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

Merge pull request #976 from Vsevolod-Livinskij/new_unordered_func

Browse Source 
New unordered function
This commit is contained in:
Dmitry Babokin
2015-02-26 09:35:57 +03:00
parent f14897101b 78bd4debc6
commit 432570c98f
1 changed files with 13 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
examples/intrinsics/knc.h
View File

@@ -1385,6 +1385,10 @@ static FORCEINLINE __vec16_i1 __unordered_float(__vec16_f a, __vec16_f b) {
return _mm512_cmpunord_ps_mask(a, b); return _mm512_cmpunord_ps_mask(a, b);
} }
static FORCEINLINE __vec16_i1 __unordered_float_and_mask(__vec16_f a, __vec16_f b, __vec16_i1 mask) {
return _mm512_mask_cmpunord_ps_mask(mask, a, b);
}
static FORCEINLINE __vec16_f __select(__vec16_i1 mask, __vec16_f a, __vec16_f b) { static FORCEINLINE __vec16_f __select(__vec16_i1 mask, __vec16_f a, __vec16_f b) {
return _mm512_mask_mov_ps(b, mask, a); return _mm512_mask_mov_ps(b, mask, a);
} }
@@ -1727,6 +1731,15 @@ static FORCEINLINE __vec16_i1 __unordered_double(__vec16_d a, __vec16_d b) {
return _mm512_kmovlhb(ret1, ret2); return _mm512_kmovlhb(ret1, ret2);
} }
static FORCEINLINE __vec16_i1 __unordered_double_and_mask(__vec16_d a, __vec16_d b, __vec16_i1 mask) {
__vec16_i1 ret1;
__vec16_i1 ret2;
__vec16_i1 tmp_m = mask;
ret1 = _mm512_mask_cmpunord_pd_mask(mask, a.v1, b.v1);
ret2 = _mm512_mask_cmpunord_pd_mask(_mm512_kswapb(tmp_m, tmp_m), a.v2, b.v2);
return _mm512_kmovlhb(ret1, ret2);
}
static FORCEINLINE __vec16_d __select(__vec16_i1 mask, __vec16_d a, __vec16_d b) { static FORCEINLINE __vec16_d __select(__vec16_i1 mask, __vec16_d a, __vec16_d b) {
__vec16_d ret; __vec16_d ret;
__vec16_i1 tmp_m = mask; __vec16_i1 tmp_m = mask;