diff --git a/src/ZSOLVER/tensorproductmultigrid_Source/communication.f90 b/src/ZSOLVER/tensorproductmultigrid_Source/communication.f90 index 3d33cf8b80fac89c56298a4a805e7074df25ac36..4273ddadf026988dfcb71676ddad44a92e4cec2b 100644 --- a/src/ZSOLVER/tensorproductmultigrid_Source/communication.f90 +++ b/src/ZSOLVER/tensorproductmultigrid_Source/communication.f90 @@ -547,6 +547,7 @@ contains end if if (LUseT) then local_sumt = 0.0_rl + !$acc kernels loop collapse(3) do iz=0,nz+1 do iy=a%icompy_min,a%icompy_max do ix=a%icompx_min,a%icompx_max @@ -1578,7 +1579,7 @@ contains ! Copy local data while waiting for data from other processes if (LUseO) b%s(0:nz+1,1:a_n,1:a_n) = a%s(0:nz+1,1:a_n,1:a_n) if (LUseT) then -#ifdef MNH_GPUDIREC +#ifdef MNH_GPUDIRECT zb_st => b%st za_st => a%st !$acc parallel loop collapse(3) @@ -1882,7 +1883,7 @@ contains ! While sending, copy local data if (LUseO) b%s(0:nz+1,1:b_n,1:b_n) = a%s(0:nz+1,1:b_n,1:b_n) if (LUseT) then -#ifdef MNH_GPUDIREC +#ifdef MNH_GPUDIRECT zb_st => b%st za_st => a%st !$acc parallel loop collapse(3) @@ -1911,7 +1912,7 @@ contains if (LUseO) call mpi_recv(b%s(0,1,1),1,interior(level,m),source_rank,recv_tag,MPI_COMM_HORIZ,stat,ierr) recv_tag = 1010 if (LUseT) then -#ifdef MNH_GPUDIREC +#ifdef MNH_GPUDIRECT ztab_interiorT_ne_m_haloTout => tab_interiorT_ne(level,m)%haloTout !$acc host_data use_device(ztab_interiorT_ne_m_haloTout) call mpi_recv(ztab_interiorT_ne_m_haloTout,size(ztab_interiorT_ne_m_haloTout), & @@ -1977,7 +1978,7 @@ contains if (LUseT) then #ifdef MNH_GPUDIRECT ztab_interiorT_se_m_haloTout => tab_interiorT_se(level,m)%haloTout - !$acc host_data use_device(ztab_interiorT_ne_m_haloTout) + !$acc host_data use_device(ztab_interiorT_se_m_haloTout) call mpi_recv(ztab_interiorT_se_m_haloTout,size(ztab_interiorT_se_m_haloTout), & MPI_DOUBLE_PRECISION,source_rank,recv_tag,MPI_COMM_HORIZ,stat,ierr) !$acc end host_data