Fix CUDA tracing when UR is built standalone

Previously, CUDA tracing only worked when the UR sources were built as part of
the CUDA PI plugin. FindCUDACupti.cmake is copied verbatim from intel/llvm.

Author: callumfare

cmake/FindCUDACupti.cmake

@@ -0,0 +1,26 @@
+# 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
+
+# This is lifted from intel-llvm's FindCUDACupti implementation
+# https://github.com/intel/llvm/blob/0cd04144d9ca83371c212e8e4709a59c968291b9/sycl/cmake/modules/FindCUDACupti.cmake
+
+macro(find_cuda_cupti_library)
+  find_library(CUDA_cupti_LIBRARY
+    NAMES cupti
+    HINTS ${CUDA_TOOLKIT_ROOT_DIR}
+          ENV CUDA_PATH
+    PATH_SUFFIXES nvidia/current lib64 lib/x64 lib
+                  ../extras/CUPTI/lib64/
+                  ../extras/CUPTI/lib/
+  )
+endmacro()
+
+macro(find_cuda_cupti_include_dir)
+  find_path(CUDA_CUPTI_INCLUDE_DIR cupti.h PATHS
+      "${CUDA_TOOLKIT_ROOT_DIR}/extras/CUPTI/include"
+      "${CUDA_INCLUDE_DIRS}/../extras/CUPTI/include"
+      "${CUDA_INCLUDE_DIRS}"
+      NO_DEFAULT_PATH)
+endmacro()

source/adapters/cuda/CMakeLists.txt

@@ -76,15 +76,46 @@ else()
   message(WARNING "CUDA adapter USM pools are disabled, set UMF_ENABLE_POOL_TRACKING to enable them")
 endif()
 
+if (UR_ENABLE_TRACING)
+  include(FindCUDACupti)
+  # The following two ifs can be removed when FindCUDA -> FindCUDAToolkit.
+  # CUDA_CUPTI_INCLUDE_DIR -> CUDAToolkit_CUPTI_INCLUDE_DIR
+  if(NOT CUDA_CUPTI_INCLUDE_DIR)
+    find_cuda_cupti_include_dir()
+  endif()
+  # CUDA_cupti_LIBRARY -> CUDAToolkit_cupti_LIBRARY
+  if(NOT CUDA_cupti_LIBRARY)
+    find_cuda_cupti_library()
+  endif()
+
+  if (NOT XPTI_INCLUDES)
+    get_target_property(XPTI_INCLUDES xpti INCLUDE_DIRECTORIES)
+  endif()
+  if (NOT XPTI_PROXY_SRC)
+    get_target_property(XPTI_SRC_DIR xpti SOURCE_DIR)
+    set(XPTI_PROXY_SRC "${XPTI_SRC_DIR}/xpti_proxy.cpp")
+  endif()
+  target_compile_definitions(${TARGET_NAME} PRIVATE
+    XPTI_ENABLE_INSTRUMENTATION
+    )
+  target_include_directories(${TARGET_NAME} PUBLIC
+    ${XPTI_INCLUDES}
+    ${CUDA_CUPTI_INCLUDE_DIR}
+  )
+  target_sources(${TARGET_NAME} PRIVATE ${XPTI_PROXY_SRC})
+endif()
+
 if (CUDA_cupti_LIBRARY)
   target_compile_definitions("ur_adapter_cuda" PRIVATE CUPTI_LIB_PATH="${CUDA_cupti_LIBRARY}")
+  list(APPEND EXTRA_LIBS ${CUDA_cupti_LIBRARY})
 endif()
 
 target_link_libraries(${TARGET_NAME} PRIVATE
     ${PROJECT_NAME}::headers
     ${PROJECT_NAME}::common
     Threads::Threads
     cudadrv
+    ${EXTRA_LIBS}
 )
 
 target_include_directories(${TARGET_NAME} PRIVATE