Merge branch 'sycl' into reuse-shadow-memory

.github/labeler.yml

@@ -31,7 +31,7 @@ loader:
 opencl:
   - changed-files:
     - any-glob-to-any-file:
-      - source/adapter/opencl/**
+      - source/adapters/opencl/**
 
 level-zero:
   - changed-files:

.github/workflows/benchmarks-nightly.yml

@@ -0,0 +1,38 @@
+name: Compute Benchmarks Nightly
+
+on:
+  schedule:
+    - cron: '0 0 * * *'  # Runs at midnight UTC every day
+
+permissions:
+  contents: read
+  pull-requests: write
+
+jobs:
+  nightly:
+    name: Compute Benchmarks Nightly level-zero
+    uses: ./.github/workflows/benchmarks-reusable.yml
+    with:
+      str_name: 'level_zero'
+      unit: 'gpu'
+      pr_no: 0
+      bench_script_params: '--save baseline'
+      sycl_config_params: ''
+      sycl_repo: 'intel/llvm'
+      sycl_commit: ''
+
+  nightly2:
+    # we need to wait until previous job is done so that the html report
+    # contains both runs
+    needs: nightly
+    name: Compute Benchmarks Nightly level-zero v2
+    uses: ./.github/workflows/benchmarks-reusable.yml
+    with:
+        str_name: 'level_zero_v2'
+        unit: 'gpu'
+        pr_no: 0
+        bench_script_params: '--save baseline-v2'
+        sycl_config_params: ''
+        sycl_repo: 'intel/llvm'
+        sycl_commit: ''
+        upload_report: true

.github/workflows/benchmarks_compute.yml → .github/workflows/benchmarks-reusable.yml

@@ -1,50 +1,39 @@
-name: Compute Benchmarks
+name: Benchmarks Reusable
 
 on:
-  # Can be triggered via manual "dispatch" (from workflow view in GitHub Actions tab)
-  workflow_dispatch:
-    # acceptable input for adapter-specific runs
+  workflow_call:
     inputs:
       str_name:
-        description: Formatted adapter name
-        type: choice
         required: true
-        default: 'level_zero'
-        options:
-          - level_zero
-          - level_zero_v2
+        type: string
       unit:
-        description: Test unit (cpu/gpu)
-        type: choice
         required: true
-        default: 'gpu'
-        options:
-          - cpu
-          - gpu
+        type: string
       pr_no:
-        description: PR number (if 0, it'll run on the main)
-        type: number
         required: true
-      bench_script_params:
-        description: Parameters passed to script executing benchmark
+        # even though this is a number, this is a workaround for issues with
+        # reusable workflow calls that result in "Unexpected value '0'" error.
         type: string
+      bench_script_params:
         required: false
+        type: string
         default: ''
       sycl_config_params:
-        description: Extra params for SYCL configuration
-        type: string
         required: false
+        type: string
         default: ''
       sycl_repo:
-        description: 'Compiler repo'
-        type: string
         required: true
+        type: string
         default: 'intel/llvm'
       sycl_commit:
-        description: 'Compiler commit'
-        type: string
         required: false
+        type: string
         default: ''
+      upload_report:
+        required: false
+        type: boolean
+        default: false
 
 permissions:
   contents: read
@@ -56,19 +45,17 @@ jobs:
     strategy:
       matrix:
         adapter: [
-          {str_name: "${{inputs.str_name}}",
-          sycl_config: "${{inputs.sycl_config_params}}",
-          unit: "${{inputs.unit}}"
+          {str_name: "${{ inputs.str_name }}",
+          sycl_config: "${{ inputs.sycl_config_params }}",
+          unit: "${{ inputs.unit }}"
           }
         ]
         build_type: [Release]
         compiler: [{c: clang, cxx: clang++}]
 
-    runs-on: "${{inputs.str_name}}_PERF"
+    runs-on: "${{ inputs.str_name }}_PERF"
 
     steps:
-    # Workspace on self-hosted runners is not cleaned automatically.
-    # We have to delete the files created outside of using actions.
     - name: Cleanup self-hosted workspace
       if: always()
       run: |
@@ -99,7 +86,8 @@ jobs:
         path: ur-repo
 
