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

packed load/store for knl

Browse Source
This commit is contained in:
Anton Mitrokhin
2015-04-02 17:24:57 +03:00
parent c22ee2b0a0
commit 81251115af
1 changed files with 12 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/intrinsics/knl.h
View File

@@ -3070,8 +3070,7 @@ static FORCEINLINE __vec16_i8 __gather_base_offsets64_i8(uint8_t *_base, uint32_
{
// TODO
__vec16_i8 ret;
for (int i = 0; i < 16; ++i)
if ((mask & (1 << i)) != 0) {
for (int i = 0; i < 16; ++i) {
int8_t *ptr = (int8_t *)(_base + scale * offsets[i]);
ret[i] = *ptr;
}
@@ -3088,10 +3087,9 @@ static FORCEINLINE __vec16_i16 __gather_base_offsets64_i16(uint8_t *_base, uint3
{
// TODO
__vec16_i16 ret;
for (int i = 0; i < 16; ++i)
if ((mask & (1 << i)) != 0) {
for (int i = 0; i < 16; i++) {
int16_t *ptr = (int16_t *)(_base + scale * offsets[i]);
ret[i] = *ptr;
ret.v[i] = *ptr;
}
return ret;
}
@@ -3198,13 +3196,13 @@ static FORCEINLINE void __scatter64_i64(__vec16_i64 ptrs, __vec16_i64 val, __vec
static FORCEINLINE int32_t __packed_load_active(uint32_t *p, __vec16_i32 *val, __vec16_i1 mask) {
__vec16_i32 v = __load<64>(val);
v = _mm512_mask_loadu_epi32(v, mask, p);
v = _mm512_mask_expandloadu_epi32(v, mask, p);
__store<64>(val, v);
return _mm_countbits_32(uint32_t(mask));
}
static FORCEINLINE int32_t __packed_store_active(uint32_t *p, __vec16_i32 val, __vec16_i1 mask) {
_mm512_mask_storeu_epi32(p, mask, val);
_mm512_mask_compressstoreu_epi32(p, mask, val);
return _mm_countbits_32(uint32_t(mask));
}