Merge branch 'main' into merge-some-main-changes-into-adapters-third-patch

Author: omarahmed1111

.github/workflows/cmake.yml

@@ -292,6 +292,13 @@ jobs:
     - name: Install prerequisites
       run: python3 -m pip install -r third_party/requirements.txt
 
+    - name: Install doxygen
+      run: |
+        $WorkingDir = $PWD.Path
+        Invoke-WebRequest -Uri https://github.com/doxygen/doxygen/releases/download/Release_1_9_8/doxygen-1.9.8.windows.x64.bin.zip -OutFile "$WorkingDir\doxygen.zip"
+        Expand-Archive -Path "$WorkingDir\doxygen.zip"
+        Add-Content $env:GITHUB_PATH "$WorkingDir\doxygen"
+
     - name: Configure CMake
       run: >
         cmake

.github/workflows/e2e_nightly.yml

@@ -0,0 +1,119 @@
+name: E2E Nightly
+
+on:
+  schedule:
+    # Run every day at 23:00 UTC
+    - cron: '0 23 * * *'
+
+jobs:
+  e2e-build-hw:
+    name: Build SYCL, UR, run E2E
+    strategy:
+      matrix:
+        adapter: [
+          {name: CUDA}
+        ]
+        build_type: [Release]
+        compiler: [{c: clang, cxx: clang++}]
+
+    runs-on: ${{matrix.adapter.name}}
+
+    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: |
+        ls -la ./
+        rm -rf ./* || true
+
+    - name: Checkout UR
+      uses: actions/checkout@v4
+      with:
+        ref: adapters
+        path: ur-repo
+
+    - name: Checkout SYCL
+      uses: actions/checkout@v4
+      with:
+        repository: intel/llvm
+        ref: sycl
+        path: sycl-repo
+
+    - name: Install pip packages
+      working-directory: ${{github.workspace}}/ur-repo
+      run: pip install -r third_party/requirements.txt
+
+    - name: Configure CMake UR
+      working-directory: ${{github.workspace}}/ur-repo
+      run: >
+        cmake
+        -B build
+        -DCMAKE_C_COMPILER=${{matrix.compiler.c}}
+        -DCMAKE_CXX_COMPILER=${{matrix.compiler.cxx}}
+        -DCMAKE_BUILD_TYPE=${{matrix.build_type}}
+        -DUR_ENABLE_TRACING=ON
+        -DUR_DEVELOPER_MODE=ON
+        -DUR_BUILD_ADAPTER_${{matrix.adapter.name}}=ON
+
+    - name: Build UR
+      run: LD_LIBRARY_PATH=${{github.workspace}}/dpcpp_compiler/lib
+        cmake --build ${{github.workspace}}/ur-repo/build -j $(nproc)
+
+    - name: Set env vars & pre setup
+      run: |
+        echo "SYCL_PREFER_UR=1" >> $GITHUB_ENV
+        echo "CUDA_LIB_PATH=/usr/local/cuda/lib64/stubs" >> $GITHUB_ENV
+        echo "LD_LIBRARY_PATH=/usr/local/cuda/compat/:/usr/local/cuda/lib64:$LD_LIBRARY_PATH" >> $GITHUB_ENV
+        source /opt/intel/oneapi/setvars.sh
+        sycl-ls
+
+    - name: Configure SYCL
+      run: >
+        python3 sycl-repo/buildbot/configure.py
+        -t ${{matrix.build_type}}
+        -o ${{github.workspace}}/sycl_build
+        --cmake-gen "Unix Makefiles"
+        --ci-defaults --cuda --hip
+        --cmake-opt="-DLLVM_INSTALL_UTILS=ON"
+        --cmake-opt="-DSYCL_PI_TESTS=OFF"
+        --cmake-opt=-DCMAKE_C_COMPILER_LAUNCHER=ccache
+        --cmake-opt=-DCMAKE_CXX_COMPILER_LAUNCHER=ccache
+
+    - name: Build SYCL
+      run: cmake --build ${{github.workspace}}/sycl_build
+
+    - name: Run check-sycl
+      # Remove after fixing SYCL test :: abi/layout_handler.cpp
+      # This issue does not affect further execution of e2e with UR.
+      continue-on-error: true
+      run: cmake --build ${{github.workspace}}/sycl_build --target check-sycl
+
+    - name: Swap UR loader and adapters
+      run: |
+        cp ${{github.workspace}}/ur-repo/build/lib/libur_loader.so* ${{github.workspace}}/sycl_build/lib/
+        cp ${{github.workspace}}/ur-repo/build/lib/libur_adapter_cuda.so* ${{github.workspace}}/sycl_build/lib/
+
+    - name: Set additional env. vars
+      run: |
+        echo "${{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: Build e2e tests
+      run: >
+        cmake
+        -GNinja
+        -B ${{github.workspace}}/build-e2e/
+        -S ${{github.workspace}}/sycl-repo/sycl/test-e2e/
+        -DSYCL_TEST_E2E_TARGETS="ext_oneapi_cuda:gpu"
+        -DCMAKE_CXX_COMPILER="$(which clang++)"
+        -DLLVM_LIT="${{github.workspace}}/sycl-repo/llvm/utils/lit/lit.py"
+
+    - name: Run e2e tests
+      run: ninja -C build-e2e check-sycl-e2e

README.md

@@ -31,7 +31,7 @@ see cmake options for details.
     - [Adapter naming convention](#adapter-naming-convention)
     - [Source code generation](#source-code-generation)
     - [Documentation](#documentation)
-
+6. [Release Process](#release-process)
 
 ## Contents of the repo
 
@@ -182,3 +182,26 @@ Code is generated using included [Python scripts](/scripts/README.md).
 
