mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-12-12 14:48:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader_recompiler: Better handling of geometry shader scenario G (#3064)

Browse Source
This commit is contained in:
TheTurtle
2025-06-09 03:31:51 +03:00
committed by GitHub
parent ae2053c487
commit c20d02dd40
8 changed files with 33 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/shader_recompiler/ir/passes/readlane_elimination_pass.cpp
View File

@@ -1,6 +1,7 @@
// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#include <unordered_map>
#include "shader_recompiler/ir/program.h"
namespace Shader::Optimization {

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

@@ -91,6 +91,19 @@ void RingAccessElimination(const IR::Program& program, const RuntimeInfo& runtim
const auto& gs_info = runtime_info.gs_info;
info.gs_copy_data = Shader::ParseCopyShader(gs_info.vs_copy);
u32 output_vertices = gs_info.output_vertices;
if (info.gs_copy_data.output_vertices &&
info.gs_copy_data.output_vertices != output_vertices) {
ASSERT_MSG(output_vertices > info.gs_copy_data.output_vertices &&
gs_info.mode == AmdGpu::Liverpool::GsMode::Mode::ScenarioG,
"Invalid geometry shader vertex configuration scenario = {}, max_vert_out = "
"{}, output_vertices = {}",
u32(gs_info.mode), output_vertices, info.gs_copy_data.output_vertices);
LOG_WARNING(Render_Vulkan, "MAX_VERT_OUT {} is larger than actual output vertices {}",
output_vertices, info.gs_copy_data.output_vertices);
output_vertices = info.gs_copy_data.output_vertices;
}
ForEachInstruction([&](IR::IREmitter& ir, IR::Inst& inst) {
const auto opcode = inst.GetOpcode();
switch (opcode) {
@@ -122,7 +135,7 @@ 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 = gs_info.output_vertices * 4u;
const auto comp_ofs = 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;