diff options
Diffstat (limited to 'src/shader_recompiler/ir_opt')
13 files changed, 308 insertions, 110 deletions
diff --git a/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp b/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp index bfd2ae650..7cff8ecdc 100644 --- a/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp +++ b/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "common/alignment.h" #include "shader_recompiler/environment.h" @@ -29,6 +28,41 @@ void AddConstantBufferDescriptor(Info& info, u32 index, u32 count) { }); } +void AddRegisterIndexedLdc(Info& info) { + info.uses_cbuf_indirect = true; + + for (u32 i = 0; i < Info::MAX_INDIRECT_CBUFS; i++) { + AddConstantBufferDescriptor(info, i, 1); + + // The shader can use any possible access size + info.constant_buffer_used_sizes[i] = 0x10'000; + } +} + +u32 GetElementSize(IR::Type& used_type, Shader::IR::Opcode opcode) { + switch (opcode) { + case IR::Opcode::GetCbufU8: + case IR::Opcode::GetCbufS8: + used_type |= IR::Type::U8; + return 1; + case IR::Opcode::GetCbufU16: + case IR::Opcode::GetCbufS16: + used_type |= IR::Type::U16; + return 2; + case IR::Opcode::GetCbufU32: + used_type |= IR::Type::U32; + return 4; + case IR::Opcode::GetCbufF32: + used_type |= IR::Type::F32; + return 4; + case IR::Opcode::GetCbufU32x2: + used_type |= IR::Type::U32x2; + return 8; + default: + throw InvalidArgument("Invalid opcode {}", opcode); + } +} + void GetPatch(Info& info, IR::Patch patch) { if (!IR::IsGeneric(patch)) { throw NotImplementedException("Reading non-generic patch {}", patch); @@ -463,42 +497,18 @@ void VisitUsages(Info& info, IR::Inst& inst) { case IR::Opcode::GetCbufU32x2: { const IR::Value index{inst.Arg(0)}; const IR::Value offset{inst.Arg(1)}; - if (!index.IsImmediate()) { - throw NotImplementedException("Constant buffer with non-immediate index"); - } - AddConstantBufferDescriptor(info, index.U32(), 1); - u32 element_size{}; - switch (inst.GetOpcode()) { - case IR::Opcode::GetCbufU8: - case IR::Opcode::GetCbufS8: - info.used_constant_buffer_types |= IR::Type::U8; - element_size = 1; - break; - case IR::Opcode::GetCbufU16: - case IR::Opcode::GetCbufS16: - info.used_constant_buffer_types |= IR::Type::U16; - element_size = 2; - break; - case IR::Opcode::GetCbufU32: - info.used_constant_buffer_types |= IR::Type::U32; - element_size = 4; - break; - case IR::Opcode::GetCbufF32: - info.used_constant_buffer_types |= IR::Type::F32; - element_size = 4; - break; - case IR::Opcode::GetCbufU32x2: - info.used_constant_buffer_types |= IR::Type::U32x2; - element_size = 8; - break; - default: - break; - } - u32& size{info.constant_buffer_used_sizes[index.U32()]}; - if (offset.IsImmediate()) { - size = Common::AlignUp(std::max(size, offset.U32() + element_size), 16u); + if (index.IsImmediate()) { + AddConstantBufferDescriptor(info, index.U32(), 1); + u32 element_size = GetElementSize(info.used_constant_buffer_types, inst.GetOpcode()); + u32& size{info.constant_buffer_used_sizes[index.U32()]}; + if (offset.IsImmediate()) { + size = Common::AlignUp(std::max(size, offset.U32() + element_size), 16u); + } else { + size = 0x10'000; + } } else { - size = 0x10'000; + AddRegisterIndexedLdc(info); + GetElementSize(info.used_indirect_cbuf_types, inst.GetOpcode()); } break; } diff --git a/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp