx != y && xx != n); + mv(x, xx); + ma_li(n, Imm32(32)); +#if JS_CODEGEN_RISCV64 + srliw(y, x, 16); + ma_branch(&L0, Equal, y, Operand(zero_reg)); + mv(x, y); + addiw(n, n, -16); + bind(&L0); + srliw(y, x, 8); + ma_branch(&L1, Equal, y, Operand(zero_reg)); + addiw(n, n, -8); + mv(x, y); + bind(&L1); + srliw(y, x, 4); + ma_branch(&L2, Equal, y, Operand(zero_reg)); + addiw(n, n, -4); + mv(x, y); + bind(&L2); + srliw(y, x, 2); + ma_branch(&L3, Equal, y, Operand(zero_reg)); + addiw(n, n, -2); + mv(x, y); + bind(&L3); + srliw(y, x, 1); + subw(rd, n, x); + ma_branch(&L4, Equal, y, Operand(zero_reg)); + addiw(rd, n, -2); + bind(&L4); +#elif JS_CODEGEN_RISCV32 + srli(y, x, 16); + ma_branch(&L0, Equal, y, Operand(zero_reg)); + mv(x, y); + addi(n, n, -16); + bind(&L0); + srli(y, x, 8); + ma_branch(&L1, Equal, y, Operand(zero_reg)); + addi(n, n, -8); + mv(x, y); + bind(&L1); + srli(y, x, 4); + ma_branch(&L2, Equal, y, Operand(zero_reg)); + addi(n, n, -4); + mv(x, y); + bind(&L2); + srli(y, x, 2); + ma_branch(&L3, Equal, y, Operand(zero_reg)); + addi(n, n, -2); + mv(x, y); + bind(&L3); + srli(y, x, 1); + sub(rd, n, x); + ma_branch(&L4, Equal, y, Operand(zero_reg)); + addi(rd, n, -2); + bind(&L4); +#endif +} + +#if JS_CODEGEN_RISCV64 +void MacroAssemblerRiscv64::Clz64(Register rd, Register xx) { + // 64 bit: count number of leading zeros. + // int n = 64; + // unsigned y; + + // y = x >>32; if (y != 0) { n = n - 32; x = y; } + // y = x >>16; if (y != 0) { n = n - 16; x = y; } + // y = x >> 8; if (y != 0) { n = n - 8; x = y; } + // y = x >> 4; if (y != 0) { n = n - 4; x = y; } + // y = x >> 2; if (y != 0) { n = n - 2; x = y; } + // y = x >> 1; if (y != 0) {rd = n - 2; return;} + // rd = n - x; + + Label L0, L1, L2, L3, L4, L5; + UseScratchRegisterScope temps(this); + Register x = rd; + Register y = temps.Acquire(); + Register n = temps.Acquire(); + MOZ_ASSERT(xx != y && xx != n); + mv(x, xx); + ma_li(n, Imm32(64)); + srli(y, x, 32); + ma_branch(&L0, Equal, y, Operand(zero_reg)); + addiw(n, n, -32); + mv(x, y); + bind(&L0); + srli(y, x, 16); + ma_branch(&L1, Equal, y, Operand(zero_reg)); + addiw(n, n, -16); + mv(x, y); + bind(&L1); + srli(y, x, 8); + ma_branch(&L2, Equal, y, Operand(zero_reg)); + addiw(n, n, -8); + mv(x, y); + bind(&L2); + srli(y, x, 4); + ma_branch(&L3, Equal, y, Operand(zero_reg)); + addiw(n, n, -4); + mv(x, y); + bind(&L3); + srli(y, x, 2); + ma_branch(&L4, Equal, y, Operand(zero_reg)); + addiw(n, n, -2); + mv(x, y); + bind(&L4); + srli(y, x, 1); + subw(rd, n, x); + ma_branch(&L5, Equal, y, Operand(zero_reg)); + addiw(rd, n, -2); + bind(&L5); +} +#endif +void MacroAssemblerRiscv64::Ctz32(Register rd, Register rs) { + // Convert trailing zeroes to trailing ones, and bits to their left + // to zeroes. + + { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_add64(scratch, rs, Operand(-1)); + ma_xor(rd, scratch, rs); + ma_and(rd, rd, scratch); + // Count number of leading zeroes. + } + Clz32(rd, rd); + { + // Subtract number of leading zeroes from 32 to get number of trailing + // ones. Remember that the trailing ones were formerly trailing zeroes. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, Imm32(32)); + ma_sub32(rd, scratch, rd); + } +} +#if JS_CODEGEN_RISCV64 +void MacroAssemblerRiscv64::Ctz64(Register rd, Register rs) { + // Convert trailing zeroes to trailing ones, and bits to their left + // to zeroes. + { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_add64(scratch, rs, Operand(-1)); + ma_xor(rd, scratch, rs); + ma_and(rd, rd, scratch); + // Count number of leading zeroes. + } + Clz64(rd, rd); + { + // Subtract number of leading zeroes from 64 to get number of trailing + // ones. Remember that the trailing ones were formerly trailing zeroes. + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, 64); + ma_sub64(rd, scratch, rd); + } +} +#endif +void MacroAssemblerRiscv64::Popcnt32(Register rd, Register rs, + Register scratch) { + MOZ_ASSERT(scratch != rs); + MOZ_ASSERT(scratch != rd); + // https://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel + // + // A generalization of the best bit counting method to integers of + // bit-widths up to 128 (parameterized by type T) is this: + // + // v = v - ((v >> 1) & (T)~(T)0/3); // temp + // v = (v & (T)~(T)0/15*3) + ((v >> 2) & (T)~(T)0/15*3); // temp + // v = (v + (v >> 4)) & (T)~(T)0/255*15; // temp + // c = (T)(v * ((T)~(T)0/255)) >> (sizeof(T) - 1) * BITS_PER_BYTE; //count + // + // There are algorithms which are faster in the cases where very few + // bits are set but the algorithm here attempts to minimize the total + // number of instructions executed even when a large number of bits + // are set. + // The number of instruction is 20. + // uint32_t B0 = 0x55555555; // (T)~(T)0/3 + // uint32_t B1 = 0x33333333; // (T)~(T)0/15*3 + // uint32_t B2 = 0x0F0F0F0F; // (T)~(T)0/255*15 + // uint32_t value = 0x01010101; // (T)~(T)0/255 + + uint32_t shift = 24; + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register value = temps.Acquire(); + MOZ_ASSERT((rd != value) && (rs != value)); + ma_li(value, 0x01010101); // value = 0x01010101; + ma_li(scratch2, 0x55555555); // B0 = 0x55555555; + ma_srl32(scratch, rs, Operand(1)); + ma_and(scratch, scratch, scratch2); + ma_sub32(scratch, rs, scratch); + ma_li(scratch2, 0x33333333); // B1 = 0x33333333; + slli(rd, scratch2, 4); + or_(scratch2, scratch2, rd); + ma_and(rd, scratch, scratch2); + ma_srl32(scratch, scratch, Operand(2)); + ma_and(scratch, scratch, scratch2); + ma_add32(scratch, rd, scratch); + ma_srl32(rd, scratch, Operand(4)); + ma_add32(rd, rd, scratch); + ma_li(scratch2, 0xF); + ma_mul32(scratch2, value, scratch2); // B2 = 0x0F0F0F0F; + ma_and(rd, rd, scratch2); + ma_mul32(rd, rd, value); + ma_srl32(rd, rd, Operand(shift)); +} + +#if JS_CODEGEN_RISCV64 +void MacroAssemblerRiscv64::Popcnt64(Register rd, Register rs, + Register scratch) { + MOZ_ASSERT(scratch != rs); + MOZ_ASSERT(scratch != rd); + // uint64_t B0 = 0x5555555555555555l; // (T)~(T)0/3 + // uint64_t B1 = 0x3333333333333333l; // (T)~(T)0/15*3 + // uint64_t B2 = 0x0F0F0F0F0F0F0F0Fl; // (T)~(T)0/255*15 + // uint64_t value = 0x0101010101010101l; // (T)~(T)0/255 + // uint64_t shift = 24; // (sizeof(T) - 1) * BITS_PER_BYTE + uint64_t shift = 24; + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + Register value = temps.Acquire(); + MOZ_ASSERT((rd != value) && (rs != value)); + ma_li(value, 0x1111111111111111l); // value = 0x1111111111111111l; + ma_li(scratch2, 5); + ma_mul64(scratch2, value, scratch2); // B0 = 0x5555555555555555l; + ma_srl64(scratch, rs, Operand(1)); + ma_and(scratch, scratch, scratch2); + ma_sub64(scratch, rs, scratch); + ma_li(scratch2, 3); + ma_mul64(scratch2, value, scratch2); // B1 = 0x3333333333333333l; + ma_and(rd, scratch, scratch2); + ma_srl64(scratch, scratch, Operand(2)); + ma_and(scratch, scratch, scratch2); + ma_add64(scratch, rd, scratch); + ma_srl64(rd, scratch, Operand(4)); + ma_add64(rd, rd, scratch); + ma_li(scratch2, 0xF); + ma_li(value, 0x0101010101010101l); // value = 0x0101010101010101l; + ma_mul64(scratch2, value, scratch2); // B2 = 0x0F0F0F0F0F0F0F0Fl; + ma_and(rd, rd, scratch2); + ma_mul64(rd, rd, value); + srli(rd, rd, 32 + shift); +} +#endif + +void MacroAssemblerRiscv64::ma_div_branch_overflow(Register rd, Register rj, + Register rk, + Label* overflow) { + ScratchRegisterScope scratch(asMasm()); + ma_mod32(scratch, rj, rk); + ma_b(scratch, scratch, overflow, Assembler::NonZero); + divw(rd, rj, rk); +} + +void MacroAssemblerRiscv64::ma_div_branch_overflow(Register rd, Register rj, + Imm32 imm, Label* overflow) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + ma_div_branch_overflow(rd, rj, scratch, overflow); +} + +void MacroAssemblerRiscv64::ma_mod_mask(Register src, Register dest, + Register hold, Register remain, + int32_t shift, Label* negZero) { + // MATH: + // We wish to compute x % (1<<y) - 1 for a known constant, y. + // First, let b = (1<<y) and C = (1<<y)-1, then think of the 32 bit + // dividend as a number in base b, namely + // c_0*1 + c_1*b + c_2*b^2 ... c_n*b^n + // now, since both addition and multiplication commute with modulus, + // x % C == (c_0 + c_1*b + ... + c_n*b^n) % C == + // (c_0 % C) + (c_1%C) * (b % C) + (c_2 % C) * (b^2 % C)... + // now, since b == C + 1, b % C == 1, and b^n % C == 1 + // this means that the whole thing simplifies to: + // c_0 + c_1 + c_2 ... c_n % C + // each c_n can easily be computed by a shift/bitextract, and the modulus + // can be maintained by simply subtracting by C whenever the number gets + // over C. + int32_t mask = (1 << shift) - 1; + Label head, negative, sumSigned, done; + + // hold holds -1 if the value was negative, 1 otherwise. + // remain holds the remaining bits that have not been processed + // SecondScratchReg serves as a temporary location to store extracted bits + // into as well as holding the trial subtraction as a temp value dest is + // the accumulator (and holds the final result) + + // move the whole value into the remain. + or_(remain, src, zero); + // Zero out the dest. + ma_li(dest, Imm32(0)); + // Set the hold appropriately. + ma_b(remain, remain, &negative, Signed, ShortJump); + ma_li(hold, Imm32(1)); + ma_branch(&head, ShortJump); + + bind(&negative); + ma_li(hold, Imm32(-1)); + subw(remain, zero, remain); + + // Begin the main loop. + bind(&head); + + UseScratchRegisterScope temps(this); + Register scratch2 = temps.Acquire(); + // Extract the bottom bits into SecondScratchReg. + ma_and(scratch2, remain, Imm32(mask)); + // Add those bits to the accumulator. + addw(dest, dest, scratch2); + // Do a trial subtraction + ma_sub32(scratch2, dest, Imm32(mask)); + // If (sum - C) > 0, store sum - C back into sum, thus performing a + // modulus. + ma_b(scratch2, Register(scratch2), &sumSigned, Signed, ShortJump); + or_(dest, scratch2, zero); + bind(&sumSigned); + // Get rid of the bits that we extracted before. + srliw(remain, remain, shift); + // If the shift produced zero, finish, otherwise, continue in the loop. + ma_b(remain, remain, &head, NonZero, ShortJump); + // Check the hold to see if we need to negate the result. + ma_b(hold, hold, &done, NotSigned, ShortJump); + + // If the hold was non-zero, negate the result to be in line with + // what JS wants + if (negZero != nullptr) { + // Jump out in case of negative zero. + ma_b(hold, hold, negZero, Zero); + subw(dest, zero, dest); + } else { + subw(dest, zero, dest); + } + + bind(&done); +} + +void MacroAssemblerRiscv64::ma_fmovz(FloatFormat fmt, FloatRegister fd, + FloatRegister fj, Register rk) { + Label done; + ma_b(rk, zero, &done, Assembler::NotEqual); + if (fmt == SingleFloat) { + fmv_s(fd, fj); + } else { + fmv_d(fd, fj); + } + bind(&done); +} + +void MacroAssemblerRiscv64::ByteSwap(Register rd, Register rs, int operand_size, + Register scratch) { + MOZ_ASSERT(scratch != rs); + MOZ_ASSERT(scratch != rd); + MOZ_ASSERT(operand_size == 4 || operand_size == 8); + if (operand_size == 4) { + // Uint32_t x1 = 0x00FF00FF; + // x0 = (x0 << 16 | x0 >> 16); + // x0 = (((x0 & x1) << 8) | ((x0 & (x1 << 8)) >> 8)); + UseScratchRegisterScope temps(this); + BlockTrampolinePoolScope block_trampoline_pool(this, 17); + MOZ_ASSERT((rd != t6) && (rs != t6)); + Register x0 = temps.Acquire(); + Register x1 = temps.Acquire(); + Register x2 = scratch; + RV_li(x1, 0x00FF00FF); + slliw(x0, rs, 16); + srliw(rd, rs, 16); + or_(x0, rd, x0); // x0 <- x0 << 16 | x0 >> 16 + and_(x2, x0, x1); // x2 <- x0 & 0x00FF00FF + slliw(x2, x2, 8); // x2 <- (x0 & x1) << 8 + slliw(x1, x1, 8); // x1 <- 0xFF00FF00 + and_(rd, x0, x1); // x0 & 0xFF00FF00 + srliw(rd, rd, 8); + or_(rd, rd, x2); // (((x0 & x1) << 8) | ((x0 & (x1 << 8)) >> 8)) + } else { + // uinx24_t x1 = 0x0000FFFF0000FFFFl; + // uinx24_t x1 = 0x00FF00FF00FF00FFl; + // x0 = (x0 << 32 | x0 >> 32); + // x0 = (x0 & x1) << 16 | (x0 & (x1 << 16)) >> 16; + // x0 = (x0 & x1) << 8 | (x0 & (x1 << 8)) >> 8; + UseScratchRegisterScope temps(this); + BlockTrampolinePoolScope block_trampoline_pool(this, 30); + MOZ_ASSERT((rd != t6) && (rs != t6)); + Register x0 = temps.Acquire(); + Register x1 = temps.Acquire(); + Register x2 = scratch; + RV_li(x1, 0x0000FFFF0000FFFFl); + slli(x0, rs, 32); + srli(rd, rs, 32); + or_(x0, rd, x0); // x0 <- x0 << 32 | x0 >> 32 + and_(x2, x0, x1); // x2 <- x0 & 0x0000FFFF0000FFFF + slli(x2, x2, 16); // x2 <- (x0 & 0x0000FFFF0000FFFF) << 16 + slli(x1, x1, 16); // x1 <- 0xFFFF0000FFFF0000 + and_(rd, x0, x1); // rd <- x0 & 0xFFFF0000FFFF0000 + srli(rd, rd, 16); // rd <- x0 & (x1 << 16)) >> 16 + or_(x0, rd, x2); // (x0 & x1) << 16 | (x0 & (x1 << 16)) >> 16; + RV_li(x1, 0x00FF00FF00FF00FFl); + and_(x2, x0, x1); // x2 <- x0 & 0x00FF00FF00FF00FF + slli(x2, x2, 8); // x2 <- (x0 & x1) << 8 + slli(x1, x1, 8); // x1 <- 0xFF00FF00FF00FF00 + and_(rd, x0, x1); + srli(rd, rd, 8); // rd <- (x0 & (x1 << 8)) >> 8 + or_(rd, rd, x2); // (((x0 & x1) << 8) | ((x0 & (x1 << 8)) >> 8)) + } +} + +template <typename F_TYPE> +void MacroAssemblerRiscv64::FloatMinMaxHelper(FPURegister dst, FPURegister src1, + FPURegister src2, + MaxMinKind kind) { + MOZ_ASSERT((std::is_same<F_TYPE, float>::value) || + (std::is_same<F_TYPE, double>::value)); + + if (src1 == src2 && dst != src1) { + if (std::is_same<float, F_TYPE>::value) { + fmv_s(dst, src1); + } else { + fmv_d(dst, src1); + } + return; + } + + Label done, nan; + + // For RISCV, fmin_s returns the other non-NaN operand as result if only one + // operand is NaN; but for JS, if any operand is NaN, result is Nan. The + // following handles the discrepency between handling of NaN between ISA and + // JS semantics + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + if (std::is_same<float, F_TYPE>::value) { + CompareIsNotNanF32(scratch, src1, src2); + } else { + CompareIsNotNanF64(scratch, src1, src2); + } + BranchFalseF(scratch, &nan); + + if (kind == MaxMinKind::kMax) { + if (std::is_same<float, F_TYPE>::value) { + fmax_s(dst, src1, src2); + } else { + fmax_d(dst, src1, src2); + } + } else { + if (std::is_same<float, F_TYPE>::value) { + fmin_s(dst, src1, src2); + } else { + fmin_d(dst, src1, src2); + } + } + jump(&done); + + bind(&nan); + // if any operand is NaN, return NaN (fadd returns NaN if any operand is NaN) + if (std::is_same<float, F_TYPE>::value) { + fadd_s(dst, src1, src2); + } else { + fadd_d(dst, src1, src2); + } + + bind(&done); +} + +void MacroAssemblerRiscv64::Float32Max(FPURegister dst, FPURegister src1, + FPURegister src2) { + comment(__FUNCTION__); + FloatMinMaxHelper<float>(dst, src1, src2, MaxMinKind::kMax); +} + +void MacroAssemblerRiscv64::Float32Min(FPURegister dst, FPURegister src1, + FPURegister src2) { + comment(__FUNCTION__); + FloatMinMaxHelper<float>(dst, src1, src2, MaxMinKind::kMin); +} + +void MacroAssemblerRiscv64::Float64Max(FPURegister dst, FPURegister src1, + FPURegister src2) { + comment(__FUNCTION__); + FloatMinMaxHelper<double>(dst, src1, src2, MaxMinKind::kMax); +} + +void MacroAssemblerRiscv64::Float64Min(FPURegister dst, FPURegister src1, + FPURegister src2) { + comment(__FUNCTION__); + FloatMinMaxHelper<double>(dst, src1, src2, MaxMinKind::kMin); +} + +void MacroAssemblerRiscv64::BranchTrueShortF(Register rs, Label* target) { + ma_branch(target, NotEqual, rs, Operand(zero_reg)); +} + +void MacroAssemblerRiscv64::BranchFalseShortF(Register rs, Label* target) { + ma_branch(target, Equal, rs, Operand(zero_reg)); +} + +void MacroAssemblerRiscv64::BranchTrueF(Register rs, Label* target) { + bool long_branch = target->bound() ? !is_near(target) : false; + if (long_branch) { + Label skip; + BranchFalseShortF(rs, &skip); + BranchLong(target); + bind(&skip); + } else { + BranchTrueShortF(rs, target); + } +} + +void MacroAssemblerRiscv64::BranchFalseF(Register rs, Label* target) { + bool long_branch = target->bound() ? !is_near(target) : false; + if (long_branch) { + Label skip; + BranchTrueShortF(rs, &skip); + BranchLong(target); + bind(&skip); + } else { + BranchFalseShortF(rs, target); + } +} + +void MacroAssemblerRiscv64::Ror(Register rd, Register rs, const Operand& rt) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + if (rt.is_reg()) { + negw(scratch, rt.rm()); + sllw(scratch, rs, scratch); + srlw(rd, rs, rt.rm()); + or_(rd, scratch, rd); + sext_w(rd, rd); + } else { + int64_t ror_value = rt.immediate() % 32; + if (ror_value == 0) { + mv(rd, rs); + return; + } else if (ror_value < 0) { + ror_value += 32; + } + srliw(scratch, rs, ror_value); + slliw(rd, rs, 32 - ror_value); + or_(rd, scratch, rd); + sext_w(rd, rd); + } +} + +void MacroAssemblerRiscv64::Dror(Register rd, Register rs, const Operand& rt) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + if (rt.is_reg()) { + negw(scratch, rt.rm()); + sll(scratch, rs, scratch); + srl(rd, rs, rt.rm()); + or_(rd, scratch, rd); + } else { + int64_t dror_value = rt.immediate() % 64; + if (dror_value == 0) { + mv(rd, rs); + return; + } else if (dror_value < 0) { + dror_value += 64; + } + srli(scratch, rs, dror_value); + slli(rd, rs, 64 - dror_value); + or_(rd, scratch, rd); + } +} + +void MacroAssemblerRiscv64::wasmLoadImpl(const wasm::MemoryAccessDesc& access, + Register memoryBase, Register ptr, + Register ptrScratch, + AnyRegister output, Register tmp) { + access.assertOffsetInGuardPages(); + uint32_t offset = access.offset(); + MOZ_ASSERT_IF(offset, ptrScratch != InvalidReg); + + // Maybe add the offset. + if (offset) { + asMasm().addPtr(ImmWord(offset), ptrScratch); + ptr = ptrScratch; + } + + asMasm().memoryBarrierBefore(access.sync()); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + FaultingCodeOffset fco; + switch (access.type()) { + case Scalar::Int8: + add(scratch, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lb(output.gpr(), scratch, 0); + break; + case Scalar::Uint8: + add(scratch, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lbu(output.gpr(), scratch, 0); + break; + case Scalar::Int16: + add(scratch, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lh(output.gpr(), scratch, 0); + break; + case Scalar::Uint16: + add(scratch, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lhu(output.gpr(), scratch, 0); + break; + case Scalar::Int32: + add(scratch, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lw(output.gpr(), scratch, 0); + break; + case Scalar::Uint32: + add(scratch, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + lwu(output.gpr(), scratch, 0); + break; + case Scalar::Float64: + add(scratch, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + fld(output.fpu(), scratch, 0); + break; + case Scalar::Float32: + add(scratch, memoryBase, ptr); + fco = FaultingCodeOffset(currentOffset()); + flw(output.fpu(), scratch, 0); + break; + default: + MOZ_CRASH("unexpected array type"); + } + + asMasm().append(access, js::wasm::TrapMachineInsnForLoad(access.byteSize()), + fco); + asMasm().memoryBarrierAfter(access.sync()); +} + +void MacroAssemblerRiscv64::wasmStoreImpl(const wasm::MemoryAccessDesc& access, + AnyRegister value, + Register memoryBase, Register ptr, + Register ptrScratch, Register tmp) { + access.assertOffsetInGuardPages(); + uint32_t offset = access.offset(); + MOZ_ASSERT_IF(offset, ptrScratch != InvalidReg); + + // Maybe add the offset. + if (offset) { + asMasm().addPtr(ImmWord(offset), ptrScratch); + ptr = ptrScratch; + } + + unsigned byteSize = access.byteSize(); + bool isSigned; + bool isFloat = false; + + switch (access.type()) { + case Scalar::Int8: + isSigned = true; + break; + case Scalar::Uint8: + isSigned = false; + break; + case Scalar::Int16: + isSigned = true; + break; + case Scalar::Uint16: + isSigned = false; + break; + case Scalar::Int32: + isSigned = true; + break; + case Scalar::Uint32: + isSigned = false; + break; + case Scalar::Int64: + isSigned = true; + break; + case Scalar::Float64: + isFloat = true; + break; + case Scalar::Float32: + isFloat = true; + break; + default: + MOZ_CRASH("unexpected array type"); + } + + BaseIndex address(memoryBase, ptr, TimesOne); + asMasm().memoryBarrierBefore(access.sync()); + FaultingCodeOffset fco; + if (isFloat) { + if (byteSize == 4) { + fco = asMasm().ma_fst_s(value.fpu(), address); + } else { + fco = asMasm().ma_fst_d(value.fpu(), address); + } + } else { + fco = asMasm().ma_store(value.gpr(), address, + static_cast<LoadStoreSize>(8 * byteSize), + isSigned ? SignExtend : ZeroExtend); + } + // Only the last emitted instruction is a memory access. + asMasm().append(access, js::wasm::TrapMachineInsnForStore(access.byteSize()), + fco); + asMasm().memoryBarrierAfter(access.sync()); +} + +void MacroAssemblerRiscv64::GenPCRelativeJumpAndLink(Register rd, + int32_t imm32) { + MOZ_ASSERT(is_int32(imm32 + 0x800)); + int32_t Hi20 = ((imm32 + 0x800) >> 12); + int32_t Lo12 = imm32 << 20 >> 20; + auipc(rd, Hi20); // Read PC + Hi20 into scratch. + jalr(rd, Lo12); // jump PC + Hi20 + Lo12 +} + +void MacroAssemblerRiscv64::BranchAndLinkShortHelper(int32_t offset, Label* L) { + MOZ_ASSERT(L == nullptr || offset == 0); + offset = GetOffset(offset, L, OffsetSize::kOffset21); + jal(offset); +} + +void MacroAssemblerRiscv64::BranchAndLinkShort(int32_t offset) { + MOZ_ASSERT(is_int21(offset)); + BranchAndLinkShortHelper(offset, nullptr); +} + +void MacroAssemblerRiscv64::BranchAndLinkShort(Label* L) { + BranchAndLinkShortHelper(0, L); +} + +void MacroAssemblerRiscv64::BranchAndLink(Label* L) { + if (L->bound()) { + if (is_near(L)) { + BranchAndLinkShort(L); + } else { + BranchAndLinkLong(L); + } + } else { + BranchAndLinkShort(L); + } +} + +void MacroAssemblerRiscv64::ma_fmv_d(FloatRegister src, ValueOperand dest) { + fmv_x_d(dest.valueReg(), src); +} + +void MacroAssemblerRiscv64::ma_fmv_d(ValueOperand src, FloatRegister dest) { + fmv_d_x(dest, src.valueReg()); +} + +void MacroAssemblerRiscv64::ma_fmv_w(FloatRegister src, ValueOperand dest) { + fmv_x_w(dest.valueReg(), src); +} + +void MacroAssemblerRiscv64::ma_fmv_w(ValueOperand src, FloatRegister dest) { + fmv_w_x(dest, src.valueReg()); +} + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/MacroAssembler-riscv64.h b/js/src/jit/riscv64/MacroAssembler-riscv64.h new file mode 100644 index 0000000000..0b4b45f67f --- /dev/null +++ b/js/src/jit/riscv64/MacroAssembler-riscv64.h @@ -0,0 +1,1249 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +// Copyright 2021 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_MacroAssembler_riscv64_h +#define jit_riscv64_MacroAssembler_riscv64_h + +#include <iterator> + +#include "jit/MoveResolver.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "wasm/WasmTypeDecls.h" + +namespace js { +namespace jit { + +static Register CallReg = t6; + +enum LiFlags { + Li64 = 0, + Li48 = 1, +}; + +class CompactBufferReader; +enum LoadStoreSize { + SizeByte = 8, + SizeHalfWord = 16, + SizeWord = 32, + SizeDouble = 64 +}; + +enum LoadStoreExtension { ZeroExtend = 0, SignExtend = 1 }; +enum JumpKind { LongJump = 0, ShortJump = 1 }; +enum FloatFormat { SingleFloat, DoubleFloat }; +class ScratchTagScope : public ScratchRegisterScope { + public: + ScratchTagScope(MacroAssembler& masm, const ValueOperand&) + : ScratchRegisterScope(masm) {} +}; + +class ScratchTagScopeRelease { + ScratchTagScope* ts_; + + public: + explicit ScratchTagScopeRelease(ScratchTagScope* ts) : ts_(ts) { + ts_->release(); + } + ~ScratchTagScopeRelease() { ts_->reacquire(); } +}; + +struct ImmTag : public Imm32 { + ImmTag(JSValueTag mask) : Imm32(int32_t(mask)) {} +}; + +class MacroAssemblerRiscv64 : public Assembler { + public: + MacroAssemblerRiscv64() {} + +#ifdef JS_SIMULATOR_RISCV64 + // See riscv64/base-constants-riscv.h DebugParameters. + void Debug(uint32_t parameters) { break_(parameters, false); } +#endif + + // Perform a downcast. Should be removed by Bug 996602. + MacroAssembler& asMasm(); + const MacroAssembler& asMasm() const; + + MoveResolver moveResolver_; + + static bool SupportsFloatingPoint() { return true; } + static bool SupportsUnalignedAccesses() { return true; } + static bool SupportsFastUnalignedFPAccesses() { return true; } + void haltingAlign(int alignment) { + // TODO(loong64): Implement a proper halting align. + nopAlign(alignment); + } + + // TODO(RISCV) Reorder parameters so out parameters come last. + bool CalculateOffset(Label* L, int32_t* offset, OffsetSize bits); + int32_t GetOffset(int32_t offset, Label* L, OffsetSize bits); + + inline void GenPCRelativeJump(Register rd, int32_t imm32) { + MOZ_ASSERT(is_int32(imm32 + 0x800)); + int32_t Hi20 = ((imm32 + 0x800) >> 12); + int32_t Lo12 = imm32 << 20 >> 20; + auipc(rd, Hi20); // Read PC + Hi20 into scratch. + jr(rd, Lo12); // jump PC + Hi20 + Lo12 + } + + // load + FaultingCodeOffset ma_load(Register dest, Address address, + LoadStoreSize size = SizeWord, + LoadStoreExtension extension = SignExtend); + FaultingCodeOffset ma_load(Register dest, const BaseIndex& src, + LoadStoreSize size = SizeWord, + LoadStoreExtension extension = SignExtend); + FaultingCodeOffset ma_loadDouble(FloatRegister dest, Address address); + FaultingCodeOffset ma_loadFloat(FloatRegister dest, Address address); + // store + FaultingCodeOffset ma_store(Register data, Address address, + LoadStoreSize size = SizeWord, + LoadStoreExtension extension = SignExtend); + FaultingCodeOffset ma_store(Register data, const BaseIndex& dest, + LoadStoreSize size = SizeWord, + LoadStoreExtension extension = SignExtend); + FaultingCodeOffset ma_store(Imm32 imm, const BaseIndex& dest, + LoadStoreSize size = SizeWord, + LoadStoreExtension extension = SignExtend); + FaultingCodeOffset ma_store(Imm32 imm, Address address, + LoadStoreSize size = SizeWord, + LoadStoreExtension extension = SignExtend); + void ma_storeDouble(FloatRegister dest, Address address); + void ma_storeFloat(FloatRegister dest, Address address); + void ma_liPatchable(Register dest, Imm32 imm); + void ma_liPatchable(Register dest, ImmPtr imm); + void ma_liPatchable(Register dest, ImmWord imm, LiFlags flags = Li48); + void ma_li(Register dest, ImmGCPtr ptr); + void ma_li(Register dest, Imm32 imm); + void ma_li(Register dest, Imm64 imm); + void ma_li(Register dest, intptr_t imm) { RV_li(dest, imm); } + void ma_li(Register dest, CodeLabel* label); + void ma_li(Register dest, ImmWord imm); + + // branches when done from within la-specific code + void ma_b(Register lhs, Register rhs, Label* l, Condition c, + JumpKind jumpKind = LongJump); + void ma_b(Register lhs, Imm32 imm, Label* l, Condition c, + JumpKind jumpKind = LongJump); + void BranchAndLinkShort(Label* L); + void BranchAndLink(Label* label); + void BranchAndLinkShort(int32_t offset); + void BranchAndLinkShortHelper(int32_t offset, Label* L); + void BranchAndLinkLong(Label* L); + void GenPCRelativeJumpAndLink(Register rd, int32_t imm32); + +#define DEFINE_INSTRUCTION(instr) \ + void instr(Register rd, Register rj, Operand rt); \ + void instr(Register rd, Register rj, Imm32 imm) { \ + instr(rd, rj, Operand(imm.value)); \ + } \ + void instr(Register rd, Imm32 imm) { instr(rd, rd, Operand(imm.value)); } \ + void instr(Register rd, Register rs) { instr(rd, rd, Operand(rs)); } + +#define DEFINE_INSTRUCTION2(instr) \ + void instr(Register rs, const Operand& rt); \ + void instr(Register rs, Register rt) { instr(rs, Operand(rt)); } \ + void instr(Register rs, Imm32 j) { instr(rs, Operand(j.value)); } + + DEFINE_INSTRUCTION(ma_and); + DEFINE_INSTRUCTION(ma_or); + DEFINE_INSTRUCTION(ma_xor); + DEFINE_INSTRUCTION(ma_nor); + DEFINE_INSTRUCTION(ma_sub32) + DEFINE_INSTRUCTION(ma_sub64) + DEFINE_INSTRUCTION(ma_add32) + DEFINE_INSTRUCTION(ma_add64) + DEFINE_INSTRUCTION(ma_div32) + DEFINE_INSTRUCTION(ma_divu32) + DEFINE_INSTRUCTION(ma_div64) + DEFINE_INSTRUCTION(ma_divu64) + DEFINE_INSTRUCTION(ma_mod32) + DEFINE_INSTRUCTION(ma_modu32) + DEFINE_INSTRUCTION(ma_mod64) + DEFINE_INSTRUCTION(ma_modu64) + DEFINE_INSTRUCTION(ma_mul32) + DEFINE_INSTRUCTION(ma_mulh32) + DEFINE_INSTRUCTION(ma_mulhu32) + DEFINE_INSTRUCTION(ma_mul64) + DEFINE_INSTRUCTION(ma_mulh64) + DEFINE_INSTRUCTION(ma_sll64) + DEFINE_INSTRUCTION(ma_sra64) + DEFINE_INSTRUCTION(ma_srl64) + DEFINE_INSTRUCTION(ma_sll32) + DEFINE_INSTRUCTION(ma_sra32) + DEFINE_INSTRUCTION(ma_srl32) + DEFINE_INSTRUCTION(ma_slt) + DEFINE_INSTRUCTION(ma_sltu) + DEFINE_INSTRUCTION(ma_sle) + DEFINE_INSTRUCTION(ma_sleu) + DEFINE_INSTRUCTION(ma_sgt) + DEFINE_INSTRUCTION(ma_sgtu) + DEFINE_INSTRUCTION(ma_sge) + DEFINE_INSTRUCTION(ma_sgeu) + DEFINE_INSTRUCTION(ma_seq) + DEFINE_INSTRUCTION(ma_sne) + + DEFINE_INSTRUCTION2(ma_seqz) + DEFINE_INSTRUCTION2(ma_snez) + DEFINE_INSTRUCTION2(ma_neg); + +#undef DEFINE_INSTRUCTION2 +#undef DEFINE_INSTRUCTION + // arithmetic based ops + void ma_add32TestOverflow(Register rd, Register rj, Register rk, + Label* overflow); + void ma_add32TestOverflow(Register rd, Register rj, Imm32 imm, + Label* overflow); + void ma_addPtrTestOverflow(Register rd, Register rj, Register rk, + Label* overflow); + void ma_addPtrTestOverflow(Register rd, Register rj, Imm32 imm, + Label* overflow); + void ma_addPtrTestOverflow(Register rd, Register rj, ImmWord imm, + Label* overflow); + void ma_addPtrTestCarry(Condition cond, Register rd, Register rj, Register rk, + Label* overflow); + void ma_addPtrTestCarry(Condition cond, Register rd, Register rj, Imm32 imm, + Label* overflow); + void ma_addPtrTestCarry(Condition cond, Register rd, Register rj, ImmWord imm, + Label* overflow); + + // subtract + void ma_sub32TestOverflow(Register rd, Register rj, Register rk, + Label* overflow); + void ma_subPtrTestOverflow(Register rd, Register rj, Register rk, + Label* overflow); + void ma_subPtrTestOverflow(Register rd, Register rj, Imm32 imm, + Label* overflow); + + // multiplies. For now, there are only few that we care about. + void ma_mulPtrTestOverflow(Register rd, Register rj, Register rk, + Label* overflow); + + // branches when done from within la-specific code + void ma_b(Register lhs, ImmWord imm, Label* l, Condition c, + JumpKind jumpKind = LongJump); + void ma_b(Register lhs, ImmPtr imm, Label* l, Condition c, + JumpKind jumpKind = LongJump); + void ma_b(Register lhs, ImmGCPtr imm, Label* l, Condition c, + JumpKind jumpKind = LongJump) { + UseScratchRegisterScope temps(this); + Register ScratchRegister = temps.Acquire(); + ma_li(ScratchRegister, imm); + ma_b(lhs, ScratchRegister, l, c, jumpKind); + } + void ma_b(Register lhs, Address addr, Label* l, Condition c, + JumpKind jumpKind = LongJump); + void ma_b(Address addr, Imm32 imm, Label* l, Condition c, + JumpKind jumpKind = LongJump); + void ma_b(Address addr, ImmGCPtr imm, Label* l, Condition c, + JumpKind jumpKind = LongJump); + void ma_b(Address addr, Register rhs, Label* l, Condition c, + JumpKind jumpKind = LongJump) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(rhs != scratch); + ma_load(scratch, addr, SizeDouble); + ma_b(scratch, rhs, l, c, jumpKind); + } + + void ma_branch(Label* target, Condition cond, Register r1, const Operand& r2, + JumpKind jumpKind = ShortJump); + + void ma_branch(Label* target, JumpKind jumpKind = ShortJump) { + ma_branch(target, Always, zero, zero, jumpKind); + } + + // fp instructions + void ma_lid(FloatRegister dest, double value); + + // fp instructions + void ma_lis(FloatRegister dest, float value); + + FaultingCodeOffset ma_fst_d(FloatRegister src, BaseIndex address); + FaultingCodeOffset ma_fst_s(FloatRegister src, BaseIndex address); + + void ma_fld_d(FloatRegister dest, const BaseIndex& src); + void ma_fld_s(FloatRegister dest, const BaseIndex& src); + + void ma_fmv_d(FloatRegister src, ValueOperand dest); + void ma_fmv_d(ValueOperand src, FloatRegister dest); + + void ma_fmv_w(FloatRegister src, ValueOperand dest); + void ma_fmv_w(ValueOperand src, FloatRegister dest); + + FaultingCodeOffset ma_fld_s(FloatRegister ft, Address address); + FaultingCodeOffset ma_fld_d(FloatRegister ft, Address address); + FaultingCodeOffset ma_fst_d(FloatRegister ft, Address address); + FaultingCodeOffset ma_fst_s(FloatRegister ft, Address address); + + // stack + void ma_pop(Register r); + void ma_push(Register r); + void ma_pop(FloatRegister f); + void ma_push(FloatRegister f); + + Condition ma_cmp(Register rd, Register lhs, Register rhs, Condition c); + Condition ma_cmp(Register rd, Register lhs, Imm32 imm, Condition c); + void ma_cmp_set(Register dst, Register lhs, ImmWord imm, Condition c); + void ma_cmp_set(Register dst, Register lhs, ImmPtr imm, Condition c); + void ma_cmp_set(Register dst, Address address, Imm32 imm, Condition c); + void ma_cmp_set(Register dst, Address address, ImmWord imm, Condition c); + + void ma_rotr_w(Register rd, Register rj, Imm32 shift); + + void ma_fmovz(FloatFormat fmt, FloatRegister fd, FloatRegister fj, + Register rk); + void ma_fmovn(FloatFormat fmt, FloatRegister fd, FloatRegister fj, + Register rk); + + // arithmetic based ops + void ma_add32TestCarry(Condition cond, Register rd, Register rj, Register rk, + Label* overflow); + void ma_add32TestCarry(Condition cond, Register rd, Register rj, Imm32 imm, + Label* overflow); + + // subtract + void ma_sub32TestOverflow(Register rd, Register rj, Imm32 imm, + Label* overflow); + + void MulOverflow32(Register dst, Register left, const Operand& right, + Register overflow); + // multiplies. For now, there are only few that we care about. + void ma_mul32TestOverflow(Register rd, Register rj, Register rk, + Label* overflow); + void ma_mul32TestOverflow(Register rd, Register rj, Imm32 imm, + Label* overflow); + + // divisions + void ma_div_branch_overflow(Register rd, Register rj, Register rk, + Label* overflow); + void ma_div_branch_overflow(Register rd, Register rj, Imm32 imm, + Label* overflow); + + // fast mod, uses scratch registers, and thus needs to be in the assembler + // implicitly assumes that we can overwrite dest at the beginning of the + // sequence + void ma_mod_mask(Register src, Register dest, Register hold, Register remain, + int32_t shift, Label* negZero = nullptr); + + // FP branches + void ma_compareF32(Register rd, DoubleCondition cc, FloatRegister cmp1, + FloatRegister cmp2); + void ma_compareF64(Register rd, DoubleCondition cc, FloatRegister cmp1, + FloatRegister cmp2); + + void CompareIsNotNanF32(Register rd, FPURegister cmp1, FPURegister cmp2); + void CompareIsNotNanF64(Register rd, FPURegister cmp1, FPURegister cmp2); + void CompareIsNanF32(Register rd, FPURegister cmp1, FPURegister cmp2); + void CompareIsNanF64(Register rd, FPURegister cmp1, FPURegister cmp2); + + void ma_call(ImmPtr dest); + + void ma_jump(ImmPtr dest); + + void jump(Label* label) { ma_branch(label); } + void jump(Register reg) { jr(reg); } + + void ma_cmp_set(Register dst, Register lhs, Register rhs, Condition c); + void ma_cmp_set(Register dst, Register lhs, Imm32 imm, Condition c); + + void computeScaledAddress(const BaseIndex& address, Register dest); + + void BranchShort(Label* L); + + void BranchShort(int32_t offset, Condition cond, Register rs, + const Operand& rt); + void BranchShort(Label* L, Condition cond, Register rs, const Operand& rt); + void BranchShortHelper(int32_t offset, Label* L); + bool BranchShortHelper(int32_t offset, Label* L, Condition cond, Register rs, + const Operand& rt); + bool BranchShortCheck(int32_t offset, Label* L, Condition cond, Register rs, + const Operand& rt); + void BranchLong(Label* L); + + // Floating point branches + void BranchTrueShortF(Register rs, Label* target); + void BranchFalseShortF(Register rs, Label* target); + + void BranchTrueF(Register rs, Label* target); + void BranchFalseF(Register rs, Label* target); + + void moveFromDoubleHi(FloatRegister src, Register dest) { + fmv_x_d(dest, src); + srli(dest, dest, 32); + } + // Bit field starts at bit pos and extending for size bits is extracted from + // rs and stored zero/sign-extended and right-justified in rt + void ExtractBits(Register rt, Register rs, uint16_t pos, uint16_t size, + bool sign_extend = false); + void ExtractBits(Register dest, Register source, Register pos, int size, + bool sign_extend = false) { + sra(dest, source, pos); + ExtractBits(dest, dest, 0, size, sign_extend); + } + + // Insert bits [0, size) of source to bits [pos, pos+size) of dest + void InsertBits(Register dest, Register source, Register pos, int size); + + // Insert bits [0, size) of source to bits [pos, pos+size) of dest + void InsertBits(Register dest, Register source, int pos, int size); + + template <typename F_TYPE> + void RoundHelper(FPURegister dst, FPURegister src, FPURegister fpu_scratch, + FPURoundingMode mode); + + template <typename TruncFunc> + void RoundFloatingPointToInteger(Register rd, FPURegister fs, Register result, + TruncFunc trunc, bool Inexact = false); + + void Clear_if_nan_d(Register rd, FPURegister fs); + void Clear_if_nan_s(Register rd, FPURegister fs); + // Convert double to unsigned word. + void Trunc_uw_d(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Convert double to signed word. + void Trunc_w_d(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Convert double to unsigned long. + void Trunc_ul_d(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Convert singled to signed long. + void Trunc_l_d(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Convert single to signed word. + void Trunc_w_s(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Convert single to unsigned word. + void Trunc_uw_s(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Convert single to unsigned long. + void Trunc_ul_s(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Convert singled to signed long. + void Trunc_l_s(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Round double functions + void Trunc_d_d(FPURegister fd, FPURegister fs, FPURegister fpu_scratch); + void Round_d_d(FPURegister fd, FPURegister fs, FPURegister fpu_scratch); + void Floor_d_d(FPURegister fd, FPURegister fs, FPURegister fpu_scratch); + void Ceil_d_d(FPURegister fd, FPURegister fs, FPURegister fpu_scratch); + + // Round float functions + void Trunc_s_s(FPURegister fd, FPURegister fs, FPURegister fpu_scratch); + void Round_s_s(FPURegister fd, FPURegister fs, FPURegister fpu_scratch); + void Floor_s_s(FPURegister fd, FPURegister fs, FPURegister fpu_scratch); + void Ceil_s_s(FPURegister fd, FPURegister fs, FPURegister fpu_scratch); + + // Round single to signed word. + void Round_w_s(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Round double to signed word. + void Round_w_d(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Ceil single to signed word. + void Ceil_w_s(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Ceil double to signed word. + void Ceil_w_d(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Floor single to signed word. + void Floor_w_s(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + // Floor double to signed word. + void Floor_w_d(Register rd, FPURegister fs, Register result = InvalidReg, + bool Inexact = false); + + void Clz32(Register rd, Register rs); + void Ctz32(Register rd, Register rs); + void Popcnt32(Register rd, Register rs, Register scratch); + + void Popcnt64(Register rd, Register rs, Register scratch); + void Ctz64(Register rd, Register rs); + void Clz64(Register rd, Register rs); + + // Change endianness + void ByteSwap(Register dest, Register src, int operand_size, + Register scratch); + + void Ror(Register rd, Register rs, const Operand& rt); + void Dror(Register rd, Register rs, const Operand& rt); + + void Float32Max(FPURegister dst, FPURegister src1, FPURegister src2); + void Float32Min(FPURegister dst, FPURegister src1, FPURegister src2); + void Float64Max(FPURegister dst, FPURegister src1, FPURegister src2); + void Float64Min(FPURegister dst, FPURegister src1, FPURegister src2); + + template <typename F> + void FloatMinMaxHelper(FPURegister dst, FPURegister src1, FPURegister src2, + MaxMinKind kind); + + inline void NegateBool(Register rd, Register rs) { xori(rd, rs, 1); } + + protected: + void wasmLoadImpl(const wasm::MemoryAccessDesc& access, Register memoryBase, + Register ptr, Register ptrScratch, AnyRegister output, + Register tmp); + void wasmStoreImpl(const wasm::MemoryAccessDesc& access, AnyRegister value, + Register memoryBase, Register ptr, Register ptrScratch, + Register tmp); +}; + +class MacroAssemblerRiscv64Compat : public MacroAssemblerRiscv64 { + public: + using MacroAssemblerRiscv64::call; + + MacroAssemblerRiscv64Compat() {} + + void convertBoolToInt32(Register src, Register dest) { + ma_and(dest, src, Imm32(0xff)); + }; + void convertInt32ToDouble(Register src, FloatRegister dest) { + fcvt_d_w(dest, src); + }; + void convertInt32ToDouble(const Address& src, FloatRegister dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_load(scratch, src, SizeWord, SignExtend); + fcvt_d_w(dest, scratch); + }; + void convertInt32ToDouble(const BaseIndex& src, FloatRegister dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(scratch != src.base); + MOZ_ASSERT(scratch != src.index); + computeScaledAddress(src, scratch); + convertInt32ToDouble(Address(scratch, src.offset), dest); + }; + void convertUInt32ToDouble(Register src, FloatRegister dest); + void convertUInt32ToFloat32(Register src, FloatRegister dest); + void convertDoubleToFloat32(FloatRegister src, FloatRegister dest); + void convertDoubleToInt32(FloatRegister src, Register dest, Label* fail, + bool negativeZeroCheck = true); + void convertDoubleToPtr(FloatRegister src, Register dest, Label* fail, + bool negativeZeroCheck = true); + void convertFloat32ToInt32(FloatRegister src, Register dest, Label* fail, + bool negativeZeroCheck = true); + + void convertFloat32ToDouble(FloatRegister src, FloatRegister dest); + void convertInt32ToFloat32(Register src, FloatRegister dest); + void convertInt32ToFloat32(const Address& src, FloatRegister dest); + + void movq(Register rj, Register rd); + + void computeEffectiveAddress(const Address& address, Register dest) { + ma_add64(dest, address.base, Imm32(address.offset)); + } + + void computeEffectiveAddress(const BaseIndex& address, Register dest) { + computeScaledAddress(address, dest); + if (address.offset) { + ma_add64(dest, dest, Imm32(address.offset)); + } + } + + void j(Label* dest) { ma_branch(dest); } + + void mov(Register src, Register dest) { addi(dest, src, 0); } + void mov(ImmWord imm, Register dest) { ma_li(dest, imm); } + void mov(ImmPtr imm, Register dest) { + mov(ImmWord(uintptr_t(imm.value)), dest); + } + void mov(CodeLabel* label, Register dest) { ma_li(dest, label); } + void mov(Register src, Address dest) { MOZ_CRASH("NYI-IC"); } + void mov(Address src, Register dest) { MOZ_CRASH("NYI-IC"); } + + void writeDataRelocation(const Value& val) { + // Raw GC pointer relocations and Value relocations both end up in + // TraceOneDataRelocation. + if (val.isGCThing()) { + gc::Cell* cell = val.toGCThing(); + if (cell && gc::IsInsideNursery(cell)) { + embedsNurseryPointers_ = true; + } + dataRelocations_.writeUnsigned(currentOffset()); + } + } + + void branch(JitCode* c) { + BlockTrampolinePoolScope block_trampoline_pool(this, 7); + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + BufferOffset bo = m_buffer.nextOffset(); + addPendingJump(bo, ImmPtr(c->raw()), RelocationKind::JITCODE); + ma_liPatchable(scratch, ImmPtr(c->raw())); + jr(scratch); + } + void branch(const Register reg) { jr(reg); } + void ret() { + ma_pop(ra); + jalr(zero_reg, ra, 0); + } + inline void retn(Imm32 n); + void push(Imm32 imm) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + ma_push(scratch); + } + void push(ImmWord imm) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + ma_push(scratch); + } + void push(ImmGCPtr imm) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + ma_li(scratch, imm); + ma_push(scratch); + } + void push(const Address& address) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + loadPtr(address, scratch); + ma_push(scratch); + } + void push(Register reg) { ma_push(reg); } + void push(FloatRegister reg) { ma_push(reg); } + void pop(Register reg) { ma_pop(reg); } + void pop(FloatRegister reg) { ma_pop(reg); } + + // Emit a branch that can be toggled to a non-operation. On LOONG64 we use + // "andi" instruction to toggle the branch. + // See ToggleToJmp(), ToggleToCmp(). + CodeOffset toggledJump(Label* label); + + // Emit a "jalr" or "nop" instruction. ToggleCall can be used to patch + // this instruction. + CodeOffset toggledCall(JitCode* target, bool enabled); + + static size_t ToggledCallSize(uint8_t* code) { + // Four instructions used in: MacroAssemblerRiscv64Compat::toggledCall + return 7 * sizeof(uint32_t); + } + + CodeOffset pushWithPatch(ImmWord imm) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + CodeOffset offset = movWithPatch(imm, scratch); + ma_push(scratch); + return offset; + } + + CodeOffset movWithPatch(ImmWord imm, Register dest) { + BlockTrampolinePoolScope block_trampoline_pool(this, 8); + CodeOffset offset = CodeOffset(currentOffset()); + ma_liPatchable(dest, imm, Li64); + return offset; + } + CodeOffset movWithPatch(ImmPtr imm, Register dest) { + BlockTrampolinePoolScope block_trampoline_pool(this, 6); + CodeOffset offset = CodeOffset(currentOffset()); + ma_liPatchable(dest, imm); + return offset; + } + + void writeCodePointer(CodeLabel* label) { + label->patchAt()->bind(currentOffset()); + label->setLinkMode(CodeLabel::RawPointer); + m_buffer.ensureSpace(sizeof(void*)); + emit(uint32_t(-1)); + emit(uint32_t(-1)); + } + + void jump(Label* label) { ma_branch(label); } + void jump(Register reg) { jr(reg); } + void jump(const Address& address) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + loadPtr(address, scratch); + jr(scratch); + } + + void jump(JitCode* code) { branch(code); } + + void jump(ImmPtr ptr) { + BufferOffset bo = m_buffer.nextOffset(); + addPendingJump(bo, ptr, RelocationKind::HARDCODED); + ma_jump(ptr); + } + + void jump(TrampolinePtr code) { jump(ImmPtr(code.value)); } + + void splitTag(Register src, Register dest) { + srli(dest, src, JSVAL_TAG_SHIFT); + } + + void splitTag(const ValueOperand& operand, Register dest) { + splitTag(operand.valueReg(), dest); + } + + void splitTagForTest(const ValueOperand& value, ScratchTagScope& tag) { + splitTag(value, tag); + } + + void moveIfZero(Register dst, Register src, Register cond) { + ScratchRegisterScope scratch(asMasm()); + MOZ_ASSERT(dst != scratch && cond != scratch); + Label done; + ma_branch(&done, NotEqual, cond, zero); + mv(dst, src); + bind(&done); + } + + void moveIfNotZero(Register dst, Register src, Register cond) { + ScratchRegisterScope scratch(asMasm()); + MOZ_ASSERT(dst != scratch && cond != scratch); + Label done; + ma_branch(&done, Equal, cond, zero); + mv(dst, src); + bind(&done); + } + // unboxing code + void unboxNonDouble(const ValueOperand& operand, Register dest, + JSValueType type) { + unboxNonDouble(operand.valueReg(), dest, type); + } + + template <typename T> + void unboxNonDouble(T src, Register dest, JSValueType type) { + MOZ_ASSERT(type != JSVAL_TYPE_DOUBLE); + if (type == JSVAL_TYPE_INT32 || type == JSVAL_TYPE_BOOLEAN) { + load32(src, dest); + return; + } + loadPtr(src, dest); + unboxNonDouble(dest, dest, type); + } + + void unboxNonDouble(Register src, Register dest, JSValueType type) { + MOZ_ASSERT(type != JSVAL_TYPE_DOUBLE); + if (type == JSVAL_TYPE_INT32 || type == JSVAL_TYPE_BOOLEAN) { + slliw(dest, src, 0); + return; + } + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + MOZ_ASSERT(scratch != src); + mov(ImmWord(JSVAL_TYPE_TO_SHIFTED_TAG(type)), scratch); + xor_(dest, src, scratch); + } + + template <typename T> + void unboxObjectOrNull(const T& src, Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_OBJECT); + static_assert(JS::detail::ValueObjectOrNullBit == + (uint64_t(0x8) << JSVAL_TAG_SHIFT)); + InsertBits(dest, zero, JSVAL_TAG_SHIFT + 3, 1); + } + + void unboxGCThingForGCBarrier(const Address& src, Register dest) { + loadPtr(src, dest); + ExtractBits(dest, dest, 0, JSVAL_TAG_SHIFT - 1); + } + void unboxGCThingForGCBarrier(const ValueOperand& src, Register dest) { + ExtractBits(dest, src.valueReg(), 0, JSVAL_TAG_SHIFT - 1); + } + + void unboxWasmAnyRefGCThingForGCBarrier(const Address& src, Register dest) { + ScratchRegisterScope scratch(asMasm()); + MOZ_ASSERT(scratch != dest); + movePtr(ImmWord(wasm::AnyRef::GCThingMask), scratch); + loadPtr(src, dest); + ma_and(dest, dest, scratch); + } + + void getWasmAnyRefGCThingChunk(Register src, Register dest) { + MOZ_ASSERT(src != dest); + movePtr(ImmWord(wasm::AnyRef::GCThingChunkMask), dest); + ma_and(dest, dest, src); + } + + // Like unboxGCThingForGCBarrier, but loads the GC thing's chunk base. + void getGCThingValueChunk(const Address& src, Register dest) { + ScratchRegisterScope scratch(asMasm()); + MOZ_ASSERT(scratch != dest); + loadPtr(src, dest); + movePtr(ImmWord(JS::detail::ValueGCThingPayloadChunkMask), scratch); + and_(dest, dest, scratch); + } + void getGCThingValueChunk(const ValueOperand& src, Register dest) { + MOZ_ASSERT(src.valueReg() != dest); + movePtr(ImmWord(JS::detail::ValueGCThingPayloadChunkMask), dest); + and_(dest, dest, src.valueReg()); + } + + void unboxInt32(const ValueOperand& operand, Register dest); + void unboxInt32(Register src, Register dest); + void unboxInt32(const Address& src, Register dest); + void unboxInt32(const BaseIndex& src, Register dest); + void unboxBoolean(const ValueOperand& operand, Register dest); + void unboxBoolean(Register src, Register dest); + void unboxBoolean(const Address& src, Register dest); + void unboxBoolean(const BaseIndex& src, Register dest); + void unboxDouble(const ValueOperand& operand, FloatRegister dest); + void unboxDouble(Register src, Register dest); + void unboxDouble(const Address& src, FloatRegister dest); + void unboxDouble(const BaseIndex& src, FloatRegister dest); + void unboxString(const ValueOperand& operand, Register dest); + void unboxString(Register src, Register dest); + void unboxString(const Address& src, Register dest); + void unboxSymbol(const ValueOperand& src, Register dest); + void unboxSymbol(Register src, Register dest); + void unboxSymbol(const Address& src, Register dest); + void unboxBigInt(const ValueOperand& operand, Register dest); + void unboxBigInt(Register src, Register dest); + void unboxBigInt(const Address& src, Register dest); + void unboxObject(const ValueOperand& src, Register dest); + void unboxObject(Register src, Register dest); + void unboxObject(const Address& src, Register dest); + void unboxObject(const BaseIndex& src, Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_OBJECT); + } + void unboxValue(const ValueOperand& src, AnyRegister dest, JSValueType type); + + void notBoolean(const ValueOperand& val) { + xori(val.valueReg(), val.valueReg(), 1); + } + + // boxing code + void boxDouble(FloatRegister src, const ValueOperand& dest, FloatRegister); + void boxNonDouble(JSValueType type, Register src, const ValueOperand& dest); + + // Extended unboxing API. If the payload is already in a register, returns + // that register. Otherwise, provides a move to the given scratch register, + // and returns that. + [[nodiscard]] Register extractObject(const Address& address, + Register scratch); + [[nodiscard]] Register extractObject(const ValueOperand& value, + Register scratch) { + unboxObject(value, scratch); + return scratch; + } + [[nodiscard]] Register extractString(const ValueOperand& value, + Register scratch) { + unboxString(value, scratch); + return scratch; + } + [[nodiscard]] Register extractSymbol(const ValueOperand& value, + Register scratch) { + unboxSymbol(value, scratch); + return scratch; + } + [[nodiscard]] Register extractInt32(const ValueOperand& value, + Register scratch) { + unboxInt32(value, scratch); + return scratch; + } + [[nodiscard]] Register extractBoolean(const ValueOperand& value, + Register scratch) { + unboxBoolean(value, scratch); + return scratch; + } + [[nodiscard]] Register extractTag(const Address& address, Register scratch); + [[nodiscard]] Register extractTag(const BaseIndex& address, Register scratch); + [[nodiscard]] Register extractTag(const ValueOperand& value, + Register scratch) { + splitTag(value, scratch); + return scratch; + } + + void ensureDouble(const ValueOperand& source, FloatRegister dest, + Label* failure); + + void boolValueToDouble(const ValueOperand& operand, FloatRegister dest); + void int32ValueToDouble(const ValueOperand& operand, FloatRegister dest); + void loadInt32OrDouble(const Address& src, FloatRegister dest); + void loadInt32OrDouble(const BaseIndex& addr, FloatRegister dest); + void loadConstantDouble(double dp, FloatRegister dest); + + void boolValueToFloat32(const ValueOperand& operand, FloatRegister dest); + void int32ValueToFloat32(const ValueOperand& operand, FloatRegister dest); + void loadConstantFloat32(float f, FloatRegister dest); + + void testNullSet(Condition cond, const ValueOperand& value, Register dest); + + void testObjectSet(Condition cond, const ValueOperand& value, Register dest); + + void testUndefinedSet(Condition cond, const ValueOperand& value, + Register dest); + + // higher level tag testing code + Address ToPayload(Address value) { return value; } + + template <typename T> + void loadUnboxedValue(const T& address, MIRType type, AnyRegister dest) { + if (dest.isFloat()) { + loadInt32OrDouble(address, dest.fpu()); + } else { + unboxNonDouble(address, dest.gpr(), ValueTypeFromMIRType(type)); + } + } + + void storeUnboxedPayload(ValueOperand value, BaseIndex address, size_t nbytes, + JSValueType type) { + switch (nbytes) { + case 8: { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + Register scratch2 = temps.Acquire(); + if (type == JSVAL_TYPE_OBJECT) { + unboxObjectOrNull(value, scratch2); + } else { + unboxNonDouble(value, scratch2, type); + } + computeEffectiveAddress(address, scratch); + sd(scratch2, scratch, 0); + return; + } + case 4: + store32(value.valueReg(), address); + return; + case 1: + store8(value.valueReg(), address); + return; + default: + MOZ_CRASH("Bad payload width"); + } + } + + void storeUnboxedPayload(ValueOperand value, Address address, size_t nbytes, + JSValueType type) { + switch (nbytes) { + case 8: { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + if (type == JSVAL_TYPE_OBJECT) { + unboxObjectOrNull(value, scratch); + } else { + unboxNonDouble(value, scratch, type); + } + storePtr(scratch, address); + return; + } + case 4: + store32(value.valueReg(), address); + return; + case 1: + store8(value.valueReg(), address); + return; + default: + MOZ_CRASH("Bad payload width"); + } + } + + void boxValue(JSValueType type, Register src, Register dest) { + MOZ_ASSERT(src != dest); + + JSValueTag tag = (JSValueTag)JSVAL_TYPE_TO_TAG(type); + ma_li(dest, Imm32(tag)); + slli(dest, dest, JSVAL_TAG_SHIFT); + if (type == JSVAL_TYPE_INT32 || type == JSVAL_TYPE_BOOLEAN) { + InsertBits(dest, src, 0, 32); + } else { + InsertBits(dest, src, 0, JSVAL_TAG_SHIFT); + } + } + + void storeValue(ValueOperand val, const Address& dest); + void storeValue(ValueOperand val, const BaseIndex& dest); + void storeValue(JSValueType type, Register reg, Address dest); + void storeValue(JSValueType type, Register reg, BaseIndex dest); + void storeValue(const Value& val, Address dest); + void storeValue(const Value& val, BaseIndex dest); + void storeValue(const Address& src, const Address& dest, Register temp) { + loadPtr(src, temp); + storePtr(temp, dest); + } + + void storePrivateValue(Register src, const Address& dest) { + storePtr(src, dest); + } + void storePrivateValue(ImmGCPtr imm, const Address& dest) { + storePtr(imm, dest); + } + + void loadValue(Address src, ValueOperand val); + void loadValue(const BaseIndex& src, ValueOperand val); + + void loadUnalignedValue(const Address& src, ValueOperand dest) { + loadValue(src, dest); + } + + void tagValue(JSValueType type, Register payload, ValueOperand dest); + + void pushValue(ValueOperand val); + void popValue(ValueOperand val); + void pushValue(const Value& val) { + if (val.isGCThing()) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + writeDataRelocation(val); + movWithPatch(ImmWord(val.asRawBits()), scratch); + push(scratch); + } else { + push(ImmWord(val.asRawBits())); + } + } + void pushValue(JSValueType type, Register reg) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + boxValue(type, reg, scratch); + push(scratch); + } + void pushValue(const Address& addr); + void pushValue(const BaseIndex& addr, Register scratch) { + loadValue(addr, ValueOperand(scratch)); + pushValue(ValueOperand(scratch)); + } + + void handleFailureWithHandlerTail(Label* profilerExitTail, + Label* bailoutTail); + + ///////////////////////////////////////////////////////////////// + // Common interface. + ///////////////////////////////////////////////////////////////// + public: + // The following functions are exposed for use in platform-shared code. + + inline void incrementInt32Value(const Address& addr); + + void move32(Imm32 imm, Register dest); + void move32(Register src, Register dest); + + void movePtr(Register src, Register dest); + void movePtr(ImmWord imm, Register dest); + void movePtr(ImmPtr imm, Register dest); + void movePtr(wasm::SymbolicAddress imm, Register dest); + void movePtr(ImmGCPtr imm, Register dest); + + FaultingCodeOffset load8SignExtend(const Address& address, Register dest); + FaultingCodeOffset load8SignExtend(const BaseIndex& src, Register dest); + + FaultingCodeOffset load8ZeroExtend(const Address& address, Register dest); + FaultingCodeOffset load8ZeroExtend(const BaseIndex& src, Register dest); + + FaultingCodeOffset load16SignExtend(const Address& address, Register dest); + FaultingCodeOffset load16SignExtend(const BaseIndex& src, Register dest); + + template <typename S> + void load16UnalignedSignExtend(const S& src, Register dest) { + load16SignExtend(src, dest); + } + + FaultingCodeOffset load16ZeroExtend(const Address& address, Register dest); + FaultingCodeOffset load16ZeroExtend(const BaseIndex& src, Register dest); + + void SignExtendByte(Register rd, Register rs) { + slli(rd, rs, xlen - 8); + srai(rd, rd, xlen - 8); + } + + void SignExtendShort(Register rd, Register rs) { + slli(rd, rs, xlen - 16); + srai(rd, rd, xlen - 16); + } + + void SignExtendWord(Register rd, Register rs) { sext_w(rd, rs); } + void ZeroExtendWord(Register rd, Register rs) { + slli(rd, rs, 32); + srli(rd, rd, 32); + } + + template <typename S> + void load16UnalignedZeroExtend(const S& src, Register dest) { + load16ZeroExtend(src, dest); + } + + FaultingCodeOffset load32(const Address& address, Register dest); + FaultingCodeOffset load32(const BaseIndex& address, Register dest); + FaultingCodeOffset load32(AbsoluteAddress address, Register dest); + FaultingCodeOffset load32(wasm::SymbolicAddress address, Register dest); + + template <typename S> + void load32Unaligned(const S& src, Register dest) { + load32(src, dest); + } + + FaultingCodeOffset load64(const Address& address, Register64 dest) { + return loadPtr(address, dest.reg); + } + FaultingCodeOffset load64(const BaseIndex& address, Register64 dest) { + return loadPtr(address, dest.reg); + } + + FaultingCodeOffset loadDouble(const Address& addr, FloatRegister dest) { + return ma_loadDouble(dest, addr); + } + FaultingCodeOffset loadDouble(const BaseIndex& src, FloatRegister dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + computeScaledAddress(src, scratch); + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + fld(dest, scratch, 0); + return fco; + } + + FaultingCodeOffset loadFloat32(const Address& addr, FloatRegister dest) { + return ma_loadFloat(dest, addr); + } + + FaultingCodeOffset loadFloat32(const BaseIndex& src, FloatRegister dest) { + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + computeScaledAddress(src, scratch); + FaultingCodeOffset fco = FaultingCodeOffset(currentOffset()); + flw(dest, scratch, 0); + return fco; + } + + template <typename S> + FaultingCodeOffset load64Unaligned(const S& src, Register64 dest) { + return load64(src, dest); + } + + FaultingCodeOffset loadPtr(const Address& address, Register dest); + FaultingCodeOffset loadPtr(const BaseIndex& src, Register dest); + FaultingCodeOffset loadPtr(AbsoluteAddress address, Register dest); + FaultingCodeOffset loadPtr(wasm::SymbolicAddress address, Register dest); + + FaultingCodeOffset loadPrivate(const Address& address, Register dest); + + FaultingCodeOffset store8(Register src, const Address& address); + FaultingCodeOffset store8(Imm32 imm, const Address& address); + FaultingCodeOffset store8(Register src, const BaseIndex& address); + FaultingCodeOffset store8(Imm32 imm, const BaseIndex& address); + + FaultingCodeOffset store16(Register src, const Address& address); + FaultingCodeOffset store16(Imm32 imm, const Address& address); + FaultingCodeOffset store16(Register src, const BaseIndex& address); + FaultingCodeOffset store16(Imm32 imm, const BaseIndex& address); + + template <typename T> + FaultingCodeOffset store16Unaligned(Register src, const T& dest) { + return store16(src, dest); + } + + FaultingCodeOffset store32(Register src, AbsoluteAddress address); + FaultingCodeOffset store32(Register src, const Address& address); + FaultingCodeOffset store32(Register src, const BaseIndex& address); + FaultingCodeOffset store32(Imm32 src, const Address& address); + FaultingCodeOffset store32(Imm32 src, const BaseIndex& address); + + // NOTE: This will use second scratch on LOONG64. Only ARM needs the + // implementation without second scratch. + void store32_NoSecondScratch(Imm32 src, const Address& address) { + store32(src, address); + } + + template <typename T> + void store32Unaligned(Register src, const T& dest) { + store32(src, dest); + } + + FaultingCodeOffset store64(Imm64 imm, Address address) { + return storePtr(ImmWord(imm.value), address); + } + FaultingCodeOffset store64(Imm64 imm, const BaseIndex& address) { + return storePtr(ImmWord(imm.value), address); + } + + FaultingCodeOffset store64(Register64 src, Address address) { + return storePtr(src.reg, address); + } + FaultingCodeOffset store64(Register64 src, const BaseIndex& address) { + return storePtr(src.reg, address); + } + + template <typename T> + FaultingCodeOffset store64Unaligned(Register64 src, const T& dest) { + return store64(src, dest); + } + + template <typename T> + FaultingCodeOffset storePtr(ImmWord imm, T address); + template <typename T> + FaultingCodeOffset storePtr(ImmPtr imm, T address); + template <typename T> + FaultingCodeOffset storePtr(ImmGCPtr imm, T address); + FaultingCodeOffset storePtr(Register src, const Address& address); + FaultingCodeOffset storePtr(Register src, const BaseIndex& address); + FaultingCodeOffset storePtr(Register src, AbsoluteAddress dest); + + void moveDouble(FloatRegister src, FloatRegister dest) { fmv_d(dest, src); } + + void zeroDouble(FloatRegister reg) { fmv_d_x(reg, zero); } + + void convertUInt64ToDouble(Register src, FloatRegister dest); + + void breakpoint(uint32_t value = 0); + + void checkStackAlignment() { +#ifdef DEBUG + Label aligned; + UseScratchRegisterScope temps(this); + Register scratch = temps.Acquire(); + andi(scratch, sp, ABIStackAlignment - 1); + ma_b(scratch, zero, &aligned, Equal, ShortJump); + breakpoint(); + bind(&aligned); +#endif + }; + + static void calculateAlignedStackPointer(void** stackPointer); + + void cmpPtrSet(Assembler::Condition cond, Address lhs, ImmPtr rhs, + Register dest); + void cmpPtrSet(Assembler::Condition cond, Register lhs, Address rhs, + Register dest); + void cmpPtrSet(Assembler::Condition cond, Address lhs, Register rhs, + Register dest); + + void cmp32Set(Assembler::Condition cond, Register lhs, Address rhs, + Register dest); + + protected: + bool buildOOLFakeExitFrame(void* fakeReturnAddr); + + void wasmLoadI64Impl(const wasm::MemoryAccessDesc& access, + Register memoryBase, Register ptr, Register ptrScratch, + Register64 output, Register tmp); + void wasmStoreI64Impl(const wasm::MemoryAccessDesc& access, Register64 value, + Register memoryBase, Register ptr, Register ptrScratch, + Register tmp); + + public: + void abiret() { jr(ra); } + + void moveFloat32(FloatRegister src, FloatRegister dest) { fmv_s(dest, src); } + + // Instrumentation for entering and leaving the profiler. + void profilerEnterFrame(Register framePtr, Register scratch); + void profilerExitFrame(); +}; + +typedef MacroAssemblerRiscv64Compat MacroAssemblerSpecific; + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_MacroAssembler_riscv64_h */ diff --git a/js/src/jit/riscv64/MoveEmitter-riscv64.cpp b/js/src/jit/riscv64/MoveEmitter-riscv64.cpp new file mode 100644 index 0000000000..79f8d176b2 --- /dev/null +++ b/js/src/jit/riscv64/MoveEmitter-riscv64.cpp @@ -0,0 +1,333 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "jit/riscv64/MoveEmitter-riscv64.h" + +#include "jit/MacroAssembler-inl.h" + +using namespace js; +using namespace js::jit; + +void MoveEmitterRiscv64::breakCycle(const MoveOperand& from, + const MoveOperand& to, MoveOp::Type type, + uint32_t slotId) { + // There is some pattern: + // (A -> B) + // (B -> A) + // + // This case handles (A -> B), which we reach first. We save B, then allow + // the original move to continue. + switch (type) { + case MoveOp::FLOAT32: + if (to.isMemory()) { + ScratchFloat32Scope fpscratch32(masm); + masm.loadFloat32(getAdjustedAddress(to), fpscratch32); + masm.storeFloat32(fpscratch32, cycleSlot(slotId)); + } else { + masm.storeFloat32(to.floatReg(), cycleSlot(slotId)); + } + break; + case MoveOp::DOUBLE: + if (to.isMemory()) { + ScratchDoubleScope fpscratch64(masm); + masm.loadDouble(getAdjustedAddress(to), fpscratch64); + masm.storeDouble(fpscratch64, cycleSlot(slotId)); + } else { + masm.storeDouble(to.floatReg(), cycleSlot(slotId)); + } + break; + case MoveOp::INT32: + if (to.isMemory()) { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.load32(getAdjustedAddress(to), scratch2); + masm.store32(scratch2, cycleSlot(0)); + } else { + masm.store32(to.reg(), cycleSlot(0)); + } + break; + case MoveOp::GENERAL: + if (to.isMemory()) { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.loadPtr(getAdjustedAddress(to), scratch2); + masm.storePtr(scratch2, cycleSlot(0)); + } else { + masm.storePtr(to.reg(), cycleSlot(0)); + } + break; + default: + MOZ_CRASH("Unexpected move type"); + } +} + +void MoveEmitterRiscv64::completeCycle(const MoveOperand& from, + const MoveOperand& to, MoveOp::Type type, + uint32_t slotId) { + // There is some pattern: + // (A -> B) + // (B -> A) + // + // This case handles (B -> A), which we reach last. We emit a move from the + // saved value of B, to A. + switch (type) { + case MoveOp::FLOAT32: + if (to.isMemory()) { + ScratchFloat32Scope fpscratch32(masm); + masm.loadFloat32(cycleSlot(slotId), fpscratch32); + masm.storeFloat32(fpscratch32, getAdjustedAddress(to)); + } else { + masm.loadFloat32(cycleSlot(slotId), to.floatReg()); + } + break; + case MoveOp::DOUBLE: + if (to.isMemory()) { + ScratchDoubleScope fpscratch64(masm); + masm.loadDouble(cycleSlot(slotId), fpscratch64); + masm.storeDouble(fpscratch64, getAdjustedAddress(to)); + } else { + masm.loadDouble(cycleSlot(slotId), to.floatReg()); + } + break; + case MoveOp::INT32: + MOZ_ASSERT(slotId == 0); + if (to.isMemory()) { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.load32(cycleSlot(0), scratch2); + masm.store32(scratch2, getAdjustedAddress(to)); + } else { + masm.load32(cycleSlot(0), to.reg()); + } + break; + case MoveOp::GENERAL: + MOZ_ASSERT(slotId == 0); + if (to.isMemory()) { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.loadPtr(cycleSlot(0), scratch2); + masm.storePtr(scratch2, getAdjustedAddress(to)); + } else { + masm.loadPtr(cycleSlot(0), to.reg()); + } + break; + default: + MOZ_CRASH("Unexpected move type"); + } +} + +void MoveEmitterRiscv64::emit(const MoveResolver& moves) { + if (moves.numCycles()) { + // Reserve stack for cycle resolution + static_assert(SpillSlotSize == 8); + masm.reserveStack(moves.numCycles() * SpillSlotSize); + pushedAtCycle_ = masm.framePushed(); + } + + for (size_t i = 0; i < moves.numMoves(); i++) { + emit(moves.getMove(i)); + } +} + +void MoveEmitterRiscv64::emit(const MoveOp& move) { + const MoveOperand& from = move.from(); + const MoveOperand& to = move.to(); + + if (move.isCycleEnd() && move.isCycleBegin()) { + // A fun consequence of aliased registers is you can have multiple + // cycles at once, and one can end exactly where another begins. + breakCycle(from, to, move.endCycleType(), move.cycleBeginSlot()); + completeCycle(from, to, move.type(), move.cycleEndSlot()); + return; + } + + if (move.isCycleEnd()) { + MOZ_ASSERT(inCycle_); + completeCycle(from, to, move.type(), move.cycleEndSlot()); + MOZ_ASSERT(inCycle_ > 0); + inCycle_--; + return; + } + + if (move.isCycleBegin()) { + breakCycle(from, to, move.endCycleType(), move.cycleBeginSlot()); + inCycle_++; + } + + switch (move.type()) { + case MoveOp::FLOAT32: + emitFloat32Move(from, to); + break; + case MoveOp::DOUBLE: + emitDoubleMove(from, to); + break; + case MoveOp::INT32: + emitInt32Move(from, to); + break; + case MoveOp::GENERAL: + emitMove(from, to); + break; + default: + MOZ_CRASH("Unexpected move type"); + } +} + +void MoveEmitterRiscv64::emitMove(const MoveOperand& from, + const MoveOperand& to) { + if (from.isGeneralReg()) { + if (to.isGeneralReg()) { + masm.movePtr(from.reg(), to.reg()); + } else if (to.isMemory()) { + masm.storePtr(from.reg(), getAdjustedAddress(to)); + } else { + MOZ_CRASH("Invalid emitMove arguments."); + } + } else if (from.isMemory()) { + if (to.isGeneralReg()) { + masm.loadPtr(getAdjustedAddress(from), to.reg()); + } else if (to.isMemory()) { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.loadPtr(getAdjustedAddress(from), scratch2); + masm.storePtr(scratch2, getAdjustedAddress(to)); + } else { + MOZ_CRASH("Invalid emitMove arguments."); + } + } else if (from.isEffectiveAddress()) { + if (to.isGeneralReg()) { + masm.computeEffectiveAddress(getAdjustedAddress(from), to.reg()); + } else if (to.isMemory()) { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.computeEffectiveAddress(getAdjustedAddress(from), scratch2); + masm.storePtr(scratch2, getAdjustedAddress(to)); + } else { + MOZ_CRASH("Invalid emitMove arguments."); + } + } else { + MOZ_CRASH("Invalid emitMove arguments."); + } +} + +void MoveEmitterRiscv64::emitInt32Move(const MoveOperand& from, + const MoveOperand& to) { + if (from.isGeneralReg()) { + if (to.isGeneralReg()) { + masm.move32(from.reg(), to.reg()); + } else if (to.isMemory()) { + masm.store32(from.reg(), getAdjustedAddress(to)); + } else { + MOZ_CRASH("Invalid emitInt32Move arguments."); + } + } else if (from.isMemory()) { + if (to.isGeneralReg()) { + masm.load32(getAdjustedAddress(from), to.reg()); + } else if (to.isMemory()) { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.load32(getAdjustedAddress(from), scratch2); + masm.store32(scratch2, getAdjustedAddress(to)); + } else { + MOZ_CRASH("Invalid emitInt32Move arguments."); + } + } else if (from.isEffectiveAddress()) { + if (to.isGeneralReg()) { + masm.computeEffectiveAddress(getAdjustedAddress(from), to.reg()); + } else if (to.isMemory()) { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.computeEffectiveAddress(getAdjustedAddress(from), scratch2); + masm.store32(scratch2, getAdjustedAddress(to)); + } else { + MOZ_CRASH("Invalid emitInt32Move arguments."); + } + } else { + MOZ_CRASH("Invalid emitInt32Move arguments."); + } +} + +void MoveEmitterRiscv64::emitFloat32Move(const MoveOperand& from, + const MoveOperand& to) { + if (from.isFloatReg()) { + if (to.isFloatReg()) { + masm.fmv_s(to.floatReg(), from.floatReg()); + } else if (to.isGeneralReg()) { + // This should only be used when passing float parameter in a1,a2,a3 + MOZ_ASSERT(to.reg() == a1 || to.reg() == a2 || to.reg() == a3); + masm.fmv_x_w(to.reg(), from.floatReg()); + } else { + MOZ_ASSERT(to.isMemory()); + masm.storeFloat32(from.floatReg(), getAdjustedAddress(to)); + } + } else if (to.isFloatReg()) { + MOZ_ASSERT(from.isMemory()); + masm.loadFloat32(getAdjustedAddress(from), to.floatReg()); + } else if (to.isGeneralReg()) { + MOZ_ASSERT(from.isMemory()); + // This should only be used when passing float parameter in a1,a2,a3 + MOZ_ASSERT(to.reg() == a1 || to.reg() == a2 || to.reg() == a3); + masm.loadPtr(getAdjustedAddress(from), to.reg()); + } else { + MOZ_ASSERT(from.isMemory()); + MOZ_ASSERT(to.isMemory()); + ScratchFloat32Scope fpscratch32(masm); + masm.loadFloat32(getAdjustedAddress(from), fpscratch32); + masm.storeFloat32(fpscratch32, getAdjustedAddress(to)); + } +} + +void MoveEmitterRiscv64::emitDoubleMove(const MoveOperand& from, + const MoveOperand& to) { + if (from.isFloatReg()) { + if (to.isFloatReg()) { + masm.fmv_d(to.floatReg(), from.floatReg()); + } else if (to.isGeneralReg()) { + masm.fmv_x_d(to.reg(), from.floatReg()); + } else { + MOZ_ASSERT(to.isMemory()); + masm.storeDouble(from.floatReg(), getAdjustedAddress(to)); + } + } else if (to.isFloatReg()) { + if (from.isMemory()) { + masm.loadDouble(getAdjustedAddress(from), to.floatReg()); + } else { + masm.fmv_d_x(to.floatReg(), from.reg()); + } + } else { + MOZ_ASSERT(from.isMemory()); + MOZ_ASSERT(to.isMemory()); + ScratchDoubleScope fpscratch64(masm); + masm.loadDouble(getAdjustedAddress(from), fpscratch64); + masm.storeDouble(fpscratch64, getAdjustedAddress(to)); + } +} + +Address MoveEmitterRiscv64::cycleSlot(uint32_t slot, uint32_t subslot) const { + int32_t offset = masm.framePushed() - pushedAtCycle_; + return Address(StackPointer, offset + slot * sizeof(double) + subslot); +} + +int32_t MoveEmitterRiscv64::getAdjustedOffset(const MoveOperand& operand) { + MOZ_ASSERT(operand.isMemoryOrEffectiveAddress()); + if (operand.base() != StackPointer) { + return operand.disp(); + } + + // Adjust offset if stack pointer has been moved. + return operand.disp() + masm.framePushed() - pushedAtStart_; +} + +Address MoveEmitterRiscv64::getAdjustedAddress(const MoveOperand& operand) { + return Address(operand.base(), getAdjustedOffset(operand)); +} + +void MoveEmitterRiscv64::assertDone() { MOZ_ASSERT(inCycle_ == 0); } + +void MoveEmitterRiscv64::finish() { + assertDone(); + + masm.freeStack(masm.framePushed() - pushedAtStart_); +} diff --git a/js/src/jit/riscv64/MoveEmitter-riscv64.h b/js/src/jit/riscv64/MoveEmitter-riscv64.h new file mode 100644 index 0000000000..34d86b5794 --- /dev/null +++ b/js/src/jit/riscv64/MoveEmitter-riscv64.h @@ -0,0 +1,70 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_MoveEmitter_riscv64_h +#define jit_riscv64_MoveEmitter_riscv64_h + +#include "mozilla/Assertions.h" +#include "jit/MacroAssembler.