hipnosuer
/
Zuma
forked from ShuriZma/suyu
1
0
Fork 0
Code Pull Requests Activity

shader: Add coarse derivatives

This commit is contained in:
FernandoS27 2021-04-17 12:51:43 +02:00 committed by ameerj
parent 04c459fc8d
commit 080857b60e
7 changed files with 28 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/shader_recompiler/backend/spirv/emit_spirv.h
View File

@ -533,4 +533,8 @@ Id EmitDPdxFine(EmitContext& ctx, Id op_a);
Id EmitDPdyFine(EmitContext& ctx, Id op_a); Id EmitDPdyFine(EmitContext& ctx, Id op_a);
Id EmitDPdxCoarse(EmitContext& ctx, Id op_a);
Id EmitDPdyCoarse(EmitContext& ctx, Id op_a);
} // namespace Shader::Backend::SPIRV } // namespace Shader::Backend::SPIRV

8
src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp
View File

@ -191,4 +191,12 @@ Id EmitDPdyFine(EmitContext& ctx, Id op_a) {
return ctx.OpDPdyFine(ctx.F32[1], op_a); return ctx.OpDPdyFine(ctx.F32[1], op_a);
} }
Id EmitDPdxCoarse(EmitContext& ctx, Id op_a) {
return ctx.OpDPdxCoarse(ctx.F32[1], op_a);
}
Id EmitDPdyCoarse(EmitContext& ctx, Id op_a) {
return ctx.OpDPdyCoarse(ctx.F32[1], op_a);
}
} // namespace Shader::Backend::SPIRV } // namespace Shader::Backend::SPIRV

8
src/shader_recompiler/frontend/ir/ir_emitter.cpp
View File

@ -1933,4 +1933,12 @@ F32 IREmitter::DPdyFine(const F32& a) {
return Inst<F32>(Opcode::DPdyFine, a); return Inst<F32>(Opcode::DPdyFine, a);
} }
F32 IREmitter::DPdxCoarse(const F32& a) {
return Inst<F32>(Opcode::DPdxCoarse, a);
}
F32 IREmitter::DPdyCoarse(const F32& a) {
return Inst<F32>(Opcode::DPdyCoarse, a);
}
} // namespace Shader::IR } // namespace Shader::IR

4
src/shader_recompiler/frontend/ir/ir_emitter.h
View File

@ -357,6 +357,10 @@ public:
[[nodiscard]] F32 DPdyFine(const F32& a); [[nodiscard]] F32 DPdyFine(const F32& a);
[[nodiscard]] F32 DPdxCoarse(const F32& a);
[[nodiscard]] F32 DPdyCoarse(const F32& a);
private: private:
IR::Block::iterator insertion_point; IR::Block::iterator insertion_point;

2
src/shader_recompiler/frontend/ir/opcodes.inc
View File

@ -513,3 +513,5 @@ OPCODE(ShuffleButterfly, U32, U32,
OPCODE(FSwizzleAdd, F32, F32, F32, U32, ) OPCODE(FSwizzleAdd, F32, F32, F32, U32, )
OPCODE(DPdxFine, F32, F32, ) OPCODE(DPdxFine, F32, F32, )
OPCODE(DPdyFine, F32, F32, ) OPCODE(DPdyFine, F32, F32, )
OPCODE(DPdxCoarse, F32, F32, )
OPCODE(DPdyCoarse, F32, F32, )

2
src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp
View File

@ -532,6 +532,8 @@ void VisitUsages(Info& info, IR::Inst& inst) {
break; break;
case IR::Opcode::DPdxFine: case IR::Opcode::DPdxFine:
case IR::Opcode::DPdyFine: case IR::Opcode::DPdyFine:
case IR::Opcode::DPdxCoarse:
case IR::Opcode::DPdyCoarse:
info.uses_derivates = true; info.uses_derivates = true;
break; break;
case IR::Opcode::LoadStorageU8: case IR::Opcode::LoadStorageU8:

8
src/shader_recompiler/ir_opt/constant_propagation_pass.cpp
View File

@ -428,19 +428,15 @@ void FoldFSwizzleAdd(IR::Block& block, IR::Inst& inst) {
if (!swizzle.IsImmediate()) { if (!swizzle.IsImmediate()) {
return; return;
} }
const IR::Value value_1{GetThroughCast(inst.Arg(0).Resolve(), IR::Opcode::BitCastF32U32)}; const IR::Value value_1{GetThroughCast(inst.Arg(0).Resolve(), IR::Opcode::BitCastF32U32)};
const IR::Value value_2{GetThroughCast(inst.Arg(1).Resolve(), IR::Opcode::BitCastF32U32)}; const IR::Value value_2{GetThroughCast(inst.Arg(1).Resolve(), IR::Opcode::BitCastF32U32)};
if (value_1.IsImmediate()) { if (value_1.IsImmediate()) {
return; return;
} }
const u32 swizzle_value{swizzle.U32()}; const u32 swizzle_value{swizzle.U32()};
if (swizzle_value != 0x99 && swizzle_value != 0xA5) { if (swizzle_value != 0x99 && swizzle_value != 0xA5) {
return; return;
} }
IR::Inst* const inst2{value_1.InstRecursive()}; IR::Inst* const inst2{value_1.InstRecursive()};
if (inst2->GetOpcode() != IR::Opcode::ShuffleButterfly) { if (inst2->GetOpcode() != IR::Opcode::ShuffleButterfly) {
return; return;
@ -449,19 +445,15 @@ void FoldFSwizzleAdd(IR::Block& block, IR::Inst& inst) {
if (value_2 != value_3) { if (value_2 != value_3) {
return; return;
} }
const IR::Value index{inst2->Arg(1)}; const IR::Value index{inst2->Arg(1)};
const IR::Value clamp{inst2->Arg(2)}; const IR::Value clamp{inst2->Arg(2)};
const IR::Value segmentation_mask{inst2->Arg(3)}; const IR::Value segmentation_mask{inst2->Arg(3)};
if (!index.IsImmediate() || !clamp.IsImmediate() || !segmentation_mask.IsImmediate()) { if (!index.IsImmediate() || !clamp.IsImmediate() || !segmentation_mask.IsImmediate()) {
return; return;
} }
if (clamp.U32() != 3 || segmentation_mask.U32() != 28) { if (clamp.U32() != 3 || segmentation_mask.U32() != 28) {
return; return;
} }
if (swizzle_value == 0x99) { if (swizzle_value == 0x99) {
// DPdxFine // DPdxFine
if (index.U32() == 1) { if (index.U32() == 1) {