From ae46ad48ed358be05bcebe01f38ed9e6850e01cd Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Tue, 9 Apr 2019 17:08:07 -0300 Subject: [PATCH 1/7] shader_ir: Avoid using static on heap-allocated objects Using static here might be faster at runtime, but it adds a heap allocation called before main. --- src/video_core/shader/shader_ir.cpp | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp index ac5112d78..5175f83c6 100644 --- a/src/video_core/shader/shader_ir.cpp +++ b/src/video_core/shader/shader_ir.cpp @@ -219,7 +219,7 @@ Node ShaderIR::GetOperandAbsNegHalf(Node value, bool absolute, bool negate) { } Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, Node op_b) { - static const std::unordered_map PredicateComparisonTable = { + const std::unordered_map PredicateComparisonTable = { {PredCondition::LessThan, OperationCode::LogicalFLessThan}, {PredCondition::Equal, OperationCode::LogicalFEqual}, {PredCondition::LessEqual, OperationCode::LogicalFLessEqual}, @@ -255,7 +255,7 @@ Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, N Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_signed, Node op_a, Node op_b) { - static const std::unordered_map PredicateComparisonTable = { + const std::unordered_map PredicateComparisonTable = { {PredCondition::LessThan, OperationCode::LogicalILessThan}, {PredCondition::Equal, OperationCode::LogicalIEqual}, {PredCondition::LessEqual, OperationCode::LogicalILessEqual}, @@ -285,7 +285,6 @@ Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_si Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, const MetaHalfArithmetic& meta, Node op_a, Node op_b) { - UNIMPLEMENTED_IF_MSG(condition == PredCondition::LessThanWithNan || condition == PredCondition::NotEqualWithNan || condition == PredCondition::LessEqualWithNan || @@ -293,7 +292,7 @@ Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition condition == PredCondition::GreaterEqualWithNan, "Unimplemented NaN comparison for half floats"); - static const std::unordered_map PredicateComparisonTable = { + const std::unordered_map PredicateComparisonTable = { {PredCondition::LessThan, OperationCode::Logical2HLessThan}, {PredCondition::Equal, OperationCode::Logical2HEqual}, {PredCondition::LessEqual, OperationCode::Logical2HLessEqual}, @@ -316,7 +315,7 @@ Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition } OperationCode ShaderIR::GetPredicateCombiner(PredOperation operation) { - static const std::unordered_map PredicateOperationTable = { + const std::unordered_map PredicateOperationTable = { {PredOperation::And, OperationCode::LogicalAnd}, {PredOperation::Or, OperationCode::LogicalOr}, {PredOperation::Xor, OperationCode::LogicalXor}, From acf618afbc834ccfd05a33205c035ecb9737b5db Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Tue, 9 Apr 2019 17:33:48 -0300 Subject: [PATCH 2/7] renderer_opengl: Implement half float NaN comparisons --- .../renderer_opengl/gl_shader_decompiler.cpp | 60 +++++++++++++------ src/video_core/shader/shader_ir.cpp | 17 ++---- src/video_core/shader/shader_ir.h | 18 ++++-- 3 files changed, 59 insertions(+), 36 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 28e490b3c..cbaa4dceb 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -1173,34 +1173,46 @@ private: return GenerateUnary(operation, "any", Type::Bool, Type::Bool2); } + template + std::string GenerateHalfComparison(Operation operation, std::string compare_op) { + std::string comparison{GenerateBinaryCall(operation, compare_op, Type::Bool2, + Type::HalfFloat, Type::HalfFloat)}; + if constexpr (!with_nan) { + return comparison; + } + return "halfFloatNanComparison(" + comparison + ", " + + VisitOperand(operation, 0, Type::HalfFloat) + ", " + + VisitOperand(operation, 1, Type::HalfFloat) + ')'; + } + + template std::string Logical2HLessThan(Operation operation) { - return GenerateBinaryCall(operation, "lessThan", Type::Bool2, Type::HalfFloat, - Type::HalfFloat); + return GenerateHalfComparison(operation, "lessThan"); } + template std::string Logical2HEqual(Operation operation) { - return GenerateBinaryCall(operation, "equal", Type::Bool2, Type::HalfFloat, - Type::HalfFloat); + return GenerateHalfComparison(operation, "equal"); } + template std::string Logical2HLessEqual(Operation operation) { - return GenerateBinaryCall(operation, "lessThanEqual", Type::Bool2, Type::HalfFloat, - Type::HalfFloat); + return GenerateHalfComparison(operation, "lessThanEqual"); } + template std::string Logical2HGreaterThan(Operation operation) { - return GenerateBinaryCall(operation, "greaterThan", Type::Bool2, Type::HalfFloat, - Type::HalfFloat); + return