diff options
Diffstat (limited to 'gfx/skia/skia/src/sksl/codegen/SkSLMetalCodeGenerator.cpp')
-rw-r--r-- | gfx/skia/skia/src/sksl/codegen/SkSLMetalCodeGenerator.cpp | 3226 |
1 files changed, 3226 insertions, 0 deletions
diff --git a/gfx/skia/skia/src/sksl/codegen/SkSLMetalCodeGenerator.cpp b/gfx/skia/skia/src/sksl/codegen/SkSLMetalCodeGenerator.cpp new file mode 100644 index 0000000000..d173fae687 --- /dev/null +++ b/gfx/skia/skia/src/sksl/codegen/SkSLMetalCodeGenerator.cpp @@ -0,0 +1,3226 @@ +/* + * Copyright 2016 Google Inc. + * + * Use of this source code is governed by a BSD-style license that can be + * found in the LICENSE file. + */ + +#include "src/sksl/codegen/SkSLMetalCodeGenerator.h" + +#include "include/core/SkSpan.h" +#include "include/core/SkTypes.h" +#include "include/private/SkSLIRNode.h" +#include "include/private/SkSLLayout.h" +#include "include/private/SkSLModifiers.h" +#include "include/private/SkSLProgramElement.h" +#include "include/private/SkSLStatement.h" +#include "include/private/SkSLString.h" +#include "include/private/base/SkTo.h" +#include "include/sksl/SkSLErrorReporter.h" +#include "include/sksl/SkSLOperator.h" +#include "include/sksl/SkSLPosition.h" +#include "src/base/SkScopeExit.h" +#include "src/sksl/SkSLAnalysis.h" +#include "src/sksl/SkSLBuiltinTypes.h" +#include "src/sksl/SkSLCompiler.h" +#include "src/sksl/SkSLContext.h" +#include "src/sksl/SkSLIntrinsicList.h" +#include "src/sksl/SkSLMemoryLayout.h" +#include "src/sksl/SkSLOutputStream.h" +#include "src/sksl/SkSLProgramSettings.h" +#include "src/sksl/SkSLUtil.h" +#include "src/sksl/analysis/SkSLProgramVisitor.h" +#include "src/sksl/ir/SkSLBinaryExpression.h" +#include "src/sksl/ir/SkSLBlock.h" +#include "src/sksl/ir/SkSLConstructor.h" +#include "src/sksl/ir/SkSLConstructorArrayCast.h" +#include "src/sksl/ir/SkSLConstructorCompound.h" +#include "src/sksl/ir/SkSLConstructorMatrixResize.h" +#include "src/sksl/ir/SkSLDoStatement.h" +#include "src/sksl/ir/SkSLExpression.h" +#include "src/sksl/ir/SkSLExpressionStatement.h" +#include "src/sksl/ir/SkSLExtension.h" +#include "src/sksl/ir/SkSLFieldAccess.h" +#include "src/sksl/ir/SkSLForStatement.h" +#include "src/sksl/ir/SkSLFunctionCall.h" +#include "src/sksl/ir/SkSLFunctionDeclaration.h" +#include "src/sksl/ir/SkSLFunctionDefinition.h" +#include "src/sksl/ir/SkSLFunctionPrototype.h" +#include "src/sksl/ir/SkSLIfStatement.h" +#include "src/sksl/ir/SkSLIndexExpression.h" +#include "src/sksl/ir/SkSLInterfaceBlock.h" +#include "src/sksl/ir/SkSLLiteral.h" +#include "src/sksl/ir/SkSLModifiersDeclaration.h" +#include "src/sksl/ir/SkSLNop.h" +#include "src/sksl/ir/SkSLPostfixExpression.h" +#include "src/sksl/ir/SkSLPrefixExpression.h" +#include "src/sksl/ir/SkSLProgram.h" +#include "src/sksl/ir/SkSLReturnStatement.h" +#include "src/sksl/ir/SkSLSetting.h" +#include "src/sksl/ir/SkSLStructDefinition.h" +#include "src/sksl/ir/SkSLSwitchCase.h" +#include "src/sksl/ir/SkSLSwitchStatement.h" +#include "src/sksl/ir/SkSLSwizzle.h" +#include "src/sksl/ir/SkSLTernaryExpression.h" +#include "src/sksl/ir/SkSLVarDeclarations.h" +#include "src/sksl/ir/SkSLVariable.h" +#include "src/sksl/ir/SkSLVariableReference.h" +#include "src/sksl/spirv.h" + +#include <algorithm> +#include <cstddef> +#include <functional> +#include <limits> +#include <memory> + +namespace SkSL { + +static const char* operator_name(Operator op) { + switch (op.kind()) { + case Operator::Kind::LOGICALXOR: return " != "; + default: return op.operatorName(); + } +} + +class MetalCodeGenerator::GlobalStructVisitor { +public: + virtual ~GlobalStructVisitor() = default; + virtual void visitInterfaceBlock(const InterfaceBlock& block, std::string_view blockName) {} + virtual void visitTexture(const Type& type, const Modifiers& modifiers, + std::string_view name) {} + virtual void visitSampler(const Type& type, std::string_view name) {} + virtual void visitConstantVariable(const VarDeclaration& decl) {} + virtual void visitNonconstantVariable(const Variable& var, const Expression* value) {} +}; + +class MetalCodeGenerator::ThreadgroupStructVisitor { +public: + virtual ~ThreadgroupStructVisitor() = default; + virtual void visitNonconstantVariable(const Variable& var) = 0; +}; + +void MetalCodeGenerator::write(std::string_view s) { + if (s.empty()) { + return; + } + if (fAtLineStart) { + for (int i = 0; i < fIndentation; i++) { + fOut->writeText(" "); + } + } + fOut->writeText(std::string(s).c_str()); + fAtLineStart = false; +} + +void MetalCodeGenerator::writeLine(std::string_view s) { + this->write(s); + fOut->writeText(fLineEnding); + fAtLineStart = true; +} + +void MetalCodeGenerator::finishLine() { + if (!fAtLineStart) { + this->writeLine(); + } +} + +void MetalCodeGenerator::writeExtension(const Extension& ext) { + this->writeLine("#extension " + std::string(ext.name()) + " : enable"); +} + +std::string MetalCodeGenerator::typeName(const Type& type) { + // we need to know the modifiers for textures + switch (type.typeKind()) { + case Type::TypeKind::kArray: + SkASSERT(!type.isUnsizedArray()); + SkASSERTF(type.columns() > 0, "invalid array size: %s", type.description().c_str()); + return String::printf("array<%s, %d>", + this->typeName(type.componentType()).c_str(), type.columns()); + + case Type::TypeKind::kVector: + return this->typeName(type.componentType()) + std::to_string(type.columns()); + + case Type::TypeKind::kMatrix: + return this->typeName(type.componentType()) + std::to_string(type.columns()) + "x" + + std::to_string(type.rows()); + + case Type::TypeKind::kSampler: + if (type.dimensions() != SpvDim2D) { + fContext.fErrors->error(Position(), "Unsupported texture dimensions"); + } + return "sampler2D"; + + case Type::TypeKind::kTexture: + switch (type.textureAccess()) { + case Type::TextureAccess::kSample: return "texture2d<half>"; + case Type::TextureAccess::kRead: return "texture2d<half, access::read>"; + case Type::TextureAccess::kWrite: return "texture2d<half, access::write>"; + case Type::TextureAccess::kReadWrite: return "texture2d<half, access::read_write>"; + default: break; + } + SkUNREACHABLE; + case Type::TypeKind::kAtomic: + // SkSL currently only supports the atomicUint type. + SkASSERT(type.matches(*fContext.fTypes.fAtomicUInt)); + return "atomic_uint"; + default: + return std::string(type.name()); + } +} + +void MetalCodeGenerator::writeStructDefinition(const StructDefinition& s) { + const Type& type = s.type(); + this->writeLine("struct " + type.displayName() + " {"); + fIndentation++; + this->writeFields(type.fields(), type.fPosition); + fIndentation--; + this->writeLine("};"); +} + +void MetalCodeGenerator::writeType(const Type& type) { + this->write(this->typeName(type)); +} + +void MetalCodeGenerator::writeExpression(const Expression& expr, Precedence parentPrecedence) { + switch (expr.kind()) { + case Expression::Kind::kBinary: + this->writeBinaryExpression(expr.as<BinaryExpression>(), parentPrecedence); + break; + case Expression::Kind::kConstructorArray: + case Expression::Kind::kConstructorStruct: + this->writeAnyConstructor(expr.asAnyConstructor(), "{", "}", parentPrecedence); + break; + case Expression::Kind::kConstructorArrayCast: + this->writeConstructorArrayCast(expr.as<ConstructorArrayCast>(), parentPrecedence); + break; + case Expression::Kind::kConstructorCompound: + this->writeConstructorCompound(expr.as<ConstructorCompound>(), parentPrecedence); + break; + case Expression::Kind::kConstructorDiagonalMatrix: + case Expression::Kind::kConstructorSplat: + this->writeAnyConstructor(expr.asAnyConstructor(), "(", ")", parentPrecedence); + break; + case Expression::Kind::kConstructorMatrixResize: + this->writeConstructorMatrixResize(expr.as<ConstructorMatrixResize>(), + parentPrecedence); + break; + case Expression::Kind::kConstructorScalarCast: + case Expression::Kind::kConstructorCompoundCast: + this->writeCastConstructor(expr.asAnyConstructor(), "(", ")", parentPrecedence); + break; + case Expression::Kind::kFieldAccess: + this->writeFieldAccess(expr.as<FieldAccess>()); + break; + case Expression::Kind::kLiteral: + this->writeLiteral(expr.as<Literal>()); + break; + case Expression::Kind::kFunctionCall: + this->writeFunctionCall(expr.as<FunctionCall>()); + break; + case Expression::Kind::kPrefix: + this->writePrefixExpression(expr.as<PrefixExpression>(), parentPrecedence); + break; + case Expression::Kind::kPostfix: + this->writePostfixExpression(expr.as<PostfixExpression>(), parentPrecedence); + break; + case Expression::Kind::kSetting: + this->writeExpression(*expr.as<Setting>().toLiteral(fContext), parentPrecedence); + break; + case Expression::Kind::kSwizzle: + this->writeSwizzle(expr.as<Swizzle>()); + break; + case Expression::Kind::kVariableReference: + this->writeVariableReference(expr.as<VariableReference>()); + break; + case Expression::Kind::kTernary: + this->writeTernaryExpression(expr.as<TernaryExpression>(), parentPrecedence); + break; + case Expression::Kind::kIndex: + this->writeIndexExpression(expr.as<IndexExpression>()); + break; + default: + SkDEBUGFAILF("unsupported expression: %s", expr.description().c_str()); + break; + } +} + +// returns true if we should pass by reference instead of by value +static bool pass_by_reference(const Type& type, const Modifiers& modifiers) { + return (modifiers.fFlags & Modifiers::kOut_Flag) && !type.isUnsizedArray(); +} + +// returns true if we need to specify an address space modifier +static bool needs_address_space(const Type& type, const Modifiers& modifiers) { + return type.isUnsizedArray() || pass_by_reference(type, modifiers); +} + +// returns true if the InterfaceBlock has the `buffer` modifier +static bool is_buffer(const InterfaceBlock& block) { + return block.var()->modifiers().fFlags & Modifiers::kBuffer_Flag; +} + +// returns true if the InterfaceBlock has the `readonly` modifier +static bool is_readonly(const InterfaceBlock& block) { + return block.var()->modifiers().fFlags & Modifiers::kReadOnly_Flag; +} + +std::string MetalCodeGenerator::getOutParamHelper(const FunctionCall& call, + const ExpressionArray& arguments, + const SkTArray<VariableReference*>& outVars) { + // It's possible for out-param function arguments to contain an out-param function call + // expression. Emit the function into a temporary stream to prevent the nested helper from + // clobbering the current helper as we recursively evaluate argument expressions. + StringStream tmpStream; + AutoOutputStream outputToExtraFunctions(this, &tmpStream, &fIndentation); + + const FunctionDeclaration& function = call.function(); + + std::string name = "_skOutParamHelper" + std::to_string(fSwizzleHelperCount++) + + "_" + function.mangledName(); + const char* separator = ""; + + // Emit a prototype for the function we'll be calling through to in our helper. + if (!function.isBuiltin()) { + this->writeFunctionDeclaration(function); + this->writeLine(";"); + } + + // Synthesize a helper function that takes the same inputs as `function`, except in places where + // `outVars` is non-null; in those places, we take the type of the VariableReference. + // + // float _skOutParamHelper0_originalFuncName(float _var0, float _var1, float& outParam) { + this->writeType(call.type()); + this->write(" "); + this->write(name); + this->write("("); + this->writeFunctionRequirementParams(function, separator); + + SkASSERT(outVars.size() == arguments.size()); + SkASSERT(SkToSizeT(outVars.size()) == function.parameters().size()); + + // We need to detect cases where the caller passes the same variable as an out-param more than + // once, and avoid reusing the variable name. (In those cases we can actually just ignore the + // redundant input parameter entirely, and not give it any name.) + SkTHashSet<const Variable*> writtenVars; + + for (int index = 0; index < arguments.size(); ++index) { + this->write(separator); + separator = ", "; + + const Variable* param = function.parameters()[index]; + this->writeModifiers(param->modifiers()); + + const Type* type = outVars[index] ? &outVars[index]->type() : &arguments[index]->type(); + this->writeType(*type); + + if (pass_by_reference(param->type(), param->modifiers())) { + this->write("&"); + } + if (outVars[index]) { + const Variable* var = outVars[index]->variable(); + if (!writtenVars.contains(var)) { + writtenVars.add(var); + + this->write(" "); + fIgnoreVariableReferenceModifiers = true; + this->writeVariableReference(*outVars[index]); + fIgnoreVariableReferenceModifiers = false; + } + } else { + this->write(" _var"); + this->write(std::to_string(index)); + } + } + this->writeLine(") {"); + + ++fIndentation; + for (int index = 0; index < outVars.size(); ++index) { + if (!outVars[index]) { + continue; + } + // float3 _var2[ = outParam.zyx]; + this->writeType(arguments[index]->type()); + this->write(" _var"); + this->write(std::to_string(index)); + + const Variable* param = function.parameters()[index]; + if (param->modifiers().fFlags & Modifiers::kIn_Flag) { + this->write(" = "); + fIgnoreVariableReferenceModifiers = true; + this->writeExpression(*arguments[index], Precedence::kAssignment); + fIgnoreVariableReferenceModifiers = false; + } + + this->writeLine(";"); + } + + // [int _skResult = ] myFunction(inputs, outputs, _globals, _var0, _var1, _var2, _var3); + bool hasResult = (call.type().name() != "void"); + if (hasResult) { + this->writeType(call.type()); + this->write(" _skResult = "); + } + + this->writeName(function.mangledName()); + this->write("("); + separator = ""; + this->writeFunctionRequirementArgs(function, separator); + + for (int index = 0; index < arguments.size(); ++index) { + this->write(separator); + separator = ", "; + + this->write("_var"); + this->write(std::to_string(index)); + } + this->writeLine(");"); + + for (int index = 0; index < outVars.size(); ++index) { + if (!outVars[index]) { + continue; + } + // outParam.zyx = _var2; + fIgnoreVariableReferenceModifiers = true; + this->writeExpression(*arguments[index], Precedence::kAssignment); + fIgnoreVariableReferenceModifiers = false; + this->write(" = _var"); + this->write(std::to_string(index)); + this->writeLine(";"); + } + + if (hasResult) { + this->writeLine("return _skResult;"); + } + + --fIndentation; + this->writeLine("}"); + + // Write the function out to `fExtraFunctions`. + write_stringstream(tmpStream, fExtraFunctions); + + return name; +} + +std::string MetalCodeGenerator::getBitcastIntrinsic(const Type& outType) { + return "as_type<" + outType.displayName() + ">"; +} + +void MetalCodeGenerator::writeFunctionCall(const FunctionCall& c) { + const FunctionDeclaration& function = c.function(); + + // Many intrinsics need to be rewritten in Metal. + if (function.isIntrinsic()) { + if (this->writeIntrinsicCall(c, function.intrinsicKind())) { + return; + } + } + + // Determine whether or not we need to emulate GLSL's out-param semantics for Metal using a + // helper function. (Specifically, out-parameters in GLSL are only written back to the original + // variable at the end of the function call; also, swizzles are supported, whereas Metal doesn't + // allow a swizzle to be passed to a `floatN&`.) + const ExpressionArray& arguments = c.arguments(); + const std::vector<Variable*>& parameters = function.parameters(); + SkASSERT(SkToSizeT(arguments.size()) == parameters.size()); + + bool foundOutParam = false; + SkSTArray<16, VariableReference*> outVars; + outVars.push_back_n(arguments.size(), (VariableReference*)nullptr); + + for (int index = 0; index < arguments.size(); ++index) { + // If this is an out parameter... + if (parameters[index]->modifiers().fFlags & Modifiers::kOut_Flag) { + // Find the expression's inner variable being written to. + Analysis::AssignmentInfo info; + // Assignability was verified at IRGeneration time, so this should always succeed. + SkAssertResult(Analysis::IsAssignable(*arguments[index], &info)); + outVars[index] = info.fAssignedVar; + foundOutParam = true; + } + } + + if (foundOutParam) { + // Out parameters need to be written back to at the end of the function. To do this, we + // synthesize a helper function which evaluates the out-param expression into a temporary + // variable, calls the original function, then writes the temp var back into the out param + // using the original out-param expression. (This lets us support things like swizzles and + // array indices.) + this->write(getOutParamHelper(c, arguments, outVars)); + } else { + this->write(function.mangledName()); + } + + this->write("("); + const char* separator = ""; + this->writeFunctionRequirementArgs(function, separator); + for (int i = 0; i < arguments.size(); ++i) { + this->write(separator); + separator = ", "; + + if (outVars[i]) { + this->writeExpression(*outVars[i], Precedence::kSequence); + } else { + this->writeExpression(*arguments[i], Precedence::kSequence); + } + } + this->write(")"); +} + +static constexpr char kInverse2x2[] = R"( +template <typename T> +matrix<T, 2, 2> mat2_inverse(matrix<T, 2, 2> m) { +return matrix<T, 2, 2>(m[1].y, -m[0].y, -m[1].x, m[0].x) * (1/determinant(m)); +} +)"; + +static constexpr char kInverse3x3[] = R"( +template <typename T> +matrix<T, 3, 3> mat3_inverse(matrix<T, 3, 3> m) { +T + a00 = m[0].x, a01 = m[0].y, a02 = m[0].z, + a10 = m[1].x, a11 = m[1].y, a12 = m[1].z, + a20 = m[2].x, a21 = m[2].y, a22 = m[2].z, + b01 = a22*a11 - a12*a21, + b11 = -a22*a10 + a12*a20, + b21 = a21*a10 - a11*a20, + det = a00*b01 + a01*b11 + a02*b21; +return matrix<T, 3, 3>( + b01, (-a22*a01 + a02*a21), ( a12*a01 - a02*a11), + b11, ( a22*a00 - a02*a20), (-a12*a00 + a02*a10), + b21, (-a21*a00 + a01*a20), ( a11*a00 - a01*a10)) * (1/det); +} +)"; + +static constexpr char kInverse4x4[] = R"( +template <typename T> +matrix<T, 4, 4> mat4_inverse(matrix<T, 4, 4> m) { +T + a00 = m[0].x, a01 = m[0].y, a02 = m[0].z, a03 = m[0].w, + a10 = m[1].x, a11 = m[1].y, a12 = m[1].z, a13 = m[1].w, + a20 = m[2].x, a21 = m[2].y, a22 = m[2].z, a23 = m[2].w, + a30 = m[3].x, a31 = m[3].y, a32 = m[3].z, a33 = m[3].w, + b00 = a00*a11 - a01*a10, + b01 = a00*a12 - a02*a10, + b02 = a00*a13 - a03*a10, + b03 = a01*a12 - a02*a11, + b04 = a01*a13 - a03*a11, + b05 = a02*a13 - a03*a12, + b06 = a20*a31 - a21*a30, + b07 = a20*a32 - a22*a30, + b08 = a20*a33 - a23*a30, + b09 = a21*a32 - a22*a31, + b10 = a21*a33 - a23*a31, + b11 = a22*a33 - a23*a32, + det = b00*b11 - b01*b10 + b02*b09 + b03*b08 - b04*b07 + b05*b06; +return matrix<T, 4, 4>( + a11*b11 - a12*b10 + a13*b09, + a02*b10 - a01*b11 - a03*b09, + a31*b05 - a32*b04 + a33*b03, + a22*b04 - a21*b05 - a23*b03, + a12*b08 - a10*b11 - a13*b07, + a00*b11 - a02*b08 + a03*b07, + a32*b02 - a30*b05 - a33*b01, + a20*b05 - a22*b02 + a23*b01, + a10*b10 - a11*b08 + a13*b06, + a01*b08 - a00*b10 - a03*b06, + a30*b04 - a31*b02 + a33*b00, + a21*b02 - a20*b04 - a23*b00, + a11*b07 - a10*b09 - a12*b06, + a00*b09 - a01*b07 + a02*b06, + a31*b01 - a30*b03 - a32*b00, + a20*b03 - a21*b01 + a22*b00) * (1/det); +} +)"; + +std::string MetalCodeGenerator::getInversePolyfill(const ExpressionArray& arguments) { + // Only use polyfills for a function taking a single-argument square matrix. + SkASSERT(arguments.size() == 1); + const Type& type = arguments.front()->type(); + if (type.isMatrix() && type.rows() == type.columns()) { + switch (type.rows()) { + case 2: + if (!fWrittenInverse2) { + fWrittenInverse2 = true; + fExtraFunctions.writeText(kInverse2x2); + } + return "mat2_inverse"; + case 3: + if (!fWrittenInverse3) { + fWrittenInverse3 = true; + fExtraFunctions.writeText(kInverse3x3); + } + return "mat3_inverse"; + case 4: + if (!fWrittenInverse4) { + fWrittenInverse4 = true; + fExtraFunctions.writeText(kInverse4x4); + } + return "mat4_inverse"; + } + } + SkDEBUGFAILF("no polyfill for inverse(%s)", type.description().c_str()); + return "inverse"; +} + +void MetalCodeGenerator::writeMatrixCompMult() { + static constexpr char kMatrixCompMult[] = R"( +template <typename T, int C, int R> +matrix<T, C, R> matrixCompMult(matrix<T, C, R> a, const matrix<T, C, R> b) { + for (int c = 0; c < C; ++c) { a[c] *= b[c]; } + return a; +} +)"; + if (!fWrittenMatrixCompMult) { + fWrittenMatrixCompMult = true; + fExtraFunctions.writeText(kMatrixCompMult); + } +} + +void MetalCodeGenerator::writeOuterProduct() { + static constexpr char kOuterProduct[] = R"( +template <typename T, int C, int R> +matrix<T, C, R> outerProduct(const vec<T, R> a, const vec<T, C> b) { + matrix<T, C, R> m; + for (int c = 0; c < C; ++c) { m[c] = a * b[c]; } + return m; +} +)"; + if (!fWrittenOuterProduct) { + fWrittenOuterProduct = true; + fExtraFunctions.writeText(kOuterProduct); + } +} + +std::string MetalCodeGenerator::getTempVariable(const Type& type) { + std::string tempVar = "_skTemp" + std::to_string(fVarCount++); + this->fFunctionHeader += " " + this->typeName(type) + " " + tempVar + ";\n"; + return tempVar; +} + +void MetalCodeGenerator::writeSimpleIntrinsic(const FunctionCall& c) { + // Write out an intrinsic function call exactly as-is. No muss no fuss. + this->write(c.function().name()); + this->writeArgumentList(c.arguments()); +} + +void MetalCodeGenerator::writeArgumentList(const ExpressionArray& arguments) { + this->write("("); + const char* separator = ""; + for (const std::unique_ptr<Expression>& arg : arguments) { + this->write(separator); + separator = ", "; + this->writeExpression(*arg, Precedence::kSequence); + } + this->write(")"); +} + +bool MetalCodeGenerator::writeIntrinsicCall(const FunctionCall& c, IntrinsicKind kind) { + const ExpressionArray& arguments = c.arguments(); + switch (kind) { + case k_read_IntrinsicKind: { + this->writeExpression(*arguments[0], Precedence::kTopLevel); + this->write(".read("); + this->writeExpression(*arguments[1], Precedence::kSequence); + this->write(")"); + return true; + } + case k_write_IntrinsicKind: { + this->writeExpression(*arguments[0], Precedence::kTopLevel); + this->write(".write("); + this->writeExpression(*arguments[2], Precedence::kSequence); + this->write(", "); + this->writeExpression(*arguments[1], Precedence::kSequence); + this->write(")"); + return true; + } + case k_width_IntrinsicKind: { + this->writeExpression(*arguments[0], Precedence::kTopLevel); + this->write(".get_width()"); + return true; + } + case k_height_IntrinsicKind: { + this->writeExpression(*arguments[0], Precedence::kTopLevel); + this->write(".get_height()"); + return true; + } + case k_mod_IntrinsicKind: { + // fmod(x, y) in metal calculates x - y * trunc(x / y) instead of x - y * floor(x / y) + std::string tmpX = this->getTempVariable(arguments[0]->type()); + std::string tmpY = this->getTempVariable(arguments[1]->type()); + this->write("(" + tmpX + " = "); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(", " + tmpY + " = "); + this->writeExpression(*arguments[1], Precedence::kSequence); + this->write(", " + tmpX + " - " + tmpY + " * floor(" + tmpX + " / " + tmpY + "))"); + return true; + } + // GLSL declares scalar versions of most geometric intrinsics, but these don't exist in MSL + case k_distance_IntrinsicKind: { + if (arguments[0]->type().columns() == 1) { + this->write("abs("); + this->writeExpression(*arguments[0], Precedence::kAdditive); + this->write(" - "); + this->writeExpression(*arguments[1], Precedence::kAdditive); + this->write(")"); + } else { + this->writeSimpleIntrinsic(c); + } + return true; + } + case k_dot_IntrinsicKind: { + if (arguments[0]->type().columns() == 1) { + this->write("("); + this->writeExpression(*arguments[0], Precedence::kMultiplicative); + this->write(" * "); + this->writeExpression(*arguments[1], Precedence::kMultiplicative); + this->write(")"); + } else { + this->writeSimpleIntrinsic(c); + } + return true; + } + case k_faceforward_IntrinsicKind: { + if (arguments[0]->type().columns() == 1) { + // ((((Nref) * (I) < 0) ? 1 : -1) * (N)) + this->write("(((("); + this->writeExpression(*arguments[2], Precedence::kSequence); + this->write(") * ("); + this->writeExpression(*arguments[1], Precedence::kSequence); + this->write(") < 0) ? 