diff options
Diffstat (limited to 'js/src/jit/mips32')
20 files changed, 12496 insertions, 0 deletions
diff --git a/js/src/jit/mips32/Architecture-mips32.cpp b/js/src/jit/mips32/Architecture-mips32.cpp new file mode 100644 index 0000000000..598551eafe --- /dev/null +++ b/js/src/jit/mips32/Architecture-mips32.cpp @@ -0,0 +1,94 @@ +/* -*- 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/mips32/Architecture-mips32.h" + +#include "jit/RegisterSets.h" + +namespace js { +namespace jit { + +const char* const Registers::RegNames[] = { + "zero", "at", "v0", "v1", "a0", "a1", "a2", "a3", "t0", "t1", "t2", + "t3", "t4", "t5", "t6", "t7", "s0", "s1", "s2", "s3", "s4", "s5", + "s6", "s7", "t8", "t9", "k0", "k1", "gp", "sp", "fp", "ra"}; + +const uint32_t Allocatable = 14; + +const Registers::SetType Registers::ArgRegMask = Registers::SharedArgRegMask; + +const Registers::SetType Registers::JSCallMask = + (1 << Registers::a2) | (1 << Registers::a3); + +const Registers::SetType Registers::CallMask = + (1 << Registers::v0) | + (1 << Registers::v1); // used for double-size returns + +FloatRegisters::Encoding FloatRegisters::FromName(const char* name) { + for (size_t i = 0; i < RegisterIdLimit; i++) { + if (strcmp(GetName(i), name) == 0) { + return Encoding(i); + } + } + + return Invalid; +} + +FloatRegister FloatRegister::doubleOverlay() const { + MOZ_ASSERT(isNotOdd()); + if (isSingle()) { + return FloatRegister(code_, Double); + } + return *this; +} + +FloatRegister FloatRegister::singleOverlay() const { + MOZ_ASSERT(isNotOdd()); + if (isDouble()) { + return FloatRegister(code_, Single); + } + return *this; +} + +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) { + // Even for single size registers save complete double register. + mod.addUnchecked((*iter).doubleOverlay()); + } + return mod.set(); +} + +uint32_t FloatRegister::GetPushSizeInBytes(const FloatRegisterSet& s) { +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + FloatRegisterSet ss = s.reduceSetForPush(); + uint64_t bits = ss.bits(); + // We are only pushing double registers. + MOZ_ASSERT((bits & 0xFFFF) == 0); + uint32_t ret = mozilla::CountPopulation32(bits) * sizeof(double); + + // Additional space needed by MacroAssembler::PushRegsInMask to ensure + // correct alignment of double values. + if (ret) { + ret += sizeof(double); + } + + return ret; +} +uint32_t FloatRegister::getRegisterDumpOffsetInBytes() { + MOZ_ASSERT(isNotOdd()); + return id() * sizeof(float); +} + +} // namespace jit +} // namespace js diff --git a/js/src/jit/mips32/Architecture-mips32.h b/js/src/jit/mips32/Architecture-mips32.h new file mode 100644 index 0000000000..b8b3869adc --- /dev/null +++ b/js/src/jit/mips32/Architecture-mips32.h @@ -0,0 +1,286 @@ +/* -*- 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_mips32_Architecture_mips32_h +#define jit_mips32_Architecture_mips32_h + +#include "mozilla/EndianUtils.h" +#include "mozilla/MathAlgorithms.h" + +#include <limits.h> +#include <stdint.h> + +#include "jit/mips-shared/Architecture-mips-shared.h" + +#include "js/Utility.h" + +namespace js { +namespace jit { + +static const uint32_t ShadowStackSpace = 4 * sizeof(uintptr_t); + +// These offsets are specific to nunboxing, and capture offsets into the +// components of a js::Value. +// Size of MIPS32 general purpose registers is 32 bits. +#if MOZ_LITTLE_ENDIAN() +static const int32_t NUNBOX32_TYPE_OFFSET = 4; +static const int32_t NUNBOX32_PAYLOAD_OFFSET = 0; +#else +static const int32_t NUNBOX32_TYPE_OFFSET = 0; +static const int32_t NUNBOX32_PAYLOAD_OFFSET = 4; +#endif + +// Size of each bailout table entry. +// For MIPS this is 2 instructions relative call. +static const uint32_t BAILOUT_TABLE_ENTRY_SIZE = 2 * sizeof(void*); + +// MIPS32 can have two types of floating-point coprocessors modes: +// - FR=0 mode/ 32-bit FPRs - Historical default, there are 32 single +// precision registers and pairs of even and odd float registers are used as +// double precision registers. Example: f0 (double) is composed of +// f0 and f1 (single). Loongson3A FPU running in this mode doesn't allow +// use of odd registers for single precision arithmetic. +// - FR=1 mode/ 64-bit FPRs - In this case, there are 32 double precision +// register which can also be used as single precision registers. More info +// https://dmz-portal.imgtec.com/wiki/MIPS_O32_ABI_-_FR0_and_FR1_Interlinking + +// Currently we enable 16 even single precision registers which can be also can +// be used as double precision registers. It enables jit code to run even on +// Loongson3A. It does not support FR=1 mode because MacroAssembler threats odd +// single precision registers as high parts of even double precision registers. +#ifdef __mips_fpr +static_assert(__mips_fpr == 32, "MIPS32 jit only supports FR=0 fpu mode."); +#endif + +class FloatRegisters : public FloatRegistersMIPSShared { + public: + static const char* GetName(uint32_t i) { + MOZ_ASSERT(i < RegisterIdLimit); + return FloatRegistersMIPSShared::GetName(Encoding(i % 32)); + } + + static Encoding FromName(const char* name); + + static const uint32_t Total = 32; + static const uint32_t TotalDouble = 16; + static const uint32_t TotalSingle = 16; + + static const uint32_t Allocatable = 30; + static const SetType AllSingleMask = (1ULL << TotalSingle) - 1; + + static const SetType AllDoubleMask = ((1ULL << TotalDouble) - 1) + << TotalSingle; + static const SetType AllMask = AllDoubleMask | AllSingleMask; + + // When saving all registers we only need to do is save double registers. + static const uint32_t TotalPhys = 16; + static const uint32_t RegisterIdLimit = 32; + + static_assert(sizeof(SetType) * 8 >= Total, + "SetType should be large enough to enumerate all registers."); + + static const SetType NonVolatileMask = + ((SetType(1) << (FloatRegisters::f20 >> 1)) | + (SetType(1) << (FloatRegisters::f22 >> 1)) | + (SetType(1) << (FloatRegisters::f24 >> 1)) | + (SetType(1) << (FloatRegisters::f26 >> 1)) | + (SetType(1) << (FloatRegisters::f28 >> 1)) | + (SetType(1) << (FloatRegisters::f30 >> 1))) * + ((1 << TotalSingle) + 1); + + static const SetType VolatileMask = AllMask & ~NonVolatileMask; + + static const SetType WrapperMask = VolatileMask; + + static const SetType NonAllocatableMask = + (SetType(1) << (FloatRegisters::f18 >> 1)) * ((1 << TotalSingle) + 1); + + static const SetType AllocatableMask = AllMask & ~NonAllocatableMask; +}; + +class FloatRegister : public FloatRegisterMIPSShared { + public: + enum RegType { + Single = 0x0, + Double = 0x1, + }; + + typedef FloatRegisters Codes; + typedef Codes::Code Code; + typedef Codes::Encoding Encoding; + + Encoding code_ : 6; + + protected: + RegType kind_ : 1; + + public: + constexpr FloatRegister(uint32_t code, RegType kind = Double) + : code_(Encoding(code)), kind_(kind) {} + constexpr FloatRegister() + : code_(FloatRegisters::invalid_freg), kind_(Double) {} + + bool operator==(const FloatRegister& other) const { + MOZ_ASSERT(!isInvalid()); + MOZ_ASSERT(!other.isInvalid()); + return kind_ == other.kind_ && code_ == other.code_; + } + bool equiv(const FloatRegister& other) const { return other.kind_ == kind_; } + size_t size() const { return (kind_ == Double) ? 8 : 4; } + size_t pushSize() const { return size(); } + + bool isNotOdd() const { return !isInvalid() && ((code_ & 1) == 0); } + + bool isSingle() const { return kind_ == Single; } + bool isDouble() const { return kind_ == Double; } + bool isInvalid() const { return code_ == FloatRegisters::invalid_freg; } + bool isSimd128() const { return false; } + + FloatRegister doubleOverlay() const; + FloatRegister singleOverlay() const; + + FloatRegister asSingle() const { return singleOverlay(); } + FloatRegister asDouble() const { return doubleOverlay(); } + FloatRegister asSimd128() const { MOZ_CRASH("NYI"); } + + Code code() const { + MOZ_ASSERT(isNotOdd()); + return Code((code_ >> 1) | (kind_ << 4)); + } + Encoding encoding() const { + MOZ_ASSERT(!isInvalid()); + return code_; + } + uint32_t id() const { + MOZ_ASSERT(!isInvalid()); + return code_; + } + static FloatRegister FromCode(uint32_t i) { + uint32_t code = i & 15; + uint32_t kind = i >> 4; + return FloatRegister(Encoding(code << 1), RegType(kind)); + } + + static FloatRegister FromIndex(uint32_t index, RegType kind) { + MOZ_ASSERT(index < 16); + return FloatRegister(Encoding(index << 1), kind); + } + + bool volatile_() const { + return !!((SetType(1) << code()) & FloatRegisters::VolatileMask); + } + const char* name() const { return FloatRegisters::GetName(code_); } + bool operator!=(const FloatRegister& other) const { + return other.kind_ != kind_ || code_ != other.code_; + } + bool aliases(const FloatRegister& other) { + MOZ_ASSERT(isNotOdd()); + return code_ == other.code_; + } + uint32_t numAliased() const { + MOZ_ASSERT(isNotOdd()); + return 2; + } + FloatRegister aliased(uint32_t aliasIdx) { + MOZ_ASSERT(isNotOdd()); + + if (aliasIdx == 0) { + return *this; + } + MOZ_ASSERT(aliasIdx == 1); + if (isDouble()) { + return singleOverlay(); + } + return doubleOverlay(); + } + uint32_t numAlignedAliased() const { + MOZ_ASSERT(isNotOdd()); + return 2; + } + FloatRegister alignedAliased(uint32_t aliasIdx) { + MOZ_ASSERT(isNotOdd()); + + if (aliasIdx == 0) { + return *this; + } + MOZ_ASSERT(aliasIdx == 1); + if (isDouble()) { + return singleOverlay(); + } + return doubleOverlay(); + } + + SetType alignedOrDominatedAliasedSet() const { + MOZ_ASSERT(isNotOdd()); + return (SetType(1) << (code_ >> 1)) * + ((1 << FloatRegisters::TotalSingle) + 1); + } + + static constexpr RegTypeName DefaultType = RegTypeName::Float64; + + template <RegTypeName = 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); + } + + static Code FromName(const char* name) { + return FloatRegisters::FromName(name); + } + static TypedRegisterSet<FloatRegister> ReduceSetForPush( + const TypedRegisterSet<FloatRegister>& s); + static uint32_t GetPushSizeInBytes(const TypedRegisterSet<FloatRegister>& s); + uint32_t getRegisterDumpOffsetInBytes(); +}; + +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; +} + +template <> +inline FloatRegister::SetType +FloatRegister::AllocatableAsIndexableSet<RegTypeName::Float32>(SetType set) { + // Single registers are not dominating any smaller registers, thus masking + // is enough to convert an allocatable set into a set of register list all + // single register available. + return set & FloatRegisters::AllSingleMask; +} + +template <> +inline FloatRegister::SetType +FloatRegister::AllocatableAsIndexableSet<RegTypeName::Float64>(SetType set) { + return set & FloatRegisters::AllDoubleMask; +} + +// In order to handle functions such as int(*)(int, double) where the first +// argument is a general purpose register, and the second argument is a floating +// point register, we have to store the double content into 2 general purpose +// registers, namely a2 and a3. +#define JS_CODEGEN_REGISTER_PAIR 1 + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_Architecture_mips32_h */ diff --git a/js/src/jit/mips32/Assembler-mips32.cpp b/js/src/jit/mips32/Assembler-mips32.cpp new file mode 100644 index 0000000000..8073b8e4ec --- /dev/null +++ b/js/src/jit/mips32/Assembler-mips32.cpp @@ -0,0 +1,369 @@ +/* -*- 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/mips32/Assembler-mips32.h" + +#include "mozilla/DebugOnly.h" +#include "mozilla/Maybe.h" + +#include "jit/AutoWritableJitCode.h" + +using mozilla::DebugOnly; + +using namespace js; +using namespace js::jit; + +ABIArgGenerator::ABIArgGenerator() + : usedArgSlots_(0), + firstArgFloatSize_(0), + useGPRForFloats_(false), + current_() {} + +ABIArg ABIArgGenerator::next(MIRType type) { + Register destReg; + switch (type) { + case MIRType::Int32: + case MIRType::Pointer: + case MIRType::RefOrNull: + case MIRType::StackResults: + if (GetIntArgReg(usedArgSlots_, &destReg)) { + current_ = ABIArg(destReg); + } else { + current_ = ABIArg(usedArgSlots_ * sizeof(intptr_t)); + } + usedArgSlots_++; + break; + case MIRType::Int64: + if (!usedArgSlots_) { + current_ = ABIArg(a0, a1); + usedArgSlots_ = 2; + } else if (usedArgSlots_ <= 2) { + current_ = ABIArg(a2, a3); + usedArgSlots_ = 4; + } else { + if (usedArgSlots_ < NumIntArgRegs) { + usedArgSlots_ = NumIntArgRegs; + } + usedArgSlots_ += usedArgSlots_ % 2; + current_ = ABIArg(usedArgSlots_ * sizeof(intptr_t)); + usedArgSlots_ += 2; + } + break; + case MIRType::Float32: + if (!usedArgSlots_) { + current_ = ABIArg(f12.asSingle()); + firstArgFloatSize_ = 1; + } else if (usedArgSlots_ == firstArgFloatSize_) { + current_ = ABIArg(f14.asSingle()); + } else if (useGPRForFloats_ && GetIntArgReg(usedArgSlots_, &destReg)) { + current_ = ABIArg(destReg); + } else { + if (usedArgSlots_ < NumIntArgRegs) { + usedArgSlots_ = NumIntArgRegs; + } + current_ = ABIArg(usedArgSlots_ * sizeof(intptr_t)); + } + usedArgSlots_++; + break; + case MIRType::Double: + if (!usedArgSlots_) { + current_ = ABIArg(f12); + usedArgSlots_ = 2; + firstArgFloatSize_ = 2; + } else if (usedArgSlots_ == firstArgFloatSize_) { + current_ = ABIArg(f14); + usedArgSlots_ = 4; + } else if (useGPRForFloats_ && usedArgSlots_ <= 2) { + current_ = ABIArg(a2, a3); + usedArgSlots_ = 4; + } else { + if (usedArgSlots_ < NumIntArgRegs) { + usedArgSlots_ = NumIntArgRegs; + } + usedArgSlots_ += usedArgSlots_ % 2; + current_ = ABIArg(usedArgSlots_ * sizeof(intptr_t)); + usedArgSlots_ += 2; + } + break; + default: + MOZ_CRASH("Unexpected argument type"); + } + return current_; +} + +uint32_t js::jit::RT(FloatRegister r) { + MOZ_ASSERT(r.id() < FloatRegisters::RegisterIdLimit); + return r.id() << RTShift; +} + +uint32_t js::jit::RD(FloatRegister r) { + MOZ_ASSERT(r.id() < FloatRegisters::RegisterIdLimit); + return r.id() << RDShift; +} + +uint32_t js::jit::RZ(FloatRegister r) { + MOZ_ASSERT(r.id() < FloatRegisters::RegisterIdLimit); + return r.id() << RZShift; +} + +uint32_t js::jit::SA(FloatRegister r) { + MOZ_ASSERT(r.id() < FloatRegisters::RegisterIdLimit); + return r.id() << SAShift; +} + +void Assembler::executableCopy(uint8_t* buffer) { + MOZ_ASSERT(isFinished); + m_buffer.executableCopy(buffer); +} + +uintptr_t Assembler::GetPointer(uint8_t* instPtr) { + Instruction* inst = (Instruction*)instPtr; + return Assembler::ExtractLuiOriValue(inst, inst->next()); +} + +static JitCode* CodeFromJump(Instruction* jump) { + uint8_t* target = (uint8_t*)Assembler::ExtractLuiOriValue(jump, jump->next()); + 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::ExtractLuiOriValue(inst, inst->next()); + void* prior = ptr; + + // 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); + } + AssemblerMIPSShared::UpdateLuiOriValue(inst, inst->next(), uint32_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); + } +} + +Assembler::Condition Assembler::UnsignedCondition(Condition cond) { + switch (cond) { + case Zero: + case NonZero: + return cond; + case LessThan: + case Below: + return Below; + case LessThanOrEqual: + case BelowOrEqual: + return BelowOrEqual; + case GreaterThan: + case Above: + return Above; + case AboveOrEqual: + case GreaterThanOrEqual: + return AboveOrEqual; + default: + MOZ_CRASH("unexpected condition"); + } +} + +Assembler::Condition Assembler::ConditionWithoutEqual(Condition cond) { + switch (cond) { + case LessThan: + case LessThanOrEqual: + return LessThan; + case Below: + case BelowOrEqual: + return Below; + case GreaterThan: + case GreaterThanOrEqual: + return GreaterThan; + case Above: + case AboveOrEqual: + return Above; + default: + MOZ_CRASH("unexpected condition"); + } +} + +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); + AssemblerMIPSShared::UpdateLuiOriValue(inst, inst->next(), + (uint32_t)(rawCode + target)); + } + } +} + +void Assembler::bind(InstImm* inst, uintptr_t branch, uintptr_t target) { + int32_t offset = target - branch; + InstImm inst_bgezal = InstImm(op_regimm, zero, rt_bgezal, BOffImm16(0)); + InstImm inst_beq = InstImm(op_beq, zero, zero, BOffImm16(0)); + + // If encoded offset is 4, then the jump must be short + if (BOffImm16(inst[0]).decode() == 4) { + MOZ_ASSERT(BOffImm16::IsInRange(offset)); + inst[0].setBOffImm16(BOffImm16(offset)); + inst[1].makeNop(); + return; + } + + // Generate the long jump for calls because return address has to be the + // address after the reserved block. + if (inst[0].encode() == inst_bgezal.encode()) { + addLongJump(BufferOffset(branch), BufferOffset(target)); + Assembler::WriteLuiOriInstructions(inst, &inst[1], ScratchRegister, + LabelBase::INVALID_OFFSET); + inst[2] = InstReg(op_special, ScratchRegister, zero, ra, ff_jalr).encode(); + // There is 1 nop after this. + return; + } + + if (BOffImm16::IsInRange(offset)) { + bool conditional = (inst[0].encode() != inst_bgezal.encode() && + inst[0].encode() != inst_beq.encode()); + + inst[0].setBOffImm16(BOffImm16(offset)); + inst[1].makeNop(); + + // Skip the trailing nops in conditional branches. + if (conditional) { + inst[2] = InstImm(op_regimm, zero, rt_bgez, BOffImm16(3 * sizeof(void*))) + .encode(); + // There are 2 nops after this + } + return; + } + + if (inst[0].encode() == inst_beq.encode()) { + // Handle long unconditional jump. + addLongJump(BufferOffset(branch), BufferOffset(target)); + Assembler::WriteLuiOriInstructions(inst, &inst[1], ScratchRegister, + LabelBase::INVALID_OFFSET); + inst[2] = InstReg(op_special, ScratchRegister, zero, zero, ff_jr).encode(); + // There is 1 nop after this. + } else { + // Handle long conditional jump. + inst[0] = invertBranch(inst[0], BOffImm16(5 * sizeof(void*))); + // No need for a "nop" here because we can clobber scratch. + addLongJump(BufferOffset(branch + sizeof(void*)), BufferOffset(target)); + Assembler::WriteLuiOriInstructions(&inst[1], &inst[2], ScratchRegister, + LabelBase::INVALID_OFFSET); + inst[3] = InstReg(op_special, ScratchRegister, zero, zero, ff_jr).encode(); + // There is 1 nop after this. + } +} + +void Assembler::processCodeLabels(uint8_t* rawCode) { + for (const CodeLabel& label : codeLabels_) { + Bind(rawCode, label); + } +} + +uint32_t Assembler::PatchWrite_NearCallSize() { return 4 * sizeof(uint32_t); } + +void Assembler::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. + Assembler::WriteLuiOriInstructions(inst, &inst[1], ScratchRegister, + (uint32_t)dest); + inst[2] = InstReg(op_special, ScratchRegister, zero, ra, ff_jalr); + inst[3] = InstNOP(); +} + +uint32_t Assembler::ExtractLuiOriValue(Instruction* inst0, Instruction* inst1) { + InstImm* i0 = (InstImm*)inst0; + InstImm* i1 = (InstImm*)inst1; + MOZ_ASSERT(i0->extractOpcode() == ((uint32_t)op_lui >> OpcodeShift)); + MOZ_ASSERT(i1->extractOpcode() == ((uint32_t)op_ori >> OpcodeShift)); + + uint32_t value = i0->extractImm16Value() << 16; + value = value | i1->extractImm16Value(); + return value; +} + +void Assembler::WriteLuiOriInstructions(Instruction* inst0, Instruction* inst1, + Register reg, uint32_t value) { + *inst0 = InstImm(op_lui, zero, reg, Imm16::Upper(Imm32(value))); + *inst1 = InstImm(op_ori, reg, reg, Imm16::Lower(Imm32(value))); +} + +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<uint32_t> value = Assembler::ExtractLuiOriValue(&inst[0], &inst[1]); + MOZ_ASSERT(value == uint32_t(expectedValue.value)); + + // Replace with new value + AssemblerMIPSShared::UpdateLuiOriValue(inst, inst->next(), + uint32_t(newValue.value)); +} + +uint32_t Assembler::ExtractInstructionImmediate(uint8_t* code) { + InstImm* inst = (InstImm*)code; + return Assembler::ExtractLuiOriValue(inst, inst->next()); +} + +void Assembler::ToggleCall(CodeLocationLabel inst_, bool enabled) { + Instruction* inst = (Instruction*)inst_.raw(); + InstImm* i0 = (InstImm*)inst; + InstImm* i1 = (InstImm*)i0->next(); + Instruction* i2 = (Instruction*)i1->next(); + + MOZ_ASSERT(i0->extractOpcode() == ((uint32_t)op_lui >> OpcodeShift)); + MOZ_ASSERT(i1->extractOpcode() == ((uint32_t)op_ori >> OpcodeShift)); + + if (enabled) { + InstReg jalr = InstReg(op_special, ScratchRegister, zero, ra, ff_jalr); + *i2 = jalr; + } else { + InstNOP nop; + *i2 = nop; + } +} diff --git a/js/src/jit/mips32/Assembler-mips32.h b/js/src/jit/mips32/Assembler-mips32.h new file mode 100644 index 0000000000..e8523547e2 --- /dev/null +++ b/js/src/jit/mips32/Assembler-mips32.h @@ -0,0 +1,261 @@ +/* -*- 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_mips32_Assembler_mips32_h +#define jit_mips32_Assembler_mips32_h + +#include <iterator> + +#include "jit/mips-shared/Assembler-mips-shared.h" + +#include "jit/mips32/Architecture-mips32.h" + +namespace js { +namespace jit { + +static constexpr Register CallTempReg4 = t4; +static constexpr Register CallTempReg5 = t5; + +static constexpr Register CallTempNonArgRegs[] = {t0, t1, t2, t3, t4}; +static const uint32_t NumCallTempNonArgRegs = std::size(CallTempNonArgRegs); + +class ABIArgGenerator { + unsigned usedArgSlots_; + unsigned firstArgFloatSize_; + // Note: This is not compliant with the system ABI. The Lowering phase + // expects to lower an MWasmParameter to only one register. + bool useGPRForFloats_; + ABIArg current_; + + public: + ABIArgGenerator(); + ABIArg next(MIRType argType); + ABIArg& current() { return current_; } + + void enforceO32ABI() { useGPRForFloats_ = true; } + + uint32_t stackBytesConsumedSoFar() const { + if (usedArgSlots_ <= 4) { + return ShadowStackSpace; + } + + return usedArgSlots_ * sizeof(intptr_t); + } + + void increaseStackOffset(uint32_t bytes) { MOZ_CRASH("NYI"); } +}; + +// These registers may be volatile or nonvolatile. +static constexpr Register ABINonArgReg0 = t0; +static constexpr Register ABINonArgReg1 = t1; +static constexpr Register ABINonArgReg2 = t2; +static constexpr Register ABINonArgReg3 = t3; + +// This register may be volatile or nonvolatile. Avoid f18 which is the +// ScratchDoubleReg. +static constexpr FloatRegister ABINonArgDoubleReg{FloatRegisters::f16, + FloatRegister::Double}; + +// These registers may be volatile or nonvolatile. +// Note: these three registers are all guaranteed to be different +static constexpr Register ABINonArgReturnReg0 = t0; +static constexpr Register ABINonArgReturnReg1 = t1; +static constexpr Register ABINonVolatileReg = s0; + +// 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 = t0; + +// TLS 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. +static constexpr Register WasmTlsReg = s5; + +// Registers used for asm.js/wasm table calls. These registers must be disjoint +// from the ABI argument registers, WasmTlsReg and each other. +static constexpr Register WasmTableCallScratchReg0 = ABINonArgReg0; +static constexpr Register WasmTableCallScratchReg1 = ABINonArgReg1; +static constexpr Register WasmTableCallSigReg = ABINonArgReg2; +static constexpr Register WasmTableCallIndexReg = ABINonArgReg3; + +// Register used as a scratch along the return path in the fast js -> wasm stub +// code. This must not overlap ReturnReg, JSReturnOperand, or WasmTlsReg. It +// must be a volatile register. +static constexpr Register WasmJitEntryReturnScratch = t1; + +static constexpr Register InterpreterPCReg = t5; + +static constexpr Register JSReturnReg_Type = a3; +static constexpr Register JSReturnReg_Data = a2; +static constexpr Register64 ReturnReg64(v1, v0); +static constexpr FloatRegister ReturnFloat32Reg = {FloatRegisters::f0, + FloatRegister::Single}; +static constexpr FloatRegister ReturnDoubleReg = {FloatRegisters::f0, + FloatRegister::Double}; +static constexpr FloatRegister ScratchFloat32Reg = {FloatRegisters::f18, + FloatRegister::Single}; +static constexpr FloatRegister ScratchDoubleReg = {FloatRegisters::f18, + FloatRegister::Double}; + +struct ScratchFloat32Scope : public AutoFloatRegisterScope { + explicit ScratchFloat32Scope(MacroAssembler& masm) + : AutoFloatRegisterScope(masm, ScratchFloat32Reg) {} +}; + +struct ScratchDoubleScope : public AutoFloatRegisterScope { + explicit ScratchDoubleScope(MacroAssembler& masm) + : AutoFloatRegisterScope(masm, ScratchDoubleReg) {} +}; + +static constexpr FloatRegister f0 = {FloatRegisters::f0, FloatRegister::Double}; +static constexpr FloatRegister f2 = {FloatRegisters::f2, FloatRegister::Double}; +static constexpr FloatRegister f4 = {FloatRegisters::f4, FloatRegister::Double}; +static constexpr FloatRegister f6 = {FloatRegisters::f6, FloatRegister::Double}; +static constexpr FloatRegister f8 = {FloatRegisters::f8, FloatRegister::Double}; +static constexpr FloatRegister f10 = {FloatRegisters::f10, + FloatRegister::Double}; +static constexpr FloatRegister f12 = {FloatRegisters::f12, + FloatRegister::Double}; +static constexpr FloatRegister f14 = {FloatRegisters::f14, + FloatRegister::Double}; +static constexpr FloatRegister f16 = {FloatRegisters::f16, + FloatRegister::Double}; +static constexpr FloatRegister f18 = {FloatRegisters::f18, + FloatRegister::Double}; +static constexpr FloatRegister f20 = {FloatRegisters::f20, + FloatRegister::Double}; +static constexpr FloatRegister f22 = {FloatRegisters::f22, + FloatRegister::Double}; +static constexpr FloatRegister f24 = {FloatRegisters::f24, + FloatRegister::Double}; +static constexpr FloatRegister f26 = {FloatRegisters::f26, + FloatRegister::Double}; +static constexpr FloatRegister f28 = {FloatRegisters::f28, + FloatRegister::Double}; +static constexpr FloatRegister f30 = {FloatRegisters::f30, + FloatRegister::Double}; + +// MIPS CPUs can only load multibyte data that is "naturally" +// four-byte-aligned, sp register should be eight-byte-aligned. +static constexpr uint32_t ABIStackAlignment = 8; +static constexpr uint32_t JitStackAlignment = 8; + +static constexpr uint32_t JitStackValueAlignment = + JitStackAlignment / sizeof(Value); +static_assert(JitStackAlignment % sizeof(Value) == 0 && + JitStackValueAlignment >= 1, + "Stack alignment should be a non-zero multiple of sizeof(Value)"); + +// TODO this is just a filler to prevent a build failure. The MIPS SIMD +// alignment requirements still need to be explored. +// TODO Copy the static_asserts from x64/x86 assembler files. +static constexpr uint32_t SimdMemoryAlignment = 8; +static constexpr uint32_t WasmStackAlignment = SimdMemoryAlignment; +static const uint32_t WasmTrapInstructionLength = 4; + +// The offsets are dynamically asserted during +// code generation in the prologue/epilogue. +static constexpr uint32_t WasmCheckedCallEntryOffset = 0u; +static constexpr uint32_t WasmCheckedTailEntryOffset = 16u; + +static constexpr Scale ScalePointer = TimesFour; + +class Assembler : public AssemblerMIPSShared { + public: + Assembler() : AssemblerMIPSShared() {} + + static Condition UnsignedCondition(Condition cond); + static Condition ConditionWithoutEqual(Condition cond); + + static uintptr_t GetPointer(uint8_t*); + + protected: + // This is used to access the odd register form the pair of single + // precision registers that make one double register. + FloatRegister getOddPair(FloatRegister reg) { + MOZ_ASSERT(reg.isDouble()); + MOZ_ASSERT(reg.id() % 2 == 0); + FloatRegister odd(reg.id() | 1, FloatRegister::Single); + return odd; + } + + public: + using AssemblerMIPSShared::bind; + + static void Bind(uint8_t* rawCode, const CodeLabel& label); + + void processCodeLabels(uint8_t* rawCode); + + static void TraceJumpRelocations(JSTracer* trc, JitCode* code, + CompactBufferReader& reader); + static void TraceDataRelocations(JSTracer* trc, JitCode* code, + CompactBufferReader& reader); + + void bind(InstImm* inst, uintptr_t branch, uintptr_t target); + + // Copy the assembly code to the given buffer, and perform any pending + // relocations relying on the target address. + void executableCopy(uint8_t* buffer); + + static uint32_t PatchWrite_NearCallSize(); + + static uint32_t ExtractLuiOriValue(Instruction* inst0, Instruction* inst1); + static void WriteLuiOriInstructions(Instruction* inst, Instruction* inst1, + Register reg, uint32_t value); + + static void PatchWrite_NearCall(CodeLocationLabel start, + CodeLocationLabel toCall); + static void PatchDataWithValueCheck(CodeLocationLabel label, ImmPtr newValue, + ImmPtr expectedValue); + static void PatchDataWithValueCheck(CodeLocationLabel label, + PatchedImmPtr newValue, + PatchedImmPtr expectedValue); + + static uint32_t ExtractInstructionImmediate(uint8_t* code); + + static void ToggleCall(CodeLocationLabel inst_, bool enabled); +}; // Assembler + +static const uint32_t NumIntArgRegs = 4; + +static inline bool GetIntArgReg(uint32_t usedArgSlots, Register* out) { + if (usedArgSlots < NumIntArgRegs) { + *out = Register::FromCode(a0.code() + usedArgSlots); + 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_mips32_Assembler_mips32_h */ diff --git a/js/src/jit/mips32/Bailouts-mips32.cpp b/js/src/jit/mips32/Bailouts-mips32.cpp new file mode 100644 index 0000000000..ac0fd14bd1 --- /dev/null +++ b/js/src/jit/mips32/Bailouts-mips32.cpp @@ -0,0 +1,57 @@ +/* -*- 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/mips32/Bailouts-mips32.h" + +#include "jit/JitFrames.h" +#include "jit/JitRuntime.h" +#include "jit/ScriptFromCalleeToken.h" +#include "vm/JSContext.h" +#include "vm/Realm.h" + +#include "vm/JSScript-inl.h" + +using namespace js; +using namespace js::jit; + +BailoutFrameInfo::BailoutFrameInfo(const JitActivationIterator& activations, + BailoutStack* bailout) + : machine_(bailout->machine()) { + uint8_t* sp = bailout->parentStackPointer(); + framePointer_ = sp + bailout->frameSize(); + topFrameSize_ = framePointer_ - sp; + + JSScript* script = + ScriptFromCalleeToken(((JitFrameLayout*)framePointer_)->calleeToken()); + JitActivation* activation = activations.activation()->asJit(); + topIonScript_ = script->ionScript(); + + attachOnJitActivation(activations); + + if (bailout->frameClass() == FrameSizeClass::None()) { + snapshotOffset_ = bailout->snapshotOffset(); + return; + } + + // Compute the snapshot offset from the bailout ID. + JSRuntime* rt = activation->compartment()->runtimeFromMainThread(); + TrampolinePtr code = rt->jitRuntime()->getBailoutTable(bailout->frameClass()); +#ifdef DEBUG + uint32_t tableSize = + rt->jitRuntime()->getBailoutTableSize(bailout->frameClass()); +#endif + uintptr_t tableOffset = bailout->tableOffset(); + uintptr_t tableStart = reinterpret_cast<uintptr_t>(code.value); + + MOZ_ASSERT(tableOffset >= tableStart && tableOffset < tableStart + tableSize); + MOZ_ASSERT((tableOffset - tableStart) % BAILOUT_TABLE_ENTRY_SIZE == 0); + + uint32_t bailoutId = + ((tableOffset - tableStart) / BAILOUT_TABLE_ENTRY_SIZE) - 1; + MOZ_ASSERT(bailoutId < BAILOUT_TABLE_SIZE); + + snapshotOffset_ = topIonScript_->bailoutToSnapshot(bailoutId); +} diff --git a/js/src/jit/mips32/Bailouts-mips32.h b/js/src/jit/mips32/Bailouts-mips32.h new file mode 100644 index 0000000000..5d9d60a482 --- /dev/null +++ b/js/src/jit/mips32/Bailouts-mips32.h @@ -0,0 +1,71 @@ +/* -*- 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_mips32_Bailouts_mips32_h +#define jit_mips32_Bailouts_mips32_h + +#include "jit/Bailouts.h" + +namespace js { +namespace jit { + +class BailoutStack { + uintptr_t frameClassId_; + // This is pushed in the bailout handler. Both entry points into the + // handler inserts their own value int lr, which is then placed onto the + // stack along with frameClassId_ above. This should be migrated to ip. + public: + union { + uintptr_t frameSize_; + uintptr_t tableOffset_; + }; + + protected: + RegisterDump::FPUArray fpregs_; + RegisterDump::GPRArray regs_; + + uintptr_t snapshotOffset_; + uintptr_t padding_; + + public: + FrameSizeClass frameClass() const { + return FrameSizeClass::FromClass(frameClassId_); + } + uintptr_t tableOffset() const { + MOZ_ASSERT(frameClass() != FrameSizeClass::None()); + return tableOffset_; + } + uint32_t frameSize() const { + if (frameClass() == FrameSizeClass::None()) { + return frameSize_; + } + return frameClass().frameSize(); + } + MachineState machine() { return MachineState::FromBailout(regs_, fpregs_); } + SnapshotOffset snapshotOffset() const { + MOZ_ASSERT(frameClass() == FrameSizeClass::None()); + return snapshotOffset_; + } + uint8_t* parentStackPointer() const { + if (frameClass() == FrameSizeClass::None()) { + return (uint8_t*)this + sizeof(BailoutStack); + } + return (uint8_t*)this + offsetof(BailoutStack, snapshotOffset_); + } + static size_t offsetOfFrameClass() { + return offsetof(BailoutStack, frameClassId_); + } + static size_t offsetOfFrameSize() { + return offsetof(BailoutStack, frameSize_); + } + static size_t offsetOfFpRegs() { return offsetof(BailoutStack, fpregs_); } + static size_t offsetOfRegs() { return offsetof(BailoutStack, regs_); } +}; + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_Bailouts_mips32_h */ diff --git a/js/src/jit/mips32/CodeGenerator-mips32.cpp b/js/src/jit/mips32/CodeGenerator-mips32.cpp new file mode 100644 index 0000000000..2cee25a7a5 --- /dev/null +++ b/js/src/jit/mips32/CodeGenerator-mips32.cpp @@ -0,0 +1,508 @@ +/* -*- 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/mips32/CodeGenerator-mips32.h" + +#include "mozilla/MathAlgorithms.h" + +#include "jit/CodeGenerator.h" +#include "jit/MIR.h" +#include "jit/MIRGraph.h" +#include "js/Conversions.h" +#include "vm/Shape.h" +#include "vm/TraceLogging.h" + +#include "jit/MacroAssembler-inl.h" +#include "jit/shared/CodeGenerator-shared-inl.h" + +using namespace js; +using namespace js::jit; + +ValueOperand CodeGeneratorMIPS::ToValue(LInstruction* ins, size_t pos) { + Register typeReg = ToRegister(ins->getOperand(pos + TYPE_INDEX)); + Register payloadReg = ToRegister(ins->getOperand(pos + PAYLOAD_INDEX)); + return ValueOperand(typeReg, payloadReg); +} + +ValueOperand CodeGeneratorMIPS::ToTempValue(LInstruction* ins, size_t pos) { + Register typeReg = ToRegister(ins->getTemp(pos + TYPE_INDEX)); + Register payloadReg = ToRegister(ins->getTemp(pos + PAYLOAD_INDEX)); + return ValueOperand(typeReg, payloadReg); +} + +void CodeGenerator::visitBox(LBox* box) { + const LDefinition* type = box->getDef(TYPE_INDEX); + + MOZ_ASSERT(!box->getOperand(0)->isConstant()); + + // For NUNBOX32, the input operand and the output payload have the same + // virtual register. All that needs to be written is the type tag for + // the type definition. + masm.move32(Imm32(MIRTypeToTag(box->type())), ToRegister(type)); +} + +void CodeGenerator::visitBoxFloatingPoint(LBoxFloatingPoint* box) { + const AnyRegister in = ToAnyRegister(box->getOperand(0)); + const ValueOperand out = ToOutValue(box); + + masm.moveValue(TypedOrValueRegister(box->type(), in), out); +} + +void CodeGenerator::visitUnbox(LUnbox* unbox) { + // Note that for unbox, the type and payload indexes are switched on the + // inputs. + MUnbox* mir = unbox->mir(); + Register type = ToRegister(unbox->type()); + + if (mir->fallible()) { + bailoutCmp32(Assembler::NotEqual, type, Imm32(MIRTypeToTag(mir->type())), + unbox->snapshot()); + } +} + +void CodeGeneratorMIPS::splitTagForTest(const ValueOperand& value, + ScratchTagScope& tag) { + MOZ_ASSERT(value.typeReg() == tag); +} + +void CodeGenerator::visitCompareI64(LCompareI64* lir) { + MCompare* mir = lir->mir(); + MOZ_ASSERT(mir->compareType() == MCompare::Compare_Int64 || + mir->compareType() == MCompare::Compare_UInt64); + + const LInt64Allocation lhs = lir->getInt64Operand(LCompareI64::Lhs); + const LInt64Allocation rhs = lir->getInt64Operand(LCompareI64::Rhs); + Register64 lhsRegs = ToRegister64(lhs); + Register output = ToRegister(lir->output()); + + bool isSigned = mir->compareType() == MCompare::Compare_Int64; + Assembler::Condition condition = JSOpToCondition(lir->jsop(), isSigned); + + if (IsConstant(rhs)) { + Imm64 imm = Imm64(ToInt64(rhs)); + masm.cmp64Set(condition, lhsRegs, imm, output); + } else { + Register64 rhsRegs = ToRegister64(rhs); + masm.cmp64Set(condition, lhsRegs, rhsRegs, output); + } +} + +void CodeGenerator::visitCompareI64AndBranch(LCompareI64AndBranch* lir) { + MCompare* mir = lir->cmpMir(); + MOZ_ASSERT(mir->compareType() == MCompare::Compare_Int64 || + mir->compareType() == MCompare::Compare_UInt64); + + const LInt64Allocation lhs = lir->getInt64Operand(LCompareI64::Lhs); + const LInt64Allocation rhs = lir->getInt64Operand(LCompareI64::Rhs); + Register64 lhsRegs = ToRegister64(lhs); + + bool isSigned = mir->compareType() == MCompare::Compare_Int64; + Assembler::Condition condition = JSOpToCondition(lir->jsop(), isSigned); + + Label* trueLabel = getJumpLabelForBranch(lir->ifTrue()); + Label* falseLabel = getJumpLabelForBranch(lir->ifFalse()); + + if (isNextBlock(lir->ifFalse()->lir())) { + falseLabel = nullptr; + } else if (isNextBlock(lir->ifTrue()->lir())) { + condition = Assembler::InvertCondition(condition); + trueLabel = falseLabel; + falseLabel = nullptr; + } + + if (IsConstant(rhs)) { + Imm64 imm = Imm64(ToInt64(rhs)); + masm.branch64(condition, lhsRegs, imm, trueLabel, falseLabel); + } else { + Register64 rhsRegs = ToRegister64(rhs); + masm.branch64(condition, lhsRegs, rhsRegs, trueLabel, falseLabel); + } +} + +void CodeGenerator::visitDivOrModI64(LDivOrModI64* lir) { + Register64 lhs = ToRegister64(lir->getInt64Operand(LDivOrModI64::Lhs)); + Register64 rhs = ToRegister64(lir->getInt64Operand(LDivOrModI64::Rhs)); + Register64 output = ToOutRegister64(lir); + + MOZ_ASSERT(output == ReturnReg64); + + Label done; + + // Handle divide by zero. + if (lir->canBeDivideByZero()) { + Label nonZero; + masm.branchTest64(Assembler::NonZero, rhs, rhs, InvalidReg, &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.branch64(Assembler::NotEqual, lhs, Imm64(INT64_MIN), ¬Overflow); + masm.branch64(Assembler::NotEqual, rhs, Imm64(-1), ¬Overflow); + if (lir->mir()->isMod()) { + masm.xor64(output, output); + } else { + masm.wasmTrap(wasm::Trap::IntegerOverflow, lir->bytecodeOffset()); + } + masm.jump(&done); + masm.bind(¬Overflow); + } + + masm.setupWasmABICall(); + masm.passABIArg(lhs.high); + masm.passABIArg(lhs.low); + masm.passABIArg(rhs.high); + masm.passABIArg(rhs.low); + + MOZ_ASSERT(gen->compilingWasm()); + if (lir->mir()->isMod()) { + masm.callWithABI(lir->bytecodeOffset(), wasm::SymbolicAddress::ModI64); + } else { + masm.callWithABI(lir->bytecodeOffset(), wasm::SymbolicAddress::DivI64); + } + MOZ_ASSERT(ReturnReg64 == output); + + masm.bind(&done); +} + +void CodeGenerator::visitUDivOrModI64(LUDivOrModI64* lir) { + Register64 lhs = ToRegister64(lir->getInt64Operand(LDivOrModI64::Lhs)); + Register64 rhs = ToRegister64(lir->getInt64Operand(LDivOrModI64::Rhs)); + + MOZ_ASSERT(ToOutRegister64(lir) == ReturnReg64); + + // Prevent divide by zero. + if (lir->canBeDivideByZero()) { + Label nonZero; + masm.branchTest64(Assembler::NonZero, rhs, rhs, InvalidReg, &nonZero); + masm.wasmTrap(wasm::Trap::IntegerDivideByZero, lir->bytecodeOffset()); + masm.bind(&nonZero); + } + + masm.setupWasmABICall(); + masm.passABIArg(lhs.high); + masm.passABIArg(lhs.low); + masm.passABIArg(rhs.high); + masm.passABIArg(rhs.low); + + MOZ_ASSERT(gen->compilingWasm()); + if (lir->mir()->isMod()) { + masm.callWithABI(lir->bytecodeOffset(), wasm::SymbolicAddress::UModI64); + } else { + masm.callWithABI(lir->bytecodeOffset(), wasm::SymbolicAddress::UDivI64); + } +} + +void CodeGeneratorMIPS::emitBigIntDiv(LBigIntDiv* ins, Register dividend, + Register divisor, Register output, + Label* fail) { + // Callers handle division by zero and integer overflow. + +#ifdef MIPSR6 + masm.as_div(/* result= */ dividend, dividend, divisor); +#else + masm.as_div(dividend, divisor); + masm.as_mflo(dividend); +#endif + + // Create and return the result. + masm.newGCBigInt(output, divisor, fail, bigIntsCanBeInNursery()); + masm.initializeBigInt(output, dividend); +} + +void CodeGeneratorMIPS::emitBigIntMod(LBigIntMod* ins, Register dividend, + Register divisor, Register output, + Label* fail) { + // Callers handle division by zero and integer overflow. + +#ifdef MIPSR6 + masm.as_mod(/* result= */ dividend, dividend, divisor); +#else + masm.as_div(dividend, divisor); + masm.as_mfhi(dividend); +#endif + + // Create and return the result. + masm.newGCBigInt(output, divisor, fail, bigIntsCanBeInNursery()); + masm.initializeBigInt(output, dividend); +} + +template <typename T> +void CodeGeneratorMIPS::emitWasmLoadI64(T* lir) { + const MWasmLoad* mir = lir->mir(); + + Register ptrScratch = InvalidReg; + if (!lir->ptrCopy()->isBogusTemp()) { + ptrScratch = ToRegister(lir->ptrCopy()); + } + + if (IsUnaligned(mir->access())) { + masm.wasmUnalignedLoadI64(mir->access(), HeapReg, ToRegister(lir->ptr()), + ptrScratch, ToOutRegister64(lir), + ToRegister(lir->getTemp(1))); + } else { + masm.wasmLoadI64(mir->access(), HeapReg, ToRegister(lir->ptr()), ptrScratch, + ToOutRegister64(lir)); + } +} + +void CodeGenerator::visitWasmLoadI64(LWasmLoadI64* lir) { + emitWasmLoadI64(lir); +} + +void CodeGenerator::visitWasmUnalignedLoadI64(LWasmUnalignedLoadI64* lir) { + emitWasmLoadI64(lir); +} + +template <typename T> +void CodeGeneratorMIPS::emitWasmStoreI64(T* lir) { + const MWasmStore* mir = lir->mir(); + + Register ptrScratch = InvalidReg; + if (!lir->ptrCopy()->isBogusTemp()) { + ptrScratch = ToRegister(lir->ptrCopy()); + } + + if (IsUnaligned(mir->access())) { + masm.wasmUnalignedStoreI64(mir->access(), ToRegister64(lir->value()), + HeapReg, ToRegister(lir->ptr()), ptrScratch, + ToRegister(lir->getTemp(1))); + } else { + masm.wasmStoreI64(mir->access(), ToRegister64(lir->value()), HeapReg, + ToRegister(lir->ptr()), ptrScratch); + } +} + +void CodeGenerator::visitWasmStoreI64(LWasmStoreI64* lir) { + emitWasmStoreI64(lir); +} + +void CodeGenerator::visitWasmUnalignedStoreI64(LWasmUnalignedStoreI64* lir) { + emitWasmStoreI64(lir); +} + +void CodeGenerator::visitWasmSelectI64(LWasmSelectI64* lir) { + MOZ_ASSERT(lir->mir()->type() == MIRType::Int64); + Register cond = ToRegister(lir->condExpr()); + const LInt64Allocation trueExpr = lir->trueExpr(); + const LInt64Allocation falseExpr = lir->falseExpr(); + + Register64 output = ToOutRegister64(lir); + + masm.move64(ToRegister64(trueExpr), output); + + if (falseExpr.low().isRegister()) { + masm.as_movz(output.low, ToRegister(falseExpr.low()), cond); + masm.as_movz(output.high, ToRegister(falseExpr.high()), cond); + } else { + Label