backup/cnc-ddraw
1
0
Fork 0
mirror of https://github.com/FunkyFr3sh/cnc-ddraw.git synced 2025-03-26 02:19:24 +01:00
Code Issues Projects Releases Activity

use AVX for less than 100kb

Browse Source
This commit is contained in:
FunkyFr3sh 2022-09-20 00:39:02 +02:00
parent b98ced6d1e
commit 234448ebee
1 changed files with 51 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
src/blt.c
View File

@ -12,7 +12,9 @@ void blt_copy(
size_t size) size_t size)
{ {
#ifdef __AVX__ #ifdef __AVX__
if (size >= 1024 * 4096 && g_blt_use_avx && !((DWORD)dst % 32) && !((DWORD)src % 32)) if (g_blt_use_avx && !((DWORD)dst % 32) && !((DWORD)src % 32))
{
if (size >= 1024 * 4096)
{ {
_mm_prefetch((const char*)(src), _MM_HINT_NTA); _mm_prefetch((const char*)(src), _MM_HINT_NTA);
@ -45,14 +47,30 @@ void blt_copy(
_mm_sfence(); _mm_sfence();
_mm256_zeroupper(); _mm256_zeroupper();
}
if (size > 0) else if (size < 1024 * 100)
{ {
memcpy(dst, src, size); while (size >= 128)
{
__m256i c0 = _mm256_load_si256(((const __m256i*)src) + 0);
__m256i c1 = _mm256_load_si256(((const __m256i*)src) + 1);
__m256i c2 = _mm256_load_si256(((const __m256i*)src) + 2);
__m256i c3 = _mm256_load_si256(((const __m256i*)src) + 3);
_mm256_store_si256((((__m256i*)dst) + 0), c0);
_mm256_store_si256((((__m256i*)dst) + 1), c1);
_mm256_store_si256((((__m256i*)dst) + 2), c2);
_mm256_store_si256((((__m256i*)dst) + 3), c3);
src += 128;
dst += 128;
size -= 128;
} }
return;
} }
#endif // __AVX__
/* memcpy below handles the remainder */
}
#endif
if (size >= 1024 * 100) if (size >= 1024 * 100)
{ {