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { + +class MacroAssemblerRiscv64; +class MoveResolver; +struct Register; + +class MoveEmitterRiscv64 { + uint32_t inCycle_; + MacroAssembler& masm; + + // Original stack push value. + uint32_t pushedAtStart_; + + // These store stack offsets to spill locations, snapshotting + // codegen->framePushed_ at the time they were allocated. They are -1 if no + // stack space has been allocated for that particular spill. + int32_t pushedAtCycle_; + + // These are registers that are available for temporary use. They may be + // assigned InvalidReg. If no corresponding spill space has been assigned, + // then these registers do not need to be spilled. + Register spilledReg_; + FloatRegister spilledFloatReg_; + + public: + explicit MoveEmitterRiscv64(MacroAssembler& m) + : inCycle_(0), + masm(m), + pushedAtStart_(masm.framePushed()), + pushedAtCycle_(-1), + spilledReg_(InvalidReg), + spilledFloatReg_(InvalidFloatReg) {} + void emit(const MoveResolver&); + void emit(const MoveOp& move); + void emitMove(const MoveOperand& from, const MoveOperand& to); + void emitInt32Move(const MoveOperand& from, const MoveOperand& to); + void emitFloat32Move(const MoveOperand& from, const MoveOperand& to); + void emitDoubleMove(const MoveOperand& from, const MoveOperand& to); + void finish(); + void assertDone(); + void setScratchRegister(Register) { MOZ_CRASH("Unimplement on riscv"); } + Address cycleSlot(uint32_t slot, uint32_t subslot = 0) const; + int32_t getAdjustedOffset(const MoveOperand& operand); + Address getAdjustedAddress(const MoveOperand& operand); + + void breakCycle(const MoveOperand& from, const MoveOperand& to, + MoveOp::Type type, uint32_t slotId); + void completeCycle(const MoveOperand& from, const MoveOperand& to, + MoveOp::Type type, uint32_t slot); +}; + +typedef MoveEmitterRiscv64 MoveEmitter; + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_MoveEmitter_riscv64_h */ diff --git a/js/src/jit/riscv64/Register-riscv64.h b/js/src/jit/riscv64/Register-riscv64.h new file mode 100644 index 0000000000..d8e8bf43af --- /dev/null +++ b/js/src/jit/riscv64/Register-riscv64.h @@ -0,0 +1,191 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_Register_riscv64_h +#define jit_riscv64_Register_riscv64_h + +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/Registers.h" +#include "jit/RegisterSets.h" + +namespace js { +namespace jit { + +static constexpr Register zero{Registers::zero}; +static constexpr Register ra{Registers::ra}; +static constexpr Register tp{Registers::tp}; +static constexpr Register sp{Registers::sp}; +static constexpr Register gp{Registers::gp}; +static constexpr Register a0{Registers::a0}; +static constexpr Register a1{Registers::a1}; +static constexpr Register a2{Registers::a2}; +static constexpr Register a3{Registers::a3}; +static constexpr Register a4{Registers::a4}; +static constexpr Register a5{Registers::a5}; +static constexpr Register a6{Registers::a6}; +static constexpr Register a7{Registers::a7}; +static constexpr Register t0{Registers::t0}; +static constexpr Register t1{Registers::t1}; +static constexpr Register t2{Registers::t2}; +static constexpr Register t3{Registers::t3}; +static constexpr Register t4{Registers::t4}; +static constexpr Register t5{Registers::t5}; +static constexpr Register t6{Registers::t6}; +static constexpr Register fp{Registers::fp}; +static constexpr Register s1{Registers::s1}; +static constexpr Register s2{Registers::s2}; +static constexpr Register s3{Registers::s3}; +static constexpr Register s4{Registers::s4}; +static constexpr Register s5{Registers::s5}; +static constexpr Register s6{Registers::s6}; +static constexpr Register s7{Registers::s7}; +static constexpr Register s8{Registers::s8}; +static constexpr Register s9{Registers::s9}; +static constexpr Register s10{Registers::s10}; +static constexpr Register s11{Registers::s11}; + +static constexpr FloatRegister ft0{FloatRegisters::f0}; +static constexpr FloatRegister ft1{FloatRegisters::f1}; +static constexpr FloatRegister ft2{FloatRegisters::f2}; +static constexpr FloatRegister ft3{FloatRegisters::f3}; +static constexpr FloatRegister ft4{FloatRegisters::f4}; +static constexpr FloatRegister ft5{FloatRegisters::f5}; +static constexpr FloatRegister ft6{FloatRegisters::f6}; +static constexpr FloatRegister ft7{FloatRegisters::f7}; +static constexpr FloatRegister fs0{FloatRegisters::f8}; +static constexpr FloatRegister fs1{FloatRegisters::f9}; +static constexpr FloatRegister fa0{FloatRegisters::f10}; +static constexpr FloatRegister fa1{FloatRegisters::f11}; +static constexpr FloatRegister fa2{FloatRegisters::f12}; +static constexpr FloatRegister fa3{FloatRegisters::f13}; +static constexpr FloatRegister fa4{FloatRegisters::f14}; +static constexpr FloatRegister fa5{FloatRegisters::f15}; +static constexpr FloatRegister fa6{FloatRegisters::f16}; +static constexpr FloatRegister fa7{FloatRegisters::f17}; +static constexpr FloatRegister fs2{FloatRegisters::f18}; +static constexpr FloatRegister fs3{FloatRegisters::f19}; +static constexpr FloatRegister fs4{FloatRegisters::f20}; +static constexpr FloatRegister fs5{FloatRegisters::f21}; +static constexpr FloatRegister fs6{FloatRegisters::f22}; +static constexpr FloatRegister fs7{FloatRegisters::f23}; +static constexpr FloatRegister fs8{FloatRegisters::f24}; +static constexpr FloatRegister fs9{FloatRegisters::f25}; +static constexpr FloatRegister fs10{FloatRegisters::f26}; +static constexpr FloatRegister fs11{FloatRegisters::f27}; +static constexpr FloatRegister ft8{FloatRegisters::f28}; +static constexpr FloatRegister ft9{FloatRegisters::f29}; +static constexpr FloatRegister ft10{FloatRegisters::f30}; +static constexpr FloatRegister ft11{FloatRegisters::f31}; + +static constexpr Register StackPointer{Registers::sp}; +static constexpr Register FramePointer{Registers::fp}; +static constexpr Register ReturnReg{Registers::a0}; +static constexpr Register ScratchRegister{Registers::s11}; +static constexpr Register64 ReturnReg64(ReturnReg); + +static constexpr FloatRegister ReturnFloat32Reg{FloatRegisters::fa0}; +static constexpr FloatRegister ReturnDoubleReg{FloatRegisters::fa0}; +#ifdef ENABLE_WASM_SIMD +static constexpr FloatRegister ReturnSimd128Reg{FloatRegisters::invalid_reg}; +static constexpr FloatRegister ScratchSimd128Reg{FloatRegisters::invalid_reg}; +#endif +static constexpr FloatRegister InvalidFloatReg{}; + +static constexpr FloatRegister ScratchFloat32Reg{FloatRegisters::ft10}; +static constexpr FloatRegister ScratchDoubleReg{FloatRegisters::ft10}; +static constexpr FloatRegister ScratchDoubleReg2{FloatRegisters::fs11}; + +static constexpr Register OsrFrameReg{Registers::a3}; +static constexpr Register PreBarrierReg{Registers::a1}; +static constexpr Register InterpreterPCReg{Registers::t0}; +static constexpr Register CallTempReg0{Registers::t0}; +static constexpr Register CallTempReg1{Registers::t1}; +static constexpr Register CallTempReg2{Registers::t2}; +static constexpr Register CallTempReg3{Registers::t3}; +static constexpr Register CallTempReg4{Registers::a6}; +static constexpr Register CallTempReg5{Registers::a7}; +static constexpr Register InvalidReg{Registers::invalid_reg}; +static constexpr Register CallTempNonArgRegs[] = {t0, t1, t2, t3}; +static const uint32_t NumCallTempNonArgRegs = std::size(CallTempNonArgRegs); + +static constexpr Register IntArgReg0{Registers::a0}; +static constexpr Register IntArgReg1{Registers::a1}; +static constexpr Register IntArgReg2{Registers::a2}; +static constexpr Register IntArgReg3{Registers::a3}; +static constexpr Register IntArgReg4{Registers::a4}; +static constexpr Register IntArgReg5{Registers::a5}; +static constexpr Register IntArgReg6{Registers::a6}; +static constexpr Register IntArgReg7{Registers::a7}; +static constexpr Register HeapReg{Registers::s7}; + +// Registers used by RegExpMatcher and RegExpExecMatch stubs (do not use +// JSReturnOperand). +static constexpr Register RegExpMatcherRegExpReg = CallTempReg0; +static constexpr Register RegExpMatcherStringReg = CallTempReg1; +static constexpr Register RegExpMatcherLastIndexReg = CallTempReg2; + +// Registers used by RegExpExecTest stub (do not use ReturnReg). +static constexpr Register RegExpExecTestRegExpReg = CallTempReg0; +static constexpr Register RegExpExecTestStringReg = CallTempReg1; + +// Registers used by RegExpSearcher stub (do not use ReturnReg). +static constexpr Register RegExpSearcherRegExpReg = CallTempReg0; +static constexpr Register RegExpSearcherStringReg = CallTempReg1; +static constexpr Register RegExpSearcherLastIndexReg = CallTempReg2; + +static constexpr Register JSReturnReg_Type{Registers::a3}; +static constexpr Register JSReturnReg_Data{Registers::s2}; +static constexpr Register JSReturnReg{Registers::a2}; +static constexpr ValueOperand JSReturnOperand = ValueOperand(JSReturnReg); + +// These registers may be volatile or nonvolatile. +static constexpr Register ABINonArgReg0{Registers::t0}; +static constexpr Register ABINonArgReg1{Registers::t1}; +static constexpr Register ABINonArgReg2{Registers::t2}; +static constexpr Register ABINonArgReg3{Registers::t3}; + +// These registers may be volatile or nonvolatile. +// Note: these three registers are all guaranteed to be different +static constexpr Register ABINonArgReturnReg0{Registers::t0}; +static constexpr Register ABINonArgReturnReg1{Registers::t1}; +static constexpr Register ABINonVolatileReg{Registers::s1}; + +// This register is guaranteed to be clobberable during the prologue and +// epilogue of an ABI call which must preserve both ABI argument, return +// and non-volatile registers. +static constexpr Register ABINonArgReturnVolatileReg{Registers::t0}; + +// This register may be volatile or nonvolatile. +// Avoid ft11 which is the scratch register. +static constexpr FloatRegister ABINonArgDoubleReg{FloatRegisters::ft11}; + +static constexpr Register WasmTableCallScratchReg0{ABINonArgReg0}; +static constexpr Register WasmTableCallScratchReg1{ABINonArgReg1}; +static constexpr Register WasmTableCallSigReg{ABINonArgReg2}; +static constexpr Register WasmTableCallIndexReg{ABINonArgReg3}; + +// Instance pointer argument register for WebAssembly functions. This must not +// alias any other register used for passing function arguments or return +// values. Preserved by WebAssembly functions. Must be nonvolatile. +static constexpr Register InstanceReg{Registers::s4}; + +static constexpr Register WasmJitEntryReturnScratch{Registers::t1}; + +static constexpr Register WasmCallRefCallScratchReg0{ABINonArgReg0}; +static constexpr Register WasmCallRefCallScratchReg1{ABINonArgReg1}; +static constexpr Register WasmCallRefReg{ABINonArgReg3}; + +static constexpr Register WasmTailCallInstanceScratchReg{ABINonArgReg1}; +static constexpr Register WasmTailCallRAScratchReg{ra}; +static constexpr Register WasmTailCallFPScratchReg{ABINonArgReg3}; + +} // namespace jit +} // namespace js + +#endif // jit_riscv64_Register_riscv64_h diff --git a/js/src/jit/riscv64/SharedICHelpers-riscv64-inl.h b/js/src/jit/riscv64/SharedICHelpers-riscv64-inl.h new file mode 100644 index 0000000000..ba70746197 --- /dev/null +++ b/js/src/jit/riscv64/SharedICHelpers-riscv64-inl.h @@ -0,0 +1,81 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_SharedICHelpers_riscv64_inl_h +#define jit_riscv64_SharedICHelpers_riscv64_inl_h + +#include "jit/SharedICHelpers.h" + +namespace js { +namespace jit { + +inline void EmitBaselineTailCallVM(TrampolinePtr target, MacroAssembler& masm, + uint32_t argSize) { +#ifdef DEBUG + Register scratch = R2.scratchReg(); + + // Compute frame size. + masm.movePtr(FramePointer, scratch); + masm.subPtr(StackPointer, scratch); + + // Store frame size without VMFunction arguments for debug assertions. + masm.subPtr(Imm32(argSize), scratch); + Address frameSizeAddr(FramePointer, + BaselineFrame::reverseOffsetOfDebugFrameSize()); + masm.store32(scratch, frameSizeAddr); + masm.addPtr(Imm32(argSize), scratch); +#endif + + // Push frame descriptor and perform the tail call. + masm.pushFrameDescriptor(FrameType::BaselineJS); + + MOZ_ASSERT(ICTailCallReg == ra); + // The return address will be pushed by the VM wrapper, for compatibility + // with direct calls. Refer to the top of generateVMWrapper(). + // ICTailCallReg (ra) already contains the return address (as we keep + // it there through the stub calls). + + masm.jump(target); +} + +inline void EmitBaselineCallVM(TrampolinePtr target, MacroAssembler& masm) { + masm.pushFrameDescriptor(FrameType::BaselineStub); + masm.call(target); +} + +inline void EmitBaselineEnterStubFrame(MacroAssembler& masm, Register scratch) { + MOZ_ASSERT(scratch != ICTailCallReg); + +#ifdef DEBUG + // Compute frame size. + masm.movePtr(FramePointer, scratch); + masm.subPtr(StackPointer, scratch); + + Address frameSizeAddr(FramePointer, + BaselineFrame::reverseOffsetOfDebugFrameSize()); + masm.store32(scratch, frameSizeAddr); +#endif + + // Note: when making changes here, don't forget to update + // BaselineStubFrame if needed. + + // Push frame descriptor and return address. + masm.PushFrameDescriptor(FrameType::BaselineJS); + masm.Push(ICTailCallReg); + + // Save old frame pointer, stack pointer and stub reg. + masm.Push(FramePointer); + masm.movePtr(StackPointer, FramePointer); + masm.Push(ICStubReg); + + // Stack should remain aligned. + masm.assertStackAlignment(sizeof(Value), 0); +} + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_SharedICHelpers_riscv64_inl_h */ diff --git a/js/src/jit/riscv64/SharedICHelpers-riscv64.h b/js/src/jit/riscv64/SharedICHelpers-riscv64.h new file mode 100644 index 0000000000..3411c6727e --- /dev/null +++ b/js/src/jit/riscv64/SharedICHelpers-riscv64.h @@ -0,0 +1,77 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_SharedICHelpers_riscv64_h +#define jit_riscv64_SharedICHelpers_riscv64_h +#include "jit/BaselineIC.h" +#include "jit/JitFrames.h" +#include "jit/MacroAssembler.h" +#include "jit/SharedICRegisters.h" +namespace js { +namespace jit { + +static const size_t ICStackValueOffset = 0; + +inline void EmitRestoreTailCallReg(MacroAssembler& masm) { + // No-op on RISC-V because ra register is always holding the return address. +} + +inline void EmitRepushTailCallReg(MacroAssembler& masm) { + // No-op on RISC-V because ra register is always holding the return address. +} +inline void EmitCallIC(MacroAssembler& masm, CodeOffset* callOffset) { + // The stub pointer must already be in ICStubReg. + // Load stubcode pointer from the ICStub. + // R2 won't be active when we call ICs, so we can use it as scratch. + masm.loadPtr(Address(ICStubReg, ICStub::offsetOfStubCode()), R2.scratchReg()); + + // Call the stubcode via a direct jump-and-link + masm.call(R2.scratchReg()); + *callOffset = CodeOffset(masm.currentOffset()); +} +inline void EmitReturnFromIC(MacroAssembler& masm) { masm.branch(ra); } +inline void EmitBaselineLeaveStubFrame(MacroAssembler& masm) { + masm.loadPtr( + Address(FramePointer, BaselineStubFrameLayout::ICStubOffsetFromFP), + ICStubReg); + + masm.movePtr(FramePointer, StackPointer); + masm.Pop(FramePointer); + + // Load the return address. + masm.Pop(ICTailCallReg); + + // Discard the frame descriptor. + { + UseScratchRegisterScope temps(&masm); + Register scratch2 = temps.Acquire(); + masm.Pop(scratch2); + } + + masm.checkStackAlignment(); +} + +inline void EmitStubGuardFailure(MacroAssembler& masm) { + // Load next stub into ICStubReg + masm.loadPtr(Address(ICStubReg, ICCacheIRStub::offsetOfNext()), ICStubReg); + + // Return address is already loaded, just jump to the next stubcode. + MOZ_ASSERT(ICTailCallReg == ra); + masm.jump(Address(ICStubReg, ICStub::offsetOfStubCode())); +} +template <typename AddrType> +inline void EmitPreBarrier(MacroAssembler& masm, const AddrType& addr, + MIRType type) { + // On RISC-V, $ra is clobbered by guardedCallPreBarrier. Save it first. + masm.push(ra); + masm.guardedCallPreBarrier(addr, type); + masm.pop(ra); +} + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_SharedICHelpers_riscv64_h */ diff --git a/js/src/jit/riscv64/SharedICRegisters-riscv64.h b/js/src/jit/riscv64/SharedICRegisters-riscv64.h new file mode 100644 index 0000000000..3dcefe51c7 --- /dev/null +++ b/js/src/jit/riscv64/SharedICRegisters-riscv64.h @@ -0,0 +1,38 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_SharedICRegisters_riscv64_h +#define jit_riscv64_SharedICRegisters_riscv64_h + +#include "jit/Registers.h" +#include "jit/RegisterSets.h" +#include "jit/riscv64/MacroAssembler-riscv64.h" + +namespace js { +namespace jit { + +// ValueOperands R0, R1, and R2. +// R0 == JSReturnReg, and R2 uses registers not preserved across calls. R1 value +// should be preserved across calls. +static constexpr ValueOperand R0(a2); +static constexpr ValueOperand R1(s1); +static constexpr ValueOperand R2(a0); + +// ICTailCallReg and ICStubReg +// These use registers that are not preserved across calls. +static constexpr Register ICTailCallReg = ra; +static constexpr Register ICStubReg = t0; + +// FloatReg0 must be equal to ReturnFloatReg. +static constexpr FloatRegister FloatReg0 = fa0; +static constexpr FloatRegister FloatReg1 = fa1; +static constexpr FloatRegister FloatReg2 = fa2; +static constexpr FloatRegister FloatReg3 = fa3; + +} // namespace jit +} // namespace js + +#endif /* jit_riscv64_SharedICRegisters_riscv64_h */ diff --git a/js/src/jit/riscv64/Simulator-riscv64.cpp b/js/src/jit/riscv64/Simulator-riscv64.cpp new file mode 100644 index 0000000000..aa0c350c7d --- /dev/null +++ b/js/src/jit/riscv64/Simulator-riscv64.cpp @@ -0,0 +1,4785 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: */ +// Copyright 2021 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +#ifdef JS_SIMULATOR_RISCV64 +# include "jit/riscv64/Simulator-riscv64.h" + +# include "mozilla/Casting.h" +# include "mozilla/FloatingPoint.h" +# include "mozilla/IntegerPrintfMacros.h" +# include "mozilla/Likely.h" +# include "mozilla/MathAlgorithms.h" + +# include <float.h> +# include <iostream> +# include <limits> + +# include "jit/AtomicOperations.h" +# include "jit/riscv64/Assembler-riscv64.h" +# include "js/Conversions.h" +# include "js/UniquePtr.h" +# include "js/Utility.h" +# include "threading/LockGuard.h" +# include "vm/JSContext.h" +# include "vm/Runtime.h" +# include "wasm/WasmInstance.h" +# include "wasm/WasmSignalHandlers.h" + +# define I8(v) static_cast<int8_t>(v) +# define I16(v) static_cast<int16_t>(v) +# define U16(v) static_cast<uint16_t>(v) +# define I32(v) static_cast<int32_t>(v) +# define U32(v) static_cast<uint32_t>(v) +# define I64(v) static_cast<int64_t>(v) +# define U64(v) static_cast<uint64_t>(v) +# define I128(v) static_cast<__int128_t>(v) +# define U128(v) static_cast<__uint128_t>(v) + +# define REGIx_FORMAT PRIx64 +# define REGId_FORMAT PRId64 + +# define I32_CHECK(v) \ + ({ \ + MOZ_ASSERT(I64(I32(v)) == I64(v)); \ + I32((v)); \ + }) + +namespace js { +namespace jit { + +bool Simulator::FLAG_trace_sim = false; +bool Simulator::FLAG_debug_sim = false; +bool Simulator::FLAG_riscv_trap_to_simulator_debugger = false; +bool Simulator::FLAG_riscv_print_watchpoint = false; + +static void UNIMPLEMENTED() { + printf("UNIMPLEMENTED instruction.\n"); + MOZ_CRASH(); +} +static void UNREACHABLE() { + printf("UNREACHABLE instruction.\n"); + MOZ_CRASH(); +} +# define UNSUPPORTED() \ + std::cout << "Unrecognized instruction [@pc=0x" << std::hex \ + << registers_[pc] << "]: 0x" << instr_.InstructionBits() \ + << std::endl; \ + printf("Unsupported instruction.\n"); \ + MOZ_CRASH(); + +static char* ReadLine(const char* prompt) { + UniqueChars result; + char lineBuf[256]; + int offset = 0; + bool keepGoing = true; + fprintf(stdout, "%s", prompt); + fflush(stdout); + while (keepGoing) { + if (fgets(lineBuf, sizeof(lineBuf), stdin) == nullptr) { + // fgets got an error. Just give up. + return nullptr; + } + int len = strlen(lineBuf); + if (len > 0 && lineBuf[len - 1] == '\n') { + // Since we read a new line we are done reading the line. This + // will exit the loop after copying this buffer into the result. + keepGoing = false; + } + if (!result) { + // Allocate the initial result and make room for the terminating '\0' + result.reset(js_pod_malloc<char>(len + 1)); + if (!result) { + return nullptr; + } + } else { + // Allocate a new result with enough room for the new addition. + int new_len = offset + len + 1; + char* new_result = js_pod_malloc<char>(new_len); + if (!new_result) { + return nullptr; + } + // Copy the existing input into the new array and set the new + // array as the result. + memcpy(new_result, result.get(), offset * sizeof(char)); + result.reset(new_result); + } + // Copy the newly read line into the result. + memcpy(result.get() + offset, lineBuf, len * sizeof(char)); + offset += len; + } + + MOZ_ASSERT(result); + result[offset] = '\0'; + return result.release(); +} + +// ----------------------------------------------------------------------------- +// Riscv assembly various constants. + +// C/C++ argument slots size. +const int kCArgSlotCount = 0; +const int kCArgsSlotsSize = kCArgSlotCount * sizeof(uintptr_t); +const int kBranchReturnOffset = 2 * kInstrSize; + +class CachePage { + public: + static const int LINE_VALID = 0; + static const int LINE_INVALID = 1; + + static const int kPageShift = 12; + static const int kPageSize = 1 << kPageShift; + static const int kPageMask = kPageSize - 1; + static const int kLineShift = 2; // The cache line is only 4 bytes right now. + static const int kLineLength = 1 << kLineShift; + static const int kLineMask = kLineLength - 1; + + CachePage() { memset(&validity_map_, LINE_INVALID, sizeof(validity_map_)); } + + char* validityByte(int offset) { + return &validity_map_[offset >> kLineShift]; + } + + char* cachedData(int offset) { return &data_[offset]; } + + private: + char data_[kPageSize]; // The cached data. + static const int kValidityMapSize = kPageSize >> kLineShift; + char validity_map_[kValidityMapSize]; // One byte per line. +}; + +// Protects the icache() and redirection() properties of the +// Simulator. +class AutoLockSimulatorCache : public LockGuard<Mutex> { + using Base = LockGuard<Mutex>; + + public: + explicit AutoLockSimulatorCache() + : Base(SimulatorProcess::singleton_->cacheLock_) {} +}; + +mozilla::Atomic<size_t, mozilla::ReleaseAcquire> + SimulatorProcess::ICacheCheckingDisableCount( + 1); // Checking is disabled by default. +SimulatorProcess* SimulatorProcess::singleton_ = nullptr; + +int64_t Simulator::StopSimAt = -1; + +static bool IsFlag(const char* found, const char* flag) { + return strlen(found) == strlen(flag) && strcmp(found, flag) == 0; +} + +Simulator* Simulator::Create() { + auto sim = MakeUnique<Simulator>(); + if (!sim) { + return nullptr; + } + + if (!sim->init()) { + return nullptr; + } + + int64_t stopAt; + char* stopAtStr = getenv("RISCV_SIM_STOP_AT"); + if (stopAtStr && sscanf(stopAtStr, "%" PRIi64, &stopAt) == 1) { + fprintf(stderr, "\nStopping simulation at icount %" PRIi64 "\n", stopAt); + Simulator::StopSimAt = stopAt; + } + char* str = getenv("RISCV_TRACE_SIM"); + if (str != nullptr && IsFlag(str, "true")) { + FLAG_trace_sim = true; + } + + return sim.release(); +} + +void Simulator::Destroy(Simulator* sim) { js_delete(sim); } + +# if JS_CODEGEN_RISCV64 +void Simulator::TraceRegWr(int64_t value, TraceType t) { + if (FLAG_trace_sim) { + union { + int64_t fmt_int64; + int32_t fmt_int32[2]; + float fmt_float[2]; + double fmt_double; + } v; + v.fmt_int64 = value; + + switch (t) { + case WORD: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int32:%" PRId32 + " uint32:%" PRIu32, + v.fmt_int64, icount_, v.fmt_int32[0], v.fmt_int32[0]); + break; + case DWORD: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int64:%" REGId_FORMAT + " uint64:%" PRIu64, + value, icount_, value, value); + break; + case FLOAT: + SNPrintF(trace_buf_, "%016" REGIx_FORMAT " (%" PRId64 ") flt:%e", + v.fmt_int64, icount_, v.fmt_float[0]); + break; + case DOUBLE: + SNPrintF(trace_buf_, "%016" REGIx_FORMAT " (%" PRId64 ") dbl:%e", + v.fmt_int64, icount_, v.fmt_double); + break; + default: + UNREACHABLE(); + } + } +} + +# elif JS_CODEGEN_RISCV32 +template <typename T> +void Simulator::TraceRegWr(T value, TraceType t) { + if (::v8::internal::FLAG_trace_sim) { + union { + int32_t fmt_int32; + float fmt_float; + double fmt_double; + } v; + if (t != DOUBLE) { + v.fmt_int32 = value; + } else { + DCHECK_EQ(sizeof(T), 8); + v.fmt_double = value; + } + switch (t) { + case WORD: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int32:%" REGId_FORMAT + " uint32:%" PRIu32, + v.fmt_int32, icount_, v.fmt_int32, v.fmt_int32); + break; + case FLOAT: + SNPrintF(trace_buf_, "%016" REGIx_FORMAT " (%" PRId64 ") flt:%e", + v.fmt_int32, icount_, v.fmt_float); + break; + case DOUBLE: + SNPrintF(trace_buf_, "%016" PRIx64 " (%" PRId64 ") dbl:%e", + static_cast<int64_t>(v.fmt_double), icount_, v.fmt_double); + break; + default: + UNREACHABLE(); + } + } +} +# endif +// The RiscvDebugger class is used by the simulator while debugging simulated +// code. +class RiscvDebugger { + public: + explicit RiscvDebugger(Simulator* sim) : sim_(sim) {} + + void Debug(); + // Print all registers with a nice formatting. + void PrintRegs(char name_prefix, int start_index, int end_index); + void printAllRegs(); + void printAllRegsIncludingFPU(); + + static const Instr kNopInstr = 0x0; + + private: + Simulator* sim_; + + int64_t GetRegisterValue(int regnum); + int64_t GetFPURegisterValue(int regnum); + float GetFPURegisterValueFloat(int regnum); + double GetFPURegisterValueDouble(int regnum); +# ifdef CAN_USE_RVV_INSTRUCTIONS + __int128_t GetVRegisterValue(int regnum); +# endif + bool GetValue(const char* desc, int64_t* value); +}; + +int64_t RiscvDebugger::GetRegisterValue(int regnum) { + if (regnum == Simulator::Register::kNumSimuRegisters) { + return sim_->get_pc(); + } else { + return sim_->getRegister(regnum); + } +} + +int64_t RiscvDebugger::GetFPURegisterValue(int regnum) { + if (regnum == Simulator::FPURegister::kNumFPURegisters) { + return sim_->get_pc(); + } else { + return sim_->getFpuRegister(regnum); + } +} + +float RiscvDebugger::GetFPURegisterValueFloat(int regnum) { + if (regnum == Simulator::FPURegister::kNumFPURegisters) { + return sim_->get_pc(); + } else { + return sim_->getFpuRegisterFloat(regnum); + } +} + +double RiscvDebugger::GetFPURegisterValueDouble(int regnum) { + if (regnum == Simulator::FPURegister::kNumFPURegisters) { + return sim_->get_pc(); + } else { + return sim_->getFpuRegisterDouble(regnum); + } +} + +# ifdef CAN_USE_RVV_INSTRUCTIONS +__int128_t RiscvDebugger::GetVRegisterValue(int regnum) { + if (regnum == kNumVRegisters) { + return sim_->get_pc(); + } else { + return sim_->get_vregister(regnum); + } +} +# endif + +bool RiscvDebugger::GetValue(const char* desc, int64_t* value) { + int regnum = Registers::FromName(desc); + int fpuregnum = FloatRegisters::FromName(desc); + + if (regnum != Registers::invalid_reg) { + *value = GetRegisterValue(regnum); + return true; + } else if (fpuregnum != FloatRegisters::invalid_reg) { + *value = GetFPURegisterValue(fpuregnum); + return true; + } else if (strncmp(desc, "0x", 2) == 0) { + return sscanf(desc + 2, "%" SCNx64, reinterpret_cast<int64_t*>(value)) == 1; + } else { + return sscanf(desc, "%" SCNu64, reinterpret_cast<int64_t*>(value)) == 1; + } +} + +# define REG_INFO(name) \ + name, GetRegisterValue(Registers::FromName(name)), \ + GetRegisterValue(Registers::FromName(name)) + +void RiscvDebugger::PrintRegs(char name_prefix, int start_index, + int end_index) { + EmbeddedVector<char, 10> name1, name2; + MOZ_ASSERT(name_prefix == 'a' || name_prefix == 't' || name_prefix == 's'); + MOZ_ASSERT(start_index >= 0 && end_index <= 99); + int num_registers = (end_index - start_index) + 1; + for (int i = 0; i < num_registers / 2; i++) { + SNPrintF(name1, "%c%d", name_prefix, start_index + 2 * i); + SNPrintF(name2, "%c%d", name_prefix, start_index + 2 * i + 1); + printf("%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT + " \t%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT " \n", + REG_INFO(name1.start()), REG_INFO(name2.start())); + } + if (num_registers % 2 == 1) { + SNPrintF(name1, "%c%d", name_prefix, end_index); + printf("%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT " \n", + REG_INFO(name1.start())); + } +} + +void RiscvDebugger::printAllRegs() { + printf("\n"); + // ra, sp, gp + printf("%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT + "\t%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT + "\t%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT "\n", + REG_INFO("ra"), REG_INFO("sp"), REG_INFO("gp")); + + // tp, fp, pc + printf("%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT + "\t%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT + "\t%3s: 0x%016" REGIx_FORMAT " %14" REGId_FORMAT "\n", + REG_INFO("tp"), REG_INFO("fp"), REG_INFO("pc")); + + // print register a0, .., a7 + PrintRegs('a', 0, 7); + // print registers s1, ..., s11 + PrintRegs('s', 1, 11); + // print registers t0, ..., t6 + PrintRegs('t', 0, 6); +} + +# undef REG_INFO + +void RiscvDebugger::printAllRegsIncludingFPU() { +# define FPU_REG_INFO(n) \ + FloatRegisters::GetName(n), GetFPURegisterValue(n), \ + GetFPURegisterValueDouble(n) + + printAllRegs(); + + printf("\n\n"); + // f0, f1, f2, ... f31. + MOZ_ASSERT(kNumFPURegisters % 2 == 0); + for (int i = 0; i < kNumFPURegisters; i += 2) + printf("%3s: 0x%016" PRIx64 " %16.4e \t%3s: 0x%016" PRIx64 " %16.4e\n", + FPU_REG_INFO(i), FPU_REG_INFO(i + 1)); +# undef FPU_REG_INFO +} + +void RiscvDebugger::Debug() { + intptr_t last_pc = -1; + bool done = false; + +# define COMMAND_SIZE 63 +# define ARG_SIZE 255 + +# define STR(a) #a +# define XSTR(a) STR(a) + + char cmd[COMMAND_SIZE + 1]; + char arg1[ARG_SIZE + 1]; + char arg2[ARG_SIZE + 1]; + char* argv[3] = {cmd, arg1, arg2}; + + // Make sure to have a proper terminating character if reaching the limit. + cmd[COMMAND_SIZE] = 0; + arg1[ARG_SIZE] = 0; + arg2[ARG_SIZE] = 0; + + while (!done && (sim_->get_pc() != Simulator::end_sim_pc)) { + if (last_pc != sim_->get_pc()) { + disasm::NameConverter converter; + disasm::Disassembler dasm(converter); + // Use a reasonably large buffer. + EmbeddedVector<char, 256> buffer; + dasm.InstructionDecode(buffer, reinterpret_cast<byte*>(sim_->get_pc())); + printf(" 0x%016" REGIx_FORMAT " %s\n", sim_->get_pc(), buffer.start()); + last_pc = sim_->get_pc(); + } + char* line = ReadLine("sim> "); + if (line == nullptr) { + break; + } else { + char* last_input = sim_->lastDebuggerInput(); + if (strcmp(line, "\n") == 0 && last_input != nullptr) { + line = last_input; + } else { + // Ownership is transferred to sim_; + sim_->setLastDebuggerInput(line); + } + // Use sscanf to parse the individual parts of the command line. At the + // moment no command expects more than two parameters. + int argc = sscanf( + line, + "%" XSTR(COMMAND_SIZE) "s " + "%" XSTR(ARG_SIZE) "s " + "%" XSTR(ARG_SIZE) "s", + cmd, arg1, arg2); + if ((strcmp(cmd, "si") == 0) || (strcmp(cmd, "stepi") == 0)) { + SimInstruction* instr = + reinterpret_cast<SimInstruction*>(sim_->get_pc()); + if (!(instr->IsTrap()) || + instr->InstructionBits() == rtCallRedirInstr) { + sim_->icount_++; + sim_->InstructionDecode( + reinterpret_cast<Instruction*>(sim_->get_pc())); + } else { + // Allow si to jump over generated breakpoints. + printf("/!\\ Jumping over generated breakpoint.\n"); + sim_->set_pc(sim_->get_pc() + kInstrSize); + } + } else if ((strcmp(cmd, "c") == 0) || (strcmp(cmd, "cont") == 0)) { + // Leave the debugger shell. + done = true; + } else if ((strcmp(cmd, "p") == 0) || (strcmp(cmd, "print") == 0)) { + if (argc == 2) { + int64_t value; + int64_t fvalue; + double dvalue; + if (strcmp(arg1, "all") == 0) { + printAllRegs(); + } else if (strcmp(arg1, "allf") == 0) { + printAllRegsIncludingFPU(); + } else { + int regnum = Registers::FromName(arg1); + int fpuregnum = FloatRegisters::FromName(arg1); +# ifdef CAN_USE_RVV_INSTRUCTIONS + int vregnum = VRegisters::FromName(arg1); +# endif + if (regnum != Registers::invalid_reg) { + value = GetRegisterValue(regnum); + printf("%s: 0x%08" REGIx_FORMAT " %" REGId_FORMAT " \n", arg1, + value, value); + } else if (fpuregnum != FloatRegisters::invalid_reg) { + fvalue = GetFPURegisterValue(fpuregnum); + dvalue = GetFPURegisterValueDouble(fpuregnum); + printf("%3s: 0x%016" PRIx64 " %16.4e\n", + FloatRegisters::GetName(fpuregnum), fvalue, dvalue); +# ifdef CAN_USE_RVV_INSTRUCTIONS + } else if (vregnum != kInvalidVRegister) { + __int128_t v = GetVRegisterValue(vregnum); + printf("\t%s:0x%016" REGIx_FORMAT "%016" REGIx_FORMAT "\n", + VRegisters::GetName(vregnum), (uint64_t)(v >> 64), + (uint64_t)v); +# endif + } else { + printf("%s unrecognized\n", arg1); + } + } + } else { + if (argc == 3) { + if (strcmp(arg2, "single") == 0) { + int64_t value; + float fvalue; + int fpuregnum = FloatRegisters::FromName(arg1); + + if (fpuregnum != FloatRegisters::invalid_reg) { + value = GetFPURegisterValue(fpuregnum); + value &= 0xFFFFFFFFUL; + fvalue = GetFPURegisterValueFloat(fpuregnum); + printf("%s: 0x%08" PRIx64 " %11.4e\n", arg1, value, fvalue); + } else { + printf("%s unrecognized\n", arg1); + } + } else { + printf("print <fpu register> single\n"); + } + } else { + printf("print <register> or print <fpu register> single\n"); + } + } + } else if ((strcmp(cmd, "po") == 0) || + (strcmp(cmd, "printobject") == 0)) { + UNIMPLEMENTED(); + } else if (strcmp(cmd, "stack") == 0 || strcmp(cmd, "mem") == 0) { + int64_t* cur = nullptr; + int64_t* end = nullptr; + int next_arg = 1; + if (argc < 2) { + printf("Need to specify <address> to memhex command\n"); + continue; + } + int64_t value; + if (!GetValue(arg1, &value)) { + printf("%s unrecognized\n", arg1); + continue; + } + cur = reinterpret_cast<int64_t*>(value); + next_arg++; + + int64_t words; + if (argc == next_arg) { + words = 10; + } else { + if (!GetValue(argv[next_arg], &words)) { + words = 10; + } + } + end = cur + words; + + while (cur < end) { + printf(" 0x%012" PRIxPTR " : 0x%016" REGIx_FORMAT + " %14" REGId_FORMAT " ", + reinterpret_cast<intptr_t>(cur), *cur, *cur); + printf("\n"); + cur++; + } + } else if ((strcmp(cmd, "watch") == 0)) { + if (argc < 2) { + printf("Need to specify <address> to mem command\n"); + continue; + } + int64_t value; + if (!GetValue(arg1, &value)) { + printf("%s unrecognized\n", arg1); + continue; + } + sim_->watch_address_ = reinterpret_cast<int64_t*>(value); + sim_->watch_value_ = *(sim_->watch_address_); + } else if ((strcmp(cmd, "disasm") == 0) || (strcmp(cmd, "dpc") == 0) || + (strcmp(cmd, "di") == 0)) { + disasm::NameConverter converter; + disasm::Disassembler dasm(converter); + // Use a reasonably large buffer. + EmbeddedVector<char, 256> buffer; + + byte* cur = nullptr; + byte* end = nullptr; + + if (argc == 1) { + cur = reinterpret_cast<byte*>(sim_->get_pc()); + end = cur + (10 * kInstrSize); + } else if (argc == 2) { + auto regnum = Registers::FromName(arg1); + if (regnum != Registers::invalid_reg || strncmp(arg1, "0x", 2) == 0) { + // The argument is an address or a register name. + sreg_t value; + if (GetValue(arg1, &value)) { + cur = reinterpret_cast<byte*>(value); + // Disassemble 10 instructions at <arg1>. + end = cur + (10 * kInstrSize); + } + } else { + // The argument is the number of instructions. + sreg_t value; + if (GetValue(arg1, &value)) { + cur = reinterpret_cast<byte*>(sim_->get_pc()); + // Disassemble <arg1> instructions. + end = cur + (value * kInstrSize); + } + } + } else { + sreg_t value1; + sreg_t value2; + if (GetValue(arg1, &value1) && GetValue(arg2, &value2)) { + cur = reinterpret_cast<byte*>(value1); + end = cur + (value2 * kInstrSize); + } + } + while (cur < end) { + dasm.InstructionDecode(buffer, cur); + printf(" 0x%08" PRIxPTR " %s\n", reinterpret_cast<intptr_t>(cur), + buffer.start()); + cur += kInstrSize; + } + } else if (strcmp(cmd, "trace") == 0) { + Simulator::FLAG_trace_sim = true; + Simulator::FLAG_riscv_print_watchpoint = true; + } else if (strcmp(cmd, "break") == 0 || strcmp(cmd, "b") == 0 || + strcmp(cmd, "tbreak") == 0) { + bool is_tbreak = strcmp(cmd, "tbreak") == 0; + if (argc == 2) { + int64_t value; + if (GetValue(arg1, &value)) { + sim_->SetBreakpoint(reinterpret_cast<SimInstruction*>(value), + is_tbreak); + } else { + printf("%s unrecognized\n", arg1); + } + } else { + sim_->ListBreakpoints(); + printf("Use `break <address>` to set or disable a breakpoint\n"); + printf( + "Use `tbreak <address>` to set or disable a temporary " + "breakpoint\n"); + } + } else if (strcmp(cmd, "flags") == 0) { + printf("No flags on RISC-V !\n"); + } else if (strcmp(cmd, "stop") == 0) { + int64_t value; + if (argc == 3) { + // Print information about all/the specified breakpoint(s). + if (strcmp(arg1, "info") == 0) { + if (strcmp(arg2, "all") == 0) { + printf("Stop information:\n"); + for (uint32_t i = kMaxWatchpointCode + 1; i <= kMaxStopCode; + i++) { + sim_->printStopInfo(i); + } + } else if (GetValue(arg2, &value)) { + sim_->printStopInfo(value); + } else { + printf("Unrecognized argument.\n"); + } + } else if (strcmp(arg1, "enable") == 0) { + // Enable all/the specified breakpoint(s). + if (strcmp(arg2, "all") == 0) { + for (uint32_t i = kMaxWatchpointCode + 1; i <= kMaxStopCode; + i++) { + sim_->enableStop(i); + } + } else if (GetValue(arg2, &value)) { + sim_->enableStop(value); + } else { + printf("Unrecognized argument.\n"); + } + } else if (strcmp(arg1, "disable") == 0) { + // Disable all/the specified breakpoint(s). + if (strcmp(arg2, "all") == 0) { + for (uint32_t i = kMaxWatchpointCode + 1; i <= kMaxStopCode; + i++) { + sim_->disableStop(i); + } + } else if (GetValue(arg2, &value)) { + sim_->disableStop(value); + } else { + printf("Unrecognized argument.\n"); + } + } + } else { + printf("Wrong usage. Use help command for more information.\n"); + } + } else if ((strcmp(cmd, "stat") == 0) || (strcmp(cmd, "st") == 0)) { + UNIMPLEMENTED(); + } else if ((strcmp(cmd, "h") == 0) || (strcmp(cmd, "help") == 0)) { + printf("cont (alias 'c')\n"); + printf(" Continue execution\n"); + printf("stepi (alias 'si')\n"); + printf(" Step one instruction\n"); + printf("print (alias 'p')\n"); + printf(" print <register>\n"); + printf(" Print register content\n"); + printf(" Use register name 'all' to print all GPRs\n"); + printf(" Use register name 'allf' to print all GPRs and FPRs\n"); + printf("printobject (alias 'po')\n"); + printf(" printobject <register>\n"); + printf(" Print an object from a register\n"); + printf("stack\n"); + printf(" stack [<words>]\n"); + printf(" Dump stack content, default dump 10 words)\n"); + printf("mem\n"); + printf(" mem <address> [<words>]\n"); + printf(" Dump memory content, default dump 10 words)\n"); + printf("watch\n"); + printf(" watch <address> \n"); + printf(" watch memory content.)\n"); + printf("flags\n"); + printf(" print flags\n"); + printf("disasm (alias 'di')\n"); + printf(" disasm [<instructions>]\n"); + printf(" disasm [<address/register>] (e.g., disasm pc) \n"); + printf(" disasm [[<address/register>] <instructions>]\n"); + printf(" Disassemble code, default is 10 instructions\n"); + printf(" from pc\n"); + printf("gdb \n"); + printf(" Return to gdb if the simulator was started with gdb\n"); + printf("break (alias 'b')\n"); + printf(" break : list all breakpoints\n"); + printf(" break <address> : set / enable / disable a breakpoint.\n"); + printf("tbreak\n"); + printf(" tbreak : list all breakpoints\n"); + printf( + " tbreak <address> : set / enable / disable a temporary " + "breakpoint.\n"); + printf(" Set a breakpoint enabled only for one stop. \n"); + printf("stop feature:\n"); + printf(" Description:\n"); + printf(" Stops are debug instructions inserted by\n"); + printf(" the Assembler::stop() function.\n"); + printf(" When hitting a stop, the Simulator will\n"); + printf(" stop and give control to the Debugger.\n"); + printf(" All stop codes are watched:\n"); + printf(" - They can be enabled / disabled: the Simulator\n"); + printf(" will / won't stop when hitting them.\n"); + printf(" - The Simulator keeps track of how many times they \n"); + printf(" are met. (See the info command.) Going over a\n"); + printf(" disabled stop still increases its counter. \n"); + printf(" Commands:\n"); + printf(" stop info all/<code> : print infos about number <code>\n"); + printf(" or all stop(s).\n"); + printf(" stop enable/disable all/<code> : enables / disables\n"); + printf(" all or number <code> stop(s)\n"); + } else { + printf("Unknown command: %s\n", cmd); + } + } + } + +# undef COMMAND_SIZE +# undef ARG_SIZE + +# undef STR +# undef XSTR +} + +void Simulator::SetBreakpoint(SimInstruction* location, bool is_tbreak) { + for (unsigned i = 0; i < breakpoints_.size(); i++) { + if (breakpoints_.at(i).location == location) { + if (breakpoints_.at(i).is_tbreak != is_tbreak) { + printf("Change breakpoint at %p to %s breakpoint\n", + reinterpret_cast<void*>(location), + is_tbreak ? "temporary" : "regular"); + breakpoints_.at(i).is_tbreak = is_tbreak; + return; + } + printf("Existing breakpoint at %p was %s\n", + reinterpret_cast<void*>(location), + breakpoints_.at(i).enabled ? "disabled" : "enabled"); + breakpoints_.at(i).enabled = !breakpoints_.at(i).enabled; + return; + } + } + Breakpoint new_breakpoint = {location, true, is_tbreak}; + breakpoints_.push_back(new_breakpoint); + printf("Set a %sbreakpoint at %p\n", is_tbreak ? "temporary " : "", + reinterpret_cast<void*>(location)); +} + +void Simulator::ListBreakpoints() { + printf("Breakpoints:\n"); + for (unsigned i = 0; i < breakpoints_.size(); i++) { + printf("%p : %s %s\n", + reinterpret_cast<void*>(breakpoints_.at(i).location), + breakpoints_.at(i).enabled ? "enabled" : "disabled", + breakpoints_.at(i).is_tbreak ? ": temporary" : ""); + } +} + +void Simulator::CheckBreakpoints() { + bool hit_a_breakpoint = false; + bool is_tbreak = false; + SimInstruction* pc_ = reinterpret_cast<SimInstruction*>(get_pc()); + for (unsigned i = 0; i < breakpoints_.size(); i++) { + if ((breakpoints_.at(i).location == pc_) && breakpoints_.at(i).enabled) { + hit_a_breakpoint = true; + if (breakpoints_.at(i).is_tbreak) { + // Disable a temporary breakpoint. + is_tbreak = true; + breakpoints_.at(i).enabled = false; + } + break; + } + } + if (hit_a_breakpoint) { + printf("Hit %sa breakpoint at %p.\n", is_tbreak ? "and disabled " : "", + reinterpret_cast<void*>(pc_)); + RiscvDebugger dbg(this); + dbg.Debug(); + } +} + +static bool AllOnOnePage(uintptr_t start, int size) { + intptr_t start_page = (start & ~CachePage::kPageMask); + intptr_t end_page = ((start + size) & ~CachePage::kPageMask); + return start_page == end_page; +} + +void Simulator::setLastDebuggerInput(char* input) { + js_free(lastDebuggerInput_); + lastDebuggerInput_ = input; +} + +static CachePage* GetCachePageLocked(SimulatorProcess::ICacheMap& i_cache, + void* page) { + SimulatorProcess::ICacheMap::AddPtr p = i_cache.lookupForAdd(page); + if (p) { + return p->value(); + } + AutoEnterOOMUnsafeRegion oomUnsafe; + CachePage* new_page = js_new<CachePage>(); + if (!new_page || !i_cache.add(p, page, new_page)) { + oomUnsafe.crash("Simulator CachePage"); + } + return new_page; +} + +// Flush from start up to and not including start + size. +static void FlushOnePageLocked(SimulatorProcess::ICacheMap& i_cache, + intptr_t start, int size) { + MOZ_ASSERT(size <= CachePage::kPageSize); + MOZ_ASSERT(AllOnOnePage(start, size - 1)); + MOZ_ASSERT((start & CachePage::kLineMask) == 0); + MOZ_ASSERT((size & CachePage::kLineMask) == 0); + void* page = reinterpret_cast<void*>(start & (~CachePage::kPageMask)); + int offset = (start & CachePage::kPageMask); + CachePage* cache_page = GetCachePageLocked(i_cache, page); + char* valid_bytemap = cache_page->validityByte(offset); + memset(valid_bytemap, CachePage::LINE_INVALID, size >> CachePage::kLineShift); +} + +static void FlushICacheLocked(SimulatorProcess::ICacheMap& i_cache, + void* start_addr, size_t size) { + intptr_t start = reinterpret_cast<intptr_t>(start_addr); + int intra_line = (start & CachePage::kLineMask); + start -= intra_line; + size += intra_line; + size = ((size - 1) | CachePage::kLineMask) + 1; + int offset = (start & CachePage::kPageMask); + while (!AllOnOnePage(start, size - 1)) { + int bytes_to_flush = CachePage::kPageSize - offset; + FlushOnePageLocked(i_cache, start, bytes_to_flush); + start += bytes_to_flush; + size -= bytes_to_flush; + MOZ_ASSERT((start & CachePage::kPageMask) == 0); + offset = 0; + } + if (size != 0) { + FlushOnePageLocked(i_cache, start, size); + } +} + +/* static */ +void SimulatorProcess::checkICacheLocked(SimInstruction* instr) { + intptr_t address = reinterpret_cast<intptr_t>(instr); + void* page = reinterpret_cast<void*>(address & (~CachePage::kPageMask)); + void* line = reinterpret_cast<void*>(address & (~CachePage::kLineMask)); + int offset = (address & CachePage::kPageMask); + CachePage* cache_page = GetCachePageLocked(icache(), page); + char* cache_valid_byte = cache_page->validityByte(offset); + bool cache_hit = (*cache_valid_byte == CachePage::LINE_VALID); + char* cached_line = cache_page->cachedData(offset & ~CachePage::kLineMask); + + if (cache_hit) { + // Check that the data in memory matches the contents of the I-cache. + int cmpret = memcmp(reinterpret_cast<void*>(instr), + cache_page->cachedData(offset), kInstrSize); + MOZ_ASSERT(cmpret == 0); + } else { + // Cache miss. Load memory into the cache. + memcpy(cached_line, line, CachePage::kLineLength); + *cache_valid_byte = CachePage::LINE_VALID; + } +} + +HashNumber SimulatorProcess::ICacheHasher::hash(const Lookup& l) { + return U32(reinterpret_cast<uintptr_t>(l)) >> 2; +} + +bool SimulatorProcess::ICacheHasher::match(const Key& k, const Lookup& l) { + MOZ_ASSERT((reinterpret_cast<intptr_t>(k) & CachePage::kPageMask) == 0); + MOZ_ASSERT((reinterpret_cast<intptr_t>(l) & CachePage::kPageMask) == 0); + return k == l; +} + +/* static */ +void SimulatorProcess::FlushICache(void* start_addr, size_t size) { + if (!ICacheCheckingDisableCount) { + AutoLockSimulatorCache als; + js::jit::FlushICacheLocked(icache(), start_addr, size); + } +} + +Simulator::Simulator() { + // Set up simulator support first. Some of this information is needed to + // setup the architecture state. + + // Note, allocation and anything that depends on allocated memory is + // deferred until init(), in order to handle OOM properly. + + stack_ = nullptr; + stackLimit_ = 0; + pc_modified_ = false; + icount_ = 0; + break_count_ = 0; + break_pc_ = nullptr; + break_instr_ = 0; + single_stepping_ = false; + single_step_callback_ = nullptr; + single_step_callback_arg_ = nullptr; + + // Set up architecture state. + // All registers are initialized to zero to start with. + for (int i = 0; i < Simulator::Register::kNumSimuRegisters; i++) { + registers_[i] = 0; + } + for (int i = 0; i < Simulator::FPURegister::kNumFPURegisters; i++) { + FPUregisters_[i] = 0; + } + FCSR_ = 0; + LLBit_ = false; + LLAddr_ = 0; + lastLLValue_ = 0; + + // The ra and pc are initialized to a known bad value that will cause an + // access violation if the simulator ever tries to execute it. + registers_[pc] = bad_ra; + registers_[ra] = bad_ra; + + for (int i = 0; i < kNumExceptions; i++) { + exceptions[i] = 0; + } + + lastDebuggerInput_ = nullptr; +} + +bool Simulator::init() { + // Allocate 2MB for the stack. Note that we will only use 1MB, see below. + static const size_t stackSize = 2 * 1024 * 1024; + stack_ = js_pod_malloc<char>(stackSize); + if (!stack_) { + return false; + } + + // Leave a safety margin of 1MB to prevent overrunning the stack when + // pushing values (total stack size is 2MB). + stackLimit_ = reinterpret_cast<uintptr_t>(stack_) + 1024 * 1024; + + // The sp is initialized to point to the bottom (high address) of the + // allocated stack area. To be safe in potential stack underflows we leave + // some buffer below. + registers_[sp] = reinterpret_cast<int64_t>(stack_) + stackSize - 64; + + return true; +} + +// When the generated code calls an external reference we need to catch that in +// the simulator. The external reference will be a function compiled for the +// host architecture. We need to call that function instead of trying to +// execute it with the simulator. We do that by redirecting the external +// reference to a swi (software-interrupt) instruction that is handled by +// the simulator. We write the original destination of the jump just at a known +// offset from the swi instruction so the simulator knows what to call. +class Redirection { + friend class SimulatorProcess; + + // sim's lock must already be held. + Redirection(void* nativeFunction, ABIFunctionType type) + : nativeFunction_(nativeFunction), + swiInstruction_(rtCallRedirInstr), + type_(type), + next_(nullptr) { + next_ = SimulatorProcess::redirection(); + if (!SimulatorProcess::ICacheCheckingDisableCount) { + FlushICacheLocked(SimulatorProcess::icache(), addressOfSwiInstruction(), + kInstrSize); + } + SimulatorProcess::setRedirection(this); + } + + public: + void* addressOfSwiInstruction() { return &swiInstruction_; } + void* nativeFunction() const { return nativeFunction_; } + ABIFunctionType type() const { return type_; } + + static Redirection* Get(void* nativeFunction, ABIFunctionType type) { + AutoLockSimulatorCache als; + + Redirection* current = SimulatorProcess::redirection(); + for (; current != nullptr; current = current->next_) { + if (current->nativeFunction_ == nativeFunction) { + MOZ_ASSERT(current->type() == type); + return current; + } + } + + // Note: we can't use js_new here because the constructor is private. + AutoEnterOOMUnsafeRegion oomUnsafe; + Redirection* redir = js_pod_malloc<Redirection>(1); + if (!redir) { + oomUnsafe.crash("Simulator redirection"); + } + new (redir) Redirection(nativeFunction, type); + return redir; + } + + static Redirection* FromSwiInstruction(Instruction* swiInstruction) { + uint8_t* addrOfSwi = reinterpret_cast<uint8_t*>(swiInstruction); + uint8_t* addrOfRedirection = + addrOfSwi - offsetof(Redirection, swiInstruction_); + return reinterpret_cast<Redirection*>(addrOfRedirection); + } + + private: + void* nativeFunction_; + uint32_t swiInstruction_; + ABIFunctionType type_; + Redirection* next_; +}; + +Simulator::~Simulator() { js_free(stack_); } + +SimulatorProcess::SimulatorProcess() + : cacheLock_(mutexid::SimulatorCacheLock), redirection_(nullptr) { + if (getenv("MIPS_SIM_ICACHE_CHECKS")) { + ICacheCheckingDisableCount = 0; + } +} + +SimulatorProcess::~SimulatorProcess() { + Redirection* r = redirection_; + while (r) { + Redirection* next = r->next_; + js_delete(r); + r = next; + } +} + +/* static */ +void* Simulator::RedirectNativeFunction(void* nativeFunction, + ABIFunctionType type) { + Redirection* redirection = Redirection::Get(nativeFunction, type); + return redirection->addressOfSwiInstruction(); +} + +// Get the active Simulator for the current thread. +Simulator* Simulator::Current() { + JSContext* cx = TlsContext.get(); + MOZ_ASSERT(CurrentThreadCanAccessRuntime(cx->runtime())); + return cx->simulator(); +} + +// Sets the register in the architecture state. It will also deal with updating +// Simulator internal state for special registers such as PC. +void Simulator::setRegister(int reg, int64_t value) { + MOZ_ASSERT((reg >= 0) && (reg < Simulator::Register::kNumSimuRegisters)); + if (reg == pc) { + pc_modified_ = true; + } + + // Zero register always holds 0. + registers_[reg] = (reg == 0) ? 0 : value; +} + +void Simulator::setFpuRegister(int fpureg, int64_t value) { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + FPUregisters_[fpureg] = value; +} + +void Simulator::setFpuRegisterLo(int fpureg, int32_t value) { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + *mozilla::BitwiseCast<int32_t*>(&FPUregisters_[fpureg]) = value; +} + +void Simulator::setFpuRegisterHi(int fpureg, int32_t value) { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + *((mozilla::BitwiseCast<int32_t*>(&FPUregisters_[fpureg])) + 1) = value; +} + +void Simulator::setFpuRegisterFloat(int fpureg, float value) { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + *mozilla::BitwiseCast<int64_t*>(&FPUregisters_[fpureg]) = box_float(value); +} + +void Simulator::setFpuRegisterFloat(int fpureg, Float32 value) { + MOZ_ASSERT((fpureg >= 0) && (fpureg < kNumFPURegisters)); + Float64 t = Float64::FromBits(box_float(value.get_bits())); + memcpy(&FPUregisters_[fpureg], &t, 8); +} + +void Simulator::setFpuRegisterDouble(int fpureg, double value) { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + *mozilla::BitwiseCast<double*>(&FPUregisters_[fpureg]) = value; +} + +void Simulator::setFpuRegisterDouble(int fpureg, Float64 value) { + MOZ_ASSERT((fpureg >= 0) && (fpureg < kNumFPURegisters)); + memcpy(&FPUregisters_[fpureg], &value, 8); +} + +// Get the register from the architecture state. This function does handle +// the special case of accessing the PC register. +int64_t Simulator::getRegister(int reg) const { + MOZ_ASSERT((reg >= 0) && (reg < Simulator::Register::kNumSimuRegisters)); + if (reg == 0) { + return 0; + } + return registers_[reg] + ((reg == pc) ? SimInstruction::kPCReadOffset : 0); +} + +int64_t Simulator::getFpuRegister(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + return FPUregisters_[fpureg]; +} + +int32_t Simulator::getFpuRegisterLo(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + return *mozilla::BitwiseCast<int32_t*>(&FPUregisters_[fpureg]); +} + +int32_t Simulator::getFpuRegisterHi(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + return *((mozilla::BitwiseCast<int32_t*>(&FPUregisters_[fpureg])) + 1); +} + +float Simulator::getFpuRegisterFloat(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + return *mozilla::BitwiseCast<float*>(&FPUregisters_[fpureg]); +} + +Float32 Simulator::getFpuRegisterFloat32(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && (fpureg < kNumFPURegisters)); + if (!is_boxed_float(FPUregisters_[fpureg])) { + return Float32::FromBits(0x7ffc0000); + } + return Float32::FromBits( + *bit_cast<uint32_t*>(const_cast<int64_t*>(&FPUregisters_[fpureg]))); +} + +double Simulator::getFpuRegisterDouble(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + return *mozilla::BitwiseCast<double*>(&FPUregisters_[fpureg]); +} + +Float64 Simulator::getFpuRegisterFloat64(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && (fpureg < kNumFPURegisters)); + return Float64::FromBits(FPUregisters_[fpureg]); +} + +void Simulator::setCallResultDouble(double result) { + setFpuRegisterDouble(fa0, result); +} + +void Simulator::setCallResultFloat(float result) { + setFpuRegisterFloat(fa0, result); +} + +void Simulator::setCallResult(int64_t res) { setRegister(a0, res); } + +void Simulator::setCallResult(__int128_t res) { + setRegister(a0, I64(res)); + setRegister(a1, I64(res >> 64)); +} + +// Raw access to the PC register. +void Simulator::set_pc(int64_t value) { + pc_modified_ = true; + registers_[pc] = value; +} + +bool Simulator::has_bad_pc() const { + return ((registers_[pc] == bad_ra) || (registers_[pc] == end_sim_pc)); +} + +// Raw access to the PC register without the special adjustment when reading. +int64_t Simulator::get_pc() const { return registers_[pc]; } + +JS::ProfilingFrameIterator::RegisterState Simulator::registerState() { + wasm::RegisterState state; + state.pc = (void*)get_pc(); + state.fp = (void*)getRegister(fp); + state.sp = (void*)getRegister(sp); + state.lr = (void*)getRegister(ra); + return state; +} + +void Simulator::HandleWasmTrap() { + uint8_t* newPC; + if (wasm::HandleIllegalInstruction(registerState(), &newPC)) { + set_pc(int64_t(newPC)); + return; + } +} + +// TODO(plind): consider making icount_ printing a flag option. +template <typename T> +void Simulator::TraceMemRd(sreg_t addr, T value, sreg_t reg_value) { + if (FLAG_trace_sim) { + if (std::is_integral<T>::value) { + switch (sizeof(T)) { + case 1: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int8:%" PRId8 + " uint8:%" PRIu8 " <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<int8_t>(value), + static_cast<uint8_t>(value), addr); + break; + case 2: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int16:%" PRId16 + " uint16:%" PRIu16 " <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<int16_t>(value), + static_cast<uint16_t>(value), addr); + break; + case 4: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int32:%" PRId32 + " uint32:%" PRIu32 " <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<int32_t>(value), + static_cast<uint32_t>(value), addr); + break; + case 8: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int64:%" PRId64 + " uint64:%" PRIu64 " <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<int64_t>(value), + static_cast<uint64_t>(value), addr); + break; + default: + UNREACHABLE(); + } + } else if (std::is_same<float, T>::value) { + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 + ") flt:%e <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<float>(value), addr); + } else if (std::is_same<double, T>::value) { + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 + ") dbl:%e <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<double>(value), addr); + } else { + UNREACHABLE(); + } + } +} + +void Simulator::TraceMemRdFloat(sreg_t addr, Float32 value, int64_t reg_value) { + if (FLAG_trace_sim) { + SNPrintF(trace_buf_, + "%016" PRIx64 " (%" PRId64 + ") flt:%e <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<float>(value.get_scalar()), addr); + } +} + +void Simulator::TraceMemRdDouble(sreg_t addr, double value, int64_t reg_value) { + if (FLAG_trace_sim) { + SNPrintF(trace_buf_, + "%016" PRIx64 " (%" PRId64 + ") dbl:%e <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<double>(value), addr); + } +} + +void Simulator::TraceMemRdDouble(sreg_t addr, Float64 value, + int64_t reg_value) { + if (FLAG_trace_sim) { + SNPrintF(trace_buf_, + "%016" PRIx64 " (%" PRId64 + ") dbl:%e <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<double>(value.get_scalar()), addr); + } +} + +template <typename T> +void Simulator::TraceMemWr(sreg_t addr, T value) { + if (FLAG_trace_sim) { + switch (sizeof(T)) { + case 1: + SNPrintF(trace_buf_, + " (%" PRIu64 ") int8:%" PRId8 + " uint8:%" PRIu8 " --> [addr: %" REGIx_FORMAT "]", + icount_, static_cast<int8_t>(value), + static_cast<uint8_t>(value), addr); + break; + case 2: + SNPrintF(trace_buf_, + " (%" PRIu64 ") int16:%" PRId16 + " uint16:%" PRIu16 " --> [addr: %" REGIx_FORMAT "]", + icount_, static_cast<int16_t>(value), + static_cast<uint16_t>(value), addr); + break; + case 4: + if (std::is_integral<T>::value) { + SNPrintF(trace_buf_, + " (%" PRIu64 ") int32:%" PRId32 + " uint32:%" PRIu32 " --> [addr: %" REGIx_FORMAT "]", + icount_, static_cast<int32_t>(value), + static_cast<uint32_t>(value), addr); + } else { + SNPrintF(trace_buf_, + " (%" PRIu64 + ") flt:%e --> [addr: %" REGIx_FORMAT "]", + icount_, static_cast<float>(value), addr); + } + break; + case 8: + if (std::is_integral<T>::value) { + SNPrintF(trace_buf_, + " (%" PRIu64 ") int64:%" PRId64 + " uint64:%" PRIu64 " --> [addr: %" REGIx_FORMAT "]", + icount_, static_cast<int64_t>(value), + static_cast<uint64_t>(value), addr); + } else { + SNPrintF(trace_buf_, + " (%" PRIu64 + ") dbl:%e --> [addr: %" REGIx_FORMAT "]", + icount_, static_cast<double>(value), addr); + } + break; + default: + UNREACHABLE(); + } + } +} + +void Simulator::TraceMemWrDouble(sreg_t addr, double value) { + if (FLAG_trace_sim) { + SNPrintF(trace_buf_, + " (%" PRIu64 + ") dbl:%e --> [addr: %" REGIx_FORMAT "]", + icount_, value, addr); + } +} + +template <typename T> +void Simulator::TraceLr(sreg_t addr, T value, sreg_t reg_value) { + if (FLAG_trace_sim) { + if (std::is_integral<T>::value) { + switch (sizeof(T)) { + case 4: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int32:%" PRId32 + " uint32:%" PRIu32 " <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<int32_t>(value), + static_cast<uint32_t>(value), addr); + break; + case 8: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRId64 ") int64:%" PRId64 + " uint64:%" PRIu64 " <-- [addr: %" REGIx_FORMAT "]", + reg_value, icount_, static_cast<int64_t>(value), + static_cast<uint64_t>(value), addr); + break; + default: + UNREACHABLE(); + } + } else { + UNREACHABLE(); + } + } +} + +template <typename T> +void Simulator::TraceSc(sreg_t addr, T value) { + if (FLAG_trace_sim) { + switch (sizeof(T)) { + case 4: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRIu64 ") int32:%" PRId32 + " uint32:%" PRIu32 " --> [addr: %" REGIx_FORMAT "]", + getRegister(rd_reg()), icount_, static_cast<int32_t>(value), + static_cast<uint32_t>(value), addr); + break; + case 8: + SNPrintF(trace_buf_, + "%016" REGIx_FORMAT " (%" PRIu64 ") int64:%" PRId64 + " uint64:%" PRIu64 " --> [addr: %" REGIx_FORMAT "]", + getRegister(rd_reg()), icount_, static_cast<int64_t>(value), + static_cast<uint64_t>(value), addr); + break; + default: + UNREACHABLE(); + } + } +} + +// TODO(RISCV): check whether the specific board supports unaligned load/store +// (determined by EEI). For now, we assume the board does not support unaligned +// load/store (e.g., trapping) +template <typename T> +T Simulator::ReadMem(sreg_t addr, Instruction* instr) { + if (handleWasmSegFault(addr, sizeof(T))) { + return -1; + } + if (addr >= 0 && addr < 0x400) { + // This has to be a nullptr-dereference, drop into debugger. + printf("Memory read from bad address: 0x%08" REGIx_FORMAT + " , pc=0x%08" PRIxPTR " \n", + addr, reinterpret_cast<intptr_t>(instr)); + DieOrDebug(); + } + T* ptr = reinterpret_cast<T*>(addr); + T value = *ptr; + return value; +} + +template <typename T> +void Simulator::WriteMem(sreg_t addr, T value, Instruction* instr) { + if (handleWasmSegFault(addr, sizeof(T))) { + value = -1; + return; + } + if (addr >= 0 && addr < 0x400) { + // This has to be a nullptr-dereference, drop into debugger. + printf("Memory write to bad address: 0x%08" REGIx_FORMAT + " , pc=0x%08" PRIxPTR " \n", + addr, reinterpret_cast<intptr_t>(instr)); + DieOrDebug(); + } + T* ptr = reinterpret_cast<T*>(addr); + if (!std::is_same<double, T>::value) { + TraceMemWr(addr, value); + } else { + TraceMemWrDouble(addr, value); + } + *ptr = value; +} + +template <> +void Simulator::WriteMem(sreg_t addr, Float32 value, Instruction* instr) { + if (handleWasmSegFault(addr, 4)) { + value = Float32(-1.0f); + return; + } + if (addr >= 0 && addr < 0x400) { + // This has to be a nullptr-dereference, drop into debugger. + printf("Memory write to bad address: 0x%08" REGIx_FORMAT + " , pc=0x%08" PRIxPTR " \n", + addr, reinterpret_cast<intptr_t>(instr)); + DieOrDebug(); + } + float* ptr = reinterpret_cast<float*>(addr); + TraceMemWr(addr, value.get_scalar()); + memcpy(ptr, &value, 4); +} + +template <> +void Simulator::WriteMem(sreg_t addr, Float64 value, Instruction* instr) { + if (handleWasmSegFault(addr, 8)) { + value = Float64(-1.0); + return; + } + if (addr >= 0 && addr < 0x400) { + // This has to be a nullptr-dereference, drop into debugger. + printf("Memory write to bad address: 0x%08" REGIx_FORMAT + " , pc=0x%08" PRIxPTR " \n", + addr, reinterpret_cast<intptr_t>(instr)); + DieOrDebug(); + } + double* ptr = reinterpret_cast<double*>(addr); + TraceMemWrDouble(addr, value.get_scalar()); + memcpy(ptr, &value, 8); +} + +uintptr_t Simulator::stackLimit() const { return stackLimit_; } + +uintptr_t* Simulator::addressOfStackLimit() { return &stackLimit_; } + +bool Simulator::overRecursed(uintptr_t newsp) const { + if (newsp == 0) { + newsp = getRegister(sp); + } + return newsp <= stackLimit(); +} + +bool Simulator::overRecursedWithExtra(uint32_t extra) const { + uintptr_t newsp = getRegister(sp) - extra; + return newsp <= stackLimit(); +} + +// Unsupported instructions use format to print an error and stop execution. +void Simulator::format(SimInstruction* instr, const char* format) { + printf("Simulator found unsupported instruction:\n 0x%016lx: %s\n", + reinterpret_cast<intptr_t>(instr), format); + MOZ_CRASH(); +} + +// Note: With the code below we assume that all runtime calls return a 64 bits +// result. If they don't, the v1 result register contains a bogus value, which +// is fine because it is caller-saved. +typedef int64_t (*Prototype_General0)(); +typedef int64_t (*Prototype_General1)(int64_t arg0); +typedef int64_t (*Prototype_General2)(int64_t arg0, int64_t arg1); +typedef int64_t (*Prototype_General3)(int64_t arg0, int64_t arg1, int64_t arg2); +typedef int64_t (*Prototype_General4)(int64_t arg0, int64_t arg1, int64_t arg2, + int64_t arg3); +typedef int64_t (*Prototype_General5)(int64_t arg0, int64_t arg1, int64_t arg2, + int64_t arg3, int64_t arg4); +typedef int64_t (*Prototype_General6)(int64_t arg0, int64_t arg1, int64_t arg2, + int64_t arg3, int64_t arg4, int64_t arg5); +typedef int64_t (*Prototype_General7)(int64_t arg0, int64_t arg1, int64_t arg2, + int64_t arg3, int64_t arg4, int64_t arg5, + int64_t arg6); +typedef int64_t (*Prototype_General8)(int64_t arg0, int64_t arg1, int64_t arg2, + int64_t arg3, int64_t arg4, int64_t arg5, + int64_t arg6, int64_t arg7); +typedef int64_t (*Prototype_GeneralGeneralGeneralInt64)(int64_t arg0, + int64_t arg1, + int64_t arg2, + int64_t arg3); +typedef int64_t (*Prototype_GeneralGeneralInt64Int64)(int64_t arg0, + int64_t arg1, + int64_t arg2, + int64_t arg3); + +typedef int64_t (*Prototype_Int_Double)(double arg0); +typedef int64_t (*Prototype_Int_IntDouble)(int64_t arg0, double arg1); +typedef int64_t (*Prototype_Int_DoubleInt)(double arg0, int64_t arg1); +typedef int64_t (*Prototype_Int_DoubleIntInt)(double arg0, int64_t arg1, + int64_t arg2); +typedef int64_t (*Prototype_Int_IntDoubleIntInt)(int64_t arg0, double arg1, + int64_t arg2, int64_t arg3); + +typedef float (*Prototype_Float32_Float32)(float arg0); +typedef int64_t (*Prototype_Int_Float32)(float arg0); +typedef float (*Prototype_Float32_Float32Float32)(float arg0, float arg1); + +typedef double (*Prototype_Double_None)(); +typedef double (*Prototype_Double_Double)(double arg0); +typedef double (*Prototype_Double_Int)(int64_t arg0); +typedef double (*Prototype_Double_DoubleInt)(double arg0, int64_t arg1); +typedef double (*Prototype_Double_IntDouble)(int64_t arg0, double arg1); +typedef double (*Prototype_Double_DoubleDouble)(double arg0, double arg1); +typedef double (*Prototype_Double_DoubleDoubleDouble)(double arg0, double arg1, + double arg2); +typedef double (*Prototype_Double_DoubleDoubleDoubleDouble)(double arg0, + double arg1, + double arg2, + double arg3); + +typedef int32_t (*Prototype_Int32_General)(int64_t); +typedef int32_t (*Prototype_Int32_GeneralInt32)(int64_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int32)(int64_t, int32_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int32Int32)(int64_t, int32_t, + int32_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int32Int32Int32)(int64_t, int32_t, + int32_t, int32_t, + int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int32Int32Int32Int32)( + int64_t, int32_t, int32_t, int32_t, int32_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int32Int32Int32General)( + int64_t, int32_t, int32_t, int32_t, int32_t, int64_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int32Int32General)( + int64_t, int32_t, int32_t, int32_t, int64_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int32General)(int64_t, int32_t, + int32_t, int64_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int32Int64Int32)(int64_t, int32_t, + int32_t, int64_t, + int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt32GeneralInt32)(int64_t, int32_t, + int64_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt32GeneralInt32Int32)( + int64_t, int32_t, int64_t, int32_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt32Int64Int64Int32)(int64_t, int32_t, + int64_t, int64_t, + int32_t); +typedef int32_t (*Prototype_Int32_GeneralGeneral)(int64_t, int64_t); +typedef int32_t (*Prototype_Int32_GeneralGeneralGeneral)(int64_t, int64_t, + int64_t); +typedef int32_t (*Prototype_Int32_GeneralGeneralInt32Int32)(int64_t, int64_t, + int32_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int32Int32)(int64_t, int64_t, + int32_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int32Int32Int32Int32)( + int64_t, int64_t, int32_t, int32_t, int32_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int32Int64Int32)(int64_t, int64_t, + int32_t, int64_t, + int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int32Int32Int32)(int64_t, int64_t, + int32_t, int32_t, + int32_t); +typedef int32_t (*Prototype_Int32_GeneralGeneralInt32Int32Int32GeneralInt32)( + int64_t, int64_t, int32_t, int32_t, int32_t, int64_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralGeneralInt32General)(int32_t, int32_t, + int32_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int32Int64General)( + int64_t, int64_t, int32_t, int64_t, int64_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int64Int64)(int64_t, int64_t, + int64_t, int64_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int64Int64Int32)(int64_t, int64_t, + int64_t, int64_t, + int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int64General)(int64_t, int64_t, + int64_t, int64_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int64Int64General)( + int64_t, int64_t, int64_t, int64_t, int64_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int64Int64Int32Int32)( + int64_t, int64_t, int64_t, int64_t, int32_t, int32_t); +typedef int64_t (*Prototype_General_GeneralInt32)(int64_t, int32_t); +typedef int64_t (*Prototype_General_GeneralInt32Int32)(int64_t, int32_t, + int32_t); +typedef int64_t (*Prototype_General_GeneralInt32General)(int64_t, int32_t, + int64_t); +typedef int64_t (*Prototype_General_GeneralInt32Int32GeneralInt32)( + int64_t, int32_t, int32_t, int64_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralGeneralInt32GeneralInt32Int32Int32)( + int64_t, int64_t, int32_t, int64_t, int32_t, int32_t, int32_t); +typedef int64_t (*Prototype_Int64_General)(int64_t); +typedef int64_t (*Prototype_Int64_GeneralInt32)(int64_t, int32_t); +typedef int64_t (*Prototype_Int64_GeneralInt64)(int64_t, int64_t); +typedef int64_t (*Prototype_Int64_GeneralInt64Int32)(int64_t, int64_t, int32_t); +typedef int32_t (*Prototype_Int32_GeneralInt64Int64General)(int64_t, int64_t, + int64_t, int64_t); +// Generated by Assembler::break_()/stop(), ebreak code is passed as immediate +// field of a subsequent LUI instruction; otherwise returns -1 +static inline uint32_t get_ebreak_code(Instruction* instr) { + MOZ_ASSERT(instr->InstructionBits() == kBreakInstr); + uint8_t* cur = reinterpret_cast<uint8_t*>(instr); + Instruction* next_instr = reinterpret_cast<Instruction*>(cur + kInstrSize); + if (next_instr->BaseOpcodeFieldRaw() == LUI) + return (next_instr->Imm20UValue()); + else + return -1; +} + +// Software interrupt instructions are used by the simulator to call into C++. +void Simulator::SoftwareInterrupt() { + // There are two instructions that could get us here, the ebreak or ecall + // instructions are "SYSTEM" class opcode distinuished by Imm12Value field w/ + // the rest of instruction fields being zero + // We first check if we met a call_rt_redirected. + if (instr_.InstructionBits() == rtCallRedirInstr) { + Redirection* redirection = Redirection::FromSwiInstruction(instr_.instr()); + uintptr_t nativeFn = + reinterpret_cast<uintptr_t>(redirection->nativeFunction()); + + intptr_t arg0 = getRegister(a0); + intptr_t arg1 = getRegister(a1); + intptr_t arg2 = getRegister(a2); + intptr_t arg3 = getRegister(a3); + intptr_t arg4 = getRegister(a4); + intptr_t arg5 = getRegister(a5); + intptr_t arg6 = getRegister(a6); + intptr_t arg7 = getRegister(a7); + + // This is dodgy but it works because the C entry stubs are never moved. + // See comment in codegen-arm.cc and bug 1242173. + intptr_t saved_ra = getRegister(ra); + + intptr_t external = + reinterpret_cast<intptr_t>(redirection->nativeFunction()); + + bool stack_aligned = (getRegister(sp) & (ABIStackAlignment - 1)) == 0; + if (!stack_aligned) { + fprintf(stderr, "Runtime call with unaligned stack!