limit max threads

Author: dzzz2001

source/module_hamilt_lcao/module_gint/gint_force_gpu.cu

@@ -92,7 +92,10 @@ void gint_fvl_gpu(const hamilt::HContainer<double>* dm,
                          dm->get_nnr() * sizeof(double),
                          cudaMemcpyHostToDevice));
 
-#pragma omp parallel num_threads(num_streams)
+#ifdef _OPENMP
+const int max_thread_num = std::min(omp_get_max_threads(), num_streams);
+#endif
+#pragma omp parallel num_threads(max_thread_num)
 {
 #ifdef _OPENMP
     const int tid = omp_get_thread_num();

source/module_hamilt_lcao/module_gint/gint_rho_gpu.cu

@@ -72,7 +72,10 @@ void gint_rho_gpu(const hamilt::HContainer<double>* dm,
                          cudaMemcpyHostToDevice));
 
 // calculate the rho for every nbzp bigcells
-#pragma omp parallel num_threads(num_streams)
+#ifdef _OPENMP
+const int max_thread_num = std::min(omp_get_max_threads(), num_streams);
+#endif
+#pragma omp parallel num_threads(max_thread_num)
 {
 #ifdef _OPENMP
     const int tid = omp_get_thread_num();

source/module_hamilt_lcao/module_gint/gint_vl_gpu.cu

@@ -74,7 +74,10 @@ void gint_vl_gpu(hamilt::HContainer<double>* hRGint,
     Cuda_Mem_Wrapper<double*> gemm_B(max_atompair_per_z, num_streams, true);
     Cuda_Mem_Wrapper<double*> gemm_C(max_atompair_per_z, num_streams, true);
 
-#pragma omp parallel num_threads(num_streams)
+#ifdef _OPENMP
+const int max_thread_num = std::min(omp_get_max_threads(), num_streams);
+#endif
+#pragma omp parallel num_threads(max_thread_num)
 {
 #ifdef _OPENMP
     const int tid = omp_get_thread_num();