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

Serialize accesses to parallelized level3 functions from multiple callers 

for #1851
This commit is contained in:
Martin Kroeker 2018-11-19 14:02:50 +01:00 committed by GitHub
parent 76a66eaac8
commit 2e6fae2aad
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 12 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
driver/level3/level3_thread.c
View File

@ -514,6 +514,10 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
*range_n, FLOAT *sa, FLOAT *sb,
BLASLONG nthreads_m, BLASLONG nthreads_n) {
#ifndef USE_OPENMP
static pthread_mutex_t level3_lock = PTHREAD_MUTEX_INITIALIZER;
#endif
blas_arg_t newarg;
#ifndef USE_ALLOC_HEAP
@ -554,6 +558,10 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
#endif
#endif
#ifndef USE_OPENMP
pthread_mutex_lock(&level3_lock);
#endif
#ifdef USE_ALLOC_HEAP
/* Dynamically allocate workspace */
job = (job_t*)malloc(MAX_CPU_NUMBER * sizeof(job_t));
@ -671,6 +679,10 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
free(job);
#endif
#ifndef USE_OPENMP
pthread_mutex_unlock(&level3_lock);
#endif
return 0;
}