b/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp index c134a12bc..826f9a54a 100644 --- a/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp +++ b/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <functional> @@ -8,7 +7,6 @@ #include <type_traits> #include "common/bit_cast.h" -#include "common/bit_util.h" #include "shader_recompiler/exception.h" #include "shader_recompiler/frontend/ir/ir_emitter.h" #include "shader_recompiler/frontend/ir/value.h" diff --git a/src/shader_recompiler/ir_opt/dead_code_elimination_pass.cpp b/src/shader_recompiler/ir_opt/dead_code_elimination_pass.cpp index 400836301..9a7d47344 100644 --- a/src/shader_recompiler/ir_opt/dead_code_elimination_pass.cpp +++ b/src/shader_recompiler/ir_opt/dead_code_elimination_pass.cpp @@ -1,25 +1,104 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#include <algorithm> + +#include <boost/container/small_vector.hpp> #include "shader_recompiler/frontend/ir/basic_block.h" #include "shader_recompiler/frontend/ir/value.h" #include "shader_recompiler/ir_opt/passes.h" namespace Shader::Optimization { - -void DeadCodeEliminationPass(IR::Program& program) { +namespace { +template <bool TEST_USES> +void DeadInstElimination(IR::Block* const block) { // We iterate over the instructions in reverse order. // This is because removing an instruction reduces the number of uses for earlier instructions. - for (IR::Block* const block : program.post_order_blocks) { - auto it{block->end()}; - while (it != block->begin()) { - --it; - if (!it->HasUses() && !it->MayHaveSideEffects()) { - it->Invalidate(); - it = block->Instructions().erase(it); + auto it{block->end()}; + while (it != block->begin()) { + --it; + if constexpr (TEST_USES) { + if (it->HasUses() || it->MayHaveSideEffects()) { + continue; + } + } + it->Invalidate(); + it = block->Instructions().erase(it); + } +} + +void DeletedPhiArgElimination(IR::Program& program, std::span<const IR::Block*> dead_blocks) { + for (IR::Block* const block : program.blocks) { + for (IR::Inst& phi : *block) { + if (!IR::IsPhi(phi)) { + continue; + } + for (size_t i = 0; i < phi.NumArgs(); ++i) { + if (std::ranges::find(dead_blocks, phi.PhiBlock(i)) == dead_blocks.end()) { + continue; + } + // Phi operand at this index is an unreachable block + phi.ErasePhiOperand(i); + --i; + } + } + } +} + +void DeadBranchElimination(IR::Program& program) { + boost::container::small_vector<const IR::Block*, 3> dead_blocks; + const auto begin_it{program.syntax_list.begin()}; + for (auto node_it = begin_it; node_it != program.syntax_list.end(); ++node_it) { + if (node_it->type != IR::AbstractSyntaxNode::Type::If) { + continue; + } + IR::Inst* const cond_ref{node_it->data.if_node.cond.Inst()}; + const IR::U1 cond{cond_ref->Arg(0)}; + if (!cond.IsImmediate()) { + continue; + } + if (cond.U1()) { + continue; + } + // False immediate condition. Remove condition ref, erase the entire branch. + cond_ref->Invalidate(); + // Account for nested if-statements within the if(false) branch + u32 nested_ifs{1u}; + while (node_it->type != IR::AbstractSyntaxNode::Type::EndIf || nested_ifs > 0) { + node_it = program.syntax_list.erase(node_it); + switch (node_it->type) { + case IR::AbstractSyntaxNode::Type::If: + ++nested_ifs; + break; + case IR::AbstractSyntaxNode::Type::EndIf: + --nested_ifs; + break; + case IR::AbstractSyntaxNode::Type::Block: { + IR::Block* const