cwpearson/mpi_test
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update readme

Browse Source
This commit is contained in:
Carl Pearson
2021-06-02 16:19:46 -04:00
parent 06192614b0
commit 5000db0508
1 changed files with 10 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
README.md
View File

@@ -10,6 +10,7 @@ If CUDA is detected, additional binaries can be built.
|`one-sided` | Test | MPI | 2 | one-sided communication |
|`one-sided-gpu`| Test | MPI+CUDA | 2 | one-sided communication on GPU buffer |
|`persistent` | Benchmark | MPI | 2 | ping-pong time for persistent Send/Recv |
|`send-recv` | Benchmark | MPI | 2 | ping-pong time for Send/Recv |
## Build
```
@@ -33,6 +34,14 @@ If any tests fails, you can re-run them individually.
Execute any binary you want using `mpirun`, or whatever is appropriate for your platform.
## Examples
### Summit
1 node: `jsrun -n 1 ./persistent`
2 nodes: `jsrun -n 2 -r 1 -a 1 ./persistent`
## Notes on specific platforms
Some Open MPIs use `long long` for their datatypes, which means we can't support ANSI C++ (`-ansi`).
Some Open MPIs use `long long` for their datatypes, which means we can't support ANSI C++ (`-ansi`).