Merge branch 'main' into georgi/reintroduce-catch-bad-alloc-cuda

oneapi-src · Sep 25, 2024 · d07ea63 · d07ea63
2 parents 5b73907 + 17aa04d
commit d07ea63
Show file tree

Hide file tree

Showing 290 changed files with 10,018 additions and 6,231 deletions.
diff --git a/.github/workflows/benchmarks_compute.yml b/.github/workflows/benchmarks_compute.yml
@@ -12,6 +12,7 @@ on:
         default: 'level_zero'
         options:
           - level_zero
+          - level_zero_v2
       unit:
         description: Test unit (cpu/gpu)
         type: choice
@@ -142,17 +143,25 @@ jobs:
         --ci-defaults ${{matrix.adapter.sycl_config}}
         --cmake-opt="-DLLVM_INSTALL_UTILS=ON"
         --cmake-opt="-DSYCL_PI_TESTS=OFF"
-        --cmake-opt="-DSYCL_PI_UR_USE_FETCH_CONTENT=OFF"
-        --cmake-opt="-DSYCL_PI_UR_SOURCE_DIR=${{github.workspace}}/ur-repo/"
+        --cmake-opt="-DSYCL_UR_USE_FETCH_CONTENT=OFF"
+        --cmake-opt="-DSYCL_UR_SOURCE_DIR=${{github.workspace}}/ur-repo/"
         --cmake-opt=-DCMAKE_C_COMPILER_LAUNCHER=ccache
         --cmake-opt=-DCMAKE_CXX_COMPILER_LAUNCHER=ccache
 
     - name: Build SYCL
       run: cmake --build ${{github.workspace}}/sycl_build -j
 
-    - name: Set oneAPI Device Selector
-      run: |
-        echo "ONEAPI_DEVICE_SELECTOR=${{ matrix.adapter.str_name }}:${{ matrix.adapter.unit }}" >> $GITHUB_ENV
+    - name: Configure UR
+      working-directory: ${{github.workspace}}/ur-repo
+      run: >
+        cmake -DCMAKE_BUILD_TYPE=Release
+        -B${{github.workspace}}/ur-repo/build
+        -DUR_BUILD_TESTS=OFF
+        -DUR_BUILD_ADAPTER_L0=ON
+        -DUR_BUILD_ADAPTER_L0_V2=ON
+
+    - name: Build UR
+      run: cmake --build ${{github.workspace}}/ur-repo/build -j $(nproc)
 
     - name: Run benchmarks
       id: benchmarks

diff --git a/.github/workflows/build-hw-reusable.yml b/.github/workflows/build-hw-reusable.yml
@@ -18,6 +18,10 @@ on:
         required: false
         type: string
         default: OFF
+      static_adapter:
+        required: false
+        type: string
+        default: OFF
 
 permissions:
   contents: read
@@ -36,7 +40,7 @@ jobs:
     strategy:
       matrix:
         adapter: [
-          {name: "${{inputs.adapter_name}}", platform: "${{inputs.platform}}", static_Loader: "${{inputs.static_loader}}"},
+          {name: "${{inputs.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++}]
@@ -49,6 +53,10 @@ jobs:
            build_type: Release
          - adapter: {static_Loader: ON}
            compiler: {c: clang, cxx: clang++}
+         - adapter: {static_adapter: ON}
+           build_type: Release
+         - adapter: {static_adapter: ON}
+           compiler: {c: clang, cxx: clang++}
 
     runs-on: ${{inputs.runner_name}}
 
@@ -76,6 +84,7 @@ jobs:
         -DUR_BUILD_TESTS=ON
         -DUR_BUILD_ADAPTER_${{matrix.adapter.name}}=ON
         -DUR_STATIC_LOADER=${{matrix.adapter.static_Loader}}
+        -DUR_STATIC_ADAPTER_${{matrix.adapter.name}}=${{matrix.adapter.static_adapter}}
         -DUR_DPCXX=${{github.workspace}}/dpcpp_compiler/bin/clang++
         -DUR_SYCL_LIBRARY_DIR=${{github.workspace}}/dpcpp_compiler/lib
         ${{ matrix.adapter.name == 'HIP' && '-DUR_CONFORMANCE_AMD_ARCH=gfx1030' || '' }}

