LLVMSPIRVLib.h: No such file or directory

[hpcpony]

Describe the bug

/opt/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/llvm/tools/spirv-to-ir-wrapper/spirv-to-ir-wrapper.cpp:22:10: fatal error: LLVMSPIRVLib.h: No such file or directory
22 | #include "LLVMSPIRVLib.h"
| ^~~~~~~~~~~~~~~~
compilation terminated.

To reproduce

git clone https://github.com/intel/llvm -b v6.3.0

cd llvm/llvm

cmake -S . -B build -DUR_BUILD_TESTS=OFF -DUR_ENABLE_TRACING=ON -DUR_BUILD_ADAPTER_CUDA=ON -DUR_BUILD_ADAPTER_HIP=OFF -DCMAKE_BUILD_TYPE=Release -DLLVM_TARGETS_TO_BUILD="NVPTX" -DCMAKE_INSTALL_PREFIX="/scr/oneapi/unified-runtime/install"

Environment

• OS: rhel 9.7
• Target device and vendor: NVPTX

Additional context

same problem in 6.2.1

[bratpiorka]

Hi @hpcpony,

I think some of the arguments you passed to CMake are invalid due to directory scope. For example, you cd into llvm/llvm, where the Unified Runtime project does not exist (it is located in the top-level directory), yet you are using Unified Runtime–specific arguments. CMake prints warnings about this:

CMake Warning:
  Manually-specified variables were not used by the project:

    UR_BUILD_ADAPTER_CUDA
    UR_BUILD_ADAPTER_HIP
    UR_BUILD_TESTS
    UR_ENABLE_TRACING

On the other hand, if you use unified-runtime as the source directory instead, then LLVM_TARGETS_TO_BUILD will not be recognized.

If you let me know what effect you want to achieve, I can help you further. If you're simply interested in setting up a build environment for CUDA, you can build the project using the python wrapper:

git clone https://github.com/intel/llvm -b v6.3.0
cd llvm
python3 buildbot/configure.py -t Release -o build --cuda --cmake-gen "Ninja"
cmake --build build

Please let me know whether this resolves your issue so I can close it, or if there's anything else I can help you with.

[hpcpony]

Sorry, my goal is to build the NVIDIA plugin (like what I used to just download from Codeplay) so I can use OneAPI with GPU(s). I tried starting from the announcement in https://www.intel.com/content/www/us/en/developer/articles/release-notes/oneapi-dpcpp/2025.html but it tries to build all the LLVM targets which is overkill (and time consuming). I'd like to just build the parts for NVIDIA (or AMD eventually) as necessary. So, I guess if you can point me in that direction I'd be most grateful.

If this is the new procedure it might also be nice to have some good compatibility information between OneAPI releases and what we should build from this repo.

[bratpiorka]

Here is how you can build the CUDA plugin so that it works with oneAPI 2025.3:

git clone https://github.com/intel/llvm -b v6.3.0 --depth=1

cd llvm
cmake -S unified-runtime -B build \
  -DUR_BUILD_TESTS=OFF \
  -DUR_ENABLE_TRACING=ON \
  -DUR_BUILD_ADAPTER_CUDA=ON \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_INSTALL_PREFIX="_install"

cmake --build build -j
cmake --install build

After the build completes, copy the _install/lib/libur* files into the oneAPI installation directory:
$ONEAPI_ROOT/compiler/latest/lib
Then initialize the oneAPI environment variables and test the setup:

source $ONEAPI_ROOT/setvars.sh
sycl-ls

Please let me know if this works for you.

[hpcpony]

Do I need to build with the Intel compiler or with a LLVM compiler? I tried with gcc and ran into some issues.

There seems to be some sort of strange issue with cmake 4.2.1:

[gpu2:6.3.0] git clone https://github.com/intel/llvm -b v6.3.0 --depth=1
Cloning into 'llvm'...
remote: Enumerating objects: 177277, done.
remote: Counting objects: 100% (177277/177277), done.
remote: Compressing objects: 100% (133656/133656), done.
remote: Total 177277 (delta 48624), reused 80501 (delta 38967), pack-reused 0 (from 0)
Receiving objects: 100% (177277/177277), 269.86 MiB | 7.96 MiB/s, done.
Resolving deltas: 100% (48624/48624), done.
Note: switching to 'd5f649b706f63b5c74e1929bc95db8de91085560'.

