Merge branch 'main' into yc/0806-exclude-shadow-from-coredump

Author: yingcong-wu

.github/workflows/benchmarks_compute.yml

@@ -151,9 +151,17 @@ jobs:
     - 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

.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' || '' }}

.github/workflows/cmake.yml

@@ -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:

CMakeLists.txt

@@ -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)

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` |

include/ur_api.h

@@ -1569,8 +1569,7 @@ typedef enum ur_device_info_t {
                                                                      ///< ::urDevicePartition
     UR_DEVICE_INFO_MAX_NUM_SUB_GROUPS = 80,                          ///< [uint32_t] max number of sub groups
     UR_DEVICE_INFO_SUB_GROUP_INDEPENDENT_FORWARD_PROGRESS = 81,      ///< [::ur_bool_t] support sub group independent forward progress
-    UR_DEVICE_INFO_SUB_GROUP_SIZES_INTEL = 82,                       ///< [uint32_t[]] return an array of sub group sizes supported on Intel
-                                                                     ///< device
+    UR_DEVICE_INFO_SUB_GROUP_SIZES_INTEL = 82,                       ///< [uint32_t[]] return an array of supported sub group sizes
     UR_DEVICE_INFO_USM_HOST_SUPPORT = 83,                            ///< [::ur_device_usm_access_capability_flags_t] support USM host memory
                                                                      ///< access
     UR_DEVICE_INFO_USM_DEVICE_SUPPORT = 84,                          ///< [::ur_device_usm_access_capability_flags_t] support USM device memory

