diff options
author | Gengxin Xie <gengxin.xie@intel.com> | 2020-03-19 14:03:11 +0800 |
---|---|---|
committer | Gengxin Xie <gengxin.xie@intel.com> | 2020-03-19 14:03:11 +0800 |
commit | 90f744f1a233b7b554d94dba47df499e750fd481 (patch) | |
tree | 0110c603b6d6fbd8ef3d0264834fcaeee037c728 /numpy | |
parent | 6964479b702119d1a2dfdf08579ac0353ed39bdd (diff) | |
download | numpy-90f744f1a233b7b554d94dba47df499e750fd481.tar.gz |
MANIT: Change some variables in AVX512F_exp_DOUBLE
Diffstat (limited to 'numpy')
-rw-r--r-- | numpy/core/src/umath/simd.inc.src | 16 |
1 files changed, 8 insertions, 8 deletions
diff --git a/numpy/core/src/umath/simd.inc.src b/numpy/core/src/umath/simd.inc.src index 0830255c7..94828c459 100644 --- a/numpy/core/src/umath/simd.inc.src +++ b/numpy/core/src/umath/simd.inc.src @@ -2710,10 +2710,10 @@ AVX512F_exp_DOUBLE(npy_double * op, const npy_intp steps) { npy_intp num_remaining_elements = array_size; - const npy_intp stride = steps / sizeof(npy_double); - const npy_int num_lanes = 64 / sizeof(npy_double); - npy_int indexarr[8]; - for (npy_int ii = 0; ii < 8; ii++) { + const npy_intp stride = steps / (npy_intp)sizeof(npy_double); + const npy_int num_lanes = 64 / (npy_intp)sizeof(npy_double); + npy_int32 indexarr[8]; + for (npy_int32 ii = 0; ii < 8; ii++) { indexarr[ii] = ii*stride; } @@ -2735,12 +2735,12 @@ AVX512F_exp_DOUBLE(npy_double * op, __m512d ones_d = _mm512_set1_pd(1.0f); __m256i vindex = _mm256_loadu_si256((__m256i*)&indexarr[0]); - __m512d mTable_top_0 = _mm512_loadu_pd(&(EXP_Table_top[0])); - __m512d mTable_top_1 = _mm512_loadu_pd(&(EXP_Table_top[8])); + __m512d mTable_top_0 = _mm512_loadu_pd(&(EXP_Table_top[8*0])); + __m512d mTable_top_1 = _mm512_loadu_pd(&(EXP_Table_top[8*1])); __m512d mTable_top_2 = _mm512_loadu_pd(&(EXP_Table_top[8*2])); __m512d mTable_top_3 = _mm512_loadu_pd(&(EXP_Table_top[8*3])); - __m512d mTable_tail_0 = _mm512_loadu_pd(&(EXP_Table_tail[0])); - __m512d mTable_tail_1 = _mm512_loadu_pd(&(EXP_Table_tail[8])); + __m512d mTable_tail_0 = _mm512_loadu_pd(&(EXP_Table_tail[8*0])); + __m512d mTable_tail_1 = _mm512_loadu_pd(&(EXP_Table_tail[8*1])); __m512d mTable_tail_2 = _mm512_loadu_pd(&(EXP_Table_tail[8*2])); __m512d mTable_tail_3 = _mm512_loadu_pd(&(EXP_Table_tail[8*3])); |