Merge 'origin/sycl' into duncan/host-to-device-img-map

Author: DBDuncan

.github/scripts/compute_benchmarks.py

@@ -42,7 +42,7 @@ permissions:
 jobs:
   e2e-build-hw:
     # Run only on upstream; forks will not have the HW
-    if: github.repository == 'oneapi-src/unified-runtime'
+    # if: github.repository == 'oneapi-src/unified-runtime'
     name: Build SYCL, UR, run Compute Benchmarks
     strategy:
       matrix:
@@ -88,6 +88,9 @@ jobs:
       with:
         path: ur-repo
 
+    - name: Install pip packages
+      run: pip install -r ${{github.workspace}}/ur-repo/third_party/requirements.txt
+
     # We need to fetch special ref for proper PR's merge commit. Note, this ref may be absent if the PR is already merged.
     - name: Fetch PR's merge commit
       if: ${{ inputs.pr_no != 0 }}
@@ -131,44 +134,13 @@ jobs:
     - name: Build SYCL
       run: cmake --build ${{github.workspace}}/sycl_build -j
 
-    - name: Set additional env. vars
-      run: |
-        echo "${{github.workspace}}/sycl_build/bin" >> $GITHUB_PATH
-        echo "LD_LIBRARY_PATH=${{github.workspace}}/sycl_build/lib:$LD_LIBRARY_PATH" >> $GITHUB_ENV
-
-    # Running (newly built) sycl-ls sets up some extra variables
-    - name: Setup SYCL variables
-      run: |
-        which clang++ sycl-ls
-        SYCL_PI_TRACE=-1 sycl-ls
-
-    - name: Checkout Compute Benchmarks
-      uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
-      with:
-        repository: intel/compute-benchmarks
-        path: compute-benchmarks-repo
-        submodules: recursive
-
-    - name: Configure Compute Benchmarks
-      run: >
-        cmake
-        -B ${{github.workspace}}/compute-benchmarks-build/
-        -S ${{github.workspace}}/compute-benchmarks-repo/
-        -DCMAKE_BUILD_TYPE=Release
-        -DBUILD_SYCL=ON
-        -DSYCL_COMPILER_ROOT=${{github.workspace}}/sycl_build
-        -DALLOW_WARNINGS=ON
-
-    - name: Build Compute Benchmarks
-      run: cmake --build ${{github.workspace}}/compute-benchmarks-build/ -j
-
     - name: Set oneAPI Device Selector
       run: |
         echo "ONEAPI_DEVICE_SELECTOR=${{ matrix.adapter.str_name }}:${{ matrix.adapter.unit }}" >> $GITHUB_ENV
 
-    - name: Run SYCL API Overhead benchmark
+    - name: Run benchmarks
       id: benchmarks
-      run: ${{ github.workspace }}/ur-repo/.github/scripts/compute_benchmarks.py ${{ github.workspace }}/compute-benchmarks-build/bin/ ${{ inputs.bench_script_params }}
+      run: numactl -N 0 ${{ github.workspace }}/ur-repo/scripts/benchmarks/main.py ~/bench_workdir ${{github.workspace}}/sycl_build ${{ inputs.bench_script_params }}
 
     - name: Add comment to PR
       uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
@@ -188,7 +160,7 @@ jobs:
           const test_status = '${{ steps.benchmarks.outcome }}';
           const job_status = '${{ job.status }}';
           const params = '${{ inputs.bench_script_params }}';
-          const body = `Compute Benchmarks ${adapter} run (with params: ${params}):\n${url}\nJob status: ${job_status}. Test status: ${test_status}.\n ${markdown}`;
+          const body = `Compute Benchmarks ${adapter} run (${params}):\n${url}\nJob status: ${job_status}. Test status: ${test_status}.\n ${markdown}`;
 
