moebiusband/MD-Bench
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Remove AVX512 reciprocal usage in AVX2 file

Browse Source 
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
This commit is contained in:
Rafael Ravedutti
2022-11-15 01:40:37 +01:00
parent efa462d0af
commit bc06220aeb
2 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
common/includes/simd/avx2_double.h
View File

@@ -60,7 +60,8 @@ static inline MD_FLOAT simd_incr_reduced_sum(MD_FLOAT *m, MD_SIMD_FLOAT v0, MD_S
}
static inline MD_SIMD_FLOAT select_by_mask(MD_SIMD_FLOAT a, MD_SIMD_MASK m) { return _mm256_and_pd(a, m); }
static inline MD_SIMD_FLOAT simd_reciprocal(MD_SIMD_FLOAT a) { return _mm256_rcp14_pd(a); }
static inline MD_SIMD_FLOAT simd_reciprocal(MD_SIMD_FLOAT a) { return _mm256_cvtps_pd(_mm_rcp_ps(_mm256_cvtpd_ps(a))); }
//static inline MD_SIMD_FLOAT simd_reciprocal(MD_SIMD_FLOAT a) { return _mm256_rcp14_pd(a); }
static inline MD_SIMD_FLOAT simd_fma(MD_SIMD_FLOAT a, MD_SIMD_FLOAT b, MD_SIMD_FLOAT c) { return _mm256_fmadd_pd(a, b, c); }
static inline MD_SIMD_FLOAT simd_masked_add(MD_SIMD_FLOAT a, MD_SIMD_FLOAT b, MD_SIMD_MASK m) { return simd_add(a, _mm256_and_pd(b, m)); }
static inline MD_SIMD_MASK simd_mask_cond_lt(MD_SIMD_FLOAT a, MD_SIMD_FLOAT b) { return _mm256_cmp_pd(a, b, _CMP_LT_OQ); }