[CP2K-user] [CP2K:17967] Re: cudaErrorStubLibrary runtime error in CP2K

[Amit Gupta]

I resolved the above error, by symlinking to libcudart as libcuds.so.1. So 
I guess this issue is closed. 
I ran in secondary issue of  `cudaErrorInsufficientDriver` so I guess there 
is something wrong with my loaded CUDA libraries, So I will probably 
recompile it again with more carefully loaded latest driver. 
Thank you

On Monday, October 31, 2022 at 10:14:26 AM UTC-5 Amit Gupta wrote:

> Could it be a compile time issue on my end? I remember toolchain requiring 
> libcuda for linking (cosma I believe), and I found issues on different 
> repositories on github, mentioning that libcuda resides in 
> lib/stubs/libcuda.so. So I linked against it, provide its location at 
> runtime in LD_LIBRARY_PATH.  
>
> On Monday, October 31, 2022 at 3:55:13 AM UTC-5 Ole Schütt wrote:
>
>> Hi Amit,
>>
>> if you are getting cudaErrorStubLibrary then the app failed to load the 
>> real driver. Presumably your local admin can help with this. See also the 
>> CUDA docs 
>> <https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html#group__CUDART__TYPES_1g3f51e3575c2178246db0a94a430e0038>
>> :
>>
>>    cudaErrorStubLibrary = 34 This indicates that the CUDA driver that 
>> the application has loaded is a stub library. Applications that run with 
>> the stub rather than a real driver loaded will result in CUDA API returning 
>> this error. 
>>
>> -Ole
>>
>> On Sunday, October 30, 2022 at 9:59:11 PM UTC+1 Amit Gupta wrote:
>>
>>> Hi,
>>> I compiled a CUDA compatible version of CP2K, but when I submit my jobs, 
>>> I get the following error:
>>>
>>> --------------------------------------------------------------------------
>>> ERROR: cudaErrorStubLibrary 
>>> /scratch/ag9288/softwares/cp2k-2022.2/src/offload/offload_library.c 49
>>> forrtl: error (76): Abort trap signal
>>> Image              PC                Routine            Line       
>>>  Source             
>>> cp2k.psmp          0000000006B83CAB  Unknown               Unknown 
>>>  Unknown
>>> libpthread-2.28.s  000014D73245EB20  Unknown               Unknown 
>>>  Unknown
>>> libc-2.28.so       000014D712D4037F  gsignal               Unknown 
>>>  Unknown
>>> libc-2.28.so       000014D712D2ADB5  abort                 Unknown 
>>>  Unknown
>>> cp2k.psmp          00000000031F8C29  offload_get_devic          49 
>>>  offload_library.c
>>> cp2k.psmp          0000000001387F46  f77_interface_mp_         276 
>>>  f77_interface.F
>>> cp2k.psmp          000000000042281D  MAIN__                    284 
>>>  cp2k.F
>>> cp2k.psmp          0000000006BED096  Unknown               Unknown 
>>>  Unknown
>>> libc-2.28.so       000014D712D2C493  __libc_start_main     Unknown 
>>>  Unknown
>>> cp2k.psmp          000000000042195E  Unknown               Unknown 
>>>  Unknown
>>>
>>> --------------------------------------------------------------------------
>>>
>>> It looks like the executable is unable to find the card, while all the 
>>> libraries are correctly loaded.
>>>
>>> Version: 
>>> Intel fortran: 19.1
>>> cuda : 11.3
>>> openmpi 4.1.1
>>> CP2K: 2022.2
>>>
>>> Steps I took to build the binary:
>>> 1. Install the toolchain:
>>> ./install_cp2k_toolchain.sh --with-intel=system --enable-cuda 
>>> --gpu-ver=V100  --with-openmpi=system  --with-libxc=install 
>>> --with-libint=install --with-fftw=install --with-mkl=system 
>>> --with-libxsmm=install  --with-elpa=install   --with-ptscotch=no   
>>> --with-superlu=no    --with-pexsi=no      --with-quip=no       
>>> --with-plumed=no    --with-sirius=no     --with-gsl=no       
>>>  --with-libvdwxc=no   --with-spglib=no    --with-hdf5=no    --with-spfft=no 
>>>      --with-spla=install   --with-cosma=install  --with-libvori=no 
>>> --with-openblas=no -j4
>>>
>>> 2. Manualy build and copy Libint, for the fortran makefile bug:
>>> https://github.com/cp2k/libint-cp2k/releases
>>> cmake -DCMAKE_C_COMPILER=icc -DCMAKE_C_FLAGS="-qopenmp -O2" 
>>> -DCMAKE_CXX_COMPILER=icpc -DCMAKE_CXX_FLAGS="-qopenmp -O2" 
>>> -DCMAKE_Fortran_COMPILER=ifort -DCMAKE_Fortran_FLAGS="-qopenmp -O2" 
>>> -DREQUIRE_CXX_API=OFF -DENABLE_FORTRAN=ON ..
>>>
>>> 3. Apply required CUDA patch to CP2K and compile:
>>> https://github.com/cp2k/cp2k/commit/ee6c3aa.patch
>>>  make -j 4 ARCH=local_cuda VERSION="ssmp sdbg psmp pdbg"
>>>
>>> PS: The job-node is different from the compile node, could that might be 
>>> an issue? Also while the compilation went fine, there was an error in 
>>> symlinking the binaries.
>>>
>>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/658a9011-fda3-4b70-a30d-9d3dd4ef2307n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20221031/317f417b/attachment-0001.htm>