 Documentation is generated from source code using Sphinx -
 see [scripts dir](/scripts/README.md) for details.
+
+## Release Process
+
+Unified Runtime releases are aligned with oneAPI releases. Once all changes
+planned for a release have been accepted, the release process is defined as:
+
+1. Create a new release branch based on the [main][main-branch] branch taking
+   the form `v<major>.<minor>.x` where `x` is a placeholder for the patch
+   version. This branch will always contain the latest patch version for a given
+   release.
+2. Create a PR to increment the CMake project version on the [main][main-branch]
+   and merge before accepting any other changes.
+3. Create a new tag based on the latest commit on the release branch taking the
+   form `v<major>.<minor>.<patch>`.
+4. Create a [new GitHub release][new-github-release] using the tag created in
+   the previous step.
+   * Prior to version 1.0, check the *Set as a pre-release* tick box.
+5. Update downstream projects to utilize the release tag. If any issues arise
+   from integration, apply any necessary hot fixes to `v<major>.<minor>.x`
+   branch and go back to step 3.
+
+[main-branch]: https://github.com/oneapi-src/unified-runtime/tree/main
+[new-github-release]: https://github.com/oneapi-src/unified-runtime/releases/new

include/ur.py

@@ -202,6 +202,10 @@ class ur_function_v(IntEnum):
     KERNEL_SUGGEST_MAX_COOPERATIVE_GROUP_COUNT_EXP = 194## Enumerator for ::urKernelSuggestMaxCooperativeGroupCountExp
     COMMAND_BUFFER_APPEND_USM_PREFETCH_EXP = 195    ## Enumerator for ::urCommandBufferAppendUSMPrefetchExp
     COMMAND_BUFFER_APPEND_USM_ADVISE_EXP = 196      ## Enumerator for ::urCommandBufferAppendUSMAdviseExp
+    PROGRAM_BUILD_EXP = 197                         ## Enumerator for ::urProgramBuildExp
+    PROGRAM_COMPILE_EXP = 198                       ## Enumerator for ::urProgramCompileExp
+    PROGRAM_LINK_EXP = 199                          ## Enumerator for ::urProgramLinkExp
+    LOADER_CONFIG_SET_CODE_LOCATION_CALLBACK = 200  ## Enumerator for ::urLoaderConfigSetCodeLocationCallback
 
 class ur_function_t(c_int):
     def __str__(self):
@@ -518,6 +522,24 @@ def __str__(self):
         return str(ur_loader_config_info_v(self.value))
 
 
+###############################################################################
+## @brief Code location data
+class ur_code_location_t(Structure):
+    _fields_ = [
+        ("functionName", c_char_p),                                     ## [in][out] Function name.
+        ("sourceFile", c_char_p),                                       ## [in][out] Source code file.
+        ("lineNumber", c_ulong),                                        ## [in][out] Source code line number.
+        ("columnNumber", c_ulong)                                       ## [in][out] Source code column number.
+    ]
+
+###############################################################################
+## @brief Code location callback with user data.
+def ur_code_location_callback_t(user_defined_callback):
+    @CFUNCTYPE(ur_code_location_t, c_void_p)
+    def ur_code_location_callback_t_wrapper(pUserData):
+        return user_defined_callback(pUserData)
+    return ur_code_location_callback_t_wrapper
+
 ###############################################################################
 ## @brief Supported adapter info
 class ur_adapter_info_v(IntEnum):
@@ -2106,10 +2128,10 @@ class ur_event_native_properties_t(Structure):
 ###############################################################################
 ## @brief Event states for all events.
 class ur_execution_info_v(IntEnum):
-    EXECUTION_INFO_COMPLETE = 0                     ## Indicates that the event has completed.
-    EXECUTION_INFO_RUNNING = 1                      ## Indicates that the device has started processing this event.
-    EXECUTION_INFO_SUBMITTED = 2                    ## Indicates that the event has been submitted by the host to the device.
-    EXECUTION_INFO_QUEUED = 3                       ## Indicates that the event has been queued, this is the initial state of
+    COMPLETE = 0                                    ## Indicates that the event has completed.
+    RUNNING = 1                                     ## Indicates that the device has started processing this event.
+    SUBMITTED = 2                                   ## Indicates that the event has been submitted by the host to the device.
+    QUEUED = 3                                      ## Indicates that the event has been queued, this is the initial state of
                                                     ## events.
 
