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

Merge pull request #1929 from martin-frbg/issue1924 

Avoid taking the root of a negative number in simple threaded syrk
This commit is contained in:
Martin Kroeker 2018-12-23 09:03:58 +01:00 committed by GitHub
parent e23366e860 f343ed65b5
commit 20033f992a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 11 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
driver/level3/syrk_thread.c
View File

@ -48,7 +48,7 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
BLASLONG width, i; BLASLONG width, i;
BLASLONG n_from, n_to; BLASLONG n_from, n_to;
double dnum, nf, nt, di; double dnum, nf, nt, di, dinum;
int num_cpu; int num_cpu;
int mask = 0; int mask = 0;
@ -109,7 +109,11 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
if (nthreads - num_cpu > 1) { if (nthreads - num_cpu > 1) {
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)(( - di + mask)/(mask+1)) * (mask+1);
else
width = (BLASLONG)(( sqrt(dinum) - di + mask)/(mask+1)) * (mask+1);
if ((width <= 0) || (width > n_to - i)) width = n_to - i; if ((width <= 0) || (width > n_to - i)) width = n_to - i;
@ -136,9 +140,7 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
nf = (double)(arg -> n - n_from); nf = (double)(arg -> n - n_from);
nt = (double)(arg -> n - n_to); nt = (double)(arg -> n - n_to);
dnum = (nt * nt - nf * nf) / (double)nthreads; dnum = (nt * nt - nf * nf) / (double)nthreads;
num_cpu = 0; num_cpu = 0;
range[0] = n_from; range[0] = n_from;
@ -149,8 +151,11 @@ int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (
if (nthreads - num_cpu > 1) { if (nthreads - num_cpu > 1) {
di = (double)(arg -> n - i); di = (double)(arg -> n - i);
width = ((BLASLONG)((-sqrt(di * di + dnum) + di) + mask)/(mask+1)) * (mask+1); dinum = di * di + dnum;
if (dinum<0)
width = ((BLASLONG)(di + mask)/(mask+1)) * (mask+1);
else
width = ((BLASLONG)((-sqrt(dinum) + di) + mask)/(mask+1)) * (mask+1);
if ((width <= 0) || (width > n_to - i)) width = n_to - i; if ((width <= 0) || (width > n_to - i)) width = n_to - i;
} else { } else {