done; + masm.ma_b(cond, cond, &done, Assembler::NonZero, ShortJump); + masm.loadPtr(ToAddress(falseExpr.low()), output.low); + masm.loadPtr(ToAddress(falseExpr.high()), output.high); + masm.bind(&done); + } +} + +void CodeGenerator::visitWasmReinterpretFromI64(LWasmReinterpretFromI64* lir) { + MOZ_ASSERT(lir->mir()->type() == MIRType::Double); + MOZ_ASSERT(lir->mir()->input()->type() == MIRType::Int64); + Register64 input = ToRegister64(lir->getInt64Operand(0)); + FloatRegister output = ToFloatRegister(lir->output()); + + masm.moveToDoubleLo(input.low, output); + masm.moveToDoubleHi(input.high, output); +} + +void CodeGenerator::visitWasmReinterpretToI64(LWasmReinterpretToI64* lir) { + MOZ_ASSERT(lir->mir()->type() == MIRType::Int64); + MOZ_ASSERT(lir->mir()->input()->type() == MIRType::Double); + FloatRegister input = ToFloatRegister(lir->getOperand(0)); + Register64 output = ToOutRegister64(lir); + + masm.moveFromDoubleLo(input, output.low); + masm.moveFromDoubleHi(input, output.high); +} + +void CodeGenerator::visitExtendInt32ToInt64(LExtendInt32ToInt64* lir) { + Register input = ToRegister(lir->input()); + Register64 output = ToOutRegister64(lir); + + if (input != output.low) { + masm.move32(input, output.low); + } + if (lir->mir()->isUnsigned()) { + masm.move32(Imm32(0), output.high); + } else { + masm.ma_sra(output.high, output.low, Imm32(31)); + } +} + +void CodeGenerator::visitWrapInt64ToInt32(LWrapInt64ToInt32* lir) { + const LInt64Allocation& input = lir->getInt64Operand(0); + Register output = ToRegister(lir->output()); + + if (lir->mir()->bottomHalf()) { + masm.move32(ToRegister(input.low()), output); + } else { + masm.move32(ToRegister(input.high()), output); + } +} + +void CodeGenerator::visitSignExtendInt64(LSignExtendInt64* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + Register64 output = ToOutRegister64(lir); + switch (lir->mode()) { + case MSignExtendInt64::Byte: + masm.move8SignExtend(input.low, output.low); + break; + case MSignExtendInt64::Half: + masm.move16SignExtend(input.low, output.low); + break; + case MSignExtendInt64::Word: + masm.move32(input.low, output.low); + break; + } + masm.ma_sra(output.high, output.low, Imm32(31)); +} + +void CodeGenerator::visitClzI64(LClzI64* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + Register64 output = ToOutRegister64(lir); + masm.clz64(input, output.low); + masm.move32(Imm32(0), output.high); +} + +void CodeGenerator::visitCtzI64(LCtzI64* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + Register64 output = ToOutRegister64(lir); + masm.ctz64(input, output.low); + masm.move32(Imm32(0), output.high); +} + +void CodeGenerator::visitNotI64(LNotI64* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + Register output = ToRegister(lir->output()); + + masm.as_or(output, input.low, input.high); + masm.cmp32Set(Assembler::Equal, output, Imm32(0), output); +} + +void CodeGenerator::visitWasmTruncateToInt64(LWasmTruncateToInt64* lir) { + FloatRegister input = ToFloatRegister(lir->input()); + FloatRegister arg = input; + Register64 output = ToOutRegister64(lir); + MWasmTruncateToInt64* mir = lir->mir(); + MIRType fromType = mir->input()->type(); + + auto* ool = new (alloc()) + OutOfLineWasmTruncateCheck(mir, input, Register64::Invalid()); + addOutOfLineCode(ool, mir); + + if (fromType == MIRType::Float32) { + arg = ScratchDoubleReg; + masm.convertFloat32ToDouble(input, arg); + } + + if (!lir->mir()->isSaturating()) { + masm.Push(input); + + masm.setupWasmABICall(); + masm.passABIArg(arg, MoveOp::DOUBLE); + + if (lir->mir()->isUnsigned()) { + masm.callWithABI(mir->bytecodeOffset(), + wasm::SymbolicAddress::TruncateDoubleToUint64); + } else { + masm.callWithABI(mir->bytecodeOffset(), + wasm::SymbolicAddress::TruncateDoubleToInt64); + } + + masm.Pop(input); + + masm.ma_xor(ScratchRegister, output.high, Imm32(0x80000000)); + masm.ma_or(ScratchRegister, output.low); + masm.ma_b(ScratchRegister, Imm32(0), ool->entry(), Assembler::Equal); + + masm.bind(ool->rejoin()); + } else { + masm.setupWasmABICall(); + masm.passABIArg(arg, MoveOp::DOUBLE); + if (lir->mir()->isUnsigned()) { + masm.callWithABI(mir->bytecodeOffset(), + wasm::SymbolicAddress::SaturatingTruncateDoubleToUint64); + } else { + masm.callWithABI(mir->bytecodeOffset(), + wasm::SymbolicAddress::SaturatingTruncateDoubleToInt64); + } + } + + MOZ_ASSERT(ReturnReg64 == output); +} + +void CodeGenerator::visitInt64ToFloatingPoint(LInt64ToFloatingPoint* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + mozilla::DebugOnly<FloatRegister> output = ToFloatRegister(lir->output()); + + MInt64ToFloatingPoint* mir = lir->mir(); + MIRType toType = mir->type(); + + masm.setupWasmABICall(); + masm.passABIArg(input.high); + masm.passABIArg(input.low); + + if (lir->mir()->isUnsigned()) { + if (toType == MIRType::Double) { + masm.callWithABI(mir->bytecodeOffset(), + wasm::SymbolicAddress::Uint64ToDouble, MoveOp::DOUBLE); + } else { + masm.callWithABI(mir->bytecodeOffset(), + wasm::SymbolicAddress::Uint64ToFloat32, MoveOp::FLOAT32); + } + } else { + if (toType == MIRType::Double) { + masm.callWithABI(mir->bytecodeOffset(), + wasm::SymbolicAddress::Int64ToDouble, MoveOp::DOUBLE); + } else { + masm.callWithABI(mir->bytecodeOffset(), + wasm::SymbolicAddress::Int64ToFloat32, MoveOp::FLOAT32); + } + } + + MOZ_ASSERT_IF(toType == MIRType::Double, *(&output) == ReturnDoubleReg); + MOZ_ASSERT_IF(toType == MIRType::Float32, *(&output) == ReturnFloat32Reg); +} + +void CodeGenerator::visitTestI64AndBranch(LTestI64AndBranch* lir) { + Register64 input = ToRegister64(lir->getInt64Operand(0)); + + branchToBlock(input.high, Imm32(0), lir->ifTrue(), Assembler::NonZero); + emitBranch(input.low, Imm32(0), Assembler::NonZero, lir->ifTrue(), + lir->ifFalse()); +} + +void CodeGenerator::visitWasmAtomicLoadI64(LWasmAtomicLoadI64* lir) { + Register ptr = ToRegister(lir->ptr()); + Register64 output = ToOutRegister64(lir); + uint32_t offset = lir->mir()->access().offset(); + + BaseIndex addr(HeapReg, ptr, TimesOne, offset); + + masm.wasmAtomicLoad64(lir->mir()->access(), addr, Register64::Invalid(), + output); +} + +void CodeGenerator::visitWasmAtomicStoreI64(LWasmAtomicStoreI64* lir) { + Register ptr = ToRegister(lir->ptr()); + Register64 value = ToRegister64(lir->value()); + Register tmp = ToRegister(lir->tmp()); + uint32_t offset = lir->mir()->access().offset(); + + BaseIndex addr(HeapReg, ptr, TimesOne, offset); + + masm.wasmAtomicStore64(lir->mir()->access(), addr, tmp, value); +} diff --git a/js/src/jit/mips32/CodeGenerator-mips32.h b/js/src/jit/mips32/CodeGenerator-mips32.h new file mode 100644 index 0000000000..161b94326f --- /dev/null +++ b/js/src/jit/mips32/CodeGenerator-mips32.h @@ -0,0 +1,60 @@ +/* -*- 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_mips32_CodeGenerator_mips32_h +#define jit_mips32_CodeGenerator_mips32_h + +#include "jit/mips-shared/CodeGenerator-mips-shared.h" + +namespace js { +namespace jit { + +class CodeGeneratorMIPS : public CodeGeneratorMIPSShared { + protected: + CodeGeneratorMIPS(MIRGenerator* gen, LIRGraph* graph, MacroAssembler* masm) + : CodeGeneratorMIPSShared(gen, graph, masm) {} + + void testNullEmitBranch(Assembler::Condition cond, const ValueOperand& value, + MBasicBlock* ifTrue, MBasicBlock* ifFalse) { + emitBranch(value.typeReg(), (Imm32)ImmType(JSVAL_TYPE_NULL), cond, ifTrue, + ifFalse); + } + void testUndefinedEmitBranch(Assembler::Condition cond, + const ValueOperand& value, MBasicBlock* ifTrue, + MBasicBlock* ifFalse) { + emitBranch(value.typeReg(), (Imm32)ImmType(JSVAL_TYPE_UNDEFINED), cond, + ifTrue, ifFalse); + } + void testObjectEmitBranch(Assembler::Condition cond, + const ValueOperand& value, MBasicBlock* ifTrue, + MBasicBlock* ifFalse) { + emitBranch(value.typeReg(), (Imm32)ImmType(JSVAL_TYPE_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 CodeGeneratorMIPS CodeGeneratorSpecific; + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_CodeGenerator_mips32_h */ diff --git a/js/src/jit/mips32/LIR-mips32.h b/js/src/jit/mips32/LIR-mips32.h new file mode 100644 index 0000000000..da68ad7464 --- /dev/null +++ b/js/src/jit/mips32/LIR-mips32.h @@ -0,0 +1,197 @@ +/* -*- 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_mips32_LIR_mips32_h +#define jit_mips32_LIR_mips32_h + +namespace js { +namespace jit { + +class LBoxFloatingPoint : public LInstructionHelper<2, 1, 1> { + MIRType type_; + + public: + LIR_HEADER(BoxFloatingPoint); + + LBoxFloatingPoint(const LAllocation& in, const LDefinition& temp, + MIRType type) + : LInstructionHelper(classOpcode), type_(type) { + setOperand(0, in); + setTemp(0, temp); + } + + MIRType type() const { return type_; } + const char* extraName() const { return StringFromMIRType(type_); } +}; + +class LUnbox : public LInstructionHelper<1, 2, 0> { + public: + LIR_HEADER(Unbox); + + LUnbox() : LInstructionHelper(classOpcode) {} + + MUnbox* mir() const { return mir_->toUnbox(); } + const LAllocation* payload() { return getOperand(0); } + const LAllocation* type() { return getOperand(1); } + const char* extraName() const { return StringFromMIRType(mir()->type()); } +}; + +class LUnboxFloatingPoint : public LInstructionHelper<1, 2, 0> { + MIRType type_; + + public: + LIR_HEADER(UnboxFloatingPoint); + + static const size_t Input = 0; + + LUnboxFloatingPoint(const LBoxAllocation& input, MIRType type) + : LInstructionHelper(classOpcode), type_(type) { + setBoxOperand(Input, input); + } + + MUnbox* mir() const { return mir_->toUnbox(); } + MIRType type() const { return type_; } + const char* extraName() const { return StringFromMIRType(type_); } +}; + +class LDivOrModI64 + : public LCallInstructionHelper<INT64_PIECES, INT64_PIECES * 2, 0> { + public: + LIR_HEADER(DivOrModI64) + + static const size_t Lhs = 0; + static const size_t Rhs = INT64_PIECES; + + LDivOrModI64(const LInt64Allocation& lhs, const LInt64Allocation& rhs) + : LCallInstructionHelper(classOpcode) { + setInt64Operand(Lhs, lhs); + setInt64Operand(Rhs, rhs); + } + + 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 LCallInstructionHelper<INT64_PIECES, INT64_PIECES * 2, 0> { + public: + LIR_HEADER(UDivOrModI64) + + static const size_t Lhs = 0; + static const size_t Rhs = INT64_PIECES; + + LUDivOrModI64(const LInt64Allocation& lhs, const LInt64Allocation& rhs) + : LCallInstructionHelper(classOpcode) { + setInt64Operand(Lhs, lhs); + setInt64Operand(Rhs, rhs); + } + 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 LWasmTruncateToInt64 : public LCallInstructionHelper<INT64_PIECES, 1, 0> { + public: + LIR_HEADER(WasmTruncateToInt64); + + explicit LWasmTruncateToInt64(const LAllocation& in) + : LCallInstructionHelper(classOpcode) { + setOperand(0, in); + } + + MWasmTruncateToInt64* mir() const { return mir_->toWasmTruncateToInt64(); } +}; + +class LInt64ToFloatingPoint + : public LCallInstructionHelper<1, INT64_PIECES, 0> { + public: + LIR_HEADER(Int64ToFloatingPoint); + + explicit LInt64ToFloatingPoint(const LInt64Allocation& in) + : LCallInstructionHelper(classOpcode) { + setInt64Operand(0, in); + } + + MInt64ToFloatingPoint* mir() const { return mir_->toInt64ToFloatingPoint(); } +}; + +class LWasmAtomicLoadI64 : public LInstructionHelper<INT64_PIECES, 1, 0> { + public: + LIR_HEADER(WasmAtomicLoadI64); + + LWasmAtomicLoadI64(const LAllocation& ptr) : LInstructionHelper(classOpcode) { + setOperand(0, ptr); + } + + const LAllocation* ptr() { return getOperand(0); } + const MWasmLoad* mir() const { return mir_->toWasmLoad(); } +}; + +class LWasmAtomicStoreI64 : public LInstructionHelper<0, 1 + INT64_PIECES, 1> { + public: + LIR_HEADER(WasmAtomicStoreI64); + + LWasmAtomicStoreI64(const LAllocation& ptr, const LInt64Allocation& value, + const LDefinition& tmp) + : LInstructionHelper(classOpcode) { + setOperand(0, ptr); + setInt64Operand(1, value); + setTemp(0, tmp); + } + + const LAllocation* ptr() { return getOperand(0); } + const LInt64Allocation value() { return getInt64Operand(1); } + const LDefinition* tmp() { return getTemp(0); } + const MWasmStore* mir() const { return mir_->toWasmStore(); } +}; + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_LIR_mips32_h */ diff --git a/js/src/jit/mips32/Lowering-mips32.cpp b/js/src/jit/mips32/Lowering-mips32.cpp new file mode 100644 index 0000000000..721491c46b --- /dev/null +++ b/js/src/jit/mips32/Lowering-mips32.cpp @@ -0,0 +1,257 @@ +/* -*- 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/mips32/Lowering-mips32.h" + +#include "jit/Lowering.h" +#include "jit/mips32/Assembler-mips32.h" +#include "jit/MIR.h" + +#include "jit/shared/Lowering-shared-inl.h" + +using namespace js; +using namespace js::jit; + +LBoxAllocation LIRGeneratorMIPS::useBoxFixed(MDefinition* mir, Register reg1, + Register reg2, bool useAtStart) { + MOZ_ASSERT(mir->type() == MIRType::Value); + MOZ_ASSERT(reg1 != reg2); + + ensureDefined(mir); + return LBoxAllocation(LUse(reg1, mir->virtualRegister(), useAtStart), + LUse(reg2, VirtualRegisterOfPayload(mir), useAtStart)); +} + +void LIRGenerator::visitBox(MBox* box) { + MDefinition* inner = box->getOperand(0); + + // If the box wrapped a double, it needs a new register. + if (IsFloatingPointType(inner->type())) { + defineBox(new (alloc()) LBoxFloatingPoint( + useRegisterAtStart(inner), tempCopy(inner, 0), inner->type()), + box); + return; + } + + if (box->canEmitAtUses()) { + emitAtUses(box); + return; + } + + if (inner->isConstant()) { + defineBox(new (alloc()) LValue(inner->toConstant()->toJSValue()), box); + return; + } + + LBox* lir = new (alloc()) LBox(use(inner), inner->type()); + + // Otherwise, we should not define a new register for the payload portion + // of the output, so bypass defineBox(). + uint32_t vreg = getVirtualRegister(); + + // Note that because we're using BogusTemp(), we do not change the type of + // the definition. We also do not define the first output as "TYPE", + // because it has no corresponding payload at (vreg + 1). Also note that + // although we copy the input's original type for the payload half of the + // definition, this is only for clarity. BogusTemp() definitions are + // ignored. + lir->setDef(0, LDefinition(vreg, LDefinition::GENERAL)); + lir->setDef(1, LDefinition::BogusTemp()); + box->setVirtualRegister(vreg); + add(lir); +} + +void LIRGenerator::visitUnbox(MUnbox* unbox) { + MDefinition* inner = unbox->getOperand(0); + + // An unbox on mips reads in a type tag (either in memory or a register) and + // a payload. Unlike most instructions consuming a box, we ask for the type + // second, so that the result can re-use the first input. + MOZ_ASSERT(inner->type() == MIRType::Value); + + ensureDefined(inner); + + if (IsFloatingPointType(unbox->type())) { + LUnboxFloatingPoint* lir = + new (alloc()) LUnboxFloatingPoint(useBox(inner), unbox->type()); + if (unbox->fallible()) { + assignSnapshot(lir, unbox->bailoutKind()); + } + define(lir, unbox); + return; + } + + // Swap the order we use the box pieces so we can re-use the payload + // register. + LUnbox* lir = new (alloc()) LUnbox; + lir->setOperand(0, usePayloadInRegisterAtStart(inner)); + lir->setOperand(1, useType(inner, LUse::REGISTER)); + + if (unbox->fallible()) { + assignSnapshot(lir, unbox->bailoutKind()); + } + + // Types and payloads form two separate intervals. If the type becomes dead + // before the payload, it could be used as a Value without the type being + // recoverable. Unbox's purpose is to eagerly kill the definition of a type + // tag, so keeping both alive (for the purpose of gcmaps) is unappealing. + // Instead, we create a new virtual register. + defineReuseInput(lir, unbox, 0); +} + +void LIRGenerator::visitReturnImpl(MDefinition* opd, bool isGenerator) { + MOZ_ASSERT(opd->type() == MIRType::Value); + + LReturn* ins = new (alloc()) LReturn(isGenerator); + ins->setOperand(0, LUse(JSReturnReg_Type)); + ins->setOperand(1, LUse(JSReturnReg_Data)); + fillBoxUses(ins, 0, opd); + add(ins); +} + +void LIRGeneratorMIPS::lowerUntypedPhiInput(MPhi* phi, uint32_t inputPosition, + LBlock* block, size_t lirIndex) { + MDefinition* operand = phi->getOperand(inputPosition); + LPhi* type = block->getPhi(lirIndex + VREG_TYPE_OFFSET); + LPhi* payload = block->getPhi(lirIndex + VREG_DATA_OFFSET); + type->setOperand( + inputPosition, + LUse(operand->virtualRegister() + VREG_TYPE_OFFSET, LUse::ANY)); + payload->setOperand(inputPosition, + LUse(VirtualRegisterOfPayload(operand), LUse::ANY)); +} + +void LIRGeneratorMIPS::defineInt64Phi(MPhi* phi, size_t lirIndex) { + LPhi* low = current->getPhi(lirIndex + INT64LOW_INDEX); + LPhi* high = current->getPhi(lirIndex + INT64HIGH_INDEX); + + uint32_t lowVreg = getVirtualRegister(); + + phi->setVirtualRegister(lowVreg); + + uint32_t highVreg = getVirtualRegister(); + MOZ_ASSERT(lowVreg + INT64HIGH_INDEX == highVreg + INT64LOW_INDEX); + + low->setDef(0, LDefinition(lowVreg, LDefinition::INT32)); + high->setDef(0, LDefinition(highVreg, LDefinition::INT32)); + annotate(high); + annotate(low); +} + +void LIRGeneratorMIPS::lowerInt64PhiInput(MPhi* phi, uint32_t inputPosition, + LBlock* block, size_t lirIndex) { + MDefinition* operand = phi->getOperand(inputPosition); + LPhi* low = block->getPhi(lirIndex + INT64LOW_INDEX); + LPhi* high = block->getPhi(lirIndex + INT64HIGH_INDEX); + low->setOperand(inputPosition, + LUse(operand->virtualRegister() + INT64LOW_INDEX, LUse::ANY)); + high->setOperand( + inputPosition, + LUse(operand->virtualRegister() + INT64HIGH_INDEX, LUse::ANY)); +} + +void LIRGeneratorMIPS::lowerTruncateDToInt32(MTruncateToInt32* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Double); + + define(new (alloc()) + LTruncateDToInt32(useRegister(opd), LDefinition::BogusTemp()), + ins); +} + +void LIRGeneratorMIPS::lowerTruncateFToInt32(MTruncateToInt32* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Float32); + + define(new (alloc()) + LTruncateFToInt32(useRegister(opd), LDefinition::BogusTemp()), + ins); +} + +void LIRGeneratorMIPS::lowerDivI64(MDiv* div) { + if (div->isUnsigned()) { + lowerUDivI64(div); + return; + } + + LDivOrModI64* lir = new (alloc()) LDivOrModI64( + useInt64RegisterAtStart(div->lhs()), useInt64RegisterAtStart(div->rhs())); + + defineReturn(lir, div); +} + +void LIRGeneratorMIPS::lowerWasmBuiltinDivI64(MWasmBuiltinDivI64* div) { + MOZ_CRASH("We don't use runtime div for this architecture"); +} + +void LIRGeneratorMIPS::lowerModI64(MMod* mod) { + if (mod->isUnsigned()) { + lowerUModI64(mod); + return; + } + + LDivOrModI64* lir = new (alloc()) LDivOrModI64( + useInt64RegisterAtStart(mod->lhs()), useInt64RegisterAtStart(mod->rhs())); + + defineReturn(lir, mod); +} + +void LIRGeneratorMIPS::lowerWasmBuiltinModI64(MWasmBuiltinModI64* mod) { + MOZ_CRASH("We don't use runtime mod for this architecture"); +} + +void LIRGeneratorMIPS::lowerUDivI64(MDiv* div) { + LUDivOrModI64* lir = new (alloc()) LUDivOrModI64( + useInt64RegisterAtStart(div->lhs()), useInt64RegisterAtStart(div->rhs())); + defineReturn(lir, div); +} + +void LIRGeneratorMIPS::lowerUModI64(MMod* mod) { + LUDivOrModI64* lir = new (alloc()) LUDivOrModI64( + useInt64RegisterAtStart(mod->lhs()), useInt64RegisterAtStart(mod->rhs())); + defineReturn(lir, mod); +} + +void LIRGeneratorMIPS::lowerBigIntDiv(MBigIntDiv* ins) { + auto* lir = new (alloc()) LBigIntDiv(useRegister(ins->lhs()), + useRegister(ins->rhs()), temp(), temp()); + define(lir, ins); + assignSafepoint(lir, ins); +} + +void LIRGeneratorMIPS::lowerBigIntMod(MBigIntMod* ins) { + auto* lir = new (alloc()) LBigIntMod(useRegister(ins->lhs()), + useRegister(ins->rhs()), temp(), temp()); + define(lir, ins); + assignSafepoint(lir, ins); +} + +void LIRGenerator::visitWasmTruncateToInt64(MWasmTruncateToInt64* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Double || opd->type() == MIRType::Float32); + + defineReturn(new (alloc()) LWasmTruncateToInt64(useRegisterAtStart(opd)), + ins); +} + +void LIRGeneratorMIPS::lowerWasmBuiltinTruncateToInt64( + MWasmBuiltinTruncateToInt64* ins) { + MOZ_CRASH("We don't use it for this architecture"); +} + +void LIRGenerator::visitInt64ToFloatingPoint(MInt64ToFloatingPoint* ins) { + MDefinition* opd = ins->input(); + MOZ_ASSERT(opd->type() == MIRType::Int64); + MOZ_ASSERT(IsFloatingPointType(ins->type())); + + defineReturn( + new (alloc()) LInt64ToFloatingPoint(useInt64RegisterAtStart(opd)), ins); +} + +void LIRGeneratorMIPS::lowerBuiltinInt64ToFloatingPoint( + MBuiltinInt64ToFloatingPoint* ins) { + MOZ_CRASH("We don't use it for this architecture"); +} diff --git a/js/src/jit/mips32/Lowering-mips32.h b/js/src/jit/mips32/Lowering-mips32.h new file mode 100644 index 0000000000..1565c84656 --- /dev/null +++ b/js/src/jit/mips32/Lowering-mips32.h @@ -0,0 +1,54 @@ +/* -*- 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_mips32_Lowering_mips32_h +#define jit_mips32_Lowering_mips32_h + +#include "jit/mips-shared/Lowering-mips-shared.h" + +namespace js { +namespace jit { + +class LIRGeneratorMIPS : public LIRGeneratorMIPSShared { + protected: + LIRGeneratorMIPS(MIRGenerator* gen, MIRGraph& graph, LIRGraph& lirGraph) + : LIRGeneratorMIPSShared(gen, graph, lirGraph) {} + + // Returns a box allocation with type set to reg1 and payload set to reg2. + LBoxAllocation useBoxFixed(MDefinition* mir, Register reg1, Register reg2, + bool useAtStart = false); + + inline LDefinition tempToUnbox() { return LDefinition::BogusTemp(); } + + void lowerUntypedPhiInput(MPhi* phi, uint32_t inputPosition, LBlock* block, + size_t lirIndex); + + void lowerInt64PhiInput(MPhi* phi, uint32_t inputPosition, LBlock* block, + size_t lirIndex); + void defineInt64Phi(MPhi* phi, size_t lirIndex); + + void lowerBuiltinInt64ToFloatingPoint(MBuiltinInt64ToFloatingPoint* ins); + void lowerWasmBuiltinTruncateToInt64(MWasmBuiltinTruncateToInt64* ins); + void lowerTruncateDToInt32(MTruncateToInt32* ins); + void lowerTruncateFToInt32(MTruncateToInt32* ins); + + void lowerDivI64(MDiv* div); + void lowerWasmBuiltinDivI64(MWasmBuiltinDivI64* div); + void lowerModI64(MMod* mod); + void lowerWasmBuiltinModI64(MWasmBuiltinModI64* mod); + void lowerUDivI64(MDiv* div); + void lowerUModI64(MMod* mod); + + void lowerBigIntDiv(MBigIntDiv* ins); + void lowerBigIntMod(MBigIntMod* ins); +}; + +typedef LIRGeneratorMIPS LIRGeneratorSpecific; + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_Lowering_mips32_h */ diff --git a/js/src/jit/mips32/MacroAssembler-mips32-inl.h b/js/src/jit/mips32/MacroAssembler-mips32-inl.h new file mode 100644 index 0000000000..7899d4c335 --- /dev/null +++ b/js/src/jit/mips32/MacroAssembler-mips32-inl.h @@ -0,0 +1,960 @@ +/* -*- 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_mips32_MacroAssembler_mips32_inl_h +#define jit_mips32_MacroAssembler_mips32_inl_h + +#include "jit/mips32/MacroAssembler-mips32.h" + +#include "vm/BigIntType.h" // JS::BigInt + +#include "jit/mips-shared/MacroAssembler-mips-shared-inl.h" + +namespace js { +namespace jit { + +//{{{ check_macroassembler_style + +void MacroAssembler::move64(Register64 src, Register64 dest) { + move32(src.low, dest.low); + move32(src.high, dest.high); +} + +void MacroAssembler::move64(Imm64 imm, Register64 dest) { + move32(Imm32(imm.value & 0xFFFFFFFFL), dest.low); + move32(Imm32((imm.value >> 32) & 0xFFFFFFFFL), dest.high); +} + +void MacroAssembler::moveDoubleToGPR64(FloatRegister src, Register64 dest) { + moveFromDoubleHi(src, dest.high); + moveFromDoubleLo(src, dest.low); +} + +void MacroAssembler::moveGPR64ToDouble(Register64 src, FloatRegister dest) { + moveToDoubleHi(src.high, dest); + moveToDoubleLo(src.low, dest); +} + +void MacroAssembler::move64To32(Register64 src, Register dest) { + if (src.low != dest) { + move32(src.low, dest); + } +} + +void MacroAssembler::move32To64ZeroExtend(Register src, Register64 dest) { + if (src != dest.low) { + move32(src, dest.low); + } + move32(Imm32(0), dest.high); +} + +void MacroAssembler::move8To64SignExtend(Register src, Register64 dest) { + move8SignExtend(src, dest.low); + move32To64SignExtend(dest.low, dest); +} + +void MacroAssembler::move16To64SignExtend(Register src, Register64 dest) { + move16SignExtend(src, dest.low); + move32To64SignExtend(dest.low, dest); +} + +void MacroAssembler::move32To64SignExtend(Register src, Register64 dest) { + if (src != dest.low) { + move32(src, dest.low); + } + ma_sra(dest.high, dest.low, Imm32(31)); +} + +void MacroAssembler::move32ZeroExtendToPtr(Register src, Register dest) { + move32(src, dest); +} + +// =============================================================== +// Load instructions + +void MacroAssembler::load32SignExtendToPtr(const Address& src, Register dest) { + load32(src, dest); +} + +// =============================================================== +// Logical instructions + +void MacroAssembler::notPtr(Register reg) { ma_not(reg, reg); } + +void MacroAssembler::andPtr(Register src, Register dest) { ma_and(dest, src); } + +void MacroAssembler::andPtr(Imm32 imm, Register dest) { ma_and(dest, imm); } + +void MacroAssembler::and64(Imm64 imm, Register64 dest) { + if (imm.low().value != int32_t(0xFFFFFFFF)) { + and32(imm.low(), dest.low); + } + if (imm.hi().value != int32_t(0xFFFFFFFF)) { + and32(imm.hi(), dest.high); + } +} + +void MacroAssembler::and64(Register64 src, Register64 dest) { + and32(src.low, dest.low); + and32(src.high, dest.high); +} + +void MacroAssembler::or64(Imm64 imm, Register64 dest) { + if (imm.low().value) { + or32(imm.low(), dest.low); + } + if (imm.hi().value) { + or32(imm.hi(), dest.high); + } +} + +void MacroAssembler::xor64(Imm64 imm, Register64 dest) { + if (imm.low().value) { + xor32(imm.low(), dest.low); + } + if (imm.hi().value) { + xor32(imm.hi(), dest.high); + } +} + +void MacroAssembler::orPtr(Register src, Register dest) { ma_or(dest, src); } + +void MacroAssembler::orPtr(Imm32 imm, Register dest) { ma_or(dest, imm); } + +void MacroAssembler::or64(Register64 src, Register64 dest) { + or32(src.low, dest.low); + or32(src.high, dest.high); +} + +void MacroAssembler::xor64(Register64 src, Register64 dest) { + ma_xor(dest.low, src.low); + ma_xor(dest.high, src.high); +} + +void MacroAssembler::xorPtr(Register src, Register dest) { ma_xor(dest, src); } + +void MacroAssembler::xorPtr(Imm32 imm, Register dest) { ma_xor(dest, imm); } + +// =============================================================== +// Swap instructions + +void MacroAssembler::byteSwap64(Register64 reg) { + byteSwap32(reg.high); + byteSwap32(reg.low); + + // swap reg.high and reg.low. + ma_xor(reg.high, reg.low); + ma_xor(reg.low, reg.high); + ma_xor(reg.high, reg.low); +} + +// =============================================================== +// Arithmetic functions + +void MacroAssembler::addPtr(Register src, Register dest) { ma_addu(dest, src); } + +void MacroAssembler::addPtr(Imm32 imm, Register dest) { ma_addu(dest, imm); } + +void MacroAssembler::addPtr(ImmWord imm, Register dest) { + addPtr(Imm32(imm.value), dest); +} + +void MacroAssembler::add64(Register64 src, Register64 dest) { + if (dest.low == src.low) { + as_sltu(ScratchRegister, src.low, zero); + as_addu(dest.low, dest.low, src.low); + } else { + as_addu(dest.low, dest.low, src.low); + as_sltu(ScratchRegister, dest.low, src.low); + } + as_addu(dest.high, dest.high, src.high); + as_addu(dest.high, dest.high, ScratchRegister); +} + +void MacroAssembler::add64(Imm32 imm, Register64 dest) { + if (Imm16::IsInSignedRange(imm.value)) { + as_addiu(dest.low, dest.low, imm.value); + as_sltiu(ScratchRegister, dest.low, imm.value); + } else { + ma_li(ScratchRegister, imm); + as_addu(dest.low, dest.low, ScratchRegister); + as_sltu(ScratchRegister, dest.low, ScratchRegister); + } + as_addu(dest.high, dest.high, ScratchRegister); +} + +void MacroAssembler::add64(Imm64 imm, Register64 dest) { + add64(imm.low(), dest); + ma_addu(dest.high, dest.high, imm.hi()); +} + +CodeOffset MacroAssembler::sub32FromStackPtrWithPatch(Register dest) { + CodeOffset offset = CodeOffset(currentOffset()); + ma_liPatchable(dest, Imm32(0)); + as_subu(dest, StackPointer, dest); + return offset; +} + +void MacroAssembler::patchSub32FromStackPtr(CodeOffset offset, Imm32 imm) { + Instruction* lui = + (Instruction*)m_buffer.getInst(BufferOffset(offset.offset())); + MOZ_ASSERT(lui->extractOpcode() == ((uint32_t)op_lui >> OpcodeShift)); + MOZ_ASSERT(lui->next()->extractOpcode() == ((uint32_t)op_ori >> OpcodeShift)); + + UpdateLuiOriValue(lui, lui->next(), imm.value); +} + +void MacroAssembler::subPtr(Register src, Register dest) { + as_subu(dest, dest, src); +} + +void MacroAssembler::subPtr(Imm32 imm, Register dest) { + ma_subu(dest, dest, imm); +} + +void MacroAssembler::sub64(Register64 src, Register64 dest) { + MOZ_ASSERT(dest.low != src.high); + MOZ_ASSERT(dest.high != src.low); + MOZ_ASSERT(dest.high != src.high); + + as_sltu(ScratchRegister, dest.low, src.low); + as_subu(dest.high, dest.high, ScratchRegister); + as_subu(dest.low, dest.low, src.low); + as_subu(dest.high, dest.high, src.high); +} + +void MacroAssembler::sub64(Imm64 imm, Register64 dest) { + if (Imm16::IsInSignedRange(imm.low().value) && + Imm16::IsInSignedRange(-imm.value)) { + as_sltiu(ScratchRegister, dest.low, imm.low().value); + as_subu(dest.high, dest.high, ScratchRegister); + as_addiu(dest.low, dest.low, -imm.value); + } else { + ma_li(SecondScratchReg, imm.low()); + as_sltu(ScratchRegister, dest.low, SecondScratchReg); + as_subu(dest.high, dest.high, ScratchRegister); + as_subu(dest.low, dest.low, SecondScratchReg); + } + ma_subu(dest.high, dest.high, imm.hi()); +} + +void MacroAssembler::mul64(Imm64 imm, const Register64& dest) { + // LOW32 = LOW(LOW(dest) * LOW(imm)); + // HIGH32 = LOW(HIGH(dest) * LOW(imm)) [multiply imm into upper bits] + // + LOW(LOW(dest) * HIGH(imm)) [multiply dest into upper bits] + // + HIGH(LOW(dest) * LOW(imm)) [carry] + + if (imm.low().value == 5) { + // Optimized case for Math.random(). + as_sll(ScratchRegister, dest.low, 2); + as_srl(SecondScratchReg, dest.low, 32 - 2); + as_addu(dest.low, ScratchRegister, dest.low); + as_sltu(ScratchRegister, dest.low, ScratchRegister); + as_addu(ScratchRegister, ScratchRegister, SecondScratchReg); + as_sll(SecondScratchReg, dest.high, 2); + as_addu(SecondScratchReg, SecondScratchReg, dest.high); + as_addu(dest.high, ScratchRegister, SecondScratchReg); + } else { + // HIGH32 = LOW(HIGH(dest) * LOW(imm)) [multiply imm into upper bits] + // + LOW(LOW(dest) * HIGH(imm)) [multiply dest into upper bits] + ma_li(ScratchRegister, imm.low()); + as_mult(dest.high, ScratchRegister); + ma_li(ScratchRegister, imm.hi()); + as_madd(dest.low, ScratchRegister); + as_mflo(dest.high); + // + HIGH(LOW(dest) * LOW(imm)) [carry] + // LOW32 = LOW(LOW(dest) * LOW(imm)); + ma_li(ScratchRegister, imm.low()); + as_multu(dest.low, ScratchRegister); + as_mfhi(ScratchRegister); + as_mflo(dest.low); + as_addu(dest.high, dest.high, ScratchRegister); + } +} + +void MacroAssembler::mul64(Imm64 imm, const Register64& dest, + const Register temp) { + // LOW32 = LOW(LOW(dest) * LOW(imm)); + // HIGH32 = LOW(HIGH(dest) * LOW(imm)) [multiply imm into upper bits] + // + LOW(LOW(dest) * HIGH(imm)) [multiply dest into upper bits] + // + HIGH(LOW(dest) * LOW(imm)) [carry] + + MOZ_ASSERT(temp != dest.high && temp != dest.low); + + // HIGH32 = LOW(HIGH(dest) * LOW(imm)) [multiply imm into upper bits] + // + LOW(LOW(dest) * HIGH(imm)) [multiply dest into upper bits] + ma_li(ScratchRegister, imm.low()); + as_mult(dest.high, ScratchRegister); + ma_li(temp, imm.hi()); + as_madd(dest.low, temp); + as_mflo(dest.high); + // + HIGH(LOW(dest) * LOW(imm)) [carry] + // LOW32 = LOW(LOW(dest) * LOW(imm)); + as_multu(dest.low, ScratchRegister); + as_mfhi(ScratchRegister); + as_mflo(dest.low); + as_addu(dest.high, dest.high, ScratchRegister); +} + +void MacroAssembler::mul64(const Register64& src, const Register64& dest, + const Register temp) { + // LOW32 = LOW(LOW(dest) * LOW(imm)); + // HIGH32 = LOW(HIGH(dest) * LOW(imm)) [multiply imm into upper bits] + // + LOW(LOW(dest) * HIGH(imm)) [multiply dest into upper bits] + // + HIGH(LOW(dest) * LOW(imm)) [carry] + + MOZ_ASSERT(dest != src); + MOZ_ASSERT(dest.low != src.high && dest.high != src.low); + + // HIGH32 = LOW(HIGH(dest) * LOW(src)) [multiply src into upper bits] + // + LOW(LOW(dest) * HIGH(src)) [multiply dest into upper bits] + as_mult(dest.high, src.low); + as_madd(dest.low, src.high); + as_mflo(dest.high); + // + HIGH(LOW(dest) * LOW(src)) [carry] + // LOW32 = LOW(LOW(dest) * LOW(src)); + as_multu(dest.low, src.low); + as_mfhi(ScratchRegister); + as_mflo(dest.low); + as_addu(dest.high, dest.high, ScratchRegister); +} + +void MacroAssembler::neg64(Register64 reg) { + as_subu(ScratchRegister, zero, reg.low); + as_sltu(ScratchRegister, reg.low, ScratchRegister); + as_subu(reg.high, zero, reg.high); + as_subu(reg.high, reg.high, ScratchRegister); +} + +void MacroAssembler::negPtr(Register reg) { as_subu(reg, zero, reg); } + +void MacroAssembler::mulBy3(Register src, Register dest) { + MOZ_ASSERT(src != ScratchRegister); + as_addu(ScratchRegister, src, src); + as_addu(dest, ScratchRegister, src); +} + +void MacroAssembler::inc64(AbsoluteAddress dest) { + ma_li(ScratchRegister, Imm32((int32_t)dest.addr)); + as_lw(SecondScratchReg, ScratchRegister, 0); + + as_addiu(SecondScratchReg, SecondScratchReg, 1); + as_sw(SecondScratchReg, ScratchRegister, 0); + + as_sltiu(SecondScratchReg, SecondScratchReg, 1); + as_lw(ScratchRegister, ScratchRegister, 4); + + as_addu(SecondScratchReg, ScratchRegister, SecondScratchReg); + + ma_li(ScratchRegister, Imm32((int32_t)dest.addr)); + as_sw(SecondScratchReg, ScratchRegister, 4); +} + +// =============================================================== +// Shift functions + +void MacroAssembler::lshiftPtr(Imm32 imm, Register dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 32); + ma_sll(dest, dest, imm); +} + +void MacroAssembler::lshiftPtr(Register src, Register dest) { + ma_sll(dest, dest, src); +} + +void MacroAssembler::lshift64(Imm32 imm, Register64 dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + ScratchRegisterScope scratch(*this); + + if (imm.value == 0) { + return; + } else if (imm.value < 32) { + as_sll(dest.high, dest.high, imm.value); + as_srl(scratch, dest.low, (32 - imm.value) % 32); + as_or(dest.high, dest.high, scratch); + as_sll(dest.low, dest.low, imm.value); + } else { + as_sll(dest.high, dest.low, imm.value - 32); + move32(Imm32(0), dest.low); + } +} + +void MacroAssembler::lshift64(Register unmaskedShift, Register64 dest) { + Label done; + ScratchRegisterScope shift(*this); + + ma_and(shift, unmaskedShift, Imm32(0x3f)); + ma_b(shift, Imm32(0), &done, Equal); + + mov(dest.low, SecondScratchReg); + ma_sll(dest.low, dest.low, shift); + as_nor(shift, zero, shift); + as_srl(SecondScratchReg, SecondScratchReg, 1); + ma_srl(SecondScratchReg, SecondScratchReg, shift); + ma_and(shift, unmaskedShift, Imm32(0x3f)); + ma_sll(dest.high, dest.high, shift); + as_or(dest.high, dest.high, SecondScratchReg); + + ma_and(SecondScratchReg, shift, Imm32(0x20)); + as_movn(dest.high, dest.low, SecondScratchReg); + as_movn(dest.low, zero, SecondScratchReg); + + bind(&done); +} + +void MacroAssembler::rshiftPtr(Imm32 imm, Register dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 32); + ma_srl(dest, dest, imm); +} + +void MacroAssembler::rshiftPtrArithmetic(Imm32 imm, Register dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 32); + ma_sra(dest, dest, imm); +} + +void MacroAssembler::rshiftPtr(Register src, Register dest) { + ma_srl(dest, dest, src); +} + +void MacroAssembler::rshift64(Imm32 imm, Register64 dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + ScratchRegisterScope scratch(*this); + + if (imm.value == 0) { + return; + } else if (imm.value < 32) { + as_srl(dest.low, dest.low, imm.value); + as_sll(scratch, dest.high, (32 - imm.value) % 32); + as_or(dest.low, dest.low, scratch); + as_srl(dest.high, dest.high, imm.value); + } else if (imm.value == 32) { + ma_move(dest.low, dest.high); + move32(Imm32(0), dest.high); + } else { + ma_srl(dest.low, dest.high, Imm32(imm.value - 32)); + move32(Imm32(0), dest.high); + } +} + +void MacroAssembler::rshift64(Register unmaskedShift, Register64 dest) { + Label done; + ScratchRegisterScope shift(*this); + + ma_and(shift, unmaskedShift, Imm32(0x3f)); + ma_b(shift, Imm32(0), &done, Equal); + + mov(dest.high, SecondScratchReg); + ma_srl(dest.high, dest.high, shift); + as_nor(shift, zero, shift); + as_sll(SecondScratchReg, SecondScratchReg, 1); + ma_sll(SecondScratchReg, SecondScratchReg, shift); + ma_and(shift, unmaskedShift, Imm32(0x3f)); + ma_srl(dest.low, dest.low, shift); + as_or(dest.low, dest.low, SecondScratchReg); + ma_and(SecondScratchReg, shift, Imm32(0x20)); + as_movn(dest.low, dest.high, SecondScratchReg); + as_movn(dest.high, zero, SecondScratchReg); + + bind(&done); +} + +void MacroAssembler::rshift64Arithmetic(Imm32 imm, Register64 dest) { + MOZ_ASSERT(0 <= imm.value && imm.value < 64); + ScratchRegisterScope scratch(*this); + + if (imm.value == 0) { + return; + } else if (imm.value < 32) { + as_srl(dest.low, dest.low, imm.value); + as_sll(scratch, dest.high, (32 - imm.value) % 32); + as_or(dest.low, dest.low, scratch); + as_sra(dest.high, dest.high, imm.value); + } else if (imm.value == 32) { + ma_move(dest.low, dest.high); + as_sra(dest.high, dest.high, 31); + } else { + as_sra(dest.low, dest.high, imm.value - 32); + as_sra(dest.high, dest.high, 31); + } +} + +void MacroAssembler::rshift64Arithmetic(Register unmaskedShift, + Register64 dest) { + Label done; + + ScratchRegisterScope shift(*this); + ma_and(shift, unmaskedShift, Imm32(0x3f)); + ma_b(shift, Imm32(0), &done, Equal); + + mov(dest.high, SecondScratchReg); + ma_sra(dest.high, dest.high, shift); + as_nor(shift, zero, shift); + as_sll(SecondScratchReg, SecondScratchReg, 1); + ma_sll(SecondScratchReg, SecondScratchReg, shift); + ma_and(shift, unmaskedShift, Imm32(0x3f)); + ma_srl(dest.low, dest.low, shift); + as_or(dest.low, dest.low, SecondScratchReg); + ma_and(SecondScratchReg, shift, Imm32(0x20)); + as_sra(shift, dest.high, 31); + as_movn(dest.low, dest.high, SecondScratchReg); + as_movn(dest.high, shift, SecondScratchReg); + + bind(&done); +} + +// =============================================================== +// Rotation functions + +void MacroAssembler::rotateLeft64(Imm32 count, Register64 input, + Register64 dest, Register temp) { + MOZ_ASSERT(temp == InvalidReg); + MOZ_ASSERT(input.low != dest.high && input.high != dest.low); + + int32_t amount = count.value & 0x3f; + if (amount > 32) { + rotateRight64(Imm32(64 - amount), input, dest, temp); + } else { + ScratchRegisterScope scratch(*this); + if (amount == 0) { + ma_move(dest.low, input.low); + ma_move(dest.high, input.high); + } else if (amount == 32) { + ma_move(scratch, input.low); + ma_move(dest.low, input.high); + ma_move(dest.high, scratch); + } else { + MOZ_ASSERT(0 < amount && amount < 32); + ma_move(scratch, input.high); + ma_sll(dest.high, input.high, Imm32(amount)); + ma_srl(SecondScratchReg, input.low, Imm32(32 - amount)); + as_or(dest.high, dest.high, SecondScratchReg); + ma_sll(dest.low, input.low, Imm32(amount)); + ma_srl(SecondScratchReg, scratch, Imm32(32 - amount)); + as_or(dest.low, dest.low, SecondScratchReg); + } + } +} + +void MacroAssembler::rotateLeft64(Register shift, Register64 src, + Register64 dest, Register temp) { + MOZ_ASSERT(temp != src.low && temp != src.high); + MOZ_ASSERT(shift != src.low && shift != src.high); + MOZ_ASSERT(temp != InvalidReg); + + ScratchRegisterScope scratch(*this); + + ma_and(scratch, shift, Imm32(0x3f)); + as_nor(SecondScratchReg, zero, scratch); + ma_sll(temp, src.low, scratch); + ma_move(scratch, src.low); + as_srl(dest.low, src.high, 1); + ma_srl(dest.low, dest.low, SecondScratchReg); + as_or(dest.low, dest.low, temp); + ma_move(SecondScratchReg, src.high); + as_srl(dest.high, scratch, 1); + ma_and(scratch, shift, Imm32(0x3f)); + ma_sll(temp, SecondScratchReg, scratch); + as_nor(SecondScratchReg, zero, scratch); + ma_srl(dest.high, dest.high, SecondScratchReg); + as_or(dest.high, dest.high, temp); + ma_and(temp, scratch, Imm32(32)); + as_movn(SecondScratchReg, dest.high, temp); + as_movn(dest.high, dest.low, temp); + as_movn(dest.low, SecondScratchReg, temp); +} + +void MacroAssembler::rotateRight64(Imm32 count, Register64 input, + Register64 dest, Register temp) { + MOZ_ASSERT(temp == InvalidReg); + MOZ_ASSERT(input.low != dest.high && input.high != dest.low); + + int32_t amount = count.value & 0x3f; + if (amount > 32) { + rotateLeft64(Imm32(64 - amount), input, dest, temp); + } else { + ScratchRegisterScope scratch(*this); + if (amount == 0) { + ma_move(dest.low, input.low); + ma_move(dest.high, input.high); + } else if (amount == 32) { + ma_move(scratch, input.low); + ma_move(dest.low, input.high); + ma_move(dest.high, scratch); + } else { + MOZ_ASSERT(0 < amount && amount < 32); + ma_move(scratch, input.high); + ma_srl(dest.high, input.high, Imm32(amount)); + ma_sll(SecondScratchReg, input.low, Imm32(32 - amount)); + as_or(dest.high, dest.high, SecondScratchReg); + ma_srl(dest.low, input.low, Imm32(amount)); + ma_sll(SecondScratchReg, scratch, Imm32(32 - amount)); + as_or(dest.low, dest.low, SecondScratchReg); + } + } +} + +void MacroAssembler::rotateRight64(Register shift, Register64 src, + Register64 dest, Register temp) { + MOZ_ASSERT(temp != src.low && temp != src.high); + MOZ_ASSERT(shift != src.low && shift != src.high); + MOZ_ASSERT(temp != InvalidReg); + + ScratchRegisterScope scratch(*this); + + ma_and(scratch, shift, Imm32(0x3f)); + as_nor(SecondScratchReg, zero, scratch); + ma_srl(temp, src.low, scratch); + ma_move(scratch, src.low); + as_sll(dest.low, src.high, 1); + ma_sll(dest.low, dest.low, SecondScratchReg); + as_or(dest.low, dest.low, temp); + ma_move(SecondScratchReg, src.high); + as_sll(dest.high, scratch, 1); + ma_and(scratch, shift, Imm32(0x3f)); + ma_srl(temp, SecondScratchReg, scratch); + as_nor(SecondScratchReg, zero, scratch); + ma_sll(dest.high, dest.high, SecondScratchReg); + as_or(dest.high, dest.high, temp); + ma_and(temp, scratch, Imm32(32)); + as_movn(SecondScratchReg, dest.high, temp); + as_movn(dest.high, dest.low, temp); + as_movn(dest.low, SecondScratchReg, temp); +} + +template <typename T1, typename T2> +void MacroAssembler::cmpPtrSet(Condition cond, T1 lhs, T2 rhs, Register dest) { + ma_cmp_set(dest, lhs, rhs, cond); +} + +template <typename T1, typename T2> +void MacroAssembler::cmp32Set(Condition cond, T1 lhs, T2 rhs, Register dest) { + ma_cmp_set(dest, lhs, rhs, cond); +} + +// =============================================================== +// Bit counting functions + +void MacroAssembler::clz64(Register64 src, Register dest) { + as_clz(ScratchRegister, src.high); + as_clz(SecondScratchReg, src.low); + as_movn(SecondScratchReg, zero, src.high); + as_addu(dest, ScratchRegister, SecondScratchReg); +} + +void MacroAssembler::ctz64(Register64 src, Register dest) { + as_movz(SecondScratchReg, src.high, src.low); + as_movn(SecondScratchReg, src.low, src.low); + ma_ctz(SecondScratchReg, SecondScratchReg); + ma_li(ScratchRegister, Imm32(0x20)); + as_movn(ScratchRegister, zero, src.low); + as_addu(dest, SecondScratchReg, ScratchRegister); +} + +void MacroAssembler::popcnt64(Register64 src, Register64 dest, Register tmp) { + MOZ_ASSERT(dest.low != tmp); + MOZ_ASSERT(dest.high != tmp); + MOZ_ASSERT(dest.low != dest.high); + + as_srl(tmp, src.low, 1); + as_srl(SecondScratchReg, src.high, 1); + ma_li(ScratchRegister, Imm32(0x55555555)); + as_and(tmp, tmp, ScratchRegister); + as_subu(tmp, src.low, tmp); + as_and(SecondScratchReg, SecondScratchReg, ScratchRegister); + as_subu(SecondScratchReg, src.high, SecondScratchReg); + ma_li(ScratchRegister, Imm32(0x33333333)); + as_and(dest.low, tmp, ScratchRegister); + as_srl(tmp, tmp, 2); + as_and(tmp, tmp, ScratchRegister); + as_addu(tmp, dest.low, tmp); + as_and(dest.high, SecondScratchReg, ScratchRegister); + as_srl(SecondScratchReg, SecondScratchReg, 2); + as_and(SecondScratchReg, SecondScratchReg, ScratchRegister); + as_addu(SecondScratchReg, dest.high, SecondScratchReg); + ma_li(ScratchRegister, Imm32(0x0F0F0F0F)); + as_addu(tmp, SecondScratchReg, tmp); + as_srl(dest.low, tmp, 4); + as_and(dest.low, dest.low, ScratchRegister); + as_and(tmp, tmp, ScratchRegister); + as_addu(dest.low, dest.low, tmp); + ma_mul(dest.low, dest.low, Imm32(0x01010101)); + as_srl(dest.low, dest.low, 24); + ma_move(dest.high, zero); +} + +// =============================================================== +// Branch functions + +void MacroAssembler::branch64(Condition cond, const Address& lhs, Imm64 val, + Label* label) { + MOZ_ASSERT(cond == Assembler::NotEqual, + "other condition codes not supported"); + + branch32(cond, lhs, val.firstHalf(), label); + branch32(cond, Address(lhs.base, lhs.offset + sizeof(uint32_t)), + val.secondHalf(), label); +} + +void MacroAssembler::branch64(Condition cond, const Address& lhs, + const Address& rhs, Register scratch, + Label* label) { + MOZ_ASSERT(cond == Assembler::NotEqual, + "other condition codes not supported"); + MOZ_ASSERT(lhs.base != scratch); + MOZ_ASSERT(rhs.base != scratch); + + load32(rhs, scratch); + branch32(cond, lhs, scratch, label); + + load32(Address(rhs.base, rhs.offset + sizeof(uint32_t)), scratch); + branch32(cond, Address(lhs.base, lhs.offset + sizeof(uint32_t)), scratch, + label); +} + +void MacroAssembler::branch64(Condition cond, Register64 lhs, Imm64 val, + Label* success, Label* fail) { + if (val.value == 0) { + switch (cond) { + case Assembler::Equal: + case Assembler::BelowOrEqual: + case Assembler::NotEqual: + case Assembler::Above: + as_or(ScratchRegister, lhs.high, lhs.low); + ma_b(ScratchRegister, ScratchRegister, success, + (cond == Assembler::Equal || cond == Assembler::BelowOrEqual) + ? Assembler::Zero + : Assembler::NonZero); + break; + case Assembler::LessThan: + case Assembler::GreaterThanOrEqual: + ma_b(lhs.high, Imm32(0), success, cond); + break; + case Assembler::LessThanOrEqual: + case Assembler::GreaterThan: + as_or(SecondScratchReg, lhs.high, lhs.low); + as_sra(ScratchRegister, lhs.high, 31); + as_sltu(ScratchRegister, ScratchRegister, SecondScratchReg); + ma_b(ScratchRegister, ScratchRegister, success, + (cond == Assembler::LessThanOrEqual) ? Assembler::Zero + : Assembler::NonZero); + break; + case Assembler::Below: + // This condition is always false. No branch required. + break; + case Assembler::AboveOrEqual: + ma_b(success); + break; + default: + MOZ_CRASH("Condition code not supported"); + } + return; + } + + Condition c = ma_cmp64(cond, lhs, val, SecondScratchReg); + ma_b(SecondScratchReg, SecondScratchReg, success, c); + if (fail) { + jump(fail); + } +} + +void MacroAssembler::branch64(Condition cond, Register64 lhs, Register64 rhs, + Label* success, Label* fail) { + Condition c = ma_cmp64(cond, lhs, rhs, SecondScratchReg); + ma_b(SecondScratchReg, SecondScratchReg, success, c); + if (fail) { + jump(fail); + } +} + +void MacroAssembler::branchPrivatePtr(Condition cond, const Address& lhs, + Register rhs, Label* label) { + branchPtr(cond, lhs, rhs, label); +} + +template <class L> +void MacroAssembler::branchTest64(Condition cond, Register64 lhs, + Register64 rhs, Register temp, L label) { + if (cond == Assembler::Zero || cond == Assembler::NonZero) { + MOZ_ASSERT(lhs.low == rhs.low); + MOZ_ASSERT(lhs.high == rhs.high); + as_or(ScratchRegister, lhs.low, lhs.high); + ma_b(ScratchRegister, ScratchRegister, label, cond); + } else if (cond == Assembler::Signed || cond == Assembler::NotSigned) { + branchTest32(cond, lhs.high, rhs.high, label); + } else { + MOZ_CRASH("Unsupported condition"); + } +} + +void MacroAssembler::branchTestUndefined(Condition cond, + const ValueOperand& value, + Label* label) { + branchTestUndefined(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestInt32(Condition cond, const ValueOperand& value, + Label* label) { + branchTestInt32(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestInt32Truthy(bool b, const ValueOperand& value, + Label* label) { + ScratchRegisterScope scratch(*this); + as_and(scratch, value.payloadReg(), value.payloadReg()); + ma_b(scratch, scratch, label, b ? NonZero : Zero); +} + +void MacroAssembler::branchTestDouble(Condition cond, Register tag, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + Condition actual = (cond == Equal) ? Below : AboveOrEqual; + ma_b(tag, ImmTag(JSVAL_TAG_CLEAR), label, actual); +} + +void MacroAssembler::branchTestDouble(Condition cond, const ValueOperand& value, + Label* label) { + branchTestDouble(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestNumber(Condition cond, const ValueOperand& value, + Label* label) { + branchTestNumber(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestBoolean(Condition cond, + const ValueOperand& value, + Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_b(value.typeReg(), ImmType(JSVAL_TYPE_BOOLEAN), label, cond); +} + +void MacroAssembler::branchTestBooleanTruthy(bool b, const ValueOperand& value, + Label* label) { + ma_b(value.payloadReg(), value.payloadReg(), label, b ? NonZero : Zero); +} + +void MacroAssembler::branchTestString(Condition cond, const ValueOperand& value, + Label* label) { + branchTestString(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestStringTruthy(bool b, const ValueOperand& value, + Label* label) { + Register string = value.payloadReg(); + SecondScratchRegisterScope scratch2(*this); + ma_lw(scratch2, Address(string, JSString::offsetOfLength())); + ma_b(scratch2, Imm32(0), label, b ? NotEqual : Equal); +} + +void MacroAssembler::branchTestSymbol(Condition cond, const ValueOperand& value, + Label* label) { + branchTestSymbol(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestBigInt(Condition cond, const BaseIndex& address, + Label* label) { + SecondScratchRegisterScope scratch2(*this); + Register tag = extractTag(address, scratch2); + branchTestBigInt(cond, tag, label); +} + +void MacroAssembler::branchTestBigInt(Condition cond, const ValueOperand& value, + Label* label) { + branchTestBigInt(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestBigIntTruthy(bool b, const ValueOperand& value, + Label* label) { + Register bi = value.payloadReg(); + SecondScratchRegisterScope scratch2(*this); + ma_lw(scratch2, Address(bi, BigInt::offsetOfDigitLength())); + ma_b(scratch2, Imm32(0), label, b ? NotEqual : Equal); +} + +void MacroAssembler::branchTestNull(Condition cond, const ValueOperand& value, + Label* label) { + branchTestNull(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestObject(Condition cond, const ValueOperand& value, + Label* label) { + branchTestObject(cond, value.typeReg(), label); +} + +void MacroAssembler::branchTestPrimitive(Condition cond, + const ValueOperand& value, + Label* label) { + branchTestPrimitive(cond, value.typeReg(), label); +} + +template <class L> +void MacroAssembler::branchTestMagic(Condition cond, const ValueOperand& value, + L label) { + ma_b(value.typeReg(), ImmTag(JSVAL_TAG_MAGIC), label, cond); +} + +void MacroAssembler::branchTestMagic(Condition cond, const Address& valaddr, + JSWhyMagic why, Label* label) { + MOZ_ASSERT(cond == Assembler::Equal || cond == Assembler::NotEqual); + + Label notMagic; + if (cond == Assembler::Equal) { + branchTestMagic(Assembler::NotEqual, valaddr, ¬Magic); + } else { + branchTestMagic(Assembler::NotEqual, valaddr, label); + } + + branch32(cond, ToPayload(valaddr), Imm32(why), label); + bind(¬Magic); +} + +void MacroAssembler::branchTruncateDoubleMaybeModUint32(FloatRegister src, + Register dest, + Label* fail) { + as_truncwd(ScratchFloat32Reg, src); + as_cfc1(ScratchRegister, Assembler::FCSR); + moveFromFloat32(ScratchFloat32Reg, dest); + ma_ext(ScratchRegister, ScratchRegister, Assembler::CauseV, 1); + ma_b(ScratchRegister, Imm32(0), fail, Assembler::NotEqual); +} + +void MacroAssembler::branchTruncateFloat32MaybeModUint32(FloatRegister src, + Register dest, + Label* fail) { + as_truncws(ScratchFloat32Reg, src); + as_cfc1(ScratchRegister, Assembler::FCSR); + moveFromFloat32(ScratchFloat32Reg, dest); + ma_ext(ScratchRegister, ScratchRegister, Assembler::CauseV, 1); + ma_b(ScratchRegister, Imm32(0), fail, Assembler::NotEqual); +} + +//}}} check_macroassembler_style +// =============================================================== + +void MacroAssemblerMIPSCompat::incrementInt32Value(const Address& addr) { + asMasm().add32(Imm32(1), ToPayload(addr)); +} + +void MacroAssemblerMIPSCompat::computeEffectiveAddress(const BaseIndex& address, + Register dest) { + computeScaledAddress(address, dest); + if (address.offset) { + asMasm().addPtr(Imm32(address.offset), dest); + } +} + +void MacroAssemblerMIPSCompat::retn(Imm32 n) { + // pc <- [sp]; sp += n + loadPtr(Address(StackPointer, 0), ra); + asMasm().addPtr(n, StackPointer); + as_jr(ra); + as_nop(); +} + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_MacroAssembler_mips32_inl_h */ diff --git a/js/src/jit/mips32/MacroAssembler-mips32.cpp b/js/src/jit/mips32/MacroAssembler-mips32.cpp new file mode 100644 index 0000000000..6a07fff2c1 --- /dev/null +++ b/js/src/jit/mips32/MacroAssembler-mips32.cpp @@ -0,0 +1,2787 @@ +/* -*- 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/mips32/MacroAssembler-mips32.h" + +#include "mozilla/DebugOnly.h" +#include "mozilla/EndianUtils.h" +#include "mozilla/MathAlgorithms.h" + +#include "jit/Bailouts.h" +#include "jit/BaselineFrame.h" +#include "jit/JitFrames.h" +#include "jit/JitRuntime.h" +#include "jit/MacroAssembler.h" +#include "jit/mips32/Simulator-mips32.h" +#include "jit/MoveEmitter.h" +#include "jit/SharedICRegisters.h" +#include "util/Memory.h" +#include "vm/JitActivation.h" // js::jit::JitActivation +#include "vm/JSContext.h" + +#include "jit/MacroAssembler-inl.h" + +using namespace js; +using namespace jit; + +using mozilla::Abs; + +static const int32_t PAYLOAD_OFFSET = NUNBOX32_PAYLOAD_OFFSET; +static const int32_t TAG_OFFSET = NUNBOX32_TYPE_OFFSET; + +static_assert(sizeof(intptr_t) == 4, "Not 64-bit clean."); + +void MacroAssemblerMIPSCompat::convertBoolToInt32(Register src, Register dest) { + // Note that C++ bool is only 1 byte, so zero extend it to clear the + // higher-order bits. + ma_and(dest, src, Imm32(0xff)); +} + +void MacroAssemblerMIPSCompat::convertInt32ToDouble(Register src, + FloatRegister dest) { + as_mtc1(src, dest); + as_cvtdw(dest, dest); +} + +void MacroAssemblerMIPSCompat::convertInt32ToDouble(const Address& src, + FloatRegister dest) { + ma_ls(dest, src); + as_cvtdw(dest, dest); +} + +void MacroAssemblerMIPSCompat::convertInt32ToDouble(const BaseIndex& src, + FloatRegister dest) { + computeScaledAddress(src, ScratchRegister); + convertInt32ToDouble(Address(ScratchRegister, src.offset), dest); +} + +void MacroAssemblerMIPSCompat::convertUInt32ToDouble(Register src, + FloatRegister dest) { + Label positive, done; + ma_b(src, src, &positive, NotSigned, ShortJump); + + const uint32_t kExponentShift = + mozilla::FloatingPoint<double>::kExponentShift - 32; + const uint32_t kExponent = + (31 + mozilla::FloatingPoint<double>::kExponentBias); + + ma_ext(SecondScratchReg, src, 31 - kExponentShift, kExponentShift); + ma_li(ScratchRegister, Imm32(kExponent << kExponentShift)); + ma_or(SecondScratchReg, ScratchRegister); + ma_sll(ScratchRegister, src, Imm32(kExponentShift + 1)); + moveToDoubleHi(SecondScratchReg, dest); + moveToDoubleLo(ScratchRegister, dest); + + ma_b(&done, ShortJump); + + bind(&positive); + convertInt32ToDouble(src, dest); + + bind(&done); +} + +void MacroAssemblerMIPSCompat::convertUInt32ToFloat32(Register src, + FloatRegister dest) { + Label positive, done; + ma_b(src, src, &positive, NotSigned, ShortJump); + + const uint32_t kExponentShift = + mozilla::FloatingPoint<double>::kExponentShift - 32; + const uint32_t kExponent = + (31 + mozilla::FloatingPoint<double>::kExponentBias); + + ma_ext(SecondScratchReg, src, 31 - kExponentShift, kExponentShift); + ma_li(ScratchRegister, Imm32(kExponent << kExponentShift)); + ma_or(SecondScratchReg, ScratchRegister); + ma_sll(ScratchRegister, src, Imm32(kExponentShift + 1)); + FloatRegister destDouble = dest.asDouble(); + moveToDoubleHi(SecondScratchReg, destDouble); + moveToDoubleLo(ScratchRegister, destDouble); + + convertDoubleToFloat32(destDouble, dest); + + ma_b(&done, ShortJump); + + bind(&positive); + convertInt32ToFloat32(src, dest); + + bind(&done); +} + +void MacroAssemblerMIPSCompat::convertDoubleToFloat32(FloatRegister src, + FloatRegister dest) { + as_cvtsd(dest, src); +} + +const int CauseBitPos = int(Assembler::CauseI); +const int CauseBitCount = 1 + int(Assembler::CauseV) - int(Assembler::CauseI); +const int CauseIOrVMask = ((1 << int(Assembler::CauseI)) | + (1 << int(Assembler::CauseV))) >> + int(Assembler::CauseI); + +// 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 MacroAssemblerMIPSCompat::convertDoubleToInt32(FloatRegister src, + Register dest, Label* fail, + bool negativeZeroCheck) { + if (negativeZeroCheck) { + moveFromDoubleHi(src, dest); + moveFromDoubleLo(src, SecondScratchReg); + ma_xor(dest, Imm32(INT32_MIN)); + ma_or(dest, SecondScratchReg); + ma_b(dest, Imm32(0), fail, Assembler::Equal); + } + + // Truncate double to int ; if result is inexact or invalid fail. + as_truncwd(ScratchFloat32Reg, src); + as_cfc1(ScratchRegister, Assembler::FCSR); + moveFromFloat32(ScratchFloat32Reg, dest); + ma_ext(ScratchRegister, ScratchRegister, CauseBitPos, CauseBitCount); + // Here adding the masking andi instruction just for a precaution. + // For the instruction of trunc.*.*, the Floating Point Exceptions can be + // only Inexact, Invalid Operation, Unimplemented Operation. + // Leaving it maybe is also ok. + as_andi(ScratchRegister, ScratchRegister, CauseIOrVMask); + ma_b(ScratchRegister, Imm32(0), fail, Assembler::NotEqual); +} + +// 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 MacroAssemblerMIPSCompat::convertFloat32ToInt32(FloatRegister src, + Register dest, Label* fail, + bool negativeZeroCheck) { + if (negativeZeroCheck) { + moveFromFloat32(src, dest); + ma_b(dest, Imm32(INT32_MIN), fail, Assembler::Equal); + } + + as_truncws(ScratchFloat32Reg, src); + as_cfc1(ScratchRegister, Assembler::FCSR); + moveFromFloat32(ScratchFloat32Reg, dest); + ma_ext(ScratchRegister, ScratchRegister, CauseBitPos, CauseBitCount); + as_andi(ScratchRegister, ScratchRegister, CauseIOrVMask); + ma_b(ScratchRegister, Imm32(0), fail, Assembler::NotEqual); +} + +void MacroAssemblerMIPSCompat::convertFloat32ToDouble(FloatRegister src, + FloatRegister dest) { + as_cvtds(dest, src); +} + +void MacroAssemblerMIPSCompat::convertInt32ToFloat32(Register src, + FloatRegister dest) { + as_mtc1(src, dest); + as_cvtsw(dest, dest); +} + +void MacroAssemblerMIPSCompat::convertInt32ToFloat32(const Address& src, + FloatRegister dest) { + ma_ls(dest, src); + as_cvtsw(dest, dest); +} + +void MacroAssemblerMIPS::ma_li(Register dest, CodeLabel* label) { + BufferOffset bo = m_buffer.nextOffset(); + ma_liPatchable(dest, ImmWord(/* placeholder */ 0)); + label->patchAt()->bind(bo.getOffset()); + label->setLinkMode(CodeLabel::MoveImmediate); +} + +void MacroAssemblerMIPS::ma_li(Register dest, ImmWord imm) { + ma_li(dest, Imm32(uint32_t(imm.value))); +} + +void MacroAssemblerMIPS::ma_liPatchable(Register dest, ImmPtr imm) { + ma_liPatchable(dest, ImmWord(uintptr_t(imm.value))); +} + +void MacroAssemblerMIPS::ma_liPatchable(Register dest, ImmWord imm) { + ma_liPatchable(dest, Imm32(int32_t(imm.value))); +} + +// Arithmetic-based ops. + +// Add. +void MacroAssemblerMIPS::ma_addTestOverflow(Register rd, Register rs, + Register rt, Label* overflow) { + MOZ_ASSERT_IF(rs == rd, rs != rt); + MOZ_ASSERT(rs != ScratchRegister); + MOZ_ASSERT(rt != ScratchRegister); + MOZ_ASSERT(rd != rt); + MOZ_ASSERT(rd != ScratchRegister); + MOZ_ASSERT(rd != SecondScratchReg); + + if (rs == rt) { + as_addu(rd, rs, rs); + as_xor(SecondScratchReg, rs, rd); + ma_b(SecondScratchReg, Imm32(0), overflow, Assembler::LessThan); + return; + } + + // If different sign, no overflow + as_xor(ScratchRegister, rs, rt); + + as_addu(rd, rs, rt); + as_nor(ScratchRegister, ScratchRegister, zero); + // If different sign, then overflow + as_xor(SecondScratchReg, rt, rd); + as_and(SecondScratchReg, SecondScratchReg, ScratchRegister); + ma_b(SecondScratchReg, Imm32(0), overflow, Assembler::LessThan); +} + +void MacroAssemblerMIPS::ma_addTestOverflow(Register rd, Register rs, Imm32 imm, + Label* overflow) { + MOZ_ASSERT(rs != ScratchRegister); + MOZ_ASSERT(rs != SecondScratchReg); + MOZ_ASSERT(rd != ScratchRegister); + MOZ_ASSERT(rd != SecondScratchReg); + + Register rs_copy = rs; + + if (imm.value > 0) { + as_nor(ScratchRegister, rs, zero); + } else if (rs == rd) { + ma_move(ScratchRegister, rs); + rs_copy = ScratchRegister; + } + + if (Imm16::IsInSignedRange(imm.value)) { + as_addiu(rd, rs, imm.value); + } else { + ma_li(SecondScratchReg, imm); + as_addu(rd, rs, SecondScratchReg); + } + + if (imm.value > 0) { + as_and(ScratchRegister, ScratchRegister, rd); + } else { + as_nor(SecondScratchReg, rd, zero); + as_and(ScratchRegister, rs_copy, SecondScratchReg); + } + + ma_b(ScratchRegister, Imm32(0), overflow, Assembler::LessThan); +} + +// Subtract. +void MacroAssemblerMIPS::ma_subTestOverflow(Register rd, Register rs, + Register rt, Label* overflow) { + // The rs == rt case should probably be folded at MIR stage. + // Happens for Number_isInteger*. Not worth specializing here. + MOZ_ASSERT_IF(rs == rd, rs != rt); + MOZ_ASSERT(rs != SecondScratchReg); + MOZ_ASSERT(rt != SecondScratchReg); + MOZ_ASSERT(rd != rt); + MOZ_ASSERT(rd != ScratchRegister); + MOZ_ASSERT(rd != SecondScratchReg); + + Register rs_copy = rs; + + if (rs == rd) { + ma_move(SecondScratchReg, rs); + rs_copy = SecondScratchReg; + } + + as_subu(rd, rs, rt); + // If same sign, no overflow + as_xor(ScratchRegister, rs_copy, rt); + // If different sign, then overflow + as_xor(SecondScratchReg, rs_copy, rd); + as_and(SecondScratchReg, SecondScratchReg, ScratchRegister); + ma_b(SecondScratchReg, Imm32(0), overflow, Assembler::LessThan); +} + +// Memory. + +void MacroAssemblerMIPS::ma_load(Register dest, Address address, + LoadStoreSize size, + LoadStoreExtension extension) { + int16_t encodedOffset; + Register base; + + if (isLoongson() && ZeroExtend != extension && + !Imm16::IsInSignedRange(address.offset)) { + ma_li(ScratchRegister, Imm32(address.offset)); + base = address.base; + + switch (size) { + case SizeByte: + as_gslbx(dest, base, ScratchRegister, 0); + break; + case SizeHalfWord: + as_gslhx(dest, base, ScratchRegister, 0); + break; + case SizeWord: + as_gslwx(dest, base, ScratchRegister, 0); + break; + case SizeDouble: + as_gsldx(dest, base, ScratchRegister, 0); + break; + default: + MOZ_CRASH("Invalid argument for ma_load"); + } + return; + } + + if (!Imm16::IsInSignedRange(address.offset)) { + ma_li(ScratchRegister, Imm32(address.offset)); + as_addu(ScratchRegister, address.base, ScratchRegister); + base = ScratchRegister; + encodedOffset = Imm16(0).encode(); + } else { + encodedOffset = Imm16(address.offset).encode(); + base = address.base; + } + + switch (size) { + case SizeByte: + if (ZeroExtend == extension) { + as_lbu(dest, base, encodedOffset); + } else { + as_lb(dest, base, encodedOffset); + } + break; + case SizeHalfWord: + if (ZeroExtend == extension) { + as_lhu(dest, base, encodedOffset); + } else { + as_lh(dest, base, encodedOffset); + } + break; + case SizeWord: + as_lw(dest, base, encodedOffset); + break; + default: + MOZ_CRASH("Invalid argument for ma_load"); + } +} + +void MacroAssemblerMIPS::ma_store(Register data, Address address, + LoadStoreSize size, + LoadStoreExtension extension) { + int16_t encodedOffset; + Register base; + + if (isLoongson() && !Imm16::IsInSignedRange(address.offset)) { + ma_li(ScratchRegister, Imm32(address.offset)); + base = address.base; + + switch (size) { + case SizeByte: + as_gssbx(data, base, ScratchRegister, 0); + break; + case SizeHalfWord: + as_gsshx(data, base, ScratchRegister, 0); + break; + case SizeWord: + as_gsswx(data, base, ScratchRegister, 0); + break; + case SizeDouble: + as_gssdx(data, base, ScratchRegister, 0); + break; + default: + MOZ_CRASH("Invalid argument for ma_store"); + } + return; + } + + if (!Imm16::IsInSignedRange(address.offset)) { + ma_li(ScratchRegister, Imm32(address.offset)); + as_addu(ScratchRegister, address.base, ScratchRegister); + base = ScratchRegister; + encodedOffset = Imm16(0).encode(); + } else { + encodedOffset = Imm16(address.offset).encode(); + base = address.base; + } + + switch (size) { + case SizeByte: + as_sb(data, base, encodedOffset); + break; + case SizeHalfWord: + as_sh(data, base, encodedOffset); + break; + case SizeWord: + as_sw(data, base, encodedOffset); + break; + default: + MOZ_CRASH("Invalid argument for ma_store"); + } +} + +void MacroAssemblerMIPSCompat::computeScaledAddress(const BaseIndex& address, + Register dest) { + int32_t shift = Imm32::ShiftOf(address.scale).value; + if (shift) { + ma_sll(ScratchRegister, address.index, Imm32(shift)); + as_addu(dest, address.base, ScratchRegister); + } else { + as_addu(dest, address.base, address.index); + } +} + +// Shortcut for when we know we're transferring 32 bits of data. +void MacroAssemblerMIPS::ma_lw(Register data, Address address) { + ma_load(data, address, SizeWord); +} + +void MacroAssemblerMIPS::ma_sw(Register data, Address address) { + ma_store(data, address, SizeWord); +} + +void MacroAssemblerMIPS::ma_sw(Imm32 imm, Address address) { + MOZ_ASSERT(address.base != ScratchRegister); + ma_li(ScratchRegister, imm); + + if (Imm16::IsInSignedRange(address.offset)) { + as_sw(ScratchRegister, address.base, address.offset); + } else { + MOZ_ASSERT(address.base != SecondScratchReg); + + ma_li(SecondScratchReg, Imm32(address.offset)); + as_addu(SecondScratchReg, address.base, SecondScratchReg); + as_sw(ScratchRegister, SecondScratchReg, 0); + } +} + +void MacroAssemblerMIPS::ma_sw(Register data, BaseIndex& address) { + ma_store(data, address, SizeWord); +} + +void MacroAssemblerMIPS::ma_pop(Register r) { + as_lw(r, StackPointer, 0); + as_addiu(StackPointer, StackPointer, sizeof(intptr_t)); +} + +void MacroAssemblerMIPS::ma_push(Register r) { + if (r == sp) { + // Pushing sp requires one more instruction. + ma_move(ScratchRegister, sp); + r = ScratchRegister; + } + + as_addiu(StackPointer, StackPointer, -sizeof(intptr_t)); + as_sw(r, StackPointer, 0); +} + +// Branches when done from within mips-specific code. +void MacroAssemblerMIPS::ma_b(Register lhs, Address addr, Label* label, + Condition c, JumpKind jumpKind) { + MOZ_ASSERT(lhs != ScratchRegister); + ma_lw(ScratchRegister, addr); + ma_b(lhs, ScratchRegister, label, c, jumpKind); +} + +void MacroAssemblerMIPS::ma_b(Address addr, Imm32 imm, Label* label, + Condition c, JumpKind jumpKind) { + ma_lw(SecondScratchReg, addr); + ma_b(SecondScratchReg, imm, label, c, jumpKind); +} + +void MacroAssemblerMIPS::ma_b(Address addr, ImmGCPtr imm, Label* label, + Condition c, JumpKind jumpKind) { + ma_lw(SecondScratchReg, addr); + ma_b(SecondScratchReg, imm, label, c, jumpKind); +} + +void MacroAssemblerMIPS::ma_bal(Label* label, DelaySlotFill delaySlotFill) { + spew("branch .Llabel %p\n", label); + if (label->bound()) { + // Generate the long jump for calls because return address has to be + // the address after the reserved block. + addLongJump(nextOffset(), BufferOffset(label->offset())); + ma_liPatchable(ScratchRegister, Imm32(LabelBase::INVALID_OFFSET)); + as_jalr(ScratchRegister); + if (delaySlotFill == FillDelaySlot) { + as_nop(); + } + return; + } + + // Second word holds a pointer to the next branch in label's chain. + uint32_t nextInChain = + label->used() ? label->offset() : LabelBase::INVALID_OFFSET; + + // Make the whole branch continous in the buffer. + m_buffer.ensureSpace(4 * sizeof(uint32_t)); + + spew("bal .Llabel %p\n", label); + BufferOffset bo = writeInst(getBranchCode(BranchIsCall).encode()); + writeInst(nextInChain); + if (!oom()) { + label->use(bo.getOffset()); + } + // Leave space for long jump. + as_nop(); + if (delaySlotFill == FillDelaySlot) { + as_nop(); + } +} + +void MacroAssemblerMIPS::branchWithCode(InstImm code, Label* label, + JumpKind jumpKind) { + spew("branch .Llabel %p", label); + MOZ_ASSERT(code.encode() != + InstImm(op_regimm, zero, rt_bgezal, BOffImm16(0)).encode()); + InstImm inst_beq = InstImm(op_beq, zero, zero, BOffImm16(0)); + + if (label->bound()) { + int32_t offset = label->offset() - m_buffer.nextOffset().getOffset(); + + if (BOffImm16::IsInRange(offset)) { + jumpKind = ShortJump; + } + + if (jumpKind == ShortJump) { + MOZ_ASSERT(BOffImm16::IsInRange(offset)); + code.setBOffImm16(BOffImm16(offset)); +#ifdef JS_JITSPEW + decodeBranchInstAndSpew(code); +#endif + writeInst(code.encode()); + as_nop(); + return; + } + + if (code.encode() == inst_beq.encode()) { + // Handle long jump + addLongJump(nextOffset(), BufferOffset(label->offset())); + ma_liPatchable(ScratchRegister, Imm32(LabelBase::INVALID_OFFSET)); + as_jr(ScratchRegister); + as_nop(); + return; + } + + // Handle long conditional branch + spew("invert branch .Llabel %p", label); + InstImm code_r = invertBranch(code, BOffImm16(5 * sizeof(uint32_t))); +#ifdef JS_JITSPEW + decodeBranchInstAndSpew(code_r); +#endif + writeInst(code_r.encode()); + + // No need for a "nop" here because we can clobber scratch. + addLongJump(nextOffset(), BufferOffset(label->offset())); + ma_liPatchable(ScratchRegister, Imm32(LabelBase::INVALID_OFFSET)); + as_jr(ScratchRegister); + as_nop(); + return; + } + + // Generate open jump and link it to a label. + + // Second word holds a pointer to the next branch in label's chain. + uint32_t nextInChain = + label->used() ? label->offset() : LabelBase::INVALID_OFFSET; + + if (jumpKind == ShortJump) { + // Make the whole branch continous in the buffer. + m_buffer.ensureSpace(2 * sizeof(uint32_t)); + + // Indicate that this is short jump with offset 4. + code.setBOffImm16(BOffImm16(4)); +#ifdef JS_JITSPEW + decodeBranchInstAndSpew(code); +#endif + BufferOffset bo = writeInst(code.encode()); + writeInst(nextInChain); + if (!oom()) { + label->use(bo.getOffset()); + } + return; + } + + bool conditional = code.encode() != inst_beq.encode(); + + // Make the whole branch continous in the buffer. + m_buffer.ensureSpace((conditional ? 5 : 4) * sizeof(uint32_t)); + +#ifdef JS_JITSPEW + decodeBranchInstAndSpew(code); +#endif + BufferOffset bo = writeInst(code.encode()); + writeInst(nextInChain); + if (!oom()) { + label->use(bo.getOffset()); + } + // Leave space for potential long jump. + as_nop(); + as_nop(); + if (conditional) { + as_nop(); + } +} + +void MacroAssemblerMIPSCompat::cmp64Set(Condition cond, Register64 lhs, + Imm64 val, Register dest) { + if (val.value == 0) { + switch (cond) { + case Assembler::Equal: + case Assembler::BelowOrEqual: + as_or(dest, lhs.high, lhs.low); + as_sltiu(dest, dest, 1); + break; + case Assembler::NotEqual: + case Assembler::Above: + as_or(dest, lhs.high, lhs.low); + as_sltu(dest, zero, dest); + break; + case Assembler::LessThan: + case Assembler::GreaterThanOrEqual: + as_slt(dest, lhs.high, zero); + if (cond == Assembler::GreaterThanOrEqual) { + as_xori(dest, dest, 1); + } + break; + case Assembler::GreaterThan: + case Assembler::LessThanOrEqual: + as_or(SecondScratchReg, lhs.high, lhs.low); + as_sra(ScratchRegister, lhs.high, 31); + as_sltu(dest, ScratchRegister, SecondScratchReg); + if (cond == Assembler::LessThanOrEqual) { + as_xori(dest, dest, 1); + } + break; + case Assembler::Below: + case Assembler::AboveOrEqual: + as_ori(dest, zero, cond == Assembler::AboveOrEqual ? 1 : 0); + break; + default: + MOZ_CRASH("Condition code not supported"); + break; + } + return; + } + + Condition c = ma_cmp64(cond, lhs, val, dest); + + switch (cond) { + // For Equal/NotEqual cond ma_cmp64 dest holds non boolean result. + case Assembler::Equal: + as_sltiu(dest, dest, 1); + break; + case Assembler::NotEqual: + as_sltu(dest, zero, dest); + break; + default: + if (c == Assembler::Zero) as_xori(dest, dest, 1); + break; + } +} + +void MacroAssemblerMIPSCompat::cmp64Set(Condition cond, Register64 lhs, + Register64 rhs, Register dest) { + Condition c = ma_cmp64(cond, lhs, rhs, dest); + + switch (cond) { + // For Equal/NotEqual cond ma_cmp64 dest holds non boolean result. + case Assembler::Equal: + as_sltiu(dest, dest, 1); + break; + case Assembler::NotEqual: + as_sltu(dest, zero, dest); + break; + default: + if (c == Assembler::Zero) as_xori(dest, dest, 1); + break; + } +} + +Assembler::Condition MacroAssemblerMIPSCompat::ma_cmp64(Condition cond, + Register64 lhs, + Register64 rhs, + Register dest) { + switch (cond) { + case Assembler::Equal: + case Assembler::NotEqual: + as_xor(SecondScratchReg, lhs.high, rhs.high); + as_xor(ScratchRegister, lhs.low, rhs.low); + as_or(dest, SecondScratchReg, ScratchRegister); + return (cond == Assembler::Equal) ? Assembler::Zero : Assembler::NonZero; + break; + case Assembler::LessThan: + case Assembler::GreaterThanOrEqual: + as_slt(SecondScratchReg, rhs.high, lhs.high); + as_sltu(ScratchRegister, lhs.low, rhs.low); + as_slt(SecondScratchReg, SecondScratchReg, ScratchRegister); + as_slt(ScratchRegister, lhs.high, rhs.high); + as_or(dest, ScratchRegister, SecondScratchReg); + return (cond == Assembler::GreaterThanOrEqual) ? Assembler::Zero + : Assembler::NonZero; + break; + case Assembler::GreaterThan: + case Assembler::LessThanOrEqual: + as_slt(SecondScratchReg, lhs.high, rhs.high); + as_sltu(ScratchRegister, rhs.low, lhs.low); + as_slt(SecondScratchReg, SecondScratchReg, ScratchRegister); + as_slt(ScratchRegister, rhs.high, lhs.high); + as_or(dest, ScratchRegister, SecondScratchReg); + return (cond == Assembler::LessThanOrEqual) ? Assembler::Zero + : Assembler::NonZero; + break; + case Assembler::Below: + case Assembler::AboveOrEqual: + as_sltu(SecondScratchReg, rhs.high, lhs.high); + as_sltu(ScratchRegister, lhs.low, rhs.low); + as_slt(SecondScratchReg, SecondScratchReg, ScratchRegister); + as_sltu(ScratchRegister, lhs.high, rhs.high); + as_or(dest, ScratchRegister, SecondScratchReg); + return (cond == Assembler::AboveOrEqual) ? Assembler::Zero + : Assembler::NonZero; + break; + case Assembler::Above: + case Assembler::BelowOrEqual: + as_sltu(SecondScratchReg, lhs.high, rhs.high); + as_sltu(ScratchRegister, rhs.low, lhs.low); + as_slt(SecondScratchReg, SecondScratchReg, ScratchRegister); + as_sltu(ScratchRegister, rhs.high, lhs.high); + as_or(dest, ScratchRegister, SecondScratchReg); + return (cond == Assembler::BelowOrEqual) ? Assembler::Zero + : Assembler::NonZero; + break; + default: + MOZ_CRASH("Condition code not supported"); + break; + } +} + +Assembler::Condition MacroAssemblerMIPSCompat::ma_cmp64(Condition cond, + Register64 lhs, + Imm64 val, + Register dest) { + MOZ_ASSERT(val.value != 0); + + switch (cond) { + case Assembler::Equal: + case Assembler::NotEqual: + ma_xor(SecondScratchReg, lhs.high, val.hi()); + ma_xor(ScratchRegister, lhs.low, val.low()); + as_or(dest, SecondScratchReg, ScratchRegister); + return (cond == Assembler::Equal) ? Assembler::Zero : Assembler::NonZero; + break; + case Assembler::LessThan: + case Assembler::GreaterThanOrEqual: + ma_li(SecondScratchReg, val.hi()); + as_slt(ScratchRegister, lhs.high, SecondScratchReg); + as_slt(SecondScratchReg, SecondScratchReg, lhs.high); + as_subu(SecondScratchReg, SecondScratchReg, ScratchRegister); + ma_li(ScratchRegister, val.low()); + as_sltu(ScratchRegister, lhs.low, ScratchRegister); + as_slt(dest, SecondScratchReg, ScratchRegister); + return (cond == Assembler::GreaterThanOrEqual) ? Assembler::Zero + : Assembler::NonZero; + break; + case Assembler::GreaterThan: + case Assembler::LessThanOrEqual: + ma_li(SecondScratchReg, val.hi()); + as_slt(ScratchRegister, SecondScratchReg, lhs.high); + as_slt(SecondScratchReg, lhs.high, SecondScratchReg); + as_subu(SecondScratchReg, SecondScratchReg, ScratchRegister); + ma_li(ScratchRegister, val.low()); + as_sltu(ScratchRegister, ScratchRegister, lhs.low); + as_slt(dest, SecondScratchReg, ScratchRegister); + return (cond == Assembler::LessThanOrEqual) ? Assembler::Zero + : Assembler::NonZero; + break; + case Assembler::Below: + case Assembler::AboveOrEqual: + ma_li(SecondScratchReg, val.hi()); + as_sltu(ScratchRegister, lhs.high, SecondScratchReg); + as_sltu(SecondScratchReg, SecondScratchReg, lhs.high); + as_subu(SecondScratchReg, SecondScratchReg, ScratchRegister); + ma_li(ScratchRegister, val.low()); + as_sltu(ScratchRegister, lhs.low, ScratchRegister); + as_slt(dest, SecondScratchReg, ScratchRegister); + return (cond == Assembler::AboveOrEqual) ? Assembler::Zero + : Assembler::NonZero; + break; + case Assembler::Above: + case Assembler::BelowOrEqual: + ma_li(SecondScratchReg, val.hi()); + as_sltu(ScratchRegister, SecondScratchReg, lhs.high); + as_sltu(SecondScratchReg, lhs.high, SecondScratchReg); + as_subu(SecondScratchReg, SecondScratchReg, ScratchRegister); + ma_li(ScratchRegister, val.low()); + as_sltu(ScratchRegister, ScratchRegister, lhs.low); + as_slt(dest, SecondScratchReg, ScratchRegister); + return (cond == Assembler::BelowOrEqual) ? Assembler::Zero + : Assembler::NonZero; + break; + default: + MOZ_CRASH("Condition code not supported"); + break; + } +} + +// fp instructions +void MacroAssemblerMIPS::ma_lid(FloatRegister dest, double value) { + struct DoubleStruct { + uint32_t lo; + uint32_t hi; + }; + DoubleStruct intStruct = mozilla::BitwiseCast<DoubleStruct>(value); +#if MOZ_BIG_ENDIAN() + std::swap(intStruct.hi, intStruct.lo); +#endif + + // put hi part of 64 bit value into the odd register + if (intStruct.hi == 0) { + moveToDoubleHi(zero, dest); + } else { + ma_li(ScratchRegister, Imm32(intStruct.hi)); + moveToDoubleHi(ScratchRegister, dest); + } + + // put low part of 64 bit value into the even register + if (intStruct.lo == 0) { + moveToDoubleLo(zero, dest); + } else { + ma_li(ScratchRegister, Imm32(intStruct.lo)); + moveToDoubleLo(ScratchRegister, dest); + } +} + +void MacroAssemblerMIPS::ma_mv(FloatRegister src, ValueOperand dest) { + moveFromDoubleLo(src, dest.payloadReg()); + moveFromDoubleHi(src, dest.typeReg()); +} + +void MacroAssemblerMIPS::ma_mv(ValueOperand src, FloatRegister dest) { + moveToDoubleLo(src.payloadReg(), dest); + moveToDoubleHi(src.typeReg(), dest); +} + +void MacroAssemblerMIPS::ma_ls(FloatRegister ft, Address address) { + if (Imm16::IsInSignedRange(address.offset)) { + as_lwc1(ft, address.base, address.offset); + } else { + MOZ_ASSERT(address.base != ScratchRegister); + ma_li(ScratchRegister, Imm32(address.offset)); + if (isLoongson()) { + as_gslsx(ft, address.base, ScratchRegister, 0); + } else { + as_addu(ScratchRegister, address.base, ScratchRegister); + as_lwc1(ft, ScratchRegister, 0); + } + } +} + +void MacroAssemblerMIPS::ma_ld(FloatRegister ft, Address address) { + if (Imm16::IsInSignedRange(address.offset)) { + as_ldc1(ft, address.base, address.offset); + } else { + MOZ_ASSERT(address.base != ScratchRegister); + ma_li(ScratchRegister, Imm32(address.offset)); + if (isLoongson()) { + as_gsldx(ft, address.base, ScratchRegister, 0); + } else { + as_addu(ScratchRegister, address.base, ScratchRegister); + as_ldc1(ft, ScratchRegister, 0); + } + } +} + +void MacroAssemblerMIPS::ma_sd(FloatRegister ft, Address address) { + if (Imm16::IsInSignedRange(address.offset)) { + as_sdc1(ft, address.base, address.offset); + } else { + MOZ_ASSERT(address.base != ScratchRegister); + ma_li(ScratchRegister, Imm32(address.offset)); + if (isLoongson()) { + as_gssdx(ft, address.base, ScratchRegister, 0); + } else { + as_addu(ScratchRegister, address.base, ScratchRegister); + as_sdc1(ft, ScratchRegister, 0); + } + } +} + +void MacroAssemblerMIPS::ma_ss(FloatRegister ft, Address address) { + if (Imm16::IsInSignedRange(address.offset)) { + as_swc1(ft, address.base, address.offset); + } else { + MOZ_ASSERT(address.base != ScratchRegister); + ma_li(ScratchRegister, Imm32(address.offset)); + if (isLoongson()) { + as_gsssx(ft, address.base, ScratchRegister, 0); + } else { + as_addu(ScratchRegister, address.base, ScratchRegister); + as_swc1(ft, ScratchRegister, 0); + } + } +} + +void MacroAssemblerMIPS::ma_ldc1WordAligned(FloatRegister ft, Register base, + int32_t off) { + MOZ_ASSERT(Imm16::IsInSignedRange(off + PAYLOAD_OFFSET) && + Imm16::IsInSignedRange(off + TAG_OFFSET)); + + as_lwc1(ft, base, off + PAYLOAD_OFFSET); + as_lwc1(getOddPair(ft), base, off + TAG_OFFSET); +} + +void MacroAssemblerMIPS::ma_sdc1WordAligned(FloatRegister ft, Register base, + int32_t off) { + MOZ_ASSERT(Imm16::IsInSignedRange(off + PAYLOAD_OFFSET) && + Imm16::IsInSignedRange(off + TAG_OFFSET)); + + as_swc1(ft, base, off + PAYLOAD_OFFSET); + as_swc1(getOddPair(ft), base, off + TAG_OFFSET); +} + +void MacroAssemblerMIPS::ma_pop(FloatRegister f) { + if (f.isDouble()) { + ma_ldc1WordAligned(f, StackPointer, 0); + } else { + as_lwc1(f, StackPointer, 0); + } + + as_addiu(StackPointer, StackPointer, f.size()); +} + +void MacroAssemblerMIPS::ma_push(FloatRegister f) { + as_addiu(StackPointer, StackPointer, -f.size()); + + if (f.isDouble()) { + ma_sdc1WordAligned(f, StackPointer, 0); + } else { + as_swc1(f, StackPointer, 0); + } +} + +bool MacroAssemblerMIPSCompat::buildOOLFakeExitFrame(void* fakeReturnAddr) { + uint32_t descriptor = MakeFrameDescriptor( + asMasm().framePushed(), FrameType::IonJS, ExitFrameLayout::Size()); + + asMasm().Push(Imm32(descriptor)); // descriptor_ + asMasm().Push(ImmPtr(fakeReturnAddr)); + + return true; +} + +void MacroAssemblerMIPSCompat::move32(Imm32 imm, Register dest) { + ma_li(dest, imm); +} + +void MacroAssemblerMIPSCompat::move32(Register src, Register dest) { + ma_move(dest, src); +} + +void MacroAssemblerMIPSCompat::movePtr(Register src, Register dest) { + ma_move(dest, src); +} +void MacroAssemblerMIPSCompat::movePtr(ImmWord imm, Register dest) { + ma_li(dest, imm); +} + +void MacroAssemblerMIPSCompat::movePtr(ImmGCPtr imm, Register dest) { + ma_li(dest, imm); +} + +void MacroAssemblerMIPSCompat::movePtr(ImmPtr imm, Register dest) { + movePtr(ImmWord(uintptr_t(imm.value)), dest); +} +void MacroAssemblerMIPSCompat::movePtr(wasm::SymbolicAddress imm, + Register dest) { + append(wasm::SymbolicAccess(CodeOffset(nextOffset().getOffset()), imm)); + ma_liPatchable(dest, ImmWord(-1)); +} + +void MacroAssemblerMIPSCompat::load8ZeroExtend(const Address& address, + Register dest) { + ma_load(dest, address, SizeByte, ZeroExtend); +} + +void MacroAssemblerMIPSCompat::load8ZeroExtend(const BaseIndex& src, + Register dest) { + ma_load(dest, src, SizeByte, ZeroExtend); +} + +void MacroAssemblerMIPSCompat::load8SignExtend(const Address& address, + Register dest) { + ma_load(dest, address, SizeByte, SignExtend); +} + +void MacroAssemblerMIPSCompat::load8SignExtend(const BaseIndex& src, + Register dest) { + ma_load(dest, src, SizeByte, SignExtend); +} + +void MacroAssemblerMIPSCompat::load16ZeroExtend(const Address& address, + Register dest) { + ma_load(dest, address, SizeHalfWord, ZeroExtend); +} + +void MacroAssemblerMIPSCompat::load16ZeroExtend(const BaseIndex& src, + Register dest) { + ma_load(dest, src, SizeHalfWord, ZeroExtend); +} + +void MacroAssemblerMIPSCompat::load16SignExtend(const Address& address, + Register dest) { + ma_load(dest, address, SizeHalfWord, SignExtend); +} + +void MacroAssemblerMIPSCompat::load16SignExtend(const BaseIndex& src, + Register dest) { + ma_load(dest, src, SizeHalfWord, SignExtend); +} + +void MacroAssemblerMIPSCompat::load32(const Address& address, Register dest) { + ma_load(dest, address, SizeWord); +} + +void MacroAssemblerMIPSCompat::load32(const BaseIndex& address, Register dest) { + ma_load(dest, address, SizeWord); +} + +void MacroAssemblerMIPSCompat::load32(AbsoluteAddress address, Register dest) { + movePtr(ImmPtr(address.addr), ScratchRegister); + load32(Address(ScratchRegister, 0), dest); +} + +void MacroAssemblerMIPSCompat::load32(wasm::SymbolicAddress address, + Register dest) { + movePtr(address, ScratchRegister); + load32(Address(ScratchRegister, 0), dest); +} + +void MacroAssemblerMIPSCompat::loadPtr(const Address& address, Register dest) { + ma_load(dest, address, SizeWord); +} + +void MacroAssemblerMIPSCompat::loadPtr(const BaseIndex& src, Register dest) { + ma_load(dest, src, SizeWord); +} + +void MacroAssemblerMIPSCompat::loadPtr(AbsoluteAddress address, Register dest) { + movePtr(ImmPtr(address.addr), ScratchRegister); + loadPtr(Address(ScratchRegister, 0), dest); +} + +void MacroAssemblerMIPSCompat::loadPtr(wasm::SymbolicAddress address, + Register dest) { + movePtr(address, ScratchRegister); + loadPtr(Address(ScratchRegister, 0), dest); +} + +void MacroAssemblerMIPSCompat::loadPrivate(const Address& address, + Register dest) { + ma_lw(dest, Address(address.base, address.