 class ur_execution_info_t(c_int):
@@ -2296,6 +2318,11 @@ class ur_exp_command_buffer_handle_t(c_void_p):
 ##        which is returned when querying device extensions.
 UR_COOPERATIVE_KERNELS_EXTENSION_STRING_EXP = "ur_exp_cooperative_kernels"
 
+###############################################################################
+## @brief The extension string which defines support for test
+##        which is returned when querying device extensions.
+UR_MULTI_DEVICE_COMPILE_EXTENSION_STRING_EXP = "ur_exp_multi_device_compile"
+
 ###############################################################################
 ## @brief Supported peer info
 class ur_exp_peer_info_v(IntEnum):
@@ -2612,6 +2639,37 @@ class ur_program_dditable_t(Structure):
         ("pfnCreateWithNativeHandle", c_void_p)                         ## _urProgramCreateWithNativeHandle_t
     ]
 
+###############################################################################
+## @brief Function-pointer for urProgramBuildExp
+if __use_win_types:
+    _urProgramBuildExp_t = WINFUNCTYPE( ur_result_t, ur_program_handle_t, c_ulong, POINTER(ur_device_handle_t), c_char_p )
+else:
+    _urProgramBuildExp_t = CFUNCTYPE( ur_result_t, ur_program_handle_t, c_ulong, POINTER(ur_device_handle_t), c_char_p )
+
+###############################################################################
+## @brief Function-pointer for urProgramCompileExp
+if __use_win_types:
+    _urProgramCompileExp_t = WINFUNCTYPE( ur_result_t, ur_program_handle_t, c_ulong, POINTER(ur_device_handle_t), c_char_p )
+else:
+    _urProgramCompileExp_t = CFUNCTYPE( ur_result_t, ur_program_handle_t, c_ulong, POINTER(ur_device_handle_t), c_char_p )
+
+###############################################################################
+## @brief Function-pointer for urProgramLinkExp
+if __use_win_types:
+    _urProgramLinkExp_t = WINFUNCTYPE( ur_result_t, ur_context_handle_t, c_ulong, POINTER(ur_device_handle_t), c_ulong, POINTER(ur_program_handle_t), c_char_p, POINTER(ur_program_handle_t) )
+else:
+    _urProgramLinkExp_t = CFUNCTYPE( ur_result_t, ur_context_handle_t, c_ulong, POINTER(ur_device_handle_t), c_ulong, POINTER(ur_program_handle_t), c_char_p, POINTER(ur_program_handle_t) )
+
+
+###############################################################################
+## @brief Table of ProgramExp functions pointers
+class ur_program_exp_dditable_t(Structure):
+    _fields_ = [
+        ("pfnBuildExp", c_void_p),                                      ## _urProgramBuildExp_t
+        ("pfnCompileExp", c_void_p),                                    ## _urProgramCompileExp_t
+        ("pfnLinkExp", c_void_p)                                        ## _urProgramLinkExp_t
+    ]
+
 ###############################################################################
 ## @brief Function-pointer for urKernelCreate
 if __use_win_types:
@@ -3843,6 +3901,7 @@ class ur_dditable_t(Structure):
         ("Context", ur_context_dditable_t),
         ("Event", ur_event_dditable_t),
         ("Program", ur_program_dditable_t),
+        ("ProgramExp", ur_program_exp_dditable_t),
         ("Kernel", ur_kernel_dditable_t),
         ("KernelExp", ur_kernel_exp_dditable_t),
         ("Sampler", ur_sampler_dditable_t),
@@ -3947,6 +4006,18 @@ def __init__(self, version : ur_api_version_t):
         self.urProgramGetNativeHandle = _urProgramGetNativeHandle_t(self.__dditable.Program.pfnGetNativeHandle)
         self.urProgramCreateWithNativeHandle = _urProgramCreateWithNativeHandle_t(self.__dditable.Program.pfnCreateWithNativeHandle)
 
+        # call driver to get function pointers
+        ProgramExp = ur_program_exp_dditable_t()
+        r = ur_result_v(self.__dll.urGetProgramExpProcAddrTable(version, byref(ProgramExp)))
+        if r != ur_result_v.SUCCESS:
+            raise Exception(r)
+        self.__dditable.ProgramExp = ProgramExp
+
+        # attach function interface to function address
+        self.urProgramBuildExp = _urProgramBuildExp_t(self.__dditable.ProgramExp.pfnBuildExp)
+        self.urProgramCompileExp = _urProgramCompileExp_t(self.__dditable.ProgramExp.pfnCompileExp)
+        self.urProgramLinkExp = _urProgramLinkExp_t(self.__dditable.ProgramExp.pfnLinkExp)
+
         # call driver to get function pointers
         Kernel = ur_kernel_dditable_t()
         r = ur_result_v(self.__dll.urGetKernelProcAddrTable(version, byref(Kernel)))