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

shader: decode upper-half byte stores through shared operand selection

Browse Source 
Co-authored-by: fxpw <fxpw@outlook.com>
This commit is contained in:
nmzik
2026-10-14 04:20:02 +02:00
parent 6abfe3bbe4
commit 8d18be71db
5 changed files with 46 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ShaderRecompilerComputeTests.cpp
+40
View File
@@ -30527,6 +30527,42 @@ TestCase BufferStoreVariants() {
O::BUFFER_STORE_DWORDX4, O::S_ENDPGM}};
}
TestCase BufferStoreByteD16Hi(bool masked) {
using O = ShaderOpcode;
TestCase test;
test.name = masked ? "BufferStoreByteD16HiMasked" : "BufferStoreByteD16HiAdjacentLanes";
AppendVMovLiteral(&test.code, 4, 0xa5c39b72u);
test.code.push_back(EncodeVop2(0x25, 25, InlineU32(4), 0));
if (masked) {
AppendSMovLiteral(&test.code, 126, 0x55555555u);
AppendSMovLiteral(&test.code, 127, 0xaaaaaaaau);
}
// Captured b21e373811f28e31 pc 0x1b8c: e0646000 80020419.
test.code.push_back(EncodeMubuf0(0x19u, 0, true, false, true));
test.code.push_back(EncodeMubuf1(4, 2, 25));
AppendEnd(&test.code);
test.initial = std::vector<u32>(20, 0x11223344u);
test.expected = test.initial;
for (u32 lane = 0; lane < 60; lane++) {
const u32 mask = !masked ? 0xffffffffu : lane < 32 ? 0x55555555u : 0xaaaaaaaau;
if (((mask >> (lane % 32)) & 1u) == 0) continue;
const u32 byte = lane + 4, shift = (byte % 4) * 8;
test.expected[byte / 4] = (test.expected[byte / 4] & ~(0xffu << shift)) | (0xc3u << shift);
}
const auto descriptor = MakeStructuredStorageBufferData(1, 64);
std::copy_n(descriptor.begin(), 4, test.user_data.begin() + 8);
test.has_user_data = true;
test.storage_buffer_range_bytes = 64;
test.compute_info.threads_num[0] = 64;
test.compute_info.threads_num[1] = test.compute_info.threads_num[2] = 1;
test.compute_info.thread_ids_num = 1;
test.compute_info.wave_size = 64;
test.has_compute_info = true;
test.opcodes = {O::V_MOV_B32, O::V_ADD_NC_U32, O::BUFFER_STORE_BYTE_D16_HI, O::S_ENDPGM};
test.forbidden_spirv = {"OpAtomicLoad", "OpAtomicCompareExchange"};
return test;
}
TestCase BufferFormatVariants() {
using O = ShaderOpcode;
@@ -39383,6 +39419,8 @@ std::vector<TestCase> MakeCases() {
AddCase(BufferLoadFormatXyRejectsPartialRecord);
AddCase(BufferStoreFormatXyRejectsPartialRecord);
AddCase(BufferStoreVariants);
cases.push_back(BufferStoreByteD16Hi(false));
cases.push_back(BufferStoreByteD16Hi(true));
AddCase(BufferFormatVariants);
AddCase(BufferLoadFormatXyzwSnapshotsOverlappingAddress);
AddCase(BufferLoadFormatXyzwResource32_32FloatAppliesXy01);
@@ -45086,6 +45124,8 @@ int main(int argc, char **argv) {
RunCase(&vulkan, BufferStoreFormatXRejectsPartialRecord());
RunCase(&vulkan, BufferStoreFormatXyRejectsPartialRecord());
RunCase(&vulkan, BufferFormatStoreVariants());
RunCase(&vulkan, BufferStoreByteD16Hi(false));
RunCase(&vulkan, BufferStoreByteD16Hi(true));
return 0;
}
if (argc == 2 && std::strcmp(argv[1], "--buffer-float16-store-only") == 0) {