You are in 'detached HEAD' state. You can look around, make experimental
changes and commit them, and you can discard any commits you make in this
state without impacting any branches by switching back to a branch.

If you want to create a new branch to retain commits you create, you may
do so (now or later) by using -c with the switch command. Example:

  git switch -c <new-branch-name>

Or undo this operation with:

  git switch -

Turn off this advice by setting config variable advice.detachedHead to false

Updating files: 100% (168594/168594), done.
[gpu2:6.3.0] cd llvm
[gpu2:llvm] cmake -S unified-runtime -B build \
  -DUR_BUILD_TESTS=OFF \
  -DUR_ENABLE_TRACING=ON \
  -DUR_BUILD_ADAPTER_CUDA=ON \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_INSTALL_PREFIX="_install"
-- The C compiler identification is GNU 11.5.0
-- The CXX compiler identification is GNU 11.5.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Performing Test CXX_HAS_FCF_PROTECTION_FULL
-- Performing Test CXX_HAS_FCF_PROTECTION_FULL - Success
-- Performing Test CXX_HAS_FSTACK_CLASH_PROTECTION
-- Performing Test CXX_HAS_FSTACK_CLASH_PROTECTION - Success
-- Found Python3: /usr/bin/python3.9 (found version "3.9.25") found components: Interpreter
-- Performing Test USING_LIBSTDCXX
-- Performing Test USING_LIBSTDCXX - Success
-- Will fetch emhash from https://github.com/ktprime/emhash
-- Will fetch parallel-hashmap from https://github.com/greg7mdp/parallel-hashmap.git
-- Detected PHMAP Version - 1.3.12
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Success
-- Found Threads: TRUE
-- Performing Test HAS_MAYBE_UNINIT
-- Performing Test HAS_MAYBE_UNINIT - Success
-- Found CUDAToolkit: /usr/local/cuda/include (found suitable version "13.1.115", minimum required is "10.1")
-- Download Unified Memory Framework from github.com
-- Will fetch Unified Memory Framework from https://github.com/oneapi-src/unified-memory-framework.git
-- CMake version: 4.2.1
-- UMF version: 1.0.0-rc2
-- Found PkgConfig: /usr/bin/pkg-config (found version "1.7.3")
-- CMAKE_GENERATOR: Unix Makefiles
-- CMAKE_PREFIX_PATH=
-- Disabling the Jemalloc Pool and tests and benchmarks that use it because jemalloc was not built/found.
-- Checking for module 'hwloc>=2.3.0'
--   Found hwloc, version 2.4.1
--     LIBHWLOC_LIBRARIES = hwloc
--     LIBHWLOC_INCLUDE_DIRS = /usr/include
--     LIBHWLOC_LIBRARY_DIRS = /usr/lib64
--     LIBHWLOC_API_VERSION = 
-- Checking for module 'cuda' using find_library()
--   Found cuda using find_library()
--     CUDA_LIBRARIES = /usr/lib64/libcuda.so
--     CUDA_INCLUDE_DIRS = 
--     CUDA_LIBRARY_DIRS = /usr/lib64
-- CUDA_INCLUDE_DIRS = /usr/local/cuda/include
-- CMAKE_BUILD_TYPE: Release
-- Checking for module 'tbb'
--   Package 'tbb', required by 'virtual:world', not found
-- Checking for module 'tbb' using find_library()
-- <tbb/scalable_allocator.h> header NOT found (set CMAKE_PREFIX_PATH to point the location)
-- tbb NOT found (set CMAKE_PREFIX_PATH to point the location)
-- Disabling tests and benchmarks that use the Scalable Pool because the TBB they require was not found.
-- Disabling the proxy library, because UMF_PROXY_LIB_BASED_ON_POOL==SCALABLE but TBB was not found
-- Performing Test C_HAS_-Wno-covered-switch-default
-- Performing Test C_HAS_-Wno-covered-switch-default - Failed
-- Performing Test CXX_HAS_-Wno-covered-switch-default
-- Performing Test CXX_HAS_-Wno-covered-switch-default - Failed
-- Found Python3: /usr/bin/python3.9 (found suitable version "3.9.25", minimum required is "3.8") found components: Interpreter
-- Adding 'docs' target for creating a documentation.
-- Could NOT find Libbacktrace (missing: LIBBACKTRACE_LIBRARIES LIBBACKTRACE_INCLUDE_DIR) 
-- Using default backtrace for validation
-- Found CUDAToolkit: /usr/local/cuda/targets/x86_64-linux/include;/usr/local/cuda/targets/x86_64-linux/include/cccl (found suitable version "13.1.115", minimum required is "10.1")
CMake Error at source/adapters/cuda/CMakeLists.txt:70 (set_target_properties):
  set_target_properties called with incorrect number of arguments.


