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

small tuning

Browse Source
This commit is contained in:
evghenii
2013-09-18 20:03:08 +03:00
parent e4b1f58595
commit 3cf63362a4
1 changed files with 3 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
examples/intrinsics/knc-i1x16.h
View File

@@ -496,22 +496,9 @@ static FORCEINLINE bool __none(__vec16_i1 mask) {
return _mm512_kortestz(mask, mask);
}
#if 0
static FORCEINLINE __vec16_i1 __equal_i1(__vec16_i1 a, __vec16_i1 b) {
#if 0
return _mm512_kand(a,b); /* this fails some short circut tests */
#else
return _mm512_knot( _mm512_kandn(a, b)); /* this fails some asin test */
#endif
return _mm512_kxnor(a,b);
}
#else /* passes all the tests */
static FORCEINLINE __vec16_i1 __equal_i1(__vec16_i1 a, __vec16_i1 b) {
__vec16_i1 r;
r.v = (a.v & b.v) | (~a.v & ~b.v);
return r;
}
#endif
static FORCEINLINE __vec16_i1 __and(__vec16_i1 a, __vec16_i1 b) {
return _mm512_kand(a, b);
}
@@ -524,7 +511,6 @@ static FORCEINLINE __vec16_i1 __or(__vec16_i1 a, __vec16_i1 b) {
return _mm512_kor(a, b);
}
#if 0
static FORCEINLINE __vec16_i1 __not(__vec16_i1 a) {
return _mm512_knot(a);
}
@@ -536,30 +522,11 @@ static FORCEINLINE __vec16_i1 __and_not1(__vec16_i1 a, __vec16_i1 b) {
static FORCEINLINE __vec16_i1 __and_not2(__vec16_i1 a, __vec16_i1 b) {
return _mm512_kandnr(a, b);
}
#else
static FORCEINLINE __vec16_i1 __not(__vec16_i1 v) {
__vec16_i1 r;
r.v = ~v.v;
return r;
}
static FORCEINLINE __vec16_i1 __and_not1(__vec16_i1 a, __vec16_i1 b) {
__vec16_i1 r;
r.v = ~a.v & b.v;
return r;
}
static FORCEINLINE __vec16_i1 __and_not2(__vec16_i1 a, __vec16_i1 b) {
__vec16_i1 r;
r.v = a.v & ~b.v;
return r;
}
#endif
static FORCEINLINE __vec16_i1 __select(__vec16_i1 mask, __vec16_i1 a,
__vec16_i1 b) {
return ((a & mask) | (b & ~mask));
//return __or(__and(a, mask), __andnr(b, mask));
// return ((a & mask) | (b & ~mask));
return __or(__and(a, mask), __and_not2(b, mask));
}
static FORCEINLINE __vec16_i1 __select(bool cond, __vec16_i1 a, __vec16_i1 b) {