raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-19 18:36:53 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: use compact integer RTZ half conversion

Browse Source
This commit is contained in:
nmzik
2026-10-17 12:23:12 +02:00
parent c2173801d4
commit 9f88043ebe
2 changed files with 93 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ShaderRecompilerComputeTests.cpp
+51
View File
@@ -25877,6 +25877,49 @@ TestCase CvtPkrtzF16F32SubnormalRoundsTowardZero() {
O::S_ENDPGM}};
}
TestCase CvtPkrtzF16F32RuntimeEdges() {
using O = ShaderOpcode;
// Independent expected half encodings, including truncation boundaries and NaN payloads.
constexpr std::array<std::array<u32, 2>, 31> inputs{{
{0x00000000u, 0x0000u}, {0x80000000u, 0x8000u},
{0x00000001u, 0x0000u}, {0x007fffffu, 0x0000u},
{0x00800000u, 0x0000u}, {0x32ffffffu, 0x0000u},
{0x33000000u, 0x0000u}, {0x337fffffu, 0x0000u},
{0x33800000u, 0x0001u}, {0x33c00000u, 0x0001u},
{0x34000000u, 0x0002u}, {0x387fc000u, 0x03ffu},
{0x387fe000u, 0x03ffu}, {0x387fffffu, 0x03ffu},
{0x38800000u, 0x0400u}, {0x38802000u, 0x0401u},
{0x3e22f983u, 0x3117u}, {0x3f800000u, 0x3c00u},
{0x3f801000u, 0x3c00u}, {0x3f801001u, 0x3c00u},
{0x3f802000u, 0x3c01u}, {0x477fe000u, 0x7bffu},
{0x477ff000u, 0x7bffu}, {0x477fffffu, 0x7bffu},
{0x47800000u, 0x7bffu}, {0x7f7fffffu, 0x7bffu},
{0x7f800000u, 0x7c00u}, {0x7f800001u, 0x7e00u},
{0x7fa02000u, 0x7f01u}, {0x7fc12000u, 0x7e09u},
{0x7fffffffu, 0x7fffu},
}};
TestCase test;
test.name = "CvtPkrtzF16F32RuntimeEdges";
for (const auto &input : inputs)
test.initial.insert(test.initial.end(), {input[0], input[0] ^ 0x80000000u});
test.expected = test.initial;
for (u32 i = 0; i < inputs.size(); ++i) {
for (u32 reg = 0; reg < 2; ++reg) {
AppendVMovU32(&test.code, 30, (i * 2u + reg) * 4u);
AppendBufferLoadDword(&test.code, reg, 30);
}
test.code.push_back(EncodeVop2(0x2f, 2, Vgpr(0), 1));
AppendStoreVgpr(&test.code, 2, static_cast<u32>(test.expected.size()));
const auto half = inputs[i][1];
test.expected.push_back(half | ((half ^ 0x8000u) << 16));
}
AppendEnd(&test.code);
test.opcodes = {O::V_MOV_B32, O::BUFFER_LOAD_DWORD, O::V_CVT_PKRTZ_F16_F32,
O::BUFFER_STORE_DWORD, O::S_ENDPGM};
test.forbidden_spirv = {"PackHalf2x16", "UnpackHalf2x16"};
return test;
}
TestCase CvtPkrtzF16F32SdwaAndOutputModifiers() {
using O = ShaderOpcode;
@@ -42213,6 +42256,7 @@ std::vector<TestCase> MakeCases() {
AddCase(Vop3pInt16Ops);
AddCase(CvtPkU8F32PacksSelectedByte);
AddCase(CvtPkrtzF16F32SubnormalRoundsTowardZero);
AddCase(CvtPkrtzF16F32RuntimeEdges);
AddCase(CvtPkrtzF16F32SdwaAndOutputModifiers);
AddCase(PackedMinMaxF16NanAndSignedZeroEdges);
AddCase(VectorMinMaxF16Ops);
@@ -48029,6 +48073,13 @@ int main(int argc, char **argv) {
std::setvbuf(stdout, nullptr, _IONBF, 0);
EnsureConfigInitialized();
CheckLeastRecentlyUsedCacheOrdering();
if (argc == 2 && std::strcmp(argv[1], "--cvt-pkrtz-only") == 0) {
VulkanHarness vulkan;
RunCase(&vulkan, CvtPkrtzF16F32SubnormalRoundsTowardZero());
RunCase(&vulkan, CvtPkrtzF16F32RuntimeEdges());
RunCase(&vulkan, CvtPkrtzF16F32SdwaAndOutputModifiers());
return 0;
}
if (argc == 2 && std::strcmp(argv[1], "--thread-dimensions-only") == 0) {
VulkanHarness vulkan;
CheckComputeThreadDimensions(vulkan);