gremling/KytyPS5
1
0 Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 08:13:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: give A16 1D and late 3D gradients their own address VGPRs

Browse Source 
Co-authored-by: shadowbeat070 <70843316+shadowbeat070@users.noreply.github.com>
This commit is contained in:
nmzikandshadowbeat070 committed 2026-10-27 04:23:01 +01:00
1 parent 0a72641678
commit 78adb9293d
6 files changed
+82 -15
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/graphics/shader/recompiler/frontend/decode/ImageOps.cpp
+16 -6
View File
@@ -319,21 +319,31 @@ uint32_t CountDmaskComponents(uint32_t dmask) {
} // namespace
ImageAddressComponent ImageAddressComponentLayout(uint32_t flags, uint32_t component) {
ImageAddressComponent ImageAddressComponentLayout(uint32_t flags, ImageDimension dimension,
uint32_t component) {
if ((flags & ImageSampleFlagA16) == 0u) return {component * 32u, 32u};
// A16 offsets, bias and PCF reference each take a whole VGPR, the bias as a half in its low
// 16 bits; the components after them pack two per VGPR.
// 16 bits. The other components pack two per VGPR, except that each 1D derivative and the last
// two 3D ones take the low half of their own VGPR.
const bool has_bias = (flags & ImageSampleFlagBias) != 0u;
const uint32_t bias_index = (flags & ImageSampleFlagOffset) != 0u ? 1u : 0u;
const uint32_t whole_vgprs = std::popcount(
flags & (ImageSampleFlagOffset | ImageSampleFlagBias | ImageSampleFlagCompare));
if (component >= whole_vgprs) return {(whole_vgprs + component) * 16u, 16u};
return {component * 32u, has_bias && component == bias_index ? 16u : 32u};
if (component < whole_vgprs) {
return {component * 32u, has_bias && component == bias_index ? 16u : 32u};
}
const uint32_t axes =
(flags & ImageSampleFlagDerivative) != 0u ? ImageGradientComponents(dimension) : 0u;
const uint32_t unpaired = axes % 2u * 2u;
const uint32_t paired = axes * 2u - unpaired;
const uint32_t index = component - whole_vgprs;
const uint32_t half = index + std::clamp(index, paired, paired + unpaired) - paired;
return {whole_vgprs * 32u + half * 16u, 16u};
}
uint32_t ImageAddressDwordCount(uint32_t flags, uint32_t components) {
uint32_t ImageAddressDwordCount(uint32_t flags, ImageDimension dimension, uint32_t components) {
if (components == 0u) return 0u;
const auto last = ImageAddressComponentLayout(flags, components - 1u);
const auto last = ImageAddressComponentLayout(flags, dimension, components - 1u);
return (last.bit_offset + last.bit_width + 31u) / 32u;
}