Mirrored_Repos/LookingGlass
1
0
Fork 0
mirror of https://github.com/gnif/LookingGlass.git synced 2025-10-12 18:38:12 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[common] add AVX/AVX2 memory copy implementations

Browse Source
This commit is contained in:
Geoffrey McRae
2023-11-19 00:09:42 +11:00
parent 6357df1a7a
commit e61678ef1b
3 changed files with 124 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
common/src/rects.c
View File

@@ -22,6 +22,45 @@
#include "common/util.h"
#include <stdlib.h>
#include <immintrin.h>
__attribute__((target("default")))
void rectCopyUnaligned(uint8_t * dst, const uint8_t * src,
int ystart, int yend, int dx, int dstPitch, int srcPitch, int width)
{
src += ystart * srcPitch + dx;
dst += ystart * dstPitch + dx;
for (int i = ystart; i < yend; ++i)
{
memcpy(dst, src, width);
src += srcPitch;
dst += dstPitch;
}
}
__attribute__((target("avx")))
void rectCopyUnaligned(uint8_t * dst, const uint8_t * src,
int ystart, int yend, int dx, int dstPitch, int srcPitch, int width)
{
src += ystart * srcPitch + dx;
dst += ystart * dstPitch + dx;
for (int i = ystart; i < yend; ++i)
{
int col;
for(col = 0; col <= width - 32; col += 32)
{
_mm_prefetch(src + col + 256, _MM_HINT_T0);
__m256i srcData = _mm256_loadu_si256((__m256i*)(src + col));
_mm256_storeu_si256((__m256i*)(dst + col), srcData);
}
for(; col < width; ++col)
dst[col] = src[col];
src += srcPitch;
dst += dstPitch;
}
}
struct Corner
{