Error compiling DPC++

[tamara-laranga]

Describe the bug

I am trying to compile DPC++ wittch cuda capabilities using the docker provided in this repository but I am encountering this error at the end of the building process. Does anyone know how to solve it?

 /llvm/libdevice/imf/../imf_impl_utils.hpp:12:10: fatal error: 'cstddef' file not found
   12 | #include <cstddef>
      |          ^~~~~~~~~
1 error generated.
make[2]: *** [tools/libdevice/CMakeFiles/imf_fallback_bf16_bc.dir/build.make:77: lib/libsycl-fallback-imf-bf16.bc] Error 1
make[1]: *** [CMakeFiles/Makefile2:110653: tools/libdevice/CMakeFiles/imf_fallback_bf16_bc.dir/all] Error 2
make: *** [Makefile:156: all] Error 2

To reproduce

The commands I am using to do so are the following ones:

$ docker run --name sycl_build -it -v /local/workspace/dir/:/src ghcr.io/intel/llvm/ubuntu2204_build /bin/bash
$ git clone https://github.com/intel/llvm.git -b sycl
$ cd llvm
$ python ./buildbot/configure.py --cuda -t release --cmake-gen "Unix Makefiles"
$ cd build
$ make install

Environment

• OS: WSL 2
• Target device and vendor: Expected to compile SYCL code to run on NVIDIA 3060 TI GPU

Additional context

No response

[tamara-laranga]

I managed to solve the previous error but now this happens

 [ 95%] Built target ur_adapter_level_zero Consolidate compiler generated dependencies of target ur_adapter_cuda 
make[2]: *** No rule to make target 'CUDA_CUDA_LIBRARY-NOTFOUND', needed by 'lib/libur_adapter_cuda.so.0.10.0'. Stop. 
make[1]: *** [CMakeFiles/Makefile2:104757: _deps/unified-runtime build/source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/all] 
Error 2 make: *** [Makefile:156: all] 
Error 2

[JackAKirk]

Depending on the host setup, there are various issues on recent nvidia docker images related to interactions with the host, whereby nvidia devices are not visible. There are also particular issues with wsl2. The following may help you.

• NOTICE: Containers losing access to GPUs with error: "Failed to initialize NVML: Unknown Error" NVIDIA/nvidia-container-toolkit#48
• "Failed to initialize NVML: Unknown Error" after random amount of time NVIDIA/nvidia-docker#1671
• nvidia-container-cli: initialization error: load library failed: libnvidia-ml.so.1 NVIDIA/nvidia-container-toolkit#154

[tamara-laranga]

I followed some of the advice but the same issue is persistent. The trouble comes when trying to compile DPC++ with --cuda capabilities in the docker provided in this repository. Was someone able to install this compiler so SYCL code can be executed in a NVIDIA GPU?

The steps I am using can be seen in more detail in the following.
https://github.com/intel/llvm/blob/sycl/sycl/doc/GetStartedGuide.md#build-dpc-toolchain-with-support-for-nvidia-cuda

[bader]

Depending on the host setup, there are various issues on recent nvidia docker images related to interactions with the host, whereby nvidia devices are not visible. There are also particular issues with wsl2. The following may help you.

• NOTICE: Containers losing access to GPUs with error: "Failed to initialize NVML: Unknown Error" NVIDIA/nvidia-container-toolkit#48
• "Failed to initialize NVML: Unknown Error" after random amount of time NVIDIA/nvidia-docker#1671
• nvidia-container-cli: initialization error: load library failed: libnvidia-ml.so.1 NVIDIA/nvidia-container-toolkit#154

@JackAKirk, to be honest, I expected docker container environment to be enough for building SYCL compiler even w/o access to GPU. It looks like I was wrong and CMake requires access to NVIDIA GPU to configure CUDA properly.

CMake successfully configured paths to most of the libraries and includes on my machine (w/o NVIDIA GPU), but I hit the same problem with building PI plug-in.

[0/2] Re-checking globbed directories...
ninja: error: 'CUDA_CUDA_LIBRARY-NOTFOUND', needed by 'lib/libpi_cuda.so', missing and no known rule to make it

Relevant parts of CMakeCache.txt.

