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

Fixed missing tmp var needed for use with gather intrinsic

Browse Source
This commit is contained in:
Jean-Luc Duprat
2012-07-11 15:43:11 -07:00
parent 216ac4b1a4
commit df18b2a150
1 changed files with 2 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
examples/intrinsics/knc.h
View File

@@ -1589,11 +1589,6 @@ CAST_BITS_SCALAR(double, int64_t)
/////////////////////////////////////////////////////////////////////////// ///////////////////////////////////////////////////////////////////////////
// various math functions // various math functions
/*
static FORCEINLINE void __fastmath() {
}
*/
static FORCEINLINE float __round_uniform_float(float v) { static FORCEINLINE float __round_uniform_float(float v) {
return roundf(v); return roundf(v);
} }
@@ -1943,7 +1938,7 @@ static FORCEINLINE void __masked_store_blend_float(void *p, __vec16_f val,
static FORCEINLINE __vec16_i32 static FORCEINLINE __vec16_i32
__gather_base_offsets32_i32(uint8_t *base, uint32_t scale, __vec16_i32 offsets, __gather_base_offsets32_i32(uint8_t *base, uint32_t scale, __vec16_i32 offsets,
__vec16_i1 mask) { __vec16_i1 mask) {
// Loop is generated by intrinsic __vec16_i32 tmp = _mm512_undefined_epi32();
__vec16_i32 ret = _mm512_mask_i32extgather_epi32(tmp, mask, offsets, base, __vec16_i32 ret = _mm512_mask_i32extgather_epi32(tmp, mask, offsets, base,
_MM_UPCONV_EPI32_NONE, scale, _MM_UPCONV_EPI32_NONE, scale,
_MM_HINT_NONE); _MM_HINT_NONE);
@@ -1953,7 +1948,7 @@ __gather_base_offsets32_i32(uint8_t *base, uint32_t scale, __vec16_i32 offsets,
static FORCEINLINE __vec16_f static FORCEINLINE __vec16_f
__gather_base_offsets32_float(uint8_t *base, uint32_t scale, __vec16_i32 offsets, __gather_base_offsets32_float(uint8_t *base, uint32_t scale, __vec16_i32 offsets,
__vec16_i1 mask) { __vec16_i1 mask) {
// Loop is generated by intrinsic __vec16_f tmp = _mm512_undefined_ps();
__vec16_f ret = _mm512_mask_i32extgather_ps(tmp, mask, offsets, base, __vec16_f ret = _mm512_mask_i32extgather_ps(tmp, mask, offsets, base,
_MM_UPCONV_PS_NONE, scale, _MM_UPCONV_PS_NONE, scale,
_MM_HINT_NONE); _MM_HINT_NONE);