[UR][Offload] Add initial Offload adapter

unified-runtime/CMakeLists.txt

@@ -47,6 +47,7 @@ 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_BUILD_ADAPTER_OFFLOAD "Build the experimental Offload 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)

unified-runtime/scripts/core/common.yml

@@ -362,3 +362,6 @@ etors:
     - name: NATIVE_CPU
       value: "5"
       desc: "The backend is Native CPU"
+    - name: OFFLOAD
+      value: "0x100"
+      desc: "The backend is liboffload"

unified-runtime/scripts/core/manifests.yml

@@ -61,3 +61,10 @@ name: native_cpu
 backend: $X_BACKEND_NATIVE_CPU
 device_types:
     - $X_DEVICE_TYPE_CPU
+--- #--------------------------------------------------------------------------
+type: manifest
+name: offload
+backend: $X_BACKEND_OFFLOAD
+device_types:
+    - $X_DEVICE_TYPE_CPU
+    - $X_DEVICE_TYPE_GPU

unified-runtime/source/adapters/CMakeLists.txt

@@ -76,9 +76,15 @@ if(UR_BUILD_ADAPTER_OPENCL OR UR_BUILD_ADAPTER_ALL)
     add_ur_adapter_subdirectory(opencl)
     list(APPEND TEMP_LIST "opencl")
 endif()
+
 if(UR_BUILD_ADAPTER_NATIVE_CPU OR UR_BUILD_ADAPTER_ALL)
     add_ur_adapter_subdirectory(native_cpu)
     list(APPEND TEMP_LIST "native_cpu")
 endif()
 
+if(UR_BUILD_ADAPTER_OFFLOAD)
+    add_ur_adapter_subdirectory(offload)
+    list(APPEND TEMP_LIST "offload")
+endif()
+
 set(UR_ADAPTERS_LIST "${TEMP_LIST}" CACHE STRING "" FORCE)

unified-runtime/source/adapters/offload/CMakeLists.txt

@@ -0,0 +1,70 @@
+# Copyright (C) 2025 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(TARGET_NAME ur_adapter_offload)
+
+set(UR_OFFLOAD_INSTALL_DIR "" CACHE PATH "Path to the directory containing libomptarget.so etc")
+if (UR_OFFLOAD_INSTALL_DIR STREQUAL "")
+    message(FATAL_ERROR "UR_OFFLOAD_INSTALL_DIR must be defined for the Offload adapter")
+endif()
+
+set(UR_OFFLOAD_INCLUDE_DIR "" CACHE PATH "Path to the directory containing LLVM headers")
+if (UR_OFFLOAD_INCLUDE_DIR STREQUAL "")
+    message(FATAL_ERROR "UR_OFFLOAD_INCLUDE_DIR must be defined for the Offload adapter")
+endif()
+
+# When targetting CUDA devices, we need a workaround to avoid sending PTX to
+# liboffload as the CUDA plugin doesn't support it yet. The workaround is to
+# simply always link the incoming program so it ends up as CUBIN. Try to find
+# the cuda driver so we can enable this where possible.
+if (NOT TARGET cudadrv)
+    find_package(CUDA 10.1)
+    add_library(cudadrv SHARED IMPORTED GLOBAL)
+    set_target_properties(
+        cudadrv PROPERTIES
+        IMPORTED_LOCATION             ${CUDA_cuda_driver_LIBRARY}
+        INTERFACE_INCLUDE_DIRECTORIES ${CUDAToolkit_INCLUDE_DIRS}
+    )
+endif()
+
+add_ur_adapter(${TARGET_NAME}
+    SHARED
+    ${CMAKE_CURRENT_SOURCE_DIR}/adapter.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/context.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/device.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/enqueue.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/event.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/kernel.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/platform.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/program.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/queue.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/ur2offload.hpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/ur_interface_loader.cpp
+    ${CMAKE_CURRENT_SOURCE_DIR}/usm.cpp
+)
+
+set_target_properties(${TARGET_NAME} PROPERTIES
+    VERSION "${PROJECT_VERSION_MAJOR}.${PROJECT_VERSION_MINOR}.${PROJECT_VERSION_PATCH}"
+    SOVERSION "${PROJECT_VERSION_MAJOR}"
+)
+
+set(ADDITIONAL_LINK_LIBS "")
+if (CUDA_cuda_driver_LIBRARY)
+    list(APPEND ADDITIONAL_LINK_LIBS cudadrv)
+    target_compile_definitions(${TARGET_NAME} PRIVATE UR_CUDA_ENABLED)
+endif()
+
+target_link_libraries(${TARGET_NAME} PRIVATE
+    ${PROJECT_NAME}::headers
+    ${PROJECT_NAME}::common
+    ${PROJECT_NAME}::umf
+    ${UR_OFFLOAD_INSTALL_DIR}/lib/libLLVMOffload.so
+    ${ADDITIONAL_LINK_LIBS}
+)
+
+target_include_directories(${TARGET_NAME} PRIVATE
+    "${UR_OFFLOAD_INCLUDE_DIR}/offload"
+    "${CMAKE_CURRENT_SOURCE_DIR}/../../"
+)

