diff options
Diffstat (limited to 'src/shader_recompiler/ir_opt')
4 files changed, 114 insertions, 23 deletions
diff --git a/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp b/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp index 826f9a54a..4d81e9336 100644 --- a/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp +++ b/src/shader_recompiler/ir_opt/constant_propagation_pass.cpp @@ -7,6 +7,7 @@ #include <type_traits> #include "common/bit_cast.h" +#include "shader_recompiler/environment.h" #include "shader_recompiler/exception.h" #include "shader_recompiler/frontend/ir/ir_emitter.h" #include "shader_recompiler/frontend/ir/value.h" @@ -515,6 +516,9 @@ void FoldBitCast(IR::Inst& inst, IR::Opcode reverse) { case IR::Attribute::PrimitiveId: case IR::Attribute::InstanceId: case IR::Attribute::VertexId: + case IR::Attribute::BaseVertex: + case IR::Attribute::BaseInstance: + case IR::Attribute::DrawID: break; default: return; @@ -644,7 +648,63 @@ void FoldFSwizzleAdd(IR::Block& block, IR::Inst& inst) { } } -void ConstantPropagation(IR::Block& block, IR::Inst& inst) { +void FoldConstBuffer(Environment& env, IR::Block& block, IR::Inst& inst) { + const IR::Value bank{inst.Arg(0)}; + const IR::Value offset{inst.Arg(1)}; + if (!bank.IsImmediate() || !offset.IsImmediate()) { + return; + } + const auto bank_value = bank.U32(); + const auto offset_value = offset.U32(); + auto replacement = env.GetReplaceConstBuffer(bank_value, offset_value); + if (!replacement) { + return; + } + const auto new_attribute = [replacement]() { + switch (*replacement) { + case ReplaceConstant::BaseInstance: + return IR::Attribute::BaseInstance; + case ReplaceConstant::BaseVertex: + return IR::Attribute::BaseVertex; + case ReplaceConstant::DrawID: + return IR::Attribute::DrawID; + default: + throw NotImplementedException("Not implemented replacement variable {}", *replacement); + } + }(); + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + if (inst.GetOpcode() == IR::Opcode::GetCbufU32) { + inst.ReplaceUsesWith(ir.GetAttributeU32(new_attribute)); + } else { + inst.ReplaceUsesWith(ir.GetAttribute(new_attribute)); + } +} + +void FoldDriverConstBuffer(Environment& env, IR::Block& block, IR::Inst& inst, u32 which_bank, + u32 offset_start = 0, u32 offset_end = std::numeric_limits<u16>::max()) { + const IR::Value bank{inst.Arg(0)}; + const IR::Value offset{inst.Arg(1)}; + if (!bank.IsImmediate() || !offset.IsImmediate()) { + return; + } + const auto bank_value = bank.U32(); + if (bank_value != which_bank) { + return; + } + const auto offset_value = offset.U32(); + if (offset_value < offset_start || offset_value >= offset_end) { + return; + } + IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; + if (inst.GetOpcode() == IR::Opcode::GetCbufU32) { + inst.ReplaceUsesWith(IR::Value{env.ReadCbufValue(bank_value, offset_value)}); + } else { + inst.ReplaceUsesWith( + IR::Value{Common::BitCast<f32>(env.ReadCbufValue(bank_value, offset_value))}); + } +} + +void ConstantPropagation(Environment& env, IR::Block& block, IR::Inst& inst) { switch (inst.GetOpcode()) { case IR::Opcode::GetRegister: return FoldGetRegister(inst); @@ -789,18 +849,28 @@ void ConstantPropagation(IR::Block& block, IR::Inst& inst) { IR::Opcode::CompositeInsertF16x4); case IR::Opcode::FSwizzleAdd: return FoldFSwizzleAdd(block, inst); + case IR::Opcode::GetCbufF32: + case IR::Opcode::GetCbufU32: + if (env.HasHLEMacroState()) { + FoldConstBuffer(env, block, inst); + } + if (env.IsPropietaryDriver()) { + FoldDriverConstBuffer(env, block, inst, 1); + } + break; default: break; } } + } // Anonymous namespace -void ConstantPropagationPass(IR::Program& program) { +void ConstantPropagationPass(Environment& env, IR::Program& program) { const auto end{program.post_order_blocks.rend()}; for (auto it = program.post_order_blocks.rbegin(); it != end; ++it) { IR::Block* const block{*it}; for (IR::Inst& inst : block->Instructions()) { - ConstantPropagation(*block, inst); + ConstantPropagation(env, *block, inst); } } } 