diff --git a/.github/workflows/cmake.yml b/.github/workflows/cmake.yml
@@ -20,7 +20,7 @@ jobs:
         compiler: [{c: gcc, cxx: g++}]
         libbacktrace: ['-DVAL_USE_LIBBACKTRACE_BACKTRACE=OFF']
         pool_tracking: ['-DUMF_ENABLE_POOL_TRACKING=ON', '-DUMF_ENABLE_POOL_TRACKING=OFF']
-        latency_tracking: ['-DUMF_ENABLE_LATENCY_TRACKING=OFF']
+        latency_tracking: ['-DUR_ENABLE_LATENCY_HISTOGRAM=OFF']
         include:
           - os: 'ubuntu-22.04'
             build_type: Release
@@ -40,7 +40,7 @@ jobs:
           - os: 'ubuntu-22.04'
             build_type: Release
             compiler: {c: clang, cxx: clang++}
-            latency_tracking: '-DUMF_ENABLE_LATENCY_TRACKING=ON'
+            latency_tracking: '-DUR_ENABLE_LATENCY_HISTOGRAM=ON'
     runs-on: ${{ (matrix.os == 'ubuntu-22.04' && github.repository_owner == 'oneapi-src') && 'intel-ubuntu-22.04' || matrix.os }}
 
     steps:
@@ -155,6 +155,7 @@ jobs:
       adapter_name: L0
       runner_name: L0
       static_loader: ON
+      static_adapter: ON
 
   opencl:
     name: OpenCL
@@ -216,7 +217,8 @@ jobs:
         os: ['windows-2019', 'windows-2022']
         adapter: [
           {name: None, var: ''}, {name: L0, var: '-DUR_BUILD_ADAPTER_L0=ON'},
-          {name: None, var: ''}, {name: L0_V2, var: '-DUR_BUILD_ADAPTER_L0_V2=ON'}
+          {name: None, var: ''}, {name: L0_V2, var: '-DUR_BUILD_ADAPTER_L0_V2=ON'},
+          {name: L0, var: '-DUR_BUILD_ADAPTER_L0=ON -DUR_STATIC_ADAPTER_L0=ON'}
         ]
 
         # TODO: building level zero loader on windows-2019 and clang-cl is currently broken
@@ -225,16 +227,25 @@ jobs:
            adapter: {name: L0, var: '-DUR_BUILD_ADAPTER_L0=ON'}
          - os: 'windows-2019'
            adapter: {name: L0_V2, var: '-DUR_BUILD_ADAPTER_L0_V2=ON'}
+         - os: 'windows-2019'
+           adapter: {name: L0, var: '-DUR_BUILD_ADAPTER_L0=ON -DUR_STATIC_ADAPTER_L0=ON'}
          - adapter: {name: L0, var: '-DUR_BUILD_ADAPTER_L0=ON'}
            compiler: {c: clang-cl, cxx: clang-cl}
          - adapter: {name: L0_V2, var: '-DUR_BUILD_ADAPTER_L0_V2=ON'}
            compiler: {c: clang-cl, cxx: clang-cl}
+         - adapter: {name: L0, var: '-DUR_BUILD_ADAPTER_L0=ON -DUR_STATIC_ADAPTER_L0=ON'}
+           compiler: {c: clang-cl, cxx: clang-cl}
 
         build_type: [Debug, Release]
         compiler: [{c: cl, cxx: cl}, {c: clang-cl, cxx: clang-cl}]
         include:
           - compiler: {c: clang-cl, cxx: clang-cl}
             toolset: "-T ClangCL"
+          - os: 'windows-2022'
+            adapter: {name: L0, var: '-DUR_BUILD_ADAPTER_L0=ON -DUR_STATIC_ADAPTER_L0=ON'}
+            build_type: 'Release'
+            compiler: {c: cl, cxx: cl}
+
     runs-on: ${{matrix.os}}
 
     steps:

diff --git a/.github/workflows/coverity.yml b/.github/workflows/coverity.yml
@@ -1,5 +1,5 @@
 #
-# Copyright (C) 2023 Intel Corporation
+# Copyright (C) 2023-2024 Intel Corporation
 #
 # Part of the Unified-Runtime Project, under the Apache License v2.0 with LLVM Exceptions.
 # See LICENSE.TXT
@@ -38,11 +38,6 @@ jobs:
       - name: Install pip packages
         run: pip install -r third_party/requirements.txt
 
-      - name: Install apt packages
-        run: |
-          sudo apt-get update
-          sudo apt-get install -y libhwloc-dev
-
       - name: Configure CMake
         run: >
           cmake

diff --git a/.github/workflows/e2e_core.yml b/.github/workflows/e2e_core.yml
@@ -130,8 +130,8 @@ jobs:
         --ci-defaults ${{matrix.adapter.config}}
         --cmake-opt="-DLLVM_INSTALL_UTILS=ON"
         --cmake-opt="-DSYCL_PI_TESTS=OFF"
-        --cmake-opt="-DSYCL_PI_UR_USE_FETCH_CONTENT=OFF"
-        --cmake-opt="-DSYCL_PI_UR_SOURCE_DIR=${{github.workspace}}/ur-repo/"
+        --cmake-opt="-DSYCL_UR_USE_FETCH_CONTENT=OFF"
+        --cmake-opt="-DSYCL_UR_SOURCE_DIR=${{github.workspace}}/ur-repo/"
         --cmake-opt=-DCMAKE_C_COMPILER_LAUNCHER=ccache
         --cmake-opt=-DCMAKE_CXX_COMPILER_LAUNCHER=ccache
 
