Mirror-bot/AnyPS5
1
0 Fork 0
mirror of https://github.com/boykopovar/AnyPS5.git synced 2026-11-07 13:39:55 +00:00
Code Issues Projects Releases Activity

Merge pull request #5167 from LeandroLP23/perf/vcmp-test-recompile

Browse Source 
test(agc): compile each v_cmp denormals program once for its eight input groups
This commit is contained in:
Adrià authored and GitHub committed 2026-10-14 20:21:08 +02:00
2 parents 32e810db49 59235d9cb0
commit f25386b90b
1 file changed
+16 -12
Download Patch File Download Diff File Expand all files Collapse all files
core/libs/prx/libSceAgcDriver/tests/execution/VCmpF32Denormals.cpp
+16 -12
View File
@@ -95,17 +95,9 @@ std::string Hex(std::uint32_t value) {
return text;
}
std::uint32_t Run(AgcDriver::VulkanDevice& device, ShaderRecompiler::ShaderFloatMode mode, std::uint32_t compare, std::uint32_t capture, std::uint32_t group) {
ShaderRecompiler::RecompileResult Compile(const AgcDriver::VulkanDevice& device, ShaderRecompiler::ShaderFloatMode mode, std::uint32_t compare, std::uint32_t capture) {
Code[6] = compare;
Code[7] = capture;
for (std::uint32_t lane = 0; lane < Threads; ++lane) {
const std::uint32_t pair = group * Threads + lane;
Input[lane * Inputs] = Values[pair / ValueCount];
Input[lane * Inputs + 1u] = Values[pair % ValueCount];
Input[lane * Inputs + 2u] = 0u;
Input[lane * Inputs + 3u] = 0u;
}
Output.fill(0xdeadbeefu);
std::vector<std::uint32_t> userData(8, 0u);
const auto input = BufferDescriptor(Input.data(), static_cast<std::uint32_t>(Input.size() * 4u));
const auto output = BufferDescriptor(Output.data(), static_cast<std::uint32_t>(Output.size() * 4u));
@@ -122,8 +114,19 @@ std::uint32_t Run(AgcDriver::VulkanDevice& device, ShaderRecompiler::ShaderFloat
};
request.useCache = false;
request.context.floatMode = mode;
const auto result = ShaderRecompiler::Recompile(request);
device.Dispatch(result, 1, 1, 1, {}, reinterpret_cast<std::uintptr_t>(code.data()));
return ShaderRecompiler::Recompile(request);
}
std::uint32_t Run(AgcDriver::VulkanDevice& device, const ShaderRecompiler::RecompileResult& program, std::uint32_t group) {
for (std::uint32_t lane = 0; lane < Threads; ++lane) {
const std::uint32_t pair = group * Threads + lane;
Input[lane * Inputs] = Values[pair / ValueCount];
Input[lane * Inputs + 1u] = Values[pair % ValueCount];
Input[lane * Inputs + 2u] = 0u;
Input[lane * Inputs + 3u] = 0u;
}
Output.fill(0xdeadbeefu);
device.Dispatch(program, 1, 1, 1, {}, reinterpret_cast<std::uintptr_t>(Code.data()));
device.WaitIdle();
std::uint32_t mask = 0u;
for (std::uint32_t lane = 0; lane < Threads; ++lane) {
@@ -135,8 +138,9 @@ std::uint32_t Run(AgcDriver::VulkanDevice& device, ShaderRecompiler::ShaderFloat
void Check(AgcDriver::VulkanDevice& device, const char* modeName, ShaderRecompiler::ShaderFloatMode mode, const Row& row, bool exec, const std::uint32_t* expected) {
const std::string name = std::string(exec ? "v_cmpx_" : "v_cmp_") + row.condition + "_f32";
const auto program = Compile(device, mode, exec ? row.cmpx : row.cmp, exec ? CaptureExec : CaptureVcc);
for (std::uint32_t group = 0; group < Groups; ++group) {
const std::uint32_t actual = Run(device, mode, exec ? row.cmpx : row.cmp, exec ? CaptureExec : CaptureVcc, group);
const std::uint32_t actual = Run(device, program, group);
if (actual == expected[group]) continue;
const std::uint32_t diff = actual ^ expected[group];
for (std::uint32_t lane = 0; lane < Threads; ++lane) {