mirrors/looking-glass
2
0
Fork 0
mirror of https://github.com/gnif/LookingGlass.git synced 2024-11-12 20:27:53 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[common] rect: fix avx alignment issue take 2

Browse source
This commit is contained in:
Geoffrey McRae 2023-11-19 18:01:34 +11:00
parent 660b4b8ec8
commit 8d27d9e2e2
1 changed files with 5 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
common/src/rects.c
View file

@ -328,18 +328,18 @@ static void rectCopyUnaligned_avx(
src += ystart * srcPitch + dx; src += ystart * srcPitch + dx;
dst += ystart * dstPitch + dx; dst += ystart * dstPitch + dx;
const int nvec = width / sizeof(__m256i); const int align = (32 - ((uintptr_t)dst & 31)) & 31;
const int rem = width % sizeof(__m256i); const int nvec = (width - align) / sizeof(__m256i);
const int align = (uintptr_t)dst & 31; const int rem = (width - align) % sizeof(__m256i);
for (int i = ystart; i < yend; ++i) for (int i = ystart; i < yend; ++i)
{ {
// copy the unaligned bytes // copy the unaligned bytes
for(int col = align; col > 0; --col) for(int col = align; col >= 0; --col)
dst[col] = src[col]; dst[col] = src[col];
const __m256i *restrict s = (__m256i*)(src + align); const __m256i *restrict s = (__m256i*)(src + align);
__m256i *restrict d = (__m256i*)ALIGN_TO((uintptr_t)dst, 32); __m256i *restrict d = (__m256i*)(dst + align);
int vec; int vec;
for(vec = nvec; vec > 3; vec -= 4) for(vec = nvec; vec > 3; vec -= 4)