//Compile device code in 64 bit mode
CUDA_64_BIT_DEVICE_CODE:BOOL=ON

//Attach the build rule to the CUDA source file.  Enable only when
// the CUDA source file is added to at most one target.
CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE:BOOL=ON

//Generate and parse .cubin files in Device mode.
CUDA_BUILD_CUBIN:BOOL=OFF

//Build in Emulation mode
CUDA_BUILD_EMULATION:BOOL=OFF

//"cudart" library
CUDA_CUDART_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcudart.so

//"cuda" library (older versions only).
CUDA_CUDA_LIBRARY:FILEPATH=CUDA_CUDA_LIBRARY-NOTFOUND

//Path to a file.
CUDA_CUPTI_INCLUDE_DIR:PATH=/usr/local/cuda/include

//Directory to put all the output files.  If blank it will default
// to the CMAKE_CURRENT_BINARY_DIR
CUDA_GENERATED_OUTPUT_DIR:PATH=

//Generated file extension
CUDA_HOST_COMPILATION_CPP:BOOL=ON

//Host side compiler used by NVCC
CUDA_HOST_COMPILER:FILEPATH=/usr/bin/cc

//Path to a program.
CUDA_NVCC_EXECUTABLE:FILEPATH=/usr/local/cuda/bin/nvcc

//Semi-colon delimit multiple arguments. during all build types.
CUDA_NVCC_FLAGS:STRING=

//Semi-colon delimit multiple arguments. during DEBUG builds.
CUDA_NVCC_FLAGS_DEBUG:STRING=

//Semi-colon delimit multiple arguments. during MINSIZEREL builds.
CUDA_NVCC_FLAGS_MINSIZEREL:STRING=

//Semi-colon delimit multiple arguments. during RELEASE builds.
CUDA_NVCC_FLAGS_RELEASE:STRING=

//Semi-colon delimit multiple arguments. during RELWITHDEBINFO
// builds.
CUDA_NVCC_FLAGS_RELWITHDEBINFO:STRING=

//"OpenCL" library
CUDA_OpenCL_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libOpenCL.so

//Propagate C/CXX_FLAGS and friends to the host compiler via -Xcompile
CUDA_PROPAGATE_HOST_FLAGS:BOOL=ON

//Path to a file.
CUDA_SDK_ROOT_DIR:PATH=CUDA_SDK_ROOT_DIR-NOTFOUND

//Compile CUDA objects with separable compilation enabled.  Requires
// CUDA 5.0+
CUDA_SEPARABLE_COMPILATION:BOOL=OFF

//Path to a file.
CUDA_TOOLKIT_INCLUDE:PATH=/usr/local/cuda/include

//Toolkit location.
CUDA_TOOLKIT_ROOT_DIR:PATH=/usr/local/cuda

//Use the static version of the CUDA runtime library if available
CUDA_USE_STATIC_CUDA_RUNTIME:BOOL=ON

//Print out the commands run while compiling the CUDA source file.
//  With the Makefile generator this defaults to VERBOSE variable
// specified on the command line, but can be forced on with this
// option.
CUDA_VERBOSE_BUILD:BOOL=OFF

//Version of CUDA as computed from nvcc.
CUDA_VERSION:STRING=12.1

//"cublas" library
CUDA_cublas_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcublas.so

//"cudadevrt" library
CUDA_cudadevrt_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcudadevrt.a

//static CUDA runtime library
CUDA_cudart_static_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcudart_static.a

//"cufft" library
CUDA_cufft_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcufft.so

//"cupti" library
CUDA_cupti_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcupti.so

//"curand" library
CUDA_curand_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcurand.so

//"cusolver" library
CUDA_cusolver_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcusolver.so

//"cusparse" library
CUDA_cusparse_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libcusparse.so

//"nppc" library
CUDA_nppc_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppc.so

//"nppial" library
CUDA_nppial_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppial.so

//"nppicc" library
CUDA_nppicc_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppicc.so

//"nppidei" library
CUDA_nppidei_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppidei.so

//"nppif" library
CUDA_nppif_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppif.so

//"nppig" library
CUDA_nppig_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppig.so

//"nppim" library
CUDA_nppim_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppim.so

//"nppist" library
CUDA_nppist_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppist.so

//"nppisu" library
CUDA_nppisu_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppisu.so

//"nppitc" library
CUDA_nppitc_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnppitc.so

