@@ -4,10 +4,12 @@ For each job step launched with a job launcher, this program prints the hardware
## Compiling
To compile, you'll need to have HIP and MPI installed, and use a OpenMP-capable compiler.
To compile, you'll need to have HIP and MPI installed, and use a OpenMP-capable compiler. Modify the Makefile accordingly.
## Usage
To run, simply launch the code with your favorite job launcher.
To run, simply launch the executable with your favorite job launcher.
> OPTIONAL: There is an `example_map.sh` script that can be modified and called "in front of" `hello_jobstep` (or any other executable really). The script uses `numactl` to map hardware threads and GPUs to node-local MPI ranks.
> NOTE: `HIP_VISIBLE_DEVICES` must be set.
> NOTE: On Lyra, the current Slurm doesn't easily allow for fine-grained process/thread placement so an example mapping script is also included in this repo. It can be modifed and called "in front of" `hello_jobstep` (or any other executable really). The script uses `numactl` to map hardware threads and GPUs to node-local MPI ranks.