block{node_it->data.block}; + DeadInstElimination<false>(block); + dead_blocks.push_back(block); + break; + } + default: + break; } } + // Erase EndIf node of the if(false) branch + node_it = program.syntax_list.erase(node_it); + // Account for loop increment + --node_it; + } + if (!dead_blocks.empty()) { + DeletedPhiArgElimination(program, std::span(dead_blocks.data(), dead_blocks.size())); + } +} +} // namespace + +void DeadCodeEliminationPass(IR::Program& program) { + DeadBranchElimination(program); + for (IR::Block* const block : program.post_order_blocks) { + DeadInstElimination<true>(block); } } diff --git a/src/shader_recompiler/ir_opt/dual_vertex_pass.cpp b/src/shader_recompiler/ir_opt/dual_vertex_pass.cpp index 055ba9c54..900f6f9fd 100644 --- a/src/shader_recompiler/ir_opt/dual_vertex_pass.cpp +++ b/src/shader_recompiler/ir_opt/dual_vertex_pass.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include "shader_recompiler/frontend/ir/ir_emitter.h" #include "shader_recompiler/ir_opt/passes.h" diff --git a/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp b/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp index 38592afd0..336338e62 100644 --- a/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp +++ b/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp @@ -1,11 +1,7 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later -#include <algorithm> -#include <compare> #include <optional> -#include <queue> #include <boost/container/flat_set.hpp> #include <boost/container/small_vector.hpp> @@ -334,7 +330,8 @@ std::optional<LowAddrInfo> TrackLowAddress(IR::Inst* inst) { /// Tries to track the storage buffer address used by a global memory instruction std::optional<StorageBufferAddr> Track(const IR::Value& value, const Bias* bias) { const auto pred{[bias](const IR::Inst* inst) -> std::optional<StorageBufferAddr> { - if (inst->GetOpcode() != IR::Opcode::GetCbufU32) { + if (inst->GetOpcode() != IR::Opcode::GetCbufU32 && + inst->GetOpcode() != IR::Opcode::GetCbufU32x2) { return std::nullopt; } const IR::Value index{inst->Arg(0)}; diff --git a/src/shader_recompiler/ir_opt/identity_removal_pass.cpp b/src/shader_recompiler/ir_opt/identity_removal_pass.cpp index e9b55f835..951534bbf 100644 --- a/src/shader_recompiler/ir_opt/identity_removal_pass.cpp +++ b/src/shader_recompiler/ir_opt/identity_removal_pass.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <vector> diff --git a/src/shader_recompiler/ir_opt/lower_fp16_to_fp32.cpp b/src/shader_recompiler/ir_opt/lower_fp16_to_fp32.cpp index 773e1f961..71e12b3e4 100644 --- a/src/shader_recompiler/ir_opt/lower_fp16_to_fp32.cpp +++ b/src/shader_recompiler/ir_opt/lower_fp16_to_fp32.cpp @@ -1,10 +1,6 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later -#include <algorithm> - -#include "shader_recompiler/frontend/ir/ir_emitter.h" #include "shader_recompiler/frontend/ir/value.h" #include "shader_recompiler/ir_opt/passes.h" diff --git a/src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp b/src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp index c2654cd9b..cdb58f46b 100644 --- a/src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp +++ b/src/shader_recompiler/ir_opt/lower_int64_to_int32.