Hybrid MPI and CUDA Parallelization for CFD Applications on Multi-GPU HPC Clusters

Algorithm 2

Asynchronous concurrent execution algorithm of CFD on the GPU.

(1)	cudaStreamCreate (&stream[j]);
(2)	Boundary_Processing_GPU<<<Block_size, Thread_size, stream[0]>>>( );
(3)	Time_Step_GPU<<<Block_size, Thread_size, stream[1]>>>( );
(4)	Grad_Initial<<<Block_Size, Thread_Size, stream[2]>>>( );
(5)	RHS_Initial<<<Block_Size, Thread_Size, stream[3]>>>( );
(6)	cudaDeviceSynchronize ( );
(7)	Grad_Primitive_Variables_GPU<<<Block_size, Thread_size, stream[0]>>>( );
(8)	Convective_Flux_GPU<<<Block_size, Thread_size, stream[1]>>>( );
(9)	cudaDeviceSynchronize ( );
(10)	Viscous_Flux_GPU<<<Block_size, Thread_size, stream[0]>>>( );
(11)	RHS_GPU<<<Block_size, Thread_size, stream[0]>>>( );
(12)	Primitive_Variables_Update_GPU<<<Block_size, Thread_size, stream[0]>>>( );
(13)	cudaDeviceSynchronize ();
(14)	cudaStreamDestroy (stream[j]);