GenerateHalfComparison(operation, "greaterThan"); } + template std::string Logical2HNotEqual(Operation operation) { - return GenerateBinaryCall(operation, "notEqual", Type::Bool2, Type::HalfFloat, - Type::HalfFloat); + return GenerateHalfComparison(operation, "notEqual"); } + template std::string Logical2HGreaterEqual(Operation operation) { - return GenerateBinaryCall(operation, "greaterThanEqual", Type::Bool2, Type::HalfFloat, - Type::HalfFloat); + return GenerateHalfComparison(operation, "greaterThanEqual"); } std::string Texture(Operation operation) { @@ -1525,12 +1537,18 @@ private: &GLSLDecompiler::LogicalNotEqual, &GLSLDecompiler::LogicalGreaterEqual, - &GLSLDecompiler::Logical2HLessThan, - &GLSLDecompiler::Logical2HEqual, - &GLSLDecompiler::Logical2HLessEqual, - &GLSLDecompiler::Logical2HGreaterThan, - &GLSLDecompiler::Logical2HNotEqual, - &GLSLDecompiler::Logical2HGreaterEqual, + &GLSLDecompiler::Logical2HLessThan, + &GLSLDecompiler::Logical2HEqual, + &GLSLDecompiler::Logical2HLessEqual, + &GLSLDecompiler::Logical2HGreaterThan, + &GLSLDecompiler::Logical2HNotEqual, + &GLSLDecompiler::Logical2HGreaterEqual, + &GLSLDecompiler::Logical2HLessThan, + &GLSLDecompiler::Logical2HEqual, + &GLSLDecompiler::Logical2HLessEqual, + &GLSLDecompiler::Logical2HGreaterThan, + &GLSLDecompiler::Logical2HNotEqual, + &GLSLDecompiler::Logical2HGreaterEqual, &GLSLDecompiler::Texture, &GLSLDecompiler::TextureLod, @@ -1633,6 +1651,12 @@ std::string GetCommonDeclarations() { "}\n\n" "vec2 toHalf2(float value) {\n" " return unpackHalf2x16(ftou(value));\n" + "}\n\n" + "bvec2 halfFloatNanComparison(bvec2 comparison, vec2 pair1, vec2 pair2) {\n" + " bvec2 is_nan1 = isnan(pair1);\n" + " bvec2 is_nan2 = isnan(pair2);\n" + " return bvec2(comparison.x || is_nan1.x || is_nan2.x, comparison.y || is_nan1.y || " + "is_nan2.y);\n" "}\n"; } diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp index 5175f83c6..5c1c591f8 100644 --- a/src/video_core/shader/shader_ir.cpp +++ b/src/video_core/shader/shader_ir.cpp @@ -285,13 +285,6 @@ Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_si Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, const MetaHalfArithmetic& meta, Node op_a, Node op_b) { - UNIMPLEMENTED_IF_MSG(condition == PredCondition::LessThanWithNan || - condition == PredCondition::NotEqualWithNan || - condition == PredCondition::LessEqualWithNan || - condition == PredCondition::GreaterThanWithNan || - condition == PredCondition::GreaterEqualWithNan, - "Unimplemented NaN comparison for half floats"); - const std::unordered_map PredicateComparisonTable = { {PredCondition::LessThan, OperationCode::Logical2HLessThan}, {PredCondition::Equal, OperationCode::Logical2HEqual}, @@ -299,11 +292,11 @@ Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition {PredCondition::GreaterThan, OperationCode::Logical2HGreaterThan}, {PredCondition::NotEqual, OperationCode::Logical2HNotEqual}, {PredCondition::GreaterEqual, OperationCode::Logical2HGreaterEqual}, - {PredCondition::LessThanWithNan, OperationCode::Logical2HLessThan}, - {PredCondition::NotEqualWithNan, OperationCode::Logical2HNotEqual}, - {PredCondition::LessEqualWithNan, OperationCode::Logical2HLessEqual}, - {PredCondition::GreaterThanWithNan, OperationCode::Logical2HGreaterThan}, - {PredCondition::GreaterEqualWithNan, OperationCode::Logical2HGreaterEqual}}; + {PredCondition::LessThanWithNan, OperationCode::Logical2HLessThanWithNan}, + {PredCondition::NotEqualWithNan, OperationCode::Logical2HNotEqualWithNan}, + {PredCondition::LessEqualWithNan, OperationCode::Logical2HLessEqualWithNan}, + {PredCondition::GreaterThanWithNan, OperationCode::Logical2HGreaterThanWithNan}, + {PredCondition::GreaterEqualWithNan, OperationCode::Logical2HGreaterEqualWithNan}}; const auto comparison{PredicateComparisonTable.find(condition)}; UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonTable.end(), diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index 4888998d3..0ae51389b 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -150,12 +150,18 @@ enum class OperationCode { LogicalUNotEqual, /// (uint a, uint b) -> bool LogicalUGreaterEqual, /// (uint a, uint b) -> bool - Logical2HLessThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HLessEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HNotEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 - Logical2HGreaterEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HLessThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HLessEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HGreaterThan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HNotEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HGreaterEqual, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HLessThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HLessEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HGreaterThanWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HNotEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 + Logical2HGreaterEqualWithNan, /// (MetaHalfArithmetic, f16vec2 a, f16vec2) -> bool2 Texture, /// (MetaTexture, float[N] coords) -> float4 TextureLod, /// (MetaTexture, float[N] coords) -> float4 From 90cbf89303d5252fdd14dba99447edc9c7c73722 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Tue, 9 Apr 2019 17:34:11 -0300 Subject: [PATCH 3/7] shader_ir/decode: Reduce severity of unimplemented half-float FTZ --- src/video_core/shader/decode/arithmetic_half.cpp | 4 +++- src/video_core/shader/decode/arithmetic_half_immediate.cpp | 4 +++- src/video_core/shader/decode/half_set.cpp | 4 +++- 3 files changed, 9 insertions(+), 3 deletions(-) diff --git a/src/video_core/shader/decode/arithmetic_half.cpp b/src/video_core/shader/decode/arithmetic_half.cpp index baee89107..8cf49314d 100644 --- a/src/video_core/shader/decode/arithmetic_half.cpp +++ b/src/video_core/shader/decode/arithmetic_half.cpp @@ -18,7 +18,9 @@ u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) { if (opcode->get().GetId() == OpCode::Id::HADD2_C || opcode->get().GetId() == OpCode::Id::HADD2_R) { - UNIMPLEMENTED_IF(instr.alu_half.ftz != 0); + if (instr.alu_half.ftz != 0) { + LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName()); + } } UNIMPLEMENTED_IF_MSG(instr.alu_half.saturate != 0, "Half float saturation not implemented"); diff --git a/src/video_core/shader/decode/arithmetic_half_immediate.cpp b/src/video_core/shader/decode/arithmetic_half_immediate.cpp index c2164ba50..4a5b0620e 100644 --- a/src/video_core/shader/decode/arithmetic_half_immediate.cpp +++ b/src/video_core/shader/decode/arithmetic_half_immediate.cpp @@ -17,7 +17,9 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) { const auto opcode = OpCode::Decode(instr); if (opcode->get().GetId() == OpCode::Id::HADD2_IMM) { - UNIMPLEMENTED_IF(instr.alu_half_imm.ftz != 0); + if (instr.alu_half_imm.ftz != 0) { + LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName()); + } } else { UNIMPLEMENTED_IF(instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None); } diff --git a/src/video_core/shader/decode/half_set.cpp b/src/video_core/shader/decode/half_set.cpp index 748368555..d07625741 100644 --- a/src/video_core/shader/decode/half_set.cpp +++ b/src/video_core/shader/decode/half_set.cpp @@ -18,7 +18,9 @@ u32 ShaderIR::DecodeHalfSet(NodeBlock& bb, u32 pc) { const Instruction instr = {program_code[pc]}; const auto opcode = OpCode::Decode(instr); - UNIMPLEMENTED_IF(instr.hset2.ftz != 0); + if (instr.hset2.ftz != 0) { + LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName()); + } // instr.hset2.type_a // instr.hset2.type_b From 64613db6052dfe0c056ce3d6ead21ae9a32f03ae Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Tue, 9 Apr 2019 18:41:41 -0300 Subject: [PATCH 4/7] shader_ir/decode: Implement half float saturation --- .../renderer_opengl/gl_shader_decompiler.cpp | 15 +++++++++++---- .../renderer_vulkan/vk_shader_decompiler.cpp | 6 ++++++ .../shader/decode/arithmetic_half_immediate.cpp | 6 ++---- src/video_core/shader/shader_ir.cpp | 9 +++++++++ src/video_core/shader/shader_ir.h | 3 +++ 5 files changed, 31 insertions(+), 8 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index cbaa4dceb..9c972fd3a 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -617,13 +617,11 @@ private: } std::string VisitOperand(Operation operation, std::size_t operand_index, Type type) { - std::string value = VisitOperand(operation, operand_index); + const std::string value = VisitOperand(operation, operand_index); switch (type) { case Type::HalfFloat: { const auto half_meta = std::get_if(&operation.GetMeta()); - if (!half_meta) { - value = "toHalf2(" + value + ')'; - } + ASSERT(half_meta); switch (half_meta->types.at(operand_index)) { case Tegra::Shader::HalfType::H0_H1: @@ -1067,6 +1065,14 @@ private: return BitwiseCastResult(value, Type::HalfFloat); } + std::string HClamp(Operation operation) { + const std::string value = VisitOperand(operation, 0, Type::HalfFloat); + const std::string min = VisitOperand(operation, 1, Type::Float); + const std::string max = VisitOperand(operation, 2, Type::Float); + const std::string clamped = "clamp(" + value + ", vec2(" + min + "), vec2(" + max + "))"; + return ApplyPrecise(operation, BitwiseCastResult(clamped, Type::HalfFloat)); + } + std::string HMergeF32(Operation