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::loadUnalignedDouble( + const wasm::MemoryAccessDesc& access, const BaseIndex& src, Register temp, + FloatRegister dest) { + MOZ_ASSERT(MOZ_LITTLE_ENDIAN(), "Wasm-only; wasm is disabled on big-endian."); + computeScaledAddress(src, SecondScratchReg); + + BufferOffset load; + if (Imm16::IsInSignedRange(src.offset) && + Imm16::IsInSignedRange(src.offset + 7)) { + load = as_lwl(temp, SecondScratchReg, src.offset + INT64LOW_OFFSET + 3); + as_lwr(temp, SecondScratchReg, src.offset + INT64LOW_OFFSET); + append(access, load.getOffset()); + moveToDoubleLo(temp, dest); + load = as_lwl(temp, SecondScratchReg, src.offset + INT64HIGH_OFFSET + 3); + as_lwr(temp, SecondScratchReg, src.offset + INT64HIGH_OFFSET); + append(access, load.getOffset()); + moveToDoubleHi(temp, dest); + } else { + ma_li(ScratchRegister, Imm32(src.offset)); + as_daddu(ScratchRegister, SecondScratchReg, ScratchRegister); + load = as_lwl(temp, ScratchRegister, INT64LOW_OFFSET + 3); + as_lwr(temp, ScratchRegister, INT64LOW_OFFSET); + append(access, load.getOffset()); + moveToDoubleLo(temp, dest); + load = as_lwl(temp, ScratchRegister, INT64HIGH_OFFSET + 3); + as_lwr(temp, ScratchRegister, INT64HIGH_OFFSET); + append(access, load.getOffset()); + moveToDoubleHi(temp, dest); + } +} + +void MacroAssemblerMIPSCompat::loadUnalignedFloat32( + const wasm::MemoryAccessDesc& access, const BaseIndex& src, Register temp, + FloatRegister dest) { + MOZ_ASSERT(MOZ_LITTLE_ENDIAN(), "Wasm-only; wasm is disabled on big-endian."); + computeScaledAddress(src, SecondScratchReg); + BufferOffset load; + if (Imm16::IsInSignedRange(src.offset) && + Imm16::IsInSignedRange(src.offset + 3)) { + load = as_lwl(temp, SecondScratchReg, src.offset + 3); + as_lwr(temp, SecondScratchReg, src.offset); + } else { + ma_li(ScratchRegister, Imm32(src.offset)); + as_daddu(ScratchRegister, SecondScratchReg, ScratchRegister); + load = as_lwl(temp, ScratchRegister, 3); + as_lwr(temp, ScratchRegister, 0); + } + append(access, load.getOffset()); + moveToFloat32(temp, dest); +} + +void MacroAssemblerMIPSCompat::store8(Imm32 imm, const Address& address) { + ma_li(SecondScratchReg, imm); + ma_store(SecondScratchReg, address, SizeByte); +} + +void MacroAssemblerMIPSCompat::store8(Register src, const Address& address) { + ma_store(src, address, SizeByte); +} + +void MacroAssemblerMIPSCompat::store8(Imm32 imm, const BaseIndex& dest) { + ma_store(imm, dest, SizeByte); +} + +void MacroAssemblerMIPSCompat::store8(Register src, const BaseIndex& dest) { + ma_store(src, dest, SizeByte); +} + +void MacroAssemblerMIPSCompat::store16(Imm32 imm, const Address& address) { + ma_li(SecondScratchReg, imm); + ma_store(SecondScratchReg, address, SizeHalfWord); +} + +void MacroAssemblerMIPSCompat::store16(Register src, const Address& address) { + ma_store(src, address, SizeHalfWord); +} + +void MacroAssemblerMIPSCompat::store16(Imm32 imm, const BaseIndex& dest) { + ma_store(imm, dest, SizeHalfWord); +} + +void MacroAssemblerMIPSCompat::store16(Register src, const BaseIndex& address) { + ma_store(src, address, SizeHalfWord); +} + +void MacroAssemblerMIPSCompat::store32(Register src, AbsoluteAddress address) { + movePtr(ImmPtr(address.addr), ScratchRegister); + store32(src, Address(ScratchRegister, 0)); +} + +void MacroAssemblerMIPSCompat::store32(Register src, const Address& address) { + ma_store(src, address, SizeWord); +} + +void MacroAssemblerMIPSCompat::store32(Imm32 src, const Address& address) { + move32(src, SecondScratchReg); + ma_store(SecondScratchReg, address, SizeWord); +} + +void MacroAssemblerMIPSCompat::store32(Imm32 imm, const BaseIndex& dest) { + ma_store(imm, dest, SizeWord); +} + +void MacroAssemblerMIPSCompat::store32(Register src, const BaseIndex& dest) { + ma_store(src, dest, SizeWord); +} + +template <typename T> +void MacroAssemblerMIPSCompat::storePtr(ImmWord imm, T address) { + ma_li(SecondScratchReg, imm); + ma_store(SecondScratchReg, address, SizeWord); +} + +template void MacroAssemblerMIPSCompat::storePtr<Address>(ImmWord imm, + Address address); +template void MacroAssemblerMIPSCompat::storePtr<BaseIndex>(ImmWord imm, + BaseIndex address); + +template <typename T> +void MacroAssemblerMIPSCompat::storePtr(ImmPtr imm, T address) { + storePtr(ImmWord(uintptr_t(imm.value)), address); +} + +template void MacroAssemblerMIPSCompat::storePtr<Address>(ImmPtr imm, + Address address); +template void MacroAssemblerMIPSCompat::storePtr<BaseIndex>(ImmPtr imm, + BaseIndex address); + +template <typename T> +void MacroAssemblerMIPSCompat::storePtr(ImmGCPtr imm, T address) { + movePtr(imm, SecondScratchReg); + storePtr(SecondScratchReg, address); +} + +template void MacroAssemblerMIPSCompat::storePtr<Address>(ImmGCPtr imm, + Address address); +template void MacroAssemblerMIPSCompat::storePtr<BaseIndex>(ImmGCPtr imm, + BaseIndex address); + +void MacroAssemblerMIPSCompat::storePtr(Register src, const Address& address) { + ma_store(src, address, SizeWord); +} + +void MacroAssemblerMIPSCompat::storePtr(Register src, + const BaseIndex& address) { + ma_store(src, address, SizeWord); +} + +void MacroAssemblerMIPSCompat::storePtr(Register src, AbsoluteAddress dest) { + movePtr(ImmPtr(dest.addr), ScratchRegister); + storePtr(src, Address(ScratchRegister, 0)); +} + +void MacroAssemblerMIPSCompat::storeUnalignedFloat32( + const wasm::MemoryAccessDesc& access, FloatRegister src, Register temp, + const BaseIndex& dest) { + MOZ_ASSERT(MOZ_LITTLE_ENDIAN(), "Wasm-only; wasm is disabled on big-endian."); + computeScaledAddress(dest, SecondScratchReg); + moveFromFloat32(src, temp); + + BufferOffset store; + if (Imm16::IsInSignedRange(dest.offset) && + Imm16::IsInSignedRange(dest.offset + 3)) { + store = as_swl(temp, SecondScratchReg, dest.offset + 3); + as_swr(temp, SecondScratchReg, dest.offset); + } else { + ma_li(ScratchRegister, Imm32(dest.offset)); + as_daddu(ScratchRegister, SecondScratchReg, ScratchRegister); + store = as_swl(temp, ScratchRegister, 3); + as_swr(temp, ScratchRegister, 0); + } + append(access, store.getOffset()); +} + +void MacroAssemblerMIPSCompat::storeUnalignedDouble( + const wasm::MemoryAccessDesc& access, FloatRegister src, Register temp, + const BaseIndex& dest) { + MOZ_ASSERT(MOZ_LITTLE_ENDIAN(), "Wasm-only; wasm is disabled on big-endian."); + computeScaledAddress(dest, SecondScratchReg); + + BufferOffset store; + if (Imm16::IsInSignedRange(dest.offset) && + Imm16::IsInSignedRange(dest.offset + 7)) { + moveFromDoubleHi(src, temp); + store = as_swl(temp, SecondScratchReg, dest.offset + INT64HIGH_OFFSET + 3); + as_swr(temp, SecondScratchReg, dest.offset + INT64HIGH_OFFSET); + moveFromDoubleLo(src, temp); + as_swl(temp, SecondScratchReg, dest.offset + INT64LOW_OFFSET + 3); + as_swr(temp, SecondScratchReg, dest.offset + INT64LOW_OFFSET); + + } else { + ma_li(ScratchRegister, Imm32(dest.offset)); + as_daddu(ScratchRegister, SecondScratchReg, ScratchRegister); + moveFromDoubleHi(src, temp); + store = as_swl(temp, ScratchRegister, INT64HIGH_OFFSET + 3); + as_swr(temp, ScratchRegister, INT64HIGH_OFFSET); + moveFromDoubleLo(src, temp); + as_swl(temp, ScratchRegister, INT64LOW_OFFSET + 3); + as_swr(temp, ScratchRegister, INT64LOW_OFFSET); + } + append(access, store.getOffset()); +} + +void MacroAssembler::clampDoubleToUint8(FloatRegister input, Register output) { + as_roundwd(ScratchDoubleReg, input); + ma_li(ScratchRegister, Imm32(255)); + as_mfc1(output, ScratchDoubleReg); + zeroDouble(ScratchDoubleReg); + as_sltiu(SecondScratchReg, output, 255); + as_colt(DoubleFloat, ScratchDoubleReg, input); + // if res > 255; res = 255; + as_movz(output, ScratchRegister, SecondScratchReg); + // if !(input > 0); res = 0; + as_movf(output, zero); +} + +// higher level tag testing code +Operand MacroAssemblerMIPSCompat::ToPayload(Operand base) { + return Operand(Register::FromCode(base.base()), base.disp() + PAYLOAD_OFFSET); +} + +Operand MacroAssemblerMIPSCompat::ToType(Operand base) { + return Operand(Register::FromCode(base.base()), base.disp() + TAG_OFFSET); +} + +void MacroAssemblerMIPSCompat::testNullSet(Condition cond, + const ValueOperand& value, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_cmp_set(dest, value.typeReg(), ImmType(JSVAL_TYPE_NULL), cond); +} + +void MacroAssemblerMIPSCompat::testObjectSet(Condition cond, + const ValueOperand& value, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_cmp_set(dest, value.typeReg(), ImmType(JSVAL_TYPE_OBJECT), cond); +} + +void MacroAssemblerMIPSCompat::testUndefinedSet(Condition cond, + const ValueOperand& value, + Register dest) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ma_cmp_set(dest, value.typeReg(), ImmType(JSVAL_TYPE_UNDEFINED), cond); +} + +// unboxing code +void MacroAssemblerMIPSCompat::unboxNonDouble(const ValueOperand& operand, + Register dest, JSValueType) { + if (operand.payloadReg() != dest) { + ma_move(dest, operand.payloadReg()); + } +} + +void MacroAssemblerMIPSCompat::unboxNonDouble(const Address& src, Register dest, + JSValueType) { + ma_lw(dest, Address(src.base, src.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::unboxNonDouble(const BaseIndex& src, + Register dest, JSValueType) { + computeScaledAddress(src, SecondScratchReg); + ma_lw(dest, Address(SecondScratchReg, src.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::unboxInt32(const ValueOperand& operand, + Register dest) { + ma_move(dest, operand.payloadReg()); +} + +void MacroAssemblerMIPSCompat::unboxInt32(const Address& src, Register dest) { + ma_lw(dest, Address(src.base, src.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::unboxBoolean(const ValueOperand& operand, + Register dest) { + ma_move(dest, operand.payloadReg()); +} + +void MacroAssemblerMIPSCompat::unboxBoolean(const Address& src, Register dest) { + ma_lw(dest, Address(src.base, src.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::unboxDouble(const ValueOperand& operand, + FloatRegister dest) { + moveToDoubleLo(operand.payloadReg(), dest); + moveToDoubleHi(operand.typeReg(), dest); +} + +void MacroAssemblerMIPSCompat::unboxDouble(const Address& src, + FloatRegister dest) { + ma_lw(ScratchRegister, Address(src.base, src.offset + PAYLOAD_OFFSET)); + moveToDoubleLo(ScratchRegister, dest); + ma_lw(ScratchRegister, Address(src.base, src.offset + TAG_OFFSET)); + moveToDoubleHi(ScratchRegister, dest); +} + +void MacroAssemblerMIPSCompat::unboxDouble(const BaseIndex& src, + FloatRegister dest) { + loadDouble(src, dest); +} + +void MacroAssemblerMIPSCompat::unboxString(const ValueOperand& operand, + Register dest) { + ma_move(dest, operand.payloadReg()); +} + +void MacroAssemblerMIPSCompat::unboxString(const Address& src, Register dest) { + ma_lw(dest, Address(src.base, src.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::unboxBigInt(const ValueOperand& operand, + Register dest) { + ma_move(dest, operand.payloadReg()); +} + +void MacroAssemblerMIPSCompat::unboxBigInt(const Address& src, Register dest) { + ma_lw(dest, Address(src.base, src.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::unboxObject(const ValueOperand& src, + Register dest) { + ma_move(dest, src.payloadReg()); +} + +void MacroAssemblerMIPSCompat::unboxObject(const Address& src, Register dest) { + ma_lw(dest, Address(src.base, src.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::unboxObjectOrNull(const Address& src, + Register dest) { + ma_lw(dest, Address(src.base, src.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::unboxValue(const ValueOperand& src, + AnyRegister dest, JSValueType) { + if (dest.isFloat()) { + Label notInt32, end; + asMasm().branchTestInt32(Assembler::NotEqual, src, ¬Int32); + convertInt32ToDouble(src.payloadReg(), dest.fpu()); + ma_b(&end, ShortJump); + bind(¬Int32); + unboxDouble(src, dest.fpu()); + bind(&end); + } else if (src.payloadReg() != dest.gpr()) { + ma_move(dest.gpr(), src.payloadReg()); + } +} + +void MacroAssemblerMIPSCompat::boxDouble(FloatRegister src, + const ValueOperand& dest, + FloatRegister) { + moveFromDoubleLo(src, dest.payloadReg()); + moveFromDoubleHi(src, dest.typeReg()); +} + +void MacroAssemblerMIPSCompat::boxNonDouble(JSValueType type, Register src, + const ValueOperand& dest) { + if (src != dest.payloadReg()) { + ma_move(dest.payloadReg(), src); + } + ma_li(dest.typeReg(), ImmType(type)); +} + +void MacroAssemblerMIPSCompat::boolValueToDouble(const ValueOperand& operand, + FloatRegister dest) { + convertBoolToInt32(operand.payloadReg(), ScratchRegister); + convertInt32ToDouble(ScratchRegister, dest); +} + +void MacroAssemblerMIPSCompat::int32ValueToDouble(const ValueOperand& operand, + FloatRegister dest) { + convertInt32ToDouble(operand.payloadReg(), dest); +} + +void MacroAssemblerMIPSCompat::boolValueToFloat32(const ValueOperand& operand, + FloatRegister dest) { + convertBoolToInt32(operand.payloadReg(), ScratchRegister); + convertInt32ToFloat32(ScratchRegister, dest); +} + +void MacroAssemblerMIPSCompat::int32ValueToFloat32(const ValueOperand& operand, + FloatRegister dest) { + convertInt32ToFloat32(operand.payloadReg(), dest); +} + +void MacroAssemblerMIPSCompat::loadConstantFloat32(float f, + FloatRegister dest) { + ma_lis(dest, f); +} + +void MacroAssemblerMIPSCompat::loadInt32OrDouble(const Address& src, + FloatRegister dest) { + Label notInt32, end; + // If it's an int, convert it to double. + ma_lw(SecondScratchReg, Address(src.base, src.offset + TAG_OFFSET)); + asMasm().branchTestInt32(Assembler::NotEqual, SecondScratchReg, ¬Int32); + ma_lw(SecondScratchReg, Address(src.base, src.offset + PAYLOAD_OFFSET)); + convertInt32ToDouble(SecondScratchReg, dest); + ma_b(&end, ShortJump); + + // Not an int, just load as double. + bind(¬Int32); + ma_ld(dest, src); + bind(&end); +} + +void MacroAssemblerMIPSCompat::loadInt32OrDouble(Register base, Register index, + FloatRegister dest, + int32_t shift) { + Label notInt32, end; + + // If it's an int, convert it to double. + + computeScaledAddress(BaseIndex(base, index, ShiftToScale(shift)), + SecondScratchReg); + // Since we only have one scratch, we need to stomp over it with the tag. + load32(Address(SecondScratchReg, TAG_OFFSET), SecondScratchReg); + asMasm().branchTestInt32(Assembler::NotEqual, SecondScratchReg, ¬Int32); + + computeScaledAddress(BaseIndex(base, index, ShiftToScale(shift)), + SecondScratchReg); + load32(Address(SecondScratchReg, PAYLOAD_OFFSET), SecondScratchReg); + convertInt32ToDouble(SecondScratchReg, dest); + ma_b(&end, ShortJump); + + // 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(BaseIndex(base, index, ShiftToScale(shift)), + SecondScratchReg); + loadDouble(Address(SecondScratchReg, 0), dest); + bind(&end); +} + +void MacroAssemblerMIPSCompat::loadConstantDouble(double dp, + FloatRegister dest) { + ma_lid(dest, dp); +} + +Register MacroAssemblerMIPSCompat::extractObject(const Address& address, + Register scratch) { + ma_lw(scratch, Address(address.base, address.offset + PAYLOAD_OFFSET)); + return scratch; +} + +Register MacroAssemblerMIPSCompat::extractTag(const Address& address, + Register scratch) { + ma_lw(scratch, Address(address.base, address.offset + TAG_OFFSET)); + return scratch; +} + +Register MacroAssemblerMIPSCompat::extractTag(const BaseIndex& address, + Register scratch) { + computeScaledAddress(address, scratch); + return extractTag(Address(scratch, address.offset), scratch); +} + +uint32_t MacroAssemblerMIPSCompat::getType(const Value& val) { + return val.toNunboxTag(); +} + +void MacroAssemblerMIPSCompat::moveData(const Value& val, Register data) { + if (val.isGCThing()) { + ma_li(data, ImmGCPtr(val.toGCThing())); + } else { + ma_li(data, Imm32(val.toNunboxPayload())); + } +} + +///////////////////////////////////////////////////////////////// +// X86/X64-common/ARM/MIPS interface. +///////////////////////////////////////////////////////////////// +void MacroAssemblerMIPSCompat::storeValue(ValueOperand val, Operand dst) { + storeValue(val, Address(Register::FromCode(dst.base()), dst.disp())); +} + +void MacroAssemblerMIPSCompat::storeValue(ValueOperand val, + const BaseIndex& dest) { + computeScaledAddress(dest, SecondScratchReg); + storeValue(val, Address(SecondScratchReg, dest.offset)); +} + +void MacroAssemblerMIPSCompat::storeValue(JSValueType type, Register reg, + BaseIndex dest) { + computeScaledAddress(dest, ScratchRegister); + + // Make sure that ma_sw doesn't clobber ScratchRegister + int32_t offset = dest.offset; + if (!Imm16::IsInSignedRange(offset)) { + ma_li(SecondScratchReg, Imm32(offset)); + as_addu(ScratchRegister, ScratchRegister, SecondScratchReg); + offset = 0; + } + + storeValue(type, reg, Address(ScratchRegister, offset)); +} + +void MacroAssemblerMIPSCompat::storeValue(ValueOperand val, + const Address& dest) { + ma_sw(val.payloadReg(), Address(dest.base, dest.offset + PAYLOAD_OFFSET)); + ma_sw(val.typeReg(), Address(dest.base, dest.offset + TAG_OFFSET)); +} + +void MacroAssemblerMIPSCompat::storeValue(JSValueType type, Register reg, + Address dest) { + MOZ_ASSERT(dest.base != SecondScratchReg); + + ma_sw(reg, Address(dest.base, dest.offset + PAYLOAD_OFFSET)); + ma_li(SecondScratchReg, ImmTag(JSVAL_TYPE_TO_TAG(type))); + ma_sw(SecondScratchReg, Address(dest.base, dest.offset + TAG_OFFSET)); +} + +void MacroAssemblerMIPSCompat::storeValue(const Value& val, Address dest) { + MOZ_ASSERT(dest.base != SecondScratchReg); + + ma_li(SecondScratchReg, Imm32(getType(val))); + ma_sw(SecondScratchReg, Address(dest.base, dest.offset + TAG_OFFSET)); + moveData(val, SecondScratchReg); + ma_sw(SecondScratchReg, Address(dest.base, dest.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::storeValue(const Value& val, BaseIndex dest) { + computeScaledAddress(dest, ScratchRegister); + + // Make sure that ma_sw doesn't clobber ScratchRegister + int32_t offset = dest.offset; + if (!Imm16::IsInSignedRange(offset)) { + ma_li(SecondScratchReg, Imm32(offset)); + as_addu(ScratchRegister, ScratchRegister, SecondScratchReg); + offset = 0; + } + storeValue(val, Address(ScratchRegister, offset)); +} + +void MacroAssemblerMIPSCompat::loadValue(const BaseIndex& addr, + ValueOperand val) { + computeScaledAddress(addr, SecondScratchReg); + loadValue(Address(SecondScratchReg, addr.offset), val); +} + +void MacroAssemblerMIPSCompat::loadValue(Address src, ValueOperand val) { + // Ensure that loading the payload does not erase the pointer to the + // Value in memory. + if (src.base != val.payloadReg()) { + ma_lw(val.payloadReg(), Address(src.base, src.offset + PAYLOAD_OFFSET)); + ma_lw(val.typeReg(), Address(src.base, src.offset + TAG_OFFSET)); + } else { + ma_lw(val.typeReg(), Address(src.base, src.offset + TAG_OFFSET)); + ma_lw(val.payloadReg(), Address(src.base, src.offset + PAYLOAD_OFFSET)); + } +} + +void MacroAssemblerMIPSCompat::tagValue(JSValueType type, Register payload, + ValueOperand dest) { + MOZ_ASSERT(dest.typeReg() != dest.payloadReg()); + if (payload != dest.payloadReg()) { + ma_move(dest.payloadReg(), payload); + } + ma_li(dest.typeReg(), ImmType(type)); +} + +void MacroAssemblerMIPSCompat::pushValue(ValueOperand val) { + // Allocate stack slots for type and payload. One for each. + asMasm().subPtr(Imm32(sizeof(Value)), StackPointer); + // Store type and payload. + storeValue(val, Address(StackPointer, 0)); +} + +void MacroAssemblerMIPSCompat::pushValue(const Address& addr) { + // Allocate stack slots for type and payload. One for each. + ma_subu(StackPointer, StackPointer, Imm32(sizeof(Value))); + // If address is based on StackPointer its offset needs to be adjusted + // to accommodate for previous stack allocation. + int32_t offset = + addr.base != StackPointer ? addr.offset : addr.offset + sizeof(Value); + // Store type and payload. + ma_lw(ScratchRegister, Address(addr.base, offset + TAG_OFFSET)); + ma_sw(ScratchRegister, Address(StackPointer, TAG_OFFSET)); + ma_lw(ScratchRegister, Address(addr.base, offset + PAYLOAD_OFFSET)); + ma_sw(ScratchRegister, Address(StackPointer, PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::popValue(ValueOperand val) { + // Load payload and type. + as_lw(val.payloadReg(), StackPointer, PAYLOAD_OFFSET); + as_lw(val.typeReg(), StackPointer, TAG_OFFSET); + // Free stack. + as_addiu(StackPointer, StackPointer, sizeof(Value)); +} + +void MacroAssemblerMIPSCompat::storePayload(const Value& val, Address dest) { + moveData(val, SecondScratchReg); + ma_sw(SecondScratchReg, Address(dest.base, dest.offset + PAYLOAD_OFFSET)); +} + +void MacroAssemblerMIPSCompat::storePayload(Register src, Address dest) { + ma_sw(src, Address(dest.base, dest.offset + PAYLOAD_OFFSET)); + return; +} + +void MacroAssemblerMIPSCompat::storePayload(const Value& val, + const BaseIndex& dest) { + MOZ_ASSERT(dest.offset == 0); + + computeScaledAddress(dest, SecondScratchReg); + + moveData(val, ScratchRegister); + + as_sw(ScratchRegister, SecondScratchReg, NUNBOX32_PAYLOAD_OFFSET); +} + +void MacroAssemblerMIPSCompat::storePayload(Register src, + const BaseIndex& dest) { + MOZ_ASSERT(dest.offset == 0); + + computeScaledAddress(dest, SecondScratchReg); + as_sw(src, SecondScratchReg, NUNBOX32_PAYLOAD_OFFSET); +} + +void MacroAssemblerMIPSCompat::storeTypeTag(ImmTag tag, Address dest) { + ma_li(SecondScratchReg, tag); + ma_sw(SecondScratchReg, Address(dest.base, dest.offset + TAG_OFFSET)); +} + +void MacroAssemblerMIPSCompat::storeTypeTag(ImmTag tag, const BaseIndex& dest) { + MOZ_ASSERT(dest.offset == 0); + + computeScaledAddress(dest, SecondScratchReg); + ma_li(ScratchRegister, tag); + as_sw(ScratchRegister, SecondScratchReg, TAG_OFFSET); +} + +void MacroAssemblerMIPSCompat::breakpoint() { as_break(0); } + +void MacroAssemblerMIPSCompat::ensureDouble(const ValueOperand& source, + FloatRegister dest, + Label* failure) { + Label isDouble, done; + asMasm().branchTestDouble(Assembler::Equal, source.typeReg(), &isDouble); + asMasm().branchTestInt32(Assembler::NotEqual, source.typeReg(), failure); + + convertInt32ToDouble(source.payloadReg(), dest); + jump(&done); + + bind(&isDouble); + unboxDouble(source, dest); + + bind(&done); +} + +void MacroAssemblerMIPSCompat::checkStackAlignment() { +#ifdef DEBUG + Label aligned; + as_andi(ScratchRegister, sp, ABIStackAlignment - 1); + ma_b(ScratchRegister, zero, &aligned, Equal, ShortJump); + as_break(BREAK_STACK_UNALIGNED); + bind(&aligned); +#endif +} + +void MacroAssemblerMIPSCompat::alignStackPointer() { + movePtr(StackPointer, SecondScratchReg); + asMasm().subPtr(Imm32(sizeof(intptr_t)), StackPointer); + asMasm().andPtr(Imm32(~(ABIStackAlignment - 1)), StackPointer); + storePtr(SecondScratchReg, Address(StackPointer, 0)); +} + +void MacroAssemblerMIPSCompat::restoreStackPointer() { + loadPtr(Address(StackPointer, 0), StackPointer); +} + +void MacroAssemblerMIPSCompat::handleFailureWithHandlerTail( + Label* profilerExitTail) { + // Reserve space for exception information. + int size = (sizeof(ResumeFromException) + ABIStackAlignment) & + ~(ABIStackAlignment - 1); + asMasm().subPtr(Imm32(size), StackPointer); + ma_move(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>( + MoveOp::GENERAL, CheckUnsafeCallWithABI::DontCheckHasExitFrame); + + Label entryFrame; + Label catch_; + Label finally; + Label return_; + Label bailout; + Label wasm; + + // Already clobbered a0, so use it... + load32(Address(StackPointer, offsetof(ResumeFromException, kind)), a0); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ResumeFromException::RESUME_ENTRY_FRAME), + &entryFrame); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ResumeFromException::RESUME_CATCH), &catch_); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ResumeFromException::RESUME_FINALLY), &finally); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ResumeFromException::RESUME_FORCED_RETURN), &return_); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ResumeFromException::RESUME_BAILOUT), &bailout); + asMasm().branch32(Assembler::Equal, a0, + Imm32(ResumeFromException::RESUME_WASM), &wasm); + + breakpoint(); // Invalid kind. + + // No exception handler. Load the error value, load the new stack pointer + // and return from the entry frame. + bind(&entryFrame); + asMasm().moveValue(MagicValue(JS_ION_ERROR), JSReturnOperand); + loadPtr(Address(StackPointer, offsetof(ResumeFromException, stackPointer)), + StackPointer); + + // We're going to be returning by the ion calling convention + ma_pop(ra); + as_jr(ra); + as_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, offsetof(ResumeFromException, target)), a0); + loadPtr(Address(StackPointer, offsetof(ResumeFromException, framePointer)), + BaselineFrameReg); + loadPtr(Address(StackPointer, offsetof(ResumeFromException, stackPointer)), + StackPointer); + jump(a0); + + // If we found a finally block, this must be a baseline frame. Push + // two values expected by JSOp::Retsub: BooleanValue(true) and the + // exception. + bind(&finally); + ValueOperand exception = ValueOperand(a1, a2); + loadValue(Address(sp, offsetof(ResumeFromException, exception)), exception); + + loadPtr(Address(sp, offsetof(ResumeFromException, target)), a0); + loadPtr(Address(sp, offsetof(ResumeFromException, framePointer)), + BaselineFrameReg); + loadPtr(Address(sp, offsetof(ResumeFromException, stackPointer)), sp); + + pushValue(BooleanValue(true)); + pushValue(exception); + jump(a0); + + // Only used in debug mode. Return BaselineFrame->returnValue() to the + // caller. + bind(&return_); + loadPtr(Address(StackPointer, offsetof(ResumeFromException, framePointer)), + BaselineFrameReg); + loadPtr(Address(StackPointer, offsetof(ResumeFromException, stackPointer)), + StackPointer); + loadValue( + Address(BaselineFrameReg, BaselineFrame::reverseOffsetOfReturnValue()), + JSReturnOperand); + ma_move(StackPointer, BaselineFrameReg); + pop(BaselineFrameReg); + + // If profiling is enabled, then update the lastProfilingFrame to refer to + // caller frame before returning. + { + Label skipProfilingInstrumentation; + // Test if profiler enabled. + AbsoluteAddress addressOfEnabled( + GetJitContext()->runtime->geckoProfiler().addressOfEnabled()); + asMasm().branch32(Assembler::Equal, addressOfEnabled, Imm32(0), + &skipProfilingInstrumentation); + jump(profilerExitTail); + bind(&skipProfilingInstrumentation); + } + + 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, offsetof(ResumeFromException, bailoutInfo)), a2); + ma_li(ReturnReg, Imm32(1)); + loadPtr(Address(sp, offsetof(ResumeFromException, target)), a1); + jump(a1); + + // 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, offsetof(ResumeFromException, framePointer)), + FramePointer); + loadPtr(Address(StackPointer, offsetof(ResumeFromException, stackPointer)), + StackPointer); + ret(); +} + +CodeOffset MacroAssemblerMIPSCompat::toggledJump(Label* label) { + CodeOffset ret(nextOffset().getOffset()); + ma_b(label); + return ret; +} + +CodeOffset MacroAssemblerMIPSCompat::toggledCall(JitCode* target, + bool enabled) { + BufferOffset bo = nextOffset(); + CodeOffset offset(bo.getOffset()); + addPendingJump(bo, ImmPtr(target->raw()), RelocationKind::JITCODE); + ma_liPatchable(ScratchRegister, ImmPtr(target->raw())); + if (enabled) { + as_jalr(ScratchRegister); + as_nop(); + } else { + as_nop(); + as_nop(); + } + MOZ_ASSERT_IF(!oom(), nextOffset().getOffset() - offset.offset() == + ToggledCallSize(nullptr)); + return offset; +} + +void MacroAssemblerMIPSCompat::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 MacroAssemblerMIPSCompat::profilerExitFrame() { + jump(GetJitContext()->runtime->jitRuntime()->getProfilerExitFrameTail()); +} + +void MacroAssembler::subFromStackPtr(Imm32 imm32) { + if (imm32.value) { + asMasm().subPtr(imm32, StackPointer); + } +} + +//{{{ check_macroassembler_style +// =============================================================== +// Stack manipulation functions. + +void MacroAssembler::PushRegsInMask(LiveRegisterSet set) { + int32_t diffF = set.fpus().getPushSizeInBytes(); + int32_t diffG = set.gprs().size() * sizeof(intptr_t); + + reserveStack(diffG); + for (GeneralRegisterBackwardIterator iter(set.gprs()); iter.more(); ++iter) { + diffG -= sizeof(intptr_t); + storePtr(*iter, Address(StackPointer, diffG)); + } + MOZ_ASSERT(diffG == 0); + +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + if (diffF > 0) { + // Double values have to be aligned. We reserve extra space so that we can + // start writing from the first aligned location. + // We reserve a whole extra double so that the buffer has even size. + ma_and(SecondScratchReg, sp, Imm32(~(ABIStackAlignment - 1))); + reserveStack(diffF); + + diffF -= sizeof(double); + + for (FloatRegisterForwardIterator iter(set.fpus().reduceSetForPush()); + iter.more(); ++iter) { + as_sdc1(*iter, SecondScratchReg, -diffF); + diffF -= sizeof(double); + } + + MOZ_ASSERT(diffF == 0); + } +} + +void MacroAssembler::PopRegsInMaskIgnore(LiveRegisterSet set, + LiveRegisterSet ignore) { + int32_t diffG = set.gprs().size() * sizeof(intptr_t); + int32_t diffF = set.fpus().getPushSizeInBytes(); + const int32_t reservedG = diffG; + const int32_t reservedF = diffF; + +#ifdef ENABLE_WASM_SIMD +# error "Needs more careful logic if SIMD is enabled" +#endif + + if (reservedF > 0) { + // Read the buffer form the first aligned location. + ma_addu(SecondScratchReg, sp, Imm32(reservedF)); + ma_and(SecondScratchReg, SecondScratchReg, Imm32(~(ABIStackAlignment - 1))); + + diffF -= sizeof(double); + + LiveFloatRegisterSet fpignore(ignore.fpus().reduceSetForPush()); + for (FloatRegisterForwardIterator iter(set.fpus().reduceSetForPush()); + iter.more(); ++iter) { + if (!ignore.has(*iter)) { + as_ldc1(*iter, SecondScratchReg, -diffF); + } + diffF -= sizeof(double); + } + freeStack(reservedF); + MOZ_ASSERT(diffF == 0); + } + + for (GeneralRegisterBackwardIterator iter(set.gprs()); iter.more(); ++iter) { + diffG -= sizeof(intptr_t); + if (!ignore.has(*iter)) { + loadPtr(Address(StackPointer, diffG), *iter); + } + } + freeStack(reservedG); + MOZ_ASSERT(diffG == 0); +} + +void MacroAssembler::storeRegsInMask(LiveRegisterSet set, Address dest, + Register scratch) { + int32_t diffF = set.fpus().getPushSizeInBytes(); + int32_t diffG = set.gprs().size() * sizeof(intptr_t); + + MOZ_ASSERT(dest.offset >= diffG + diffF); + MOZ_ASSERT(dest.base == StackPointer); + + 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 + + if (diffF > 0) { + computeEffectiveAddress(dest, scratch); + ma_and(scratch, scratch, Imm32(~(ABIStackAlignment - 1))); + + diffF -= sizeof(double); + + for (FloatRegisterForwardIterator iter(set.fpus().reduceSetForPush()); + iter.more(); ++iter) { + as_sdc1(*iter, scratch, -diffF); + diffF -= sizeof(double); + } + MOZ_ASSERT(diffF == 0); + } +} +// =============================================================== +// ABI function calls. + +void MacroAssembler::setupUnalignedABICall(Register scratch) { + MOZ_ASSERT(!IsCompilingWasm(), "wasm should only use aligned ABI calls"); + setupNativeABICall(); + dynamicAlignment_ = true; + + ma_move(scratch, StackPointer); + + // 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::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(*this); + emitter.emit(moveResolver_); + emitter.finish(); + } + + assertStackAlignment(ABIStackAlignment); +} + +void MacroAssembler::callWithABIPost(uint32_t stackAdjust, MoveOp::Type 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, MoveOp::Type result) { + // Load the callee in t9, no instruction between the lw and call + // should clobber it. Note that we can't use fun.base because it may + // be one of the IntArg registers clobbered before the call. + ma_move(t9, fun); + uint32_t stackAdjust; + callWithABIPre(&stackAdjust); + call(t9); + callWithABIPost(stackAdjust, result); +} + +void MacroAssembler::callWithABINoProfiler(const Address& fun, + MoveOp::Type result) { + // Load the callee in t9, as above. + loadPtr(Address(fun.base, fun.offset), t9); + uint32_t stackAdjust; + callWithABIPre(&stackAdjust); + call(t9); + callWithABIPost(stackAdjust, result); +} +// =============================================================== +// Move instructions + +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)) { + mov(ImmWord(MIRTypeToTag(type)), dest.typeReg()); + if (reg.gpr() != dest.payloadReg()) { + move32(reg.gpr(), dest.payloadReg()); + } + return; + } + + ScratchDoubleScope scratch(*this); + 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) { + Register s0 = src.typeReg(); + Register s1 = src.payloadReg(); + Register d0 = dest.typeReg(); + Register d1 = dest.payloadReg(); + + // Either one or both of the source registers could be the same as a + // destination register. + if (s1 == d0) { + if (s0 == d1) { + // If both are, this is just a swap of two registers. + ScratchRegisterScope scratch(*this); + MOZ_ASSERT(d1 != scratch); + MOZ_ASSERT(d0 != scratch); + move32(d1, scratch); + move32(d0, d1); + move32(scratch, d0); + return; + } + // If only one is, copy that source first. + std::swap(s0, s1); + std::swap(d0, d1); + } + + if (s0 != d0) { + move32(s0, d0); + } + if (s1 != d1) { + move32(s1, d1); + } +} + +void MacroAssembler::moveValue(const Value& src, const ValueOperand& dest) { + move32(Imm32(src.toNunboxTag()), dest.typeReg()); + if (src.isGCThing()) { + movePtr(ImmGCPtr(src.toGCThing()), dest.payloadReg()); + } else { + move32(Imm32(src.toNunboxPayload()), dest.payloadReg()); + } +} + +// =============================================================== +// Branch functions + +void MacroAssembler::branchValueIsNurseryCell(Condition cond, + const Address& address, + Register temp, Label* label) { + MOZ_ASSERT(cond == Assembler::Equal || cond == Assembler::NotEqual); + + Label done; + + branchTestGCThing(Assembler::NotEqual, address, + cond == Assembler::Equal ? &done : label); + + loadPtr(address, temp); + branchPtrInNurseryChunk(cond, temp, InvalidReg, label); + + bind(&done); +} + +void MacroAssembler::branchValueIsNurseryCell(Condition cond, + ValueOperand value, Register temp, + Label* label) { + MOZ_ASSERT(cond == Assembler::Equal || cond == Assembler::NotEqual); + + Label done; + + branchTestGCThing(Assembler::NotEqual, value, + cond == Assembler::Equal ? &done : label); + branchPtrInNurseryChunk(cond, value.payloadReg(), temp, label); + + bind(&done); +} + +void MacroAssembler::branchTestValue(Condition cond, const ValueOperand& lhs, + const Value& rhs, Label* label) { + MOZ_ASSERT(cond == Equal || cond == NotEqual); + ScratchRegisterScope scratch(*this); + moveData(rhs, scratch); + + if (cond == Equal) { + Label done; + ma_b(lhs.payloadReg(), scratch, &done, NotEqual, ShortJump); + { ma_b(lhs.typeReg(), Imm32(getType(rhs)), label, Equal); } + bind(&done); + } else { + ma_b(lhs.payloadReg(), scratch, label, NotEqual); + + ma_b(lhs.typeReg(), Imm32(getType(rhs)), label, NotEqual); + } +} + +// ======================================================================== +// Memory access primitives. +template <typename T> +void MacroAssembler::storeUnboxedValue(const ConstantOrRegister& value, + MIRType valueType, const T& dest, + MIRType slotType) { + if (valueType == MIRType::Double) { + storeDouble(value.reg().typedReg().fpu(), dest); + return; + } + + // Store the type tag if needed. + if (valueType != slotType) { + storeTypeTag(ImmType(ValueTypeFromMIRType(valueType)), dest); + } + + // Store the payload. + if (value.constant()) { + storePayload(value.value(), dest); + } else { + storePayload(value.reg().typedReg().gpr(), dest); + } +} + +template void MacroAssembler::storeUnboxedValue(const ConstantOrRegister& value, + MIRType valueType, + const Address& dest, + MIRType slotType); +template void MacroAssembler::storeUnboxedValue( + const ConstantOrRegister& value, MIRType valueType, + const BaseObjectElementIndex& dest, MIRType slotType); + +void MacroAssembler::PushBoxed(FloatRegister reg) { Push(reg); } + +void MacroAssembler::wasmBoundsCheck32(Condition cond, Register index, + Register boundsCheckLimit, + Label* label) { + ma_b(index, boundsCheckLimit, label, cond); +} + +void MacroAssembler::wasmBoundsCheck32(Condition cond, Register index, + Address boundsCheckLimit, Label* label) { + SecondScratchRegisterScope scratch2(*this); + load32(boundsCheckLimit, SecondScratchReg); + ma_b(index, SecondScratchReg, label, cond); +} + +void MacroAssembler::wasmTruncateDoubleToUInt32(FloatRegister input, + Register output, + bool isSaturating, + Label* oolEntry) { + Label done; + + as_truncwd(ScratchFloat32Reg, input); + ma_li(ScratchRegister, Imm32(INT32_MAX)); + moveFromFloat32(ScratchFloat32Reg, output); + + // For numbers in -1.[ : ]INT32_MAX range do nothing more + ma_b(output, ScratchRegister, &done, Assembler::Below, ShortJump); + + loadConstantDouble(double(INT32_MAX + 1ULL), ScratchDoubleReg); + ma_li(ScratchRegister, Imm32(INT32_MIN)); + as_subd(ScratchDoubleReg, input, ScratchDoubleReg); + as_truncwd(ScratchFloat32Reg, ScratchDoubleReg); + as_cfc1(SecondScratchReg, Assembler::FCSR); + moveFromFloat32(ScratchFloat32Reg, output); + ma_ext(SecondScratchReg, SecondScratchReg, Assembler::CauseV, 1); + ma_addu(output, ScratchRegister); + + ma_b(SecondScratchReg, Imm32(0), oolEntry, Assembler::NotEqual); + + bind(&done); +} + +void MacroAssembler::wasmTruncateFloat32ToUInt32(FloatRegister input, + Register output, + bool isSaturating, + Label* oolEntry) { + Label done; + + as_truncws(ScratchFloat32Reg, input); + ma_li(ScratchRegister, Imm32(INT32_MAX)); + moveFromFloat32(ScratchFloat32Reg, output); + // For numbers in -1.