Aquila-Consortium/libsharp2
0
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

perform autotuning on the fly

Browse source
This commit is contained in:
Martin Reinecke 2012-07-12 16:23:27 +02:00
parent fd69f89ae2
commit 7928e13156
6 changed files with 91 additions and 28 deletions
Download patch file Download diff file Expand all files Collapse all files

6
libsharp/sharp_vecsupport.h
View file

@ -137,10 +137,8 @@ typedef __m256d Tv;
#ifdef __FMA4__
#define vfmaeq(a,b,c) a=_mm256_macc_pd(b,c,a)
#define vfmseq(a,b,c) a=_mm256_nmacc_pd(b,c,a)
#define vfmaaeq(a,b,c,d,e) \
a=_mm256_macc_pd(d,e,_mm256_macc_pd(b,c,a))
#define vfmaseq(a,b,c,d,e) \
a=_mm256_nmacc_pd(d,e,_mm256_macc_pd(b,c,a))
#define vfmaaeq(a,b,c,d,e) a=_mm256_macc_pd(d,e,_mm256_macc_pd(b,c,a))
#define vfmaseq(a,b,c,d,e) a=_mm256_nmacc_pd(d,e,_mm256_macc_pd(b,c,a))
#else
#define vfmaeq(a,b,c) a=_mm256_add_pd(a,_mm256_mul_pd(b,c))
#define vfmseq(a,b,c) a=_mm256_sub_pd(a,_mm256_mul_pd(b,c))