Merge branch 'main' into try-catch

Author: Hugh Delaney

.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/build-fuzz-reusable.yml

@@ -41,14 +41,12 @@ 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${{github.workspace}}/build
@@ -60,7 +58,6 @@ jobs:
         -DUR_USE_ASAN=ON
         -DUR_USE_UBSAN=ON
         -DUR_BUILD_ADAPTER_L0=ON
-        -DUR_USE_CFI=OFF
         -DUR_LEVEL_ZERO_LOADER_LIBRARY=${{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

@@ -82,8 +82,6 @@ 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${{github.workspace}}/build
@@ -96,7 +94,6 @@ jobs:
         -DUR_BUILD_ADAPTER_${{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++

.github/workflows/cmake.yml

@@ -256,10 +256,12 @@ jobs:
            compiler: {c: clang-cl, cxx: clang-cl}
 
         build_type: [Debug, Release]
-        compiler: [{c: cl, cxx: cl}, {c: clang-cl, cxx: clang-cl}]
+        # TODO: clang-cl seems to be fully broken (https://github.com/oneapi-src/unified-runtime/issues/2348)
+        #compiler: [{c: cl, cxx: cl}, {c: clang-cl, cxx: clang-cl}]
+        compiler: [{c: cl, cxx: cl}]
         include:
-          - compiler: {c: clang-cl, cxx: clang-cl}
-            toolset: "-T ClangCL"
+          #- 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'

.github/workflows/e2e_core.yml

@@ -190,7 +190,7 @@ jobs:
 
     - name: Run e2e tests
       id: tests
-      run: ninja -C build-e2e check-sycl-e2e
+      run: ninja -C build-e2e check-sycl-e2e || echo "e2e tests have failed. Ignoring failure."
 
     # FIXME: Requires pull-request: write permissions but this is only granted
     # on pull requests from forks if using pull_request_target workflow

CMakeLists.txt

@@ -16,19 +16,6 @@ include(CheckCXXSourceCompiles)
 include(CMakePackageConfigHelpers)
 include(CTest)
 
-list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
-include(helpers)
-
-if(CMAKE_SYSTEM_NAME STREQUAL Darwin)
-    set(Python3_FIND_FRAMEWORK NEVER)
-    set(Python3_FIND_STRATEGY LOCATION)
-endif()
-
-find_package(Python3 COMPONENTS Interpreter REQUIRED)
-
-set(CMAKE_CXX_STANDARD 17)
-set(CMAKE_CXX_STANDARD_REQUIRED YES)
-
 # Build Options
 option(UR_BUILD_EXAMPLES "Build example applications." ON)
 option(UR_BUILD_TESTS "Build unit tests." ON)
@@ -81,6 +68,19 @@ set(UR_ADAPTER_HIP_SOURCE_DIR "" CACHE PATH
 set(UR_ADAPTER_NATIVE_CPU_SOURCE_DIR "" CACHE PATH
     "Path to external 'native_cpu' adapter source dir")
 
+list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/cmake")
+include(helpers)
+
+if(CMAKE_SYSTEM_NAME STREQUAL Darwin)
+    set(Python3_FIND_FRAMEWORK NEVER)
+    set(Python3_FIND_STRATEGY LOCATION)
+endif()
+
+find_package(Python3 COMPONENTS Interpreter REQUIRED)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED YES)
+
 # There's little reason not to generate the compile_commands.json
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 

cmake/FetchLevelZero.cmake

@@ -7,6 +7,8 @@ set(UR_LEVEL_ZERO_LOADER_LIBRARY "" CACHE FILEPATH "Path of the Level Zero Loade
 set(UR_LEVEL_ZERO_INCLUDE_DIR "" CACHE FILEPATH "Directory containing the Level Zero Headers")
 set(UR_LEVEL_ZERO_LOADER_REPO "" CACHE STRING "Github repo to get the Level Zero loader sources from")
 set(UR_LEVEL_ZERO_LOADER_TAG "" CACHE STRING " GIT tag of the Level Loader taken from github repo")
+set(UR_COMPUTE_RUNTIME_REPO "" CACHE STRING "Github repo to get the compute runtime sources from")
+set(UR_COMPUTE_RUNTIME_TAG "" CACHE STRING " GIT tag of the compute runtime taken from github repo")
 