[ : ]INT32_MAX range do nothing more + ma_b(output, ScratchRegister, &done, Assembler::Below, ShortJump); + + loadConstantFloat32(float(INT32_MAX + 1ULL), ScratchFloat32Reg); + ma_li(ScratchRegister, Imm32(INT32_MIN)); + as_subs(ScratchFloat32Reg, input, ScratchFloat32Reg); + as_truncws(ScratchFloat32Reg, ScratchFloat32Reg); + as_cfc1(SecondScratchReg, Assembler::FCSR); + moveFromFloat32(ScratchFloat32Reg, output); + ma_ext(SecondScratchReg, SecondScratchReg, Assembler::CauseV, 1); + ma_addu(output, ScratchRegister); + + // Guard against negative values that result in 0 due the precision loss. + as_sltiu(ScratchRegister, output, 1); + ma_or(SecondScratchReg, ScratchRegister); + + ma_b(SecondScratchReg, Imm32(0), oolEntry, Assembler::NotEqual); + + bind(&done); +} + +void MacroAssembler::wasmLoadI64(const wasm::MemoryAccessDesc& access, + Register memoryBase, Register ptr, + Register ptrScratch, Register64 output) { + wasmLoadI64Impl(access, memoryBase, ptr, ptrScratch, output, InvalidReg); +} + +void MacroAssembler::wasmUnalignedLoadI64(const wasm::MemoryAccessDesc& access, + Register memoryBase, Register ptr, + Register ptrScratch, + Register64 output, Register tmp) { + wasmLoadI64Impl(access, memoryBase, ptr, ptrScratch, output, tmp); +} + +void MacroAssembler::wasmStoreI64(const wasm::MemoryAccessDesc& access, + Register64 value, Register memoryBase, + Register ptr, Register ptrScratch) { + wasmStoreI64Impl(access, value, memoryBase, ptr, ptrScratch, InvalidReg); +} + +void MacroAssembler::wasmUnalignedStoreI64(const wasm::MemoryAccessDesc& access, + Register64 value, + Register memoryBase, Register ptr, + Register ptrScratch, Register tmp) { + wasmStoreI64Impl(access, value, memoryBase, ptr, ptrScratch, tmp); +} + +void MacroAssemblerMIPSCompat::wasmLoadI64Impl( + const wasm::MemoryAccessDesc& access, Register memoryBase, Register ptr, + Register ptrScratch, Register64 output, Register tmp) { + uint32_t offset = access.offset(); + MOZ_ASSERT_IF(offset, ptrScratch != InvalidReg); + + MOZ_ASSERT(!access.isZeroExtendSimd128Load()); + MOZ_ASSERT(!access.isSplatSimd128Load()); + MOZ_ASSERT(!access.isWidenSimd128Load()); + + // Maybe add the offset. + if (offset) { + asMasm().movePtr(ptr, ptrScratch); + asMasm().addPtr(Imm32(offset), ptrScratch); + ptr = ptrScratch; + } + + unsigned byteSize = access.byteSize(); + bool isSigned; + + 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; + default: + MOZ_CRASH("unexpected array type"); + } + + BaseIndex address(memoryBase, ptr, TimesOne); + MOZ_ASSERT(INT64LOW_OFFSET == 0); + if (IsUnaligned(access)) { + MOZ_ASSERT(tmp != InvalidReg); + if (byteSize <= 4) { + asMasm().ma_load_unaligned(access, output.low, address, tmp, + static_cast<LoadStoreSize>(8 * byteSize), + isSigned ? SignExtend : ZeroExtend); + if (!isSigned) { + asMasm().move32(Imm32(0), output.high); + } else { + asMasm().ma_sra(output.high, output.low, Imm32(31)); + } + } else { + MOZ_ASSERT(output.low != ptr); + asMasm().ma_load_unaligned(access, output.low, address, tmp, SizeWord, + ZeroExtend); + asMasm().ma_load_unaligned( + access, output.high, + BaseIndex(HeapReg, ptr, TimesOne, INT64HIGH_OFFSET), tmp, SizeWord, + SignExtend); + } + return; + } + + asMasm().memoryBarrierBefore(access.sync()); + if (byteSize <= 4) { + asMasm().ma_load(output.low, address, + static_cast<LoadStoreSize>(8 * byteSize), + isSigned ? SignExtend : ZeroExtend); + asMasm().append(access, asMasm().size() - 4); + if (!isSigned) { + asMasm().move32(Imm32(0), output.high); + } else { + asMasm().ma_sra(output.high, output.low, Imm32(31)); + } + } else { + MOZ_ASSERT(output.low != ptr); + asMasm().ma_load(output.low, BaseIndex(HeapReg, ptr, TimesOne), SizeWord); + asMasm().append(access, asMasm().size() - 4); + asMasm().ma_load(output.high, + BaseIndex(HeapReg, ptr, TimesOne, INT64HIGH_OFFSET), + SizeWord); + asMasm().append(access, asMasm().size() - 4); + } + asMasm().memoryBarrierAfter(access.sync()); +} + +void MacroAssemblerMIPSCompat::wasmStoreI64Impl( + const wasm::MemoryAccessDesc& access, Register64 value, Register memoryBase, + Register ptr, Register ptrScratch, Register tmp) { + uint32_t offset = access.offset(); + MOZ_ASSERT(offset < asMasm().wasmMaxOffsetGuardLimit()); + MOZ_ASSERT_IF(offset, ptrScratch != InvalidReg); + + // Maybe add the offset. + if (offset) { + asMasm().addPtr(Imm32(offset), ptrScratch); + ptr = ptrScratch; + } + + unsigned byteSize = access.byteSize(); + bool isSigned; + 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; + default: + MOZ_CRASH("unexpected array type"); + } + + MOZ_ASSERT(INT64LOW_OFFSET == 0); + BaseIndex address(memoryBase, ptr, TimesOne); + if (IsUnaligned(access)) { + MOZ_ASSERT(tmp != InvalidReg); + if (byteSize <= 4) { + asMasm().ma_store_unaligned(access, value.low, address, tmp, + static_cast<LoadStoreSize>(8 * byteSize), + isSigned ? SignExtend : ZeroExtend); + } else { + asMasm().ma_store_unaligned( + access, value.high, + BaseIndex(HeapReg, ptr, TimesOne, INT64HIGH_OFFSET), tmp, SizeWord, + SignExtend); + asMasm().ma_store_unaligned(access, value.low, address, tmp, SizeWord, + ZeroExtend); + } + return; + } + + asMasm().memoryBarrierBefore(access.sync()); + if (byteSize <= 4) { + asMasm().ma_store(value.low, address, + static_cast<LoadStoreSize>(8 * byteSize)); + asMasm().append(access, asMasm().size() - 4); + } else { + asMasm().ma_store(value.high, + BaseIndex(HeapReg, ptr, TimesOne, INT64HIGH_OFFSET), + SizeWord); + asMasm().append(access, asMasm().size() - 4); + asMasm().ma_store(value.low, address, SizeWord); + } + asMasm().memoryBarrierAfter(access.sync()); +} + +static void EnterAtomic64Region(MacroAssembler& masm, + const wasm::MemoryAccessDesc& access, + Register addr, Register spinlock, + Register scratch) { + masm.movePtr(wasm::SymbolicAddress::js_jit_gAtomic64Lock, spinlock); + + masm.append(access, masm.size()); + masm.as_lbu( + zero, addr, + 7); // Force memory trap on invalid access before we enter the spinlock. + + Label tryLock; + + masm.memoryBarrier(MembarFull); + + masm.bind(&tryLock); + + masm.as_ll(scratch, spinlock, 0); + masm.ma_b(scratch, scratch, &tryLock, Assembler::NonZero, ShortJump); + masm.ma_li(scratch, Imm32(1)); + masm.as_sc(scratch, spinlock, 0); + masm.ma_b(scratch, scratch, &tryLock, Assembler::Zero, ShortJump); + + masm.memoryBarrier(MembarFull); +} + +static void ExitAtomic64Region(MacroAssembler& masm, Register spinlock) { + masm.memoryBarrier(MembarFull); + masm.as_sw(zero, spinlock, 0); + masm.memoryBarrier(MembarFull); +} + +template <typename T> +static void AtomicLoad64(MacroAssembler& masm, + const wasm::MemoryAccessDesc& access, const T& mem, + Register64 temp, Register64 output) { + MOZ_ASSERT(temp.low == InvalidReg && temp.high == InvalidReg); + + masm.computeEffectiveAddress(mem, SecondScratchReg); + + EnterAtomic64Region(masm, access, /* addr= */ SecondScratchReg, + /* spinlock= */ ScratchRegister, + /* scratch= */ output.low); + + masm.load64(Address(SecondScratchReg, 0), output); + + ExitAtomic64Region(masm, /* spinlock= */ ScratchRegister); +} + +void MacroAssembler::wasmAtomicLoad64(const wasm::MemoryAccessDesc& access, + const Address& mem, Register64 temp, + Register64 output) { + AtomicLoad64(*this, access, mem, temp, output); +} + +void MacroAssembler::wasmAtomicLoad64(const wasm::MemoryAccessDesc& access, + const BaseIndex& mem, Register64 temp, + Register64 output) { + AtomicLoad64(*this, access, mem, temp, output); +} + +template <typename T> +void MacroAssemblerMIPSCompat::wasmAtomicStore64( + const wasm::MemoryAccessDesc& access, const T& mem, Register temp, + Register64 value) { + computeEffectiveAddress(mem, SecondScratchReg); + + EnterAtomic64Region(asMasm(), access, /* addr= */ SecondScratchReg, + /* spinlock= */ ScratchRegister, /* scratch= */ temp); + + store64(value, Address(SecondScratchReg, 0)); + + ExitAtomic64Region(asMasm(), /* spinlock= */ ScratchRegister); +} + +template void MacroAssemblerMIPSCompat::wasmAtomicStore64( + const wasm::MemoryAccessDesc& access, const Address& mem, Register temp, + Register64 value); +template void MacroAssemblerMIPSCompat::wasmAtomicStore64( + const wasm::MemoryAccessDesc& access, const BaseIndex& mem, Register temp, + Register64 value); + +template <typename T> +static void WasmCompareExchange64(MacroAssembler& masm, + const wasm::MemoryAccessDesc& access, + const T& mem, Register64 expect, + Register64 replace, Register64 output) { + MOZ_ASSERT(output != expect); + MOZ_ASSERT(output != replace); + + Label exit; + + masm.computeEffectiveAddress(mem, SecondScratchReg); + Address addr(SecondScratchReg, 0); + + EnterAtomic64Region(masm, access, /* addr= */ SecondScratchReg, + /* spinlock= */ ScratchRegister, + /* scratch= */ output.low); + + masm.load64(addr, output); + + masm.ma_b(output.low, expect.low, &exit, Assembler::NotEqual, ShortJump); + masm.ma_b(output.high, expect.high, &exit, Assembler::NotEqual, ShortJump); + masm.store64(replace, addr); + masm.bind(&exit); + ExitAtomic64Region(masm, /* spinlock= */ ScratchRegister); +} + +void MacroAssembler::wasmCompareExchange64(const wasm::MemoryAccessDesc& access, + const Address& mem, + Register64 expect, + Register64 replace, + Register64 output) { + WasmCompareExchange64(*this, access, mem, expect, replace, output); +} + +void MacroAssembler::wasmCompareExchange64(const wasm::MemoryAccessDesc& access, + const BaseIndex& mem, + Register64 expect, + Register64 replace, + Register64 output) { + WasmCompareExchange64(*this, access, mem, expect, replace, output); +} + +template <typename T> +static void WasmAtomicExchange64(MacroAssembler& masm, + const wasm::MemoryAccessDesc& access, + const T& mem, Register64 src, + Register64 output) { + masm.computeEffectiveAddress(mem, SecondScratchReg); + Address addr(SecondScratchReg, 0); + + EnterAtomic64Region(masm, access, /* addr= */ SecondScratchReg, + /* spinlock= */ ScratchRegister, + /* scratch= */ output.low); + + masm.load64(addr, output); + masm.store64(src, addr); + + ExitAtomic64Region(masm, /* spinlock= */ ScratchRegister); +} + +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); +} + +template <typename T> +static void AtomicFetchOp64(MacroAssembler& masm, + const wasm::MemoryAccessDesc& access, AtomicOp op, + Register64 value, const T& mem, Register64 temp, + Register64 output) { + masm.computeEffectiveAddress(mem, SecondScratchReg); + + EnterAtomic64Region(masm, access, /* addr= */ SecondScratchReg, + /* spinlock= */ ScratchRegister, + /* scratch= */ output.low); + + masm.load64(Address(SecondScratchReg, 0), output); + + switch (op) { + case AtomicFetchAddOp: + masm.as_addu(temp.low, output.low, value.low); + masm.as_sltu(temp.high, temp.low, output.low); + masm.as_addu(temp.high, temp.high, output.high); + masm.as_addu(temp.high, temp.high, value.high); + break; + case AtomicFetchSubOp: + masm.as_sltu(temp.high, output.low, value.low); + masm.as_subu(temp.high, output.high, temp.high); + masm.as_subu(temp.low, output.low, value.low); + masm.as_subu(temp.high, temp.high, value.high); + break; + case AtomicFetchAndOp: + masm.as_and(temp.low, output.low, value.low); + masm.as_and(temp.high, output.high, value.high); + break; + case AtomicFetchOrOp: + masm.as_or(temp.low, output.low, value.low); + masm.as_or(temp.high, output.high, value.high); + break; + case AtomicFetchXorOp: + masm.as_xor(temp.low, output.low, value.low); + masm.as_xor(temp.high, output.high, value.high); + break; + default: + MOZ_CRASH(); + } + + masm.store64(temp, Address(SecondScratchReg, 0)); + + ExitAtomic64Region(masm, /* spinlock= */ ScratchRegister); +} + +void MacroAssembler::wasmAtomicFetchOp64(const wasm::MemoryAccessDesc& access, + AtomicOp op, Register64 value, + const Address& mem, Register64 temp, + Register64 output) { + AtomicFetchOp64(*this, access, 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, op, value, mem, temp, output); +} + +// ======================================================================== +// Convert floating point. + +static const double TO_DOUBLE_HIGH_SCALE = 0x100000000; + +bool MacroAssembler::convertUInt64ToDoubleNeedsTemp() { return false; } + +void MacroAssembler::convertUInt64ToDouble(Register64 src, FloatRegister dest, + Register temp) { + MOZ_ASSERT(temp == Register::Invalid()); + convertUInt32ToDouble(src.high, dest); + loadConstantDouble(TO_DOUBLE_HIGH_SCALE, ScratchDoubleReg); + mulDouble(ScratchDoubleReg, dest); + convertUInt32ToDouble(src.low, ScratchDoubleReg); + addDouble(ScratchDoubleReg, dest); +} + +void MacroAssembler::convertInt64ToDouble(Register64 src, FloatRegister dest) { + convertInt32ToDouble(src.high, dest); + loadConstantDouble(TO_DOUBLE_HIGH_SCALE, ScratchDoubleReg); + mulDouble(ScratchDoubleReg, dest); + convertUInt32ToDouble(src.low, ScratchDoubleReg); + addDouble(ScratchDoubleReg, dest); +} + +//}}} check_macroassembler_style diff --git a/js/src/jit/mips32/MacroAssembler-mips32.h b/js/src/jit/mips32/MacroAssembler-mips32.h new file mode 100644 index 0000000000..142fcdc800 --- /dev/null +++ b/js/src/jit/mips32/MacroAssembler-mips32.h @@ -0,0 +1,802 @@ +/* -*- 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_mips32_MacroAssembler_mips32_h +#define jit_mips32_MacroAssembler_mips32_h + +#include "mozilla/EndianUtils.h" + +#include "jit/mips-shared/MacroAssembler-mips-shared.h" +#include "jit/MoveResolver.h" +#include "vm/BytecodeUtil.h" +#include "wasm/WasmTypes.h" + +namespace js { +namespace jit { + +struct ImmTag : public Imm32 { + ImmTag(JSValueTag mask) : Imm32(int32_t(mask)) {} +}; + +struct ImmType : public ImmTag { + ImmType(JSValueType type) : ImmTag(JSVAL_TYPE_TO_TAG(type)) {} +}; + +static constexpr ValueOperand JSReturnOperand{JSReturnReg_Type, + JSReturnReg_Data}; +static const ValueOperand softfpReturnOperand = ValueOperand(v1, v0); + +static const int defaultShift = 3; +static_assert(1 << defaultShift == sizeof(JS::Value), + "The defaultShift is wrong"); + +static const uint32_t LOW_32_MASK = (1LL << 32) - 1; +#if MOZ_LITTLE_ENDIAN() +static const int32_t LOW_32_OFFSET = 0; +static const int32_t HIGH_32_OFFSET = 4; +#else +static const int32_t LOW_32_OFFSET = 4; +static const int32_t HIGH_32_OFFSET = 0; +#endif + +// See documentation for ScratchTagScope and ScratchTagScopeRelease in +// MacroAssembler-x64.h. + +class ScratchTagScope { + const ValueOperand& v_; + + public: + ScratchTagScope(MacroAssembler&, const ValueOperand& v) : v_(v) {} + operator Register() { return v_.typeReg(); } + void release() {} + void reacquire() {} +}; + +class ScratchTagScopeRelease { + public: + explicit ScratchTagScopeRelease(ScratchTagScope*) {} +}; + +class MacroAssemblerMIPS : public MacroAssemblerMIPSShared { + public: + using MacroAssemblerMIPSShared::ma_b; + using MacroAssemblerMIPSShared::ma_cmp_set; + using MacroAssemblerMIPSShared::ma_ld; + using MacroAssemblerMIPSShared::ma_li; + using MacroAssemblerMIPSShared::ma_liPatchable; + using MacroAssemblerMIPSShared::ma_load; + using MacroAssemblerMIPSShared::ma_ls; + using MacroAssemblerMIPSShared::ma_sd; + using MacroAssemblerMIPSShared::ma_ss; + using MacroAssemblerMIPSShared::ma_store; + using MacroAssemblerMIPSShared::ma_subTestOverflow; + + void ma_li(Register dest, CodeLabel* label); + + void ma_li(Register dest, ImmWord imm); + void ma_liPatchable(Register dest, ImmPtr imm); + void ma_liPatchable(Register dest, ImmWord imm); + + // load + void ma_load(Register dest, Address address, LoadStoreSize size = SizeWord, + LoadStoreExtension extension = SignExtend); + + // store + void ma_store(Register data, Address address, LoadStoreSize size = SizeWord, + LoadStoreExtension extension = SignExtend); + + // arithmetic based ops + // add + void ma_addTestOverflow(Register rd, Register rs, Register rt, + Label* overflow); + void ma_addTestOverflow(Register rd, Register rs, Imm32 imm, Label* overflow); + + // subtract + void ma_subTestOverflow(Register rd, Register rs, Register rt, + Label* overflow); + + // memory + // shortcut for when we know we're transferring 32 bits of data + void ma_lw(Register data, Address address); + + void ma_sw(Register data, Address address); + void ma_sw(Imm32 imm, Address address); + void ma_sw(Register data, BaseIndex& address); + + void ma_pop(Register r); + void ma_push(Register r); + + void branchWithCode(InstImm code, Label* label, JumpKind jumpKind); + // branches when done from within mips-specific code + void ma_b(Register lhs, ImmWord imm, Label* l, Condition c, + JumpKind jumpKind = LongJump) { + ma_b(lhs, Imm32(uint32_t(imm.value)), l, c, jumpKind); + } + void ma_b(Address addr, ImmWord imm, Label* l, Condition c, + JumpKind jumpKind = LongJump) { + ma_b(addr, Imm32(uint32_t(imm.value)), 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) { + MOZ_ASSERT(rhs != ScratchRegister); + ma_lw(ScratchRegister, addr); + ma_b(ScratchRegister, rhs, l, c, jumpKind); + } + + void ma_bal(Label* l, DelaySlotFill delaySlotFill = FillDelaySlot); + + // fp instructions + void ma_lid(FloatRegister dest, double value); + + void ma_mv(FloatRegister src, ValueOperand dest); + void ma_mv(ValueOperand src, FloatRegister dest); + + void ma_ls(FloatRegister ft, Address address); + void ma_ld(FloatRegister ft, Address address); + void ma_sd(FloatRegister ft, Address address); + void ma_ss(FloatRegister ft, Address address); + + void ma_ldc1WordAligned(FloatRegister ft, Register base, int32_t off); + void ma_sdc1WordAligned(FloatRegister ft, Register base, int32_t off); + + void ma_pop(FloatRegister f); + void ma_push(FloatRegister f); + + void ma_cmp_set(Register dst, Register lhs, ImmWord imm, Condition c) { + ma_cmp_set(dst, lhs, Imm32(uint32_t(imm.value)), c); + } + void ma_cmp_set(Register dst, Register lhs, ImmPtr imm, Condition c) { + ma_cmp_set(dst, lhs, ImmWord(uintptr_t(imm.value)), c); + } + void ma_cmp_set(Register dst, Register lhs, Address addr, Condition c) { + MOZ_ASSERT(lhs != ScratchRegister); + ma_lw(ScratchRegister, addr); + ma_cmp_set(dst, lhs, ScratchRegister, c); + } + void ma_cmp_set(Register dst, Address lhs, Register rhs, Condition c) { + MOZ_ASSERT(rhs != ScratchRegister); + ma_lw(ScratchRegister, lhs); + ma_cmp_set(dst, ScratchRegister, rhs, c); + } + void ma_cmp_set(Register dst, Address lhs, ImmPtr imm, Condition c) { + ma_lw(SecondScratchReg, lhs); + ma_cmp_set(dst, SecondScratchReg, imm, c); + } + + // These fuctions abstract the access to high part of the double precision + // float register. It is intended to work on both 32 bit and 64 bit + // floating point coprocessor. + // :TODO: (Bug 985881) Modify this for N32 ABI to use mthc1 and mfhc1 + void moveToDoubleHi(Register src, FloatRegister dest) { + as_mtc1(src, getOddPair(dest)); + } + void moveFromDoubleHi(FloatRegister src, Register dest) { + as_mfc1(dest, getOddPair(src)); + } +}; + +class MacroAssembler; + +class MacroAssemblerMIPSCompat : public MacroAssemblerMIPS { + public: + using MacroAssemblerMIPS::call; + + MacroAssemblerMIPSCompat() {} + + void convertBoolToInt32(Register source, Register dest); + void convertInt32ToDouble(Register src, FloatRegister dest); + void convertInt32ToDouble(const Address& src, FloatRegister dest); + void convertInt32ToDouble(const BaseIndex& src, FloatRegister 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 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 computeScaledAddress(const BaseIndex& address, Register dest); + + void computeEffectiveAddress(const Address& address, Register dest) { + ma_addu(dest, address.base, Imm32(address.offset)); + } + + inline void computeEffectiveAddress(const BaseIndex& address, Register dest); + + void j(Label* dest) { ma_b(dest); } + + void mov(Register src, Register dest) { as_ori(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 branch(JitCode* c) { + BufferOffset bo = m_buffer.nextOffset(); + addPendingJump(bo, ImmPtr(c->raw()), RelocationKind::JITCODE); + ma_liPatchable(ScratchRegister, ImmPtr(c->raw())); + as_jr(ScratchRegister); + as_nop(); + } + void branch(const Register reg) { + as_jr(reg); + as_nop(); + } + void nop() { as_nop(); } + void ret() { + ma_pop(ra); + as_jr(ra); + as_nop(); + } + inline void retn(Imm32 n); + void push(Imm32 imm) { + ma_li(ScratchRegister, imm); + ma_push(ScratchRegister); + } + void push(ImmWord imm) { + ma_li(ScratchRegister, imm); + ma_push(ScratchRegister); + } + void push(ImmGCPtr imm) { + ma_li(ScratchRegister, imm); + ma_push(ScratchRegister); + } + void push(const Address& address) { + loadPtr(address, ScratchRegister); + ma_push(ScratchRegister); + } + 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 MIPS 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: MacroAssemblerMIPSCompat::toggledCall + return 4 * sizeof(uint32_t); + } + + CodeOffset pushWithPatch(ImmWord imm) { + CodeOffset label = movWithPatch(imm, ScratchRegister); + ma_push(ScratchRegister); + return label; + } + + CodeOffset movWithPatch(ImmWord imm, Register dest) { + CodeOffset label = CodeOffset(currentOffset()); + ma_liPatchable(dest, imm); + return label; + } + CodeOffset movWithPatch(ImmPtr imm, Register dest) { + return movWithPatch(ImmWord(uintptr_t(imm.value)), dest); + } + + void writeCodePointer(CodeLabel* label) { + BufferOffset off = writeInst(-1); + label->patchAt()->bind(off.getOffset()); + label->setLinkMode(CodeLabel::RawPointer); + } + + void jump(Label* label) { ma_b(label); } + void jump(Register reg) { + as_jr(reg); + as_nop(); + } + void jump(const Address& address) { + loadPtr(address, ScratchRegister); + as_jr(ScratchRegister); + as_nop(); + } + + 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 negl(Register reg) { ma_negu(reg, reg); } + + void splitTagForTest(const ValueOperand& value, ScratchTagScope& tag) { + MOZ_ASSERT(value.typeReg() == tag); + } + + // unboxing code + void unboxNonDouble(const ValueOperand& operand, Register dest, JSValueType); + void unboxNonDouble(const Address& src, Register dest, JSValueType); + void unboxNonDouble(const BaseIndex& src, Register dest, JSValueType); + void unboxInt32(const ValueOperand& operand, Register dest); + void unboxInt32(const Address& src, Register dest); + void unboxBoolean(const ValueOperand& operand, Register dest); + void unboxBoolean(const Address& src, Register dest); + void unboxDouble(const ValueOperand& operand, FloatRegister dest); + void unboxDouble(const Address& src, FloatRegister dest); + void unboxDouble(const BaseIndex& src, FloatRegister dest); + void unboxString(const ValueOperand& operand, Register dest); + void unboxString(const Address& src, Register dest); + void unboxBigInt(const ValueOperand& operand, Register dest); + void unboxBigInt(const Address& src, Register dest); + void unboxObject(const ValueOperand& src, Register dest); + void unboxObject(const Address& src, Register dest); + void unboxObject(const BaseIndex& src, Register dest) { + unboxNonDouble(src, dest, JSVAL_TYPE_OBJECT); + } + void unboxObjectOrNull(const Address& src, Register dest); + void unboxValue(const ValueOperand& src, AnyRegister dest, JSValueType); + + void unboxGCThingForGCBarrier(const Address& src, Register dest) { + unboxObject(src, dest); + } + void unboxGCThingForGCBarrier(const ValueOperand& src, Register dest) { + unboxObject(src, dest); + } + + void notBoolean(const ValueOperand& val) { + as_xori(val.payloadReg(), val.payloadReg(), 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) { + return value.payloadReg(); + } + [[nodiscard]] Register extractString(const ValueOperand& value, + Register scratch) { + return value.payloadReg(); + } + [[nodiscard]] Register extractSymbol(const ValueOperand& value, + Register scratch) { + return value.payloadReg(); + } + [[nodiscard]] Register extractInt32(const ValueOperand& value, + Register scratch) { + return value.payloadReg(); + } + [[nodiscard]] Register extractBoolean(const ValueOperand& value, + Register scratch) { + return value.payloadReg(); + } + [[nodiscard]] Register extractTag(const Address& address, Register scratch); + [[nodiscard]] Register extractTag(const BaseIndex& address, Register scratch); + [[nodiscard]] Register extractTag(const ValueOperand& value, + Register scratch) { + return value.typeReg(); + } + + void boolValueToDouble(const ValueOperand& operand, FloatRegister dest); + void int32ValueToDouble(const ValueOperand& operand, FloatRegister dest); + void loadInt32OrDouble(const Address& address, FloatRegister dest); + void loadInt32OrDouble(Register base, Register index, FloatRegister dest, + int32_t shift = defaultShift); + 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 + Operand ToPayload(Operand base); + Address ToPayload(Address base) { + return ToPayload(Operand(base)).toAddress(); + } + + BaseIndex ToPayload(BaseIndex base) { + return BaseIndex(base.base, base.index, base.scale, + base.offset + NUNBOX32_PAYLOAD_OFFSET); + } + + protected: + Operand ToType(Operand base); + Address ToType(Address base) { return ToType(Operand(base)).toAddress(); } + + uint32_t getType(const Value& val); + void moveData(const Value& val, Register data); + + public: + void moveValue(const Value& val, Register type, Register data); + + void loadUnboxedValue(Address address, MIRType type, AnyRegister dest) { + if (dest.isFloat()) { + loadInt32OrDouble(address, dest.fpu()); + } else { + ma_lw(dest.gpr(), ToPayload(address)); + } + } + + void loadUnboxedValue(BaseIndex address, MIRType type, AnyRegister dest) { + if (dest.isFloat()) { + loadInt32OrDouble(address.base, address.index, dest.fpu(), address.scale); + } else { + load32(ToPayload(address), dest.gpr()); + } + } + + template <typename T> + void storeUnboxedValue(ConstantOrRegister value, MIRType valueType, + const T& dest, MIRType slotType); + + template <typename T> + void storeUnboxedPayload(ValueOperand value, T address, size_t nbytes, + JSValueType) { + switch (nbytes) { + case 4: + store32(value.payloadReg(), address); + return; + case 1: + store8(value.payloadReg(), address); + return; + default: + MOZ_CRASH("Bad payload width"); + } + } + + void moveValue(const Value& val, const ValueOperand& dest); + + void moveValue(const ValueOperand& src, const ValueOperand& dest) { + Register s0 = src.typeReg(), d0 = dest.typeReg(), s1 = src.payloadReg(), + d1 = dest.payloadReg(); + + // Either one or both of the source registers could be the same as a + // destination register. + if (s1 == d0) { + if (s0 == d1) { + // If both are, this is just a swap of two registers. + MOZ_ASSERT(d1 != ScratchRegister); + MOZ_ASSERT(d0 != ScratchRegister); + move32(d1, ScratchRegister); + move32(d0, d1); + move32(ScratchRegister, d0); + return; + } + // If only one is, copy that source first. + std::swap(s0, s1); + std::swap(d0, d1); + } + + if (s0 != d0) { + move32(s0, d0); + } + if (s1 != d1) { + move32(s1, d1); + } + } + + void storeValue(ValueOperand val, Operand dst); + void storeValue(ValueOperand val, const BaseIndex& dest); + void storeValue(JSValueType type, Register reg, BaseIndex dest); + void storeValue(ValueOperand val, const Address& dest); + void storeValue(JSValueType type, Register reg, Address 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) { + load32(ToType(src), temp); + store32(temp, ToType(dest)); + + load32(ToPayload(src), temp); + store32(temp, ToPayload(dest)); + } + + void loadValue(Address src, ValueOperand val); + void loadValue(Operand dest, ValueOperand val) { + loadValue(dest.toAddress(), val); + } + void loadValue(const BaseIndex& addr, 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); +#if MOZ_LITTLE_ENDIAN() + void pushValue(const Value& val) { + push(Imm32(val.toNunboxTag())); + if (val.isGCThing()) { + push(ImmGCPtr(val.toGCThing())); + } else { + push(Imm32(val.toNunboxPayload())); + } + } + void pushValue(JSValueType type, Register reg) { + push(ImmTag(JSVAL_TYPE_TO_TAG(type))); + ma_push(reg); + } +#else + void pushValue(const Value& val) { + if (val.isGCThing()) { + push(ImmGCPtr(val.toGCThing())); + } else { + push(Imm32(val.toNunboxPayload())); + } + push(Imm32(val.toNunboxTag())); + } + void pushValue(JSValueType type, Register reg) { + ma_push(reg); + push(ImmTag(JSVAL_TYPE_TO_TAG(type))); + } +#endif + void pushValue(const Address& addr); + + void storePayload(const Value& val, Address dest); + void storePayload(Register src, Address dest); + void storePayload(const Value& val, const BaseIndex& dest); + void storePayload(Register src, const BaseIndex& dest); + void storeTypeTag(ImmTag tag, Address dest); + void storeTypeTag(ImmTag tag, const BaseIndex& dest); + + void handleFailureWithHandlerTail(Label* profilerExitTail); + + template <typename T> + void wasmAtomicStore64(const wasm::MemoryAccessDesc& access, const T& mem, + Register temp, Register64 value); + + ///////////////////////////////////////////////////////////////// + // 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); + + void load8SignExtend(const Address& address, Register dest); + void load8SignExtend(const BaseIndex& src, Register dest); + + void load8ZeroExtend(const Address& address, Register dest); + void load8ZeroExtend(const BaseIndex& src, Register dest); + + void load16SignExtend(const Address& address, Register dest); + void load16SignExtend(const BaseIndex& src, Register dest); + + template <typename S> + void load16UnalignedSignExtend(const S& src, Register dest) { + ma_load_unaligned(dest, src, SizeHalfWord, SignExtend); + } + + void load16ZeroExtend(const Address& address, Register dest); + void load16ZeroExtend(const BaseIndex& src, Register dest); + + template <typename S> + void load16UnalignedZeroExtend(const S& src, Register dest) { + ma_load_unaligned(dest, src, SizeHalfWord, ZeroExtend); + } + + void load32(const Address& address, Register dest); + void load32(const BaseIndex& address, Register dest); + void load32(AbsoluteAddress address, Register dest); + void load32(wasm::SymbolicAddress address, Register dest); + + template <typename S> + void load32Unaligned(const S& src, Register dest) { + ma_load_unaligned(dest, src); + } + + void load64(const Address& address, Register64 dest) { + load32(LowWord(address), dest.low); + load32(HighWord(address), dest.high); + } + void load64(const BaseIndex& address, Register64 dest) { + load32(LowWord(address), dest.low); + load32(HighWord(address), dest.high); + } + + template <typename S> + void load64Unaligned(const S& src, Register64 dest) { + ma_load_unaligned(dest.low, LowWord(src)); + ma_load_unaligned(dest.high, HighWord(src)); + } + + void loadPtr(const Address& address, Register dest); + void loadPtr(const BaseIndex& src, Register dest); + void loadPtr(AbsoluteAddress address, Register dest); + void loadPtr(wasm::SymbolicAddress address, Register dest); + + void loadPrivate(const Address& address, Register dest); + + void loadUnalignedDouble(const wasm::MemoryAccessDesc& access, + const BaseIndex& src, Register temp, + FloatRegister dest); + + void loadUnalignedFloat32(const wasm::MemoryAccessDesc& access, + const BaseIndex& src, Register temp, + FloatRegister dest); + + void store8(Register src, const Address& address); + void store8(Imm32 imm, const Address& address); + void store8(Register src, const BaseIndex& address); + void store8(Imm32 imm, const BaseIndex& address); + + void store16(Register src, const Address& address); + void store16(Imm32 imm, const Address& address); + void store16(Register src, const BaseIndex& address); + void store16(Imm32 imm, const BaseIndex& address); + + template <typename T> + void store16Unaligned(Register src, const T& dest) { + ma_store_unaligned(src, dest, SizeHalfWord); + } + + void store32(Register src, AbsoluteAddress address); + void store32(Register src, const Address& address); + void store32(Register src, const BaseIndex& address); + void store32(Imm32 src, const Address& address); + void store32(Imm32 src, const BaseIndex& address); + + // NOTE: This will use second scratch on MIPS. 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) { + ma_store_unaligned(src, dest); + } + + void store64(Register64 src, Address address) { + store32(src.low, Address(address.base, address.offset + LOW_32_OFFSET)); + store32(src.high, Address(address.base, address.offset + HIGH_32_OFFSET)); + } + void store64(Register64 src, const BaseIndex& address) { + store32(src.low, Address(address.base, address.offset + LOW_32_OFFSET)); + store32(src.high, Address(address.base, address.offset + HIGH_32_OFFSET)); + } + + void store64(Imm64 imm, Address address) { + store32(imm.low(), Address(address.base, address.offset + LOW_32_OFFSET)); + store32(imm.hi(), Address(address.base, address.offset + HIGH_32_OFFSET)); + } + void store64(Imm64 imm, const BaseIndex& address) { + store32(imm.low(), Address(address.base, address.offset + LOW_32_OFFSET)); + store32(imm.hi(), Address(address.base, address.offset + HIGH_32_OFFSET)); + } + + template <typename T> + void store64Unaligned(Register64 src, const T& dest) { + ma_store_unaligned(src.low, LowWord(dest)); + ma_store_unaligned(src.high, HighWord(dest)); + } + + template <typename T> + void storePtr(ImmWord imm, T address); + template <typename T> + void storePtr(ImmPtr imm, T address); + template <typename T> + void storePtr(ImmGCPtr imm, T address); + void storePtr(Register src, const Address& address); + void storePtr(Register src, const BaseIndex& address); + void storePtr(Register src, AbsoluteAddress dest); + + void storeUnalignedFloat32(const wasm::MemoryAccessDesc& access, + FloatRegister src, Register temp, + const BaseIndex& dest); + void storeUnalignedDouble(const wasm::MemoryAccessDesc& access, + FloatRegister src, Register temp, + const BaseIndex& dest); + + void moveDouble(FloatRegister src, FloatRegister dest) { as_movd(dest, src); } + + void zeroDouble(FloatRegister reg) { + moveToDoubleLo(zero, reg); + moveToDoubleHi(zero, reg); + } + + void breakpoint(); + + void checkStackAlignment(); + + void alignStackPointer(); + void restoreStackPointer(); + static void calculateAlignedStackPointer(void** stackPointer); + + // If source is a double, load it into dest. If source is int32, + // convert it to double. Else, branch to failure. + void ensureDouble(const ValueOperand& source, FloatRegister dest, + Label* failure); + + void cmp64Set(Condition cond, Register64 lhs, Register64 rhs, Register dest); + void cmp64Set(Condition cond, Register64 lhs, Imm64 val, Register dest); + + protected: + bool buildOOLFakeExitFrame(void* fakeReturnAddr); + + void enterAtomic64Region(Register addr, Register spinlock, Register tmp); + void exitAtomic64Region(Register spinlock); + 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); + Condition ma_cmp64(Condition cond, Register64 lhs, Register64 rhs, + Register dest); + Condition ma_cmp64(Condition cond, Register64 lhs, Imm64 val, Register dest); + + public: + void lea(Operand addr, Register dest) { + ma_addu(dest, addr.baseReg(), Imm32(addr.disp())); + } + + void abiret() { + as_jr(ra); + as_nop(); + } + + void ma_storeImm(Imm32 imm, const Address& addr) { ma_sw(imm, addr); } + + void moveFloat32(FloatRegister src, FloatRegister dest) { + as_movs(dest, src); + } + void loadWasmGlobalPtr(uint32_t globalDataOffset, Register dest) { + loadPtr(Address(WasmTlsReg, + offsetof(wasm::TlsData, globalArea) + globalDataOffset), + dest); + } + void loadWasmPinnedRegsFromTls() { + loadPtr(Address(WasmTlsReg, offsetof(wasm::TlsData, memoryBase)), HeapReg); + } + + // Instrumentation for entering and leaving the profiler. + void profilerEnterFrame(Register framePtr, Register scratch); + void profilerExitFrame(); +}; + +typedef MacroAssemblerMIPSCompat MacroAssemblerSpecific; + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_MacroAssembler_mips32_h */ diff --git a/js/src/jit/mips32/MoveEmitter-mips32.cpp b/js/src/jit/mips32/MoveEmitter-mips32.cpp new file mode 100644 index 0000000000..2f52c73899 --- /dev/null +++ b/js/src/jit/mips32/MoveEmitter-mips32.cpp @@ -0,0 +1,152 @@ +/* -*- 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/mips32/MoveEmitter-mips32.h" + +#include "jit/MacroAssembler-inl.h" + +using namespace js; +using namespace js::jit; + +void MoveEmitterMIPS::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()) { + FloatRegister temp = ScratchFloat32Reg; + masm.loadFloat32(getAdjustedAddress(to), temp); + // Since it is uncertain if the load will be aligned or not + // just fill both of them with the same value. + masm.storeFloat32(temp, cycleSlot(slotId, 0)); + masm.storeFloat32(temp, cycleSlot(slotId, 4)); + } else { + // Just always store the largest possible size. + masm.storeDouble(to.floatReg().doubleOverlay(), cycleSlot(slotId, 0)); + } + break; + case MoveOp::DOUBLE: + if (to.isMemory()) { + FloatRegister temp = ScratchDoubleReg; + masm.loadDouble(getAdjustedAddress(to), temp); + masm.storeDouble(temp, cycleSlot(slotId, 0)); + } else { + masm.storeDouble(to.floatReg(), cycleSlot(slotId, 0)); + } + break; + case MoveOp::INT32: + MOZ_ASSERT(sizeof(uintptr_t) == sizeof(int32_t)); + [[fallthrough]]; + case MoveOp::GENERAL: + if (to.isMemory()) { + Register temp = tempReg(); + masm.loadPtr(getAdjustedAddress(to), temp); + masm.storePtr(temp, cycleSlot(0, 0)); + } else { + // Second scratch register should not be moved by MoveEmitter. + MOZ_ASSERT(to.reg() != spilledReg_); + masm.storePtr(to.reg(), cycleSlot(0, 0)); + } + break; + default: + MOZ_CRASH("Unexpected move type"); + } +} + +void MoveEmitterMIPS::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()) { + FloatRegister temp = ScratchFloat32Reg; + masm.loadFloat32(cycleSlot(slotId, 0), temp); + masm.storeFloat32(temp, getAdjustedAddress(to)); + } else { + uint32_t offset = 0; + if (from.floatReg().numAlignedAliased() == 1) { + offset = sizeof(float); + } + masm.loadFloat32(cycleSlot(slotId, offset), to.floatReg()); + } + break; + case MoveOp::DOUBLE: + if (to.isMemory()) { + FloatRegister temp = ScratchDoubleReg; + masm.loadDouble(cycleSlot(slotId, 0), temp); + masm.storeDouble(temp, getAdjustedAddress(to)); + } else { + masm.loadDouble(cycleSlot(slotId, 0), to.floatReg()); + } + break; + case MoveOp::INT32: + MOZ_ASSERT(sizeof(uintptr_t) == sizeof(int32_t)); + [[fallthrough]]; + case MoveOp::GENERAL: + MOZ_ASSERT(slotId == 0); + if (to.isMemory()) { + Register temp = tempReg(); + masm.loadPtr(cycleSlot(0, 0), temp); + masm.storePtr(temp, getAdjustedAddress(to)); + } else { + // Second scratch register should not be moved by MoveEmitter. + MOZ_ASSERT(to.reg() != spilledReg_); + masm.loadPtr(cycleSlot(0, 0), to.reg()); + } + break; + default: + MOZ_CRASH("Unexpected move type"); + } +} + +void MoveEmitterMIPS::emitDoubleMove(const MoveOperand& from, + const MoveOperand& to) { + if (from.isFloatReg()) { + if (to.isFloatReg()) { + masm.moveDouble(from.floatReg(), to.floatReg()); + } else if (to.isGeneralRegPair()) { + // Used for passing double parameter in a2,a3 register pair. + // Two moves are added for one double parameter by + // MacroAssembler::passABIArg + MOZ_ASSERT(to.evenReg() == a2 && to.oddReg() == a3, + "Invalid emitDoubleMove arguments."); + masm.moveFromDoubleLo(from.floatReg(), a2); + masm.moveFromDoubleHi(from.floatReg(), a3); + } else { + MOZ_ASSERT(to.isMemory()); + masm.storeDouble(from.floatReg(), getAdjustedAddress(to)); + } + } else if (to.isFloatReg()) { + MOZ_ASSERT(from.isMemory()); + masm.loadDouble(getAdjustedAddress(from), to.floatReg()); + } else if (to.isGeneralRegPair()) { + // Used for passing double parameter in a2,a3 register pair. + // Two moves are added for one double parameter by + // MacroAssembler::passABIArg + MOZ_ASSERT(from.isMemory()); + MOZ_ASSERT(to.evenReg() == a2 && to.oddReg() == a3, + "Invalid emitDoubleMove arguments."); + masm.loadPtr(getAdjustedAddress(from), a2); + masm.loadPtr( + Address(from.base(), getAdjustedOffset(from) + sizeof(uint32_t)), a3); + } else { + MOZ_ASSERT(from.isMemory()); + MOZ_ASSERT(to.isMemory()); + masm.loadDouble(getAdjustedAddress(from), ScratchDoubleReg); + masm.storeDouble(ScratchDoubleReg, getAdjustedAddress(to)); + } +} diff --git a/js/src/jit/mips32/MoveEmitter-mips32.h b/js/src/jit/mips32/MoveEmitter-mips32.h new file mode 100644 index 0000000000..4a669e9991 --- /dev/null +++ b/js/src/jit/mips32/MoveEmitter-mips32.h @@ -0,0 +1,31 @@ +/* -*- 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_mips32_MoveEmitter_mips32_h +#define jit_mips32_MoveEmitter_mips32_h + +#include "jit/mips-shared/MoveEmitter-mips-shared.h" + +namespace js { +namespace jit { + +class MoveEmitterMIPS : public MoveEmitterMIPSShared { + void emitDoubleMove(const MoveOperand& from, const MoveOperand& to); + void breakCycle(const MoveOperand& from, const MoveOperand& to, + MoveOp::Type type, uint32_t slot); + void completeCycle(const MoveOperand& from, const MoveOperand& to, + MoveOp::Type type, uint32_t slot); + + public: + MoveEmitterMIPS(MacroAssembler& masm) : MoveEmitterMIPSShared(masm) {} +}; + +typedef MoveEmitterMIPS MoveEmitter; + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_MoveEmitter_mips32_h */ diff --git a/js/src/jit/mips32/SharedICRegisters-mips32.h b/js/src/jit/mips32/SharedICRegisters-mips32.h new file mode 100644 index 0000000000..b8686b3521 --- /dev/null +++ b/js/src/jit/mips32/SharedICRegisters-mips32.h @@ -0,0 +1,48 @@ +/* -*- 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_mips32_SharedICRegisters_mips32_h +#define jit_mips32_SharedICRegisters_mips32_h + +#include "jit/mips32/Assembler-mips32.h" +#include "jit/Registers.h" +#include "jit/RegisterSets.h" + +namespace js { +namespace jit { + +static constexpr Register BaselineFrameReg = s5; +static constexpr Register BaselineStackReg = sp; + +static constexpr ValueOperand R0(a3, a2); +static constexpr ValueOperand R1(s7, s6); +static constexpr ValueOperand R2(t7, t6); + +// ICTailCallReg and ICStubReg +// These use registers that are not preserved across calls. +static constexpr Register ICTailCallReg = ra; +static constexpr Register ICStubReg = t5; + +static constexpr Register ExtractTemp0 = InvalidReg; +static constexpr Register ExtractTemp1 = InvalidReg; + +// Register used internally by MacroAssemblerMIPS. +static constexpr Register BaselineSecondScratchReg = SecondScratchReg; + +// Note that ICTailCallReg is actually just the link register. +// In MIPS code emission, we do not clobber ICTailCallReg since we keep +// the return address for calls there. + +// FloatReg0 must be equal to ReturnFloatReg. +static constexpr FloatRegister FloatReg0 = f0; +static constexpr FloatRegister FloatReg1 = f2; +static constexpr FloatRegister FloatReg2 = f4; +static constexpr FloatRegister FloatReg3 = f6; + +} // namespace jit +} // namespace js + +#endif /* jit_mips32_SharedICRegisters_mips32_h */ diff --git a/js/src/jit/mips32/Simulator-mips32.cpp b/js/src/jit/mips32/Simulator-mips32.cpp new file mode 100644 index 0000000000..5efc0f20db --- /dev/null +++ b/js/src/jit/mips32/Simulator-mips32.cpp @@ -0,0 +1,3631 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +// Copyright 2011 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. + +#include "jit/mips32/Simulator-mips32.h" + +#include "mozilla/Casting.h" +#include "mozilla/FloatingPoint.h" +#include "mozilla/Likely.h" +#include "mozilla/MathAlgorithms.h" + +#include <float.h> + +#include "jit/AtomicOperations.h" +#include "jit/mips32/Assembler-mips32.h" +#include "js/UniquePtr.h" +#include "js/Utility.