add sve ztrsm

This commit is contained in:
Bine Brank
2022-01-15 22:27:25 +01:00
parent aaa2b1a861
commit f1315288a8
13 changed files with 539 additions and 24 deletions

View File

@@ -58,8 +58,9 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG offset, FLOAT
svbool_t pn = svwhilelt_b64(js, n);
int n_active = svcntp_b64(svptrue_b64(), pn);
#else
int32_t N = n;
int32_t js = 0;
svbool_t pn = svwhilelt_b32(js, n);
svbool_t pn = svwhilelt_b32(js, N);
int n_active = svcntp_b32(svptrue_b32(), pn);
#endif
do {
@@ -83,7 +84,11 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG offset, FLOAT
ii += n_active;
} else {
if (ii < jj) {
#ifdef DOUBLE
svfloat64_t aj_vec = svld1(pn, ao);
#else
svfloat32_t aj_vec = svld1(pn, ao);
#endif
svst1(pn, b, aj_vec);
}
ao += lda;
@@ -103,7 +108,7 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG offset, FLOAT
n_active = svcntp_b64(svptrue_b64(), pn);
} while (svptest_any(svptrue_b64(), pn));
#else
pn = svwhilelt_b32(js, n);
pn = svwhilelt_b32(js, N);
n_active = svcntp_b32(svptrue_b32(), pn);
} while (svptest_any(svptrue_b32(), pn));
#endif