1 : -1) * ("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write("))"); + } else { + this->writeSimpleIntrinsic(c); + } + return true; + } + case k_length_IntrinsicKind: { + this->write(arguments[0]->type().columns() == 1 ? "abs(" : "length("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_normalize_IntrinsicKind: { + this->write(arguments[0]->type().columns() == 1 ? "sign(" : "normalize("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_packUnorm2x16_IntrinsicKind: { + this->write("pack_float_to_unorm2x16("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_unpackUnorm2x16_IntrinsicKind: { + this->write("unpack_unorm2x16_to_float("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_packSnorm2x16_IntrinsicKind: { + this->write("pack_float_to_snorm2x16("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_unpackSnorm2x16_IntrinsicKind: { + this->write("unpack_snorm2x16_to_float("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_packUnorm4x8_IntrinsicKind: { + this->write("pack_float_to_unorm4x8("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_unpackUnorm4x8_IntrinsicKind: { + this->write("unpack_unorm4x8_to_float("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_packSnorm4x8_IntrinsicKind: { + this->write("pack_float_to_snorm4x8("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_unpackSnorm4x8_IntrinsicKind: { + this->write("unpack_snorm4x8_to_float("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_packHalf2x16_IntrinsicKind: { + this->write("as_type<uint>(half2("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write("))"); + return true; + } + case k_unpackHalf2x16_IntrinsicKind: { + this->write("float2(as_type<half2>("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write("))"); + return true; + } + case k_floatBitsToInt_IntrinsicKind: + case k_floatBitsToUint_IntrinsicKind: + case k_intBitsToFloat_IntrinsicKind: + case k_uintBitsToFloat_IntrinsicKind: { + this->write(this->getBitcastIntrinsic(c.type())); + this->write("("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_degrees_IntrinsicKind: { + this->write("(("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(") * 57.2957795)"); + return true; + } + case k_radians_IntrinsicKind: { + this->write("(("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(") * 0.0174532925)"); + return true; + } + case k_dFdx_IntrinsicKind: { + this->write("dfdx"); + this->writeArgumentList(c.arguments()); + return true; + } + case k_dFdy_IntrinsicKind: { + if (!fRTFlipName.empty()) { + this->write("(" + fRTFlipName + ".y * dfdy"); + } else { + this->write("(dfdy"); + } + this->writeArgumentList(c.arguments()); + this->write(")"); + return true; + } + case k_inverse_IntrinsicKind: { + this->write(this->getInversePolyfill(arguments)); + this->writeArgumentList(c.arguments()); + return true; + } + case k_inversesqrt_IntrinsicKind: { + this->write("rsqrt"); + this->writeArgumentList(c.arguments()); + return true; + } + case k_atan_IntrinsicKind: { + this->write(c.arguments().size() == 2 ? "atan2" : "atan"); + this->writeArgumentList(c.arguments()); + return true; + } + case k_reflect_IntrinsicKind: { + if (arguments[0]->type().columns() == 1) { + // We need to synthesize `I - 2 * N * I * N`. + std::string tmpI = this->getTempVariable(arguments[0]->type()); + std::string tmpN = this->getTempVariable(arguments[1]->type()); + + // (_skTempI = ... + this->write("(" + tmpI + " = "); + this->writeExpression(*arguments[0], Precedence::kSequence); + + // , _skTempN = ... + this->write(", " + tmpN + " = "); + this->writeExpression(*arguments[1], Precedence::kSequence); + + // , _skTempI - 2 * _skTempN * _skTempI * _skTempN) + this->write(", " + tmpI + " - 2 * " + tmpN + " * " + tmpI + " * " + tmpN + ")"); + } else { + this->writeSimpleIntrinsic(c); + } + return true; + } + case k_refract_IntrinsicKind: { + if (arguments[0]->type().columns() == 1) { + // Metal does implement refract for vectors; rather than reimplementing refract from + // scratch, we can replace the call with `refract(float2(I,0), float2(N,0), eta).x`. + this->write("(refract(float2("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(", 0), float2("); + this->writeExpression(*arguments[1], Precedence::kSequence); + this->write(", 0), "); + this->writeExpression(*arguments[2], Precedence::kSequence); + this->write(").x)"); + } else { + this->writeSimpleIntrinsic(c); + } + return true; + } + case k_roundEven_IntrinsicKind: { + this->write("rint"); + this->writeArgumentList(c.arguments()); + return true; + } + case k_bitCount_IntrinsicKind: { + this->write("popcount("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write(")"); + return true; + } + case k_findLSB_IntrinsicKind: { + // Create a temp variable to store the expression, to avoid double-evaluating it. + std::string skTemp = this->getTempVariable(arguments[0]->type()); + std::string exprType = this->typeName(arguments[0]->type()); + + // ctz returns numbits(type) on zero inputs; GLSL documents it as generating -1 instead. + // Use select to detect zero inputs and force a -1 result. + + // (_skTemp1 = (.....), select(ctz(_skTemp1), int4(-1), _skTemp1 == int4(0))) + this->write("("); + this->write(skTemp); + this->write(" = ("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write("), select(ctz("); + this->write(skTemp); + this->write("), "); + this->write(exprType); + this->write("(-1), "); + this->write(skTemp); + this->write(" == "); + this->write(exprType); + this->write("(0)))"); + return true; + } + case k_findMSB_IntrinsicKind: { + // Create a temp variable to store the expression, to avoid double-evaluating it. + std::string skTemp1 = this->getTempVariable(arguments[0]->type()); + std::string exprType = this->typeName(arguments[0]->type()); + + // GLSL findMSB is actually quite different from Metal's clz: + // - For signed negative numbers, it returns the first zero bit, not the first one bit! + // - For an empty input (0/~0 depending on sign), findMSB gives -1; clz is numbits(type) + + // (_skTemp1 = (.....), + this->write("("); + this->write(skTemp1); + this->write(" = ("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write("), "); + + // Signed input types might be negative; we need another helper variable to negate the + // input (since we can only find one bits, not zero bits). + std::string skTemp2; + if (arguments[0]->type().isSigned()) { + // ... _skTemp2 = (select(_skTemp1, ~_skTemp1, _skTemp1 < 0)), + skTemp2 = this->getTempVariable(arguments[0]->type()); + this->write(skTemp2); + this->write(" = (select("); + this->write(skTemp1); + this->write(", ~"); + this->write(skTemp1); + this->write(", "); + this->write(skTemp1); + this->write(" < 0)), "); + } else { + skTemp2 = skTemp1; + } + + // ... select(int4(clz(_skTemp2)), int4(-1), _skTemp2 == int4(0))) + this->write("select("); + this->write(this->typeName(c.type())); + this->write("(clz("); + this->write(skTemp2); + this->write(")), "); + this->write(this->typeName(c.type())); + this->write("(-1), "); + this->write(skTemp2); + this->write(" == "); + this->write(exprType); + this->write("(0)))"); + return true; + } + case k_sign_IntrinsicKind: { + if (arguments[0]->type().componentType().isInteger()) { + // Create a temp variable to store the expression, to avoid double-evaluating it. + std::string skTemp = this->getTempVariable(arguments[0]->type()); + std::string exprType = this->typeName(arguments[0]->type()); + + // (_skTemp = (.....), + this->write("("); + this->write(skTemp); + this->write(" = ("); + this->writeExpression(*arguments[0], Precedence::kSequence); + this->write("), "); + + // ... select(select(int4(0), int4(-1), _skTemp < 0), int4(1), _skTemp > 0)) + this->write("select(select("); + this->write(exprType); + this->write("(0), "); + this->write(exprType); + this->write("(-1), "); + this->write(skTemp); + this->write(" < 0), "); + this->write(exprType); + this->write("(1), "); + this->write(skTemp); + this->write(" > 0))"); + } else { + this->writeSimpleIntrinsic(c); + } + return true; + } + case k_matrixCompMult_IntrinsicKind: { + this->writeMatrixCompMult(); + this->writeSimpleIntrinsic(c); + return true; + } + case k_outerProduct_IntrinsicKind: { + this->writeOuterProduct(); + this->writeSimpleIntrinsic(c); + return true; + } + case k_mix_IntrinsicKind: { + SkASSERT(c.arguments().size() == 3); + if (arguments[2]->type().componentType().isBoolean()) { + // The Boolean forms of GLSL mix() use the select() intrinsic in Metal. + this->write("select"); + this->writeArgumentList(c.arguments()); + return true; + } + // The basic form of mix() is supported by Metal as-is. + this->writeSimpleIntrinsic(c); + return true; + } + case k_equal_IntrinsicKind: + case k_greaterThan_IntrinsicKind: + case k_greaterThanEqual_IntrinsicKind: + case k_lessThan_IntrinsicKind: + case k_lessThanEqual_IntrinsicKind: + case k_notEqual_IntrinsicKind: { + this->write("("); + this->writeExpression(*c.arguments()[0], Precedence::kRelational); + switch (kind) { + case k_equal_IntrinsicKind: + this->write(" == "); + break; + case k_notEqual_IntrinsicKind: + this->write(" != "); + break; + case k_lessThan_IntrinsicKind: + this->write(" < "); + break; + case k_lessThanEqual_IntrinsicKind: + this->write(" <= "); + break; + case k_greaterThan_IntrinsicKind: + this->write(" > "); + break; + case k_greaterThanEqual_IntrinsicKind: + this->write(" >= "); + break; + default: + SK_ABORT("unsupported comparison intrinsic kind"); + } + this->writeExpression(*c.arguments()[1], Precedence::kRelational); + this->write(")"); + return true; + } + case k_storageBarrier_IntrinsicKind: + this->write("threadgroup_barrier(mem_flags::mem_device)"); + return true; + case k_workgroupBarrier_IntrinsicKind: + this->write("threadgroup_barrier(mem_flags::mem_threadgroup)"); + return true; + case k_atomicAdd_IntrinsicKind: + this->write("atomic_fetch_add_explicit(&"); + this->writeExpression(*c.arguments()[0], Precedence::kSequence); + this->write(", "); + this->writeExpression(*c.arguments()[1], Precedence::kSequence); + this->write(", memory_order_relaxed)"); + return true; + case k_atomicLoad_IntrinsicKind: + this->write("atomic_load_explicit(&"); + this->writeExpression(*c.arguments()[0], Precedence::kSequence); + this->write(", memory_order_relaxed)"); + return true; + case k_atomicStore_IntrinsicKind: + this->write("atomic_store_explicit(&"); + this->writeExpression(*c.arguments()[0], Precedence::kSequence); + this->write(", "); + this->writeExpression(*c.arguments()[1], Precedence::kSequence); + this->write(", memory_order_relaxed)"); + return true; + default: + return false; + } +} + +// Assembles a matrix of type floatRxC by resizing another matrix named `x0`. +// Cells that don't exist in the source matrix will be populated with identity-matrix values. +void MetalCodeGenerator::assembleMatrixFromMatrix(const Type& sourceMatrix, int rows, int columns) { + SkASSERT(rows <= 4); + SkASSERT(columns <= 4); + + std::string matrixType = this->typeName(sourceMatrix.componentType()); + + const char* separator = ""; + for (int c = 0; c < columns; ++c) { + fExtraFunctions.printf("%s%s%d(", separator, matrixType.c_str(), rows); + separator = "), "; + + // Determine how many values to take from the source matrix for this row. + int swizzleLength = 0; + if (c < sourceMatrix.columns()) { + swizzleLength = std::min<>(rows, sourceMatrix.rows()); + } + + // Emit all the values from the source matrix row. + bool firstItem; + switch (swizzleLength) { + case 0: firstItem = true; break; + case 1: firstItem = false; fExtraFunctions.printf("x0[%d].x", c); break; + case 2: firstItem = false; fExtraFunctions.printf("x0[%d].xy", c); break; + case 3: firstItem = false; fExtraFunctions.printf("x0[%d].xyz", c); break; + case 4: firstItem = false; fExtraFunctions.printf("x0[%d].xyzw", c); break; + default: SkUNREACHABLE; + } + + // Emit the placeholder identity-matrix cells. + for (int r = swizzleLength; r < rows; ++r) { + fExtraFunctions.printf("%s%s", firstItem ? "" : ", ", (r == c) ? "1.0" : "0.0"); + firstItem = false; + } + } + + fExtraFunctions.writeText(")"); +} + +// Assembles a matrix of type floatCxR by concatenating an arbitrary mix of values, named `x0`, +// `x1`, etc. An error is written if the expression list don't contain exactly C*R scalars. +void MetalCodeGenerator::assembleMatrixFromExpressions(const AnyConstructor& ctor, + int columns, int rows) { + SkASSERT(rows <= 4); + SkASSERT(columns <= 4); + + std::string matrixType = this->typeName(ctor.type().componentType()); + size_t argIndex = 0; + int argPosition = 0; + auto args = ctor.argumentSpan(); + + static constexpr char kSwizzle[] = "xyzw"; + const char* separator = ""; + for (int c = 0; c < columns; ++c) { + fExtraFunctions.printf("%s%s%d(", separator, matrixType.c_str(), rows); + separator = "), "; + + const char* columnSeparator = ""; + for (int r = 0; r < rows;) { + fExtraFunctions.writeText(columnSeparator); + columnSeparator = ", "; + + if (argIndex < args.size()) { + const Type& argType = args[argIndex]->type(); + switch (argType.typeKind()) { + case Type::TypeKind::kScalar: { + fExtraFunctions.printf("x%zu", argIndex); + ++r; + ++argPosition; + break; + } + case Type::TypeKind::kVector: { + fExtraFunctions.printf("x%zu.", argIndex); + do { + fExtraFunctions.write8(kSwizzle[argPosition]); + ++r; + ++argPosition; + } while (r < rows && argPosition < argType.columns()); + break; + } + case Type::TypeKind::kMatrix: { + fExtraFunctions.printf("x%zu[%d].", argIndex, argPosition / argType.rows()); + do { + fExtraFunctions.write8(kSwizzle[argPosition]); + ++r; + ++argPosition; + } while (r < rows && (argPosition % argType.rows()) != 0); + break; + } + default: { + SkDEBUGFAIL("incorrect type of argument for matrix constructor"); + fExtraFunctions.writeText("<error>"); + break; + } + } + + if (argPosition >= argType.columns() * argType.rows()) { + ++argIndex; + argPosition = 0; + } + } else { + SkDEBUGFAIL("not enough arguments for matrix constructor"); + fExtraFunctions.writeText("<error>"); + } + } + } + + if (argPosition != 0 || argIndex != args.size()) { + SkDEBUGFAIL("incorrect number of arguments for matrix constructor"); + fExtraFunctions.writeText(", <error>"); + } + + fExtraFunctions.writeText(")"); +} + +// Generates a constructor for 'matrix' which reorganizes the input arguments into the proper shape. +// Keeps track of previously generated constructors so that we won't generate more than one +// constructor for any given permutation of input argument types. Returns the name of the +// generated constructor method. +std::string MetalCodeGenerator::getMatrixConstructHelper(const