Implementation and Optimization of a CFD Solver Using Overlapped Meshes on Multiple MIC Coprocessors

Algorithm 1

Communication optimization algorithm.

(1)	!$omp parallel do private(idev, ib, … …)
(2)	do idev = 0, 1
(3)	repeat
(4)	offload target(mic:idev): set_boundary_condition for each block
(5)	if(icycle 1) offload target(mic:idev) wait(sgr(idev)): set_CRI_to_domain
(6)	offload target(mic:idev) exchange_interface_data
(7)	do ib = 1, nb(idev)
(8)	offload target(mic:idev): spatial_step
(9)	offload target(mic:idev): temporal_step
(10)	offload target(mic:idev): compute_CPI
(11)	offload_transfer target(mic:idev) out() signal((ib))
(12)	end do
(13)	master thread: offload_wait all related to each device
(14)	master thread on CPU: exchange_interpolation_data
(15)	master thread: offload_transfer target(mic:idev) in() signal(sgr(idev))
(16)	until convergence
(17)	!$omp end parallel do