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) + +namespace js { +namespace jit { + +static const Instr kCallRedirInstr = + op_special | MAX_BREAK_CODE << FunctionBits | ff_break; + +// Utils functions. +static bool HaveSameSign(int32_t a, int32_t b) { return ((a ^ b) >= 0); } + +static uint32_t GetFCSRConditionBit(uint32_t cc) { + if (cc == 0) { + return 23; + } else { + return 24 + cc; + } +} + +static const int32_t kRegisterskMaxValue = 0x7fffffff; +static const int32_t kRegisterskMinValue = 0x80000000; + +// ----------------------------------------------------------------------------- +// MIPS assembly various constants. + +class SimInstruction { + public: + enum { + kInstrSize = 4, + // On MIPS PC cannot actually be directly accessed. We behave as if PC was + // always the value of the current instruction being executed. + kPCReadOffset = 0 + }; + + // Get the raw instruction bits. + inline Instr instructionBits() const { + 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) & ((2 << (hi - lo)) - 1); + } + + // Instruction type. + enum Type { kRegisterType, kImmediateType, kJumpType, kUnsupported = -1 }; + + // Get the encoding type of the instruction. + Type instructionType() const; + + // Accessors for the different named fields used in the MIPS encoding. + inline OpcodeField opcodeValue() const { + return static_cast<OpcodeField>( + bits(OpcodeShift + OpcodeBits - 1, OpcodeShift)); + } + + inline int rsValue() const { + MOZ_ASSERT(instructionType() == kRegisterType || + instructionType() == kImmediateType); + return bits(RSShift + RSBits - 1, RSShift); + } + + inline int rtValue() const { + MOZ_ASSERT(instructionType() == kRegisterType || + instructionType() == kImmediateType); + return bits(RTShift + RTBits - 1, RTShift); + } + + inline int rdValue() const { + MOZ_ASSERT(instructionType() == kRegisterType); + return bits(RDShift + RDBits - 1, RDShift); + } + + inline int saValue() const { + MOZ_ASSERT(instructionType() == kRegisterType); + return bits(SAShift + SABits - 1, SAShift); + } + + inline int functionValue() const { + MOZ_ASSERT(instructionType() == kRegisterType || + instructionType() == kImmediateType); + return bits(FunctionShift + FunctionBits - 1, FunctionShift); + } + + inline int fdValue() const { return bits(FDShift + FDBits - 1, FDShift); } + + inline int fsValue() const { return bits(FSShift + FSBits - 1, FSShift); } + + inline int ftValue() const { return bits(FTShift + FTBits - 1, FTShift); } + + inline int frValue() const { return bits(FRShift + FRBits - 1, FRShift); } + + // Float Compare condition code instruction bits. + inline int fcccValue() const { + return bits(FCccShift + FCccBits - 1, FCccShift); + } + + // Float Branch condition code instruction bits. + inline int fbccValue() const { + return bits(FBccShift + FBccBits - 1, FBccShift); + } + + // Float Branch true/false instruction bit. + inline int fbtrueValue() const { + return bits(FBtrueShift + FBtrueBits - 1, FBtrueShift); + } + + // Return the fields at their original place in the instruction encoding. + inline OpcodeField opcodeFieldRaw() const { + return static_cast<OpcodeField>(instructionBits() & OpcodeMask); + } + + inline int rsFieldRaw() const { + MOZ_ASSERT(instructionType() == kRegisterType || + instructionType() == kImmediateType); + return instructionBits() & RSMask; + } + + // Same as above function, but safe to call within instructionType(). + inline int rsFieldRawNoAssert() const { return instructionBits() & RSMask; } + + inline int rtFieldRaw() const { + MOZ_ASSERT(instructionType() == kRegisterType || + instructionType() == kImmediateType); + return instructionBits() & RTMask; + } + + inline int rdFieldRaw() const { + MOZ_ASSERT(instructionType() == kRegisterType); + return instructionBits() & RDMask; + } + + inline int saFieldRaw() const { + MOZ_ASSERT(instructionType() == kRegisterType); + return instructionBits() & SAMask; + } + + inline int functionFieldRaw() const { + return instructionBits() & FunctionMask; + } + + // Get the secondary field according to the opcode. + inline int secondaryValue() const { + OpcodeField op = opcodeFieldRaw(); + switch (op) { + case op_special: + case op_special2: + return functionValue(); + case op_cop1: + return rsValue(); + case op_regimm: + return rtValue(); + default: + return ff_null; + } + } + + inline int32_t imm16Value() const { + MOZ_ASSERT(instructionType() == kImmediateType); + return bits(Imm16Shift + Imm16Bits - 1, Imm16Shift); + } + + inline int32_t imm26Value() const { + MOZ_ASSERT(instructionType() == kJumpType); + return bits(Imm26Shift + Imm26Bits - 1, Imm26Shift); + } + + // Say if the instruction should not be used in a branch delay slot. + bool isForbiddenInBranchDelay() const; + // Say if the instruction 'links'. e.g. jal, bal. + bool isLinkingInstruction() const; + // Say if the instruction is a debugger break/trap. + bool isTrap() const; + + private: + SimInstruction() = delete; + SimInstruction(const SimInstruction& other) = delete; + void operator=(const SimInstruction& other) = delete; +}; + +bool SimInstruction::isForbiddenInBranchDelay() const { + const int op = opcodeFieldRaw(); + switch (op) { + case op_j: + case op_jal: + case op_beq: + case op_bne: + case op_blez: + case op_bgtz: + case op_beql: + case op_bnel: + case op_blezl: + case op_bgtzl: + return true; + case op_regimm: + switch (rtFieldRaw()) { + case rt_bltz: + case rt_bgez: + case rt_bltzal: + case rt_bgezal: + return true; + default: + return false; + }; + break; + case op_special: + switch (functionFieldRaw()) { + case ff_jr: + case ff_jalr: + return true; + default: + return false; + }; + break; + default: + return false; + } +} + +bool SimInstruction::isLinkingInstruction() const { + const int op = opcodeFieldRaw(); + switch (op) { + case op_jal: + return true; + case op_regimm: + switch (rtFieldRaw()) { + case rt_bgezal: + case rt_bltzal: + return true; + default: + return false; + }; + case op_special: + switch (functionFieldRaw()) { + case ff_jalr: + return true; + default: + return false; + }; + default: + return false; + }; +} + +bool SimInstruction::isTrap() const { + if (opcodeFieldRaw() != op_special) { + return false; + } else { + switch (functionFieldRaw()) { + case ff_break: + return instructionBits() != kCallRedirInstr; + case ff_tge: + case ff_tgeu: + case ff_tlt: + case ff_tltu: + case ff_teq: + case ff_tne: + return bits(15, 6) != kWasmTrapCode; + default: + return false; + }; + } +} + +SimInstruction::Type SimInstruction::instructionType() const { + switch (opcodeFieldRaw()) { + case op_special: + switch (functionFieldRaw()) { + case ff_jr: + case ff_jalr: + case ff_break: + case ff_sll: + case ff_srl: + case ff_sra: + case ff_sllv: + case ff_srlv: + case ff_srav: + case ff_mfhi: + case ff_mflo: + case ff_mult: + case ff_multu: + case ff_div: + case ff_divu: + case ff_add: + case ff_addu: + case ff_sub: + case ff_subu: + case ff_and: + case ff_or: + case ff_xor: + case ff_nor: + case ff_slt: + case ff_sltu: + case ff_tge: + case ff_tgeu: + case ff_tlt: + case ff_tltu: + case ff_teq: + case ff_tne: + case ff_movz: + case ff_movn: + case ff_movci: + case ff_sync: + return kRegisterType; + default: + return kUnsupported; + }; + break; + case op_special2: + switch (functionFieldRaw()) { + case ff_mul: + case ff_madd: + case ff_maddu: + case ff_clz: + return kRegisterType; + default: + return kUnsupported; + }; + break; + case op_special3: + switch (functionFieldRaw()) { + case ff_ins: + case ff_ext: + case ff_bshfl: + return kRegisterType; + default: + return kUnsupported; + }; + break; + case op_cop1: // Coprocessor instructions. + switch (rsFieldRawNoAssert()) { + case rs_bc1: // Branch on coprocessor condition. + return kImmediateType; + default: + return kRegisterType; + }; + break; + case op_cop1x: + return kRegisterType; + // 16 bits Immediate type instructions. e.g.: addi dest, src, imm16. + case op_regimm: + case op_beq: + case op_bne: + case op_blez: + case op_bgtz: + case op_addi: + case op_addiu: + case op_slti: + case op_sltiu: + case op_andi: + case op_ori: + case op_xori: + case op_lui: + case op_beql: + case op_bnel: + case op_blezl: + case op_bgtzl: + case op_lb: + case op_lh: + case op_lwl: + case op_lw: + case op_lbu: + case op_lhu: + case op_lwr: + case op_sb: + case op_sh: + case op_swl: + case op_sw: + case op_swr: + case op_lwc1: + case op_ldc1: + case op_swc1: + case op_sdc1: + case op_ll: + case op_sc: + return kImmediateType; + // 26 bits immediate type instructions. e.g.: j imm26. + case op_j: + case op_jal: + return kJumpType; + default: + return kUnsupported; + } + return kUnsupported; +} + +// C/C++ argument slots size. +const int kCArgSlotCount = 4; +const int kCArgsSlotsSize = kCArgSlotCount * SimInstruction::kInstrSize; +const int kBranchReturnOffset = 2 * SimInstruction::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: + AutoLockSimulatorCache() : Base(SimulatorProcess::singleton_->cacheLock_) {} +}; + +mozilla::Atomic<size_t, mozilla::ReleaseAcquire> + SimulatorProcess::ICacheCheckingDisableCount( + 1); // Checking is disabled by default. +SimulatorProcess* SimulatorProcess::singleton_ = nullptr; + +int Simulator::StopSimAt = -1; + +Simulator* Simulator::Create() { + auto sim = MakeUnique<Simulator>(); + if (!sim) { + return nullptr; + } + + if (!sim->init()) { + return nullptr; + } + + char* stopAtStr = getenv("MIPS_SIM_STOP_AT"); + int64_t stopAt; + if (stopAtStr && sscanf(stopAtStr, "%lld", &stopAt) == 1) { + fprintf(stderr, "\nStopping simulation at icount %lld\n", stopAt); + Simulator::StopSimAt = stopAt; + } + + return sim.release(); +} + +void Simulator::Destroy(Simulator* sim) { js_delete(sim); } + +// The MipsDebugger class is used by the simulator while debugging simulated +// code. +class MipsDebugger { + public: + explicit MipsDebugger(Simulator* sim) : sim_(sim) {} + + void stop(SimInstruction* instr); + void debug(); + // Print all registers with a nice formatting. + void printAllRegs(); + void printAllRegsIncludingFPU(); + + private: + // We set the breakpoint code to 0xfffff to easily recognize it. + static const Instr kBreakpointInstr = op_special | ff_break | 0xfffff << 6; + static const Instr kNopInstr = op_special | ff_sll; + + Simulator* sim_; + + int32_t getRegisterValue(int regnum); + int32_t getFPURegisterValueInt(int regnum); + int64_t getFPURegisterValueLong(int regnum); + float getFPURegisterValueFloat(int regnum); + double getFPURegisterValueDouble(int regnum); + bool getValue(const char* desc, int32_t* value); + + // Set or delete a breakpoint. Returns true if successful. + bool setBreakpoint(SimInstruction* breakpc); + bool deleteBreakpoint(SimInstruction* breakpc); + + // Undo and redo all breakpoints. This is needed to bracket disassembly and + // execution to skip past breakpoints when run from the debugger. + void undoBreakpoints(); + void redoBreakpoints(); +}; + +static void UNSUPPORTED() { + printf("Unsupported instruction.\n"); + MOZ_CRASH(); +} + +void MipsDebugger::stop(SimInstruction* instr) { + // Get the stop code. + uint32_t code = instr->bits(25, 6); + // Retrieve the encoded address, which comes just after this stop. + char* msg = + *reinterpret_cast<char**>(sim_->get_pc() + SimInstruction::kInstrSize); + // Update this stop description. + if (!sim_->watchedStops_[code].desc_) { + sim_->watchedStops_[code].desc_ = msg; + } + // Print the stop message and code if it is not the default code. + if (code != kMaxStopCode) { + printf("Simulator hit stop %u: %s\n", code, msg); + } else { + printf("Simulator hit %s\n", msg); + } + sim_->set_pc(sim_->get_pc() + 2 * SimInstruction::kInstrSize); + debug(); +} + +int32_t MipsDebugger::getRegisterValue(int regnum) { + if (regnum == kPCRegister) { + return sim_->get_pc(); + } + return sim_->getRegister(regnum); +} + +int32_t MipsDebugger::getFPURegisterValueInt(int regnum) { + return sim_->getFpuRegister(regnum); +} + +int64_t MipsDebugger::getFPURegisterValueLong(int regnum) { + return sim_->getFpuRegisterLong(regnum); +} + +float MipsDebugger::getFPURegisterValueFloat(int regnum) { + return sim_->getFpuRegisterFloat(regnum); +} + +double MipsDebugger::getFPURegisterValueDouble(int regnum) { + return sim_->getFpuRegisterDouble(regnum); +} + +bool MipsDebugger::getValue(const char* desc, int32_t* value) { + Register reg = Register::FromName(desc); + if (reg != InvalidReg) { + *value = getRegisterValue(reg.code()); + return true; + } + + if (strncmp(desc, "0x", 2) == 0) { + return sscanf(desc, "%x", reinterpret_cast<uint32_t*>(value)) == 1; + } + return sscanf(desc, "%i", value) == 1; +} + +bool MipsDebugger::setBreakpoint(SimInstruction* breakpc) { + // Check if a breakpoint can be set. If not return without any side-effects. + if (sim_->break_pc_ != nullptr) { + return false; + } + + // Set the breakpoint. + sim_->break_pc_ = breakpc; + sim_->break_instr_ = breakpc->instructionBits(); + // Not setting the breakpoint instruction in the code itself. It will be set + // when the debugger shell continues. + return true; +} + +bool MipsDebugger::deleteBreakpoint(SimInstruction* breakpc) { + if (sim_->break_pc_ != nullptr) { + sim_->break_pc_->setInstructionBits(sim_->break_instr_); + } + + sim_->break_pc_ = nullptr; + sim_->break_instr_ = 0; + return true; +} + +void MipsDebugger::undoBreakpoints() { + if (sim_->break_pc_) { + sim_->break_pc_->setInstructionBits(sim_->break_instr_); + } +} + +void MipsDebugger::redoBreakpoints() { + if (sim_->break_pc_) { + sim_->break_pc_->setInstructionBits(kBreakpointInstr); + } +} + +void MipsDebugger::printAllRegs() { + int32_t value; + for (uint32_t i = 0; i < Registers::Total; i++) { + value = getRegisterValue(i); + printf("%3s: 0x%08x %10d ", Registers::GetName(i), value, value); + + if (i % 2) { + printf("\n"); + } + } + printf("\n"); + + value = getRegisterValue(Simulator::LO); + printf(" LO: 0x%08x %10d ", value, value); + value = getRegisterValue(Simulator::HI); + printf(" HI: 0x%08x %10d\n", value, value); + value = getRegisterValue(Simulator::pc); + printf(" pc: 0x%08x\n", value); +} + +void MipsDebugger::printAllRegsIncludingFPU() { + printAllRegs(); + + printf("\n\n"); + // f0, f1, f2, ... f31. + for (uint32_t i = 0; i < FloatRegisters::RegisterIdLimit; i++) { + if (i & 0x1) { + printf("%3s: 0x%08x\tflt: %-8.4g\n", FloatRegisters::GetName(i), + getFPURegisterValueInt(i), getFPURegisterValueFloat(i)); + } else { + printf("%3s: 0x%08x\tflt: %-8.4g\tdbl: %-16.4g\n", + FloatRegisters::GetName(i), getFPURegisterValueInt(i), + getFPURegisterValueFloat(i), getFPURegisterValueDouble(i)); + } + } +} + +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(); +} + +static void DisassembleInstruction(uint32_t pc) { + uint8_t* bytes = reinterpret_cast<uint8_t*>(pc); + char hexbytes[256]; + sprintf(hexbytes, "0x%x 0x%x 0x%x 0x%x", bytes[0], bytes[1], bytes[2], + bytes[3]); + char llvmcmd[1024]; + sprintf(llvmcmd, + "bash -c \"echo -n '%p'; echo '%s' | " + "llvm-mc -disassemble -arch=mipsel -mcpu=mips32r2 | " + "grep -v pure_instructions | grep -v .text\"", + static_cast<void*>(bytes), hexbytes); + if (system(llvmcmd)) { + printf("Cannot disassemble instruction.\n"); + } +} + +void MipsDebugger::debug() { + intptr_t lastPC = -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; + + // Undo all set breakpoints while running in the debugger shell. This will + // make them invisible to all commands. + undoBreakpoints(); + + while (!done && (sim_->get_pc() != Simulator::end_sim_pc)) { + if (lastPC != sim_->get_pc()) { + DisassembleInstruction(sim_->get_pc()); + lastPC = 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()) { + sim_->instructionDecode( + reinterpret_cast<SimInstruction*>(sim_->get_pc())); + } else { + // Allow si to jump over generated breakpoints. + printf("/!\\ Jumping over generated breakpoint.\n"); + sim_->set_pc(sim_->get_pc() + SimInstruction::kInstrSize); + } + } else if ((strcmp(cmd, "c") == 0) || (strcmp(cmd, "cont") == 0)) { + // Execute the one instruction we broke at with breakpoints disabled. + sim_->instructionDecode( + reinterpret_cast<SimInstruction*>(sim_->get_pc())); + // Leave the debugger shell. + done = true; + } else if ((strcmp(cmd, "p") == 0) || (strcmp(cmd, "print") == 0)) { + if (argc == 2) { + int32_t value; + if (strcmp(arg1, "all") == 0) { + printAllRegs(); + } else if (strcmp(arg1, "allf") == 0) { + printAllRegsIncludingFPU(); + } else { + Register reg = Register::FromName(arg1); + FloatRegisters::Code fCode = FloatRegister::FromName(arg1); + if (reg != InvalidReg) { + value = getRegisterValue(reg.code()); + printf("%s: 0x%08x %d \n", arg1, value, value); + } else if (fCode != FloatRegisters::Invalid) { + if (fCode & 0x1) { + printf("%3s: 0x%08x\tflt: %-8.4g\n", + FloatRegisters::GetName(fCode), + getFPURegisterValueInt(fCode), + getFPURegisterValueFloat(fCode)); + } else { + printf("%3s: 0x%08x\tflt: %-8.4g\tdbl: %-16.4g\n", + FloatRegisters::GetName(fCode), + getFPURegisterValueInt(fCode), + getFPURegisterValueFloat(fCode), + getFPURegisterValueDouble(fCode)); + } + } else { + printf("%s unrecognized\n", arg1); + } + } + } else { + printf("print <register> or print <fpu register> single\n"); + } + } else if (strcmp(cmd, "stack") == 0 || strcmp(cmd, "mem") == 0) { + int32_t* cur = nullptr; + int32_t* end = nullptr; + int next_arg = 1; + + if (strcmp(cmd, "stack") == 0) { + cur = reinterpret_cast<int32_t*>(sim_->getRegister(Simulator::sp)); + } else { // Command "mem". + int32_t value; + if (!getValue(arg1, &value)) { + printf("%s unrecognized\n", arg1); + continue; + } + cur = reinterpret_cast<int32_t*>(value); + next_arg++; + } + + int32_t words; + if (argc == next_arg) { + words = 10; + } else { + if (!getValue(argv[next_arg], &words)) { + words = 10; + } + } + end = cur + words; + + while (cur < end) { + printf(" %p: 0x%08x %10d", cur, *cur, *cur); + printf("\n"); + cur++; + } + + } else if ((strcmp(cmd, "disasm") == 0) || (strcmp(cmd, "dpc") == 0) || + (strcmp(cmd, "di") == 0)) { + uint8_t* cur = nullptr; + uint8_t* end = nullptr; + + if (argc == 1) { + cur = reinterpret_cast<uint8_t*>(sim_->get_pc()); + end = cur + (10 * SimInstruction::kInstrSize); + } else if (argc == 2) { + Register reg = Register::FromName(arg1); + if (reg != InvalidReg || strncmp(arg1, "0x", 2) == 0) { + // The argument is an address or a register name. + int32_t value; + if (getValue(arg1, &value)) { + cur = reinterpret_cast<uint8_t*>(value); + // Disassemble 10 instructions at <arg1>. + end = cur + (10 * SimInstruction::kInstrSize); + } + } else { + // The argument is the number of instructions. + int32_t value; + if (getValue(arg1, &value)) { + cur = reinterpret_cast<uint8_t*>(sim_->get_pc()); + // Disassemble <arg1> instructions. + end = cur + (value * SimInstruction::kInstrSize); + } + } + } else { + int32_t value1; + int32_t value2; + if (getValue(arg1, &value1) && getValue(arg2, &value2)) { + cur = reinterpret_cast<uint8_t*>(value1); + end = cur + (value2 * SimInstruction::kInstrSize); + } + } + + while (cur < end) { + DisassembleInstruction(uint32_t(cur)); + cur += SimInstruction::kInstrSize; + } + } else if (strcmp(cmd, "gdb") == 0) { + printf("relinquishing control to gdb\n"); + asm("int $3"); + printf("regaining control from gdb\n"); + } else if (strcmp(cmd, "break") == 0) { + if (argc == 2) { + int32_t value; + if (getValue(arg1, &value)) { + if (!setBreakpoint(reinterpret_cast<SimInstruction*>(value))) { + printf("setting breakpoint failed\n"); + } + } else { + printf("%s unrecognized\n", arg1); + } + } else { + printf("break <address>\n"); + } + } else if (strcmp(cmd, "del") == 0) { + if (!deleteBreakpoint(nullptr)) { + printf("deleting breakpoint failed\n"); + } + } else if (strcmp(cmd, "flags") == 0) { + printf("No flags on MIPS !\n"); + } else if (strcmp(cmd, "stop") == 0) { + int32_t value; + intptr_t stop_pc = sim_->get_pc() - 2 * SimInstruction::kInstrSize; + SimInstruction* stop_instr = reinterpret_cast<SimInstruction*>(stop_pc); + SimInstruction* msg_address = reinterpret_cast<SimInstruction*>( + stop_pc + SimInstruction::kInstrSize); + if ((argc == 2) && (strcmp(arg1, "unstop") == 0)) { + // Remove the current stop. + if (sim_->isStopInstruction(stop_instr)) { + stop_instr->setInstructionBits(kNopInstr); + msg_address->setInstructionBits(kNopInstr); + } else { + printf("Not at debugger stop.\n"); + } + } else 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, "h") == 0) || (strcmp(cmd, "help") == 0)) { + printf("cont\n"); + printf(" continue execution (alias 'c')\n"); + printf("stepi\n"); + printf(" step one instruction (alias 'si')\n"); + printf("print <register>\n"); + printf(" print register content (alias 'p')\n"); + printf(" use register name 'all' to print all registers\n"); + printf("printobject <register>\n"); + printf(" print an object from a register (alias 'po')\n"); + printf("stack [<words>]\n"); + printf(" dump stack content, default dump 10 words)\n"); + printf("mem <address> [<words>]\n"); + printf(" dump memory content, default dump 10 words)\n"); + printf("flags\n"); + printf(" print flags\n"); + printf("disasm [<instructions>]\n"); + printf("disasm [<address/register>]\n"); + printf("disasm [[<address/register>] <instructions>]\n"); + printf(" disassemble code, default is 10 instructions\n"); + printf(" from pc (alias 'di')\n"); + printf("gdb\n"); + printf(" enter gdb\n"); + printf("break <address>\n"); + printf(" set a break point on the address\n"); + printf("del\n"); + printf(" delete the breakpoint\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 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"); + printf(" stop unstop\n"); + printf(" ignore the stop instruction at the current location\n"); + printf(" from now on\n"); + } else { + printf("Unknown command: %s\n", cmd); + } + } + } + + // Add all the breakpoints back to stop execution and enter the debugger + // shell when hit. + redoBreakpoints(); + +#undef COMMAND_SIZE +#undef ARG_SIZE + +#undef STR +#undef XSTR +} + +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), + SimInstruction::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 static_cast<uint32_t>(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 < 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<int32_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_(kCallRedirInstr), + type_(type), + next_(nullptr) { + next_ = SimulatorProcess::redirection(); + if (!SimulatorProcess::ICacheCheckingDisableCount) { + FlushICacheLocked(SimulatorProcess::icache(), addressOfSwiInstruction(), + SimInstruction::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(SimInstruction* 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, int32_t value) { + MOZ_ASSERT((reg >= 0) && (reg < Register::kNumSimuRegisters)); + if (reg == pc) { + pc_modified_ = true; + } + + // Zero register always holds 0. + registers_[reg] = (reg == 0) ? 0 : value; +} + +void Simulator::setFpuRegister(int fpureg, int32_t value) { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + FPUregisters_[fpureg] = value; +} + +void Simulator::setFpuRegisterFloat(int fpureg, float value) { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + *mozilla::BitwiseCast<float*>(&FPUregisters_[fpureg]) = value; +} + +void Simulator::setFpuRegisterDouble(int fpureg, double value) { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters) && + ((fpureg % 2) == 0)); + *mozilla::BitwiseCast<double*>(&FPUregisters_[fpureg]) = value; +} + +// Get the register from the architecture state. This function does handle +// the special case of accessing the PC register. +int32_t Simulator::getRegister(int reg) const { + MOZ_ASSERT((reg >= 0) && (reg < Register::kNumSimuRegisters)); + if (reg == 0) { + return 0; + } + return registers_[reg] + ((reg == pc) ? SimInstruction::kPCReadOffset : 0); +} + +double Simulator::getDoubleFromRegisterPair(int reg) { + MOZ_ASSERT((reg >= 0) && (reg < Register::kNumSimuRegisters) && + ((reg % 2) == 0)); + + double dm_val = 0.0; + // Read the bits from the unsigned integer register_[] array + // into the double precision floating point value and return it. + memcpy(&dm_val, ®isters_[reg], sizeof(dm_val)); + return (dm_val); +} + +int32_t Simulator::getFpuRegister(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + return FPUregisters_[fpureg]; +} + +int64_t Simulator::getFpuRegisterLong(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters) && + ((fpureg % 2) == 0)); + return *mozilla::BitwiseCast<int64_t*>( + const_cast<int32_t*>(&FPUregisters_[fpureg])); +} + +float Simulator::getFpuRegisterFloat(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters)); + return *mozilla::BitwiseCast<float*>( + const_cast<int32_t*>(&FPUregisters_[fpureg])); +} + +double Simulator::getFpuRegisterDouble(int fpureg) const { + MOZ_ASSERT((fpureg >= 0) && + (fpureg < Simulator::FPURegister::kNumFPURegisters) && + ((fpureg % 2) == 0)); + return *mozilla::BitwiseCast<double*>( + const_cast<int32_t*>(&FPUregisters_[fpureg])); +} + +// Runtime FP routines take up to two double arguments and zero +// or one integer arguments. All are constructed here, +// from a0-a3 or f12 and f14. +void Simulator::getFpArgs(double* x, double* y, int32_t* z) { + *x = getFpuRegisterDouble(12); + *y = getFpuRegisterDouble(14); + *z = getRegister(a2); +} + +void Simulator::getFpFromStack(int32_t* stack, double* x) { + MOZ_ASSERT(stack); + MOZ_ASSERT(x); + memcpy(x, stack, sizeof(double)); +} + +void Simulator::setCallResultDouble(double result) { + setFpuRegisterDouble(f0, result); +} + +void Simulator::setCallResultFloat(float result) { + setFpuRegisterFloat(f0, result); +} + +void Simulator::setCallResult(int64_t res) { + setRegister(v0, static_cast<int32_t>(res)); + setRegister(v1, static_cast<int32_t>(res >> 32)); +} + +// Helper functions for setting and testing the FCSR register's bits. +void Simulator::setFCSRBit(uint32_t cc, bool value) { + if (value) { + FCSR_ |= (1 << cc); + } else { + FCSR_ &= ~(1 << cc); + } +} + +bool Simulator::testFCSRBit(uint32_t cc) { return FCSR_ & (1 << cc); } + +// Sets the rounding error codes in FCSR based on the result of the rounding. +// Returns true if the operation was invalid. +bool Simulator::setFCSRRoundError(double original, double rounded) { + bool ret = false; + + setFCSRBit(kFCSRInexactCauseBit, false); + setFCSRBit(kFCSRUnderflowCauseBit, false); + setFCSRBit(kFCSROverflowCauseBit, false); + setFCSRBit(kFCSRInvalidOpCauseBit, false); + + if (!std::isfinite(original) || !std::isfinite(rounded)) { + setFCSRBit(kFCSRInvalidOpFlagBit, true); + setFCSRBit(kFCSRInvalidOpCauseBit, true); + ret = true; + } + + if (original != rounded) { + setFCSRBit(kFCSRInexactFlagBit, true); + setFCSRBit(kFCSRInexactCauseBit, true); + } + + if (rounded < DBL_MIN && rounded > -DBL_MIN && rounded != 0) { + setFCSRBit(kFCSRUnderflowFlagBit, true); + setFCSRBit(kFCSRUnderflowCauseBit, true); + ret = true; + } + + if (rounded > INT_MAX || rounded < INT_MIN) { + setFCSRBit(kFCSROverflowFlagBit, true); + setFCSRBit(kFCSROverflowCauseBit, true); + // The reference is not really clear but it seems this is required: + setFCSRBit(kFCSRInvalidOpFlagBit, true); + setFCSRBit(kFCSRInvalidOpCauseBit, true); + ret = true; + } + + return ret; +} + +// Raw access to the PC register. +void Simulator::set_pc(int32_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. +int32_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; +} + +// MIPS memory instructions (except lwl/r and swl/r) trap on unaligned memory +// access enabling the OS to handle them via trap-and-emulate. +// Note that simulator runs have the runtime system running directly on the host +// system and only generated code is executed in the simulator. +// Since the host is typically IA32 it will not trap on unaligned memory access. +// We assume that that executing correct generated code will not produce +// unaligned memory access, so we explicitly check for address alignment and +// trap. Note that trapping does not occur when executing wasm code, which +// requires that unaligned memory access provides correct result. +int Simulator::readW(uint32_t addr, SimInstruction* instr) { + if (handleWasmSegFault(addr, 4)) { + return -1; + } + + if ((addr & kPointerAlignmentMask) == 0 || + wasm::InCompiledCode(reinterpret_cast<void*>(get_pc()))) { + intptr_t* ptr = reinterpret_cast<intptr_t*>(addr); + return *ptr; + } + printf("Unaligned read at 0x%08x, pc=0x%08" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +void Simulator::writeW(uint32_t addr, int value, SimInstruction* instr) { + if (handleWasmSegFault(addr, 4)) { + return; + } + + if ((addr & kPointerAlignmentMask) == 0 || + wasm::InCompiledCode(reinterpret_cast<void*>(get_pc()))) { + intptr_t* ptr = reinterpret_cast<intptr_t*>(addr); + LLBit_ = false; + *ptr = value; + return; + } + printf("Unaligned write at 0x%08x, pc=0x%08" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); +} + +double Simulator::readD(uint32_t addr, SimInstruction* instr) { + if (handleWasmSegFault(addr, 8)) { + return NAN; + } + + if ((addr & kDoubleAlignmentMask) == 0 || + wasm::InCompiledCode(reinterpret_cast<void*>(get_pc()))) { + double* ptr = reinterpret_cast<double*>(addr); + return *ptr; + } + printf("Unaligned (double) read at 0x%08x, pc=0x%08" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +void Simulator::writeD(uint32_t addr, double value, SimInstruction* instr) { + if (handleWasmSegFault(addr, 8)) { + return; + } + + if ((addr & kDoubleAlignmentMask) == 0 || + wasm::InCompiledCode(reinterpret_cast<void*>(get_pc()))) { + double* ptr = reinterpret_cast<double*>(addr); + LLBit_ = false; + *ptr = value; + return; + } + printf("Unaligned (double) write at 0x%08x, pc=0x%08" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); +} + +uint16_t Simulator::readHU(uint32_t addr, SimInstruction* instr) { + if (handleWasmSegFault(addr, 2)) { + return 0xffff; + } + + if ((addr & 1) == 0 || + wasm::InCompiledCode(reinterpret_cast<void*>(get_pc()))) { + uint16_t* ptr = reinterpret_cast<uint16_t*>(addr); + return *ptr; + } + printf("Unaligned unsigned halfword read at 0x%08x, pc=0x%08" PRIxPTR "\n", + addr, reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +int16_t Simulator::readH(uint32_t addr, SimInstruction* instr) { + if (handleWasmSegFault(addr, 2)) { + return -1; + } + + if ((addr & 1) == 0 || + wasm::InCompiledCode(reinterpret_cast<void*>(get_pc()))) { + int16_t* ptr = reinterpret_cast<int16_t*>(addr); + return *ptr; + } + printf("Unaligned signed halfword read at 0x%08x, pc=0x%08" PRIxPTR "\n", + addr, reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +void Simulator::writeH(uint32_t addr, uint16_t value, SimInstruction* instr) { + if (handleWasmSegFault(addr, 2)) { + return; + } + + if ((addr & 1) == 0 || + wasm::InCompiledCode(reinterpret_cast<void*>(get_pc()))) { + uint16_t* ptr = reinterpret_cast<uint16_t*>(addr); + LLBit_ = false; + *ptr = value; + return; + } + printf("Unaligned unsigned halfword write at 0x%08x, pc=0x%08" PRIxPTR "\n", + addr, reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); +} + +void Simulator::writeH(uint32_t addr, int16_t value, SimInstruction* instr) { + if (handleWasmSegFault(addr, 2)) { + return; + } + + if ((addr & 1) == 0 || + wasm::InCompiledCode(reinterpret_cast<void*>(get_pc()))) { + int16_t* ptr = reinterpret_cast<int16_t*>(addr); + LLBit_ = false; + *ptr = value; + return; + } + printf("Unaligned halfword write at 0x%08x, pc=0x%08" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); +} + +uint32_t Simulator::readBU(uint32_t addr) { + if (handleWasmSegFault(addr, 1)) { + return 0xff; + } + + uint8_t* ptr = reinterpret_cast<uint8_t*>(addr); + return *ptr; +} + +int32_t Simulator::readB(uint32_t addr) { + if (handleWasmSegFault(addr, 1)) { + return -1; + } + + int8_t* ptr = reinterpret_cast<int8_t*>(addr); + return *ptr; +} + +void Simulator::writeB(uint32_t addr, uint8_t value) { + if (handleWasmSegFault(addr, 1)) { + return; + } + + uint8_t* ptr = reinterpret_cast<uint8_t*>(addr); + LLBit_ = false; + *ptr = value; +} + +void Simulator::writeB(uint32_t addr, int8_t value) { + if (handleWasmSegFault(addr, 1)) { + return; + } + + int8_t* ptr = reinterpret_cast<int8_t*>(addr); + LLBit_ = false; + *ptr = value; +} + +int Simulator::loadLinkedW(uint32_t addr, SimInstruction* instr) { + if ((addr & kPointerAlignmentMask) == 0) { + if (handleWasmSegFault(addr, 1)) { + 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 read at 0x%08x, pc=0x%08" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +int Simulator::storeConditionalW(uint32_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%08x, pc=0x%08" PRIxPTR ", expected: 0x%08x\n", + addr, reinterpret_cast<intptr_t>(instr), LLAddr_); + MOZ_CRASH(); + } + + if ((addr & kPointerAlignmentMask) == 0) { + SharedMem<int32_t*> ptr = + SharedMem<int32_t*>::shared(reinterpret_cast<int32_t*>(addr)); + + if (!LLBit_) { + return 0; + } + + LLBit_ = false; + LLAddr_ = 0; + int32_t expected = lastLLValue_; + int32_t old = + AtomicOperations::compareExchangeSeqCst(ptr, expected, int32_t(value)); + return (old == expected) ? 1 : 0; + } + printf("Unaligned SC at 0x%08x, pc=0x%08" PRIxPTR "\n", addr, + reinterpret_cast<intptr_t>(instr)); + MOZ_CRASH(); + return 0; +} + +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%08" PRIxPTR ": %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)(int32_t arg0); +typedef int64_t (*Prototype_General2)(int32_t arg0, int32_t arg1); +typedef int64_t (*Prototype_General3)(int32_t arg0, int32_t arg1, int32_t arg2); +typedef int64_t (*Prototype_General4)(int32_t arg0, int32_t arg1, int32_t arg2, + int32_t arg3); +typedef int64_t (*Prototype_General5)(int32_t arg0, int32_t arg1, int32_t arg2, + int32_t arg3, int32_t arg4); +typedef int64_t (*Prototype_General6)(int32_t arg0, int32_t arg1, int32_t arg2, + int32_t arg3, int32_t arg4, int32_t arg5); +typedef int64_t (*Prototype_General7)(int32_t arg0, int32_t arg1, int32_t arg2, + int32_t arg3, int32_t arg4, int32_t arg5, + int32_t arg6); +typedef int64_t (*Prototype_General8)(int32_t arg0, int32_t arg1, int32_t arg2, + int32_t arg3, int32_t arg4, int32_t arg5, + int32_t arg6, int32_t arg7); +typedef int64_t (*Prototype_GeneralGeneralGeneralInt64)(int32_t arg0, + int32_t arg1, + int32_t arg2, + int64_t arg3); +typedef int64_t (*Prototype_GeneralGeneralInt64Int64)(int32_t arg0, + int32_t arg1, + int64_t arg2, + int64_t arg3); + +typedef double (*Prototype_Double_None)(); +typedef double (*Prototype_Double_Double)(double arg0); +typedef double (*Prototype_Double_Int)(int32_t arg0); +typedef int32_t (*Prototype_Int_Double)(double arg0); +typedef int64_t (*Prototype_Int64_Double)(double arg0); +typedef int32_t (*Prototype_Int_DoubleIntInt)(double arg0, int32_t arg1, + int32_t arg2); +typedef int32_t (*Prototype_Int_IntDoubleIntInt)(int32_t arg0, double arg1, + int32_t arg2, int32_t arg3); +typedef float (*Prototype_Float32_Float32)(float arg0); +typedef int32_t (*Prototype_Int_Float32)(float arg0); +typedef float (*Prototype_Float32_Float32Float32)(float arg0, float arg1); +typedef float (*Prototype_Float32_IntInt)(int arg0, int arg1); + +typedef double (*Prototype_Double_DoubleInt)(double arg0, int32_t arg1); +typedef double (*Prototype_Double_IntInt)(int32_t arg0, int32_t arg1); +typedef double (*Prototype_Double_IntDouble)(int32_t arg0, double arg1); +typedef double (*Prototype_Double_DoubleDouble)(double arg0, double arg1); +typedef int32_t (*Prototype_Int_IntDouble)(int32_t 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); + +static int64_t MakeInt64(int32_t first, int32_t second) { + // Little-endian order. + return ((int64_t)second << 32) | (uint32_t)first; +} + +// Software interrupt instructions are used by the simulator to call into C++. +void Simulator::softwareInterrupt(SimInstruction* instr) { + int32_t func = instr->functionFieldRaw(); + uint32_t code = (func == ff_break) ? instr->bits(25, 6) : -1; + + // We first check if we met a call_rt_redirected. + if (instr->instructionBits() == kCallRedirInstr) { +#if !defined(USES_O32_ABI) + MOZ_CRASH("Only O32 ABI supported."); +#else + Redirection* redirection = Redirection::FromSwiInstruction(instr); + int32_t arg0 = getRegister(a0); + int32_t arg1 = getRegister(a1); + int32_t arg2 = getRegister(a2); + int32_t arg3 = getRegister(a3); + + int32_t* stack_pointer = reinterpret_cast<int32_t*>(getRegister(sp)); + // Args 4 and 5 are on the stack after the reserved space for args 0..3. + int32_t arg4 = stack_pointer[4]; + int32_t arg5 = stack_pointer[5]; + + // This is dodgy but it works because the C entry stubs are never moved. + // See comment in codegen-arm.cc and bug 1242173. + int32_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); + } + + switch (redirection->type()) { + case Args_General0: { + Prototype_General0 target = + reinterpret_cast<Prototype_General0>(external); + int64_t result = target(); + setCallResult(result); + break; + } + case Args_General1: { + Prototype_General1 target = + reinterpret_cast<Prototype_General1>(external); + int64_t result = target(arg0); + setCallResult(result); + break; + } + case Args_General2: { + Prototype_General2 target = + reinterpret_cast<Prototype_General2>(external); + int64_t result = target(arg0, arg1); + setCallResult(result); + break; + } + case Args_General3: { + Prototype_General3 target = + reinterpret_cast<Prototype_General3>(external); + int64_t result = target(arg0, arg1, arg2); + setCallResult(result); + break; + } + case Args_General4: { + Prototype_General4 target = + reinterpret_cast<Prototype_General4>(external); + int64_t result = target(arg0, arg1, arg2, arg3); + setCallResult(result); + break; + } + case Args_General5: { + Prototype_General5 target = + reinterpret_cast<Prototype_General5>(external); + int64_t result = target(arg0, arg1, arg2, arg3, arg4); + setCallResult(result); + break; + } + case Args_General6: { + Prototype_General6 target = + reinterpret_cast<Prototype_General6>(external); + int64_t result = target(arg0, arg1, arg2, arg3, arg4, arg5); + setCallResult(result); + break; + } + case Args_General7: { + Prototype_General7 target = + reinterpret_cast<Prototype_General7>(external); + int32_t arg6 = stack_pointer[6]; + int64_t result = target(arg0, arg1, arg2, arg3, arg4, arg5, arg6); + setCallResult(result); + break; + } + case Args_General8: { + Prototype_General8 target = + reinterpret_cast<Prototype_General8>(external); + int32_t arg6 = stack_pointer[6]; + int32_t arg7 = stack_pointer[7]; + int64_t result = target(arg0, arg1, arg2, arg3, arg4, arg5, arg6, arg7); + setCallResult(result); + break; + } + case Args_Double_None: { + Prototype_Double_None target = + reinterpret_cast<Prototype_Double_None>(external); + double dresult = target(); + setCallResultDouble(dresult); + break; + } + case Args_Int_Double: { + double dval0, dval1; + int32_t ival; + getFpArgs(&dval0, &dval1, &ival); + Prototype_Int_Double target = + reinterpret_cast<Prototype_Int_Double>(external); + int32_t res = target(dval0); + setRegister(v0, res); + break; + } + case Args_Int_GeneralGeneralGeneralInt64: { + Prototype_GeneralGeneralGeneralInt64 target = + reinterpret_cast<Prototype_GeneralGeneralGeneralInt64>(external); + // The int64 arg is not split across register and stack + int64_t result = target(arg0, arg1, arg2, MakeInt64(arg4, arg5)); + setCallResult(result); + break; + } + case Args_Int_GeneralGeneralInt64Int64: { + Prototype_GeneralGeneralInt64Int64 target = + reinterpret_cast<Prototype_GeneralGeneralInt64Int64>(external); + int64_t result = + target(arg0, arg1, MakeInt64(arg2, arg3), MakeInt64(arg4, arg5)); + setCallResult(result); + break; + } + case Args_Int64_Double: { + double dval0, dval1; + int32_t ival; + getFpArgs(&dval0, &dval1, &ival); + Prototype_Int64_Double target = + reinterpret_cast<Prototype_Int64_Double>(external); + int64_t result = target(dval0); + setCallResult(result); + break; + } + case Args_Int_DoubleIntInt: { + double dval = getFpuRegisterDouble(12); + Prototype_Int_DoubleIntInt target = + reinterpret_cast<Prototype_Int_DoubleIntInt>(external); + int32_t res = target(dval, arg2, arg3); + setRegister(v0, res); + break; + } + case Args_Int_IntDoubleIntInt: { + double dval = getDoubleFromRegisterPair(a2); + Prototype_Int_IntDoubleIntInt target = + reinterpret_cast<Prototype_Int_IntDoubleIntInt>(external); + int32_t res = target(arg0, dval, arg4, arg5); + setRegister(v0, res); + break; + } + case Args_Double_Double: { + double dval0, dval1; + int32_t ival; + getFpArgs(&dval0, &dval1, &ival); + Prototype_Double_Double target = + reinterpret_cast<Prototype_Double_Double>(external); + double dresult = target(dval0); + setCallResultDouble(dresult); + break; + } + case Args_Float32_Float32: { + float fval0; + fval0 = getFpuRegisterFloat(12); + Prototype_Float32_Float32 target = + reinterpret_cast<Prototype_Float32_Float32>(external); + float fresult = target(fval0); + setCallResultFloat(fresult); + break; + } + case Args_Int_Float32: { + float fval0; + fval0 = getFpuRegisterFloat(12); + Prototype_Int_Float32 target = + reinterpret_cast<Prototype_Int_Float32>(external); + int32_t result = target(fval0); + setRegister(v0, result); + break; + } + case Args_Float32_Float32Float32: { + float fval0; + float fval1; + fval0 = getFpuRegisterFloat(12); + fval1 = getFpuRegisterFloat(14); + Prototype_Float32_Float32Float32 target = + reinterpret_cast<Prototype_Float32_Float32Float32>(external); + float fresult = target(fval0, fval1); + setCallResultFloat(fresult); + break; + } + case Args_Float32_IntInt: { + Prototype_Float32_IntInt target = + reinterpret_cast<Prototype_Float32_IntInt>(external); + float fresult = target(arg0, arg1); + setCallResultFloat(fresult); + break; + } + case Args_Double_Int: { + Prototype_Double_Int target = + reinterpret_cast<Prototype_Double_Int>(external); + double dresult = target(arg0); + setCallResultDouble(dresult); + break; + } + case Args_Double_IntInt: { + Prototype_Double_IntInt target = + reinterpret_cast<Prototype_Double_IntInt>(external); + double dresult = target(arg0, arg1); + setCallResultDouble(dresult); + break; + } + case Args_Double_DoubleInt: { + double dval0, dval1; + int32_t ival; + getFpArgs(&dval0, &dval1, &ival); + Prototype_Double_DoubleInt target = + reinterpret_cast<Prototype_Double_DoubleInt>(external); + double dresult = target(dval0, ival); + setCallResultDouble(dresult); + break; + } + case Args_Double_DoubleDouble: { + double dval0, dval1; + int32_t ival; + getFpArgs(&dval0, &dval1, &ival); + Prototype_Double_DoubleDouble target = + reinterpret_cast<Prototype_Double_DoubleDouble>(external); + double dresult = target(dval0, dval1); + setCallResultDouble(dresult); + break; + } + case Args_Double_IntDouble: { + int32_t ival = getRegister(a0); + double dval0 = getDoubleFromRegisterPair(a2); + Prototype_Double_IntDouble target = + reinterpret_cast<Prototype_Double_IntDouble>(external); + double dresult = target(ival, dval0); + setCallResultDouble(dresult); + break; + } + case Args_Int_IntDouble: { + int32_t ival = getRegister(a0); + double dval0 = getDoubleFromRegisterPair(a2); + Prototype_Int_IntDouble target = + reinterpret_cast<Prototype_Int_IntDouble>(external); + int32_t result = target(ival, dval0); + setRegister(v0, result); + break; + } + case Args_Double_DoubleDoubleDouble: { + double dval0, dval1, dval2; + int32_t ival; + getFpArgs(&dval0, &dval1, &ival); + // the last argument is on stack + getFpFromStack(stack_pointer + 4, &dval2); + Prototype_Double_DoubleDoubleDouble target = + reinterpret_cast<Prototype_Double_DoubleDoubleDouble>(external); + double dresult = target(dval0, dval1, dval2); + setCallResultDouble(dresult); + break; + } + case Args_Double_DoubleDoubleDoubleDouble: { + double dval0, dval1, dval2, dval3; + int32_t ival; + getFpArgs(&dval0, &dval1, &ival); + // the two last arguments are on stack + getFpFromStack(stack_pointer + 4, &dval2); + getFpFromStack(stack_pointer + 6, &dval3); + Prototype_Double_DoubleDoubleDoubleDouble target = + reinterpret_cast<Prototype_Double_DoubleDoubleDoubleDouble>( + external); + double dresult = target(dval0, dval1, dval2, dval3); + setCallResultDouble(dresult); + break; + } + default: + MOZ_CRASH("call"); + } + + if (single_stepping_) { + single_step_callback_(single_step_callback_arg_, this, nullptr); + } + + setRegister(ra, saved_ra); + set_pc(getRegister(ra)); +#endif + } else if (func == ff_break && code <= kMaxStopCode) { + if (isWatchpoint(code)) { + printWatchpoint(code); + } else { + increaseStopCounter(code); + handleStop(code, instr); + } + } else { + switch (func) { + case ff_tge: + case ff_tgeu: + case ff_tlt: + case ff_tltu: + case ff_teq: + case ff_tne: + if (instr->bits(15, 6) == kWasmTrapCode) { + uint8_t* newPC; + if (wasm::HandleIllegalInstruction(registerState(), &newPC)) { + set_pc(int32_t(newPC)); + return; + } + } + }; + // All remaining break_ codes, and all traps are handled here. + MipsDebugger dbg(this); + dbg.debug(); + } +} + +// Stop helper functions. +bool Simulator::isWatchpoint(uint32_t code) { + return (code <= kMaxWatchpointCode); +} + +void Simulator::printWatchpoint(uint32_t code) { + MipsDebugger dbg(this); + ++break_count_; + printf( + "\n---- break %d marker: %3d (instr count: %8d) ----------" + "----------------------------------", + code, break_count_, icount_); + dbg.printAllRegs(); // Print registers and continue running. +} + +void Simulator::handleStop(uint32_t code, SimInstruction* instr) { + // Stop if it is enabled, otherwise go on jumping over the stop + // and the message address. + if (isEnabledStop(code)) { + MipsDebugger dbg(this); + dbg.stop(instr); + } else { + set_pc(get_pc() + 2 * SimInstruction::kInstrSize); + } +} + +bool Simulator::isStopInstruction(SimInstruction* instr) { + int32_t func = instr->functionFieldRaw(); + uint32_t code = static_cast<uint32_t>(instr->bits(25, 6)); + return (func == ff_break) && code > kMaxWatchpointCode && + 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::signalExceptions() { + for (int i = 1; i < kNumExceptions; i++) { + if (exceptions[i] != 0) { + MOZ_CRASH("Error: Exception raised."); + } + } +} + +// Handle execution based on instruction types. +void Simulator::configureTypeRegister(SimInstruction* instr, int32_t& alu_out, + int64_t& i64hilo, uint64_t& u64hilo, + int32_t& next_pc, + int32_t& return_addr_reg, + bool& do_interrupt) { + // Every local variable declared here needs to be const. + // This is to make sure that changed values are sent back to + // decodeTypeRegister correctly. + + // Instruction fields. + const OpcodeField op = instr->opcodeFieldRaw(); + const int32_t rs_reg = instr->rsValue(); + const int32_t rs = getRegister(rs_reg); + const uint32_t rs_u = static_cast<uint32_t>(rs); + const int32_t rt_reg = instr->rtValue(); + const int32_t rt = getRegister(rt_reg); + const uint32_t rt_u = static_cast<uint32_t>(rt); + const int32_t rd_reg = instr->rdValue(); + const uint32_t sa = instr->saValue(); + + const int32_t fs_reg = instr->fsValue(); + + // ---------- Configuration. + switch (op) { + case op_cop1: // Coprocessor instructions. + switch (instr->rsFieldRaw()) { + case rs_bc1: // Handled in DecodeTypeImmed, should never come here. + MOZ_CRASH(); + break; + case rs_cfc1: + // At the moment only FCSR is supported. + MOZ_ASSERT(fs_reg == kFCSRRegister); + alu_out = FCSR_; + break; + case rs_mfc1: + alu_out = getFpuRegister(fs_reg); + break; + case rs_mfhc1: + MOZ_CRASH(); + break; + case rs_ctc1: + case rs_mtc1: + case rs_mthc1: + // Do the store in the execution step. + break; + case rs_s: + case rs_d: + case rs_w: + case rs_l: + case rs_ps: + // Do everything in the execution step. + break; + default: + MOZ_CRASH(); + } + break; + case op_cop1x: + break; + case op_special: + switch (instr->functionFieldRaw()) { + case ff_jr: + case ff_jalr: + next_pc = getRegister(instr->rsValue()); + return_addr_reg = instr->rdValue(); + break; + case ff_sll: + alu_out = rt << sa; + break; + case ff_srl: + if (rs_reg == 0) { + // Regular logical right shift of a word by a fixed number of + // bits instruction. RS field is always equal to 0. + alu_out = rt_u >> sa; + } else { + // Logical right-rotate of a word by a fixed number of bits. This + // is special case of SRL instruction, added in MIPS32 Release 2. + // RS field is equal to 00001. + alu_out = (rt_u >> sa) | (rt_u << (32 - sa)); + } + break; + case ff_sra: + alu_out = rt >> sa; + break; + case ff_sllv: + alu_out = rt << rs; + break; + case ff_srlv: + if (sa == 0) { + // Regular logical right-shift of a word by a variable number of + // bits instruction. SA field is always equal to 0. + alu_out = rt_u >> rs; + } else { + // Logical right-rotate of a word by a variable number of bits. + // This is special case od SRLV instruction, added in MIPS32 + // Release 2. SA field is equal to 00001. + alu_out = (rt_u >> rs_u) | (rt_u << (32 - rs_u)); + } + break; + case ff_srav: + alu_out = rt >> rs; + break; + case ff_mfhi: + alu_out = getRegister(HI); + break; + case ff_mflo: + alu_out = getRegister(LO); + break; + case ff_mult: + i64hilo = static_cast<int64_t>(rs) * static_cast<int64_t>(rt); + break; + case ff_multu: + u64hilo = static_cast<uint64_t>(rs_u) * static_cast<uint64_t>(rt_u); + break; + case ff_add: + if (HaveSameSign(rs, rt)) { + if (rs > 0) { + exceptions[kIntegerOverflow] = rs > (kRegisterskMaxValue - rt); + } else if (rs < 0) { + exceptions[kIntegerUnderflow] = rs < (kRegisterskMinValue - rt); + } + } + alu_out = rs + rt; + break; + case ff_addu: + alu_out = rs + rt; + break; + case ff_sub: + if (!HaveSameSign(rs, rt)) { + if (rs > 0) { + exceptions[kIntegerOverflow] = rs > (kRegisterskMaxValue + rt); + } else if (rs < 0) { + exceptions[kIntegerUnderflow] = rs < (kRegisterskMinValue + rt); + } + } + alu_out = rs - rt; + break; + case ff_subu: + alu_out = rs - rt; + break; + case ff_and: + alu_out = rs & rt; + break; + case ff_or: + alu_out = rs | rt; + break; + case ff_xor: + alu_out = rs ^ rt; + break; + case ff_nor: + alu_out = ~(rs | rt); + break; + case ff_slt: + alu_out = rs < rt ? 