Explore pinning buffer arrays used in communication

[semi-h]

Pinning the buffer arrays may help with latencies in MPI communication and improve strong scaling performance on CUDA backend.

x3d2/src/backend/cuda/backend.f90

Lines 40 to 46 in 27b14d6

	real(dp), device, allocatable, dimension(:, :, :) :: &
	u_recv_s_dev, u_recv_e_dev, u_send_s_dev, u_send_e_dev, &
	v_recv_s_dev, v_recv_e_dev, v_send_s_dev, v_send_e_dev, &
	w_recv_s_dev, w_recv_e_dev, w_send_s_dev, w_send_e_dev, &
	du_send_s_dev, du_send_e_dev, du_recv_s_dev, du_recv_e_dev, &
	dud_send_s_dev, dud_send_e_dev, dud_recv_s_dev, dud_recv_e_dev, &
	d2u_send_s_dev, d2u_send_e_dev, d2u_recv_s_dev, d2u_recv_e_dev

[semi-h]

Add pinned attribute to these array declarations so that communication is faster
https://developer.nvidia.com/blog/how-optimize-data-transfers-cuda-fortran/