operation) { return "float(toHalf2(" + Visit(operation[0]) + ")[0])"; } @@ -1501,6 +1507,7 @@ private: &GLSLDecompiler::Fma, &GLSLDecompiler::Absolute, &GLSLDecompiler::HNegate, + &GLSLDecompiler::HClamp, &GLSLDecompiler::HMergeF32, &GLSLDecompiler::HMergeH0, &GLSLDecompiler::HMergeH1, diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp index e0a6f5e87..6ecb0bcb0 100644 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp @@ -744,6 +744,11 @@ private: return {}; } + Id HClamp(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + Id HMergeF32(Operation operation) { UNIMPLEMENTED(); return {}; @@ -1216,6 +1221,7 @@ private: &SPIRVDecompiler::Ternary<&Module::OpFma, Type::HalfFloat>, &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::HalfFloat>, &SPIRVDecompiler::HNegate, + &SPIRVDecompiler::HClamp, &SPIRVDecompiler::HMergeF32, &SPIRVDecompiler::HMergeH0, &SPIRVDecompiler::HMergeH1, diff --git a/src/video_core/shader/decode/arithmetic_half_immediate.cpp b/src/video_core/shader/decode/arithmetic_half_immediate.cpp index 4a5b0620e..cf4bc0432 100644 --- a/src/video_core/shader/decode/arithmetic_half_immediate.cpp +++ b/src/video_core/shader/decode/arithmetic_half_immediate.cpp @@ -23,8 +23,6 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) { } else { UNIMPLEMENTED_IF(instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None); } - UNIMPLEMENTED_IF_MSG(instr.alu_half_imm.saturate != 0, - "Half float immediate saturation not implemented"); Node op_a = GetRegister(instr.gpr8); op_a = GetOperandAbsNegHalf(op_a, instr.alu_half_imm.abs_a, instr.alu_half_imm.negate_a); @@ -43,10 +41,10 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) { return Immediate(0); } }(); + + value = GetSaturatedHalfFloat(value, instr.alu_half_imm.saturate); value = HalfMerge(GetRegister(instr.gpr0), value, instr.alu_half_imm.merge); - SetRegister(bb, instr.gpr0, value); - return pc; } diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp index 5c1c591f8..d158b4bfd 100644 --- a/src/video_core/shader/shader_ir.cpp +++ b/src/video_core/shader/shader_ir.cpp @@ -218,6 +218,15 @@ Node ShaderIR::GetOperandAbsNegHalf(Node value, bool absolute, bool negate) { return value; } +Node ShaderIR::GetSaturatedHalfFloat(Node value, bool saturate) { + if (!saturate) { + return value; + } + const Node positive_zero = Immediate(std::copysignf(0, 1)); + const Node positive_one = Immediate(1.0f); + return Operation(OperationCode::HClamp, HALF_NO_PRECISE, value, positive_zero, positive_one); +} + Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, Node op_b) { const std::unordered_map PredicateComparisonTable = { {PredCondition::LessThan, OperationCode::LogicalFLessThan}, diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index 0ae51389b..d329da58d 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -114,6 +114,7 @@ enum class OperationCode { HFma, /// (MetaHalfArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2 HAbsolute, /// (f16vec2 a) -> f16vec2 HNegate, /// (f16vec2 a, bool first, bool second) -> f16vec2 + HClamp, /// (f16vec2 src, float min, float max) -> f16vec2 HMergeF32, /// (f16vec2 src) -> float HMergeH0, /// (f16vec2 dest, f16vec2 src) -> f16vec2 HMergeH1, /// (f16vec2 dest, f16vec2 src) -> f16vec2 @@ -716,6 +717,8 @@ private: Node HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge); /// Conditionally absolute/negated half float pair. Absolute is applied first Node GetOperandAbsNegHalf(Node value, bool absolute, bool negate); + /// Conditionally saturates a half float pair + Node GetSaturatedHalfFloat(Node value, bool saturate = true); /// Returns a predicate comparing two floats Node GetPredicateComparisonFloat(Tegra::Shader::PredCondition condition, Node op_a, Node op_b); From abcbcb1b2a437021f87c15c63de862437b920989 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Mon, 15 Apr 2019 19:15:55 -0300 Subject: [PATCH 5/7] gl_shader_decompiler: Fix MrgH0 decompilation GLSL decompilation for HMergeH0 was wrong. This addresses that issue. --- src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 9c972fd3a..01e9433c8 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -1078,8 +1078,8 @@ private: } std::string HMergeH0(Operation operation) { - return "fromHalf2(vec2(toHalf2(" + Visit(operation[0]) + ")[1], toHalf2(" + - Visit(operation[1]) + ")[0]))"; + return "fromHalf2(vec2(toHalf2(" + Visit(operation[1]) + ")[0], toHalf2(" + + Visit(operation[0]) + ")[1]))"; } std::string HMergeH1(Operation operation) { From f43995ec535042d24fa1af637642a2974bfa4773 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Mon, 15 Apr 2019 19:48:11 -0300 