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

Author: omarahmed1111

.github/CODEOWNERS

@@ -10,6 +10,9 @@ source/adapters/hip             @oneapi-src/unified-runtime-hip-write
 # OpenCL adapter
 source/adapters/opencl          @oneapi-src/unified-runtime-opencl-write
 
+# Native CPU adapter
+source/adapters/native_cpu          @oneapi-src/unified-runtime-native-cpu-write
+
 # Command-buffer experimental feature
 source/adapters/**/command_buffer.*  @oneapi-src/unified-runtime-command-buffer-write
 scripts/core/EXP-COMMAND-BUFFER.rst  @oneapi-src/unified-runtime-command-buffer-write

CMakeLists.txt

@@ -40,6 +40,7 @@ option(UR_BUILD_ADAPTER_L0 "build level 0 adapter from SYCL" OFF)
 option(UR_BUILD_ADAPTER_OPENCL "build opencl adapter from SYCL" OFF)
 option(UR_BUILD_ADAPTER_CUDA "build cuda adapter from SYCL" OFF)
 option(UR_BUILD_ADAPTER_HIP "build hip adapter from SYCL" OFF)
+option(UR_BUILD_ADAPTER_NATIVE_CPU "build native_cpu adapter from SYCL" OFF)
 option(UR_BUILD_EXAMPLE_CODEGEN "Build the codegen example." OFF)
 option(VAL_USE_LIBBACKTRACE_BACKTRACE "enable libbacktrace validation backtrace for linux" OFF)
 set(UR_DPCXX "" CACHE FILEPATH "Path of the DPC++ compiler executable")

README.md

@@ -134,6 +134,7 @@ List of options provided by CMake:
 | UR_BUILD_ADAPTER_OPENCL | Fetch and use opencl adapter from SYCL                 | ON/OFF     | OFF     |
 | UR_BUILD_ADAPTER_CUDA   | Fetch and use cuda adapter from SYCL                   | ON/OFF     | OFF     |
 | UR_BUILD_ADAPTER_HIP    | Fetch and use hip adapter from SYCL                    | ON/OFF     | OFF     |
+| UR_BUILD_ADAPTER_NATIVE_CPU | Fetch and use native-cpu adapter from SYCL         | ON/OFF     | OFF     |
 | UR_HIP_PLATFORM         | Build hip adapter for AMD or NVIDIA platform           | AMD/NVIDIA | AMD     |
 | UR_ENABLE_COMGR         | Enable comgr lib usage           | AMD/NVIDIA | AMD     |
 | UR_DPCXX | Path of the DPC++ compiler executable to build CTS device binaries | File path | `""` |

source/adapters/CMakeLists.txt

@@ -47,3 +47,6 @@ endif()
 if(UR_BUILD_ADAPTER_OPENCL)
     add_subdirectory(opencl)
 endif()
+if(UR_BUILD_ADAPTER_NATIVE_CPU)
+    add_subdirectory(native_cpu)
+endif()

source/adapters/cuda/ur_interface_loader.cpp

@@ -395,6 +395,8 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetEnqueueExpProcAddrTable(
     return result;
   }
 
+  pDdiTable->pfnCooperativeKernelLaunchExp = nullptr;
+
   return UR_RESULT_SUCCESS;
 }
 
@@ -405,6 +407,8 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetKernelExpProcAddrTable(
     return result;
   }
 
+  pDdiTable->pfnSuggestMaxCooperativeGroupCountExp = nullptr;
+
   return UR_RESULT_SUCCESS;
 }
 

source/adapters/hip/ur_interface_loader.cpp

@@ -349,6 +349,8 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetEnqueueExpProcAddrTable(
     return result;
   }
 
+  pDdiTable->pfnCooperativeKernelLaunchExp = nullptr;
+
   return UR_RESULT_SUCCESS;
 }
 
@@ -359,6 +361,8 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetKernelExpProcAddrTable(
     return result;
   }
 
+  pDdiTable->pfnSuggestMaxCooperativeGroupCountExp = nullptr;
+
   return UR_RESULT_SUCCESS;
 }
 

source/adapters/level_zero/ur_interface_loader.cpp

@@ -11,6 +11,8 @@
 #include <ur_api.h>
 #include <ur_ddi.h>
 
+namespace {
+
 ur_result_t validateProcInputs(ur_api_version_t version, void *pDdiTable) {
   if (nullptr == pDdiTable) {
     return UR_RESULT_ERROR_INVALID_NULL_POINTER;
@@ -22,6 +24,11 @@ ur_result_t validateProcInputs(ur_api_version_t version, void *pDdiTable) {
   }
   return UR_RESULT_SUCCESS;
 }
+} // namespace
+
+#if defined(__cplusplus)
+extern "C" {
+#endif
 
