Merge pull request #3936 from martin-frbg/issue3933

Observe any stricter (OpenMP) thread count limit imposed by openblas_set_num_threads()

common_thread.h

@@ -53,6 +53,7 @@ extern void goto_set_num_threads(int nthreads);
 /* Global Parameter */
 extern int blas_cpu_number;
 extern int blas_num_threads;
+extern int blas_num_threads_set;
 extern int blas_omp_linked;
 
 #define BLAS_LEGACY	0x8000U
@@ -137,9 +138,13 @@ typedef struct blas_queue {
 extern int blas_server_avail;
 
 static __inline int num_cpu_avail(int level) {
+int openmp_nthreads;
 
 #ifdef USE_OPENMP
-	int openmp_nthreads=omp_get_max_threads();
+	if (blas_num_threads_set == 0)
+	openmp_nthreads=omp_get_max_threads();
+	else
+	openmp_nthreads=blas_cpu_number;
 #endif
 
 #ifndef USE_OPENMP 

driver/others/blas_server_omp.c

@@ -100,6 +100,8 @@ static void adjust_thread_buffers() {
 
 void goto_set_num_threads(int num_threads) {
 
+  blas_num_threads_set = 1;
+  if (num_threads < 0) blas_num_threads_set = 0;
   if (num_threads < 1) num_threads = blas_num_threads;
 
   if (num_threads > MAX_CPU_NUMBER) num_threads = MAX_CPU_NUMBER;

driver/others/memory.c

@@ -422,6 +422,8 @@ This value is equal or large than blas_cpu_number. This means some threads are s
 */
 int blas_num_threads = 0;
 
+int blas_num_threads_set = 0;
+
 int  goto_get_num_procs  (void) {
   return blas_cpu_number;
 }
@@ -1994,6 +1996,8 @@ This value is equal or large than blas_cpu_number. This means some threads are s
 */
 int blas_num_threads = 0;
 
+int blas_num_threads_set = 0;
+
 int  goto_get_num_procs  (void) {
   return blas_cpu_number;
 }

driver/others/memory_qalloc.c

@@ -283,6 +283,7 @@ The numbers of threads in the thread pool.
 This value is equal or large than blas_cpu_number. This means some threads are sleep.
 */
 int blas_num_threads = 0;
+int blas_num_threads_set = 0;
 
 int  goto_get_num_procs  (void) {
   return blas_cpu_number;