gremling/KytyPS5
1
0 Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 00:12:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: count one SGPR for lane reads and wave32 lane masks in call targets

Browse Source 
Co-authored-by: Ali Almohaya <AliAlmohaya@gmail.com>
Co-authored-by: Senaxx <michelbeld@gmail.com>
Co-authored-by: Mac <me@itsmemac.com>
This commit is contained in:
nmzikAli AlmohayaSenaxxMac
 committed 2026-10-26 05:32:02 +01:00
1 parent 153dadd0b3
commit 7524e8f79e
5 files changed
+33 -12
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/graphics/shader/recompiler/ShaderRecompiler.cpp
+7 -4
View File
@@ -467,6 +467,7 @@ Decoder::Program DecodeFusedProgram(std::span<const uint32_t> front, std::span<c
void PrepareCallTargets(ShaderSource& source, const CompileOptions& options) {
using namespace IR;
EXIT_NOT_IMPLEMENTED(options.stage != ShaderType::Compute);
source.wave_size = options.wave_size;
for (uint32_t i = 0; i < source.decoded.instructions.size(); ++i) {
if (source.decoded.instructions[i].opcode != Decoder::Opcode::S_SWAPPC_B64) continue;
source.calls.push_back({.instruction = i});
@@ -503,7 +504,8 @@ void PrepareCallTargets(ShaderSource& source, const CompileOptions& options) {
};
Register = [&](uint32_t before, uint32_t code) -> Value {
std::vector<uint32_t> crossed_calls;
const auto index = CFG::FindScalarDefinition(source.decoded, graph, before, code, crossed_calls);
const auto index = CFG::FindScalarDefinition(source.decoded, graph, before, code,
source.wave_size, crossed_calls);
for (const auto crossed: crossed_calls) {
std::ranges::find(source.calls, crossed, &ShaderSource::Call::instruction)->preserved_registers.set(code);
}
@@ -702,10 +704,11 @@ const Decoder::Program& RefreshShaderSource(ShaderSource& source, const IR::SrtR
EXIT_IF(function_pc > UINT32_MAX - function_size);
const auto function_offset = static_cast<uint32_t>(linked.code.size());
for (auto inst: function) {
EXIT_IF(CFG::MayWriteScalarRegister(inst, call.dst.reg) ||
CFG::MayWriteScalarRegister(inst, call.dst.reg + 1u));
EXIT_IF(CFG::MayWriteScalarRegister(inst, call.dst.reg, source.wave_size) ||
CFG::MayWriteScalarRegister(inst, call.dst.reg + 1u, source.wave_size));
for (uint32_t code = 0; code < site.preserved_registers.size(); ++code) {
if (site.preserved_registers[code] && CFG::MayWriteScalarRegister(inst, code))
if (site.preserved_registers[code] &&
CFG::MayWriteScalarRegister(inst, code, source.wave_size))
EXIT("shader call at pc 0x%08x clobbers scalar target dependency s%u", call.pc, code);
}
linked.code.insert(linked.code.end(), inst.raw, inst.raw + inst.word_count);