     - name: Install pip packages
-      run: pip install -r ${{github.workspace}}/ur-repo/third_party/requirements.txt
+      run: |
+        pip install --force-reinstall -r ${{github.workspace}}/ur-repo/third_party/benchmark_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
@@ -169,13 +157,15 @@ jobs:
       run: cmake --install ${{github.workspace}}/ur_build
 
     - name: Run benchmarks
+      working-directory: ${{ github.workspace }}/ur-repo/
       id: benchmarks
       run: >
-        numactl -N 0 ${{ github.workspace }}/ur-repo/scripts/benchmarks/main.py 
-        ~/bench_workdir 
-        ${{github.workspace}}/sycl_build
-        ${{github.workspace}}/ur_install
-        ${{ matrix.adapter.str_name }}
+        numactl -N 0 ${{ github.workspace }}/ur-repo/scripts/benchmarks/main.py
+        ~/bench_workdir
+        --sycl ${{ github.workspace }}/sycl_build
+        --ur ${{ github.workspace }}/ur_install
+        --adapter ${{ matrix.adapter.str_name }}
+        ${{ inputs.upload_report && '--output-html' || '' }}
         ${{ inputs.bench_script_params }}
 
     - name: Add comment to PR
@@ -186,7 +176,7 @@ jobs:
           let markdown = ""
           try {
             const fs = require('fs');
-            markdown = fs.readFileSync('benchmark_results.md', 'utf8');
+            markdown = fs.readFileSync('ur-repo/benchmark_results.md', 'utf8');
           } catch(err) {
           }
 
@@ -204,3 +194,10 @@ jobs:
             repo: context.repo.repo,
             body: body
           })
+
+    - name: Upload HTML report
+      if: ${{ always() && inputs.upload_report }}
+      uses: actions/cache/save@6849a6489940f00c2f30c0fb92c6274307ccb58a # v4.1.2
+      with:
+        path: ur-repo/benchmark_results.html
+        key: benchmark-results-${{ matrix.adapter.str_name }}-${{ github.run_id }}

.github/workflows/benchmarks.yml

@@ -0,0 +1,68 @@
+name: Compute Benchmarks
+
+on:
+  workflow_dispatch:
+    inputs:
+      str_name:
+        description: Formatted adapter name
+        type: choice
+        required: true
+        default: 'level_zero'
+        options:
+          - level_zero
+          - level_zero_v2
+      unit:
+        description: Test unit (cpu/gpu)
+        type: choice
+        required: true
+        default: 'gpu'
+        options:
+          - cpu
+          - gpu
+      pr_no:
+        description: PR number (if 0, it'll run on the main)
+        type: number
+        required: true
+      bench_script_params:
+        description: Parameters passed to script executing benchmark
+        type: string
+        required: false
+        default: ''
+      sycl_config_params:
+        description: Extra params for SYCL configuration
+        type: string
+        required: false
+        default: ''
+      sycl_repo:
+        description: 'Compiler repo'
+        type: string
+        required: true
+        default: 'intel/llvm'
+      sycl_commit:
+        description: 'Compiler commit'
+        type: string
+        required: false
+        default: ''
+      upload_report:
+        description: 'Upload HTML report'
+        type: boolean
+        required: false
+        default: false
+
+permissions:
+  contents: read
+  pull-requests: write
+
+jobs:
+  manual:
+    name: Compute Benchmarks
+    uses: ./.github/workflows/benchmarks-reusable.yml
+    with:
+      str_name: ${{ inputs.str_name }}
+      unit: ${{ inputs.unit }}
+      pr_no: ${{ inputs.pr_no }}
+      bench_script_params: ${{ inputs.bench_script_params }}
+      sycl_config_params: ${{ inputs.sycl_config_params }}
+      sycl_repo: ${{ inputs.sycl_repo }}
+      sycl_commit: ${{ inputs.sycl_commit }}
+      upload_report: ${{ inputs.upload_report }}

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

@@ -41,12 +41,14 @@ jobs:
 
     - name: Build level zero with gcc
       run: |
-        git clone -b v1.17.6 --depth=1 https://github.com/oneapi-src/level-zero.git ${{github.workspace}}/level-zero
+        git clone -b v1.18.5 --depth=1 https://github.com/oneapi-src/level-zero.git ${{github.workspace}}/level-zero
         cd ${{github.workspace}}/level-zero
         cmake -B build -DCMAKE_BUILD_TYPE=${{matrix.build_type}} -DCMAKE_C_COMPILER=gcc -DCMAKE_CXX_COMPILER=g++
         cmake --build build -j $(nproc)
 
