[CI] workaround MI300 node failure (#717)

* update workflow

This is a combination of 3 commits.

update

install upstream

rm scripts

* disable upstream

* try rocm label

* use latest

* test MI300 node

* remove upstream install code

Author: micmelesse

.github/workflows/amd_perf_kernel_Integration_tests.yml

@@ -4,13 +4,10 @@ on:
   workflow_dispatch:
   pull_request:
     branches: [main_perf]
-  merge_group:
-    branches: [main_perf]
-    types: [checks_requested]
 
 concurrency:
   group: ${{ github.ref }}
-  cancel-in-progress: ${{ github.ref != 'refs/heads/main_perf' }}
+  cancel-in-progress: true
 
 permissions: read-all
 
@@ -39,24 +36,8 @@ jobs:
             exit 1
           fi
 
-  Runner-Preparation-AMD:
-    runs-on: ubuntu-latest
-    timeout-minutes: 30
-    outputs:
-      matrix-HIP: ${{ steps.set-matrix.outputs.matrix-HIP }}
-    steps:
-      - name: Prepare runner matrix
-        id: set-matrix
-        run: |
-          if [ x"${{ github.repository }}" == x"ROCm/triton" ]; then
-            echo '::set-output name=matrix-HIP::[["self-hosted", "gfx942"]]'
-          else
-            echo '::set-output name=matrix-HIP::[["ubuntu-latest"]]'
-          fi
-
   pre-commit:
     name: pre-commit (code formatting)
-    needs: Runner-Preparation-AMD
     runs-on: ubuntu-latest
     steps:
       - name: Checkout
@@ -92,21 +73,20 @@ jobs:
           git diff
 
   Integration-Tests-AMD:
-    needs: Runner-Preparation-AMD
-    if: needs.Runner-Preparation-AMD.outputs.matrix-HIP != ''
     runs-on: ${{ matrix.runner }}
-    timeout-minutes: 90
     strategy:
       matrix:
-        runner: ${{ fromJson(needs.Runner-Preparation-AMD.outputs.matrix-HIP) }}
-        triton_version: [local, upstream]
+        runner: [gfx942]
       fail-fast: false # disables failing the entire job when one matrix entry fails
     container:
       image: rocm/pytorch:latest
       options: --device=/dev/kfd --device=/dev/dri --security-opt seccomp=unconfined --group-add video --user root
     steps:
       - name: Checkout
         uses: actions/checkout@v4
+      - name: Show Device Info
+        run: |
+          rocminfo | grep gfx
       - name: Clear cache
         run: |
           rm -rf ~/.triton
@@ -119,23 +99,17 @@ jobs:
         run: |
           python3 -m pip install --upgrade pip
           python3 -m pip install lit matplotlib pandas
-      - name: Install Triton (local or upstream)
+      - name: Uninstall Triton
+        run : |
+          pip uninstall -y triton
+          rm -rf ~/.triton
+          rm -rf ./triton/python/build
+      - name: Install Triton
         run: |
-          if [ "${{ matrix.triton_version }}" = "local" ]; then
-            echo "Installing Local Triton..."
-            pip uninstall -y triton
-            cd python
-            pip install -v -e .
-            pip install numpy==1.26.4
-          else
-            echo "Installing Upstream Triton..."
-            pip uninstall -y triton
-            pip install matplotlib pandas pytest
-            git clone https://github.com/triton-lang/triton
-            cd triton
-            pip install --verbose -e python
-            cd ..
-          fi
+          pip uninstall -y triton
+          cd python
+          pip install -v -e .
+          pip install numpy==1.26.4
       - name: Show Triton version
         run: |
           pip show triton