Hui Zhou
Hui Zhou
> In fact, I have not studied the performance benefit of issuing multiple segments. I probably should have done that. My premise is the performance being mostly receiver side bound....
I think the conclusion is we need to use UCX's GPU direct, potentially with a dedicated IPC AM protocol path or wait until UCX am add that feature. With that...
I am pretty sure it is not intentional to try symmetric allocation on the shared window since it only has a single allocation anyway. I believe this is either a...
@ashwinraghu Could you test https://github.com/pmodels/mpich/pull/5966?
> @hzhou there is an additional warning showing up for which I am unsure how to address: > > ``` > src/mpi/datatype/datatype_impl.c:27:46: warning: performing pointer subtraction with a null pointer...
test:mpich/custom netmod: ch4:ofi config: vci
Not quite reliable. While often works, sometime I get ``` ~/work/pull_requests/2204_get_failed/test/mpi/ft$ mpirun -disable-auto-cleanup -l -n 2 ./recvdead [0] MPIR_update_failed_procs: {1} [0] No Errors ~/work/pull_requests/2204_get_failed/test/mpi/ft$ mpirun -disable-auto-cleanup -l -n 2 ./recvdead...
test:mpich/ch3/most test:mpich/ch4/most
test:mpich/ch4/ofi
test:mpich/ch4/ofi