try rocprof

.github/workflows/amd_workflow.yml

@@ -36,10 +36,13 @@ jobs:
         pip install --upgrade pip
         pip install --pre pytorch-triton-rocm==3.1.0+cf34004b8a torch==2.6.0.dev20241023+rocm6.2 --index-url https://download.pytorch.org/whl/nightly/rocm6.2
 
-    - name: Run script
+    - name: Run script with rocprof
       run: |
         source ${VENV_DIR}/bin/activate
-        python "${{ github.event.inputs.filename }}" > training.log 2>&1
+        mkdir -p rocprof_output
+        rocprof --stats --timestamp on --hip-trace --roctx-trace \
+          --output-dir rocprof_output \
+          python "${{ github.event.inputs.filename }}" > training.log 2>&1
 
     - name: Upload artifacts
       uses: actions/upload-artifact@v3
@@ -49,3 +52,4 @@ jobs:
         path: |
           training.log
           ${{ github.event.inputs.filename }}
+          rocprof_output/