Mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-05-25 21:03:18 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

spirv: Flush denormals if possible (#1302)

Browse source
This commit is contained in:
TheTurtle 2024-10-10 17:47:39 +03:00 • committed by GitHub
parent 56e8ed7833
commit 100036aecf
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
14 changed files with 130 additions and 98 deletions
Download patch file Download diff file Expand all files Collapse all files

15
src/shader_recompiler/ir/passes/ring_access_elimination.cpp
View file

@ -1,7 +1,7 @@
// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include "shader_recompiler/frontend/translate/translate.h"
#include "shader_recompiler/ir/ir_emitter.h"
#include "shader_recompiler/ir/opcodes.h"
#include "shader_recompiler/ir/program.h"
#include "shader_recompiler/ir/reg.h"
@ -11,6 +11,8 @@ namespace Shader::Optimization {
void RingAccessElimination(const IR::Program& program, const RuntimeInfo& runtime_info,
Stage stage) {
auto& info = program.info;
const auto& ForEachInstruction = [&](auto func) {
for (IR::Block* block : program.blocks) {
for (IR::Inst& inst : block->Instructions()) {
@ -52,6 +54,9 @@ void RingAccessElimination(const IR::Program& program, const RuntimeInfo& runtim
break;
}
case Stage::Geometry: {
const auto& gs_info = runtime_info.gs_info;
info.gs_copy_data = Shader::ParseCopyShader(gs_info.vs_copy);
ForEachInstruction([&](IR::IREmitter& ir, IR::Inst& inst) {
const auto opcode = inst.GetOpcode();
switch (opcode) {
@ -81,12 +86,12 @@ void RingAccessElimination(const IR::Program& program, const RuntimeInfo& runtim
const auto offset = inst.Flags<IR::BufferInstInfo>().inst_offset.Value();
const auto data = ir.BitCast<IR::F32>(IR::U32{inst.Arg(2)});
const auto comp_ofs = runtime_info.gs_info.output_vertices * 4u;
const auto output_size = comp_ofs * runtime_info.gs_info.out_vertex_data_size;
const auto comp_ofs = gs_info.output_vertices * 4u;
const auto output_size = comp_ofs * gs_info.out_vertex_data_size;
const auto vc_read_ofs = (((offset / comp_ofs) * comp_ofs) % output_size) * 16u;
const auto& it = runtime_info.gs_info.copy_data.attr_map.find(vc_read_ofs);
ASSERT(it != runtime_info.gs_info.copy_data.attr_map.cend());
const auto& it = info.gs_copy_data.attr_map.find(vc_read_ofs);
ASSERT(it != info.gs_copy_data.attr_map.cend());
const auto& [attr, comp] = it->second;
inst.ReplaceOpcode(IR::Opcode::SetAttribute);

21
src/shader_recompiler/ir/reg.h
View file

@ -10,20 +10,6 @@
namespace Shader::IR {
enum class FpRoundMode : u32 {
NearestEven = 0,
PlusInf = 1,
MinInf = 2,
ToZero = 3,
};
enum class FpDenormMode : u32 {
InOutFlush = 0,
InAllowOutFlush = 1,
InFlushOutAllow = 2,
InOutAllow = 3,
};
enum class FloatClassFunc : u32 {
SignalingNan = 1 << 0,
QuietNan = 1 << 1,
@ -41,13 +27,6 @@ enum class FloatClassFunc : u32 {
};
DECLARE_ENUM_FLAG_OPERATORS(FloatClassFunc)
union Mode {
BitField<0, 4, FpRoundMode> fp_round;
BitField<4, 2, FpDenormMode> fp_denorm_single;
BitField<6, 2, FpDenormMode> fp_denorm_double;
BitField<8, 1, u32> dx10_clamp;
};
union TextureInstInfo {
u32 raw;
BitField<0, 1, u32> is_depth;