\n"); + MOZ_CRASH(); + } + + if (single_stepping_) { + single_step_callback_(single_step_callback_arg_, this, nullptr); + } + if (FLAG_trace_sim) { + printf( + "Call to host function at %p with args %ld, %ld, %ld, %ld, %ld, %ld, " + "%ld, %ld\n", + reinterpret_cast<void*>(external), arg0, arg1, arg2, arg3, arg4, arg5, + arg6, arg7); + } + switch (redirection->type()) { + case Args_General0: { + Prototype_General0 target = + reinterpret_cast<Prototype_General0>(external); + int64_t result = target(); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setCallResult(result); + break; + } + case Args_General1: { + Prototype_General1 target = + reinterpret_cast<Prototype_General1>(external); + int64_t result = target(arg0); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setCallResult(result); + break; + } + case Args_General2: { + Prototype_General2 target = + reinterpret_cast<Prototype_General2>(external); + int64_t result = target(arg0, arg1); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setCallResult(result); + break; + } + case Args_General3: { + Prototype_General3 target = + reinterpret_cast<Prototype_General3>(external); + int64_t result = target(arg0, arg1, arg2); + if (FLAG_trace_sim) printf("ret %ld\n", result); + if (external == intptr_t(&js::wasm::Instance::wake_m32)) { + result = int32_t(result); + } + setCallResult(result); + break; + } + case Args_General4: { + Prototype_General4 target = + reinterpret_cast<Prototype_General4>(external); + int64_t result = target(arg0, arg1, arg2, arg3); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setCallResult(result); + break; + } + case Args_General5: { + Prototype_General5 target = + reinterpret_cast<Prototype_General5>(external); + int64_t result = target(arg0, arg1, arg2, arg3, arg4); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setCallResult(result); + break; + } + case Args_General6: { + Prototype_General6 target = + reinterpret_cast<Prototype_General6>(external); + int64_t result = target(arg0, arg1, arg2, arg3, arg4, arg5); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setCallResult(result); + break; + } + case Args_General7: { + Prototype_General7 target = + reinterpret_cast<Prototype_General7>(external); + int64_t result = target(arg0, arg1, arg2, arg3, arg4, arg5, arg6); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setCallResult(result); + break; + } + case Args_General8: { + Prototype_General8 target = + reinterpret_cast<Prototype_General8>(external); + int64_t result = target(arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setCallResult(result); + break; + } + case Args_Double_None: { + Prototype_Double_None target = + reinterpret_cast<Prototype_Double_None>(external); + double dresult = target(); + if (FLAG_trace_sim) printf("ret %f\n", dresult); + setCallResultDouble(dresult); + break; + } + case Args_Int_Double: { + double dval0 = getFpuRegisterDouble(fa0); + Prototype_Int_Double target = + reinterpret_cast<Prototype_Int_Double>(external); + int64_t result = target(dval0); + if (FLAG_trace_sim) printf("ret %ld\n", result); + if (external == intptr_t((int32_t(*)(double))JS::ToInt32)) { + result = int32_t(result); + } + setRegister(a0, result); + break; + } + case Args_Int_GeneralGeneralGeneralInt64: { + Prototype_GeneralGeneralGeneralInt64 target = + reinterpret_cast<Prototype_GeneralGeneralGeneralInt64>(external); + int64_t result = target(arg0, arg1, arg2, arg3); + if (FLAG_trace_sim) printf("ret %ld\n", result); + if (external == intptr_t(&js::wasm::Instance::wait_i32_m32)) { + result = int32_t(result); + } + setRegister(a0, result); + break; + } + case Args_Int_GeneralGeneralInt64Int64: { + Prototype_GeneralGeneralInt64Int64 target = + reinterpret_cast<Prototype_GeneralGeneralInt64Int64>(external); + int64_t result = target(arg0, arg1, arg2, arg3); + if (FLAG_trace_sim) printf("ret %ld\n", result); + if (external == intptr_t(&js::wasm::Instance::wait_i64_m32)) { + result = int32_t(result); + } + setRegister(a0, result); + break; + } + case Args_Int_DoubleInt: { + double dval = getFpuRegisterDouble(fa0); + Prototype_Int_DoubleInt target = + reinterpret_cast<Prototype_Int_DoubleInt>(external); + int64_t result = target(dval, arg0); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setRegister(a0, result); + break; + } + case Args_Int_DoubleIntInt: { + double dval = getFpuRegisterDouble(fa0); + Prototype_Int_DoubleIntInt target = + reinterpret_cast<Prototype_Int_DoubleIntInt>(external); + int64_t result = target(dval, arg1, arg2); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setRegister(a0, result); + break; + } + case Args_Int_IntDoubleIntInt: { + double dval = getFpuRegisterDouble(fa0); + Prototype_Int_IntDoubleIntInt target = + reinterpret_cast<Prototype_Int_IntDoubleIntInt>(external); + int64_t result = target(arg0, dval, arg2, arg3); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setRegister(a0, result); + break; + } + case Args_Double_Double: { + double dval0 = getFpuRegisterDouble(fa0); + Prototype_Double_Double target = + reinterpret_cast<Prototype_Double_Double>(external); + double dresult = target(dval0); + if (FLAG_trace_sim) printf("ret %f\n", dresult); + setCallResultDouble(dresult); + break; + } + case Args_Float32_Float32: { + float fval0; + fval0 = getFpuRegisterFloat(fa0); + Prototype_Float32_Float32 target = + reinterpret_cast<Prototype_Float32_Float32>(external); + float fresult = target(fval0); + if (FLAG_trace_sim) printf("ret %f\n", fresult); + setCallResultFloat(fresult); + break; + } + case Args_Int_Float32: { + float fval0; + fval0 = getFpuRegisterFloat(fa0); + Prototype_Int_Float32 target = + reinterpret_cast<Prototype_Int_Float32>(external); + int64_t result = target(fval0); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setRegister(a0, result); + break; + } + case Args_Float32_Float32Float32: { + float fval0; + float fval1; + fval0 = getFpuRegisterFloat(fa0); + fval1 = getFpuRegisterFloat(fa1); + Prototype_Float32_Float32Float32 target = + reinterpret_cast<Prototype_Float32_Float32Float32>(external); + float fresult = target(fval0, fval1); + if (FLAG_trace_sim) printf("ret %f\n", fresult); + setCallResultFloat(fresult); + break; + } + case Args_Double_Int: { + Prototype_Double_Int target = + reinterpret_cast<Prototype_Double_Int>(external); + double dresult = target(arg0); + if (FLAG_trace_sim) printf("ret %f\n", dresult); + setCallResultDouble(dresult); + break; + } + case Args_Double_DoubleInt: { + double dval0 = getFpuRegisterDouble(fa0); + Prototype_Double_DoubleInt target = + reinterpret_cast<Prototype_Double_DoubleInt>(external); + double dresult = target(dval0, arg0); + if (FLAG_trace_sim) printf("ret %f\n", dresult); + setCallResultDouble(dresult); + break; + } + case Args_Double_DoubleDouble: { + double dval0 = getFpuRegisterDouble(fa0); + double dval1 = getFpuRegisterDouble(fa1); + Prototype_Double_DoubleDouble target = + reinterpret_cast<Prototype_Double_DoubleDouble>(external); + double dresult = target(dval0, dval1); + if (FLAG_trace_sim) printf("ret %f\n", dresult); + setCallResultDouble(dresult); + break; + } + case Args_Double_IntDouble: { + double dval0 = getFpuRegisterDouble(fa0); + Prototype_Double_IntDouble target = + reinterpret_cast<Prototype_Double_IntDouble>(external); + double dresult = target(arg0, dval0); + if (FLAG_trace_sim) printf("ret %f\n", dresult); + setCallResultDouble(dresult); + break; + } + case Args_Int_IntDouble: { + double dval0 = getFpuRegisterDouble(fa0); + Prototype_Int_IntDouble target = + reinterpret_cast<Prototype_Int_IntDouble>(external); + int64_t result = target(arg0, dval0); + if (FLAG_trace_sim) printf("ret %ld\n", result); + setRegister(a0, result); + break; + } + case Args_Double_DoubleDoubleDouble: { + double dval0 = getFpuRegisterDouble(fa0); + double dval1 = getFpuRegisterDouble(fa1); + double dval2 = getFpuRegisterDouble(fa2); + Prototype_Double_DoubleDoubleDouble target = + reinterpret_cast<Prototype_Double_DoubleDoubleDouble>(external); + double dresult = target(dval0, dval1, dval2); + if (FLAG_trace_sim) printf("ret %f\n", dresult); + setCallResultDouble(dresult); + break; + } + case Args_Double_DoubleDoubleDoubleDouble: { + double dval0 = getFpuRegisterDouble(fa0); + double dval1 = getFpuRegisterDouble(fa1); + double dval2 = getFpuRegisterDouble(fa2); + double dval3 = getFpuRegisterDouble(fa3); + Prototype_Double_DoubleDoubleDoubleDouble target = + reinterpret_cast<Prototype_Double_DoubleDoubleDoubleDouble>( + external); + double dresult = target(dval0, dval1, dval2, dval3); + if (FLAG_trace_sim) printf("ret %f\n", dresult); + setCallResultDouble(dresult); + break; + } + case Args_Int32_General: { + int32_t ret = reinterpret_cast<Prototype_Int32_General>(nativeFn)(arg0); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32: { + int32_t ret = reinterpret_cast<Prototype_Int32_GeneralInt32>(nativeFn)( + arg0, I32(arg1)); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int32: { + int32_t ret = reinterpret_cast<Prototype_Int32_GeneralInt32Int32>( + nativeFn)(arg0, I32(arg1), I32(arg2)); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int32Int32: { + int32_t ret = reinterpret_cast<Prototype_Int32_GeneralInt32Int32Int32>( + nativeFn)(arg0, I32(arg1), I32(arg2), I32(arg3)); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int32Int32Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt32Int32Int32Int32>( + nativeFn)(arg0, I32(arg1), I32(arg2), I32(arg3), I32(arg4)); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int32Int32Int32Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt32Int32Int32Int32Int32>( + nativeFn)(arg0, I32(arg1), I32(arg2), I32(arg3), I32(arg4), + I32(arg5)); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int32Int32Int32General: { + int32_t ret = reinterpret_cast< + Prototype_Int32_GeneralInt32Int32Int32Int32General>(nativeFn)( + arg0, I32(arg1), I32(arg2), I32(arg3), I32(arg4), arg5); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int32Int32General: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt32Int32Int32General>( + nativeFn)(arg0, I32(arg1), I32(arg2), I32(arg3), arg4); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int32General: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt32Int32General>( + nativeFn)(arg0, I32(arg1), I32(arg2), arg3); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int32Int64Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt32Int32Int64Int32>( + nativeFn)(arg0, I32(arg1), I32(arg2), arg3, I32(arg4)); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32GeneralInt32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt32GeneralInt32>( + nativeFn)(arg0, I32(arg1), arg2, I32(arg3)); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32GeneralInt32Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt32GeneralInt32Int32>( + nativeFn)(arg0, I32(arg1), arg2, I32(arg3), I32(arg4)); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralGeneral: { + int32_t ret = reinterpret_cast<Prototype_Int32_GeneralGeneral>( + nativeFn)(arg0, arg1); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralInt32Int64Int64Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt32Int64Int64Int32>( + nativeFn)(arg0, I32(arg1), arg2, arg3, I32(arg4)); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralGeneralGeneral: { + int32_t ret = reinterpret_cast<Prototype_Int32_GeneralGeneralGeneral>( + nativeFn)(arg0, arg1, arg2); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_Int32_GeneralGeneralInt32Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralGeneralInt32Int32>( + nativeFn)(arg0, arg1, I32(arg2), I32(arg3)); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralInt64Int32Int32: { + int32_t ret = reinterpret_cast<Prototype_Int32_GeneralInt64Int32Int32>( + nativeFn)(arg0, arg1, I32(arg2), I32(arg3)); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralInt64Int32Int32Int32Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt64Int32Int32Int32Int32>( + nativeFn)(arg0, arg1, I32(arg2), I32(arg3), I32(arg4), + I32(arg5)); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralInt64Int32Int64Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt64Int32Int64Int32>( + nativeFn)(arg0, arg1, I32(arg2), arg3, I32(arg4)); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralInt64Int32Int64General: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt64Int32Int64General>( + nativeFn)(arg0, arg1, I32(arg2), arg3, arg4); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralInt64Int64Int64: { + int32_t ret = reinterpret_cast<Prototype_Int32_GeneralInt64Int64Int64>( + nativeFn)(arg0, arg1, arg2, arg3); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralInt64Int64Int64Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt64Int64Int64Int32>( + nativeFn)(arg0, arg1, arg2, arg3, I32(arg4)); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralInt64Int64General: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt64Int64General>( + nativeFn)(arg0, arg1, arg2, arg3); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralInt64Int64Int64General: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt64Int64Int64General>( + nativeFn)(arg0, arg1, arg2, arg3, arg4); + if (FLAG_trace_sim) printf("ret %d\n", ret); + setRegister(a0, I64(ret)); + break; + } + case Args_General_GeneralInt32: { + int64_t ret = reinterpret_cast<Prototype_General_GeneralInt32>( + nativeFn)(arg0, I32(arg1)); + if (FLAG_trace_sim) printf("ret %ld\n", ret); + setRegister(a0, ret); + break; + } + case js::jit::Args_Int32_GeneralInt64Int64Int64Int32Int32: { + int32_t ret = + reinterpret_cast<Prototype_Int32_GeneralInt64Int64Int64Int32Int32>( + nativeFn)(arg0, arg1, arg2, arg3, I32(arg4), I32(arg5)); + setRegister(a0, I64(ret)); + break; + } + case Args_General_GeneralInt32Int32: { + int64_t ret = reinterpret_cast<Prototype_General_GeneralInt32Int32>( + nativeFn)(arg0, I32(arg1), I32(arg2)); + if (FLAG_trace_sim) printf("ret %ld\n", ret); + setRegister(a0, ret); + break; + } + case Args_General_GeneralInt32General: { + int64_t ret = reinterpret_cast<Prototype_General_GeneralInt32General>( + nativeFn)(arg0, I32(arg1), arg2); + if (FLAG_trace_sim) printf("ret %ld\n", ret); + setRegister(a0, ret); + break; + } + case js::jit::Args_General_GeneralInt32Int32GeneralInt32: { + int64_t ret = + reinterpret_cast<Prototype_General_GeneralInt32Int32GeneralInt32>( + nativeFn)(arg0, I32(arg1), I32(arg2), arg3, I32(arg4)); + setRegister(a0, ret); + break; + } + case js::jit::Args_Int32_GeneralGeneralInt32Int32Int32GeneralInt32: { + int32_t ret = reinterpret_cast< + Prototype_Int32_GeneralGeneralInt32Int32Int32GeneralInt32>( + nativeFn)(arg0, arg1, I32(arg2), I32(arg3), I32(arg4), arg5, + I32(arg6)); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int32_GeneralGeneralInt32General: { + Prototype_Int32_GeneralGeneralInt32General target = + reinterpret_cast<Prototype_Int32_GeneralGeneralInt32General>( + external); + int64_t result = target(I32(arg0), I32(arg1), I32(arg2), I32(arg3)); + setRegister(a0, I64(result)); + break; + } + case js::jit::Args_Int32_GeneralGeneralInt32GeneralInt32Int32Int32: { + int64_t arg6 = getRegister(a6); + int32_t ret = reinterpret_cast< + Prototype_Int32_GeneralGeneralInt32GeneralInt32Int32Int32>( + nativeFn)(arg0, arg1, I32(arg2), arg3, I32(arg4), I32(arg5), + I32(arg6)); + setRegister(a0, I64(ret)); + break; + } + case js::jit::Args_Int64_General: { + int64_t ret = reinterpret_cast<Prototype_Int64_General>(nativeFn)(arg0); + if (FLAG_trace_sim) printf("ret %ld\n", ret); + setRegister(a0, ret); + break; + } + case js::jit::Args_Int64_GeneralInt32: { + int64_t ret = reinterpret_cast<Prototype_Int64_GeneralInt32>(nativeFn)( + arg0, I32(arg1)); + setRegister(a0, ret); + break; + } + case js::jit::Args_Int64_GeneralInt64: { + int64_t ret = reinterpret_cast<Prototype_Int64_GeneralInt64>(nativeFn)( + arg0, arg1); + setRegister(a0, ret); + break; + } + case js::jit::Args_Int64_GeneralInt64Int32: { + int64_t ret = reinterpret_cast<Prototype_Int64_GeneralInt64Int32>( + nativeFn)(arg0, arg1, I32(arg2)); + setRegister(a0, ret); + break; + } + default: + MOZ_CRASH("Unknown function type."); + } + + if (single_stepping_) { + single_step_callback_(single_step_callback_arg_, this, nullptr); + } + + setRegister(ra, saved_ra); + set_pc(getRegister(ra)); + + } else if (instr_.InstructionBits() == kBreakInstr && + (get_ebreak_code(instr_.instr()) <= kMaxStopCode)) { + uint32_t code = get_ebreak_code(instr_.instr()); + if (isWatchpoint(code)) { + printWatchpoint(code); + } else if (IsTracepoint(code)) { + if (!FLAG_debug_sim) { + MOZ_CRASH("Add --debug-sim when tracepoint instruction is used.\n"); + } + // printf("%d %d %d %d %d %d %d\n", code, code & LOG_TRACE, code & + // LOG_REGS, + // code & kDebuggerTracingDirectivesMask, TRACE_ENABLE, + // TRACE_DISABLE, kDebuggerTracingDirectivesMask); + switch (code & kDebuggerTracingDirectivesMask) { + case TRACE_ENABLE: + if (code & LOG_TRACE) { + FLAG_trace_sim = true; + } + if (code & LOG_REGS) { + RiscvDebugger dbg(this); + dbg.printAllRegs(); + } + break; + case TRACE_DISABLE: + if (code & LOG_TRACE) { + FLAG_trace_sim = false; + } + break; + default: + UNREACHABLE(); + } + } else { + increaseStopCounter(code); + handleStop(code); + } + } else { + // uint8_t code = get_ebreak_code(instr_.instr()) - kMaxStopCode - 1; + // switch (LNode::Opcode(code)) { + // #define EMIT_OP(OP, ...) \ +// case LNode::Opcode::OP:\ +// std::cout << #OP << std::endl; \ +// break; + // LIR_OPCODE_LIST(EMIT_OP); + // #undef EMIT_OP + // } + DieOrDebug(); + } +} + +// Stop helper functions. +bool Simulator::isWatchpoint(uint32_t code) { + return (code <= kMaxWatchpointCode); +} + +bool Simulator::IsTracepoint(uint32_t code) { + return (code <= kMaxTracepointCode && code > kMaxWatchpointCode); +} + +void Simulator::printWatchpoint(uint32_t code) { + RiscvDebugger dbg(this); + ++break_count_; + if (FLAG_riscv_print_watchpoint) { + printf("\n---- break %d marker: %20" PRIi64 " (instr count: %20" PRIi64 + ") ----\n", + code, break_count_, icount_); + dbg.printAllRegs(); // Print registers and continue running. + } +} + +void Simulator::handleStop(uint32_t code) { + // Stop if it is enabled, otherwise go on jumping over the stop + // and the message address. + if (isEnabledStop(code)) { + RiscvDebugger dbg(this); + dbg.Debug(); + } else { + set_pc(get_pc() + 2 * kInstrSize); + } +} + +bool Simulator::isStopInstruction(SimInstruction* instr) { + if (instr->InstructionBits() != kBreakInstr) return false; + int32_t code = get_ebreak_code(instr->instr()); + return code != -1 && static_cast<uint32_t>(code) > kMaxWatchpointCode && + static_cast<uint32_t>(code) <= kMaxStopCode; +} + +bool Simulator::isEnabledStop(uint32_t code) { + MOZ_ASSERT(code <= kMaxStopCode); + MOZ_ASSERT(code > kMaxWatchpointCode); + return !(watchedStops_[code].count_ & kStopDisabledBit); +} + +void Simulator::enableStop(uint32_t code) { + if (!isEnabledStop(code)) { + watchedStops_[code].count_ &= ~kStopDisabledBit; + } +} + +void Simulator::disableStop(uint32_t code) { + if (isEnabledStop(code)) { + watchedStops_[code].count_ |= kStopDisabledBit; + } +} + +void Simulator::increaseStopCounter(uint32_t code) { + MOZ_ASSERT(code <= kMaxStopCode); + if ((watchedStops_[code].count_ & ~(1 << 31)) == 0x7fffffff) { + printf( + "Stop counter for code %i has overflowed.\n" + "Enabling this code and reseting the counter to 0.\n", + code); + watchedStops_[code].count_ = 0; + enableStop(code); + } else { + watchedStops_[code].count_++; + } +} + +// Print a stop status. +void Simulator::printStopInfo(uint32_t code) { + if (code <= kMaxWatchpointCode) { + printf("That is a watchpoint, not a stop.\n"); + return; + } else if (code > kMaxStopCode) { + printf("Code too large, only %u stops can be used\n", kMaxStopCode + 1); + return; + } + const char* state = isEnabledStop(code) ? "Enabled" : "Disabled"; + int32_t count = watchedStops_[code].count_ & ~kStopDisabledBit; + // Don't print the state of unused breakpoints. + if (count != 0) { + if (watchedStops_[code].desc_) { + printf("stop %i - 0x%x: \t%s, \tcounter = %i, \t%s\n", code, code, state, + count, watchedStops_[code].desc_); + } else { + printf("stop %i - 0x%x: \t%s, \tcounter = %i\n", code, code, state, + count); + } + } +} + +void Simulator::SignalException(Exception e) { + printf("Error: Exception %i raised.", static_cast<int>(e)); + MOZ_CRASH(); +} + +// TODO(plind): refactor this messy debug code when we do unaligned access. +void Simulator::DieOrDebug() { + if (FLAG_riscv_trap_to_simulator_debugger) { + RiscvDebugger dbg(this); + dbg.Debug(); + } else { + MOZ_CRASH("Die"); + } +} + +// Executes the current instruction. +void Simulator::InstructionDecode(Instruction* instr) { + // if (FLAG_check_icache) { + // CheckICache(SimulatorProcess::icache(), instr); + // } + pc_modified_ = false; + + EmbeddedVector<char, 256> buffer; + + if (FLAG_trace_sim || FLAG_debug_sim) { + SNPrintF(trace_buf_, " "); + disasm::NameConverter converter; + disasm::Disassembler dasm(converter); + // Use a reasonably large buffer. + dasm.InstructionDecode(buffer, reinterpret_cast<byte*>(instr)); + + // printf("EXECUTING 0x%08" PRIxPTR " %-44s\n", + // reinterpret_cast<intptr_t>(instr), buffer.begin()); + } + + instr_ = instr; + switch (instr_.InstructionType()) { + case Instruction::kRType: + DecodeRVRType(); + break; + case Instruction::kR4Type: + DecodeRVR4Type(); + break; + case Instruction::kIType: + DecodeRVIType(); + break; + case Instruction::kSType: + DecodeRVSType(); + break; + case Instruction::kBType: + DecodeRVBType(); + break; + case Instruction::kUType: + DecodeRVUType(); + break; + case Instruction::kJType: + DecodeRVJType(); + break; + case Instruction::kCRType: + DecodeCRType(); + break; + case Instruction::kCAType: + DecodeCAType(); + break; + case Instruction::kCJType: + DecodeCJType(); + break; + case Instruction::kCBType: + DecodeCBType(); + break; + case Instruction::kCIType: + DecodeCIType(); + break; + case Instruction::kCIWType: + DecodeCIWType(); + break; + case Instruction::kCSSType: + DecodeCSSType(); + break; + case Instruction::kCLType: + DecodeCLType(); + break; + case Instruction::kCSType: + DecodeCSType(); + break; +# ifdef CAN_USE_RVV_INSTRUCTIONS + case Instruction::kVType: + DecodeVType(); + break; +# endif + default: + UNSUPPORTED(); + } + + if (FLAG_trace_sim) { + printf(" 0x%012" PRIxPTR " %-44s\t%s\n", + reinterpret_cast<intptr_t>(instr), buffer.start(), + trace_buf_.start()); + } + + if (!pc_modified_) { + setRegister(pc, reinterpret_cast<sreg_t>(instr) + instr->InstructionSize()); + } + + if (watch_address_ != nullptr) { + printf(" 0x%012" PRIxPTR " : 0x%016" REGIx_FORMAT " %14" REGId_FORMAT + " \n", + reinterpret_cast<intptr_t>(watch_address_), *watch_address_, + *watch_address_); + if (watch_value_ != *watch_address_) { + RiscvDebugger dbg(this); + dbg.Debug(); + watch_value_ = *watch_address_; + } + } +} + +void Simulator::enable_single_stepping(SingleStepCallback cb, void* arg) { + single_stepping_ = true; + single_step_callback_ = cb; + single_step_callback_arg_ = arg; + single_step_callback_(single_step_callback_arg_, this, (void*)get_pc()); +} + +void Simulator::disable_single_stepping() { + if (!single_stepping_) { + return; + } + single_step_callback_(single_step_callback_arg_, this, (void*)get_pc()); + single_stepping_ = false; + single_step_callback_ = nullptr; + single_step_callback_arg_ = nullptr; +} + +template <bool enableStopSimAt> +void Simulator::execute() { + if (single_stepping_) { + single_step_callback_(single_step_callback_arg_, this, nullptr); + } + + // Get the PC to simulate. Cannot use the accessor here as we need the + // raw PC value and not the one used as input to arithmetic instructions. + int64_t program_counter = get_pc(); + + while (program_counter != end_sim_pc) { + if (enableStopSimAt && (icount_ == Simulator::StopSimAt)) { + RiscvDebugger dbg(this); + dbg.Debug(); + } + if (single_stepping_) { + single_step_callback_(single_step_callback_arg_, this, + (void*)program_counter); + } + Instruction* instr = reinterpret_cast<Instruction*>(program_counter); + InstructionDecode(instr); + icount_++; + program_counter = get_pc(); + } + + if (single_stepping_) { + single_step_callback_(single_step_callback_arg_, this, nullptr); + } +} + +// RISCV Instruction Decode Routine +void Simulator::DecodeRVRType() { + switch (instr_.InstructionBits() & kRTypeMask) { + case RO_ADD: { + set_rd(sext_xlen(rs1() + rs2())); + break; + } + case RO_SUB: { + set_rd(sext_xlen(rs1() - rs2())); + break; + } + case RO_SLL: { + set_rd(sext_xlen(rs1() << (rs2() & (xlen - 1)))); + break; + } + case RO_SLT: { + set_rd(sreg_t(rs1()) < sreg_t(rs2())); + break; + } + case RO_SLTU: { + set_rd(reg_t(rs1()) < reg_t(rs2())); + break; + } + case RO_XOR: { + set_rd(rs1() ^ rs2()); + break; + } + case RO_SRL: { + set_rd(sext_xlen(zext_xlen(rs1()) >> (rs2() & (xlen - 1)))); + break; + } + case RO_SRA: { + set_rd(sext_xlen(sext_xlen(rs1()) >> (rs2() & (xlen - 1)))); + break; + } + case RO_OR: { + set_rd(rs1() | rs2()); + break; + } + case RO_AND: { + set_rd(rs1() & rs2()); + break; + } +# ifdef JS_CODEGEN_RISCV64 + case RO_ADDW: { + set_rd(sext32(rs1() + rs2())); + break; + } + case RO_SUBW: { + set_rd(sext32(rs1() - rs2())); + break; + } + case RO_SLLW: { + set_rd(sext32(rs1() << (rs2() & 0x1F))); + break; + } + case RO_SRLW: { + set_rd(sext32(uint32_t(rs1()) >> (rs2() & 0x1F))); + break; + } + case RO_SRAW: { + set_rd(sext32(int32_t(rs1()) >> (rs2() & 0x1F))); + break; + } +# endif /* JS_CODEGEN_RISCV64 */ + // TODO(riscv): Add RISCV M extension macro + case RO_MUL: { + set_rd(rs1() * rs2()); + break; + } + case RO_MULH: { + set_rd(mulh(rs1(), rs2())); + break; + } + case RO_MULHSU: { + set_rd(mulhsu(rs1(), rs2())); + break; + } + case RO_MULHU: { + set_rd(mulhu(rs1(), rs2())); + break; + } + case RO_DIV: { + sreg_t lhs = sext_xlen(rs1()); + sreg_t rhs = sext_xlen(rs2()); + if (rhs == 0) { + set_rd(-1); + } else if (lhs == INTPTR_MIN && rhs == -1) { + set_rd(lhs); + } else { + set_rd(sext_xlen(lhs / rhs)); + } + break; + } + case RO_DIVU: { + reg_t lhs = zext_xlen(rs1()); + reg_t rhs = zext_xlen(rs2()); + if (rhs == 0) { + set_rd(UINTPTR_MAX); + } else { + set_rd(zext_xlen(lhs / rhs)); + } + break; + } + case RO_REM: { + sreg_t lhs = sext_xlen(rs1()); + sreg_t rhs = sext_xlen(rs2()); + if (rhs == 0) { + set_rd(lhs); + } else if (lhs == INTPTR_MIN && rhs == -1) { + set_rd(0); + } else { + set_rd(sext_xlen(lhs % rhs)); + } + break; + } + case RO_REMU: { + reg_t lhs = zext_xlen(rs1()); + reg_t rhs = zext_xlen(rs2()); + if (rhs == 0) { + set_rd(lhs); + } else { + set_rd(zext_xlen(lhs % rhs)); + } + break; + } +# ifdef JS_CODEGEN_RISCV64 + case RO_MULW: { + set_rd(sext32(sext32(rs1()) * sext32(rs2()))); + break; + } + case RO_DIVW: { + sreg_t lhs = sext32(rs1()); + sreg_t rhs = sext32(rs2()); + if (rhs == 0) { + set_rd(-1); + } else if (lhs == INT32_MIN && rhs == -1) { + set_rd(lhs); + } else { + set_rd(sext32(lhs / rhs)); + } + break; + } + case RO_DIVUW: { + reg_t lhs = zext32(rs1()); + reg_t rhs = zext32(rs2()); + if (rhs == 0) { + set_rd(UINT32_MAX); + } else { + set_rd(zext32(lhs / rhs)); + } + break; + } + case RO_REMW: { + sreg_t lhs = sext32(rs1()); + sreg_t rhs = sext32(rs2()); + if (rhs == 0) { + set_rd(lhs); + } else if (lhs == INT32_MIN && rhs == -1) { + set_rd(0); + } else { + set_rd(sext32(lhs % rhs)); + } + break; + } + case RO_REMUW: { + reg_t lhs = zext32(rs1()); + reg_t rhs = zext32(rs2()); + if (rhs == 0) { + set_rd(zext32(lhs)); + } else { + set_rd(zext32(lhs % rhs)); + } + break; + } +# endif /*JS_CODEGEN_RISCV64*/ + // TODO(riscv): End Add RISCV M extension macro + default: { + switch (instr_.BaseOpcode()) { + case AMO: + DecodeRVRAType(); + break; + case OP_FP: + DecodeRVRFPType(); + break; + default: + UNSUPPORTED(); + } + } + } +} + +template <typename T> +T Simulator::FMaxMinHelper(T a, T b, MaxMinKind kind) { + // set invalid bit for signaling nan + if ((a == std::numeric_limits<T>::signaling_NaN()) || + (b == std::numeric_limits<T>::signaling_NaN())) { + set_csr_bits(csr_fflags, kInvalidOperation); + } + + T result = 0; + if (std::isnan(a) && std::isnan(b)) { + result = std::numeric_limits<float>::quiet_NaN(); + } else if (std::isnan(a)) { + result = b; + } else if (std::isnan(b)) { + result = a; + } else if (b == a) { // Handle -0.0 == 0.0 case. + if (kind == MaxMinKind::kMax) { + result = std::signbit(b) ? a : b; + } else { + result = std::signbit(b) ? b : a; + } + } else { + result = (kind == MaxMinKind::kMax) ? fmax(a, b) : fmin(a, b); + } + + return result; +} + +float Simulator::RoundF2FHelper(float input_val, int rmode) { + if (rmode == DYN) rmode = get_dynamic_rounding_mode(); + + float rounded = 0; + switch (rmode) { + case RNE: { // Round to Nearest, tiest to Even + rounded = floorf(input_val); + float error = input_val - rounded; + + // Take care of correctly handling the range [-0.5, -0.0], which must + // yield -0.0. + if ((-0.5 <= input_val) && (input_val < 0.0)) { + rounded = -0.0; + + // If the error is greater than 0.5, or is equal to 0.5 and the integer + // result is odd, round up. + } else if ((error > 0.5) || + ((error == 0.5) && (std::fmod(rounded, 2) != 0))) { + rounded++; + } + break; + } + case RTZ: // Round towards Zero + rounded = std::truncf(input_val); + break; + case RDN: // Round Down (towards -infinity) + rounded = floorf(input_val); + break; + case RUP: // Round Up (towards +infinity) + rounded = ceilf(input_val); + break; + case RMM: // Round to Nearest, tiest to Max Magnitude + rounded = std::roundf(input_val); + break; + default: + UNREACHABLE(); + } + + return rounded; +} + +double Simulator::RoundF2FHelper(double input_val, int rmode) { + if (rmode == DYN) rmode = get_dynamic_rounding_mode(); + + double rounded = 0; + switch (rmode) { + case RNE: { // Round to Nearest, tiest to Even + rounded = std::floor(input_val); + double error = input_val - rounded; + + // Take care of correctly handling the range [-0.5, -0.0], which must + // yield -0.0. + if ((-0.5 <= input_val) && (input_val < 0.0)) { + rounded = -0.0; + + // If the error is greater than 0.5, or is equal to 0.5 and the integer + // result is odd, round up. + } else if ((error > 0.5) || + ((error == 0.5) && (std::fmod(rounded, 2) != 0))) { + rounded++; + } + break; + } + case RTZ: // Round towards Zero + rounded = std::trunc(input_val); + break; + case RDN: // Round Down (towards -infinity) + rounded = std::floor(input_val); + break; + case RUP: // Round Up (towards +infinity) + rounded = std::ceil(input_val); + break; + case RMM: // Round to Nearest, tiest to Max Magnitude + rounded = std::round(input_val); + break; + default: + UNREACHABLE(); + } + return rounded; +} + +// convert rounded floating-point to integer types, handle input values that +// are out-of-range, underflow, or NaN, and set appropriate fflags +template <typename I_TYPE, typename F_TYPE> +I_TYPE Simulator::RoundF2IHelper(F_TYPE original, int rmode) { + MOZ_ASSERT(std::is_integral<I_TYPE>::value); + + MOZ_ASSERT((std::is_same<F_TYPE, float>::value || + std::is_same<F_TYPE, double>::value)); + + I_TYPE max_i = std::numeric_limits<I_TYPE>::max(); + I_TYPE min_i = std::numeric_limits<I_TYPE>::min(); + + if (!std::isfinite(original)) { + set_fflags(kInvalidOperation); + if (std::isnan(original) || + original == std::numeric_limits<F_TYPE>::infinity()) { + return max_i; + } else { + MOZ_ASSERT(original == -std::numeric_limits<F_TYPE>::infinity()); + return min_i; + } + } + + F_TYPE rounded = RoundF2FHelper(original, rmode); + if (original != rounded) set_fflags(kInexact); + + if (!std::isfinite(rounded)) { + set_fflags(kInvalidOperation); + if (std::isnan(rounded) || + rounded == std::numeric_limits<F_TYPE>::infinity()) { + return max_i; + } else { + MOZ_ASSERT(rounded == -std::numeric_limits<F_TYPE>::infinity()); + return min_i; + } + } + + // Since integer max values are either all 1s (for unsigned) or all 1s + // except for sign-bit (for signed), they cannot be represented precisely in + // floating point, in order to precisely tell whether the rounded floating + // point is within the max range, we compare against (max_i+1) which would + // have a single 1 w/ many trailing zeros + float max_i_plus_1 = + std::is_same<uint64_t, I_TYPE>::value + ? 0x1p64f // uint64_t::max + 1 cannot be represented in integers, + // so use its float representation directly + : static_cast<float>(static_cast<uint64_t>(max_i) + 1); + if (rounded >= max_i_plus_1) { + set_fflags(kOverflow | kInvalidOperation); + return max_i; + } + + // Since min_i (either 0 for unsigned, or for signed) is represented + // precisely in floating-point, comparing rounded directly against min_i + if (rounded <= min_i) { + if (rounded < min_i) set_fflags(kOverflow | kInvalidOperation); + return min_i; + } + + F_TYPE underflow_fval = + std::is_same<F_TYPE, float>::value ? FLT_MIN : DBL_MIN; + if (rounded < underflow_fval && rounded > -underflow_fval && rounded != 0) { + set_fflags(kUnderflow); + } + + return static_cast<I_TYPE>(rounded); +} + +template <typename T> +static int64_t FclassHelper(T value) { + switch (std::fpclassify(value)) { + case FP_INFINITE: + return (std::signbit(value) ? kNegativeInfinity : kPositiveInfinity); + case FP_NAN: + return (isSnan(value) ? kSignalingNaN : kQuietNaN); + case FP_NORMAL: + return (std::signbit(value) ? kNegativeNormalNumber + : kPositiveNormalNumber); + case FP_SUBNORMAL: + return (std::signbit(value) ? kNegativeSubnormalNumber + : kPositiveSubnormalNumber); + case FP_ZERO: + return (std::signbit(value) ? kNegativeZero : kPositiveZero); + default: + UNREACHABLE(); + } + UNREACHABLE(); + return FP_ZERO; +} + +template <typename T> +bool Simulator::CompareFHelper(T input1, T input2, FPUCondition cc) { + MOZ_ASSERT(std::is_floating_point<T>::value); + bool result = false; + switch (cc) { + case LT: + case LE: + // FLT, FLE are signaling compares + if (std::isnan(input1) || std::isnan(input2)) { + set_fflags(kInvalidOperation); + result = false; + } else { + result = (cc == LT) ? (input1 < input2) : (input1 <= input2); + } + break; + case EQ: + if (std::numeric_limits<T>::signaling_NaN() == input1 || + std::numeric_limits<T>::signaling_NaN() == input2) { + set_fflags(kInvalidOperation); + } + if (std::isnan(input1) || std::isnan(input2)) { + result = false; + } else { + result = (input1 == input2); + } + break; + case NE: + if (std::numeric_limits<T>::signaling_NaN() == input1 || + std::numeric_limits<T>::signaling_NaN() == input2) { + set_fflags(kInvalidOperation); + } + if (std::isnan(input1) || std::isnan(input2)) { + result = true; + } else { + result = (input1 != input2); + } + break; + default: + UNREACHABLE(); + } + return result; +} + +template <typename T> +static inline bool is_invalid_fmul(T src1, T src2) { + return (isinf(src1) && src2 == static_cast<T>(0.0)) || + (src1 == static_cast<T>(0.0) && isinf(src2)); +} + +template <typename T> +static inline bool is_invalid_fadd(T src1, T src2) { + return (isinf(src1) && isinf(src2) && + std::signbit(src1) != std::signbit(src2)); +} + +template <typename T> +static inline bool is_invalid_fsub(T src1, T src2) { + return (isinf(src1) && isinf(src2) && + std::signbit(src1) == std::signbit(src2)); +} + +template <typename T> +static inline bool is_invalid_fdiv(T src1, T src2) { + return ((src1 == 0 && src2 == 0) || (isinf(src1) && isinf(src2))); +} + +template <typename T> +static inline bool is_invalid_fsqrt(T src1) { + return (src1 < 0); +} + +int Simulator::loadLinkedW(uint64_t addr, SimInstruction* instr) { + if ((addr & 3) == 0) { + if (handleWasmSegFault(addr, 4)) { + return -1; + } + + volatile int32_t* ptr = reinterpret_cast<volatile int32_t*>(addr); + int32_t value = *ptr; + lastLLValue_ = value; + LLAddr_ = addr; + // Note that any memory write or "external" interrupt should reset this + // value to false. + LLBit_ = true; + return value; + } + printf("Unaligned write at 0x%016" PRIx64 ", pc=0x%016" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +int Simulator::storeConditionalW(uint64_t addr, int value, + SimInstruction* instr) { + // Correct behavior in this case, as defined by architecture, is to just + // return 0, but there is no point at allowing that. It is certainly an + // indicator of a bug. + if (addr != LLAddr_) { + printf("SC to bad address: 0x%016" PRIx64 ", pc=0x%016" PRIx64 + ", expected: 0x%016" PRIx64 "\n", + addr, reinterpret_cast<intptr_t>(instr), LLAddr_); + MOZ_CRASH(); + } + + if ((addr & 3) == 0) { + SharedMem<int32_t*> ptr = + SharedMem<int32_t*>::shared(reinterpret_cast<int32_t*>(addr)); + + if (!LLBit_) { + return 1; + } + + LLBit_ = false; + LLAddr_ = 0; + int32_t expected = int32_t(lastLLValue_); + int32_t old = + AtomicOperations::compareExchangeSeqCst(ptr, expected, int32_t(value)); + return (old == expected) ? 0 : 1; + } + printf("Unaligned SC at 0x%016" PRIx64 ", pc=0x%016" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +int64_t Simulator::loadLinkedD(uint64_t addr, SimInstruction* instr) { + if ((addr & kPointerAlignmentMask) == 0) { + if (handleWasmSegFault(addr, 8)) { + return -1; + } + + volatile int64_t* ptr = reinterpret_cast<volatile int64_t*>(addr); + int64_t value = *ptr; + lastLLValue_ = value; + LLAddr_ = addr; + // Note that any memory write or "external" interrupt should reset this + // value to false. + LLBit_ = true; + return value; + } + printf("Unaligned write at 0x%016" PRIx64 ", pc=0x%016" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +int Simulator::storeConditionalD(uint64_t addr, int64_t value, + SimInstruction* instr) { + // Correct behavior in this case, as defined by architecture, is to just + // return 0, but there is no point at allowing that. It is certainly an + // indicator of a bug. + if (addr != LLAddr_) { + printf("SC to bad address: 0x%016" PRIx64 ", pc=0x%016" PRIx64 + ", expected: 0x%016" PRIx64 "\n", + addr, reinterpret_cast<intptr_t>(instr), LLAddr_); + MOZ_CRASH(); + } + + if ((addr & kPointerAlignmentMask) == 0) { + SharedMem<int64_t*> ptr = + SharedMem<int64_t*>::shared(reinterpret_cast<int64_t*>(addr)); + + if (!LLBit_) { + return 1; + } + + LLBit_ = false; + LLAddr_ = 0; + int64_t expected = lastLLValue_; + int64_t old = + AtomicOperations::compareExchangeSeqCst(ptr, expected, int64_t(value)); + return (old == expected) ? 0 : 1; + } + printf("Unaligned SC at 0x%016" PRIx64 ", pc=0x%016" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +void Simulator::DecodeRVRAType() { + // TODO(riscv): Add macro for RISCV A extension + // Special handling for A extension instructions because it uses func5 + // For all A extension instruction, V8 simulator is pure sequential. No + // Memory address lock or other synchronizaiton behaviors. + switch (instr_.InstructionBits() & kRATypeMask) { + case RO_LR_W: { + sreg_t addr = rs1(); + set_rd(loadLinkedW(addr, &instr_)); + TraceLr(addr, getRegister(rd_reg()), getRegister(rd_reg())); + break; + } + case RO_SC_W: { + sreg_t addr = rs1(); + auto value = static_cast<int32_t>(rs2()); + auto result = + storeConditionalW(addr, static_cast<int32_t>(rs2()), &instr_); + set_rd(result); + if (!result) { + TraceSc(addr, value); + } + break; + } + case RO_AMOSWAP_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<uint32_t>( + rs1(), [&](uint32_t lhs) { return (uint32_t)rs2(); }, instr_.instr(), + WORD))); + break; + } + case RO_AMOADD_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<uint32_t>( + rs1(), [&](uint32_t lhs) { return lhs + (uint32_t)rs2(); }, + instr_.instr(), WORD))); + break; + } + case RO_AMOXOR_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<uint32_t>( + rs1(), [&](uint32_t lhs) { return lhs ^ (uint32_t)rs2(); }, + instr_.instr(), WORD))); + break; + } + case RO_AMOAND_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<uint32_t>( + rs1(), [&](uint32_t lhs) { return lhs & (uint32_t)rs2(); }, + instr_.instr(), WORD))); + break; + } + case RO_AMOOR_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<uint32_t>( + rs1(), [&](uint32_t lhs) { return lhs | (uint32_t)rs2(); }, + instr_.instr(), WORD))); + break; + } + case RO_AMOMIN_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<int32_t>( + rs1(), [&](int32_t lhs) { return std::min(lhs, (int32_t)rs2()); }, + instr_.instr(), WORD))); + break; + } + case RO_AMOMAX_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<int32_t>( + rs1(), [&](int32_t lhs) { return std::max(lhs, (int32_t)rs2()); }, + instr_.instr(), WORD))); + break; + } + case RO_AMOMINU_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<uint32_t>( + rs1(), [&](uint32_t lhs) { return std::min(lhs, (uint32_t)rs2()); }, + instr_.instr(), WORD))); + break; + } + case RO_AMOMAXU_W: { + if ((rs1() & 0x3) != 0) { + DieOrDebug(); + } + set_rd(sext32(amo<uint32_t>( + rs1(), [&](uint32_t lhs) { return std::max(lhs, (uint32_t)rs2()); }, + instr_.instr(), WORD))); + break; + } +# ifdef JS_CODEGEN_RISCV64 + case RO_LR_D: { + sreg_t addr = rs1(); + set_rd(loadLinkedD(addr, &instr_)); + TraceLr(addr, getRegister(rd_reg()), getRegister(rd_reg())); + break; + } + case RO_SC_D: { + sreg_t addr = rs1(); + auto value = static_cast<int64_t>(rs2()); + auto result = + storeConditionalD(addr, static_cast<int64_t>(rs2()), &instr_); + set_rd(result); + if (!result) { + TraceSc(addr, value); + } + break; + } + case RO_AMOSWAP_D: { + set_rd(amo<int64_t>( + rs1(), [&](int64_t lhs) { return rs2(); }, instr_.instr(), DWORD)); + break; + } + case RO_AMOADD_D: { + set_rd(amo<int64_t>( + rs1(), [&](int64_t lhs) { return lhs + rs2(); }, instr_.instr(), + DWORD)); + break; + } + case RO_AMOXOR_D: { + set_rd(amo<int64_t>( + rs1(), [&](int64_t lhs) { return lhs ^ rs2(); }, instr_.instr(), + DWORD)); + break; + } + case RO_AMOAND_D: { + set_rd(amo<int64_t>( + rs1(), [&](int64_t lhs) { return lhs & rs2(); }, instr_.instr(), + DWORD)); + break; + } + case RO_AMOOR_D: { + set_rd(amo<int64_t>( + rs1(), [&](int64_t lhs) { return lhs | rs2(); }, instr_.instr(), + DWORD)); + break; + } + case RO_AMOMIN_D: { + set_rd(amo<int64_t>( + rs1(), [&](int64_t lhs) { return std::min(lhs, rs2()); }, + instr_.instr(), DWORD)); + break; + } + case RO_AMOMAX_D: { + set_rd(amo<int64_t>( + rs1(), [&](int64_t lhs) { return std::max(lhs, rs2()); }, + instr_.instr(), DWORD)); + break; + } + case RO_AMOMINU_D: { + set_rd(amo<uint64_t>( + rs1(), [&](uint64_t lhs) { return std::min(lhs, (uint64_t)rs2()); }, + instr_.instr(), DWORD)); + break; + } + case RO_AMOMAXU_D: { + set_rd(amo<uint64_t>( + rs1(), [&](uint64_t lhs) { return std::max(lhs, (uint64_t)rs2()); }, + instr_.instr(), DWORD)); + break; + } +# endif /*JS_CODEGEN_RISCV64*/ + // TODO(riscv): End Add macro for RISCV A extension + default: { + UNSUPPORTED(); + } + } +} + +void Simulator::DecodeRVRFPType() { + // OP_FP instructions (F/D) uses func7 first. Some further uses func3 and + // rs2() + + // kRATypeMask is only for func7 + switch (instr_.InstructionBits() & kRFPTypeMask) { + // TODO(riscv): Add macro for RISCV F extension + case RO_FADD_S: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs1, float frs2) { + if (is_invalid_fadd(frs1, frs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else { + return frs1 + frs2; + } + }; + set_frd(CanonicalizeFPUOp2<float>(fn)); + break; + } + case RO_FSUB_S: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs1, float frs2) { + if (is_invalid_fsub(frs1, frs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else { + return frs1 - frs2; + } + }; + set_frd(CanonicalizeFPUOp2<float>(fn)); + break; + } + case RO_FMUL_S: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs1, float frs2) { + if (is_invalid_fmul(frs1, frs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else { + return frs1 * frs2; + } + }; + set_frd(CanonicalizeFPUOp2<float>(fn)); + break; + } + case RO_FDIV_S: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs1, float frs2) { + if (is_invalid_fdiv(frs1, frs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else if (frs2 == 0.0f) { + this->set_fflags(kDivideByZero); + return (std::signbit(frs1) == std::signbit(frs2) + ? std::numeric_limits<float>::infinity() + : -std::numeric_limits<float>::infinity()); + } else { + return frs1 / frs2; + } + }; + set_frd(CanonicalizeFPUOp2<float>(fn)); + break; + } + case RO_FSQRT_S: { + if (instr_.Rs2Value() == 0b00000) { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs) { + if (is_invalid_fsqrt(frs)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else { + return std::sqrt(frs); + } + }; + set_frd(CanonicalizeFPUOp1<float>(fn)); + } else { + UNSUPPORTED(); + } + break; + } + case RO_FSGNJ_S: { // RO_FSGNJN_S RO_FSQNJX_S + switch (instr_.Funct3Value()) { + case 0b000: { // RO_FSGNJ_S + set_frd(fsgnj32(frs1_boxed(), frs2_boxed(), false, false)); + break; + } + case 0b001: { // RO_FSGNJN_S + set_frd(fsgnj32(frs1_boxed(), frs2_boxed(), true, false)); + break; + } + case 0b010: { // RO_FSQNJX_S + set_frd(fsgnj32(frs1_boxed(), frs2_boxed(), false, true)); + break; + } + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FMIN_S: { // RO_FMAX_S + switch (instr_.Funct3Value()) { + case 0b000: { // RO_FMIN_S + set_frd(FMaxMinHelper(frs1(), frs2(), MaxMinKind::kMin)); + break; + } + case 0b001: { // RO_FMAX_S + set_frd(FMaxMinHelper(frs1(), frs2(), MaxMinKind::kMax)); + break; + } + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FCVT_W_S: { // RO_FCVT_WU_S , 64F RO_FCVT_L_S RO_FCVT_LU_S + float original_val = frs1(); + switch (instr_.Rs2Value()) { + case 0b00000: { // RO_FCVT_W_S + set_rd(RoundF2IHelper<int32_t>(original_val, instr_.RoundMode())); + break; + } + case 0b00001: { // RO_FCVT_WU_S + set_rd(sext32( + RoundF2IHelper<uint32_t>(original_val, instr_.RoundMode()))); + break; + } +# ifdef JS_CODEGEN_RISCV64 + case 0b00010: { // RO_FCVT_L_S + set_rd(RoundF2IHelper<int64_t>(original_val, instr_.RoundMode())); + break; + } + case 0b00011: { // RO_FCVT_LU_S + set_rd(RoundF2IHelper<uint64_t>(original_val, instr_.RoundMode())); + break; + } +# endif /* JS_CODEGEN_RISCV64 */ + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FMV: { // RO_FCLASS_S + switch (instr_.Funct3Value()) { + case 0b000: { + if (instr_.Rs2Value() == 0b00000) { + // RO_FMV_X_W + set_rd(sext32(getFpuRegister(rs1_reg()))); + } else { + UNSUPPORTED(); + } + break; + } + case 0b001: { // RO_FCLASS_S + set_rd(FclassHelper(frs1())); + break; + } + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FLE_S: { // RO_FEQ_S RO_FLT_S RO_FLE_S + switch (instr_.Funct3Value()) { + case 0b010: { // RO_FEQ_S + set_rd(CompareFHelper(frs1(), frs2(), EQ)); + break; + } + case 0b001: { // RO_FLT_S + set_rd(CompareFHelper(frs1(), frs2(), LT)); + break; + } + case 0b000: { // RO_FLE_S + set_rd(CompareFHelper(frs1(), frs2(), LE)); + break; + } + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FCVT_S_W: { // RO_FCVT_S_WU , 64F RO_FCVT_S_L RO_FCVT_S_LU + switch (instr_.Rs2Value()) { + case 0b00000: { // RO_FCVT_S_W + set_frd(static_cast<float>((int32_t)rs1())); + break; + } + case 0b00001: { // RO_FCVT_S_WU + set_frd(static_cast<float>((uint32_t)rs1())); + break; + } +# ifdef JS_CODEGEN_RISCV64 + case 0b00010: { // RO_FCVT_S_L + set_frd(static_cast<float>((int64_t)rs1())); + break; + } + case 0b00011: { // RO_FCVT_S_LU + set_frd(static_cast<float>((uint64_t)rs1())); + break; + } +# endif /* JS_CODEGEN_RISCV64 */ + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FMV_W_X: { + if (instr_.Funct3Value() == 0b000) { + // since FMV preserves source bit-pattern, no need to canonize + Float32 result = Float32::FromBits((uint32_t)rs1()); + set_frd(result); + } else { + UNSUPPORTED(); + } + break; + } + // TODO(riscv): Add macro for RISCV D extension + case RO_FADD_D: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs1, double drs2) { + if (is_invalid_fadd(drs1, drs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else { + return drs1 + drs2; + } + }; + set_drd(CanonicalizeFPUOp2<double>(fn)); + break; + } + case RO_FSUB_D: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs1, double drs2) { + if (is_invalid_fsub(drs1, drs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else { + return drs1 - drs2; + } + }; + set_drd(CanonicalizeFPUOp2<double>(fn)); + break; + } + case RO_FMUL_D: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs1, double drs2) { + if (is_invalid_fmul(drs1, drs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else { + return drs1 * drs2; + } + }; + set_drd(CanonicalizeFPUOp2<double>(fn)); + break; + } + case RO_FDIV_D: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs1, double drs2) { + if (is_invalid_fdiv(drs1, drs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else if (drs2 == 0.0) { + this->set_fflags(kDivideByZero); + return (std::signbit(drs1) == std::signbit(drs2) + ? std::numeric_limits<double>::infinity() + : -std::numeric_limits<double>::infinity()); + } else { + return drs1 / drs2; + } + }; + set_drd(CanonicalizeFPUOp2<double>(fn)); + break; + } + case RO_FSQRT_D: { + if (instr_.Rs2Value() == 0b00000) { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs) { + if (is_invalid_fsqrt(drs)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else { + return std::sqrt(drs); + } + }; + set_drd(CanonicalizeFPUOp1<double>(fn)); + } else { + UNSUPPORTED(); + } + break; + } + case RO_FSGNJ_D: { // RO_FSGNJN_D RO_FSQNJX_D + switch (instr_.Funct3Value()) { + case 0b000: { // RO_FSGNJ_D + set_drd(fsgnj64(drs1_boxed(), drs2_boxed(), false, false)); + break; + } + case 0b001: { // RO_FSGNJN_D + set_drd(fsgnj64(drs1_boxed(), drs2_boxed(), true, false)); + break; + } + case 0b010: { // RO_FSQNJX_D + set_drd(fsgnj64(drs1_boxed(), drs2_boxed(), false, true)); + break; + } + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FMIN_D: { // RO_FMAX_D + switch (instr_.Funct3Value()) { + case 0b000: { // RO_FMIN_D + set_drd(FMaxMinHelper(drs1(), drs2(), MaxMinKind::kMin)); + break; + } + case 0b001: { // RO_FMAX_D + set_drd(FMaxMinHelper(drs1(), drs2(), MaxMinKind::kMax)); + break; + } + default: { + UNSUPPORTED(); + } + } + break; + } + case (RO_FCVT_S_D & kRFPTypeMask): { + if (instr_.Rs2Value() == 0b00001) { + auto fn = [](double drs) { return static_cast<float>(drs); }; + set_frd(CanonicalizeDoubleToFloatOperation(fn)); + } else { + UNSUPPORTED(); + } + break; + } + case RO_FCVT_D_S: { + if (instr_.Rs2Value() == 0b00000) { + auto fn = [](float frs) { return static_cast<double>(frs); }; + set_drd(CanonicalizeFloatToDoubleOperation(fn)); + } else { + UNSUPPORTED(); + } + break; + } + case RO_FLE_D: { // RO_FEQ_D RO_FLT_D RO_FLE_D + switch (instr_.Funct3Value()) { + case 0b010: { // RO_FEQ_S + set_rd(CompareFHelper(drs1(), drs2(), EQ)); + break; + } + case 0b001: { // RO_FLT_D + set_rd(CompareFHelper(drs1(), drs2(), LT)); + break; + } + case 0b000: { // RO_FLE_D + set_rd(CompareFHelper(drs1(), drs2(), LE)); + break; + } + default: { + UNSUPPORTED(); + } + } + break; + } + case (RO_FCLASS_D & kRFPTypeMask): { // RO_FCLASS_D , 64D RO_FMV_X_D + if (instr_.Rs2Value() != 0b00000) { + UNSUPPORTED(); + } + switch (instr_.Funct3Value()) { + case 0b001: { // RO_FCLASS_D + set_rd(FclassHelper(drs1())); + break; + } +# ifdef JS_CODEGEN_RISCV64 + case 0b000: { // RO_FMV_X_D + set_rd(bit_cast<int64_t>(drs1())); + break; + } +# endif /* JS_CODEGEN_RISCV64 */ + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FCVT_W_D: { // RO_FCVT_WU_D , 64F RO_FCVT_L_D RO_FCVT_LU_D + double original_val = drs1(); + switch (instr_.Rs2Value()) { + case 0b00000: { // RO_FCVT_W_D + set_rd(RoundF2IHelper<int32_t>(original_val, instr_.RoundMode())); + break; + } + case 0b00001: { // RO_FCVT_WU_D + set_rd(sext32( + RoundF2IHelper<uint32_t>(original_val, instr_.RoundMode()))); + break; + } +# ifdef JS_CODEGEN_RISCV64 + case 0b00010: { // RO_FCVT_L_D + set_rd(RoundF2IHelper<int64_t>(original_val, instr_.RoundMode())); + break; + } + case 0b00011: { // RO_FCVT_LU_D + set_rd(RoundF2IHelper<uint64_t>(original_val, instr_.RoundMode())); + break; + } +# endif /* JS_CODEGEN_RISCV64 */ + default: { + UNSUPPORTED(); + } + } + break; + } + case RO_FCVT_D_W: { // RO_FCVT_D_WU , 64F RO_FCVT_D_L RO_FCVT_D_LU + switch (instr_.Rs2Value()) { + case 0b00000: { // RO_FCVT_D_W + set_drd((int32_t)rs1()); + break; + } + case 0b00001: { // RO_FCVT_D_WU + set_drd((uint32_t)rs1()); + break; + } +# ifdef JS_CODEGEN_RISCV64 + case 0b00010: { // RO_FCVT_D_L + set_drd((int64_t)rs1()); + break; + } + case 0b00011: { // RO_FCVT_D_LU + set_drd((uint64_t)rs1()); + break; + } +# endif /* JS_CODEGEN_RISCV64 */ + default: { + UNSUPPORTED(); + } + } + break; + } +# ifdef JS_CODEGEN_RISCV64 + case RO_FMV_D_X: { + if (instr_.Funct3Value() == 0b000 && instr_.Rs2Value() == 0b00000) { + // Since FMV preserves source bit-pattern, no need to canonize + set_drd(bit_cast<double>(rs1())); + } else { + UNSUPPORTED(); + } + break; + } +# endif /* JS_CODEGEN_RISCV64 */ + default: { + UNSUPPORTED(); + } + } +} + +void Simulator::DecodeRVR4Type() { + switch (instr_.InstructionBits() & kR4TypeMask) { + // TODO(riscv): use F Extension macro block + case RO_FMADD_S: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs1, float frs2, float frs3) { + if (is_invalid_fmul(frs1, frs2) || is_invalid_fadd(frs1 * frs2, frs3)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else { + return std::fma(frs1, frs2, frs3); + } + }; + set_frd(CanonicalizeFPUOp3<float>(fn)); + break; + } + case RO_FMSUB_S: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs1, float frs2, float frs3) { + if (is_invalid_fmul(frs1, frs2) || is_invalid_fsub(frs1 * frs2, frs3)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else { + return std::fma(frs1, frs2, -frs3); + } + }; + set_frd(CanonicalizeFPUOp3<float>(fn)); + break; + } + case RO_FNMSUB_S: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs1, float frs2, float frs3) { + if (is_invalid_fmul(frs1, frs2) || is_invalid_fsub(frs3, frs1 * frs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else { + return -std::fma(frs1, frs2, -frs3); + } + }; + set_frd(CanonicalizeFPUOp3<float>(fn)); + break; + } + case RO_FNMADD_S: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](float frs1, float frs2, float frs3) { + if (is_invalid_fmul(frs1, frs2) || is_invalid_fadd(frs1 * frs2, frs3)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<float>::quiet_NaN(); + } else { + return -std::fma(frs1, frs2, frs3); + } + }; + set_frd(CanonicalizeFPUOp3<float>(fn)); + break; + } + // TODO(riscv): use F Extension macro block + case RO_FMADD_D: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs1, double drs2, double drs3) { + if (is_invalid_fmul(drs1, drs2) || is_invalid_fadd(drs1 * drs2, drs3)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else { + return std::fma(drs1, drs2, drs3); + } + }; + set_drd(CanonicalizeFPUOp3<double>(fn)); + break; + } + case RO_FMSUB_D: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs1, double drs2, double drs3) { + if (is_invalid_fmul(drs1, drs2) || is_invalid_fsub(drs1 * drs2, drs3)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else { + return std::fma(drs1, drs2, -drs3); + } + }; + set_drd(CanonicalizeFPUOp3<double>(fn)); + break; + } + case RO_FNMSUB_D: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs1, double drs2, double drs3) { + if (is_invalid_fmul(drs1, drs2) || is_invalid_fsub(drs3, drs1 * drs2)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else { + return -std::fma(drs1, drs2, -drs3); + } + }; + set_drd(CanonicalizeFPUOp3<double>(fn)); + break; + } + case RO_FNMADD_D: { + // TODO(riscv): use rm value (round mode) + auto fn = [this](double drs1, double drs2, double drs3) { + if (is_invalid_fmul(drs1, drs2) || is_invalid_fadd(drs1 * drs2, drs3)) { + this->set_fflags(kInvalidOperation); + return std::numeric_limits<double>::quiet_NaN(); + } else { + return -std::fma(drs1, drs2, drs3); + } + }; + set_drd(CanonicalizeFPUOp3<double>(fn)); + break; + } + default: + UNSUPPORTED(); + } +} + +# ifdef CAN_USE_RVV_INSTRUCTIONS +bool Simulator::DecodeRvvVL() { + uint32_t instr_temp = + instr_.InstructionBits() & (kRvvMopMask | kRvvNfMask | kBaseOpcodeMask); + if (RO_V_VL == instr_temp) { + if (!(instr_.InstructionBits() & (kRvvRs2Mask))) { + switch (instr_.vl_vs_width()) { + case 8: { + RVV_VI_LD(0, (i * nf + fn), int8, false); + break; + } + case 16: { + RVV_VI_LD(0, (i * nf + fn), int16, false); + break; + } + case 32: { + RVV_VI_LD(0, (i * nf + fn), int32, false); + break; + } + case 64: { + RVV_VI_LD(0, (i * nf + fn), int64, false); + break; + } + default: + UNIMPLEMENTED_RISCV(); + break; + } + return true; + } else { + UNIMPLEMENTED_RISCV(); + return true; + } + } else if (RO_V_VLS == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else if (RO_V_VLX == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else if (RO_V_VLSEG2 == instr_temp || RO_V_VLSEG3 == instr_temp || + RO_V_VLSEG4 == instr_temp || RO_V_VLSEG5 == instr_temp || + RO_V_VLSEG6 == instr_temp || RO_V_VLSEG7 == instr_temp || + RO_V_VLSEG8 == instr_temp) { + if (!(instr_.InstructionBits() & (kRvvRs2Mask))) { + UNIMPLEMENTED_RISCV(); + return true; + } else { + UNIMPLEMENTED_RISCV(); + return true; + } + } else if (RO_V_VLSSEG2 == instr_temp || RO_V_VLSSEG3 == instr_temp || + RO_V_VLSSEG4 == instr_temp || RO_V_VLSSEG5 == instr_temp || + RO_V_VLSSEG6 == instr_temp || RO_V_VLSSEG7 == instr_temp || + RO_V_VLSSEG8 == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else if (RO_V_VLXSEG2 == instr_temp || RO_V_VLXSEG3 == instr_temp || + RO_V_VLXSEG4 == instr_temp || RO_V_VLXSEG5 == instr_temp || + RO_V_VLXSEG6 == instr_temp || RO_V_VLXSEG7 == instr_temp || + RO_V_VLXSEG8 == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else { + return false; + } +} + +bool Simulator::DecodeRvvVS() { + uint32_t instr_temp = + instr_.InstructionBits() & (kRvvMopMask | kRvvNfMask | kBaseOpcodeMask); + if (RO_V_VS == instr_temp) { + if (!