--   UR_FORMAT_CPP_STYLE not set. Targets: 'generate' and 'check-generated' are not available
-- Configuring incomplete, errors occurred!

but when I used an older cmake 3.30.1 it got further but then ran into a bigger error:

[gpu2:llvm] cmake -S unified-runtime -B build \
  -DUR_BUILD_TESTS=OFF \
  -DUR_ENABLE_TRACING=ON \
  -DUR_BUILD_ADAPTER_CUDA=ON \
  -DCMAKE_BUILD_TYPE=Release \
  -DCMAKE_INSTALL_PREFIX="_install"
-- The C compiler identification is GNU 11.5.0
-- The CXX compiler identification is GNU 11.5.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Performing Test CXX_HAS_FCF_PROTECTION_FULL
-- Performing Test CXX_HAS_FCF_PROTECTION_FULL - Success
-- Performing Test CXX_HAS_FSTACK_CLASH_PROTECTION
-- Performing Test CXX_HAS_FSTACK_CLASH_PROTECTION - Success
-- Found Python3: /usr/bin/python3.9 (found version "3.9.25") found components: Interpreter
-- Performing Test USING_LIBSTDCXX
-- Performing Test USING_LIBSTDCXX - Success
-- Will fetch emhash from https://github.com/ktprime/emhash
-- Will fetch parallel-hashmap from https://github.com/greg7mdp/parallel-hashmap.git
-- Detected PHMAP Version - 1.3.12
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Success
-- Found Threads: TRUE
-- Performing Test HAS_MAYBE_UNINIT
-- Performing Test HAS_MAYBE_UNINIT - Success
-- Found CUDAToolkit: /usr/local/cuda/targets/x86_64-linux/include (found suitable version "13.1.115", minimum required is "10.1")
-- Download Unified Memory Framework from github.com
-- Will fetch Unified Memory Framework from https://github.com/oneapi-src/unified-memory-framework.git
-- CMake version: 3.30.2
-- UMF version: 1.0.0-rc2
-- Found PkgConfig: /usr/bin/pkg-config (found version "1.7.3")
-- CMAKE_GENERATOR: Unix Makefiles
-- CMAKE_PREFIX_PATH=
-- Disabling the Jemalloc Pool and tests and benchmarks that use it because jemalloc was not built/found.
-- Checking for module 'hwloc>=2.3.0'
--   Found hwloc, version 2.4.1
--     LIBHWLOC_LIBRARIES = hwloc
--     LIBHWLOC_INCLUDE_DIRS = /usr/include
--     LIBHWLOC_LIBRARY_DIRS = /usr/lib64
--     LIBHWLOC_API_VERSION = 
-- Checking for module 'cuda' using find_library()
--   Found cuda using find_library()
--     CUDA_LIBRARIES = /usr/lib64/libcuda.so
--     CUDA_INCLUDE_DIRS = 
--     CUDA_LIBRARY_DIRS = /usr/lib64
-- CUDA_INCLUDE_DIRS = /usr/local/cuda/targets/x86_64-linux/include
-- CMAKE_BUILD_TYPE: Release
-- Checking for module 'tbb'
--   Package 'tbb', required by 'virtual:world', not found
-- Checking for module 'tbb' using find_library()
-- <tbb/scalable_allocator.h> header NOT found (set CMAKE_PREFIX_PATH to point the location)
-- tbb NOT found (set CMAKE_PREFIX_PATH to point the location)
-- Disabling tests and benchmarks that use the Scalable Pool because the TBB they require was not found.
-- Disabling the proxy library, because UMF_PROXY_LIB_BASED_ON_POOL==SCALABLE but TBB was not found
-- Performing Test C_HAS_-Wno-covered-switch-default
-- Performing Test C_HAS_-Wno-covered-switch-default - Failed
-- Performing Test CXX_HAS_-Wno-covered-switch-default
-- Performing Test CXX_HAS_-Wno-covered-switch-default - Failed
-- Found Python3: /usr/bin/python3.9 (found suitable version "3.9.25", minimum required is "3.8") found components: Interpreter
-- Adding 'docs' target for creating a documentation.
-- Could NOT find Libbacktrace (missing: LIBBACKTRACE_LIBRARIES LIBBACKTRACE_INCLUDE_DIR) 
-- Using default backtrace for validation
--   UR_FORMAT_CPP_STYLE not set. Targets: 'generate' and 'check-generated' are not available
-- Configuring done (23.1s)
-- Generating done (2.6s)
-- Build files have been written to: /src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/build
[gpu2:llvm] cmake --build build -j 1
[  0%] Building CXX object _deps/xpti-build/src/CMakeFiles/xpti.dir/xpti_proxy.cpp.o
[  1%] Linking CXX static library ../../../lib/libxpti.a
[  1%] Built target xpti
[  2%] Building CXX object _deps/xptifw-build/src/CMakeFiles/xptifw.dir/xpti_trace_framework.cpp.o