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 336338e62..9101722ba 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 @@ -11,6 +11,7 @@ #include "shader_recompiler/frontend/ir/breadth_first_search.h" #include "shader_recompiler/frontend/ir/ir_emitter.h" #include "shader_recompiler/frontend/ir/value.h" +#include "shader_recompiler/host_translate_info.h" #include "shader_recompiler/ir_opt/passes.h" namespace Shader::Optimization { @@ -402,7 +403,7 @@ void CollectStorageBuffers(IR::Block& block, IR::Inst& inst, StorageInfo& info) } /// Returns the offset in indices (not bytes) for an equivalent storage instruction -IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer) { +IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer, u32 alignment) { IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)}; IR::U32 offset; if (const std::optional<LowAddrInfo> low_addr{TrackLowAddress(&inst)}) { @@ -415,7 +416,10 @@ IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer } // Subtract the least significant 32 bits from the guest offset. The result is the storage // buffer offset in bytes. - const IR::U32 low_cbuf{ir.GetCbuf(ir.Imm32(buffer.index), ir.Imm32(buffer.offset))}; + IR::U32 low_cbuf{ir.GetCbuf(ir.Imm32(buffer.index), ir.Imm32(buffer.offset))}; + + // Align the offset base to match the host alignment requirements + low_cbuf = ir.BitwiseAnd(low_cbuf, ir.Imm32(~(alignment - 1U))); return ir.ISub(offset, low_cbuf); } @@ -510,7 +514,7 @@ void Replace(IR::Block& block, IR::Inst& inst, const IR::U32& storage_index, } } // Anonymous namespace -void GlobalMemoryToStorageBufferPass(IR::Program& program) { +void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info) { StorageInfo info; for (IR::Block* const block : program.post_order_blocks) { for (IR::Inst& inst : block->Instructions()) { @@ -534,7 +538,8 @@ void GlobalMemoryToStorageBufferPass(IR::Program& program) { const IR::U32 index{IR::Value{static_cast<u32>(info.set.index_of(it))}}; IR::Block* const block{storage_inst.block}; IR::Inst* const inst{storage_inst.inst}; - const IR::U32 offset{StorageOffset(*block, *inst, storage_buffer)}; + const IR::U32 offset{ + StorageOffset(*block, *inst, storage_buffer, host_info.min_ssbo_alignment)}; Replace(*block, *inst, index, offset); } } diff --git a/src/shader_recompiler/ir_opt/passes.h b/src/shader_recompiler/ir_opt/passes.h index 11bfe801a..4ffad1172 100644 --- a/src/shader_recompiler/ir_opt/passes.h +++ b/src/shader_recompiler/ir_opt/passes.h @@ -13,9 +13,9 @@ struct HostTranslateInfo; namespace Shader::Optimization { void CollectShaderInfoPass(Environment& env, IR::Program& program); -void ConstantPropagationPass(IR::Program& program); +void ConstantPropagationPass(Environment& env, IR::Program& program); void DeadCodeEliminationPass(IR::Program& program); -void GlobalMemoryToStorageBufferPass(IR::Program& program); +void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info); void IdentityRemovalPass(IR::Program& program); void LowerFp16ToFp32(IR::Program& program); void LowerInt64ToInt32(IR::Program& program); diff --git a/src/shader_recompiler/ir_opt/texture_pass.cpp b/src/shader_recompiler/ir_opt/texture_pass.cpp index f5c86fcb1..d374c976a 100644 --- a/src/shader_recompiler/ir_opt/texture_pass.cpp +++ b/src/shader_recompiler/ir_opt/texture_pass.cpp @@ -355,21 +355,21 @@ TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) { }; } -TextureType ReadTextureType(Environment& env, const ConstBufferAddr& cbuf) { +u32 GetTextureHandle(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) << cbuf.shift_left}; const u32 rhs_raw{env.ReadCbufValue(secondary_index, secondary_offset) << cbuf.secondary_shift_left}; - return env.ReadTextureType(lhs_raw | rhs_raw); + return lhs_raw | rhs_raw; +} + +TextureType ReadTextureType(Environment& env, const ConstBufferAddr& cbuf) { + return env.ReadTextureType(GetTextureHandle(env, cbuf)); } TexturePixelFormat