Graph/NativeCommand/cuda_record_async_malloc.cpp failing on CUDA even with expected CUDA version

[sarnex]

Describe the bug

I updated the AWS image we use in the Nightly for CUDA to use CUDA 13, so now the runtime CUDA version check in the test passes, but the test still fails.

Note our self-hosted runners like the run in precommit are not yet updated to CUDA 13 so the runtime version check will fail and the test will be skipped

https://github.com/intel/llvm/actions/runs/22292260783/job/64482549481

  FAIL: SYCL :: Graph/NativeCommand/cuda_record_async_malloc.cpp (1292 of 2421)
  ******************** TEST 'SYCL :: Graph/NativeCommand/cuda_record_async_malloc.cpp' FAILED ********************
  Exit Code: -6
  
  Command Output (stdout):
  --
  # RUN: at line 1
  /__w/llvm/llvm/toolchain/bin//clang++ -Werror  -fsycl -fsycl-targets=nvptx64-nvidia-cuda  /__w/llvm/llvm/llvm/sycl/test-e2e/Graph/NativeCommand/cuda_record_async_malloc.cpp -o /__w/llvm/llvm/build-e2e/Graph/NativeCommand/Output/cuda_record_async_malloc.cpp.tmp.out  -L/usr/local/cuda/lib64 -lcuda  -I/usr/local/cuda/include
  # executed command: /__w/llvm/llvm/toolchain/bin//clang++ -Werror -fsycl -fsycl-targets=nvptx64-nvidia-cuda /__w/llvm/llvm/llvm/sycl/test-e2e/Graph/NativeCommand/cuda_record_async_malloc.cpp -o /__w/llvm/llvm/build-e2e/Graph/NativeCommand/Output/cuda_record_async_malloc.cpp.tmp.out -L/usr/local/cuda/lib64 -lcuda -I/usr/local/cuda/include
  # note: command had no output on stdout or stderr
  # RUN: at line 2
  env SYCL_UR_CUDA_ENABLE_IMAGE_SUPPORT=1 ONEAPI_DEVICE_SELECTOR=cuda:gpu  /__w/llvm/llvm/build-e2e/Graph/NativeCommand/Output/cuda_record_async_malloc.cpp.tmp.out
  # executed command: env SYCL_UR_CUDA_ENABLE_IMAGE_SUPPORT=1 ONEAPI_DEVICE_SELECTOR=cuda:gpu /__w/llvm/llvm/build-e2e/Graph/NativeCommand/Output/cuda_record_async_malloc.cpp.tmp.out
  # .---command stderr------------
  # | <CUDA>[ERROR]: 
  # | UR CUDA ERROR:
  # | 	Value:           801
  # | 	Name:            CUDA_ERROR_NOT_SUPPORTED
  # | 	Description:     operation not supported
  # | 	Function:        urCommandBufferAppendNativeCommandExp
  # | 	Source Location: /__w/llvm/llvm/src/unified-runtime/source/adapters/cuda/command_buffer.cpp:1541
  # | 
  # | terminate called after throwing an instance of 'sycl::_V1::exception'
  # |   what():  cuda backend failed with error: 44 (UR_RESULT_ERROR_UNSUPPORTED_FEATURE)
  # `-----------------------------
  # error: command failed with exit status: -6

To reproduce

No response

Environment

• OS: [e.g Windows/Linux]
• Target device and vendor: [e.g. Intel GPU]
• DPC++ version: [e.g. commit hash or output of clang++ --version]
• Dependencies version: [e.g. the output of sycl-ls --verbose]

Additional context

No response

[dm-vodopyanov]

@sarnex the tag "Confirmed" is not applicable for CUDA\HIP issues, removed it.

[sarnex]

Ah ok thanks, I didn't know we don't want JIRA issues for these. Won't use it in the future.

[bratpiorka]

@sarnex hi I'm currently working on supporting CUDA 13 (both driver and toolkit). BTW I will create a JIRA in URT for this issue