[  3%] Linking CXX shared library ../../../lib/libxptifw.so
[  3%] Built target xptifw
[  4%] Building CXX object source/common/CMakeFiles/ur_common.dir/ur_util.cpp.o
[  4%] Building CXX object source/common/CMakeFiles/ur_common.dir/offload_bundle_parser.cpp.o
[  5%] Building CXX object source/common/CMakeFiles/ur_common.dir/linux/ur_lib_loader.cpp.o
[  6%] Linking CXX static library ../../lib/libur_common.a
[  6%] Built target ur_common
[  7%] Building C object _deps/unified-memory-framework-build/src/utils/CMakeFiles/umf_utils.dir/utils_common.c.o
[  8%] Building C object _deps/unified-memory-framework-build/src/utils/CMakeFiles/umf_utils.dir/utils_log.c.o
[  9%] Building C object _deps/unified-memory-framework-build/src/utils/CMakeFiles/umf_utils.dir/utils_load_library.c.o
[  9%] Building C object _deps/unified-memory-framework-build/src/utils/CMakeFiles/umf_utils.dir/utils_posix_common.c.o
[ 10%] Building C object _deps/unified-memory-framework-build/src/utils/CMakeFiles/umf_utils.dir/utils_posix_concurrency.c.o
[ 11%] Building C object _deps/unified-memory-framework-build/src/utils/CMakeFiles/umf_utils.dir/utils_linux_common.c.o
[ 11%] Linking C static library ../../../../lib/libumf_utils.a
[ 11%] Built target umf_utils
[ 11%] Building C object _deps/unified-memory-framework-build/src/base_alloc/CMakeFiles/umf_ba.dir/base_alloc_global.c.o
[ 12%] Building C object _deps/unified-memory-framework-build/src/base_alloc/CMakeFiles/umf_ba.dir/base_alloc.c.o
[ 13%] Building C object _deps/unified-memory-framework-build/src/base_alloc/CMakeFiles/umf_ba.dir/base_alloc_linear.c.o
[ 14%] Building C object _deps/unified-memory-framework-build/src/base_alloc/CMakeFiles/umf_ba.dir/base_alloc_linux.c.o
[ 14%] Linking C static library ../../../../lib/libumf_ba.a
[ 14%] Built target umf_ba
[ 15%] Building C object _deps/unified-memory-framework-build/src/coarse/CMakeFiles/umf_coarse.dir/coarse.c.o
[ 16%] Building C object _deps/unified-memory-framework-build/src/coarse/CMakeFiles/umf_coarse.dir/__/__/__/unified-memory-framework-src/src/ravl/ravl.c.o
[ 16%] Linking C static library ../../../../lib/libumf_coarse.a
[ 16%] Built target umf_coarse
[ 17%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/ctl/ctl.c.o
[ 17%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/libumf.c.o
[ 18%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/ipc.c.o
[ 19%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/ipc_cache.c.o
[ 20%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memory_pool.c.o
[ 20%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memory_provider.c.o
[ 21%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memory_provider_get_last_failed.c.o
[ 22%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memtarget.c.o
[ 22%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/mempolicy.c.o
[ 23%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memspace.c.o
[ 24%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memspaces/memspace_host_all.c.o
[ 25%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memspaces/memspace_highest_capacity.c.o
[ 25%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memspaces/memspace_highest_bandwidth.c.o
[ 26%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memspaces/memspace_lowest_latency.c.o
[ 27%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memspaces/memspace_numa.c.o
[ 27%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/provider/provider_cuda.c.o
[ 28%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/provider/provider_devdax_memory.c.o
[ 29%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/provider/provider_file_memory.c.o
[ 30%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/provider/provider_fixed_memory.c.o
[ 30%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/provider/provider_level_zero.c.o
[ 31%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/provider/provider_os_memory.c.o