 # Copy Level Zero loader/headers locally to the build to avoid leaking their path.
 set(LEVEL_ZERO_COPY_DIR ${CMAKE_CURRENT_BINARY_DIR}/level_zero_loader)
@@ -40,7 +42,7 @@ if (NOT DEFINED LEVEL_ZERO_LIBRARY OR NOT DEFINED LEVEL_ZERO_INCLUDE_DIR)
         set(UR_LEVEL_ZERO_LOADER_REPO "https://github.com/oneapi-src/level-zero.git")
     endif()
     if (UR_LEVEL_ZERO_LOADER_TAG STREQUAL "")
-        set(UR_LEVEL_ZERO_LOADER_TAG v1.18.5)
+        set(UR_LEVEL_ZERO_LOADER_TAG v1.19.2)
     endif()
 
     # Disable due to a bug https://github.com/oneapi-src/level-zero/issues/104
@@ -87,8 +89,31 @@ target_link_libraries(LevelZeroLoader
     INTERFACE "${LEVEL_ZERO_LIB_NAME}"
 )
 
+file(GLOB LEVEL_ZERO_LOADER_API_HEADERS "${LEVEL_ZERO_INCLUDE_DIR}/*.h")
+file(COPY ${LEVEL_ZERO_LOADER_API_HEADERS} DESTINATION ${LEVEL_ZERO_INCLUDE_DIR}/level_zero)
 add_library(LevelZeroLoader-Headers INTERFACE)
 target_include_directories(LevelZeroLoader-Headers
-    INTERFACE "$<BUILD_INTERFACE:${LEVEL_ZERO_INCLUDE_DIR}>"
+    INTERFACE "$<BUILD_INTERFACE:${LEVEL_ZERO_INCLUDE_DIR};${LEVEL_ZERO_INCLUDE_DIR}/level_zero>"
+              "$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>"
+)
+
+if (UR_COMPUTE_RUNTIME_REPO STREQUAL "")
+set(UR_COMPUTE_RUNTIME_REPO "https://github.com/intel/compute-runtime.git")
+endif()
+if (UR_COMPUTE_RUNTIME_TAG STREQUAL "")
+set(UR_COMPUTE_RUNTIME_TAG 24.39.31294.12)
+endif()
+include(FetchContent)
+# Sparse fetch only the dir with level zero headers to avoid pulling in the entire compute-runtime.
+FetchContentSparse_Declare(compute-runtime-level-zero-headers ${UR_COMPUTE_RUNTIME_REPO} "${UR_COMPUTE_RUNTIME_TAG}" "level_zero/include")
+FetchContent_GetProperties(compute-runtime-level-zero-headers)
+if(NOT compute-runtime-level-zero-headers_POPULATED)
+  FetchContent_Populate(compute-runtime-level-zero-headers)
+endif()
+add_library(ComputeRuntimeLevelZero-Headers INTERFACE)
+set(COMPUTE_RUNTIME_LEVEL_ZERO_INCLUDE "${compute-runtime-level-zero-headers_SOURCE_DIR}/../..")
+message(STATUS "Level Zero Adapter: Using Level Zero headers from ${COMPUTE_RUNTIME_LEVEL_ZERO_INCLUDE}")
+target_include_directories(ComputeRuntimeLevelZero-Headers
+    INTERFACE "$<BUILD_INTERFACE:${COMPUTE_RUNTIME_LEVEL_ZERO_INCLUDE}>"
               "$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>"
 )

cmake/helpers.cmake

@@ -63,6 +63,12 @@ if(CMAKE_SYSTEM_NAME STREQUAL Linux)
     check_cxx_compiler_flag("-fstack-clash-protection" CXX_HAS_FSTACK_CLASH_PROTECTION)
 endif()
 
