raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

Keep shader user data inline for cached programs

Browse Source
This commit is contained in:
nmzik
2026-09-23 10:49:08 +04:00
parent 5d1ad31c49
commit 03c55e5712
4 changed files with 48 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/renderer/pipeline/pipelineCache.cpp
+4 -3
View File
@@ -276,14 +276,15 @@ struct PipelineCache::ProgramCache {
stage = ShaderType::Compute;
}
const auto user_data = std::span(params.user_data).first(params.user_data_count);
lookup_key.stage = stage;
lookup_key.hash = params.hash;
lookup_key.user_data_count = static_cast<uint32_t>(params.user_data.size());
lookup_key.user_data_count = params.user_data_count;
lookup_key.code_size = static_cast<uint32_t>(params.code.size());
BuildStageStaticKey(input_info, lookup_key.static_state);
auto entry = programs.find(lookup_key);
const ShaderRecompiler::IR::SrtRuntime runtime {
.user_data = params.user_data,
.user_data = user_data,
.shader_base = params.Base(),
.read_specialization_memory = ReadShaderGuestMemory,
};
@@ -329,7 +330,7 @@ struct PipelineCache::ProgramCache {
ShaderRecompiler::CompileOptions options;
options.stage = stage;
options.shader_hash = params.hash;
options.user_data = params.user_data;
options.user_data = user_data;
options.back_code = params.back_code;
options.dump_ir = Config::GetShaderLogDirection() != Config::LogDirection::Silent;
options.early_dump = options.dump_ir;
src/graphics/shader/shader.cpp
+15 -11
View File
@@ -108,7 +108,7 @@ static uint64_t GetDeclaredShaderHash(uint64_t shader_addr) {
static ShaderParams GetShaderParams(uint64_t shader_addr, const char* label, uint64_t declared_hash,
std::span<const uint32_t> user_data,
const ShaderMappedData& data) {
const ShaderMappedData& data, uint32_t user_data_base = 0) {
if (data.code_size_bytes == 0 || data.code_size_bytes % sizeof(uint32_t) != 0) {
EXIT("%s hash=0x%016" PRIx64 " shader=0x%016" PRIx64
" has invalid AGC shader_size=0x%08" PRIx32 "\n",
@@ -116,12 +116,16 @@ static ShaderParams GetShaderParams(uint64_t shader_addr, const char* label, uin
}
const auto code_words = data.code_size_bytes / sizeof(uint32_t);
const auto code = std::span {reinterpret_cast<const uint32_t*>(shader_addr), code_words};
return {
.code = code,
.user_data = std::vector<uint32_t>(user_data.begin(), user_data.end()),
.hash = declared_hash != 0 ? declared_hash
: XXH3_64bits(code.data(), code.size_bytes()),
ShaderParams params {
.code = code,
.user_data_count = static_cast<uint32_t>(user_data.size()) + user_data_base,
.hash = declared_hash != 0 ? declared_hash
: XXH3_64bits(code.data(), code.size_bytes()),
};
EXIT_IF(user_data.size() > HW::UserSgprInfo::SGPRS_MAX ||
params.user_data_count > params.user_data.size());
std::copy(user_data.begin(), user_data.end(), params.user_data.begin() + user_data_base);
return params;
}
#if 0
@@ -756,11 +760,13 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
const HW::UserConfig& user_config, ShaderVertexInputInfo& info) {
const auto& sh = context.GetShaderRegisters();
const auto data = ShaderGetMappedData(regs.es_regs.data_addr, "ShaderGetInputInfoVS():");
const bool merged = (context.GetShaderStages() & 0x20u) != 0;
auto params = GetShaderParams(
regs.es_regs.data_addr, "ShaderRecompiler VS",
GetDeclaredShaderHash(regs.es_regs.data_addr),
std::span<const uint32_t>(regs.gs_user_sgpr.value, regs.gs_regs.rsrc2.user_sgpr), data);
if ((context.GetShaderStages() & 0x20u) == 0) {
std::span<const uint32_t>(regs.gs_user_sgpr.value, regs.gs_regs.rsrc2.user_sgpr), data,
merged ? 8u : 0u);
if (!merged) {
if (!ShaderGetStaticVertexInputInfo(regs.es_regs.data_addr, regs.gs_user_sgpr,
regs.gs_regs.rsrc2.user_sgpr, sh, data, info)) {
EXIT("failed to prepare vertex shader program\n");
@@ -770,7 +776,6 @@ ShaderParams PrepareProgram(const HW::VertexShaderInfo& regs, const HW::Context&
}
// NGG user SGPRs start at s8; a separately compiled GS back half also receives
// its user-data pointer in s0:s1.
params.user_data.insert(params.user_data.begin(), 8u, 0u);
info = {};
info.logical_stage = ShaderType::Mesh;
info.pa_cl_vs_out_cntl = sh.m_paClVsOutCntl;
@@ -841,14 +846,13 @@ PrepareTessellationPrograms(const HW::VertexShaderInfo& regs, const HW::Context&
GetShaderParams(regs.ls_regs.data_addr, "ShaderRecompiler LS",
GetDeclaredShaderHash(regs.ls_regs.data_addr), local_users, local),
GetShaderParams(regs.hs_regs.data_addr, "ShaderRecompiler HS",
GetDeclaredShaderHash(regs.hs_regs.data_addr), local_users, control),
GetDeclaredShaderHash(regs.hs_regs.data_addr), local_users, control, 8u),
GetShaderParams(regs.es_regs.data_addr, "ShaderRecompiler TES",
GetDeclaredShaderHash(regs.es_regs.data_addr), evaluation_users,
evaluation),
};
// The fused HS back half receives its separate user-data address in s0:s1.
// RDNA2 reserves s0:s7 before the native HS user SGPRs.
params[1].user_data.insert(params[1].user_data.begin(), 8u, 0u);
params[1].user_data[0] = static_cast<uint32_t>(regs.hs_regs.user_data_addr);
params[1].user_data[1] = static_cast<uint32_t>(regs.hs_regs.user_data_addr >> 32u);
src/graphics/shader/shaderCompiler.h
+3 -1
View File
@@ -3,6 +3,7 @@
#include "graphics/shader/shader.h"
#include <array>
#include <span>
#include <vector>
@@ -15,7 +16,8 @@ class UserConfig;
struct ShaderParams {
std::span<const uint32_t> code;
std::vector<uint32_t> user_data;
std::array<uint32_t, 40> user_data {}; // 32 user SGPRs plus the merged-stage s0:s7 prefix.
uint32_t user_data_count = 0;
uint64_t hash = 0;
std::span<const uint32_t> back_code;