[ 32%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/provider/provider_tracking.c.o
[ 32%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/critnib/critnib.c.o
[ 34%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/ravl/ravl.c.o
[ 35%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/pool/pool_disjoint.c.o
[ 35%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/pool/pool_jemalloc.c.o
[ 36%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/pool/pool_proxy.c.o
[ 37%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/pool/pool_scalable.c.o
[ 38%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/topology.c.o
[ 38%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/memtargets/memtarget_numa.c.o
[ 39%] Building C object _deps/unified-memory-framework-build/src/CMakeFiles/umf.dir/libumf_linux.c.o
[ 40%] Linking C shared library ../../../lib/libumf.so
[ 40%] Built target umf
[ 41%] Building CXX object source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/ur_interface_loader.cpp.o
[ 42%] Building CXX object source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/adapter.cpp.o
[ 43%] Building CXX object source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/async_alloc.cpp.o
[ 43%] Building CXX object source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/command_buffer.cpp.o
[ 44%] Building CXX object source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/common.cpp.o
[ 45%] Building CXX object source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/context.cpp.o
[ 45%] Building CXX object source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/device.cpp.o
[ 46%] Building CXX object source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/enqueue.cpp.o
In file included from /src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.hpp:12,
                 from /src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp:11:
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp: In function ‘void setCuMemAdvise(CUdeviceptr, size_t, ur_usm_advice_flags_t, CUdevice)’:
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp:67:65: error: could not convert ‘Device’ from ‘CUdevice’ {aka ‘int’} to ‘CUmemLocation’ {aka ‘CUmemLocation_st’}
   67 |       UR_CHECK_ERROR(cuMemAdvise(DevPtr, Size, FlagPair.second, Device));
      |                                                                 ^~~~~~
      |                                                                 |
      |                                                                 CUdevice {aka int}
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/common.hpp:40:16: note: in definition of macro ‘UR_CHECK_ERROR’
   40 |   checkErrorUR(Result, __func__, __LINE__, __FILE__)
      |                ^~~~~~
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp:85:65: error: could not convert ‘-1’ from ‘CUdevice’ {aka ‘int’} to ‘CUmemLocation’ {aka ‘CUmemLocation_st’}
   85 |       UR_CHECK_ERROR(cuMemAdvise(DevPtr, Size, FlagPair.second, CU_DEVICE_CPU));
      |                                                                 ^~~~~~~~~~~~~
      |                                                                 |
      |                                                                 CUdevice {aka int}
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/common.hpp:40:16: note: in definition of macro ‘UR_CHECK_ERROR’
   40 |   checkErrorUR(Result, __func__, __LINE__, __FILE__)
      |                ^~~~~~
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp: In function ‘ur_result_t urEnqueueUSMPrefetch(ur_queue_handle_t, const void*, size_t, ur_usm_migration_flags_t, uint32_t, ur_event_handle_t_* const*, ur_event_handle_t_**)’:
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp:1554:64: error: could not convert ‘Device->ur_device_handle_t_::get()’ from ‘ur_device_handle_t_::native_type’ {aka ‘int’} to ‘CUmemLocation’ {aka ‘CUmemLocation_st’}
