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

Merge pull request #4425 from martin-frbg/issue2392 

Add BLAS extension openblas_set_num_threads_local()
This commit is contained in:
Martin Kroeker 2024-01-14 21:57:57 +01:00 committed by GitHub
parent b3341527ad 152a6c43b6
commit a782103b9c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 23 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
cblas.h
View File

@ -12,6 +12,7 @@ extern "C" {
/*Set the number of threads on runtime.*/ /*Set the number of threads on runtime.*/
void openblas_set_num_threads(int num_threads); void openblas_set_num_threads(int num_threads);
void goto_set_num_threads(int num_threads); void goto_set_num_threads(int num_threads);
int openblas_set_num_threads_local(int num_threads);
/*Get the number of threads on runtime.*/ /*Get the number of threads on runtime.*/
int openblas_get_num_threads(void); int openblas_get_num_threads(void);

7
common_thread.h
View File

@ -137,19 +137,20 @@ typedef struct blas_queue {
extern int blas_server_avail; extern int blas_server_avail;
extern int blas_omp_number_max; extern int blas_omp_number_max;
extern int blas_omp_threads_local;
static __inline int num_cpu_avail(int level) { static __inline int num_cpu_avail(int level) {
#ifdef USE_OPENMP #ifdef USE_OPENMP
int openmp_nthreads; int openmp_nthreads;
openmp_nthreads=omp_get_max_threads(); openmp_nthreads=omp_get_max_threads();
if (omp_in_parallel()) openmp_nthreads = blas_omp_threads_local;
#endif #endif
#ifndef USE_OPENMP #ifndef USE_OPENMP
if (blas_cpu_number == 1 if (blas_cpu_number == 1
#endif #else
#ifdef USE_OPENMP if (openmp_nthreads == 1
if (openmp_nthreads == 1 || omp_in_parallel()
#endif #endif
) return 1; ) return 1;

2
driver/others/blas_server.c
View File

@ -113,6 +113,8 @@ extern unsigned int openblas_thread_timeout(void);
/* We need this global for checking if initialization is finished. */ /* We need this global for checking if initialization is finished. */
int blas_server_avail __attribute__((aligned(ATTRIBUTE_SIZE))) = 0; int blas_server_avail __attribute__((aligned(ATTRIBUTE_SIZE))) = 0;
int blas_omp_threads_local = 1;
/* Local Variables */ /* Local Variables */
#if defined(USE_PTHREAD_LOCK) #if defined(USE_PTHREAD_LOCK)
static pthread_mutex_t server_lock = PTHREAD_MUTEX_INITIALIZER; static pthread_mutex_t server_lock = PTHREAD_MUTEX_INITIALIZER;

1
driver/others/blas_server_omp.c
View File

@ -69,6 +69,7 @@
int blas_server_avail = 0; int blas_server_avail = 0;
int blas_omp_number_max = 0; int blas_omp_number_max = 0;
int blas_omp_threads_local = 1;
extern int openblas_omp_adaptive_env(void); extern int openblas_omp_adaptive_env(void);

2
driver/others/blas_server_win32.c
View File

@ -59,6 +59,8 @@ static CRITICAL_SECTION queue_lock;
/* We need this global for checking if initialization is finished. */ /* We need this global for checking if initialization is finished. */
int blas_server_avail = 0; int blas_server_avail = 0;
int blas_omp_threads_local = 1;
/* Local Variables */ /* Local Variables */
static BLASULONG server_lock = 0; static BLASULONG server_lock = 0;

13
driver/others/openblas_set_num_threads.c
View File

@ -36,11 +36,20 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#ifdef SMP_SERVER #ifdef SMP_SERVER
extern void openblas_set_num_threads(int num_threads) ; extern void openblas_set_num_threads(int num_threads) ;
extern int openblas_get_num_threads(void) ;
void openblas_set_num_threads_(int* num_threads){ void openblas_set_num_threads_(int* num_threads){
openblas_set_num_threads(*num_threads); openblas_set_num_threads(*num_threads);
} }
int openblas_set_num_threads_local(int num_threads){
int ret = openblas_get_num_threads();
openblas_set_num_threads(num_threads);
blas_omp_threads_local=num_threads;
return ret;
}
#else #else
//Single thread //Single thread
@ -50,4 +59,8 @@ void openblas_set_num_threads(int num_threads) {
void openblas_set_num_threads_(int* num_threads){ void openblas_set_num_threads_(int* num_threads){
} }
int openblas_set_num_threads_local(int num_threads){
return 1;
}
#endif #endif