Merge pull request #9694 from ameerj/txq-mips

shader_recompiler: TXQ: Skip QueryLevels when possible
This commit is contained in:
liamwhite 2023-01-29 12:27:26 -05:00 committed by GitHub
commit 208e635f37
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
11 changed files with 37 additions and 29 deletions

View File

@ -532,7 +532,7 @@ void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
} }
void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
ScalarS32 lod) { ScalarS32 lod, [[maybe_unused]] const IR::Value& skip_mips) {
const auto info{inst.Flags<IR::TextureInstInfo>()}; const auto info{inst.Flags<IR::TextureInstInfo>()};
const std::string texture{Texture(ctx, info, index)}; const std::string texture{Texture(ctx, info, index)};
const std::string_view type{TextureType(info)}; const std::string_view type{TextureType(info)};

View File

@ -581,7 +581,7 @@ void EmitImageGatherDref(EmitContext& ctx, IR::Inst& inst, const IR::Value& inde
void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
const IR::Value& coord, const IR::Value& offset, ScalarS32 lod, ScalarS32 ms); const IR::Value& coord, const IR::Value& offset, ScalarS32 lod, ScalarS32 ms);
void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
ScalarS32 lod); ScalarS32 lod, const IR::Value& skip_mips);
void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, Register coord); void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, Register coord);
void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
const IR::Value& coord, const IR::Value& derivatives, const IR::Value& coord, const IR::Value& derivatives,

View File

@ -460,27 +460,27 @@ void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
} }
void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view lod) { std::string_view lod, const IR::Value& skip_mips_val) {
const auto info{inst.Flags<IR::TextureInstInfo>()}; const auto info{inst.Flags<IR::TextureInstInfo>()};
const auto texture{Texture(ctx, info, index)}; const auto texture{Texture(ctx, info, index)};
const bool skip_mips{skip_mips_val.U1()};
const auto mips{
[&] { return skip_mips ? "0u" : fmt::format("uint(textureQueryLevels({}))", texture); }};
switch (info.type) { switch (info.type) {
case TextureType::Color1D: case TextureType::Color1D:
return ctx.AddU32x4( return ctx.AddU32x4("{}=uvec4(uint(textureSize({},int({}))),0u,0u,{});", inst, texture, lod,
"{}=uvec4(uint(textureSize({},int({}))),0u,0u,uint(textureQueryLevels({})));", inst, mips());
texture, lod, texture);
case TextureType::ColorArray1D: case TextureType::ColorArray1D:
case TextureType::Color2D: case TextureType::Color2D:
case TextureType::ColorCube: case TextureType::ColorCube:
case TextureType::Color2DRect: case TextureType::Color2DRect:
return ctx.AddU32x4( return ctx.AddU32x4("{}=uvec4(uvec2(textureSize({},int({}))),0u,{});", inst, texture, lod,
"{}=uvec4(uvec2(textureSize({},int({}))),0u,uint(textureQueryLevels({})));", inst, mips());
texture, lod, texture);
case TextureType::ColorArray2D: case TextureType::ColorArray2D:
case TextureType::Color3D: case TextureType::Color3D:
case TextureType::ColorArrayCube: case TextureType::ColorArrayCube:
return ctx.AddU32x4( return ctx.AddU32x4("{}=uvec4(uvec3(textureSize({},int({}))),{});", inst, texture, lod,
"{}=uvec4(uvec3(textureSize({},int({}))),uint(textureQueryLevels({})));", inst, texture, mips());
lod, texture);
case TextureType::Buffer: case TextureType::Buffer:
throw NotImplementedException("EmitImageQueryDimensions Texture buffers"); throw NotImplementedException("EmitImageQueryDimensions Texture buffers");
} }

View File

@ -654,7 +654,7 @@ void EmitImageFetch(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords, std::string_view offset, std::string_view lod, std::string_view coords, std::string_view offset, std::string_view lod,
std::string_view ms); std::string_view ms);
void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageQueryDimensions(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view lod); std::string_view lod, const IR::Value& skip_mips);
void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageQueryLod(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,
std::string_view coords); std::string_view coords);
void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index, void EmitImageGradient(EmitContext& ctx, IR::Inst& inst, const IR::Value& index,

View File

@ -445,11 +445,13 @@ Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id c
TextureImage(ctx, info, index), coords, operands.MaskOptional(), operands.Span()); TextureImage(ctx, info, index), coords, operands.MaskOptional(), operands.Span());
} }
Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod) { Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod,
const IR::Value& skip_mips_val) {
const auto info{inst->Flags<IR::TextureInstInfo>()}; const auto info{inst->Flags<IR::TextureInstInfo>()};
const Id image{TextureImage(ctx, info, index)}; const Id image{TextureImage(ctx, info, index)};
const Id zero{ctx.u32_zero_value}; const Id zero{ctx.u32_zero_value};
const auto mips{[&] { return ctx.OpImageQueryLevels(ctx.U32[1], image); }}; const bool skip_mips{skip_mips_val.U1()};
const auto mips{[&] { return skip_mips ? zero : ctx.OpImageQueryLevels(ctx.U32[1], image); }};
switch (info.type) { switch (info.type) {
case TextureType::Color1D: case TextureType::Color1D:
return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[1], image, lod), return ctx.OpCompositeConstruct(ctx.U32[4], ctx.OpImageQuerySizeLod(ctx.U32[1], image, lod),

