raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-22 03:46:53 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: compute DPP permute, rotate, mirror and share lanes with bit-field and XOR ops

Browse Source
This commit is contained in:
nmzik
2026-10-22 00:40:07 +02:00
parent 6123419437
commit 09a9654458
2 changed files with 57 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ShaderRecompilerComputeTests.cpp
+33
View File
@@ -27298,6 +27298,35 @@ TestCase VectorDppRowShare(u32 wave_size) {
return test;
}
// row_mirror, row_half_mirror and row_ror read lanes within the row; v1 = 100 + lane.
TestCase VectorDppMirrorAndRotate(u32 wave_size) {
using O = ShaderOpcode;
TestCase test;
test.name = wave_size == 32 ? "VectorDppMirrorAndRotateWave32" : "VectorDppMirrorAndRotateWave64";
auto &code = test.code;
AppendVMovU32(&code, 1, 100);
code.push_back(EncodeVop2(0x25, 1, Vgpr(0), 1));
for (const u32 control : {0x140u, 0x141u, 0x121u, 0x125u, 0x12fu}) {
code.push_back(EncodeVop1(0x01, 2, 250));
code.push_back(EncodeVop1Dpp(1, control));
AppendStoreVgprAtLaneDwordOffset(&code, 2, 0, u32(test.expected.size()));
for (u32 lane = 0; lane < wave_size; ++lane) {
const u32 source = control == 0x140u ? (lane | 15u) - (lane & 15u)
: control == 0x141u ? (lane | 7u) - (lane & 7u)
: (lane & ~15u) | ((lane - (control & 15u)) & 15u);
test.expected.push_back(100 + source);
}
}
AppendEnd(&code);
test.opcodes = {O::V_MOV_B32, O::V_ADD_NC_U32, O::BUFFER_STORE_DWORD, O::S_ENDPGM};
test.compute_info.wave_size = wave_size;
test.compute_info.threads_num[0] = wave_size;
test.compute_info.threads_num[1] = test.compute_info.threads_num[2] = 1;
test.compute_info.thread_ids_num = 1;
test.has_compute_info = true;
return test;
}
TestCase VectorDppBankMaskPreservesDestination() {
using O = ShaderOpcode;
@@ -43058,6 +43087,8 @@ std::vector<TestCase> MakeCases() {
AddCase(VectorDppRowXmask);
cases.push_back(VectorDppRowShare(32));
cases.push_back(VectorDppRowShare(64));
cases.push_back(VectorDppMirrorAndRotate(32));
cases.push_back(VectorDppMirrorAndRotate(64));
cases.push_back(ComputeWave64Minimum(false));
cases.push_back(ComputeWave64Minimum(true));
AddCase(VectorDppBankMaskPreservesDestination);
@@ -49148,6 +49179,8 @@ int main(int argc, char **argv) {
RunCase(&vulkan, VectorDppRowXmask());
RunCase(&vulkan, VectorDppRowShare(32));
RunCase(&vulkan, VectorDppRowShare(64));
RunCase(&vulkan, VectorDppMirrorAndRotate(32));
RunCase(&vulkan, VectorDppMirrorAndRotate(64));
RunCase(&vulkan, VectorDppBankMaskPreservesDestination());
RunCase(&vulkan, VectorDppBoundsControlZeroPreservesDestination());
return 0;