include/ur_api_funcs.def

@@ -0,0 +1,212 @@
+
+/*
+ *
+ * Copyright (C) 2024 Intel Corporation
+ *
+ * Part of the Unified-Runtime Project, under the Apache License v2.0 with LLVM Exceptions.
+ * See LICENSE.TXT
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ *
+ * @file ur_api_funcs.def
+ * @version v0.11-r0
+ *
+ */
+
+ // Auto-generated file, do not edit.
+
+_UR_API(urPlatformGet)
+_UR_API(urPlatformGetInfo)
+_UR_API(urPlatformGetNativeHandle)
+_UR_API(urPlatformCreateWithNativeHandle)
+_UR_API(urPlatformGetApiVersion)
+_UR_API(urPlatformGetBackendOption)
+_UR_API(urContextCreate)
+_UR_API(urContextRetain)
+_UR_API(urContextRelease)
+_UR_API(urContextGetInfo)
+_UR_API(urContextGetNativeHandle)
+_UR_API(urContextCreateWithNativeHandle)
+_UR_API(urContextSetExtendedDeleter)
+_UR_API(urEventGetInfo)
+_UR_API(urEventGetProfilingInfo)
+_UR_API(urEventWait)
+_UR_API(urEventRetain)
+_UR_API(urEventRelease)
+_UR_API(urEventGetNativeHandle)
+_UR_API(urEventCreateWithNativeHandle)
+_UR_API(urEventSetCallback)
+_UR_API(urProgramCreateWithIL)
+_UR_API(urProgramCreateWithBinary)
+_UR_API(urProgramBuild)
+_UR_API(urProgramCompile)
+_UR_API(urProgramLink)
+_UR_API(urProgramRetain)
+_UR_API(urProgramRelease)
+_UR_API(urProgramGetFunctionPointer)
+_UR_API(urProgramGetGlobalVariablePointer)
+_UR_API(urProgramGetInfo)
+_UR_API(urProgramGetBuildInfo)
+_UR_API(urProgramSetSpecializationConstants)
+_UR_API(urProgramGetNativeHandle)
+_UR_API(urProgramCreateWithNativeHandle)
+_UR_API(urProgramBuildExp)
+_UR_API(urProgramCompileExp)
+_UR_API(urProgramLinkExp)
+_UR_API(urKernelCreate)
+_UR_API(urKernelGetInfo)
+_UR_API(urKernelGetGroupInfo)
+_UR_API(urKernelGetSubGroupInfo)
+_UR_API(urKernelRetain)
+_UR_API(urKernelRelease)
+_UR_API(urKernelGetNativeHandle)
+_UR_API(urKernelCreateWithNativeHandle)
+_UR_API(urKernelGetSuggestedLocalWorkSize)
+_UR_API(urKernelSetArgValue)
+_UR_API(urKernelSetArgLocal)
+_UR_API(urKernelSetArgPointer)
+_UR_API(urKernelSetExecInfo)
+_UR_API(urKernelSetArgSampler)
+_UR_API(urKernelSetArgMemObj)
+_UR_API(urKernelSetSpecializationConstants)
+_UR_API(urKernelSuggestMaxCooperativeGroupCountExp)
+_UR_API(urQueueGetInfo)
+_UR_API(urQueueCreate)
+_UR_API(urQueueRetain)
+_UR_API(urQueueRelease)
+_UR_API(urQueueGetNativeHandle)
+_UR_API(urQueueCreateWithNativeHandle)
+_UR_API(urQueueFinish)
+_UR_API(urQueueFlush)
+_UR_API(urSamplerCreate)
+_UR_API(urSamplerRetain)
+_UR_API(urSamplerRelease)
+_UR_API(urSamplerGetInfo)
+_UR_API(urSamplerGetNativeHandle)
+_UR_API(urSamplerCreateWithNativeHandle)
+_UR_API(urMemImageCreate)
+_UR_API(urMemBufferCreate)
+_UR_API(urMemRetain)
+_UR_API(urMemRelease)
+_UR_API(urMemBufferPartition)
+_UR_API(urMemGetNativeHandle)
+_UR_API(urMemBufferCreateWithNativeHandle)
+_UR_API(urMemImageCreateWithNativeHandle)
+_UR_API(urMemGetInfo)
+_UR_API(urMemImageGetInfo)
+_UR_API(urPhysicalMemCreate)
+_UR_API(urPhysicalMemRetain)
+_UR_API(urPhysicalMemRelease)
+_UR_API(urAdapterGet)
+_UR_API(urAdapterRelease)
+_UR_API(urAdapterRetain)
+_UR_API(urAdapterGetLastError)
+_UR_API(urAdapterGetInfo)
+_UR_API(urEnqueueKernelLaunch)
+_UR_API(urEnqueueEventsWait)
+_UR_API(urEnqueueEventsWaitWithBarrier)
+_UR_API(urEnqueueMemBufferRead)
+_UR_API(urEnqueueMemBufferWrite)
+_UR_API(urEnqueueMemBufferReadRect)
+_UR_API(urEnqueueMemBufferWriteRect)
+_UR_API(urEnqueueMemBufferCopy)
+_UR_API(urEnqueueMemBufferCopyRect)
+_UR_API(urEnqueueMemBufferFill)
+_UR_API(urEnqueueMemImageRead)
+_UR_API(urEnqueueMemImageWrite)
+_UR_API(urEnqueueMemImageCopy)
+_UR_API(urEnqueueMemBufferMap)
+_UR_API(urEnqueueMemUnmap)