 1554 |         cuMemPrefetchAsync((CUdeviceptr)pMem, size, Device->get(), CuStream));
      |                                                     ~~~~~~~~~~~^~
      |                                                                |
      |                                                                ur_device_handle_t_::native_type {aka int}
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/common.hpp:40:16: note: in definition of macro ‘UR_CHECK_ERROR’
   40 |   checkErrorUR(Result, __func__, __LINE__, __FILE__)
      |                ^~~~~~
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp: In function ‘ur_result_t urEnqueueUSMAdvise(ur_queue_handle_t, const void*, size_t, ur_usm_advice_flags_t, ur_event_handle_t_**)’:
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp:1625:58: error: could not convert ‘hQueue->ur_queue_handle_t_::<anonymous>.stream_queue_t<CUstream_st*, 128, 64, CUevent_st*>::getDevice()->ur_device_handle_t_::get()’ from ‘ur_device_handle_t_::native_type’ {aka ‘int’} to ‘CUmemLocation’ {aka ‘CUmemLocation_st’}
 1625 |                                  hQueue->getDevice()->get()));
      |                                  ~~~~~~~~~~~~~~~~~~~~~~~~^~
      |                                                          |
      |                                                          ur_device_handle_t_::native_type {aka int}
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/common.hpp:40:16: note: in definition of macro ‘UR_CHECK_ERROR’
   40 |   checkErrorUR(Result, __func__, __LINE__, __FILE__)
      |                ^~~~~~
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp:1628:58: error: could not convert ‘hQueue->ur_queue_handle_t_::<anonymous>.stream_queue_t<CUstream_st*, 128, 64, CUevent_st*>::getDevice()->ur_device_handle_t_::get()’ from ‘ur_device_handle_t_::native_type’ {aka ‘int’} to ‘CUmemLocation’ {aka ‘CUmemLocation_st’}
 1628 |                                  hQueue->getDevice()->get()));
      |                                  ~~~~~~~~~~~~~~~~~~~~~~~~^~
      |                                                          |
      |                                                          ur_device_handle_t_::native_type {aka int}
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/common.hpp:40:16: note: in definition of macro ‘UR_CHECK_ERROR’
   40 |   checkErrorUR(Result, __func__, __LINE__, __FILE__)
      |                ^~~~~~
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/enqueue.cpp:1631:58: error: could not convert ‘hQueue->ur_queue_handle_t_::<anonymous>.stream_queue_t<CUstream_st*, 128, 64, CUevent_st*>::getDevice()->ur_device_handle_t_::get()’ from ‘ur_device_handle_t_::native_type’ {aka ‘int’} to ‘CUmemLocation’ {aka ‘CUmemLocation_st’}
 1631 |                                  hQueue->getDevice()->get()));
      |                                  ~~~~~~~~~~~~~~~~~~~~~~~~^~
      |                                                          |
      |                                                          ur_device_handle_t_::native_type {aka int}
/src/Companies/Intel/Software/OneAPI/NVIDIA/6.3.0/llvm/unified-runtime/source/adapters/cuda/common.hpp:40:16: note: in definition of macro ‘UR_CHECK_ERROR’
   40 |   checkErrorUR(Result, __func__, __LINE__, __FILE__)
      |                ^~~~~~
gmake[2]: *** [source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/build.make:174: source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/enqueue.cpp.o] Error 1
gmake[1]: *** [CMakeFiles/Makefile2:1708: source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/all] Error 2
gmake: *** [Makefile:146: all] Error 2

I couldn't find an immediate answer as to why there's the type mismatch.

[bratpiorka]

for

CMake Error at source/adapters/cuda/CMakeLists.txt:70 (set_target_properties):
  set_target_properties called with incorrect number of arguments.

please apply 3c2678c