libsmm_acc_gpu_blas_init possibly calls omp_get_num_threads outside of parallel region

[dev-zero]

dbcsr/src/acc/libsmm_acc/libsmm_acc_init.cpp

Lines 37 to 41 in 21dae0f

	#if defined _OPENMP
	int nthreads = omp_get_num_threads();
	#else
	int nthreads = 1;
	#endif

dbcsr/src/acc/libsmm_acc/libsmm_acc_init.cpp

Lines 80 to 85 in 21dae0f

	// deallocate memory for acc_blas handles
	#if defined _OPENMP
	int nthreads = omp_get_num_threads();
	#else
	int nthreads = 1;
	#endif

... if the init/finalize is meant to be called from inside a parallel region, then guarding by omp_in_parallel() might be a good idea, but if called from outside a #parallel then we always get one thread only.