1 : 0; + break; + case ff_sltu: + alu_out = rs_u < rt_u ? 1 : 0; + break; + // Break and trap instructions. + case ff_break: + do_interrupt = true; + break; + case ff_tge: + do_interrupt = rs >= rt; + break; + case ff_tgeu: + do_interrupt = rs_u >= rt_u; + break; + case ff_tlt: + do_interrupt = rs < rt; + break; + case ff_tltu: + do_interrupt = rs_u < rt_u; + break; + case ff_teq: + do_interrupt = rs == rt; + break; + case ff_tne: + do_interrupt = rs != rt; + break; + case ff_movn: + case ff_movz: + case ff_movci: + case ff_sync: + // No action taken on decode. + break; + case ff_div: + case ff_divu: + // div and divu never raise exceptions. + break; + default: + MOZ_CRASH(); + } + break; + case op_special2: + switch (instr->functionFieldRaw()) { + case ff_mul: + alu_out = rs_u * rt_u; // Only the lower 32 bits are kept. + break; + case ff_mult: + i64hilo = static_cast<int64_t>(rs) * static_cast<int64_t>(rt); + break; + case ff_multu: + u64hilo = static_cast<uint64_t>(rs_u) * static_cast<uint64_t>(rt_u); + break; + case ff_madd: + i64hilo += static_cast<int64_t>(rs) * static_cast<int64_t>(rt); + break; + case ff_maddu: + u64hilo += static_cast<uint64_t>(rs_u) * static_cast<uint64_t>(rt_u); + break; + case ff_clz: + alu_out = rs_u ? __builtin_clz(rs_u) : 32; + break; + default: + MOZ_CRASH(); + } + break; + case op_special3: + switch (instr->functionFieldRaw()) { + case ff_ins: { // Mips32r2 instruction. + // Interpret rd field as 5-bit msb of insert. + uint16_t msb = rd_reg; + // Interpret sa field as 5-bit lsb of insert. + uint16_t lsb = sa; + uint16_t size = msb - lsb + 1; + uint32_t mask = (1 << size) - 1; + alu_out = (rt_u & ~(mask << lsb)) | ((rs_u & mask) << lsb); + break; + } + case ff_ext: { // Mips32r2 instruction. + // Interpret rd field as 5-bit msb of extract. + uint16_t msb = rd_reg; + // Interpret sa field as 5-bit lsb of extract. + uint16_t lsb = sa; + uint16_t size = msb + 1; + uint32_t mask = (1 << size) - 1; + alu_out = (rs_u & (mask << lsb)) >> lsb; + break; + } + case ff_bshfl: { // Mips32r2 instruction. + if (16 == sa) { // seb + alu_out = I32(I8(rt)); + } else if (24 == sa) { // seh + alu_out = I32(I16(rt)); + } else { + MOZ_CRASH(); + } + break; + } + default: + MOZ_CRASH(); + } + break; + default: + MOZ_CRASH(); + } +} + +void Simulator::decodeTypeRegister(SimInstruction* instr) { + // Instruction fields. + const OpcodeField op = instr->opcodeFieldRaw(); + const int32_t rs_reg = instr->rsValue(); + const int32_t rs = getRegister(rs_reg); + const uint32_t rs_u = static_cast<uint32_t>(rs); + const int32_t rt_reg = instr->rtValue(); + const int32_t rt = getRegister(rt_reg); + const uint32_t rt_u = static_cast<uint32_t>(rt); + const int32_t rd_reg = instr->rdValue(); + + const int32_t fr_reg = instr->frValue(); + const int32_t fs_reg = instr->fsValue(); + const int32_t ft_reg = instr->ftValue(); + const int32_t fd_reg = instr->fdValue(); + int64_t i64hilo = 0; + uint64_t u64hilo = 0; + + // ALU output. + // It should not be used as is. Instructions using it should always + // initialize it first. + int32_t alu_out = 0x12345678; + + // For break and trap instructions. + bool do_interrupt = false; + + // For jr and jalr. + // Get current pc. + int32_t current_pc = get_pc(); + // Next pc + int32_t next_pc = 0; + int32_t return_addr_reg = 31; + + // Set up the variables if needed before executing the instruction. + configureTypeRegister(instr, alu_out, i64hilo, u64hilo, next_pc, + return_addr_reg, do_interrupt); + + // ---------- Raise exceptions triggered. + signalExceptions(); + + // ---------- Execution. + switch (op) { + case op_cop1: + switch (instr->rsFieldRaw()) { + case rs_bc1: // Branch on coprocessor condition. + MOZ_CRASH(); + break; + case rs_cfc1: + case rs_mfc1: + setRegister(rt_reg, alu_out); + break; + case rs_mfhc1: + MOZ_CRASH(); + break; + case rs_ctc1: + // At the moment only FCSR is supported. + MOZ_ASSERT(fs_reg == kFCSRRegister); + FCSR_ = registers_[rt_reg]; + break; + case rs_mtc1: + FPUregisters_[fs_reg] = registers_[rt_reg]; + break; + case rs_mthc1: + MOZ_CRASH(); + break; + case rs_s: + float f, ft_value, fs_value; + uint32_t cc, fcsr_cc; + fs_value = getFpuRegisterFloat(fs_reg); + ft_value = getFpuRegisterFloat(ft_reg); + cc = instr->fcccValue(); + fcsr_cc = GetFCSRConditionBit(cc); + switch (instr->functionFieldRaw()) { + case ff_add_fmt: + setFpuRegisterFloat(fd_reg, fs_value + ft_value); + break; + case ff_sub_fmt: + setFpuRegisterFloat(fd_reg, fs_value - ft_value); + break; + case ff_mul_fmt: + setFpuRegisterFloat(fd_reg, fs_value * ft_value); + break; + case ff_div_fmt: + setFpuRegisterFloat(fd_reg, fs_value / ft_value); + break; + case ff_abs_fmt: + setFpuRegisterFloat(fd_reg, fabsf(fs_value)); + break; + case ff_mov_fmt: + setFpuRegisterFloat(fd_reg, fs_value); + break; + case ff_neg_fmt: + setFpuRegisterFloat(fd_reg, -fs_value); + break; + case ff_sqrt_fmt: + setFpuRegisterFloat(fd_reg, sqrtf(fs_value)); + break; + case ff_c_un_fmt: + setFCSRBit(fcsr_cc, + mozilla::IsNaN(fs_value) || mozilla::IsNaN(ft_value)); + break; + case ff_c_eq_fmt: + setFCSRBit(fcsr_cc, (fs_value == ft_value)); + break; + case ff_c_ueq_fmt: + setFCSRBit(fcsr_cc, + (fs_value == ft_value) || (mozilla::IsNaN(fs_value) || + mozilla::IsNaN(ft_value))); + break; + case ff_c_olt_fmt: + setFCSRBit(fcsr_cc, (fs_value < ft_value)); + break; + case ff_c_ult_fmt: + setFCSRBit(fcsr_cc, + (fs_value < ft_value) || (mozilla::IsNaN(fs_value) || + mozilla::IsNaN(ft_value))); + break; + case ff_c_ole_fmt: + setFCSRBit(fcsr_cc, (fs_value <= ft_value)); + break; + case ff_c_ule_fmt: + setFCSRBit(fcsr_cc, + (fs_value <= ft_value) || (mozilla::IsNaN(fs_value) || + mozilla::IsNaN(ft_value))); + break; + case ff_cvt_d_fmt: + f = getFpuRegisterFloat(fs_reg); + setFpuRegisterDouble(fd_reg, static_cast<double>(f)); + break; + case ff_cvt_w_fmt: // Convert float to word. + // Rounding modes are not yet supported. + MOZ_ASSERT((FCSR_ & 3) == 0); + // In rounding mode 0 it should behave like ROUND. + [[fallthrough]]; + case ff_round_w_fmt: { // Round double to word (round half to + // even). + float rounded = std::floor(fs_value + 0.5); + int32_t result = static_cast<int32_t>(rounded); + if ((result & 1) != 0 && result - fs_value == 0.5) { + // If the number is halfway between two integers, + // round to the even one. + result--; + } + setFpuRegister(fd_reg, result); + if (setFCSRRoundError(fs_value, rounded)) { + setFpuRegister(fd_reg, kFPUInvalidResult); + } + break; + } + case ff_trunc_w_fmt: { // Truncate float to word (round towards 0). + float rounded = truncf(fs_value); + int32_t result = static_cast<int32_t>(rounded); + setFpuRegister(fd_reg, result); + if (setFCSRRoundError(fs_value, rounded)) { + setFpuRegister(fd_reg, kFPUInvalidResult); + } + break; + } + case ff_floor_w_fmt: { // Round float to word towards negative + // infinity. + float rounded = std::floor(fs_value); + int32_t result = static_cast<int32_t>(rounded); + setFpuRegister(fd_reg, result); + if (setFCSRRoundError(fs_value, rounded)) { + setFpuRegister(fd_reg, kFPUInvalidResult); + } + break; + } + case ff_ceil_w_fmt: { // Round double to word towards positive + // infinity. + float rounded = std::ceil(fs_value); + int32_t result = static_cast<int32_t>(rounded); + setFpuRegister(fd_reg, result); + if (setFCSRRoundError(fs_value, rounded)) { + setFpuRegister(fd_reg, kFPUInvalidResult); + } + break; + } + case ff_cvt_l_fmt: + case ff_round_l_fmt: + case ff_trunc_l_fmt: + case ff_floor_l_fmt: + case ff_ceil_l_fmt: + case ff_cvt_ps_s: + case ff_c_f_fmt: + MOZ_CRASH(); + break; + case ff_movf_fmt: + // location of cc field in MOVF is equal to float branch + // instructions + cc = instr->fbccValue(); + fcsr_cc = GetFCSRConditionBit(cc); + if (testFCSRBit(fcsr_cc)) { + setFpuRegisterFloat(fd_reg, getFpuRegisterFloat(fs_reg)); + } + break; + case ff_movz_fmt: + if (rt == 0) { + setFpuRegisterFloat(fd_reg, getFpuRegisterFloat(fs_reg)); + } + break; + case ff_movn_fmt: + if (rt != 0) { + setFpuRegisterFloat(fd_reg, getFpuRegisterFloat(fs_reg)); + } + break; + default: + MOZ_CRASH(); + } + break; + case rs_d: + double dt_value, ds_value; + ds_value = getFpuRegisterDouble(fs_reg); + cc = instr->fcccValue(); + fcsr_cc = GetFCSRConditionBit(cc); + switch (instr->functionFieldRaw()) { + case ff_add_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFpuRegisterDouble(fd_reg, ds_value + dt_value); + break; + case ff_sub_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFpuRegisterDouble(fd_reg, ds_value - dt_value); + break; + case ff_mul_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFpuRegisterDouble(fd_reg, ds_value * dt_value); + break; + case ff_div_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFpuRegisterDouble(fd_reg, ds_value / dt_value); + break; + case ff_abs_fmt: + setFpuRegisterDouble(fd_reg, fabs(ds_value)); + break; + case ff_mov_fmt: + setFpuRegisterDouble(fd_reg, ds_value); + break; + case ff_neg_fmt: + setFpuRegisterDouble(fd_reg, -ds_value); + break; + case ff_sqrt_fmt: + setFpuRegisterDouble(fd_reg, sqrt(ds_value)); + break; + case ff_c_un_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFCSRBit(fcsr_cc, + mozilla::IsNaN(ds_value) || mozilla::IsNaN(dt_value)); + break; + case ff_c_eq_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFCSRBit(fcsr_cc, (ds_value == dt_value)); + break; + case ff_c_ueq_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFCSRBit(fcsr_cc, + (ds_value == dt_value) || (mozilla::IsNaN(ds_value) || + mozilla::IsNaN(dt_value))); + break; + case ff_c_olt_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFCSRBit(fcsr_cc, (ds_value < dt_value)); + break; + case ff_c_ult_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFCSRBit(fcsr_cc, + (ds_value < dt_value) || (mozilla::IsNaN(ds_value) || + mozilla::IsNaN(dt_value))); + break; + case ff_c_ole_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFCSRBit(fcsr_cc, (ds_value <= dt_value)); + break; + case ff_c_ule_fmt: + dt_value = getFpuRegisterDouble(ft_reg); + setFCSRBit(fcsr_cc, + (ds_value <= dt_value) || (mozilla::IsNaN(ds_value) || + mozilla::IsNaN(dt_value))); + break; + case ff_cvt_w_fmt: // Convert double to word. + // Rounding modes are not yet supported. + MOZ_ASSERT((FCSR_ & 3) == 0); + // In rounding mode 0 it should behave like ROUND. + [[fallthrough]]; + case ff_round_w_fmt: { // Round double to word (round half to + // even). + double rounded = std::floor(ds_value + 0.5); + int32_t result = static_cast<int32_t>(rounded); + if ((result & 1) != 0 && result - ds_value == 0.5) { + // If the number is halfway between two integers, + // round to the even one. + result--; + } + setFpuRegister(fd_reg, result); + if (setFCSRRoundError(ds_value, rounded)) { + setFpuRegister(fd_reg, kFPUInvalidResult); + } + break; + } + case ff_trunc_w_fmt: { // Truncate double to word (round towards + // 0). + double rounded = trunc(ds_value); + int32_t result = static_cast<int32_t>(rounded); + setFpuRegister(fd_reg, result); + if (setFCSRRoundError(ds_value, rounded)) { + setFpuRegister(fd_reg, kFPUInvalidResult); + } + break; + } + case ff_floor_w_fmt: { // Round double to word towards negative + // infinity. + double rounded = std::floor(ds_value); + int32_t result = static_cast<int32_t>(rounded); + setFpuRegister(fd_reg, result); + if (setFCSRRoundError(ds_value, rounded)) { + setFpuRegister(fd_reg, kFPUInvalidResult); + } + break; + } + case ff_ceil_w_fmt: { // Round double to word towards positive + // infinity. + double rounded = std::ceil(ds_value); + int32_t result = static_cast<int32_t>(rounded); + setFpuRegister(fd_reg, result); + if (setFCSRRoundError(ds_value, rounded)) { + setFpuRegister(fd_reg, kFPUInvalidResult); + } + break; + } + case ff_cvt_s_fmt: // Convert double to float (single). + setFpuRegisterFloat(fd_reg, static_cast<float>(ds_value)); + break; + case ff_cvt_l_fmt: + case ff_trunc_l_fmt: + case ff_round_l_fmt: + case ff_floor_l_fmt: + case ff_ceil_l_fmt: + case ff_c_f_fmt: + MOZ_CRASH(); + break; + case ff_movf_fmt: + // location of cc field in MOVF is equal to float branch + // instructions + cc = instr->fbccValue(); + fcsr_cc = GetFCSRConditionBit(cc); + if (testFCSRBit(fcsr_cc)) { + setFpuRegisterDouble(fd_reg, getFpuRegisterDouble(fs_reg)); + } + break; + case ff_movz_fmt: + if (rt == 0) { + setFpuRegisterDouble(fd_reg, getFpuRegisterDouble(fs_reg)); + } + break; + case ff_movn_fmt: + if (rt != 0) { + setFpuRegisterDouble(fd_reg, getFpuRegisterDouble(fs_reg)); + } + break; + default: + MOZ_CRASH(); + } + break; + case rs_w: + switch (instr->functionFieldRaw()) { + case ff_cvt_s_fmt: // Convert word to float (single). + alu_out = getFpuRegister(fs_reg); + setFpuRegisterFloat(fd_reg, static_cast<float>(alu_out)); + break; + case ff_cvt_d_fmt: // Convert word to double. + alu_out = getFpuRegister(fs_reg); + setFpuRegisterDouble(fd_reg, static_cast<double>(alu_out)); + break; + default: + MOZ_CRASH(); + } + break; + case rs_l: + switch (instr->functionFieldRaw()) { + case ff_cvt_d_fmt: + case ff_cvt_s_fmt: + MOZ_CRASH(); + break; + default: + MOZ_CRASH(); + } + break; + case rs_ps: + break; + default: + MOZ_CRASH(); + } + break; + case op_cop1x: + switch (instr->functionFieldRaw()) { + case ff_madd_s: + float fr, ft, fs; + fr = getFpuRegisterFloat(fr_reg); + fs = getFpuRegisterFloat(fs_reg); + ft = getFpuRegisterFloat(ft_reg); + setFpuRegisterFloat(fd_reg, fs * ft + fr); + break; + case ff_madd_d: + double dr, dt, ds; + dr = getFpuRegisterDouble(fr_reg); + ds = getFpuRegisterDouble(fs_reg); + dt = getFpuRegisterDouble(ft_reg); + setFpuRegisterDouble(fd_reg, ds * dt + dr); + break; + default: + MOZ_CRASH(); + } + break; + case op_special: + switch (instr->functionFieldRaw()) { + case ff_jr: { + SimInstruction* branch_delay_instr = + reinterpret_cast<SimInstruction*>(current_pc + + SimInstruction::kInstrSize); + branchDelayInstructionDecode(branch_delay_instr); + set_pc(next_pc); + pc_modified_ = true; + break; + } + case ff_jalr: { + SimInstruction* branch_delay_instr = + reinterpret_cast<SimInstruction*>(current_pc + + SimInstruction::kInstrSize); + setRegister(return_addr_reg, + current_pc + 2 * SimInstruction::kInstrSize); + branchDelayInstructionDecode(branch_delay_instr); + set_pc(next_pc); + pc_modified_ = true; + break; + } + // Instructions using HI and LO registers. + case ff_mult: + setRegister(LO, static_cast<int32_t>(i64hilo & 0xffffffff)); + setRegister(HI, static_cast<int32_t>(i64hilo >> 32)); + break; + case ff_multu: + setRegister(LO, static_cast<int32_t>(u64hilo & 0xffffffff)); + setRegister(HI, static_cast<int32_t>(u64hilo >> 32)); + break; + case ff_div: + // Divide by zero and overflow was not checked in the configuration + // step - div and divu do not raise exceptions. On division by 0 + // the result will be UNPREDICTABLE. On overflow (INT_MIN/-1), + // return INT_MIN which is what the hardware does. + if (rs == INT_MIN && rt == -1) { + setRegister(LO, INT_MIN); + setRegister(HI, 0); + } else if (rt != 0) { + setRegister(LO, rs / rt); + setRegister(HI, rs % rt); + } + break; + case ff_divu: + if (rt_u != 0) { + setRegister(LO, rs_u / rt_u); + setRegister(HI, rs_u % rt_u); + } + break; + // Break and trap instructions. + case ff_break: + case ff_tge: + case ff_tgeu: + case ff_tlt: + case ff_tltu: + case ff_teq: + case ff_tne: + if (do_interrupt) { + softwareInterrupt(instr); + } + break; + case ff_sync: + switch (instr->bits(10, 6)) { + case 0x0: + case 0x4: + case 0x10: + case 0x11: + case 0x12: + case 0x13: + AtomicOperations::fenceSeqCst(); + break; + default: + MOZ_CRASH(); + } + break; + // Conditional moves. + case ff_movn: + if (rt) setRegister(rd_reg, rs); + break; + case ff_movci: { + uint32_t cc = instr->fbccValue(); + uint32_t fcsr_cc = GetFCSRConditionBit(cc); + if (instr->bit(16)) { // Read Tf bit. + if (testFCSRBit(fcsr_cc)) setRegister(rd_reg, rs); + } else { + if (!testFCSRBit(fcsr_cc)) setRegister(rd_reg, rs); + } + break; + } + case ff_movz: + if (!rt) setRegister(rd_reg, rs); + break; + default: // For other special opcodes we do the default operation. + setRegister(rd_reg, alu_out); + } + break; + case op_special2: + switch (instr->functionFieldRaw()) { + case ff_mul: + setRegister(rd_reg, alu_out); + // HI and LO are UNPREDICTABLE after the operation. + setRegister(LO, Unpredictable); + setRegister(HI, Unpredictable); + break; + case ff_madd: + setRegister( + LO, getRegister(LO) + static_cast<int32_t>(i64hilo & 0xffffffff)); + setRegister(HI, + getRegister(HI) + static_cast<int32_t>(i64hilo >> 32)); + break; + case ff_maddu: + setRegister( + LO, getRegister(LO) + static_cast<int32_t>(u64hilo & 0xffffffff)); + setRegister(HI, + getRegister(HI) + static_cast<int32_t>(u64hilo >> 32)); + break; + default: // For other special2 opcodes we do the default operation. + setRegister(rd_reg, alu_out); + } + break; + case op_special3: + switch (instr->functionFieldRaw()) { + case ff_ins: + // Ins instr leaves result in Rt, rather than Rd. + setRegister(rt_reg, alu_out); + break; + case ff_ext: + // Ext instr leaves result in Rt, rather than Rd. + setRegister(rt_reg, alu_out); + break; + case ff_bshfl: + setRegister(rd_reg, alu_out); + break; + default: + MOZ_CRASH(); + } + break; + // Unimplemented opcodes raised an error in the configuration step before, + // so we can use the default here to set the destination register in + // common cases. + default: + setRegister(rd_reg, alu_out); + } +} + +// Type 2: instructions using a 16 bytes immediate. (e.g. addi, beq). +void Simulator::decodeTypeImmediate(SimInstruction* instr) { + // Instruction fields. + OpcodeField op = instr->opcodeFieldRaw(); + int32_t rs = getRegister(instr->rsValue()); + uint32_t rs_u = static_cast<uint32_t>(rs); + int32_t rt_reg = instr->rtValue(); // Destination register. + int32_t rt = getRegister(rt_reg); + int16_t imm16 = instr->imm16Value(); + + int32_t ft_reg = instr->ftValue(); // Destination register. + + // Zero extended immediate. + uint32_t oe_imm16 = 0xffff & imm16; + // Sign extended immediate. + int32_t se_imm16 = imm16; + + // Get current pc. + int32_t current_pc = get_pc(); + // Next pc. + int32_t next_pc = bad_ra; + + // Used for conditional branch instructions. + bool do_branch = false; + bool execute_branch_delay_instruction = false; + + // Used for arithmetic instructions. + int32_t alu_out = 0; + // Floating point. + double fp_out = 0.0; + uint32_t cc, cc_value, fcsr_cc; + + // Used for memory instructions. + uint32_t addr = 0x0; + // Value to be written in memory. + uint32_t mem_value = 0x0; + + // ---------- Configuration (and execution for op_regimm). + switch (op) { + // ------------- op_cop1. Coprocessor instructions. + case op_cop1: + switch (instr->rsFieldRaw()) { + case rs_bc1: // Branch on coprocessor condition. + cc = instr->fbccValue(); + fcsr_cc = GetFCSRConditionBit(cc); + cc_value = testFCSRBit(fcsr_cc); + do_branch = (instr->fbtrueValue()) ? cc_value : !cc_value; + execute_branch_delay_instruction = true; + // Set next_pc. + if (do_branch) { + next_pc = current_pc + (imm16 << 2) + SimInstruction::kInstrSize; + } else { + next_pc = current_pc + kBranchReturnOffset; + } + break; + default: + MOZ_CRASH(); + } + break; + // ------------- op_regimm class. + case op_regimm: + switch (instr->rtFieldRaw()) { + case rt_bltz: + do_branch = (rs < 0); + break; + case rt_bltzal: + do_branch = rs < 0; + break; + case rt_bgez: + do_branch = rs >= 0; + break; + case rt_bgezal: + do_branch = rs >= 0; + break; + default: + MOZ_CRASH(); + } + switch (instr->rtFieldRaw()) { + case rt_bltz: + case rt_bltzal: + case rt_bgez: + case rt_bgezal: + // Branch instructions common part. + execute_branch_delay_instruction = true; + // Set next_pc. + if (do_branch) { + next_pc = current_pc + (imm16 << 2) + SimInstruction::kInstrSize; + if (instr->isLinkingInstruction()) { + setRegister(31, current_pc + kBranchReturnOffset); + } + } else { + next_pc = current_pc + kBranchReturnOffset; + } + break; + default: + break; + } + break; // case op_regimm. + // ------------- Branch instructions. + // When comparing to zero, the encoding of rt field is always 0, so we + // don't need to replace rt with zero. + case op_beq: + do_branch = (rs == rt); + break; + case op_bne: + do_branch = rs != rt; + break; + case op_blez: + do_branch = rs <= 0; + break; + case op_bgtz: + do_branch = rs > 0; + break; + // ------------- Arithmetic instructions. + case op_addi: + if (HaveSameSign(rs, se_imm16)) { + if (rs > 0) { + exceptions[kIntegerOverflow] = rs > (kRegisterskMaxValue - se_imm16); + } else if (rs < 0) { + exceptions[kIntegerUnderflow] = rs < (kRegisterskMinValue - se_imm16); + } + } + alu_out = rs + se_imm16; + break; + case op_addiu: + alu_out = rs + se_imm16; + break; + case op_slti: + alu_out = (rs < se_imm16) ? 1 : 0; + break; + case op_sltiu: + alu_out = (rs_u < static_cast<uint32_t>(se_imm16)) ? 1 : 0; + break; + case op_andi: + alu_out = rs & oe_imm16; + break; + case op_ori: + alu_out = rs | oe_imm16; + break; + case op_xori: + alu_out = rs ^ oe_imm16; + break; + case op_lui: + alu_out = (oe_imm16 << 16); + break; + // ------------- Memory instructions. + case op_lb: + addr = rs + se_imm16; + alu_out = readB(addr); + break; + case op_lh: + addr = rs + se_imm16; + alu_out = readH(addr, instr); + break; + case op_lwl: { + // al_offset is offset of the effective address within an aligned word. + uint8_t al_offset = (rs + se_imm16) & kPointerAlignmentMask; + uint8_t byte_shift = kPointerAlignmentMask - al_offset; + uint32_t mask = (1 << byte_shift * 8) - 1; + addr = rs + se_imm16 - al_offset; + alu_out = readW(addr, instr); + alu_out <<= byte_shift * 8; + alu_out |= rt & mask; + break; + } + case op_lw: + addr = rs + se_imm16; + alu_out = readW(addr, instr); + break; + case op_lbu: + addr = rs + se_imm16; + alu_out = readBU(addr); + break; + case op_lhu: + addr = rs + se_imm16; + alu_out = readHU(addr, instr); + break; + case op_lwr: { + // al_offset is offset of the effective address within an aligned word. + uint8_t al_offset = (rs + se_imm16) & kPointerAlignmentMask; + uint8_t byte_shift = kPointerAlignmentMask - al_offset; + uint32_t mask = al_offset ? (~0 << (byte_shift + 1) * 8) : 0; + addr = rs + se_imm16 - al_offset; + alu_out = readW(addr, instr); + alu_out = static_cast<uint32_t>(alu_out) >> al_offset * 8; + alu_out |= rt & mask; + break; + } + case op_sb: + addr = rs + se_imm16; + break; + case op_sh: + addr = rs + se_imm16; + break; + case op_swl: { + uint8_t al_offset = (rs + se_imm16) & kPointerAlignmentMask; + uint8_t byte_shift = kPointerAlignmentMask - al_offset; + uint32_t mask = byte_shift ? (~0 << (al_offset + 1) * 8) : 0; + addr = rs + se_imm16 - al_offset; + mem_value = readW(addr, instr) & mask; + mem_value |= static_cast<uint32_t>(rt) >> byte_shift * 8; + break; + } + case op_sw: + addr = rs + se_imm16; + break; + case op_swr: { + uint8_t al_offset = (rs + se_imm16) & kPointerAlignmentMask; + uint32_t mask = (1 << al_offset * 8) - 1; + addr = rs + se_imm16 - al_offset; + mem_value = readW(addr, instr); + mem_value = (rt << al_offset * 8) | (mem_value & mask); + break; + } + case op_lwc1: + addr = rs + se_imm16; + alu_out = readW(addr, instr); + break; + case op_ldc1: + addr = rs + se_imm16; + fp_out = readD(addr, instr); + break; + case op_swc1: + case op_sdc1: + addr = rs + se_imm16; + break; + case op_ll: + addr = rs + se_imm16; + alu_out = loadLinkedW(addr, instr); + break; + case op_sc: + addr = rs + se_imm16; + alu_out = storeConditionalW(addr, rt, instr); + break; + default: + MOZ_CRASH(); + } + + // ---------- Raise exceptions triggered. + signalExceptions(); + + // ---------- Execution. + switch (op) { + // ------------- Branch instructions. + case op_beq: + case op_bne: + case op_blez: + case op_bgtz: + // Branch instructions common part. + execute_branch_delay_instruction = true; + // Set next_pc. + if (do_branch) { + next_pc = current_pc + (imm16 << 2) + SimInstruction::kInstrSize; + if (instr->isLinkingInstruction()) { + setRegister(31, current_pc + 2 * SimInstruction::kInstrSize); + } + } else { + next_pc = current_pc + 2 * SimInstruction::kInstrSize; + } + break; + // ------------- Arithmetic instructions. + case op_addi: + case op_addiu: + case op_slti: + case op_sltiu: + case op_andi: + case op_ori: + case op_xori: + case op_lui: + setRegister(rt_reg, alu_out); + break; + // ------------- Memory instructions. + case op_lb: + case op_lh: + case op_lwl: + case op_lw: + case op_lbu: + case op_lhu: + case op_lwr: + case op_ll: + case op_sc: + setRegister(rt_reg, alu_out); + break; + case op_sb: + writeB(addr, static_cast<int8_t>(rt)); + break; + case op_sh: + writeH(addr, static_cast<uint16_t>(rt), instr); + break; + case op_swl: + writeW(addr, mem_value, instr); + break; + case op_sw: + writeW(addr, rt, instr); + break; + case op_swr: + writeW(addr, mem_value, instr); + break; + case op_lwc1: + setFpuRegister(ft_reg, alu_out); + break; + case op_ldc1: + setFpuRegisterDouble(ft_reg, fp_out); + break; + case op_swc1: + addr = rs + se_imm16; + writeW(addr, getFpuRegister(ft_reg), instr); + break; + case op_sdc1: + addr = rs + se_imm16; + writeD(addr, getFpuRegisterDouble(ft_reg), instr); + break; + default: + break; + } + + if (execute_branch_delay_instruction) { + // Execute branch delay slot + // We don't check for end_sim_pc. First it should not be met as the current + // pc is valid. Secondly a jump should always execute its branch delay slot. + SimInstruction* branch_delay_instr = reinterpret_cast<SimInstruction*>( + current_pc + SimInstruction::kInstrSize); + branchDelayInstructionDecode(branch_delay_instr); + } + + // If needed update pc after the branch delay execution. + if (next_pc != bad_ra) { + set_pc(next_pc); + } +} + +// Type 3: instructions using a 26 bytes immediate. (e.g. j, jal). +void Simulator::decodeTypeJump(SimInstruction* instr) { + // Get current pc. + int32_t current_pc = get_pc(); + // Get unchanged bits of pc. + int32_t pc_high_bits = current_pc & 0xf0000000; + // Next pc. + int32_t next_pc = pc_high_bits | (instr->imm26Value() << 2); + + // Execute branch delay slot. + // We don't check for end_sim_pc. First it should not be met as the current pc + // is valid. Secondly a jump should always execute its branch delay slot. + SimInstruction* branch_delay_instr = reinterpret_cast<SimInstruction*>( + current_pc + SimInstruction::kInstrSize); + branchDelayInstructionDecode(branch_delay_instr); + + // Update pc and ra if necessary. + // Do this after the branch delay execution. + if (instr->isLinkingInstruction()) { + setRegister(31, current_pc + 2 * SimInstruction::kInstrSize); + } + set_pc(next_pc); + pc_modified_ = true; +} + +// Executes the current instruction. +void Simulator::instructionDecode(SimInstruction* instr) { + if (!SimulatorProcess::ICacheCheckingDisableCount) { + AutoLockSimulatorCache als; + SimulatorProcess::checkICacheLocked(instr); + } + pc_modified_ = false; + + switch (instr->instructionType()) { + case SimInstruction::kRegisterType: + decodeTypeRegister(instr); + break; + case SimInstruction::kImmediateType: + decodeTypeImmediate(instr); + break; + case SimInstruction::kJumpType: + decodeTypeJump(instr); + break; + default: + UNSUPPORTED(); + } + if (!pc_modified_) { + setRegister(pc, + reinterpret_cast<int32_t>(instr) + SimInstruction::kInstrSize); + } +} + +void Simulator::branchDelayInstructionDecode(SimInstruction* instr) { + if (instr->instructionBits() == NopInst) { + // Short-cut generic nop instructions. They are always valid and they + // never change the simulator state. + return; + } + + if (instr->isForbiddenInBranchDelay()) { + MOZ_CRASH("Eror:Unexpected opcode in a branch delay slot."); + } + instructionDecode(instr); +} + +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. + int program_counter = get_pc(); + + while (program_counter != end_sim_pc) { + if (enableStopSimAt && (icount_ == Simulator::StopSimAt)) { + MipsDebugger dbg(this); + dbg.debug(); + } else { + if (single_stepping_) { + single_step_callback_(single_step_callback_arg_, this, + (void*)program_counter); + } + SimInstruction* instr = + reinterpret_cast<SimInstruction*>(program_counter); + instructionDecode(instr); + icount_++; + } + program_counter = get_pc(); + } + + if (single_stepping_) { + single_step_callback_(single_step_callback_arg_, this, nullptr); + } +} + +void Simulator::callInternal(uint8_t* entry) { + // Prepare to execute the code at entry. + setRegister(pc, reinterpret_cast<int32_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. + // The code below assumes that r9 is not used as sb (static base) in + // simulator code and therefore is regarded as a callee-saved register. + int32_t s0_val = getRegister(s0); + int32_t s1_val = getRegister(s1); + int32_t s2_val = getRegister(s2); + int32_t s3_val = getRegister(s3); + int32_t s4_val = getRegister(s4); + int32_t s5_val = getRegister(s5); + int32_t s6_val = getRegister(s6); + int32_t s7_val = getRegister(s7); + int32_t gp_val = getRegister(gp); + int32_t sp_val = getRegister(sp); + int32_t fp_val = getRegister(fp); + + // Set up the callee-saved registers with a known value. To be able to check + // that they are preserved properly across JS execution. + int32_t callee_saved_value = icount_; + setRegister(s0, callee_saved_value); + setRegister(s1, callee_saved_value); + setRegister(s2, callee_saved_value); + setRegister(s3, callee_saved_value); + setRegister(s4, callee_saved_value); + setRegister(s5, callee_saved_value); + setRegister(s6, callee_saved_value); + setRegister(s7, callee_saved_value); + setRegister(gp, callee_saved_value); + setRegister(fp, 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(s0)); + MOZ_ASSERT(callee_saved_value == getRegister(s1)); + MOZ_ASSERT(callee_saved_value == getRegister(s2)); + MOZ_ASSERT(callee_saved_value == getRegister(s3)); + MOZ_ASSERT(callee_saved_value == getRegister(s4)); + MOZ_ASSERT(callee_saved_value == getRegister(s5)); + MOZ_ASSERT(callee_saved_value == getRegister(s6)); + MOZ_ASSERT(callee_saved_value == getRegister(s7)); + MOZ_ASSERT(callee_saved_value == getRegister(gp)); + MOZ_ASSERT(callee_saved_value == getRegister(fp)); + + // Restore callee-saved registers with the original value. + setRegister(s0, s0_val); + setRegister(s1, s1_val); + setRegister(s2, s2_val); + setRegister(s3, s3_val); + setRegister(s4, s4_val); + setRegister(s5, s5_val); + setRegister(s6, s6_val); + setRegister(s7, s7_val); + setRegister(gp, gp_val); + setRegister(sp, sp_val); + setRegister(fp, fp_val); +} + +int32_t Simulator::call(uint8_t* entry, int argument_count, ...) { + va_list parameters; + va_start(parameters, argument_count); + + int original_stack = getRegister(sp); + // Compute position of stack on entry to generated code. + int entry_stack = original_stack; + if (argument_count > kCArgSlotCount) { + entry_stack = entry_stack - argument_count * sizeof(int32_t); + } else { + entry_stack = entry_stack - kCArgsSlotsSize; + } + + entry_stack &= ~(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, int32_t)); + } else { + stack_argument[i] = va_arg(parameters, int32_t); + } + } + + va_end(parameters); + setRegister(sp, entry_stack); + + callInternal(entry); + + // Pop stack passed arguments. + MOZ_ASSERT(entry_stack == getRegister(sp)); + setRegister(sp, original_stack); + + int32_t result = getRegister(v0); + 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_; } diff --git a/js/src/jit/mips32/Simulator-mips32.h b/js/src/jit/mips32/Simulator-mips32.h new file mode 100644 index 0000000000..79acef3e63 --- /dev/null +++ b/js/src/jit/mips32/Simulator-mips32.h @@ -0,0 +1,526 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +// Copyright 2011 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_mips32_Simulator_mips32_h +#define jit_mips32_Simulator_mips32_h + +#ifdef JS_SIMULATOR_MIPS32 + +# include "mozilla/Atomics.h" + +# include "jit/IonTypes.h" +# include "js/ProfilingFrameIterator.h" +# include "threading/Thread.h" +# include "vm/MutexIDs.h" +# include "wasm/WasmSignalHandlers.h" + +namespace js { + +namespace jit { + +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 = 4; +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 = 34; + +// 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; + +// 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; + +// On MIPS 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 kMaxWatchpointCode = 31; +const uint32_t kMaxStopCode = 127; +const uint32_t kWasmTrapCode = 6; + +// ----------------------------------------------------------------------------- +// Utility functions + +typedef uint32_t Instr; +class SimInstruction; + +// Per thread simulator state. +class Simulator { + friend class MipsDebugger; + + public: + // Registers are declared in order. See "See MIPS Run Linux" chapter 2. + enum Register { + no_reg = -1, + zero_reg = 0, + at, + v0, + v1, + a0, + a1, + a2, + a3, + t0, + t1, + t2, + t3, + t4, + t5, + t6, + t7, + s0, + s1, + s2, + s3, + s4, + s5, + s6, + s7, + t8, + t9, + k0, + k1, + gp, + sp, + s8, + ra, + // LO, HI, and pc. + LO, + HI, + pc, // pc must be the last register. + kNumSimuRegisters, + // aliases + fp = s8 + }; + + // Coprocessor registers. + enum FPURegister { + f0, + f1, + f2, + f3, + f4, + f5, + f6, + f7, + f8, + f9, + f10, + f11, + f12, + f13, + f14, + f15, // f12 and f14 are arguments FPURegisters. + f16, + f17, + f18, + f19, + f20, + f21, + f22, + f23, + f24, + f25, + f26, + f27, + f28, + f29, + f30, + f31, + kNumFPURegisters + }; + + // 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(); + + // 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, int32_t value); + int32_t getRegister(int reg) const; + double getDoubleFromRegisterPair(int reg); + // Same for FPURegisters. + void setFpuRegister(int fpureg, int32_t value); + void setFpuRegisterFloat(int fpureg, float value); + void setFpuRegisterFloat(int fpureg, int64_t value); + void setFpuRegisterDouble(int fpureg, double value); + void setFpuRegisterDouble(int fpureg, int64_t value); + int32_t getFpuRegister(int fpureg) const; + int64_t getFpuRegisterLong(int fpureg) const; + float getFpuRegisterFloat(int fpureg) const; + double getFpuRegisterDouble(int fpureg) const; + void setFCSRBit(uint32_t cc, bool value); + bool testFCSRBit(uint32_t cc); + bool setFCSRRoundError(double original, double rounded); + + // Special case of set_register and get_register to access the raw PC value. + void set_pc(int32_t value); + int32_t get_pc() const; + + 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. + int32_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. + inline uint32_t readBU(uint32_t addr); + inline int32_t readB(uint32_t addr); + inline void writeB(uint32_t addr, uint8_t value); + inline void writeB(uint32_t addr, int8_t value); + + inline uint16_t readHU(uint32_t addr, SimInstruction* instr); + inline int16_t readH(uint32_t addr, SimInstruction* instr); + // Note: Overloaded on the sign of the value. + inline void writeH(uint32_t addr, uint16_t value, SimInstruction* instr); + inline void writeH(uint32_t addr, int16_t value, SimInstruction* instr); + + inline int readW(uint32_t addr, SimInstruction* instr); + inline void writeW(uint32_t addr, int value, SimInstruction* instr); + + inline double readD(uint32_t addr, SimInstruction* instr); + inline void writeD(uint32_t addr, double value, SimInstruction* instr); + + inline int32_t loadLinkedW(uint32_t addr, SimInstruction* instr); + inline int32_t storeConditionalW(uint32_t addr, int32_t value, + SimInstruction* instr); + + // Executing is handled based on the instruction type. + void decodeTypeRegister(SimInstruction* instr); + + // Helper function for decodeTypeRegister. + void configureTypeRegister(SimInstruction* instr, int32_t& alu_out, + int64_t& i64hilo, uint64_t& u64hilo, + int32_t& next_pc, int32_t& return_addr_reg, + bool& do_interrupt); + + void decodeTypeImmediate(SimInstruction* instr); + void decodeTypeJump(SimInstruction* instr); + + // Used for breakpoints and traps. + void softwareInterrupt(SimInstruction* instr); + + // Stop helper functions. + bool isWatchpoint(uint32_t code); + void printWatchpoint(uint32_t code); + void handleStop(uint32_t code, SimInstruction* instr); + 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); + + JS::ProfilingFrameIterator::RegisterState registerState(); + + // 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(int32_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(int32_t(newPC)); + return true; + } + + // Executes