hwpplayer1/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

shader: Refactor half floating instructions

Browse source
This commit is contained in:
FernandoS27 2021-03-20 21:22:21 +01:00 committed by ameerj
parent f91859efd2
commit e802512d8e
4 changed files with 84 additions and 58 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/shader_recompiler/CMakeLists.txt
View file

@ -78,6 +78,8 @@ add_library(shader_recompiler STATIC
frontend/maxwell/translate/impl/floating_point_range_reduction.cpp
frontend/maxwell/translate/impl/floating_point_set_predicate.cpp
frontend/maxwell/translate/impl/half_floating_point_add.cpp
frontend/maxwell/translate/impl/half_floating_point_helper.cpp
frontend/maxwell/translate/impl/half_floating_point_helper.h
frontend/maxwell/translate/impl/impl.cpp
frontend/maxwell/translate/impl/impl.h
frontend/maxwell/translate/impl/integer_add.cpp

60
src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_add.cpp
View file

@ -2,66 +2,10 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_helper.h"
namespace Shader::Maxwell {
namespace {
enum class Merge : u64 {
H1_H0,
F32,
MRG_H0,
MRG_H1,
};
enum class Swizzle : u64 {
H1_H0,
F32,
H0_H0,
H1_H1,
};
std::pair<IR::F16F32F64, IR::F16F32F64> Extract(IR::IREmitter& ir, IR::U32 value, Swizzle swizzle) {
switch (swizzle) {
case Swizzle::H1_H0: {
const IR::Value vector{ir.UnpackFloat2x16(value)};
return {IR::F16{ir.CompositeExtract(vector, 0)}, IR::F16{ir.CompositeExtract(vector, 1)}};
}
case Swizzle::H0_H0: {
const IR::F16 scalar{ir.CompositeExtract(ir.UnpackFloat2x16(value), 0)};
return {scalar, scalar};
}
case Swizzle::H1_H1: {
const IR::F16 scalar{ir.CompositeExtract(ir.UnpackFloat2x16(value), 1)};
return {scalar, scalar};
}
case Swizzle::F32: {
const IR::F32 scalar{ir.BitCast<IR::F32>(value)};
return {scalar, scalar};
}
}
throw InvalidArgument("Invalid swizzle {}", swizzle);
}
IR::U32 MergeResult(IR::IREmitter& ir, IR::Reg dest, const IR::F16& lhs, const IR::F16& rhs,
Merge merge) {
switch (merge) {
case Merge::H1_H0:
return ir.PackFloat2x16(ir.CompositeConstruct(lhs, rhs));
case Merge::F32:
return ir.BitCast<IR::U32, IR::F32>(ir.FPConvert(32, lhs));
case Merge::MRG_H0:
case Merge::MRG_H1: {
const IR::Value vector{ir.UnpackFloat2x16(ir.GetReg(dest))};
const bool h0{merge == Merge::MRG_H0};
const IR::F16& insert{h0 ? lhs : rhs};
return ir.PackFloat2x16(ir.CompositeInsert(vector, insert, h0 ? 0 : 1));
}
}
throw InvalidArgument("Invalid merge {}", merge);
}
void HADD2(TranslatorVisitor& v, u64 insn, Merge merge, bool ftz, bool sat, bool abs_a, bool neg_a,
Swizzle swizzle_a, bool abs_b, bool neg_b, Swizzle swizzle_b, const IR::U32& src_b) {
@ -122,7 +66,7 @@ void HADD2(TranslatorVisitor& v, u64 insn, bool sat, bool abs_b, bool neg_b, Swi
HADD2(v, insn, hadd2.merge, hadd2.ftz != 0, sat, hadd2.abs_a != 0, hadd2.neg_a != 0,
hadd2.swizzle_a, abs_b, neg_b, swizzle_b, src_b);
}
} // Anonymous namespace
} // namespace
void TranslatorVisitor::HADD2_reg(u64 insn) {
union {

49
src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_helper.cpp Normal file
View file

@ -0,0 +1,49 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_helper.h"
namespace Shader::Maxwell {
std::pair<IR::F16F32F64, IR::F16F32F64> Extract(IR::IREmitter& ir, IR::U32 value, Swizzle swizzle) {
switch (swizzle) {
case Swizzle::H1_H0: {
const IR::Value vector{ir.UnpackFloat2x16(value)};
return {IR::F16{ir.CompositeExtract(vector, 0)}, IR::F16{ir.CompositeExtract(vector, 1)}};
}
case Swizzle::H0_H0: {
const IR::F16 scalar{ir.CompositeExtract(ir.UnpackFloat2x16(value), 0)};
return {scalar, scalar};
}
case Swizzle::H1_H1: {
const IR::F16 scalar{ir.CompositeExtract(ir.UnpackFloat2x16(value), 1)};
return {scalar, scalar};
}
case Swizzle::F32: {
const IR::F32 scalar{ir.BitCast<IR::F32>(value)};
return {scalar, scalar};
}
}
throw InvalidArgument("Invalid swizzle {}", swizzle);
}
IR::U32 MergeResult(IR::IREmitter& ir, IR::Reg dest, const IR::F16& lhs, const IR::F16& rhs,
Merge merge) {
switch (merge) {
case Merge::H1_H0:
return ir.PackFloat2x16(ir.CompositeConstruct(lhs, rhs));
case Merge::F32:
return ir.BitCast<IR::U32, IR::F32>(ir.FPConvert(32, lhs));
case Merge::MRG_H0:
case Merge::MRG_H1: {
const IR::Value vector{ir.UnpackFloat2x16(ir.GetReg(dest))};
const bool h0{merge == Merge::MRG_H0};
const IR::F16& insert{h0 ? lhs : rhs};
return ir.PackFloat2x16(ir.CompositeInsert(vector, insert, h0 ? 0 : 1));
}
}
throw InvalidArgument("Invalid merge {}", merge);
}
} // namespace Shader::Maxwell

31
src/shader_recompiler/frontend/maxwell/translate/impl/half_floating_point_helper.h Normal file
View file

@ -0,0 +1,31 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/common_types.h"
#include "shader_recompiler/exception.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
namespace Shader::Maxwell {
enum class Merge : u64 {
H1_H0,
F32,
MRG_H0,
MRG_H1,
};
enum class Swizzle : u64 {
H1_H0,
F32,
H0_H0,
H1_H1,
};
std::pair<IR::F16F32F64, IR::F16F32F64> Extract(IR::IREmitter& ir, IR::U32 value, Swizzle swizzle);
IR::U32 MergeResult(IR::IREmitter& ir, IR::Reg dest, const IR::F16& lhs, const IR::F16& rhs,
Merge merge);
} // namespace Shader::Maxwell