Merge pull request #4163 from martin-frbg/issue4017

Rework OpenMP thread count limit handling

Author: martin-frbg

common_thread.h

@@ -53,7 +53,6 @@ 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
@@ -136,15 +135,13 @@ typedef struct blas_queue {
 #ifdef SMP_SERVER
 
 extern int blas_server_avail;
+extern int blas_omp_number_max;
 
 static __inline int num_cpu_avail(int level) {
 
 #ifdef USE_OPENMP
 int openmp_nthreads;
-	if (blas_num_threads_set == 0)
 	openmp_nthreads=omp_get_max_threads();
-	else
-	openmp_nthreads=blas_cpu_number;
 #endif
 
 #ifndef USE_OPENMP 
@@ -156,7 +153,13 @@ int openmp_nthreads;
       ) return 1;        
 
 #ifdef USE_OPENMP
-  if (blas_cpu_number != openmp_nthreads) {
+     if (openmp_nthreads > blas_omp_number_max){
+#ifdef DEBUG
+     fprintf(stderr,"WARNING - more OpenMP threads requested (%d) than available (%d)\n",openmp_nthreads,blas_omp_number_max);
+#endif
+     openmp_nthreads = blas_omp_number_max;
+     }
+     if (blas_cpu_number != openmp_nthreads) {
 	  goto_set_num_threads(openmp_nthreads);
   }
 #endif

driver/others/blas_server_omp.c

@@ -68,6 +68,7 @@
 #endif
 
 int blas_server_avail = 0;
+int blas_omp_number_max = 0;
 
 extern int openblas_omp_adaptive_env();
 
@@ -100,8 +101,6 @@ 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;
@@ -125,6 +124,8 @@ void openblas_set_num_threads(int num_threads) {
 }
 
 int blas_thread_init(void){
+if(blas_omp_number_max <= 0)
+  blas_omp_number_max = omp_get_max_threads();
 
   blas_get_cpu_number();
 

driver/others/memory.c

@@ -422,8 +422,6 @@ 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;
 }
@@ -1996,8 +1994,6 @@ 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

@@ -283,7 +283,6 @@ 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;