           github.rest.issues.createComment({
             issue_number: pr_no,

.github/workflows/benchmarks_compute.yml

@@ -21,9 +21,9 @@ jobs:
       config: ""
       unit: "gpu"
       # Failing tests
-      xfail: "ESIMD/preemption.cpp;Matrix/SG32/element_wise_all_ops.cpp;Matrix/SG32/get_coord_int8_matB.cpp;Matrix/element_wise_all_ops.cpp;Matrix/element_wise_all_ops_1d.cpp;Matrix/element_wise_all_ops_1d_cont.cpp;Matrix/element_wise_all_ops_scalar.cpp;Matrix/element_wise_ops.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_apply_bf16.cpp;Matrix/joint_matrix_apply_two_matrices.cpp;Matrix/joint_matrix_bfloat16.cpp;Matrix/joint_matrix_bfloat16_array.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;ProgramManager/uneven_kernel_split.cpp"
+      xfail: "Matrix/SG32/get_coord_int8_matB.cpp;Matrix/element_wise_all_ops_1d.cpp;Matrix/element_wise_all_ops_1d_cont.cpp;Matrix/element_wise_all_ops_scalar.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;Plugin/level_zero_barrier_optimization.cpp"
       # Flaky tests
       filter_out: "UserDefinedReductions/user_defined_reductions.cpp"
       # These runners by default spawn upwards of 260 workers.
       # We also add a time out just in case some test hangs
-      extra_lit_flags: "-sv -j 100 --max-time 600"
+      extra_lit_flags: "--param gpu-intel-pvc=True -sv -j 100 --max-time 600"

.github/workflows/e2e_level_zero.yml

@@ -0,0 +1,63 @@
+---
+name: Multi Device testing
+
+on: [push, pull_request]
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+permissions:
+  contents: read
+
+jobs:
+  examples:
+    name: Multi Device testing
+    if: github.repository == 'oneapi-src/unified-runtime'  # run only on upstream; forks won't have the HW
+    strategy:
+      matrix:
+        adapter: [
+          {name: L0}
+        ]
+        build_type: [Debug, Release]
+        compiler: [{c: gcc, cxx: g++}] # TODO: investigate why memory-adapter-level_zero hangs with clang
+
+    runs-on: "${{matrix.adapter.name}}_2T"
+
+    steps:
+    - uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
+
+    - name: Install pip packages
+      run: pip install -r third_party/requirements.txt
+
+    # TODO: enable once test failure are fixed/ignored
+    # - name: Download DPC++
+    #   run: |
+    #     wget -O ${{github.workspace}}/dpcpp_compiler.tar.gz https://github.com/intel/llvm/releases/download/nightly-2024-01-29/sycl_linux.tar.gz
+    #     mkdir dpcpp_compiler
+    #     tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
+
+    - name: Configure CMake
+      shell: bash -el {0}
+      run: >
+        cmake
+        -B${{github.workspace}}/build
+        -DCMAKE_C_COMPILER=${{matrix.compiler.c}}
+        -DCMAKE_CXX_COMPILER=${{matrix.compiler.cxx}}
+        -DCMAKE_BUILD_TYPE=${{matrix.build_type}}
+        -DUR_DEVELOPER_MODE=ON
+        -DUR_BUILD_TESTS=ON
+        -DUR_BUILD_ADAPTER_${{matrix.adapter.name}}=ON
+        -DUR_TEST_DEVICES_COUNT=2
+
+    - name: Build
+      run: cmake --build ${{github.workspace}}/build -j $(nproc)
+
+    - name: Test adapter specific
+      working-directory: ${{github.workspace}}/build
+      run: ctest -C ${{matrix.build_type}} --output-on-failure -L "adapter-specific" -E "test-adapter-level_zero_multi_queue" --timeout 180
+      # TODO: investigate why test-adapter-level_zero_multi_queue fails on newer driver
+
+    - name: Test adapters
+      working-directory: ${{github.workspace}}/build
+      run: env UR_CTS_ADAPTER_PLATFORM="${{matrix.adapter.platform}}" ctest -C ${{matrix.build_type}} --output-on-failure -L "conformance" --timeout 180