AnyConstructor& c) { + const Type& type = c.type(); + int columns = type.columns(); + int rows = type.rows(); + auto args = c.argumentSpan(); + std::string typeName = this->typeName(type); + + // Create the helper-method name and use it as our lookup key. + std::string name = String::printf("%s_from", typeName.c_str()); + for (const std::unique_ptr<Expression>& expr : args) { + String::appendf(&name, "_%s", this->typeName(expr->type()).c_str()); + } + + // If a helper-method has not been synthesized yet, create it now. + if (!fHelpers.contains(name)) { + fHelpers.add(name); + + // Unlike GLSL, Metal requires that matrices are initialized with exactly R vectors of C + // components apiece. (In Metal 2.0, you can also supply R*C scalars, but you still cannot + // supply a mixture of scalars and vectors.) + fExtraFunctions.printf("%s %s(", typeName.c_str(), name.c_str()); + + size_t argIndex = 0; + const char* argSeparator = ""; + for (const std::unique_ptr<Expression>& expr : args) { + fExtraFunctions.printf("%s%s x%zu", argSeparator, + this->typeName(expr->type()).c_str(), argIndex++); + argSeparator = ", "; + } + + fExtraFunctions.printf(") {\n return %s(", typeName.c_str()); + + if (args.size() == 1 && args.front()->type().isMatrix()) { + this->assembleMatrixFromMatrix(args.front()->type(), rows, columns); + } else { + this->assembleMatrixFromExpressions(c, columns, rows); + } + + fExtraFunctions.writeText(");\n}\n"); + } + return name; +} + +bool MetalCodeGenerator::matrixConstructHelperIsNeeded(const ConstructorCompound& c) { + SkASSERT(c.type().isMatrix()); + + // GLSL is fairly free-form about inputs to its matrix constructors, but Metal is not; it + // expects exactly R vectors of C components apiece. (Metal 2.0 also allows a list of R*C + // scalars.) Some cases are simple to translate and so we handle those inline--e.g. a list of + // scalars can be constructed trivially. In more complex cases, we generate a helper function + // that converts our inputs into a properly-shaped matrix. + // A matrix construct helper method is always used if any input argument is a matrix. + // Helper methods are also necessary when any argument would span multiple rows. For instance: + // + // float2 x = (1, 2); + // float3x2(x, 3, 4, 5, 6) = | 1 3 5 | = no helper needed; conversion can be done inline + // | 2 4 6 | + // + // float2 x = (2, 3); + // float3x2(1, x, 4, 5, 6) = | 1 3 5 | = x spans multiple rows; a helper method will be used + // | 2 4 6 | + // + // float4 x = (1, 2, 3, 4); + // float2x2(x) = | 1 3 | = x spans multiple rows; a helper method will be used + // | 2 4 | + // + + int position = 0; + for (const std::unique_ptr<Expression>& expr : c.arguments()) { + // If an input argument is a matrix, we need a helper function. + if (expr->type().isMatrix()) { + return true; + } + position += expr->type().columns(); + if (position > c.type().rows()) { + // An input argument would span multiple rows; a helper function is required. + return true; + } + if (position == c.type().rows()) { + // We've advanced to the end of a row. Wrap to the start of the next row. + position = 0; + } + } + + return false; +} + +void MetalCodeGenerator::writeConstructorMatrixResize(const ConstructorMatrixResize& c, + Precedence parentPrecedence) { + // Matrix-resize via casting doesn't natively exist in Metal at all, so we always need to use a + // matrix-construct helper here. + this->write(this->getMatrixConstructHelper(c)); + this->write("("); + this->writeExpression(*c.argument(), Precedence::kSequence); + this->write(")"); +} + +void MetalCodeGenerator::writeConstructorCompound(const ConstructorCompound& c, + Precedence parentPrecedence) { + if (c.type().isVector()) { + this->writeConstructorCompoundVector(c, parentPrecedence); + } else if (c.type().isMatrix()) { + this->writeConstructorCompoundMatrix(c, parentPrecedence); + } else { + fContext.fErrors->error(c.fPosition, "unsupported compound constructor"); + } +} + +void MetalCodeGenerator::writeConstructorArrayCast(const ConstructorArrayCast& c, + Precedence parentPrecedence) { + const Type& inType = c.argument()->type().componentType(); + const Type& outType = c.type().componentType(); + std::string inTypeName = this->typeName(inType); + std::string outTypeName = this->typeName(outType); + + std::string name = "array_of_" + outTypeName + "_from_" + inTypeName; + if (!fHelpers.contains(name)) { + fHelpers.add(name); + fExtraFunctions.printf(R"( +template <size_t N> +array<%s, N> %s(thread const array<%s, N>& x) { + array<%s, N> result; + for (int i = 0; i < N; ++i) { + result[i] = %s(x[i]); + } + return result; +} +)", + outTypeName.c_str(), name.c_str(), inTypeName.c_str(), + outTypeName.c_str(), + outTypeName.c_str()); + } + + this->write(name); + this->write("("); + this->writeExpression(*c.argument(), Precedence::kSequence); + this->write(")"); +} + +std::string MetalCodeGenerator::getVectorFromMat2x2ConstructorHelper(const Type& matrixType) { + SkASSERT(matrixType.isMatrix()); + SkASSERT(matrixType.rows() == 2); + SkASSERT(matrixType.columns() == 2); + + std::string baseType = this->typeName(matrixType.componentType()); + std::string name = String::printf("%s4_from_%s2x2", baseType.c_str(), baseType.c_str()); + if (!fHelpers.contains(name)) { + fHelpers.add(name); + + fExtraFunctions.printf(R"( +%s4 %s(%s2x2 x) { + return %s4(x[0].xy, x[1].xy); +} +)", baseType.c_str(), name.c_str(), baseType.c_str(), baseType.c_str()); + } + + return name; +} + +void MetalCodeGenerator::writeConstructorCompoundVector(const ConstructorCompound& c, + Precedence parentPrecedence) { + SkASSERT(c.type().isVector()); + + // Metal supports constructing vectors from a mix of scalars and vectors, but not matrices. + // GLSL supports vec4(mat2x2), so we detect that case here and emit a helper function. + if (c.type().columns() == 4 && c.argumentSpan().size() == 1) { + const Expression& expr = *c.argumentSpan().front(); + if (expr.type().isMatrix()) { + this->write(this->getVectorFromMat2x2ConstructorHelper(expr.type())); + this->write("("); + this->writeExpression(expr, Precedence::kSequence); + this->write(")"); + return; + } + } + + this->writeAnyConstructor(c, "(", ")", parentPrecedence); +} + +void MetalCodeGenerator::writeConstructorCompoundMatrix(const ConstructorCompound& c, + Precedence parentPrecedence) { + SkASSERT(c.type().isMatrix()); + + // Emit and invoke a matrix-constructor helper method if one is necessary. + if (this->matrixConstructHelperIsNeeded(c)) { + this->write(this->getMatrixConstructHelper(c)); + this->write("("); + const char* separator = ""; + for (const std::unique_ptr<Expression>& expr : c.arguments()) { + this->write(separator); + separator = ", "; + this->writeExpression(*expr, Precedence::kSequence); + } + this->write(")"); + return; + } + + // Metal doesn't allow creating matrices by passing in scalars and vectors in a jumble; it + // requires your scalars to be grouped up into columns. Because `matrixConstructHelperIsNeeded` + // returned false, we know that none of our scalars/vectors "wrap" across across a column, so we + // can group our inputs up and synthesize a constructor for each column. + const Type& matrixType = c.type(); + const Type& columnType = matrixType.componentType().toCompound( + fContext, /*columns=*/matrixType.rows(), /*rows=*/1); + + this->writeType(matrixType); + this->write("("); + const char* separator = ""; + int scalarCount = 0; + for (const std::unique_ptr<Expression>& arg : c.arguments()) { + this->write(separator); + separator = ", "; + if (arg->type().columns() < matrixType.rows()) { + // Write a `floatN(` constructor to group scalars and smaller vectors together. + if (!scalarCount) { + this->writeType(columnType); + this->write("("); + } + scalarCount += arg->type().columns(); + } + this->writeExpression(*arg, Precedence::kSequence); + if (scalarCount && scalarCount == matrixType.rows()) { + // Close our `floatN(...` constructor block from above. + this->write(")"); + scalarCount = 0; + } + } + this->write(")"); +} + +void MetalCodeGenerator::writeAnyConstructor(const AnyConstructor& c, + const char* leftBracket, + const char* rightBracket, + Precedence parentPrecedence) { + this->writeType(c.type()); + this->write(leftBracket); + const char* separator = ""; + for (const std::unique_ptr<Expression>& arg : c.argumentSpan()) { + this->write(separator); + separator = ", "; + this->writeExpression(*arg, Precedence::kSequence); + } + this->write(rightBracket); +} + +void MetalCodeGenerator::writeCastConstructor(const AnyConstructor& c, + const char* leftBracket, + const char* rightBracket, + Precedence parentPrecedence) { + return this->writeAnyConstructor(c, leftBracket, rightBracket, parentPrecedence); +} + +void MetalCodeGenerator::writeFragCoord() { + if (!fRTFlipName.empty()) { + this->write("float4(_fragCoord.x, "); + this->write(fRTFlipName.c_str()); + this->write(".x + "); + this->write(fRTFlipName.c_str()); + this->write(".y * _fragCoord.y, 0.0, _fragCoord.w)"); + } else { + this->write("float4(_fragCoord.x, _fragCoord.y, 0.0, _fragCoord.w)"); + } +} + +static bool is_compute_builtin(const Variable& var) { + switch (var.modifiers().fLayout.fBuiltin) { + case SK_NUMWORKGROUPS_BUILTIN: + case SK_WORKGROUPID_BUILTIN: + case SK_LOCALINVOCATIONID_BUILTIN: + case SK_GLOBALINVOCATIONID_BUILTIN: + case SK_LOCALINVOCATIONINDEX_BUILTIN: + return true; + default: + break; + } + return false; +} + +// true if the var is part of the Inputs struct +static bool is_input(const Variable& var) { + SkASSERT(var.storage() == VariableStorage::kGlobal); + return var.modifiers().fFlags & Modifiers::kIn_Flag && + (var.modifiers().fLayout.fBuiltin == -1 || is_compute_builtin(var)) && + var.type().typeKind() != Type::TypeKind::kTexture; +} + +// true if the var is part of the Outputs struct +static bool is_output(const Variable& var) { + SkASSERT(var.storage() == VariableStorage::kGlobal); + // inout vars get written into the Inputs struct, so we exclude them from Outputs + return (var.modifiers().fFlags & Modifiers::kOut_Flag) && + !(var.modifiers().fFlags & Modifiers::kIn_Flag) && + var.modifiers().fLayout.fBuiltin == -1 && + var.type().typeKind() != Type::TypeKind::kTexture; +} + +// true if the var is part of the Uniforms struct +static bool is_uniforms(const Variable& var) { + SkASSERT(var.storage() == VariableStorage::kGlobal); + return var.modifiers().fFlags & Modifiers::kUniform_Flag && + var.type().typeKind() != Type::TypeKind::kSampler; +} + +// true if the var is part of the Threadgroups struct +static bool is_threadgroup(const Variable& var) { + SkASSERT(var.storage() == VariableStorage::kGlobal); + return var.modifiers().fFlags & Modifiers::kWorkgroup_Flag; +} + +// true if the var is part of the Globals struct +static bool is_in_globals(const Variable& var) { + SkASSERT(var.storage() == VariableStorage::kGlobal); + return !(var.modifiers().fFlags & Modifiers::kConst_Flag); +} + +void MetalCodeGenerator::writeVariableReference(const VariableReference& ref) { + // When assembling out-param helper functions, we copy variables into local clones with matching + // names. We never want to prepend "_in." or "_globals." when writing these variables since + // we're actually targeting the clones. + if (fIgnoreVariableReferenceModifiers) { + this->writeName(ref.variable()->mangledName()); + return; + } + + switch (ref.variable()->modifiers().fLayout.fBuiltin) { + case SK_FRAGCOLOR_BUILTIN: + this->write("_out.sk_FragColor"); + break; + case SK_FRAGCOORD_BUILTIN: + this->writeFragCoord(); + break; + case SK_VERTEXID_BUILTIN: + this->write("sk_VertexID"); + break; + case SK_INSTANCEID_BUILTIN: + this->write("sk_InstanceID"); + break; + case SK_CLOCKWISE_BUILTIN: + // We'd set the front facing winding in the MTLRenderCommandEncoder to be counter + // clockwise to match Skia convention. + if (!fRTFlipName.empty()) { + this->write("(" + fRTFlipName + ".y < 0 ? _frontFacing : !