one instruction. + void instructionDecode(SimInstruction* instr); + // Execute one instruction placed in a branch delay slot. + void branchDelayInstructionDecode(SimInstruction* instr); + + public: + static int StopSimAt; + + // Runtime call support. + static void* RedirectNativeFunction(void* nativeFunction, + ABIFunctionType type); + + private: + enum Exception { + kNone, + kIntegerOverflow, + kIntegerUnderflow, + kDivideByZero, + kNumExceptions + }; + int16_t exceptions[kNumExceptions]; + + // Exceptions. + void signalExceptions(); + + // Handle arguments and return value for runtime FP functions. + void getFpArgs(double* x, double* y, int32_t* z); + void getFpFromStack(int32_t* stack, double* x); + + void setCallResultDouble(double result); + void setCallResultFloat(float result); + void setCallResult(int64_t res); + + void callInternal(uint8_t* entry); + + // Architecture state. + // Registers. + int32_t registers_[kNumSimuRegisters]; + // Coprocessor Registers. + int32_t FPUregisters_[kNumFPURegisters]; + // FPU control register. + uint32_t FCSR_; + + bool LLBit_; + uint32_t LLAddr_; + int32_t lastLLValue_; + + // Simulator support. + char* stack_; + uintptr_t stackLimit_; + bool pc_modified_; + int icount_; + int break_count_; + + // Debugger input. + char* lastDebuggerInput_; + + // Registered breakpoints. + SimInstruction* break_pc_; + Instr break_instr_; + + // 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_; + + 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_MIPS32 */ + +#endif /* jit_mips32_Simulator_mips32_h */ diff --git a/js/src/jit/mips32/Trampoline-mips32.cpp b/js/src/jit/mips32/Trampoline-mips32.cpp new file mode 100644 index 0000000000..7506953cd6 --- /dev/null +++ b/js/src/jit/mips32/Trampoline-mips32.cpp @@ -0,0 +1,1345 @@ +/* -*- 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 "mozilla/DebugOnly.h" + +#include "jit/Bailouts.h" +#include "jit/BaselineFrame.h" +#include "jit/CalleeToken.h" +#include "jit/JitFrames.h" +#include "jit/JitRuntime.h" +#include "jit/JitSpewer.h" +#include "jit/mips-shared/SharedICHelpers-mips-shared.h" +#include "jit/mips32/Bailouts-mips32.h" +#ifdef JS_ION_PERF +# include "jit/PerfSpewer.h" +#endif +#include "jit/VMFunctions.h" +#include "vm/JitActivation.h" // js::jit::JitActivation +#include "vm/JSContext.h" +#include "vm/Realm.h" + +#include "jit/MacroAssembler-inl.h" + +using namespace js; +using namespace js::jit; + +static_assert(sizeof(uintptr_t) == sizeof(uint32_t), "Not 64-bit clean."); + +struct EnterJITRegs { + double f30; + double f28; + double f26; + double f24; + double f22; + double f20; + + // non-volatile registers. + uintptr_t ra; + uintptr_t fp; + uintptr_t s7; + uintptr_t s6; + uintptr_t s5; + uintptr_t s4; + uintptr_t s3; + uintptr_t s2; + uintptr_t s1; + uintptr_t s0; +}; + +struct EnterJITArgs { + // First 4 argumet placeholders + void* jitcode; // <- sp points here when function is entered. + int maxArgc; + Value* maxArgv; + InterpreterFrame* fp; + + // Arguments on stack + CalleeToken calleeToken; + JSObject* scopeChain; + size_t numStackValues; + Value* vp; +}; + +static void GenerateReturn(MacroAssembler& masm, int returnCode) { + MOZ_ASSERT(masm.framePushed() == sizeof(EnterJITRegs)); + + // Restore non-volatile registers + masm.as_lw(s0, StackPointer, offsetof(EnterJITRegs, s0)); + masm.as_lw(s1, StackPointer, offsetof(EnterJITRegs, s1)); + masm.as_lw(s2, StackPointer, offsetof(EnterJITRegs, s2)); + masm.as_lw(s3, StackPointer, offsetof(EnterJITRegs, s3)); + masm.as_lw(s4, StackPointer, offsetof(EnterJITRegs, s4)); + masm.as_lw(s5, StackPointer, offsetof(EnterJITRegs, s5)); + masm.as_lw(s6, StackPointer, offsetof(EnterJITRegs, s6)); + masm.as_lw(s7, StackPointer, offsetof(EnterJITRegs, s7)); + masm.as_lw(fp, StackPointer, offsetof(EnterJITRegs, fp)); + masm.as_lw(ra, StackPointer, offsetof(EnterJITRegs, ra)); + + // Restore non-volatile floating point registers + masm.as_ldc1(f20, StackPointer, offsetof(EnterJITRegs, f20)); + masm.as_ldc1(f22, StackPointer, offsetof(EnterJITRegs, f22)); + masm.as_ldc1(f24, StackPointer, offsetof(EnterJITRegs, f24)); + masm.as_ldc1(f26, StackPointer, offsetof(EnterJITRegs, f26)); + masm.as_ldc1(f28, StackPointer, offsetof(EnterJITRegs, f28)); + masm.as_ldc1(f30, StackPointer, offsetof(EnterJITRegs, f30)); + + masm.freeStack(sizeof(EnterJITRegs)); + + masm.branch(ra); +} + +static void GeneratePrologue(MacroAssembler& masm) { + // Save non-volatile registers. These must be saved by the trampoline, + // rather than the JIT'd code, because they are scanned by the conservative + // scanner. + masm.reserveStack(sizeof(EnterJITRegs)); + masm.as_sw(s0, StackPointer, offsetof(EnterJITRegs, s0)); + masm.as_sw(s1, StackPointer, offsetof(EnterJITRegs, s1)); + masm.as_sw(s2, StackPointer, offsetof(EnterJITRegs, s2)); + masm.as_sw(s3, StackPointer, offsetof(EnterJITRegs, s3)); + masm.as_sw(s4, StackPointer, offsetof(EnterJITRegs, s4)); + masm.as_sw(s5, StackPointer, offsetof(EnterJITRegs, s5)); + masm.as_sw(s6, StackPointer, offsetof(EnterJITRegs, s6)); + masm.as_sw(s7, StackPointer, offsetof(EnterJITRegs, s7)); + masm.as_sw(fp, StackPointer, offsetof(EnterJITRegs, fp)); + masm.as_sw(ra, StackPointer, offsetof(EnterJITRegs, ra)); + + masm.as_sdc1(f20, StackPointer, offsetof(EnterJITRegs, f20)); + masm.as_sdc1(f22, StackPointer, offsetof(EnterJITRegs, f22)); + masm.as_sdc1(f24, StackPointer, offsetof(EnterJITRegs, f24)); + masm.as_sdc1(f26, StackPointer, offsetof(EnterJITRegs, f26)); + masm.as_sdc1(f28, StackPointer, offsetof(EnterJITRegs, f28)); + masm.as_sdc1(f30, StackPointer, offsetof(EnterJITRegs, f30)); +} + +/* + * This method generates a trampoline for a c++ function with the following + * signature: + * void enter(void* code, int argc, Value* argv, InterpreterFrame* fp, + * CalleeToken calleeToken, JSObject* scopeChain, Value* vp) + * ...using standard EABI calling convention + */ +void JitRuntime::generateEnterJIT(JSContext* cx, MacroAssembler& masm) { + enterJITOffset_ = startTrampolineCode(masm); + + const Register reg_code = a0; + const Register reg_argc = a1; + const Register reg_argv = a2; + const mozilla::DebugOnly<Register> reg_frame = a3; + + MOZ_ASSERT(OsrFrameReg == reg_frame); + + GeneratePrologue(masm); + + const Address slotToken( + sp, sizeof(EnterJITRegs) + offsetof(EnterJITArgs, calleeToken)); + const Address slotVp(sp, sizeof(EnterJITRegs) + offsetof(EnterJITArgs, vp)); + + // Save stack pointer into s4 + masm.movePtr(StackPointer, s4); + + // Load calleeToken into s2. + masm.loadPtr(slotToken, s2); + + // Save stack pointer as baseline frame. + masm.movePtr(StackPointer, BaselineFrameReg); + + // Load the number of actual arguments into s3. + masm.loadPtr(slotVp, s3); + masm.unboxInt32(Address(s3, 0), s3); + + /*************************************************************** + Loop over argv vector, push arguments onto stack in reverse order + ***************************************************************/ + + // if we are constructing, that also needs to include newTarget + { + Label noNewTarget; + masm.branchTest32(Assembler::Zero, s2, + Imm32(CalleeToken_FunctionConstructing), &noNewTarget); + + masm.add32(Imm32(1), reg_argc); + + masm.bind(&noNewTarget); + } + + masm.as_sll(s0, reg_argc, 3); // s0 = argc * 8 + masm.addPtr(reg_argv, s0); // s0 = argv + argc * 8 + + // 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(s0, reg_argv, &footer, Assembler::BelowOrEqual, ShortJump); + { + masm.bind(&header); + + masm.subPtr(Imm32(2 * sizeof(uintptr_t)), s0); + masm.subPtr(Imm32(2 * sizeof(uintptr_t)), StackPointer); + + ValueOperand value = ValueOperand(s6, s7); + masm.loadValue(Address(s0, 0), value); + masm.storeValue(value, Address(StackPointer, 0)); + + masm.ma_b(s0, reg_argv, &header, Assembler::Above, ShortJump); + } + masm.bind(&footer); + + masm.subPtr(Imm32(2 * sizeof(uintptr_t)), StackPointer); + masm.storePtr(s3, + Address(StackPointer, sizeof(uintptr_t))); // actual arguments + masm.storePtr(s2, Address(StackPointer, 0)); // callee token + + masm.subPtr(StackPointer, s4); + masm.makeFrameDescriptor(s4, FrameType::CppToJSJit, JitFrameLayout::Size()); + masm.push(s4); // descriptor + + CodeLabel returnLabel; + CodeLabel oomReturnLabel; + { + // Handle Interpreter -> Baseline OSR. + AllocatableGeneralRegisterSet regs(GeneralRegisterSet::All()); + regs.take(OsrFrameReg); + regs.take(BaselineFrameReg); + regs.take(reg_code); + regs.take(ReturnReg); + + const Address slotNumStackValues( + BaselineFrameReg, + sizeof(EnterJITRegs) + offsetof(EnterJITArgs, numStackValues)); + const Address slotScopeChain( + BaselineFrameReg, + sizeof(EnterJITRegs) + offsetof(EnterJITArgs, scopeChain)); + + Label notOsr; + masm.ma_b(OsrFrameReg, OsrFrameReg, ¬Osr, Assembler::Zero, ShortJump); + + Register scratch = regs.takeAny(); + + Register numStackValues = regs.takeAny(); + masm.load32(slotNumStackValues, numStackValues); + + // 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(BaselineFrameReg, Address(StackPointer, 0)); + + // Reserve frame. + Register framePtr = BaselineFrameReg; + masm.subPtr(Imm32(BaselineFrame::Size()), StackPointer); + masm.movePtr(StackPointer, framePtr); + + // Reserve space for locals and stack values. + masm.ma_sll(scratch, numStackValues, Imm32(3)); + masm.subPtr(scratch, StackPointer); + + // Enter exit frame. + masm.addPtr( + Imm32(BaselineFrame::Size() + BaselineFrame::FramePointerOffset), + scratch); + masm.makeFrameDescriptor(scratch, FrameType::BaselineJS, + ExitFrameLayout::Size()); + + // Push frame descriptor and fake return address. + masm.reserveStack(2 * sizeof(uintptr_t)); + masm.storePtr( + scratch, Address(StackPointer, sizeof(uintptr_t))); // Frame descriptor + masm.storePtr(zero, Address(StackPointer, 0)); // fake return address + + // 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(BaselineFrameReg); // BaselineFrame + masm.passABIArg(OsrFrameReg); // InterpreterFrame + masm.passABIArg(numStackValues); + masm.callWithABI<Fn, jit::InitBaselineFrameForOsr>( + MoveOp::GENERAL, CheckUnsafeCallWithABI::DontCheckHasExitFrame); + + regs.add(OsrFrameReg); + regs.take(JSReturnOperand); + 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.addPtr(Imm32(BaselineFrame::Size()), framePtr); + masm.branchIfFalseBool(ReturnReg, &error); + + // If OSR-ing, then emit instrumentation for setting lastProfilerFrame + // if profiler instrumentation is enabled. + { + Label skipProfilingInstrumentation; + Register realFramePtr = numStackValues; + AbsoluteAddress addressOfEnabled( + cx->runtime()->geckoProfiler().addressOfEnabled()); + masm.branch32(Assembler::Equal, addressOfEnabled, Imm32(0), + &skipProfilingInstrumentation); + masm.ma_addu(realFramePtr, framePtr, Imm32(sizeof(void*))); + masm.profilerEnterFrame(realFramePtr, 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.ma_li(scratch, &oomReturnLabel); + masm.jump(scratch); + + masm.bind(¬Osr); + // Load the scope chain in R1. + MOZ_ASSERT(R1.scratchReg() != reg_code); + masm.loadPtr(slotScopeChain, R1.scratchReg()); + } + + // The call will push the return address on the stack, thus we check that + // the stack would be aligned once the call is complete. + masm.assertStackAlignment(JitStackAlignment, 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); + masm.addCodeLabel(oomReturnLabel); + } + + // Pop arguments off the stack. + // s0 <- 8*argc (size of all arguments we pushed on the stack) + masm.pop(s0); + masm.rshiftPtr(Imm32(FRAMESIZE_SHIFT), s0); + masm.addPtr(s0, StackPointer); + + // Store the returned value into the slotVp + masm.loadPtr(slotVp, s1); + masm.storeValue(JSReturnOperand, Address(s1, 0)); + + // Restore non-volatile registers and return. + GenerateReturn(masm, ShortJump); +} + +void JitRuntime::generateInvalidator(MacroAssembler& masm, Label* bailoutTail) { + invalidatorOffset_ = startTrampolineCode(masm); + + // NOTE: Members ionScript_ and osiPointReturnAddress_ of + // InvalidationBailoutStack are already on the stack. + static const uint32_t STACK_DATA_SIZE = + sizeof(InvalidationBailoutStack) - 2 * sizeof(uintptr_t); + + // Stack has to be alligned here. If not, we will have to fix it. + masm.checkStackAlignment(); + + // Make room for data on stack. + masm.subPtr(Imm32(STACK_DATA_SIZE), StackPointer); + + // Save general purpose registers + for (uint32_t i = 0; i < Registers::Total; i++) { + Address address = + Address(StackPointer, InvalidationBailoutStack::offsetOfRegs() + + i * sizeof(uintptr_t)); + masm.storePtr(Register::FromCode(i), address); + } + + // Save floating point registers + // We can use as_sd because stack is alligned. + for (uint32_t i = 0; i < FloatRegisters::TotalDouble; i++) { + masm.as_sdc1( + FloatRegister::FromIndex(i, FloatRegister::Double), StackPointer, + InvalidationBailoutStack::offsetOfFpRegs() + i * sizeof(double)); + } + + // Pass pointer to InvalidationBailoutStack structure. + masm.movePtr(StackPointer, a0); + + // Reserve place for return value and BailoutInfo pointer + masm.subPtr(Imm32(2 * sizeof(uintptr_t)), StackPointer); + // Pass pointer to return value. + masm.ma_addu(a1, StackPointer, Imm32(sizeof(uintptr_t))); + // Pass pointer to BailoutInfo + masm.movePtr(StackPointer, a2); + + using Fn = bool (*)(InvalidationBailoutStack * sp, size_t * frameSizeOut, + BaselineBailoutInfo * *info); + masm.setupAlignedABICall(); + masm.passABIArg(a0); + masm.passABIArg(a1); + masm.passABIArg(a2); + masm.callWithABI<Fn, InvalidationBailout>( + MoveOp::GENERAL, CheckUnsafeCallWithABI::DontCheckOther); + + masm.loadPtr(Address(StackPointer, 0), a2); + masm.loadPtr(Address(StackPointer, sizeof(uintptr_t)), a1); + // Remove the return address, the IonScript, the register state + // (InvaliationBailoutStack) and the space that was allocated for the + // return value. + masm.addPtr(Imm32(sizeof(InvalidationBailoutStack) + 2 * sizeof(uintptr_t)), + StackPointer); + // remove the space that this frame was using before the bailout + // (computed by InvalidationBailout) + masm.addPtr(a1, StackPointer); + + // Jump to shared bailout tail. The BailoutInfo pointer has to be in r2. + masm.jump(bailoutTail); +} + +void JitRuntime::generateArgumentsRectifier(MacroAssembler& masm, + ArgumentsRectifierKind kind) { + switch (kind) { + case ArgumentsRectifierKind::Normal: + argumentsRectifierOffset_ = startTrampolineCode(masm); + break; + case ArgumentsRectifierKind::TrialInlining: + trialInliningArgumentsRectifierOffset_ = startTrampolineCode(masm); + break; + } + masm.pushReturnAddress(); + + Register numActArgsReg = t6; + Register calleeTokenReg = t7; + Register numArgsReg = t5; + + // Load the number of actual arguments into numActArgsReg + masm.loadPtr( + Address(StackPointer, RectifierFrameLayout::offsetOfNumActualArgs()), + numActArgsReg); + + // Load the number of |undefined|s to push into t1. + masm.loadPtr( + Address(StackPointer, RectifierFrameLayout::offsetOfCalleeToken()), + calleeTokenReg); + + // Copy the number of actual arguments into s3. + masm.mov(numActArgsReg, s3); + + masm.mov(calleeTokenReg, numArgsReg); + masm.andPtr(Imm32(CalleeTokenMask), numArgsReg); + masm.load16ZeroExtend(Address(numArgsReg, JSFunction::offsetOfNargs()), + numArgsReg); + + masm.as_subu(t1, numArgsReg, s3); + + // Get the topmost argument. + masm.ma_sll(t0, s3, Imm32(3)); // t0 <- nargs * 8 + masm.as_addu(t2, sp, t0); // t2 <- sp + nargs * 8 + masm.addPtr(Imm32(sizeof(RectifierFrameLayout)), t2); + + { + Label notConstructing; + + masm.branchTest32(Assembler::Zero, calleeTokenReg, + Imm32(CalleeToken_FunctionConstructing), + ¬Constructing); + + // Add sizeof(Value) to overcome |this| + masm.subPtr(Imm32(sizeof(Value)), StackPointer); + masm.load32(Address(t2, NUNBOX32_TYPE_OFFSET + sizeof(Value)), t0); + masm.store32(t0, Address(StackPointer, NUNBOX32_TYPE_OFFSET)); + masm.load32(Address(t2, NUNBOX32_PAYLOAD_OFFSET + sizeof(Value)), t0); + masm.store32(t0, Address(StackPointer, NUNBOX32_PAYLOAD_OFFSET)); + + // Include the newly pushed newTarget value in the frame size + // calculated below. + masm.add32(Imm32(1), numArgsReg); + + masm.bind(¬Constructing); + } + + // Push undefined. + masm.moveValue(UndefinedValue(), ValueOperand(t3, t4)); + { + Label undefLoopTop; + masm.bind(&undefLoopTop); + + masm.subPtr(Imm32(sizeof(Value)), StackPointer); + masm.storeValue(ValueOperand(t3, t4), Address(StackPointer, 0)); + masm.sub32(Imm32(1), t1); + + masm.ma_b(t1, t1, &undefLoopTop, Assembler::NonZero, ShortJump); + } + + // Push arguments, |nargs| + 1 times (to include |this|). + { + Label copyLoopTop, initialSkip; + + masm.ma_b(&initialSkip, ShortJump); + + masm.bind(©LoopTop); + masm.subPtr(Imm32(sizeof(Value)), t2); + masm.sub32(Imm32(1), s3); + + masm.bind(&initialSkip); + + MOZ_ASSERT(sizeof(Value) == 2 * sizeof(uint32_t)); + // Read argument and push to stack. + masm.subPtr(Imm32(sizeof(Value)), StackPointer); + masm.load32(Address(t2, NUNBOX32_TYPE_OFFSET), t0); + masm.store32(t0, Address(StackPointer, NUNBOX32_TYPE_OFFSET)); + masm.load32(Address(t2, NUNBOX32_PAYLOAD_OFFSET), t0); + masm.store32(t0, Address(StackPointer, NUNBOX32_PAYLOAD_OFFSET)); + + masm.ma_b(s3, s3, ©LoopTop, Assembler::NonZero, ShortJump); + } + + // translate the framesize from values into bytes + masm.ma_addu(t0, numArgsReg, Imm32(1)); + masm.lshiftPtr(Imm32(3), t0); + + // Construct sizeDescriptor. + masm.makeFrameDescriptor(t0, FrameType::Rectifier, JitFrameLayout::Size()); + + // Construct JitFrameLayout. + masm.subPtr(Imm32(3 * sizeof(uintptr_t)), StackPointer); + // Push actual arguments. + masm.storePtr(numActArgsReg, Address(StackPointer, 2 * sizeof(uintptr_t))); + // Push callee token. + masm.storePtr(calleeTokenReg, Address(StackPointer, sizeof(uintptr_t))); + // Push frame descriptor. + masm.storePtr(t0, Address(StackPointer, 0)); + + // Call the target function. + masm.andPtr(Imm32(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; + } + + // arg1 + // ... + // argN + // num actual args + // callee token + // sizeDescriptor <- sp now + // return address + + // Remove the rectifier frame. + // t0 <- descriptor with FrameType. + masm.loadPtr(Address(StackPointer, 0), t0); + masm.rshiftPtr(Imm32(FRAMESIZE_SHIFT), t0); // t0 <- descriptor. + + // Discard descriptor, calleeToken and number of actual arguments. + masm.addPtr(Imm32(3 * sizeof(uintptr_t)), StackPointer); + + // arg1 + // ... + // argN <- sp now; t0 <- frame descriptor + // num actual args + // callee token + // sizeDescriptor + // return address + + // Discard pushed arguments. + masm.addPtr(t0, StackPointer); + + masm.ret(); +} + +// NOTE: Members snapshotOffset_ and padding_ of BailoutStack +// are not stored in PushBailoutFrame(). +static const uint32_t bailoutDataSize = + sizeof(BailoutStack) - 2 * sizeof(uintptr_t); +static const uint32_t bailoutInfoOutParamSize = 2 * sizeof(uintptr_t); + +/* There are two different stack layouts when doing bailout. They are + * represented via class BailoutStack. + * + * - First case is when bailout is done trough bailout table. In this case + * table offset is stored in $ra (look at JitRuntime::generateBailoutTable()) + * and thunk code should save it on stack. In this case frameClassId_ cannot + * be NO_FRAME_SIZE_CLASS_ID. Members snapshotOffset_ and padding_ are not on + * the stack. + * + * - Other case is when bailout is done via out of line code (lazy bailout). + * In this case frame size is stored in $ra (look at + * CodeGeneratorMIPS::generateOutOfLineCode()) and thunk code should save it + * on stack. Other difference is that members snapshotOffset_ and padding_ are + * pushed to the stack by CodeGeneratorMIPS::visitOutOfLineBailout(). Field + * frameClassId_ is forced to be NO_FRAME_SIZE_CLASS_ID + * (See: JitRuntime::generateBailoutHandler). + */ +static void PushBailoutFrame(MacroAssembler& masm, uint32_t frameClass, + Register spArg) { + // Make sure that alignment is proper. + masm.checkStackAlignment(); + + // Make room for data. + masm.subPtr(Imm32(bailoutDataSize), StackPointer); + + // Save general purpose registers. + for (uint32_t i = 0; i < Registers::Total; i++) { + uint32_t off = BailoutStack::offsetOfRegs() + i * sizeof(uintptr_t); + masm.storePtr(Register::FromCode(i), Address(StackPointer, off)); + } + +#ifdef ENABLE_WASM_SIMD + // What to do for SIMD? +# error "Needs more careful logic if SIMD is enabled" +#endif + + // Save floating point registers + // We can use as_sdc1 because stack is alligned. + for (uint32_t i = 0; i < FloatRegisters::TotalDouble; i++) { + masm.as_sdc1(FloatRegister::FromIndex(i, FloatRegister::Double), + StackPointer, + BailoutStack::offsetOfFpRegs() + i * sizeof(double)); + } + + // Store the frameSize_ or tableOffset_ stored in ra + // See: JitRuntime::generateBailoutTable() + // See: CodeGeneratorMIPS::generateOutOfLineCode() + masm.storePtr(ra, Address(StackPointer, BailoutStack::offsetOfFrameSize())); + + // Put frame class to stack + masm.storePtr(ImmWord(frameClass), + Address(StackPointer, BailoutStack::offsetOfFrameClass())); + + // Put pointer to BailoutStack as first argument to the Bailout() + masm.movePtr(StackPointer, spArg); +} + +static void GenerateBailoutThunk(MacroAssembler& masm, uint32_t frameClass, + Label* bailoutTail) { + PushBailoutFrame(masm, frameClass, a0); + + // Put pointer to BailoutInfo + masm.subPtr(Imm32(bailoutInfoOutParamSize), StackPointer); + masm.storePtr(ImmPtr(nullptr), Address(StackPointer, 0)); + masm.movePtr(StackPointer, a1); + + using Fn = bool (*)(BailoutStack * sp, BaselineBailoutInfo * *info); + masm.setupAlignedABICall(); + masm.passABIArg(a0); + masm.passABIArg(a1); + masm.callWithABI<Fn, Bailout>(MoveOp::GENERAL, + CheckUnsafeCallWithABI::DontCheckOther); + + // Get BailoutInfo pointer + masm.loadPtr(Address(StackPointer, 0), a2); + + // Remove both the bailout frame and the topmost Ion frame's stack. + if (frameClass == NO_FRAME_SIZE_CLASS_ID) { + // Load frameSize from stack + masm.loadPtr(Address(StackPointer, bailoutInfoOutParamSize + + BailoutStack::offsetOfFrameSize()), + a1); + + // Remove complete BailoutStack class and data after it + masm.addPtr(Imm32(sizeof(BailoutStack) + bailoutInfoOutParamSize), + StackPointer); + // Remove frame size srom stack + masm.addPtr(a1, StackPointer); + } else { + uint32_t frameSize = FrameSizeClass::FromClass(frameClass).frameSize(); + // Remove the data this fuction added and frame size. + masm.addPtr(Imm32(bailoutDataSize + bailoutInfoOutParamSize + frameSize), + StackPointer); + } + + // Jump to shared bailout tail. The BailoutInfo pointer has to be in a2. + masm.jump(bailoutTail); +} + +JitRuntime::BailoutTable JitRuntime::generateBailoutTable(MacroAssembler& masm, + Label* bailoutTail, + uint32_t frameClass) { + uint32_t offset = startTrampolineCode(masm); + + Label bailout; + for (size_t i = 0; i < BAILOUT_TABLE_SIZE; i++) { + // Calculate offset to the end of table + int32_t offset = (BAILOUT_TABLE_SIZE - i) * BAILOUT_TABLE_ENTRY_SIZE; + + // We use the 'ra' as table offset later in GenerateBailoutThunk + masm.as_bal(BOffImm16(offset)); + masm.nop(); + } + masm.bind(&bailout); + + GenerateBailoutThunk(masm, frameClass, bailoutTail); + + return BailoutTable(offset, masm.currentOffset() - offset); +} + +void JitRuntime::generateBailoutHandler(MacroAssembler& masm, + Label* bailoutTail) { + bailoutHandlerOffset_ = startTrampolineCode(masm); + + GenerateBailoutThunk(masm, NO_FRAME_SIZE_CLASS_ID, bailoutTail); +} + +bool JitRuntime::generateVMWrapper(JSContext* cx, MacroAssembler& masm, + const VMFunctionData& f, DynFn nativeFun, + uint32_t* wrapperOffset) { + *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); + + // If it isn't a tail call, then the return address needs to be saved + if (f.expectTailCall == NonTailCall) { + masm.pushReturnAddress(); + } + + // We're aligned to an exit frame, so link it up. + masm.loadJSContext(cxreg); + masm.enterExitFrame(cxreg, regs.getAny(), &f); + + // Save the base of the argument set stored on the stack. + Register argsBase = InvalidReg; + if (f.explicitArgs) { + argsBase = t1; // Use temporary register. + regs.take(argsBase); + masm.ma_addu(argsBase, StackPointer, + Imm32(ExitFrameLayout::SizeWithFooter())); + } + uint32_t framePushedBeforeAlignStack = masm.framePushed(); + masm.alignStackPointer(); + masm.setFramePushed(0); + + // Reserve space for the outparameter. Reserve sizeof(Value) for every + // case so that stack stays aligned. + uint32_t outParamSize = 0; + switch (f.outParam) { + case Type_Value: + outParamSize = sizeof(Value); + masm.reserveStack(outParamSize); + break; + + case Type_Handle: { + uint32_t pushed = masm.framePushed(); + masm.PushEmptyRooted(f.outParamRootType); + outParamSize = masm.framePushed() - pushed; + } break; + + case Type_Bool: + case Type_Int32: + MOZ_ASSERT(sizeof(uintptr_t) == sizeof(uint32_t)); + [[fallthrough]]; + case Type_Pointer: + outParamSize = sizeof(uintptr_t); + masm.reserveStack(outParamSize); + break; + + case Type_Double: + outParamSize = sizeof(double); + masm.reserveStack(outParamSize); + break; + default: + MOZ_ASSERT(f.outParam == Type_Void); + break; + } + + uint32_t outParamOffset = 0; + if (f.outParam != Type_Void) { + // Make sure that stack is double aligned after outParam. + MOZ_ASSERT(outParamSize <= sizeof(double)); + outParamOffset += sizeof(double) - outParamSize; + } + // Reserve stack for double sized args that are copied to be aligned. + outParamOffset += f.doubleByRefArgs() * sizeof(double); + + Register doubleArgs = t0; + masm.reserveStack(outParamOffset); + masm.movePtr(StackPointer, doubleArgs); + + if (!generateTLEnterVM(masm, f)) { + return false; + } + + masm.setupAlignedABICall(); + masm.passABIArg(cxreg); + + size_t argDisp = 0; + size_t doubleArgDisp = 0; + + // Copy any arguments. + for (uint32_t explicitArg = 0; explicitArg < f.explicitArgs; explicitArg++) { + switch (f.argProperties(explicitArg)) { + case VMFunctionData::WordByValue: + masm.passABIArg(MoveOperand(argsBase, argDisp), MoveOp::GENERAL); + argDisp += sizeof(uint32_t); + break; + case VMFunctionData::DoubleByValue: + // Values should be passed by reference, not by value, so we + // assert that the argument is a double-precision float. + MOZ_ASSERT(f.argPassedInFloatReg(explicitArg)); + masm.passABIArg(MoveOperand(argsBase, argDisp), MoveOp::DOUBLE); + argDisp += sizeof(double); + break; + case VMFunctionData::WordByRef: + masm.passABIArg( + MoveOperand(argsBase, argDisp, MoveOperand::EFFECTIVE_ADDRESS), + MoveOp::GENERAL); + argDisp += sizeof(uint32_t); + break; + case VMFunctionData::DoubleByRef: + // Copy double sized argument to aligned place. + masm.ma_ldc1WordAligned(ScratchDoubleReg, argsBase, argDisp); + masm.as_sdc1(ScratchDoubleReg, doubleArgs, doubleArgDisp); + masm.passABIArg(MoveOperand(doubleArgs, doubleArgDisp, + MoveOperand::EFFECTIVE_ADDRESS), + MoveOp::GENERAL); + doubleArgDisp += sizeof(double); + argDisp += sizeof(double); + break; + } + } + + MOZ_ASSERT_IF(f.outParam != Type_Void, doubleArgDisp + sizeof(double) == + outParamOffset + outParamSize); + + // Copy the implicit outparam, if any. + if (f.outParam != Type_Void) { + masm.passABIArg( + MoveOperand(doubleArgs, outParamOffset, MoveOperand::EFFECTIVE_ADDRESS), + MoveOp::GENERAL); + } + + masm.callWithABI(nativeFun, MoveOp::GENERAL, + CheckUnsafeCallWithABI::DontCheckHasExitFrame); + + if (!generateTLExitVM(masm, f)) { + return false; + } + + // Test for failure. + switch (f.failType()) { + case Type_Object: + masm.branchTestPtr(Assembler::Zero, v0, v0, masm.failureLabel()); + break; + case Type_Bool: + // Called functions return bools, which are 0/false and non-zero/true + masm.branchIfFalseBool(v0, masm.failureLabel()); + break; + case Type_Void: + break; + default: + MOZ_CRASH("unknown failure kind"); + } + + masm.freeStack(outParamOffset); + + // Load the outparam and free any allocated stack. + switch (f.outParam) { + case Type_Handle: + masm.popRooted(f.outParamRootType, ReturnReg, JSReturnOperand); + break; + + case Type_Value: + masm.loadValue(Address(StackPointer, 0), JSReturnOperand); + masm.freeStack(sizeof(Value)); + break; + + case Type_Int32: + MOZ_ASSERT(sizeof(uintptr_t) == sizeof(uint32_t)); + [[fallthrough]]; + case Type_Pointer: + masm.load32(Address(StackPointer, 0), ReturnReg); + masm.freeStack(sizeof(uintptr_t)); + break; + + case Type_Bool: + masm.load8ZeroExtend(Address(StackPointer, 0), ReturnReg); + masm.freeStack(sizeof(uintptr_t)); + break; + + case Type_Double: + if (JitOptions.supportsFloatingPoint) { + masm.as_ldc1(ReturnDoubleReg, StackPointer, 0); + } else { + masm.assumeUnreachable( + "Unable to load into float reg, with no FP support."); + } + masm.freeStack(sizeof(double)); + break; + + default: + MOZ_ASSERT(f.outParam == Type_Void); + break; + } + + masm.restoreStackPointer(); + masm.setFramePushed(framePushedBeforeAlignStack); + + masm.leaveExitFrame(); + masm.retn(Imm32(sizeof(ExitFrameLayout) + + f.explicitStackSlots() * sizeof(uintptr_t) + + f.extraValuesToPop * sizeof(Value))); + + return true; +} + +uint32_t JitRuntime::generatePreBarrier(JSContext* cx, MacroAssembler& masm, + MIRType type) { + 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; + if (JitOptions.supportsFloatingPoint) { + save.set() = RegisterSet(GeneralRegisterSet(Registers::VolatileMask), + FloatRegisterSet(FloatRegisters::VolatileMask)); + } else { + save.set() = RegisterSet(GeneralRegisterSet(Registers::VolatileMask), + FloatRegisterSet()); + } + save.add(ra); + masm.PushRegsInMask(save); + + masm.movePtr(ImmPtr(cx->runtime()), a0); + + masm.setupUnalignedABICall(a2); + masm.passABIArg(a0); + masm.passABIArg(a1); + masm.callWithABI(JitMarkFunction(type)); + + save.take(AnyRegister(ra)); + masm.PopRegsInMask(save); + masm.ret(); + + masm.bind(&noBarrier); + masm.pop(temp3); + masm.pop(temp2); + masm.pop(temp1); + masm.abiret(); + + return offset; +} + +void JitRuntime::generateExceptionTailStub(MacroAssembler& masm, + Label* profilerExitTail) { + exceptionTailOffset_ = startTrampolineCode(masm); + + masm.bind(masm.failureLabel()); + masm.handleFailureWithHandlerTail(profilerExitTail); +} + +void JitRuntime::generateBailoutTailStub(MacroAssembler& masm, + Label* bailoutTail) { + bailoutTailOffset_ = startTrampolineCode(masm); + masm.bind(bailoutTail); + + masm.generateBailoutTail(a1, a2); +} + +void JitRuntime::generateProfilerExitFrameTailStub(MacroAssembler& masm, + Label* profilerExitTail) { + profilerExitFrameTailOffset_ = startTrampolineCode(masm); + masm.bind(profilerExitTail); + + Register scratch1 = t0; + Register scratch2 = t1; + Register scratch3 = t2; + Register scratch4 = t3; + + // + // The code generated below expects that the current stack pointer points + // to an Ion or Baseline frame, at the state it would be immediately + // before a ret(). Thus, after this stub's business is done, it executes + // a ret() and returns directly to the caller script, on behalf of the + // callee script that jumped to this code. + // + // Thus the expected stack is: + // + // StackPointer ----+ + // v + // ..., ActualArgc, CalleeToken, Descriptor, ReturnAddr + // MEM-HI MEM-LOW + // + // + // The generated jitcode is responsible for overwriting the + // jitActivation->lastProfilingFrame field with a pointer to the previous + // Ion or Baseline jit-frame that was pushed before this one. It is also + // responsible for overwriting jitActivation->lastProfilingCallSite with + // the return address into that frame. The frame could either be an + // immediate "caller" frame, or it could be a frame in a previous + // JitActivation (if the current frame was entered from C++, and the C++ + // was entered by some caller jit-frame further down the stack). + // + // So this jitcode is responsible for "walking up" the jit stack, finding + // the previous Ion or Baseline JS frame, and storing its address and the + // return address into the appropriate fields on the current jitActivation. + // + // There are a fixed number of different path types that can lead to the + // current frame, which is either a baseline or ion frame: + // + // <Baseline-Or-Ion> + // ^ + // | + // ^--- Ion + // | + // ^--- Baseline Stub <---- Baseline + // | + // ^--- Argument Rectifier + // | ^ + // | | + // | ^--- Ion + // | | + // | ^--- Baseline Stub <---- Baseline + // | + // ^--- Entry Frame (From C++) + // + Register actReg = scratch4; + masm.loadJSContext(actReg); + masm.loadPtr(Address(actReg, offsetof(JSContext, profilingActivation_)), + actReg); + + Address lastProfilingFrame(actReg, + JitActivation::offsetOfLastProfilingFrame()); + Address lastProfilingCallSite(actReg, + JitActivation::offsetOfLastProfilingCallSite()); + +#ifdef DEBUG + // Ensure that frame we are exiting is current lastProfilingFrame + { + masm.loadPtr(lastProfilingFrame, scratch1); + Label checkOk; + masm.branchPtr(Assembler::Equal, scratch1, ImmWord(0), &checkOk); + masm.branchPtr(Assembler::Equal, StackPointer, scratch1, &checkOk); + masm.assumeUnreachable( + "Mismatch between stored lastProfilingFrame and current stack " + "pointer."); + masm.bind(&checkOk); + } +#endif + + // Load the frame descriptor into |scratch1|, figure out what to do depending + // on its type. + masm.loadPtr(Address(StackPointer, JitFrameLayout::offsetOfDescriptor()), + scratch1); + + // Going into the conditionals, we will have: + // FrameDescriptor.size in scratch1 + // FrameDescriptor.type in scratch2 + masm.ma_and(scratch2, scratch1, Imm32((1 << FRAMETYPE_BITS) - 1)); + masm.rshiftPtr(Imm32(FRAMESIZE_SHIFT), scratch1); + + // Handling of each case is dependent on FrameDescriptor.type + Label handle_IonJS; + Label handle_BaselineStub; + Label handle_Rectifier; + Label handle_IonICCall; + Label handle_Entry; + Label end; + + masm.branch32(Assembler::Equal, scratch2, Imm32(FrameType::IonJS), + &handle_IonJS); + masm.branch32(Assembler::Equal, scratch2, Imm32(FrameType::BaselineJS), + &handle_IonJS); + masm.branch32(Assembler::Equal, scratch2, Imm32(FrameType::BaselineStub), + &handle_BaselineStub); + masm.branch32(Assembler::Equal, scratch2, Imm32(FrameType::Rectifier), + &handle_Rectifier); + masm.branch32(Assembler::Equal, scratch2, Imm32(FrameType::IonICCall), + &handle_IonICCall); + masm.branch32(Assembler::Equal, scratch2, Imm32(FrameType::CppToJSJit), + &handle_Entry); + + // The WasmToJSJit is just another kind of entry. + masm.branch32(Assembler::Equal, scratch2, Imm32(FrameType::WasmToJSJit), + &handle_Entry); + + masm.assumeUnreachable( + "Invalid caller frame type when exiting from Ion frame."); + + // + // FrameType::IonJS + // + // Stack layout: + // ... + // Ion-Descriptor + // Prev-FP ---> Ion-ReturnAddr + // ... previous frame data ... |- Descriptor.Size + // ... arguments ... | + // ActualArgc | + // CalleeToken |- JitFrameLayout::Size() + // Descriptor | + // FP -----> ReturnAddr | + // + masm.bind(&handle_IonJS); + { + // |scratch1| contains Descriptor.size + + // returning directly to an IonJS frame. Store return addr to frame + // in lastProfilingCallSite. + masm.loadPtr(Address(StackPointer, JitFrameLayout::offsetOfReturnAddress()), + scratch2); + masm.storePtr(scratch2, lastProfilingCallSite); + + // Store return frame in lastProfilingFrame. + // scratch2 := StackPointer + Descriptor.size*1 + JitFrameLayout::Size(); + masm.as_addu(scratch2, StackPointer, scratch1); + masm.ma_addu(scratch2, scratch2, Imm32(JitFrameLayout::Size())); + masm.storePtr(scratch2, lastProfilingFrame); + masm.ret(); + } + + // + // FrameType::BaselineStub + // + // Look past the stub and store the frame pointer to + // the baselineJS frame prior to it. + // + // Stack layout: + // ... + // BL-Descriptor + // Prev-FP ---> BL-ReturnAddr + // +-----> BL-PrevFramePointer + // | ... BL-FrameData ... + // | BLStub-Descriptor + // | BLStub-ReturnAddr + // | BLStub-StubPointer | + // +------ BLStub-SavedFramePointer |- Descriptor.Size + // ... arguments ... | + // ActualArgc | + // CalleeToken |- JitFrameLayout::Size() + // Descriptor | + // FP -----> ReturnAddr | + // + // We take advantage of the fact that the stub frame saves the frame + // pointer pointing to the baseline frame, so a bunch of calculation can + // be avoided. + // + masm.bind(&handle_BaselineStub); + { + masm.as_addu(scratch3, StackPointer, scratch1); + Address stubFrameReturnAddr( + scratch3, JitFrameLayout::Size() + + BaselineStubFrameLayout::offsetOfReturnAddress()); + masm.loadPtr(stubFrameReturnAddr, scratch2); + masm.storePtr(scratch2, lastProfilingCallSite); + + Address stubFrameSavedFramePtr( + scratch3, JitFrameLayout::Size() - (2 * sizeof(void*))); + masm.loadPtr(stubFrameSavedFramePtr, scratch2); + masm.addPtr(Imm32(sizeof(void*)), scratch2); // Skip past BL-PrevFramePtr + masm.storePtr(scratch2, lastProfilingFrame); + masm.ret(); + } + + // + // FrameType::Rectifier + // + // The rectifier frame can be preceded by either an IonJS, a BaselineStub, + // or a CppToJSJit/WasmToJSJit frame. + // + // Stack layout if caller of rectifier was Ion or CppToJSJit/WasmToJSJit: + // + // Ion-Descriptor + // Ion-ReturnAddr + // ... ion frame data ... |- Rect-Descriptor.Size + // < COMMON LAYOUT > + // + // Stack layout if caller of rectifier was Baseline: + // + // BL-Descriptor + // Prev-FP ---> BL-ReturnAddr + // +-----> BL-SavedFramePointer + // | ... baseline frame data ... + // | BLStub-Descriptor + // | BLStub-ReturnAddr + // | BLStub-StubPointer | + // +------ BLStub-SavedFramePointer |- Rect-Descriptor.Size + // ... args to rectifier ... | + // < COMMON LAYOUT > + // + // Common stack layout: + // + // ActualArgc | + // CalleeToken |- IonRectitiferFrameLayout::Size() + // Rect-Descriptor | + // Rect-ReturnAddr | + // ... rectifier data & args ... |- Descriptor.Size + // ActualArgc | + // CalleeToken |- JitFrameLayout::Size() + // Descriptor | + // FP -----> ReturnAddr | + // + masm.bind(&handle_Rectifier); + { + // scratch2 := StackPointer + Descriptor.size*1 + JitFrameLayout::Size(); + masm.as_addu(scratch2, StackPointer, scratch1); + masm.add32(Imm32(JitFrameLayout::Size()), scratch2); + masm.loadPtr(Address(scratch2, RectifierFrameLayout::offsetOfDescriptor()), + scratch3); + masm.ma_srl(scratch1, scratch3, Imm32(FRAMESIZE_SHIFT)); + masm.and32(Imm32((1 << FRAMETYPE_BITS) - 1), scratch3); + + // Now |scratch1| contains Rect-Descriptor.Size + // and |scratch2| points to Rectifier frame + // and |scratch3| contains Rect-Descriptor.Type + + masm.assertRectifierFrameParentType(scratch3); + + // Check for either Ion or BaselineStub frame. + Label notIonFrame; + masm.branch32(Assembler::NotEqual, scratch3, Imm32(FrameType::IonJS), + ¬IonFrame); + + // Handle Rectifier <- IonJS + // scratch3 := RectFrame[ReturnAddr] + masm.loadPtr( + Address(scratch2, RectifierFrameLayout::offsetOfReturnAddress()), + scratch3); + masm.storePtr(scratch3, lastProfilingCallSite); + + // scratch3 := RectFrame + Rect-Descriptor.Size + + // RectifierFrameLayout::Size() + masm.as_addu(scratch3, scratch2, scratch1); + masm.add32(Imm32(RectifierFrameLayout::Size()), scratch3); + masm.storePtr(scratch3, lastProfilingFrame); + masm.ret(); + + masm.bind(¬IonFrame); + + // Check for either BaselineStub or a CppToJSJit/WasmToJSJit entry + // frame. + masm.branch32(Assembler::NotEqual, scratch3, Imm32(FrameType::BaselineStub), + &handle_Entry); + + // Handle Rectifier <- BaselineStub <- BaselineJS + masm.as_addu(scratch3, scratch2, scratch1); + Address stubFrameReturnAddr( + scratch3, RectifierFrameLayout::Size() + + BaselineStubFrameLayout::offsetOfReturnAddress()); + masm.loadPtr(stubFrameReturnAddr, scratch2); + masm.storePtr(scratch2, lastProfilingCallSite); + + Address stubFrameSavedFramePtr( + scratch3, RectifierFrameLayout::Size() - (2 * sizeof(void*))); + masm.loadPtr(stubFrameSavedFramePtr, scratch2); + masm.addPtr(Imm32(sizeof(void*)), scratch2); + masm.storePtr(scratch2, lastProfilingFrame); + masm.ret(); + } + + // FrameType::IonICCall + // + // The caller is always an IonJS frame. + // + // Ion-Descriptor + // Ion-ReturnAddr + // ... ion frame data ... |- CallFrame-Descriptor.Size + // StubCode | + // ICCallFrame-Descriptor |- IonICCallFrameLayout::Size() + // ICCallFrame-ReturnAddr | + // ... call frame data & args ... |- Descriptor.Size + // ActualArgc | + // CalleeToken |- JitFrameLayout::Size() + // Descriptor | + // FP -----> ReturnAddr | + masm.bind(&handle_IonICCall); + { + // scratch2 := StackPointer + Descriptor.size + JitFrameLayout::Size() + masm.as_addu(scratch2, StackPointer, scratch1); + masm.addPtr(Imm32(JitFrameLayout::Size()), scratch2); + + // scratch3 := ICCallFrame-Descriptor.Size + masm.loadPtr(Address(scratch2, IonICCallFrameLayout::offsetOfDescriptor()), + scratch3); +#ifdef DEBUG + // Assert previous frame is an IonJS frame. + masm.movePtr(scratch3, scratch1); + masm.and32(Imm32((1 << FRAMETYPE_BITS) - 1), scratch1); + { + Label checkOk; + masm.branch32(Assembler::Equal, scratch1, Imm32(FrameType::IonJS), + &checkOk); + masm.assumeUnreachable("IonICCall frame must be preceded by IonJS frame"); + masm.bind(&checkOk); + } +#endif + masm.rshiftPtr(Imm32(FRAMESIZE_SHIFT), scratch3); + + // lastProfilingCallSite := ICCallFrame-ReturnAddr + masm.loadPtr( + Address(scratch2, IonICCallFrameLayout::offsetOfReturnAddress()), + scratch1); + masm.storePtr(scratch1, lastProfilingCallSite); + + // lastProfilingFrame := ICCallFrame + ICCallFrame-Descriptor.Size + + // IonICCallFrameLayout::Size() + masm.as_addu(scratch1, scratch2, scratch3); + masm.addPtr(Imm32(IonICCallFrameLayout::Size()), scratch1); + masm.storePtr(scratch1, lastProfilingFrame); + masm.ret(); + } + + // + // FrameType::CppToJSJit / FrameType::WasmToJSJit + // + // If at an entry frame, store null into both fields. + // A fast-path wasm->jit transition frame is an entry frame from the point + // of view of the JIT. + // + masm.bind(&handle_Entry); + { + masm.movePtr(ImmPtr(nullptr), scratch1); + masm.storePtr(scratch1, lastProfilingCallSite); + masm.storePtr(scratch1, lastProfilingFrame); + masm.ret(); + } +} |