+if (UR_USE_CFI AND UR_USE_ASAN)
+    message(WARNING "Both UR_USE_CFI and UR_USE_ASAN are ON. "
+        "Due to build errors, this is unsupported; CFI checks will be disabled")
+    set(UR_USE_CFI OFF)
+endif()
+
 if (UR_USE_CFI)
     set(SAVED_CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS})
     set(CMAKE_REQUIRED_FLAGS "-flto -fvisibility=hidden")

include/ur_api.h

@@ -1705,6 +1705,8 @@ typedef enum ur_device_info_t {
     UR_DEVICE_INFO_ENQUEUE_NATIVE_COMMAND_SUPPORT_EXP = 0x2020,      ///< [::ur_bool_t] returns true if the device supports enqueueing of native
                                                                      ///< work
     UR_DEVICE_INFO_LOW_POWER_EVENTS_EXP = 0x2021,                    ///< [::ur_bool_t] returns true if the device supports low-power events.
+    UR_DEVICE_INFO_2D_BLOCK_ARRAY_CAPABILITIES_EXP = 0x2022,         ///< [::ur_exp_device_2d_block_array_capability_flags_t] return a bit-field
+                                                                     ///< of Intel GPU 2D block array capabilities
     /// @cond
     UR_DEVICE_INFO_FORCE_UINT32 = 0x7fffffff
     /// @endcond
@@ -1730,7 +1732,7 @@ typedef enum ur_device_info_t {
 ///     - ::UR_RESULT_ERROR_INVALID_NULL_HANDLE
 ///         + `NULL == hDevice`
 ///     - ::UR_RESULT_ERROR_INVALID_ENUMERATION
-///         + `::UR_DEVICE_INFO_LOW_POWER_EVENTS_EXP < propName`
+///         + `::UR_DEVICE_INFO_2D_BLOCK_ARRAY_CAPABILITIES_EXP < propName`
 ///     - ::UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION
 ///         + If `propName` is not supported by the adapter.
 ///     - ::UR_RESULT_ERROR_INVALID_SIZE
@@ -7428,6 +7430,27 @@ urEnqueueWriteHostPipe(
                                               ///< an element of the phEventWaitList array.
 );
 
+#if !defined(__GNUC__)
+#pragma endregion
+#endif
+// Intel 'oneAPI' Unified Runtime Experimental device descriptor for querying Intel device 2D block array capabilities
+#if !defined(__GNUC__)
+#pragma region 2d_block_array_capabilities_(experimental)
+#endif
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Intel GPU 2D block array capabilities
+typedef uint32_t ur_exp_device_2d_block_array_capability_flags_t;
+typedef enum ur_exp_device_2d_block_array_capability_flag_t {
+    UR_EXP_DEVICE_2D_BLOCK_ARRAY_CAPABILITY_FLAG_LOAD = UR_BIT(0),  ///< Load instructions are supported
+    UR_EXP_DEVICE_2D_BLOCK_ARRAY_CAPABILITY_FLAG_STORE = UR_BIT(1), ///< Store instructions are supported
+    /// @cond
+    UR_EXP_DEVICE_2D_BLOCK_ARRAY_CAPABILITY_FLAG_FORCE_UINT32 = 0x7fffffff
+    /// @endcond
+
+} ur_exp_device_2d_block_array_capability_flag_t;
+/// @brief Bit Mask for validating ur_exp_device_2d_block_array_capability_flags_t
+#define UR_EXP_DEVICE_2D_BLOCK_ARRAY_CAPABILITY_FLAGS_MASK 0xfffffffc
+
 #if !defined(__GNUC__)
 #pragma endregion
 #endif

include/ur_print.h

@@ -874,6 +874,14 @@ UR_APIEXPORT ur_result_t UR_APICALL urPrintMapFlags(enum ur_map_flag_t value, ch
 ///         - `buff_size < out_size`
 UR_APIEXPORT ur_result_t UR_APICALL urPrintUsmMigrationFlags(enum ur_usm_migration_flag_t value, char *buffer, const size_t buff_size, size_t *out_size);
 
+///////////////////////////////////////////////////////////////////////////////
+/// @brief Print ur_exp_device_2d_block_array_capability_flag_t enum
+/// @returns
+///     - ::UR_RESULT_SUCCESS
+///     - ::UR_RESULT_ERROR_INVALID_SIZE
+///         - `buff_size < out_size`
+UR_APIEXPORT ur_result_t UR_APICALL urPrintExpDevice_2dBlockArrayCapabilityFlags(enum ur_exp_device_2d_block_array_capability_flag_t value, char *buffer, const size_t buff_size, size_t *out_size);
+
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Print ur_exp_image_copy_flag_t enum
 /// @returns