Merge branch 'main' into fast-sub-group-transpose

Author: victor-eds

.github/actions/setup-pytorch/action.yml

@@ -16,6 +16,9 @@ inputs:
   ref:
     description: Branch, tag, commit id
     default: ""
+  mode:
+    description: Source or wheels
+    default: source
 runs:
   using: "composite"
   steps:
@@ -71,7 +74,7 @@ runs:
     - name: Generate PyTorch cache key
       shell: bash
       run: |
-        PYTORCH_CACHE_KEY=$(echo $PYTHON_VERSION $PYTORCH_COMMIT_ID ${{ hashFiles('scripts/patch-pytorch.sh') }} | sha256sum - | cut -d\  -f1)
+        PYTORCH_CACHE_KEY=$(echo $PYTHON_VERSION $PYTORCH_COMMIT_ID ${{ hashFiles('scripts/patch-pytorch.sh') }} ${{ inputs.mode }} | sha256sum - | cut -d\  -f1)
         echo "PYTORCH_CACHE_KEY=$PYTORCH_CACHE_KEY" | tee -a "$GITHUB_ENV"
 
     - name: Load PyTorch from a cache
@@ -90,11 +93,12 @@ runs:
       with:
         repository: ${{ env.PYTORCH_REPO }}
         ref: ${{ env.PYTORCH_COMMIT_ID }}
-        submodules: recursive
+        # To build PyTorch from source we need all submodules, they are not required for benchmarks
+        submodules: ${{ inputs.mode == 'source' && 'recursive' || 'false' }}
         path: pytorch
 
     - name: Apply additional PR patches
-      if: ${{ steps.pytorch-cache.outputs.status == 'miss' && inputs.repository == 'pytorch/pytorch' }}
+      if: ${{ steps.pytorch-cache.outputs.status == 'miss' && inputs.repository == 'pytorch/pytorch' && inputs.mode == 'source' }}
       shell: bash
       run: |
         cd pytorch
@@ -108,7 +112,7 @@ runs:
         pip install 'numpy<2.0.0'
 
     - name: Build PyTorch
-      if: ${{ steps.pytorch-cache.outputs.status == 'miss' }}
+      if: ${{ steps.pytorch-cache.outputs.status == 'miss' && inputs.mode == 'source' }}
       shell: bash
       run: |
         source ${{ inputs.oneapi }}/setvars.sh
@@ -117,11 +121,24 @@ runs:
         pip install -r requirements.txt
         python setup.py bdist_wheel
 
-    - name: Install PyTorch
+    - name: Install PyTorch (built from source)
+      if: ${{ inputs.mode == 'source' }}
       shell: bash
       run: |
         source ${{ inputs.oneapi }}/setvars.sh
         pip install pytorch/dist/*.whl
+
+    - name: Install PyTorch (from wheels)
+      if: ${{ inputs.mode == 'wheels' }}
+      shell: bash
+      run: |
+        source ${{ inputs.oneapi }}/setvars.sh
+        pip install torch --index-url https://download.pytorch.org/whl/nightly/xpu
+
+    - name: Get PyTorch version
+      shell: bash
+      run: |
+        source ${{ inputs.oneapi }}/setvars.sh
         PYTORCH_VERSION="$(python -c 'import torch;print(torch.__version__)')"
         echo "PYTORCH_VERSION=$PYTORCH_VERSION" | tee -a "$GITHUB_ENV"
 

.github/models/accuracy/huggingface.txt

@@ -0,0 +1,10 @@
+BlenderbotSmallForCausalLM
+CamemBert
+DistilBertForMaskedLM
+ElectraForCausalLM
+GPT2ForSequenceClassification
+MobileBertForMaskedLM
+MobileBertForQuestionAnswering
+Speech2Text2ForCausalLM
+T5ForConditionalGeneration
+XLNetLMHeadModel

.github/models/accuracy/timm_models.txt

@@ -0,0 +1,6 @@
+beit_base_patch16_224
+fbnetv3_b
+gmixer_24_224
+mobilenetv3_large_100
+nfnet_l0
+tnt_s_patch16_224