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <utility> diff --git a/src/shader_recompiler/ir_opt/passes.h b/src/shader_recompiler/ir_opt/passes.h index f877c7ba0..6ff8e4266 100644 --- a/src/shader_recompiler/ir_opt/passes.h +++ b/src/shader_recompiler/ir_opt/passes.h @@ -1,13 +1,9 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #pragma once -#include <span> - #include "shader_recompiler/environment.h" -#include "shader_recompiler/frontend/ir/basic_block.h" #include "shader_recompiler/frontend/ir/program.h" namespace Shader::Optimization { diff --git a/src/shader_recompiler/ir_opt/rescaling_pass.cpp b/src/shader_recompiler/ir_opt/rescaling_pass.cpp index c28500dd1..9198fa5f2 100644 --- a/src/shader_recompiler/ir_opt/rescaling_pass.cpp +++ b/src/shader_recompiler/ir_opt/rescaling_pass.cpp @@ -1,8 +1,6 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later -#include "common/alignment.h" #include "common/settings.h" #include "shader_recompiler/environment.h" #include "shader_recompiler/frontend/ir/ir_emitter.h" @@ -18,6 +16,7 @@ namespace { switch (type) { case TextureType::Color2D: case TextureType::ColorArray2D: + case TextureType::Color2DRect: return true; case TextureType::Color1D: case TextureType::ColorArray1D: @@ -134,7 +133,8 @@ void PatchImageQueryDimensions(IR::Block& block, IR::Inst& inst) { const IR::U1 is_scaled{ir.IsTextureScaled(ir.Imm32(info.descriptor_index))}; switch (info.type) { case TextureType::Color2D: - case TextureType::ColorArray2D: { + case TextureType::ColorArray2D: + case TextureType::Color2DRect: { const IR::Value new_inst{&*block.PrependNewInst(it, inst)}; const IR::U32 width{DownScale(ir, is_scaled, IR::U32{ir.CompositeExtract(new_inst, 0)})}; const IR::U32 height{DownScale(ir, is_scaled, IR::U32{ir.CompositeExtract(new_inst, 1)})}; @@ -165,6 +165,7 @@ void ScaleIntegerComposite(IR::IREmitter& ir, IR::Inst& inst, const IR::U1& is_s const IR::U32 y{Scale(ir, is_scaled, IR::U32{ir.CompositeExtract(composite, 1)})}; switch (info.type) { case TextureType::Color2D: + case TextureType::Color2DRect: inst.SetArg(index, ir.CompositeConstruct(x, y)); break; case TextureType::ColorArray2D: { @@ -183,6 +184,32 @@ void ScaleIntegerComposite(IR::IREmitter& ir, IR::Inst& inst, const IR::U1& is_s } } +void ScaleIntegerOffsetComposite(IR::IREmitter& ir, IR::Inst& inst, const IR::U1& is_scaled, + size_t index) { + const IR::Value composite{inst.Arg(index)}; + if (composite.IsEmpty()) { + return; + } + const auto info{inst.Flags<IR::TextureInstInfo>()}; + const IR::U32 x{Scale(ir, is_scaled, IR::U32{ir.CompositeExtract(composite, 0)})}; + const IR::U32 y{Scale(ir, is_scaled, IR::U32{ir.CompositeExtract(composite, 1)})}; + switch (info.type) { + case TextureType::ColorArray2D: + case TextureType::Color2D: + case TextureType::Color2DRect: + inst.SetArg(index, ir.CompositeConstruct(x, y)); + break; + case TextureType::Color1D: + case TextureType::ColorArray1D: + case TextureType::Color3D: + case TextureType::ColorCube: + case TextureType::ColorArrayCube: + case TextureType::Buffer: + // Nothing to patch here + break; + } +} + void SubScaleCoord(IR::IREmitter& ir, IR::Inst& inst, const IR::U1& is_scaled) { const auto info{inst.Flags<IR::TextureInstInfo>()}; const IR::Value coord{inst.Arg(1)}; @@ -193,6 +220,7 @@ void SubScaleCoord(IR::IREmitter& ir, IR::Inst& inst, const IR::U1& is_scaled) { const IR::U32 scaled_y{SubScale(ir, is_scaled, coord_y, IR::Attribute::PositionY)}; switch (info.type) { case TextureType::Color2D: + case TextureType::Color2DRect: inst.SetArg(1, ir.CompositeConstruct(scaled_x, scaled_y)); break; case TextureType::ColorArray2D: { @@ -220,7 +248,7 @@ void SubScaleImageFetch(IR::Block& block, IR::Inst& inst) { const IR::U1 is_scaled{ir.IsTextureScaled(ir.Imm32(info.descriptor_index))}; SubScaleCoord(ir, inst, is_scaled); // Scale ImageFetch offset - ScaleIntegerComposite(ir, inst, is_scaled, 2); + ScaleIntegerOffsetComposite(ir, inst, is_scaled, 2); } void SubScaleImageRead(IR::Block& block, IR::Inst& inst) { @@ -242,7 +270,7 @@ void PatchImageFetch(IR::Block& block, IR::Inst& inst) { const IR::U1 is_scaled{ir.IsTextureScaled(ir.Imm32(info.descriptor_index))}; ScaleIntegerComposite(ir, inst, is_scaled, 1); // Scale ImageFetch offset - ScaleIntegerComposite(ir, inst, is_scaled, 2); + ScaleIntegerOffsetComposite(ir, inst, is_scaled, 2); } void PatchImageRead(IR::Block& block, IR::Inst& inst) { diff --git a/src/shader_recompiler/ir_opt/ssa_rewrite_pass.cpp b/src/shader_recompiler/ir_opt/ssa_rewrite_pass.cpp index 87aa09358..d0b145860 100644 --- a/src/shader_recompiler/ir_opt/ssa_rewrite_pass.cpp +++ b/src/shader_recompiler/ir_opt/ssa_rewrite_pass.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later // This file implements the SSA rewriting algorithm proposed in // @@ -20,7 +19,6 @@ #include <vector> #include <boost/container/flat_map.hpp> -#include <boost/container/flat_set.hpp> #include "shader_recompiler/frontend/ir/basic_block.h" #include "shader_recompiler/frontend/ir/opcodes.h" diff --git a/src/shader_recompiler/ir_opt/texture_pass.cpp b/src/shader_recompiler/ir_opt/texture_pass.cpp index 96c997a58..e8be58357 100644 --- a/src/shader_recompiler/ir_opt/texture_pass.cpp +++ b/src/shader_recompiler/ir_opt/texture_pass.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <algorithm> #include <bit> @@ -20,8 +19,10 @@ namespace { struct ConstBufferAddr { u32 index; u32 offset; + u32 shift_left; u32 secondary_index; u32 secondary_offset; + u32 secondary_shift_left; IR::U32 dynamic_offset; u32 count; bool has_secondary; @@ -173,19 +174,41 @@ bool IsTextureInstruction(const IR::Inst& inst) { return IndexedInstruction(inst) != IR::Opcode::Void; } -std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst); +std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env); -std::optional<ConstBufferAddr> Track(const IR::Value& value) { - return IR::BreadthFirstSearch(value, TryGetConstBuffer); +std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env) { + return IR::BreadthFirstSearch( + value, [&env](const IR::Inst* inst) { return TryGetConstBuffer(inst, env); }); } -std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { +std::optional<u32> TryGetConstant(IR::Value& value, Environment& env) { + const IR::Inst* inst = value.InstRecursive(); + if (inst->GetOpcode() != IR::Opcode::GetCbufU32) { + return std::nullopt; + } + const IR::Value index{inst->Arg(0)}; + const IR::Value offset{inst->Arg(1)}; + if (!index.IsImmediate()) { + return std::nullopt; + } + if (!offset.IsImmediate()) { + return std::nullopt; + } + const auto index_number = index.U32(); + if (index_number != 1) { + return std::nullopt; + } + const auto offset_number = offset.U32(); + return env.ReadCbufValue(index_number, offset_number); +} + +std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env) { switch (inst->GetOpcode()) { default: return std::nullopt; case IR::Opcode::BitwiseOr32: { - std::optional lhs{Track(inst->Arg(0))}; - std::optional rhs{Track(inst->Arg(1))}; + std::optional lhs{Track(inst->Arg(0), env)}; + std::optional rhs{Track(inst->Arg(1), env)}; if (!lhs || !rhs) { return std::nullopt; } @@ -195,19 +218,62 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { if (lhs->count > 1 || rhs->count > 1) { return std::nullopt; } - if (lhs->index > rhs->index || lhs->offset > rhs->offset) { + if (lhs->shift_left > 0 || lhs->index > rhs->index || lhs->offset > rhs->offset) { std::swap(lhs, rhs); } return ConstBufferAddr{ .index = lhs->index, .offset = lhs->offset, + .shift_left = lhs->shift_left, .secondary_index = rhs->index, .secondary_offset = rhs->offset, + .secondary_shift_left = rhs->shift_left, .dynamic_offset = {}, .count = 1, .has_secondary = true, }; } + case IR::Opcode::ShiftLeftLogical32: { + const IR::Value shift{inst->Arg(1)}; + if (!shift.IsImmediate()) { + return std::nullopt; + } + std::optional lhs{Track(inst->Arg(0), env)}; + if (lhs) { + lhs->shift_left = shift.U32(); + } + return lhs; + break; + } + case IR::Opcode::BitwiseAnd32: { + IR::Value op1{inst->Arg(0)}; + IR::Value op2{inst->Arg(1)}; + if (op1.IsImmediate()) { + std::swap(op1, op2); + } + if (!op2.IsImmediate() && !op1.IsImmediate()) { + do { + auto try_index = TryGetConstant(op1, env); + if (try_index) { + op1 = op2; + op2 = IR::Value{*try_index}; + break; + } + auto try_index_2 = TryGetConstant(op2, env); + if (try_index_2) { + op2 = IR::Value{*try_index_2}; + break; + } + return std::nullopt; + } while (false); + } + std::optional lhs{Track(op1, env)}; + if (lhs) { + lhs->shift_left = static_cast<u32>(std::countr_zero(op2.U32())); + } + return lhs; + break; + } case IR::Opcode::GetCbufU32x2: case IR::Opcode::GetCbufU32: break; @@ -223,8 +289,10 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { return ConstBufferAddr{ .index = index.U32(), .offset = offset.U32(), + .shift_left = 0, .secondary_index = 0, .secondary_offset = 0, + .secondary_shift_left = 0, .dynamic_offset = {}, .count = 1, .has_secondary = false, @@ -248,8 +316,10 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { return ConstBufferAddr{ .index = index.U32(), .offset = base_offset, + .shift_left = 0, .secondary_index = 0, .secondary_offset = 0, + .secondary_shift_left = 0, .dynamic_offset = dynamic_offset, .count = 8, .has_secondary = false, @@ -259,7 +329,7 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst) { TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) { ConstBufferAddr addr; if (IsBindless(inst)) { - const std::optional<ConstBufferAddr> track_addr{Track(inst.Arg(0))}; + const std::optional<ConstBufferAddr> track_addr{Track(inst.Arg(0), env)}; if (!track_addr) { throw NotImplementedException("Failed to track bindless texture constant buffer"); } @@ -268,8 +338,10 @@ TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) { addr = ConstBufferAddr{ .index = env.TextureBoundBuffer(), .offset = inst.Arg(0).U32(), + .shift_left = 0, .secondary_index = 0, .secondary_offset = 0, + .secondary_shift_left = 0, .dynamic_offset = {}, .count = 1, .has_secondary = false, @@ -285,8 +357,9 @@ TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) { TextureType ReadTextureType(Environment& env, const ConstBufferAddr& cbuf) { const u32 secondary_index{cbuf.has_secondary ? cbuf.secondary_index : cbuf.index}; const u32 