Investigate tracking more advanced MPI routines - Githubissues

huttered40 / critter

Critical path analysis of MPI parallel programs

BSD 2-Clause "Simplified" License

2 stars 1 forks source link

Investigate tracking more advanced MPI routines #4

Closed huttered40 closed 4 years ago

huttered40 commented 5 years ago

Currently, we've been focusing on the 5 MPI routines used in cacqr2. The other routines (both those implemented in critter and those that have not been implemented yet) need to either be debugged or implemented.

Add benchmarks for:

[x] MPI_Barrier
[ ] MPI_Gather
[ ] MPI_Gatherv
[ ] MPI_Allgatherv
[ ] MPI_Scatter
[ ] MPI_Scatterv
[ ] MPI_Reduce_scatter
[ ] MPI_Alltoall
[ ] MPI_Alltoallv
[x] MPI_Send
[x] MPI_Recv
[x] MPI_Isend
[x] MPI_Irecv
[x] MPI_Sendrecv

Investigate possible errors in:

[x] MPI_Barrier
[x] MPI_Gather
[x] MPI_Gatherv
[x] MPI_Allgatherv
[x] MPI_Scatter
[x] MPI_Scatterv
[ ] MPI_Reduce_scatter
[x] MPI_Alltoall
[x] MPI_Alltoallv
[x] MPI_Send
[x] MPI_Recv
[x] MPI_Isend
[x] MPI_Irecv
[x] MPI_Sendrecv

Investigate possibility of tracking:

[ ] MPI non-blocking collectives
[ ] MPI one-sided routines

huttered40 commented 4 years ago

I have added support for nonblocking collectives, and various MPI_Send variants. Nonblocking p2p communication has been tested.

I'd rather leave benchmarking these primitives to a separate issue, if at all.