.github/models/accuracy/torchbench.txt

@@ -0,0 +1,4 @@
+resnet50
+hf_T5
+densenet121
+mobilenet_v2

.github/workflows/auto-update-translator-cid.yml

@@ -24,6 +24,7 @@ jobs:
         uses: actions/checkout@v4
         with:
           fetch-depth: 0
+          token: ${{ secrets.BOT_TOKEN }}
 
       # search the opened PR that has branch "bot/update_translator_cid", no new PR as long as the target PR is open
       - name: Search the opened PR that has branch ${{ env.PR_BRANCH }}
@@ -98,8 +99,8 @@ jobs:
         env:
           GITHUB_TOKEN: ${{ secrets.BOT_TOKEN }}
         run: |
-          git config --global user.name 'github-actions[bot]'
-          git config --global user.email 'github-actions[bot]@users.noreply.github.com'
+          git config --global user.name 'glados-intel'
+          git config --global user.email 'glados@intel.com'
 
           if git show-ref --verify --quiet "refs/remotes/origin/${PR_BRANCH}"; then
             echo "${PR_BRANCH} already exist, deleting..."

.github/workflows/build-test-gpu.yml

@@ -12,6 +12,13 @@ on:
         description: PyTorch ref, keep empty for default
         type: string
         default: ""
+      pytorch_mode:
+        description: PyTorch mode, source or wheels
+        type: choice
+        options:
+          - source
+          - wheels
+        default: source
       upload_test_reports:
         description: Upload test reports
         type: boolean
@@ -46,6 +53,7 @@ jobs:
       device: ${{ inputs.runner_label }}
       runner_label: ${{ inputs.runner_label }}
       pytorch_ref: ${{ inputs.pytorch_ref }}
+      pytorch_mode: ${{ inputs.pytorch_mode || 'source' }}
       python_version: ${{ matrix.python }}
       upload_test_reports: ${{ inputs.upload_test_reports }}
       ignore_errors: ${{ inputs.ignore_errors }}

.github/workflows/build-test-reusable.yml

@@ -20,6 +20,10 @@ on:
         description: PyTorch ref, keep empty for default
         type: string
         default: ""
+      pytorch_mode:
+        description: PyTorch mode, source or wheels
+        type: string
+        default: "source"
       python_version:
         description: Python version
         type: string
@@ -96,6 +100,7 @@ jobs:
         with:
           repository: pytorch/pytorch
           ref: ${{ inputs.pytorch_ref }}
+          mode: ${{ inputs.pytorch_mode }}
 
       - name: Install pass_rate dependencies
         run: |
@@ -154,7 +159,6 @@ jobs:
           echo "TRANSFORMERS_VERSION=$TRANSFORMERS_VERSION" | tee -a $GITHUB_ENV
 
       - name: Install transformers
-        if: ${{ inputs.python_version != '3.12' }}
         uses: ./.github/actions/install-dependency
         with:
           package: transformers
@@ -163,7 +167,6 @@ jobs:
           try-tag-prefix: v
 
       - name: Run E2E test
-        if: ${{ inputs.python_version != '3.12' }}
         run: |
           # Set WORKSPACE for inductor_xpu_test.sh to make sure it creates "inductor_log" outside of pytorch cloned directory
           export WORKSPACE=$GITHUB_WORKSPACE

.github/workflows/build-test.yml

@@ -12,6 +12,13 @@ on:
         description: PyTorch ref, keep empty for default
         type: string
         default: ""
+      pytorch_mode:
+        description: PyTorch mode, source or wheels
+        type: choice
+        options:
+          - source
+          - wheels
+        default: source
       upload_test_reports:
         description: Upload test reports
         type: boolean
@@ -120,6 +127,7 @@ jobs:
       driver_version: ${{ matrix.driver }}
       runner_label: ${{ inputs.runner_label }}
       pytorch_ref: ${{ inputs.pytorch_ref }}
+      pytorch_mode: ${{ inputs.pytorch_mode || 'source' }}
       python_version: ${{ matrix.python }}
       upload_test_reports: ${{ inputs.upload_test_reports || false }}
       ignore_errors: ${{ inputs.ignore_errors || false }}

