-
Notifications
You must be signed in to change notification settings - Fork 795
Closed
Labels
Description
Describe the bug
While compiling from source, an error occurred while compiling llvm/sycl/plugins/cuda/pi_cuda.cpp
.
llvm/sycl/plugins/cuda/pi_cuda.cpp:1957:35: error: use of undeclared identifier 'cuDeviceGetUuid_v2'; did you mean 'cuDeviceGetUuid'? sycl::detail::pi::assertion(cuDeviceGetUuid_v2(&uuid, device->get()) == ^~~~~~~~~~~~~~~~~~ cuDeviceGetUuid
To Reproduce
- Run configure.py:
CUDA_LIB_PATH=/usr/local/cuda-10.2/lib64/stubs/ CC=clang CXX=clang++ python $DPCPP_HOME/llvm/buildbot/configure.py --cuda --cmake-opt="-DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda-10.2" --cmake-gen "Ninja" -o ./build-ibm --host-target PowerPC --use-lld
- Try to compile:
CUDA_LIB_PATH=/usr/local/cuda-10.2/lib64/stubs/ CC=clang CXX=clang++ python $DPCPP_HOME/llvm/buildbot/compile.py -o ./build-ibm -j 40
- While compiling the error above occurred. By reading the relevant code snippet I guess that the reason for this error is that since I do not have a version of CUDA above 11.4 then
cuDeviceGetUuid_v2
is not available at compile time.
Environment (please complete the following information):
- OS: RHEL 7.5
- Target device and vendor: IBM Power CPU and Nvidia Volta V100.
- DPC++ version: Commit hash from git log: 10e07a3
- Dependencies version: cuda 10.2
Additional context
N/A