Subject: [PATCH 6/7] shader_ir/decode: Fix half float pre-operations and remove MetaHalfArithmetic Operations done before the main half float operation (like HAdd) were managing a packed value instead of the unpacked one. Adding an unpacked operation allows us to drop the per-operand MetaHalfArithmetic entry, simplifying the code overall. --- .../renderer_opengl/gl_shader_decompiler.cpp | 51 +++++++++---------- .../renderer_vulkan/vk_shader_decompiler.cpp | 12 +++-- .../shader/decode/arithmetic_half.cpp | 11 ++-- .../decode/arithmetic_half_immediate.cpp | 7 ++- src/video_core/shader/decode/half_set.cpp | 12 ++--- .../shader/decode/half_set_predicate.cpp | 8 +-- src/video_core/shader/decode/hfma2.cpp | 12 ++--- src/video_core/shader/shader_ir.cpp | 18 ++++--- src/video_core/shader/shader_ir.h | 26 ++++------ 9 files changed, 72 insertions(+), 85 deletions(-) diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 01e9433c8..44ad21b7f 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -121,14 +121,10 @@ std::string GetTopologyName(Tegra::Shader::OutputTopology topology) { /// Returns true if an object has to be treated as precise bool IsPrecise(Operation operand) { - const auto& meta = operand.GetMeta(); - + const auto& meta{operand.GetMeta()}; if (const auto arithmetic = std::get_if(&meta)) { return arithmetic->precise; } - if (const auto half_arithmetic = std::get_if(&meta)) { - return half_arithmetic->precise; - } return false; } @@ -617,26 +613,7 @@ private: } std::string VisitOperand(Operation operation, std::size_t operand_index, Type type) { - const std::string value = VisitOperand(operation, operand_index); - switch (type) { - case Type::HalfFloat: { - const auto half_meta = std::get_if(&operation.GetMeta()); - ASSERT(half_meta); - - switch (half_meta->types.at(operand_index)) { - case Tegra::Shader::HalfType::H0_H1: - return "toHalf2(" + value + ')'; - case Tegra::Shader::HalfType::F32: - return "vec2(" + value + ')'; - case Tegra::Shader::HalfType::H0_H0: - return "vec2(toHalf2(" + value + ")[0])"; - case Tegra::Shader::HalfType::H1_H1: - return "vec2(toHalf2(" + value + ")[1])"; - } - } - default: - return CastOperand(value, type); - } + return CastOperand(VisitOperand(operation, operand_index), type); } std::string CastOperand(const std::string& value, Type type) const { @@ -650,9 +627,7 @@ private: case Type::Uint: return "ftou(" + value + ')'; case Type::HalfFloat: - // Can't be handled as a stand-alone value - UNREACHABLE(); - return value; + return "toHalf2(" + value + ')'; } UNREACHABLE(); return value; @@ -1073,6 +1048,25 @@ private: return ApplyPrecise(operation, BitwiseCastResult(clamped, Type::HalfFloat)); } + std::string HUnpack(Operation operation) { + const std::string operand{VisitOperand(operation, 0, Type::HalfFloat)}; + const auto value = [&]() -> std::string { + switch (std::get(operation.GetMeta())) { + case Tegra::Shader::HalfType::H0_H1: + return operand; + case Tegra::Shader::HalfType::F32: + return "vec2(fromHalf2(" + operand + "))"; + case Tegra::Shader::HalfType::H0_H0: + return "vec2(" + operand + "[0])"; + case Tegra::Shader::HalfType::H1_H1: + return "vec2(" + operand + "[1])"; + } + UNREACHABLE(); + return "0"; + }(); + return "fromHalf2(" + value + ')'; + } + std::string HMergeF32(Operation operation) { return "float(toHalf2(" + Visit(operation[0]) + ")[0])"; } @@ -1508,6 +1502,7 @@ private: &GLSLDecompiler::Absolute, &GLSLDecompiler::HNegate, &GLSLDecompiler::HClamp, + &GLSLDecompiler::HUnpack, &GLSLDecompiler::HMergeF32, &GLSLDecompiler::HMergeH0, &GLSLDecompiler::HMergeH1, diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp index 6ecb0bcb0..beff33ba1 100644 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp @@ -76,14 +76,10 @@ constexpr u32 GetGenericAttributeLocation(Attribute::Index attribute) { /// Returns true if an object has to be treated as precise bool IsPrecise(Operation operand) { - const auto& meta = operand.GetMeta(); - + const auto& meta{operand.GetMeta()}; if (std::holds_alternative(meta)) { return std::get(meta).precise; } - if (std::holds_alternative(meta)) { - return std::get(meta).precise; - } return false; } @@ -749,6 +745,11 @@ private: return {}; } + Id HUnpack(Operation operation) { + UNIMPLEMENTED(); + return {}; + } + Id HMergeF32(Operation operation) { UNIMPLEMENTED(); return {}; @@ -1222,6 +1223,7 @@ private: &SPIRVDecompiler::Unary<&Module::OpFAbs, Type::HalfFloat>, &SPIRVDecompiler::HNegate, &SPIRVDecompiler::HClamp, + &SPIRVDecompiler::HUnpack, &SPIRVDecompiler::HMergeF32, &SPIRVDecompiler::HMergeH0, &SPIRVDecompiler::HMergeH1, diff --git