99
1010#include " avx512-16bit-common.h"
1111
12- typedef union
13- {
12+ typedef union {
1413 _Float16 f_;
1514 uint16_t i_;
1615} Fp16Bits;
@@ -74,8 +73,8 @@ struct zmm_vector<_Float16> {
7473 }
7574 static zmm_t mask_mov (zmm_t x, opmask_t mask, zmm_t y)
7675 {
77- return _mm512_castsi512_ph (
78- _mm512_mask_mov_epi16 ( _mm512_castph_si512 (x), mask, _mm512_castph_si512 (y)));
76+ return _mm512_castsi512_ph (_mm512_mask_mov_epi16 (
77+ _mm512_castph_si512 (x), mask, _mm512_castph_si512 (y)));
7978 }
8079 static void mask_storeu (void *mem, opmask_t mask, zmm_t x)
8180 {
@@ -104,8 +103,10 @@ struct zmm_vector<_Float16> {
104103 template <uint8_t mask>
105104 static zmm_t shuffle (zmm_t zmm)
106105 {
107- __m512i temp = _mm512_shufflehi_epi16 (_mm512_castph_si512 (zmm), (_MM_PERM_ENUM)mask);
108- return _mm512_castsi512_ph (_mm512_shufflelo_epi16 (temp, (_MM_PERM_ENUM)mask));
106+ __m512i temp = _mm512_shufflehi_epi16 (_mm512_castph_si512 (zmm),
107+ (_MM_PERM_ENUM)mask);
108+ return _mm512_castsi512_ph (
109+ _mm512_shufflelo_epi16 (temp, (_MM_PERM_ENUM)mask));
109110 }
110111 static void storeu (void *mem, zmm_t x)
111112 {
@@ -123,7 +124,7 @@ X86_SIMD_SORT_INLINE int64_t replace_nan_with_inf(_Float16 *arr,
123124 if (arrsize < 32 ) {
124125 loadmask = (0x00000001 << arrsize) - 0x00000001 ;
125126 in_zmm = _mm512_castsi512_ph (
126- _mm512_maskz_loadu_epi16 (loadmask, arr));
127+ _mm512_maskz_loadu_epi16 (loadmask, arr));
127128 }
128129 else {
129130 in_zmm = _mm512_loadu_ph (arr);
@@ -140,7 +141,7 @@ X86_SIMD_SORT_INLINE int64_t replace_nan_with_inf(_Float16 *arr,
140141X86_SIMD_SORT_INLINE void
141142replace_inf_with_nan (_Float16 *arr, int64_t arrsize, int64_t nan_count)
142143{
143- memset (arr + arrsize - nan_count, 0xFF , nan_count* 2 );
144+ memset (arr + arrsize - nan_count, 0xFF , nan_count * 2 );
144145}
145146
146147template <>
0 commit comments