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: sample packed normalized images with one bilinear gather

Browse Source
This commit is contained in:
nmzik
2026-10-12 23:51:22 +02:00
parent c11254a996
commit 56738aa4e1
7 changed files with 225 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ShaderRecompilerComputeTests.cpp
+81
View File
@@ -34625,6 +34625,83 @@ TestCase ImageSamplePackedUintConvertsSampleAndGather() {
return test;
}
template <bool sparse> TestCase ImageSamplePackedUnormBilinear() {
using O = ShaderOpcode;
namespace D = ShaderRecompiler::Decoder;
const std::vector<u32> unnormalized = {EncodeMimg0(0x27, 0x7) | (1u << 12u),
EncodeMimg1(0, 20, 0, 2)};
D::Instruction decoded;
D::DecodeInstruction(unnormalized, 0, decoded);
Require("ImageSamplePackedUnormBilinear", "coordinate mode",
decoded.image_sample_flags ==
(D::ImageSampleFlagLevelZero | D::ImageSampleFlagUnnormalized),
"unnormalized instruction coordinates lost their sampling restriction");
constexpr u32 channels[4][3] = {{0, 0, 0}, {2047, 0, 512},
{0, 2047, 1023}, {2047, 2047, 256}};
constexpr float coordinates[][2] = {{0.5f, 0.5f}, {0.375f, 0.625f},
{-0.25f, 1.25f}, {0.25f, 0.25f},
{0.75f, 0.25f}, {0.0f, 0.5f}};
TestCase test;
test.name = sparse ? "ImageSamplePackedUnormBilinearSparse" : "ImageSamplePackedUnormBilinear";
test.image_width = test.image_height = 2;
test.sampled_image_format = vk::Format::eR32Uint;
test.sampled_image_dwords_per_pixel = 1;
for (const auto &pixel : channels) {
test.sampled_image_rgba.push_back(pixel[0] | (pixel[1] << 11u) | (pixel[2] << 22u));
}
test.user_data = MakeSampledTextureData(Prospero::BufferFormat::k11_11_10UNorm);
test.user_data[1] |= 1u << 30u;
test.user_data[2] = 1u << 14u;
test.user_data[8] = 0x92u;
test.user_data[9] = 0x00fff000u;
test.user_data[10] = 0x0a500000u; // Captured linear guest sampler; gather uses integer texels.
test.has_user_data = true;
test.image_descriptor_swizzle = DstSel(6, 4, 5, 1);
test.expected_force_point_sampler = true;
test.expected_float_tolerance = 0.000005f;
auto &code = test.code;
code.push_back(EncodeSop1(0x04, 12, 126));
for (const auto &uv : coordinates) {
AppendVMovLiteral(&code, 20, std::bit_cast<u32>(uv[0]));
AppendVMovLiteral(&code, 21, std::bit_cast<u32>(uv[1]));
code.push_back(EncodeMimg0(0x27, sparse ? 0x5 : 0xf));
code.push_back(EncodeMimg1(0, 20, 0, 2));
const float x = uv[0] * 2.0f - 0.5f, y = uv[1] * 2.0f - 0.5f;
const int ix = static_cast<int>(std::floor(x)), iy = static_cast<int>(std::floor(y));
const float fx = x - std::floor(x), fy = y - std::floor(y);
float rgba[4] = {0, 0, 0, 1};
for (u32 channel = 0; channel < 3; ++channel) {
for (int dy = 0; dy < 2; ++dy) for (int dx = 0; dx < 2; ++dx) {
const auto pixel = std::clamp(iy + dy, 0, 1) * 2 + std::clamp(ix + dx, 0, 1);
rgba[channel] += channels[pixel][channel] / (channel == 2 ? 1023.0f : 2047.0f) *
(dx ? fx : 1.0f - fx) * (dy ? fy : 1.0f - fy);
}
}
constexpr u32 swizzle[4] = {2, 0, 1, 3};
u32 reg = 0;
for (u32 component = 0; component < 4; ++component) {
if (sparse && component != 0 && component != 2) continue;
AppendStoreVgpr(&code, reg++, test.expected.size());
test.expected.push_back(std::bit_cast<u32>(rgba[swizzle[component]]));
}
}
AppendVMovLiteral(&code, 0, 0x3f400000u);
code.push_back(EncodeSop1(0x04, 126, InlineU32(0)));
code.push_back(EncodeMimg0(0x27, 0x1));
code.push_back(EncodeMimg1(0, 20, 0, 2));
code.push_back(EncodeSop1(0x04, 126, 12));
AppendStoreVgpr(&code, 0, test.expected.size());
test.expected.push_back(0x3f400000u);
AppendEnd(&code);
test.user_data[50] = test.expected.size() * sizeof(u32);
test.user_data[51] = 3u << 28u;
test.opcodes = {O::V_MOV_B32, O::S_MOV_B64, O::IMAGE_SAMPLE,
O::BUFFER_STORE_DWORD, O::S_ENDPGM};
test.required_spirv = {"OpImageGather", "OpBitFieldUExtract", "OpConvertUToF"};
test.forbidden_spirv = {"OpImageFetch", "OpImageSample", "OpImageRead", "OpImageWrite"};
return test;
}
template <bool rg, bool snorm = false> TestCase ImageSamplePacked8() {
using O = ShaderOpcode;
TestCase test;
@@ -38570,6 +38647,8 @@ std::vector<TestCase> MakeCases() {
AddCase(ImageLoadR32SintUsesSignedSampledImage);
AddCase(ImageLoadPackedUintUnpacksAndSwizzles);
AddCase(ImageSamplePackedUintConvertsSampleAndGather);
AddCase(ImageSamplePackedUnormBilinear<false>);
AddCase(ImageSamplePackedUnormBilinear<true>);
AddCase(ImageSamplePacked8<false>);
AddCase(ImageSamplePacked8<true>);
AddCase(ImageSamplePacked8<false, true>);
@@ -44197,6 +44276,8 @@ int main(int argc, char **argv) {
vulkan.CheckBgra16Readback();
RunCase(&vulkan, ImageLoadPackedUintUnpacksAndSwizzles());
RunCase(&vulkan, ImageSamplePackedUintConvertsSampleAndGather());
RunCase(&vulkan, ImageSamplePackedUnormBilinear<false>());
RunCase(&vulkan, ImageSamplePackedUnormBilinear<true>());
RunCase(&vulkan, ImageStorePackedUintSaturatesChannels());
RunCase(&vulkan, ImageStorePackedUintHonorsSparseDmask());
RunCase(&vulkan, ImageStorePackedUnorm<false>());