unified-runtime/source/adapters/offload/adapter.cpp

@@ -0,0 +1,109 @@
+//===----------- adapter.cpp - LLVM Offload Adapter  ----------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#include <OffloadAPI.h>
+#include <atomic>
+#include <cstdint>
+#include <unordered_set>
+
+#include "adapter.hpp"
+#include "device.hpp"
+#include "platform.hpp"
+#include "ur/ur.hpp"
+#include "ur_api.h"
+
+ur_adapter_handle_t_ Adapter{};
+
+// Initialize liboffload and perform the initial platform and device discovery
+ur_result_t ur_adapter_handle_t_::init() {
+  auto Res = olInit();
+  (void)Res;
+
+  // Discover every platform and device
+  Res = olIterateDevices(
+      [](ol_device_handle_t D, void *UserData) {
+        auto *Platforms =
+            reinterpret_cast<decltype(Adapter.Platforms) *>(UserData);
+
+        ol_platform_handle_t Platform;
+        olGetDeviceInfo(D, OL_DEVICE_INFO_PLATFORM, sizeof(Platform),
+                        &Platform);
+        ol_platform_backend_t Backend;
+        olGetPlatformInfo(Platform, OL_PLATFORM_INFO_BACKEND, sizeof(Backend),
+                          &Backend);
+        if (Backend == OL_PLATFORM_BACKEND_HOST) {
+          Adapter.HostDevice = D;
+        } else if (Backend != OL_PLATFORM_BACKEND_UNKNOWN) {
+          auto URPlatform =
+              std::find_if(Platforms->begin(), Platforms->end(), [&](auto &P) {
+                return P.OffloadPlatform == Platform;
+              });
+
+          if (URPlatform == Platforms->end()) {
+            URPlatform =
+                Platforms->insert(URPlatform, ur_platform_handle_t_(Platform));
+          }
+
+          URPlatform->Devices.push_back(ur_device_handle_t_{&*URPlatform, D});
+        }
+        return false;
+      },
+      &Adapter.Platforms);
+
+  (void)Res;
+
+  return UR_RESULT_SUCCESS;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL urAdapterGet(
+    uint32_t, ur_adapter_handle_t *phAdapters, uint32_t *pNumAdapters) {
+  if (phAdapters) {
+    if (++Adapter.RefCount == 1) {
+      Adapter.init();
+    }
+    *phAdapters = &Adapter;
+  }
+  if (pNumAdapters) {
+    *pNumAdapters = 1;
+  }
+  return UR_RESULT_SUCCESS;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL urAdapterRelease(ur_adapter_handle_t) {
+  if (--Adapter.RefCount == 0) {
+    // This can crash when tracing is enabled.
+    // olShutDown();
+  };
+  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 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_BACKEND_OFFLOAD);
+  case UR_ADAPTER_INFO_REFERENCE_COUNT:
+    return ReturnValue(Adapter.RefCount.load());
+  default:
+    return UR_RESULT_ERROR_INVALID_ENUMERATION;
+  }
+
+  return UR_RESULT_SUCCESS;
+}

