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

now using the ASSUME_ALIGNED flag in knc.h

Browse Source
This commit is contained in:
Ingo Wald
2012-09-17 15:54:00 +02:00
parent d492af7bc0
commit d2312b1fbd
1 changed files with 5 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
examples/intrinsics/knc.h
View File

@@ -2123,7 +2123,7 @@ static FORCEINLINE __vec16_i32 __masked_load_i32<64>(void *p, __vec16_i1 mask) {
// Currently, when a pseudo_gather is converted into a masked load, it has to be unaligned
static FORCEINLINE __vec16_i32 __masked_load_i32(void *p, __vec16_i1 mask) {
#if 1
#if ASSUME_ALIGNED
# pragma message("masked_load_i32 currently works only for _aligned_ loads")
return _mm512_mask_load_epi32(__vec16_i32(),mask.m,p);
#else
@@ -2165,7 +2165,7 @@ template <> static FORCEINLINE __vec16_f __masked_load_float<64>(void *p, __vec1
#endif
static FORCEINLINE __vec16_f __masked_load_float(void *p, __vec16_i1 mask) {
#if 1
#if ASSUME_ALIGNED
# pragma message("masked_load_i32 currently works only for _aligned_ loads")
return _mm512_mask_load_ps(__vec16_f(),mask.m,p);
#else
@@ -2241,7 +2241,7 @@ template <> static FORCEINLINE void __masked_store_i32<64>(void *p, __vec16_i32
#endif
static FORCEINLINE void __masked_store_i32(void *p, __vec16_i32 val, __vec16_i1 mask) {
#if 1
#if ASSUME_ALIGNED
# pragma message("masked_store_i32 currently works only for _aligned_ stores")
_mm512_mask_store_epi32(p,mask.m,val);
#else
@@ -2279,7 +2279,7 @@ template <> static FORCEINLINE void __masked_store_float<64>(void *p, __vec16_f
#endif
static FORCEINLINE void __masked_store_float(void *p, __vec16_f val, __vec16_i1 mask) {
#if 1
#if ASSUME_ALIGNED
# pragma message("masked_store_float currently works only for _aligned_ stores")
// std::cout << "masked store to " << p << " mask " << ((int*)mask.m) << std::endl;
@@ -2318,7 +2318,7 @@ template <> static FORCEINLINE void __masked_store_double<64>(void *p, __vec16_d
#endif
static FORCEINLINE void __masked_store_double(void *p, __vec16_d val, __vec16_i1 mask) {
#if 1
#if ASSUME_ALIGNED
# pragma message("masked_store_double currently works only for _aligned_ stores")
_mm512_mask_store_pd(((float*)p)+00,mask.m8.m1,val.v1);
_mm512_mask_store_pd(((float*)p)+16,mask.m8.m2,val.v2);