diff options
Diffstat (limited to 'src/shader_recompiler/frontend')
| -rw-r--r-- | src/shader_recompiler/frontend/ir/ir_emitter.cpp | 133 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/ir/ir_emitter.h | 21 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/ir/microinstruction.cpp | 73 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/ir/microinstruction.h | 22 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/ir/modifiers.h | 10 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/ir/opcodes.cpp | 2 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/ir/opcodes.inc | 569 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/ir/reg.h | 11 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/ir/value.h | 1 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/maxwell/maxwell.inc | 4 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/maxwell/program.cpp | 1 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp | 8 | ||||
| -rw-r--r-- | src/shader_recompiler/frontend/maxwell/translate/impl/texture_sample.cpp | 232 |
13 files changed, 772 insertions, 315 deletions
diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.cpp b/src/shader_recompiler/frontend/ir/ir_emitter.cpp index f38b46bac..ae3354c66 100644 --- a/src/shader_recompiler/frontend/ir/ir_emitter.cpp +++ b/src/shader_recompiler/frontend/ir/ir_emitter.cpp @@ -7,11 +7,24 @@ #include "shader_recompiler/frontend/ir/value.h" namespace Shader::IR { - -[[noreturn]] static void ThrowInvalidType(Type type) { +namespace { +[[noreturn]] void ThrowInvalidType(Type type) { throw InvalidArgument("Invalid type {}", type); } +Value MakeLodClampPair(IREmitter& ir, const F32& bias_lod, const F32& lod_clamp) { + if (!bias_lod.IsEmpty() && !lod_clamp.IsEmpty()) { + return ir.CompositeConstruct(bias_lod, lod_clamp); + } else if (!bias_lod.IsEmpty()) { + return bias_lod; + } else if (!lod_clamp.IsEmpty()) { + return lod_clamp; + } else { + return Value{}; + } +} +} // Anonymous namespace + U1 IREmitter::Imm1(bool value) const { return U1{Value{value}}; } @@ -261,6 +274,10 @@ U1 IREmitter::GetOverflowFromOp(const Value& op) { return Inst<U1>(Opcode::GetOverflowFromOp, op); } +U1 IREmitter::GetSparseFromOp(const Value& op) { + return Inst<U1>(Opcode::GetSparseFromOp, op); +} + F16F32F64 IREmitter::FPAdd(const F16F32F64& a, const F16F32F64& b, FpControl control) { if (a.Type() != a.Type()) { throw InvalidArgument("Mismatching types {} and {}", a.Type(), b.Type()); @@ -1035,6 +1052,82 @@ U32U64 IREmitter::ConvertFToI(size_t bitsize, bool is_signed, const F16F32F64& v } } +F16F32F64 IREmitter::ConvertSToF(size_t bitsize, const U32U64& value) { + switch (bitsize) { + case 16: + switch (value.Type()) { + case Type::U32: + return Inst<F16>(Opcode::ConvertF16S32, value); + case Type::U64: + return Inst<F16>(Opcode::ConvertF16S64, value); + default: + ThrowInvalidType(value.Type()); + } + case 32: + switch (value.Type()) { + case Type::U32: + return Inst<F32>(Opcode::ConvertF32S32, value); + case Type::U64: + return Inst<F32>(Opcode::ConvertF32S64, value); + default: + ThrowInvalidType(value.Type()); + } + case 64: + switch (value.Type()) { + case Type::U32: + return Inst<F16>(Opcode::ConvertF64S32, value); + case Type::U64: + return Inst<F16>(Opcode::ConvertF64S64, value); + default: + ThrowInvalidType(value.Type()); + } + default: + throw InvalidArgument("Invalid destination bitsize {}", bitsize); + } +} + +F16F32F64 IREmitter::ConvertUToF(size_t bitsize, const U32U64& value) { + switch (bitsize) { + case 16: + switch (value.Type()) { + case Type::U32: + return Inst<F16>(Opcode::ConvertF16U32, value); + case Type::U64: + return Inst<F16>(Opcode::ConvertF16U64, value); + default: + ThrowInvalidType(value.Type()); + } + case 32: + switch (value.Type()) { + case Type::U32: + return Inst<F32>(Opcode::ConvertF32U32, value); + case Type::U64: + return Inst<F32>(Opcode::ConvertF32U64, value); + default: + ThrowInvalidType(value.Type()); + } + case 64: + switch (value.Type()) { + case Type::U32: + return Inst<F16>(Opcode::ConvertF64U32, value); + case Type::U64: + return Inst<F16>(Opcode::ConvertF64U64, value); + default: + ThrowInvalidType(value.Type()); + } + default: + throw InvalidArgument("Invalid destination bitsize {}", bitsize); + } +} + +F16F32F64 IREmitter::ConvertIToF(size_t bitsize, bool is_signed, const U32U64& value) { + if (is_signed) { + return ConvertSToF(bitsize, value); + } else { + return ConvertUToF(bitsize, value); + } +} + U32U64 IREmitter::UConvert(size_t result_bitsize, const U32U64& value) { switch (result_bitsize) { case 32: @@ -1107,4 +1200,40 @@ F16F32F64 IREmitter::FPConvert(size_t result_bitsize, const F16F32F64& value) { throw NotImplementedException("Conversion from {} to {} bits", value.Type(), result_bitsize); } +Value IREmitter::ImageSampleImplicitLod(const Value& handle, const Value& coords, const F32& bias, + const Value& offset, const F32& lod_clamp, + TextureInstInfo info) { + const Value bias_lc{MakeLodClampPair(*this, bias, lod_clamp)}; + const Opcode op{handle.IsImmediate() ? Opcode::BoundImageSampleImplicitLod + : Opcode::BindlessImageSampleImplicitLod}; + return Inst(op, Flags{info}, handle, coords, bias_lc, offset); +} + +Value IREmitter::ImageSampleExplicitLod(const Value& handle, const Value& coords, const F32& lod, + const Value& offset, const F32& lod_clamp, + TextureInstInfo info) { + const Value lod_lc{MakeLodClampPair(*this, lod, lod_clamp)}; + const Opcode op{handle.IsImmediate() ? Opcode::BoundImageSampleExplicitLod + : Opcode::BindlessImageSampleExplicitLod}; + return Inst(op, Flags{info}, handle, coords, lod_lc, offset); +} + +F32 IREmitter::ImageSampleDrefImplicitLod(const Value& handle, const Value& coords, const F32& dref, + const F32& bias, const Value& offset, + const F32& lod_clamp, TextureInstInfo info) { + const Value bias_lc{MakeLodClampPair(*this, bias, lod_clamp)}; + const Opcode op{handle.IsImmediate() ? Opcode::BoundImageSampleDrefImplicitLod + : Opcode::BindlessImageSampleDrefImplicitLod}; + return Inst<F32>(op, Flags{info}, handle, coords, dref, bias_lc, offset); +} + +F32 IREmitter::ImageSampleDrefExplicitLod(const Value& handle, const Value& coords, const F32& dref, + const F32& lod, const Value& offset, const F32& lod_clamp, + TextureInstInfo info) { + const Value lod_lc{MakeLodClampPair(*this, lod, lod_clamp)}; + const Opcode