3
0
Fork 0
forked from suyu/suyu

shader: Implement TLD

This commit is contained in:
FernandoS27 2021-03-26 20:51:05 +01:00 committed by ameerj
parent 7a1c14269e
commit dc1a9a3bed
8 changed files with 174 additions and 16 deletions

View file

@ -133,6 +133,7 @@ add_library(shader_recompiler STATIC
frontend/maxwell/translate/impl/texture_fetch_swizzled.cpp
frontend/maxwell/translate/impl/texture_gather_swizzled.cpp
frontend/maxwell/translate/impl/texture_gather.cpp
frontend/maxwell/translate/impl/texture_load.cpp
frontend/maxwell/translate/impl/texture_query.cpp
frontend/maxwell/translate/impl/video_helper.cpp
frontend/maxwell/translate/impl/video_helper.h

View file

@ -254,7 +254,7 @@ Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id c
const auto info{inst->Flags<IR::TextureInstInfo>()};
const ImageOperands operands(offset, lod, ms);
return Emit(&EmitContext::OpImageSparseFetch, &EmitContext::OpImageFetch, ctx, inst, ctx.F32[4],
Texture(ctx, index), coords, operands.Mask(), operands.Span());
TextureImage(ctx, index), coords, operands.Mask(), operands.Span());
}
Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod) {

View file

@ -378,7 +378,7 @@ OPCODE(BindlessImageSampleDrefImplicitLod, F32, U32,
OPCODE(BindlessImageSampleDrefExplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, )
OPCODE(BindlessImageGather, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BindlessImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, )
OPCODE(BindlessImageFetch, F32x4, U32, Opaque, U32, U32, )
OPCODE(BindlessImageFetch, F32x4, U32, Opaque, Opaque, Opaque, Opaque, )
OPCODE(BindlessImageQueryDimensions, U32x4, U32, U32, )
OPCODE(BoundImageSampleImplicitLod, F32x4, U32, Opaque, Opaque, Opaque, )
@ -387,7 +387,7 @@ OPCODE(BoundImageSampleDrefImplicitLod, F32, U32,
OPCODE(BoundImageSampleDrefExplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, )
OPCODE(BoundImageGather, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BoundImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, )
OPCODE(BoundImageFetch, F32x4, U32, Opaque, U32, U32, )
OPCODE(BoundImageFetch, F32x4, U32, Opaque, Opaque, Opaque, Opaque, )
OPCODE(BoundImageQueryDimensions, U32x4, U32, U32, )
OPCODE(ImageSampleImplicitLod, F32x4, U32, Opaque, Opaque, Opaque, )
@ -396,7 +396,7 @@ OPCODE(ImageSampleDrefImplicitLod, F32, U32,
OPCODE(ImageSampleDrefExplicitLod, F32, U32, Opaque, F32, Opaque, Opaque, )
OPCODE(ImageGather, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(ImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, )
OPCODE(ImageFetch, F32x4, U32, Opaque, U32, U32, )
OPCODE(ImageFetch, F32x4, U32, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageQueryDimensions, U32x4, U32, U32, )
// Warp operations

View file

@ -252,8 +252,8 @@ INST(SYNC, "SYNC", "1111 0000 1111 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---")
INST(TLD, "TLD", "1101 1100 ---- ----")
INST(TLD_b, "TLD (b)", "1101 1101 ---- ----")
INST(TLD4, "TLD4", "1100 10-- ---- ----")
INST(TLD4_b, "TLD4 (b)", "1101 1110 11-- ----")
INST(TLD4S, "TLD4S", "1101 1111 -0-- ----")

View file

@ -313,14 +313,6 @@ void TranslatorVisitor::SYNC(u64) {
ThrowNotImplemented(Opcode::SYNC);
}
void TranslatorVisitor::TLD(u64) {
ThrowNotImplemented(Opcode::TLD);
}
void TranslatorVisitor::TLD_b(u64) {
ThrowNotImplemented(Opcode::TLD_b);
}
void TranslatorVisitor::TLDS(u64) {
ThrowNotImplemented(Opcode::TLDS);
}