View File

@ -539,7 +539,8 @@ Id EmitImageGatherDref(EmitContext& ctx, IR::Inst* inst, const IR::Value& index,
const IR::Value& offset, const IR::Value& offset2, Id dref); const IR::Value& offset, const IR::Value& offset2, Id dref);
Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, Id offset, Id EmitImageFetch(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, Id offset,
Id lod, Id ms); Id lod, Id ms);
Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod); Id EmitImageQueryDimensions(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id lod,
const IR::Value& skip_mips);
Id EmitImageQueryLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords); Id EmitImageQueryLod(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords);
Id EmitImageGradient(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords, Id EmitImageGradient(EmitContext& ctx, IR::Inst* inst, const IR::Value& index, Id coords,
Id derivates, Id offset, Id lod_clamp); Id derivates, Id offset, Id lod_clamp);

View File

@ -1846,15 +1846,16 @@ Value IREmitter::ImageFetch(const Value& handle, const Value& coords, const Valu
return Inst(op, Flags{info}, handle, coords, offset, lod, multisampling); return Inst(op, Flags{info}, handle, coords, offset, lod, multisampling);
} }
Value IREmitter::ImageQueryDimension(const Value& handle, const IR::U32& lod) { Value IREmitter::ImageQueryDimension(const Value& handle, const IR::U32& lod,
const IR::U1& skip_mips) {
const Opcode op{handle.IsImmediate() ? Opcode::BoundImageQueryDimensions const Opcode op{handle.IsImmediate() ? Opcode::BoundImageQueryDimensions
: Opcode::BindlessImageQueryDimensions}; : Opcode::BindlessImageQueryDimensions};
return Inst(op, handle, lod); return Inst(op, handle, lod, skip_mips);
} }
Value IREmitter::ImageQueryDimension(const Value& handle, const IR::U32& lod, Value IREmitter::ImageQueryDimension(const Value& handle, const IR::U32& lod,
TextureInstInfo info) { const IR::U1& skip_mips, TextureInstInfo info) {
return Inst(Opcode::ImageQueryDimensions, Flags{info}, handle, lod); return Inst(Opcode::ImageQueryDimensions, Flags{info}, handle, lod, skip_mips);
} }
Value IREmitter::ImageQueryLod(const Value& handle, const Value& coords, TextureInstInfo info) { Value IREmitter::ImageQueryLod(const Value& handle, const Value& coords, TextureInstInfo info) {

View File

@ -320,9 +320,10 @@ public:
[[nodiscard]] F32 ImageSampleDrefExplicitLod(const Value& handle, const Value& coords, [[nodiscard]] F32 ImageSampleDrefExplicitLod(const Value& handle, const Value& coords,
const F32& dref, const F32& lod, const F32& dref, const F32& lod,
const Value& offset, TextureInstInfo info); const Value& offset, TextureInstInfo info);
[[nodiscard]] Value ImageQueryDimension(const Value& handle, const IR::U32& lod);
[[nodiscard]] Value ImageQueryDimension(const Value& handle, const IR::U32& lod, [[nodiscard]] Value ImageQueryDimension(const Value& handle, const IR::U32& lod,
TextureInstInfo info); const IR::U1& skip_mips);
[[nodiscard]] Value ImageQueryDimension(const Value& handle, const IR::U32& lod,
const IR::U1& skip_mips, TextureInstInfo info);
[[nodiscard]] Value ImageQueryLod(const Value& handle, const Value& coords, [[nodiscard]] Value ImageQueryLod(const Value& handle, const Value& coords,
TextureInstInfo info); TextureInstInfo info);