@@ -157,7 +157,7 @@ jobs:
     - name: Setup SYCL variables
       run: |
         which clang++ sycl-ls
-        SYCL_PI_TRACE=-1 sycl-ls
+        SYCL_UR_TRACE=-1 sycl-ls
 
     - name: Build e2e tests
       run: >
@@ -169,10 +169,6 @@ jobs:
         -DCMAKE_CXX_COMPILER="$(which clang++)"
         -DLLVM_LIT="${{github.workspace}}/sycl-repo/llvm/utils/lit/lit.py"
 
-    - name: Set LIT_XFAIL_NOT
-      if: inputs.xfail_not != ''
-      run: echo "LIT_XFAIL_NOT=${{inputs.xfail_not}}" >> $GITHUB_ENV
-
     - name: Set LIT_XFAIL
       if: inputs.xfail != ''
       run: echo "LIT_XFAIL=${{inputs.xfail}}" >> $GITHUB_ENV
@@ -181,6 +177,10 @@ jobs:
       if: inputs.filter_out != ''
       run: echo "LIT_FILTER_OUT=${{inputs.filter_out}}" >> $GITHUB_ENV
 
+    - name: Set LIT_XFAIL_NOT
+      if: inputs.xfail_not != ''
+      run: echo "LIT_XFAIL_NOT=${{inputs.xfail_not}}" >> $GITHUB_ENV
+
     # TODO: remove once intel/llvm lit tests can properly recognize the GPU
     - name: Configure hardware platform feature for L0
       if: matrix.adapter.name == 'L0'

diff --git a/.github/workflows/e2e_cuda.yml b/.github/workflows/e2e_cuda.yml
@@ -21,3 +21,4 @@ jobs:
       config: "--cuda"
       unit: "gpu"
       extra_lit_flags: "-sv --max-time=3600"
+      xfail: "Regression/device_num.cpp"
diff --git a/.github/workflows/e2e_level_zero.yml b/.github/workflows/e2e_level_zero.yml
@@ -21,9 +21,11 @@ jobs:
       config: ""
       unit: "gpu"
       # Failing tests
-      xfail: "Config/select_device.cpp;DeviceCodeSplit/grf.cpp;ESIMD/grf.cpp;KernelAndProgram/target_register_alloc_mode.cpp;Matrix/SG32/get_coord_int8_matB.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_prefetch.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;Plugin/level_zero_barrier_optimization.cpp"
+      xfail: "DeviceCodeSplit/grf.cpp;ESIMD/mask_expand_load.cpp;KernelAndProgram/target_register_alloc_mode.cpp;Matrix/SG32/get_coord_int8_matB.cpp;Matrix/get_coord_int8_matB.cpp;Matrix/joint_matrix_prefetch.cpp;Matrix/joint_matrix_rowmajorA_rowmajorB.cpp;ESIMD/mask_expand_load.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_bf16_fill_k_cache_OOB.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_out_bounds.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_prefetch.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_rowmajorA_rowmajorB.cpp;Matrix/element_wise_all_ops_1d.cpp;Matrix/element_wise_all_ops_1d_cont.cpp;Matrix/element_wise_all_ops_scalar.cpp;Matrix/joint_matrix_bf16_fill_k_cache_OOB.cpp;Matrix/joint_matrix_bf16_fill_k_cache_prefetch.cpp;Matrix/joint_matrix_out_bounds.cpp;Matrix/joint_matrix_unaligned_k.cpp;Matrix/SPVCooperativeMatrix/SG32/get_coord_int8_matB.cpp;Matrix/SPVCooperativeMatrix/element_wise_all_ops_1d.cpp;Matrix/SPVCooperativeMatrix/element_wise_all_ops_1d_cont.cpp;Matrix/SPVCooperativeMatrix/element_wise_all_ops_scalar.cpp;Matrix/SPVCooperativeMatrix/element_wise_ops.cpp;Matrix/SPVCooperativeMatrix/get_coord_int8_matB.cpp;Matrix/SPVCooperativeMatrix/joint_matrix_bf16_fill_k_cache_SLM.cpp;Matrix/joint_matrix_bf16_fill_k_cache_SLM.cpp"
+      # Unexpectedly Passed Tests
+      xfail_not: ""
       # Flaky tests
-      filter_out: "ESIMD/named_barriers/loop_extended.cpp"
+      filter_out: "Basic/accessor/accessor.cpp|DeviceArchitecture/device_architecture_comparison_on_device_aot.cpp|Graph/Explicit/interop-level-zero-launch-kernel.cpp|Graph/RecordReplay/interop-level-zero-launch-kernel.cpp|syclcompat/launch/launch_policy_lmem.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: "--param gpu-intel-pvc=True --param gpu-intel-pvc-1T=True -sv -j 100 --max-time=3600"
diff --git a/.github/workflows/e2e_opencl.yml b/.github/workflows/e2e_opencl.yml
@@ -20,5 +20,5 @@ jobs:
       prefix: ""
       config: ""
       unit: "cpu"
-      xfail: "AOT/double.cpp;AOT/half.cpp;AOT/reqd-sg-size.cpp;Basic/built-ins/marray_geometric.cpp;KernelCompiler/kernel_compiler_spirv.cpp;KernelCompiler/opencl_queries.cpp"
+      xfail: "AOT/double.cpp;AOT/half.cpp;AOT/reqd-sg-size.cpp;Basic/built-ins/marray_geometric.cpp;KernelCompiler/kernel_compiler_spirv.cpp;KernelCompiler/opencl_queries.cpp;NonUniformGroups/ballot_group.cpp;NonUniformGroups/ballot_group_algorithms.cpp;NonUniformGroups/fixed_size_group_algorithms.cpp;NonUniformGroups/opportunistic_group.cpp;NonUniformGroups/opportunistic_group_algorithms.cpp;NonUniformGroups/tangle_group.cpp;NonUniformGroups/tangle_group_algorithms.cpp"
       extra_lit_flags: "-sv --max-time=3600"
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -4,7 +4,7 @@
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 
 cmake_minimum_required(VERSION 3.20.0 FATAL_ERROR)
-project(unified-runtime VERSION 0.10.0)
+project(unified-runtime VERSION 0.11.0)
 