(instr_.InstructionBits() & (kRvvRs2Mask))) { + switch (instr_.vl_vs_width()) { + case 8: { + RVV_VI_ST(0, (i * nf + fn), uint8, false); + break; + } + case 16: { + RVV_VI_ST(0, (i * nf + fn), uint16, false); + break; + } + case 32: { + RVV_VI_ST(0, (i * nf + fn), uint32, false); + break; + } + case 64: { + RVV_VI_ST(0, (i * nf + fn), uint64, false); + break; + } + default: + UNIMPLEMENTED_RISCV(); + break; + } + } else { + UNIMPLEMENTED_RISCV(); + } + return true; + } else if (RO_V_VSS == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else if (RO_V_VSX == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else if (RO_V_VSU == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else if (RO_V_VSSEG2 == instr_temp || RO_V_VSSEG3 == instr_temp || + RO_V_VSSEG4 == instr_temp || RO_V_VSSEG5 == instr_temp || + RO_V_VSSEG6 == instr_temp || RO_V_VSSEG7 == instr_temp || + RO_V_VSSEG8 == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else if (RO_V_VSSSEG2 == instr_temp || RO_V_VSSSEG3 == instr_temp || + RO_V_VSSSEG4 == instr_temp || RO_V_VSSSEG5 == instr_temp || + RO_V_VSSSEG6 == instr_temp || RO_V_VSSSEG7 == instr_temp || + RO_V_VSSSEG8 == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else if (RO_V_VSXSEG2 == instr_temp || RO_V_VSXSEG3 == instr_temp || + RO_V_VSXSEG4 == instr_temp || RO_V_VSXSEG5 == instr_temp || + RO_V_VSXSEG6 == instr_temp || RO_V_VSXSEG7 == instr_temp || + RO_V_VSXSEG8 == instr_temp) { + UNIMPLEMENTED_RISCV(); + return true; + } else { + return false; + } +} +# endif + +void Simulator::DecodeRVIType() { + switch (instr_.InstructionBits() & kITypeMask) { + case RO_JALR: { + set_rd(get_pc() + kInstrSize); + // Note: No need to shift 2 for JALR's imm12, but set lowest bit to 0. + sreg_t next_pc = (rs1() + imm12()) & ~sreg_t(1); + set_pc(next_pc); + break; + } + case RO_LB: { + sreg_t addr = rs1() + imm12(); + int8_t val = ReadMem<int8_t>(addr, instr_.instr()); + set_rd(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rd_reg())); + break; + } + case RO_LH: { + sreg_t addr = rs1() + imm12(); + int16_t val = ReadMem<int16_t>(addr, instr_.instr()); + set_rd(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rd_reg())); + break; + } + case RO_LW: { + sreg_t addr = rs1() + imm12(); + int32_t val = ReadMem<int32_t>(addr, instr_.instr()); + set_rd(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rd_reg())); + break; + } + case RO_LBU: { + sreg_t addr = rs1() + imm12(); + uint8_t val = ReadMem<uint8_t>(addr, instr_.instr()); + set_rd(zext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rd_reg())); + break; + } + case RO_LHU: { + sreg_t addr = rs1() + imm12(); + uint16_t val = ReadMem<uint16_t>(addr, instr_.instr()); + set_rd(zext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rd_reg())); + break; + } +# ifdef JS_CODEGEN_RISCV64 + case RO_LWU: { + int64_t addr = rs1() + imm12(); + uint32_t val = ReadMem<uint32_t>(addr, instr_.instr()); + set_rd(zext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rd_reg())); + break; + } + case RO_LD: { + int64_t addr = rs1() + imm12(); + int64_t val = ReadMem<int64_t>(addr, instr_.instr()); + set_rd(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rd_reg())); + break; + } +# endif /*JS_CODEGEN_RISCV64*/ + case RO_ADDI: { + set_rd(sext_xlen(rs1() + imm12())); + break; + } + case RO_SLTI: { + set_rd(sreg_t(rs1()) < sreg_t(imm12())); + break; + } + case RO_SLTIU: { + set_rd(reg_t(rs1()) < reg_t(imm12())); + break; + } + case RO_XORI: { + set_rd(imm12() ^ rs1()); + break; + } + case RO_ORI: { + set_rd(imm12() | rs1()); + break; + } + case RO_ANDI: { + set_rd(imm12() & rs1()); + break; + } + case RO_SLLI: { + require(shamt6() < xlen); + set_rd(sext_xlen(rs1() << shamt6())); + break; + } + case RO_SRLI: { // RO_SRAI + if (!instr_.IsArithShift()) { + require(shamt6() < xlen); + set_rd(sext_xlen(zext_xlen(rs1()) >> shamt6())); + } else { + require(shamt6() < xlen); + set_rd(sext_xlen(sext_xlen(rs1()) >> shamt6())); + } + break; + } +# ifdef JS_CODEGEN_RISCV64 + case RO_ADDIW: { + set_rd(sext32(rs1() + imm12())); + break; + } + case RO_SLLIW: { + set_rd(sext32(rs1() << shamt5())); + break; + } + case RO_SRLIW: { // RO_SRAIW + if (!instr_.IsArithShift()) { + set_rd(sext32(uint32_t(rs1()) >> shamt5())); + } else { + set_rd(sext32(int32_t(rs1()) >> shamt5())); + } + break; + } +# endif /*JS_CODEGEN_RISCV64*/ + case RO_FENCE: { + // DO nothing in sumulator + break; + } + case RO_ECALL: { // RO_EBREAK + if (instr_.Imm12Value() == 0) { // ECALL + SoftwareInterrupt(); + } else if (instr_.Imm12Value() == 1) { // EBREAK + uint8_t code = get_ebreak_code(instr_.instr()); + if (code == kWasmTrapCode) { + HandleWasmTrap(); + } + SoftwareInterrupt(); + } else { + UNSUPPORTED(); + } + break; + } + // TODO(riscv): use Zifencei Standard Extension macro block + case RO_FENCE_I: { + // spike: flush icache. + break; + } + // TODO(riscv): use Zicsr Standard Extension macro block + case RO_CSRRW: { + if (rd_reg() != zero_reg) { + set_rd(zext_xlen(read_csr_value(csr_reg()))); + } + write_csr_value(csr_reg(), rs1()); + break; + } + case RO_CSRRS: { + set_rd(zext_xlen(read_csr_value(csr_reg()))); + if (rs1_reg() != zero_reg) { + set_csr_bits(csr_reg(), rs1()); + } + break; + } + case RO_CSRRC: { + set_rd(zext_xlen(read_csr_value(csr_reg()))); + if (rs1_reg() != zero_reg) { + clear_csr_bits(csr_reg(), rs1()); + } + break; + } + case RO_CSRRWI: { + if (rd_reg() != zero_reg) { + set_rd(zext_xlen(read_csr_value(csr_reg()))); + } + if (csr_reg() == csr_cycle) { + if (imm5CSR() == kWasmTrapCode) { + HandleWasmTrap(); + return; + } + } + write_csr_value(csr_reg(), imm5CSR()); + break; + } + case RO_CSRRSI: { + set_rd(zext_xlen(read_csr_value(csr_reg()))); + if (imm5CSR() != 0) { + set_csr_bits(csr_reg(), imm5CSR()); + } + break; + } + case RO_CSRRCI: { + set_rd(zext_xlen(read_csr_value(csr_reg()))); + if (imm5CSR() != 0) { + clear_csr_bits(csr_reg(), imm5CSR()); + } + break; + } + // TODO(riscv): use F Extension macro block + case RO_FLW: { + sreg_t addr = rs1() + imm12(); + uint32_t val = ReadMem<uint32_t>(addr, instr_.instr()); + set_frd(Float32::FromBits(val), false); + TraceMemRdFloat(addr, Float32::FromBits(val), getFpuRegister(frd_reg())); + break; + } + // TODO(riscv): use D Extension macro block + case RO_FLD: { + sreg_t addr = rs1() + imm12(); + uint64_t val = ReadMem<uint64_t>(addr, instr_.instr()); + set_drd(Float64::FromBits(val), false); + TraceMemRdDouble(addr, Float64::FromBits(val), getFpuRegister(frd_reg())); + break; + } + default: { +# ifdef CAN_USE_RVV_INSTRUCTIONS + if (!DecodeRvvVL()) { + UNSUPPORTED(); + } + break; +# else + UNSUPPORTED(); +# endif + } + } +} + +void Simulator::DecodeRVSType() { + switch (instr_.InstructionBits() & kSTypeMask) { + case RO_SB: + WriteMem<uint8_t>(rs1() + s_imm12(), (uint8_t)rs2(), instr_.instr()); + break; + case RO_SH: + WriteMem<uint16_t>(rs1() + s_imm12(), (uint16_t)rs2(), instr_.instr()); + break; + case RO_SW: + WriteMem<uint32_t>(rs1() + s_imm12(), (uint32_t)rs2(), instr_.instr()); + break; +# ifdef JS_CODEGEN_RISCV64 + case RO_SD: + WriteMem<uint64_t>(rs1() + s_imm12(), (uint64_t)rs2(), instr_.instr()); + break; +# endif /*JS_CODEGEN_RISCV64*/ + // TODO(riscv): use F Extension macro block + case RO_FSW: { + WriteMem<Float32>(rs1() + s_imm12(), getFpuRegisterFloat32(rs2_reg()), + instr_.instr()); + break; + } + // TODO(riscv): use D Extension macro block + case RO_FSD: { + WriteMem<Float64>(rs1() + s_imm12(), getFpuRegisterFloat64(rs2_reg()), + instr_.instr()); + break; + } + default: +# ifdef CAN_USE_RVV_INSTRUCTIONS + if (!DecodeRvvVS()) { + UNSUPPORTED(); + } + break; +# else + UNSUPPORTED(); +# endif + } +} + +void Simulator::DecodeRVBType() { + switch (instr_.InstructionBits() & kBTypeMask) { + case RO_BEQ: + if (rs1() == rs2()) { + int64_t next_pc = get_pc() + boffset(); + set_pc(next_pc); + } + break; + case RO_BNE: + if (rs1() != rs2()) { + int64_t next_pc = get_pc() + boffset(); + set_pc(next_pc); + } + break; + case RO_BLT: + if (rs1() < rs2()) { + int64_t next_pc = get_pc() + boffset(); + set_pc(next_pc); + } + break; + case RO_BGE: + if (rs1() >= rs2()) { + int64_t next_pc = get_pc() + boffset(); + set_pc(next_pc); + } + break; + case RO_BLTU: + if ((reg_t)rs1() < (reg_t)rs2()) { + int64_t next_pc = get_pc() + boffset(); + set_pc(next_pc); + } + break; + case RO_BGEU: + if ((reg_t)rs1() >= (reg_t)rs2()) { + int64_t next_pc = get_pc() + boffset(); + set_pc(next_pc); + } + break; + default: + UNSUPPORTED(); + } +} +void Simulator::DecodeRVUType() { + // U Type doesn't have additoinal mask + switch (instr_.BaseOpcodeFieldRaw()) { + case LUI: + set_rd(u_imm20()); + break; + case AUIPC: + set_rd(sext_xlen(u_imm20() + get_pc())); + break; + default: + UNSUPPORTED(); + } +} +void Simulator::DecodeRVJType() { + // J Type doesn't have additional mask + switch (instr_.BaseOpcodeValue()) { + case JAL: { + set_rd(get_pc() + kInstrSize); + int64_t next_pc = get_pc() + imm20J(); + set_pc(next_pc); + break; + } + default: + UNSUPPORTED(); + } +} +void Simulator::DecodeCRType() { + switch (instr_.RvcFunct4Value()) { + case 0b1000: + if (instr_.RvcRs1Value() != 0 && instr_.RvcRs2Value() == 0) { // c.jr + set_pc(rvc_rs1()); + } else if (instr_.RvcRdValue() != 0 && + instr_.RvcRs2Value() != 0) { // c.mv + set_rvc_rd(sext_xlen(rvc_rs2())); + } else { + UNSUPPORTED(); + } + break; + case 0b1001: + if (instr_.RvcRs1Value() == 0 && instr_.RvcRs2Value() == 0) { // c.ebreak + DieOrDebug(); + } else if (instr_.RvcRdValue() != 0 && + instr_.RvcRs2Value() == 0) { // c.jalr + setRegister(ra, get_pc() + kShortInstrSize); + set_pc(rvc_rs1()); + } else if (instr_.RvcRdValue() != 0 && + instr_.RvcRs2Value() != 0) { // c.add + set_rvc_rd(sext_xlen(rvc_rs1() + rvc_rs2())); + } else { + UNSUPPORTED(); + } + break; + default: + UNSUPPORTED(); + } +} + +void Simulator::DecodeCAType() { + switch (instr_.InstructionBits() & kCATypeMask) { + case RO_C_SUB: + set_rvc_rs1s(sext_xlen(rvc_rs1s() - rvc_rs2s())); + break; + case RO_C_XOR: + set_rvc_rs1s(rvc_rs1s() ^ rvc_rs2s()); + break; + case RO_C_OR: + set_rvc_rs1s(rvc_rs1s() | rvc_rs2s()); + break; + case RO_C_AND: + set_rvc_rs1s(rvc_rs1s() & rvc_rs2s()); + break; +# if JS_CODEGEN_RISCV64 + case RO_C_SUBW: + set_rvc_rs1s(sext32(rvc_rs1s() - rvc_rs2s())); + break; + case RO_C_ADDW: + set_rvc_rs1s(sext32(rvc_rs1s() + rvc_rs2s())); + break; +# endif + default: + UNSUPPORTED(); + } +} + +void Simulator::DecodeCIType() { + switch (instr_.RvcOpcode()) { + case RO_C_NOP_ADDI: + if (instr_.RvcRdValue() == 0) // c.nop + break; + else // c.addi + set_rvc_rd(sext_xlen(rvc_rs1() + rvc_imm6())); + break; +# if JS_CODEGEN_RISCV64 + case RO_C_ADDIW: + set_rvc_rd(sext32(rvc_rs1() + rvc_imm6())); + break; +# endif + case RO_C_LI: + set_rvc_rd(sext_xlen(rvc_imm6())); + break; + case RO_C_LUI_ADD: + if (instr_.RvcRdValue() == 2) { + // c.addi16sp + int64_t value = getRegister(sp) + rvc_imm6_addi16sp(); + setRegister(sp, value); + } else if (instr_.RvcRdValue() != 0 && instr_.RvcRdValue() != 2) { + // c.lui + set_rvc_rd(rvc_u_imm6()); + } else { + UNSUPPORTED(); + } + break; + case RO_C_SLLI: + set_rvc_rd(sext_xlen(rvc_rs1() << rvc_shamt6())); + break; + case RO_C_FLDSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_ldsp(); + uint64_t val = ReadMem<uint64_t>(addr, instr_.instr()); + set_rvc_drd(Float64::FromBits(val), false); + TraceMemRdDouble(addr, Float64::FromBits(val), + getFpuRegister(rvc_frd_reg())); + break; + } +# if JS_CODEGEN_RISCV64 + case RO_C_LWSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_lwsp(); + int64_t val = ReadMem<int32_t>(addr, instr_.instr()); + set_rvc_rd(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rvc_rd_reg())); + break; + } + case RO_C_LDSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_ldsp(); + int64_t val = ReadMem<int64_t>(addr, instr_.instr()); + set_rvc_rd(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rvc_rd_reg())); + break; + } +# elif JS_CODEGEN_RISCV32 + case RO_C_FLWSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_ldsp(); + uint32_t val = ReadMem<uint32_t>(addr, instr_.instr()); + set_rvc_frd(Float32::FromBits(val), false); + TraceMemRdFloat(addr, Float32::FromBits(val), + getFpuRegister(rvc_frd_reg())); + break; + } + case RO_C_LWSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_lwsp(); + int32_t val = ReadMem<int32_t>(addr, instr_.instr()); + set_rvc_rd(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rvc_rd_reg())); + break; + } +# endif + default: + UNSUPPORTED(); + } +} + +void Simulator::DecodeCIWType() { + switch (instr_.RvcOpcode()) { + case RO_C_ADDI4SPN: { + set_rvc_rs2s(getRegister(sp) + rvc_imm8_addi4spn()); + break; + default: + UNSUPPORTED(); + } + } +} + +void Simulator::DecodeCSSType() { + switch (instr_.RvcOpcode()) { + case RO_C_FSDSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_sdsp(); + WriteMem<Float64>(addr, getFpuRegisterFloat64(rvc_rs2_reg()), + instr_.instr()); + break; + } +# if JS_CODEGEN_RISCV32 + case RO_C_FSWSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_sdsp(); + WriteMem<Float32>(addr, getFpuRegisterFloat32(rvc_rs2_reg()), + instr_.instr()); + break; + } +# endif + case RO_C_SWSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_swsp(); + WriteMem<int32_t>(addr, (int32_t)rvc_rs2(), instr_.instr()); + break; + } +# if JS_CODEGEN_RISCV64 + case RO_C_SDSP: { + sreg_t addr = getRegister(sp) + rvc_imm6_sdsp(); + WriteMem<int64_t>(addr, (int64_t)rvc_rs2(), instr_.instr()); + break; + } +# endif + default: + UNSUPPORTED(); + } +} + +void Simulator::DecodeCLType() { + switch (instr_.RvcOpcode()) { + case RO_C_LW: { + sreg_t addr = rvc_rs1s() + rvc_imm5_w(); + int64_t val = ReadMem<int32_t>(addr, instr_.instr()); + set_rvc_rs2s(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rvc_rs2s_reg())); + break; + } + case RO_C_FLD: { + sreg_t addr = rvc_rs1s() + rvc_imm5_d(); + uint64_t val = ReadMem<uint64_t>(addr, instr_.instr()); + set_rvc_drs2s(Float64::FromBits(val), false); + break; + } +# if JS_CODEGEN_RISCV64 + case RO_C_LD: { + sreg_t addr = rvc_rs1s() + rvc_imm5_d(); + int64_t val = ReadMem<int64_t>(addr, instr_.instr()); + set_rvc_rs2s(sext_xlen(val), false); + TraceMemRd(addr, val, getRegister(rvc_rs2s_reg())); + break; + } +# elif JS_CODEGEN_RISCV32 + case RO_C_FLW: { + sreg_t addr = rvc_rs1s() + rvc_imm5_d(); + uint32_t val = ReadMem<uint32_t>(addr, instr_.instr()); + set_rvc_frs2s(Float32::FromBits(val), false); + break; + } +# endif + default: + UNSUPPORTED(); + } +} + +void Simulator::DecodeCSType() { + switch (instr_.RvcOpcode()) { + case RO_C_SW: { + sreg_t addr = rvc_rs1s() + rvc_imm5_w(); + WriteMem<int32_t>(addr, (int32_t)rvc_rs2s(), instr_.instr()); + break; + } +# if JS_CODEGEN_RISCV64 + case RO_C_SD: { + sreg_t addr = rvc_rs1s() + rvc_imm5_d(); + WriteMem<int64_t>(addr, (int64_t)rvc_rs2s(), instr_.instr()); + break; + } +# endif + case RO_C_FSD: { + sreg_t addr = rvc_rs1s() + rvc_imm5_d(); + WriteMem<double>(addr, static_cast<double>(rvc_drs2s()), instr_.instr()); + break; + } + default: + UNSUPPORTED(); + } +} + +void Simulator::DecodeCJType() { + switch (instr_.RvcOpcode()) { + case RO_C_J: { + set_pc(get_pc() + instr_.RvcImm11CJValue()); + break; + } + default: + UNSUPPORTED(); + } +} + +void Simulator::DecodeCBType() { + switch (instr_.RvcOpcode()) { + case RO_C_BNEZ: + if (rvc_rs1() != 0) { + sreg_t next_pc = get_pc() + rvc_imm8_b(); + set_pc(next_pc); + } + break; + case RO_C_BEQZ: + if (rvc_rs1() == 0) { + sreg_t next_pc = get_pc() + rvc_imm8_b(); + set_pc(next_pc); + } + break; + case RO_C_MISC_ALU: + if (instr_.RvcFunct2BValue() == 0b00) { // c.srli + set_rvc_rs1s(sext_xlen(sext_xlen(rvc_rs1s()) >> rvc_shamt6())); + } else if (instr_.RvcFunct2BValue() == 0b01) { // c.srai + require(rvc_shamt6() < xlen); + set_rvc_rs1s(sext_xlen(sext_xlen(rvc_rs1s()) >> rvc_shamt6())); + } else if (instr_.RvcFunct2BValue() == 0b10) { // c.andi + set_rvc_rs1s(rvc_imm6() & rvc_rs1s()); + } else { + UNSUPPORTED(); + } + break; + default: + UNSUPPORTED(); + } +} + +void Simulator::callInternal(uint8_t* entry) { + // Prepare to execute the code at entry. + setRegister(pc, reinterpret_cast<int64_t>(entry)); + // Put down marker for end of simulation. The simulator will stop simulation + // when the PC reaches this value. By saving the "end simulation" value into + // the LR the simulation stops when returning to this call point. + setRegister(ra, end_sim_pc); + // Remember the values of callee-saved registers. + intptr_t s0_val = getRegister(Simulator::Register::fp); + intptr_t s1_val = getRegister(Simulator::Register::s1); + intptr_t s2_val = getRegister(Simulator::Register::s2); + intptr_t s3_val = getRegister(Simulator::Register::s3); + intptr_t s4_val = getRegister(Simulator::Register::s4); + intptr_t s5_val = getRegister(Simulator::Register::s5); + intptr_t s6_val = getRegister(Simulator::Register::s6); + intptr_t s7_val = getRegister(Simulator::Register::s7); + intptr_t s8_val = getRegister(Simulator::Register::s8); + intptr_t s9_val = getRegister(Simulator::Register::s9); + intptr_t s10_val = getRegister(Simulator::Register::s10); + intptr_t s11_val = getRegister(Simulator::Register::s11); + intptr_t gp_val = getRegister(Simulator::Register::gp); + intptr_t sp_val = getRegister(Simulator::Register::sp); + + // Set up the callee-saved registers with a known value. To be able to check + // that they are preserved properly across JS execution. If this value is + // small int, it should be SMI. + intptr_t callee_saved_value = icount_; + setRegister(Simulator::Register::fp, callee_saved_value); + setRegister(Simulator::Register::s1, callee_saved_value); + setRegister(Simulator::Register::s2, callee_saved_value); + setRegister(Simulator::Register::s3, callee_saved_value); + setRegister(Simulator::Register::s4, callee_saved_value); + setRegister(Simulator::Register::s5, callee_saved_value); + setRegister(Simulator::Register::s6, callee_saved_value); + setRegister(Simulator::Register::s7, callee_saved_value); + setRegister(Simulator::Register::s8, callee_saved_value); + setRegister(Simulator::Register::s9, callee_saved_value); + setRegister(Simulator::Register::s10, callee_saved_value); + setRegister(Simulator::Register::s11, callee_saved_value); + setRegister(Simulator::Register::gp, callee_saved_value); + + // Start the simulation. + if (Simulator::StopSimAt != -1) { + execute<true>(); + } else { + execute<false>(); + } + + // Check that the callee-saved registers have been preserved. + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::fp)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s1)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s2)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s3)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s4)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s5)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s6)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s7)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s8)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s9)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s10)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::s11)); + MOZ_ASSERT(callee_saved_value == getRegister(Simulator::Register::gp)); + + // Restore callee-saved registers with the original value. + setRegister(Simulator::Register::fp, s0_val); + setRegister(Simulator::Register::s1, s1_val); + setRegister(Simulator::Register::s2, s2_val); + setRegister(Simulator::Register::s3, s3_val); + setRegister(Simulator::Register::s4, s4_val); + setRegister(Simulator::Register::s5, s5_val); + setRegister(Simulator::Register::s6, s6_val); + setRegister(Simulator::Register::s7, s7_val); + setRegister(Simulator::Register::s8, s8_val); + setRegister(Simulator::Register::s9, s9_val); + setRegister(Simulator::Register::s10, s10_val); + setRegister(Simulator::Register::s11, s11_val); + setRegister(Simulator::Register::gp, gp_val); + setRegister(Simulator::Register::sp, sp_val); +} + +int64_t Simulator::call(uint8_t* entry, int argument_count, ...) { + va_list parameters; + va_start(parameters, argument_count); + + int64_t original_stack = getRegister(sp); + // Compute position of stack on entry to generated code. + int64_t entry_stack = original_stack; + if (argument_count > kCArgSlotCount) { + entry_stack = entry_stack - argument_count * sizeof(int64_t); + } else { + entry_stack = entry_stack - kCArgsSlotsSize; + } + + entry_stack &= ~U64(ABIStackAlignment - 1); + + intptr_t* stack_argument = reinterpret_cast<intptr_t*>(entry_stack); + + // Setup the arguments. + for (int i = 0; i < argument_count; i++) { + js::jit::Register argReg; + if (GetIntArgReg(i, &argReg)) { + setRegister(argReg.code(), va_arg(parameters, int64_t)); + } else { + stack_argument[i] = va_arg(parameters, int64_t); + } + } + + va_end(parameters); + setRegister(sp, entry_stack); + + callInternal(entry); + + // Pop stack passed arguments. + MOZ_ASSERT(entry_stack == getRegister(sp)); + setRegister(sp, original_stack); + + int64_t result = getRegister(a0); + return result; +} + +uintptr_t Simulator::pushAddress(uintptr_t address) { + int new_sp = getRegister(sp) - sizeof(uintptr_t); + uintptr_t* stack_slot = reinterpret_cast<uintptr_t*>(new_sp); + *stack_slot = address; + setRegister(sp, new_sp); + return new_sp; +} + +uintptr_t Simulator::popAddress() { + int current_sp = getRegister(sp); + uintptr_t* stack_slot = reinterpret_cast<uintptr_t*>(current_sp); + uintptr_t address = *stack_slot; + setRegister(sp, current_sp + sizeof(uintptr_t)); + return address; +} + +} // namespace jit +} // namespace js + +js::jit::Simulator* JSContext::simulator() const { return simulator_; } + +#endif // JS_SIMULATOR_RISCV64 diff --git a/js/src/jit/riscv64/Simulator-riscv64.h b/js/src/jit/riscv64/Simulator-riscv64.h new file mode 100644 index 0000000000..1a0a1636c3 --- /dev/null +++ b/js/src/jit/riscv64/Simulator-riscv64.h @@ -0,0 +1,1282 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: */ +// Copyright 2021 the V8 project authors. All rights reserved. +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * Redistributions in binary form must reproduce the above +// copyright notice, this list of conditions and the following +// disclaimer in the documentation and/or other materials provided +// with the distribution. +// * Neither the name of Google Inc. nor the names of its +// contributors may be used to endorse or promote products derived +// from this software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT +// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, +// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT +// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, +// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY +// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +#ifndef jit_riscv64_Simulator_riscv64_h +#define jit_riscv64_Simulator_riscv64_h + +#ifdef JS_SIMULATOR_RISCV64 +# include "mozilla/Atomics.h" + +# include <vector> + +# include "jit/IonTypes.h" +# include "jit/riscv64/constant/Constant-riscv64.h" +# include "jit/riscv64/constant/util-riscv64.h" +# include "jit/riscv64/disasm/Disasm-riscv64.h" +# include "js/ProfilingFrameIterator.h" +# include "threading/Thread.h" +# include "vm/MutexIDs.h" +# include "wasm/WasmSignalHandlers.h" + +namespace js { + +namespace jit { + +template <class Dest, class Source> +inline Dest bit_cast(const Source& source) { + static_assert(sizeof(Dest) == sizeof(Source), + "bit_cast requires source and destination to be the same size"); + static_assert(std::is_trivially_copyable<Dest>::value, + "bit_cast requires the destination type to be copyable"); + static_assert(std::is_trivially_copyable<Source>::value, + "bit_cast requires the source type to be copyable"); + + Dest dest; + memcpy(&dest, &source, sizeof(dest)); + return dest; +} + +# define ASSERT_TRIVIALLY_COPYABLE(T) \ + static_assert(std::is_trivially_copyable<T>::value, \ + #T " should be trivially copyable") +# define ASSERT_NOT_TRIVIALLY_COPYABLE(T) \ + static_assert(!std::is_trivially_copyable<T>::value, \ + #T " should not be trivially copyable") + +constexpr uint32_t kHoleNanUpper32 = 0xFFF7FFFF; +constexpr uint32_t kHoleNanLower32 = 0xFFF7FFFF; + +constexpr uint64_t kHoleNanInt64 = + (static_cast<uint64_t>(kHoleNanUpper32) << 32) | kHoleNanLower32; +// Safety wrapper for a 32-bit floating-point value to make sure we don't lose +// the exact bit pattern during deoptimization when passing this value. +class Float32 { + public: + Float32() = default; + + // This constructor does not guarantee that bit pattern of the input value + // is preserved if the input is a NaN. + explicit Float32(float value) : bit_pattern_(bit_cast<uint32_t>(value)) { + // Check that the provided value is not a NaN, because the bit pattern of a + // NaN may be changed by a bit_cast, e.g. for signalling NaNs on + // ia32. + MOZ_ASSERT(!std::isnan(value)); + } + + uint32_t get_bits() const { return bit_pattern_; } + + float get_scalar() const { return bit_cast<float>(bit_pattern_); } + + bool is_nan() const { + // Even though {get_scalar()} might flip the quiet NaN bit, it's ok here, + // because this does not change the is_nan property. + return std::isnan(get_scalar()); + } + + // Return a pointer to the field storing the bit pattern. Used in code + // generation tests to store generated values there directly. + uint32_t* get_bits_address() { return &bit_pattern_; } + + static constexpr Float32 FromBits(uint32_t bits) { return Float32(bits); } + + private: + uint32_t bit_pattern_ = 0; + + explicit constexpr Float32(uint32_t bit_pattern) + : bit_pattern_(bit_pattern) {} +}; + +ASSERT_TRIVIALLY_COPYABLE(Float32); + +// Safety wrapper for a 64-bit floating-point value to make sure we don't lose +// the exact bit pattern during deoptimization when passing this value. +// TODO(ahaas): Unify this class with Double in double.h +class Float64 { + public: + Float64() = default; + + // This constructor does not guarantee that bit pattern of the input value + // is preserved if the input is a NaN. + explicit Float64(double value) : bit_pattern_(bit_cast<uint64_t>(value)) { + // Check that the provided value is not a NaN, because the bit pattern of a + // NaN may be changed by a bit_cast, e.g. for signalling NaNs on + // ia32. + MOZ_ASSERT(!std::isnan(value)); + } + + uint64_t get_bits() const { return bit_pattern_; } + double get_scalar() const { return bit_cast<double>(bit_pattern_); } + bool is_hole_nan() const { return bit_pattern_ == kHoleNanInt64; } + bool is_nan() const { + // Even though {get_scalar()} might flip the quiet NaN bit, it's ok here, + // because this does not change the is_nan property. + return std::isnan(get_scalar()); + } + + // Return a pointer to the field storing the bit pattern. Used in code + // generation tests to store generated values there directly. + uint64_t* get_bits_address() { return &bit_pattern_; } + + static constexpr Float64 FromBits(uint64_t bits) { return Float64(bits); } + + private: + uint64_t bit_pattern_ = 0; + + explicit constexpr Float64(uint64_t bit_pattern) + : bit_pattern_(bit_pattern) {} +}; + +ASSERT_TRIVIALLY_COPYABLE(Float64); + +class JitActivation; + +class Simulator; +class Redirection; +class CachePage; +class AutoLockSimulator; + +// When the SingleStepCallback is called, the simulator is about to execute +// sim->get_pc() and the current machine state represents the completed +// execution of the previous pc. +typedef void (*SingleStepCallback)(void* arg, Simulator* sim, void* pc); + +const intptr_t kPointerAlignment = 8; +const intptr_t kPointerAlignmentMask = kPointerAlignment - 1; + +const intptr_t kDoubleAlignment = 8; +const intptr_t kDoubleAlignmentMask = kDoubleAlignment - 1; + +// Number of general purpose registers. +const int kNumRegisters = 32; + +// In the simulator, the PC register is simulated as the 34th register. +const int kPCRegister = 32; + +// Number coprocessor registers. +const int kNumFPURegisters = 32; + +// FPU (coprocessor 1) control registers. Currently only FCSR is implemented. +const int kFCSRRegister = 31; +const int kInvalidFPUControlRegister = -1; +const uint32_t kFPUInvalidResult = static_cast<uint32_t>(1 << 31) - 1; +const uint64_t kFPUInvalidResult64 = static_cast<uint64_t>(1ULL << 63) - 1; + +// FCSR constants. +const uint32_t kFCSRInexactFlagBit = 2; +const uint32_t kFCSRUnderflowFlagBit = 3; +const uint32_t kFCSROverflowFlagBit = 4; +const uint32_t kFCSRDivideByZeroFlagBit = 5; +const uint32_t kFCSRInvalidOpFlagBit = 6; + +const uint32_t kFCSRInexactCauseBit = 12; +const uint32_t kFCSRUnderflowCauseBit = 13; +const uint32_t kFCSROverflowCauseBit = 14; +const uint32_t kFCSRDivideByZeroCauseBit = 15; +const uint32_t kFCSRInvalidOpCauseBit = 16; + +const uint32_t kFCSRInexactFlagMask = 1 << kFCSRInexactFlagBit; +const uint32_t kFCSRUnderflowFlagMask = 1 << kFCSRUnderflowFlagBit; +const uint32_t kFCSROverflowFlagMask = 1 << kFCSROverflowFlagBit; +const uint32_t kFCSRDivideByZeroFlagMask = 1 << kFCSRDivideByZeroFlagBit; +const uint32_t kFCSRInvalidOpFlagMask = 1 << kFCSRInvalidOpFlagBit; + +const uint32_t kFCSRFlagMask = + kFCSRInexactFlagMask | kFCSRUnderflowFlagMask | kFCSROverflowFlagMask | + kFCSRDivideByZeroFlagMask | kFCSRInvalidOpFlagMask; + +const uint32_t kFCSRExceptionFlagMask = kFCSRFlagMask ^ kFCSRInexactFlagMask; + +// ----------------------------------------------------------------------------- +// Utility types and functions for RISCV +# ifdef JS_CODEGEN_RISCV32 +using sreg_t = int32_t; +using reg_t = uint32_t; +using freg_t = uint64_t; +using sfreg_t = int64_t; +# elif JS_CODEGEN_RISCV64 +using sreg_t = int64_t; +using reg_t = uint64_t; +using freg_t = uint64_t; +using sfreg_t = int64_t; +# else +# error "Cannot detect Riscv's bitwidth" +# endif + +# define sext32(x) ((sreg_t)(int32_t)(x)) +# define zext32(x) ((reg_t)(uint32_t)(x)) + +# ifdef JS_CODEGEN_RISCV64 +# define sext_xlen(x) (((sreg_t)(x) << (64 - xlen)) >> (64 - xlen)) +# define zext_xlen(x) (((reg_t)(x) << (64 - xlen)) >> (64 - xlen)) +# elif JS_CODEGEN_RISCV32 +# define sext_xlen(x) (((sreg_t)(x) << (32 - xlen)) >> (32 - xlen)) +# define zext_xlen(x) (((reg_t)(x) << (32 - xlen)) >> (32 - xlen)) +# endif + +# define BIT(n) (0x1LL << n) +# define QUIET_BIT_S(nan) (bit_cast<int32_t>(nan) & BIT(22)) +# define QUIET_BIT_D(nan) (bit_cast<int64_t>(nan) & BIT(51)) +static inline bool isSnan(float fp) { return !QUIET_BIT_S(fp); } +static inline bool isSnan(double fp) { return !QUIET_BIT_D(fp); } +# undef QUIET_BIT_S +# undef QUIET_BIT_D + +# ifdef JS_CODEGEN_RISCV64 +inline uint64_t mulhu(uint64_t a, uint64_t b) { + __uint128_t full_result = ((__uint128_t)a) * ((__uint128_t)b); + return full_result >> 64; +} + +inline int64_t mulh(int64_t a, int64_t b) { + __int128_t full_result = ((__int128_t)a) * ((__int128_t)b); + return full_result >> 64; +} + +inline int64_t mulhsu(int64_t a, uint64_t b) { + __int128_t full_result = ((__int128_t)a) * ((__uint128_t)b); + return full_result >> 64; +} +# elif JS_CODEGEN_RISCV32 +inline uint32_t mulhu(uint32_t a, uint32_t b) { + uint64_t full_result = ((uint64_t)a) * ((uint64_t)b); + uint64_t upper_part = full_result >> 32; + return (uint32_t)upper_part; +} + +inline int32_t mulh(int32_t a, int32_t b) { + int64_t full_result = ((int64_t)a) * ((int64_t)b); + int64_t upper_part = full_result >> 32; + return (int32_t)upper_part; +} + +inline int32_t mulhsu(int32_t a, uint32_t b) { + int64_t full_result = ((int64_t)a) * ((uint64_t)b); + int64_t upper_part = full_result >> 32; + return (int32_t)upper_part; +} +# endif + +// Floating point helpers +# define F32_SIGN ((uint32_t)1 << 31) +union u32_f32 { + uint32_t u; + float f; +}; +inline float fsgnj32(float rs1, float rs2, bool n, bool x) { + u32_f32 a = {.f = rs1}, b = {.f = rs2}; + u32_f32 res; + res.u = (a.u & ~F32_SIGN) | ((((x) ? a.u + : (n) ? F32_SIGN + : 0) ^ + b.u) & + F32_SIGN); + return res.f; +} + +inline Float32 fsgnj32(Float32 rs1, Float32 rs2, bool n, bool x) { + u32_f32 a = {.u = rs1.get_bits()}, b = {.u = rs2.get_bits()}; + u32_f32 res; + if (x) { // RO_FSQNJX_S + res.u = (a.u & ~F32_SIGN) | ((a.u ^ b.u) & F32_SIGN); + } else { + if (n) { // RO_FSGNJN_S + res.u = (a.u & ~F32_SIGN) | ((F32_SIGN ^ b.u) & F32_SIGN); + } else { // RO_FSGNJ_S + res.u = (a.u & ~F32_SIGN) | ((0 ^ b.u) & F32_SIGN); + } + } + return Float32::FromBits(res.u); +} +# define F64_SIGN ((uint64_t)1 << 63) +union u64_f64 { + uint64_t u; + double d; +}; +inline double fsgnj64(double rs1, double rs2, bool n, bool x) { + u64_f64 a = {.d = rs1}, b = {.d = rs2}; + u64_f64 res; + res.u = (a.u & ~F64_SIGN) | ((((x) ? a.u + : (n) ? F64_SIGN + : 0) ^ + b.u) & + F64_SIGN); + return res.d; +} + +inline Float64 fsgnj64(Float64 rs1, Float64 rs2, bool n, bool x) { + u64_f64 a = {.d = rs1.get_scalar()}, b = {.d = rs2.get_scalar()}; + u64_f64 res; + if (x) { // RO_FSQNJX_D + res.u = (a.u & ~F64_SIGN) | ((a.u ^ b.u) & F64_SIGN); + } else { + if (n) { // RO_FSGNJN_D + res.u = (a.u & ~F64_SIGN) | ((F64_SIGN ^ b.u) & F64_SIGN); + } else { // RO_FSGNJ_D + res.u = (a.u & ~F64_SIGN) | ((0 ^ b.u) & F64_SIGN); + } + } + return Float64::FromBits(res.u); +} +inline bool is_boxed_float(int64_t v) { return (uint32_t)((v >> 32) + 1) == 0; } +inline int64_t box_float(float v) { + return (0xFFFFFFFF00000000 | bit_cast<int32_t>(v)); +} + +inline uint64_t box_float(uint32_t v) { return (0xFFFFFFFF00000000 | v); } + +// ----------------------------------------------------------------------------- +// Utility functions + +class SimInstructionBase : public InstructionBase { + public: + Type InstructionType() const { return type_; } + inline Instruction* instr() const { return instr_; } + inline int32_t operand() const { return operand_; } + + protected: + SimInstructionBase() : operand_(-1), instr_(nullptr), type_(kUnsupported) {} + explicit SimInstructionBase(Instruction* instr) {} + + int32_t operand_; + Instruction* instr_; + Type type_; + + private: + SimInstructionBase& operator=(const SimInstructionBase&) = delete; +}; + +class SimInstruction : public InstructionGetters<SimInstructionBase> { + public: + SimInstruction() {} + + explicit SimInstruction(Instruction* instr) { *this = instr; } + + SimInstruction& operator=(Instruction* instr) { + operand_ = *reinterpret_cast<const int32_t*>(instr); + instr_ = instr; + type_ = InstructionBase::InstructionType(); + MOZ_ASSERT(reinterpret_cast<void*>(&operand_) == this); + return *this; + } +}; + +// Per thread simulator state. +class Simulator { + friend class RiscvDebugger; + + public: + static bool FLAG_riscv_trap_to_simulator_debugger; + static bool FLAG_trace_sim; + static bool FLAG_debug_sim; + static bool FLAG_riscv_print_watchpoint; + // Registers are declared in order. + enum Register { + no_reg = -1, + x0 = 0, + x1, + x2, + x3, + x4, + x5, + x6, + x7, + x8, + x9, + x10, + x11, + x12, + x13, + x14, + x15, + x16, + x17, + x18, + x19, + x20, + x21, + x22, + x23, + x24, + x25, + x26, + x27, + x28, + x29, + x30, + x31, + pc, + kNumSimuRegisters, + // alias + zero = x0, + ra = x1, + sp = x2, + gp = x3, + tp = x4, + t0 = x5, + t1 = x6, + t2 = x7, + fp = x8, + s1 = x9, + a0 = x10, + a1 = x11, + a2 = x12, + a3 = x13, + a4 = x14, + a5 = x15, + a6 = x16, + a7 = x17, + s2 = x18, + s3 = x19, + s4 = x20, + s5 = x21, + s6 = x22, + s7 = x23, + s8 = x24, + s9 = x25, + s10 = x26, + s11 = x27, + t3 = x28, + t4 = x29, + t5 = x30, + t6 = x31, + }; + + // Coprocessor registers. + enum FPURegister { + f0, + f1, + f2, + f3, + f4, + f5, + f6, + f7, + f8, + f9, + f10, + f11, + f12, + f13, + f14, + f15, + f16, + f17, + f18, + f19, + f20, + f21, + f22, + f23, + f24, + f25, + f26, + f27, + f28, + f29, + f30, + f31, + kNumFPURegisters, + // alias + ft0 = f0, + ft1 = f1, + ft2 = f2, + ft3 = f3, + ft4 = f4, + ft5 = f5, + ft6 = f6, + ft7 = f7, + fs0 = f8, + fs1 = f9, + fa0 = f10, + fa1 = f11, + fa2 = f12, + fa3 = f13, + fa4 = f14, + fa5 = f15, + fa6 = f16, + fa7 = f17, + fs2 = f18, + fs3 = f19, + fs4 = f20, + fs5 = f21, + fs6 = f22, + fs7 = f23, + fs8 = f24, + fs9 = f25, + fs10 = f26, + fs11 = f27, + ft8 = f28, + ft9 = f29, + ft10 = f30, + ft11 = f31 + }; + + // Returns nullptr on OOM. + static Simulator* Create(); + + static void Destroy(Simulator* simulator); + + // Constructor/destructor are for internal use only; use the static methods + // above. + Simulator(); + ~Simulator(); + + // RISCV decoding routine + void DecodeRVRType(); + void DecodeRVR4Type(); + void DecodeRVRFPType(); // Special routine for R/OP_FP type + void DecodeRVRAType(); // Special routine for R/AMO type + void DecodeRVIType(); + void DecodeRVSType(); + void DecodeRVBType(); + void DecodeRVUType(); + void DecodeRVJType(); + void DecodeCRType(); + void DecodeCAType(); + void DecodeCIType(); + void DecodeCIWType(); + void DecodeCSSType(); + void DecodeCLType(); + void DecodeCSType(); + void DecodeCJType(); + void DecodeCBType(); +# ifdef CAN_USE_RVV_INSTRUCTIONS + void DecodeVType(); + void DecodeRvvIVV(); + void DecodeRvvIVI(); + void DecodeRvvIVX(); + void DecodeRvvMVV(); + void DecodeRvvMVX(); + void DecodeRvvFVV(); + void DecodeRvvFVF(); + bool DecodeRvvVL(); + bool DecodeRvvVS(); +# endif + // The currently executing Simulator instance. Potentially there can be one + // for each native thread. + static Simulator* Current(); + + static inline uintptr_t StackLimit() { + return Simulator::Current()->stackLimit(); + } + + uintptr_t* addressOfStackLimit(); + + // Accessors for register state. Reading the pc value adheres to the MIPS + // architecture specification and is off by a 8 from the currently executing + // instruction. + void setRegister(int reg, int64_t value); + int64_t getRegister(int reg) const; + // Same for FPURegisters. + void setFpuRegister(int fpureg, int64_t value); + void setFpuRegisterLo(int fpureg, int32_t value); + void setFpuRegisterHi(int fpureg, int32_t value); + void setFpuRegisterFloat(int fpureg, float value); + void setFpuRegisterDouble(int fpureg, double value); + void setFpuRegisterFloat(int fpureg, Float32 value); + void setFpuRegisterDouble(int fpureg, Float64 value); + + int64_t getFpuRegister(int fpureg) const; + int32_t getFpuRegisterLo(int fpureg) const; + int32_t getFpuRegisterHi(int fpureg) const; + float getFpuRegisterFloat(int fpureg) const; + double getFpuRegisterDouble(int fpureg) const; + Float32 getFpuRegisterFloat32(int fpureg) const; + Float64 getFpuRegisterFloat64(int fpureg) const; + + inline int16_t shamt6() const { return (imm12() & 0x3F); } + inline int16_t shamt5() const { return (imm12() & 0x1F); } + inline int16_t rvc_shamt6() const { return instr_.RvcShamt6(); } + inline int32_t s_imm12() const { return instr_.StoreOffset(); } + inline int32_t u_imm20() const { return instr_.Imm20UValue() << 12; } + inline int32_t rvc_u_imm6() const { return instr_.RvcImm6Value() << 12; } + inline void require(bool check) { + if (!check) { + SignalException(kIllegalInstruction); + } + } + + // Special case of setRegister and getRegister to access the raw PC value. + void set_pc(int64_t value); + int64_t get_pc() const; + + SimInstruction instr_; + // RISCV utlity API to access register value + // Helpers for data value tracing. + enum TraceType { + BYTE, + HALF, + WORD, +# if JS_CODEGEN_RISCV64 + DWORD, +# endif + FLOAT, + DOUBLE, + // FLOAT_DOUBLE, + // WORD_DWORD + }; + inline int32_t rs1_reg() const { return instr_.Rs1Value(); } + inline sreg_t rs1() const { return getRegister(rs1_reg()); } + inline float frs1() const { return getFpuRegisterFloat(rs1_reg()); } + inline double drs1() const { return getFpuRegisterDouble(rs1_reg()); } + inline Float32 frs1_boxed() const { return getFpuRegisterFloat32(rs1_reg()); } + inline Float64 drs1_boxed() const { return getFpuRegisterFloat64(rs1_reg()); } + inline int32_t rs2_reg() const { return instr_.Rs2Value(); } + inline sreg_t rs2() const { return getRegister(rs2_reg()); } + inline float frs2() const { return getFpuRegisterFloat(rs2_reg()); } + inline double drs2() const { return getFpuRegisterDouble(rs2_reg()); } + inline Float32 frs2_boxed() const { return getFpuRegisterFloat32(rs2_reg()); } + inline Float64 drs2_boxed() const { return getFpuRegisterFloat64(rs2_reg()); } + inline int32_t rs3_reg() const { return instr_.Rs3Value(); } + inline sreg_t rs3() const { return getRegister(rs3_reg()); } + inline float frs3() const { return getFpuRegisterFloat(rs3_reg()); } + inline double drs3() const { return getFpuRegisterDouble(rs3_reg()); } + inline Float32 frs3_boxed() const { return getFpuRegisterFloat32(rs3_reg()); } + inline Float64 drs3_boxed() const { return getFpuRegisterFloat64(rs3_reg()); } + inline int32_t rd_reg() const { return instr_.RdValue(); } + inline int32_t frd_reg() const { return instr_.RdValue(); } + inline int32_t rvc_rs1_reg() const { return instr_.RvcRs1Value(); } + inline sreg_t rvc_rs1() const { return getRegister(rvc_rs1_reg()); } + inline int32_t rvc_rs2_reg() const { return instr_.RvcRs2Value(); } + inline sreg_t rvc_rs2() const { return getRegister(rvc_rs2_reg()); } + inline double rvc_drs2() const { return getFpuRegisterDouble(rvc_rs2_reg()); } + inline int32_t rvc_rs1s_reg() const { return instr_.RvcRs1sValue(); } + inline sreg_t rvc_rs1s() const { return getRegister(rvc_rs1s_reg()); } + inline int32_t rvc_rs2s_reg() const { return instr_.RvcRs2sValue(); } + inline sreg_t rvc_rs2s() const { return getRegister(rvc_rs2s_reg()); } + inline double rvc_drs2s() const { + return getFpuRegisterDouble(rvc_rs2s_reg()); + } + inline int32_t rvc_rd_reg() const { return instr_.RvcRdValue(); } + inline int32_t rvc_frd_reg() const { return instr_.RvcRdValue(); } + inline int16_t boffset() const { return instr_.BranchOffset(); } + inline int16_t imm12() const { return instr_.Imm12Value(); } + inline int32_t imm20J() const { return instr_.Imm20JValue(); } + inline int32_t imm5CSR() const { return instr_.Rs1Value(); } + inline int16_t csr_reg() const { return instr_.CsrValue(); } + inline int16_t rvc_imm6() const { return instr_.RvcImm6Value(); } + inline int16_t rvc_imm6_addi16sp() const { + return instr_.RvcImm6Addi16spValue(); + } + inline int16_t rvc_imm8_addi4spn() const { + return instr_.RvcImm8Addi4spnValue(); + } + inline int16_t rvc_imm6_lwsp() const { return instr_.RvcImm6LwspValue(); } + inline int16_t rvc_imm6_ldsp() const { return instr_.RvcImm6LdspValue(); } + inline int16_t rvc_imm6_swsp() const { return instr_.RvcImm6SwspValue(); } + inline int16_t rvc_imm6_sdsp() const { return instr_.RvcImm6SdspValue(); } + inline int16_t rvc_imm5_w() const { return instr_.RvcImm5WValue(); } + inline int16_t rvc_imm5_d() const { return instr_.RvcImm5DValue(); } + inline int16_t rvc_imm8_b() const { return instr_.RvcImm8BValue(); } + + // Helper for debugging memory access. + inline void DieOrDebug(); + +# if JS_CODEGEN_RISCV32 + template <typename T> + void TraceRegWr(T value, TraceType t = WORD); +# elif JS_CODEGEN_RISCV64 + void TraceRegWr(sreg_t value, TraceType t = DWORD); +# endif + void TraceMemWr(sreg_t addr, sreg_t value, TraceType t); + template <typename T> + void TraceMemRd(sreg_t addr, T value, sreg_t reg_value); + void TraceMemRdDouble(sreg_t addr, double value, int64_t reg_value); + void TraceMemRdDouble(sreg_t addr, Float64 value, int64_t reg_value); + void TraceMemRdFloat(sreg_t addr, Float32 value, int64_t reg_value); + + template <typename T> + void TraceLr(sreg_t addr, T value, sreg_t reg_value); + + template <typename T> + void TraceSc(sreg_t addr, T value); + + template <typename T> + void TraceMemWr(sreg_t addr, T value); + void TraceMemWrDouble(sreg_t addr, double value); + + inline void set_rd(sreg_t value, bool trace = true) { + setRegister(rd_reg(), value); +# if JS_CODEGEN_RISCV64 + if (trace) TraceRegWr(getRegister(rd_reg()), DWORD); +# elif JS_CODEGEN_RISCV32 + if (trace) TraceRegWr(getRegister(rd_reg()), WORD); +# endif + } + inline void set_frd(float value, bool trace = true) { + setFpuRegisterFloat(rd_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rd_reg()), FLOAT); + } + inline void set_frd(Float32 value, bool trace = true) { + setFpuRegisterFloat(rd_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rd_reg()), FLOAT); + } + inline void set_drd(double value, bool trace = true) { + setFpuRegisterDouble(rd_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rd_reg()), DOUBLE); + } + inline void set_drd(Float64 value, bool trace = true) { + setFpuRegisterDouble(rd_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rd_reg()), DOUBLE); + } + inline void set_rvc_rd(sreg_t value, bool trace = true) { + setRegister(rvc_rd_reg(), value); +# if JS_CODEGEN_RISCV64 + if (trace) TraceRegWr(getRegister(rvc_rd_reg()), DWORD); +# elif JS_CODEGEN_RISCV32 + if (trace) TraceRegWr(getRegister(rvc_rd_reg()), WORD); +# endif + } + inline void set_rvc_rs1s(sreg_t value, bool trace = true) { + setRegister(rvc_rs1s_reg(), value); +# if JS_CODEGEN_RISCV64 + if (trace) TraceRegWr(getRegister(rvc_rs1s_reg()), DWORD); +# elif JS_CODEGEN_RISCV32 + if (trace) TraceRegWr(getRegister(rvc_rs1s_reg()), WORD); +# endif + } + inline void set_rvc_rs2(sreg_t value, bool trace = true) { + setRegister(rvc_rs2_reg(), value); +# if JS_CODEGEN_RISCV64 + if (trace) TraceRegWr(getRegister(rvc_rs2_reg()), DWORD); +# elif JS_CODEGEN_RISCV32 + if (trace) TraceRegWr(getRegister(rvc_rs2_reg()), WORD); +# endif + } + inline void set_rvc_drd(double value, bool trace = true) { + setFpuRegisterDouble(rvc_rd_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rvc_rd_reg()), DOUBLE); + } + inline void set_rvc_drd(Float64 value, bool trace = true) { + setFpuRegisterDouble(rvc_rd_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rvc_rd_reg()), DOUBLE); + } + inline void set_rvc_frd(Float32 value, bool trace = true) { + setFpuRegisterFloat(rvc_rd_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rvc_rd_reg()), DOUBLE); + } + inline void set_rvc_rs2s(sreg_t value, bool trace = true) { + setRegister(rvc_rs2s_reg(), value); +# if JS_CODEGEN_RISCV64 + if (trace) TraceRegWr(getRegister(rvc_rs2s_reg()), DWORD); +# elif JS_CODEGEN_RISCV32 + if (trace) TraceRegWr(getRegister(rvc_rs2s_reg()), WORD); +# endif + } + inline void set_rvc_drs2s(double value, bool trace = true) { + setFpuRegisterDouble(rvc_rs2s_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rvc_rs2s_reg()), DOUBLE); + } + inline void set_rvc_drs2s(Float64 value, bool trace = true) { + setFpuRegisterDouble(rvc_rs2s_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rvc_rs2s_reg()), DOUBLE); + } + + inline void set_rvc_frs2s(Float32 value, bool trace = true) { + setFpuRegisterFloat(rvc_rs2s_reg(), value); + if (trace) TraceRegWr(getFpuRegister(rvc_rs2s_reg()), FLOAT); + } + + uint32_t get_dynamic_rounding_mode() { return read_csr_value(csr_frm); } + + // helper functions to read/write/set/clear CRC values/bits + uint32_t read_csr_value(uint32_t csr) { + switch (csr) { + case csr_fflags: // Floating-Point Accrued Exceptions (RW) + return (FCSR_ & kFcsrFlagsMask); + case csr_frm: // Floating-Point Dynamic Rounding Mode (RW) + return (FCSR_ & kFcsrFrmMask) >> kFcsrFrmShift; + case csr_fcsr: // Floating-Point Control and Status Register (RW) + return (FCSR_ & kFcsrMask); + default: + MOZ_CRASH("UNIMPLEMENTED"); + } + } + + void write_csr_value(uint32_t csr, reg_t val) { + uint32_t value = (uint32_t)val; + switch (csr) { + case csr_fflags: // Floating-Point Accrued Exceptions (RW) + MOZ_ASSERT(value <= ((1 << kFcsrFlagsBits) - 1)); + FCSR_ = (FCSR_ & (~kFcsrFlagsMask)) | value; + break; + case csr_frm: // Floating-Point Dynamic Rounding Mode (RW) + MOZ_ASSERT(value <= ((1 << kFcsrFrmBits) - 1)); + FCSR_ = (FCSR_ & (~kFcsrFrmMask)) | (value << kFcsrFrmShift); + break; + case csr_fcsr: // Floating-Point Control and Status Register (RW) + MOZ_ASSERT(value <= ((1 << kFcsrBits) - 1)); + FCSR_ = (FCSR_ & (~kFcsrMask)) | value; + break; + default: + MOZ_CRASH("UNIMPLEMENTED"); + } + } + + void set_csr_bits(uint32_t csr, reg_t val) { + uint32_t value = (uint32_t)val; + switch (csr) { + case csr_fflags: // Floating-Point Accrued Exceptions (RW) + MOZ_ASSERT(value <= ((1 << kFcsrFlagsBits) - 1)); + FCSR_ = FCSR_ | value; + break; + case csr_frm: // Floating-Point Dynamic Rounding Mode (RW) + MOZ_ASSERT(value <= ((1 << kFcsrFrmBits) - 1)); + FCSR_ = FCSR_ | (value << kFcsrFrmShift); + break; + case csr_fcsr: // Floating-Point Control and Status Register (RW) + MOZ_ASSERT(value <= ((1 << kFcsrBits) - 1)); + FCSR_ = FCSR_ | value; + break; + default: + MOZ_CRASH("UNIMPLEMENTED"); + } + } + + void clear_csr_bits(uint32_t csr, reg_t val) { + uint32_t value = (uint32_t)val; + switch (csr) { + case csr_fflags: // Floating-Point Accrued Exceptions (RW) + MOZ_ASSERT(value <= ((1 << kFcsrFlagsBits) - 1)); + FCSR_ = FCSR_ & (~value); + break; + case csr_frm: // Floating-Point Dynamic Rounding Mode (RW) + MOZ_ASSERT(value <= ((1 << kFcsrFrmBits) - 1)); + FCSR_ = FCSR_ & (~(value << kFcsrFrmShift)); + break; + case csr_fcsr: // Floating-Point Control and Status Register (RW) + MOZ_ASSERT(value <= ((1 << kFcsrBits) - 1)); + FCSR_ = FCSR_ & (~value); + break; + default: + MOZ_CRASH("UNIMPLEMENTED"); + } + } + + bool test_fflags_bits(uint32_t mask) { + return (FCSR_ & kFcsrFlagsMask & mask) != 0; + } + + void set_fflags(uint32_t flags) { set_csr_bits(csr_fflags, flags); } + void clear_fflags(int32_t flags) { clear_csr_bits(csr_fflags, flags); } + + float RoundF2FHelper(float input_val, int rmode); + double RoundF2FHelper(double input_val, int rmode); + template <typename I_TYPE, typename F_TYPE> + I_TYPE RoundF2IHelper(F_TYPE original, int rmode); + + template <typename T> + T FMaxMinHelper(T a, T b, MaxMinKind kind); + + template <typename T> + bool CompareFHelper(T input1, T input2, FPUCondition cc); + + template <typename T> + T get_pc_as() const { + return reinterpret_cast<T>(get_pc()); + } + + void enable_single_stepping(SingleStepCallback cb, void* arg); + void disable_single_stepping(); + + // Accessor to the internal simulator stack area. + uintptr_t stackLimit() const; + bool overRecursed(uintptr_t newsp = 0) const; + bool overRecursedWithExtra(uint32_t extra) const; + + // Executes MIPS instructions until the PC reaches end_sim_pc. + template <bool enableStopSimAt> + void execute(); + + // Sets up the simulator state and grabs the result on return. + int64_t call(uint8_t* entry, int argument_count, ...); + + // Push an address onto the JS stack. + uintptr_t pushAddress(uintptr_t address); + + // Pop an address from the JS stack. + uintptr_t popAddress(); + + // Debugger input. + void setLastDebuggerInput(char* input); + char* lastDebuggerInput() { return lastDebuggerInput_; } + + // Returns true if pc register contains one of the 'SpecialValues' defined + // below (bad_ra, end_sim_pc). + bool has_bad_pc() const; + + private: + enum SpecialValues { + // Known bad pc value to ensure that the simulator does not execute + // without being properly setup. + bad_ra = -1, + // A pc value used to signal the simulator to stop execution. Generally + // the ra is set to this value on transition from native C code to + // simulated execution, so that the simulator can "return" to the native + // C code. + end_sim_pc = -2, + // Unpredictable value. + Unpredictable = 0xbadbeaf + }; + + bool init(); + + // Unsupported instructions use Format to print an error and stop execution. + void format(SimInstruction* instr, const char* format); + + // Read and write memory. + // RISCV Memory read/write methods + template <typename T> + T ReadMem(sreg_t addr, Instruction* instr); + template <typename T> + void WriteMem(sreg_t addr, T value, Instruction* instr); + template <typename T, typename OP> + T amo(sreg_t addr, OP f, Instruction* instr, TraceType t) { + auto lhs = ReadMem<T>(addr, instr); + // TODO(RISCV): trace memory read for AMO + WriteMem<T>(addr, (T)f(lhs), instr); + return lhs; + } + + inline int32_t loadLinkedW(uint64_t addr, SimInstruction* instr); + inline int storeConditionalW(uint64_t addr, int32_t value, + SimInstruction* instr); + + inline int64_t loadLinkedD(uint64_t addr, SimInstruction* instr); + inline int storeConditionalD(uint64_t addr, int64_t value, + SimInstruction* instr); + + // Used for breakpoints and traps. + void SoftwareInterrupt(); + + // Stop helper functions. + bool isWatchpoint(uint32_t code); + bool IsTracepoint(uint32_t code); + void printWatchpoint(uint32_t code); + void handleStop(uint32_t code); + bool isStopInstruction(SimInstruction* instr); + bool isEnabledStop(uint32_t code); + void enableStop(uint32_t code); + void disableStop(uint32_t code); + void increaseStopCounter(uint32_t code); + void printStopInfo(uint32_t code); + + // Simulator breakpoints. + struct Breakpoint { + SimInstruction* location; + bool enabled; + bool is_tbreak; + }; + std::vector<Breakpoint> breakpoints_; + void SetBreakpoint(SimInstruction* breakpoint, bool is_tbreak); + void ListBreakpoints(); + void CheckBreakpoints(); + + JS::ProfilingFrameIterator::RegisterState registerState(); + void HandleWasmTrap(); + + // Handle any wasm faults, returning true if the fault was handled. + // This method is rather hot so inline the normal (no-wasm) case. + bool MOZ_ALWAYS_INLINE handleWasmSegFault(uint64_t addr, unsigned numBytes) { + if (MOZ_LIKELY(!js::wasm::CodeExists)) { + return false; + } + + uint8_t* newPC; + if (!js::wasm::MemoryAccessTraps(registerState(), (uint8_t*)addr, numBytes, + &newPC)) { + return false; + } + + LLBit_ = false; + set_pc(int64_t(newPC)); + return true; + } + + // Executes one instruction. + void InstructionDecode(Instruction* instr); + + // ICache. + // static void CheckICache(base::CustomMatcherHashMap* i_cache, + // Instruction* instr); + // static void FlushOnePage(base::CustomMatcherHashMap* i_cache, intptr_t + // start, + // size_t size); + // static CachePage* GetCachePage(base::CustomMatcherHashMap* i_cache, + // void* page); + template <typename T, typename Func> + inline T CanonicalizeFPUOpFMA(Func fn, T dst, T src1, T src2) { + static_assert(std::is_floating_point<T>::value); + auto alu_out = fn(dst, src1, src2); + // if any input or result is NaN, the result is quiet_NaN + if (std::isnan(alu_out) || std::isnan(src1) || std::isnan(src2) || + std::isnan(dst)) { + // signaling_nan sets kInvalidOperation bit + if (isSnan(alu_out) || isSnan(src1) || isSnan(src2) || isSnan(dst)) + set_fflags(kInvalidOperation); + alu_out = std::numeric_limits<T>::quiet_NaN(); + } + return alu_out; + } + + template <typename T, typename Func> + inline T CanonicalizeFPUOp3(Func fn) { + static_assert(std::is_floating_point<T>::value); + T src1 = std::is_same<float, T>::value ? frs1() : drs1(); + T src2 = std::is_same<float, T>::value ? frs2() : drs2(); + T src3 = std::is_same<float, T>::value ? frs3() : drs3(); + auto alu_out = fn(src1, src2, src3); + // if any input or result is NaN, the result is quiet_NaN + if (std::isnan(alu_out) || std::isnan(src1) || std::isnan(src2) || + std::isnan(src3)) { + // signaling_nan sets kInvalidOperation bit + if (isSnan(alu_out) || isSnan(src1) || isSnan(src2) || isSnan(src3)) + set_fflags(kInvalidOperation); + alu_out = std::numeric_limits<T>::quiet_NaN(); + } + return alu_out; + } + + template <typename T, typename Func> + inline T CanonicalizeFPUOp2(Func fn) { + static_assert(std::is_floating_point<T>::value); + T src1 = std::is_same<float, T>::value ? frs1() : drs1(); + T src2 = std::is_same<float, T>::value ? frs2() : drs2(); + auto alu_out = fn(src1, src2); + // if any input or result is NaN, the result is quiet_NaN + if (std::isnan(alu_out) || std::isnan(src1) || std::isnan(src2)) { + // signaling_nan sets kInvalidOperation bit + if (isSnan(alu_out) || isSnan(src1) || isSnan(src2)) + set_fflags(kInvalidOperation); + alu_out = std::numeric_limits<T>::quiet_NaN(); + } + return alu_out; + } + + template <typename T, typename Func> + inline T CanonicalizeFPUOp1(Func fn) { + static_assert(std::is_floating_point<T>::value); + T src1 = std::is_same<float, T>::value ? frs1() : drs1(); + auto alu_out = fn(src1); + // if any input or result is NaN, the result is quiet_NaN + if (std::isnan(alu_out) || std::isnan(src1)) { + // signaling_nan sets kInvalidOperation bit + if (isSnan(alu_out) || isSnan(src1)) set_fflags(kInvalidOperation); + alu_out = std::numeric_limits<T>::quiet_NaN(); + } + return alu_out; + } + + template <typename Func> + inline float CanonicalizeDoubleToFloatOperation(Func fn) { + float alu_out = fn(drs1()); + if (std::isnan(alu_out) || std::isnan(drs1())) + alu_out = std::numeric_limits<float>::quiet_NaN(); + return alu_out; + } + + template <typename Func> + inline float CanonicalizeDoubleToFloatOperation(Func fn, double frs) { + float alu_out = fn(frs); + if (std::isnan(alu_out) || std::isnan(drs1())) + alu_out = std::numeric_limits<float>::quiet_NaN(); + return alu_out; + } + + template <typename Func> + inline float CanonicalizeFloatToDoubleOperation(Func fn, float frs) { + double alu_out = fn(frs); + if (std::isnan(alu_out) || std::isnan(frs1())) + alu_out = std::numeric_limits<double>::quiet_NaN(); + return alu_out; + } + + template <typename Func> + inline float CanonicalizeFloatToDoubleOperation(Func fn) { + double alu_out = fn(frs1()); + if (std::isnan(alu_out) || std::isnan(frs1())) + alu_out = std::numeric_limits<double>::quiet_NaN(); + return alu_out; + } + + public: + static int64_t StopSimAt; + + // Runtime call support. + static void* RedirectNativeFunction(void* nativeFunction, + ABIFunctionType type); + + private: + enum Exception { + none, + kIntegerOverflow, + kIntegerUnderflow, + kDivideByZero, + kNumExceptions, + // RISCV illegual instruction exception + kIllegalInstruction, + }; + int16_t exceptions[kNumExceptions]; + + // Exceptions. + void SignalException(Exception e); + + // Handle return value for runtime FP functions. + void setCallResultDouble(double result); + void setCallResultFloat(float result); + void setCallResult(int64_t res); + void setCallResult(__int128 res); + + void callInternal(uint8_t* entry); + + // Architecture state. + // Registers. + int64_t registers_[kNumSimuRegisters]; + // Coprocessor Registers. + int64_t FPUregisters_[kNumFPURegisters]; + // FPU control register. + uint32_t FCSR_; + + bool LLBit_; + uintptr_t LLAddr_; + int64_t lastLLValue_; + + // Simulator support. + char* stack_; + uintptr_t stackLimit_; + bool pc_modified_; + int64_t icount_; + int64_t break_count_; + + // Debugger input. + char* lastDebuggerInput_; + + intptr_t* watch_address_ = nullptr; + intptr_t watch_value_ = 0; + + // Registered breakpoints. + SimInstruction* break_pc_; + Instr break_instr_; + EmbeddedVector<char, 256> trace_buf_; + + // Single-stepping support + bool single_stepping_; + SingleStepCallback single_step_callback_; + void* single_step_callback_arg_; + + // A stop is watched if its code is less than kNumOfWatchedStops. + // Only watched stops support enabling/disabling and the counter feature. + static const uint32_t kNumOfWatchedStops = 256; + + // Stop is disabled if bit 31 is set. + static const uint32_t kStopDisabledBit = 1U << 31; + + // A stop is enabled, meaning the simulator will stop when meeting the + // instruction, if bit 31 of watchedStops_[code].count is unset. + // The value watchedStops_[code].count & ~(1 << 31) indicates how many times + // the breakpoint was hit or gone through. + struct StopCountAndDesc { + uint32_t count_; + char* desc_; + }; + StopCountAndDesc watchedStops_[kNumOfWatchedStops]; +}; + +// Process wide simulator state. +class SimulatorProcess { + friend class Redirection; + friend class AutoLockSimulatorCache; + + private: + // ICache checking. + struct ICacheHasher { + typedef void* Key; + typedef void* Lookup; + static HashNumber hash(const Lookup& l); + static bool match(const Key& k, const Lookup& l); + }; + + public: + typedef HashMap<void*, CachePage*, ICacheHasher, SystemAllocPolicy> ICacheMap; + + static mozilla::Atomic<size_t, mozilla::ReleaseAcquire> + ICacheCheckingDisableCount; + static void FlushICache(void* start, size_t size); + + static void checkICacheLocked(SimInstruction* instr); + + static bool initialize() { + singleton_ = js_new<SimulatorProcess>(); + return singleton_; + } + static void destroy() { + js_delete(singleton_); + singleton_ = nullptr; + } + + SimulatorProcess(); + ~SimulatorProcess(); + + private: + static SimulatorProcess* singleton_; + + // This lock creates a critical section around 'redirection_' and + // 'icache_', which are referenced both by the execution engine + // and by the off-thread compiler (see Redirection::Get in the cpp file). + Mutex cacheLock_ MOZ_UNANNOTATED; + + Redirection* redirection_; + ICacheMap icache_; + + public: + static ICacheMap& icache() { + // Technically we need the lock to access the innards of the + // icache, not to take its address, but the latter condition + // serves as a useful complement to the former. + singleton_->cacheLock_.assertOwnedByCurrentThread(); + return singleton_->icache_; + } + + static Redirection* redirection() { + singleton_->cacheLock_.assertOwnedByCurrentThread(); + return singleton_->redirection_; + } + + static void setRedirection(js::jit::Redirection* redirection) { + singleton_->cacheLock_.assertOwnedByCurrentThread(); + singleton_->redirection_ = redirection; + } +}; + +} // namespace jit +} // namespace js + +#endif /* JS_SIMULATOR_MIPS64 */ + +#endif /* jit_riscv64_Simulator_riscv64_h */ diff --git a/js/src/jit/riscv64/Trampoline-riscv64.cpp b/js/src/jit/riscv64/Trampoline-riscv64.cpp new file mode 100644 index 0000000000..9ddcb2e89c --- /dev/null +++ b/js/src/jit/riscv64/Trampoline-riscv64.cpp @@ -0,0 +1,785 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "jit/Bailouts.h" +#include "jit/BaselineFrame.h" +#include "jit/CalleeToken.h" +#include "jit/JitFrames.h" +#include "jit/JitRuntime.h" +#ifdef JS_ION_PERF +# include "jit/PerfSpewer.h" +#endif +#include "jit/riscv64/SharedICRegisters-riscv64.h" +#include "jit/VMFunctions.h" +#include "vm/JitActivation.h" // js::jit::JitActivation +#include "vm/JSContext.h" + +#include "jit/MacroAssembler-inl.h" + +using namespace js; +using namespace js::jit; + +// This file includes stubs for generating the JIT trampolines when there is no +// JIT backend, and also includes implementations for assorted random things +// which can't be implemented in headers. + +// All registers to save and restore. This includes the stack pointer, since we +// use the ability to reference register values on the stack by index. +static const LiveRegisterSet AllRegs = + LiveRegisterSet(GeneralRegisterSet(Registers::AllMask), + FloatRegisterSet(FloatRegisters::AllMask)); + +static void PushBailoutFrame(MacroAssembler& masm, Register spArg) { + // Push the frameSize_ stored in ra + // See: CodeGeneratorRiscv64::generateOutOfLineCode() + masm.push(ra); + + // Push registers such that we can access them from [base + code]. + masm.PushRegsInMask(AllRegs); + + // Put pointer to BailoutStack as first argument to the Bailout() + masm.movePtr(StackPointer, spArg); +} + +struct EnterJITRegs { + double fs11; + double fs10; + double fs9; + double fs8; + double fs7; + double fs6; + double fs5; + double fs4; + double fs3; + double fs2; + double fs1; + double fs0; + + // uintptr_t align; + + // non-volatile registers. + uint64_t ra; + uint64_t sp; + uint64_t fp; + uint64_t gp; + uint64_t s11; + uint64_t s10; + uint64_t s9; + uint64_t s8; + uint64_t s7; + uint64_t s6; + uint64_t s5; + uint64_t s4; + uint64_t s3; + uint64_t s2; + uint64_t s1; + // Save reg_vp(a7) on stack, use it after call jit code. + uint64_t a7; +}; + +static void GenerateReturn(MacroAssembler& masm, int returnCode) { + MOZ_ASSERT(masm.framePushed() == sizeof(EnterJITRegs)); + + // Restore non-volatile registers + masm.ld(s1, StackPointer, offsetof(EnterJITRegs, s1)); + masm.ld(s2, StackPointer, offsetof(EnterJITRegs, s2)); + masm.ld(s3, StackPointer, offsetof(EnterJITRegs, s3)); + masm.ld(s4, StackPointer, offsetof(EnterJITRegs, s4)); + masm.ld(s5, StackPointer, offsetof(EnterJITRegs, s5)); + masm.ld(s6, StackPointer, offsetof(EnterJITRegs, s6)); + masm.ld(s7, StackPointer, offsetof(EnterJITRegs, s7)); + masm.ld(s8, StackPointer, offsetof(EnterJITRegs, s8)); + masm.ld(s9, StackPointer, offsetof(EnterJITRegs, s9)); + masm.ld(s10, StackPointer, offsetof(EnterJITRegs, s10)); + masm.ld(s11, StackPointer, offsetof(EnterJITRegs, s11)); + masm.ld(gp, StackPointer, offsetof(EnterJITRegs, gp)); + masm.ld(fp, StackPointer, offsetof(EnterJITRegs, fp)); + masm.ld(sp, StackPointer, offsetof(EnterJITRegs, sp)); + masm.ld(ra, StackPointer, offsetof(EnterJITRegs, ra)); + + // Restore non-volatile floating point registers + masm.fld(fs11, StackPointer, offsetof(EnterJITRegs, fs11)); + masm.fld(fs10, StackPointer, offsetof(EnterJITRegs, fs10)); + masm.fld(fs9, StackPointer, offsetof(EnterJITRegs, fs9)); + masm.fld(fs8, StackPointer, offsetof(EnterJITRegs, fs8)); + masm.fld(fs7, StackPointer, offsetof(EnterJITRegs, fs7)); + masm.fld(fs6, StackPointer, offsetof(EnterJITRegs, fs6)); + masm.fld(fs5, StackPointer, offsetof(EnterJITRegs, fs5)); + masm.fld(fs4, StackPointer, offsetof(EnterJITRegs, fs4)); + masm.fld(fs3, StackPointer, offsetof(EnterJITRegs, fs3)); + masm.fld(fs2, StackPointer, offsetof(EnterJITRegs, fs2)); + masm.fld(fs1, StackPointer, offsetof(EnterJITRegs, fs1)); + masm.fld(fs0, StackPointer, offsetof(EnterJITRegs, fs0)); + + masm.freeStack(sizeof(EnterJITRegs)); + + masm.branch(ra); +} + +static void GeneratePrologue(MacroAssembler& masm) { + masm.reserveStack(sizeof(EnterJITRegs)); + + masm.sd(s1, StackPointer, offsetof(EnterJITRegs, s1)); + masm.sd(s2, StackPointer, offsetof(EnterJITRegs, s2)); + masm.sd(s3, StackPointer, offsetof(EnterJITRegs, s3)); + masm.sd(s4, StackPointer, offsetof(EnterJITRegs, s4)); + masm.sd(s5, StackPointer, offsetof(EnterJITRegs, s5)); + masm.sd(s6, StackPointer, offsetof(EnterJITRegs, s6)); + masm.sd(s7, StackPointer, offsetof(EnterJITRegs, s7)); + masm.sd(s8, StackPointer, offsetof(EnterJITRegs, s8)); + masm.sd(s9, StackPointer, offsetof(EnterJITRegs, s9)); + masm.sd(s10, StackPointer, offsetof(EnterJITRegs, s10)); + masm.sd(s11, StackPointer, offsetof(EnterJITRegs, s11)); + masm.sd(gp, StackPointer, offsetof(EnterJITRegs, gp)); + masm.sd(fp, StackPointer, offsetof(EnterJITRegs, fp)); + masm.sd(sp, StackPointer, offsetof(EnterJITRegs, sp)); + masm.sd(ra, StackPointer, offsetof(EnterJITRegs, ra)); + masm.sd(a7, StackPointer, offsetof(EnterJITRegs, a7)); + + masm.fsd(fs11, StackPointer, offsetof(EnterJITRegs, fs11)); + masm.fsd(fs10, StackPointer, offsetof(EnterJITRegs, fs10)); + masm.fsd(fs9, StackPointer, offsetof(EnterJITRegs, fs9)); + masm.fsd(fs8, StackPointer, offsetof(EnterJITRegs, fs8)); + masm.fsd(fs7, StackPointer, offsetof(EnterJITRegs, fs7)); + masm.fsd(fs6, StackPointer, offsetof(EnterJITRegs, fs6)); + masm.fsd(fs5, StackPointer, offsetof(EnterJITRegs, fs5)); + masm.fsd(fs4, StackPointer, offsetof(EnterJITRegs, fs4)); + masm.fsd(fs3, StackPointer, offsetof(EnterJITRegs, fs3)); + masm.fsd(fs2, StackPointer, offsetof(EnterJITRegs, fs2)); + masm.fsd(fs1, StackPointer, offsetof(EnterJITRegs, fs1)); + masm.fsd(fs0, StackPointer, offsetof(EnterJITRegs, fs0)); +} + +static void GenerateBailoutThunk(MacroAssembler& masm, Label* bailoutTail) { + PushBailoutFrame(masm, a0); + + // Make space for Bailout's bailoutInfo outparam. + masm.reserveStack(sizeof(void*)); + masm.movePtr(StackPointer, a1); + + // Call the bailout function. + using Fn = bool (*)(BailoutStack* sp, BaselineBailoutInfo** info); + masm.setupUnalignedABICall(a2); + masm.passABIArg(a0); + masm.passABIArg(a1); + masm.callWithABI<Fn, Bailout>(ABIType::General, + CheckUnsafeCallWithABI::DontCheckOther); + + // Get the bailoutInfo outparam. + masm.pop(a2); + + // Remove both the bailout frame and the topmost Ion frame's stack. + masm.moveToStackPtr(FramePointer); + + // Jump to shared bailout tail. The BailoutInfo pointer has to be in a2. + masm.jump(bailoutTail); +} + +// Generates a trampoline for calling Jit compiled code from a C++ function. +// The trampoline use the EnterJitCode signature, with the standard x64 fastcall +// calling convention. +void JitRuntime::generateEnterJIT(JSContext* cx, MacroAssembler& masm) { + AutoCreatedBy acb(masm, "JitRuntime::generateEnterJIT"); + + enterJITOffset_ = startTrampolineCode(masm); + + const Register reg_code = IntArgReg0; + const Register reg_argc = IntArgReg1; + const Register reg_argv = IntArgReg2; + const mozilla::DebugOnly<Register> reg_frame = IntArgReg3; + const Register reg_token = IntArgReg4; + const Register reg_chain = IntArgReg5; + const Register reg_values = IntArgReg6; + const Register reg_vp = IntArgReg7; + + MOZ_ASSERT(OsrFrameReg == reg_frame); + + GeneratePrologue(masm); + + // Save stack pointer as baseline frame. + masm.movePtr(StackPointer, FramePointer); + + // Load the number of actual arguments into s3. + masm.unboxInt32(Address(reg_vp, 0), s3); + + /*************************************************************** + Loop over argv vector, push arguments onto stack in reverse order + ***************************************************************/ + + // if we are constructing, that also needs to include newTarget + JitSpew(JitSpew_Codegen, "__Line__: %d", __LINE__); + { + Label noNewTarget; + masm.branchTest32(Assembler::Zero, reg_token, + Imm32(CalleeToken_FunctionConstructing), &noNewTarget); + + masm.add32(Imm32(1), reg_argc); + + masm.bind(&noNewTarget); + } + JitSpew(JitSpew_Codegen, "__Line__: %d", __LINE__); + // Make stack algined + masm.ma_and(s2, reg_argc, Imm32(1)); + masm.ma_sub64(s1, zero, Imm32(sizeof(Value))); + Label no_zero; + masm.ma_branch(&no_zero, Assembler::Condition::Equal, s2, Operand(0)); + masm.mv(s1, zero); + masm.bind(&no_zero); + masm.ma_add64(StackPointer, StackPointer, s1); + + masm.slli(s2, reg_argc, 3); // Value* argv + masm.addPtr(reg_argv, s2); // s2 = &argv[argc] + JitSpew(JitSpew_Codegen, "__Line__: %d", __LINE__); + // Loop over arguments, copying them from an unknown buffer onto the Ion + // stack so they can be accessed from JIT'ed code. + Label header, footer; + // If there aren't any arguments, don't do anything + masm.ma_b(s2, reg_argv, &footer, Assembler::BelowOrEqual, ShortJump); + { + masm.bind(&header); + + masm.subPtr(Imm32(sizeof(Value)), s2); + masm.subPtr(Imm32(sizeof(Value)), StackPointer); + + ValueOperand value = ValueOperand(s6); + masm.loadValue(Address(s2, 0), value); + masm.storeValue(value, Address(StackPointer, 0)); + + masm.ma_b(s2, reg_argv, &header, Assembler::Above, ShortJump); + } + masm.bind(&footer); + JitSpew(JitSpew_Codegen, "__Line__: %d", __LINE__); + masm.push(reg_token); + masm.pushFrameDescriptorForJitCall(FrameType::CppToJSJit, s3, s3); + + CodeLabel returnLabel; + Label oomReturnLabel; + { + // Handle Interpreter -> Baseline OSR. + AllocatableGeneralRegisterSet regs(GeneralRegisterSet::All()); + MOZ_ASSERT(!regs.has(FramePointer)); + regs.take(OsrFrameReg); + regs.take(reg_code); + MOZ_ASSERT(!regs.has(ReturnReg), "ReturnReg matches reg_code"); + + Label notOsr; + masm.ma_b(OsrFrameReg, OsrFrameReg, ¬Osr, Assembler::Zero, ShortJump); + + Register numStackValues = reg_values; + regs.take(numStackValues); + Register scratch = regs.takeAny(); + + // Push return address. + masm.subPtr(Imm32(sizeof(uintptr_t)), StackPointer); + masm.ma_li(scratch, &returnLabel); + masm.storePtr(scratch, Address(StackPointer, 0)); + + // Push previous frame pointer. + masm.subPtr(Imm32(sizeof(uintptr_t)), StackPointer); + masm.storePtr(FramePointer, Address(StackPointer, 0)); + + // Reserve frame. + Register framePtr = FramePointer; + masm.movePtr(StackPointer, framePtr); + masm.subPtr(Imm32(BaselineFrame::Size()), StackPointer); + + Register framePtrScratch = regs.takeAny(); + masm.movePtr(sp, framePtrScratch); + + // Reserve space for locals and stack values. + masm.slli(scratch, numStackValues, 3); + masm.subPtr(scratch, StackPointer); + + // Enter exit frame. + masm.reserveStack(3 * sizeof(uintptr_t)); + masm.storePtr( + ImmWord(MakeFrameDescriptor(FrameType::BaselineJS)), + Address(StackPointer, 2 * sizeof(uintptr_t))); // Frame descriptor + masm.storePtr( + zero, Address(StackPointer, sizeof(uintptr_t))); // fake return address + masm.storePtr(FramePointer, Address(StackPointer, 0)); + + // No GC things to mark, push a bare token. + masm.loadJSContext(scratch); + masm.enterFakeExitFrame(scratch, scratch, ExitFrameType::Bare); + + masm.reserveStack(2 * sizeof(uintptr_t)); + masm.storePtr(framePtr, + Address(StackPointer, sizeof(uintptr_t))); // BaselineFrame + masm.storePtr(reg_code, Address(StackPointer, 0)); // jitcode + + using Fn = bool (*)(BaselineFrame* frame, InterpreterFrame* interpFrame, + uint32_t numStackValues); + masm.setupUnalignedABICall(scratch); + masm.passABIArg(framePtrScratch); // BaselineFrame + masm.passABIArg(OsrFrameReg); // InterpreterFrame + masm.passABIArg(numStackValues); + masm.callWithABI<Fn, jit::InitBaselineFrameForOsr>( + ABIType::General, CheckUnsafeCallWithABI::DontCheckHasExitFrame); + + regs.add(OsrFrameReg); + Register jitcode = regs.takeAny(); + masm.loadPtr(Address(StackPointer, 0), jitcode); + masm.loadPtr(Address(StackPointer, sizeof(uintptr_t)), framePtr); + masm.freeStack(2 * sizeof(uintptr_t)); + + Label error; + masm.freeStack(ExitFrameLayout::SizeWithFooter()); + masm.branchIfFalseBool(ReturnReg, &error); + + // If OSR-ing, then emit instrumentation for setting lastProfilerFrame + // if profiler instrumentation is enabled. + { + Label skipProfilingInstrumentation; + AbsoluteAddress addressOfEnabled( + cx->runtime()->geckoProfiler().addressOfEnabled()); + masm.branch32(Assembler::Equal, addressOfEnabled, Imm32(0), + &skipProfilingInstrumentation); + masm.profilerEnterFrame(framePtr, scratch); + masm.bind(&skipProfilingInstrumentation); + } + + masm.jump(jitcode); + + // OOM: load error value, discard return address and previous frame + // pointer and return. + masm.bind(&error); + masm.movePtr(framePtr, StackPointer); + masm.addPtr(Imm32(2 * sizeof(uintptr_t)), StackPointer); + masm.moveValue(MagicValue(JS_ION_ERROR), JSReturnOperand); + masm.jump(&oomReturnLabel); + + masm.bind(¬Osr); + // Load the scope chain in R1. + MOZ_ASSERT(R1.scratchReg() != reg_code); + masm.ma_or(R1.scratchReg(), reg_chain, zero); + } + JitSpew(JitSpew_Codegen, "__Line__: %d", __LINE__); + // The call will push the return address and frame pointer on the stack, thus + // we check that the stack would be aligned once the call is complete. + masm.assertStackAlignment(JitStackAlignment, 2 * sizeof(uintptr_t)); + + // Call the function with pushing return address to stack. + masm.callJitNoProfiler(reg_code); + + { + // Interpreter -> Baseline OSR will return here. + masm.bind(&returnLabel); + masm.addCodeLabel(returnLabel); + masm.bind(&oomReturnLabel); + } + + // Discard arguments and padding. Set sp to the address of the EnterJITRegs + // on the stack. + masm.mov(FramePointer, StackPointer); + + // Store the returned value into the vp + masm.ld(reg_vp, StackPointer, offsetof(EnterJITRegs, a7)); + masm.storeValue(JSReturnOperand, Address(reg_vp, 0)); + JitSpew(JitSpew_Codegen, "__Line__: %d", __LINE__); + // Restore non-volatile registers and return. + GenerateReturn(masm, ShortJump); +} + +// static +mozilla::Maybe<::JS::ProfilingFrameIterator::RegisterState> +JitRuntime::getCppEntryRegisters(JitFrameLayout* frameStackAddress) { + return mozilla::Nothing{}; +} + +void JitRuntime::generateInvalidator(MacroAssembler& masm, Label* bailoutTail) { + AutoCreatedBy acb(masm, "JitRuntime::generateInvalidator"); + + invalidatorOffset_ = startTrampolineCode(masm); + + // Stack has to be alligned here. If not, we will have to fix it. + masm.checkStackAlignment(); + + // Push registers such that we can access them from [base + code]. + masm.PushRegsInMask(AllRegs); + + // Pass pointer to InvalidationBailoutStack structure. + masm.movePtr(StackPointer, a0); + + // Reserve place for BailoutInfo pointer. Two words to ensure alignment for + // setupAlignedABICall. + masm.subPtr(Imm32(2 * sizeof(uintptr_t)), StackPointer); + // Pass pointer to BailoutInfo + masm.movePtr(StackPointer, a1); + + using Fn = bool (*)(InvalidationBailoutStack* sp, BaselineBailoutInfo** info); + masm.setupAlignedABICall(); + masm.passABIArg(a0); + masm.passABIArg(a1); + masm.callWithABI<Fn, InvalidationBailout>( + ABIType::General, CheckUnsafeCallWithABI::DontCheckOther); + + masm.pop(a2); + + // Pop the machine state and the dead frame. + masm.moveToStackPtr(FramePointer); + + // Jump to shared bailout tail. The BailoutInfo pointer has to be in r2. + masm.jump(bailoutTail); +} + +void JitRuntime::generateArgumentsRectifier(MacroAssembler& masm, + ArgumentsRectifierKind kind) { + // Do not erase the frame pointer in this function. + + AutoCreatedBy acb(masm, "JitRuntime::generateArgumentsRectifier"); + + switch (kind) { + case ArgumentsRectifierKind::Normal: + argumentsRectifierOffset_ = startTrampolineCode(masm); + break; + case ArgumentsRectifierKind::TrialInlining: + trialInliningArgumentsRectifierOffset_ = startTrampolineCode(masm); + break; + } + masm.pushReturnAddress(); + // Caller: + // [arg2] [arg1] [this] [[argc] [callee] [descr] [raddr]] <- sp + + // Frame prologue. + // + // NOTE: if this changes, fix the Baseline bailout code too! + // See BaselineStackBuilder::calculatePrevFramePtr and + // BaselineStackBuilder::buildRectifierFrame (in BaselineBailouts.cpp). + masm.push(FramePointer); + masm.mov(StackPointer, FramePointer); + + // Load argc. + masm.loadNumActualArgs(FramePointer, s3); + + Register numActArgsReg = a6; + Register calleeTokenReg = a7; + Register numArgsReg = a5; + + // Load |nformals| into numArgsReg. + masm.loadPtr( + Address(FramePointer, RectifierFrameLayout::offsetOfCalleeToken()), + calleeTokenReg); + masm.mov(calleeTokenReg, numArgsReg); + masm.andPtr(Imm32(uint32_t(CalleeTokenMask)), numArgsReg); + masm.loadFunctionArgCount(numArgsReg, numArgsReg); + + // Stash another copy in t3, since we are going to do destructive operations + // on numArgsReg + masm.mov(numArgsReg, t3); + + static_assert( + CalleeToken_FunctionConstructing == 1, + "Ensure that we can use the constructing bit to count the value"); + masm.mov(calleeTokenReg, t2); + masm.ma_and(t2, t2, Imm32(uint32_t(CalleeToken_FunctionConstructing))); + + // Including |this|, and |new.target|, there are (|nformals| + 1 + + // isConstructing) arguments to push to the stack. Then we push a + // JitFrameLayout. We compute the padding expressed in the number of extra + // |undefined| values to push on the stack. + static_assert( + sizeof(JitFrameLayout) % JitStackAlignment == 0, + "No need to consider the JitFrameLayout for aligning the stack"); + static_assert( + JitStackAlignment % sizeof(Value) == 0, + "Ensure that we can pad the stack by pushing extra UndefinedValue"); + + MOZ_ASSERT(mozilla::IsPowerOfTwo(JitStackValueAlignment)); + masm.add32( + Imm32(JitStackValueAlignment - 1 /* for padding */ + 1 /* for |this| */), + numArgsReg); + masm.add32(t2, numArgsReg); + masm.and32(Imm32(~(JitStackValueAlignment - 1)), numArgsReg); + + // Load the number of |undefined|s to push into t1. Subtract 1 for |this|. + masm.ma_sub64(t1, numArgsReg, s3); + masm.sub32(Imm32(1), t1); + + // Caller: + // [arg2] [arg1] [this] [ [argc] [callee] [descr] [raddr] ] <- sp + // '--- s3 ----' + // + // Rectifier frame: + // [fp'] [undef] [undef] [undef] [arg2] [arg1] [this] [ [argc] [callee] + // [descr] [raddr] ] + // '-------- t1 ---------' '--- s3 ----' + + // Copy number of actual arguments into numActArgsReg. + masm.mov(s3, numActArgsReg); + + masm.moveValue(UndefinedValue(), ValueOperand(t0)); + + // Push undefined. (including the padding) + { + Label undefLoopTop; + + masm.bind(&undefLoopTop); + masm.sub32(Imm32(1), t1); + masm.subPtr(Imm32(sizeof(Value)), StackPointer); + masm.storeValue(ValueOperand(t0), Address(StackPointer, 0)); + + masm.ma_b(t1, t1, &undefLoopTop, Assembler::NonZero, ShortJump); + } + + // Get the topmost argument. + static_assert(sizeof(Value) == 8, "TimesEight is used to skip arguments"); + + // Get the topmost argument. + masm.slli(t0, s3, 3); // t0 <- nargs * 8 + masm.ma_add64(t1, FramePointer, t0); // t1 <- fp(saved sp) + nargs * 8 + masm.addPtr(Imm32(sizeof(RectifierFrameLayout)), t1); + + // Push arguments, |nargs| + 1 times (to include |this|). + masm.addPtr(Imm32(1), s3); + { + Label copyLoopTop; + + masm.bind(©LoopTop); + masm.sub32(Imm32(1), s3); + masm.subPtr(Imm32(sizeof(Value)), StackPointer); + masm.loadValue(Address(t1, 0), ValueOperand(t0)); + masm.storeValue(ValueOperand(t0), Address(StackPointer, 0)); + masm.subPtr(Imm32(sizeof(Value)), t1); + + masm.ma_b(s3, s3, ©LoopTop, Assembler::NonZero, ShortJump); + } + + // if constructing, copy newTarget + { + Label notConstructing; + + masm.branchTest32(Assembler::Zero, calleeTokenReg, + Imm32(CalleeToken_FunctionConstructing), + ¬Constructing); + + // thisFrame[numFormals] = prevFrame[argc] + ValueOperand newTarget(t0); + + // Load vp[argc]. Add sizeof(Value) for |this|. + BaseIndex newTargetSrc(FramePointer, numActArgsReg, TimesEight, + sizeof(RectifierFrameLayout) + sizeof(Value)); + masm.loadValue(newTargetSrc, newTarget); + + // Again, 1 for |this| + BaseIndex newTargetDest(StackPointer, t3, TimesEight, sizeof(Value)); + masm.storeValue(newTarget, newTargetDest); + + masm.bind(¬Constructing); + } + + // Caller: + // [arg2] [arg1] [this] [ [argc] [callee] [descr] [raddr] ] + // + // + // Rectifier frame: + // [fp'] <- fp [undef] [undef] [undef] [arg2] [arg1] [this] <- sp [ [argc] + // [callee] [descr] [raddr] ] + // + + // Construct JitFrameLayout. + masm.push(calleeTokenReg); + masm.pushFrameDescriptorForJitCall(FrameType::Rectifier, numActArgsReg, + numActArgsReg); + + // Call the target function. + masm.andPtr(Imm32(uint32_t(CalleeTokenMask)), calleeTokenReg); + switch (kind) { + case ArgumentsRectifierKind::Normal: + masm.loadJitCodeRaw(calleeTokenReg, t1); + argumentsRectifierReturnOffset_ = masm.callJitNoProfiler(t1); + break; + case ArgumentsRectifierKind::TrialInlining: + Label noBaselineScript, done; + masm.loadBaselineJitCodeRaw(calleeTokenReg, t1, &noBaselineScript); + masm.callJitNoProfiler(t1); + masm.jump(&done); + + // See BaselineCacheIRCompiler::emitCallInlinedFunction. + masm.bind(&noBaselineScript); + masm.loadJitCodeRaw(calleeTokenReg, t1); + masm.callJitNoProfiler(t1); + masm.bind(&done); + break; + } + + masm.mov(FramePointer, StackPointer); + masm.pop(FramePointer); + masm.ret(); +} + +void JitRuntime::generateBailoutHandler(MacroAssembler& masm, + Label* bailoutTail) { + AutoCreatedBy acb(masm, "JitRuntime::generateBailoutHandler"); + + bailoutHandlerOffset_ = startTrampolineCode(masm); + + GenerateBailoutThunk(masm, bailoutTail); +} + +uint32_t JitRuntime::generatePreBarrier(JSContext* cx, MacroAssembler& masm, + MIRType type) { + AutoCreatedBy acb(masm, "JitRuntime::generatePreBarrier"); + + uint32_t offset = startTrampolineCode(masm); + + MOZ_ASSERT(PreBarrierReg == a1); + Register temp1 = a0; + Register temp2 = a2; + Register temp3 = a3; + masm.push(temp1); + masm.push(temp2); + masm.push(temp3); + + Label noBarrier; + masm.emitPreBarrierFastPath(cx->runtime(), type, temp1, temp2, temp3, + &noBarrier); + + // Call into C++ to mark this GC thing. + masm.pop(temp3); + masm.pop(temp2); + masm.pop(temp1); + + LiveRegisterSet save; + save.set() = RegisterSet(GeneralRegisterSet(Registers::VolatileMask), + FloatRegisterSet(FloatRegisters::VolatileMask)); + masm.push(ra); + masm.PushRegsInMask(save); + + masm.movePtr(ImmPtr(cx->runtime()), a0); + + masm.setupUnalignedABICall(a2); + masm.passABIArg(a0); + masm.passABIArg(a1); + masm.callWithABI(JitPreWriteBarrier(type)); + + masm.PopRegsInMask(save); + masm.ret(); + + masm.bind(&noBarrier); + masm.pop(temp3); + masm.pop(temp2); + masm.pop(temp1); + masm.abiret(); + + return offset; +} + +void JitRuntime::generateBailoutTailStub(MacroAssembler& masm, + Label* bailoutTail) { + AutoCreatedBy acb(masm, "JitRuntime::generateBailoutTailStub"); + + masm.bind(bailoutTail); + masm.generateBailoutTail(a1, a2); +} + +bool JitRuntime::generateVMWrapper(JSContext* cx, MacroAssembler& masm, + VMFunctionId id, const VMFunctionData& f, + DynFn nativeFun, uint32_t* wrapperOffset) { + AutoCreatedBy acb(masm, "JitRuntime::generateVMWrapper"); + + *wrapperOffset = startTrampolineCode(masm); + + // Avoid conflicts with argument registers while discarding the result after + // the function call. + AllocatableGeneralRegisterSet regs(Register::Codes::WrapperMask); + + static_assert( + (Register::Codes::VolatileMask & ~Register::Codes::WrapperMask) == 0, + "Wrapper register set should be a superset of Volatile register set."); + + // The context is the first argument; a0 is the first argument register. + Register cxreg = a0; + regs.take(cxreg); + + // On link-register platforms, it is the responsibility of the VM *callee* to + // push the return address, while the caller must ensure that the address + // is stored in ra on entry. This allows the VM wrapper to work with both + // direct calls and tail calls. + masm.pushReturnAddress(); + + // Push the frame pointer to finish the exit frame, then link it up. + masm.Push(FramePointer); + masm.moveStackPtrTo(FramePointer); + masm.loadJSContext(cxreg); + masm.enterExitFrame(cxreg, regs.getAny(), id); + + // Reserve space for the outparameter. + masm.reserveVMFunctionOutParamSpace(f); + + masm.setupUnalignedABICallDontSaveRestoreSP(); + masm.passABIArg(cxreg); + + size_t argDisp = ExitFrameLayout::Size(); + + // Copy any arguments. + for (uint32_t explicitArg = 0; explicitArg < f.explicitArgs; explicitArg++) { + switch (f.argProperties(explicitArg)) { + case VMFunctionData::WordByValue: + if (f.argPassedInFloatReg(explicitArg)) { + masm.passABIArg(MoveOperand(FramePointer, argDisp), ABIType::Float64); + } else { + masm.passABIArg(MoveOperand(FramePointer, argDisp), ABIType::General); + } + argDisp += sizeof(void*); + break; + case VMFunctionData::WordByRef: + masm.passABIArg(MoveOperand(FramePointer, argDisp, + MoveOperand::Kind::EffectiveAddress), + ABIType::General); + argDisp += sizeof(void*); + break; + case VMFunctionData::DoubleByValue: + case VMFunctionData::DoubleByRef: + MOZ_CRASH("NYI: riscv callVM should not be used with 128bits values."); + break; + } + } + + // Copy the semi-implicit outparam, if any. + // It is not a C++-abi outparam, which would get passed in the + // outparam register, but a real parameter to the function, which + // was stack-allocated above. + const int32_t outParamOffset = + -int32_t(ExitFooterFrame::Size()) - f.sizeOfOutParamStackSlot(); + if (f.outParam != Type_Void) { + masm.passABIArg(MoveOperand(FramePointer, outParamOffset, + MoveOperand::Kind::EffectiveAddress), + ABIType::General); + } + + masm.callWithABI(nativeFun, ABIType::General, + CheckUnsafeCallWithABI::DontCheckHasExitFrame); + + // Test for failure. + switch (f.failType()) { + case Type_Cell: + masm.branchTestPtr(Assembler::Zero, a0, a0, masm.failureLabel()); + break; + case Type_Bool: + // Called functions return bools, which are 0/false and non-zero/true + masm.branchIfFalseBool(a0, masm.failureLabel()); + break; + case Type_Void: + break; + default: + MOZ_CRASH("unknown failure kind"); + } + + // Load the outparam. + masm.loadVMFunctionOutParam(f, Address(FramePointer, outParamOffset)); + + // Pop frame and restore frame pointer. + masm.moveToStackPtr(FramePointer); + masm.pop(FramePointer); + + // Return. Subtract sizeof(void*) for the frame pointer. + masm.retn(Imm32(sizeof(ExitFrameLayout) - sizeof(void*) + + f.explicitStackSlots() * sizeof(void*) + + f.extraValuesToPop * sizeof(Value))); + + return true; +} diff --git a/js/src/jit/riscv64/constant/Base-constant-riscv.cpp b/js/src/jit/riscv64/constant/Base-constant-riscv.cpp new file mode 100644 index 0000000000..9658689775 --- /dev/null +++ b/js/src/jit/riscv64/constant/Base-constant-riscv.cpp @@ -0,0 +1,247 @@ +// Copyright 2021 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/constant/Base-constant-riscv.h" + +#include "mozilla/Assertions.h" +#include "mozilla/Types.h" + +#include <stdio.h> + +#include "jit/riscv64/constant/Constant-riscv-c.h" +#include "jit/riscv64/constant/Constant-riscv-d.h" +#include "jit/riscv64/constant/Constant-riscv-f.h" +#include "jit/riscv64/constant/Constant-riscv-i.h" +#include "jit/riscv64/constant/Constant-riscv-m.h" +#include "jit/riscv64/constant/Constant-riscv-v.h" +#include "jit/riscv64/constant/Constant-riscv-zicsr.h" +#include "jit/riscv64/constant/Constant-riscv-zifencei.h" +#include "jit/riscv64/Simulator-riscv64.h" +namespace js { +namespace jit { + +int32_t ImmBranchMaxForwardOffset(OffsetSize bits) { + return (1 << (bits - 1)) - 1; +} + +bool InstructionBase::IsShortInstruction() const { + uint8_t FirstByte = *reinterpret_cast<const uint8_t*>(this); + return (FirstByte & 0x03) <= C2; +} + +template <class T> +int InstructionGetters<T>::RvcRdValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + return this->Bits(kRvcRdShift + kRvcRdBits - 1, kRvcRdShift); +} + +template <class T> +int InstructionGetters<T>::RvcRs2Value() const { + MOZ_ASSERT(this->IsShortInstruction()); + return this->Bits(kRvcRs2Shift + kRvcRs2Bits - 1, kRvcRs2Shift); +} + +template <class T> +int InstructionGetters<T>::RvcRs1sValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + return 0b1000 + this->Bits(kRvcRs1sShift + kRvcRs1sBits - 1, kRvcRs1sShift); +} + +template <class T> +int InstructionGetters<T>::RvcRs2sValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + return 0b1000 + this->Bits(kRvcRs2sShift + kRvcRs2sBits - 1, kRvcRs2sShift); +} + +template <class T> +inline int InstructionGetters<T>::RvcFunct6Value() const { + MOZ_ASSERT(this->IsShortInstruction()); + return this->Bits(kRvcFunct6Shift + kRvcFunct6Bits - 1, kRvcFunct6Shift); +} + +template <class T> +inline int InstructionGetters<T>::RvcFunct4Value() const { + MOZ_ASSERT(this->IsShortInstruction()); + return this->Bits(kRvcFunct4Shift + kRvcFunct4Bits - 1, kRvcFunct4Shift); +} + +template <class T> +inline int InstructionGetters<T>::RvcFunct3Value() const { + MOZ_ASSERT(this->IsShortInstruction()); + return this->Bits(kRvcFunct3Shift + kRvcFunct3Bits - 1, kRvcFunct3Shift); +} + +template <class T> +inline int InstructionGetters<T>::RvcFunct2Value() const { + MOZ_ASSERT(this->IsShortInstruction()); + return this->Bits(kRvcFunct2Shift + kRvcFunct2Bits - 1, kRvcFunct2Shift); +} + +template <class T> +inline int InstructionGetters<T>::RvcFunct2BValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + return this->Bits(kRvcFunct2BShift + kRvcFunct2Bits - 1, kRvcFunct2BShift); +} + +template <class T> +uint32_t InstructionGetters<T>::Rvvzimm() const { + if ((this->InstructionBits() & + (kBaseOpcodeMask | kFunct3Mask | 0x80000000)) == RO_V_VSETVLI) { + uint32_t Bits = this->InstructionBits(); + uint32_t zimm = Bits & kRvvZimmMask; + return zimm >> kRvvZimmShift; + } else { + MOZ_ASSERT((this->InstructionBits() & + (kBaseOpcodeMask | kFunct3Mask | 0xC0000000)) == RO_V_VSETIVLI); + uint32_t Bits = this->InstructionBits(); + uint32_t zimm = Bits & kRvvZimmMask; + return (zimm >> kRvvZimmShift) & 0x3FF; + } +} + +template <class T> +uint32_t InstructionGetters<T>::Rvvuimm() const { + MOZ_ASSERT((this->InstructionBits() & + (kBaseOpcodeMask | kFunct3Mask | 0xC0000000)) == RO_V_VSETIVLI); + uint32_t Bits = this->InstructionBits(); + uint32_t uimm = Bits & kRvvUimmMask; + return uimm >> kRvvUimmShift; +} + +template class InstructionGetters<InstructionBase>; +#ifdef JS_SIMULATOR_RISCV64 +template class InstructionGetters<SimInstructionBase>; +#endif + +OffsetSize InstructionBase::GetOffsetSize() const { + if (IsIllegalInstruction()) { + MOZ_CRASH("IllegalInstruction"); + } + if (IsShortInstruction()) { + switch (InstructionBits() & kRvcOpcodeMask) { + case RO_C_J: + return kOffset11; + case RO_C_BEQZ: + case RO_C_BNEZ: + return kOffset9; + default: + MOZ_CRASH("IllegalInstruction"); + } + } else { + switch (InstructionBits() & kBaseOpcodeMask) { + case BRANCH: + return kOffset13; + case JAL: + return kOffset21; + default: + MOZ_CRASH("IllegalInstruction"); + } + } +} + +InstructionBase::Type InstructionBase::InstructionType() const { + if (IsIllegalInstruction()) { + return kUnsupported; + } + // RV64C Instruction + if (IsShortInstruction()) { + switch (InstructionBits() & kRvcOpcodeMask) { + case RO_C_ADDI4SPN: + return kCIWType; + case RO_C_FLD: + case RO_C_LW: +#ifdef JS_CODEGEN_RISCV64 + case RO_C_LD: +#endif + return kCLType; + case RO_C_FSD: + case RO_C_SW: +#ifdef JS_CODEGEN_RISCV64 + case RO_C_SD: +#endif + return kCSType; + case RO_C_NOP_ADDI: + case RO_C_LI: +#ifdef JS_CODEGEN_RISCV64 + case RO_C_ADDIW: +#endif + case RO_C_LUI_ADD: + return kCIType; + case RO_C_MISC_ALU: + if (Bits(11, 10) != 0b11) + return kCBType; + else + return kCAType; + case RO_C_J: + return kCJType; + case RO_C_BEQZ: + case RO_C_BNEZ: + return kCBType; + case RO_C_SLLI: + case RO_C_FLDSP: + case RO_C_LWSP: +#ifdef JS_CODEGEN_RISCV64 + case RO_C_LDSP: +#endif + return kCIType; + case RO_C_JR_MV_ADD: + return kCRType; + case RO_C_FSDSP: + case RO_C_SWSP: +#ifdef JS_CODEGEN_RISCV64 + case RO_C_SDSP: +#endif + return kCSSType; + default: + break; + } + } else { + // RISCV routine + switch (InstructionBits() & kBaseOpcodeMask) { + case LOAD: + return kIType; + case LOAD_FP: + return kIType; + case MISC_MEM: + return kIType; + case OP_IMM: + return kIType; + case AUIPC: + return kUType; + case OP_IMM_32: + return kIType; + case STORE: + return kSType; + case STORE_FP: + return kSType; + case AMO: + return kRType; + case OP: + return kRType; + case LUI: + return kUType; + case OP_32: + return kRType; + case MADD: + case MSUB: + case NMSUB: + case NMADD: + return kR4Type; + case OP_FP: + return kRType; + case BRANCH: + return kBType; + case JALR: + return kIType; + case JAL: + return kJType; + case SYSTEM: + return kIType; + case OP_V: + return kVType; + } + } + return kUnsupported; +} +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/constant/Base-constant-riscv.h b/js/src/jit/riscv64/constant/Base-constant-riscv.h new file mode 100644 index 0000000000..929ccd67b5 --- /dev/null +++ b/js/src/jit/riscv64/constant/Base-constant-riscv.h @@ -0,0 +1,1057 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Base_constant_riscv__h_ +#define jit_riscv64_constant_Base_constant_riscv__h_ +namespace js { +namespace jit { + +// On RISC-V Simulator breakpoints can have different codes: +// - Breaks between 0 and kMaxWatchpointCode are treated as simple watchpoints, +// the simulator will run through them and print the registers. +// - Breaks between kMaxWatchpointCode and kMaxStopCode are treated as stop() +// instructions (see Assembler::stop()). +// - Breaks larger than kMaxStopCode are simple breaks, dropping you into the +// debugger. +const uint32_t kMaxTracepointCode = 63; +const uint32_t kMaxWatchpointCode = 31; +const uint32_t kMaxStopCode = 127; +const uint32_t kWasmTrapCode = 6; +static_assert(kMaxWatchpointCode < kMaxStopCode); +static_assert(kMaxTracepointCode < kMaxStopCode); + +// Debug parameters. +// +// For example: +// +// __ Debug(TRACE_ENABLE | LOG_TRACE); +// starts tracing: set v8_flags.trace-sim is true. +// __ Debug(TRACE_ENABLE | LOG_REGS); +// PrintAllregs. +// __ Debug(TRACE_DISABLE | LOG_TRACE); +// stops tracing: set v8_flags.trace-sim is false. +const uint32_t kDebuggerTracingDirectivesMask = 0b111 << 3; +enum DebugParameters : uint32_t { + NO_PARAM = 1 << 5, + BREAK = 1 << 0, + LOG_TRACE = 1 << 1, + LOG_REGS = 1 << 2, + LOG_ALL = LOG_TRACE, + // Trace control. + TRACE_ENABLE = 1 << 3 | NO_PARAM, + TRACE_DISABLE = 1 << 4 | NO_PARAM, +}; +// On RISCV all instructions are 32 bits, except for RVC. +using Instr = int32_t; +using ShortInstr = int16_t; +typedef unsigned char byte; +// ----- Fields offset and length. +// RISCV constants +const int kBaseOpcodeShift = 0; +const int kBaseOpcodeBits = 7; +const int kFunct7Shift = 25; +const int kFunct7Bits = 7; +const int kFunct5Shift = 27; +const int kFunct5Bits = 5; +const int kFunct3Shift = 12; +const int kFunct3Bits = 3; +const int kFunct2Shift = 25; +const int kFunct2Bits = 2; +const int kRs1Shift = 15; +const int kRs1Bits = 5; +const int kVs1Shift = 15; +const int kVs1Bits = 5; +const int kVs2Shift = 20; +const int kVs2Bits = 5; +const int kVdShift = 7; +const int kVdBits = 5; +const int kRs2Shift = 20; +const int kRs2Bits = 5; +const int kRs3Shift = 27; +const int kRs3Bits = 5; +const int kRdShift = 7; +const int kRdBits = 5; +const int kRlShift = 25; +const int kAqShift = 26; +const int kImm12Shift = 20; +const int kImm12Bits = 12; +const int kImm11Shift = 2; +const int kImm11Bits = 11; +const int kShamtShift = 20; +const int kShamtBits = 5; +const int kShamtWShift = 20; +// FIXME: remove this once we have a proper way to handle the wide shift amount +const int kShamtWBits = 6; +const int kArithShiftShift = 30; +const int kImm20Shift = 12; +const int kImm20Bits = 20; +const int kCsrShift = 20; +const int kCsrBits = 12; +const int kMemOrderBits = 4; +const int kPredOrderShift = 24; +const int kSuccOrderShift = 20; + +// for C extension +const int kRvcFunct4Shift = 12; +const int kRvcFunct4Bits = 4; +const int kRvcFunct3Shift = 13; +const int kRvcFunct3Bits = 3; +const int kRvcRs1Shift = 7; +const int kRvcRs1Bits = 5; +const int kRvcRs2Shift = 2; +const int kRvcRs2Bits = 5; +const int kRvcRdShift = 7; +const int kRvcRdBits = 5; +const int kRvcRs1sShift = 7; +const int kRvcRs1sBits = 3; +const int kRvcRs2sShift = 2; +const int kRvcRs2sBits = 3; +const int kRvcFunct2Shift = 5; +const int kRvcFunct2BShift = 10; +const int kRvcFunct2Bits = 2; +const int kRvcFunct6Shift = 10; +const int kRvcFunct6Bits = 6; + +const uint32_t kRvcOpcodeMask = + 0b11 | (((1 << kRvcFunct3Bits) - 1) << kRvcFunct3Shift); +const uint32_t kRvcFunct3Mask = + (((1 << kRvcFunct3Bits) - 1) << kRvcFunct3Shift); +const uint32_t kRvcFunct4Mask = + (((1 << kRvcFunct4Bits) - 1) << kRvcFunct4Shift); +const uint32_t kRvcFunct6Mask = + (((1 << kRvcFunct6Bits) - 1) << kRvcFunct6Shift); +const uint32_t kRvcFunct2Mask = + (((1 << kRvcFunct2Bits) - 1) << kRvcFunct2Shift); +const uint32_t kRvcFunct2BMask = + (((1 << kRvcFunct2Bits) - 1) << kRvcFunct2BShift); +const uint32_t kCRTypeMask = kRvcOpcodeMask | kRvcFunct4Mask; +const uint32_t kCSTypeMask = kRvcOpcodeMask | kRvcFunct6Mask; +const uint32_t kCATypeMask = kRvcOpcodeMask | kRvcFunct6Mask | kRvcFunct2Mask; +const uint32_t kRvcBImm8Mask = (((1 << 5) - 1) << 2) | (((1 << 3) - 1) << 10); + +// RISCV Instruction bit masks +const uint32_t kBaseOpcodeMask = ((1 << kBaseOpcodeBits) - 1) + << kBaseOpcodeShift; +const uint32_t kFunct3Mask = ((1 << kFunct3Bits) - 1) << kFunct3Shift; +const uint32_t kFunct5Mask = ((1 << kFunct5Bits) - 1) << kFunct5Shift; +const uint32_t kFunct7Mask = ((1 << kFunct7Bits) - 1) << kFunct7Shift; +const uint32_t kFunct2Mask = 0b11 << kFunct7Shift; +const uint32_t kRTypeMask = kBaseOpcodeMask | kFunct3Mask | kFunct7Mask; +const uint32_t kRATypeMask = kBaseOpcodeMask | kFunct3Mask | kFunct5Mask; +const uint32_t kRFPTypeMask = kBaseOpcodeMask | kFunct7Mask; +const uint32_t kR4TypeMask = kBaseOpcodeMask | kFunct3Mask | kFunct2Mask; +const uint32_t kITypeMask = kBaseOpcodeMask | kFunct3Mask; +const uint32_t kSTypeMask = kBaseOpcodeMask | kFunct3Mask; +const uint32_t kBTypeMask = kBaseOpcodeMask | kFunct3Mask; +const uint32_t kUTypeMask = kBaseOpcodeMask; +const uint32_t kJTypeMask = kBaseOpcodeMask; +const uint32_t kRs1FieldMask = ((1 << kRs1Bits) - 1) << kRs1Shift; +const uint32_t kRs2FieldMask = ((1 << kRs2Bits) - 1) << kRs2Shift; +const uint32_t kRs3FieldMask = ((1 << kRs3Bits) - 1) << kRs3Shift; +const uint32_t kRdFieldMask = ((1 << kRdBits) - 1) << kRdShift; +const uint32_t kBImm12Mask = kFunct7Mask | kRdFieldMask; +const uint32_t kImm20Mask = ((1 << kImm20Bits) - 1) << kImm20Shift; +const uint32_t kImm12Mask = ((1 << kImm12Bits) - 1) << kImm12Shift; +const uint32_t kImm11Mask = ((1 << kImm11Bits) - 1) << kImm11Shift; +const uint32_t kImm31_12Mask = ((1 << 20) - 1) << 12; +const uint32_t kImm19_0Mask = ((1 << 20) - 1); + +// for RVV extension +#define RVV_LMUL(V) \ + V(m1) \ + V(m2) \ + V(m4) \ + V(m8) \ + V(RESERVERD) \ + V(mf8) \ + V(mf4) \ + V(mf2) + +enum Vlmul { +#define DEFINE_FLAG(name) name, + RVV_LMUL(DEFINE_FLAG) +#undef DEFINE_FLAG +}; + +#define RVV_SEW(V) \ + V(E8) \ + V(E16) \ + V(E32) \ + V(E64) + +#define DEFINE_FLAG(name) name, +enum VSew { + RVV_SEW(DEFINE_FLAG) +#undef DEFINE_FLAG +}; + +constexpr int kRvvELEN = 64; +constexpr int kRvvVLEN = 128; +constexpr int kRvvSLEN = kRvvVLEN; +const int kRvvFunct6Shift = 26; +const int kRvvFunct6Bits = 6; +const uint32_t kRvvFunct6Mask = + (((1 << kRvvFunct6Bits) - 1) << kRvvFunct6Shift); + +const int kRvvVmBits = 1; +const int kRvvVmShift = 25; +const uint32_t kRvvVmMask = (((1 << kRvvVmBits) - 1) << kRvvVmShift); + +const int kRvvVs2Bits = 5; +const int kRvvVs2Shift = 20; +const uint32_t kRvvVs2Mask = (((1 << kRvvVs2Bits) - 1) << kRvvVs2Shift); + +const int kRvvVs1Bits = 5; +const int kRvvVs1Shift = 15; +const uint32_t kRvvVs1Mask = (((1 << kRvvVs1Bits) - 1) << kRvvVs1Shift); + +const int kRvvRs1Bits = kRvvVs1Bits; +const int kRvvRs1Shift = kRvvVs1Shift; +const uint32_t kRvvRs1Mask = (((1 << kRvvRs1Bits) - 1) << kRvvRs1Shift); + +const int kRvvRs2Bits = 5; +const int kRvvRs2Shift = 20; +const uint32_t kRvvRs2Mask = (((1 << kRvvRs2Bits) - 1) << kRvvRs2Shift); + +const int kRvvImm5Bits = kRvvVs1Bits; +const int kRvvImm5Shift = kRvvVs1Shift; +const uint32_t kRvvImm5Mask = (((1 << kRvvImm5Bits) - 1) << kRvvImm5Shift); + +const int kRvvVdBits = 5; +const int kRvvVdShift = 7; +const uint32_t kRvvVdMask = (((1 << kRvvVdBits) - 1) << kRvvVdShift); + +const int kRvvRdBits = kRvvVdBits; +const int kRvvRdShift = kRvvVdShift; +const uint32_t kRvvRdMask = (((1 << kRvvRdBits) - 1) << kRvvRdShift); + +const int kRvvZimmBits = 11; +const int kRvvZimmShift = 20; +const uint32_t kRvvZimmMask = (((1 << kRvvZimmBits) - 1) << kRvvZimmShift); + +const int kRvvUimmShift = kRvvRs1Shift; +const int kRvvUimmBits = kRvvRs1Bits; +const uint32_t kRvvUimmMask = (((1 << kRvvUimmBits) - 1) << kRvvUimmShift); + +const int kRvvWidthBits = 3; +const int kRvvWidthShift = 12; +const uint32_t kRvvWidthMask = (((1 << kRvvWidthBits) - 1) << kRvvWidthShift); + +const int kRvvMopBits = 2; +const int kRvvMopShift = 26; +const uint32_t kRvvMopMask = (((1 << kRvvMopBits) - 1) << kRvvMopShift); + +const int kRvvMewBits = 1; +const int kRvvMewShift = 28; +const uint32_t kRvvMewMask = (((1 << kRvvMewBits) - 1) << kRvvMewShift); + +const int kRvvNfBits = 3; +const int kRvvNfShift = 29; +const uint32_t kRvvNfMask = (((1 << kRvvNfBits) - 1) << kRvvNfShift); + +const int kNopByte = 0x00000013; + +enum BaseOpcode : uint32_t { + LOAD = 0b0000011, // I form: LB LH LW LBU LHU + LOAD_FP = 0b0000111, // I form: FLW FLD FLQ + MISC_MEM = 0b0001111, // I special form: FENCE FENCE.I + OP_IMM = 0b0010011, // I form: ADDI SLTI SLTIU XORI ORI ANDI SLLI SRLI SRAI + // Note: SLLI/SRLI/SRAI I form first, then func3 001/101 => R type + AUIPC = 0b0010111, // U form: AUIPC + OP_IMM_32 = 0b0011011, // I form: ADDIW SLLIW SRLIW SRAIW + // Note: SRLIW SRAIW I form first, then func3 101 special shift encoding + STORE = 0b0100011, // S form: SB SH SW SD + STORE_FP = 0b0100111, // S form: FSW FSD FSQ + AMO = 0b0101111, // R form: All A instructions + OP = 0b0110011, // R: ADD SUB SLL SLT SLTU XOR SRL SRA OR AND and 32M set + LUI = 0b0110111, // U form: LUI + OP_32 = 0b0111011, // R: ADDW SUBW SLLW SRLW SRAW MULW DIVW DIVUW REMW REMUW + MADD = 0b1000011, // R4 type: FMADD.S FMADD.D FMADD.Q + MSUB = 0b1000111, // R4 type: FMSUB.S FMSUB.D FMSUB.Q + NMSUB = 0b1001011, // R4 type: FNMSUB.S FNMSUB.D FNMSUB.Q + NMADD = 0b1001111, // R4 type: FNMADD.S FNMADD.D FNMADD.Q + OP_FP = 0b1010011, // R type: Q ext + BRANCH = 0b1100011, // B form: BEQ BNE, BLT, BGE, BLTU BGEU + JALR = 0b1100111, // I form: JALR + JAL = 0b1101111, // J form: JAL + SYSTEM = 0b1110011, // I form: ECALL EBREAK Zicsr ext + OP_V = 0b1010111, // V form: RVV + + // C extension + C0 = 0b00, + C1 = 0b01, + C2 = 0b10, + FUNCT2_0 = 0b00, + FUNCT2_1 = 0b01, + FUNCT2_2 = 0b10, + FUNCT2_3 = 0b11, +}; + +// ----- Emulated conditions. +// On RISC-V we use this enum to abstract from conditional branch instructions. +// The 'U' prefix is used to specify unsigned comparisons. +// Opposite conditions must be paired as odd/even numbers +// because 'NegateCondition' function flips LSB to negate condition. +enum RiscvCondition { // Any value < 0 is considered no_condition. + overflow = 0, + no_overflow = 1, + Uless = 2, + Ugreater_equal = 3, + Uless_equal = 4, + Ugreater = 5, + equal = 6, + not_equal = 7, // Unordered or Not Equal. + less = 8, + greater_equal = 9, + less_equal = 10, + greater = 11, + cc_always = 12, + + // Aliases. + eq = equal, + ne = not_equal, + ge = greater_equal, + lt = less, + gt = greater, + le = less_equal, + al = cc_always, + ult = Uless, + uge = Ugreater_equal, + ule = Uless_equal, + ugt = Ugreater, +}; + +// ----- Coprocessor conditions. +enum FPUCondition { + kNoFPUCondition = -1, + EQ = 0x02, // Ordered and Equal + NE = 0x03, // Unordered or Not Equal + LT = 0x04, // Ordered and Less Than + GE = 0x05, // Ordered and Greater Than or Equal + LE = 0x06, // Ordered and Less Than or Equal + GT = 0x07, // Ordered and Greater Than +}; + +enum CheckForInexactConversion { + kCheckForInexactConversion, + kDontCheckForInexactConversion +}; + +enum class MaxMinKind : int { kMin = 0, kMax = 1 }; + +// ---------------------------------------------------------------------------- +// RISCV flags + +enum ControlStatusReg { + csr_fflags = 0x001, // Floating-Point Accrued Exceptions (RW) + csr_frm = 0x002, // Floating-Point Dynamic Rounding Mode (RW) + csr_fcsr = 0x003, // Floating-Point Control and Status Register (RW) + csr_cycle = 0xc00, // Cycle counter for RDCYCLE instruction (RO) + csr_time = 0xc01, // Timer for RDTIME instruction (RO) + csr_instret = 0xc02, // Insns-retired counter for RDINSTRET instruction (RO) + csr_cycleh = 0xc80, // Upper 32 bits of cycle, RV32I only (RO) + csr_timeh = 0xc81, // Upper 32 bits of time, RV32I only (RO) + csr_instreth = 0xc82 // Upper 32 bits of instret, RV32I only (RO) +}; + +enum FFlagsMask { + kInvalidOperation = 0b10000, // NV: Invalid + kDivideByZero = 0b1000, // DZ: Divide by Zero + kOverflow = 0b100, // OF: Overflow + kUnderflow = 0b10, // UF: Underflow + kInexact = 0b1 // NX: Inexact +}; + +enum FPURoundingMode { + RNE = 0b000, // Round to Nearest, ties to Even + RTZ = 0b001, // Round towards Zero + RDN = 0b010, // Round Down (towards -infinity) + RUP = 0b011, // Round Up (towards +infinity) + RMM = 0b100, // Round to Nearest, tiest to Max Magnitude + DYN = 0b111 // In instruction's rm field, selects dynamic rounding mode; + // In Rounding Mode register, Invalid +}; + +enum MemoryOdering { + PSI = 0b1000, // PI or SI + PSO = 0b0100, // PO or SO + PSR = 0b0010, // PR or SR + PSW = 0b0001, // PW or SW + PSIORW = PSI | PSO | PSR | PSW +}; + +const int kFloat32ExponentBias = 127; +const int kFloat32MantissaBits = 23; +const int kFloat32ExponentBits = 8; +const int kFloat64ExponentBias = 1023; +const int kFloat64MantissaBits = 52; +const int kFloat64ExponentBits = 11; + +enum FClassFlag { + kNegativeInfinity = 1, + kNegativeNormalNumber = 1 << 1, + kNegativeSubnormalNumber = 1 << 2, + kNegativeZero = 1 << 3, + kPositiveZero = 1 << 4, + kPositiveSubnormalNumber = 1 << 5, + kPositiveNormalNumber = 1 << 6, + kPositiveInfinity = 1 << 7, + kSignalingNaN = 1 << 8, + kQuietNaN = 1 << 9 +}; + +enum OffsetSize : uint32_t { + kOffset21 = 21, // RISCV jal + kOffset12 = 12, // RISCV imm12 + kOffset20 = 20, // RISCV imm20 + kOffset13 = 13, // RISCV branch + kOffset32 = 32, // RISCV auipc + instr_I + kOffset11 = 11, // RISCV C_J + kOffset9 = 9, // RISCV compressed branch +}; + +// The classes of immediate branch ranges, in order of increasing range. +// Note that CondBranchType and CompareBranchType have the same range. +enum ImmBranchRangeType { + CondBranchRangeType, // + UncondBranchRangeType, // + UnknownBranchRangeType, + + // Number of 'short-range' branch range types. + // We don't consider unconditional branches 'short-range'. + NumShortBranchRangeTypes = UnknownBranchRangeType +}; + +inline ImmBranchRangeType OffsetSizeToImmBranchRangeType(OffsetSize bits) { + switch (bits) { + case kOffset21: + return UncondBranchRangeType; + case kOffset13: + return CondBranchRangeType; + default: + MOZ_CRASH("Unimplement"); + } +} + +inline OffsetSize ImmBranchRangeTypeToOffsetSize(ImmBranchRangeType type) { + switch (type) { + case CondBranchRangeType: + return kOffset13; + case UncondBranchRangeType: + return kOffset21; + default: + MOZ_CRASH("Unimplement"); + } +} + +int32_t ImmBranchMaxForwardOffset(OffsetSize bits); + +inline int32_t ImmBranchMaxForwardOffset(ImmBranchRangeType type) { + return ImmBranchMaxForwardOffset(ImmBranchRangeTypeToOffsetSize(type)); +} +// ----------------------------------------------------------------------------- +// Specific instructions, constants, and masks. +// These constants are declared in assembler-riscv64.cc, as they use named +// registers and other constants. + +// An Illegal instruction +const Instr kIllegalInstr = 0; // All other bits are 0s (i.e., ecall) +// An ECALL instruction, used for redirected real time call +const Instr rtCallRedirInstr = SYSTEM; // All other bits are 0s (i.e., ecall) +// An EBreak instruction, used for debugging and semi-hosting +const Instr kBreakInstr = SYSTEM | 1 << kImm12Shift; // ebreak + +constexpr uint8_t kInstrSize = 4; +constexpr uint8_t kShortInstrSize = 2; +constexpr uint8_t kInstrSizeLog2 = 2; + +class InstructionBase { + public: + enum { + // On RISC-V, PC cannot actually be directly accessed. We behave as if PC + // was always the value of the current instruction being executed. + kPCReadOffset = 0 + }; + + // Instruction type. + enum Type { + kRType, + kR4Type, // Special R4 for Q extension + kIType, + kSType, + kBType, + kUType, + kJType, + // C extension + kCRType, + kCIType, + kCSSType, + kCIWType, + kCLType, + kCSType, + kCAType, + kCBType, + kCJType, + // V extension + kVType, + kVLType, + kVSType, + kVAMOType, + kVIVVType, + kVFVVType, + kVMVVType, + kVIVIType, + kVIVXType, + kVFVFType, + kVMVXType, + kVSETType, + kUnsupported = -1 + }; + + inline bool IsIllegalInstruction() const { + uint16_t FirstHalfWord = *reinterpret_cast<const uint16_t*>(this); + return FirstHalfWord == 0; + } + + bool IsShortInstruction() const; + + inline uint8_t InstructionSize() const { + return (this->IsShortInstruction()) ? kShortInstrSize : kInstrSize; + } + + // Get the raw instruction bits. + inline Instr InstructionBits() const { + if (this->IsShortInstruction()) { + return 0x0000FFFF & (*reinterpret_cast<const ShortInstr*>(this)); + } + return *reinterpret_cast<const Instr*>(this); + } + + // Set the raw instruction bits to value. + inline void SetInstructionBits(Instr value) { + *reinterpret_cast<Instr*>(this) = value; + } + + // Read one particular bit out of the instruction bits. + inline int Bit(int nr) const { return (InstructionBits() >> nr) & 1; } + + // Read a bit field out of the instruction bits. + inline int Bits(int hi, int lo) const { + return (InstructionBits() >> lo) & ((2U << (hi - lo)) - 1); + } + + // Accessors for the different named fields used in the RISC-V encoding. + inline BaseOpcode BaseOpcodeValue() const { + return static_cast<BaseOpcode>( + Bits(kBaseOpcodeShift + kBaseOpcodeBits - 1, kBaseOpcodeShift)); + } + + // Return the fields at their original place in the instruction encoding. + inline BaseOpcode BaseOpcodeFieldRaw() const { + return static_cast<BaseOpcode>(InstructionBits() & kBaseOpcodeMask); + } + + // Safe to call within R-type instructions + inline int Funct7FieldRaw() const { return InstructionBits() & kFunct7Mask; } + + // Safe to call within R-, I-, S-, or B-type instructions + inline int Funct3FieldRaw() const { return InstructionBits() & kFunct3Mask; } + + // Safe to call within R-, I-, S-, or B-type instructions + inline int Rs1FieldRawNoAssert() const { + return InstructionBits() & kRs1FieldMask; + } + + // Safe to call within R-, S-, or B-type instructions + inline int Rs2FieldRawNoAssert() const { + return InstructionBits() & kRs2FieldMask; + } + + // Safe to call within R4-type instructions + inline int Rs3FieldRawNoAssert() const { + return InstructionBits() & kRs3FieldMask; + } + + inline int32_t ITypeBits() const { return InstructionBits() & kITypeMask; } + + inline int32_t InstructionOpcodeType() const { + if (IsShortInstruction()) { + return InstructionBits() & kRvcOpcodeMask; + } else { + return InstructionBits() & kBaseOpcodeMask; + } + } + + // Get the encoding type of the instruction. + Type InstructionType() const; + OffsetSize GetOffsetSize() const; + inline ImmBranchRangeType GetImmBranchRangeType() const { + return OffsetSizeToImmBranchRangeType(GetOffsetSize()); + } + + protected: + InstructionBase() {} +}; + +template <class T> +class InstructionGetters : public T { + public: + // Say if the instruction is a break or a trap. + bool IsTrap() const; + + inline int BaseOpcode() const { + return this->InstructionBits() & kBaseOpcodeMask; + } + + inline int RvcOpcode() const { + MOZ_ASSERT(this->IsShortInstruction()); + return this->InstructionBits() & kRvcOpcodeMask; + } + + inline int Rs1Value() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kRType || + this->InstructionType() == InstructionBase::kR4Type || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kSType || + this->InstructionType() == InstructionBase::kBType || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kVType); + return this->Bits(kRs1Shift + kRs1Bits - 1, kRs1Shift); + } + + inline int Rs2Value() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kRType || + this->InstructionType() == InstructionBase::kR4Type || + this->InstructionType() == InstructionBase::kSType || + this->InstructionType() == InstructionBase::kBType || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kVType); + return this->Bits(kRs2Shift + kRs2Bits - 1, kRs2Shift); + } + + inline int Rs3Value() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kR4Type); + return this->Bits(kRs3Shift + kRs3Bits - 1, kRs3Shift); + } + + inline int Vs1Value() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kVType || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kSType); + return this->Bits(kVs1Shift + kVs1Bits - 1, kVs1Shift); + } + + inline int Vs2Value() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kVType || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kSType); + return this->Bits(kVs2Shift + kVs2Bits - 1, kVs2Shift); + } + + inline int VdValue() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kVType || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kSType); + return this->Bits(kVdShift + kVdBits - 1, kVdShift); + } + + inline int RdValue() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kRType || + this->InstructionType() == InstructionBase::kR4Type || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kSType || + this->InstructionType() == InstructionBase::kUType || + this->InstructionType() == InstructionBase::kJType || + this->InstructionType() == InstructionBase::kVType); + return this->Bits(kRdShift + kRdBits - 1, kRdShift); + } + + inline int RvcRs1Value() const { return this->RvcRdValue(); } + + int RvcRdValue() const; + + int RvcRs2Value() const; + + int RvcRs1sValue() const; + + int RvcRs2sValue() const; + + int Funct7Value() const; + + inline int Funct3Value() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kRType || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kSType || + this->InstructionType() == InstructionBase::kBType); + return this->Bits(kFunct3Shift + kFunct3Bits - 1, kFunct3Shift); + } + + inline int Funct5Value() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kRType && + this->BaseOpcode() == OP_FP); + return this->Bits(kFunct5Shift + kFunct5Bits - 1, kFunct5Shift); + } + + int RvcFunct6Value() const; + + int RvcFunct4Value() const; + + int RvcFunct3Value() const; + + int RvcFunct2Value() const; + + int RvcFunct2BValue() const; + + inline int CsrValue() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kIType && + this->BaseOpcode() == SYSTEM); + return (this->Bits(kCsrShift + kCsrBits - 1, kCsrShift)); + } + + inline int RoundMode() const { + MOZ_ASSERT((this->InstructionType() == InstructionBase::kRType || + this->InstructionType() == InstructionBase::kR4Type) && + this->BaseOpcode() == OP_FP); + return this->Bits(kFunct3Shift + kFunct3Bits - 1, kFunct3Shift); + } + + inline int MemoryOrder(bool is_pred) const { + MOZ_ASSERT((this->InstructionType() == InstructionBase::kIType && + this->BaseOpcode() == MISC_MEM)); + if (is_pred) { + return this->Bits(kPredOrderShift + kMemOrderBits - 1, kPredOrderShift); + } else { + return this->Bits(kSuccOrderShift + kMemOrderBits - 1, kSuccOrderShift); + } + } + + inline int Imm12Value() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kIType); + int Value = this->Bits(kImm12Shift + kImm12Bits - 1, kImm12Shift); + return Value << 20 >> 20; + } + + inline int32_t Imm12SExtValue() const { + int32_t Value = this->Imm12Value() << 20 >> 20; + return Value; + } + + inline int BranchOffset() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kBType); + // | imm[12|10:5] | rs2 | rs1 | funct3 | imm[4:1|11] | opcode | + // 31 25 11 7 + uint32_t Bits = this->InstructionBits(); + int16_t imm13 = ((Bits & 0xf00) >> 7) | ((Bits & 0x7e000000) >> 20) | + ((Bits & 0x80) << 4) | ((Bits & 0x80000000) >> 19); + return imm13 << 19 >> 19; + } + + inline int StoreOffset() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kSType); + // | imm[11:5] | rs2 | rs1 | funct3 | imm[4:0] | opcode | + // 31 25 11 7 + uint32_t Bits = this->InstructionBits(); + int16_t imm12 = ((Bits & 0xf80) >> 7) | ((Bits & 0xfe000000) >> 20); + return imm12 << 20 >> 20; + } + + inline int Imm20UValue() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kUType); + // | imm[31:12] | rd | opcode | + // 31 12 + int32_t Bits = this->InstructionBits(); + return Bits >> 12; + } + + inline int Imm20JValue() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kJType); + // | imm[20|10:1|11|19:12] | rd | opcode | + // 31 12 + uint32_t Bits = this->InstructionBits(); + int32_t imm20 = ((Bits & 0x7fe00000) >> 20) | ((Bits & 0x100000) >> 9) | + (Bits & 0xff000) | ((Bits & 0x80000000) >> 11); + return imm20 << 11 >> 11; + } + + inline bool IsArithShift() const { + // Valid only for right shift operations + MOZ_ASSERT((this->BaseOpcode() == OP || this->BaseOpcode() == OP_32 || + this->BaseOpcode() == OP_IMM || + this->BaseOpcode() == OP_IMM_32) && + this->Funct3Value() == 0b101); + return this->InstructionBits() & 0x40000000; + } + + inline int Shamt() const { + // Valid only for shift instructions (SLLI, SRLI, SRAI) + MOZ_ASSERT((this->InstructionBits() & kBaseOpcodeMask) == OP_IMM && + (this->Funct3Value() == 0b001 || this->Funct3Value() == 0b101)); + // | 0A0000 | shamt | rs1 | funct3 | rd | opcode | + // 31 25 20 + return this->Bits(kImm12Shift + 5, kImm12Shift); + } + + inline int Shamt32() const { + // Valid only for shift instructions (SLLIW, SRLIW, SRAIW) + MOZ_ASSERT((this->InstructionBits() & kBaseOpcodeMask) == OP_IMM_32 && + (this->Funct3Value() == 0b001 || this->Funct3Value() == 0b101)); + // | 0A00000 | shamt | rs1 | funct3 | rd | opcode | + // 31 24 20 + return this->Bits(kImm12Shift + 4, kImm12Shift); + } + + inline int RvcImm6Value() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | imm[5] | rs1/rd | imm[4:0] | opcode | + // 15 12 6 2 + uint32_t Bits = this->InstructionBits(); + int32_t imm6 = ((Bits & 0x1000) >> 7) | ((Bits & 0x7c) >> 2); + return imm6 << 26 >> 26; + } + + inline int RvcImm6Addi16spValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | nzimm[9] | 2 | nzimm[4|6|8:7|5] | opcode | + // 15 12 6 2 + uint32_t Bits = this->InstructionBits(); + int32_t imm10 = ((Bits & 0x1000) >> 3) | ((Bits & 0x40) >> 2) | + ((Bits & 0x20) << 1) | ((Bits & 0x18) << 4) | + ((Bits & 0x4) << 3); + MOZ_ASSERT(imm10 != 0); + return imm10 << 22 >> 22; + } + + inline int RvcImm8Addi4spnValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | nzimm[11] | rd' | opcode | + // 15 13 5 2 + uint32_t Bits = this->InstructionBits(); + int32_t uimm10 = ((Bits & 0x20) >> 2) | ((Bits & 0x40) >> 4) | + ((Bits & 0x780) >> 1) | ((Bits & 0x1800) >> 7); + MOZ_ASSERT(uimm10 != 0); + return uimm10; + } + + inline int RvcShamt6() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | nzuimm[5] | rs1/rd | nzuimm[4:0] | opcode | + // 15 12 6 2 + int32_t imm6 = this->RvcImm6Value(); + return imm6 & 0x3f; + } + + inline int RvcImm6LwspValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | uimm[5] | rs1 | uimm[4:2|7:6] | opcode | + // 15 12 6 2 + uint32_t Bits = this->InstructionBits(); + int32_t imm8 = + ((Bits & 0x1000) >> 7) | ((Bits & 0x70) >> 2) | ((Bits & 0xc) << 4); + return imm8; + } + + inline int RvcImm6LdspValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | uimm[5] | rs1 | uimm[4:3|8:6] | opcode | + // 15 12 6 2 + uint32_t Bits = this->InstructionBits(); + int32_t imm9 = + ((Bits & 0x1000) >> 7) | ((Bits & 0x60) >> 2) | ((Bits & 0x1c) << 4); + return imm9; + } + + inline int RvcImm6SwspValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | uimm[5:2|7:6] | rs2 | opcode | + // 15 12 7 + uint32_t Bits = this->InstructionBits(); + int32_t imm8 = ((Bits & 0x1e00) >> 7) | ((Bits & 0x180) >> 1); + return imm8; + } + + inline int RvcImm6SdspValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | uimm[5:3|8:6] | rs2 | opcode | + // 15 12 7 + uint32_t Bits = this->InstructionBits(); + int32_t imm9 = ((Bits & 0x1c00) >> 7) | ((Bits & 0x380) >> 1); + return imm9; + } + + inline int RvcImm5WValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | imm[5:3] | rs1 | imm[2|6] | rd | opcode | + // 15 12 10 6 4 2 + uint32_t Bits = this->InstructionBits(); + int32_t imm7 = + ((Bits & 0x1c00) >> 7) | ((Bits & 0x40) >> 4) | ((Bits & 0x20) << 1); + return imm7; + } + + inline int RvcImm5DValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | imm[5:3] | rs1 | imm[7:6] | rd | opcode | + // 15 12 10 6 4 2 + uint32_t Bits = this->InstructionBits(); + int32_t imm8 = ((Bits & 0x1c00) >> 7) | ((Bits & 0x60) << 1); + return imm8; + } + + inline int RvcImm11CJValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | [11|4|9:8|10|6|7|3:1|5] | opcode | + // 15 12 2 + uint32_t Bits = this->InstructionBits(); + int32_t imm12 = ((Bits & 0x4) << 3) | ((Bits & 0x38) >> 2) | + ((Bits & 0x40) << 1) | ((Bits & 0x80) >> 1) | + ((Bits & 0x100) << 2) | ((Bits & 0x600) >> 1) | + ((Bits & 0x800) >> 7) | ((Bits & 0x1000) >> 1); + return imm12 << 20 >> 20; + } + + inline int RvcImm8BValue() const { + MOZ_ASSERT(this->IsShortInstruction()); + // | funct3 | imm[8|4:3] | rs1` | imm[7:6|2:1|5] | opcode | + // 15 12 10 7 2 + uint32_t Bits = this->InstructionBits(); + int32_t imm9 = ((Bits & 0x4) << 3) | ((Bits & 0x18) >> 2) | + ((Bits & 0x60) << 1) | ((Bits & 0xc00) >> 7) | + ((Bits & 0x1000) >> 4); + return imm9 << 23 >> 23; + } + + inline int vl_vs_width() { + int width = 0; + if ((this->InstructionBits() & kBaseOpcodeMask) != LOAD_FP && + (this->InstructionBits() & kBaseOpcodeMask) != STORE_FP) + return -1; + switch (this->InstructionBits() & (kRvvWidthMask | kRvvMewMask)) { + case 0x0: + width = 8; + break; + case 0x00005000: + width = 16; + break; + case 0x00006000: + width = 32; + break; + case 0x00007000: + width = 64; + break; + case 0x10000000: + width = 128; + break; + case 0x10005000: + width = 256; + break; + case 0x10006000: + width = 512; + break; + case 0x10007000: + width = 1024; + break; + default: + width = -1; + break; + } + return width; + } + + uint32_t Rvvzimm() const; + + uint32_t Rvvuimm() const; + + inline uint32_t RvvVsew() const { + uint32_t zimm = this->Rvvzimm(); + uint32_t vsew = (zimm >> 3) & 0x7; + return vsew; + } + + inline uint32_t RvvVlmul() const { + uint32_t zimm = this->Rvvzimm(); + uint32_t vlmul = zimm & 0x7; + return vlmul; + } + + inline uint8_t RvvVM() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kVType || + this->InstructionType() == InstructionBase::kIType || + this->InstructionType() == InstructionBase::kSType); + return this->Bits(kRvvVmShift + kRvvVmBits - 1, kRvvVmShift); + } + + inline const char* RvvSEW() const { + uint32_t vsew = this->RvvVsew(); + switch (vsew) { +#define CAST_VSEW(name) \ + case name: \ + return #name; + RVV_SEW(CAST_VSEW) + default: + return "unknown"; +#undef CAST_VSEW + } + } + + inline const char* RvvLMUL() const { + uint32_t vlmul = this->RvvVlmul(); + switch (vlmul) { +#define CAST_VLMUL(name) \ + case name: \ + return #name; + RVV_LMUL(CAST_VLMUL) + default: + return "unknown"; +#undef CAST_VLMUL + } + } + +#define sext(x, len) (((int32_t)(x) << (32 - len)) >> (32 - len)) +#define zext(x, len) (((uint32_t)(x) << (32 - len)) >> (32 - len)) + + inline int32_t RvvSimm5() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kVType); + return sext(this->Bits(kRvvImm5Shift + kRvvImm5Bits - 1, kRvvImm5Shift), + kRvvImm5Bits); + } + + inline uint32_t RvvUimm5() const { + MOZ_ASSERT(this->InstructionType() == InstructionBase::kVType); + uint32_t imm = this->Bits(kRvvImm5Shift + kRvvImm5Bits - 1, kRvvImm5Shift); + return zext(imm, kRvvImm5Bits); + } +#undef sext +#undef zext + inline bool AqValue() const { return this->Bits(kAqShift, kAqShift); } + + inline bool RlValue() const { return this->Bits(kRlShift, kRlShift); } +}; + +class Instruction : public InstructionGetters<InstructionBase> { + public: + // Instructions are read of out a code stream. The only way to get a + // reference to an instruction is to convert a pointer. There is no way + // to allocate or create instances of class Instruction. + // Use the At(pc) function to create references to Instruction. + static Instruction* At(byte* pc) { + return reinterpret_cast<Instruction*>(pc); + } + + private: + // We need to prevent the creation of instances of class Instruction. + Instruction() = delete; + Instruction(const Instruction&) = delete; + Instruction& operator=(const Instruction&) = delete; +}; + +// ----------------------------------------------------------------------------- +// Instructions. + +template <class P> +bool InstructionGetters<P>::IsTrap() const { + return (this->InstructionBits() == kBreakInstr); +} + +} // namespace jit +} // namespace js +#endif // jit_riscv64_constant_Base_constant_riscv__h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-a.h b/js/src/jit/riscv64/constant/Constant-riscv-a.h new file mode 100644 index 0000000000..718e607240 --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-a.h @@ -0,0 +1,43 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_a_h_ +#define jit_riscv64_constant_Constant_riscv64_a_h_ + +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { + +enum OpcodeRISCVA : uint32_t { + // RV32A Standard Extension + RO_LR_W = AMO | (0b010 << kFunct3Shift) | (0b00010 << kFunct5Shift), + RO_SC_W = AMO | (0b010 << kFunct3Shift) | (0b00011 << kFunct5Shift), + RO_AMOSWAP_W = AMO | (0b010 << kFunct3Shift) | (0b00001 << kFunct5Shift), + RO_AMOADD_W = AMO | (0b010 << kFunct3Shift) | (0b00000 << kFunct5Shift), + RO_AMOXOR_W = AMO | (0b010 << kFunct3Shift) | (0b00100 << kFunct5Shift), + RO_AMOAND_W = AMO | (0b010 << kFunct3Shift) | (0b01100 << kFunct5Shift), + RO_AMOOR_W = AMO | (0b010 << kFunct3Shift) | (0b01000 << kFunct5Shift), + RO_AMOMIN_W = AMO | (0b010 << kFunct3Shift) | (0b10000 << kFunct5Shift), + RO_AMOMAX_W = AMO | (0b010 << kFunct3Shift) | (0b10100 << kFunct5Shift), + RO_AMOMINU_W = AMO | (0b010 << kFunct3Shift) | (0b11000 << kFunct5Shift), + RO_AMOMAXU_W = AMO | (0b010 << kFunct3Shift) | (0b11100 << kFunct5Shift), + +#ifdef JS_CODEGEN_RISCV64 + // RV64A Standard Extension (in addition to RV32A) + RO_LR_D = AMO | (0b011 << kFunct3Shift) | (0b00010 << kFunct5Shift), + RO_SC_D = AMO | (0b011 << kFunct3Shift) | (0b00011 << kFunct5Shift), + RO_AMOSWAP_D = AMO | (0b011 << kFunct3Shift) | (0b00001 << kFunct5Shift), + RO_AMOADD_D = AMO | (0b011 << kFunct3Shift) | (0b00000 << kFunct5Shift), + RO_AMOXOR_D = AMO | (0b011 << kFunct3Shift) | (0b00100 << kFunct5Shift), + RO_AMOAND_D = AMO | (0b011 << kFunct3Shift) | (0b01100 << kFunct5Shift), + RO_AMOOR_D = AMO | (0b011 << kFunct3Shift) | (0b01000 << kFunct5Shift), + RO_AMOMIN_D = AMO | (0b011 << kFunct3Shift) | (0b10000 << kFunct5Shift), + RO_AMOMAX_D = AMO | (0b011 << kFunct3Shift) | (0b10100 << kFunct5Shift), + RO_AMOMINU_D = AMO | (0b011 << kFunct3Shift) | (0b11000 << kFunct5Shift), + RO_AMOMAXU_D = AMO | (0b011 << kFunct3Shift) | (0b11100 << kFunct5Shift), +#endif // JS_CODEGEN_RISCV64 +}; +} // namespace jit +} // namespace js + +#endif // jit_riscv64_constant_Constant_riscv64_a_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-c.h b/js/src/jit/riscv64/constant/Constant-riscv-c.h new file mode 100644 index 0000000000..a7d4792f5f --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-c.h @@ -0,0 +1,61 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_c_h_ +#define jit_riscv64_constant_Constant_riscv64_c_h_ + +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { +enum OpcodeRISCVC : uint32_t { + + RO_C_ADDI4SPN = C0 | (0b000 << kRvcFunct3Shift), + RO_C_ADDI16SP = C1 | (0b011 << kRvcFunct3Shift), + RO_C_LW = C0 | (0b010 << kRvcFunct3Shift), + RO_C_SW = C0 | (0b110 << kRvcFunct3Shift), + RO_C_NOP_ADDI = C1 | (0b000 << kRvcFunct3Shift), + RO_C_LI = C1 | (0b010 << kRvcFunct3Shift), + RO_C_SUB = C1 | (0b100011 << kRvcFunct6Shift) | (FUNCT2_0 << kRvcFunct2Shift), + RO_C_XOR = C1 | (0b100011 << kRvcFunct6Shift) | (FUNCT2_1 << kRvcFunct2Shift), + RO_C_OR = C1 | (0b100011 << kRvcFunct6Shift) | (FUNCT2_2 << kRvcFunct2Shift), + RO_C_AND = C1 | (0b100011 << kRvcFunct6Shift) | (FUNCT2_3 << kRvcFunct2Shift), + RO_C_LUI_ADD = C1 | (0b011 << kRvcFunct3Shift), + RO_C_MISC_ALU = C1 | (0b100 << kRvcFunct3Shift), + RO_C_J = C1 | (0b101 << kRvcFunct3Shift), + RO_C_BEQZ = C1 | (0b110 << kRvcFunct3Shift), + RO_C_BNEZ = C1 | (0b111 << kRvcFunct3Shift), + RO_C_SLLI = C2 | (0b000 << kRvcFunct3Shift), + RO_C_LWSP = C2 | (0b010 << kRvcFunct3Shift), + RO_C_JR_MV_ADD = C2 | (0b100 << kRvcFunct3Shift), + RO_C_JR = C2 | (0b1000 << kRvcFunct4Shift), + RO_C_MV = C2 | (0b1000 << kRvcFunct4Shift), + RO_C_EBREAK = C2 | (0b1001 << kRvcFunct4Shift), + RO_C_JALR = C2 | (0b1001 << kRvcFunct4Shift), + RO_C_ADD = C2 | (0b1001 << kRvcFunct4Shift), + RO_C_SWSP = C2 | (0b110 << kRvcFunct3Shift), + + RO_C_FSD = C0 | (0b101 << kRvcFunct3Shift), + RO_C_FLD = C0 | (0b001 << kRvcFunct3Shift), + RO_C_FLDSP = C2 | (0b001 << kRvcFunct3Shift), + RO_C_FSDSP = C2 | (0b101 << kRvcFunct3Shift), +#ifdef JS_CODEGEN_RISCV64 + RO_C_LD = C0 | (0b011 << kRvcFunct3Shift), + RO_C_SD = C0 | (0b111 << kRvcFunct3Shift), + RO_C_LDSP = C2 | (0b011 << kRvcFunct3Shift), + RO_C_SDSP = C2 | (0b111 << kRvcFunct3Shift), + RO_C_ADDIW = C1 | (0b001 << kRvcFunct3Shift), + RO_C_SUBW = + C1 | (0b100111 << kRvcFunct6Shift) | (FUNCT2_0 << kRvcFunct2Shift), + RO_C_ADDW = + C1 | (0b100111 << kRvcFunct6Shift) | (FUNCT2_1 << kRvcFunct2Shift), +#endif +#ifdef JS_CODEGEN_RISCV32 + RO_C_FLWSP = C2 | (0b011 << kRvcFunct3Shift), + RO_C_FSWSP = C2 | (0b111 << kRvcFunct3Shift), + RO_C_FLW = C0 | (0b011 << kRvcFunct3Shift), + RO_C_FSW = C0 | (0b111 << kRvcFunct3Shift), +#endif +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_constant_Constant_riscv64_c_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-d.h b/js/src/jit/riscv64/constant/Constant-riscv-d.h new file mode 100644 index 0000000000..d97e44ffe5 --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-d.h @@ -0,0 +1,55 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_d_h_ +#define jit_riscv64_constant_Constant_riscv64_d_h_ +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { + +enum OpcodeRISCVD : uint32_t { + // RV32D Standard Extension + RO_FLD = LOAD_FP | (0b011 << kFunct3Shift), + RO_FSD = STORE_FP | (0b011 << kFunct3Shift), + RO_FMADD_D = MADD | (0b01 << kFunct2Shift), + RO_FMSUB_D = MSUB | (0b01 << kFunct2Shift), + RO_FNMSUB_D = NMSUB | (0b01 << kFunct2Shift), + RO_FNMADD_D = NMADD | (0b01 << kFunct2Shift), + RO_FADD_D = OP_FP | (0b0000001 << kFunct7Shift), + RO_FSUB_D = OP_FP | (0b0000101 << kFunct7Shift), + RO_FMUL_D = OP_FP | (0b0001001 << kFunct7Shift), + RO_FDIV_D = OP_FP | (0b0001101 << kFunct7Shift), + RO_FSQRT_D = OP_FP | (0b0101101 << kFunct7Shift) | (0b00000 << kRs2Shift), + RO_FSGNJ_D = OP_FP | (0b000 << kFunct3Shift) | (0b0010001 << kFunct7Shift), + RO_FSGNJN_D = OP_FP | (0b001 << kFunct3Shift) | (0b0010001 << kFunct7Shift), + RO_FSQNJX_D = OP_FP | (0b010 << kFunct3Shift) | (0b0010001 << kFunct7Shift), + RO_FMIN_D = OP_FP | (0b000 << kFunct3Shift) | (0b0010101 << kFunct7Shift), + RO_FMAX_D = OP_FP | (0b001 << kFunct3Shift) | (0b0010101 << kFunct7Shift), + RO_FCVT_S_D = OP_FP | (0b0100000 << kFunct7Shift) | (0b00001 << kRs2Shift), + RO_FCVT_D_S = OP_FP | (0b0100001 << kFunct7Shift) | (0b00000 << kRs2Shift), + RO_FEQ_D = OP_FP | (0b010 << kFunct3Shift) | (0b1010001 << kFunct7Shift), + RO_FLT_D = OP_FP | (0b001 << kFunct3Shift) | (0b1010001 << kFunct7Shift), + RO_FLE_D = OP_FP | (0b000 << kFunct3Shift) | (0b1010001 << kFunct7Shift), + RO_FCLASS_D = OP_FP | (0b001 << kFunct3Shift) | (0b1110001 << kFunct7Shift) | + (0b00000 << kRs2Shift), + RO_FCVT_W_D = OP_FP | (0b1100001 << kFunct7Shift) | (0b00000 << kRs2Shift), + RO_FCVT_WU_D = OP_FP | (0b1100001 << kFunct7Shift) | (0b00001 << kRs2Shift), + RO_FCVT_D_W = OP_FP | (0b1101001 << kFunct7Shift) | (0b00000 << kRs2Shift), + RO_FCVT_D_WU = OP_FP | (0b1101001 << kFunct7Shift) | (0b00001 << kRs2Shift), + +#ifdef JS_CODEGEN_RISCV64 + // RV64D Standard Extension (in addition to RV32D) + RO_FCVT_L_D = OP_FP | (0b1100001 << kFunct7Shift) | (0b00010 << kRs2Shift), + RO_FCVT_LU_D = OP_FP | (0b1100001 << kFunct7Shift) | (0b00011 << kRs2Shift), + RO_FMV_X_D = OP_FP | (0b000 << kFunct3Shift) | (0b1110001 << kFunct7Shift) | + (0b00000 << kRs2Shift), + RO_FCVT_D_L = OP_FP | (0b1101001 << kFunct7Shift) | (0b00010 << kRs2Shift), + RO_FCVT_D_LU = OP_FP | (0b1101001 << kFunct7Shift) | (0b00011 << kRs2Shift), + RO_FMV_D_X = OP_FP | (0b000 << kFunct3Shift) | (0b1111001 << kFunct7Shift) | + (0b00000 << kRs2Shift), +#endif +}; +} // namespace jit +} // namespace js + +#endif // jit_riscv64_constant_Constant_riscv64_a_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-f.h b/js/src/jit/riscv64/constant/Constant-riscv-f.h new file mode 100644 index 0000000000..28c96394e2 --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-f.h @@ -0,0 +1,51 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_f_h_ +#define jit_riscv64_constant_Constant_riscv64_f_h_ +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { + +enum OpcodeRISCVF : uint32_t { + // RV32F Standard Extension + RO_FLW = LOAD_FP | (0b010 << kFunct3Shift), + RO_FSW = STORE_FP | (0b010 << kFunct3Shift), + RO_FMADD_S = MADD | (0b00 << kFunct2Shift), + RO_FMSUB_S = MSUB | (0b00 << kFunct2Shift), + RO_FNMSUB_S = NMSUB | (0b00 << kFunct2Shift), + RO_FNMADD_S = NMADD | (0b00 << kFunct2Shift), + RO_FADD_S = OP_FP | (0b0000000 << kFunct7Shift), + RO_FSUB_S = OP_FP | (0b0000100 << kFunct7Shift), + RO_FMUL_S = OP_FP | (0b0001000 << kFunct7Shift), + RO_FDIV_S = OP_FP | (0b0001100 << kFunct7Shift), + RO_FSQRT_S = OP_FP | (0b0101100 << kFunct7Shift) | (0b00000 << kRs2Shift), + RO_FSGNJ_S = OP_FP | (0b000 << kFunct3Shift) | (0b0010000 << kFunct7Shift), + RO_FSGNJN_S = OP_FP | (0b001 << kFunct3Shift) | (0b0010000 << kFunct7Shift), + RO_FSQNJX_S = OP_FP | (0b010 << kFunct3Shift) | (0b0010000 << kFunct7Shift), + RO_FMIN_S = OP_FP | (0b000 << kFunct3Shift) | (0b0010100 << kFunct7Shift), + RO_FMAX_S = OP_FP | (0b001 << kFunct3Shift) | (0b0010100 << kFunct7Shift), + RO_FCVT_W_S = OP_FP | (0b1100000 << kFunct7Shift) | (0b00000 << kRs2Shift), + RO_FCVT_WU_S = OP_FP | (0b1100000 << kFunct7Shift) | (0b00001 << kRs2Shift), + RO_FMV = OP_FP | (0b1110000 << kFunct7Shift) | (0b000 << kFunct3Shift) | + (0b00000 << kRs2Shift), + RO_FEQ_S = OP_FP | (0b010 << kFunct3Shift) | (0b1010000 << kFunct7Shift), + RO_FLT_S = OP_FP | (0b001 << kFunct3Shift) | (0b1010000 << kFunct7Shift), + RO_FLE_S = OP_FP | (0b000 << kFunct3Shift) | (0b1010000 << kFunct7Shift), + RO_FCLASS_S = OP_FP | (0b001 << kFunct3Shift) | (0b1110000 << kFunct7Shift), + RO_FCVT_S_W = OP_FP | (0b1101000 << kFunct7Shift) | (0b00000 << kRs2Shift), + RO_FCVT_S_WU = OP_FP | (0b1101000 << kFunct7Shift) | (0b00001 << kRs2Shift), + RO_FMV_W_X = OP_FP | (0b000 << kFunct3Shift) | (0b1111000 << kFunct7Shift), + +#ifdef JS_CODEGEN_RISCV64 + // RV64F Standard Extension (in addition to RV32F) + RO_FCVT_L_S = OP_FP | (0b1100000 << kFunct7Shift) | (0b00010 << kRs2Shift), + RO_FCVT_LU_S = OP_FP | (0b1100000 << kFunct7Shift) | (0b00011 << kRs2Shift), + RO_FCVT_S_L = OP_FP | (0b1101000 << kFunct7Shift) | (0b00010 << kRs2Shift), + RO_FCVT_S_LU = OP_FP | (0b1101000 << kFunct7Shift) | (0b00011 << kRs2Shift), +#endif // JS_CODEGEN_RISCV64 +}; +} // namespace jit +} // namespace js + +#endif // jit_riscv64_constant_Constant_riscv64_f_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-i.h b/js/src/jit/riscv64/constant/Constant-riscv-i.h new file mode 100644 index 0000000000..586ffd8a14 --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-i.h @@ -0,0 +1,73 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_i_h_ +#define jit_riscv64_constant_Constant_riscv64_i_h_ +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { + +enum OpcodeRISCV32I : uint32_t { + // Note use RO (RiscV Opcode) prefix + // RV32I Base Instruction Set + RO_LUI = LUI, + RO_AUIPC = AUIPC, + RO_JAL = JAL, + RO_JALR = JALR | (0b000 << kFunct3Shift), + RO_BEQ = BRANCH | (0b000 << kFunct3Shift), + RO_BNE = BRANCH | (0b001 << kFunct3Shift), + RO_BLT = BRANCH | (0b100 << kFunct3Shift), + RO_BGE = BRANCH | (0b101 << kFunct3Shift), + RO_BLTU = BRANCH | (0b110 << kFunct3Shift), + RO_BGEU = BRANCH | (0b111 << kFunct3Shift), + RO_LB = LOAD | (0b000 << kFunct3Shift), + RO_LH = LOAD | (0b001 << kFunct3Shift), + RO_LW = LOAD | (0b010 << kFunct3Shift), + RO_LBU = LOAD | (0b100 << kFunct3Shift), + RO_LHU = LOAD | (0b101 << kFunct3Shift), + RO_SB = STORE | (0b000 << kFunct3Shift), + RO_SH = STORE | (0b001 << kFunct3Shift), + RO_SW = STORE | (0b010 << kFunct3Shift), + RO_ADDI = OP_IMM | (0b000 << kFunct3Shift), + RO_SLTI = OP_IMM | (0b010 << kFunct3Shift), + RO_SLTIU = OP_IMM | (0b011 << kFunct3Shift), + RO_XORI = OP_IMM | (0b100 << kFunct3Shift), + RO_ORI = OP_IMM | (0b110 << kFunct3Shift), + RO_ANDI = OP_IMM | (0b111 << kFunct3Shift), + RO_SLLI = OP_IMM | (0b001 << kFunct3Shift), + RO_SRLI = OP_IMM | (0b101 << kFunct3Shift), + // RO_SRAI = OP_IMM | (0b101 << kFunct3Shift), // Same as SRLI, use func7 + RO_ADD = OP | (0b000 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_SUB = OP | (0b000 << kFunct3Shift) | (0b0100000 << kFunct7Shift), + RO_SLL = OP | (0b001 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_SLT = OP | (0b010 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_SLTU = OP | (0b011 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_XOR = OP | (0b100 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_SRL = OP | (0b101 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_SRA = OP | (0b101 << kFunct3Shift) | (0b0100000 << kFunct7Shift), + RO_OR = OP | (0b110 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_AND = OP | (0b111 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_FENCE = MISC_MEM | (0b000 << kFunct3Shift), + RO_ECALL = SYSTEM | (0b000 << kFunct3Shift), +// RO_EBREAK = SYSTEM | (0b000 << kFunct3Shift), // Same as ECALL, use imm12 + +#if JS_CODEGEN_RISCV64 + // RV64I Base Instruction Set (in addition to RV32I) + RO_LWU = LOAD | (0b110 << kFunct3Shift), + RO_LD = LOAD | (0b011 << kFunct3Shift), + RO_SD = STORE | (0b011 << kFunct3Shift), + RO_ADDIW = OP_IMM_32 | (0b000 << kFunct3Shift), + RO_SLLIW = OP_IMM_32 | (0b001 << kFunct3Shift), + RO_SRLIW = OP_IMM_32 | (0b101 << kFunct3Shift), + // RO_SRAIW = OP_IMM_32 | (0b101 << kFunct3Shift), // Same as SRLIW, use func7 + RO_ADDW = OP_32 | (0b000 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_SUBW = OP_32 | (0b000 << kFunct3Shift) | (0b0100000 << kFunct7Shift), + RO_SLLW = OP_32 | (0b001 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_SRLW = OP_32 | (0b101 << kFunct3Shift) | (0b0000000 << kFunct7Shift), + RO_SRAW = OP_32 | (0b101 << kFunct3Shift) | (0b0100000 << kFunct7Shift), +#endif +}; +} // namespace jit +} // namespace js + +#endif // jit_riscv64_constant_Constant_riscv64_i_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-m.h b/js/src/jit/riscv64/constant/Constant-riscv-m.h new file mode 100644 index 0000000000..81a69dab41 --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-m.h @@ -0,0 +1,34 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_m_h_ +#define jit_riscv64_constant_Constant_riscv64_m_h_ + +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { + +enum OpcodeRISCVM : uint32_t { + // RV32M Standard Extension + RO_MUL = OP | (0b000 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_MULH = OP | (0b001 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_MULHSU = OP | (0b010 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_MULHU = OP | (0b011 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_DIV = OP | (0b100 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_DIVU = OP | (0b101 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_REM = OP | (0b110 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_REMU = OP | (0b111 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + +#ifdef JS_CODEGEN_RISCV64 + // RV64M Standard Extension (in addition to RV32M) + RO_MULW = OP_32 | (0b000 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_DIVW = OP_32 | (0b100 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_DIVUW = OP_32 | (0b101 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_REMW = OP_32 | (0b110 << kFunct3Shift) | (0b0000001 << kFunct7Shift), + RO_REMUW = OP_32 | (0b111 << kFunct3Shift) | (0b0000001 << kFunct7Shift), +#endif +}; +} // namespace jit +} // namespace js + +#endif // jit_riscv64_extension_CONSTANT_RISCV_M_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-v.h b/js/src/jit/riscv64/constant/Constant-riscv-v.h new file mode 100644 index 0000000000..cca3540efd --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-v.h @@ -0,0 +1,508 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_v_h_ +#define jit_riscv64_constant_Constant_riscv64_v_h_ +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { + +namespace RVV { +enum TailAgnosticType { + ta = 0x1, // Tail agnostic + tu = 0x0, // Tail undisturbed +}; + +enum MaskAgnosticType { + ma = 0x1, // Mask agnostic + mu = 0x0, // Mask undisturbed +}; +enum MaskType { + Mask = 0x0, // use the mask + NoMask = 0x1, +}; +} // namespace RVV + +enum OpcodeRISCVV : uint32_t { + // RVV Extension + OP_IVV = OP_V | (0b000 << kFunct3Shift), + OP_FVV = OP_V | (0b001 << kFunct3Shift), + OP_MVV = OP_V | (0b010 << kFunct3Shift), + OP_IVI = OP_V | (0b011 << kFunct3Shift), + OP_IVX = OP_V | (0b100 << kFunct3Shift), + OP_FVF = OP_V | (0b101 << kFunct3Shift), + OP_MVX = OP_V | (0b110 << kFunct3Shift), + + RO_V_VSETVLI = OP_V | (0b111 << kFunct3Shift) | 0b0 << 31, + RO_V_VSETIVLI = OP_V | (0b111 << kFunct3Shift) | 0b11 << 30, + RO_V_VSETVL = OP_V | (0b111 << kFunct3Shift) | 0b1 << 31, + + // RVV LOAD/STORE + RO_V_VL = LOAD_FP | (0b00 << kRvvMopShift) | (0b000 << kRvvNfShift), + RO_V_VLS = LOAD_FP | (0b10 << kRvvMopShift) | (0b000 << kRvvNfShift), + RO_V_VLX = LOAD_FP | (0b11 << kRvvMopShift) | (0b000 << kRvvNfShift), + + RO_V_VS = STORE_FP | (0b00 << kRvvMopShift) | (0b000 << kRvvNfShift), + RO_V_VSS = STORE_FP | (0b10 << kRvvMopShift) | (0b000 << kRvvNfShift), + RO_V_VSX = STORE_FP | (0b11 << kRvvMopShift) | (0b000 << kRvvNfShift), + RO_V_VSU = STORE_FP | (0b01 << kRvvMopShift) | (0b000 << kRvvNfShift), + // THE kFunct6Shift is mop + RO_V_VLSEG2 = LOAD_FP | (0b00 << kRvvMopShift) | (0b001 << kRvvNfShift), + RO_V_VLSEG3 = LOAD_FP | (0b00 << kRvvMopShift) | (0b010 << kRvvNfShift), + RO_V_VLSEG4 = LOAD_FP | (0b00 << kRvvMopShift) | (0b011 << kRvvNfShift), + RO_V_VLSEG5 = LOAD_FP | (0b00 << kRvvMopShift) | (0b100 << kRvvNfShift), + RO_V_VLSEG6 = LOAD_FP | (0b00 << kRvvMopShift) | (0b101 << kRvvNfShift), + RO_V_VLSEG7 = LOAD_FP | (0b00 << kRvvMopShift) | (0b110 << kRvvNfShift), + RO_V_VLSEG8 = LOAD_FP | (0b00 << kRvvMopShift) | (0b111 << kRvvNfShift), + + RO_V_VSSEG2 = STORE_FP | (0b00 << kRvvMopShift) | (0b001 << kRvvNfShift), + RO_V_VSSEG3 = STORE_FP | (0b00 << kRvvMopShift) | (0b010 << kRvvNfShift), + RO_V_VSSEG4 = STORE_FP | (0b00 << kRvvMopShift) | (0b011 << kRvvNfShift), + RO_V_VSSEG5 = STORE_FP | (0b00 << kRvvMopShift) | (0b100 << kRvvNfShift), + RO_V_VSSEG6 = STORE_FP | (0b00 << kRvvMopShift) | (0b101 << kRvvNfShift), + RO_V_VSSEG7 = STORE_FP | (0b00 << kRvvMopShift) | (0b110 << kRvvNfShift), + RO_V_VSSEG8 = STORE_FP | (0b00 << kRvvMopShift) | (0b111 << kRvvNfShift), + + RO_V_VLSSEG2 = LOAD_FP | (0b10 << kRvvMopShift) | (0b001 << kRvvNfShift), + RO_V_VLSSEG3 = LOAD_FP | (0b10 << kRvvMopShift) | (0b010 << kRvvNfShift), + RO_V_VLSSEG4 = LOAD_FP | (0b10 << kRvvMopShift) | (0b011 << kRvvNfShift), + RO_V_VLSSEG5 = LOAD_FP | (0b10 << kRvvMopShift) | (0b100 << kRvvNfShift), + RO_V_VLSSEG6 = LOAD_FP | (0b10 << kRvvMopShift) | (0b101 << kRvvNfShift), + RO_V_VLSSEG7 = LOAD_FP | (0b10 << kRvvMopShift) | (0b110 << kRvvNfShift), + RO_V_VLSSEG8 = LOAD_FP | (0b10 << kRvvMopShift) | (0b111 << kRvvNfShift), + + RO_V_VSSSEG2 = STORE_FP | (0b10 << kRvvMopShift) | (0b001 << kRvvNfShift), + RO_V_VSSSEG3 = STORE_FP | (0b10 << kRvvMopShift) | (0b010 << kRvvNfShift), + RO_V_VSSSEG4 = STORE_FP | (0b10 << kRvvMopShift) | (0b011 << kRvvNfShift), + RO_V_VSSSEG5 = STORE_FP | (0b10 << kRvvMopShift) | (0b100 << kRvvNfShift), + RO_V_VSSSEG6 = STORE_FP | (0b10 << kRvvMopShift) | (0b101 << kRvvNfShift), + RO_V_VSSSEG7 = STORE_FP | (0b10 << kRvvMopShift) | (0b110 << kRvvNfShift), + RO_V_VSSSEG8 = STORE_FP | (0b10 << kRvvMopShift) | (0b111 << kRvvNfShift), + + RO_V_VLXSEG2 = LOAD_FP | (0b11 << kRvvMopShift) | (0b001 << kRvvNfShift), + RO_V_VLXSEG3 = LOAD_FP | (0b11 << kRvvMopShift) | (0b010 << kRvvNfShift), + RO_V_VLXSEG4 = LOAD_FP | (0b11 << kRvvMopShift) | (0b011 << kRvvNfShift), + RO_V_VLXSEG5 = LOAD_FP | (0b11 << kRvvMopShift) | (0b100 << kRvvNfShift), + RO_V_VLXSEG6 = LOAD_FP | (0b11 << kRvvMopShift) | (0b101 << kRvvNfShift), + RO_V_VLXSEG7 = LOAD_FP | (0b11 << kRvvMopShift) | (0b110 << kRvvNfShift), + RO_V_VLXSEG8 = LOAD_FP | (0b11 << kRvvMopShift) | (0b111 << kRvvNfShift), + + RO_V_VSXSEG2 = STORE_FP | (0b11 << kRvvMopShift) | (0b001 << kRvvNfShift), + RO_V_VSXSEG3 = STORE_FP | (0b11 << kRvvMopShift) | (0b010 << kRvvNfShift), + RO_V_VSXSEG4 = STORE_FP | (0b11 << kRvvMopShift) | (0b011 << kRvvNfShift), + RO_V_VSXSEG5 = STORE_FP | (0b11 << kRvvMopShift) | (0b100 << kRvvNfShift), + RO_V_VSXSEG6 = STORE_FP | (0b11 << kRvvMopShift) | (0b101 << kRvvNfShift), + RO_V_VSXSEG7 = STORE_FP | (0b11 << kRvvMopShift) | (0b110 << kRvvNfShift), + RO_V_VSXSEG8 = STORE_FP | (0b11 << kRvvMopShift) | (0b111 << kRvvNfShift), + + // RVV Vector Arithmetic Instruction + VADD_FUNCT6 = 0b000000, + RO_V_VADD_VI = OP_IVI | (VADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VADD_VV = OP_IVV | (VADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VADD_VX = OP_IVX | (VADD_FUNCT6 << kRvvFunct6Shift), + + VSUB_FUNCT6 = 0b000010, + RO_V_VSUB_VX = OP_IVX | (VSUB_FUNCT6 << kRvvFunct6Shift), + RO_V_VSUB_VV = OP_IVV | (VSUB_FUNCT6 << kRvvFunct6Shift), + + VDIVU_FUNCT6 = 0b100000, + RO_V_VDIVU_VX = OP_MVX | (VDIVU_FUNCT6 << kRvvFunct6Shift), + RO_V_VDIVU_VV = OP_MVV | (VDIVU_FUNCT6 << kRvvFunct6Shift), + + VDIV_FUNCT6 = 0b100001, + RO_V_VDIV_VX = OP_MVX | (VDIV_FUNCT6 << kRvvFunct6Shift), + RO_V_VDIV_VV = OP_MVV | (VDIV_FUNCT6 << kRvvFunct6Shift), + + VREMU_FUNCT6 = 0b100010, + RO_V_VREMU_VX = OP_MVX | (VREMU_FUNCT6 << kRvvFunct6Shift), + RO_V_VREMU_VV = OP_MVV | (VREMU_FUNCT6 << kRvvFunct6Shift), + + VREM_FUNCT6 = 0b100011, + RO_V_VREM_VX = OP_MVX | (VREM_FUNCT6 << kRvvFunct6Shift), + RO_V_VREM_VV = OP_MVV | (VREM_FUNCT6 << kRvvFunct6Shift), + + VMULHU_FUNCT6 = 0b100100, + RO_V_VMULHU_VX = OP_MVX | (VMULHU_FUNCT6 << kRvvFunct6Shift), + RO_V_VMULHU_VV = OP_MVV | (VMULHU_FUNCT6 << kRvvFunct6Shift), + + VMUL_FUNCT6 = 0b100101, + RO_V_VMUL_VX = OP_MVX | (VMUL_FUNCT6 << kRvvFunct6Shift), + RO_V_VMUL_VV = OP_MVV | (VMUL_FUNCT6 << kRvvFunct6Shift), + + VWMUL_FUNCT6 = 0b111011, + RO_V_VWMUL_VX = OP_MVX | (VWMUL_FUNCT6 << kRvvFunct6Shift), + RO_V_VWMUL_VV = OP_MVV | (VWMUL_FUNCT6 << kRvvFunct6Shift), + + VWMULU_FUNCT6 = 0b111000, + RO_V_VWMULU_VX = OP_MVX | (VWMULU_FUNCT6 << kRvvFunct6Shift), + RO_V_VWMULU_VV = OP_MVV | (VWMULU_FUNCT6 << kRvvFunct6Shift), + + VMULHSU_FUNCT6 = 0b100110, + RO_V_VMULHSU_VX = OP_MVX | (VMULHSU_FUNCT6 << kRvvFunct6Shift), + RO_V_VMULHSU_VV = OP_MVV | (VMULHSU_FUNCT6 << kRvvFunct6Shift), + + VMULH_FUNCT6 = 0b100111, + RO_V_VMULH_VX = OP_MVX | (VMULH_FUNCT6 << kRvvFunct6Shift), + RO_V_VMULH_VV = OP_MVV | (VMULH_FUNCT6 << kRvvFunct6Shift), + + VWADD_FUNCT6 = 0b110001, + RO_V_VWADD_VV = OP_MVV | (VWADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VWADD_VX = OP_MVX | (VWADD_FUNCT6 << kRvvFunct6Shift), + + VWADDU_FUNCT6 = 0b110000, + RO_V_VWADDU_VV = OP_MVV | (VWADDU_FUNCT6 << kRvvFunct6Shift), + RO_V_VWADDU_VX = OP_MVX | (VWADDU_FUNCT6 << kRvvFunct6Shift), + + VWADDUW_FUNCT6 = 0b110101, + RO_V_VWADDUW_VX = OP_MVX | (VWADDUW_FUNCT6 << kRvvFunct6Shift), + RO_V_VWADDUW_VV = OP_MVV | (VWADDUW_FUNCT6 << kRvvFunct6Shift), + + VCOMPRESS_FUNCT6 = 0b010111, + RO_V_VCOMPRESS_VV = OP_MVV | (VCOMPRESS_FUNCT6 << kRvvFunct6Shift), + + VSADDU_FUNCT6 = 0b100000, + RO_V_VSADDU_VI = OP_IVI | (VSADDU_FUNCT6 << kRvvFunct6Shift), + RO_V_VSADDU_VV = OP_IVV | (VSADDU_FUNCT6 << kRvvFunct6Shift), + RO_V_VSADDU_VX = OP_IVX | (VSADDU_FUNCT6 << kRvvFunct6Shift), + + VSADD_FUNCT6 = 0b100001, + RO_V_VSADD_VI = OP_IVI | (VSADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VSADD_VV = OP_IVV | (VSADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VSADD_VX = OP_IVX | (VSADD_FUNCT6 << kRvvFunct6Shift), + + VSSUB_FUNCT6 = 0b100011, + RO_V_VSSUB_VV = OP_IVV | (VSSUB_FUNCT6 << kRvvFunct6Shift), + RO_V_VSSUB_VX = OP_IVX | (VSSUB_FUNCT6 << kRvvFunct6Shift), + + VSSUBU_FUNCT6 = 0b100010, + RO_V_VSSUBU_VV = OP_IVV | (VSSUBU_FUNCT6 << kRvvFunct6Shift), + RO_V_VSSUBU_VX = OP_IVX | (VSSUBU_FUNCT6 << kRvvFunct6Shift), + + VRSUB_FUNCT6 = 0b000011, + RO_V_VRSUB_VX = OP_IVX | (VRSUB_FUNCT6 << kRvvFunct6Shift), + RO_V_VRSUB_VI = OP_IVI | (VRSUB_FUNCT6 << kRvvFunct6Shift), + + VMINU_FUNCT6 = 0b000100, + RO_V_VMINU_VX = OP_IVX | (VMINU_FUNCT6 << kRvvFunct6Shift), + RO_V_VMINU_VV = OP_IVV | (VMINU_FUNCT6 << kRvvFunct6Shift), + + VMIN_FUNCT6 = 0b000101, + RO_V_VMIN_VX = OP_IVX | (VMIN_FUNCT6 << kRvvFunct6Shift), + RO_V_VMIN_VV = OP_IVV | (VMIN_FUNCT6 << kRvvFunct6Shift), + + VMAXU_FUNCT6 = 0b000110, + RO_V_VMAXU_VX = OP_IVX | (VMAXU_FUNCT6 << kRvvFunct6Shift), + RO_V_VMAXU_VV = OP_IVV | (VMAXU_FUNCT6 << kRvvFunct6Shift), + + VMAX_FUNCT6 = 0b000111, + RO_V_VMAX_VX = OP_IVX | (VMAX_FUNCT6 << kRvvFunct6Shift), + RO_V_VMAX_VV = OP_IVV | (VMAX_FUNCT6 << kRvvFunct6Shift), + + VAND_FUNCT6 = 0b001001, + RO_V_VAND_VI = OP_IVI | (VAND_FUNCT6 << kRvvFunct6Shift), + RO_V_VAND_VV = OP_IVV | (VAND_FUNCT6 << kRvvFunct6Shift), + RO_V_VAND_VX = OP_IVX | (VAND_FUNCT6 << kRvvFunct6Shift), + + VOR_FUNCT6 = 0b001010, + RO_V_VOR_VI = OP_IVI | (VOR_FUNCT6 << kRvvFunct6Shift), + RO_V_VOR_VV = OP_IVV | (VOR_FUNCT6 << kRvvFunct6Shift), + RO_V_VOR_VX = OP_IVX | (VOR_FUNCT6 << kRvvFunct6Shift), + + VXOR_FUNCT6 = 0b001011, + RO_V_VXOR_VI = OP_IVI | (VXOR_FUNCT6 << kRvvFunct6Shift), + RO_V_VXOR_VV = OP_IVV | (VXOR_FUNCT6 << kRvvFunct6Shift), + RO_V_VXOR_VX = OP_IVX | (VXOR_FUNCT6 << kRvvFunct6Shift), + + VRGATHER_FUNCT6 = 0b001100, + RO_V_VRGATHER_VI = OP_IVI | (VRGATHER_FUNCT6 << kRvvFunct6Shift), + RO_V_VRGATHER_VV = OP_IVV | (VRGATHER_FUNCT6 << kRvvFunct6Shift), + RO_V_VRGATHER_VX = OP_IVX | (VRGATHER_FUNCT6 << kRvvFunct6Shift), + + VMV_FUNCT6 = 0b010111, + RO_V_VMV_VI = OP_IVI | (VMV_FUNCT6 << kRvvFunct6Shift), + RO_V_VMV_VV = OP_IVV | (VMV_FUNCT6 << kRvvFunct6Shift), + RO_V_VMV_VX = OP_IVX | (VMV_FUNCT6 << kRvvFunct6Shift), + RO_V_VFMV_VF = OP_FVF | (VMV_FUNCT6 << kRvvFunct6Shift), + + RO_V_VMERGE_VI = RO_V_VMV_VI, + RO_V_VMERGE_VV = RO_V_VMV_VV, + RO_V_VMERGE_VX = RO_V_VMV_VX, + + VMSEQ_FUNCT6 = 0b011000, + RO_V_VMSEQ_VI = OP_IVI | (VMSEQ_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSEQ_VV = OP_IVV | (VMSEQ_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSEQ_VX = OP_IVX | (VMSEQ_FUNCT6 << kRvvFunct6Shift), + + VMSNE_FUNCT6 = 0b011001, + RO_V_VMSNE_VI = OP_IVI | (VMSNE_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSNE_VV = OP_IVV | (VMSNE_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSNE_VX = OP_IVX | (VMSNE_FUNCT6 << kRvvFunct6Shift), + + VMSLTU_FUNCT6 = 0b011010, + RO_V_VMSLTU_VV = OP_IVV | (VMSLTU_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSLTU_VX = OP_IVX | (VMSLTU_FUNCT6 << kRvvFunct6Shift), + + VMSLT_FUNCT6 = 0b011011, + RO_V_VMSLT_VV = OP_IVV | (VMSLT_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSLT_VX = OP_IVX | (VMSLT_FUNCT6 << kRvvFunct6Shift), + + VMSLE_FUNCT6 = 0b011101, + RO_V_VMSLE_VI = OP_IVI | (VMSLE_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSLE_VV = OP_IVV | (VMSLE_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSLE_VX = OP_IVX | (VMSLE_FUNCT6 << kRvvFunct6Shift), + + VMSLEU_FUNCT6 = 0b011100, + RO_V_VMSLEU_VI = OP_IVI | (VMSLEU_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSLEU_VV = OP_IVV | (VMSLEU_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSLEU_VX = OP_IVX | (VMSLEU_FUNCT6 << kRvvFunct6Shift), + + VMSGTU_FUNCT6 = 0b011110, + RO_V_VMSGTU_VI = OP_IVI | (VMSGTU_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSGTU_VX = OP_IVX | (VMSGTU_FUNCT6 << kRvvFunct6Shift), + + VMSGT_FUNCT6 = 0b011111, + RO_V_VMSGT_VI = OP_IVI | (VMSGT_FUNCT6 << kRvvFunct6Shift), + RO_V_VMSGT_VX = OP_IVX | (VMSGT_FUNCT6 << kRvvFunct6Shift), + + VSLIDEUP_FUNCT6 = 0b001110, + RO_V_VSLIDEUP_VI = OP_IVI | (VSLIDEUP_FUNCT6 << kRvvFunct6Shift), + RO_V_VSLIDEUP_VX = OP_IVX | (VSLIDEUP_FUNCT6 << kRvvFunct6Shift), + + VSLIDEDOWN_FUNCT6 = 0b001111, + RO_V_VSLIDEDOWN_VI = OP_IVI | (VSLIDEDOWN_FUNCT6 << kRvvFunct6Shift), + RO_V_VSLIDEDOWN_VX = OP_IVX | (VSLIDEDOWN_FUNCT6 << kRvvFunct6Shift), + + VSRL_FUNCT6 = 0b101000, + RO_V_VSRL_VI = OP_IVI | (VSRL_FUNCT6 << kRvvFunct6Shift), + RO_V_VSRL_VV = OP_IVV | (VSRL_FUNCT6 << kRvvFunct6Shift), + RO_V_VSRL_VX = OP_IVX | (VSRL_FUNCT6 << kRvvFunct6Shift), + + VSRA_FUNCT6 = 0b101001, + RO_V_VSRA_VI = OP_IVI | (VSRA_FUNCT6 << kRvvFunct6Shift), + RO_V_VSRA_VV = OP_IVV | (VSRA_FUNCT6 << kRvvFunct6Shift), + RO_V_VSRA_VX = OP_IVX | (VSRA_FUNCT6 << kRvvFunct6Shift), + + VSLL_FUNCT6 = 0b100101, + RO_V_VSLL_VI = OP_IVI | (VSLL_FUNCT6 << kRvvFunct6Shift), + RO_V_VSLL_VV = OP_IVV | (VSLL_FUNCT6 << kRvvFunct6Shift), + RO_V_VSLL_VX = OP_IVX | (VSLL_FUNCT6 << kRvvFunct6Shift), + + VSMUL_FUNCT6 = 0b100111, + RO_V_VSMUL_VV = OP_IVV | (VSMUL_FUNCT6 << kRvvFunct6Shift), + RO_V_VSMUL_VX = OP_IVX | (VSMUL_FUNCT6 << kRvvFunct6Shift), + + VADC_FUNCT6 = 0b010000, + RO_V_VADC_VI = OP_IVI | (VADC_FUNCT6 << kRvvFunct6Shift), + RO_V_VADC_VV = OP_IVV | (VADC_FUNCT6 << kRvvFunct6Shift), + RO_V_VADC_VX = OP_IVX | (VADC_FUNCT6 << kRvvFunct6Shift), + + VMADC_FUNCT6 = 0b010001, + RO_V_VMADC_VI = OP_IVI | (VMADC_FUNCT6 << kRvvFunct6Shift), + RO_V_VMADC_VV = OP_IVV | (VMADC_FUNCT6 << kRvvFunct6Shift), + RO_V_VMADC_VX = OP_IVX | (VMADC_FUNCT6 << kRvvFunct6Shift), + + VWXUNARY0_FUNCT6 = 0b010000, + VRXUNARY0_FUNCT6 = 0b010000, + VMUNARY0_FUNCT6 = 0b010100, + + RO_V_VWXUNARY0 = OP_MVV | (VWXUNARY0_FUNCT6 << kRvvFunct6Shift), + RO_V_VRXUNARY0 = OP_MVX | (VRXUNARY0_FUNCT6 << kRvvFunct6Shift), + RO_V_VMUNARY0 = OP_MVV | (VMUNARY0_FUNCT6 << kRvvFunct6Shift), + + VID_V = 0b10001, + + VXUNARY0_FUNCT6 = 0b010010, + RO_V_VXUNARY0 = OP_MVV | (VXUNARY0_FUNCT6 << kRvvFunct6Shift), + + VWFUNARY0_FUNCT6 = 0b010000, + RO_V_VFMV_FS = OP_FVV | (VWFUNARY0_FUNCT6 << kRvvFunct6Shift), + + VRFUNARY0_FUNCT6 = 0b010000, + RO_V_VFMV_SF = OP_FVF | (VRFUNARY0_FUNCT6 << kRvvFunct6Shift), + + VREDMAXU_FUNCT6 = 0b000110, + RO_V_VREDMAXU = OP_MVV | (VREDMAXU_FUNCT6 << kRvvFunct6Shift), + VREDMAX_FUNCT6 = 0b000111, + RO_V_VREDMAX = OP_MVV | (VREDMAX_FUNCT6 << kRvvFunct6Shift), + + VREDMINU_FUNCT6 = 0b000100, + RO_V_VREDMINU = OP_MVV | (VREDMINU_FUNCT6 << kRvvFunct6Shift), + VREDMIN_FUNCT6 = 0b000101, + RO_V_VREDMIN = OP_MVV | (VREDMIN_FUNCT6 << kRvvFunct6Shift), + + VFUNARY0_FUNCT6 = 0b010010, + RO_V_VFUNARY0 = OP_FVV | (VFUNARY0_FUNCT6 << kRvvFunct6Shift), + VFUNARY1_FUNCT6 = 0b010011, + RO_V_VFUNARY1 = OP_FVV | (VFUNARY1_FUNCT6 << kRvvFunct6Shift), + + VFCVT_XU_F_V = 0b00000, + VFCVT_X_F_V = 0b00001, + VFCVT_F_XU_V = 0b00010, + VFCVT_F_X_V = 0b00011, + VFWCVT_XU_F_V = 0b01000, + VFWCVT_X_F_V = 0b01001, + VFWCVT_F_XU_V = 0b01010, + VFWCVT_F_X_V = 0b01011, + VFWCVT_F_F_V = 0b01100, + VFNCVT_F_F_W = 0b10100, + VFNCVT_X_F_W = 0b10001, + VFNCVT_XU_F_W = 0b10000, + + VFCLASS_V = 0b10000, + VFSQRT_V = 0b00000, + VFRSQRT7_V = 0b00100, + VFREC7_V = 0b00101, + + VFADD_FUNCT6 = 0b000000, + RO_V_VFADD_VV = OP_FVV | (VFADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VFADD_VF = OP_FVF | (VFADD_FUNCT6 << kRvvFunct6Shift), + + VFSUB_FUNCT6 = 0b000010, + RO_V_VFSUB_VV = OP_FVV | (VFSUB_FUNCT6 << kRvvFunct6Shift), + RO_V_VFSUB_VF = OP_FVF | (VFSUB_FUNCT6 << kRvvFunct6Shift), + + VFDIV_FUNCT6 = 0b100000, + RO_V_VFDIV_VV = OP_FVV | (VFDIV_FUNCT6 << kRvvFunct6Shift), + RO_V_VFDIV_VF = OP_FVF | (VFDIV_FUNCT6 << kRvvFunct6Shift), + + VFMUL_FUNCT6 = 0b100100, + RO_V_VFMUL_VV = OP_FVV | (VFMUL_FUNCT6 << kRvvFunct6Shift), + RO_V_VFMUL_VF = OP_FVF | (VFMUL_FUNCT6 << kRvvFunct6Shift), + + // Vector Widening Floating-Point Add/Subtract Instructions + VFWADD_FUNCT6 = 0b110000, + RO_V_VFWADD_VV = OP_FVV | (VFWADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWADD_VF = OP_FVF | (VFWADD_FUNCT6 << kRvvFunct6Shift), + + VFWSUB_FUNCT6 = 0b110010, + RO_V_VFWSUB_VV = OP_FVV | (VFWSUB_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWSUB_VF = OP_FVF | (VFWSUB_FUNCT6 << kRvvFunct6Shift), + + VFWADD_W_FUNCT6 = 0b110100, + RO_V_VFWADD_W_VV = OP_FVV | (VFWADD_W_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWADD_W_VF = OP_FVF | (VFWADD_W_FUNCT6 << kRvvFunct6Shift), + + VFWSUB_W_FUNCT6 = 0b110110, + RO_V_VFWSUB_W_VV = OP_FVV | (VFWSUB_W_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWSUB_W_VF = OP_FVF | (VFWSUB_W_FUNCT6 << kRvvFunct6Shift), + + // Vector Widening Floating-Point Reduction Instructions + VFWREDUSUM_FUNCT6 = 0b110001, + RO_V_VFWREDUSUM_VV = OP_FVV | (VFWREDUSUM_FUNCT6 << kRvvFunct6Shift), + + VFWREDOSUM_FUNCT6 = 0b110011, + RO_V_VFWREDOSUM_VV = OP_FVV | (VFWREDOSUM_FUNCT6 << kRvvFunct6Shift), + + // Vector Widening Floating-Point Multiply + VFWMUL_FUNCT6 = 0b111000, + RO_V_VFWMUL_VV = OP_FVV | (VFWMUL_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWMUL_VF = OP_FVF | (VFWMUL_FUNCT6 << kRvvFunct6Shift), + + VMFEQ_FUNCT6 = 0b011000, + RO_V_VMFEQ_VV = OP_FVV | (VMFEQ_FUNCT6 << kRvvFunct6Shift), + RO_V_VMFEQ_VF = OP_FVF | (VMFEQ_FUNCT6 << kRvvFunct6Shift), + + VMFNE_FUNCT6 = 0b011100, + RO_V_VMFNE_VV = OP_FVV | (VMFNE_FUNCT6 << kRvvFunct6Shift), + RO_V_VMFNE_VF = OP_FVF | (VMFNE_FUNCT6 << kRvvFunct6Shift), + + VMFLT_FUNCT6 = 0b011011, + RO_V_VMFLT_VV = OP_FVV | (VMFLT_FUNCT6 << kRvvFunct6Shift), + RO_V_VMFLT_VF = OP_FVF | (VMFLT_FUNCT6 << kRvvFunct6Shift), + + VMFLE_FUNCT6 = 0b011001, + RO_V_VMFLE_VV = OP_FVV | (VMFLE_FUNCT6 << kRvvFunct6Shift), + RO_V_VMFLE_VF = OP_FVF | (VMFLE_FUNCT6 << kRvvFunct6Shift), + + VMFGE_FUNCT6 = 0b011111, + RO_V_VMFGE_VF = OP_FVF | (VMFGE_FUNCT6 << kRvvFunct6Shift), + + VMFGT_FUNCT6 = 0b011101, + RO_V_VMFGT_VF = OP_FVF | (VMFGT_FUNCT6 << kRvvFunct6Shift), + + VFMAX_FUNCT6 = 0b000110, + RO_V_VFMAX_VV = OP_FVV | (VFMAX_FUNCT6 << kRvvFunct6Shift), + RO_V_VFMAX_VF = OP_FVF | (VFMAX_FUNCT6 << kRvvFunct6Shift), + + VFREDMAX_FUNCT6 = 0b0001111, + RO_V_VFREDMAX_VV = OP_FVV | (VFREDMAX_FUNCT6 << kRvvFunct6Shift), + + VFMIN_FUNCT6 = 0b000100, + RO_V_VFMIN_VV = OP_FVV | (VFMIN_FUNCT6 << kRvvFunct6Shift), + RO_V_VFMIN_VF = OP_FVF | (VFMIN_FUNCT6 << kRvvFunct6Shift), + + VFSGNJ_FUNCT6 = 0b001000, + RO_V_VFSGNJ_VV = OP_FVV | (VFSGNJ_FUNCT6 << kRvvFunct6Shift), + RO_V_VFSGNJ_VF = OP_FVF | (VFSGNJ_FUNCT6 << kRvvFunct6Shift), + + VFSGNJN_FUNCT6 = 0b001001, + RO_V_VFSGNJN_VV = OP_FVV | (VFSGNJN_FUNCT6 << kRvvFunct6Shift), + RO_V_VFSGNJN_VF = OP_FVF | (VFSGNJN_FUNCT6 << kRvvFunct6Shift), + + VFSGNJX_FUNCT6 = 0b001010, + RO_V_VFSGNJX_VV = OP_FVV | (VFSGNJX_FUNCT6 << kRvvFunct6Shift), + RO_V_VFSGNJX_VF = OP_FVF | (VFSGNJX_FUNCT6 << kRvvFunct6Shift), + + VFMADD_FUNCT6 = 0b101000, + RO_V_VFMADD_VV = OP_FVV | (VFMADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VFMADD_VF = OP_FVF | (VFMADD_FUNCT6 << kRvvFunct6Shift), + + VFNMADD_FUNCT6 = 0b101001, + RO_V_VFNMADD_VV = OP_FVV | (VFNMADD_FUNCT6 << kRvvFunct6Shift), + RO_V_VFNMADD_VF = OP_FVF | (VFNMADD_FUNCT6 << kRvvFunct6Shift), + + VFMSUB_FUNCT6 = 0b101010, + RO_V_VFMSUB_VV = OP_FVV | (VFMSUB_FUNCT6 << kRvvFunct6Shift), + RO_V_VFMSUB_VF = OP_FVF | (VFMSUB_FUNCT6 << kRvvFunct6Shift), + + VFNMSUB_FUNCT6 = 0b101011, + RO_V_VFNMSUB_VV = OP_FVV | (VFNMSUB_FUNCT6 << kRvvFunct6Shift), + RO_V_VFNMSUB_VF = OP_FVF | (VFNMSUB_FUNCT6 << kRvvFunct6Shift), + + VFMACC_FUNCT6 = 0b101100, + RO_V_VFMACC_VV = OP_FVV | (VFMACC_FUNCT6 << kRvvFunct6Shift), + RO_V_VFMACC_VF = OP_FVF | (VFMACC_FUNCT6 << kRvvFunct6Shift), + + VFNMACC_FUNCT6 = 0b101101, + RO_V_VFNMACC_VV = OP_FVV | (VFNMACC_FUNCT6 << kRvvFunct6Shift), + RO_V_VFNMACC_VF = OP_FVF | (VFNMACC_FUNCT6 << kRvvFunct6Shift), + + VFMSAC_FUNCT6 = 0b101110, + RO_V_VFMSAC_VV = OP_FVV | (VFMSAC_FUNCT6 << kRvvFunct6Shift), + RO_V_VFMSAC_VF = OP_FVF | (VFMSAC_FUNCT6 << kRvvFunct6Shift), + + VFNMSAC_FUNCT6 = 0b101111, + RO_V_VFNMSAC_VV = OP_FVV | (VFNMSAC_FUNCT6 << kRvvFunct6Shift), + RO_V_VFNMSAC_VF = OP_FVF | (VFNMSAC_FUNCT6 << kRvvFunct6Shift), + + // Vector Widening Floating-Point Fused Multiply-Add Instructions + VFWMACC_FUNCT6 = 0b111100, + RO_V_VFWMACC_VV = OP_FVV | (VFWMACC_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWMACC_VF = OP_FVF | (VFWMACC_FUNCT6 << kRvvFunct6Shift), + + VFWNMACC_FUNCT6 = 0b111101, + RO_V_VFWNMACC_VV = OP_FVV | (VFWNMACC_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWNMACC_VF = OP_FVF | (VFWNMACC_FUNCT6 << kRvvFunct6Shift), + + VFWMSAC_FUNCT6 = 0b111110, + RO_V_VFWMSAC_VV = OP_FVV | (VFWMSAC_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWMSAC_VF = OP_FVF | (VFWMSAC_FUNCT6 << kRvvFunct6Shift), + + VFWNMSAC_FUNCT6 = 0b111111, + RO_V_VFWNMSAC_VV = OP_FVV | (VFWNMSAC_FUNCT6 << kRvvFunct6Shift), + RO_V_VFWNMSAC_VF = OP_FVF | (VFWNMSAC_FUNCT6 << kRvvFunct6Shift), + + VNCLIP_FUNCT6 = 0b101111, + RO_V_VNCLIP_WV = OP_IVV | (VNCLIP_FUNCT6 << kRvvFunct6Shift), + RO_V_VNCLIP_WX = OP_IVX | (VNCLIP_FUNCT6 << kRvvFunct6Shift), + RO_V_VNCLIP_WI = OP_IVI | (VNCLIP_FUNCT6 << kRvvFunct6Shift), + + VNCLIPU_FUNCT6 = 0b101110, + RO_V_VNCLIPU_WV = OP_IVV | (VNCLIPU_FUNCT6 << kRvvFunct6Shift), + RO_V_VNCLIPU_WX = OP_IVX | (VNCLIPU_FUNCT6 << kRvvFunct6Shift), + RO_V_VNCLIPU_WI = OP_IVI | (VNCLIPU_FUNCT6 << kRvvFunct6Shift), +}; +} // namespace jit +} // namespace js + +#endif // jit_riscv64_constant_Constant_riscv64_v_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-zicsr.h b/js/src/jit/riscv64/constant/Constant-riscv-zicsr.h new file mode 100644 index 0000000000..6fecfa3d92 --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-zicsr.h @@ -0,0 +1,30 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_zicsr_h_ +#define jit_riscv64_constant_Constant_riscv64_zicsr_h_ + +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { +// RISCV CSR related bit mask and shift +const int kFcsrFlagsBits = 5; +const uint32_t kFcsrFlagsMask = (1 << kFcsrFlagsBits) - 1; +const int kFcsrFrmBits = 3; +const int kFcsrFrmShift = kFcsrFlagsBits; +const uint32_t kFcsrFrmMask = ((1 << kFcsrFrmBits) - 1) << kFcsrFrmShift; +const int kFcsrBits = kFcsrFlagsBits + kFcsrFrmBits; +const uint32_t kFcsrMask = kFcsrFlagsMask | kFcsrFrmMask; + +enum OpcodeRISCVZICSR : uint32_t { + // RV32/RV64 Zicsr Standard Extension + RO_CSRRW = SYSTEM | (0b001 << kFunct3Shift), + RO_CSRRS = SYSTEM | (0b010 << kFunct3Shift), + RO_CSRRC = SYSTEM | (0b011 << kFunct3Shift), + RO_CSRRWI = SYSTEM | (0b101 << kFunct3Shift), + RO_CSRRSI = SYSTEM | (0b110 << kFunct3Shift), + RO_CSRRCI = SYSTEM | (0b111 << kFunct3Shift), +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_constant_Constant_riscv64_zicsr_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv-zifencei.h b/js/src/jit/riscv64/constant/Constant-riscv-zifencei.h new file mode 100644 index 0000000000..be01cd0ae0 --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv-zifencei.h @@ -0,0 +1,15 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_Constant_riscv64_zifencei_h_ +#define jit_riscv64_constant_Constant_riscv64_zifencei_h_ + +#include "jit/riscv64/constant/Base-constant-riscv.h" +namespace js { +namespace jit { +enum OpcodeRISCVIFENCEI : uint32_t { + RO_FENCE_I = MISC_MEM | (0b001 << kFunct3Shift), +}; +} +} // namespace js +#endif // jit_riscv64_constant_Constant_riscv64_zifencei_h_ diff --git a/js/src/jit/riscv64/constant/Constant-riscv64.h b/js/src/jit/riscv64/constant/Constant-riscv64.h new file mode 100644 index 0000000000..b9b1f894e7 --- /dev/null +++ b/js/src/jit/riscv64/constant/Constant-riscv64.h @@ -0,0 +1,68 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef jit_riscv64_constant_Constant_riscv64_h +#define jit_riscv64_constant_Constant_riscv64_h +#include "mozilla/Assertions.h" +#include "mozilla/Types.h" + +#include <stdio.h> + +#include "jit/riscv64/constant/Base-constant-riscv.h" +#include "jit/riscv64/constant/Constant-riscv-a.h" +#include "jit/riscv64/constant/Constant-riscv-c.h" +#include "jit/riscv64/constant/Constant-riscv-d.h" +#include "jit/riscv64/constant/Constant-riscv-f.h" +#include "jit/riscv64/constant/Constant-riscv-i.h" +#include "jit/riscv64/constant/Constant-riscv-m.h" +#include "jit/riscv64/constant/Constant-riscv-v.h" +#include "jit/riscv64/constant/Constant-riscv-zicsr.h" +#include "jit/riscv64/constant/Constant-riscv-zifencei.h" + +namespace js { +namespace jit { + +// A reasonable (ie, safe) buffer size for the disassembly of a single +// instruction. +const int ReasonableBufferSize = 256; + +// Difference between address of current opcode and value read from pc +// register. +static constexpr int kPcLoadDelta = 4; + +// Bits available for offset field in branches +static constexpr int kBranchOffsetBits = 13; + +// Bits available for offset field in jump +static constexpr int kJumpOffsetBits = 21; + +// Bits available for offset field in compresed jump +static constexpr int kCJalOffsetBits = 12; + +// Bits available for offset field in 4 branch +static constexpr int kCBranchOffsetBits = 9; + +// Max offset for b instructions with 12-bit offset field (multiple of 2) +static constexpr int kMaxBranchOffset = (1 << (kBranchOffsetBits - 1)) - 1; + +static constexpr int kCBranchOffset = (1 << (kCBranchOffsetBits - 1)) - 1; +// Max offset for jal instruction with 20-bit offset field (multiple of 2) +static constexpr int kMaxJumpOffset = (1 << (kJumpOffsetBits - 1)) - 1; + +static constexpr int kCJumpOffset = (1 << (kCJalOffsetBits - 1)) - 1; + +static constexpr int kTrampolineSlotsSize = 2 * kInstrSize; + +static_assert(kCJalOffsetBits == kOffset12); +static_assert(kCBranchOffsetBits == kOffset9); +static_assert(kJumpOffsetBits == kOffset21); +static_assert(kBranchOffsetBits == kOffset13); +// Vector as used by the original code to allow for minimal modification. +// Functions exactly like a character array with helper methods. +} // namespace jit +} // namespace js + +#endif // jit_riscv64_constant_Constant_riscv64_h diff --git a/js/src/jit/riscv64/constant/util-riscv64.h b/js/src/jit/riscv64/constant/util-riscv64.h new file mode 100644 index 0000000000..089e0f3b94 --- /dev/null +++ b/js/src/jit/riscv64/constant/util-riscv64.h @@ -0,0 +1,82 @@ + +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_constant_util_riscv64__h_ +#define jit_riscv64_constant_util_riscv64__h_ +#include <stdarg.h> +#include <stdio.h> +#include <string.h> +namespace js { +namespace jit { +template <typename T> +class V8Vector { + public: + V8Vector() : start_(nullptr), length_(0) {} + V8Vector(T* data, int length) : start_(data), length_(length) { + MOZ_ASSERT(length == 0 || (length > 0 && data != nullptr)); + } + + // Returns the length of the vector. + int length() const { return length_; } + + // Returns the pointer to the start of the data in the vector. + T* start() const { return start_; } + + // Access individual vector elements - checks bounds in debug mode. + T& operator[](int index) const { + MOZ_ASSERT(0 <= index && index < length_); + return start_[index]; + } + + inline V8Vector<T> operator+(int offset) { + MOZ_ASSERT(offset < length_); + return V8Vector<T>(start_ + offset, length_ - offset); + } + + private: + T* start_; + int length_; +}; + +template <typename T, int kSize> +class EmbeddedVector : public V8Vector<T> { + public: + EmbeddedVector() : V8Vector<T>(buffer_, kSize) {} + + explicit EmbeddedVector(T initial_value) : V8Vector<T>(buffer_, kSize) { + for (int i = 0; i < kSize; ++i) { + buffer_[i] = initial_value; + } + } + + // When copying, make underlying Vector to reference our buffer. + EmbeddedVector(const EmbeddedVector& rhs) : V8Vector<T>(rhs) { + MemCopy(buffer_, rhs.buffer_, sizeof(T) * kSize); + this->set_start(buffer_); + } + + EmbeddedVector& operator=(const EmbeddedVector& rhs) { + if (this == &rhs) return *this; + V8Vector<T>::operator=(rhs); + MemCopy(buffer_, rhs.buffer_, sizeof(T) * kSize); + this->set_start(buffer_); + return *this; + } + + private: + T buffer_[kSize]; +}; + +// Helper function for printing to a Vector. +static inline int MOZ_FORMAT_PRINTF(2, 3) + SNPrintF(V8Vector<char> str, const char* format, ...) { + va_list args; + va_start(args, format); + int result = vsnprintf(str.start(), str.length(), format, args); + va_end(args); + return result; +} +} // namespace jit +} // namespace js +#endif diff --git a/js/src/jit/riscv64/disasm/Disasm-riscv64.cpp b/js/src/jit/riscv64/disasm/Disasm-riscv64.cpp new file mode 100644 index 0000000000..bd9770d074 --- /dev/null +++ b/js/src/jit/riscv64/disasm/Disasm-riscv64.cpp @@ -0,0 +1,2155 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + */ +// Copyright 2011 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +// A Disassembler object is used to disassemble a block of code instruction by +// instruction. The default implementation of the NameConverter object can be +// overriden to modify register names or to do symbol lookup on addresses. +// +// The example below will disassemble a block of code and print it to stdout. +// +// disasm::NameConverter converter; +// disasm::Disassembler d(converter); +// for (uint8_t* pc = begin; pc < end;) { +// disasm::EmbeddedVector<char, disasm::ReasonableBufferSize> buffer; +// uint8_t* prev_pc = pc; +// pc += d.InstructionDecode(buffer, pc); +// printf("%p %08x %s\n", +// prev_pc, *reinterpret_cast<int32_t*>(prev_pc), buffer); +// } +// +// The Disassembler class also has a convenience method to disassemble a block +// of code into a FILE*, meaning that the above functionality could also be +// achieved by just calling Disassembler::Disassemble(stdout, begin, end); + +#include "jit/riscv64/disasm/Disasm-riscv64.h" + +#include <stdarg.h> +#include <stdio.h> +#include <string.h> + +#include "jit/riscv64/Assembler-riscv64.h" + +namespace js { +namespace jit { +namespace disasm { + +#define UNSUPPORTED_RISCV() printf("Unsupported instruction %d.\n", __LINE__) +//------------------------------------------------------------------------------ + +// Decoder decodes and disassembles instructions into an output buffer. +// It uses the converter to convert register names and call destinations into +// more informative description. +class Decoder { + public: + Decoder(const disasm::NameConverter& converter, V8Vector<char> out_buffer) + : converter_(converter), out_buffer_(out_buffer), out_buffer_pos_(0) { + out_buffer_[out_buffer_pos_] = '\0'; + } + + ~Decoder() {} + + // Writes one disassembled instruction into 'buffer' (0-terminated). + // Returns the length of the disassembled machine instruction in bytes. + int InstructionDecode(uint8_t* instruction); + + static bool IsConstantPoolAt(uint8_t* instr_ptr); + static int ConstantPoolSizeAt(uint8_t* instr_ptr); + + private: + // Bottleneck functions to print into the out_buffer. + void PrintChar(const char ch); + void Print(const char* str); + + // Printing of common values. + void PrintRegister(int reg); + void PrintFPURegister(int freg); + void PrintVRegister(int reg); + void PrintFPUStatusRegister(int freg); + void PrintRs1(Instruction* instr); + void PrintRs2(Instruction* instr); + void PrintRd(Instruction* instr); + void PrintUimm(Instruction* instr); + void PrintVs1(Instruction* instr); + void PrintVs2(Instruction* instr); + void PrintVd(Instruction* instr); + void PrintFRs1(Instruction* instr); + void PrintFRs2(Instruction* instr); + void PrintFRs3(Instruction* instr); + void PrintFRd(Instruction* instr); + void PrintImm12(Instruction* instr); + void PrintImm12X(Instruction* instr); + void PrintImm20U(Instruction* instr); + void PrintImm20J(Instruction* instr); + void PrintShamt(Instruction* instr); + void PrintShamt32(Instruction* instr); + void PrintRvcImm6(Instruction* instr); + void PrintRvcImm6U(Instruction* instr); + void PrintRvcImm6Addi16sp(Instruction* instr); + void PrintRvcShamt(Instruction* instr); + void PrintRvcImm6Ldsp(Instruction* instr); + void PrintRvcImm6Lwsp(Instruction* instr); + void PrintRvcImm6Sdsp(Instruction* instr); + void PrintRvcImm6Swsp(Instruction* instr); + void PrintRvcImm5W(Instruction* instr); + void PrintRvcImm5D(Instruction* instr); + void PrintRvcImm8Addi4spn(Instruction* instr); + void PrintRvcImm11CJ(Instruction* instr); + void PrintRvcImm8B(Instruction* instr); + void PrintRvvVm(Instruction* instr); + void PrintAcquireRelease(Instruction* instr); + void PrintBranchOffset(Instruction* instr); + void PrintStoreOffset(Instruction* instr); + void PrintCSRReg(Instruction* instr); + void PrintRvvSEW(Instruction* instr); + void PrintRvvLMUL(Instruction* instr); + void PrintRvvSimm5(Instruction* instr); + void PrintRvvUimm5(Instruction* instr); + void PrintRoundingMode(Instruction* instr); + void PrintMemoryOrder(Instruction* instr, bool is_pred); + + // Each of these functions decodes one particular instruction type. + void DecodeRType(Instruction* instr); + void DecodeR4Type(Instruction* instr); + void DecodeRAType(Instruction* instr); + void DecodeRFPType(Instruction* instr); + void DecodeIType(Instruction* instr); + void DecodeSType(Instruction* instr); + void DecodeBType(Instruction* instr); + void DecodeUType(Instruction* instr); + void DecodeJType(Instruction* instr); + void DecodeCRType(Instruction* instr); + void DecodeCAType(Instruction* instr); + void DecodeCIType(Instruction* instr); + void DecodeCIWType(Instruction* instr); + void DecodeCSSType(Instruction* instr); + void DecodeCLType(Instruction* instr); + void DecodeCSType(Instruction* instr); + void DecodeCJType(Instruction* instr); + void DecodeCBType(Instruction* instr); + + // Printing of instruction name. + void PrintInstructionName(Instruction* instr); + void PrintTarget(Instruction* instr); + + // Handle formatting of instructions and their options. + int FormatRegister(Instruction* instr, const char* option); + int FormatFPURegisterOrRoundMode(Instruction* instr, const char* option); + int FormatRvcRegister(Instruction* instr, const char* option); + int FormatRvcImm(Instruction* instr, const char* option); + int FormatOption(Instruction* instr, const char* option); + void Format(Instruction* instr, const char* format); + void Unknown(Instruction* instr); + + int switch_sew(Instruction* instr); + int switch_nf(Instruction* instr); + + const disasm::NameConverter& converter_; + V8Vector<char> out_buffer_; + int out_buffer_pos_; + + // Disallow copy and assign. + Decoder(const Decoder&) = delete; + void operator=(const Decoder&) = delete; +}; + +// Support for assertions in the Decoder formatting functions. +#define STRING_STARTS_WITH(string, compare_string) \ + (strncmp(string, compare_string, strlen(compare_string)) == 0) + +// Append the ch to the output buffer. +void Decoder::PrintChar(const char ch) { out_buffer_[out_buffer_pos_++] = ch; } + +// Append the str to the output buffer. +void Decoder::Print(const char* str) { + char cur = *str++; + while (cur != '\0' && (out_buffer_pos_ < int(out_buffer_.length() - 1))) { + PrintChar(cur); + cur = *str++; + } + out_buffer_[out_buffer_pos_] = 0; +} + +int Decoder::switch_nf(Instruction* instr) { + int nf = 0; + switch (instr->InstructionBits() & kRvvNfMask) { + case 0x20000000: + nf = 2; + break; + case 0x40000000: + nf = 3; + break; + case 0x60000000: + nf = 4; + break; + case 0x80000000: + nf = 5; + break; + case 0xa0000000: + nf = 6; + break; + case 0xc0000000: + nf = 7; + break; + case 0xe0000000: + nf = 8; + break; + } + return nf; +} + +int Decoder::switch_sew(Instruction* instr) { + int width = 0; + if ((instr->InstructionBits() & kBaseOpcodeMask) != LOAD_FP && + (instr->InstructionBits() & kBaseOpcodeMask) != STORE_FP) + return -1; + switch (instr->InstructionBits() & (kRvvWidthMask | kRvvMewMask)) { + case 0x0: + width = 8; + break; + case 0x00005000: + width = 16; + break; + case 0x00006000: + width = 32; + break; + case 0x00007000: + width = 64; + break; + case 0x10000000: + width = 128; + break; + case 0x10005000: + width = 256; + break; + case 0x10006000: + width = 512; + break; + case 0x10007000: + width = 1024; + break; + default: + width = -1; + break; + } + return width; +} + +// Handle all register based formatting in this function to reduce the +// complexity of FormatOption. +int Decoder::FormatRegister(Instruction* instr, const char* format) { + MOZ_ASSERT(format[0] == 'r'); + if (format[1] == 's') { // 'rs[12]: Rs register. + if (format[2] == '1') { + int reg = instr->Rs1Value(); + PrintRegister(reg); + return 3; + } else if (format[2] == '2') { + int reg = instr->Rs2Value(); + PrintRegister(reg); + return 3; + } + MOZ_CRASH(); + } else if (format[1] == 'd') { // 'rd: rd register. + int reg = instr->RdValue(); + PrintRegister(reg); + return 2; + } + MOZ_CRASH(); +} + +// Handle all FPUregister based formatting in this function to reduce the +// complexity of FormatOption. +int Decoder::FormatFPURegisterOrRoundMode(Instruction* instr, + const char* format) { + MOZ_ASSERT(format[0] == 'f'); + if (format[1] == 's') { // 'fs[1-3]: Rs register. + if (format[2] == '1') { + int reg = instr->Rs1Value(); + PrintFPURegister(reg); + return 3; + } else if (format[2] == '2') { + int reg = instr->Rs2Value(); + PrintFPURegister(reg); + return 3; + } else if (format[2] == '3') { + int reg = instr->Rs3Value(); + PrintFPURegister(reg); + return 3; + } + MOZ_CRASH(); + } else if (format[1] == 'd') { // 'fd: fd register. + int reg = instr->RdValue(); + PrintFPURegister(reg); + return 2; + } else if (format[1] == 'r') { // 'frm + MOZ_ASSERT(STRING_STARTS_WITH(format, "frm")); + PrintRoundingMode(instr); + return 3; + } + MOZ_CRASH(); +} + +// Handle all C extension register based formatting in this function to reduce +// the complexity of FormatOption. +int Decoder::FormatRvcRegister(Instruction* instr, const char* format) { + MOZ_ASSERT(format[0] == 'C'); + MOZ_ASSERT(format[1] == 'r' || format[1] == 'f'); + if (format[2] == 's') { // 'Crs[12]: Rs register. + if (format[3] == '1') { + if (format[4] == 's') { // 'Crs1s: 3-bits register + int reg = instr->RvcRs1sValue(); + if (format[1] == 'r') { + PrintRegister(reg); + } else if (format[1] == 'f') { + PrintFPURegister(reg); + } + return 5; + } + int reg = instr->RvcRs1Value(); + if (format[1] == 'r') { + PrintRegister(reg); + } else if (format[1] == 'f') { + PrintFPURegister(reg); + } + return 4; + } else if (format[3] == '2') { + if (format[4] == 's') { // 'Crs2s: 3-bits register + int reg = instr->RvcRs2sValue(); + if (format[1] == 'r') { + PrintRegister(reg); + } else if (format[1] == 'f') { + PrintFPURegister(reg); + } + return 5; + } + int reg = instr->RvcRs2Value(); + if (format[1] == 'r') { + PrintRegister(reg); + } else if (format[1] == 'f') { + PrintFPURegister(reg); + } + return 4; + } + MOZ_CRASH(); + } else if (format[2] == 'd') { // 'Crd: rd register. + int reg = instr->RvcRdValue(); + if (format[1] == 'r') { + PrintRegister(reg); + } else if (format[1] == 'f') { + PrintFPURegister(reg); + } + return 3; + } + MOZ_CRASH(); +} + +// Handle all C extension immediates based formatting in this function to reduce +// the complexity of FormatOption. +int Decoder::FormatRvcImm(Instruction* instr, const char* format) { + // TODO(riscv): add other rvc imm format + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm")); + if (format[4] == '6') { + if (format[5] == 'U') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm6U")); + PrintRvcImm6U(instr); + return 6; + } else if (format[5] == 'A') { + if (format[9] == '1' && format[10] == '6') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm6Addi16sp")); + PrintRvcImm6Addi16sp(instr); + return 13; + } + MOZ_CRASH(); + } else if (format[5] == 'L') { + if (format[6] == 'd') { + if (format[7] == 's') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm6Ldsp")); + PrintRvcImm6Ldsp(instr); + return 9; + } + } else if (format[6] == 'w') { + if (format[7] == 's') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm6Lwsp")); + PrintRvcImm6Lwsp(instr); + return 9; + } + } + MOZ_CRASH(); + } else if (format[5] == 'S') { + if (format[6] == 'w') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm6Swsp")); + PrintRvcImm6Swsp(instr); + return 9; + } else if (format[6] == 'd') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm6Sdsp")); + PrintRvcImm6Sdsp(instr); + return 9; + } + MOZ_CRASH(); + } + PrintRvcImm6(instr); + return 5; + } else if (format[4] == '5') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm5")); + if (format[5] == 'W') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm5W")); + PrintRvcImm5W(instr); + return 6; + } else if (format[5] == 'D') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm5D")); + PrintRvcImm5D(instr); + return 6; + } + MOZ_CRASH(); + } else if (format[4] == '8') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm8")); + if (format[5] == 'A') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm8Addi4spn")); + PrintRvcImm8Addi4spn(instr); + return 13; + } else if (format[5] == 'B') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm8B")); + PrintRvcImm8B(instr); + return 6; + } + MOZ_CRASH(); + } else if (format[4] == '1') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm1")); + if (format[5] == '1') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cimm11CJ")); + PrintRvcImm11CJ(instr); + return 8; + } + MOZ_CRASH(); + } + MOZ_CRASH(); +} + +// FormatOption takes a formatting string and interprets it based on +// the current instructions. The format string points to the first +// character of the option string (the option escape has already been +// consumed by the caller.) FormatOption returns the number of +// characters that were consumed from the formatting string. +int Decoder::FormatOption(Instruction* instr, const char* format) { + switch (format[0]) { + case 'C': { // `C extension + if (format[1] == 'r' || format[1] == 'f') { + return FormatRvcRegister(instr, format); + } else if (format[1] == 'i') { + return FormatRvcImm(instr, format); + } else if (format[1] == 's') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "Cshamt")); + PrintRvcShamt(instr); + return 6; + } + MOZ_CRASH(); + } + case 'c': { // `csr: CSR registers + if (format[1] == 's') { + if (format[2] == 'r') { + PrintCSRReg(instr); + return 3; + } + } + MOZ_CRASH(); + } + case 'i': { // 'imm12, 'imm12x, 'imm20U, or 'imm20J: Immediates. + if (format[3] == '1') { + if (format[4] == '2') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "imm12")); + if (format[5] == 'x') { + PrintImm12X(instr); + return 6; + } + PrintImm12(instr); + return 5; + } + } else if (format[3] == '2' && format[4] == '0') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "imm20")); + switch (format[5]) { + case 'U': + MOZ_ASSERT(STRING_STARTS_WITH(format, "imm20U")); + PrintImm20U(instr); + break; + case 'J': + MOZ_ASSERT(STRING_STARTS_WITH(format, "imm20J")); + PrintImm20J(instr); + break; + } + return 6; + } + MOZ_CRASH(); + } + case 'o': { // 'offB or 'offS: Offsets. + if (format[3] == 'B') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "offB")); + PrintBranchOffset(instr); + return 4; + } else if (format[3] == 'S') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "offS")); + PrintStoreOffset(instr); + return 4; + } + MOZ_CRASH(); + } + case 'r': { // 'r: registers. + return FormatRegister(instr, format); + } + case 'f': { // 'f: FPUregisters or `frm + return FormatFPURegisterOrRoundMode(instr, format); + } + case 'a': { // 'a: Atomic acquire and release. + PrintAcquireRelease(instr); + return 1; + } + case 'p': { // `pre + MOZ_ASSERT(STRING_STARTS_WITH(format, "pre")); + PrintMemoryOrder(instr, true); + return 3; + } + case 's': { // 's32 or 's64: Shift amount. + if (format[1] == '3') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "s32")); + PrintShamt32(instr); + return 3; + } else if (format[1] == '6') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "s64")); + PrintShamt(instr); + return 3; + } else if (format[1] == 'u') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "suc")); + PrintMemoryOrder(instr, false); + return 3; + } else if (format[1] == 'e') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "sew")); + PrintRvvSEW(instr); + return 3; + } else if (format[1] == 'i') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "simm5")); + PrintRvvSimm5(instr); + return 5; + } + MOZ_CRASH(); + } + case 'v': { + if (format[1] == 'd') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "vd")); + PrintVd(instr); + return 2; + } else if (format[2] == '1') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "vs1")); + PrintVs1(instr); + return 3; + } else if (format[2] == '2') { + MOZ_ASSERT(STRING_STARTS_WITH(format, "vs2")); + PrintVs2(instr); + return 3; + } else { + MOZ_ASSERT(STRING_STARTS_WITH(format, "vm")); + PrintRvvVm(instr); + return 2; + } + } + case 'l': { + MOZ_ASSERT(STRING_STARTS_WITH(format, "lmul")); + PrintRvvLMUL(instr); + return 4; + } + case 'u': { + if (STRING_STARTS_WITH(format, "uimm5")) { + PrintRvvUimm5(instr); + return 5; + } else { + MOZ_ASSERT(STRING_STARTS_WITH(format, "uimm")); + PrintUimm(instr); + return 4; + } + } + case 't': { // 'target: target of branch instructions' + MOZ_ASSERT(STRING_STARTS_WITH(format, "target")); + PrintTarget(instr); + return 6; + } + } + MOZ_CRASH(); +} + +// Format takes a formatting string for a whole instruction and prints it into +// the output buffer. All escaped options are handed to FormatOption to be +// parsed further. +void Decoder::Format(Instruction* instr, const char* format) { + char cur = *format++; + while ((cur != 0) && (out_buffer_pos_ < (out_buffer_.length() - 1))) { + if (cur == '\'') { // Single quote is used as the formatting escape. + format += FormatOption(instr, format); + } else { + out_buffer_[out_buffer_pos_++] = cur; + } + cur = *format++; + } + out_buffer_[out_buffer_pos_] = '\0'; +} + +// The disassembler may end up decoding data inlined in the code. We do not want +// it to crash if the data does not ressemble any known instruction. +#define VERIFY(condition) \ + if (!(condition)) { \ + Unknown(instr); \ + return; \ + } + +// For currently unimplemented decodings the disassembler calls Unknown(instr) +// which will just print "unknown" of the instruction bits. +void Decoder::Unknown(Instruction* instr) { Format(instr, "unknown"); } + +// Print the register name according to the active name converter. +void Decoder::PrintRegister(int reg) { + Print(converter_.NameOfCPURegister(reg)); +} + +void Decoder::PrintVRegister(int reg) { UNSUPPORTED_RISCV(); } + +void Decoder::PrintRs1(Instruction* instr) { + int reg = instr->Rs1Value(); + PrintRegister(reg); +} + +void Decoder::PrintRs2(Instruction* instr) { + int reg = instr->Rs2Value(); + PrintRegister(reg); +} + +void Decoder::PrintRd(Instruction* instr) { + int reg = instr->RdValue(); + PrintRegister(reg); +} + +void Decoder::PrintUimm(Instruction* instr) { + int val = instr->Rs1Value(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "0x%x", val); +} + +void Decoder::PrintVs1(Instruction* instr) { + int reg = instr->Vs1Value(); + PrintVRegister(reg); +} + +void Decoder::PrintVs2(Instruction* instr) { + int reg = instr->Vs2Value(); + PrintVRegister(reg); +} + +void Decoder::PrintVd(Instruction* instr) { + int reg = instr->VdValue(); + PrintVRegister(reg); +} + +// Print the FPUregister name according to the active name converter. +void Decoder::PrintFPURegister(int freg) { + Print(converter_.NameOfXMMRegister(freg)); +} + +void Decoder::PrintFRs1(Instruction* instr) { + int reg = instr->Rs1Value(); + PrintFPURegister(reg); +} + +void Decoder::PrintFRs2(Instruction* instr) { + int reg = instr->Rs2Value(); + PrintFPURegister(reg); +} + +void Decoder::PrintFRs3(Instruction* instr) { + int reg = instr->Rs3Value(); + PrintFPURegister(reg); +} + +void Decoder::PrintFRd(Instruction* instr) { + int reg = instr->RdValue(); + PrintFPURegister(reg); +} + +void Decoder::PrintImm12X(Instruction* instr) { + int32_t imm = instr->Imm12Value(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "0x%x", imm); +} + +void Decoder::PrintImm12(Instruction* instr) { + int32_t imm = instr->Imm12Value(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintTarget(Instruction* instr) { + // if (Assembler::IsJalr(instr->InstructionBits())) { + // if (Assembler::IsAuipc((instr - 4)->InstructionBits()) && + // (instr - 4)->RdValue() == instr->Rs1Value()) { + // int32_t imm = Assembler::BrachlongOffset((instr - + // 4)->InstructionBits(), + // instr->InstructionBits()); + // const char* target = + // converter_.NameOfAddress(reinterpret_cast<byte*>(instr - 4) + imm); + // out_buffer_pos_ += + // SNPrintF(out_buffer_ + out_buffer_pos_, " -> %s", target); + // return; + // } + // } +} + +void Decoder::PrintBranchOffset(Instruction* instr) { + int32_t imm = instr->BranchOffset(); + const char* target = + converter_.NameOfAddress(reinterpret_cast<byte*>(instr) + imm); + out_buffer_pos_ += + SNPrintF(out_buffer_ + out_buffer_pos_, "%d -> %s", imm, target); +} + +void Decoder::PrintStoreOffset(Instruction* instr) { + int32_t imm = instr->StoreOffset(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvvSEW(Instruction* instr) { + const char* sew = instr->RvvSEW(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%s", sew); +} + +void Decoder::PrintRvvLMUL(Instruction* instr) { + const char* lmul = instr->RvvLMUL(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%s", lmul); +} + +void Decoder::PrintRvvSimm5(Instruction* instr) { + const int simm5 = instr->RvvSimm5(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", simm5); +} + +void Decoder::PrintRvvUimm5(Instruction* instr) { + const uint32_t uimm5 = instr->RvvUimm5(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%u", uimm5); +} + +void Decoder::PrintImm20U(Instruction* instr) { + int32_t imm = instr->Imm20UValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "0x%x", imm); +} + +void Decoder::PrintImm20J(Instruction* instr) { + int32_t imm = instr->Imm20JValue(); + const char* target = + converter_.NameOfAddress(reinterpret_cast<byte*>(instr) + imm); + out_buffer_pos_ += + SNPrintF(out_buffer_ + out_buffer_pos_, "%d -> %s", imm, target); +} + +void Decoder::PrintShamt(Instruction* instr) { + int32_t imm = instr->Shamt(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintShamt32(Instruction* instr) { + int32_t imm = instr->Shamt32(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm6(Instruction* instr) { + int32_t imm = instr->RvcImm6Value(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm6U(Instruction* instr) { + int32_t imm = instr->RvcImm6Value() & 0xFFFFF; + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "0x%x", imm); +} + +void Decoder::PrintRvcImm6Addi16sp(Instruction* instr) { + int32_t imm = instr->RvcImm6Addi16spValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcShamt(Instruction* instr) { + int32_t imm = instr->RvcShamt6(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm6Ldsp(Instruction* instr) { + int32_t imm = instr->RvcImm6LdspValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm6Lwsp(Instruction* instr) { + int32_t imm = instr->RvcImm6LwspValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm6Swsp(Instruction* instr) { + int32_t imm = instr->RvcImm6SwspValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm6Sdsp(Instruction* instr) { + int32_t imm = instr->RvcImm6SdspValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm5W(Instruction* instr) { + int32_t imm = instr->RvcImm5WValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm5D(Instruction* instr) { + int32_t imm = instr->RvcImm5DValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm8Addi4spn(Instruction* instr) { + int32_t imm = instr->RvcImm8Addi4spnValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm11CJ(Instruction* instr) { + int32_t imm = instr->RvcImm11CJValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvcImm8B(Instruction* instr) { + int32_t imm = instr->RvcImm8BValue(); + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%d", imm); +} + +void Decoder::PrintRvvVm(Instruction* instr) { + uint8_t imm = instr->RvvVM(); + if (imm == 0) { + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, " v0.t"); + } +} + +void Decoder::PrintAcquireRelease(Instruction* instr) { + bool aq = instr->AqValue(); + bool rl = instr->RlValue(); + if (aq || rl) { + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "."); + } + if (aq) { + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "aq"); + } + if (rl) { + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "rl"); + } +} + +void Decoder::PrintCSRReg(Instruction* instr) { + int32_t csr_reg = instr->CsrValue(); + std::string s; + switch (csr_reg) { + case csr_fflags: // Floating-Point Accrued Exceptions (RW) + s = "csr_fflags"; + break; + case csr_frm: // Floating-Point Dynamic Rounding Mode (RW) + s = "csr_frm"; + break; + case csr_fcsr: // Floating-Point Control and Status Register (RW) + s = "csr_fcsr"; + break; + case csr_cycle: + s = "csr_cycle"; + break; + case csr_time: + s = "csr_time"; + break; + case csr_instret: + s = "csr_instret"; + break; + case csr_cycleh: + s = "csr_cycleh"; + break; + case csr_timeh: + s = "csr_timeh"; + break; + case csr_instreth: + s = "csr_instreth"; + break; + default: + MOZ_CRASH(); + } + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%s", s.c_str()); +} + +void Decoder::PrintRoundingMode(Instruction* instr) { + int frm = instr->RoundMode(); + std::string s; + switch (frm) { + case RNE: + s = "RNE"; + break; + case RTZ: + s = "RTZ"; + break; + case RDN: + s = "RDN"; + break; + case RUP: + s = "RUP"; + break; + case RMM: + s = "RMM"; + break; + case DYN: + s = "DYN"; + break; + default: + MOZ_CRASH(); + } + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%s", s.c_str()); +} + +void Decoder::PrintMemoryOrder(Instruction* instr, bool is_pred) { + int memOrder = instr->MemoryOrder(is_pred); + std::string s; + if ((memOrder & PSI) == PSI) { + s += "i"; + } + if ((memOrder & PSO) == PSO) { + s += "o"; + } + if ((memOrder & PSR) == PSR) { + s += "r"; + } + if ((memOrder & PSW) == PSW) { + s += "w"; + } + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%s", s.c_str()); +} + +// Printing of instruction name. +void Decoder::PrintInstructionName(Instruction* instr) {} + +// RISCV Instruction Decode Routine +void Decoder::DecodeRType(Instruction* instr) { + switch (instr->InstructionBits() & kRTypeMask) { + case RO_ADD: + Format(instr, "add 'rd, 'rs1, 'rs2"); + break; + case RO_SUB: + if (instr->Rs1Value() == zero.code()) + Format(instr, "neg 'rd, 'rs2"); + else + Format(instr, "sub 'rd, 'rs1, 'rs2"); + break; + case RO_SLL: + Format(instr, "sll 'rd, 'rs1, 'rs2"); + break; + case RO_SLT: + if (instr->Rs2Value() == zero.code()) + Format(instr, "sltz 'rd, 'rs1"); + else if (instr->Rs1Value() == zero.code()) + Format(instr, "sgtz 'rd, 'rs2"); + else + Format(instr, "slt 'rd, 'rs1, 'rs2"); + break; + case RO_SLTU: + if (instr->Rs1Value() == zero.code()) + Format(instr, "snez 'rd, 'rs2"); + else + Format(instr, "sltu 'rd, 'rs1, 'rs2"); + break; + case RO_XOR: + Format(instr, "xor 'rd, 'rs1, 'rs2"); + break; + case RO_SRL: + Format(instr, "srl 'rd, 'rs1, 'rs2"); + break; + case RO_SRA: + Format(instr, "sra 'rd, 'rs1, 'rs2"); + break; + case RO_OR: + Format(instr, "or 'rd, 'rs1, 'rs2"); + break; + case RO_AND: + Format(instr, "and 'rd, 'rs1, 'rs2"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_ADDW: + Format(instr, "addw 'rd, 'rs1, 'rs2"); + break; + case RO_SUBW: + if (instr->Rs1Value() == zero.code()) + Format(instr, "negw 'rd, 'rs2"); + else + Format(instr, "subw 'rd, 'rs1, 'rs2"); + break; + case RO_SLLW: + Format(instr, "sllw 'rd, 'rs1, 'rs2"); + break; + case RO_SRLW: + Format(instr, "srlw 'rd, 'rs1, 'rs2"); + break; + case RO_SRAW: + Format(instr, "sraw 'rd, 'rs1, 'rs2"); + break; +#endif /* JS_CODEGEN_RISCV64 */ + // TODO(riscv): Add RISCV M extension macro + case RO_MUL: + Format(instr, "mul 'rd, 'rs1, 'rs2"); + break; + case RO_MULH: + Format(instr, "mulh 'rd, 'rs1, 'rs2"); + break; + case RO_MULHSU: + Format(instr, "mulhsu 'rd, 'rs1, 'rs2"); + break; + case RO_MULHU: + Format(instr, "mulhu 'rd, 'rs1, 'rs2"); + break; + case RO_DIV: + Format(instr, "div 'rd, 'rs1, 'rs2"); + break; + case RO_DIVU: + Format(instr, "divu 'rd, 'rs1, 'rs2"); + break; + case RO_REM: + Format(instr, "rem 'rd, 'rs1, 'rs2"); + break; + case RO_REMU: + Format(instr, "remu 'rd, 'rs1, 'rs2"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_MULW: + Format(instr, "mulw 'rd, 'rs1, 'rs2"); + break; + case RO_DIVW: + Format(instr, "divw 'rd, 'rs1, 'rs2"); + break; + case RO_DIVUW: + Format(instr, "divuw 'rd, 'rs1, 'rs2"); + break; + case RO_REMW: + Format(instr, "remw 'rd, 'rs1, 'rs2"); + break; + case RO_REMUW: + Format(instr, "remuw 'rd, 'rs1, 'rs2"); + break; +#endif /*JS_CODEGEN_RISCV64*/ + // TODO(riscv): End Add RISCV M extension macro + default: { + switch (instr->BaseOpcode()) { + case AMO: + DecodeRAType(instr); + break; + case OP_FP: + DecodeRFPType(instr); + break; + default: + UNSUPPORTED_RISCV(); + } + } + } +} + +void Decoder::DecodeRAType(Instruction* instr) { + // TODO(riscv): Add macro for RISCV A extension + // Special handling for A extension instructions because it uses func5 + // For all A extension instruction, V8 simulator is pure sequential. No + // Memory address lock or other synchronizaiton behaviors. + switch (instr->InstructionBits() & kRATypeMask) { + case RO_LR_W: + Format(instr, "lr.w'a 'rd, ('rs1)"); + break; + case RO_SC_W: + Format(instr, "sc.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOSWAP_W: + Format(instr, "amoswap.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOADD_W: + Format(instr, "amoadd.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOXOR_W: + Format(instr, "amoxor.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOAND_W: + Format(instr, "amoand.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOOR_W: + Format(instr, "amoor.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOMIN_W: + Format(instr, "amomin.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOMAX_W: + Format(instr, "amomax.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOMINU_W: + Format(instr, "amominu.w'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOMAXU_W: + Format(instr, "amomaxu.w'a 'rd, 'rs2, ('rs1)"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_LR_D: + Format(instr, "lr.d'a 'rd, ('rs1)"); + break; + case RO_SC_D: + Format(instr, "sc.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOSWAP_D: + Format(instr, "amoswap.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOADD_D: + Format(instr, "amoadd.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOXOR_D: + Format(instr, "amoxor.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOAND_D: + Format(instr, "amoand.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOOR_D: + Format(instr, "amoor.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOMIN_D: + Format(instr, "amomin.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOMAX_D: + Format(instr, "amoswap.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOMINU_D: + Format(instr, "amominu.d'a 'rd, 'rs2, ('rs1)"); + break; + case RO_AMOMAXU_D: + Format(instr, "amomaxu.d'a 'rd, 'rs2, ('rs1)"); + break; +#endif /*JS_CODEGEN_RISCV64*/ + // TODO(riscv): End Add macro for RISCV A extension + default: { + UNSUPPORTED_RISCV(); + } + } +} + +void Decoder::DecodeRFPType(Instruction* instr) { + // OP_FP instructions (F/D) uses func7 first. Some further uses fun3 and rs2() + + // kRATypeMask is only for func7 + switch (instr->InstructionBits() & kRFPTypeMask) { + // TODO(riscv): Add macro for RISCV F extension + case RO_FADD_S: + Format(instr, "fadd.s 'fd, 'fs1, 'fs2"); + break; + case RO_FSUB_S: + Format(instr, "fsub.s 'fd, 'fs1, 'fs2"); + break; + case RO_FMUL_S: + Format(instr, "fmul.s 'fd, 'fs1, 'fs2"); + break; + case RO_FDIV_S: + Format(instr, "fdiv.s 'fd, 'fs1, 'fs2"); + break; + case RO_FSQRT_S: + Format(instr, "fsqrt.s 'fd, 'fs1"); + break; + case RO_FSGNJ_S: { // RO_FSGNJN_S RO_FSGNJX_S + switch (instr->Funct3Value()) { + case 0b000: // RO_FSGNJ_S + if (instr->Rs1Value() == instr->Rs2Value()) + Format(instr, "fmv.s 'fd, 'fs1"); + else + Format(instr, "fsgnj.s 'fd, 'fs1, 'fs2"); + break; + case 0b001: // RO_FSGNJN_S + if (instr->Rs1Value() == instr->Rs2Value()) + Format(instr, "fneg.s 'fd, 'fs1"); + else + Format(instr, "fsgnjn.s 'fd, 'fs1, 'fs2"); + break; + case 0b010: // RO_FSGNJX_S + if (instr->Rs1Value() == instr->Rs2Value()) + Format(instr, "fabs.s 'fd, 'fs1"); + else + Format(instr, "fsgnjx.s 'fd, 'fs1, 'fs2"); + break; + default: + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FMIN_S: { // RO_FMAX_S + switch (instr->Funct3Value()) { + case 0b000: // RO_FMIN_S + Format(instr, "fmin.s 'fd, 'fs1, 'fs2"); + break; + case 0b001: // RO_FMAX_S + Format(instr, "fmax.s 'fd, 'fs1, 'fs2"); + break; + default: + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FCVT_W_S: { // RO_FCVT_WU_S , 64F RO_FCVT_L_S RO_FCVT_LU_S + switch (instr->Rs2Value()) { + case 0b00000: // RO_FCVT_W_S + Format(instr, "fcvt.w.s ['frm] 'rd, 'fs1"); + break; + case 0b00001: // RO_FCVT_WU_S + Format(instr, "fcvt.wu.s ['frm] 'rd, 'fs1"); + break; +#ifdef JS_CODEGEN_RISCV64 + case 0b00010: // RO_FCVT_L_S + Format(instr, "fcvt.l.s ['frm] 'rd, 'fs1"); + break; + case 0b00011: // RO_FCVT_LU_S + Format(instr, "fcvt.lu.s ['frm] 'rd, 'fs1"); + break; +#endif /* JS_CODEGEN_RISCV64 */ + default: + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FMV: { // RO_FCLASS_S + if (instr->Rs2Value() != 0b00000) { + UNSUPPORTED_RISCV(); + } + switch (instr->Funct3Value()) { + case 0b000: // RO_FMV_X_W + Format(instr, "fmv.x.w 'rd, 'fs1"); + break; + case 0b001: // RO_FCLASS_S + Format(instr, "fclass.s 'rd, 'fs1"); + break; + default: + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FLE_S: { // RO_FEQ_S RO_FLT_S RO_FLE_S + switch (instr->Funct3Value()) { + case 0b010: // RO_FEQ_S + Format(instr, "feq.s 'rd, 'fs1, 'fs2"); + break; + case 0b001: // RO_FLT_S + Format(instr, "flt.s 'rd, 'fs1, 'fs2"); + break; + case 0b000: // RO_FLE_S + Format(instr, "fle.s 'rd, 'fs1, 'fs2"); + break; + default: + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FCVT_S_W: { // RO_FCVT_S_WU , 64F RO_FCVT_S_L RO_FCVT_S_LU + switch (instr->Rs2Value()) { + case 0b00000: // RO_FCVT_S_W + Format(instr, "fcvt.s.w 'fd, 'rs1"); + break; + case 0b00001: // RO_FCVT_S_WU + Format(instr, "fcvt.s.wu 'fd, 'rs1"); + break; +#ifdef JS_CODEGEN_RISCV64 + case 0b00010: // RO_FCVT_S_L + Format(instr, "fcvt.s.l 'fd, 'rs1"); + break; + case 0b00011: // RO_FCVT_S_LU + Format(instr, "fcvt.s.lu 'fd, 'rs1"); + break; +#endif /* JS_CODEGEN_RISCV64 */ + default: { + UNSUPPORTED_RISCV(); + } + } + break; + } + case RO_FMV_W_X: { + if (instr->Funct3Value() == 0b000) { + Format(instr, "fmv.w.x 'fd, 'rs1"); + } else { + UNSUPPORTED_RISCV(); + } + break; + } + // TODO(riscv): Add macro for RISCV D extension + case RO_FADD_D: + Format(instr, "fadd.d 'fd, 'fs1, 'fs2"); + break; + case RO_FSUB_D: + Format(instr, "fsub.d 'fd, 'fs1, 'fs2"); + break; + case RO_FMUL_D: + Format(instr, "fmul.d 'fd, 'fs1, 'fs2"); + break; + case RO_FDIV_D: + Format(instr, "fdiv.d 'fd, 'fs1, 'fs2"); + break; + case RO_FSQRT_D: { + if (instr->Rs2Value() == 0b00000) { + Format(instr, "fsqrt.d 'fd, 'fs1"); + } else { + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FSGNJ_D: { // RO_FSGNJN_D RO_FSGNJX_D + switch (instr->Funct3Value()) { + case 0b000: // RO_FSGNJ_D + if (instr->Rs1Value() == instr->Rs2Value()) + Format(instr, "fmv.d 'fd, 'fs1"); + else + Format(instr, "fsgnj.d 'fd, 'fs1, 'fs2"); + break; + case 0b001: // RO_FSGNJN_D + if (instr->Rs1Value() == instr->Rs2Value()) + Format(instr, "fneg.d 'fd, 'fs1"); + else + Format(instr, "fsgnjn.d 'fd, 'fs1, 'fs2"); + break; + case 0b010: // RO_FSGNJX_D + if (instr->Rs1Value() == instr->Rs2Value()) + Format(instr, "fabs.d 'fd, 'fs1"); + else + Format(instr, "fsgnjx.d 'fd, 'fs1, 'fs2"); + break; + default: + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FMIN_D: { // RO_FMAX_D + switch (instr->Funct3Value()) { + case 0b000: // RO_FMIN_D + Format(instr, "fmin.d 'fd, 'fs1, 'fs2"); + break; + case 0b001: // RO_FMAX_D + Format(instr, "fmax.d 'fd, 'fs1, 'fs2"); + break; + default: + UNSUPPORTED_RISCV(); + } + break; + } + case (RO_FCVT_S_D & kRFPTypeMask): { + if (instr->Rs2Value() == 0b00001) { + Format(instr, "fcvt.s.d ['frm] 'fd, 'fs1"); + } else { + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FCVT_D_S: { + if (instr->Rs2Value() == 0b00000) { + Format(instr, "fcvt.d.s 'fd, 'fs1"); + } else { + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FLE_D: { // RO_FEQ_D RO_FLT_D RO_FLE_D + switch (instr->Funct3Value()) { + case 0b010: // RO_FEQ_S + Format(instr, "feq.d 'rd, 'fs1, 'fs2"); + break; + case 0b001: // RO_FLT_D + Format(instr, "flt.d 'rd, 'fs1, 'fs2"); + break; + case 0b000: // RO_FLE_D + Format(instr, "fle.d 'rd, 'fs1, 'fs2"); + break; + default: + UNSUPPORTED_RISCV(); + } + break; + } + case (RO_FCLASS_D & kRFPTypeMask): { // RO_FCLASS_D , 64D RO_FMV_X_D + if (instr->Rs2Value() != 0b00000) { + UNSUPPORTED_RISCV(); + break; + } + switch (instr->Funct3Value()) { + case 0b001: // RO_FCLASS_D + Format(instr, "fclass.d 'rd, 'fs1"); + break; +#ifdef JS_CODEGEN_RISCV64 + case 0b000: // RO_FMV_X_D + Format(instr, "fmv.x.d 'rd, 'fs1"); + break; +#endif /* JS_CODEGEN_RISCV64 */ + default: + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FCVT_W_D: { // RO_FCVT_WU_D , 64F RO_FCVT_L_D RO_FCVT_LU_D + switch (instr->Rs2Value()) { + case 0b00000: // RO_FCVT_W_D + Format(instr, "fcvt.w.d ['frm] 'rd, 'fs1"); + break; + case 0b00001: // RO_FCVT_WU_D + Format(instr, "fcvt.wu.d ['frm] 'rd, 'fs1"); + break; +#ifdef JS_CODEGEN_RISCV64 + case 0b00010: // RO_FCVT_L_D + Format(instr, "fcvt.l.d ['frm] 'rd, 'fs1"); + break; + case 0b00011: // RO_FCVT_LU_D + Format(instr, "fcvt.lu.d ['frm] 'rd, 'fs1"); + break; +#endif /* JS_CODEGEN_RISCV64 */ + default: + UNSUPPORTED_RISCV(); + } + break; + } + case RO_FCVT_D_W: { // RO_FCVT_D_WU , 64F RO_FCVT_D_L RO_FCVT_D_LU + switch (instr->Rs2Value()) { + case 0b00000: // RO_FCVT_D_W + Format(instr, "fcvt.d.w 'fd, 'rs1"); + break; + case 0b00001: // RO_FCVT_D_WU + Format(instr, "fcvt.d.wu 'fd, 'rs1"); + break; +#ifdef JS_CODEGEN_RISCV64 + case 0b00010: // RO_FCVT_D_L + Format(instr, "fcvt.d.l 'fd, 'rs1"); + break; + case 0b00011: // RO_FCVT_D_LU + Format(instr, "fcvt.d.lu 'fd, 'rs1"); + break; +#endif /* JS_CODEGEN_RISCV64 */ + default: + UNSUPPORTED_RISCV(); + } + break; + } +#ifdef JS_CODEGEN_RISCV64 + case RO_FMV_D_X: { + if (instr->Funct3Value() == 0b000 && instr->Rs2Value() == 0b00000) { + Format(instr, "fmv.d.x 'fd, 'rs1"); + } else { + UNSUPPORTED_RISCV(); + } + break; + } +#endif /* JS_CODEGEN_RISCV64 */ + default: { + UNSUPPORTED_RISCV(); + } + } +} + +void Decoder::DecodeR4Type(Instruction* instr) { + switch (instr->InstructionBits() & kR4TypeMask) { + // TODO(riscv): use F Extension macro block + case RO_FMADD_S: + Format(instr, "fmadd.s 'fd, 'fs1, 'fs2, 'fs3"); + break; + case RO_FMSUB_S: + Format(instr, "fmsub.s 'fd, 'fs1, 'fs2, 'fs3"); + break; + case RO_FNMSUB_S: + Format(instr, "fnmsub.s 'fd, 'fs1, 'fs2, 'fs3"); + break; + case RO_FNMADD_S: + Format(instr, "fnmadd.s 'fd, 'fs1, 'fs2, 'fs3"); + break; + // TODO(riscv): use F Extension macro block + case RO_FMADD_D: + Format(instr, "fmadd.d 'fd, 'fs1, 'fs2, 'fs3"); + break; + case RO_FMSUB_D: + Format(instr, "fmsub.d 'fd, 'fs1, 'fs2, 'fs3"); + break; + case RO_FNMSUB_D: + Format(instr, "fnmsub.d 'fd, 'fs1, 'fs2, 'fs3"); + break; + case RO_FNMADD_D: + Format(instr, "fnmadd.d 'fd, 'fs1, 'fs2, 'fs3"); + break; + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeIType(Instruction* instr) { + switch (instr->InstructionBits() & kITypeMask) { + case RO_JALR: + if (instr->RdValue() == zero.code() && instr->Rs1Value() == ra.code() && + instr->Imm12Value() == 0) + Format(instr, "ret"); + else if (instr->RdValue() == zero.code() && instr->Imm12Value() == 0) + Format(instr, "jr 'rs1"); + else if (instr->RdValue() == ra.code() && instr->Imm12Value() == 0) + Format(instr, "jalr 'rs1"); + else + Format(instr, "jalr 'rd, 'imm12('rs1)"); + break; + case RO_LB: + Format(instr, "lb 'rd, 'imm12('rs1)"); + break; + case RO_LH: + Format(instr, "lh 'rd, 'imm12('rs1)"); + break; + case RO_LW: + Format(instr, "lw 'rd, 'imm12('rs1)"); + break; + case RO_LBU: + Format(instr, "lbu 'rd, 'imm12('rs1)"); + break; + case RO_LHU: + Format(instr, "lhu 'rd, 'imm12('rs1)"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_LWU: + Format(instr, "lwu 'rd, 'imm12('rs1)"); + break; + case RO_LD: + Format(instr, "ld 'rd, 'imm12('rs1)"); + break; +#endif /*JS_CODEGEN_RISCV64*/ + case RO_ADDI: + if (instr->Imm12Value() == 0) { + if (instr->RdValue() == zero.code() && instr->Rs1Value() == zero.code()) + Format(instr, "nop"); + else + Format(instr, "mv 'rd, 'rs1"); + } else if (instr->Rs1Value() == zero.code()) { + Format(instr, "li 'rd, 'imm12"); + } else { + Format(instr, "addi 'rd, 'rs1, 'imm12"); + } + break; + case RO_SLTI: + Format(instr, "slti 'rd, 'rs1, 'imm12"); + break; + case RO_SLTIU: + if (instr->Imm12Value() == 1) + Format(instr, "seqz 'rd, 'rs1"); + else + Format(instr, "sltiu 'rd, 'rs1, 'imm12"); + break; + case RO_XORI: + if (instr->Imm12Value() == -1) + Format(instr, "not 'rd, 'rs1"); + else + Format(instr, "xori 'rd, 'rs1, 'imm12x"); + break; + case RO_ORI: + Format(instr, "ori 'rd, 'rs1, 'imm12x"); + break; + case RO_ANDI: + Format(instr, "andi 'rd, 'rs1, 'imm12x"); + break; + case RO_SLLI: + Format(instr, "slli 'rd, 'rs1, 's64"); + break; + case RO_SRLI: { // RO_SRAI + if (!instr->IsArithShift()) { + Format(instr, "srli 'rd, 'rs1, 's64"); + } else { + Format(instr, "srai 'rd, 'rs1, 's64"); + } + break; + } +#ifdef JS_CODEGEN_RISCV64 + case RO_ADDIW: + if (instr->Imm12Value() == 0) + Format(instr, "sext.w 'rd, 'rs1"); + else + Format(instr, "addiw 'rd, 'rs1, 'imm12"); + break; + case RO_SLLIW: + Format(instr, "slliw 'rd, 'rs1, 's32"); + break; + case RO_SRLIW: { // RO_SRAIW + if (!instr->IsArithShift()) { + Format(instr, "srliw 'rd, 'rs1, 's32"); + } else { + Format(instr, "sraiw 'rd, 'rs1, 's32"); + } + break; + } +#endif /*JS_CODEGEN_RISCV64*/ + case RO_FENCE: + if (instr->MemoryOrder(true) == PSIORW && + instr->MemoryOrder(false) == PSIORW) + Format(instr, "fence"); + else + Format(instr, "fence 'pre, 'suc"); + break; + case RO_ECALL: { // RO_EBREAK + if (instr->Imm12Value() == 0) { // ECALL + Format(instr, "ecall"); + } else if (instr->Imm12Value() == 1) { // EBREAK + Format(instr, "ebreak"); + } else { + UNSUPPORTED_RISCV(); + } + break; + } + // TODO(riscv): use Zifencei Standard Extension macro block + case RO_FENCE_I: + Format(instr, "fence.i"); + break; + // TODO(riscv): use Zicsr Standard Extension macro block + // FIXME(RISC-V): Add special formatting for CSR registers + case RO_CSRRW: + if (instr->CsrValue() == csr_fcsr) { + if (instr->RdValue() == zero.code()) + Format(instr, "fscsr 'rs1"); + else + Format(instr, "fscsr 'rd, 'rs1"); + } else if (instr->CsrValue() == csr_frm) { + if (instr->RdValue() == zero.code()) + Format(instr, "fsrm 'rs1"); + else + Format(instr, "fsrm 'rd, 'rs1"); + } else if (instr->CsrValue() == csr_fflags) { + if (instr->RdValue() == zero.code()) + Format(instr, "fsflags 'rs1"); + else + Format(instr, "fsflags 'rd, 'rs1"); + } else if (instr->RdValue() == zero.code()) { + Format(instr, "csrw 'csr, 'rs1"); + } else { + Format(instr, "csrrw 'rd, 'csr, 'rs1"); + } + break; + case RO_CSRRS: + if (instr->Rs1Value() == zero.code()) { + switch (instr->CsrValue()) { + case csr_instret: + Format(instr, "rdinstret 'rd"); + break; + case csr_instreth: + Format(instr, "rdinstreth 'rd"); + break; + case csr_time: + Format(instr, "rdtime 'rd"); + break; + case csr_timeh: + Format(instr, "rdtimeh 'rd"); + break; + case csr_cycle: + Format(instr, "rdcycle 'rd"); + break; + case csr_cycleh: + Format(instr, "rdcycleh 'rd"); + break; + case csr_fflags: + Format(instr, "frflags 'rd"); + break; + case csr_frm: + Format(instr, "frrm 'rd"); + break; + case csr_fcsr: + Format(instr, "frcsr 'rd"); + break; + default: + MOZ_CRASH(); + } + } else if (instr->Rs1Value() == zero.code()) { + Format(instr, "csrr 'rd, 'csr"); + } else if (instr->RdValue() == zero.code()) { + Format(instr, "csrs 'csr, 'rs1"); + } else { + Format(instr, "csrrs 'rd, 'csr, 'rs1"); + } + break; + case RO_CSRRC: + if (instr->RdValue() == zero.code()) + Format(instr, "csrc 'csr, 'rs1"); + else + Format(instr, "csrrc 'rd, 'csr, 'rs1"); + break; + case RO_CSRRWI: + if (instr->RdValue() == zero.code()) + Format(instr, "csrwi 'csr, 'uimm"); + else + Format(instr, "csrrwi 'rd, 'csr, 'uimm"); + break; + case RO_CSRRSI: + if (instr->RdValue() == zero.code()) + Format(instr, "csrsi 'csr, 'uimm"); + else + Format(instr, "csrrsi 'rd, 'csr, 'uimm"); + break; + case RO_CSRRCI: + if (instr->RdValue() == zero.code()) + Format(instr, "csrci 'csr, 'uimm"); + else + Format(instr, "csrrci 'rd, 'csr, 'uimm"); + break; + // TODO(riscv): use F Extension macro block + case RO_FLW: + Format(instr, "flw 'fd, 'imm12('rs1)"); + break; + // TODO(riscv): use D Extension macro block + case RO_FLD: + Format(instr, "fld 'fd, 'imm12('rs1)"); + break; + default: +#ifdef CAN_USE_RVV_INSTRUCTIONS + if (instr->vl_vs_width() != -1) { + DecodeRvvVL(instr); + } else { + UNSUPPORTED_RISCV(); + } + break; +#else + UNSUPPORTED_RISCV(); +#endif + } +} + +void Decoder::DecodeSType(Instruction* instr) { + switch (instr->InstructionBits() & kSTypeMask) { + case RO_SB: + Format(instr, "sb 'rs2, 'offS('rs1)"); + break; + case RO_SH: + Format(instr, "sh 'rs2, 'offS('rs1)"); + break; + case RO_SW: + Format(instr, "sw 'rs2, 'offS('rs1)"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_SD: + Format(instr, "sd 'rs2, 'offS('rs1)"); + break; +#endif /*JS_CODEGEN_RISCV64*/ + // TODO(riscv): use F Extension macro block + case RO_FSW: + Format(instr, "fsw 'fs2, 'offS('rs1)"); + break; + // TODO(riscv): use D Extension macro block + case RO_FSD: + Format(instr, "fsd 'fs2, 'offS('rs1)"); + break; + default: +#ifdef CAN_USE_RVV_INSTRUCTIONS + if (instr->vl_vs_width() != -1) { + DecodeRvvVS(instr); + } else { + UNSUPPORTED_RISCV(); + } + break; +#else + UNSUPPORTED_RISCV(); +#endif + } +} + +void Decoder::DecodeBType(Instruction* instr) { + switch (instr->InstructionBits() & kBTypeMask) { + case RO_BEQ: + Format(instr, "beq 'rs1, 'rs2, 'offB"); + break; + case RO_BNE: + Format(instr, "bne 'rs1, 'rs2, 'offB"); + break; + case RO_BLT: + Format(instr, "blt 'rs1, 'rs2, 'offB"); + break; + case RO_BGE: + Format(instr, "bge 'rs1, 'rs2, 'offB"); + break; + case RO_BLTU: + Format(instr, "bltu 'rs1, 'rs2, 'offB"); + break; + case RO_BGEU: + Format(instr, "bgeu 'rs1, 'rs2, 'offB"); + break; + default: + UNSUPPORTED_RISCV(); + } +} +void Decoder::DecodeUType(Instruction* instr) { + // U Type doesn't have additional mask + switch (instr->BaseOpcodeFieldRaw()) { + case LUI: + Format(instr, "lui 'rd, 'imm20U"); + break; + case AUIPC: + Format(instr, "auipc 'rd, 'imm20U"); + break; + default: + UNSUPPORTED_RISCV(); + } +} +// namespace jit +void Decoder::DecodeJType(Instruction* instr) { + // J Type doesn't have additional mask + switch (instr->BaseOpcodeValue()) { + case JAL: + if (instr->RdValue() == zero.code()) + Format(instr, "j 'imm20J"); + else if (instr->RdValue() == ra.code()) + Format(instr, "jal 'imm20J"); + else + Format(instr, "jal 'rd, 'imm20J"); + break; + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCRType(Instruction* instr) { + switch (instr->RvcFunct4Value()) { + case 0b1000: + if (instr->RvcRs1Value() != 0 && instr->RvcRs2Value() == 0) + Format(instr, "jr 'Crs1"); + else if (instr->RvcRdValue() != 0 && instr->RvcRs2Value() != 0) + Format(instr, "mv 'Crd, 'Crs2"); + else + UNSUPPORTED_RISCV(); + break; + case 0b1001: + if (instr->RvcRs1Value() == 0 && instr->RvcRs2Value() == 0) + Format(instr, "ebreak"); + else if (instr->RvcRdValue() != 0 && instr->RvcRs2Value() == 0) + Format(instr, "jalr 'Crs1"); + else if (instr->RvcRdValue() != 0 && instr->RvcRs2Value() != 0) + Format(instr, "add 'Crd, 'Crd, 'Crs2"); + else + UNSUPPORTED_RISCV(); + break; + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCAType(Instruction* instr) { + switch (instr->InstructionBits() & kCATypeMask) { + case RO_C_SUB: + Format(instr, "sub 'Crs1s, 'Crs1s, 'Crs2s"); + break; + case RO_C_XOR: + Format(instr, "xor 'Crs1s, 'Crs1s, 'Crs2s"); + break; + case RO_C_OR: + Format(instr, "or 'Crs1s, 'Crs1s, 'Crs2s"); + break; + case RO_C_AND: + Format(instr, "and 'Crs1s, 'Crs1s, 'Crs2s"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_C_SUBW: + Format(instr, "subw 'Crs1s, 'Crs1s, 'Crs2s"); + break; + case RO_C_ADDW: + Format(instr, "addw 'Crs1s, 'Crs1s, 'Crs2s"); + break; +#endif + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCIType(Instruction* instr) { + switch (instr->RvcOpcode()) { + case RO_C_NOP_ADDI: + if (instr->RvcRdValue() == 0) + Format(instr, "nop"); + else + Format(instr, "addi 'Crd, 'Crd, 'Cimm6"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_C_ADDIW: + Format(instr, "addiw 'Crd, 'Crd, 'Cimm6"); + break; +#endif + case RO_C_LI: + Format(instr, "li 'Crd, 'Cimm6"); + break; + case RO_C_LUI_ADD: + if (instr->RvcRdValue() == 