ReadTexturePixelFormat(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)}; - return env.ReadTexturePixelFormat(lhs_raw | rhs_raw); + return env.ReadTexturePixelFormat(GetTextureHandle(env, cbuf)); } class Descriptors { @@ -386,8 +386,10 @@ public: return Add(texture_buffer_descriptors, desc, [&desc](const auto& existing) { return desc.cbuf_index == existing.cbuf_index && desc.cbuf_offset == existing.cbuf_offset && + desc.shift_left == existing.shift_left && desc.secondary_cbuf_index == existing.secondary_cbuf_index && desc.secondary_cbuf_offset == existing.secondary_cbuf_offset && + desc.secondary_shift_left == existing.secondary_shift_left && desc.count == existing.count && desc.size_shift == existing.size_shift && desc.has_secondary == existing.has_secondary; }); @@ -405,15 +407,20 @@ public: } u32 Add(const TextureDescriptor& desc) { - return Add(texture_descriptors, desc, [&desc](const auto& existing) { + const u32 index{Add(texture_descriptors, desc, [&desc](const auto& existing) { return desc.type == existing.type && desc.is_depth == existing.is_depth && desc.has_secondary == existing.has_secondary && desc.cbuf_index == existing.cbuf_index && desc.cbuf_offset == existing.cbuf_offset && + desc.shift_left == existing.shift_left && desc.secondary_cbuf_index == existing.secondary_cbuf_index && desc.secondary_cbuf_offset == existing.secondary_cbuf_offset && + desc.secondary_shift_left == existing.secondary_shift_left && desc.count == existing.count && desc.size_shift == existing.size_shift; - }); + })}; + // TODO: Read this from TIC + texture_descriptors[index].is_multisample |= desc.is_multisample; + return index; } u32 Add(const ImageDescriptor& desc) { @@ -452,7 +459,8 @@ void PatchImageSampleImplicitLod(IR::Block& block, IR::Inst& inst) { 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); + const IR::U1 skip_mips{ir.Imm1(true)}; + const IR::Value texture_size = ir.ImageQueryDimension(handle, lod, skip_mips, info); inst.SetArg( 1, ir.CompositeConstruct( ir.FPMul(IR::F32(ir.CompositeExtract(coord, 0)), @@ -486,10 +494,10 @@ void PatchTexelFetch(IR::Block& block, IR::Inst& inst, TexturePixelFormat pixel_ const IR::F32 w(ir.CompositeExtract(new_inst, 3)); const IR::F16F32F64 max_value(ir.Imm32(get_max_value())); const IR::Value converted = - ir.CompositeConstruct(ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::S32>(x)), max_value), - ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::S32>(y)), max_value), - ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::S32>(z)), max_value), - ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::S32>(w)), max_value)); + ir.CompositeConstruct(ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(x)), max_value), + ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(y)), max_value), + ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(z)), max_value), + ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(w)), max_value)); inst.ReplaceUsesWith(converted); } } // Anonymous namespace @@ -524,6 +532,7 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo const auto& cbuf{texture_inst.cbuf}; auto flags{inst->Flags<IR::TextureInstInfo>()}; + bool is_multisample{false}; switch (inst->GetOpcode()) { case IR::Opcode::ImageQueryDimensions: flags.type.Assign(ReadTextureType(env, cbuf)); @@ -538,6 +547,12 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo } break; case IR::Opcode::ImageFetch: + if (flags.type == TextureType::Color2D || flags.type == TextureType::Color2DRect || + flags.type == TextureType::ColorArray2D) { + is_multisample = !inst->Arg(4).IsEmpty(); + } else { + inst->SetArg(4, IR::U32{}); + } if (flags.type != TextureType::Color1D) { break; } @@ -613,6 +628,7 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo index = descriptors.Add(TextureDescriptor{ .type = flags.type, .is_depth = flags.is_depth != 0, + .is_multisample = is_multisample, .has_secondary = cbuf.has_secondary, .cbuf_index = cbuf.index, .cbuf_offset = cbuf.offset, |