raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-26 12:56:54 +01:00
Code Issues Packages Projects Releases Wiki Activity

shader: apply the LOD clamp of image_sample _cl variants

Browse Source 
Co-authored-by: Senaxx <michelbeld@gmail.com>
This commit is contained in:
nmzik
2026-10-26 03:33:26 +01:00
parent b4209dde1b
commit 4ebae87a1b
5 changed files with 70 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ShaderRecompilerComputeTests.cpp
+47 -1
View File
@@ -21559,8 +21559,9 @@ private:
"shaderStorageImageWriteWithoutFormat is not supported");
Require("VulkanHarness", "dispatch",
available_features.shaderImageGatherExtended == true &&
available_features.shaderResourceMinLod == true &&
available_derivatives.computeDerivativeGroupQuads == true,
"image gather or compute derivative quads are not supported");
"image gather, MinLod or compute derivative quads are not supported");
Require("VulkanHarness", "dispatch",
available_features11.storageBuffer16BitAccess == true &&
available_features12.storageBuffer8BitAccess == true,
@@ -21694,6 +21695,7 @@ private:
device_features.shaderStorageImageWriteWithoutFormat = true;
device_features.shaderStorageImageMultisample = available_features.shaderStorageImageMultisample;
device_features.shaderImageGatherExtended = true;
device_features.shaderResourceMinLod = true;
device_features.sampleRateShading = true;
device_features.shaderInt64 = true;
device_features.shaderFloat64 = available_features.shaderFloat64;
@@ -39224,6 +39226,48 @@ TestCase ImageGatherExplicitLod() {
return test;
}
TestCase ImageSampleLodClamp() {
using O = ShaderOpcode;
// _cl clamps the computed LOD from below: compute's implicit LOD and zero
// derivatives would both select mip 0, so a clamp of 1 must read mip 1.
std::vector<u32> code;
AppendVMovU32(&code, 20, 0);
AppendVMovU32(&code, 21, 0);
AppendVMovU32(&code, 22, 0);
AppendVMovU32(&code, 23, 0);
AppendVMovLiteral(&code, 24, 0x3f000000u);
AppendVMovLiteral(&code, 25, 0x3f000000u);
AppendVMovLiteral(&code, 26, 0x3f800000u);
code.push_back(EncodeMimg0(0x21, 0x1));
code.push_back(EncodeMimg1(0, 24, 0, 2));
code.push_back(EncodeMimg0(0x23, 0x1));
code.push_back(EncodeMimg1(1, 20, 0, 2));
AppendStoreVgpr(&code, 0, 0);
AppendStoreVgpr(&code, 1, 1);
AppendEnd(&code);
TestCase test;
test.name = "ImageSampleLodClamp";
test.code = std::move(code);
test.opcodes = {O::V_MOV_B32, O::IMAGE_SAMPLE, O::BUFFER_STORE_DWORD, O::S_ENDPGM};
test.expected = std::vector<u32>(2, 0x41000000u);
test.sampled_image_rgba_mips = {std::vector<u32>(4 * 4 * 4, 0x3f800000u),
std::vector<u32>(2 * 2 * 4, 0x41000000u)};
test.user_data = MakeSampledTextureData(Prospero::BufferFormat::k32_32_32_32Float);
test.user_data[1] |= 3u << 30u;
test.user_data[2] = 3u << 14u;
test.user_data[3] |= 1u << 16u;
test.user_data[8] = 0x92u;
test.user_data[9] = 0xfff000u;
test.user_data[10] = 0x05000000u;
test.user_data[50] = test.expected.size() * sizeof(u32);
test.user_data[51] = 3u << 28u;
test.has_user_data = true;
test.required_spirv = {"MinLod"};
return test;
}
template <u32 mode> TestCase ImageGatherLodPerLane() {
using O = ShaderOpcode;
constexpr std::array names{
@@ -44788,6 +44832,7 @@ std::vector<TestCase> MakeCases() {
AddCase(ImageSampleAndGather);
AddCase(ImageCubeGradientsPreserveDerivatives);
AddCase(ImageGatherExplicitLod);
AddCase(ImageSampleLodClamp);
AddCase(ImageGatherLodPerLane<0>);
AddCase(ImageGatherLodPerLane<1>);
AddCase(ImageGatherLodPerLane<2>);
@@ -51205,6 +51250,7 @@ int main(int argc, char **argv) {
VulkanHarness vulkan;
vulkan.CheckComparisonDepthTexture();
RunCase(&vulkan, ImageGatherExplicitLod());
RunCase(&vulkan, ImageSampleLodClamp());
RunCase(&vulkan, ImageGatherLodPerLane<0>());
RunCase(&vulkan, ImageGatherLodPerLane<1>());
RunCase(&vulkan, ImageGatherLodPerLane<2>());