floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

modify sve zgemmcopy kernels

This commit is contained in:
Bine Brank 2022-01-05 09:07:28 +01:00
parent d30157d891
commit 87537b8c55
2 changed files with 1 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
kernel/arm64/zgemm_ncopy_sve_v1.c
View File

@ -47,7 +47,6 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
IFLOAT *aoffset, *aoffset1, *boffset; IFLOAT *aoffset, *aoffset1, *boffset;
svint64_t lda_vec = svindex_s64(0LL, lda * 2); svint64_t lda_vec = svindex_s64(0LL, lda * 2);
uint64_t sve_size = svcntd();
aoffset = a; aoffset = a;
boffset = b; boffset = b;
@ -67,7 +66,7 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
aoffset1 += 2; aoffset1 += 2;
boffset += active * 2; boffset += active * 2;
} }
aoffset += sve_size * lda * 2; aoffset += active * lda * 2;
j += svcntd(); j += svcntd();
pg = svwhilelt_b64(j, n); pg = svwhilelt_b64(j, n);

2
kernel/arm64/zgemm_tcopy_sve_v1.c
View File

@ -46,8 +46,6 @@ int CNAME(BLASLONG m, BLASLONG n, IFLOAT *a, BLASLONG lda, IFLOAT *b){
BLASLONG j; BLASLONG j;
IFLOAT *aoffset, *aoffset1, *boffset; IFLOAT *aoffset, *aoffset1, *boffset;
uint64_t sve_size = svcntd();
aoffset = a; aoffset = a;
boffset = b; boffset = b;