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

Fix potentiol domain error in sqrt

This commit is contained in:
Martin Kroeker 2020-09-05 09:44:33 +02:00 committed by GitHub
parent 9d1ea75aa0
commit 330044d821
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 15 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
driver/level3/level3_syrk_threaded.c
View File

@ -526,7 +526,7 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
BLASLONG width, i, j, k; BLASLONG width, i, j, k;
BLASLONG n, n_from, n_to; BLASLONG n, n_from, n_to;
int mode, mask; int mode, mask;
double dnum; double dnum, di, dinum;
if ((nthreads == 1) || (args -> n < nthreads * SWITCH_RATIO)) { if ((nthreads == 1) || (args -> n < nthreads * SWITCH_RATIO)) {
SYRK_LOCAL(args, range_m, range_n, sa, sb, 0); SYRK_LOCAL(args, range_m, range_n, sa, sb, 0);
@ -601,9 +601,14 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
if (nthreads - num_cpu > 1) { if (nthreads - num_cpu > 1) {
double di = (double)i; di = (double)i;
width = (((BLASLONG)((sqrt(di * di + dnum) - di) + mask)/(mask+1)) * (mask+1) ); dinum = di * di + dnum;
if (dinum > 0)
width = (((BLASLONG)((sqrt(dinum) - di) + mask)/(mask+1)) * (mask+1) );
else
width = (((BLASLONG)(- di + mask)/(mask+1)) * (mask+1) );
if (num_cpu == 0) width = n - (((n - width)/(mask+1)) * (mask+1) ); if (num_cpu == 0) width = n - (((n - width)/(mask+1)) * (mask+1) );
@ -643,9 +648,14 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
if (nthreads - num_cpu > 1) { if (nthreads - num_cpu > 1) {
double di = (double)i; di = (double)i;
dinum = di * di +dnum;
if (dinum > 0)
width = (((BLASLONG)((sqrt(di * di + dnum) - di) + mask)/(mask+1)) * (mask+1)); width = (((BLASLONG)((sqrt(di * di + dnum) - di) + mask)/(mask+1)) * (mask+1));
else
width = (((BLASLONG)(- di + mask)/(mask+1)) * (mask+1));
if ((width > n - i) || (width < mask)) width = n - i; if ((width > n - i) || (width < mask)) width = n - i;