op{handle.IsImmediate() ? Opcode::BoundImageSampleDrefExplicitLod + : Opcode::BindlessImageSampleDrefExplicitLod}; + return Inst<F32>(op, Flags{info}, handle, coords, dref, lod_lc, offset); +} + } // namespace Shader::IR diff --git a/src/shader_recompiler/frontend/ir/ir_emitter.h b/src/shader_recompiler/frontend/ir/ir_emitter.h index 6e29bf0e2..cb2a7710a 100644 --- a/src/shader_recompiler/frontend/ir/ir_emitter.h +++ b/src/shader_recompiler/frontend/ir/ir_emitter.h @@ -91,6 +91,7 @@ public: [[nodiscard]] U1 GetSignFromOp(const Value& op); [[nodiscard]] U1 GetCarryFromOp(const Value& op); [[nodiscard]] U1 GetOverflowFromOp(const Value& op); + [[nodiscard]] U1 GetSparseFromOp(const Value& op); [[nodiscard]] Value CompositeConstruct(const Value& e1, const Value& e2); [[nodiscard]] Value CompositeConstruct(const Value& e1, const Value& e2, const Value& e3); @@ -159,7 +160,7 @@ public: [[nodiscard]] U32 BitFieldInsert(const U32& base, const U32& insert, const U32& offset, const U32& count); [[nodiscard]] U32 BitFieldExtract(const U32& base, const U32& offset, const U32& count, - bool is_signed); + bool is_signed = false); [[nodiscard]] U32 BitReverse(const U32& value); [[nodiscard]] U32 BitCount(const U32& value); [[nodiscard]] U32 BitwiseNot(const U32& value); @@ -186,10 +187,28 @@ public: [[nodiscard]] U32U64 ConvertFToS(size_t bitsize, const F16F32F64& value); [[nodiscard]] U32U64 ConvertFToU(size_t bitsize, const F16F32F64& value); [[nodiscard]] U32U64 ConvertFToI(size_t bitsize, bool is_signed, const F16F32F64& value); + [[nodiscard]] F16F32F64 ConvertSToF(size_t bitsize, const U32U64& value); + [[nodiscard]] F16F32F64 ConvertUToF(size_t bitsize, const U32U64& value); + [[nodiscard]] F16F32F64 ConvertIToF(size_t bitsize, bool is_signed, const U32U64& value); [[nodiscard]] U32U64 UConvert(size_t result_bitsize, const U32U64& value); [[nodiscard]] F16F32F64 FPConvert(size_t result_bitsize, const F16F32F64& value); + [[nodiscard]] Value ImageSampleImplicitLod(const Value& handle, const Value& coords, + const F32& bias, const Value& offset, + const F32& lod_clamp, TextureInstInfo info); + [[nodiscard]] Value ImageSampleExplicitLod(const Value& handle, const Value& coords, + const F32& lod, const Value& offset, + const F32& lod_clamp, TextureInstInfo info); + [[nodiscard]] F32 ImageSampleDrefImplicitLod(const Value& handle, const Value& coords, + const F32& dref, const F32& bias, + const Value& offset, const F32& lod_clamp, + TextureInstInfo info); + [[nodiscard]] F32 ImageSampleDrefExplicitLod(const Value& handle, const Value& coords, + const F32& dref, const F32& lod, + const Value& offset, const F32& lod_clamp, + TextureInstInfo info); + private: IR::Block::iterator insertion_point; diff --git a/src/shader_recompiler/frontend/ir/microinstruction.cpp b/src/shader_recompiler/frontend/ir/microinstruction.cpp index d6a9be87d..88e186f21 100644 --- a/src/shader_recompiler/frontend/ir/microinstruction.cpp +++ b/src/shader_recompiler/frontend/ir/microinstruction.cpp @@ -10,26 +10,27 @@ #include "shader_recompiler/frontend/ir/type.h" namespace Shader::IR { - -static void CheckPseudoInstruction(IR::Inst* inst, IR::Opcode opcode) { +namespace { +void CheckPseudoInstruction(IR::Inst* inst, IR::Opcode opcode) { if (inst && inst->Opcode() != opcode) { throw LogicError("Invalid pseudo-instruction"); } } -static void SetPseudoInstruction(IR::Inst*& dest_inst, IR::Inst* pseudo_inst) { +void SetPseudoInstruction(IR::Inst*& dest_inst, IR::Inst* pseudo_inst) { if (dest_inst) { throw LogicError("Only one of each type of pseudo-op allowed"); } dest_inst = pseudo_inst; } -static void RemovePseudoInstruction(IR::Inst*& inst, IR::Opcode expected_opcode) { +void RemovePseudoInstruction(IR::Inst*& inst, IR::Opcode expected_opcode) { if (inst->Opcode() != expected_opcode) { throw LogicError("Undoing use of invalid pseudo-op"); } inst = nullptr; } +} // Anonymous namespace Inst::Inst(IR::Opcode op_, u32 flags_) noexcept : op{op_}, flags{flags_} { if (op == Opcode::Phi) { @@ -82,6 +83,7 @@ bool Inst::IsPseudoInstruction() const noexcept { case Opcode::GetSignFromOp: case Opcode::GetCarryFromOp: case Opcode::GetOverflowFromOp: + case Opcode::GetSparseFromOp: return true; default: return false; @@ -96,25 +98,26 @@ bool Inst::AreAllArgsImmediates() const { [](const IR::Value& value) { return value.IsImmediate(); }); } -bool Inst::HasAssociatedPseudoOperation() const noexcept { - return zero_inst || sign_inst || carry_inst || overflow_inst; -} - Inst* Inst::GetAssociatedPseudoOperation(IR::Opcode opcode) { - // This is faster than doing a search through the block. + if (!associated_insts) { + return nullptr; + } switch (opcode) { case Opcode::GetZeroFromOp: - CheckPseudoInstruction(zero_inst, Opcode::GetZeroFromOp); - return zero_inst; + CheckPseudoInstruction(associated_insts->zero_inst, Opcode::GetZeroFromOp); + return associated_insts->zero_inst; case Opcode::GetSignFromOp: - CheckPseudoInstruction(sign_inst, Opcode::GetSignFromOp); - return sign_inst; + CheckPseudoInstruction(associated_insts->sign_inst, Opcode::GetSignFromOp); + return associated_insts->sign_inst; case Opcode::GetCarryFromOp: - CheckPseudoInstruction(carry_inst, Opcode::GetCarryFromOp); - return carry_inst; + CheckPseudoInstruction(associated_insts->carry_inst, Opcode::GetCarryFromOp); + return associated_insts->carry_inst; case Opcode::GetOverflowFromOp: - CheckPseudoInstruction(overflow_inst, Opcode::GetOverflowFromOp); - return overflow_inst; + CheckPseudoInstruction(associated_insts->overflow_inst, Opcode::GetOverflowFromOp); + return associated_insts->overflow_inst; + case Opcode::GetSparseFromOp: + CheckPseudoInstruction(associated_insts->sparse_inst, Opcode::GetSparseFromOp); + return associated_insts->sparse_inst; default: throw InvalidArgument("{} is not a pseudo-instruction", opcode); } @@ -220,22 +223,37 @@ void Inst::ReplaceOpcode(IR::Opcode opcode) { op = opcode; } +void AllocAssociatedInsts(std::unique_ptr<AssociatedInsts>& associated_insts) { + if (!associated_insts) { + associated_insts = std::make_unique<AssociatedInsts>(); + } +} + void Inst::Use(const Value& value) { Inst* const inst{value.Inst()}; ++inst->use_count; + std::unique_ptr<AssociatedInsts>& assoc_inst{inst->associated_insts}; switch (op) { case Opcode::GetZeroFromOp: - SetPseudoInstruction(inst->zero_inst, this); + AllocAssociatedInsts(assoc_inst); + SetPseudoInstruction(assoc_inst->zero_inst, this); break; case