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

Merge branch 'knl-support' of https://github.com/Vsevolod-Livinskij/ispc into knl-support

Browse Source
This commit is contained in:
Anton Mitrokhin
2015-03-26 17:46:30 +03:00
parent d33100a11c 89ded54929
commit dac1ba44e4
1 changed files with 11 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
examples/intrinsics/knl.h
View File

@@ -666,7 +666,6 @@ static FORCEINLINE __vec16_i16 __cast_trunc(__vec16_i16, const __vec16_i32 i32)
__vec16_i16 ret;
//__vec16_i32 i32_trunk = _mm512_and_epi32(i32, __smear_i32<__vec16_i32>(65535));
//_mm512_extstore_epi32(ret.v, i32_trunk, _MM_DOWNCONV_EPI32_UINT16, _MM_HINT_NONE);
_mm512_mask_cvtepi32_storeu_epi16(ret.v, 0xFFFF, i32);
return ret;
}
@@ -2465,8 +2464,7 @@ static FORCEINLINE double __ceil_uniform_double(double v) {
}
static FORCEINLINE __vec16_f __round_varying_float(__vec16_f v) {
return _mm512_round_ps(v, _MM_ROUND_MODE_NEAREST, _MM_EXPADJ_NONE);
return _mm512_cvtepi32_ps(_mm512_cvtps_epi32(v));
}
static FORCEINLINE __vec16_f __floor_varying_float(__vec16_f v) {
@@ -2778,17 +2776,14 @@ static FORCEINLINE __vec16_d __masked_load_double(void *p, __vec16_i1 mask) {
}
static FORCEINLINE void __masked_store_i8(void *p, const __vec16_i8 &val, __vec16_i1 mask) {
__vec16_i32 tmp = _mm512_extload_epi32(&val, _MM_UPCONV_EPI32_SINT8, _MM_BROADCAST32_NONE, _MM_HINT_NONE);
__m128i val_t;
#ifdef ISPC_FORCE_ALIGNED_MEMORY
_mm512_mask_extstore_epi32(p, mask, tmp, _MM_DOWNCONV_EPI32_SINT8,_MM_HINT_NONE);
val_t = _mm_load_si128((__m128i *)val.v);
#else
__vec16_i32 tmp_;
tmp_.v = _mm512_extloadunpacklo_epi32(tmp_.v, p, _MM_UPCONV_EPI32_SINT8, _MM_HINT_NONE);
tmp_.v = _mm512_extloadunpackhi_epi32(tmp_.v, (uint8_t*)p+64, _MM_UPCONV_EPI32_SINT8, _MM_HINT_NONE);
tmp_.v = _mm512_mask_mov_epi32(tmp_.v, mask, tmp.v);
_mm512_extpackstorelo_epi32(p, tmp_.v, _MM_DOWNCONV_EPI32_SINT8, _MM_HINT_NONE);
_mm512_extpackstorehi_epi32((uint8_t*)p+64, tmp_.v, _MM_DOWNCONV_EPI32_SINT8, _MM_HINT_NONE);
val_t = _mm_loadu_si128((__m128i *)val.v);
#endif
__vec16_i32 tmp = _mm512_cvtepi8_epi32(val_t);
_mm512_mask_cvtepi32_storeu_epi8(p, mask, tmp);
}
static FORCEINLINE __vec16_i8 __masked_load_i8(void *p, __vec16_i1 mask) {
@@ -2834,21 +2829,14 @@ static FORCEINLINE void __scatter_base_offsets32_i16(uint8_t *b, uint32_t scale,
static FORCEINLINE void __masked_store_i16(void *p, const __vec16_i16 &val, __vec16_i1 mask) {
__vec16_i32 tmp = _mm512_extload_epi32(&val, _MM_UPCONV_EPI32_SINT16, _MM_BROADCAST32_NONE, _MM_HINT_NONE);
__m256i val_t;
#ifdef ISPC_FORCE_ALIGNED_MEMORY
_mm512_mask_extstore_epi32(p, mask, tmp, _MM_DOWNCONV_EPI32_SINT16, _MM_HINT_NONE);
val_t = _mm256_load_si256((__m256i *)val.v);
#else
#if 0 // TODO: both implementations seem to work, need to test which one is faster
_mm512_mask_i32extscatter_epi32 (p, mask, __vec16_i32(0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15), tmp, _MM_DOWNCONV_EPI32_SINT16, sizeof(uint16_t), _MM_HINT_NONE);
#else
__vec16_i32 tmp_;
tmp_.v = _mm512_extloadunpacklo_epi32(tmp_.v, p, _MM_UPCONV_EPI32_SINT16, _MM_HINT_NONE);
tmp_.v = _mm512_extloadunpackhi_epi32(tmp_.v, (uint8_t*)p+64, _MM_UPCONV_EPI32_SINT16, _MM_HINT_NONE);
tmp_.v = _mm512_mask_mov_epi32(tmp_.v, mask, tmp.v);
_mm512_extpackstorelo_epi32(p, tmp_.v, _MM_DOWNCONV_EPI32_SINT16, _MM_HINT_NONE);
_mm512_extpackstorehi_epi32((uint8_t*)p+64, tmp_.v, _MM_DOWNCONV_EPI32_SINT16, _MM_HINT_NONE);
#endif // if 0
val_t = _mm256_loadu_si256((__m256i *)val.v);
#endif
__vec16_i32 tmp = _mm512_cvtepi16_epi32(val_t);
_mm512_mask_cvtepi32_storeu_epi16(p, mask, tmp);
}
static FORCEINLINE __vec16_i16 __masked_load_i16(void *p, __vec16_i1 mask) {