a/src/video_core/shader/decode/arithmetic_half.cpp b/src/video_core/shader/decode/arithmetic_half.cpp index 8cf49314d..9467f9417 100644 --- a/src/video_core/shader/decode/arithmetic_half.cpp +++ b/src/video_core/shader/decode/arithmetic_half.cpp @@ -29,9 +29,8 @@ u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) { const bool negate_b = opcode->get().GetId() != OpCode::Id::HMUL2_C && instr.alu_half.negate_b != 0; - const Node op_a = GetOperandAbsNegHalf(GetRegister(instr.gpr8), instr.alu_half.abs_a, negate_a); - - // instr.alu_half.type_a + Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half.type_a); + op_a = GetOperandAbsNegHalf(op_a, instr.alu_half.abs_a, negate_a); Node op_b = [&]() { switch (opcode->get().GetId()) { @@ -46,17 +45,17 @@ u32 ShaderIR::DecodeArithmeticHalf(NodeBlock& bb, u32 pc) { return Immediate(0); } }(); + op_b = UnpackHalfFloat(op_b, instr.alu_half.type_b); op_b = GetOperandAbsNegHalf(op_b, instr.alu_half.abs_b, negate_b); Node value = [&]() { - MetaHalfArithmetic meta{true, {instr.alu_half_imm.type_a, instr.alu_half.type_b}}; switch (opcode->get().GetId()) { case OpCode::Id::HADD2_C: case OpCode::Id::HADD2_R: - return Operation(OperationCode::HAdd, meta, op_a, op_b); + return Operation(OperationCode::HAdd, PRECISE, op_a, op_b); case OpCode::Id::HMUL2_C: case OpCode::Id::HMUL2_R: - return Operation(OperationCode::HMul, meta, op_a, op_b); + return Operation(OperationCode::HMul, PRECISE, op_a, op_b); default: UNIMPLEMENTED_MSG("Unhandled half float instruction: {}", opcode->get().GetName()); return Immediate(0); diff --git a/src/video_core/shader/decode/arithmetic_half_immediate.cpp b/src/video_core/shader/decode/arithmetic_half_immediate.cpp index cf4bc0432..fbcd35b18 100644 --- a/src/video_core/shader/decode/arithmetic_half_immediate.cpp +++ b/src/video_core/shader/decode/arithmetic_half_immediate.cpp @@ -24,18 +24,17 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) { UNIMPLEMENTED_IF(instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None); } - Node op_a = GetRegister(instr.gpr8); + Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half_imm.type_a); op_a = GetOperandAbsNegHalf(op_a, instr.alu_half_imm.abs_a, instr.alu_half_imm.negate_a); const Node op_b = UnpackHalfImmediate(instr, true); Node value = [&]() { - MetaHalfArithmetic meta{true, {instr.alu_half_imm.type_a}}; switch (opcode->get().GetId()) { case OpCode::Id::HADD2_IMM: - return Operation(OperationCode::HAdd, meta, op_a, op_b); + return Operation(OperationCode::HAdd, PRECISE, op_a, op_b); case OpCode::Id::HMUL2_IMM: - return Operation(OperationCode::HMul, meta, op_a, op_b); + return Operation(OperationCode::HMul, PRECISE, op_a, op_b); default: UNREACHABLE(); return Immediate(0); diff --git a/src/video_core/shader/decode/half_set.cpp b/src/video_core/shader/decode/half_set.cpp index d07625741..1dd94bf9d 100644 --- a/src/video_core/shader/decode/half_set.cpp +++ b/src/video_core/shader/decode/half_set.cpp @@ -22,9 +22,9 @@ u32 ShaderIR::DecodeHalfSet(NodeBlock& bb, u32 pc) { LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName()); } - // instr.hset2.type_a - // instr.hset2.type_b - Node op_a = GetRegister(instr.gpr8); + Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hset2.type_a); + op_a = GetOperandAbsNegHalf(op_a, instr.hset2.abs_a, instr.hset2.negate_a); + Node op_b = [&]() { switch (opcode->get().GetId()) { case OpCode::Id::HSET2_R: @@ -34,14 +34,12 @@ u32 ShaderIR::DecodeHalfSet(NodeBlock& bb, u32 pc) { return Immediate(0); } }(); - - op_a = GetOperandAbsNegHalf(op_a, instr.hset2.abs_a, instr.hset2.negate_a); + op_b = UnpackHalfFloat(op_b, instr.hset2.type_b); op_b = GetOperandAbsNegHalf(op_b, instr.hset2.abs_b, instr.hset2.negate_b); const Node second_pred = GetPredicate(instr.hset2.pred39, instr.hset2.neg_pred); - MetaHalfArithmetic meta{false, {instr.hset2.type_a, instr.hset2.type_b}}; - const Node comparison_pair = GetPredicateComparisonHalf(instr.hset2.cond, meta, op_a, op_b); + const Node comparison_pair = GetPredicateComparisonHalf(instr.hset2.cond, op_a, op_b); const OperationCode combiner = GetPredicateCombiner(instr.hset2.op); diff --git a/src/video_core/shader/decode/half_set_predicate.cpp b/src/video_core/shader/decode/half_set_predicate.cpp index e68512692..6e59eb650 100644 --- a/src/video_core/shader/decode/half_set_predicate.cpp +++ b/src/video_core/shader/decode/half_set_predicate.cpp @@ -19,10 +19,10 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) { UNIMPLEMENTED_IF(instr.hsetp2.ftz != 0); - Node op_a = GetRegister(instr.gpr8); + Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hsetp2.type_a); op_a = GetOperandAbsNegHalf(op_a, instr.hsetp2.abs_a, instr.hsetp2.negate_a); - const Node op_b = [&]() { + Node op_b = [&]() { switch (opcode->get().GetId()) { case OpCode::Id::HSETP2_R: return GetOperandAbsNegHalf(GetRegister(instr.gpr20), instr.hsetp2.abs_a, @@ -32,6 +32,7 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) { return Immediate(0); } }(); + op_b = UnpackHalfFloat(op_b, instr.hsetp2.type_b); // We can't use the constant predicate as destination. ASSERT(instr.hsetp2.pred3 != static_cast(Pred::UnusedIndex)); @@ -42,8 +43,7 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) { const OperationCode pair_combiner = instr.hsetp2.h_and ? OperationCode::LogicalAll2 : OperationCode::LogicalAny2; - MetaHalfArithmetic meta = {false, {instr.hsetp2.type_a, instr.hsetp2.type_b}}; - const Node comparison = GetPredicateComparisonHalf(instr.hsetp2.cond, meta, op_a, op_b); + const Node comparison = GetPredicateComparisonHalf(instr.hsetp2.cond, op_a, op_b); const Node first_pred = Operation(pair_combiner, comparison); // Set the primary predicate to the result of Predicate OP SecondPredicate diff --git a/src/video_core/shader/decode/hfma2.cpp b/src/video_core/shader/decode/hfma2.cpp index 7a07c5ec6..5c1becce5 100644 --- a/src/video_core/shader/decode/hfma2.cpp +++ b/src/video_core/shader/decode/hfma2.cpp @@ -27,10 +27,6 @@ u32 ShaderIR::DecodeHfma2(NodeBlock& bb, u32 pc) { } constexpr auto identity = HalfType::H0_H1; - - const HalfType type_a = instr.hfma2.type_a; - const Node op_a = GetRegister(instr.gpr8); - bool neg_b{}, neg_c{}; auto [saturate, type_b, op_b, type_c, op_c] = [&]() -> std::tuple { @@ -62,11 +58,11 @@ u32 ShaderIR::DecodeHfma2(NodeBlock& bb, u32 pc) { }(); UNIMPLEMENTED_IF_MSG(saturate, "HFMA2 saturation is not implemented"); - op_b = GetOperandAbsNegHalf(op_b, false, neg_b); - op_c = GetOperandAbsNegHalf(op_c, false, neg_c); + const Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hfma2.type_a); + op_b = GetOperandAbsNegHalf(UnpackHalfFloat(op_b, type_b), false, neg_b); + op_c = GetOperandAbsNegHalf(UnpackHalfFloat(op_c, type_c), false, neg_c); - MetaHalfArithmetic meta{true, {type_a, type_b, type_c}}; - Node value = Operation(OperationCode::HFma, meta, op_a, op_b, op_c); + Node value = Operation(OperationCode::HFma, PRECISE, op_a, op_b, op_c); value = HalfMerge(GetRegister(instr.gpr0), value, instr.hfma2.merge); SetRegister(bb, instr.gpr0, value); diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp index d158b4bfd..17f2f711c 100644 --- a/src/video_core/shader/shader_ir.cpp +++ b/src/video_core/shader/shader_ir.cpp @@ -189,7 +189,11 @@ Node ShaderIR::UnpackHalfImmediate(Instruction instr, bool has_negation) { const Node first_negate = GetPredicate(instr.half_imm.first_negate != 0); const Node second_negate = GetPredicate(instr.half_imm.second_negate != 0); - return Operation(OperationCode::HNegate, HALF_NO_PRECISE, value, first_negate, second_negate); + return Operation(OperationCode::HNegate, NO_PRECISE, value, first_negate, second_negate); +} + +Node ShaderIR::UnpackHalfFloat(Node value, Tegra::Shader::HalfType type) { + return Operation(OperationCode::HUnpack, type, value); } Node ShaderIR::HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge) { @@ -209,10 +213,10 @@ Node ShaderIR::HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge) { Node ShaderIR::GetOperandAbsNegHalf(Node value, bool absolute, bool negate) { if (absolute) { - value = Operation(OperationCode::HAbsolute, HALF_NO_PRECISE, value); + value = Operation(OperationCode::HAbsolute, NO_PRECISE, value); } if (negate) { - value = Operation(OperationCode::HNegate, HALF_NO_PRECISE, value, GetPredicate(true), + value = Operation(OperationCode::HNegate, NO_PRECISE, value, GetPredicate(true), GetPredicate(true)); } return value; @@ -224,7 +228,7 @@ Node ShaderIR::GetSaturatedHalfFloat(Node value, bool saturate) { } const Node positive_zero = Immediate(std::copysignf(0, 1)); const Node positive_one = Immediate(1.0f); - return Operation(OperationCode::HClamp, HALF_NO_PRECISE, value, positive_zero, positive_one); + return Operation(OperationCode::HClamp, NO_PRECISE, value, positive_zero, positive_one); } Node ShaderIR::GetPredicateComparisonFloat(PredCondition condition, Node op_a, Node op_b) { @@ -292,8 +296,8 @@ Node ShaderIR::GetPredicateComparisonInteger(PredCondition condition, bool is_si return predicate; } -Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, - const MetaHalfArithmetic& meta, Node op_a, Node op_b) { +Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, Node op_a, + Node op_b) { const std::unordered_map PredicateComparisonTable = { {PredCondition::LessThan, OperationCode::Logical2HLessThan}, {PredCondition::Equal, OperationCode::Logical2HEqual}, @@ -311,7 +315,7 @@ Node ShaderIR::GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition UNIMPLEMENTED_IF_MSG(comparison == PredicateComparisonTable.end(), "Unknown predicate comparison operation"); - const Node predicate = Operation(comparison->second, meta, op_a, op_b); + const Node predicate = Operation(comparison->second, NO_PRECISE, op_a, op_b); return predicate; } diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h index d329da58d..88957e970 100644 --- a/src/video_core/shader/shader_ir.h +++ b/src/video_core/shader/shader_ir.h @@ -109,12 +109,13 @@ enum class OperationCode { UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint UBitCount, /// (MetaArithmetic, uint) -> uint - HAdd, /// (MetaHalfArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 - HMul, /// (MetaHalfArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 - HFma, /// (MetaHalfArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2 + HAdd, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 + HMul, /// (MetaArithmetic, f16vec2 a, f16vec2 b) -> f16vec2 + HFma, /// (MetaArithmetic, f16vec2 a, f16vec2 b, f16vec2 c) -> f16vec2 HAbsolute, /// (f16vec2 a) -> f16vec2 HNegate, /// (f16vec2 a, bool first, bool second) -> f16vec2 HClamp, /// (f16vec2 src, float min, float max) -> f16vec2 + HUnpack, /// (Tegra::Shader::HalfType, T value) -> f16vec2 HMergeF32, /// (f16vec2 src) -> float HMergeH0, /// (f16vec2 dest, f16vec2 src) -> f16vec2 HMergeH1, /// (f16vec2 dest, f16vec2 src) -> f16vec2 @@ -287,13 +288,6 @@ struct MetaArithmetic { bool precise{}; }; -struct MetaHalfArithmetic { - bool precise{}; - std::array types = {Tegra::Shader::HalfType::H0_H1, - Tegra::Shader::HalfType::H0_H1, - Tegra::Shader::HalfType::H0_H1}; -}; - struct MetaTexture { const Sampler& sampler; Node array{}; @@ -305,11 +299,10 @@ struct MetaTexture { u32 element{}; }; -constexpr MetaArithmetic PRECISE = {true}; -constexpr MetaArithmetic NO_PRECISE = {false}; -constexpr MetaHalfArithmetic HALF_NO_PRECISE = {false}; +inline constexpr MetaArithmetic PRECISE = {true}; +inline constexpr MetaArithmetic NO_PRECISE = {false}; -using Meta = std::variant; +using Meta = std::variant; /// Holds any kind of operation that can be done in the IR class OperationNode final { @@ -713,6 +706,8 @@ private: /// Unpacks a half immediate from an instruction Node UnpackHalfImmediate(Tegra::Shader::Instruction instr, bool has_negation); + /// Unpacks a binary value into a half float pair with a type format + Node UnpackHalfFloat(Node value, Tegra::Shader::HalfType type); /// Merges a half pair into another value Node HalfMerge(Node dest, Node src, Tegra::Shader::HalfMerge merge); /// Conditionally absolute/negated half float pair. Absolute is applied first @@ -726,8 +721,7 @@ private: Node GetPredicateComparisonInteger(Tegra::Shader::PredCondition condition, bool is_signed, Node op_a, Node op_b); /// Returns a predicate comparing two half floats. meta consumes how both pairs will be compared - Node GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, - const MetaHalfArithmetic& meta, Node op_a, Node op_b); + Node GetPredicateComparisonHalf(Tegra::Shader::PredCondition condition, Node op_a, Node op_b); /// Returns a predicate combiner operation OperationCode GetPredicateCombiner(Tegra::Shader::PredOperation operation); From ef8245bed21dbe9b496f7d06b0e1eb487e553dbf Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Mon, 15 Apr 2019 21:32:24 -0300 Subject: [PATCH 7/7] vk_shader_decompiler: Add missing operations --- src/video_core/renderer_vulkan/vk_shader_decompiler.cpp | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp index beff33ba1..f50aae00d 100644 --- a/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp +++ b/src/video_core/renderer_vulkan/vk_shader_decompiler.cpp @@ -1260,6 +1260,13 @@ private: &SPIRVDecompiler::Binary<&Module::OpINotEqual, Type::Bool, Type::Uint>, &SPIRVDecompiler::Binary<&Module::OpUGreaterThanEqual, Type::Bool, Type::Uint>, + &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThan, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdNotEqual, Type::Bool, Type::HalfFloat>, + &SPIRVDecompiler::Binary<&Module::OpFOrdGreaterThanEqual, Type::Bool, Type::HalfFloat>, + // TODO(Rodrigo): Should these use the OpFUnord* variants? &SPIRVDecompiler::Binary<&Module::OpFOrdLessThan, Type::Bool, Type::HalfFloat>, &SPIRVDecompiler::Binary<&Module::OpFOrdEqual, Type::Bool, Type::HalfFloat>, &SPIRVDecompiler::Binary<&Module::OpFOrdLessThanEqual, Type::Bool, Type::HalfFloat>,