secondary_offset{cbuf.has_secondary ? cbuf.secondary_offset : cbuf.offset}; - const u32 lhs_raw{env.ReadCbufValue(cbuf.index, cbuf.offset)}; - const u32 rhs_raw{env.ReadCbufValue(secondary_index, secondary_offset)}; + const u32 lhs_raw{env.ReadCbufValue(cbuf.index, cbuf.offset) << cbuf.shift_left}; + const u32 rhs_raw{env.ReadCbufValue(secondary_index, secondary_offset) + << cbuf.secondary_shift_left}; return env.ReadTextureType(lhs_raw | rhs_raw); } @@ -363,6 +436,21 @@ private: TextureDescriptors& texture_descriptors; ImageDescriptors& image_descriptors; }; + +void PatchImageSampleImplicitLod(IR::Block& block, IR::Inst& inst) { + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + const auto info{inst.Flags<IR::TextureInstInfo>()}; + const IR::Value coord(inst.Arg(1)); + const IR::Value handle(ir.Imm32(0)); + const IR::U32 lod{ir.Imm32(0)}; + const IR::Value texture_size = ir.ImageQueryDimension(handle, lod, info); + inst.SetArg( + 1, ir.CompositeConstruct( + ir.FPMul(IR::F32(ir.CompositeExtract(coord, 0)), + ir.FPRecip(ir.ConvertUToF(32, 32, ir.CompositeExtract(texture_size, 0)))), + ir.FPMul(IR::F32(ir.CompositeExtract(coord, 1)), + ir.FPRecip(ir.ConvertUToF(32, 32, ir.CompositeExtract(texture_size, 1)))))); +} } // Anonymous namespace void TexturePass(Environment& env, IR::Program& program) { @@ -400,6 +488,14 @@ void TexturePass(Environment& env, IR::Program& program) { flags.type.Assign(ReadTextureType(env, cbuf)); inst->SetFlags(flags); break; + case IR::Opcode::ImageSampleImplicitLod: + if (flags.type != TextureType::Color2D) { + break; + } + if (ReadTextureType(env, cbuf) == TextureType::Color2DRect) { + PatchImageSampleImplicitLod(*texture_inst.block, *texture_inst.inst); + } + break; case IR::Opcode::ImageFetch: if (flags.type != TextureType::Color1D) { break; @@ -465,8 +561,10 @@ void TexturePass(Environment& env, IR::Program& program) { .has_secondary = cbuf.has_secondary, .cbuf_index = cbuf.index, .cbuf_offset = cbuf.offset, + .shift_left = cbuf.shift_left, .secondary_cbuf_index = cbuf.secondary_index, .secondary_cbuf_offset = cbuf.secondary_offset, + .secondary_shift_left = cbuf.secondary_shift_left, .count = cbuf.count, .size_shift = DESCRIPTOR_SIZE_SHIFT, }); @@ -477,8 +575,10 @@ void TexturePass(Environment& env, IR::Program& program) { .has_secondary = cbuf.has_secondary, .cbuf_index = cbuf.index, .cbuf_offset = cbuf.offset, + .shift_left = cbuf.shift_left, .secondary_cbuf_index = cbuf.secondary_index, .secondary_cbuf_offset = cbuf.secondary_offset, + .secondary_shift_left = cbuf.secondary_shift_left, .count = cbuf.count, .size_shift = DESCRIPTOR_SIZE_SHIFT, }); diff --git a/src/shader_recompiler/ir_opt/verification_pass.cpp b/src/shader_recompiler/ir_opt/verification_pass.cpp index 975d5aadf..f89f4ac28 100644 --- a/src/shader_recompiler/ir_opt/verification_pass.cpp +++ b/src/shader_recompiler/ir_opt/verification_pass.cpp @@ -1,6 +1,5 @@ -// Copyright 2021 yuzu Emulator Project -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. +// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later #include <map> #include <set> @@ -44,7 +43,7 @@ static void ValidateUses(const IR::Program& program) { } } } - for (const auto [inst, uses] : actual_uses) { + for (const auto& [inst, uses] : actual_uses) { if (inst->UseCount() != uses) { throw LogicError("Invalid uses in block: {}", IR::DumpProgram(program)); } |