_frontFacing)"); + } else { + this->write("_frontFacing"); + } + break; + default: + const Variable& var = *ref.variable(); + if (var.storage() == Variable::Storage::kGlobal) { + if (is_input(var)) { + this->write("_in."); + } else if (is_output(var)) { + this->write("_out."); + } else if (is_uniforms(var)) { + this->write("_uniforms."); + } else if (is_threadgroup(var)) { + this->write("_threadgroups."); + } else if (is_in_globals(var)) { + this->write("_globals."); + } + } + this->writeName(var.mangledName()); + } +} + +void MetalCodeGenerator::writeIndexExpression(const IndexExpression& expr) { + // Metal does not seem to handle assignment into `vec.zyx[i]` properly--it compiles, but the + // results are wrong. We rewrite the expression as `vec[uint3(2,1,0)[i]]` instead. (Filed with + // Apple as FB12055941.) + if (expr.base()->is<Swizzle>()) { + const Swizzle& swizzle = expr.base()->as<Swizzle>(); + if (swizzle.components().size() > 1) { + this->writeExpression(*swizzle.base(), Precedence::kPostfix); + this->write("[uint" + std::to_string(swizzle.components().size()) + "("); + auto separator = SkSL::String::Separator(); + for (int8_t component : swizzle.components()) { + this->write(separator()); + this->write(std::to_string(component)); + } + this->write(")["); + this->writeExpression(*expr.index(), Precedence::kTopLevel); + this->write("]]"); + return; + } + } + + this->writeExpression(*expr.base(), Precedence::kPostfix); + this->write("["); + this->writeExpression(*expr.index(), Precedence::kTopLevel); + this->write("]"); +} + +void MetalCodeGenerator::writeFieldAccess(const FieldAccess& f) { + const Type::Field* field = &f.base()->type().fields()[f.fieldIndex()]; + if (FieldAccess::OwnerKind::kDefault == f.ownerKind()) { + this->writeExpression(*f.base(), Precedence::kPostfix); + this->write("."); + } + switch (field->fModifiers.fLayout.fBuiltin) { + case SK_POSITION_BUILTIN: + this->write("_out.sk_Position"); + break; + case SK_POINTSIZE_BUILTIN: + this->write("_out.sk_PointSize"); + break; + default: + if (FieldAccess::OwnerKind::kAnonymousInterfaceBlock == f.ownerKind()) { + this->write("_globals."); + this->write(fInterfaceBlockNameMap[fInterfaceBlockMap[field]]); + this->write("->"); + } + this->writeName(field->fName); + } +} + +void MetalCodeGenerator::writeSwizzle(const Swizzle& swizzle) { + this->writeExpression(*swizzle.base(), Precedence::kPostfix); + this->write("."); + for (int c : swizzle.components()) { + SkASSERT(c >= 0 && c <= 3); + this->write(&("x\0y\0z\0w\0"[c * 2])); + } +} + +void MetalCodeGenerator::writeMatrixTimesEqualHelper(const Type& left, const Type& right, + const Type& result) { + SkASSERT(left.isMatrix()); + SkASSERT(right.isMatrix()); + SkASSERT(result.isMatrix()); + + std::string key = "Matrix *= " + this->typeName(left) + ":" + this->typeName(right); + + if (!fHelpers.contains(key)) { + fHelpers.add(key); + fExtraFunctions.printf("thread %s& operator*=(thread %s& left, thread const %s& right) {\n" + " left = left * right;\n" + " return left;\n" + "}\n", + this->typeName(result).c_str(), this->typeName(left).c_str(), + this->typeName(right).c_str()); + } +} + +void MetalCodeGenerator::writeMatrixEqualityHelpers(const Type& left, const Type& right) { + SkASSERT(left.isMatrix()); + SkASSERT(right.isMatrix()); + SkASSERT(left.rows() == right.rows()); + SkASSERT(left.columns() == right.columns()); + + std::string key = "Matrix == " + this->typeName(left) + ":" + this->typeName(right); + + if (!fHelpers.contains(key)) { + fHelpers.add(key); + fExtraFunctionPrototypes.printf(R"( +thread bool operator==(const %s left, const %s right); +thread bool operator!=(const %s left, const %s right); +)", + this->typeName(left).c_str(), + this->typeName(right).c_str(), + this->typeName(left).c_str(), + this->typeName(right).c_str()); + + fExtraFunctions.printf( + "thread bool operator==(const %s left, const %s right) {\n" + " return ", + this->typeName(left).c_str(), this->typeName(right).c_str()); + + const char* separator = ""; + for (int index=0; index<left.columns(); ++index) { + fExtraFunctions.printf("%sall(left[%d] == right[%d])", separator, index, index); + separator = " &&\n "; + } + + fExtraFunctions.printf( + ";\n" + "}\n" + "thread bool operator!=(const %s left, const %s right) {\n" + " return !(left == right);\n" + "}\n", + this->typeName(left).c_str(), this->typeName(right).c_str()); + } +} + +void MetalCodeGenerator::writeMatrixDivisionHelpers(const Type& type) { + SkASSERT(type.isMatrix()); + + std::string key = "Matrix / " + this->typeName(type); + + if (!fHelpers.contains(key)) { + fHelpers.add(key); + std::string typeName = this->typeName(type); + + fExtraFunctions.printf( + "thread %s operator/(const %s left, const %s right) {\n" + " return %s(", + typeName.c_str(), typeName.c_str(), typeName.c_str(), typeName.c_str()); + + const char* separator = ""; + for (int index=0; index<type.columns(); ++index) { + fExtraFunctions.printf("%sleft[%d] / right[%d]", separator, index, index); + separator = ", "; + } + + fExtraFunctions.printf(");\n" + "}\n" + "thread %s& operator/=(thread %s& left, thread const %s& right) {\n" + " left = left / right;\n" + " return left;\n" + "}\n", + typeName.c_str(), typeName.c_str(), typeName.c_str()); + } +} + +void MetalCodeGenerator::writeArrayEqualityHelpers(const Type& type) { + SkASSERT(type.isArray()); + + // If the array's component type needs a helper as well, we need to emit that one first. + this->writeEqualityHelpers(type.componentType(), type.componentType()); + + std::string key = "ArrayEquality []"; + if (!fHelpers.contains(key)) { + fHelpers.add(key); + fExtraFunctionPrototypes.writeText(R"( +template <typename T1, typename T2> +bool operator==(const array_ref<T1> left, const array_ref<T2> right); +template <typename T1, typename T2> +bool operator!=(const array_ref<T1> left, const array_ref<T2> right); +)"); + fExtraFunctions.writeText(R"( +template <typename T1, typename T2> +bool operator==(const array_ref<T1> left, const array_ref<T2> right) { + if (left.size() != right.size()) { + return false; + } + for (size_t index = 0; index < left.size(); ++index) { + if (!all(left[index] == right[index])) { + return false; + } + } + return true; +} + +template <typename T1, typename T2> +bool operator!=(const array_ref<T1> left, const array_ref<T2> right) { + return !(left == right); +} +)"); + } +} + +void MetalCodeGenerator::writeStructEqualityHelpers(const Type& type) { + SkASSERT(type.isStruct()); + std::string key = "StructEquality " + this->typeName(type); + + if (!fHelpers.contains(key)) { + fHelpers.add(key); + // If one of the struct's fields needs a helper as well, we need to emit that one first. + for (const Type::Field& field : type.fields()) { + this->writeEqualityHelpers(*field.fType, *field.fType); + } + + // Write operator== and operator!= for this struct, since those are assumed to exist in SkSL + // and GLSL but do not exist by default in Metal. + fExtraFunctionPrototypes.printf(R"( +thread bool operator==(thread const %s& left, thread const %s& right); +thread bool operator!=(thread const %s& left, thread const %s& right); +)", + this->typeName(type).c_str(), + this->typeName(type).c_str(), + this->typeName(type).c_str(), + this->typeName(type).c_str()); + + fExtraFunctions.printf( + "thread bool operator==(thread const %s& left, thread const %s& right) {\n" + " return ", + this->typeName(type).c_str(), + this->typeName(type).c_str()); + + const char* separator = ""; + for (const Type::Field& field : type.fields()) { + if (field.fType->isArray()) { + fExtraFunctions.printf( + "%s(make_array_ref(left.%.*s) == make_array_ref(right.%.*s))", + separator, + (int)field.fName.size(), field.fName.data(), + (int)field.fName.size(), field.fName.data()); + } else { + fExtraFunctions.printf("%sall(left.%.*s == right.%.*s)", + separator, + (int)field.fName.size(), field.fName.data(), + (int)field.fName.size(), field.fName.data()); + } + separator = " &&\n "; + } + fExtraFunctions.printf( + ";\n" + "}\n" + "thread bool operator!=(thread const %s& left, thread const %s& right) {\n" + " return !(left == right);\n" + "}\n", + this->typeName(type).c_str(), + this->typeName(type).c_str()); + } +} + +void MetalCodeGenerator::writeEqualityHelpers(const Type& leftType, const Type& rightType) { + if (leftType.isArray() && rightType.isArray()) { + this->writeArrayEqualityHelpers(leftType); + return; + } + if (leftType.isStruct() && rightType.isStruct()) { + this->writeStructEqualityHelpers(leftType); + return; + } + if (leftType.isMatrix() && rightType.isMatrix()) { + this->writeMatrixEqualityHelpers(leftType, rightType); + return; + } +} + +void MetalCodeGenerator::writeNumberAsMatrix(const Expression& expr, const Type& matrixType) { + SkASSERT(expr.type().isNumber()); + SkASSERT(matrixType.isMatrix()); + + // Componentwise multiply the scalar against a matrix of the desired size which contains all 1s. + this->write("("); + this->writeType(matrixType); + this->write("("); + + const char* separator = ""; + for (int index = matrixType.slotCount(); index--;) { + this->write(separator); + this->write("1.0"); + separator = ", "; + } + + this->write(") * "); + this->writeExpression(expr, Precedence::kMultiplicative); + this->write(")"); +} + +void MetalCodeGenerator::writeBinaryExpressionElement(const Expression& expr, + Operator op, + const Expression& other, + Precedence precedence) { + bool needMatrixSplatOnScalar = other.type().isMatrix() && expr.type().isNumber() && + op.isValidForMatrixOrVector() && + op.removeAssignment().kind() != Operator::Kind::STAR; + if (needMatrixSplatOnScalar) { + this->writeNumberAsMatrix(expr, other.type()); + } else if (op.isEquality() && expr.type().isArray()) { + this->write("make_array_ref("); + this->writeExpression(expr, precedence); + this->write(")"); + } else { + this->writeExpression(expr, precedence); + } +} + +void MetalCodeGenerator::writeBinaryExpression(const BinaryExpression& b, + Precedence parentPrecedence) { + const Expression& left = *b.left(); + const Expression& right = *b.right(); + const Type& leftType = left.type(); + const Type& rightType = right.type(); + Operator op = b.getOperator(); + Precedence precedence = op.getBinaryPrecedence(); + bool needParens = precedence >= parentPrecedence; + switch (op.kind()) { + case Operator::Kind::EQEQ: + this->writeEqualityHelpers(leftType, rightType); + if (leftType.isVector()) { + this->write("all"); + needParens = true; + } + break; + case Operator::Kind::NEQ: + this->writeEqualityHelpers(leftType, rightType); + if (leftType.isVector()) { + this->write("any"); + needParens = true; + } + break; + default: + break; + } + if (leftType.isMatrix() && rightType.isMatrix() && op.kind() == Operator::Kind::STAREQ) { + this->writeMatrixTimesEqualHelper(leftType, rightType, b.type()); + } + if (op.removeAssignment().kind() == Operator::Kind::SLASH && + ((leftType.isMatrix() && rightType.isMatrix()) || + (leftType.isScalar() && rightType.isMatrix()) || + (leftType.isMatrix() && rightType.isScalar()))) { + this->writeMatrixDivisionHelpers(leftType.isMatrix() ? leftType : rightType); + } + + if (needParens) { + this->write("("); + } + + this->writeBinaryExpressionElement(left, op, right, precedence); + + if (op.kind() != Operator::Kind::EQ && op.isAssignment() && + left.kind() == Expression::Kind::kSwizzle && !Analysis::HasSideEffects(left)) { + // This doesn't compile in Metal: + // float4 x = float4(1); + // x.xy *= float2x2(...); + // with the error message "non-const reference cannot bind to vector element", + // but switching it to x.xy = x.xy * float2x2(...) fixes it. We perform this tranformation + // as long as the LHS has no side effects, and hope for the best otherwise. + this->write(" = "); + this->writeExpression(left, Precedence::kAssignment); + this->write(operator_name(op.removeAssignment())); + precedence = op.removeAssignment().getBinaryPrecedence(); + } else { + this->write(operator_name(op)); + } + + this->writeBinaryExpressionElement(right, op, left, precedence); + + if (needParens) { + this->write(")"); + } +} + +void MetalCodeGenerator::writeTernaryExpression(const TernaryExpression& t, + Precedence parentPrecedence) { + if (Precedence::kTernary >= parentPrecedence) { + this->write("("); + } + this->writeExpression(*t.test(), Precedence::kTernary); + this->write(" ? "); + this->writeExpression(*t.ifTrue(), Precedence::kTernary); + this->write(" : "); + this->writeExpression(*t.ifFalse(), Precedence::kTernary); + if (Precedence::kTernary >= parentPrecedence) { + this->write(")"); + } +} + +void MetalCodeGenerator::writePrefixExpression(const PrefixExpression& p, + Precedence parentPrecedence) { + // According to the MSL specification, the arithmetic unary operators (+ and –) do not act + // upon matrix type operands. We treat the unary "+" as NOP for all operands. + const Operator op = p.getOperator(); + if (op.kind() == Operator::Kind::PLUS) { + return this->writeExpression(*p.operand(), Precedence::kPrefix); + } + + const bool matrixNegation = + op.kind() == Operator::Kind::MINUS && p.operand()->type().isMatrix(); + const bool needParens = Precedence::kPrefix >= parentPrecedence || matrixNegation; + + if (needParens) { + this->write("("); + } + + // Transform the unary "-" on a matrix type to a multiplication by -1. + if (matrixNegation) { + this->write("-1.0 * "); + } else { + this->write(p.getOperator().tightOperatorName()); + } + this->writeExpression(*p.operand(), Precedence::kPrefix); + + if (needParens) { + this->write(")"); + } +} + +void MetalCodeGenerator::writePostfixExpression(const PostfixExpression& p, + Precedence parentPrecedence) { + if (Precedence::kPostfix >= parentPrecedence) { + this->write("("); + } + this->writeExpression(*p.operand(), Precedence::kPostfix); + this->write(p.getOperator().tightOperatorName()); + if (Precedence::kPostfix >= parentPrecedence) { + this->write(")"); + } +} + +void MetalCodeGenerator::writeLiteral(const Literal& l) { + const Type& type = l.type(); + if (type.isFloat()) { + this->write(l.description(OperatorPrecedence::kTopLevel)); + if (!l.type().highPrecision()) { + this->write("h"); + } + return; + } + if (type.isInteger()) { + if (type.matches(*fContext.fTypes.fUInt)) { + this->write(std::to_string(l.intValue() & 0xffffffff)); + this->write("u"); + } else if (type.matches(*fContext.fTypes.fUShort)) { + this->write(std::to_string(l.intValue() & 0xffff)); + this->write("u"); + } else { + this->write(std::to_string(l.intValue())); + } + return; + } + SkASSERT(type.isBoolean()); + this->write(l.description(OperatorPrecedence::kTopLevel)); +} + +void MetalCodeGenerator::writeFunctionRequirementArgs(const FunctionDeclaration& f, + const char*& separator) { + Requirements requirements = this->requirements(f); + if (requirements & kInputs_Requirement) { + this->write(separator); + this->write("_in"); + separator = ", "; + } + if (requirements & kOutputs_Requirement) { + this->write(separator); + this->write("_out"); + separator = ", "; + } + if (requirements & kUniforms_Requirement) { + this->write(separator); + this->write("_uniforms"); + separator = ", "; + } + if (requirements & kGlobals_Requirement) { + this->write(separator); + this->write("_globals"); + separator = ", "; + } + if (requirements & kFragCoord_Requirement) { + this->write(separator); + this->write("_fragCoord"); + separator = ", "; + } + if (requirements & kThreadgroups_Requirement) { + this->write(separator); + this->write("_threadgroups"); + separator = ", "; + } +} + +void MetalCodeGenerator::writeFunctionRequirementParams(const FunctionDeclaration& f, + const char*& separator) { + Requirements requirements = this->requirements(f); + if (requirements & kInputs_Requirement) { + this->write(separator); + this->write("Inputs _in"); + separator = ", "; + } + if (requirements & kOutputs_Requirement) { + this->write(separator); + this->write("thread Outputs& _out"); + separator = ", "; + } + if (requirements & kUniforms_Requirement) { + this->write(separator); + this->write("Uniforms _uniforms"); + separator = ", "; + } + if (requirements & kGlobals_Requirement) { + this->write(separator); + this->write("thread Globals& _globals"); + separator = ", "; + } + if (requirements & kFragCoord_Requirement) { + this->write(separator); + this->write("float4 _fragCoord"); + separator = ", "; + } + if (requirements & kThreadgroups_Requirement) { + this->write(separator); + this->write("threadgroup Threadgroups& _threadgroups"); + separator = ", "; + } +} + +int MetalCodeGenerator::getUniformBinding(const Modifiers& m) { + return (m.fLayout.fBinding >= 0) ? m.fLayout.fBinding + : fProgram.fConfig->fSettings.fDefaultUniformBinding; +} + +int MetalCodeGenerator::getUniformSet(const Modifiers& m) { + return (m.fLayout.fSet >= 0) ? m.fLayout.fSet + : fProgram.fConfig->fSettings.fDefaultUniformSet; +} + +bool MetalCodeGenerator::writeFunctionDeclaration(const FunctionDeclaration& f) { + fRTFlipName = fProgram.fInputs.fUseFlipRTUniform + ? "_globals._anonInterface0->" SKSL_RTFLIP_NAME + : ""; + const char* separator = ""; + if (f.isMain()) { + if (ProgramConfig::IsFragment(fProgram.fConfig->fKind)) { + this->write("fragment Outputs fragmentMain"); + } else if (ProgramConfig::IsVertex(fProgram.fConfig->fKind)) { + this->write("vertex Outputs vertexMain"); + } else if (ProgramConfig::IsCompute(fProgram.fConfig->fKind)) { + this->write("kernel void computeMain"); + } else { + fContext.fErrors->error(Position(), "unsupported kind of program"); + return false; + } + this->write("("); + if (!ProgramConfig::IsCompute(fProgram.fConfig->fKind)) { + this->write("Inputs _in [[stage_in]]"); + separator = ", "; + } + if (-1 != fUniformBuffer) { + this->write(separator); + this->write("constant Uniforms& _uniforms [[buffer(" + + std::to_string(fUniformBuffer) + ")]]"); + separator = ", "; + } + for (const ProgramElement* e : fProgram.elements()) { + if (e->is<GlobalVarDeclaration>()) { + const GlobalVarDeclaration& decls = e->as<GlobalVarDeclaration>(); + const VarDeclaration& decl = decls.varDeclaration(); + const Variable* var = decl.var(); + const SkSL::Type::TypeKind varKind = var->type().typeKind(); + + if (varKind == Type::TypeKind::kSampler || varKind == Type::TypeKind::kTexture) { + if (var->type().dimensions() != SpvDim2D) { + // Not yet implemented--Skia currently only uses 2D textures. + fContext.fErrors->error(decls.fPosition, "Unsupported texture dimensions"); + return false; + } + + int binding = getUniformBinding(var->modifiers()); + this->write(separator); + separator = ", "; + + if (varKind == Type::TypeKind::kSampler) { + this->writeType(var->type().textureType()); + this->write(" "); + this->writeName(var->mangledName()); + this->write(kTextureSuffix); + this->write(" [[texture("); + this->write(std::to_string(binding)); + this->write(")]], sampler "); + this->writeName(var->mangledName()); + this->write(kSamplerSuffix); + this->write(" [[sampler("); + this->write(std::to_string(binding)); + this->write(")]]"); + } else { + SkASSERT(varKind == Type::TypeKind::kTexture); + this->writeType(var->type()); + this->write(" "); + this->writeName(var->mangledName()); + this->write(" [[texture("); + this->write(std::to_string(binding)); + this->write(")]]"); + } + } else if (ProgramConfig::IsCompute(fProgram.fConfig->fKind)) { + std::string type, attr; + switch (var->modifiers().fLayout.fBuiltin) { + case SK_NUMWORKGROUPS_BUILTIN: + type = "uint3 "; + attr = " [[threadgroups_per_grid]]"; + break; + case SK_WORKGROUPID_BUILTIN: + type = "uint3 "; + attr = " [[threadgroup_position_in_grid]]"; + break; + case SK_LOCALINVOCATIONID_BUILTIN: + type = "uint3 "; + attr = " [[thread_position_in_threadgroup]]"; + break; + case SK_GLOBALINVOCATIONID_BUILTIN: + type = "uint3 "; + attr = " [[thread_position_in_grid]]"; + break; + case SK_LOCALINVOCATIONINDEX_BUILTIN: + type = "uint "; + attr = " [[thread_index_in_threadgroup]]"; + break; + default: + break; + } + if (!attr.empty()) { + this->write(separator); + this->write(type); + this->write(var->name()); + this->write(attr); + separator = ", "; + } + } + } else if (e->is<InterfaceBlock>()) { + const InterfaceBlock& intf = e->as<InterfaceBlock>(); + if (intf.typeName() == "sk_PerVertex") { + continue; + } + this->write(separator); + if (is_readonly(intf)) { + this->write("const "); + } + this->write(is_buffer(intf) ? "device " : "constant "); + this->writeType(intf.var()->type()); + this->write("& " ); + this->write(fInterfaceBlockNameMap[&intf]); + this->write(" [[buffer("); + this->write(std::to_string(this->getUniformBinding(intf.var()->modifiers()))); + this->write(")]]"); + separator = ", "; + } + } + if (ProgramConfig::IsFragment(fProgram.fConfig->fKind)) { + if (fProgram.fInputs.fUseFlipRTUniform && fInterfaceBlockNameMap.empty()) { + this->write(separator); + this->write("constant sksl_synthetic_uniforms& _anonInterface0 [[buffer(1)]]"); + fRTFlipName = "_anonInterface0." SKSL_RTFLIP_NAME; + separator = ", "; + } + this->write(separator); + this->write("bool _frontFacing [[front_facing]]"); + this->write(", float4 _fragCoord [[position]]"); + separator = ", "; + } else if (ProgramConfig::IsVertex(fProgram.fConfig->fKind)) { + this->write(separator); + this->write("uint sk_VertexID [[vertex_id]], uint sk_InstanceID [[instance_id]]"); + separator = ", "; + } + } else { + this->writeType(f.returnType()); + this->write(" "); + this->writeName(f.mangledName()); + this->write("("); + this->writeFunctionRequirementParams(f, separator); + } + for (const Variable* param : f.parameters()) { + if (f.isMain() && param->modifiers().fLayout.fBuiltin != -1) { + continue; + } + this->write(separator); + separator = ", "; + this->writeModifiers(param->modifiers()); + this->writeType(param->type()); + if (pass_by_reference(param->type(), param->modifiers())) { + this->write("&"); + } + this->write(" "); + this->writeName(param->mangledName()); + } + this->write(")"); + return true; +} + +void MetalCodeGenerator::writeFunctionPrototype(const FunctionPrototype& f) { + this->writeFunctionDeclaration(f.declaration()); + this->writeLine(";"); +} + +static bool is_block_ending_with_return(const Statement* stmt) { + // This function detects (potentially nested) blocks that end in a return statement. + if (!stmt->is<Block>()) { + return false; + } + const StatementArray& block = stmt->as<Block>().children(); + for (int index = block.size(); index--; ) { + stmt = block[index].get(); + if (stmt->is<ReturnStatement>()) { + return true; + } + if (stmt->is<Block>()) { + return is_block_ending_with_return(stmt); + } + if (!stmt->is<Nop>()) { + break; + } + } + return false; +} + +void MetalCodeGenerator::writeComputeMainInputs() { + // Compute shaders only have input variables (e.g. sk_GlobalInvocationID) and access program + // inputs/outputs via the Globals and Uniforms structs. We collect the allowed "in" parameters + // into an Input struct here, since the rest of the code expects the normal _in / _out pattern. + this->write("Inputs _in = { "); + const char* separator = ""; + for (const ProgramElement* e : fProgram.elements()) { + if (e->is<GlobalVarDeclaration>()) { + const GlobalVarDeclaration& decls = e->as<GlobalVarDeclaration>(); + const Variable* var = decls.varDeclaration().var(); + if (is_input(*var)) { + this->write(separator); + separator = ", "; + this->writeName(var->mangledName()); + } + } + } + this->writeLine(" };"); +} + +void MetalCodeGenerator::writeFunction(const FunctionDefinition& f) { + SkASSERT(!fProgram.fConfig->fSettings.fFragColorIsInOut); + + if (!this->writeFunctionDeclaration(f.declaration())) { + return; + } + + fCurrentFunction = &f.declaration(); + SkScopeExit clearCurrentFunction([&] { fCurrentFunction = nullptr; }); + + this->writeLine(" {"); + + if (f.declaration().isMain()) { + fIndentation++; + this->writeGlobalInit(); + if (ProgramConfig::IsCompute(fProgram.fConfig->fKind)) { + this->writeThreadgroupInit(); + this->writeComputeMainInputs(); + } + else { + this->writeLine("Outputs _out;"); + this->writeLine("(void)_out;"); + } + fIndentation--; + } + + fFunctionHeader.clear(); + StringStream buffer; + { + AutoOutputStream outputToBuffer(this, &buffer); + fIndentation++; + for (const std::unique_ptr<Statement>& stmt : f.body()->as<Block>().children()) { + if (!stmt->isEmpty()) { + this->writeStatement(*stmt); + this->finishLine(); + } + } + if (f.declaration().isMain()) { + // If the main function doesn't end with a return, we need to synthesize one here. + if (!is_block_ending_with_return(f.body().get())) { + this->writeReturnStatementFromMain(); + this->finishLine(); + } + } + fIndentation--; + this->writeLine("}"); + } + this->write(fFunctionHeader); + this->write(buffer.str()); +} + +void MetalCodeGenerator::writeModifiers(const Modifiers& modifiers) { + if (ProgramConfig::IsCompute(fProgram.fConfig->fKind) && + (modifiers.fFlags & (Modifiers::kIn_Flag | Modifiers::kOut_Flag))) { + this->write("device "); + } else if (modifiers.fFlags & Modifiers::kOut_Flag) { + this->write("thread "); + } + if (modifiers.fFlags & Modifiers::kConst_Flag) { + this->write("const "); + } +} + +void MetalCodeGenerator::writeInterfaceBlock(const InterfaceBlock& intf) { + if (intf.typeName() == "sk_PerVertex") { + return; + } + const Type* structType = &intf.var()->type().componentType(); + this->writeModifiers(intf.var()->modifiers()); + this->write("struct "); + this->writeType(*structType); + this->writeLine(" {"); + fIndentation++; + this->writeFields(structType->fields(), structType->fPosition, &intf); + if (fProgram.fInputs.fUseFlipRTUniform) { + this->writeLine("float2 " SKSL_RTFLIP_NAME ";"); + } + fIndentation--; + this->write("}"); + if (intf.instanceName().size()) { + this->write(" "); + this->write(intf.instanceName()); + if (intf.arraySize() > 0) { + this->write("["); + this->write(std::to_string(intf.arraySize())); + this->write("]"); + } + fInterfaceBlockNameMap.set(&intf, intf.instanceName()); + } else { + fInterfaceBlockNameMap.set(&intf, *fProgram.fSymbols->takeOwnershipOfString( + "_anonInterface" + std::to_string(fAnonInterfaceCount++))); + } + this->writeLine(";"); +} + +void MetalCodeGenerator::writeFields(const std::vector<Type::Field>& fields, Position parentPos, + const InterfaceBlock* parentIntf) { + MemoryLayout memoryLayout(MemoryLayout::Standard::kMetal); + int currentOffset = 0; + for (const Type::Field& field : fields) { + int fieldOffset = field.fModifiers.fLayout.fOffset; + const Type* fieldType = field.fType; + if (!memoryLayout.isSupported(*fieldType)) { + fContext.fErrors->error(parentPos, "type '" + std::string(fieldType->name()) + + "' is not permitted here"); + return; + } + if (fieldOffset != -1) { + if (currentOffset > fieldOffset) { + fContext.fErrors->error(field.fPosition, + "offset of field '" + std::string(field.fName) + + "' must be at least " + std::to_string(currentOffset)); + return; + } else if (currentOffset < fieldOffset) { + this->write("char pad"); + this->write(std::to_string(fPaddingCount++)); + this->write("["); + this->write(std::to_string(fieldOffset - currentOffset)); + this->writeLine("];"); + currentOffset = fieldOffset; + } + int alignment = memoryLayout.alignment(*fieldType); + if (fieldOffset % alignment) { + fContext.fErrors->error(field.fPosition, + "offset of field '" + std::string(field.fName) + + "' must be a multiple of " + std::to_string(alignment)); + return; + } + } + if (fieldType->isUnsizedArray()) { + // An unsized array always appears as the last member of a storage block. We declare + // it as a one-element array and allow dereferencing past the capacity. + // TODO(armansito): This is because C++ does not support flexible array members like C99 + // does. This generally works but it can lead to UB as compilers are free to insert + // padding past the first element of the array. An alternative approach is to declare + // the struct without the unsized array member and replace variable references with a + // buffer offset calculation based on sizeof(). + this->writeModifiers(field.fModifiers); + this->writeType(fieldType->componentType()); + this->write(" "); + this->writeName(field.fName); + this->write("[1]"); + } else { + size_t fieldSize = memoryLayout.size(*fieldType); + if (fieldSize > static_cast<size_t>(std::numeric_limits<int>::max() - currentOffset)) { + fContext.fErrors->error(parentPos, "field offset overflow"); + return; + } + currentOffset += fieldSize; + this->writeModifiers(field.fModifiers); + this->writeType(*fieldType); + this->write(" "); + this->writeName(field.fName); + } + this->writeLine(";"); + if (parentIntf) { + fInterfaceBlockMap.set(&field, parentIntf); + } + } +} + +void MetalCodeGenerator::writeVarInitializer(const Variable& var, const Expression& value) { + this->writeExpression(value, Precedence::kTopLevel); +} + +void MetalCodeGenerator::writeName(std::string_view name) { + if (fReservedWords.contains(name)) { + this->write("_"); // adding underscore before name to avoid conflict with reserved words + } + this->write(name); +} + +void MetalCodeGenerator::writeVarDeclaration(const VarDeclaration& varDecl) { + this->writeModifiers(varDecl.var()->modifiers()); + this->writeType(varDecl.var()->type()); + this->write(" "); + this->writeName(varDecl.var()->mangledName()); + if (varDecl.value()) { + this->write(" = "); + this->writeVarInitializer(*varDecl.var(), *varDecl.value()); + } + this->write(";"); +} + +void MetalCodeGenerator::writeStatement(const Statement& s) { + switch (s.kind()) { + case Statement::Kind::kBlock: + this->writeBlock(s.as<Block>()); + break; + case Statement::Kind::kExpression: + this->writeExpressionStatement(s.as<ExpressionStatement>()); + break; + case Statement::Kind::kReturn: + this->writeReturnStatement(s.as<ReturnStatement>()); + break; + case Statement::Kind::kVarDeclaration: + this->writeVarDeclaration(s.as<VarDeclaration>()); + break; + case Statement::Kind::kIf: + this->writeIfStatement(s.as<IfStatement>()); + break; + case Statement::Kind::kFor: + this->writeForStatement(s.as<ForStatement>()); + break; + case Statement::Kind::kDo: + this->writeDoStatement(s.as<DoStatement>()); + break; + case Statement::Kind::kSwitch: + this->writeSwitchStatement(s.as<SwitchStatement>()); + break; + case Statement::Kind::kBreak: + this->write("break;"); + break; + case Statement::Kind::kContinue: + this->write("continue;"); + break; + case Statement::Kind::kDiscard: + this->write("discard_fragment();"); + break; + case Statement::Kind::kNop: + this->write(";"); + break; + default: + SkDEBUGFAILF("unsupported statement: %s", s.description().c_str()); + break; + } +} + +void MetalCodeGenerator::writeBlock(const Block& b) { + // Write scope markers if this block is a scope, or if the block is empty (since we need to emit + // something here to make the code valid). + bool isScope = b.isScope() || b.isEmpty(); + if (isScope) { + this->writeLine("{"); + fIndentation++; + } + for (const std::unique_ptr<Statement>& stmt : b.children()) { + if (!stmt->isEmpty()) { + this->writeStatement(*stmt); + this->finishLine(); + } + } + if (isScope) { + fIndentation--; + this->write("}"); + } +} + +void MetalCodeGenerator::writeIfStatement(const IfStatement& stmt) { + this->write("if ("); + this->writeExpression(*stmt.test(), Precedence::kTopLevel); + this->write(") "); + this->writeStatement(*stmt.ifTrue()); + if (stmt.ifFalse()) { + this->write(" else "); + this->writeStatement(*stmt.ifFalse()); + } +} + +void MetalCodeGenerator::writeForStatement(const ForStatement& f) { + // Emit loops of the form 'for(;test;)' as 'while(test)', which is probably how they started + if (!f.initializer() && f.test() && !f.next()) { + this->write("while ("); + this->writeExpression(*f.test(), Precedence::kTopLevel); + this->write(") "); + this->writeStatement(*f.statement()); + return; + } + + this->write("for ("); + if (f.initializer() && !f.initializer()->isEmpty()) { + this->writeStatement(*f.initializer()); + } else { + this->write("; "); + } + if (f.test()) { + this->writeExpression(*f.test(), Precedence::kTopLevel); + } + this->write("; "); + if (f.next()) { + this->writeExpression(*f.next(), Precedence::kTopLevel); + } + this->write(") "); + this->writeStatement(*f.statement()); +} + +void MetalCodeGenerator::writeDoStatement(const DoStatement& d) { + this->write("do "); + this->writeStatement(*d.statement()); + this->write(" while ("); + this->writeExpression(*d.test(), Precedence::kTopLevel); + this->write(");"); +} + +void MetalCodeGenerator::writeExpressionStatement(const ExpressionStatement& s) { + if (fProgram.fConfig->fSettings.fOptimize && !Analysis::HasSideEffects(*s.expression())) { + // Don't emit dead expressions. + return; + } + this->writeExpression(*s.expression(), Precedence::kTopLevel); + this->write(";"); +} + +void MetalCodeGenerator::writeSwitchStatement(const SwitchStatement& s) { + this->write("switch ("); + this->writeExpression(*s.value(), Precedence::kTopLevel); + this->writeLine(") {"); + fIndentation++; + for (const std::unique_ptr<Statement>& stmt : s.cases()) { + const SwitchCase& c = stmt->as<SwitchCase>(); + if (c.isDefault()) { + this->writeLine("default:"); + } else { + this->write("case "); + this->write(std::to_string(c.value())); + this->writeLine(":"); + } + if (!c.statement()->isEmpty()) { + fIndentation++; + this->writeStatement(*c.statement()); + this->finishLine(); + fIndentation--; + } + } + fIndentation--; + this->write("}"); +} + +void MetalCodeGenerator::writeReturnStatementFromMain() { + // main functions in Metal return a magic _out parameter that doesn't exist in SkSL. + if (ProgramConfig::IsVertex(fProgram.fConfig->fKind) || + ProgramConfig::IsFragment(fProgram.fConfig->fKind)) { + this->write("return _out;"); + } else if (ProgramConfig::IsCompute(fProgram.fConfig->fKind)) { + this->write("return;"); + } else { + SkDEBUGFAIL("unsupported kind of program"); + } +} + +void MetalCodeGenerator::writeReturnStatement(const ReturnStatement& r) { + if (fCurrentFunction && fCurrentFunction->isMain()) { + if (r.expression()) { + if (r.expression()->type().matches(*fContext.fTypes.fHalf4)) { + this->write("_out.sk_FragColor = "); + this->writeExpression(*r.expression(), Precedence::kTopLevel); + this->writeLine(";"); + } else { + fContext.fErrors->error(r.fPosition, + "Metal does not support returning '" + + r.expression()->type().description() + "' from main()"); + } + } + this->writeReturnStatementFromMain(); + return; + } + + this->write("return"); + if (r.expression()) { + this->write(" "); + this->writeExpression(*r.expression(), Precedence::kTopLevel); + } + this->write(";"); +} + +void MetalCodeGenerator::writeHeader() { + this->write("#include <metal_stdlib>\n"); + this->write("#include <simd/simd.h>\n"); + this->write("using namespace metal;\n"); +} + +void MetalCodeGenerator::writeSampler2DPolyfill() { + class : public GlobalStructVisitor { + public: + void visitSampler(const Type&, std::string_view) override { + if (fWrotePolyfill) { + return; + } + fWrotePolyfill = true; + + std::string polyfill = SkSL::String::printf(R"( +struct sampler2D { + texture2d<half> tex; + sampler smp; +}; +half4 sample(sampler2D i, float2 p, float b=%g) { return i.tex.sample(i.smp, p, bias(b)); } +half4 sample(sampler2D i, float3 p, float b=%g) { return i.tex.sample(i.smp, p.xy / p.z, bias(b)); } +half4 sampleLod(sampler2D i, float2 p, float lod) { return i.tex.sample(i.smp, p, level(lod)); } +half4 sampleLod(sampler2D i, float3 p, float lod) { + return i.tex.sample(i.smp, p.xy / p.z, level(lod)); +} +half4 sampleGrad(sampler2D i, float2 p, float2 dPdx, float2 dPdy) { + return i.tex.sample(i.smp, p, gradient2d(dPdx, dPdy)); +} + +)", + fTextureBias, + fTextureBias); + fCodeGen->write(polyfill.c_str()); + } + + MetalCodeGenerator* fCodeGen = nullptr; + float fTextureBias = 0.0f; + bool fWrotePolyfill = false; + } visitor; + + visitor.fCodeGen = this; + visitor.fTextureBias = fProgram.fConfig->fSettings.fSharpenTextures ? kSharpenTexturesBias + : 0.0f; + this->visitGlobalStruct(&visitor); +} + +void MetalCodeGenerator::writeUniformStruct() { + for (const ProgramElement* e : fProgram.elements()) { + if (e->is<GlobalVarDeclaration>()) { + const GlobalVarDeclaration& decls = e->as<GlobalVarDeclaration>(); + const Variable& var = *decls.varDeclaration().var(); + if (var.modifiers().fFlags & Modifiers::kUniform_Flag && + var.type().typeKind() != Type::TypeKind::kSampler && + var.type().typeKind() != Type::TypeKind::kTexture) { + int uniformSet = this->getUniformSet(var.modifiers()); + // Make sure that the program's uniform-set value is consistent throughout. + if (-1 == fUniformBuffer) { + this->write("struct Uniforms {\n"); + fUniformBuffer = uniformSet; + } else if (uniformSet != fUniformBuffer) { + fContext.fErrors->error(decls.fPosition, + "Metal backend requires all uniforms to have the same " + "'layout(set=...)'"); + } + this->write(" "); + this->writeType(var.type()); + this->write(" "); + this->writeName(var.mangledName()); + this->write(";\n"); + } + } + } + if (-1 != fUniformBuffer) { + this->write("};\n"); + } +} + +void MetalCodeGenerator::writeInputStruct() { + this->write("struct Inputs {\n"); + for (const ProgramElement* e : fProgram.elements()) { + if (e->is<GlobalVarDeclaration>()) { + const GlobalVarDeclaration& decls = e->as<GlobalVarDeclaration>(); + const Variable& var = *decls.varDeclaration().var(); + if (is_input(var)) { + this->write(" "); + if (ProgramConfig::IsCompute(fProgram.fConfig->fKind) && + needs_address_space(var.type(), var.modifiers())) { + // TODO: address space support + this->write("device "); + } + this->writeType(var.type()); + if (pass_by_reference(var.type(), var.modifiers())) { + this->write("&"); + } + this->write(" "); + this->writeName(var.mangledName()); + if (-1 != var.modifiers().fLayout.fLocation) { + if (ProgramConfig::IsVertex(fProgram.fConfig->fKind)) { + this->write(" [[attribute(" + + std::to_string(var.modifiers().fLayout.fLocation) + ")]]"); + } else if (ProgramConfig::IsFragment(fProgram.fConfig->fKind)) { + this->write(" [[user(locn" + + std::to_string(var.modifiers().fLayout.fLocation) + ")]]"); + } + } + this->write(";\n"); + } + } + } + this->write("};\n"); +} + +void MetalCodeGenerator::writeOutputStruct() { + this->write("struct Outputs {\n"); + if (ProgramConfig::IsVertex(fProgram.fConfig->fKind)) { + this->write(" float4 sk_Position [[position]];\n"); + } else if (ProgramConfig::IsFragment(fProgram.fConfig->fKind)) { + this->write(" half4 sk_FragColor [[color(0)]];\n"); + } + for (const ProgramElement* e : fProgram.elements()) { + if (e->is<GlobalVarDeclaration>()) { + const GlobalVarDeclaration& decls = e->as<GlobalVarDeclaration>(); + const Variable& var = *decls.varDeclaration().var(); + if (is_output(var)) { + this->write(" "); + if (ProgramConfig::IsCompute(fProgram.fConfig->fKind) && + needs_address_space(var.type(), var.modifiers())) { + // TODO: address space support + this->write("device "); + } + this->writeType(var.type()); + if (ProgramConfig::IsCompute(fProgram.fConfig->fKind) && + pass_by_reference(var.type(), var.modifiers())) { + this->write("&"); + } + this->write(" "); + this->writeName(var.mangledName()); + + int location = var.modifiers().fLayout.fLocation; + if (!ProgramConfig::IsCompute(fProgram.fConfig->fKind) && location < 0 && + var.type().typeKind() != Type::TypeKind::kTexture) { + fContext.fErrors->error(var.fPosition, + "Metal