unified-runtime/source/adapters/offload/adapter.hpp

@@ -0,0 +1,32 @@
+//===----------- adapter.hpp - LLVM Offload Adapter  ----------------------===//
+//
+// Copyright (C) 2025 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
+//
+//===----------------------------------------------------------------------===//
+
+#pragma once
+
+#include <atomic>
+#include <cstdint>
+#include <unordered_set>
+
+#include <OffloadAPI.h>
+
+#include "common.hpp"
+#include "logger/ur_logger.hpp"
+#include "platform.hpp"
+
+struct ur_adapter_handle_t_ : ur::offload::handle_base {
+  std::atomic_uint32_t RefCount = 0;
+  logger::Logger &Logger = logger::get_logger("offload");
+  ol_device_handle_t HostDevice = nullptr;
+  std::vector<ur_platform_handle_t_> Platforms;
+
+  ur_result_t init();
+};
+
+extern ur_adapter_handle_t_ Adapter;

unified-runtime/source/adapters/offload/common.hpp

@@ -0,0 +1,25 @@
+//===----------- common.hpp - LLVM Offload Adapter  -----------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+
+#pragma once
+
+#include "ur/ur.hpp"
+#include <atomic>
+
+namespace ur::offload {
+struct ddi_getter {
+  const static ur_dditable_t *value();
+};
+using handle_base = ur::handle_base<ur::offload::ddi_getter>;
+} // namespace ur::offload
+
+struct RefCounted : ur::offload::handle_base {
+  std::atomic_uint32_t RefCount = 1;
+};

unified-runtime/source/adapters/offload/context.cpp

@@ -0,0 +1,38 @@
+//===----------- context.cpp - LLVM Offload Adapter  ----------------------===//
+//
+// Copyright (C) 2025 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 "context.hpp"
+#include <ur_api.h>
+
+UR_APIEXPORT ur_result_t UR_APICALL urContextCreate(
+    uint32_t DeviceCount, const ur_device_handle_t *phDevices,
+    const ur_context_properties_t *, ur_context_handle_t *phContext) {
+  if (DeviceCount > 1) {
+    return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+  }
+
+  auto Ctx = new ur_context_handle_t_(*phDevices);
+  *phContext = Ctx;
+  return UR_RESULT_SUCCESS;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL
+urContextRetain(ur_context_handle_t hContext) {
+  hContext->RefCount++;
+  return UR_RESULT_SUCCESS;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL
+urContextRelease(ur_context_handle_t hContext) {
+  if (--hContext->RefCount == 0) {
+    delete hContext;
+  }
+  return UR_RESULT_SUCCESS;
+}

unified-runtime/source/adapters/offload/context.hpp

@@ -0,0 +1,26 @@
+//===----------- context.hpp - LLVM Offload Adapter  ----------------------===//
+//
+// Copyright (C) 2025 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
+//
+//===----------------------------------------------------------------------===//
+
+#pragma once
+
+#include "common.hpp"
+#include <OffloadAPI.h>
+#include <unordered_map>
+#include <ur_api.h>
+
+struct ur_context_handle_t_ : RefCounted {
+  ur_context_handle_t_(ur_device_handle_t hDevice) : Device{hDevice} {
+    urDeviceRetain(Device);
+  }
+  ~ur_context_handle_t_() { urDeviceRelease(Device); }
+
+  ur_device_handle_t Device;
+  std::unordered_map<void *, ol_alloc_type_t> AllocTypeMap;
+};