 # Check if unified runtime is built as a standalone project.
 if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR OR UR_STANDALONE_BUILD)
@@ -52,6 +52,7 @@ option(UR_BUILD_ADAPTER_HIP "Build the HIP adapter" OFF)
 option(UR_BUILD_ADAPTER_NATIVE_CPU "Build the Native-CPU adapter" OFF)
 option(UR_BUILD_ADAPTER_ALL "Build all currently supported adapters" OFF)
 option(UR_BUILD_ADAPTER_L0_V2 "Build the (experimental) Level-Zero v2 adapter" OFF)
+option(UR_STATIC_ADAPTER_L0 "Build the Level-Zero adapter as static and embed in the loader" OFF)
 option(UR_BUILD_EXAMPLE_CODEGEN "Build the codegen example." OFF)
 option(VAL_USE_LIBBACKTRACE_BACKTRACE "enable libbacktrace validation backtrace for linux" OFF)
 option(UR_ENABLE_ASSERTIONS "Enable assertions for all build types" OFF)

diff --git a/README.md b/README.md
@@ -145,7 +145,7 @@ List of options provided by CMake:
 | UR_DEVICE_CODE_EXTRACTOR | Path of the `clang-offload-extract` executable from the DPC++ package, required for CTS device binaries | File path | `"${dirname(UR_DPCXX)}/clang-offload-extract"` |
 | UR_DPCXX_BUILD_FLAGS | Build flags to pass to DPC++ when compiling device programs | Space-separated options list | `""` |
 | UR_SYCL_LIBRARY_DIR | Path of the SYCL runtime library directory to build CTS device binaries | Directory path | `""` |
-| UR_HIP_ROCM_DIR | Path of the default ROCm HIP installation | Directory path | `/opt/rocm` |
+| UR_HIP_ROCM_DIR | Path of the default ROCm HIP installation | Directory path | `$ENV{ROCM_PATH}` or `/opt/rocm` |
 | UR_HIP_INCLUDE_DIR | Path of the ROCm HIP include directory | Directory path | `${UR_HIP_ROCM_DIR}/include` |
 | UR_HIP_HSA_INCLUDE_DIRS | Path of the ROCm HSA include directory | Directory path | `${UR_HIP_ROCM_DIR}/hsa/include;${UR_HIP_ROCM_DIR}/include` |
 | UR_HIP_LIB_DIR | Path of the ROCm HIP library directory | Directory path | `${UR_HIP_ROCM_DIR}/lib` |

diff --git a/cmake/helpers.cmake b/cmake/helpers.cmake
@@ -70,6 +70,7 @@ function(add_ur_target_compile_options name)
         )
         if (CMAKE_BUILD_TYPE STREQUAL "Release")
             target_compile_definitions(${name} PRIVATE -D_FORTIFY_SOURCE=2)
+            target_compile_options(${name} PRIVATE -fvisibility=hidden)
         endif()
         if(UR_DEVELOPER_MODE)
             target_compile_options(${name} PRIVATE