aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

store_ps fixed

Browse Source
This commit is contained in:
Anton Mitrokhin
2015-03-25 19:12:28 +03:00
parent fc8a32425c
commit 1aa8309a9e
1 changed files with 3 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
examples/intrinsics/knl.h
View File

@@ -1455,8 +1455,7 @@ template <int ALIGN> static FORCEINLINE void __store(__vec16_f *p, __vec16_f v)
#ifdef ISPC_FORCE_ALIGNED_MEMORY #ifdef ISPC_FORCE_ALIGNED_MEMORY
_mm512_store_ps(p, v); _mm512_store_ps(p, v);
#else #else
_mm512_extpackstorelo_ps(p, v, _MM_DOWNCONV_PS_NONE, _MM_HINT_NONE); _mm512_storeu_ps(p, v);
_mm512_extpackstorehi_ps((uint8_t*)p+64, v, _MM_DOWNCONV_PS_NONE, _MM_HINT_NONE);
#endif #endif
} }
#if 0 #if 0
@@ -3106,17 +3105,11 @@ static FORCEINLINE void __masked_store_i32(void *p, __vec16_i32 val, __vec16_i1
#endif #endif
} }
static FORCEINLINE void __masked_store_float(void *p, __vec16_f val, static FORCEINLINE void __masked_store_float(void *p, __vec16_f val, __vec16_i1 mask) {
__vec16_i1 mask) {
#ifdef ISPC_FORCE_ALIGNED_MEMORY #ifdef ISPC_FORCE_ALIGNED_MEMORY
_mm512_mask_store_ps(p, mask, val.v); _mm512_mask_store_ps(p, mask, val.v);
#else #else
__vec16_f tmp; _mm512_mask_storeu_ps(p, mask, val.v);
tmp.v = _mm512_extloadunpacklo_ps(tmp.v, p, _MM_UPCONV_PS_NONE, _MM_HINT_NONE);
tmp.v = _mm512_extloadunpackhi_ps(tmp.v, (uint8_t*)p+64, _MM_UPCONV_PS_NONE, _MM_HINT_NONE);
tmp.v = _mm512_mask_mov_ps(tmp.v, mask, val.v);
_mm512_extpackstorelo_ps(p, tmp.v, _MM_DOWNCONV_PS_NONE, _MM_HINT_NONE);
_mm512_extpackstorehi_ps((uint8_t*)p+64, tmp.v, _MM_DOWNCONV_PS_NONE, _MM_HINT_NONE);
#endif #endif
} }