//"npps" library
CUDA_npps_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnpps.so

//"nvToolsExt" library
CUDA_nvToolsExt_LIBRARY:FILEPATH=/usr/local/cuda/lib64/libnvToolsExt.so

//Path to a library.
CUDA_rt_LIBRARY:FILEPATH=/usr/lib/x86_64-linux-gnu/librt.a

and

//ADVANCED property for variable: CUDA_64_BIT_DEVICE_CODE
CUDA_64_BIT_DEVICE_CODE-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE
CUDA_ATTACH_VS_BUILD_RULE_TO_CUDA_FILE-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_BUILD_CUBIN
CUDA_BUILD_CUBIN-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_BUILD_EMULATION
CUDA_BUILD_EMULATION-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_CUDART_LIBRARY
CUDA_CUDART_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_CUDA_LIBRARY
CUDA_CUDA_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_GENERATED_OUTPUT_DIR
CUDA_GENERATED_OUTPUT_DIR-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_HOST_COMPILATION_CPP
CUDA_HOST_COMPILATION_CPP-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_NVCC_EXECUTABLE
CUDA_NVCC_EXECUTABLE-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_NVCC_FLAGS
CUDA_NVCC_FLAGS-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_NVCC_FLAGS_DEBUG
CUDA_NVCC_FLAGS_DEBUG-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_NVCC_FLAGS_MINSIZEREL
CUDA_NVCC_FLAGS_MINSIZEREL-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_NVCC_FLAGS_RELEASE
CUDA_NVCC_FLAGS_RELEASE-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_NVCC_FLAGS_RELWITHDEBINFO
CUDA_NVCC_FLAGS_RELWITHDEBINFO-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_OpenCL_LIBRARY
CUDA_OpenCL_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_PROPAGATE_HOST_FLAGS
CUDA_PROPAGATE_HOST_FLAGS-ADVANCED:INTERNAL=1
//This is the value of the last time CUDA_SDK_ROOT_DIR was set
// successfully.
CUDA_SDK_ROOT_DIR_INTERNAL:INTERNAL=CUDA_SDK_ROOT_DIR-NOTFOUND
//ADVANCED property for variable: CUDA_SEPARABLE_COMPILATION
CUDA_SEPARABLE_COMPILATION-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_TOOLKIT_INCLUDE
CUDA_TOOLKIT_INCLUDE-ADVANCED:INTERNAL=1
//This is the value of the last time CUDA_TOOLKIT_ROOT_DIR was
// set successfully.
CUDA_TOOLKIT_ROOT_DIR_INTERNAL:INTERNAL=/usr/local/cuda
//This is the value of the last time CUDA_TOOLKIT_TARGET_DIR was
// set successfully.
CUDA_TOOLKIT_TARGET_DIR_INTERNAL:INTERNAL=/usr/local/cuda
//ADVANCED property for variable: CUDA_VERBOSE_BUILD
CUDA_VERBOSE_BUILD-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_VERSION
CUDA_VERSION-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_cublas_LIBRARY
CUDA_cublas_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_cudadevrt_LIBRARY
CUDA_cudadevrt_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_cudart_static_LIBRARY
CUDA_cudart_static_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_cufft_LIBRARY
CUDA_cufft_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_cupti_LIBRARY
CUDA_cupti_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_curand_LIBRARY
CUDA_curand_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_cusolver_LIBRARY
CUDA_cusolver_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_cusparse_LIBRARY
CUDA_cusparse_LIBRARY-ADVANCED:INTERNAL=1
//Location of make2cmake.cmake
CUDA_make2cmake:INTERNAL=/usr/share/cmake-3.22/Modules/FindCUDA/make2cmake.cmake
//ADVANCED property for variable: CUDA_nppc_LIBRARY
CUDA_nppc_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppial_LIBRARY
CUDA_nppial_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppicc_LIBRARY
CUDA_nppicc_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppidei_LIBRARY
CUDA_nppidei_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppif_LIBRARY
CUDA_nppif_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppig_LIBRARY
CUDA_nppig_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppim_LIBRARY
CUDA_nppim_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppist_LIBRARY
CUDA_nppist_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppisu_LIBRARY
CUDA_nppisu_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nppitc_LIBRARY
CUDA_nppitc_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_npps_LIBRARY
CUDA_npps_LIBRARY-ADVANCED:INTERNAL=1
//ADVANCED property for variable: CUDA_nvToolsExt_LIBRARY
CUDA_nvToolsExt_LIBRARY-ADVANCED:INTERNAL=1
//Location of parse_cubin.cmake
CUDA_parse_cubin:INTERNAL=/usr/share/cmake-3.22/Modules/FindCUDA/parse_cubin.cmake
//Location of run_nvcc.cmake
CUDA_run_nvcc:INTERNAL=/usr/share/cmake-3.22/Modules/FindCUDA/run_nvcc.cmake