2) + Format(instr, "addi sp, sp, 'Cimm6Addi16sp"); + else if (instr->RvcRdValue() != 0 && instr->RvcRdValue() != 2) + Format(instr, "lui 'Crd, 'Cimm6U"); + else + UNSUPPORTED_RISCV(); + break; + case RO_C_SLLI: + Format(instr, "slli 'Crd, 'Crd, 'Cshamt"); + break; + case RO_C_FLDSP: + Format(instr, "fld 'Cfd, 'Cimm6Ldsp(sp)"); + break; + case RO_C_LWSP: + Format(instr, "lw 'Crd, 'Cimm6Lwsp(sp)"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_C_LDSP: + Format(instr, "ld 'Crd, 'Cimm6Ldsp(sp)"); + break; +#elif defined(JS_CODEGEN_RISCV32) + case RO_C_FLWSP: + Format(instr, "flw 'Cfd, 'Cimm6Ldsp(sp)"); + break; +#endif + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCIWType(Instruction* instr) { + switch (instr->RvcOpcode()) { + case RO_C_ADDI4SPN: + Format(instr, "addi 'Crs2s, sp, 'Cimm8Addi4spn"); + break; + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCSSType(Instruction* instr) { + switch (instr->RvcOpcode()) { + case RO_C_SWSP: + Format(instr, "sw 'Crs2, 'Cimm6Swsp(sp)"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_C_SDSP: + Format(instr, "sd 'Crs2, 'Cimm6Sdsp(sp)"); + break; +#elif defined(JS_CODEGEN_RISCV32) + case RO_C_FSWSP: + Format(instr, "fsw 'Cfs2, 'Cimm6Sdsp(sp)"); + break; +#endif + case RO_C_FSDSP: + Format(instr, "fsd 'Cfs2, 'Cimm6Sdsp(sp)"); + break; + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCLType(Instruction* instr) { + switch (instr->RvcOpcode()) { + case RO_C_FLD: + Format(instr, "fld 'Cfs2s, 'Cimm5D('Crs1s)"); + break; + case RO_C_LW: + Format(instr, "lw 'Crs2s, 'Cimm5W('Crs1s)"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_C_LD: + Format(instr, "ld 'Crs2s, 'Cimm5D('Crs1s)"); + break; +#elif defined(JS_CODEGEN_RISCV32) + case RO_C_FLW: + Format(instr, "fld 'Cfs2s, 'Cimm5D('Crs1s)"); + break; +#endif + + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCSType(Instruction* instr) { + switch (instr->RvcOpcode()) { + case RO_C_FSD: + Format(instr, "fsd 'Cfs2s, 'Cimm5D('Crs1s)"); + break; + case RO_C_SW: + Format(instr, "sw 'Crs2s, 'Cimm5W('Crs1s)"); + break; +#ifdef JS_CODEGEN_RISCV64 + case RO_C_SD: + Format(instr, "sd 'Crs2s, 'Cimm5D('Crs1s)"); + break; +#elif defined(JS_CODEGEN_RISCV32) + case RO_C_FSW: + Format(instr, "fsw 'Cfs2s, 'Cimm5D('Crs1s)"); + break; +#endif + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCJType(Instruction* instr) { + switch (instr->RvcOpcode()) { + case RO_C_J: + Format(instr, "j 'Cimm11CJ"); + break; + default: + UNSUPPORTED_RISCV(); + } +} + +void Decoder::DecodeCBType(Instruction* instr) { + switch (instr->RvcOpcode()) { + case RO_C_BNEZ: + Format(instr, "bnez 'Crs1s, x0, 'Cimm8B"); + break; + case RO_C_BEQZ: + Format(instr, "beqz 'Crs1s, x0, 'Cimm8B"); + break; + case RO_C_MISC_ALU: + if (instr->RvcFunct2BValue() == 0b00) + Format(instr, "srli 'Crs1s, 'Crs1s, 'Cshamt"); + else if (instr->RvcFunct2BValue() == 0b01) + Format(instr, "srai 'Crs1s, 'Crs1s, 'Cshamt"); + else if (instr->RvcFunct2BValue() == 0b10) + Format(instr, "andi 'Crs1s, 'Crs1s, 'Cimm6"); + else + UNSUPPORTED_RISCV(); + break; + default: + UNSUPPORTED_RISCV(); + } +} + +#undef VERIFIY + +bool Decoder::IsConstantPoolAt(uint8_t* instr_ptr) { + UNSUPPORTED_RISCV(); + MOZ_CRASH(); +} + +int Decoder::ConstantPoolSizeAt(uint8_t* instr_ptr) { + UNSUPPORTED_RISCV(); + MOZ_CRASH(); +} + +// Disassemble the instruction at *instr_ptr into the output buffer. +int Decoder::InstructionDecode(byte* instr_ptr) { + Instruction* instr = Instruction::At(instr_ptr); + // Print raw instruction bytes. + out_buffer_pos_ += SNPrintF(out_buffer_ + out_buffer_pos_, "%08x ", + instr->InstructionBits()); + switch (instr->InstructionType()) { + case Instruction::kRType: + DecodeRType(instr); + break; + case Instruction::kR4Type: + DecodeR4Type(instr); + break; + case Instruction::kIType: + DecodeIType(instr); + break; + case Instruction::kSType: + DecodeSType(instr); + break; + case Instruction::kBType: + DecodeBType(instr); + break; + case Instruction::kUType: + DecodeUType(instr); + break; + case Instruction::kJType: + DecodeJType(instr); + break; + case Instruction::kCRType: + DecodeCRType(instr); + break; + case Instruction::kCAType: + DecodeCAType(instr); + break; + case Instruction::kCJType: + DecodeCJType(instr); + break; + case Instruction::kCIType: + DecodeCIType(instr); + break; + case Instruction::kCIWType: + DecodeCIWType(instr); + break; + case Instruction::kCSSType: + DecodeCSSType(instr); + break; + case Instruction::kCLType: + DecodeCLType(instr); + break; + case Instruction::kCSType: + DecodeCSType(instr); + break; + case Instruction::kCBType: + DecodeCBType(instr); + break; +#ifdef CAN_USE_RVV_INSTRUCTIONS + case Instruction::kVType: + DecodeVType(instr); + break; +#endif + default: + Format(instr, "UNSUPPORTED"); + UNSUPPORTED_RISCV(); + } + return instr->InstructionSize(); +} + +} // namespace disasm + +#undef STRING_STARTS_WITH +#undef VERIFY + +//------------------------------------------------------------------------------ + +namespace disasm { + +const char* NameConverter::NameOfAddress(uint8_t* addr) const { + SNPrintF(tmp_buffer_, "%p", addr); + return tmp_buffer_.start(); +} + +const char* NameConverter::NameOfConstant(uint8_t* addr) const { + return NameOfAddress(addr); +} + +const char* NameConverter::NameOfCPURegister(int reg) const { + return Registers::GetName(reg); +} + +const char* NameConverter::NameOfByteCPURegister(int reg) const { + MOZ_CRASH(" RISC-V does not have the concept of a byte register."); +} + +const char* NameConverter::NameOfXMMRegister(int reg) const { + return FloatRegisters::GetName(reg); +} + +const char* NameConverter::NameInCode(uint8_t* addr) const { + // The default name converter is called for unknown code. So we will not try + // to access any memory. + return ""; +} + +//------------------------------------------------------------------------------ + +Disassembler::Disassembler(const NameConverter& converter) + : converter_(converter) {} + +Disassembler::~Disassembler() {} + +int Disassembler::InstructionDecode(V8Vector<char> buffer, + uint8_t* instruction) { + Decoder d(converter_, buffer); + return d.InstructionDecode(instruction); +} + +int Disassembler::ConstantPoolSizeAt(uint8_t* instruction) { + return Decoder::ConstantPoolSizeAt(instruction); +} + +void Disassembler::Disassemble(FILE* f, uint8_t* begin, uint8_t* end) { + NameConverter converter; + Disassembler d(converter); + for (uint8_t* pc = begin; pc < end;) { + EmbeddedVector<char, ReasonableBufferSize> buffer; + buffer[0] = '\0'; + uint8_t* prev_pc = pc; + pc += d.InstructionDecode(buffer, pc); + fprintf(f, "%p %08x %s\n", prev_pc, + *reinterpret_cast<int32_t*>(prev_pc), buffer.start()); + } +} + +} // namespace disasm +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/disasm/Disasm-riscv64.h b/js/src/jit/riscv64/disasm/Disasm-riscv64.h new file mode 100644 index 0000000000..0548523f6b --- /dev/null +++ b/js/src/jit/riscv64/disasm/Disasm-riscv64.h @@ -0,0 +1,74 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + */ +// Copyright 2007-2008 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef jit_riscv64_disasm_Disasm_riscv64_h +#define jit_riscv64_disasm_Disasm_riscv64_h + +#include "mozilla/Assertions.h" +#include "mozilla/Types.h" + +#include <stdio.h> + +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/constant/util-riscv64.h" +namespace js { +namespace jit { +namespace disasm { + +typedef unsigned char byte; + +// Interface and default implementation for converting addresses and +// register-numbers to text. The default implementation is machine +// specific. +class NameConverter { + public: + virtual ~NameConverter() {} + virtual const char* NameOfCPURegister(int reg) const; + virtual const char* NameOfByteCPURegister(int reg) const; + virtual const char* NameOfXMMRegister(int reg) const; + virtual const char* NameOfAddress(byte* addr) const; + virtual const char* NameOfConstant(byte* addr) const; + virtual const char* NameInCode(byte* addr) const; + + protected: + EmbeddedVector<char, 128> tmp_buffer_; +}; + +// A generic Disassembler interface +class Disassembler { + public: + // Caller deallocates converter. + explicit Disassembler(const NameConverter& converter); + + virtual ~Disassembler(); + + // Writes one disassembled instruction into 'buffer' (0-terminated). + // Returns the length of the disassembled machine instruction in bytes. + int InstructionDecode(V8Vector<char> buffer, uint8_t* instruction); + + // Returns -1 if instruction does not mark the beginning of a constant pool, + // or the number of entries in the constant pool beginning here. + int ConstantPoolSizeAt(byte* instruction); + + // Write disassembly into specified file 'f' using specified NameConverter + // (see constructor). + static void Disassemble(FILE* f, uint8_t* begin, uint8_t* end); + + private: + const NameConverter& converter_; + + // Disallow implicit constructors. + Disassembler() = delete; + Disassembler(const Disassembler&) = delete; + void operator=(const Disassembler&) = delete; +}; + +} // namespace disasm +} // namespace jit +} // namespace js + +#endif // jit_riscv64_disasm_Disasm_riscv64_h diff --git a/js/src/jit/riscv64/extension/base-assembler-riscv.cc b/js/src/jit/riscv64/extension/base-assembler-riscv.cc new file mode 100644 index 0000000000..a64cc818b3 --- /dev/null +++ b/js/src/jit/riscv64/extension/base-assembler-riscv.cc @@ -0,0 +1,517 @@ +// Copyright (c) 1994-2006 Sun Microsystems Inc. +// All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// - Redistributions of source code must retain the above copyright notice, +// this list of conditions and the following disclaimer. +// +// - Redistribution in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution. +// +// - Neither the name of Sun Microsystems or the names of contributors may +// be used to endorse or promote products derived from this software without +// specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS +// IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, +// THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR +// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +// The original source code covered by the above license above has been +// modified significantly by Google Inc. +// Copyright 2021 the V8 project authors. All rights reserved. + +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/Architecture-riscv64.h" + +namespace js { +namespace jit { + +int ToNumber(Register reg) { + MOZ_ASSERT(reg.code() < Registers::Total && reg.code() >= 0); + const int kNumbers[] = { + 0, // zero_reg + 1, // ra + 2, // sp + 3, // gp + 4, // tp + 5, // t0 + 6, // t1 + 7, // t2 + 8, // s0/fp + 9, // s1 + 10, // a0 + 11, // a1 + 12, // a2 + 13, // a3 + 14, // a4 + 15, // a5 + 16, // a6 + 17, // a7 + 18, // s2 + 19, // s3 + 20, // s4 + 21, // s5 + 22, // s6 + 23, // s7 + 24, // s8 + 25, // s9 + 26, // s10 + 27, // s11 + 28, // t3 + 29, // t4 + 30, // t5 + 31, // t6 + }; + return kNumbers[reg.code()]; +} + +Register ToRegister(uint32_t num) { + MOZ_ASSERT(num >= 0 && num < Registers::Total); + const Register kRegisters[] = { + zero_reg, ra, sp, gp, tp, t0, t1, t2, fp, s1, a0, a1, a2, a3, a4, a5, + a6, a7, s2, s3, s4, s5, s6, s7, s8, s9, s10, s11, t3, t4, t5, t6}; + return kRegisters[num]; +} + +// ----- Top-level instruction formats match those in the ISA manual +// (R, I, S, B, U, J). These match the formats defined in the compiler +void AssemblerRiscvBase::GenInstrR(uint8_t funct7, uint8_t funct3, + BaseOpcode opcode, Register rd, Register rs1, + Register rs2) { + MOZ_ASSERT(is_uint7(funct7) && is_uint3(funct3)); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct7 << kFunct7Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrR(uint8_t funct7, uint8_t funct3, + BaseOpcode opcode, FPURegister rd, + FPURegister rs1, FPURegister rs2) { + MOZ_ASSERT(is_uint7(funct7) && is_uint3(funct3)); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct7 << kFunct7Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrR(uint8_t funct7, uint8_t funct3, + BaseOpcode opcode, Register rd, + FPURegister rs1, Register rs2) { + MOZ_ASSERT(is_uint7(funct7) && is_uint3(funct3)); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct7 << kFunct7Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrR(uint8_t funct7, uint8_t funct3, + BaseOpcode opcode, FPURegister rd, + Register rs1, Register rs2) { + MOZ_ASSERT(is_uint7(funct7) && is_uint3(funct3)); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct7 << kFunct7Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrR(uint8_t funct7, uint8_t funct3, + BaseOpcode opcode, FPURegister rd, + FPURegister rs1, Register rs2) { + MOZ_ASSERT(is_uint7(funct7) && is_uint3(funct3)); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct7 << kFunct7Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrR(uint8_t funct7, uint8_t funct3, + BaseOpcode opcode, Register rd, + FPURegister rs1, FPURegister rs2) { + MOZ_ASSERT(is_uint7(funct7) && is_uint3(funct3)); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct7 << kFunct7Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrR4(uint8_t funct2, BaseOpcode opcode, + Register rd, Register rs1, Register rs2, + Register rs3, FPURoundingMode frm) { + MOZ_ASSERT(is_uint2(funct2) && is_uint3(frm)); + Instr instr = opcode | (rd.code() << kRdShift) | (frm << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct2 << kFunct2Shift) | (rs3.code() << kRs3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrR4(uint8_t funct2, BaseOpcode opcode, + FPURegister rd, FPURegister rs1, + FPURegister rs2, FPURegister rs3, + FPURoundingMode frm) { + MOZ_ASSERT(is_uint2(funct2) && is_uint3(frm)); + Instr instr = opcode | (rd.code() << kRdShift) | (frm << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct2 << kFunct2Shift) | (rs3.code() << kRs3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrRAtomic(uint8_t funct5, bool aq, bool rl, + uint8_t funct3, Register rd, + Register rs1, Register rs2) { + MOZ_ASSERT(is_uint5(funct5) && is_uint3(funct3)); + Instr instr = AMO | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (rl << kRlShift) | (aq << kAqShift) | (funct5 << kFunct5Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrRFrm(uint8_t funct7, BaseOpcode opcode, + Register rd, Register rs1, Register rs2, + FPURoundingMode frm) { + MOZ_ASSERT(is_uint3(frm)); + Instr instr = opcode | (rd.code() << kRdShift) | (frm << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (rs2.code() << kRs2Shift) | + (funct7 << kFunct7Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrI(uint8_t funct3, BaseOpcode opcode, + Register rd, Register rs1, int16_t imm12) { + MOZ_ASSERT(is_uint3(funct3) && (is_uint12(imm12) || is_int12(imm12))); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (imm12 << kImm12Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrI(uint8_t funct3, BaseOpcode opcode, + FPURegister rd, Register rs1, + int16_t imm12) { + MOZ_ASSERT(is_uint3(funct3) && (is_uint12(imm12) || is_int12(imm12))); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (imm12 << kImm12Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrIShift(bool arithshift, uint8_t funct3, + BaseOpcode opcode, Register rd, + Register rs1, uint8_t shamt) { + MOZ_ASSERT(is_uint3(funct3) && is_uint6(shamt)); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (shamt << kShamtShift) | + (arithshift << kArithShiftShift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrIShiftW(bool arithshift, uint8_t funct3, + BaseOpcode opcode, Register rd, + Register rs1, uint8_t shamt) { + MOZ_ASSERT(is_uint3(funct3) && is_uint5(shamt)); + Instr instr = opcode | (rd.code() << kRdShift) | (funct3 << kFunct3Shift) | + (rs1.code() << kRs1Shift) | (shamt << kShamtWShift) | + (arithshift << kArithShiftShift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrS(uint8_t funct3, BaseOpcode opcode, + Register rs1, Register rs2, int16_t imm12) { + MOZ_ASSERT(is_uint3(funct3) && is_int12(imm12)); + Instr instr = opcode | ((imm12 & 0x1f) << 7) | // bits 4-0 + (funct3 << kFunct3Shift) | (rs1.code() << kRs1Shift) | + (rs2.code() << kRs2Shift) | + ((imm12 & 0xfe0) << 20); // bits 11-5 + emit(instr); +} + +void AssemblerRiscvBase::GenInstrS(uint8_t funct3, BaseOpcode opcode, + Register rs1, FPURegister rs2, + int16_t imm12) { + MOZ_ASSERT(is_uint3(funct3) && is_int12(imm12)); + Instr instr = opcode | ((imm12 & 0x1f) << 7) | // bits 4-0 + (funct3 << kFunct3Shift) | (rs1.code() << kRs1Shift) | + (rs2.code() << kRs2Shift) | + ((imm12 & 0xfe0) << 20); // bits 11-5 + emit(instr); +} + +void AssemblerRiscvBase::GenInstrB(uint8_t funct3, BaseOpcode opcode, + Register rs1, Register rs2, int16_t imm13) { + MOZ_ASSERT(is_uint3(funct3) && is_int13(imm13) && ((imm13 & 1) == 0)); + Instr instr = opcode | ((imm13 & 0x800) >> 4) | // bit 11 + ((imm13 & 0x1e) << 7) | // bits 4-1 + (funct3 << kFunct3Shift) | (rs1.code() << kRs1Shift) | + (rs2.code() << kRs2Shift) | + ((imm13 & 0x7e0) << 20) | // bits 10-5 + ((imm13 & 0x1000) << 19); // bit 12 + emit(instr); +} + +void AssemblerRiscvBase::GenInstrU(BaseOpcode opcode, Register rd, + int32_t imm20) { + MOZ_ASSERT((is_int20(imm20) || is_uint20(imm20))); + Instr instr = opcode | (rd.code() << kRdShift) | (imm20 << kImm20Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrJ(BaseOpcode opcode, Register rd, + int32_t imm21) { + MOZ_ASSERT(is_int21(imm21) && ((imm21 & 1) == 0)); + Instr instr = opcode | (rd.code() << kRdShift) | + (imm21 & 0xff000) | // bits 19-12 + ((imm21 & 0x800) << 9) | // bit 11 + ((imm21 & 0x7fe) << 20) | // bits 10-1 + ((imm21 & 0x100000) << 11); // bit 20 + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCR(uint8_t funct4, BaseOpcode opcode, + Register rd, Register rs2) { + MOZ_ASSERT(is_uint4(funct4)); + ShortInstr instr = opcode | (rs2.code() << kRvcRs2Shift) | + (rd.code() << kRvcRdShift) | (funct4 << kRvcFunct4Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCA(uint8_t funct6, BaseOpcode opcode, + Register rd, uint8_t funct, Register rs2) { + MOZ_ASSERT(is_uint6(funct6) && is_uint2(funct)); + ShortInstr instr = opcode | ((rs2.code() & 0x7) << kRvcRs2sShift) | + ((rd.code() & 0x7) << kRvcRs1sShift) | + (funct6 << kRvcFunct6Shift) | (funct << kRvcFunct2Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCI(uint8_t funct3, BaseOpcode opcode, + Register rd, int8_t imm6) { + MOZ_ASSERT(is_uint3(funct3) && is_int6(imm6)); + ShortInstr instr = opcode | ((imm6 & 0x1f) << 2) | + (rd.code() << kRvcRdShift) | ((imm6 & 0x20) << 7) | + (funct3 << kRvcFunct3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCIU(uint8_t funct3, BaseOpcode opcode, + Register rd, uint8_t uimm6) { + MOZ_ASSERT(is_uint3(funct3) && is_uint6(uimm6)); + ShortInstr instr = opcode | ((uimm6 & 0x1f) << 2) | + (rd.code() << kRvcRdShift) | ((uimm6 & 0x20) << 7) | + (funct3 << kRvcFunct3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCIU(uint8_t funct3, BaseOpcode opcode, + FPURegister rd, uint8_t uimm6) { + MOZ_ASSERT(is_uint3(funct3) && is_uint6(uimm6)); + ShortInstr instr = opcode | ((uimm6 & 0x1f) << 2) | + (rd.code() << kRvcRdShift) | ((uimm6 & 0x20) << 7) | + (funct3 << kRvcFunct3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCIW(uint8_t funct3, BaseOpcode opcode, + Register rd, uint8_t uimm8) { + MOZ_ASSERT(is_uint3(funct3) && is_uint8(uimm8)); + ShortInstr instr = opcode | ((uimm8) << 5) | + ((rd.code() & 0x7) << kRvcRs2sShift) | + (funct3 << kRvcFunct3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCSS(uint8_t funct3, BaseOpcode opcode, + Register rs2, uint8_t uimm6) { + MOZ_ASSERT(is_uint3(funct3) && is_uint6(uimm6)); + ShortInstr instr = opcode | (uimm6 << 7) | (rs2.code() << kRvcRs2Shift) | + (funct3 << kRvcFunct3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCSS(uint8_t funct3, BaseOpcode opcode, + FPURegister rs2, uint8_t uimm6) { + MOZ_ASSERT(is_uint3(funct3) && is_uint6(uimm6)); + ShortInstr instr = opcode | (uimm6 << 7) | (rs2.code() << kRvcRs2Shift) | + (funct3 << kRvcFunct3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCL(uint8_t funct3, BaseOpcode opcode, + Register rd, Register rs1, uint8_t uimm5) { + MOZ_ASSERT(is_uint3(funct3) && is_uint5(uimm5)); + ShortInstr instr = opcode | ((uimm5 & 0x3) << 5) | + ((rd.code() & 0x7) << kRvcRs2sShift) | + ((uimm5 & 0x1c) << 8) | (funct3 << kRvcFunct3Shift) | + ((rs1.code() & 0x7) << kRvcRs1sShift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCL(uint8_t funct3, BaseOpcode opcode, + FPURegister rd, Register rs1, + uint8_t uimm5) { + MOZ_ASSERT(is_uint3(funct3) && is_uint5(uimm5)); + ShortInstr instr = opcode | ((uimm5 & 0x3) << 5) | + ((rd.code() & 0x7) << kRvcRs2sShift) | + ((uimm5 & 0x1c) << 8) | (funct3 << kRvcFunct3Shift) | + ((rs1.code() & 0x7) << kRvcRs1sShift); + emit(instr); +} +void AssemblerRiscvBase::GenInstrCJ(uint8_t funct3, BaseOpcode opcode, + uint16_t uint11) { + MOZ_ASSERT(is_uint11(uint11)); + ShortInstr instr = opcode | (funct3 << kRvcFunct3Shift) | (uint11 << 2); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCS(uint8_t funct3, BaseOpcode opcode, + Register rs2, Register rs1, uint8_t uimm5) { + MOZ_ASSERT(is_uint3(funct3) && is_uint5(uimm5)); + ShortInstr instr = opcode | ((uimm5 & 0x3) << 5) | + ((rs2.code() & 0x7) << kRvcRs2sShift) | + ((uimm5 & 0x1c) << 8) | (funct3 << kRvcFunct3Shift) | + ((rs1.code() & 0x7) << kRvcRs1sShift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCS(uint8_t funct3, BaseOpcode opcode, + FPURegister rs2, Register rs1, + uint8_t uimm5) { + MOZ_ASSERT(is_uint3(funct3) && is_uint5(uimm5)); + ShortInstr instr = opcode | ((uimm5 & 0x3) << 5) | + ((rs2.code() & 0x7) << kRvcRs2sShift) | + ((uimm5 & 0x1c) << 8) | (funct3 << kRvcFunct3Shift) | + ((rs1.code() & 0x7) << kRvcRs1sShift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCB(uint8_t funct3, BaseOpcode opcode, + Register rs1, uint8_t uimm8) { + MOZ_ASSERT(is_uint3(funct3) && is_uint8(uimm8)); + ShortInstr instr = opcode | ((uimm8 & 0x1f) << 2) | ((uimm8 & 0xe0) << 5) | + ((rs1.code() & 0x7) << kRvcRs1sShift) | + (funct3 << kRvcFunct3Shift); + emit(instr); +} + +void AssemblerRiscvBase::GenInstrCBA(uint8_t funct3, uint8_t funct2, + BaseOpcode opcode, Register rs1, + int8_t imm6) { + MOZ_ASSERT(is_uint3(funct3) && is_uint2(funct2) && is_int6(imm6)); + ShortInstr instr = opcode | ((imm6 & 0x1f) << 2) | ((imm6 & 0x20) << 7) | + ((rs1.code() & 0x7) << kRvcRs1sShift) | + (funct3 << kRvcFunct3Shift) | (funct2 << 10); + emit(instr); +} +// ----- Instruction class templates match those in the compiler + +void AssemblerRiscvBase::GenInstrBranchCC_rri(uint8_t funct3, Register rs1, + Register rs2, int16_t imm13) { + GenInstrB(funct3, BRANCH, rs1, rs2, imm13); +} + +void AssemblerRiscvBase::GenInstrLoad_ri(uint8_t funct3, Register rd, + Register rs1, int16_t imm12) { + GenInstrI(funct3, LOAD, rd, rs1, imm12); +} + +void AssemblerRiscvBase::GenInstrStore_rri(uint8_t funct3, Register rs1, + Register rs2, int16_t imm12) { + GenInstrS(funct3, STORE, rs1, rs2, imm12); +} + +void AssemblerRiscvBase::GenInstrALU_ri(uint8_t funct3, Register rd, + Register rs1, int16_t imm12) { + GenInstrI(funct3, OP_IMM, rd, rs1, imm12); +} + +void AssemblerRiscvBase::GenInstrShift_ri(bool arithshift, uint8_t funct3, + Register rd, Register rs1, + uint8_t shamt) { + MOZ_ASSERT(is_uint6(shamt)); + GenInstrI(funct3, OP_IMM, rd, rs1, (arithshift << 10) | shamt); +} + +void AssemblerRiscvBase::GenInstrALU_rr(uint8_t funct7, uint8_t funct3, + Register rd, Register rs1, + Register rs2) { + GenInstrR(funct7, funct3, OP, rd, rs1, rs2); +} + +void AssemblerRiscvBase::GenInstrCSR_ir(uint8_t funct3, Register rd, + ControlStatusReg csr, Register rs1) { + GenInstrI(funct3, SYSTEM, rd, rs1, csr); +} + +void AssemblerRiscvBase::GenInstrCSR_ii(uint8_t funct3, Register rd, + ControlStatusReg csr, uint8_t imm5) { + GenInstrI(funct3, SYSTEM, rd, ToRegister(imm5), csr); +} + +void AssemblerRiscvBase::GenInstrShiftW_ri(bool arithshift, uint8_t funct3, + Register rd, Register rs1, + uint8_t shamt) { + GenInstrIShiftW(arithshift, funct3, OP_IMM_32, rd, rs1, shamt); +} + +void AssemblerRiscvBase::GenInstrALUW_rr(uint8_t funct7, uint8_t funct3, + Register rd, Register rs1, + Register rs2) { + GenInstrR(funct7, funct3, OP_32, rd, rs1, rs2); +} + +void AssemblerRiscvBase::GenInstrPriv(uint8_t funct7, Register rs1, + Register rs2) { + GenInstrR(funct7, 0b000, SYSTEM, ToRegister(0UL), rs1, rs2); +} + +void AssemblerRiscvBase::GenInstrLoadFP_ri(uint8_t funct3, FPURegister rd, + Register rs1, int16_t imm12) { + GenInstrI(funct3, LOAD_FP, rd, rs1, imm12); +} + +void AssemblerRiscvBase::GenInstrStoreFP_rri(uint8_t funct3, Register rs1, + FPURegister rs2, int16_t imm12) { + GenInstrS(funct3, STORE_FP, rs1, rs2, imm12); +} + +void AssemblerRiscvBase::GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, + FPURegister rd, FPURegister rs1, + FPURegister rs2) { + GenInstrR(funct7, funct3, OP_FP, rd, rs1, rs2); +} + +void AssemblerRiscvBase::GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, + FPURegister rd, Register rs1, + Register rs2) { + GenInstrR(funct7, funct3, OP_FP, rd, rs1, rs2); +} + +void AssemblerRiscvBase::GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, + FPURegister rd, FPURegister rs1, + Register rs2) { + GenInstrR(funct7, funct3, OP_FP, rd, rs1, rs2); +} + +void AssemblerRiscvBase::GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, + Register rd, FPURegister rs1, + Register rs2) { + GenInstrR(funct7, funct3, OP_FP, rd, rs1, rs2); +} + +void AssemblerRiscvBase::GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, + Register rd, FPURegister rs1, + FPURegister rs2) { + GenInstrR(funct7, funct3, OP_FP, rd, rs1, rs2); +} + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/base-assembler-riscv.h b/js/src/jit/riscv64/extension/base-assembler-riscv.h new file mode 100644 index 0000000000..cb3083d365 --- /dev/null +++ b/js/src/jit/riscv64/extension/base-assembler-riscv.h @@ -0,0 +1,219 @@ +// Copyright (c) 1994-2006 Sun Microsystems Inc. +// All Rights Reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions are +// met: +// +// - Redistributions of source code must retain the above copyright notice, +// this list of conditions and the following disclaimer. +// +// - Redistribution in binary form must reproduce the above copyright +// notice, this list of conditions and the following disclaimer in the +// documentation and/or other materials provided with the distribution. +// +// - Neither the name of Sun Microsystems or the names of contributors may +// be used to endorse or promote products derived from this software without +// specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS +// IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, +// THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR +// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR +// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +// The original source code covered by the above license above has been +// modified significantly by Google Inc. +// Copyright 2021 the V8 project authors. All rights reserved. + +#ifndef jit_riscv64_extension_Base_assembler_riscv_h +#define jit_riscv64_extension_Base_assembler_riscv_h + +#include <memory> +#include <set> +#include <stdio.h> + +#include "jit/Label.h" +#include "jit/riscv64/Architecture-riscv64.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Register-riscv64.h" + +#define xlen (uint8_t(sizeof(void*) * 8)) + +#define kBitsPerByte 8UL +// Check number width. +inline constexpr bool is_intn(int64_t x, unsigned n) { + MOZ_ASSERT((0 < n) && (n < 64)); + int64_t limit = static_cast<int64_t>(1) << (n - 1); + return (-limit <= x) && (x < limit); +} + +inline constexpr bool is_uintn(int64_t x, unsigned n) { + MOZ_ASSERT((0 < n) && (n < (sizeof(x) * kBitsPerByte))); + return !(x >> n); +} +#undef kBitsPerByte +// clang-format off +#define INT_1_TO_63_LIST(V) \ + V(1) V(2) V(3) V(4) V(5) V(6) V(7) V(8) V(9) V(10) \ + V(11) V(12) V(13) V(14) V(15) V(16) V(17) V(18) V(19) V(20) \ + V(21) V(22) V(23) V(24) V(25) V(26) V(27) V(28) V(29) V(30) \ + V(31) V(32) V(33) V(34) V(35) V(36) V(37) V(38) V(39) V(40) \ + V(41) V(42) V(43) V(44) V(45) V(46) V(47) V(48) V(49) V(50) \ + V(51) V(52) V(53) V(54) V(55) V(56) V(57) V(58) V(59) V(60) \ + V(61) V(62) V(63) +// clang-format on + +#define DECLARE_IS_INT_N(N) \ + inline constexpr bool is_int##N(int64_t x) { return is_intn(x, N); } + +#define DECLARE_IS_UINT_N(N) \ + template <class T> \ + inline constexpr bool is_uint##N(T x) { \ + return is_uintn(x, N); \ + } +INT_1_TO_63_LIST(DECLARE_IS_INT_N) +INT_1_TO_63_LIST(DECLARE_IS_UINT_N) + +#undef DECLARE_IS_INT_N +#undef INT_1_TO_63_LIST + +namespace js { +namespace jit { + +typedef FloatRegister FPURegister; +#define zero_reg zero + +#define DEBUG_PRINTF(...) \ + if (FLAG_riscv_debug) { \ + std::printf(__VA_ARGS__); \ + } + +int ToNumber(Register reg); +Register ToRegister(uint32_t num); + +class AssemblerRiscvBase { + protected: + virtual int32_t branch_offset_helper(Label* L, OffsetSize bits) = 0; + + virtual void emit(Instr x) = 0; + virtual void emit(ShortInstr x) = 0; + virtual void emit(uint64_t x) = 0; + virtual uint32_t currentOffset() = 0; + // Instruction generation. + + // ----- Top-level instruction formats match those in the ISA manual + // (R, I, S, B, U, J). These match the formats defined in LLVM's + // RISCVInstrFormats.td. + void GenInstrR(uint8_t funct7, uint8_t funct3, BaseOpcode opcode, Register rd, + Register rs1, Register rs2); + void GenInstrR(uint8_t funct7, uint8_t funct3, BaseOpcode opcode, + FPURegister rd, FPURegister rs1, FPURegister rs2); + void GenInstrR(uint8_t funct7, uint8_t funct3, BaseOpcode opcode, Register rd, + FPURegister rs1, Register rs2); + void GenInstrR(uint8_t funct7, uint8_t funct3, BaseOpcode opcode, + FPURegister rd, Register rs1, Register rs2); + void GenInstrR(uint8_t funct7, uint8_t funct3, BaseOpcode opcode, + FPURegister rd, FPURegister rs1, Register rs2); + void GenInstrR(uint8_t funct7, uint8_t funct3, BaseOpcode opcode, Register rd, + FPURegister rs1, FPURegister rs2); + void GenInstrR4(uint8_t funct2, BaseOpcode opcode, Register rd, Register rs1, + Register rs2, Register rs3, FPURoundingMode frm); + void GenInstrR4(uint8_t funct2, BaseOpcode opcode, FPURegister rd, + FPURegister rs1, FPURegister rs2, FPURegister rs3, + FPURoundingMode frm); + void GenInstrRAtomic(uint8_t funct5, bool aq, bool rl, uint8_t funct3, + Register rd, Register rs1, Register rs2); + void GenInstrRFrm(uint8_t funct7, BaseOpcode opcode, Register rd, + Register rs1, Register rs2, FPURoundingMode frm); + void GenInstrI(uint8_t funct3, BaseOpcode opcode, Register rd, Register rs1, + int16_t imm12); + void GenInstrI(uint8_t funct3, BaseOpcode opcode, FPURegister rd, + Register rs1, int16_t imm12); + void GenInstrIShift(bool arithshift, uint8_t funct3, BaseOpcode opcode, + Register rd, Register rs1, uint8_t shamt); + void GenInstrIShiftW(bool arithshift, uint8_t funct3, BaseOpcode opcode, + Register rd, Register rs1, uint8_t shamt); + void GenInstrS(uint8_t funct3, BaseOpcode opcode, Register rs1, Register rs2, + int16_t imm12); + void GenInstrS(uint8_t funct3, BaseOpcode opcode, Register rs1, + FPURegister rs2, int16_t imm12); + void GenInstrB(uint8_t funct3, BaseOpcode opcode, Register rs1, Register rs2, + int16_t imm12); + void GenInstrU(BaseOpcode opcode, Register rd, int32_t imm20); + void GenInstrJ(BaseOpcode opcode, Register rd, int32_t imm20); + void GenInstrCR(uint8_t funct4, BaseOpcode opcode, Register rd, Register rs2); + void GenInstrCA(uint8_t funct6, BaseOpcode opcode, Register rd, uint8_t funct, + Register rs2); + void GenInstrCI(uint8_t funct3, BaseOpcode opcode, Register rd, int8_t imm6); + void GenInstrCIU(uint8_t funct3, BaseOpcode opcode, Register rd, + uint8_t uimm6); + void GenInstrCIU(uint8_t funct3, BaseOpcode opcode, FPURegister rd, + uint8_t uimm6); + void GenInstrCIW(uint8_t funct3, BaseOpcode opcode, Register rd, + uint8_t uimm8); + void GenInstrCSS(uint8_t funct3, BaseOpcode opcode, FPURegister rs2, + uint8_t uimm6); + void GenInstrCSS(uint8_t funct3, BaseOpcode opcode, Register rs2, + uint8_t uimm6); + void GenInstrCL(uint8_t funct3, BaseOpcode opcode, Register rd, Register rs1, + uint8_t uimm5); + void GenInstrCL(uint8_t funct3, BaseOpcode opcode, FPURegister rd, + Register rs1, uint8_t uimm5); + void GenInstrCS(uint8_t funct3, BaseOpcode opcode, Register rs2, Register rs1, + uint8_t uimm5); + void GenInstrCS(uint8_t funct3, BaseOpcode opcode, FPURegister rs2, + Register rs1, uint8_t uimm5); + void GenInstrCJ(uint8_t funct3, BaseOpcode opcode, uint16_t uint11); + void GenInstrCB(uint8_t funct3, BaseOpcode opcode, Register rs1, + uint8_t uimm8); + void GenInstrCBA(uint8_t funct3, uint8_t funct2, BaseOpcode opcode, + Register rs1, int8_t imm6); + + // ----- Instruction class templates match those in LLVM's RISCVInstrInfo.td + void GenInstrBranchCC_rri(uint8_t funct3, Register rs1, Register rs2, + int16_t imm12); + void GenInstrLoad_ri(uint8_t funct3, Register rd, Register rs1, + int16_t imm12); + void GenInstrStore_rri(uint8_t funct3, Register rs1, Register rs2, + int16_t imm12); + void GenInstrALU_ri(uint8_t funct3, Register rd, Register rs1, int16_t imm12); + void GenInstrShift_ri(bool arithshift, uint8_t funct3, Register rd, + Register rs1, uint8_t shamt); + void GenInstrALU_rr(uint8_t funct7, uint8_t funct3, Register rd, Register rs1, + Register rs2); + void GenInstrCSR_ir(uint8_t funct3, Register rd, ControlStatusReg csr, + Register rs1); + void GenInstrCSR_ii(uint8_t funct3, Register rd, ControlStatusReg csr, + uint8_t rs1); + void GenInstrShiftW_ri(bool arithshift, uint8_t funct3, Register rd, + Register rs1, uint8_t shamt); + void GenInstrALUW_rr(uint8_t funct7, uint8_t funct3, Register rd, + Register rs1, Register rs2); + void GenInstrPriv(uint8_t funct7, Register rs1, Register rs2); + void GenInstrLoadFP_ri(uint8_t funct3, FPURegister rd, Register rs1, + int16_t imm12); + void GenInstrStoreFP_rri(uint8_t funct3, Register rs1, FPURegister rs2, + int16_t imm12); + void GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, FPURegister rd, + FPURegister rs1, FPURegister rs2); + void GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, FPURegister rd, + Register rs1, Register rs2); + void GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, FPURegister rd, + FPURegister rs1, Register rs2); + void GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, Register rd, + FPURegister rs1, Register rs2); + void GenInstrALUFP_rr(uint8_t funct7, uint8_t funct3, Register rd, + FPURegister rs1, FPURegister rs2); +}; + +} // namespace jit +} // namespace js + +#endif // jit_riscv64_extension_Base_assembler_riscv_h diff --git a/js/src/jit/riscv64/extension/base-riscv-i.cc b/js/src/jit/riscv64/extension/base-riscv-i.cc new file mode 100644 index 0000000000..2ee8877eb1 --- /dev/null +++ b/js/src/jit/riscv64/extension/base-riscv-i.cc @@ -0,0 +1,351 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/extension/base-riscv-i.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { + +void AssemblerRISCVI::lui(Register rd, int32_t imm20) { + GenInstrU(LUI, rd, imm20); +} + +void AssemblerRISCVI::auipc(Register rd, int32_t imm20) { + GenInstrU(AUIPC, rd, imm20); +} + +// Jumps + +void AssemblerRISCVI::jal(Register rd, int32_t imm21) { + GenInstrJ(JAL, rd, imm21); +} + +void AssemblerRISCVI::jalr(Register rd, Register rs1, int16_t imm12) { + GenInstrI(0b000, JALR, rd, rs1, imm12); +} + +// Branches + +void AssemblerRISCVI::beq(Register rs1, Register rs2, int16_t imm13) { + GenInstrBranchCC_rri(0b000, rs1, rs2, imm13); +} + +void AssemblerRISCVI::bne(Register rs1, Register rs2, int16_t imm13) { + GenInstrBranchCC_rri(0b001, rs1, rs2, imm13); +} + +void AssemblerRISCVI::blt(Register rs1, Register rs2, int16_t imm13) { + GenInstrBranchCC_rri(0b100, rs1, rs2, imm13); +} + +void AssemblerRISCVI::bge(Register rs1, Register rs2, int16_t imm13) { + GenInstrBranchCC_rri(0b101, rs1, rs2, imm13); +} + +void AssemblerRISCVI::bltu(Register rs1, Register rs2, int16_t imm13) { + GenInstrBranchCC_rri(0b110, rs1, rs2, imm13); +} + +void AssemblerRISCVI::bgeu(Register rs1, Register rs2, int16_t imm13) { + GenInstrBranchCC_rri(0b111, rs1, rs2, imm13); +} + +// Loads + +void AssemblerRISCVI::lb(Register rd, Register rs1, int16_t imm12) { + GenInstrLoad_ri(0b000, rd, rs1, imm12); +} + +void AssemblerRISCVI::lh(Register rd, Register rs1, int16_t imm12) { + GenInstrLoad_ri(0b001, rd, rs1, imm12); +} + +void AssemblerRISCVI::lw(Register rd, Register rs1, int16_t imm12) { + GenInstrLoad_ri(0b010, rd, rs1, imm12); +} + +void AssemblerRISCVI::lbu(Register rd, Register rs1, int16_t imm12) { + GenInstrLoad_ri(0b100, rd, rs1, imm12); +} + +void AssemblerRISCVI::lhu(Register rd, Register rs1, int16_t imm12) { + GenInstrLoad_ri(0b101, rd, rs1, imm12); +} + +// Stores + +void AssemblerRISCVI::sb(Register source, Register base, int16_t imm12) { + GenInstrStore_rri(0b000, base, source, imm12); +} + +void AssemblerRISCVI::sh(Register source, Register base, int16_t imm12) { + GenInstrStore_rri(0b001, base, source, imm12); +} + +void AssemblerRISCVI::sw(Register source, Register base, int16_t imm12) { + GenInstrStore_rri(0b010, base, source, imm12); +} + +// Arithmetic with immediate + +void AssemblerRISCVI::addi(Register rd, Register rs1, int16_t imm12) { + GenInstrALU_ri(0b000, rd, rs1, imm12); +} + +void AssemblerRISCVI::slti(Register rd, Register rs1, int16_t imm12) { + GenInstrALU_ri(0b010, rd, rs1, imm12); +} + +void AssemblerRISCVI::sltiu(Register rd, Register rs1, int16_t imm12) { + GenInstrALU_ri(0b011, rd, rs1, imm12); +} + +void AssemblerRISCVI::xori(Register rd, Register rs1, int16_t imm12) { + GenInstrALU_ri(0b100, rd, rs1, imm12); +} + +void AssemblerRISCVI::ori(Register rd, Register rs1, int16_t imm12) { + GenInstrALU_ri(0b110, rd, rs1, imm12); +} + +void AssemblerRISCVI::andi(Register rd, Register rs1, int16_t imm12) { + GenInstrALU_ri(0b111, rd, rs1, imm12); +} + +void AssemblerRISCVI::slli(Register rd, Register rs1, uint8_t shamt) { + GenInstrShift_ri(0, 0b001, rd, rs1, shamt & 0x3f); +} + +void AssemblerRISCVI::srli(Register rd, Register rs1, uint8_t shamt) { + GenInstrShift_ri(0, 0b101, rd, rs1, shamt & 0x3f); +} + +void AssemblerRISCVI::srai(Register rd, Register rs1, uint8_t shamt) { + GenInstrShift_ri(1, 0b101, rd, rs1, shamt & 0x3f); +} + +// Arithmetic + +void AssemblerRISCVI::add(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000000, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVI::sub(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0100000, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVI::sll(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000000, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVI::slt(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000000, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVI::sltu(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000000, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVI::xor_(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000000, 0b100, rd, rs1, rs2); +} + +void AssemblerRISCVI::srl(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000000, 0b101, rd, rs1, rs2); +} + +void AssemblerRISCVI::sra(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0100000, 0b101, rd, rs1, rs2); +} + +void AssemblerRISCVI::or_(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000000, 0b110, rd, rs1, rs2); +} + +void AssemblerRISCVI::and_(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000000, 0b111, rd, rs1, rs2); +} + +// Memory fences + +void AssemblerRISCVI::fence(uint8_t pred, uint8_t succ) { + MOZ_ASSERT(is_uint4(pred) && is_uint4(succ)); + uint16_t imm12 = succ | (pred << 4) | (0b0000 << 8); + GenInstrI(0b000, MISC_MEM, ToRegister(0UL), ToRegister(0UL), imm12); +} + +void AssemblerRISCVI::fence_tso() { + uint16_t imm12 = (0b0011) | (0b0011 << 4) | (0b1000 << 8); + GenInstrI(0b000, MISC_MEM, ToRegister(0UL), ToRegister(0UL), imm12); +} + +// Environment call / break + +void AssemblerRISCVI::ecall() { + GenInstrI(0b000, SYSTEM, ToRegister(0UL), ToRegister(0UL), 0); +} + +void AssemblerRISCVI::ebreak() { + GenInstrI(0b000, SYSTEM, ToRegister(0UL), ToRegister(0UL), 1); +} + +// This is a de facto standard (as set by GNU binutils) 32-bit unimplemented +// instruction (i.e., it should always trap, if your implementation has invalid +// instruction traps). +void AssemblerRISCVI::unimp() { + GenInstrI(0b001, SYSTEM, ToRegister(0), ToRegister(0), 0b110000000000); +} + +bool AssemblerRISCVI::IsBranch(Instr instr) { + return (instr & kBaseOpcodeMask) == BRANCH; +} + +bool AssemblerRISCVI::IsJump(Instr instr) { + int Op = instr & kBaseOpcodeMask; + return Op == JAL || Op == JALR; +} + +bool AssemblerRISCVI::IsNop(Instr instr) { return instr == kNopByte; } + +bool AssemblerRISCVI::IsJal(Instr instr) { + return (instr & kBaseOpcodeMask) == JAL; +} + +bool AssemblerRISCVI::IsJalr(Instr instr) { + return (instr & kBaseOpcodeMask) == JALR; +} + +bool AssemblerRISCVI::IsLui(Instr instr) { + return (instr & kBaseOpcodeMask) == LUI; +} +bool AssemblerRISCVI::IsAuipc(Instr instr) { + return (instr & kBaseOpcodeMask) == AUIPC; +} +bool AssemblerRISCVI::IsAddi(Instr instr) { + return (instr & (kBaseOpcodeMask | kFunct3Mask)) == RO_ADDI; +} +bool AssemblerRISCVI::IsOri(Instr instr) { + return (instr & (kBaseOpcodeMask | kFunct3Mask)) == RO_ORI; +} +bool AssemblerRISCVI::IsSlli(Instr instr) { + return (instr & (kBaseOpcodeMask | kFunct3Mask)) == RO_SLLI; +} + +int AssemblerRISCVI::JumpOffset(Instr instr) { + int32_t imm21 = ((instr & 0x7fe00000) >> 20) | ((instr & 0x100000) >> 9) | + (instr & 0xff000) | ((instr & 0x80000000) >> 11); + imm21 = imm21 << 11 >> 11; + return imm21; +} + +int AssemblerRISCVI::JalrOffset(Instr instr) { + MOZ_ASSERT(IsJalr(instr)); + int32_t imm12 = static_cast<int32_t>(instr & kImm12Mask) >> 20; + return imm12; +} + +int AssemblerRISCVI::AuipcOffset(Instr instr) { + MOZ_ASSERT(IsAuipc(instr)); + int32_t imm20 = static_cast<int32_t>(instr & kImm20Mask); + return imm20; +} + +bool AssemblerRISCVI::IsLw(Instr instr) { + return (instr & (kBaseOpcodeMask | kFunct3Mask)) == RO_LW; +} + +int AssemblerRISCVI::LoadOffset(Instr instr) { +#if JS_CODEGEN_RISCV64 + MOZ_ASSERT(IsLd(instr)); +#elif V8_TARGET_ARCH_RISCV32 + MOZ_ASSERT(IsLw(instr)); +#endif + int32_t imm12 = static_cast<int32_t>(instr & kImm12Mask) >> 20; + return imm12; +} + +#ifdef JS_CODEGEN_RISCV64 + +bool AssemblerRISCVI::IsAddiw(Instr instr) { + return (instr & (kBaseOpcodeMask | kFunct3Mask)) == RO_ADDIW; +} + +bool AssemblerRISCVI::IsLd(Instr instr) { + return (instr & (kBaseOpcodeMask | kFunct3Mask)) == RO_LD; +} + +void AssemblerRISCVI::lwu(Register rd, Register rs1, int16_t imm12) { + GenInstrLoad_ri(0b110, rd, rs1, imm12); +} + +void AssemblerRISCVI::ld(Register rd, Register rs1, int16_t imm12) { + GenInstrLoad_ri(0b011, rd, rs1, imm12); +} + +void AssemblerRISCVI::sd(Register source, Register base, int16_t imm12) { + GenInstrStore_rri(0b011, base, source, imm12); +} + +void AssemblerRISCVI::addiw(Register rd, Register rs1, int16_t imm12) { + GenInstrI(0b000, OP_IMM_32, rd, rs1, imm12); +} + +void AssemblerRISCVI::slliw(Register rd, Register rs1, uint8_t shamt) { + GenInstrShiftW_ri(0, 0b001, rd, rs1, shamt & 0x1f); +} + +void AssemblerRISCVI::srliw(Register rd, Register rs1, uint8_t shamt) { + GenInstrShiftW_ri(0, 0b101, rd, rs1, shamt & 0x1f); +} + +void AssemblerRISCVI::sraiw(Register rd, Register rs1, uint8_t shamt) { + GenInstrShiftW_ri(1, 0b101, rd, rs1, shamt & 0x1f); +} + +void AssemblerRISCVI::addw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0000000, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVI::subw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0100000, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVI::sllw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0000000, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVI::srlw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0000000, 0b101, rd, rs1, rs2); +} + +void AssemblerRISCVI::sraw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0100000, 0b101, rd, rs1, rs2); +} + +#endif + +int AssemblerRISCVI::BranchOffset(Instr instr) { + // | imm[12] | imm[10:5] | rs2 | rs1 | funct3 | imm[4:1|11] | opcode | + // 31 25 11 7 + int32_t imm13 = ((instr & 0xf00) >> 7) | ((instr & 0x7e000000) >> 20) | + ((instr & 0x80) << 4) | ((instr & 0x80000000) >> 19); + imm13 = imm13 << 19 >> 19; + return imm13; +} + +int AssemblerRISCVI::BrachlongOffset(Instr auipc, Instr instr_I) { + MOZ_ASSERT(reinterpret_cast<Instruction*>(&instr_I)->InstructionType() == + InstructionBase::kIType); + MOZ_ASSERT(IsAuipc(auipc)); + MOZ_ASSERT(((auipc & kRdFieldMask) >> kRdShift) == + ((instr_I & kRs1FieldMask) >> kRs1Shift)); + int32_t imm_auipc = AuipcOffset(auipc); + int32_t imm12 = static_cast<int32_t>(instr_I & kImm12Mask) >> 20; + int32_t offset = imm12 + imm_auipc; + return offset; +} + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/base-riscv-i.h b/js/src/jit/riscv64/extension/base-riscv-i.h new file mode 100644 index 0000000000..cca342c960 --- /dev/null +++ b/js/src/jit/riscv64/extension/base-riscv-i.h @@ -0,0 +1,273 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_extension_Base_riscv_i_h_ +#define jit_riscv64_extension_Base_riscv_i_h_ +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/extension/base-assembler-riscv.h" +namespace js { +namespace jit { + +class AssemblerRISCVI : public AssemblerRiscvBase { + public: + void lui(Register rd, int32_t imm20); + void auipc(Register rd, int32_t imm20); + + // Jumps + void jal(Register rd, int32_t imm20); + void jalr(Register rd, Register rs1, int16_t imm12); + + // Branches + void beq(Register rs1, Register rs2, int16_t imm12); + void bne(Register rs1, Register rs2, int16_t imm12); + void blt(Register rs1, Register rs2, int16_t imm12); + void bge(Register rs1, Register rs2, int16_t imm12); + void bltu(Register rs1, Register rs2, int16_t imm12); + void bgeu(Register rs1, Register rs2, int16_t imm12); + // Loads + void lb(Register rd, Register rs1, int16_t imm12); + void lh(Register rd, Register rs1, int16_t imm12); + void lw(Register rd, Register rs1, int16_t imm12); + void lbu(Register rd, Register rs1, int16_t imm12); + void lhu(Register rd, Register rs1, int16_t imm12); + + // Stores + void sb(Register source, Register base, int16_t imm12); + void sh(Register source, Register base, int16_t imm12); + void sw(Register source, Register base, int16_t imm12); + + // Arithmetic with immediate + void addi(Register rd, Register rs1, int16_t imm12); + void slti(Register rd, Register rs1, int16_t imm12); + void sltiu(Register rd, Register rs1, int16_t imm12); + void xori(Register rd, Register rs1, int16_t imm12); + void ori(Register rd, Register rs1, int16_t imm12); + void andi(Register rd, Register rs1, int16_t imm12); + void slli(Register rd, Register rs1, uint8_t shamt); + void srli(Register rd, Register rs1, uint8_t shamt); + void srai(Register rd, Register rs1, uint8_t shamt); + + // Arithmetic + void add(Register rd, Register rs1, Register rs2); + void sub(Register rd, Register rs1, Register rs2); + void sll(Register rd, Register rs1, Register rs2); + void slt(Register rd, Register rs1, Register rs2); + void sltu(Register rd, Register rs1, Register rs2); + void xor_(Register rd, Register rs1, Register rs2); + void srl(Register rd, Register rs1, Register rs2); + void sra(Register rd, Register rs1, Register rs2); + void or_(Register rd, Register rs1, Register rs2); + void and_(Register rd, Register rs1, Register rs2); + + // Other pseudo instructions that are not part of RISCV pseudo assemly + void nor(Register rd, Register rs, Register rt) { + or_(rd, rs, rt); + not_(rd, rd); + } + + // Memory fences + void fence(uint8_t pred, uint8_t succ); + void fence_tso(); + + // Environment call / break + void ecall(); + void ebreak(); + + void sync() { fence(0b1111, 0b1111); } + + // This is a de facto standard (as set by GNU binutils) 32-bit unimplemented + // instruction (i.e., it should always trap, if your implementation has + // invalid instruction traps). + void unimp(); + + static int JumpOffset(Instr instr); + static int AuipcOffset(Instr instr); + static int JalrOffset(Instr instr); + static int LoadOffset(Instr instr); + static int BranchOffset(Instr instr); + static int BrachlongOffset(Instr auipc, Instr instr_I); + static inline Instr SetBranchOffset(int32_t pos, int32_t target_pos, + Instr instr) { + int32_t imm = target_pos - pos; + MOZ_ASSERT((imm & 1) == 0); + MOZ_ASSERT(is_intn(imm, kBranchOffsetBits)); + + instr &= ~kBImm12Mask; + int32_t imm12 = ((imm & 0x800) >> 4) | // bit 11 + ((imm & 0x1e) << 7) | // bits 4-1 + ((imm & 0x7e0) << 20) | // bits 10-5 + ((imm & 0x1000) << 19); // bit 12 + + return instr | (imm12 & kBImm12Mask); + } + + static inline Instr SetJalOffset(int32_t pos, int32_t target_pos, + Instr instr) { + MOZ_ASSERT(IsJal(instr)); + int32_t imm = target_pos - pos; + MOZ_ASSERT((imm & 1) == 0); + MOZ_ASSERT(is_intn(imm, kJumpOffsetBits)); + + instr &= ~kImm20Mask; + int32_t imm20 = (imm & 0xff000) | // bits 19-12 + ((imm & 0x800) << 9) | // bit 11 + ((imm & 0x7fe) << 20) | // bits 10-1 + ((imm & 0x100000) << 11); // bit 20 + + return instr | (imm20 & kImm20Mask); + } + + static inline Instr SetJalrOffset(int32_t offset, Instr instr) { + MOZ_ASSERT(IsJalr(instr)); + MOZ_ASSERT(is_int12(offset)); + instr &= ~kImm12Mask; + int32_t imm12 = offset << kImm12Shift; + MOZ_ASSERT(IsJalr(instr | (imm12 & kImm12Mask))); + MOZ_ASSERT(JalrOffset(instr | (imm12 & kImm12Mask)) == offset); + return instr | (imm12 & kImm12Mask); + } + + static inline Instr SetLoadOffset(int32_t offset, Instr instr) { +#if JS_CODEGEN_RISCV64 + MOZ_ASSERT(IsLd(instr)); +#elif JS_CODEGEN_RISCV32 + MOZ_ASSERT(IsLw(instr)); +#endif + MOZ_ASSERT(is_int12(offset)); + instr &= ~kImm12Mask; + int32_t imm12 = offset << kImm12Shift; + return instr | (imm12 & kImm12Mask); + } + + static inline Instr SetAuipcOffset(int32_t offset, Instr instr) { + MOZ_ASSERT(IsAuipc(instr)); + MOZ_ASSERT(is_int20(offset)); + instr = (instr & ~kImm31_12Mask) | ((offset & kImm19_0Mask) << 12); + return instr; + } + + // Check if an instruction is a branch of some kind. + static bool IsBranch(Instr instr); + static bool IsNop(Instr instr); + static bool IsJump(Instr instr); + static bool IsJal(Instr instr); + static bool IsJalr(Instr instr); + static bool IsLui(Instr instr); + static bool IsAuipc(Instr instr); + static bool IsAddi(Instr instr); + static bool IsOri(Instr instr); + static bool IsSlli(Instr instr); + static bool IsLw(Instr instr); + + inline int32_t branch_offset(Label* L) { + return branch_offset_helper(L, OffsetSize::kOffset13); + } + inline int32_t jump_offset(Label* L) { + return branch_offset_helper(L, OffsetSize::kOffset21); + } + + // Branches + void beq(Register rs1, Register rs2, Label* L) { + beq(rs1, rs2, branch_offset(L)); + } + void bne(Register rs1, Register rs2, Label* L) { + bne(rs1, rs2, branch_offset(L)); + } + void blt(Register rs1, Register rs2, Label* L) { + blt(rs1, rs2, branch_offset(L)); + } + void bge(Register rs1, Register rs2, Label* L) { + bge(rs1, rs2, branch_offset(L)); + } + void bltu(Register rs1, Register rs2, Label* L) { + bltu(rs1, rs2, branch_offset(L)); + } + void bgeu(Register rs1, Register rs2, Label* L) { + bgeu(rs1, rs2, branch_offset(L)); + } + + void beqz(Register rs, int16_t imm13) { beq(rs, zero_reg, imm13); } + void beqz(Register rs1, Label* L) { beqz(rs1, branch_offset(L)); } + void bnez(Register rs, int16_t imm13) { bne(rs, zero_reg, imm13); } + void bnez(Register rs1, Label* L) { bnez(rs1, branch_offset(L)); } + void blez(Register rs, int16_t imm13) { bge(zero_reg, rs, imm13); } + void blez(Register rs1, Label* L) { blez(rs1, branch_offset(L)); } + void bgez(Register rs, int16_t imm13) { bge(rs, zero_reg, imm13); } + void bgez(Register rs1, Label* L) { bgez(rs1, branch_offset(L)); } + void bltz(Register rs, int16_t imm13) { blt(rs, zero_reg, imm13); } + void bltz(Register rs1, Label* L) { bltz(rs1, branch_offset(L)); } + void bgtz(Register rs, int16_t imm13) { blt(zero_reg, rs, imm13); } + + void bgtz(Register rs1, Label* L) { bgtz(rs1, branch_offset(L)); } + void bgt(Register rs1, Register rs2, int16_t imm13) { blt(rs2, rs1, imm13); } + void bgt(Register rs1, Register rs2, Label* L) { + bgt(rs1, rs2, branch_offset(L)); + } + void ble(Register rs1, Register rs2, int16_t imm13) { bge(rs2, rs1, imm13); } + void ble(Register rs1, Register rs2, Label* L) { + ble(rs1, rs2, branch_offset(L)); + } + void bgtu(Register rs1, Register rs2, int16_t imm13) { + bltu(rs2, rs1, imm13); + } + void bgtu(Register rs1, Register rs2, Label* L) { + bgtu(rs1, rs2, branch_offset(L)); + } + void bleu(Register rs1, Register rs2, int16_t imm13) { + bgeu(rs2, rs1, imm13); + } + void bleu(Register rs1, Register rs2, Label* L) { + bleu(rs1, rs2, branch_offset(L)); + } + + void j(int32_t imm21) { jal(zero_reg, imm21); } + void j(Label* L) { j(jump_offset(L)); } + void b(Label* L) { j(L); } + void jal(int32_t imm21) { jal(ra, imm21); } + void jal(Label* L) { jal(jump_offset(L)); } + void jr(Register rs) { jalr(zero_reg, rs, 0); } + void jr(Register rs, int32_t imm12) { jalr(zero_reg, rs, imm12); } + void jalr(Register rs, int32_t imm12) { jalr(ra, rs, imm12); } + void jalr(Register rs) { jalr(ra, rs, 0); } + void call(int32_t offset) { + auipc(ra, (offset >> 12) + ((offset & 0x800) >> 11)); + jalr(ra, ra, offset << 20 >> 20); + } + + void mv(Register rd, Register rs) { addi(rd, rs, 0); } + void not_(Register rd, Register rs) { xori(rd, rs, -1); } + void neg(Register rd, Register rs) { sub(rd, zero_reg, rs); } + void seqz(Register rd, Register rs) { sltiu(rd, rs, 1); } + void snez(Register rd, Register rs) { sltu(rd, zero_reg, rs); } + void sltz(Register rd, Register rs) { slt(rd, rs, zero_reg); } + void sgtz(Register rd, Register rs) { slt(rd, zero_reg, rs); } + +#if JS_CODEGEN_RISCV64 + void lwu(Register rd, Register rs1, int16_t imm12); + void ld(Register rd, Register rs1, int16_t imm12); + void sd(Register source, Register base, int16_t imm12); + void addiw(Register rd, Register rs1, int16_t imm12); + void slliw(Register rd, Register rs1, uint8_t shamt); + void srliw(Register rd, Register rs1, uint8_t shamt); + void sraiw(Register rd, Register rs1, uint8_t shamt); + void addw(Register rd, Register rs1, Register rs2); + void subw(Register rd, Register rs1, Register rs2); + void sllw(Register rd, Register rs1, Register rs2); + void srlw(Register rd, Register rs1, Register rs2); + void sraw(Register rd, Register rs1, Register rs2); + void negw(Register rd, Register rs) { subw(rd, zero_reg, rs); } + void sext_w(Register rd, Register rs) { addiw(rd, rs, 0); } + + static bool IsAddiw(Instr instr); + static bool IsLd(Instr instr); +#endif +}; + +} // namespace jit +} // namespace js + +#endif // jit_riscv64_extension_Base_riscv_I_h_ diff --git a/js/src/jit/riscv64/extension/extension-riscv-a.cc b/js/src/jit/riscv64/extension/extension-riscv-a.cc new file mode 100644 index 0000000000..ead355fc0a --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-a.cc @@ -0,0 +1,123 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/extension/extension-riscv-a.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { + +// RV32A Standard Extension +void AssemblerRISCVA::lr_w(bool aq, bool rl, Register rd, Register rs1) { + GenInstrRAtomic(0b00010, aq, rl, 0b010, rd, rs1, zero_reg); +} + +void AssemblerRISCVA::sc_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b00011, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoswap_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b00001, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoadd_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b00000, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoxor_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b00100, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoand_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b01100, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoor_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b01000, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amomin_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b10000, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amomax_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b10100, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amominu_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b11000, aq, rl, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVA::amomaxu_w(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b11100, aq, rl, 0b010, rd, rs1, rs2); +} + +// RV64A Standard Extension (in addition to RV32A) +#ifdef JS_CODEGEN_RISCV64 +void AssemblerRISCVA::lr_d(bool aq, bool rl, Register rd, Register rs1) { + GenInstrRAtomic(0b00010, aq, rl, 0b011, rd, rs1, zero_reg); +} + +void AssemblerRISCVA::sc_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b00011, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoswap_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b00001, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoadd_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b00000, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoxor_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b00100, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoand_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b01100, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amoor_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b01000, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amomin_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b10000, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amomax_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b10100, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amominu_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b11000, aq, rl, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVA::amomaxu_d(bool aq, bool rl, Register rd, Register rs1, + Register rs2) { + GenInstrRAtomic(0b11100, aq, rl, 0b011, rd, rs1, rs2); +} +#endif +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/extension-riscv-a.h b/js/src/jit/riscv64/extension/extension-riscv-a.h new file mode 100644 index 0000000000..442a4f5bba --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-a.h @@ -0,0 +1,46 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file." +#ifndef jit_riscv64_extension_Extension_riscv_a_h_ +#define jit_riscv64_extension_Extension_riscv_a_h_ +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/Register-riscv64.h" +namespace js { +namespace jit { +class AssemblerRISCVA : public AssemblerRiscvBase { + // RV32A Standard Extension + public: + void lr_w(bool aq, bool rl, Register rd, Register rs1); + void sc_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoswap_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoadd_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoxor_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoand_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoor_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amomin_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amomax_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amominu_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amomaxu_w(bool aq, bool rl, Register rd, Register rs1, Register rs2); + +#ifdef JS_CODEGEN_RISCV64 + // RV64A Standard Extension (in addition to RV32A) + void lr_d(bool aq, bool rl, Register rd, Register rs1); + void sc_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoswap_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoadd_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoxor_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoand_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amoor_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amomin_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amomax_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amominu_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); + void amomaxu_d(bool aq, bool rl, Register rd, Register rs1, Register rs2); +#endif +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_extension_Extension_riscv_A_h_ diff --git a/js/src/jit/riscv64/extension/extension-riscv-c.cc b/js/src/jit/riscv64/extension/extension-riscv-c.cc new file mode 100644 index 0000000000..714753a0e0 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-c.cc @@ -0,0 +1,275 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/extension/extension-riscv-c.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { +// RV64C Standard Extension +void AssemblerRISCVC::c_nop() { GenInstrCI(0b000, C1, zero_reg, 0); } + +void AssemblerRISCVC::c_addi(Register rd, int8_t imm6) { + MOZ_ASSERT(rd != zero_reg && imm6 != 0); + GenInstrCI(0b000, C1, rd, imm6); +} + +#ifdef JS_CODEGEN_RISCV64 +void AssemblerRISCVC::c_addiw(Register rd, int8_t imm6) { + MOZ_ASSERT(rd != zero_reg); + GenInstrCI(0b001, C1, rd, imm6); +} +#endif + +void AssemblerRISCVC::c_addi16sp(int16_t imm10) { + MOZ_ASSERT(is_int10(imm10) && (imm10 & 0xf) == 0); + uint8_t uimm6 = ((imm10 & 0x200) >> 4) | (imm10 & 0x10) | + ((imm10 & 0x40) >> 3) | ((imm10 & 0x180) >> 6) | + ((imm10 & 0x20) >> 5); + GenInstrCIU(0b011, C1, sp, uimm6); +} + +void AssemblerRISCVC::c_addi4spn(Register rd, int16_t uimm10) { + MOZ_ASSERT(is_uint10(uimm10) && (uimm10 != 0)); + uint8_t uimm8 = ((uimm10 & 0x4) >> 1) | ((uimm10 & 0x8) >> 3) | + ((uimm10 & 0x30) << 2) | ((uimm10 & 0x3c0) >> 4); + GenInstrCIW(0b000, C0, rd, uimm8); +} + +void AssemblerRISCVC::c_li(Register rd, int8_t imm6) { + MOZ_ASSERT(rd != zero_reg); + GenInstrCI(0b010, C1, rd, imm6); +} + +void AssemblerRISCVC::c_lui(Register rd, int8_t imm6) { + MOZ_ASSERT(rd != zero_reg && rd != sp && imm6 != 0); + GenInstrCI(0b011, C1, rd, imm6); +} + +void AssemblerRISCVC::c_slli(Register rd, uint8_t shamt6) { + MOZ_ASSERT(rd != zero_reg && shamt6 != 0); + GenInstrCIU(0b000, C2, rd, shamt6); +} + +void AssemblerRISCVC::c_fldsp(FPURegister rd, uint16_t uimm9) { + MOZ_ASSERT(is_uint9(uimm9) && (uimm9 & 0x7) == 0); + uint8_t uimm6 = (uimm9 & 0x38) | ((uimm9 & 0x1c0) >> 6); + GenInstrCIU(0b001, C2, rd, uimm6); +} + +#ifdef JS_CODEGEN_RISCV64 +void AssemblerRISCVC::c_ldsp(Register rd, uint16_t uimm9) { + MOZ_ASSERT(rd != zero_reg && is_uint9(uimm9) && (uimm9 & 0x7) == 0); + uint8_t uimm6 = (uimm9 & 0x38) | ((uimm9 & 0x1c0) >> 6); + GenInstrCIU(0b011, C2, rd, uimm6); +} +#endif + +void AssemblerRISCVC::c_lwsp(Register rd, uint16_t uimm8) { + MOZ_ASSERT(rd != zero_reg && is_uint8(uimm8) && (uimm8 & 0x3) == 0); + uint8_t uimm6 = (uimm8 & 0x3c) | ((uimm8 & 0xc0) >> 6); + GenInstrCIU(0b010, C2, rd, uimm6); +} + +void AssemblerRISCVC::c_jr(Register rs1) { + MOZ_ASSERT(rs1 != zero_reg); + GenInstrCR(0b1000, C2, rs1, zero_reg); +} + +void AssemblerRISCVC::c_mv(Register rd, Register rs2) { + MOZ_ASSERT(rd != zero_reg && rs2 != zero_reg); + GenInstrCR(0b1000, C2, rd, rs2); +} + +void AssemblerRISCVC::c_ebreak() { GenInstrCR(0b1001, C2, zero_reg, zero_reg); } + +void AssemblerRISCVC::c_jalr(Register rs1) { + MOZ_ASSERT(rs1 != zero_reg); + GenInstrCR(0b1001, C2, rs1, zero_reg); +} + +void AssemblerRISCVC::c_add(Register rd, Register rs2) { + MOZ_ASSERT(rd != zero_reg && rs2 != zero_reg); + GenInstrCR(0b1001, C2, rd, rs2); +} + +// CA Instructions +void AssemblerRISCVC::c_sub(Register rd, Register rs2) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs2.code() & 0b11000) == 0b01000)); + GenInstrCA(0b100011, C1, rd, 0b00, rs2); +} + +void AssemblerRISCVC::c_xor(Register rd, Register rs2) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs2.code() & 0b11000) == 0b01000)); + GenInstrCA(0b100011, C1, rd, 0b01, rs2); +} + +void AssemblerRISCVC::c_or(Register rd, Register rs2) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs2.code() & 0b11000) == 0b01000)); + GenInstrCA(0b100011, C1, rd, 0b10, rs2); +} + +void AssemblerRISCVC::c_and(Register rd, Register rs2) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs2.code() & 0b11000) == 0b01000)); + GenInstrCA(0b100011, C1, rd, 0b11, rs2); +} + +#ifdef JS_CODEGEN_RISCV64 +void AssemblerRISCVC::c_subw(Register rd, Register rs2) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs2.code() & 0b11000) == 0b01000)); + GenInstrCA(0b100111, C1, rd, 0b00, rs2); +} + +void AssemblerRISCVC::c_addw(Register rd, Register rs2) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs2.code() & 0b11000) == 0b01000)); + GenInstrCA(0b100111, C1, rd, 0b01, rs2); +} +#endif + +void AssemblerRISCVC::c_swsp(Register rs2, uint16_t uimm8) { + MOZ_ASSERT(is_uint8(uimm8) && (uimm8 & 0x3) == 0); + uint8_t uimm6 = (uimm8 & 0x3c) | ((uimm8 & 0xc0) >> 6); + GenInstrCSS(0b110, C2, rs2, uimm6); +} + +#ifdef JS_CODEGEN_RISCV64 +void AssemblerRISCVC::c_sdsp(Register rs2, uint16_t uimm9) { + MOZ_ASSERT(is_uint9(uimm9) && (uimm9 & 0x7) == 0); + uint8_t uimm6 = (uimm9 & 0x38) | ((uimm9 & 0x1c0) >> 6); + GenInstrCSS(0b111, C2, rs2, uimm6); +} +#endif + +void AssemblerRISCVC::c_fsdsp(FPURegister rs2, uint16_t uimm9) { + MOZ_ASSERT(is_uint9(uimm9) && (uimm9 & 0x7) == 0); + uint8_t uimm6 = (uimm9 & 0x38) | ((uimm9 & 0x1c0) >> 6); + GenInstrCSS(0b101, C2, rs2, uimm6); +} + +// CL Instructions + +void AssemblerRISCVC::c_lw(Register rd, Register rs1, uint16_t uimm7) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs1.code() & 0b11000) == 0b01000) && is_uint7(uimm7) && + ((uimm7 & 0x3) == 0)); + uint8_t uimm5 = + ((uimm7 & 0x4) >> 1) | ((uimm7 & 0x40) >> 6) | ((uimm7 & 0x38) >> 1); + GenInstrCL(0b010, C0, rd, rs1, uimm5); +} + +#ifdef JS_CODEGEN_RISCV64 +void AssemblerRISCVC::c_ld(Register rd, Register rs1, uint16_t uimm8) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs1.code() & 0b11000) == 0b01000) && is_uint8(uimm8) && + ((uimm8 & 0x7) == 0)); + uint8_t uimm5 = ((uimm8 & 0x38) >> 1) | ((uimm8 & 0xc0) >> 6); + GenInstrCL(0b011, C0, rd, rs1, uimm5); +} +#endif + +void AssemblerRISCVC::c_fld(FPURegister rd, Register rs1, uint16_t uimm8) { + MOZ_ASSERT(((rd.code() & 0b11000) == 0b01000) && + ((rs1.code() & 0b11000) == 0b01000) && is_uint8(uimm8) && + ((uimm8 & 0x7) == 0)); + uint8_t uimm5 = ((uimm8 & 0x38) >> 1) | ((uimm8 & 0xc0) >> 6); + GenInstrCL(0b001, C0, rd, rs1, uimm5); +} + +// CS Instructions + +void AssemblerRISCVC::c_sw(Register rs2, Register rs1, uint16_t uimm7) { + MOZ_ASSERT(((rs2.code() & 0b11000) == 0b01000) && + ((rs1.code() & 0b11000) == 0b01000) && is_uint7(uimm7) && + ((uimm7 & 0x3) == 0)); + uint8_t uimm5 = + ((uimm7 & 0x4) >> 1) | ((uimm7 & 0x40) >> 6) | ((uimm7 & 0x38) >> 1); + GenInstrCS(0b110, C0, rs2, rs1, uimm5); +} + +#ifdef JS_CODEGEN_RISCV64 +void AssemblerRISCVC::c_sd(Register rs2, Register rs1, uint16_t uimm8) { + MOZ_ASSERT(((rs2.code() & 0b11000) == 0b01000) && + ((rs1.code() & 0b11000) == 0b01000) && is_uint8(uimm8) && + ((uimm8 & 0x7) == 0)); + uint8_t uimm5 = ((uimm8 & 0x38) >> 1) | ((uimm8 & 0xc0) >> 6); + GenInstrCS(0b111, C0, rs2, rs1, uimm5); +} +#endif + +void AssemblerRISCVC::c_fsd(FPURegister rs2, Register rs1, uint16_t uimm8) { + MOZ_ASSERT(((rs2.code() & 0b11000) == 0b01000) && + ((rs1.code() & 0b11000) == 0b01000) && is_uint8(uimm8) && + ((uimm8 & 0x7) == 0)); + uint8_t uimm5 = ((uimm8 & 0x38) >> 1) | ((uimm8 & 0xc0) >> 6); + GenInstrCS(0b101, C0, rs2, rs1, uimm5); +} + +// CJ Instructions + +void AssemblerRISCVC::c_j(int16_t imm12) { + MOZ_ASSERT(is_int12(imm12)); + int16_t uimm11 = ((imm12 & 0x800) >> 1) | ((imm12 & 0x400) >> 4) | + ((imm12 & 0x300) >> 1) | ((imm12 & 0x80) >> 3) | + ((imm12 & 0x40) >> 1) | ((imm12 & 0x20) >> 5) | + ((imm12 & 0x10) << 5) | (imm12 & 0xe); + GenInstrCJ(0b101, C1, uimm11); +} + +// CB Instructions + +void AssemblerRISCVC::c_bnez(Register rs1, int16_t imm9) { + MOZ_ASSERT(((rs1.code() & 0b11000) == 0b01000) && is_int9(imm9)); + uint8_t uimm8 = ((imm9 & 0x20) >> 5) | ((imm9 & 0x6)) | ((imm9 & 0xc0) >> 3) | + ((imm9 & 0x18) << 2) | ((imm9 & 0x100) >> 1); + GenInstrCB(0b111, C1, rs1, uimm8); +} + +void AssemblerRISCVC::c_beqz(Register rs1, int16_t imm9) { + MOZ_ASSERT(((rs1.code() & 0b11000) == 0b01000) && is_int9(imm9)); + uint8_t uimm8 = ((imm9 & 0x20) >> 5) | ((imm9 & 0x6)) | ((imm9 & 0xc0) >> 3) | + ((imm9 & 0x18) << 2) | ((imm9 & 0x100) >> 1); + GenInstrCB(0b110, C1, rs1, uimm8); +} + +void AssemblerRISCVC::c_srli(Register rs1, int8_t shamt6) { + MOZ_ASSERT(((rs1.code() & 0b11000) == 0b01000) && is_int6(shamt6)); + GenInstrCBA(0b100, 0b00, C1, rs1, shamt6); +} + +void AssemblerRISCVC::c_srai(Register rs1, int8_t shamt6) { + MOZ_ASSERT(((rs1.code() & 0b11000) == 0b01000) && is_int6(shamt6)); + GenInstrCBA(0b100, 0b01, C1, rs1, shamt6); +} + +void AssemblerRISCVC::c_andi(Register rs1, int8_t imm6) { + MOZ_ASSERT(((rs1.code() & 0b11000) == 0b01000) && is_int6(imm6)); + GenInstrCBA(0b100, 0b10, C1, rs1, imm6); +} + +bool AssemblerRISCVC::IsCJal(Instr instr) { + return (instr & kRvcOpcodeMask) == RO_C_J; +} + +bool AssemblerRISCVC::IsCBranch(Instr instr) { + int Op = instr & kRvcOpcodeMask; + return Op == RO_C_BNEZ || Op == RO_C_BEQZ; +} + +int AssemblerRISCVC::CJumpOffset(Instr instr) { + int32_t imm12 = ((instr & 0x4) << 3) | ((instr & 0x38) >> 2) | + ((instr & 0x40) << 1) | ((instr & 0x80) >> 1) | + ((instr & 0x100) << 2) | ((instr & 0x600) >> 1) | + ((instr & 0x800) >> 7) | ((instr & 0x1000) >> 1); + imm12 = imm12 << 20 >> 20; + return imm12; +} + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/extension-riscv-c.h b/js/src/jit/riscv64/extension/extension-riscv-c.h new file mode 100644 index 0000000000..655141cb30 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-c.h @@ -0,0 +1,77 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_extension_Extension_riscv_c_h_ +#define jit_riscv64_extension_Extension_riscv_c_h_ +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/Register-riscv64.h" +namespace js { +namespace jit { +class AssemblerRISCVC : public AssemblerRiscvBase { + // RV64C Standard Extension + public: + void c_nop(); + void c_addi(Register rd, int8_t imm6); + + void c_addi16sp(int16_t imm10); + void c_addi4spn(Register rd, int16_t uimm10); + void c_li(Register rd, int8_t imm6); + void c_lui(Register rd, int8_t imm6); + void c_slli(Register rd, uint8_t shamt6); + void c_lwsp(Register rd, uint16_t uimm8); + void c_jr(Register rs1); + void c_mv(Register rd, Register rs2); + void c_ebreak(); + void c_jalr(Register rs1); + void c_j(int16_t imm12); + void c_add(Register rd, Register rs2); + void c_sub(Register rd, Register rs2); + void c_and(Register rd, Register rs2); + void c_xor(Register rd, Register rs2); + void c_or(Register rd, Register rs2); + void c_swsp(Register rs2, uint16_t uimm8); + void c_lw(Register rd, Register rs1, uint16_t uimm7); + void c_sw(Register rs2, Register rs1, uint16_t uimm7); + void c_bnez(Register rs1, int16_t imm9); + void c_beqz(Register rs1, int16_t imm9); + void c_srli(Register rs1, int8_t shamt6); + void c_srai(Register rs1, int8_t shamt6); + void c_andi(Register rs1, int8_t imm6); + + void c_fld(FPURegister rd, Register rs1, uint16_t uimm8); + void c_fsd(FPURegister rs2, Register rs1, uint16_t uimm8); + void c_fldsp(FPURegister rd, uint16_t uimm9); + void c_fsdsp(FPURegister rs2, uint16_t uimm9); +#ifdef JS_CODEGEN_RISCV64 + void c_ld(Register rd, Register rs1, uint16_t uimm8); + void c_sd(Register rs2, Register rs1, uint16_t uimm8); + void c_subw(Register rd, Register rs2); + void c_addw(Register rd, Register rs2); + void c_addiw(Register rd, int8_t imm6); + void c_ldsp(Register rd, uint16_t uimm9); + void c_sdsp(Register rs2, uint16_t uimm9); +#endif + + int CJumpOffset(Instr instr); + + static bool IsCBranch(Instr instr); + static bool IsCJal(Instr instr); + + inline int16_t cjump_offset(Label* L) { + return (int16_t)branch_offset_helper(L, OffsetSize::kOffset11); + } + inline int32_t cbranch_offset(Label* L) { + return branch_offset_helper(L, OffsetSize::kOffset9); + } + + void c_j(Label* L) { c_j(cjump_offset(L)); } + void c_bnez(Register rs1, Label* L) { c_bnez(rs1, cbranch_offset(L)); } + void c_beqz(Register rs1, Label* L) { c_beqz(rs1, cbranch_offset(L)); } +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_extension_Extension_riscv_C_h_ diff --git a/js/src/jit/riscv64/extension/extension-riscv-d.cc b/js/src/jit/riscv64/extension/extension-riscv-d.cc new file mode 100644 index 0000000000..cb728baf12 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-d.cc @@ -0,0 +1,167 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/extension/extension-riscv-d.