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

Keep both PGI/SUN and default code paths to avoid breaking Clang/WIndows

This commit is contained in:
Martin Kroeker 2019-08-28 18:07:44 +02:00 committed by GitHub
parent 3a55dca2dc
commit 11c59acfb1
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 17 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
kernel/x86_64/zdot.c
View File

@ -181,11 +181,19 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
#if defined(SMP) #if defined(SMP)
int nthreads; int nthreads;
FLOAT dummy_alpha; FLOAT dummy_alpha;
#if defined(C_PGI) || defined(C_SUN)
FLOAT zdotr=0., zdoti=0.; FLOAT zdotr=0., zdoti=0.;
#endif
#endif #endif
OPENBLAS_COMPLEX_FLOAT zdot; OPENBLAS_COMPLEX_FLOAT zdot;
#if defined(C_PGI) || defined(C_SUN)
zdot=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0); zdot=OPENBLAS_MAKE_COMPLEX_FLOAT(0.0,0.0);
#else
CREAL(zdot) = 0.0;
CIMAG(zdot) = 0.0;
#endif
#if defined(SMP) #if defined(SMP)
if (inc_x == 0 || inc_y == 0 || n <= 10000) if (inc_x == 0 || inc_y == 0 || n <= 10000)
nthreads = 1; nthreads = 1;
@ -211,17 +219,23 @@ OPENBLAS_COMPLEX_FLOAT CNAME(BLASLONG n, FLOAT *x, BLASLONG inc_x, FLOAT *y, BLA
ptr = (OPENBLAS_COMPLEX_FLOAT *)result; ptr = (OPENBLAS_COMPLEX_FLOAT *)result;
for (i = 0; i < nthreads; i++) { for (i = 0; i < nthreads; i++) {
#if defined(C_PGI) || defined(C_SUN)
zdotr += CREAL(*ptr); zdotr += CREAL(*ptr);
zdoti += CIMAG(*ptr); zdoti += CIMAG(*ptr);
// CREAL(zdot) = CREAL(zdot) + CREAL(*ptr); #else
// CIMAG(zdot) = CIMAG(zdot) + CIMAG(*ptr); CREAL(zdot) = CREAL(zdot) + CREAL(*ptr);
CIMAG(zdot) = CIMAG(zdot) + CIMAG(*ptr);
#endif
ptr = (void *)(((char *)ptr) + sizeof(double) * 2); ptr = (void *)(((char *)ptr) + sizeof(double) * 2);
} }
#if defined(C_PGI) || defined(C_SUN)
zdot = OPENBLAS_MAKE_COMPLEX_FLOAT(zdotr,zdoti); zdot = OPENBLAS_MAKE_COMPLEX_FLOAT(zdotr,zdoti);
#endif
} }
#else #else
zdot_compute(n, x, inc_x, y, inc_y, &zdot); zdot_compute(n, x, inc_x, y, inc_y, &zdot);
#endif #endif
return zdot; return zdot;
} }