I see that we pass --gpus all flag to docker, but only to run SYCL applications on NV GPU. The build works in our CI w/o this flag.

[tamara-laranga]

So you wat we need to pass --gpus all flag to docker in this command?

docker run --name sycl_build --gpus all -it -v /local/workspace/dir/:/src ghcr.io/intel/llvm/ubuntu2204_build /bin/bash

Because if I try that I can't even create the docker container, this error is reported:

docker: Error response from daemon: failed to create task for container: failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: error during container init: error running hook #0: error running hook: exit status 1, stdout: , stderr: Auto-detected mode as 'legacy'
nvidia-container-cli: initialization error: load library failed: libnvidia-ml.so.1: cannot open shared object file: no such file or directory: unknown.

[JackAKirk]

Depending on the host setup, there are various issues on recent nvidia docker images related to interactions with the host, whereby nvidia devices are not visible. There are also particular issues with wsl2. The following may help you.

• NOTICE: Containers losing access to GPUs with error: "Failed to initialize NVML: Unknown Error" NVIDIA/nvidia-container-toolkit#48
• "Failed to initialize NVML: Unknown Error" after random amount of time NVIDIA/nvidia-docker#1671
• nvidia-container-cli: initialization error: load library failed: libnvidia-ml.so.1 NVIDIA/nvidia-container-toolkit#154

@JackAKirk, to be honest, I expected docker container environment to be enough for building SYCL compiler even w/o access to GPU. It looks like I was wrong and CMake requires access to NVIDIA GPU to configure CUDA properly.

CMake successfully configured paths to most of the libraries and includes on my machine (w/o NVIDIA GPU), but I hit the same problem with building PI plug-in.

Yeah I remember getting past this but then still nvidia-smi did not detect the device, although this is when I was trying to get 12.5 docker image to work.

The situation with Nvidia docker is continuously changing from what I can tell. They seem to recommend using https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/index.html now; and since they have switched to this there seems to be quite a lot of bug reports. I don't fully understand how the CI is working for 11.7 and 12.1 versions.
But I was happy that at least 12.1 works, since I think this is all we care about at the moment. And I thought it would be sensible to deal with the situation at the point when we will need to upgrade the CI for new features (Probably cuda 13, which I guess must be arriving sometime in the second half of this year).
My worry was that if I get it to work with 12.5, then there will be different issues to deal with for cuda 13.

[tamara-laranga]

I was relying on the docker provided in this repository because I wasn't able to install DPC++ with cuda backend. I tryied it on Windows, WSL 2 and in an Ubuntu Laptop.

[JackAKirk]

I was relying on the docker provided in this repository because I wasn't able to install DPC++ with cuda backend. I tryied it on Windows, WSL 2 and in an Ubuntu Laptop.

In that case I would avoid using docker and concentrate on ensuring you are building correctly. Please consult the getStartedGuide for building dpc++ from source with cuda support: https://github.com/intel/llvm/blob/sycl/sycl/doc/GetStartedGuide.md#build-dpc-toolchain-with-support-for-nvidia-cuda

Make sure you are following the windows specific instructions in this section:
https://github.com/intel/llvm/blob/sycl/sycl/doc/GetStartedGuide.md#create-dpc-workspace

Some details given on the codeplay plugin documentation regarding compiling with the cuda backend are also relevant to the open source build and may also be helpful to you:
https://developer.codeplay.com/products/oneapi/nvidia/2024.2.0/guides/get-started-guide-nvidia#use-dpc-to-target-nvidia-gpus

However these instructions are aimed at linux support, since the codeplay plugins don't support windows yet, but the opensource does, so please focus your attention on the open source documentation linked above.