.github/workflows/e2e-reusable.yml

@@ -218,15 +218,15 @@ jobs:
           cd pytorch
 
           # if "only_one_model" is set, then test this model
-          # if "models" == "subset", then test the models from .github/models/performance/{suite}.txt
+          # if "models" == "subset", then test the models from .github/models/{accuracy,performance}/{suite}.txt
           # otherwise test all models.
 
           if [[ "${{ inputs.only_one_model }}" ]]; then
             bash -e $GITHUB_WORKSPACE/scripts/inductor_xpu_test.sh ${{ inputs.suite }} ${{ inputs.dtype }} ${{ inputs.mode }} ${{ inputs.test_mode }} xpu 0 static 1 0 ${{ inputs.only_one_model }}
           elif [[ "${{ inputs.models }}" == "subset" ]]; then
             while read model; do
               bash -e $GITHUB_WORKSPACE/scripts/inductor_xpu_test.sh ${{ inputs.suite }} ${{ inputs.dtype }} ${{ inputs.mode }} ${{ inputs.test_mode }} xpu 0 static 1 0 $model
-            done < $GITHUB_WORKSPACE/.github/models/performance/${{ inputs.suite }}.txt
+            done < $GITHUB_WORKSPACE/.github/models/${{ inputs.test_mode }}/${{ inputs.suite }}.txt
           else
             bash -e $GITHUB_WORKSPACE/scripts/inductor_xpu_test.sh ${{ inputs.suite }} ${{ inputs.dtype }} ${{ inputs.mode }} ${{ inputs.test_mode }} xpu 0 static 1 0
           fi

.github/workflows/integration-tests.yml

@@ -226,7 +226,7 @@ jobs:
       - name: Install pip dependencies
         run: |
           python3 -m pip install --upgrade pip
-          python3 -m pip install wheel cmake==3.24 ninja pytest-xdist lit
+          python3 -m pip install cython setuptools wheel cmake==3.24 ninja pytest-forked pytest-xdist lit
       - name: Install Triton
         env:
           TRITON_BUILD_WITH_CCACHE: "true"
@@ -250,8 +250,9 @@ jobs:
             echo "Coult not find '${SHARED_LIB_DIR}'" ; exit -1
           fi
           cd python/test/unit
-          python3 -m pytest -s -n 8 --ignore=hopper/test_flashattention.py --ignore=language/test_line_info.py --ignore=language/test_subprocess.py
+          python3 -m pytest -s -n 8 --ignore=hopper/test_flashattention.py --ignore=language/test_line_info.py --ignore=language/test_subprocess.py --ignore=test_debug.py
           python3 -m pytest -s -n 8 language/test_subprocess.py
+          python3 -m pytest -s -n 8 test_debug.py --forked
           # Run test_line_info.py separately with TRITON_DISABLE_LINE_INFO=0
           TRITON_DISABLE_LINE_INFO=0 python3 -m pytest -s language/test_line_info.py
           # Run hopper/test_flashattention.py separately to avoid out of gpu memory
@@ -407,7 +408,10 @@ jobs:
           pytest --capture=tee-sys -rfs python/tutorials/06-fused-attention.py
           cd python/test/unit
           pytest --capture=tee-sys -rfs -n 16 language runtime \
-                 --ignore=language/test_line_info.py
+                 --ignore=language/test_line_info.py \
+                 --ignore=test_debug.py
+          # TODO: uncomment
+          # pytest --capture=tee-sys -rfs test_debug.py
           TRITON_ALWAYS_COMPILE=1 TRITON_DISABLE_LINE_INFO=0 LLVM_PASS_PLUGIN_PATH=${SHARED_LIB_DIR}/libGPUHello.so \
           pytest --capture=tee-sys -rfs -vvv instrumentation/test_gpuhello.py