+_UR_API(urEnqueueUSMFill)
+_UR_API(urEnqueueUSMMemcpy)
+_UR_API(urEnqueueUSMPrefetch)
+_UR_API(urEnqueueUSMAdvise)
+_UR_API(urEnqueueUSMFill2D)
+_UR_API(urEnqueueUSMMemcpy2D)
+_UR_API(urEnqueueDeviceGlobalVariableWrite)
+_UR_API(urEnqueueDeviceGlobalVariableRead)
+_UR_API(urEnqueueReadHostPipe)
+_UR_API(urEnqueueWriteHostPipe)
+_UR_API(urEnqueueKernelLaunchCustomExp)
+_UR_API(urEnqueueCooperativeKernelLaunchExp)
+_UR_API(urEnqueueTimestampRecordingExp)
+_UR_API(urEnqueueNativeCommandExp)
+_UR_API(urBindlessImagesUnsampledImageHandleDestroyExp)
+_UR_API(urBindlessImagesSampledImageHandleDestroyExp)
+_UR_API(urBindlessImagesImageAllocateExp)
+_UR_API(urBindlessImagesImageFreeExp)
+_UR_API(urBindlessImagesUnsampledImageCreateExp)
+_UR_API(urBindlessImagesSampledImageCreateExp)
+_UR_API(urBindlessImagesImageCopyExp)
+_UR_API(urBindlessImagesImageGetInfoExp)
+_UR_API(urBindlessImagesMipmapGetLevelExp)
+_UR_API(urBindlessImagesMipmapFreeExp)
+_UR_API(urBindlessImagesImportExternalMemoryExp)
+_UR_API(urBindlessImagesMapExternalArrayExp)
+_UR_API(urBindlessImagesMapExternalLinearMemoryExp)
+_UR_API(urBindlessImagesReleaseExternalMemoryExp)
+_UR_API(urBindlessImagesImportExternalSemaphoreExp)
+_UR_API(urBindlessImagesReleaseExternalSemaphoreExp)
+_UR_API(urBindlessImagesWaitExternalSemaphoreExp)
+_UR_API(urBindlessImagesSignalExternalSemaphoreExp)
+_UR_API(urUSMHostAlloc)
+_UR_API(urUSMDeviceAlloc)
+_UR_API(urUSMSharedAlloc)
+_UR_API(urUSMFree)
+_UR_API(urUSMGetMemAllocInfo)
+_UR_API(urUSMPoolCreate)
+_UR_API(urUSMPoolRetain)
+_UR_API(urUSMPoolRelease)
+_UR_API(urUSMPoolGetInfo)
+_UR_API(urUSMPitchedAllocExp)
+_UR_API(urUSMImportExp)
+_UR_API(urUSMReleaseExp)
+_UR_API(urCommandBufferCreateExp)
+_UR_API(urCommandBufferRetainExp)
+_UR_API(urCommandBufferReleaseExp)
+_UR_API(urCommandBufferFinalizeExp)
+_UR_API(urCommandBufferAppendKernelLaunchExp)
+_UR_API(urCommandBufferAppendUSMMemcpyExp)
+_UR_API(urCommandBufferAppendUSMFillExp)
+_UR_API(urCommandBufferAppendMemBufferCopyExp)
+_UR_API(urCommandBufferAppendMemBufferWriteExp)
+_UR_API(urCommandBufferAppendMemBufferReadExp)
+_UR_API(urCommandBufferAppendMemBufferCopyRectExp)
+_UR_API(urCommandBufferAppendMemBufferWriteRectExp)
+_UR_API(urCommandBufferAppendMemBufferReadRectExp)
+_UR_API(urCommandBufferAppendMemBufferFillExp)
+_UR_API(urCommandBufferAppendUSMPrefetchExp)
+_UR_API(urCommandBufferAppendUSMAdviseExp)
+_UR_API(urCommandBufferEnqueueExp)
+_UR_API(urCommandBufferRetainCommandExp)
+_UR_API(urCommandBufferReleaseCommandExp)
+_UR_API(urCommandBufferUpdateKernelLaunchExp)
+_UR_API(urCommandBufferGetInfoExp)
+_UR_API(urCommandBufferCommandGetInfoExp)
+_UR_API(urUsmP2PEnablePeerAccessExp)
+_UR_API(urUsmP2PDisablePeerAccessExp)
+_UR_API(urUsmP2PPeerAccessGetInfoExp)
+_UR_API(urVirtualMemGranularityGetInfo)
+_UR_API(urVirtualMemReserve)
+_UR_API(urVirtualMemFree)
+_UR_API(urVirtualMemMap)
+_UR_API(urVirtualMemUnmap)
+_UR_API(urVirtualMemSetAccess)
+_UR_API(urVirtualMemGetInfo)
+_UR_API(urDeviceGet)
+_UR_API(urDeviceGetInfo)
+_UR_API(urDeviceRetain)
+_UR_API(urDeviceRelease)
+_UR_API(urDevicePartition)
+_UR_API(urDeviceSelectBinary)
+_UR_API(urDeviceGetNativeHandle)
+_UR_API(urDeviceCreateWithNativeHandle)
+_UR_API(urDeviceGetGlobalTimestamps)
+_UR_API(urLoaderConfigCreate)
+_UR_API(urLoaderConfigEnableLayer)
+_UR_API(urLoaderConfigGetInfo)
+_UR_API(urLoaderConfigRelease)
+_UR_API(urLoaderConfigRetain)
+_UR_API(urLoaderConfigSetCodeLocationCallback)
+_UR_API(urLoaderConfigSetMockingEnabled)
+_UR_API(urLoaderInit)
+_UR_API(urLoaderTearDown)