View File

@ -482,7 +482,7 @@ OPCODE(BindlessImageSampleDrefExplicitLod, F32, U32,
OPCODE(BindlessImageGather, F32x4, U32, Opaque, Opaque, Opaque, ) OPCODE(BindlessImageGather, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BindlessImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, ) OPCODE(BindlessImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, )
OPCODE(BindlessImageFetch, F32x4, U32, Opaque, Opaque, U32, Opaque, ) OPCODE(BindlessImageFetch, F32x4, U32, Opaque, Opaque, U32, Opaque, )
OPCODE(BindlessImageQueryDimensions, U32x4, U32, U32, ) OPCODE(BindlessImageQueryDimensions, U32x4, U32, U32, U1, )
OPCODE(BindlessImageQueryLod, F32x4, U32, Opaque, ) OPCODE(BindlessImageQueryLod, F32x4, U32, Opaque, )
OPCODE(BindlessImageGradient, F32x4, U32, Opaque, Opaque, Opaque, Opaque, ) OPCODE(BindlessImageGradient, F32x4, U32, Opaque, Opaque, Opaque, Opaque, )
OPCODE(BindlessImageRead, U32x4, U32, Opaque, ) OPCODE(BindlessImageRead, U32x4, U32, Opaque, )
@ -495,7 +495,7 @@ OPCODE(BoundImageSampleDrefExplicitLod, F32, U32,
OPCODE(BoundImageGather, F32x4, U32, Opaque, Opaque, Opaque, ) OPCODE(BoundImageGather, F32x4, U32, Opaque, Opaque, Opaque, )
OPCODE(BoundImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, ) OPCODE(BoundImageGatherDref, F32x4, U32, Opaque, Opaque, Opaque, F32, )
OPCODE(BoundImageFetch, F32x4, U32, Opaque, Opaque, U32, Opaque, ) OPCODE(BoundImageFetch, F32x4, U32, Opaque, Opaque, U32, Opaque, )
OPCODE(BoundImageQueryDimensions, U32x4, U32, U32, ) OPCODE(BoundImageQueryDimensions, U32x4, U32, U32, U1, )
OPCODE(BoundImageQueryLod, F32x4, U32, Opaque, ) OPCODE(BoundImageQueryLod, F32x4, U32, Opaque, )
OPCODE(BoundImageGradient, F32x4, U32, Opaque, Opaque, Opaque, Opaque, ) OPCODE(BoundImageGradient, F32x4, U32, Opaque, Opaque, Opaque, Opaque, )
OPCODE(BoundImageRead, U32x4, U32, Opaque, ) OPCODE(BoundImageRead, U32x4, U32, Opaque, )
@ -508,7 +508,7 @@ OPCODE(ImageSampleDrefExplicitLod, F32, Opaq
OPCODE(ImageGather, F32x4, Opaque, Opaque, Opaque, Opaque, ) OPCODE(ImageGather, F32x4, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageGatherDref, F32x4, Opaque, Opaque, Opaque, Opaque, F32, ) OPCODE(ImageGatherDref, F32x4, Opaque, Opaque, Opaque, Opaque, F32, )
OPCODE(ImageFetch, F32x4, Opaque, Opaque, Opaque, U32, Opaque, ) OPCODE(ImageFetch, F32x4, Opaque, Opaque, Opaque, U32, Opaque, )
OPCODE(ImageQueryDimensions, U32x4, Opaque, U32, ) OPCODE(ImageQueryDimensions, U32x4, Opaque, U32, U1, )
OPCODE(ImageQueryLod, F32x4, Opaque, Opaque, ) OPCODE(ImageQueryLod, F32x4, Opaque, Opaque, )
OPCODE(ImageGradient, F32x4, Opaque, Opaque, Opaque, Opaque, Opaque, ) OPCODE(ImageGradient, F32x4, Opaque, Opaque, Opaque, Opaque, Opaque, )
OPCODE(ImageRead, U32x4, Opaque, Opaque, ) OPCODE(ImageRead, U32x4, Opaque, Opaque, )

View File

@ -15,11 +15,13 @@ enum class Mode : u64 {
SamplePos = 5, SamplePos = 5,
}; };
IR::Value Query(TranslatorVisitor& v, const IR::U32& handle, Mode mode, IR::Reg src_reg) { IR::Value Query(TranslatorVisitor& v, const IR::U32& handle, Mode mode, IR::Reg src_reg, u64 mask) {
switch (mode) { switch (mode) {
case Mode::Dimension: { case Mode::Dimension: {
const bool needs_num_mips{((mask >> 3) & 1) != 0};
const IR::U1 skip_mips{v.ir.Imm1(!needs_num_mips)};
const IR::U32 lod{v.X(src_reg)}; const IR::U32 lod{v.X(src_reg)};
return v.ir.ImageQueryDimension(handle, lod); return v.ir.ImageQueryDimension(handle, lod, skip_mips);
} }
case Mode::TextureType: case Mode::TextureType:
case Mode::SamplePos: case Mode::SamplePos:
@ -46,7 +48,7 @@ void Impl(TranslatorVisitor& v, u64 insn, std::optional<u32> cbuf_offset) {
handle = v.X(src_reg); handle = v.X(src_reg);
++src_reg; ++src_reg;
} }
const IR::Value query{Query(v, handle, txq.mode, src_reg)}; const IR::Value query{Query(v, handle, txq.mode, src_reg, txq.mask)};
IR::Reg dest_reg{txq.dest_reg}; IR::Reg dest_reg{txq.dest_reg};
for (int element = 0; element < 4; ++element) { for (int element = 0; element < 4; ++element) {
if (((txq.mask >> element) & 1) == 0) { if (((txq.mask >> element) & 1) == 0) {

View File

@ -452,7 +452,8 @@ void PatchImageSampleImplicitLod(IR::Block& block, IR::Inst& inst) {
const IR::Value coord(inst.Arg(1)); const IR::Value coord(inst.Arg(1));
const IR::Value handle(ir.Imm32(0)); const IR::Value handle(ir.Imm32(0));
const IR::U32 lod{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( inst.SetArg(
1, ir.CompositeConstruct( 1, ir.CompositeConstruct(
ir.FPMul(IR::F32(ir.CompositeExtract(coord, 0)), ir.FPMul(IR::F32(ir.CompositeExtract(coord, 0)),