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { +// RV32D Standard Extension + +void AssemblerRISCVD::fld(FPURegister rd, Register rs1, int16_t imm12) { + GenInstrLoadFP_ri(0b011, rd, rs1, imm12); +} + +void AssemblerRISCVD::fsd(FPURegister source, Register base, int16_t imm12) { + GenInstrStoreFP_rri(0b011, base, source, imm12); +} + +void AssemblerRISCVD::fmadd_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm) { + GenInstrR4(0b01, MADD, rd, rs1, rs2, rs3, frm); +} + +void AssemblerRISCVD::fmsub_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm) { + GenInstrR4(0b01, MSUB, rd, rs1, rs2, rs3, frm); +} + +void AssemblerRISCVD::fnmsub_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm) { + GenInstrR4(0b01, NMSUB, rd, rs1, rs2, rs3, frm); +} + +void AssemblerRISCVD::fnmadd_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm) { + GenInstrR4(0b01, NMADD, rd, rs1, rs2, rs3, frm); +} + +void AssemblerRISCVD::fadd_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0000001, frm, rd, rs1, rs2); +} + +void AssemblerRISCVD::fsub_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0000101, frm, rd, rs1, rs2); +} + +void AssemblerRISCVD::fmul_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0001001, frm, rd, rs1, rs2); +} + +void AssemblerRISCVD::fdiv_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0001101, frm, rd, rs1, rs2); +} + +void AssemblerRISCVD::fsqrt_d(FPURegister rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0101101, frm, rd, rs1, zero_reg); +} + +void AssemblerRISCVD::fsgnj_d(FPURegister rd, FPURegister rs1, + FPURegister rs2) { + GenInstrALUFP_rr(0b0010001, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVD::fsgnjn_d(FPURegister rd, FPURegister rs1, + FPURegister rs2) { + GenInstrALUFP_rr(0b0010001, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVD::fsgnjx_d(FPURegister rd, FPURegister rs1, + FPURegister rs2) { + GenInstrALUFP_rr(0b0010001, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVD::fmin_d(FPURegister rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b0010101, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVD::fmax_d(FPURegister rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b0010101, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVD::fcvt_s_d(FPURegister rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0100000, frm, rd, rs1, ToRegister(1)); +} + +void AssemblerRISCVD::fcvt_d_s(FPURegister rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0100001, frm, rd, rs1, zero_reg); +} + +void AssemblerRISCVD::feq_d(Register rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b1010001, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVD::flt_d(Register rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b1010001, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVD::fle_d(Register rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b1010001, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVD::fclass_d(Register rd, FPURegister rs1) { + GenInstrALUFP_rr(0b1110001, 0b001, rd, rs1, zero_reg); +} + +void AssemblerRISCVD::fcvt_w_d(Register rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1100001, frm, rd, rs1, zero_reg); +} + +void AssemblerRISCVD::fcvt_wu_d(Register rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1100001, frm, rd, rs1, ToRegister(1)); +} + +void AssemblerRISCVD::fcvt_d_w(FPURegister rd, Register rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1101001, frm, rd, rs1, zero_reg); +} + +void AssemblerRISCVD::fcvt_d_wu(FPURegister rd, Register rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1101001, frm, rd, rs1, ToRegister(1)); +} + +#ifdef JS_CODEGEN_RISCV64 +// RV64D Standard Extension (in addition to RV32D) + +void AssemblerRISCVD::fcvt_l_d(Register rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1100001, frm, rd, rs1, ToRegister(2)); +} + +void AssemblerRISCVD::fcvt_lu_d(Register rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1100001, frm, rd, rs1, ToRegister(3)); +} + +void AssemblerRISCVD::fmv_x_d(Register rd, FPURegister rs1) { + GenInstrALUFP_rr(0b1110001, 0b000, rd, rs1, zero_reg); +} + +void AssemblerRISCVD::fcvt_d_l(FPURegister rd, Register rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1101001, frm, rd, rs1, ToRegister(2)); +} + +void AssemblerRISCVD::fcvt_d_lu(FPURegister rd, Register rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1101001, frm, rd, rs1, ToRegister(3)); +} + +void AssemblerRISCVD::fmv_d_x(FPURegister rd, Register rs1) { + GenInstrALUFP_rr(0b1111001, 0b000, rd, rs1, zero_reg); +} +#endif + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/extension-riscv-d.h b/js/src/jit/riscv64/extension/extension-riscv-d.h new file mode 100644 index 0000000000..8497c0ca63 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-d.h @@ -0,0 +1,68 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_extension_Extension_riscv_d_h_ +#define jit_riscv64_extension_Extension_riscv_d_h_ +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/Register-riscv64.h" +namespace js { +namespace jit { +class AssemblerRISCVD : public AssemblerRiscvBase { + // RV32D Standard Extension + public: + void fld(FPURegister rd, Register rs1, int16_t imm12); + void fsd(FPURegister source, Register base, int16_t imm12); + void fmadd_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm = RNE); + void fmsub_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm = RNE); + void fnmsub_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm = RNE); + void fnmadd_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm = RNE); + void fadd_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm = RNE); + void fsub_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm = RNE); + void fmul_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm = RNE); + void fdiv_d(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm = RNE); + void fsqrt_d(FPURegister rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fsgnj_d(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fsgnjn_d(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fsgnjx_d(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fmin_d(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fmax_d(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fcvt_s_d(FPURegister rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fcvt_d_s(FPURegister rd, FPURegister rs1, FPURoundingMode frm = RNE); + void feq_d(Register rd, FPURegister rs1, FPURegister rs2); + void flt_d(Register rd, FPURegister rs1, FPURegister rs2); + void fle_d(Register rd, FPURegister rs1, FPURegister rs2); + void fclass_d(Register rd, FPURegister rs1); + void fcvt_w_d(Register rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fcvt_wu_d(Register rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fcvt_d_w(FPURegister rd, Register rs1, FPURoundingMode frm = RNE); + void fcvt_d_wu(FPURegister rd, Register rs1, FPURoundingMode frm = RNE); + +#ifdef JS_CODEGEN_RISCV64 + // RV64D Standard Extension (in addition to RV32D) + void fcvt_l_d(Register rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fcvt_lu_d(Register rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fmv_x_d(Register rd, FPURegister rs1); + void fcvt_d_l(FPURegister rd, Register rs1, FPURoundingMode frm = RNE); + void fcvt_d_lu(FPURegister rd, Register rs1, FPURoundingMode frm = RNE); + void fmv_d_x(FPURegister rd, Register rs1); +#endif + + void fmv_d(FPURegister rd, FPURegister rs) { fsgnj_d(rd, rs, rs); } + void fabs_d(FPURegister rd, FPURegister rs) { fsgnjx_d(rd, rs, rs); } + void fneg_d(FPURegister rd, FPURegister rs) { fsgnjn_d(rd, rs, rs); } +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_extension_Extension_riscv_D_h_ diff --git a/js/src/jit/riscv64/extension/extension-riscv-f.cc b/js/src/jit/riscv64/extension/extension-riscv-f.cc new file mode 100644 index 0000000000..44e1fdc495 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-f.cc @@ -0,0 +1,158 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/extension/extension-riscv-f.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { + +// RV32F Standard Extension + +void AssemblerRISCVF::flw(FPURegister rd, Register rs1, int16_t imm12) { + GenInstrLoadFP_ri(0b010, rd, rs1, imm12); +} + +void AssemblerRISCVF::fsw(FPURegister source, Register base, int16_t imm12) { + GenInstrStoreFP_rri(0b010, base, source, imm12); +} + +void AssemblerRISCVF::fmadd_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm) { + GenInstrR4(0b00, MADD, rd, rs1, rs2, rs3, frm); +} + +void AssemblerRISCVF::fmsub_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm) { + GenInstrR4(0b00, MSUB, rd, rs1, rs2, rs3, frm); +} + +void AssemblerRISCVF::fnmsub_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm) { + GenInstrR4(0b00, NMSUB, rd, rs1, rs2, rs3, frm); +} + +void AssemblerRISCVF::fnmadd_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm) { + GenInstrR4(0b00, NMADD, rd, rs1, rs2, rs3, frm); +} + +void AssemblerRISCVF::fadd_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0000000, frm, rd, rs1, rs2); +} + +void AssemblerRISCVF::fsub_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0000100, frm, rd, rs1, rs2); +} + +void AssemblerRISCVF::fmul_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0001000, frm, rd, rs1, rs2); +} + +void AssemblerRISCVF::fdiv_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0001100, frm, rd, rs1, rs2); +} + +void AssemblerRISCVF::fsqrt_s(FPURegister rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b0101100, frm, rd, rs1, zero_reg); +} + +void AssemblerRISCVF::fsgnj_s(FPURegister rd, FPURegister rs1, + FPURegister rs2) { + GenInstrALUFP_rr(0b0010000, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVF::fsgnjn_s(FPURegister rd, FPURegister rs1, + FPURegister rs2) { + GenInstrALUFP_rr(0b0010000, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVF::fsgnjx_s(FPURegister rd, FPURegister rs1, + FPURegister rs2) { + GenInstrALUFP_rr(0b0010000, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVF::fmin_s(FPURegister rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b0010100, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVF::fmax_s(FPURegister rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b0010100, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVF::fcvt_w_s(Register rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1100000, frm, rd, rs1, zero_reg); +} + +void AssemblerRISCVF::fcvt_wu_s(Register rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1100000, frm, rd, rs1, ToRegister(1)); +} + +void AssemblerRISCVF::fmv_x_w(Register rd, FPURegister rs1) { + GenInstrALUFP_rr(0b1110000, 0b000, rd, rs1, zero_reg); +} + +void AssemblerRISCVF::feq_s(Register rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b1010000, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVF::flt_s(Register rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b1010000, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVF::fle_s(Register rd, FPURegister rs1, FPURegister rs2) { + GenInstrALUFP_rr(0b1010000, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVF::fclass_s(Register rd, FPURegister rs1) { + GenInstrALUFP_rr(0b1110000, 0b001, rd, rs1, zero_reg); +} + +void AssemblerRISCVF::fcvt_s_w(FPURegister rd, Register rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1101000, frm, rd, rs1, zero_reg); +} + +void AssemblerRISCVF::fcvt_s_wu(FPURegister rd, Register rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1101000, frm, rd, rs1, ToRegister(1)); +} + +void AssemblerRISCVF::fmv_w_x(FPURegister rd, Register rs1) { + GenInstrALUFP_rr(0b1111000, 0b000, rd, rs1, zero_reg); +} + +#ifdef JS_CODEGEN_RISCV64 +// RV64F Standard Extension (in addition to RV32F) + +void AssemblerRISCVF::fcvt_l_s(Register rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1100000, frm, rd, rs1, ToRegister(2)); +} + +void AssemblerRISCVF::fcvt_lu_s(Register rd, FPURegister rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1100000, frm, rd, rs1, ToRegister(3)); +} + +void AssemblerRISCVF::fcvt_s_l(FPURegister rd, Register rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1101000, frm, rd, rs1, ToRegister(2)); +} + +void AssemblerRISCVF::fcvt_s_lu(FPURegister rd, Register rs1, + FPURoundingMode frm) { + GenInstrALUFP_rr(0b1101000, frm, rd, rs1, ToRegister(3)); +} +#endif + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/extension-riscv-f.h b/js/src/jit/riscv64/extension/extension-riscv-f.h new file mode 100644 index 0000000000..3ab46ffcf6 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-f.h @@ -0,0 +1,66 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#ifndef jit_riscv64_extension_Extension_riscv_f_h_ +#define jit_riscv64_extension_Extension_riscv_f_h_ +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/Register-riscv64.h" +namespace js { +namespace jit { +class AssemblerRISCVF : public AssemblerRiscvBase { + // RV32F Standard Extension + public: + void flw(FPURegister rd, Register rs1, int16_t imm12); + void fsw(FPURegister source, Register base, int16_t imm12); + void fmadd_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm = RNE); + void fmsub_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm = RNE); + void fnmsub_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm = RNE); + void fnmadd_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURegister rs3, FPURoundingMode frm = RNE); + void fadd_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm = RNE); + void fsub_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm = RNE); + void fmul_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm = RNE); + void fdiv_s(FPURegister rd, FPURegister rs1, FPURegister rs2, + FPURoundingMode frm = RNE); + void fsqrt_s(FPURegister rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fsgnj_s(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fsgnjn_s(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fsgnjx_s(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fmin_s(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fmax_s(FPURegister rd, FPURegister rs1, FPURegister rs2); + void fcvt_w_s(Register rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fcvt_wu_s(Register rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fmv_x_w(Register rd, FPURegister rs1); + void feq_s(Register rd, FPURegister rs1, FPURegister rs2); + void flt_s(Register rd, FPURegister rs1, FPURegister rs2); + void fle_s(Register rd, FPURegister rs1, FPURegister rs2); + void fclass_s(Register rd, FPURegister rs1); + void fcvt_s_w(FPURegister rd, Register rs1, FPURoundingMode frm = RNE); + void fcvt_s_wu(FPURegister rd, Register rs1, FPURoundingMode frm = RNE); + void fmv_w_x(FPURegister rd, Register rs1); + +#ifdef JS_CODEGEN_RISCV64 + // RV64F Standard Extension (in addition to RV32F) + void fcvt_l_s(Register rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fcvt_lu_s(Register rd, FPURegister rs1, FPURoundingMode frm = RNE); + void fcvt_s_l(FPURegister rd, Register rs1, FPURoundingMode frm = RNE); + void fcvt_s_lu(FPURegister rd, Register rs1, FPURoundingMode frm = RNE); +#endif + + void fmv_s(FPURegister rd, FPURegister rs) { fsgnj_s(rd, rs, rs); } + void fabs_s(FPURegister rd, FPURegister rs) { fsgnjx_s(rd, rs, rs); } + void fneg_s(FPURegister rd, FPURegister rs) { fsgnjn_s(rd, rs, rs); } +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_extension_Extension_riscv_F_h_ diff --git a/js/src/jit/riscv64/extension/extension-riscv-m.cc b/js/src/jit/riscv64/extension/extension-riscv-m.cc new file mode 100644 index 0000000000..b5fcd6c34c --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-m.cc @@ -0,0 +1,68 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/extension/extension-riscv-m.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { +// RV32M Standard Extension + +void AssemblerRISCVM::mul(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000001, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVM::mulh(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000001, 0b001, rd, rs1, rs2); +} + +void AssemblerRISCVM::mulhsu(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000001, 0b010, rd, rs1, rs2); +} + +void AssemblerRISCVM::mulhu(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000001, 0b011, rd, rs1, rs2); +} + +void AssemblerRISCVM::div(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000001, 0b100, rd, rs1, rs2); +} + +void AssemblerRISCVM::divu(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000001, 0b101, rd, rs1, rs2); +} + +void AssemblerRISCVM::rem(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000001, 0b110, rd, rs1, rs2); +} + +void AssemblerRISCVM::remu(Register rd, Register rs1, Register rs2) { + GenInstrALU_rr(0b0000001, 0b111, rd, rs1, rs2); +} + +#ifdef JS_CODEGEN_RISCV64 +// RV64M Standard Extension (in addition to RV32M) + +void AssemblerRISCVM::mulw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0000001, 0b000, rd, rs1, rs2); +} + +void AssemblerRISCVM::divw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0000001, 0b100, rd, rs1, rs2); +} + +void AssemblerRISCVM::divuw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0000001, 0b101, rd, rs1, rs2); +} + +void AssemblerRISCVM::remw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0000001, 0b110, rd, rs1, rs2); +} + +void AssemblerRISCVM::remuw(Register rd, Register rs1, Register rs2) { + GenInstrALUW_rr(0b0000001, 0b111, rd, rs1, rs2); +} +#endif +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/extension-riscv-m.h b/js/src/jit/riscv64/extension/extension-riscv-m.h new file mode 100644 index 0000000000..7c2c932516 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-m.h @@ -0,0 +1,37 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef jit_riscv64_extension_Extension_riscv_m_h_ +#define jit_riscv64_extension_Extension_riscv_m_h_ +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/Register-riscv64.h" +namespace js { +namespace jit { +class AssemblerRISCVM : public AssemblerRiscvBase { + // RV32M Standard Extension + public: + void mul(Register rd, Register rs1, Register rs2); + void mulh(Register rd, Register rs1, Register rs2); + void mulhsu(Register rd, Register rs1, Register rs2); + void mulhu(Register rd, Register rs1, Register rs2); + void div(Register rd, Register rs1, Register rs2); + void divu(Register rd, Register rs1, Register rs2); + void rem(Register rd, Register rs1, Register rs2); + void remu(Register rd, Register rs1, Register rs2); +#ifdef JS_CODEGEN_RISCV64 + // RV64M Standard Extension (in addition to RV32M) + void mulw(Register rd, Register rs1, Register rs2); + void divw(Register rd, Register rs1, Register rs2); + void divuw(Register rd, Register rs1, Register rs2); + void remw(Register rd, Register rs1, Register rs2); + void remuw(Register rd, Register rs1, Register rs2); +#endif +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_extension_Extension_riscv_M_h_ diff --git a/js/src/jit/riscv64/extension/extension-riscv-v.cc b/js/src/jit/riscv64/extension/extension-riscv-v.cc new file mode 100644 index 0000000000..c7241158e0 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-v.cc @@ -0,0 +1,891 @@ + +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#include "jit/riscv64/extension/extension-riscv-v.h" + +#ifdef CAN_USE_RVV +# include "src/codegen/assembler.h" +# include "jit/riscv64/constant/Constant-riscv64.h" +# include "jit/riscv64/extension/register-riscv.h" + +namespace js { +namespace jit { + +// RVV + +void AssemblerRISCVV::vredmaxu_vs(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask) { + GenInstrV(VREDMAXU_FUNCT6, OP_MVV, vd, vs1, vs2, mask); +} + +void AssemblerRISCVV::vredmax_vs(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask) { + GenInstrV(VREDMAX_FUNCT6, OP_MVV, vd, vs1, vs2, mask); +} + +void AssemblerRISCVV::vredmin_vs(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask) { + GenInstrV(VREDMIN_FUNCT6, OP_MVV, vd, vs1, vs2, mask); +} + +void AssemblerRISCVV::vredminu_vs(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask) { + GenInstrV(VREDMINU_FUNCT6, OP_MVV, vd, vs1, vs2, mask); +} + +void AssemblerRISCVV::vmv_vv(VRegister vd, VRegister vs1) { + GenInstrV(VMV_FUNCT6, OP_IVV, vd, vs1, v0, NoMask); +} + +void AssemblerRISCVV::vmv_vx(VRegister vd, Register rs1) { + GenInstrV(VMV_FUNCT6, OP_IVX, vd, rs1, v0, NoMask); +} + +void AssemblerRISCVV::vmv_vi(VRegister vd, uint8_t simm5) { + GenInstrV(VMV_FUNCT6, vd, simm5, v0, NoMask); +} + +void AssemblerRISCVV::vmv_xs(Register rd, VRegister vs2) { + GenInstrV(VWXUNARY0_FUNCT6, OP_MVV, rd, 0b00000, vs2, NoMask); +} + +void AssemblerRISCVV::vmv_sx(VRegister vd, Register rs1) { + GenInstrV(VRXUNARY0_FUNCT6, OP_MVX, vd, rs1, v0, NoMask); +} + +void AssemblerRISCVV::vmerge_vv(VRegister vd, VRegister vs1, VRegister vs2) { + GenInstrV(VMV_FUNCT6, OP_IVV, vd, vs1, vs2, Mask); +} + +void AssemblerRISCVV::vmerge_vx(VRegister vd, Register rs1, VRegister vs2) { + GenInstrV(VMV_FUNCT6, OP_IVX, vd, rs1, vs2, Mask); +} + +void AssemblerRISCVV::vmerge_vi(VRegister vd, uint8_t imm5, VRegister vs2) { + GenInstrV(VMV_FUNCT6, vd, imm5, vs2, Mask); +} + +void AssemblerRISCVV::vadc_vv(VRegister vd, VRegister vs1, VRegister vs2) { + GenInstrV(VADC_FUNCT6, OP_IVV, vd, vs1, vs2, Mask); +} + +void AssemblerRISCVV::vadc_vx(VRegister vd, Register rs1, VRegister vs2) { + GenInstrV(VADC_FUNCT6, OP_IVX, vd, rs1, vs2, Mask); +} + +void AssemblerRISCVV::vadc_vi(VRegister vd, uint8_t imm5, VRegister vs2) { + GenInstrV(VADC_FUNCT6, vd, imm5, vs2, Mask); +} + +void AssemblerRISCVV::vmadc_vv(VRegister vd, VRegister vs1, VRegister vs2) { + GenInstrV(VMADC_FUNCT6, OP_IVV, vd, vs1, vs2, Mask); +} + +void AssemblerRISCVV::vmadc_vx(VRegister vd, Register rs1, VRegister vs2) { + GenInstrV(VMADC_FUNCT6, OP_IVX, vd, rs1, vs2, Mask); +} + +void AssemblerRISCVV::vmadc_vi(VRegister vd, uint8_t imm5, VRegister vs2) { + GenInstrV(VMADC_FUNCT6, vd, imm5, vs2, Mask); +} + +void AssemblerRISCVV::vrgather_vv(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask) { + DCHECK_NE(vd, vs1); + DCHECK_NE(vd, vs2); + GenInstrV(VRGATHER_FUNCT6, OP_IVV, vd, vs1, vs2, mask); +} + +void AssemblerRISCVV::vrgather_vi(VRegister vd, VRegister vs2, int8_t imm5, + MaskType mask) { + DCHECK_NE(vd, vs2); + GenInstrV(VRGATHER_FUNCT6, vd, imm5, vs2, mask); +} + +void AssemblerRISCVV::vrgather_vx(VRegister vd, VRegister vs2, Register rs1, + MaskType mask) { + DCHECK_NE(vd, vs2); + GenInstrV(VRGATHER_FUNCT6, OP_IVX, vd, rs1, vs2, mask); +} + +void AssemblerRISCVV::vwaddu_wx(VRegister vd, VRegister vs2, Register rs1, + MaskType mask) { + GenInstrV(VWADDUW_FUNCT6, OP_MVX, vd, rs1, vs2, mask); +} + +void AssemblerRISCVV::vid_v(VRegister vd, MaskType mask) { + GenInstrV(VMUNARY0_FUNCT6, OP_MVV, vd, VID_V, v0, mask); +} + +# define DEFINE_OPIVV(name, funct6) \ + void AssemblerRISCVV::name##_vv(VRegister vd, VRegister vs2, \ + VRegister vs1, MaskType mask) { \ + GenInstrV(funct6, OP_IVV, vd, vs1, vs2, mask); \ + } + +# define DEFINE_OPFVV(name, funct6) \ + void AssemblerRISCVV::name##_vv(VRegister vd, VRegister vs2, \ + VRegister vs1, MaskType mask) { \ + GenInstrV(funct6, OP_FVV, vd, vs1, vs2, mask); \ + } + +# define DEFINE_OPFWV(name, funct6) \ + void AssemblerRISCVV::name##_wv(VRegister vd, VRegister vs2, \ + VRegister vs1, MaskType mask) { \ + GenInstrV(funct6, OP_FVV, vd, vs1, vs2, mask); \ + } + +# define DEFINE_OPFRED(name, funct6) \ + void AssemblerRISCVV::name##_vs(VRegister vd, VRegister vs2, \ + VRegister vs1, MaskType mask) { \ + GenInstrV(funct6, OP_FVV, vd, vs1, vs2, mask); \ + } + +# define DEFINE_OPIVX(name, funct6) \ + void AssemblerRISCVV::name##_vx(VRegister vd, VRegister vs2, Register rs1, \ + MaskType mask) { \ + GenInstrV(funct6, OP_IVX, vd, rs1, vs2, mask); \ + } + +# define DEFINE_OPIVI(name, funct6) \ + void AssemblerRISCVV::name##_vi(VRegister vd, VRegister vs2, int8_t imm5, \ + MaskType mask) { \ + GenInstrV(funct6, vd, imm5, vs2, mask); \ + } + +# define DEFINE_OPMVV(name, funct6) \ + void AssemblerRISCVV::name##_vv(VRegister vd, VRegister vs2, \ + VRegister vs1, MaskType mask) { \ + GenInstrV(funct6, OP_MVV, vd, vs1, vs2, mask); \ + } + +// void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, VRegister vd, Register +// rs1, +// VRegister vs2, MaskType mask = NoMask); +# define DEFINE_OPMVX(name, funct6) \ + void AssemblerRISCVV::name##_vx(VRegister vd, VRegister vs2, Register rs1, \ + MaskType mask) { \ + GenInstrV(funct6, OP_MVX, vd, rs1, vs2, mask); \ + } + +# define DEFINE_OPFVF(name, funct6) \ + void AssemblerRISCVV::name##_vf(VRegister vd, VRegister vs2, \ + FPURegister fs1, MaskType mask) { \ + GenInstrV(funct6, OP_FVF, vd, fs1, vs2, mask); \ + } + +# define DEFINE_OPFWF(name, funct6) \ + void AssemblerRISCVV::name##_wf(VRegister vd, VRegister vs2, \ + FPURegister fs1, MaskType mask) { \ + GenInstrV(funct6, OP_FVF, vd, fs1, vs2, mask); \ + } + +# define DEFINE_OPFVV_FMA(name, funct6) \ + void AssemblerRISCVV::name##_vv(VRegister vd, VRegister vs1, \ + VRegister vs2, MaskType mask) { \ + GenInstrV(funct6, OP_FVV, vd, vs1, vs2, mask); \ + } + +# define DEFINE_OPFVF_FMA(name, funct6) \ + void AssemblerRISCVV::name##_vf(VRegister vd, FPURegister fs1, \ + VRegister vs2, MaskType mask) { \ + GenInstrV(funct6, OP_FVF, vd, fs1, vs2, mask); \ + } + +// vector integer extension +# define DEFINE_OPMVV_VIE(name, vs1) \ + void AssemblerRISCVV::name(VRegister vd, VRegister vs2, MaskType mask) { \ + GenInstrV(VXUNARY0_FUNCT6, OP_MVV, vd, vs1, vs2, mask); \ + } + +void AssemblerRISCVV::vfmv_vf(VRegister vd, FPURegister fs1, MaskType mask) { + GenInstrV(VMV_FUNCT6, OP_FVF, vd, fs1, v0, mask); +} + +void AssemblerRISCVV::vfmv_fs(FPURegister fd, VRegister vs2) { + GenInstrV(VWFUNARY0_FUNCT6, OP_FVV, fd, v0, vs2, NoMask); +} + +void AssemblerRISCVV::vfmv_sf(VRegister vd, FPURegister fs) { + GenInstrV(VRFUNARY0_FUNCT6, OP_FVF, vd, fs, v0, NoMask); +} + +DEFINE_OPIVV(vadd, VADD_FUNCT6) +DEFINE_OPIVX(vadd, VADD_FUNCT6) +DEFINE_OPIVI(vadd, VADD_FUNCT6) +DEFINE_OPIVV(vsub, VSUB_FUNCT6) +DEFINE_OPIVX(vsub, VSUB_FUNCT6) +DEFINE_OPMVX(vdiv, VDIV_FUNCT6) +DEFINE_OPMVX(vdivu, VDIVU_FUNCT6) +DEFINE_OPMVX(vmul, VMUL_FUNCT6) +DEFINE_OPMVX(vmulhu, VMULHU_FUNCT6) +DEFINE_OPMVX(vmulhsu, VMULHSU_FUNCT6) +DEFINE_OPMVX(vmulh, VMULH_FUNCT6) +DEFINE_OPMVV(vdiv, VDIV_FUNCT6) +DEFINE_OPMVV(vdivu, VDIVU_FUNCT6) +DEFINE_OPMVV(vmul, VMUL_FUNCT6) +DEFINE_OPMVV(vmulhu, VMULHU_FUNCT6) +DEFINE_OPMVV(vmulhsu, VMULHSU_FUNCT6) +DEFINE_OPMVV(vwmul, VWMUL_FUNCT6) +DEFINE_OPMVV(vwmulu, VWMULU_FUNCT6) +DEFINE_OPMVV(vmulh, VMULH_FUNCT6) +DEFINE_OPMVV(vwadd, VWADD_FUNCT6) +DEFINE_OPMVV(vwaddu, VWADDU_FUNCT6) +DEFINE_OPMVV(vcompress, VCOMPRESS_FUNCT6) +DEFINE_OPIVX(vsadd, VSADD_FUNCT6) +DEFINE_OPIVV(vsadd, VSADD_FUNCT6) +DEFINE_OPIVI(vsadd, VSADD_FUNCT6) +DEFINE_OPIVX(vsaddu, VSADDU_FUNCT6) +DEFINE_OPIVV(vsaddu, VSADDU_FUNCT6) +DEFINE_OPIVI(vsaddu, VSADDU_FUNCT6) +DEFINE_OPIVX(vssub, VSSUB_FUNCT6) +DEFINE_OPIVV(vssub, VSSUB_FUNCT6) +DEFINE_OPIVX(vssubu, VSSUBU_FUNCT6) +DEFINE_OPIVV(vssubu, VSSUBU_FUNCT6) +DEFINE_OPIVX(vrsub, VRSUB_FUNCT6) +DEFINE_OPIVI(vrsub, VRSUB_FUNCT6) +DEFINE_OPIVV(vminu, VMINU_FUNCT6) +DEFINE_OPIVX(vminu, VMINU_FUNCT6) +DEFINE_OPIVV(vmin, VMIN_FUNCT6) +DEFINE_OPIVX(vmin, VMIN_FUNCT6) +DEFINE_OPIVV(vmaxu, VMAXU_FUNCT6) +DEFINE_OPIVX(vmaxu, VMAXU_FUNCT6) +DEFINE_OPIVV(vmax, VMAX_FUNCT6) +DEFINE_OPIVX(vmax, VMAX_FUNCT6) +DEFINE_OPIVV(vand, VAND_FUNCT6) +DEFINE_OPIVX(vand, VAND_FUNCT6) +DEFINE_OPIVI(vand, VAND_FUNCT6) +DEFINE_OPIVV(vor, VOR_FUNCT6) +DEFINE_OPIVX(vor, VOR_FUNCT6) +DEFINE_OPIVI(vor, VOR_FUNCT6) +DEFINE_OPIVV(vxor, VXOR_FUNCT6) +DEFINE_OPIVX(vxor, VXOR_FUNCT6) +DEFINE_OPIVI(vxor, VXOR_FUNCT6) + +DEFINE_OPIVX(vslidedown, VSLIDEDOWN_FUNCT6) +DEFINE_OPIVI(vslidedown, VSLIDEDOWN_FUNCT6) +DEFINE_OPIVX(vslideup, VSLIDEUP_FUNCT6) +DEFINE_OPIVI(vslideup, VSLIDEUP_FUNCT6) + +DEFINE_OPIVV(vmseq, VMSEQ_FUNCT6) +DEFINE_OPIVX(vmseq, VMSEQ_FUNCT6) +DEFINE_OPIVI(vmseq, VMSEQ_FUNCT6) + +DEFINE_OPIVV(vmsne, VMSNE_FUNCT6) +DEFINE_OPIVX(vmsne, VMSNE_FUNCT6) +DEFINE_OPIVI(vmsne, VMSNE_FUNCT6) + +DEFINE_OPIVV(vmsltu, VMSLTU_FUNCT6) +DEFINE_OPIVX(vmsltu, VMSLTU_FUNCT6) + +DEFINE_OPIVV(vmslt, VMSLT_FUNCT6) +DEFINE_OPIVX(vmslt, VMSLT_FUNCT6) + +DEFINE_OPIVV(vmsle, VMSLE_FUNCT6) +DEFINE_OPIVX(vmsle, VMSLE_FUNCT6) +DEFINE_OPIVI(vmsle, VMSLE_FUNCT6) + +DEFINE_OPIVV(vmsleu, VMSLEU_FUNCT6) +DEFINE_OPIVX(vmsleu, VMSLEU_FUNCT6) +DEFINE_OPIVI(vmsleu, VMSLEU_FUNCT6) + +DEFINE_OPIVI(vmsgt, VMSGT_FUNCT6) +DEFINE_OPIVX(vmsgt, VMSGT_FUNCT6) + +DEFINE_OPIVI(vmsgtu, VMSGTU_FUNCT6) +DEFINE_OPIVX(vmsgtu, VMSGTU_FUNCT6) + +DEFINE_OPIVV(vsrl, VSRL_FUNCT6) +DEFINE_OPIVX(vsrl, VSRL_FUNCT6) +DEFINE_OPIVI(vsrl, VSRL_FUNCT6) + +DEFINE_OPIVV(vsra, VSRA_FUNCT6) +DEFINE_OPIVX(vsra, VSRA_FUNCT6) +DEFINE_OPIVI(vsra, VSRA_FUNCT6) + +DEFINE_OPIVV(vsll, VSLL_FUNCT6) +DEFINE_OPIVX(vsll, VSLL_FUNCT6) +DEFINE_OPIVI(vsll, VSLL_FUNCT6) + +DEFINE_OPIVV(vsmul, VSMUL_FUNCT6) +DEFINE_OPIVX(vsmul, VSMUL_FUNCT6) + +DEFINE_OPFVV(vfadd, VFADD_FUNCT6) +DEFINE_OPFVF(vfadd, VFADD_FUNCT6) +DEFINE_OPFVV(vfsub, VFSUB_FUNCT6) +DEFINE_OPFVF(vfsub, VFSUB_FUNCT6) +DEFINE_OPFVV(vfdiv, VFDIV_FUNCT6) +DEFINE_OPFVF(vfdiv, VFDIV_FUNCT6) +DEFINE_OPFVV(vfmul, VFMUL_FUNCT6) +DEFINE_OPFVF(vfmul, VFMUL_FUNCT6) +DEFINE_OPFVV(vmfeq, VMFEQ_FUNCT6) +DEFINE_OPFVV(vmfne, VMFNE_FUNCT6) +DEFINE_OPFVV(vmflt, VMFLT_FUNCT6) +DEFINE_OPFVV(vmfle, VMFLE_FUNCT6) +DEFINE_OPFVV(vfmax, VFMAX_FUNCT6) +DEFINE_OPFVV(vfmin, VFMIN_FUNCT6) + +// Vector Widening Floating-Point Add/Subtract Instructions +DEFINE_OPFVV(vfwadd, VFWADD_FUNCT6) +DEFINE_OPFVF(vfwadd, VFWADD_FUNCT6) +DEFINE_OPFVV(vfwsub, VFWSUB_FUNCT6) +DEFINE_OPFVF(vfwsub, VFWSUB_FUNCT6) +DEFINE_OPFWV(vfwadd, VFWADD_W_FUNCT6) +DEFINE_OPFWF(vfwadd, VFWADD_W_FUNCT6) +DEFINE_OPFWV(vfwsub, VFWSUB_W_FUNCT6) +DEFINE_OPFWF(vfwsub, VFWSUB_W_FUNCT6) + +// Vector Widening Floating-Point Reduction Instructions +DEFINE_OPFVV(vfwredusum, VFWREDUSUM_FUNCT6) +DEFINE_OPFVV(vfwredosum, VFWREDOSUM_FUNCT6) + +// Vector Widening Floating-Point Multiply +DEFINE_OPFVV(vfwmul, VFWMUL_FUNCT6) +DEFINE_OPFVF(vfwmul, VFWMUL_FUNCT6) + +DEFINE_OPFRED(vfredmax, VFREDMAX_FUNCT6) + +DEFINE_OPFVV(vfsngj, VFSGNJ_FUNCT6) +DEFINE_OPFVF(vfsngj, VFSGNJ_FUNCT6) +DEFINE_OPFVV(vfsngjn, VFSGNJN_FUNCT6) +DEFINE_OPFVF(vfsngjn, VFSGNJN_FUNCT6) +DEFINE_OPFVV(vfsngjx, VFSGNJX_FUNCT6) +DEFINE_OPFVF(vfsngjx, VFSGNJX_FUNCT6) + +// Vector Single-Width Floating-Point Fused Multiply-Add Instructions +DEFINE_OPFVV_FMA(vfmadd, VFMADD_FUNCT6) +DEFINE_OPFVF_FMA(vfmadd, VFMADD_FUNCT6) +DEFINE_OPFVV_FMA(vfmsub, VFMSUB_FUNCT6) +DEFINE_OPFVF_FMA(vfmsub, VFMSUB_FUNCT6) +DEFINE_OPFVV_FMA(vfmacc, VFMACC_FUNCT6) +DEFINE_OPFVF_FMA(vfmacc, VFMACC_FUNCT6) +DEFINE_OPFVV_FMA(vfmsac, VFMSAC_FUNCT6) +DEFINE_OPFVF_FMA(vfmsac, VFMSAC_FUNCT6) +DEFINE_OPFVV_FMA(vfnmadd, VFNMADD_FUNCT6) +DEFINE_OPFVF_FMA(vfnmadd, VFNMADD_FUNCT6) +DEFINE_OPFVV_FMA(vfnmsub, VFNMSUB_FUNCT6) +DEFINE_OPFVF_FMA(vfnmsub, VFNMSUB_FUNCT6) +DEFINE_OPFVV_FMA(vfnmacc, VFNMACC_FUNCT6) +DEFINE_OPFVF_FMA(vfnmacc, VFNMACC_FUNCT6) +DEFINE_OPFVV_FMA(vfnmsac, VFNMSAC_FUNCT6) +DEFINE_OPFVF_FMA(vfnmsac, VFNMSAC_FUNCT6) + +// Vector Widening Floating-Point Fused Multiply-Add Instructions +DEFINE_OPFVV_FMA(vfwmacc, VFWMACC_FUNCT6) +DEFINE_OPFVF_FMA(vfwmacc, VFWMACC_FUNCT6) +DEFINE_OPFVV_FMA(vfwnmacc, VFWNMACC_FUNCT6) +DEFINE_OPFVF_FMA(vfwnmacc, VFWNMACC_FUNCT6) +DEFINE_OPFVV_FMA(vfwmsac, VFWMSAC_FUNCT6) +DEFINE_OPFVF_FMA(vfwmsac, VFWMSAC_FUNCT6) +DEFINE_OPFVV_FMA(vfwnmsac, VFWNMSAC_FUNCT6) +DEFINE_OPFVF_FMA(vfwnmsac, VFWNMSAC_FUNCT6) + +// Vector Narrowing Fixed-Point Clip Instructions +DEFINE_OPIVV(vnclip, VNCLIP_FUNCT6) +DEFINE_OPIVX(vnclip, VNCLIP_FUNCT6) +DEFINE_OPIVI(vnclip, VNCLIP_FUNCT6) +DEFINE_OPIVV(vnclipu, VNCLIPU_FUNCT6) +DEFINE_OPIVX(vnclipu, VNCLIPU_FUNCT6) +DEFINE_OPIVI(vnclipu, VNCLIPU_FUNCT6) + +// Vector Integer Extension +DEFINE_OPMVV_VIE(vzext_vf8, 0b00010) +DEFINE_OPMVV_VIE(vsext_vf8, 0b00011) +DEFINE_OPMVV_VIE(vzext_vf4, 0b00100) +DEFINE_OPMVV_VIE(vsext_vf4, 0b00101) +DEFINE_OPMVV_VIE(vzext_vf2, 0b00110) +DEFINE_OPMVV_VIE(vsext_vf2, 0b00111) + +# undef DEFINE_OPIVI +# undef DEFINE_OPIVV +# undef DEFINE_OPIVX +# undef DEFINE_OPFVV +# undef DEFINE_OPFWV +# undef DEFINE_OPFVF +# undef DEFINE_OPFWF +# undef DEFINE_OPFVV_FMA +# undef DEFINE_OPFVF_FMA +# undef DEFINE_OPMVV_VIE + +void AssemblerRISCVV::vsetvli(Register rd, Register rs1, VSew vsew, Vlmul vlmul, + TailAgnosticType tail, MaskAgnosticType mask) { + int32_t zimm = GenZimm(vsew, vlmul, tail, mask); + Instr instr = OP_V | ((rd.code() & 0x1F) << kRvvRdShift) | (0x7 << 12) | + ((rs1.code() & 0x1F) << kRvvRs1Shift) | + (((uint32_t)zimm << kRvvZimmShift) & kRvvZimmMask) | 0x0 << 31; + emit(instr); +} + +void AssemblerRISCVV::vsetivli(Register rd, uint8_t uimm, VSew vsew, + Vlmul vlmul, TailAgnosticType tail, + MaskAgnosticType mask) { + MOZ_ASSERT(is_uint5(uimm)); + int32_t zimm = GenZimm(vsew, vlmul, tail, mask) & 0x3FF; + Instr instr = OP_V | ((rd.code() & 0x1F) << kRvvRdShift) | (0x7 << 12) | + ((uimm & 0x1F) << kRvvUimmShift) | + (((uint32_t)zimm << kRvvZimmShift) & kRvvZimmMask) | 0x3 << 30; + emit(instr); +} + +void AssemblerRISCVV::vsetvl(Register rd, Register rs1, Register rs2) { + Instr instr = OP_V | ((rd.code() & 0x1F) << kRvvRdShift) | (0x7 << 12) | + ((rs1.code() & 0x1F) << kRvvRs1Shift) | + ((rs2.code() & 0x1F) << kRvvRs2Shift) | 0x40 << 25; + emit(instr); +} + +uint8_t vsew_switch(VSew vsew) { + uint8_t width; + switch (vsew) { + case E8: + width = 0b000; + break; + case E16: + width = 0b101; + break; + case E32: + width = 0b110; + break; + default: + width = 0b111; + break; + } + return width; +} + +// OPIVV OPFVV OPMVV +void AssemblerRISCVV::GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, + VRegister vd, VRegister vs1, VRegister vs2, + MaskType mask) { + MOZ_ASSERT(opcode == OP_MVV || opcode == OP_FVV || opcode == OP_IVV); + Instr instr = (funct6 << kRvvFunct6Shift) | opcode | (mask << kRvvVmShift) | + ((vd.code() & 0x1F) << kRvvVdShift) | + ((vs1.code() & 0x1F) << kRvvVs1Shift) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} + +void AssemblerRISCVV::GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, + VRegister vd, int8_t vs1, VRegister vs2, + MaskType mask) { + MOZ_ASSERT(opcode == OP_MVV || opcode == OP_FVV || opcode == OP_IVV); + Instr instr = (funct6 << kRvvFunct6Shift) | opcode | (mask << kRvvVmShift) | + ((vd.code() & 0x1F) << kRvvVdShift) | + ((vs1 & 0x1F) << kRvvVs1Shift) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} +// OPMVV OPFVV +void AssemblerRISCVV::GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, + Register rd, VRegister vs1, VRegister vs2, + MaskType mask) { + MOZ_ASSERT(opcode == OP_MVV || opcode == OP_FVV); + Instr instr = (funct6 << kRvvFunct6Shift) | opcode | (mask << kRvvVmShift) | + ((rd.code() & 0x1F) << kRvvVdShift) | + ((vs1.code() & 0x1F) << kRvvVs1Shift) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} + +// OPFVV +void AssemblerRISCVV::GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, + FPURegister fd, VRegister vs1, VRegister vs2, + MaskType mask) { + MOZ_ASSERT(opcode == OP_FVV); + Instr instr = (funct6 << kRvvFunct6Shift) | opcode | (mask << kRvvVmShift) | + ((fd.code() & 0x1F) << kRvvVdShift) | + ((vs1.code() & 0x1F) << kRvvVs1Shift) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} + +// OPIVX OPMVX +void AssemblerRISCVV::GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, + VRegister vd, Register rs1, VRegister vs2, + MaskType mask) { + MOZ_ASSERT(opcode == OP_IVX || opcode == OP_MVX); + Instr instr = (funct6 << kRvvFunct6Shift) | opcode | (mask << kRvvVmShift) | + ((vd.code() & 0x1F) << kRvvVdShift) | + ((rs1.code() & 0x1F) << kRvvRs1Shift) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} + +// OPFVF +void AssemblerRISCVV::GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, + VRegister vd, FPURegister fs1, VRegister vs2, + MaskType mask) { + MOZ_ASSERT(opcode == OP_FVF); + Instr instr = (funct6 << kRvvFunct6Shift) | opcode | (mask << kRvvVmShift) | + ((vd.code() & 0x1F) << kRvvVdShift) | + ((fs1.code() & 0x1F) << kRvvRs1Shift) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} + +// OPMVX +void AssemblerRISCVV::GenInstrV(uint8_t funct6, Register rd, Register rs1, + VRegister vs2, MaskType mask) { + Instr instr = (funct6 << kRvvFunct6Shift) | OP_MVX | (mask << kRvvVmShift) | + ((rd.code() & 0x1F) << kRvvVdShift) | + ((rs1.code() & 0x1F) << kRvvRs1Shift) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} +// OPIVI +void AssemblerRISCVV::GenInstrV(uint8_t funct6, VRegister vd, int8_t imm5, + VRegister vs2, MaskType mask) { + MOZ_ASSERT(is_uint5(imm5) || is_int5(imm5)); + Instr instr = (funct6 << kRvvFunct6Shift) | OP_IVI | (mask << kRvvVmShift) | + ((vd.code() & 0x1F) << kRvvVdShift) | + (((uint32_t)imm5 << kRvvImm5Shift) & kRvvImm5Mask) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} + +// VL VS +void AssemblerRISCVV::GenInstrV(BaseOpcode opcode, uint8_t width, VRegister vd, + Register rs1, uint8_t umop, MaskType mask, + uint8_t IsMop, bool IsMew, uint8_t Nf) { + MOZ_ASSERT(opcode == LOAD_FP || opcode == STORE_FP); + Instr instr = opcode | ((vd.code() << kRvvVdShift) & kRvvVdMask) | + ((width << kRvvWidthShift) & kRvvWidthMask) | + ((rs1.code() << kRvvRs1Shift) & kRvvRs1Mask) | + ((umop << kRvvRs2Shift) & kRvvRs2Mask) | + ((mask << kRvvVmShift) & kRvvVmMask) | + ((IsMop << kRvvMopShift) & kRvvMopMask) | + ((IsMew << kRvvMewShift) & kRvvMewMask) | + ((Nf << kRvvNfShift) & kRvvNfMask); + emit(instr); +} +void AssemblerRISCVV::GenInstrV(BaseOpcode opcode, uint8_t width, VRegister vd, + Register rs1, Register rs2, MaskType mask, + uint8_t IsMop, bool IsMew, uint8_t Nf) { + MOZ_ASSERT(opcode == LOAD_FP || opcode == STORE_FP); + Instr instr = opcode | ((vd.code() << kRvvVdShift) & kRvvVdMask) | + ((width << kRvvWidthShift) & kRvvWidthMask) | + ((rs1.code() << kRvvRs1Shift) & kRvvRs1Mask) | + ((rs2.code() << kRvvRs2Shift) & kRvvRs2Mask) | + ((mask << kRvvVmShift) & kRvvVmMask) | + ((IsMop << kRvvMopShift) & kRvvMopMask) | + ((IsMew << kRvvMewShift) & kRvvMewMask) | + ((Nf << kRvvNfShift) & kRvvNfMask); + emit(instr); +} +// VL VS AMO +void AssemblerRISCVV::GenInstrV(BaseOpcode opcode, uint8_t width, VRegister vd, + Register rs1, VRegister vs2, MaskType mask, + uint8_t IsMop, bool IsMew, uint8_t Nf) { + MOZ_ASSERT(opcode == LOAD_FP || opcode == STORE_FP || opcode == AMO); + Instr instr = opcode | ((vd.code() << kRvvVdShift) & kRvvVdMask) | + ((width << kRvvWidthShift) & kRvvWidthMask) | + ((rs1.code() << kRvvRs1Shift) & kRvvRs1Mask) | + ((vs2.code() << kRvvRs2Shift) & kRvvRs2Mask) | + ((mask << kRvvVmShift) & kRvvVmMask) | + ((IsMop << kRvvMopShift) & kRvvMopMask) | + ((IsMew << kRvvMewShift) & kRvvMewMask) | + ((Nf << kRvvNfShift) & kRvvNfMask); + emit(instr); +} +// vmv_xs vcpop_m vfirst_m +void AssemblerRISCVV::GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, + Register rd, uint8_t vs1, VRegister vs2, + MaskType mask) { + MOZ_ASSERT(opcode == OP_MVV); + Instr instr = (funct6 << kRvvFunct6Shift) | opcode | (mask << kRvvVmShift) | + ((rd.code() & 0x1F) << kRvvVdShift) | + ((vs1 & 0x1F) << kRvvVs1Shift) | + ((vs2.code() & 0x1F) << kRvvVs2Shift); + emit(instr); +} + +void AssemblerRISCVV::vl(VRegister vd, Register rs1, uint8_t lumop, VSew vsew, + MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, lumop, mask, 0b00, 0, 0b000); +} +void AssemblerRISCVV::vls(VRegister vd, Register rs1, Register rs2, VSew vsew, + MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b000); +} +void AssemblerRISCVV::vlx(VRegister vd, Register rs1, VRegister vs2, VSew vsew, + MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, vs2, mask, 0b11, 0, 0); +} + +void AssemblerRISCVV::vs(VRegister vd, Register rs1, uint8_t sumop, VSew vsew, + MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, sumop, mask, 0b00, 0, 0b000); +} +void AssemblerRISCVV::vss(VRegister vs3, Register rs1, Register rs2, VSew vsew, + MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vs3, rs1, rs2, mask, 0b10, 0, 0b000); +} + +void AssemblerRISCVV::vsx(VRegister vd, Register rs1, VRegister vs2, VSew vsew, + MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, vs2, mask, 0b11, 0, 0b000); +} +void AssemblerRISCVV::vsu(VRegister vd, Register rs1, VRegister vs2, VSew vsew, + MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, vs2, mask, 0b01, 0, 0b000); +} + +void AssemblerRISCVV::vlseg2(VRegister vd, Register rs1, uint8_t lumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, lumop, mask, 0b00, 0, 0b001); +} + +void AssemblerRISCVV::vlseg3(VRegister vd, Register rs1, uint8_t lumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, lumop, mask, 0b00, 0, 0b010); +} + +void AssemblerRISCVV::vlseg4(VRegister vd, Register rs1, uint8_t lumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, lumop, mask, 0b00, 0, 0b011); +} + +void AssemblerRISCVV::vlseg5(VRegister vd, Register rs1, uint8_t lumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, lumop, mask, 0b00, 0, 0b100); +} + +void AssemblerRISCVV::vlseg6(VRegister vd, Register rs1, uint8_t lumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, lumop, mask, 0b00, 0, 0b101); +} + +void AssemblerRISCVV::vlseg7(VRegister vd, Register rs1, uint8_t lumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, lumop, mask, 0b00, 0, 0b110); +} + +void AssemblerRISCVV::vlseg8(VRegister vd, Register rs1, uint8_t lumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, lumop, mask, 0b00, 0, 0b111); +} +void AssemblerRISCVV::vsseg2(VRegister vd, Register rs1, uint8_t sumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, sumop, mask, 0b00, 0, 0b001); +} +void AssemblerRISCVV::vsseg3(VRegister vd, Register rs1, uint8_t sumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, sumop, mask, 0b00, 0, 0b010); +} +void AssemblerRISCVV::vsseg4(VRegister vd, Register rs1, uint8_t sumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, sumop, mask, 0b00, 0, 0b011); +} +void AssemblerRISCVV::vsseg5(VRegister vd, Register rs1, uint8_t sumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, sumop, mask, 0b00, 0, 0b100); +} +void AssemblerRISCVV::vsseg6(VRegister vd, Register rs1, uint8_t sumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, sumop, mask, 0b00, 0, 0b101); +} +void AssemblerRISCVV::vsseg7(VRegister vd, Register rs1, uint8_t sumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, sumop, mask, 0b00, 0, 0b110); +} +void AssemblerRISCVV::vsseg8(VRegister vd, Register rs1, uint8_t sumop, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, sumop, mask, 0b00, 0, 0b111); +} + +void AssemblerRISCVV::vlsseg2(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b001); +} +void AssemblerRISCVV::vlsseg3(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b010); +} +void AssemblerRISCVV::vlsseg4(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b011); +} +void AssemblerRISCVV::vlsseg5(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b100); +} +void AssemblerRISCVV::vlsseg6(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b101); +} +void AssemblerRISCVV::vlsseg7(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b110); +} +void AssemblerRISCVV::vlsseg8(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b111); +} +void AssemblerRISCVV::vssseg2(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b001); +} +void AssemblerRISCVV::vssseg3(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b010); +} +void AssemblerRISCVV::vssseg4(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b011); +} +void AssemblerRISCVV::vssseg5(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b100); +} +void AssemblerRISCVV::vssseg6(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b101); +} +void AssemblerRISCVV::vssseg7(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b110); +} +void AssemblerRISCVV::vssseg8(VRegister vd, Register rs1, Register rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b10, 0, 0b111); +} + +void AssemblerRISCVV::vlxseg2(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b001); +} +void AssemblerRISCVV::vlxseg3(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b010); +} +void AssemblerRISCVV::vlxseg4(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b011); +} +void AssemblerRISCVV::vlxseg5(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b100); +} +void AssemblerRISCVV::vlxseg6(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b101); +} +void AssemblerRISCVV::vlxseg7(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b110); +} +void AssemblerRISCVV::vlxseg8(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(LOAD_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b111); +} +void AssemblerRISCVV::vsxseg2(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b001); +} +void AssemblerRISCVV::vsxseg3(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b010); +} +void AssemblerRISCVV::vsxseg4(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b011); +} +void AssemblerRISCVV::vsxseg5(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b100); +} +void AssemblerRISCVV::vsxseg6(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b101); +} +void AssemblerRISCVV::vsxseg7(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b110); +} +void AssemblerRISCVV::vsxseg8(VRegister vd, Register rs1, VRegister rs2, + VSew vsew, MaskType mask) { + uint8_t width = vsew_switch(vsew); + GenInstrV(STORE_FP, width, vd, rs1, rs2, mask, 0b11, 0, 0b111); +} + +void AssemblerRISCVV::vfirst_m(Register rd, VRegister vs2, MaskType mask) { + GenInstrV(VWXUNARY0_FUNCT6, OP_MVV, rd, 0b10001, vs2, mask); +} + +void AssemblerRISCVV::vcpop_m(Register rd, VRegister vs2, MaskType mask) { + GenInstrV(VWXUNARY0_FUNCT6, OP_MVV, rd, 0b10000, vs2, mask); +} + +LoadStoreLaneParams::LoadStoreLaneParams(MachineRepresentation rep, + uint8_t laneidx) { + switch (rep) { + case MachineRepresentation::kWord8: + *this = LoadStoreLaneParams(laneidx, 8, kRvvVLEN / 16); + break; + case MachineRepresentation::kWord16: + *this = LoadStoreLaneParams(laneidx, 16, kRvvVLEN / 8); + break; + case MachineRepresentation::kWord32: + *this = LoadStoreLaneParams(laneidx, 32, kRvvVLEN / 4); + break; + case MachineRepresentation::kWord64: + *this = LoadStoreLaneParams(laneidx, 64, kRvvVLEN / 2); + break; + default: + UNREACHABLE(); + } +} + +} // namespace jit +} // namespace js +#endif diff --git a/js/src/jit/riscv64/extension/extension-riscv-v.h b/js/src/jit/riscv64/extension/extension-riscv-v.h new file mode 100644 index 0000000000..8f04f24c56 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-v.h @@ -0,0 +1,484 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef jit_riscv64_extension_Extension_riscv_v_h_ +#define jit_riscv64_extension_Extension_riscv_v_h_ +#ifdef CAN_USE_RVV +# include "jit/riscv64/Architecture-riscv64.h" +# include "jit/riscv64/constant/Constant-riscv64.h" +# include "jit/riscv64/extension/base-assembler-riscv.h" + +namespace js { +namespace jit { + +class AssemblerRISCVV : public AssemblerRiscvBase { + public: + // RVV + static int32_t GenZimm(VSew vsew, Vlmul vlmul, TailAgnosticType tail = tu, + MaskAgnosticType mask = mu) { + return (mask << 7) | (tail << 6) | ((vsew & 0x7) << 3) | (vlmul & 0x7); + } + + void vl(VRegister vd, Register rs1, uint8_t lumop, VSew vsew, + MaskType mask = NoMask); + void vls(VRegister vd, Register rs1, Register rs2, VSew vsew, + MaskType mask = NoMask); + void vlx(VRegister vd, Register rs1, VRegister vs3, VSew vsew, + MaskType mask = NoMask); + + void vs(VRegister vd, Register rs1, uint8_t sumop, VSew vsew, + MaskType mask = NoMask); + void vss(VRegister vd, Register rs1, Register rs2, VSew vsew, + MaskType mask = NoMask); + void vsx(VRegister vd, Register rs1, VRegister vs3, VSew vsew, + MaskType mask = NoMask); + + void vsu(VRegister vd, Register rs1, VRegister vs3, VSew vsew, + MaskType mask = NoMask); + +# define SegInstr(OP) \ + void OP##seg2(ARG); \ + void OP##seg3(ARG); \ + void OP##seg4(ARG); \ + void OP##seg5(ARG); \ + void OP##seg6(ARG); \ + void OP##seg7(ARG); \ + void OP##seg8(ARG); + +# define ARG \ + VRegister vd, Register rs1, uint8_t lumop, VSew vsew, MaskType mask = NoMask + + SegInstr(vl) SegInstr(vs) +# undef ARG + +# define ARG \ + VRegister vd, Register rs1, Register rs2, VSew vsew, MaskType mask = NoMask + + SegInstr(vls) SegInstr(vss) +# undef ARG + +# define ARG \ + VRegister vd, Register rs1, VRegister rs2, VSew vsew, MaskType mask = NoMask + + SegInstr(vsx) SegInstr(vlx) +# undef ARG +# undef SegInstr + + // RVV Vector Arithmetic Instruction + + void vmv_vv(VRegister vd, VRegister vs1); + void vmv_vx(VRegister vd, Register rs1); + void vmv_vi(VRegister vd, uint8_t simm5); + void vmv_xs(Register rd, VRegister vs2); + void vmv_sx(VRegister vd, Register rs1); + void vmerge_vv(VRegister vd, VRegister vs1, VRegister vs2); + void vmerge_vx(VRegister vd, Register rs1, VRegister vs2); + void vmerge_vi(VRegister vd, uint8_t imm5, VRegister vs2); + + void vredmaxu_vs(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask = NoMask); + void vredmax_vs(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask = NoMask); + void vredmin_vs(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask = NoMask); + void vredminu_vs(VRegister vd, VRegister vs2, VRegister vs1, + MaskType mask = NoMask); + + void vadc_vv(VRegister vd, VRegister vs1, VRegister vs2); + void vadc_vx(VRegister vd, Register rs1, VRegister vs2); + void vadc_vi(VRegister vd, uint8_t imm5, VRegister vs2); + + void vmadc_vv(VRegister vd, VRegister vs1, VRegister vs2); + void vmadc_vx(VRegister vd, Register rs1, VRegister vs2); + void vmadc_vi(VRegister vd, uint8_t imm5, VRegister vs2); + + void vfmv_vf(VRegister vd, FPURegister fs1, MaskType mask = NoMask); + void vfmv_fs(FPURegister fd, VRegister vs2); + void vfmv_sf(VRegister vd, FPURegister fs); + + void vwaddu_wx(VRegister vd, VRegister vs2, Register rs1, + MaskType mask = NoMask); + void vid_v(VRegister vd, MaskType mask = Mask); + +# define DEFINE_OPIVV(name, funct6) \ + void name##_vv(VRegister vd, VRegister vs2, VRegister vs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPIVX(name, funct6) \ + void name##_vx(VRegister vd, VRegister vs2, Register rs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPIVI(name, funct6) \ + void name##_vi(VRegister vd, VRegister vs2, int8_t imm5, \ + MaskType mask = NoMask); + +# define DEFINE_OPMVV(name, funct6) \ + void name##_vv(VRegister vd, VRegister vs2, VRegister vs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPMVX(name, funct6) \ + void name##_vx(VRegister vd, VRegister vs2, Register rs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPFVV(name, funct6) \ + void name##_vv(VRegister vd, VRegister vs2, VRegister vs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPFWV(name, funct6) \ + void name##_wv(VRegister vd, VRegister vs2, VRegister vs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPFRED(name, funct6) \ + void name##_vs(VRegister vd, VRegister vs2, VRegister vs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPFVF(name, funct6) \ + void name##_vf(VRegister vd, VRegister vs2, FPURegister fs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPFWF(name, funct6) \ + void name##_wf(VRegister vd, VRegister vs2, FPURegister fs1, \ + MaskType mask = NoMask); + +# define DEFINE_OPFVV_FMA(name, funct6) \ + void name##_vv(VRegister vd, VRegister vs1, VRegister vs2, \ + MaskType mask = NoMask); + +# define DEFINE_OPFVF_FMA(name, funct6) \ + void name##_vf(VRegister vd, FPURegister fs1, VRegister vs2, \ + MaskType mask = NoMask); + +# define DEFINE_OPMVV_VIE(name) \ + void name(VRegister vd, VRegister vs2, MaskType mask = NoMask); + + DEFINE_OPIVV(vadd, VADD_FUNCT6) + DEFINE_OPIVX(vadd, VADD_FUNCT6) + DEFINE_OPIVI(vadd, VADD_FUNCT6) + DEFINE_OPIVV(vsub, VSUB_FUNCT6) + DEFINE_OPIVX(vsub, VSUB_FUNCT6) + DEFINE_OPMVX(vdiv, VDIV_FUNCT6) + DEFINE_OPMVX(vdivu, VDIVU_FUNCT6) + DEFINE_OPMVX(vmul, VMUL_FUNCT6) + DEFINE_OPMVX(vmulhu, VMULHU_FUNCT6) + DEFINE_OPMVX(vmulhsu, VMULHSU_FUNCT6) + DEFINE_OPMVX(vmulh, VMULH_FUNCT6) + DEFINE_OPMVV(vdiv, VDIV_FUNCT6) + DEFINE_OPMVV(vdivu, VDIVU_FUNCT6) + DEFINE_OPMVV(vmul, VMUL_FUNCT6) + DEFINE_OPMVV(vmulhu, VMULHU_FUNCT6) + DEFINE_OPMVV(vmulhsu, VMULHSU_FUNCT6) + DEFINE_OPMVV(vmulh, VMULH_FUNCT6) + DEFINE_OPMVV(vwmul, VWMUL_FUNCT6) + DEFINE_OPMVV(vwmulu, VWMULU_FUNCT6) + DEFINE_OPMVV(vwaddu, VWADDU_FUNCT6) + DEFINE_OPMVV(vwadd, VWADD_FUNCT6) + DEFINE_OPMVV(vcompress, VCOMPRESS_FUNCT6) + DEFINE_OPIVX(vsadd, VSADD_FUNCT6) + DEFINE_OPIVV(vsadd, VSADD_FUNCT6) + DEFINE_OPIVI(vsadd, VSADD_FUNCT6) + DEFINE_OPIVX(vsaddu, VSADD_FUNCT6) + DEFINE_OPIVV(vsaddu, VSADDU_FUNCT6) + DEFINE_OPIVI(vsaddu, VSADDU_FUNCT6) + DEFINE_OPIVX(vssub, VSSUB_FUNCT6) + DEFINE_OPIVV(vssub, VSSUB_FUNCT6) + DEFINE_OPIVX(vssubu, VSSUBU_FUNCT6) + DEFINE_OPIVV(vssubu, VSSUBU_FUNCT6) + DEFINE_OPIVX(vrsub, VRSUB_FUNCT6) + DEFINE_OPIVI(vrsub, VRSUB_FUNCT6) + DEFINE_OPIVV(vminu, VMINU_FUNCT6) + DEFINE_OPIVX(vminu, VMINU_FUNCT6) + DEFINE_OPIVV(vmin, VMIN_FUNCT6) + DEFINE_OPIVX(vmin, VMIN_FUNCT6) + DEFINE_OPIVV(vmaxu, VMAXU_FUNCT6) + DEFINE_OPIVX(vmaxu, VMAXU_FUNCT6) + DEFINE_OPIVV(vmax, VMAX_FUNCT6) + DEFINE_OPIVX(vmax, VMAX_FUNCT6) + DEFINE_OPIVV(vand, VAND_FUNCT6) + DEFINE_OPIVX(vand, VAND_FUNCT6) + DEFINE_OPIVI(vand, VAND_FUNCT6) + DEFINE_OPIVV(vor, VOR_FUNCT6) + DEFINE_OPIVX(vor, VOR_FUNCT6) + DEFINE_OPIVI(vor, VOR_FUNCT6) + DEFINE_OPIVV(vxor, VXOR_FUNCT6) + DEFINE_OPIVX(vxor, VXOR_FUNCT6) + DEFINE_OPIVI(vxor, VXOR_FUNCT6) + DEFINE_OPIVV(vrgather, VRGATHER_FUNCT6) + DEFINE_OPIVX(vrgather, VRGATHER_FUNCT6) + DEFINE_OPIVI(vrgather, VRGATHER_FUNCT6) + + DEFINE_OPIVX(vslidedown, VSLIDEDOWN_FUNCT6) + DEFINE_OPIVI(vslidedown, VSLIDEDOWN_FUNCT6) + DEFINE_OPIVX(vslideup, VSLIDEUP_FUNCT6) + DEFINE_OPIVI(vslideup, VSLIDEUP_FUNCT6) + + DEFINE_OPIVV(vmseq, VMSEQ_FUNCT6) + DEFINE_OPIVX(vmseq, VMSEQ_FUNCT6) + DEFINE_OPIVI(vmseq, VMSEQ_FUNCT6) + + DEFINE_OPIVV(vmsne, VMSNE_FUNCT6) + DEFINE_OPIVX(vmsne, VMSNE_FUNCT6) + DEFINE_OPIVI(vmsne, VMSNE_FUNCT6) + + DEFINE_OPIVV(vmsltu, VMSLTU_FUNCT6) + DEFINE_OPIVX(vmsltu, VMSLTU_FUNCT6) + + DEFINE_OPIVV(vmslt, VMSLT_FUNCT6) + DEFINE_OPIVX(vmslt, VMSLT_FUNCT6) + + DEFINE_OPIVV(vmsle, VMSLE_FUNCT6) + DEFINE_OPIVX(vmsle, VMSLE_FUNCT6) + DEFINE_OPIVI(vmsle, VMSLE_FUNCT6) + + DEFINE_OPIVV(vmsleu, VMSLEU_FUNCT6) + DEFINE_OPIVX(vmsleu, VMSLEU_FUNCT6) + DEFINE_OPIVI(vmsleu, VMSLEU_FUNCT6) + + DEFINE_OPIVI(vmsgt, VMSGT_FUNCT6) + DEFINE_OPIVX(vmsgt, VMSGT_FUNCT6) + + DEFINE_OPIVI(vmsgtu, VMSGTU_FUNCT6) + DEFINE_OPIVX(vmsgtu, VMSGTU_FUNCT6) + + DEFINE_OPIVV(vsrl, VSRL_FUNCT6) + DEFINE_OPIVX(vsrl, VSRL_FUNCT6) + DEFINE_OPIVI(vsrl, VSRL_FUNCT6) + + DEFINE_OPIVV(vsra, VSRA_FUNCT6) + DEFINE_OPIVX(vsra, VSRA_FUNCT6) + DEFINE_OPIVI(vsra, VSRA_FUNCT6) + + DEFINE_OPIVV(vsll, VSLL_FUNCT6) + DEFINE_OPIVX(vsll, VSLL_FUNCT6) + DEFINE_OPIVI(vsll, VSLL_FUNCT6) + + DEFINE_OPIVV(vsmul, VSMUL_FUNCT6) + DEFINE_OPIVX(vsmul, VSMUL_FUNCT6) + + DEFINE_OPFVV(vfadd, VFADD_FUNCT6) + DEFINE_OPFVF(vfadd, VFADD_FUNCT6) + DEFINE_OPFVV(vfsub, VFSUB_FUNCT6) + DEFINE_OPFVF(vfsub, VFSUB_FUNCT6) + DEFINE_OPFVV(vfdiv, VFDIV_FUNCT6) + DEFINE_OPFVF(vfdiv, VFDIV_FUNCT6) + DEFINE_OPFVV(vfmul, VFMUL_FUNCT6) + DEFINE_OPFVF(vfmul, VFMUL_FUNCT6) + + // Vector Widening Floating-Point Add/Subtract Instructions + DEFINE_OPFVV(vfwadd, VFWADD_FUNCT6) + DEFINE_OPFVF(vfwadd, VFWADD_FUNCT6) + DEFINE_OPFVV(vfwsub, VFWSUB_FUNCT6) + DEFINE_OPFVF(vfwsub, VFWSUB_FUNCT6) + DEFINE_OPFWV(vfwadd, VFWADD_W_FUNCT6) + DEFINE_OPFWF(vfwadd, VFWADD_W_FUNCT6) + DEFINE_OPFWV(vfwsub, VFWSUB_W_FUNCT6) + DEFINE_OPFWF(vfwsub, VFWSUB_W_FUNCT6) + + // Vector Widening Floating-Point Reduction Instructions + DEFINE_OPFVV(vfwredusum, VFWREDUSUM_FUNCT6) + DEFINE_OPFVV(vfwredosum, VFWREDOSUM_FUNCT6) + + // Vector Widening Floating-Point Multiply + DEFINE_OPFVV(vfwmul, VFWMUL_FUNCT6) + DEFINE_OPFVF(vfwmul, VFWMUL_FUNCT6) + + DEFINE_OPFVV(vmfeq, VMFEQ_FUNCT6) + DEFINE_OPFVV(vmfne, VMFNE_FUNCT6) + DEFINE_OPFVV(vmflt, VMFLT_FUNCT6) + DEFINE_OPFVV(vmfle, VMFLE_FUNCT6) + DEFINE_OPFVV(vfmax, VMFMAX_FUNCT6) + DEFINE_OPFVV(vfmin, VMFMIN_FUNCT6) + DEFINE_OPFRED(vfredmax, VFREDMAX_FUNCT6) + + DEFINE_OPFVV(vfsngj, VFSGNJ_FUNCT6) + DEFINE_OPFVF(vfsngj, VFSGNJ_FUNCT6) + DEFINE_OPFVV(vfsngjn, VFSGNJN_FUNCT6) + DEFINE_OPFVF(vfsngjn, VFSGNJN_FUNCT6) + DEFINE_OPFVV(vfsngjx, VFSGNJX_FUNCT6) + DEFINE_OPFVF(vfsngjx, VFSGNJX_FUNCT6) + + // Vector Single-Width Floating-Point Fused Multiply-Add Instructions + DEFINE_OPFVV_FMA(vfmadd, VFMADD_FUNCT6) + DEFINE_OPFVF_FMA(vfmadd, VFMADD_FUNCT6) + DEFINE_OPFVV_FMA(vfmsub, VFMSUB_FUNCT6) + DEFINE_OPFVF_FMA(vfmsub, VFMSUB_FUNCT6) + DEFINE_OPFVV_FMA(vfmacc, VFMACC_FUNCT6) + DEFINE_OPFVF_FMA(vfmacc, VFMACC_FUNCT6) + DEFINE_OPFVV_FMA(vfmsac, VFMSAC_FUNCT6) + DEFINE_OPFVF_FMA(vfmsac, VFMSAC_FUNCT6) + DEFINE_OPFVV_FMA(vfnmadd, VFNMADD_FUNCT6) + DEFINE_OPFVF_FMA(vfnmadd, VFNMADD_FUNCT6) + DEFINE_OPFVV_FMA(vfnmsub, VFNMSUB_FUNCT6) + DEFINE_OPFVF_FMA(vfnmsub, VFNMSUB_FUNCT6) + DEFINE_OPFVV_FMA(vfnmacc, VFNMACC_FUNCT6) + DEFINE_OPFVF_FMA(vfnmacc, VFNMACC_FUNCT6) + DEFINE_OPFVV_FMA(vfnmsac, VFNMSAC_FUNCT6) + DEFINE_OPFVF_FMA(vfnmsac, VFNMSAC_FUNCT6) + + // Vector Widening Floating-Point Fused Multiply-Add Instructions + DEFINE_OPFVV_FMA(vfwmacc, VFWMACC_FUNCT6) + DEFINE_OPFVF_FMA(vfwmacc, VFWMACC_FUNCT6) + DEFINE_OPFVV_FMA(vfwnmacc, VFWNMACC_FUNCT6) + DEFINE_OPFVF_FMA(vfwnmacc, VFWNMACC_FUNCT6) + DEFINE_OPFVV_FMA(vfwmsac, VFWMSAC_FUNCT6) + DEFINE_OPFVF_FMA(vfwmsac, VFWMSAC_FUNCT6) + DEFINE_OPFVV_FMA(vfwnmsac, VFWNMSAC_FUNCT6) + DEFINE_OPFVF_FMA(vfwnmsac, VFWNMSAC_FUNCT6) + + // Vector Narrowing Fixed-Point Clip Instructions + DEFINE_OPIVV(vnclip, VNCLIP_FUNCT6) + DEFINE_OPIVX(vnclip, VNCLIP_FUNCT6) + DEFINE_OPIVI(vnclip, VNCLIP_FUNCT6) + DEFINE_OPIVV(vnclipu, VNCLIPU_FUNCT6) + DEFINE_OPIVX(vnclipu, VNCLIPU_FUNCT6) + DEFINE_OPIVI(vnclipu, VNCLIPU_FUNCT6) + + // Vector Integer Extension + DEFINE_OPMVV_VIE(vzext_vf8) + DEFINE_OPMVV_VIE(vsext_vf8) + DEFINE_OPMVV_VIE(vzext_vf4) + DEFINE_OPMVV_VIE(vsext_vf4) + DEFINE_OPMVV_VIE(vzext_vf2) + DEFINE_OPMVV_VIE(vsext_vf2) + +# undef DEFINE_OPIVI +# undef DEFINE_OPIVV +# undef DEFINE_OPIVX +# undef DEFINE_OPMVV +# undef DEFINE_OPMVX +# undef DEFINE_OPFVV +# undef DEFINE_OPFWV +# undef DEFINE_OPFVF +# undef DEFINE_OPFWF +# undef DEFINE_OPFVV_FMA +# undef DEFINE_OPFVF_FMA +# undef DEFINE_OPMVV_VIE +# undef DEFINE_OPFRED + +# define DEFINE_VFUNARY(name, funct6, vs1) \ + void name(VRegister vd, VRegister vs2, MaskType mask = NoMask) { \ + GenInstrV(funct6, OP_FVV, vd, vs1, vs2, mask); \ + } + + DEFINE_VFUNARY(vfcvt_xu_f_v, VFUNARY0_FUNCT6, VFCVT_XU_F_V) + DEFINE_VFUNARY(vfcvt_x_f_v, VFUNARY0_FUNCT6, VFCVT_X_F_V) + DEFINE_VFUNARY(vfcvt_f_x_v, VFUNARY0_FUNCT6, VFCVT_F_X_V) + DEFINE_VFUNARY(vfcvt_f_xu_v, VFUNARY0_FUNCT6, VFCVT_F_XU_V) + DEFINE_VFUNARY(vfwcvt_xu_f_v, VFUNARY0_FUNCT6, VFWCVT_XU_F_V) + DEFINE_VFUNARY(vfwcvt_x_f_v, VFUNARY0_FUNCT6, VFWCVT_X_F_V) + DEFINE_VFUNARY(vfwcvt_f_x_v, VFUNARY0_FUNCT6, VFWCVT_F_X_V) + DEFINE_VFUNARY(vfwcvt_f_xu_v, VFUNARY0_FUNCT6, VFWCVT_F_XU_V) + DEFINE_VFUNARY(vfwcvt_f_f_v, VFUNARY0_FUNCT6, VFWCVT_F_F_V) + + DEFINE_VFUNARY(vfncvt_f_f_w, VFUNARY0_FUNCT6, VFNCVT_F_F_W) + DEFINE_VFUNARY(vfncvt_x_f_w, VFUNARY0_FUNCT6, VFNCVT_X_F_W) + DEFINE_VFUNARY(vfncvt_xu_f_w, VFUNARY0_FUNCT6, VFNCVT_XU_F_W) + + DEFINE_VFUNARY(vfclass_v, VFUNARY1_FUNCT6, VFCLASS_V) + DEFINE_VFUNARY(vfsqrt_v, VFUNARY1_FUNCT6, VFSQRT_V) + DEFINE_VFUNARY(vfrsqrt7_v, VFUNARY1_FUNCT6, VFRSQRT7_V) + DEFINE_VFUNARY(vfrec7_v, VFUNARY1_FUNCT6, VFREC7_V) +# undef DEFINE_VFUNARY + + void vnot_vv(VRegister dst, VRegister src, MaskType mask = NoMask) { + vxor_vi(dst, src, -1, mask); + } + + void vneg_vv(VRegister dst, VRegister src, MaskType mask = NoMask) { + vrsub_vx(dst, src, zero_reg, mask); + } + + void vfneg_vv(VRegister dst, VRegister src, MaskType mask = NoMask) { + vfsngjn_vv(dst, src, src, mask); + } + void vfabs_vv(VRegister dst, VRegister src, MaskType mask = NoMask) { + vfsngjx_vv(dst, src, src, mask); + } + void vfirst_m(Register rd, VRegister vs2, MaskType mask = NoMask); + + void vcpop_m(Register rd, VRegister vs2, MaskType mask = NoMask); + + protected: + void vsetvli(Register rd, Register rs1, VSew vsew, Vlmul vlmul, + TailAgnosticType tail = tu, MaskAgnosticType mask = mu); + + void vsetivli(Register rd, uint8_t uimm, VSew vsew, Vlmul vlmul, + TailAgnosticType tail = tu, MaskAgnosticType mask = mu); + + inline void vsetvlmax(Register rd, VSew vsew, Vlmul vlmul, + TailAgnosticType tail = tu, + MaskAgnosticType mask = mu) { + vsetvli(rd, zero_reg, vsew, vlmul, tu, mu); + } + + inline void vsetvl(VSew vsew, Vlmul vlmul, TailAgnosticType tail = tu, + MaskAgnosticType mask = mu) { + vsetvli(zero_reg, zero_reg, vsew, vlmul, tu, mu); + } + + void vsetvl(Register rd, Register rs1, Register rs2); + + // ----------------------------RVV------------------------------------------ + // vsetvl + void GenInstrV(Register rd, Register rs1, Register rs2); + // vsetvli + void GenInstrV(Register rd, Register rs1, uint32_t zimm); + // OPIVV OPFVV OPMVV + void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, VRegister vd, + VRegister vs1, VRegister vs2, MaskType mask = NoMask); + void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, VRegister vd, int8_t vs1, + VRegister vs2, MaskType mask = NoMask); + void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, VRegister vd, + VRegister vs2, MaskType mask = NoMask); + // OPMVV OPFVV + void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, Register rd, + VRegister vs1, VRegister vs2, MaskType mask = NoMask); + // OPFVV + void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, FPURegister fd, + VRegister vs1, VRegister vs2, MaskType mask = NoMask); + + // OPIVX OPMVX + void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, VRegister vd, + Register rs1, VRegister vs2, MaskType mask = NoMask); + // OPFVF + void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, VRegister vd, + FPURegister fs1, VRegister vs2, MaskType mask = NoMask); + // OPMVX + void GenInstrV(uint8_t funct6, Register rd, Register rs1, VRegister vs2, + MaskType mask = NoMask); + // OPIVI + void GenInstrV(uint8_t funct6, VRegister vd, int8_t simm5, VRegister vs2, + MaskType mask = NoMask); + + // VL VS + void GenInstrV(BaseOpcode opcode, uint8_t width, VRegister vd, Register rs1, + uint8_t umop, MaskType mask, uint8_t IsMop, bool IsMew, + uint8_t Nf); + + void GenInstrV(BaseOpcode opcode, uint8_t width, VRegister vd, Register rs1, + Register rs2, MaskType mask, uint8_t IsMop, bool IsMew, + uint8_t Nf); + // VL VS AMO + void GenInstrV(BaseOpcode opcode, uint8_t width, VRegister vd, Register rs1, + VRegister vs2, MaskType mask, uint8_t IsMop, bool IsMew, + uint8_t Nf); + // vmv_xs vcpop_m vfirst_m + void GenInstrV(uint8_t funct6, OpcodeRISCVV opcode, Register rd, uint8_t vs1, + VRegister vs2, MaskType mask); +}; + +class LoadStoreLaneParams { + public: + int sz; + uint8_t laneidx; + + LoadStoreLaneParams(MachineRepresentation rep, uint8_t laneidx); + + private: + LoadStoreLaneParams(uint8_t laneidx, int sz, int lanes) + : sz(sz), laneidx(laneidx % lanes) {} +}; +} // namespace jit +} // namespace js +#endif +#endif // jit_riscv64_extension_Extension_riscv_V_h_ diff --git a/js/src/jit/riscv64/extension/extension-riscv-zicsr.cc b/js/src/jit/riscv64/extension/extension-riscv-zicsr.cc new file mode 100644 index 0000000000..68b95ac0ea --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-zicsr.cc @@ -0,0 +1,48 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/extension/extension-riscv-zicsr.h" + +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Register-riscv64.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { + +void AssemblerRISCVZicsr::csrrw(Register rd, ControlStatusReg csr, + Register rs1) { + GenInstrCSR_ir(0b001, rd, csr, rs1); +} + +void AssemblerRISCVZicsr::csrrs(Register rd, ControlStatusReg csr, + Register rs1) { + GenInstrCSR_ir(0b010, rd, csr, rs1); +} + +void AssemblerRISCVZicsr::csrrc(Register rd, ControlStatusReg csr, + Register rs1) { + GenInstrCSR_ir(0b011, rd, csr, rs1); +} + +void AssemblerRISCVZicsr::csrrwi(Register rd, ControlStatusReg csr, + uint8_t imm5) { + GenInstrCSR_ii(0b101, rd, csr, imm5); +} + +void AssemblerRISCVZicsr::csrrsi(Register rd, ControlStatusReg csr, + uint8_t imm5) { + GenInstrCSR_ii(0b110, rd, csr, imm5); +} + +void AssemblerRISCVZicsr::csrrci(Register rd, ControlStatusReg csr, + uint8_t imm5) { + GenInstrCSR_ii(0b111, rd, csr, imm5); +} + +void AssemblerRISCVZicsr::illegal_trap(uint8_t code) { + csrrwi(zero, csr_cycle, code); +} + +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/extension-riscv-zicsr.h b/js/src/jit/riscv64/extension/extension-riscv-zicsr.h new file mode 100644 index 0000000000..0a571fd7e3 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-zicsr.h @@ -0,0 +1,60 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef jit_riscv64_extension_Extension_riscv_zicsr_h_ +#define jit_riscv64_extension_Extension_riscv_zicsr_h_ +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/Register-riscv64.h" +namespace js { +namespace jit { + +class AssemblerRISCVZicsr : public AssemblerRiscvBase { + public: + // CSR + void csrrw(Register rd, ControlStatusReg csr, Register rs1); + void csrrs(Register rd, ControlStatusReg csr, Register rs1); + void csrrc(Register rd, ControlStatusReg csr, Register rs1); + void csrrwi(Register rd, ControlStatusReg csr, uint8_t imm5); + void csrrsi(Register rd, ControlStatusReg csr, uint8_t imm5); + void csrrci(Register rd, ControlStatusReg csr, uint8_t imm5); + + // illegal_trap + void illegal_trap(uint8_t code); + + // Read instructions-retired counter + void rdinstret(Register rd) { csrrs(rd, csr_instret, zero_reg); } + void rdinstreth(Register rd) { csrrs(rd, csr_instreth, zero_reg); } + void rdcycle(Register rd) { csrrs(rd, csr_cycle, zero_reg); } + void rdcycleh(Register rd) { csrrs(rd, csr_cycleh, zero_reg); } + void rdtime(Register rd) { csrrs(rd, csr_time, zero_reg); } + void rdtimeh(Register rd) { csrrs(rd, csr_timeh, zero_reg); } + + void csrr(Register rd, ControlStatusReg csr) { csrrs(rd, csr, zero_reg); } + void csrw(ControlStatusReg csr, Register rs) { csrrw(zero_reg, csr, rs); } + void csrs(ControlStatusReg csr, Register rs) { csrrs(zero_reg, csr, rs); } + void csrc(ControlStatusReg csr, Register rs) { csrrc(zero_reg, csr, rs); } + + void csrwi(ControlStatusReg csr, uint8_t imm) { csrrwi(zero_reg, csr, imm); } + void csrsi(ControlStatusReg csr, uint8_t imm) { csrrsi(zero_reg, csr, imm); } + void csrci(ControlStatusReg csr, uint8_t imm) { csrrci(zero_reg, csr, imm); } + + void frcsr(Register rd) { csrrs(rd, csr_fcsr, zero_reg); } + void fscsr(Register rd, Register rs) { csrrw(rd, csr_fcsr, rs); } + void fscsr(Register rs) { csrrw(zero_reg, csr_fcsr, rs); } + + void frrm(Register rd) { csrrs(rd, csr_frm, zero_reg); } + void fsrm(Register rd, Register rs) { csrrw(rd, csr_frm, rs); } + void fsrm(Register rs) { csrrw(zero_reg, csr_frm, rs); } + + void frflags(Register rd) { csrrs(rd, csr_fflags, zero_reg); } + void fsflags(Register rd, Register rs) { csrrw(rd, csr_fflags, rs); } + void fsflags(Register rs) { csrrw(zero_reg, csr_fflags, rs); } +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_extension_Extension_riscv_zicsr_h_ diff --git a/js/src/jit/riscv64/extension/extension-riscv-zifencei.cc b/js/src/jit/riscv64/extension/extension-riscv-zifencei.cc new file mode 100644 index 0000000000..ec8080b0cb --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-zifencei.cc @@ -0,0 +1,17 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. +#include "jit/riscv64/extension/extension-riscv-zifencei.h" + +#include "jit/riscv64/extension/base-assembler-riscv.h" +#include "jit/riscv64/constant/Constant-riscv64.h" +#include "jit/riscv64/Assembler-riscv64.h" +#include "jit/riscv64/Architecture-riscv64.h" +namespace js { +namespace jit { + +void AssemblerRISCVZifencei::fence_i() { + GenInstrI(0b001, MISC_MEM, ToRegister(0), ToRegister(0), 0); +} +} // namespace jit +} // namespace js diff --git a/js/src/jit/riscv64/extension/extension-riscv-zifencei.h b/js/src/jit/riscv64/extension/extension-riscv-zifencei.h new file mode 100644 index 0000000000..a245320ec4 --- /dev/null +++ b/js/src/jit/riscv64/extension/extension-riscv-zifencei.h @@ -0,0 +1,20 @@ +// Copyright 2022 the V8 project authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. + +#ifndef jit_riscv64_extension_Extension_riscv_zifencei_h_ +#define jit_riscv64_extension_Extension_riscv_zifencei_h_ +#include "mozilla/Assertions.h" + +#include <stdint.h> + +#include "jit/riscv64/extension/base-assembler-riscv.h" +namespace js { +namespace jit { +class AssemblerRISCVZifencei : public AssemblerRiscvBase { + public: + void fence_i(); +}; +} // namespace jit +} // namespace js +#endif // jit_riscv64_extension_Extension_riscv_zifencei_h_ |