Opcode::GetSignFromOp: - SetPseudoInstruction(inst->sign_inst, this); + AllocAssociatedInsts(assoc_inst); + SetPseudoInstruction(assoc_inst->sign_inst, this); break; case Opcode::GetCarryFromOp: - SetPseudoInstruction(inst->carry_inst, this); + AllocAssociatedInsts(assoc_inst); + SetPseudoInstruction(assoc_inst->carry_inst, this); break; case Opcode::GetOverflowFromOp: - SetPseudoInstruction(inst->overflow_inst, this); + AllocAssociatedInsts(assoc_inst); + SetPseudoInstruction(assoc_inst->overflow_inst, this); + break; + case Opcode::GetSparseFromOp: + AllocAssociatedInsts(assoc_inst); + SetPseudoInstruction(assoc_inst->sparse_inst, this); break; default: break; @@ -246,18 +264,23 @@ void Inst::UndoUse(const Value& value) { Inst* const inst{value.Inst()}; --inst->use_count; + std::unique_ptr<AssociatedInsts>& assoc_inst{inst->associated_insts}; switch (op) { case Opcode::GetZeroFromOp: - RemovePseudoInstruction(inst->zero_inst, Opcode::GetZeroFromOp); + AllocAssociatedInsts(assoc_inst); + RemovePseudoInstruction(assoc_inst->zero_inst, Opcode::GetZeroFromOp); break; case Opcode::GetSignFromOp: - RemovePseudoInstruction(inst->sign_inst, Opcode::GetSignFromOp); + AllocAssociatedInsts(assoc_inst); + RemovePseudoInstruction(assoc_inst->sign_inst, Opcode::GetSignFromOp); break; case Opcode::GetCarryFromOp: - RemovePseudoInstruction(inst->carry_inst, Opcode::GetCarryFromOp); + AllocAssociatedInsts(assoc_inst); + RemovePseudoInstruction(assoc_inst->carry_inst, Opcode::GetCarryFromOp); break; case Opcode::GetOverflowFromOp: - RemovePseudoInstruction(inst->overflow_inst, Opcode::GetOverflowFromOp); + AllocAssociatedInsts(assoc_inst); + RemovePseudoInstruction(assoc_inst->overflow_inst, Opcode::GetOverflowFromOp); break; default: break; diff --git a/src/shader_recompiler/frontend/ir/microinstruction.h b/src/shader_recompiler/frontend/ir/microinstruction.h index 321393dd7..d5336c438 100644 --- a/src/shader_recompiler/frontend/ir/microinstruction.h +++ b/src/shader_recompiler/frontend/ir/microinstruction.h @@ -22,7 +22,7 @@ namespace Shader::IR { class Block; -constexpr size_t MAX_ARG_COUNT = 4; +struct AssociatedInsts; class Inst : public boost::intrusive::list_base_hook<> { public: @@ -50,6 +50,11 @@ public: return op; } + /// Determines if there is a pseudo-operation associated with this instruction. + [[nodiscard]] bool HasAssociatedPseudoOperation() const noexcept { + return associated_insts != nullptr; + } + /// Determines whether or not this instruction may have side effects. [[nodiscard]] bool MayHaveSideEffects() const noexcept; @@ -60,8 +65,6 @@ public: /// Determines if all arguments of this instruction are immediates. [[nodiscard]] bool AreAllArgsImmediates() const; - /// Determines if there is a pseudo-operation associated with this instruction. - [[nodiscard]] bool HasAssociatedPseudoOperation() const noexcept; /// Gets a pseudo-operation associated with this instruction [[nodiscard]] Inst* GetAssociatedPseudoOperation(IR::Opcode opcode); @@ -122,14 +125,21 @@ private: u32 definition{}; union { NonTriviallyDummy dummy{}; - std::array<Value, MAX_ARG_COUNT> args; std::vector<std::pair<Block*, Value>> phi_args; + std::array<Value, 5> args; + }; + std::unique_ptr<AssociatedInsts> associated_insts; +}; +static_assert(sizeof(Inst) <= 128, "Inst size unintentionally increased"); + +struct AssociatedInsts { + union { + Inst* sparse_inst; + Inst* zero_inst{}; }; - Inst* zero_inst{}; Inst* sign_inst{}; Inst* carry_inst{}; Inst* overflow_inst{}; }; -static_assert(sizeof(Inst) <= 128, "Inst size unintentionally increased its size"); } // namespace Shader::IR diff --git a/src/shader_recompiler/frontend/ir/modifiers.h b/src/shader_recompiler/frontend/ir/modifiers.h index 44652eae7..ad07700ae 100644 --- a/src/shader_recompiler/frontend/ir/modifiers.h +++ b/src/shader_recompiler/frontend/ir/modifiers.h @@ -4,7 +4,9 @@ #pragma once +#include "common/bit_field.h" #include "common/common_types.h" +#include "shader_recompiler/shader_info.h" namespace Shader::IR { @@ -30,4 +32,12 @@ struct FpControl { }; static_assert(sizeof(FpControl) <= sizeof(u32)); +union TextureInstInfo { + u32 raw; + BitField<0, 8, TextureType> type; + BitField<8, 1, u32> has_bias; + BitField<16, 1, u32> has_lod_clamp; +}; +static_assert(sizeof(TextureInstInfo) <= sizeof(u32)); + } // namespace Shader::IR diff --git a/src/shader_recompiler/frontend/ir/opcodes.cpp b/src/shader_recompiler/frontend/ir/opcodes.cpp index 1f188411a..8492a13d5 100644 --- a/src/shader_recompiler/frontend/ir/opcodes.cpp +++ b/src/shader_recompiler/frontend/ir/opcodes.cpp @@ -14,7 +14,7 @@ namespace { struct OpcodeMeta { std::string_view name; Type type; - std::array<Type, 4> arg_types; + std::array<Type, 5> arg_types; }; using enum Type; diff --git a/src/shader_recompiler/frontend/ir/opcodes.inc b/src/shader_recompiler/frontend/ir/opcodes.inc index c4e72c84d..aa011fab1 100644 --- a/src/shader_recompiler/frontend/ir/opcodes.inc +++ b/src/shader_recompiler/frontend/ir/opcodes.inc @@ -2,301 +2,330 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -// opcode name, return type, arg1 type, arg2 type, arg3 type, arg4 type, ... -OPCODE(Phi, Opaque, ) -OPCODE(Identity, Opaque, Opaque, ) -OPCODE(Void, Void, ) +// opcode name, return type, arg1 type, arg2 type, arg3 type, arg4 type, arg4 type, ... +OPCODE(Phi, Opaque, ) +OPCODE(Identity, Opaque, Opaque, ) +OPCODE(Void, Void, ) // Control flow -OPCODE(Branch, Void, Label, ) -OPCODE(BranchConditional, Void, U1, Label, Label, ) -OPCODE(LoopMerge, Void, Label, Label, ) -OPCODE(SelectionMerge, Void, Label, ) -OPCODE(Return, Void, ) +OPCODE(Branch, Void, Label, ) +OPCODE(BranchConditional, Void, U1, Label, Label, ) +OPCODE(LoopMerge, Void, Label, Label, ) +OPCODE(SelectionMerge, Void, Label, ) +OPCODE(Return, Void, ) // Context getters/setters -OPCODE(GetRegister, U32, Reg, ) -OPCODE(SetRegister, Void, Reg, U32, ) -OPCODE(GetPred, U1, Pred, ) -OPCODE(SetPred, Void, Pred, U1, ) -OPCODE(GetGotoVariable, U1, U32, ) -OPCODE(SetGotoVariable, Void, U32, U1, ) -OPCODE(GetCbuf, U32, U32, U32, ) -OPCODE(GetAttribute, U32, Attribute, ) -OPCODE(SetAttribute, Void, Attribute, U32, ) -OPCODE(GetAttributeIndexed, U32, U32, ) -OPCODE(SetAttributeIndexed, Void, U32, U32, ) -OPCODE(GetZFlag, U1, Void, ) -OPCODE(GetSFlag, U1, Void, ) -OPCODE(GetCFlag, U1, Void, ) -OPCODE(GetOFlag, U1, Void, ) -OPCODE(SetZFlag, Void, U1, ) -OPCODE(SetSFlag, Void, U1, ) -OPCODE(SetCFlag, Void, U1, ) -OPCODE(SetOFlag, Void, U1, ) -OPCODE(WorkgroupId, U32x3, ) -OPCODE(LocalInvocationId, U32x3, ) +OPCODE(GetRegister, U32, Reg, ) +OPCODE(SetRegister, Void, Reg, U32, ) +OPCODE(GetPred, U1, Pred, ) +OPCODE(SetPred, Void, Pred, U1, ) +OPCODE(GetGotoVariable, U1, U32, ) +OPCODE(SetGotoVariable, Void, U32, U1, ) +OPCODE(GetCbuf, U32, U32, U32, ) +OPCODE(GetAttribute, U32, Attribute, ) +OPCODE(SetAttribute, Void, Attribute, U32, ) +OPCODE(GetAttributeIndexed, U32, U32, ) +OPCODE(SetAttributeIndexed, Void, U32, U32, ) +OPCODE(GetZFlag, U1, Void, ) +OPCODE(GetSFlag, U1, Void, ) +OPCODE(GetCFlag, U1, Void, ) +OPCODE(GetOFlag, U1, Void, ) +OPCODE(SetZFlag, Void, U1, ) +OPCODE(SetSFlag, Void, U1, ) +OPCODE(SetCFlag, Void, U1, ) +OPCODE(SetOFlag, Void, U1, ) +OPCODE(WorkgroupId, U32x3, ) +OPCODE(LocalInvocationId, U32x3, ) // Undefined -OPCODE(UndefU1, U1, ) -OPCODE(UndefU8, U8, ) -OPCODE(UndefU16, U16, ) -OPCODE(UndefU32, U32, ) -OPCODE(UndefU64, U64, ) +OPCODE(UndefU1, U1, ) +OPCODE(UndefU8, U8, ) +OPCODE(UndefU16, U16, ) +OPCODE(UndefU32, U32, ) +OPCODE(UndefU64, U64, ) // Memory operations -OPCODE(LoadGlobalU8, U32, U64, ) -OPCODE(LoadGlobalS8, U32, U64, ) -OPCODE(LoadGlobalU16, U32, U64, ) -OPCODE(LoadGlobalS16, U32, U64, ) -OPCODE(LoadGlobal32, U32, U64, ) -OPCODE(LoadGlobal64, U32x2, U64, ) -OPCODE(LoadGlobal128, U32x4, U64, ) -OPCODE(WriteGlobalU8, Void, U64, U32, ) -OPCODE(WriteGlobalS8, Void, U64, U32, ) -OPCODE(WriteGlobalU16, Void, U64, U32, ) -OPCODE(WriteGlobalS16, Void, U64, U32, ) -OPCODE(WriteGlobal32, Void, U64, U32, ) -OPCODE(WriteGlobal64, Void, U64, U32x2, ) -OPCODE(WriteGlobal128, Void, U64, U32x4, ) +OPCODE(LoadGlobalU8, U32, U64, ) +OPCODE(LoadGlobalS8, U32, U64, ) +OPCODE(LoadGlobalU16, U32, U64, ) +OPCODE(LoadGlobalS16, U32, U64, ) +OPCODE(LoadGlobal32, U32, U64, ) +OPCODE(LoadGlobal64, U32x2, U64, ) +OPCODE(LoadGlobal128, U32x4, U64, ) +OPCODE(WriteGlobalU8, Void, U64, U32, ) +OPCODE(WriteGlobalS8, Void, U64, U32, ) +OPCODE(WriteGlobalU16, Void, U64, U32, ) +OPCODE(WriteGlobalS16, Void, U64, U32, ) +OPCODE(WriteGlobal32, Void, U64, U32, ) +OPCODE(WriteGlobal64, Void, U64, U32x2, ) +OPCODE(WriteGlobal128, Void, U64, U32x4, ) // Storage buffer operations -OPCODE(LoadStorageU8, U32, U32, U32, ) -OPCODE(LoadStorageS8, U32, U32, U32, ) -OPCODE(LoadStorageU16, U32, U32, U32, ) -OPCODE(LoadStorageS16, U32, U32, U32, ) -OPCODE(LoadStorage32, U32, U32, U32, ) -OPCODE(LoadStorage64, U32x2, U32, U32, ) -OPCODE(LoadStorage128, U32x4, U32, U32, ) -OPCODE(WriteStorageU8, Void, U32, U32, U32, ) -OPCODE(WriteStorageS8, Void, U32, U32, U32, ) -OPCODE(WriteStorageU16, Void, U32, U32, U32, ) -OPCODE(WriteStorageS16, Void, U32, U32, U32, ) -OPCODE(WriteStorage32, Void, U32, U32, U32, ) -OPCODE(WriteStorage64, Void, U32, U32, U32x2, ) -OPCODE(WriteStorage128, Void, U32, U32, U32x4, ) +OPCODE(LoadStorageU8, U32, U32, U32, ) +OPCODE(LoadStorageS8, U32, U32, U32, ) +OPCODE(LoadStorageU16, U32, U32, U32, ) +OPCODE(LoadStorageS16, U32, U32, U32, ) +OPCODE(LoadStorage32, U32, U32, U32, ) +OPCODE(LoadStorage64, U32x2, U32, U32, ) +OPCODE(LoadStorage128, U32x4, U32, U32, ) +OPCODE(WriteStorageU8, Void, U32, U32, U32, ) +OPCODE(WriteStorageS8, Void, U32, U32, U32, ) +OPCODE(WriteStorageU16, Void, U32, U32, U32, ) +OPCODE(WriteStorageS16, Void, U32, U32, U32, ) +OPCODE(WriteStorage32, Void, U32, U32, U32, ) +OPCODE(WriteStorage64, Void, U32, U32, U32x2, ) +OPCODE(WriteStorage128, Void, U32, U32, U32x4, ) // Vector utility -OPCODE(CompositeConstructU32x2, U32x2, U32, U32, ) -OPCODE(CompositeConstructU32x3, U32x3, U32, U32, U32, ) -OPCODE(CompositeConstructU32x4, U32x4, U32, U32, U32, U32, ) -OPCODE(CompositeExtractU32x2, U32, U32x2, U32, ) -OPCODE(CompositeExtractU32x3, U32, U32x3, U32, ) -OPCODE(CompositeExtractU32x4, U32, U32x4, U32, ) -OPCODE(CompositeInsertU32x2, U32x2, U32x2, U32, U32, ) -OPCODE(CompositeInsertU32x3, U32x3, U32x3, U32, U32, ) -OPCODE(CompositeInsertU32x4, U32x4, U32x4, U32, U32, ) -OPCODE(CompositeConstructF16x2, F16x2, F16, F16, ) -OPCODE(CompositeConstructF16x3, F16x3, F16, F16, F16, ) -OPCODE(CompositeConstructF16x4, F16x4, F16, F16, F16, F16, ) -OPCODE(CompositeExtractF16x2, F16, F16x2, U32, ) -OPCODE(CompositeExtractF16x3, F16, F16x3, U32, ) -OPCODE(CompositeExtractF16x4, F16, F16x4, U32, ) -OPCODE(CompositeInsertF16x2, F16x2, F16x2, F16, U32, ) -OPCODE(CompositeInsertF16x3, F16x3, F16x3, F16, U32, ) -OPCODE(CompositeInsertF16x4, F16x4, F16x4, F16, U32, ) -OPCODE(CompositeConstructF32x2, F32x2, F32, F32, ) -OPCODE(CompositeConstructF32x3, F32x3, F32, F32, F32, ) -OPCODE(CompositeConstructF32x4, F32x4, F32, F32, F32, F32, ) -OPCODE(CompositeExtractF32x2, F32, F32x2, U32, ) -OPCODE(CompositeExtractF32x3, F32, F32x3, U32, ) -OPCODE(CompositeExtractF32x4, F32, F32x4, U32, ) -OPCODE(CompositeInsertF32x2, F32x2, F32x2, F32, U32, ) -OPCODE(CompositeInsertF32x3, F32x3, F32x3, F32, U32, ) -OPCODE(CompositeInsertF32x4, F32x4, F32x4, F32, U32, ) -OPCODE(CompositeConstructF64x2, F64x2, F64, F64, ) -OPCODE(CompositeConstructF64x3, F64x3, F64, F64, F64, ) -OPCODE(CompositeConstructF64x4, F64x4, F64, F64, F64, F64, ) -OPCODE(CompositeExtractF64x2, F64, F64x2, U32, ) -OPCODE(CompositeExtractF64x3, F64, F64x3, U32, ) -OPCODE(CompositeExtractF64x4, F64, F64x4, U32, ) -OPCODE(CompositeInsertF64x2, F64x2, F64x2, F64, U32, ) -OPCODE(CompositeInsertF64x3, F64x3, F64x3, F64, U32, ) -OPCODE(CompositeInsertF64x4, F64x4, F64x4, F64, U32, ) +OPCODE(CompositeConstructU32x2, U32x2, U32, U32, ) +OPCODE(CompositeConstructU32x3, U32x3, U32, U32, U32, ) +OPCODE(CompositeConstructU32x4, U32x4, U32, U32, U32, U32, ) +OPCODE(CompositeExtractU32x2, U32, U32x2, U32, ) +OPCODE(CompositeExtractU32x3, U32, U32x3, U32, ) +OPCODE(CompositeExtractU32x4, U32, U32x4, U32, ) +OPCODE(CompositeInsertU32x2, U32x2, U32x2, U32, U32, ) +OPCODE(CompositeInsertU32x3, U32x3, U32x3, U32, U32, ) +OPCODE(CompositeInsertU32x4, U32x4, U32x4, U32, U32, ) +OPCODE(CompositeConstructF16x2, F16x2, F16, F16, ) +OPCODE(CompositeConstructF16x3, F16x3, F16, F16, F16, ) +OPCODE(CompositeConstructF16x4, F16x4, F16, F16, F16, F16, ) +OPCODE(CompositeExtractF16x2, F16, F16x2, U32, ) +OPCODE(CompositeExtractF16x3, F16, F16x3, U32, ) +OPCODE(CompositeExtractF16x4, F16, F16x4, U32, ) +OPCODE(CompositeInsertF16x2, F16x2, F16x2, F16, U32, ) +OPCODE(CompositeInsertF16x3, F16x3, F16x3, F16, U32, ) +OPCODE(CompositeInsertF16x4, F16x4, F16x4, F16, U32, ) +OPCODE(CompositeConstructF32x2, F32x2, F32, F32, ) +OPCODE(CompositeConstructF32x3, F32x3, F32, F32, F32, ) +OPCODE(CompositeConstructF32x4, F32x4, F32, F32, F32, F32, ) +OPCODE(CompositeExtractF32x2, F32, F32x2, U32, ) +OPCODE(CompositeExtractF32x3, F32, F32x3, U32, ) +OPCODE(CompositeExtractF32x4, F32, F32x4, U32, ) +OPCODE(CompositeInsertF32x2, F32x2, F32x2, F32, U32, ) +OPCODE(CompositeInsertF32x3, F32x3, F32x3, F32, U32, ) +OPCODE(CompositeInsertF32x4, F32x4, F32x4, F32, U32, ) +OPCODE(CompositeConstructF64x2, F64x2, F64, F64, ) +OPCODE(CompositeConstructF64x3, F64x3, F64, F64, F64, ) +OPCODE(CompositeConstructF64x4, F64x4, F64, F64, F64, F64, ) +OPCODE(CompositeExtractF64x2, F64, F64x2, U32, ) +OPCODE(CompositeExtractF64x3, F64, F64x3, U32, ) +OPCODE(CompositeExtractF64x4, F64, F64x4, U32, ) +OPCODE(CompositeInsertF64x2, F64x2, F64x2, F64, U32, ) +OPCODE(CompositeInsertF64x3, F64x3, F64x3, F64, U32, ) +OPCODE(CompositeInsertF64x4, F64x4, F64x4, F64, U32, ) // Select operations -OPCODE(SelectU1, U1, U1, U1, U1, ) -OPCODE(SelectU8, U8, U1, U8, U8, ) -OPCODE(SelectU16, U16, U1, U16, U16, ) -OPCODE(SelectU32, U32, U1, U32, U32, ) -OPCODE(SelectU64, U64, U1, U64, U64, ) -OPCODE(SelectF16, F16, U1, F16, F16, ) -OPCODE(SelectF32, F32, U1, F32, F32, ) +OPCODE(SelectU1, U1, U1, U1, U1, ) +OPCODE(SelectU8, U8, U1, U8, U8, ) +OPCODE(SelectU16, U16, U1, U16, U16, ) +OPCODE(SelectU32, U32, U1, U32, U32, ) +OPCODE(SelectU64, U64, U1, U64, U64, ) +OPCODE(SelectF16, F16, U1, F16, F16, ) +OPCODE(SelectF32, F32, U1, F32, F32, ) // Bitwise conversions -OPCODE(BitCastU16F16, U16, F16, ) -OPCODE(BitCastU32F32, U32, F32, ) -OPCODE(BitCastU64F64, U64, F64, ) -OPCODE(BitCastF16U16, F16, U16, ) -OPCODE(BitCastF32U32, F32, U32, ) -OPCODE(BitCastF64U64, F64, U64, ) -OPCODE(PackUint2x32, U64, U32x2, ) -OPCODE(UnpackUint2x32, U32x2, U64, ) -OPCODE(PackFloat2x16, U32, F16x2, ) -OPCODE(UnpackFloat2x16, F16x2, U32, ) -OPCODE(PackHalf2x16, U32, F32x2, ) -OPCODE(UnpackHalf2x16, F32x2, U32, ) -OPCODE(PackDouble2x32, F64, U32x2, ) -OPCODE(UnpackDouble2x32, U32x2, F64, ) +OPCODE(BitCastU16F16, U16, F16, ) +OPCODE(BitCastU32F32, U32, F32, ) +OPCODE(BitCastU64F64, U64, F64, ) +OPCODE(BitCastF16U16, F16, U16, ) +OPCODE(BitCastF32U32, F32, U32, ) +OPCODE(BitCastF64U64, F64, U64, ) +OPCODE(PackUint2x32, U64, U32x2, ) +OPCODE(UnpackUint2x32, U32x2, U64, ) +OPCODE(PackFloat2x16, U32, F16x2, ) +OPCODE(UnpackFloat2x16, F16x2, U32, ) +OPCODE(PackHalf2x16, U32, F32x2, ) +OPCODE(UnpackHalf2x16, F32x2, U32, ) +OPCODE(PackDouble2x32, F64, U32x2, ) +OPCODE(UnpackDouble2x32, U32x2, F64, ) // Pseudo-operation, handled specially at final emit -OPCODE(GetZeroFromOp, U1, Opaque, ) -OPCODE(GetSignFromOp, U1, Opaque, ) -OPCODE(GetCarryFromOp, U1, Opaque, ) -OPCODE(GetOverflowFromOp, U1, Opaque, ) +OPCODE(GetZeroFromOp, U1, Opaque, ) +OPCODE(GetSignFromOp, U1, Opaque, ) +OPCODE(GetCarryFromOp, U1, Opaque, ) +OPCODE(GetOverflowFromOp, U1, Opaque, ) +OPCODE(GetSparseFromOp, U1, Opaque, ) // Floating-point operations -OPCODE(FPAbs16, F16, F16, ) -OPCODE(FPAbs32, F32, F32, ) -OPCODE(FPAbs64, F64, F64, ) -OPCODE(FPAdd16, F16, F16, F16, ) -OPCODE(FPAdd32, F32, F32, F32, ) -OPCODE(FPAdd64, F64, F64, F64, ) -OPCODE(FPFma16, F16, F16, F16, F16, ) -OPCODE(FPFma32, F32, F32, F32, F32, ) -OPCODE(FPFma64, F64, F64, F64, F64, ) -OPCODE(FPMax32, F32, F32, F32, ) -OPCODE(FPMax64, F64, F64, F64, ) -OPCODE(FPMin32, F32, F32, F32, ) -OPCODE(FPMin64, F64, F64, F64, ) -OPCODE(FPMul16, F16, F16, F16, ) -OPCODE(FPMul32, F32, F32, F32, ) -OPCODE(FPMul64, F64, F64, F64, ) -OPCODE(FPNeg16, F16, F16, ) -OPCODE(FPNeg32, F32, F32, ) -OPCODE(FPNeg64, F64, F64, ) -OPCODE(FPRecip32, F32, F32, ) -OPCODE(FPRecip64, F64, F64, ) -OPCODE(FPRecipSqrt32, F32, F32, ) -OPCODE(FPRecipSqrt64, F64, F64, ) -OPCODE(FPSqrt, F32, F32, ) -OPCODE(FPSin, F32, F32, ) -OPCODE(FPExp2, F32, F32, ) -OPCODE(FPCos, F32, F32, ) -OPCODE(FPLog2, F32, F32, ) -OPCODE(FPSaturate16, F16, F16, ) -OPCODE(FPSaturate32, F32, F32, ) -OPCODE(FPSaturate64, F64, F64, ) -OPCODE(FPRoundEven16, F16, F16, ) -OPCODE(FPRoundEven32, F32, F32, ) -OPCODE(FPRoundEven64, F64, F64, ) -OPCODE(FPFloor16, F16, F16, ) -OPCODE(FPFloor32, F32, F32, ) -OPCODE(FPFloor64, F64, F64, ) -OPCODE(FPCeil16, F16, F16, ) -OPCODE(FPCeil32, F32, F32, ) -OPCODE(FPCeil64, F64, F64, ) -OPCODE(FPTrunc16, F16, F16, ) -OPCODE(FPTrunc32, F32, F32, ) -OPCODE(FPTrunc64, F64, F64, ) +OPCODE(FPAbs16, F16, F16, ) +OPCODE(FPAbs32, F32, F32, ) +OPCODE(FPAbs64, F64, F64, ) +OPCODE(FPAdd16, F16, F16, F16, ) +OPCODE(FPAdd32, F32, F32, F32, ) +OPCODE(FPAdd64, F64, F64, F64, ) +OPCODE(FPFma16, F16, F16, F16, F16, ) +OPCODE(FPFma32, F32, F32, F32, F32, ) +OPCODE(FPFma64, F64, F64, F64, F64, ) +OPCODE(FPMax32, F32, F32, F32, ) +OPCODE(FPMax64, F64, F64, F64, ) +OPCODE(FPMin32, F32, F32, F32, ) +OPCODE(FPMin64, F64, F64, F64, ) +OPCODE(FPMul16, F16, F16, F16, ) +OPCODE(FPMul32, F32, F32, F32, ) +OPCODE(FPMul64, F64, F64, F64, ) +OPCODE(FPNeg16, F16, F16, ) +OPCODE(FPNeg32, F32, F32, ) +OPCODE(FPNeg64, F64, F64, ) +OPCODE(FPRecip32, F32, F32, ) +OPCODE(FPRecip64, F64, F64, ) +OPCODE(FPRecipSqrt32, F32, F32, ) +OPCODE(FPRecipSqrt64, F64, F64, ) +OPCODE(FPSqrt, F32, F32, ) +OPCODE(FPSin, F32, F32, ) +OPCODE(FPExp2, F32, F32, ) +OPCODE(FPCos, F32, F32, ) +OPCODE(FPLog2, F32, F32, ) +OPCODE(FPSaturate16, F16, F16, ) +OPCODE(FPSaturate32, F32, F32, ) +OPCODE(FPSaturate64, F64, F64, ) +OPCODE(FPRoundEven16, F16, F16, ) +OPCODE(FPRoundEven32, F32, F32, ) +OPCODE(FPRoundEven64, F64, F64, ) +OPCODE(FPFloor16, F16, F16, ) +OPCODE(FPFloor32, F32, F32, ) +OPCODE(FPFloor64, F64, F64, ) +OPCODE(FPCeil16, F16, F16, ) +OPCODE(FPCeil32, F32, F32, ) +OPCODE(FPCeil64, F64, F64, ) +OPCODE(FPTrunc16, F16, F16, ) +OPCODE(FPTrunc32, F32, F32, ) +OPCODE(FPTrunc64, F64, F64, ) -OPCODE(FPOrdEqual16, U1, F16, F16, ) -OPCODE(FPOrdEqual32, U1, F32, F32, ) -OPCODE(FPOrdEqual64, U1, F64, F64, ) -OPCODE(FPUnordEqual16, U1, F16, F16, ) -OPCODE(FPUnordEqual32, U1, F32, F32, ) -OPCODE(FPUnordEqual64, U1, F64, F64, ) -OPCODE(FPOrdNotEqual16, U1, F16, F16, ) -OPCODE(FPOrdNotEqual32, U1, F32, F32, ) -OPCODE(FPOrdNotEqual64, U1, F64, F64, ) -OPCODE(FPUnordNotEqual16, U1, F16, F16, ) -OPCODE(FPUnordNotEqual32, U1, F32, F32, ) -OPCODE(FPUnordNotEqual64, U1, F64, F64, ) -OPCODE(FPOrdLessThan16, U1, F16, F16, ) -OPCODE(FPOrdLessThan32, U1, F32, F32, ) -OPCODE(FPOrdLessThan64, U1, F64, F64, ) -OPCODE(FPUnordLessThan16, U1, F16, F16, ) -OPCODE(FPUnordLessThan32, U1, F32, F32, ) -OPCODE(FPUnordLessThan64, U1, F64, F64, ) -OPCODE(FPOrdGreaterThan16, U1, F16, F16, ) -OPCODE(FPOrdGreaterThan32, U1, F32, F32, ) -OPCODE(FPOrdGreaterThan64, U1, F64, F64, ) -OPCODE(FPUnordGreaterThan16, U1, F16, F16, ) -OPCODE(FPUnordGreaterThan32, U1, F32, F32, ) -OPCODE(FPUnordGreaterThan64, U1, F64, F64, ) -OPCODE(FPOrdLessThanEqual16, U1, F16, F16, ) -OPCODE(FPOrdLessThanEqual32, U1, F32, F32, ) -OPCODE(FPOrdLessThanEqual64, U1, F64, F64, ) -OPCODE(FPUnordLessThanEqual16, U1, F16, F16, ) -OPCODE(FPUnordLessThanEqual32, U1, F32, F32, ) -OPCODE(FPUnordLessThanEqual64, U1, F64, F64, ) -OPCODE(FPOrdGreaterThanEqual16, U1, F16, F16, ) -OPCODE(FPOrdGreaterThanEqual32, U1, F32, F32, ) -OPCODE(FPOrdGreaterThanEqual64, U1, F64, F64, ) -OPCODE(FPUnordGreaterThanEqual16, U1, F16, F16, ) -OPCODE(FPUnordGreaterThanEqual32, U1, F32, F32, ) -OPCODE(FPUnordGreaterThanEqual64, U1, F64, F64, ) +OPCODE(FPOrdEqual16, U1, F16, F16, ) +OPCODE(FPOrdEqual32, U1, F32, F32, ) +OPCODE(FPOrdEqual64, U1, F64, F64, ) +OPCODE(FPUnordEqual16, U1, F16, F16, ) +OPCODE(FPUnordEqual32, U1, F32, F32, ) +OPCODE(FPUnordEqual64, U1, F64, F64, ) +OPCODE(FPOrdNotEqual16, U1, F16, F16, ) +OPCODE(FPOrdNotEqual32, U1, F32, F32, ) +OPCODE(FPOrdNotEqual64, U1, F64, F64, ) +OPCODE(FPUnordNotEqual16, U1, F16, F16, ) +OPCODE(FPUnordNotEqual32, U1, F32, F32, ) +OPCODE(FPUnordNotEqual64, U1, F64, F64, ) +OPCODE(FPOrdLessThan16, U1, F16, F16, ) +OPCODE(FPOrdLessThan32, U1, F32, F32, ) +OPCODE(FPOrdLessThan64, U1, F64, F64, ) +OPCODE(FPUnordLessThan16, U1, F16, F16, ) +OPCODE(FPUnordLessThan32, U1, F32, F32, ) +OPCODE(FPUnordLessThan64, U1, F64, F64, ) +OPCODE(FPOrdGreaterThan16, U1, F16, F16, ) +OPCODE(FPOrdGreaterThan32, U1, F32, F32, ) +OPCODE(FPOrdGreaterThan64, U1, F64, F64, ) +OPCODE(FPUnordGreaterThan16, U1, F16, F16, ) +OPCODE(FPUnordGreaterThan32, U1, F32, F32, ) +OPCODE(FPUnordGreaterThan64, U1, F64, F64, ) +OPCODE(FPOrdLessThanEqual16, U1, F16, F16, ) +OPCODE(FPOrdLessThanEqual32, U1, F32, F32, ) +OPCODE(FPOrdLessThanEqual64, U1, F64, F64, ) +OPCODE(FPUnordLessThanEqual16, U1, F16, F16, ) +OPCODE(FPUnordLessThanEqual32, U1, F32, F32, ) +OPCODE(FPUnordLessThanEqual64, U1, F64, F64, ) +OPCODE(FPOrdGreaterThanEqual16, U1, F16, F16, ) +OPCODE(FPOrdGreaterThanEqual32, U1, F32, F32, ) +OPCODE(FPOrdGreaterThanEqual64, U1, F64, F64, ) +OPCODE(FPUnordGreaterThanEqual16, U1, F16, F16, ) +OPCODE(FPUnordGreaterThanEqual32, U1, F32, F32, ) +OPCODE(FPUnordGreaterThanEqual64, U1, F64, F64, ) // Integer operations -OPCODE(IAdd32, U32, U32, U32, ) -OPCODE(IAdd64, U64, U64, U64, ) -OPCODE(ISub32, U32, U32, U32, ) -OPCODE(ISub64, U64, U64, U64, ) -OPCODE(IMul32, U32, U32, U32, ) -OPCODE(INeg32, U32, U32, ) -OPCODE(INeg64, U64, U64, ) -OPCODE(IAbs32, U32, U32, ) -OPCODE(ShiftLeftLogical32, U32, U32, U32, ) -OPCODE(ShiftLeftLogical64, U64, U64, U32, ) -OPCODE(ShiftRightLogical32, U32, U32, U32, ) -OPCODE(ShiftRightLogical64, U64, U64, U32, ) -OPCODE(ShiftRightArithmetic32, U32, U32, U32, ) -OPCODE(ShiftRightArithmetic64, U64, U64, U32, ) -OPCODE(BitwiseAnd32, U32, U32, U32, ) -OPCODE(BitwiseOr32, U32, U32, U32, ) -OPCODE(BitwiseXor32, U32, U32, U32, ) -OPCODE(BitFieldInsert, U32, U32, U32, U32, U32, ) -OPCODE(BitFieldSExtract, U32, U32, U32, U32, ) -OPCODE(BitFieldUExtract, U32, U32, U32, U32, ) -OPCODE(BitReverse32, U32, U32, ) -OPCODE(BitCount32, U32, U32, ) -OPCODE(BitwiseNot32, U32, U32, ) +OPCODE(IAdd32, U32, U32, U32, ) +OPCODE(IAdd64, U64, U64, U64, ) +OPCODE(ISub32, U32, U32, U32, ) +OPCODE(ISub64, U64, U64, U64, ) +OPCODE(IMul32, U32, U32, U32, ) +OPCODE(INeg32, U32, U32, ) +OPCODE(INeg64, U64, U64, ) +OPCODE(IAbs32, U32, U32, ) +OPCODE(ShiftLeftLogical32, U32, U32, U32, ) +OPCODE(ShiftLeftLogical64, U64, U64, U32, ) +OPCODE(ShiftRightLogical32, U32, U32, U32, ) +OPCODE(ShiftRightLogical64, U64, U64, U32, ) +OPCODE(ShiftRightArithmetic32, U32, U32, U32, ) +OPCODE(ShiftRightArithmetic64, U64, U64, U32, ) +OPCODE(BitwiseAnd32, U32, U32, U32, ) +OPCODE(BitwiseOr32, U32, U32, U32, ) +OPCODE(BitwiseXor32, U32, U32, U32, ) +OPCODE(BitFieldInsert, U32, U32, U32, U32, U32, ) +OPCODE(BitFieldSExtract, U32, U32, U32, U32, ) +OPCODE(BitFieldUExtract, U32, U32, U32, U32, ) +OPCODE(BitReverse32, U32, U32, ) +OPCODE(BitCount32, U32, U32, ) +OPCODE(BitwiseNot32, U32, U32, ) -OPCODE(FindSMsb32, U32, U32, ) -OPCODE(FindUMsb32, U32, U32, ) -OPCODE(SMin32, U32, U32, U32, ) -OPCODE(UMin32, U32, U32, U32, ) -OPCODE(SMax32, U32, U32, U32, ) -OPCODE(UMax32, U32, U32, U32, ) -OPCODE(SLessThan, U1, U32, U32, ) -OPCODE(ULessThan, U1, U32, U32, ) -OPCODE(IEqual, U1, U32, U32, ) -OPCODE(SLessThanEqual, U1, U32, U32, ) -OPCODE(ULessThanEqual, U1, U32, U32, ) -OPCODE(SGreaterThan, U1, U32, U32, ) -OPCODE(UGreaterThan, U1, U32, U32, ) -OPCODE(INotEqual, U1, U32, U32, ) -OPCODE(SGreaterThanEqual, U1, U32, U32, ) -OPCODE(UGreaterThanEqual, U1, U32, U32, ) +OPCODE(FindSMsb32, U32, U32, ) +OPCODE(FindUMsb32, U32, U32, ) +OPCODE(SMin32, U32, U32, U32, ) +OPCODE(UMin32, U32, U32, U32, ) +OPCODE(SMax32, U32, U32, U32, ) +OPCODE(UMax32, U32, U32, U32, ) +OPCODE(SLessThan, U1, U32, U32, ) +OPCODE(ULessThan, U1, U32, U32, ) +OPCODE(IEqual, U1, U32, U32, ) +OPCODE(SLessThanEqual, U1, U32, U32, ) +OPCODE(ULessThanEqual, U1, U32, U32, ) +OPCODE(SGreaterThan, U1, U32, U32, ) +OPCODE(UGreaterThan, U1, U32, U32, ) +OPCODE(INotEqual, U1, U32, U32, ) +OPCODE(SGreaterThanEqual, U1, U32, U32, ) +OPCODE(UGreaterThanEqual, U1, U32, U32, ) // Logical operations -OPCODE(LogicalOr, U1, U1, U1, ) -OPCODE(LogicalAnd, U1, U1, U1, ) -OPCODE(LogicalXor, U1, U1, U1, ) -OPCODE(LogicalNot, U1, U1, ) +OPCODE(LogicalOr, U1, U1, U1, ) +OPCODE(LogicalAnd, U1, U1, U1, ) +OPCODE(LogicalXor, U1, U1, U1, ) +OPCODE(LogicalNot, U1, U1, ) // Conversion operations -OPCODE(ConvertS16F16, U32, F16, ) -OPCODE(ConvertS16F32, U32, F32, ) -OPCODE(ConvertS16F64, U32, F64, ) -OPCODE(ConvertS32F16, U32, F16, ) -OPCODE(ConvertS32F32, U32, F32, ) -OPCODE(ConvertS32F64, U32, F64, ) -OPCODE(ConvertS64F16, U64, F16, ) -OPCODE(ConvertS64F32, U64, F32, ) -OPCODE(ConvertS64F64, U64, F64, ) -OPCODE(ConvertU16F16, U32, F16, ) -OPCODE(ConvertU16F32, U32, F32, ) -OPCODE(ConvertU16F64, U32, F64, ) -OPCODE(ConvertU32F16, U32, F16, ) -OPCODE(ConvertU32F32, U32, F32, ) -OPCODE(ConvertU32F64, U32, F64, ) -OPCODE(ConvertU64F16, U64, F16, ) -OPCODE(ConvertU64F32, U64, F32, ) -OPCODE(ConvertU64F64, U64, F64, ) -OPCODE(ConvertU64U32, U64, U32, ) -OPCODE(ConvertU32U64, U32, U64, ) -OPCODE(ConvertF16F32, F16, F32, ) -OPCODE(ConvertF32F16, F32, F16, ) -OPCODE(ConvertF32F64, F32, F64, ) -OPCODE(ConvertF64F32, F64, F32, ) +OPCODE(ConvertS16F16, U32, F16, ) +OPCODE(ConvertS16F32, U32, F32, ) +OPCODE(ConvertS16F64, U32, F64, ) +OPCODE(ConvertS32F16, U32, F16, ) +OPCODE(ConvertS32F32, U32, F32, ) +OPCODE(ConvertS32F64, U32, F64, ) +OPCODE(ConvertS64F16, U64, F16, ) +OPCODE(ConvertS64F32, U64, F32, ) +OPCODE(ConvertS64F64, U64, F64, ) +OPCODE(ConvertU16F16, U32, F16, ) +OPCODE(ConvertU16F32, U32, F32, ) +OPCODE(ConvertU16F64, U32, F64, ) +OPCODE(ConvertU32F16, U32, F16, ) +OPCODE(ConvertU32F32, U32, F32, ) +OPCODE(ConvertU32F64, U32, F64, ) +OPCODE(ConvertU64F16, U64, F16, ) +OPCODE(ConvertU64F32, U64, F32, ) +OPCODE(ConvertU64F64, U64, F64, ) +OPCODE(ConvertU64U32, U64, U32, ) +OPCODE(ConvertU32U64, U32, U64, ) +OPCODE(ConvertF16F32, F16, F32, ) +OPCODE(ConvertF32F16, F32, F16, ) +OPCODE(ConvertF32F64, F32, F64, ) +OPCODE(ConvertF64F32, F64, F32, ) +OPCODE(ConvertF16S32, F16, U32, ) +OPCODE(ConvertF16S64, F16, U64, ) +OPCODE(ConvertF16U32, F16, U32, ) +OPCODE(ConvertF16U64, F16, U64, ) +OPCODE(ConvertF32S32, F32, U32, ) +OPCODE(ConvertF32S64, F32, U64, ) +OPCODE(ConvertF32U32, F32, U32, ) +OPCODE(ConvertF32U64, F32, U64, ) +OPCODE(ConvertF64S32, F64, U32, ) +OPCODE(ConvertF64S64, F64, U64, ) +OPCODE(ConvertF64U32, F64, U32, ) +OPCODE(ConvertF64U64, F64, U64, ) + +// Image operations +OPCODE(BindlessImageSampleImplicitLod, F32x4, U32, Opaque, Opaque, Opaque, ) +OPCODE(BindlessImageSampleExplicitLod, F32x4, U32, Opaque, Opaque, Opaque, ) +OPCODE(BindlessImageSampleDrefImplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, ) +OPCODE(BindlessImageSampleDrefExplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, ) + +OPCODE(BoundImageSampleImplicitLod, F32x4, U32, Opaque, Opaque, Opaque, ) +OPCODE(BoundImageSampleExplicitLod, F32x4, U32, Opaque, Opaque, Opaque, ) +OPCODE(BoundImageSampleDrefImplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, ) +OPCODE(BoundImageSampleDrefExplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, ) + +OPCODE(ImageSampleImplicitLod, F32x4, U32, Opaque, Opaque, Opaque, ) +OPCODE(ImageSampleExplicitLod, F32x4, U32, Opaque, Opaque, Opaque, ) +OPCODE(ImageSampleDrefImplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, ) +OPCODE(ImageSampleDrefExplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, ) diff --git a/src/shader_recompiler/frontend/ir/reg.h b/src/shader_recompiler/frontend/ir/reg.h index 771094eb9..8fea05f7b 100644 --- a/src/shader_recompiler/frontend/ir/reg.h +++ b/src/shader_recompiler/frontend/ir/reg.h @@ -293,6 +293,17 @@ constexpr size_t NUM_REGS = 256; return reg + (-num); } +[[nodiscard]] constexpr Reg operator++(Reg& reg) { + reg = reg + 1; + return reg; +} + +[[nodiscard]] constexpr Reg operator++(Reg& reg, int) { + const Reg copy{reg}; + reg = reg + 1; + return copy; +} + [[nodiscard]] constexpr size_t RegIndex(Reg reg) noexcept { return static_cast<size_t>(reg); } diff --git a/src/shader_recompiler/frontend/ir/value.h b/src/shader_recompiler/frontend/ir/value.h index 9b7e1480b..3602883d6 100644 --- a/src/shader_recompiler/frontend/ir/value.h +++ b/src/shader_recompiler/frontend/ir/value.h @@ -75,6 +75,7 @@ private: f64 imm_f64; }; }; +static_assert(std::is_trivially_copyable_v<Value>); template <IR::Type type_> class TypedValue : public Value { diff --git a/src/shader_recompiler/frontend/maxwell/maxwell.inc b/src/shader_recompiler/frontend/maxwell/maxwell.inc index 5d0b91598..f2a2ff331 100644 --- a/src/shader_recompiler/frontend/maxwell/maxwell.inc +++ b/src/shader_recompiler/frontend/maxwell/maxwell.inc @@ -249,8 +249,8 @@ INST(SULD, "SULD", "1110 1011 000- ----") INST(SURED, "SURED", "1110 1011 010- ----") INST(SUST, "SUST", "1110 1011 001- ----") INST(SYNC, "SYNC", "1111 0000 1111 1---") -INST(TEX, "TEX", "1100 00-- --11 1---") -INST(TEX_b, "TEX (b)", "1101 1110 1011 1---") +INST(TEX, "TEX", "1100 0--- ---- ----") +INST(TEX_b, "TEX (b)", "1101 1110 10-- ----") INST(TEXS, "TEXS", "1101 -00- ---- ----") INST(TLD, "TLD", "1101 1100 --11 1---") INST(TLD_b, "TLD (b)", "1101 1101 --11 1---") diff --git a/src/shader_recompiler/frontend/maxwell/program.cpp b/src/shader_recompiler/frontend/maxwell/program.cpp index dbfc04f75..b270bbccd 100644 --- a/src/shader_recompiler/frontend/maxwell/program.cpp +++ b/src/shader_recompiler/frontend/maxwell/program.cpp @@ -62,6 +62,7 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo Optimization::SsaRewritePass(function.post_order_blocks); } Optimization::GlobalMemoryToStorageBufferPass(program); + Optimization::TexturePass(env, program); for (IR::Function& function : functions) { Optimization::PostOrderInvoke(Optimization::ConstantPropagationPass, function); Optimization::PostOrderInvoke(Optimization::DeadCodeEliminationPass, function); diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp index fc6030e04..ff429c126 100644 --- a/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp +++ b/src/shader_recompiler/frontend/maxwell/translate/impl/not_implemented.cpp @@ -585,14 +585,6 @@ void TranslatorVisitor::SYNC(u64) { ThrowNotImplemented(Opcode::SYNC); } -void TranslatorVisitor::TEX(u64) { - ThrowNotImplemented(Opcode::TEX); -} - -void TranslatorVisitor::TEX_b(u64) { - ThrowNotImplemented(Opcode::TEX_b); -} - void TranslatorVisitor::TEXS(u64) { ThrowNotImplemented(Opcode::TEXS); } diff --git a/src/shader_recompiler/frontend/maxwell/translate/impl/texture_sample.cpp b/src/shader_recompiler/frontend/maxwell/translate/impl/texture_sample.cpp new file mode 100644 index 000000000..98d9f4c64 --- /dev/null +++ b/src/shader_recompiler/frontend/maxwell/translate/impl/texture_sample.cpp @@ -0,0 +1,232 @@ +// Copyright 2021 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <optional> + +#include "common/bit_field.h" +#include "common/common_types.h" +#include "shader_recompiler/frontend/ir/modifiers.h" +#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h" + +namespace Shader::Maxwell { +namespace { +enum class Blod : u64 { + None, + LZ, + LB, + LL, + INVALIDBLOD4, + INVALIDBLOD5, + LBA, + LLA, +}; + +enum class TextureType : u64 { + _1D, + ARRAY_1D, + _2D, + ARRAY_2D, + _3D, + ARRAY_3D, + CUBE, + ARRAY_CUBE, +}; + +Shader::TextureType GetType(TextureType type, bool dc) { + switch (type) { + case TextureType::_1D: + return dc ? Shader::TextureType::Shadow1D : Shader::TextureType::Color1D; + case TextureType::ARRAY_1D: + return dc ? Shader::TextureType::ShadowArray1D : Shader::TextureType::ColorArray1D; + case TextureType::_2D: + return dc ? Shader::TextureType::Shadow2D : Shader::TextureType::Color2D; + case TextureType::ARRAY_2D: + return dc ? Shader::TextureType::ShadowArray2D : Shader::TextureType::ColorArray2D; + case TextureType::_3D: + return dc ? Shader::TextureType::Shadow3D : Shader::TextureType::Color3D; + case TextureType::ARRAY_3D: + throw NotImplementedException("3D array texture type"); + case TextureType::CUBE: + return dc ? Shader::TextureType::ShadowCube : Shader::TextureType::ColorCube; + case TextureType::ARRAY_CUBE: + return dc ? Shader::TextureType::ShadowArrayCube : Shader::TextureType::ColorArrayCube; + } + throw NotImplementedException("Invalid texture type {}", type); +} + +IR::Value MakeCoords(TranslatorVisitor& v, IR::Reg reg, TextureType type) { + const auto read_array{[&]() -> IR::F32 { return v.ir.ConvertUToF(32, v.X(reg)); }}; + switch (type) { + case TextureType::_1D: + return v.F(reg); + case TextureType::ARRAY_1D: + return v.ir.CompositeConstruct(read_array(), v.F(reg + 1)); + case TextureType::_2D: + return v.ir.CompositeConstruct(v.F(reg), v.F(reg + 1)); + case TextureType::ARRAY_2D: + return v.ir.CompositeConstruct(read_array(), v.F(reg + 1), v.F(reg + 2)); + case TextureType::_3D: + return v.ir.CompositeConstruct(v.F(reg), v.F(reg + 1), v.F(reg + 2)); + case TextureType::ARRAY_3D: + throw NotImplementedException("3D array texture type"); + case TextureType::CUBE: + return v.ir.CompositeConstruct(v.F(reg), v.F(reg + 1), v.F(reg + 2)); + case TextureType::ARRAY_CUBE: + return v.ir.CompositeConstruct(read_array(), v.F(reg + 1), v.F(reg + 2), v.F(reg + 3)); + } + throw NotImplementedException("Invalid texture type {}", type); +} + +IR::F32 MakeLod(TranslatorVisitor& v, IR::Reg& reg, Blod blod) { + switch (blod) { + case Blod::None: + return v.ir.Imm32(0.0f); + case Blod::LZ: + return v.ir.Imm32(0.0f); + case Blod::LB: + case Blod::LL: + case Blod::LBA: + case Blod::LLA: + return v.F(reg++); + case Blod::INVALIDBLOD4: + case Blod::INVALIDBLOD5: + break; + } + throw NotImplementedException("Invalid blod {}", blod); +} + +IR::Value MakeOffset(TranslatorVisitor& v, IR::Reg& reg, TextureType type) { + const IR::U32 value{v.X(reg++)}; + switch (type) { + case TextureType::_1D: + case TextureType::ARRAY_1D: + return v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(4)); + case TextureType::_2D: + case TextureType::ARRAY_2D: + return v.ir.CompositeConstruct(v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(4)), + v.ir.BitFieldExtract(value, v.ir.Imm32(4), v.ir.Imm32(4))); + case TextureType::_3D: + case TextureType::ARRAY_3D: + return v.ir.CompositeConstruct(v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(4)), + v.ir.BitFieldExtract(value, v.ir.Imm32(4), v.ir.Imm32(4)), + v.ir.BitFieldExtract(value, v.ir.Imm32(8), v.ir.Imm32(4))); + case TextureType::CUBE: + case TextureType::ARRAY_CUBE: + throw NotImplementedException("Illegal offset on CUBE sample"); + } + throw NotImplementedException("Invalid texture type {}", type); +} + +bool HasExplicitLod(Blod blod) { + switch (blod) { + case Blod::LL: + case Blod::LLA: + case Blod::LZ: + return true; + default: + return false; + } +} + +void Impl(TranslatorVisitor& v, u64 insn, bool aoffi, Blod blod, bool lc, + std::optional<u32> cbuf_offset) { + union { + u64 raw; + BitField<35, 1, u64> ndv; + BitField<49, 1, u64> nodep; + BitField<50, 1, u64> dc; + BitField<51, 3, IR::Pred> sparse_pred; + BitField<0, 8, IR::Reg> dest_reg; + BitField<8, 8, IR::Reg> coord_reg; + BitField<20, 8, IR::Reg> meta_reg; + BitField<28, 3, TextureType> type; + BitField<31, 4, u64> mask; + } const tex{insn}; + + if (lc) { + throw NotImplementedException("LC"); + } + const IR::Value coords{MakeCoords(v, tex.coord_reg, tex.type)}; + + IR::Reg meta_reg{tex.meta_reg}; + IR::Value handle; + IR::Value offset; + IR::F32 dref; + IR::F32 lod_clamp; + if (cbuf_offset) { + handle = v.ir.Imm32(*cbuf_offset); + } else { + handle = v.X(meta_reg++); + } + const IR::F32 lod{MakeLod(v, meta_reg, blod)}; + if (aoffi) { + offset = MakeOffset(v, meta_reg, tex.type); + } + if (tex.dc != 0) { + dref = v.F(meta_reg++); + } + IR::TextureInstInfo info{}; + info.type.Assign(GetType(tex.type, tex.dc != 0)); + info.has_bias.Assign(blod == Blod::LB || blod == Blod::LBA ? 1 : 0); + info.has_lod_clamp.Assign(lc ? 1 : 0); + + const IR::Value sample{[&]() -> IR::Value { + if (tex.dc == 0) { + if (HasExplicitLod(blod)) { + return v.ir.ImageSampleExplicitLod(handle, coords, lod, offset, lod_clamp, info); + } else { + return v.ir.ImageSampleImplicitLod(handle, coords, lod, offset, lod_clamp, info); + } + } + if (HasExplicitLod(blod)) { + return v.ir.ImageSampleDrefExplicitLod(handle, coords, dref, lod, offset, lod_clamp, + info); + } else { + return v.ir.ImageSampleDrefImplicitLod(handle, coords, dref, lod, offset, lod_clamp, + info); + } + }()}; + + for (int element = 0; element < 4; ++element) { + if (((tex.mask >> element) & 1) == 0) { + continue; + } + IR::F32 value; + if (tex.dc != 0) { + value = element < 3 ? IR::F32{sample} : v.ir.Imm32(1.0f); + } else { + value = IR::F32{v.ir.CompositeExtract(sample, element)}; + } + v.F(tex.dest_reg + element, value); + } + if (tex.sparse_pred != IR::Pred::PT) { + v.ir.SetPred(tex.sparse_pred, v.ir.LogicalNot(v.ir.GetSparseFromOp(sample))); + } +} +} // Anonymous namespace + +void TranslatorVisitor::TEX(u64 insn) { + union { + u64 raw; + BitField<54, 1, u64> aoffi; + BitField<55, 3, Blod> blod; + BitField<58, 1, u64> lc; + BitField<36, 13, u64> cbuf_offset; + } const tex{insn}; + + Impl(*this, insn, tex.aoffi != 0, tex.blod, tex.lc != 0, static_cast<u32>(tex.cbuf_offset)); +} + +void TranslatorVisitor::TEX_b(u64 insn) { + union { + u64 raw; + BitField<36, 1, u64> aoffi; + BitField<37, 3, Blod> blod; + BitField<40, 1, u64> lc; + } const tex{insn}; + + Impl(*this, insn, tex.aoffi != 0, tex.blod, tex.lc != 0, std::nullopt); +} + +} // namespace Shader::Maxwell |