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

Merge pull request #1875 from martin-frbg/issue1851 

Serialize accesses to parallelized level3 functions from multiple cal…
This commit is contained in:
Martin Kroeker 2018-11-25 20:53:46 +01:00 committed by GitHub
parent 5393759a98 113cb00b95
commit f72fdf525c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 25 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
driver/level3/level3_thread.c
View File

@ -528,6 +528,15 @@ 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
#ifndef OS_WINDOWS
static pthread_mutex_t level3_lock = PTHREAD_MUTEX_INITIALIZER;
#else
CRITICAL_SECTION level3_lock;
InitializeCriticalSection((PCRITICAL_SECTION)&level3_lock);
#endif
#endif
blas_arg_t newarg;
#ifndef USE_ALLOC_HEAP
@ -568,6 +577,14 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
#endif
#endif
#ifndef USE_OPENMP
#ifndef OS_WINDOWS
pthread_mutex_lock(&level3_lock);
#else
EnterCriticalSection((PCRITICAL_SECTION)&level3_lock);
#endif
#endif
#ifdef USE_ALLOC_HEAP
/* Dynamically allocate workspace */
job = (job_t*)malloc(MAX_CPU_NUMBER * sizeof(job_t));
@ -693,6 +710,14 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
free(job);
#endif
#ifndef USE_OPENMP
#ifndef OS_WINDOWS
pthread_mutex_unlock(&level3_lock);
#else
LeaveCriticalSection((PCRITICAL_SECTION)&level3_lock);
#endif
#endif
return 0;
}