Skip to content

Revert changes to CUDA_HOME #41

Revert changes to CUDA_HOME

Revert changes to CUDA_HOME #41

name: MI250X Performance Profiler
on:
push:
branches: [ gpu_abstraction ]
jobs:
Container_Setup_and_Execution:
runs-on: [self-hosted, linux, a100]
name: Container Setup and Execution
steps:
- name: Generate runner token
id: generate_token
run: |
TOKEN=$(curl -XPOST -fsSL \
-H "Authorization: token ${{ secrets.PAT }}" \
-H "Accept: application/vnd.github.v3+json" \
"https://api.github.com/repos/${{ github.repository }}/actions/runners/registration-token" \
| grep -o '"token": *"[^"]*"' | cut -d '"' -f 4)
echo "token=$TOKEN" >> $GITHUB_OUTPUT
- name: SSH and run Docker container
env:
SSH_PRIVATE_KEY: ${{ secrets.SSH_KEY }}
MI250X_PROFILING_HOST: ${{ secrets.MI250X_PROFILING_HOST }}
MI250X_PROFILING_USER: ${{ secrets.MI250X_PROFILING_USER }}
HPC_ACCOUNT: ${{ secrets.HPC_ACCOUNT }}
HPC_PROJECT: ${{ secrets.HPC_PROJECT }}
SINGULARITY_CACHEDIR: /scratch/$HPC_ACCOUNT/$MI250X_PROFILING_USER/
SINGULARITY_TMPDIR: /scratch/$HPC_ACCOUNT/$MI250X_PROFILING_USER/
continue-on-error: true
run: |
echo "$SSH_PRIVATE_KEY" > id_rsa
chmod 600 id_rsa
ssh -o StrictHostKeyChecking=no -i id_rsa $MI250X_PROFILING_USER@$MI250X_PROFILING_HOST "\
cd /scratch/$HPC_ACCOUNT/$MI250X_PROFILING_USER/ && \
singularity pull --force oras://ghcr.io/${{ github.repository_owner }}/github_runner_mi250x:latest && \
srun --account=$HPC_ACCOUNT -p $HPC_PROJECT --gpus=1 --time=03:00:00 singularity run --rocm \
--env GITHUB_TOKEN=${{ steps.generate_token.outputs.token }} \
--env REPO_URL=https://github.com/${{ github.repository }} \
--env RUNNER_NAME=github_runner_mi250x \
--env GITHUB_RUNNER_TAGS='Linux,x64,mi250x' \
--env RUNNER_URL=https://github.com/actions/runner/releases/download/v2.303.0/actions-runner-linux-x64-2.303.0.tar.gz \
github_runner_mi250x_latest.sif"
HIP_MI250X_Profiling:
runs-on: [self-hosted, linux, mi250x]
name: HIP MI250X Profiling
env:
HIP_NAME_PREFIX: hip_AMD-Epyc-7A53_MI250X_gcc-11.2.1_rocm-5.2.3
ENABLE_CI_PROFILER: 1
MADGRAPH4GPU_DB_SECRET: ${{ secrets.MADGRAPH4GPU_DB_SECRET }}
steps:
- uses: actions/checkout@v2
- name: Runs HIP performanceProfiler.py script
run: cd tools/profiling/;
python3 performanceProfiler.py -l 'HIP' -b 'master'
- name: Uploads workplace_mg4gpu directory as an artifact
uses: actions/upload-artifact@v3
with:
name: profiling-results
path: tools/profiling/workplace_mg4gpu
Upload_JSON_files:
needs: HIP_MI250X_Profiling
runs-on: [self-hosted, linux]
name: Upload JSON files to DB
env:
HIP_NAME_PREFIX: hip_AMD-Epyc-7A53_MI250X_gcc-11.2.1_rocm-5.2.3
ENABLE_CI_PROFILER: 1
MADGRAPH4GPU_DB_SECRET: ${{ secrets.MADGRAPH4GPU_DB_SECRET }}
steps:
- uses: actions/checkout@v2
- name: Download artifact containing profiling data
uses: actions/download-artifact@v3
with:
name: profiling-results
path: tools/profiling
- name: Uploads HIP JSON files to DB
run: cd tools/profiling; python3 sendData.py --absLayer HIP --profiler 1 --branch master