     - name: Configure CMake
+      # CFI sanitization (or flto?) seems to cause linking to fail
+      # https://github.com/oneapi-src/unified-runtime/issues/2323
       run: >
         cmake
         -B${{github.workspace}}/build
@@ -58,6 +60,7 @@ jobs:
         -DUR_USE_ASAN=ON
         -DUR_USE_UBSAN=ON
         -DUR_BUILD_ADAPTER_L0=ON
+        -DUR_USE_CFI=OFF
         -DUR_LEVEL_ZERO_LOADER_LIBRARY=${{github.workspace}}/level-zero/build/lib/libze_loader.so
         -DUR_LEVEL_ZERO_INCLUDE_DIR=${{github.workspace}}/level-zero/include/
         -DUR_DPCXX=${{github.workspace}}/dpcpp_compiler/bin/clang++

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

@@ -7,10 +7,15 @@ on:
       adapter_name:
         required: true
         type: string
+      other_adapter_name:
+        required: false
+        type: string
+        default: ""
       runner_name:
         required: true
         type: string
       platform:
+        description: "Platform string, `UR_CTS_ADAPTER_PLATFORM` will be set to this."
         required: false
         type: string
         default: ""
@@ -39,9 +44,13 @@ jobs:
     if: github.repository == 'oneapi-src/unified-runtime'  # run only on upstream; forks won't have the HW
     strategy:
       matrix:
-        adapter: [
-          {name: "${{inputs.adapter_name}}", platform: "${{inputs.platform}}", static_Loader: "${{inputs.static_loader}}", static_adapter: "${{inputs.static_loader}}"},
-        ]
+        adapter: [{
+          name: "${{inputs.adapter_name}}",
+          other_name: "${{inputs.other_adapter_name}}",
+          platform: "${{inputs.platform}}",
+          static_Loader: "${{inputs.static_loader}}",
+          static_adapter: "${{inputs.static_loader}}"
+        }]
         build_type: [Debug, Release]
         compiler: [{c: gcc, cxx: g++}, {c: clang, cxx: clang++}]
         # TODO: The latest L0 loader segfaults when built with clang.
@@ -73,6 +82,8 @@ jobs:
         tar -xvf ${{github.workspace}}/dpcpp_compiler.tar.gz -C dpcpp_compiler
 
     - name: Configure CMake
+      # CFI sanitization seems to fail on our CUDA nodes
+      # https://github.com/oneapi-src/unified-runtime/issues/2309
       run: >
         cmake
         -B${{github.workspace}}/build
@@ -82,8 +93,10 @@ jobs:
         -DUR_ENABLE_TRACING=ON
         -DUR_DEVELOPER_MODE=ON
         -DUR_BUILD_TESTS=ON
-        -DUR_CONFORMANCE_TEST_LOADER=OFF
         -DUR_BUILD_ADAPTER_${{matrix.adapter.name}}=ON
+        -DUR_CONFORMANCE_TEST_LOADER=${{ matrix.adapter.other_name != '' && 'ON' || 'OFF' }}
+        ${{ matrix.adapter.other_name != '' && format('-DUR_BUILD_ADAPTER_{0}=ON', matrix.adapter.other_name) || '' }}
+        -DUR_USE_CFI=${{ matrix.adapter.name == 'CUDA' && 'OFF' || 'ON' }}
         -DUR_STATIC_LOADER=${{matrix.adapter.static_Loader}}
         -DUR_STATIC_ADAPTER_${{matrix.adapter.name}}=${{matrix.adapter.static_adapter}}
         -DUR_DPCXX=${{github.workspace}}/dpcpp_compiler/bin/clang++
@@ -103,6 +116,8 @@ jobs:
     - name: Test adapter specific
       working-directory: ${{github.workspace}}/build
       run: ctest -C ${{matrix.build_type}} --output-on-failure -L "adapter-specific" --timeout 180
+      # Don't run adapter specific tests when building multiple adapters
+      if: ${{ matrix.adapter.other_name == '' }}
 
     - name: Test adapters
       working-directory: ${{github.workspace}}/build