scripts/benchmarks/benches/base.py

@@ -16,9 +16,12 @@
 class Benchmark:
     def __init__(self, directory):
         self.directory = directory
+        self.adapter_path = os.path.join(options.ur_dir, 'build', 'lib', f"libur_adapter_{options.ur_adapter_name}.so")
 
     def run_bench(self, command, env_vars):
-        return run(command=command, env_vars=env_vars, add_sycl=True, cwd=options.benchmark_cwd).stdout.decode()
+        env_vars_with_forced_adapter = env_vars.copy()
+        env_vars_with_forced_adapter.update({'UR_ADAPTERS_FORCE_LOAD': self.adapter_path})
+        return run(command=command, env_vars=env_vars_with_forced_adapter, add_sycl=True, cwd=options.benchmark_cwd).stdout.decode()
 
     def create_data_path(self, name):
         data_path = os.path.join(self.directory, "data", name)

scripts/benchmarks/benches/compute.py

@@ -15,7 +15,6 @@ class ComputeBench:
     def __init__(self, directory):
         self.directory = directory
         self.built = False
-        self.adapter_short_name = {'level_zero' : 'L0', "level_zero_v2" : 'L0_V2'}
         return
 
     def setup(self):
@@ -35,19 +34,16 @@ def setup(self):
             f"-DALLOW_WARNINGS=ON",
             f"-DBUILD_UR=ON",
             f"-DUR_BUILD_TESTS=OFF",
-            f"-DUR_BUILD_ADAPTER_L0=ON",
             f"-DUR_BUILD_TESTS=OFF",
             f"-DUMF_DISABLE_HWLOC=ON",
             f"-DBENCHMARK_UR_SOURCE_DIR={options.ur_dir}",
-            f"-DUR_BUILD_ADAPTER_{self.adapter_short_name[options.ur_adapter_name]}=ON"
         ]
         run(configure_command, add_sycl=True)
 
         run(f"cmake --build {build_path} -j", add_sycl=True)
 
         self.built = True
         self.bins = os.path.join(build_path, 'bin')
-        self.libs = os.path.join(build_path, 'lib')
 
 class ComputeBenchmark(Benchmark):
     def __init__(self, bench, name, test):
@@ -130,9 +126,6 @@ def name(self):
         order = "in order" if self.ioq else "out of order"
         return f"api_overhead_benchmark_ur SubmitKernel {order}"
 
-    def extra_env_vars(self) -> dict:
-        return {"UR_ADAPTERS_FORCE_LOAD" : os.path.join(self.bench.libs, f"libur_adapter_{options.ur_adapter_name}.so")}
-
     def bin_args(self) -> list[str]:
         return [
             f"--Ioq={self.ioq}",