Skip to content

Fix GPU aware MPI #202

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
LudwigBoess wants to merge 2 commits into
base: 1.4.0rc
Choose a base branch
from
Draft

Fix GPU aware MPI #202

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2aeef37
add explicit Kokkos::fence() before mpi to secure against problem wit…
LudwigBoess May 8, 2026
2e16419
remove duplicate `em` comm
LudwigBoess May 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 15 additions & 0 deletions src/framework/containers/particles_comm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,6 +28,11 @@ namespace ntt {
npart_t nsend,
npart_t nrecv,
npart_t offset) {
#if defined(DEVICE_ENABLED)
// guard for Intel GPUs.
// Should be a null-operation for other architectures.
Kokkos::fence();
#endif
#if !defined(DEVICE_ENABLED) || defined(GPU_AWARE_MPI)
MPI_Sendrecv(send_arr.data(),
nsend,
Expand Down Expand Up @@ -98,6 +103,11 @@ namespace ntt {

template <typename T>
void send(array_t<T*>& send_arr, int send_rank, npart_t nsend) {
#if defined(DEVICE_ENABLED)
// guard for Intel GPUs.
// Should be a null-operation for other architectures.
Kokkos::fence();
#endif
#if !defined(DEVICE_ENABLED) || defined(GPU_AWARE_MPI)
MPI_Send(send_arr.data(), nsend, mpi::get_type<T>(), send_rank, 0, MPI_COMM_WORLD);
#else
Expand All @@ -109,6 +119,11 @@ namespace ntt {

template <typename T>
void recv(array_t<T*>& recv_arr, int recv_rank, npart_t nrecv, npart_t offset) {
#if defined(DEVICE_ENABLED)
// guard for Intel GPUs.
// Should be a null-operation for other architectures.
Kokkos::fence();
#endif
#if !defined(DEVICE_ENABLED) || defined(GPU_AWARE_MPI)
MPI_Recv(recv_arr.data() + offset,
nrecv,
Expand Down
15 changes: 15 additions & 0 deletions src/framework/domain/comm_mpi.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,6 +33,11 @@ namespace comm {
int recv_rank,
ncells_t nsend,
ncells_t nrecv) {
#if defined(DEVICE_ENABLED)
// guard for Intel GPUs.
// Should be a null-operation for other architectures.
Kokkos::fence();
#endif
#if !defined(DEVICE_ENABLED) || defined(GPU_AWARE_MPI)
MPI_Sendrecv(send_arr.data(),
nsend,
Expand Down Expand Up @@ -68,6 +73,11 @@ namespace comm {

template <unsigned short D>
void send(ndarray_t<D>& send_arr, int send_rank, ncells_t nsend) {
#if defined(DEVICE_ENABLED)
// guard for Intel GPUs.
// Should be a null-operation for other architectures.
Kokkos::fence();
#endif
#if !defined(DEVICE_ENABLED) || defined(GPU_AWARE_MPI)
MPI_Send(send_arr.data(), nsend, mpi::get_type<real_t>(), send_rank, 0, MPI_COMM_WORLD);
#else
Expand All @@ -84,6 +94,11 @@ namespace comm {

template <unsigned short D>
void recv(ndarray_t<D>& recv_arr, int recv_rank, ncells_t nrecv) {
#if defined(DEVICE_ENABLED)
// guard for Intel GPUs.
// Should be a null-operation for other architectures.
Kokkos::fence();
#endif
#if !defined(DEVICE_ENABLED) || defined(GPU_AWARE_MPI)
MPI_Recv(recv_arr.data(),
nrecv,
Expand Down
13 changes: 0 additions & 13 deletions src/framework/domain/metadomain_comm.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -349,19 +349,6 @@ namespace ntt {
false);
}
} else {
if (comm_em) {
comm::CommunicateField<M::Dim, 6>(domain.index(),
domain.fields.em,
domain.fields.em,
send_ind,
recv_ind,
send_rank,
recv_rank,
send_slice,
recv_slice,
comp_range_fld,
false);
}
if (comm_j) {
comm::CommunicateField<M::Dim, 3>(domain.index(),
domain.fields.cur,
Expand Down
Loading