Yidhra/DirectXTex
2
0
Fork 0
mirror of https://github.com/microsoft/DirectXTex.git synced 2025-07-12 05:00:14 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

BC7 CPU codec optimization to skip mode 7 for opaque blocks

Browse Source
This commit is contained in:
Chuck Walbourn 2018-07-03 19:33:07 -07:00
parent a884753044
commit ba30da0945
1 changed files with 10 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
DirectXTex/BC6HBC7.cpp
View File

@ -2733,6 +2733,7 @@ void D3DX_BC7::Encode(DWORD flags, const HDRColorA* const pIn)
D3DX_BC7 final = *this; D3DX_BC7 final = *this;
EncodeParams EP(pIn); EncodeParams EP(pIn);
float fMSEBest = FLT_MAX; float fMSEBest = FLT_MAX;
uint32_t alphaMask = 0xFF;
for (size_t i = 0; i < NUM_PIXELS_PER_BLOCK; ++i) for (size_t i = 0; i < NUM_PIXELS_PER_BLOCK; ++i)
{ {
@ -2740,8 +2741,11 @@ void D3DX_BC7::Encode(DWORD flags, const HDRColorA* const pIn)
EP.aLDRPixels[i].g = uint8_t(std::max<float>(0.0f, std::min<float>(255.0f, pIn[i].g * 255.0f + 0.01f))); EP.aLDRPixels[i].g = uint8_t(std::max<float>(0.0f, std::min<float>(255.0f, pIn[i].g * 255.0f + 0.01f)));
EP.aLDRPixels[i].b = uint8_t(std::max<float>(0.0f, std::min<float>(255.0f, pIn[i].b * 255.0f + 0.01f))); EP.aLDRPixels[i].b = uint8_t(std::max<float>(0.0f, std::min<float>(255.0f, pIn[i].b * 255.0f + 0.01f)));
EP.aLDRPixels[i].a = uint8_t(std::max<float>(0.0f, std::min<float>(255.0f, pIn[i].a * 255.0f + 0.01f))); EP.aLDRPixels[i].a = uint8_t(std::max<float>(0.0f, std::min<float>(255.0f, pIn[i].a * 255.0f + 0.01f)));
alphaMask &= EP.aLDRPixels[i].a;
} }
const bool bHasAlpha = (alphaMask != 0xFF);
for (EP.uMode = 0; EP.uMode < 8 && fMSEBest > 0; ++EP.uMode) for (EP.uMode = 0; EP.uMode < 8 && fMSEBest > 0; ++EP.uMode)
{ {
if (!(flags & BC_FLAGS_USE_3SUBSETS) && (EP.uMode == 0 || EP.uMode == 2)) if (!(flags & BC_FLAGS_USE_3SUBSETS) && (EP.uMode == 0 || EP.uMode == 2))
@ -2756,6 +2760,12 @@ void D3DX_BC7::Encode(DWORD flags, const HDRColorA* const pIn)
continue; continue;
} }
if ((!bHasAlpha) && (EP.uMode == 7))
{
// There is no value in using mode 7 for completely opaque blocks (the other 2 subset modes handle this case for opaque blocks), so skip it for a small perf win.
continue;
}
const size_t uShapes = size_t(1) << ms_aInfo[EP.uMode].uPartitionBits; const size_t uShapes = size_t(1) << ms_aInfo[EP.uMode].uPartitionBits;
assert(uShapes <= BC7_MAX_SHAPES); assert(uShapes <= BC7_MAX_SHAPES);
_Analysis_assume_(uShapes <= BC7_MAX_SHAPES); _Analysis_assume_(uShapes <= BC7_MAX_SHAPES);