 UR_DLLEXPORT ur_result_t UR_APICALL urGetGlobalProcAddrTable(
     ur_api_version_t version, ///< [in] API version requested
@@ -435,6 +442,8 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetEnqueueExpProcAddrTable(
     return result;
   }
 
+  pDdiTable->pfnCooperativeKernelLaunchExp = nullptr;
+
   return UR_RESULT_SUCCESS;
 }
 
@@ -445,5 +454,10 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetKernelExpProcAddrTable(
     return result;
   }
 
+  pDdiTable->pfnSuggestMaxCooperativeGroupCountExp = nullptr;
+
   return UR_RESULT_SUCCESS;
 }
+#if defined(__cplusplus)
+} // extern "C"
+#endif

source/adapters/native_cpu/.clang-format

@@ -0,0 +1,4 @@
+---
+Language: Cpp
+BasedOnStyle: LLVM
+...

source/adapters/native_cpu/CMakeLists.txt

@@ -0,0 +1,58 @@
+# Copyright (C) 2023 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
+
+set(NATIVE_CPU_DIR "${CMAKE_CURRENT_SOURCE_DIR}" CACHE PATH "Native CPU adapter directory")
+
+set(TARGET_NAME ur_adapter_native_cpu)
+
+add_ur_adapter(${TARGET_NAME}
+        SHARED
+        ${CMAKE_CURRENT_SOURCE_DIR}/adapter.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/command_buffer.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/common.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/common.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/context.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/context.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/device.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/device.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/enqueue.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/event.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/image.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/kernel.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/kernel.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/memory.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/memory.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/nativecpu_state.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/platform.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/platform.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/program.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/program.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/queue.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/queue.hpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/sampler.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/ur_interface_loader.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/usm_p2p.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/usm.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/../../ur/ur.cpp
+        ${CMAKE_CURRENT_SOURCE_DIR}/../../ur/ur.hpp
+)
+
+set_target_properties(${TARGET_NAME} PROPERTIES
+        VERSION "${PROJECT_VERSION_MAJOR}.${PROJECT_VERSION_MINOR}.${PROJECT_VERSION_PATCH}"
+        SOVERSION "${PROJECT_VERSION_MAJOR}"
+)
+
+find_package(Threads REQUIRED)
+
+target_link_libraries(${TARGET_NAME} PRIVATE
+        ${PROJECT_NAME}::headers
+        ${PROJECT_NAME}::common
+        ${PROJECT_NAME}::unified_malloc_framework
+        Threads::Threads
+)
+
+target_include_directories(${TARGET_NAME} PRIVATE
+        "${CMAKE_CURRENT_SOURCE_DIR}/../../"
+)

source/adapters/native_cpu/adapter.cpp

@@ -0,0 +1,64 @@
+//===---------------- adapter.cpp - Native CPU Adapter --------------------===//
+//
+// Copyright (C) 2023 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
+//
+//===----------------------------------------------------------------------===//
+
+#include "common.hpp"
+#include "ur_api.h"
+
+struct ur_adapter_handle_t_ {
+  std::atomic<uint32_t> RefCount = 0;
+} Adapter;
+
+UR_APIEXPORT ur_result_t UR_APICALL urAdapterGet(
+    uint32_t, ur_adapter_handle_t *phAdapters, uint32_t *pNumAdapters) {
+  if (phAdapters) {
+    Adapter.RefCount++;
+    *phAdapters = &Adapter;
+  }
+  if (pNumAdapters) {
+    *pNumAdapters = 1;
+  }
+  return UR_RESULT_SUCCESS;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL urAdapterRelease(ur_adapter_handle_t) {
+  Adapter.RefCount--;
+  return UR_RESULT_SUCCESS;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL urAdapterRetain(ur_adapter_handle_t) {
+  Adapter.RefCount++;
+  return UR_RESULT_SUCCESS;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL urAdapterGetLastError(
+    ur_adapter_handle_t, const char **ppMessage, int32_t *pError) {
+  *ppMessage = ErrorMessage;
+  *pError = ErrorMessageCode;
+  return UR_RESULT_SUCCESS;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL urAdapterGetInfo(ur_adapter_handle_t,
+                                                     ur_adapter_info_t propName,
+                                                     size_t propSize,
+                                                     void *pPropValue,
+                                                     size_t *pPropSizeRet) {
+  UrReturnHelper ReturnValue(propSize, pPropValue, pPropSizeRet);
+
+  switch (propName) {
+  case UR_ADAPTER_INFO_BACKEND:
+    return ReturnValue(UR_ADAPTER_BACKEND_NATIVE_CPU);
+  case UR_ADAPTER_INFO_REFERENCE_COUNT:
+    return ReturnValue(Adapter.RefCount.load());
+  default:
+    return UR_RESULT_ERROR_INVALID_ENUMERATION;
+  }
+
+  return UR_RESULT_SUCCESS;
+}