out variables must have 'layout(location=...)'"); + } else if (ProgramConfig::IsVertex(fProgram.fConfig->fKind)) { + this->write(" [[user(locn" + std::to_string(location) + ")]]"); + } else if (ProgramConfig::IsFragment(fProgram.fConfig->fKind)) { + this->write(" [[color(" + std::to_string(location) + ")"); + int colorIndex = var.modifiers().fLayout.fIndex; + if (colorIndex) { + this->write(", index(" + std::to_string(colorIndex) + ")"); + } + this->write("]]"); + } + this->write(";\n"); + } + } + } + if (ProgramConfig::IsVertex(fProgram.fConfig->fKind)) { + this->write(" float sk_PointSize [[point_size]];\n"); + } + this->write("};\n"); +} + +void MetalCodeGenerator::writeInterfaceBlocks() { + bool wroteInterfaceBlock = false; + for (const ProgramElement* e : fProgram.elements()) { + if (e->is<InterfaceBlock>()) { + this->writeInterfaceBlock(e->as<InterfaceBlock>()); + wroteInterfaceBlock = true; + } + } + if (!wroteInterfaceBlock && fProgram.fInputs.fUseFlipRTUniform) { + this->writeLine("struct sksl_synthetic_uniforms {"); + this->writeLine(" float2 " SKSL_RTFLIP_NAME ";"); + this->writeLine("};"); + } +} + +void MetalCodeGenerator::writeStructDefinitions() { + for (const ProgramElement* e : fProgram.elements()) { + if (e->is<StructDefinition>()) { + this->writeStructDefinition(e->as<StructDefinition>()); + } + } +} + +void MetalCodeGenerator::writeConstantVariables() { + class : public GlobalStructVisitor { + public: + void visitConstantVariable(const VarDeclaration& decl) override { + fCodeGen->write("constant "); + fCodeGen->writeVarDeclaration(decl); + fCodeGen->finishLine(); + } + + MetalCodeGenerator* fCodeGen = nullptr; + } visitor; + + visitor.fCodeGen = this; + this->visitGlobalStruct(&visitor); +} + +void MetalCodeGenerator::visitGlobalStruct(GlobalStructVisitor* visitor) { + for (const ProgramElement* element : fProgram.elements()) { + if (element->is<InterfaceBlock>()) { + const auto* ib = &element->as<InterfaceBlock>(); + if (ib->typeName() != "sk_PerVertex") { + visitor->visitInterfaceBlock(*ib, fInterfaceBlockNameMap[ib]); + } + continue; + } + if (!element->is<GlobalVarDeclaration>()) { + continue; + } + const GlobalVarDeclaration& global = element->as<GlobalVarDeclaration>(); + const VarDeclaration& decl = global.varDeclaration(); + const Variable& var = *decl.var(); + if (var.type().typeKind() == Type::TypeKind::kSampler) { + visitor->visitSampler(var.type(), var.mangledName()); + continue; + } + if (var.type().typeKind() == Type::TypeKind::kTexture) { + visitor->visitTexture(var.type(), var.modifiers(), var.mangledName()); + continue; + } + if (!(var.modifiers().fFlags & ~Modifiers::kConst_Flag) && + var.modifiers().fLayout.fBuiltin == -1) { + if (is_in_globals(var)) { + // Visit a regular global variable. + visitor->visitNonconstantVariable(var, decl.value().get()); + } else { + // Visit a constant-expression variable. + SkASSERT(var.modifiers().fFlags & Modifiers::kConst_Flag); + visitor->visitConstantVariable(decl); + } + } + } +} + +void MetalCodeGenerator::writeGlobalStruct() { + class : public GlobalStructVisitor { + public: + void visitInterfaceBlock(const InterfaceBlock& block, + std::string_view blockName) override { + this->addElement(); + fCodeGen->write(" "); + if (is_readonly(block)) { + fCodeGen->write("const "); + } + fCodeGen->write(is_buffer(block) ? "device " : "constant "); + fCodeGen->write(block.typeName()); + fCodeGen->write("* "); + fCodeGen->writeName(blockName); + fCodeGen->write(";\n"); + } + void visitTexture(const Type& type, const Modifiers& modifiers, + std::string_view name) override { + this->addElement(); + fCodeGen->write(" "); + fCodeGen->writeType(type); + fCodeGen->write(" "); + fCodeGen->writeName(name); + fCodeGen->write(";\n"); + } + void visitSampler(const Type&, std::string_view name) override { + this->addElement(); + fCodeGen->write(" sampler2D "); + fCodeGen->writeName(name); + fCodeGen->write(";\n"); + } + void visitConstantVariable(const VarDeclaration& decl) override { + // Constants aren't added to the global struct. + } + void visitNonconstantVariable(const Variable& var, const Expression* value) override { + this->addElement(); + fCodeGen->write(" "); + fCodeGen->writeModifiers(var.modifiers()); + fCodeGen->writeType(var.type()); + fCodeGen->write(" "); + fCodeGen->writeName(var.mangledName()); + fCodeGen->write(";\n"); + } + void addElement() { + if (fFirst) { + fCodeGen->write("struct Globals {\n"); + fFirst = false; + } + } + void finish() { + if (!fFirst) { + fCodeGen->writeLine("};"); + fFirst = true; + } + } + + MetalCodeGenerator* fCodeGen = nullptr; + bool fFirst = true; + } visitor; + + visitor.fCodeGen = this; + this->visitGlobalStruct(&visitor); + visitor.finish(); +} + +void MetalCodeGenerator::writeGlobalInit() { + class : public GlobalStructVisitor { + public: + void visitInterfaceBlock(const InterfaceBlock& blockType, + std::string_view blockName) override { + this->addElement(); + fCodeGen->write("&"); + fCodeGen->writeName(blockName); + } + void visitTexture(const Type&, const Modifiers& modifiers, std::string_view name) override { + this->addElement(); + fCodeGen->writeName(name); + } + void visitSampler(const Type&, std::string_view name) override { + this->addElement(); + fCodeGen->write("{"); + fCodeGen->writeName(name); + fCodeGen->write(kTextureSuffix); + fCodeGen->write(", "); + fCodeGen->writeName(name); + fCodeGen->write(kSamplerSuffix); + fCodeGen->write("}"); + } + void visitConstantVariable(const VarDeclaration& decl) override { + // Constant-expression variables aren't put in the global struct. + } + void visitNonconstantVariable(const Variable& var, const Expression* value) override { + this->addElement(); + if (value) { + fCodeGen->writeVarInitializer(var, *value); + } else { + fCodeGen->write("{}"); + } + } + void addElement() { + if (fFirst) { + fCodeGen->write("Globals _globals{"); + fFirst = false; + } else { + fCodeGen->write(", "); + } + } + void finish() { + if (!fFirst) { + fCodeGen->writeLine("};"); + fCodeGen->writeLine("(void)_globals;"); + } + } + MetalCodeGenerator* fCodeGen = nullptr; + bool fFirst = true; + } visitor; + + visitor.fCodeGen = this; + this->visitGlobalStruct(&visitor); + visitor.finish(); +} + +void MetalCodeGenerator::visitThreadgroupStruct(ThreadgroupStructVisitor* visitor) { + for (const ProgramElement* element : fProgram.elements()) { + if (!element->is<GlobalVarDeclaration>()) { + continue; + } + const GlobalVarDeclaration& global = element->as<GlobalVarDeclaration>(); + const VarDeclaration& decl = global.varDeclaration(); + const Variable& var = *decl.var(); + if (var.modifiers().fFlags & Modifiers::kWorkgroup_Flag) { + SkASSERT(!decl.value()); + SkASSERT(!(var.modifiers().fFlags & Modifiers::kConst_Flag)); + visitor->visitNonconstantVariable(var); + } + } +} + +void MetalCodeGenerator::writeThreadgroupStruct() { + class : public ThreadgroupStructVisitor { + public: + void visitNonconstantVariable(const Variable& var) override { + this->addElement(); + fCodeGen->write(" "); + fCodeGen->writeModifiers(var.modifiers()); + fCodeGen->writeType(var.type()); + fCodeGen->write(" "); + fCodeGen->writeName(var.mangledName()); + fCodeGen->write(";\n"); + } + void addElement() { + if (fFirst) { + fCodeGen->write("struct Threadgroups {\n"); + fFirst = false; + } + } + void finish() { + if (!fFirst) { + fCodeGen->writeLine("};"); + fFirst = true; + } + } + + MetalCodeGenerator* fCodeGen = nullptr; + bool fFirst = true; + } visitor; + + visitor.fCodeGen = this; + this->visitThreadgroupStruct(&visitor); + visitor.finish(); +} + +void MetalCodeGenerator::writeThreadgroupInit() { + class : public ThreadgroupStructVisitor { + public: + void visitNonconstantVariable(const Variable& var) override { + this->addElement(); + fCodeGen->write("{}"); + } + void addElement() { + if (fFirst) { + fCodeGen->write("threadgroup Threadgroups _threadgroups{"); + fFirst = false; + } else { + fCodeGen->write(", "); + } + } + void finish() { + if (!fFirst) { + fCodeGen->writeLine("};"); + fCodeGen->writeLine("(void)_threadgroups;"); + } + } + MetalCodeGenerator* fCodeGen = nullptr; + bool fFirst = true; + } visitor; + + visitor.fCodeGen = this; + this->visitThreadgroupStruct(&visitor); + visitor.finish(); +} + +void MetalCodeGenerator::writeProgramElement(const ProgramElement& e) { + switch (e.kind()) { + case ProgramElement::Kind::kExtension: + break; + case ProgramElement::Kind::kGlobalVar: + break; + case ProgramElement::Kind::kInterfaceBlock: + // handled in writeInterfaceBlocks, do nothing + break; + case ProgramElement::Kind::kStructDefinition: + // Handled in writeStructDefinitions. Do nothing. + break; + case ProgramElement::Kind::kFunction: + this->writeFunction(e.as<FunctionDefinition>()); + break; + case ProgramElement::Kind::kFunctionPrototype: + this->writeFunctionPrototype(e.as<FunctionPrototype>()); + break; + case ProgramElement::Kind::kModifiers: + this->writeModifiers(e.as<ModifiersDeclaration>().modifiers()); + this->writeLine(";"); + break; + default: + SkDEBUGFAILF("unsupported program element: %s\n", e.description().c_str()); + break; + } +} + +MetalCodeGenerator::Requirements MetalCodeGenerator::requirements(const Statement* s) { + class RequirementsVisitor : public ProgramVisitor { + public: + using ProgramVisitor::visitStatement; + + bool visitExpression(const Expression& e) override { + switch (e.kind()) { + case Expression::Kind::kFunctionCall: { + const FunctionCall& f = e.as<FunctionCall>(); + fRequirements |= fCodeGen->requirements(f.function()); + break; + } + case Expression::Kind::kFieldAccess: { + const FieldAccess& f = e.as<FieldAccess>(); + if (f.ownerKind() == FieldAccess::OwnerKind::kAnonymousInterfaceBlock) { + fRequirements |= kGlobals_Requirement; + return false; // don't recurse into the base variable + } + break; + } + case Expression::Kind::kVariableReference: { + const Variable& var = *e.as<VariableReference>().variable(); + + if (var.modifiers().fLayout.fBuiltin == SK_FRAGCOORD_BUILTIN) { + fRequirements |= kGlobals_Requirement | kFragCoord_Requirement; + } else if (var.storage() == Variable::Storage::kGlobal) { + if (is_input(var)) { + fRequirements |= kInputs_Requirement; + } else if (is_output(var)) { + fRequirements |= kOutputs_Requirement; + } else if (is_uniforms(var)) { + fRequirements |= kUniforms_Requirement; + } else if (is_threadgroup(var)) { + fRequirements |= kThreadgroups_Requirement; + } else if (is_in_globals(var)) { + fRequirements |= kGlobals_Requirement; + } + } + break; + } + default: + break; + } + return INHERITED::visitExpression(e); + } + + MetalCodeGenerator* fCodeGen; + Requirements fRequirements = kNo_Requirements; + using INHERITED = ProgramVisitor; + }; + + RequirementsVisitor visitor; + if (s) { + visitor.fCodeGen = this; + visitor.visitStatement(*s); + } + return visitor.fRequirements; +} + +MetalCodeGenerator::Requirements MetalCodeGenerator::requirements(const FunctionDeclaration& f) { + Requirements* found = fRequirements.find(&f); + if (!found) { + fRequirements.set(&f, kNo_Requirements); + for (const ProgramElement* e : fProgram.elements()) { + if (e->is<FunctionDefinition>()) { + const FunctionDefinition& def = e->as<FunctionDefinition>(); + if (&def.declaration() == &f) { + Requirements reqs = this->requirements(def.body().get()); + fRequirements.set(&f, reqs); + return reqs; + } + } + } + // We never found a definition for this declared function, but it's legal to prototype a + // function without ever giving a definition, as long as you don't call it. + return kNo_Requirements; + } + return *found; +} + +bool MetalCodeGenerator::generateCode() { + StringStream header; + { + AutoOutputStream outputToHeader(this, &header, &fIndentation); + this->writeHeader(); + this->writeConstantVariables(); + this->writeSampler2DPolyfill(); + this->writeStructDefinitions(); + this->writeUniformStruct(); + this->writeInputStruct(); + if (!ProgramConfig::IsCompute(fProgram.fConfig->fKind)) { + this->writeOutputStruct(); + } + this->writeInterfaceBlocks(); + this->writeGlobalStruct(); + this->writeThreadgroupStruct(); + + // Emit prototypes for every built-in function; these aren't always added in perfect order. + for (const ProgramElement* e : fProgram.fSharedElements) { + if (e->is<FunctionDefinition>()) { + this->writeFunctionDeclaration(e->as<FunctionDefinition>().declaration()); + this->writeLine(";"); + } + } + } + StringStream body; + { + AutoOutputStream outputToBody(this, &body, &fIndentation); + + for (const ProgramElement* e : fProgram.elements()) { + this->writeProgramElement(*e); + } + } + write_stringstream(header, *fOut); + write_stringstream(fExtraFunctionPrototypes, *fOut); + write_stringstream(fExtraFunctions, *fOut); + write_stringstream(body, *fOut); + return fContext.fErrors->errorCount() == 0; +} + +} // namespace SkSL |