[tamara-laranga]

In the WSL I am getting this error, does someone know how to solve it?

llvm/build/_deps/unified-runtime-src/source/adapters/cuda/image.cpp: In function ‘ur_result_t urTextureCreate(ur_sampler_handle_t, const ur_image_desc_t*, const CUDA_RESOURCE_DESC&, ur_exp_image_handle_t_**)’:
/mnt/c/Users/tamar/llvm/build/_deps/unified-runtime-src/source/adapters/cuda/image.cpp:314:29: error: ‘CU_TRSF_SEAMLESS_CUBEMAP’ was not declared in this scope
  314 |       ImageTexDesc.flags |= CU_TRSF_SEAMLESS_CUBEMAP;
      |                             ^~~~~~~~~~~~~~~~~~~~~~~~
make[2]: *** [_deps/unified-runtime-build/source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/build.make:202: _deps/unified-runtime-build/source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/image.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:104714: _deps/unified-runtime-build/source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/all] Error 2
make: *** [Makefile:156: all] Error 2

and the cuda version I am using is this one:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Thu_Nov_18_09:45:30_PST_2021
Cuda compilation tools, release 11.5, V11.5.119
Build cuda_11.5.r11.5/compiler.30672275_0

[JackAKirk]

In the WSL I am getting this error, does someone know how to solve it?

llvm/build/_deps/unified-runtime-src/source/adapters/cuda/image.cpp: In function ‘ur_result_t urTextureCreate(ur_sampler_handle_t, const ur_image_desc_t*, const CUDA_RESOURCE_DESC&, ur_exp_image_handle_t_**)’:
/mnt/c/Users/tamar/llvm/build/_deps/unified-runtime-src/source/adapters/cuda/image.cpp:314:29: error: ‘CU_TRSF_SEAMLESS_CUBEMAP’ was not declared in this scope
  314 |       ImageTexDesc.flags |= CU_TRSF_SEAMLESS_CUBEMAP;
      |                             ^~~~~~~~~~~~~~~~~~~~~~~~
make[2]: *** [_deps/unified-runtime-build/source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/build.make:202: _deps/unified-runtime-build/source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/image.cpp.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:104714: _deps/unified-runtime-build/source/adapters/cuda/CMakeFiles/ur_adapter_cuda.dir/all] Error 2
make: *** [Makefile:156: all] Error 2

and the cuda version I am using is this one:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Thu_Nov_18_09:45:30_PST_2021
Cuda compilation tools, release 11.5, V11.5.119
Build cuda_11.5.r11.5/compiler.30672275_0

@intel/bindless-images-reviewers

Do you think you have made changes recently that break support for cuda 11.5?

[JackAKirk]

I see the issue. I'll fix this. I should have caught it in my review of the cubemaps PR.

[tamara-laranga]

Could you please let me know when is fixed?

[JackAKirk]

Could you please let me know when is fixed?

This is fixed by oneapi-src/unified-runtime#1822

It will take a little while for this to be pulled down to intel/llvm.
If you want a build to work right now with cuda 11.5 then please change the following file in your local copy of intel/llvm with the following patch:

+++ b/sycl/plugins/unified_runtime/CMakeLists.txt
@@ -99,13 +99,13 @@ if(SYCL_PI_UR_USE_FETCH_CONTENT)
       CACHE PATH "Path to external '${name}' adapter source dir" FORCE)
   endfunction()
 
-  set(UNIFIED_RUNTIME_REPO "https://github.com/oneapi-src/unified-runtime.git")
+  set(UNIFIED_RUNTIME_REPO "https://github.com/JackAKirk/unified-runtime.git")
   # commit aaf08109f2a05adb61f50478824ae2739526daee
   # Author: Ben Tracy <ben.tracy@codeplay.com>
   # Date:   Mon Jun 17 14:49:53 2024 +0100
   #     [CMDBUF] Fix coverity issue in command buffers
   #     - Fix incorrect conditions for copy engine usage that were reported on coverity.
-  set(UNIFIED_RUNTIME_TAG aaf08109f2a05adb61f50478824ae2739526daee)
+  set(UNIFIED_RUNTIME_TAG 056c1e20b640089585fd1ea9da47830ae2e157e9)
 
   fetch_adapter_source(level_zero
     ${UNIFIED_RUNTIME_REPO}