Mirrored_Repos/LookingGlass
1
0
Fork 0
mirror of https://github.com/gnif/LookingGlass.git synced 2024-11-25 23:07:18 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[common] fixed bug in memcpySSE skipping remaining bytes

Browse Source
This commit is contained in:
Geoffrey McRae 2017-12-30 18:39:24 +11:00
parent e854723aa3
commit a6d2fe73ae
1 changed files with 26 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
common/memcpySSE.h
View File

@ -99,26 +99,35 @@ static inline void memcpySSE(void * dst, const void * src, size_t length)
_mm_stream_si128(_dst + 7, v7); _mm_stream_si128(_dst + 7, v7);
} }
const size_t remain = ((length - sselen) & ~0xF) >> 4; const size_t remain = length - sselen;
switch (remain) switch (remain & ~0xF)
{ {
case 7: v0 = _mm_load_si128(_src++); case 112: v0 = _mm_load_si128(_src++);
case 6: v1 = _mm_load_si128(_src++); case 96: v1 = _mm_load_si128(_src++);
case 5: v2 = _mm_load_si128(_src++); case 80: v2 = _mm_load_si128(_src++);
case 4: v3 = _mm_load_si128(_src++); case 64: v3 = _mm_load_si128(_src++);
case 3: v4 = _mm_load_si128(_src++); case 48: v4 = _mm_load_si128(_src++);
case 2: v5 = _mm_load_si128(_src++); case 32: v5 = _mm_load_si128(_src++);
case 1: v6 = _mm_load_si128(_src++); case 16: v6 = _mm_load_si128(_src++);
} }
switch (remain) switch (remain & ~0xF)
{ {
case 7: _mm_stream_si128(_dst++, v0); case 112: _mm_stream_si128(_dst++, v0);
case 6: _mm_stream_si128(_dst++, v1); case 96: _mm_stream_si128(_dst++, v1);
case 5: _mm_stream_si128(_dst++, v2); case 80: _mm_stream_si128(_dst++, v2);
case 4: _mm_stream_si128(_dst++, v3); case 64: _mm_stream_si128(_dst++, v3);
case 3: _mm_stream_si128(_dst++, v4); case 48: _mm_stream_si128(_dst++, v4);
case 2: _mm_stream_si128(_dst++, v5); case 32: _mm_stream_si128(_dst++, v5);
case 1: _mm_stream_si128(_dst++, v6); case 16: _mm_stream_si128(_dst++, v6);
}
// copy any remaining data
if (remain & 0xF)
{
uint8_t * rsrc = (uint8_t *)_src;
uint8_t * rdst = (uint8_t *)_dst;
for (size_t i = remain & 0xF; i > 0; --i)
*rdst++ = *rsrc++;
} }
} }