raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 19:07:01 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: implement 32-bit RDNA2 IMAGE_ATOMIC_CMPSWAP (#952)

Browse Source 
* shader: implement 32-bit RDNA2 image atomic compare-and-swap

* shader: share atomic emission and correct image barriers

---------

Co-authored-by: nmzik <Nmzik@mail.ru>
This commit is contained in:
Louis
2026-10-01 23:15:53 +02:00
committed by GitHub
parent 4428640b44
commit b3e419ff11
12 changed files with 163 additions and 102 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ShaderRecompilerComputeTests.cpp
+53
View File
@@ -30512,6 +30512,51 @@ TestCase ImageStoreAndAtomicShareTypedBinding() {
return test;
}
// RDNA 2 MIMG opcode 0x10: replacement at VDATA, comparison at VDATA+1.
TestCase ImageAtomicCompareSwapGlcAndExec() {
TestCase test;
test.name = "ImageAtomicCompareSwapGlcAndExec";
auto &code = test.code;
code.push_back(EncodeSop1(0x04, 12, 126)); // Preserve EXEC.
AppendVMovU32(&code, 20, 2);
AppendVMovU32(&code, 21, 1);
const auto exchange = [&](u32 replacement, u32 comparison, bool glc, u32 output) {
AppendVMovLiteral(&code, 0, replacement);
AppendVMovLiteral(&code, 1, comparison);
code.push_back(EncodeMimg0(0x10, 3, 0, glc));
code.push_back(EncodeMimg1(0, 20));
AppendStoreVgpr(&code, 0, output);
AppendStoreVgpr(&code, 1, output + 1);
};
exchange(0x12345678u, 0xdeadbeefu, true, 0); // Match; return original bits.
exchange(0x87654321u, 0xdeadbeefu, true, 2); // Mismatch; preserve texel.
exchange(0xabcdef01u, 0x12345678u, false, 4); // GLC=0 preserves VDATA.
AppendVMovLiteral(&code, 0, 0x11111111u);
AppendVMovLiteral(&code, 1, 0xabcdef01u);
code.push_back(EncodeSop1(0x04, 126, InlineU32(0))); // Disable EXEC.
code.push_back(EncodeMimg0(0x10, 3, 0, true));
code.push_back(EncodeMimg1(0, 20));
code.push_back(EncodeSop1(0x04, 126, 12));
AppendStoreVgpr(&code, 0, 6);
AppendStoreVgpr(&code, 1, 7);
AppendEnd(&code);
test.expected = {0xdeadbeefu, 0xdeadbeefu, 0x12345678u, 0xdeadbeefu,
0xabcdef01u, 0x12345678u, 0x11111111u, 0xabcdef01u};
test.opcodes = {ShaderOpcode::V_MOV_B32, ShaderOpcode::S_MOV_B64,
ShaderOpcode::IMAGE_ATOMIC_CMPSWAP, ShaderOpcode::BUFFER_STORE_DWORD,
ShaderOpcode::S_ENDPGM};
test.user_data = MakeStorageTextureData(Prospero::BufferFormat::k32UInt);
test.has_user_data = true;
test.storage_image_r32ui = std::vector<u32>(16, 0);
test.storage_image_r32ui[6] = 0xdeadbeefu;
test.expected_storage_image_r32ui = test.storage_image_r32ui;
test.expected_storage_image_r32ui[6] = 0xabcdef01u;
// Device scope, AcquireRelease | ImageMemory (0x808).
test.required_spirv = {"OpAtomicCompareExchange", "OpImageTexelPointer",
"OpMemoryBarrier %uint_1 %uint_2056"};
return test;
}
TestCase ImageAtomicSwapReturnsPreviousTexel() {
using O = ShaderOpcode;
@@ -31661,6 +31706,7 @@ std::vector<TestCase> MakeCases() {
AddCase(ImageStorePackedUintHonorsSparseDmask);
AddCase(ComputeTgSizeSgprUsesWaveMetadata);
AddCase(ImageStoreAndAtomicShareTypedBinding);
AddCase(ImageAtomicCompareSwapGlcAndExec);
AddCase(ImageAtomicSwapReturnsPreviousTexel);
AddCase(ImageStoreAndAtomicUseSeparateBindings);
AddCase(ImageAtomicVariants);
@@ -36636,6 +36682,13 @@ int main(int argc, char **argv) {
RunCase(&vulkan, BufferAtomicCmpSwapExactRaw());
return 0;
}
if (argc == 2 && std::strcmp(argv[1], "--image-cmpswap-only") == 0) {
VulkanHarness vulkan;
RunCase(&vulkan, ImageAtomicCompareSwapGlcAndExec());
RunCase(&vulkan, ImageAtomicSwapReturnsPreviousTexel());
RunCase(&vulkan, ImageAtomicVariants());
return 0;
}
if (argc == 2 && std::strcmp(argv[1], "--ds-atomics-only") == 0) {
VulkanHarness vulkan;
for (const auto &test : MakeCases()) {