View file

@ -0,0 +1,165 @@
// 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 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::U32 { return v.ir.BitFieldExtract(v.X(reg), v.ir.Imm32(0), v.ir.Imm32(16)); }};
switch (type) {
case TextureType::_1D:
return v.X(reg);
case TextureType::ARRAY_1D:
return v.ir.CompositeConstruct(v.X(reg + 1), read_array());
case TextureType::_2D:
return v.ir.CompositeConstruct(v.X(reg), v.X(reg + 1));
case TextureType::ARRAY_2D:
return v.ir.CompositeConstruct(v.X(reg + 1), v.X(reg + 2), read_array());
case TextureType::_3D:
return v.ir.CompositeConstruct(v.X(reg), v.X(reg + 1), v.X(reg + 2));
case TextureType::ARRAY_3D:
throw NotImplementedException("3D array texture type");
case TextureType::CUBE:
return v.ir.CompositeConstruct(v.X(reg), v.X(reg + 1), v.X(reg + 2));
case TextureType::ARRAY_CUBE:
return v.ir.CompositeConstruct(v.X(reg + 1), v.X(reg + 2), v.X(reg + 3), read_array());
}
throw NotImplementedException("Invalid texture type {}", type);
}
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), true);
case TextureType::_2D:
case TextureType::ARRAY_2D:
return v.ir.CompositeConstruct(
v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(4), true),
v.ir.BitFieldExtract(value, v.ir.Imm32(4), v.ir.Imm32(4), true));
case TextureType::_3D:
case TextureType::ARRAY_3D:
return v.ir.CompositeConstruct(
v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(4), true),
v.ir.BitFieldExtract(value, v.ir.Imm32(4), v.ir.Imm32(4), true),
v.ir.BitFieldExtract(value, v.ir.Imm32(8), v.ir.Imm32(4), true));
case TextureType::CUBE:
case TextureType::ARRAY_CUBE:
throw NotImplementedException("Illegal offset on CUBE sample");
}
throw NotImplementedException("Invalid texture type {}", type);
}
void Impl(TranslatorVisitor& v, u64 insn, bool is_bindless) {
union {
u64 raw;
BitField<49, 1, u64> nodep;
BitField<55, 1, u64> lod;
BitField<50, 1, u64> multisample;
BitField<35, 1, u64> aoffi;
BitField<54, 1, u64> clamp;
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;
BitField<36, 13, u64> cbuf_offset;
} const tld{insn};
const IR::Value coords{MakeCoords(v, tld.coord_reg, tld.type)};
IR::Reg meta_reg{tld.meta_reg};
IR::Value handle;
IR::Value offset;
IR::U32 lod;
IR::U32 multisample;
if (!is_bindless) {
handle = v.ir.Imm32(static_cast<u32>(tld.cbuf_offset.Value() * 4));
} else {
handle = v.X(meta_reg++);
}
if (tld.lod != 0) {
lod = v.X(meta_reg++);
}
if (tld.aoffi != 0) {
offset = MakeOffset(v, meta_reg, tld.type);
}
if (tld.multisample != 0) {
multisample = v.X(meta_reg++);
}
if (tld.clamp != 0) {
throw NotImplementedException("TLD.CL - CLAMP is not implmented");
}
IR::TextureInstInfo info{};
info.type.Assign(GetType(tld.type, false));
const IR::Value sample{[&]() -> IR::Value {
return v.ir.ImageFetch(handle, coords, offset, lod, multisample, info);
}()};
IR::Reg dest_reg{tld.dest_reg};
for (size_t element = 0; element < 4; ++element) {
if (((tld.mask >> element) & 1) == 0) {
continue;
}
v.F(dest_reg, IR::F32{v.ir.CompositeExtract(sample, element)});
++dest_reg;
}
if (tld.sparse_pred != IR::Pred::PT) {
v.ir.SetPred(tld.sparse_pred, v.ir.LogicalNot(v.ir.GetSparseFromOp(sample)));
}
}
} // Anonymous namespace
void TranslatorVisitor::TLD(u64 insn) {
Impl(*this, insn, false);
}
void TranslatorVisitor::TLD_b(u64 insn) {
Impl(*this, insn, true);
}
} // namespace Shader::Maxwell

View file

@ -382,6 +382,7 @@ void VisitUsages(Info& info, IR::Inst& inst) {
case IR::Opcode::ImageSampleDrefExplicitLod:
case IR::Opcode::ImageGather:
case IR::Opcode::ImageGatherDref:
case IR::Opcode::ImageFetch:
case IR::Opcode::ImageQueryDimensions: {
const TextureType type{inst.Flags<IR::TextureInstInfo>().type};
info.uses_sampled_1d |= type == TextureType::Color1D || type == TextureType::ColorArray1D ||

View file

@ -64,12 +64,11 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
}
const auto it = std::ranges::lower_bound(map_ranges, gpu_addr, {}, &MapRange::first);
if (it != map_ranges.end()) {
ASSERT(it->first == gpu_addr);
// ASSERT(it->first == gpu_addr);
map_ranges.erase(it);
} else {
UNREACHABLE_MSG("Unmapping non-existent GPU address=0x{:x}", gpu_addr);
}
const auto submapped_ranges = GetSubmappedRange(gpu_addr, size);
for (const auto& map : submapped_ranges) {