Try to compile an OpenMP cp2k version with CUDA support results in invalid option error

[Ole]

Hi Sebastian,

since you don't have "__CUDA_PROFILING" among your DFLAGS, the file cuda_nvtx_cu.cu should be completely empty after preprocessing. Therefore, I would suspect that the error originates from your cuda setup. Notice e.g. that nvcc calls your c++ host-compiler in the background.

-Ole



Am Sonntag, 29. Juni 2014 12:05:31 UTC+2 schrieb Sebastian Müller:
>
> Hello,
>
> I tried to compile a cp2k version with CUDA 6.0 and get the following 
> output on stderr running the make command:
>
>
> lexer.l:143: warning, Regel kann nicht passen
>> /home/Sebastian/programme/cp2k/cp2k-2.5.1_cuda/makefiles/../src/tmc_tree_types.F:12:34: 
>> warning: backslash and newline separated by space [enabled by default]
>> /home/Sebastian/programme/cp2k/cp2k-2.5.1_cuda/makefiles/../src/tmc_tree_types.F:14:36: 
>> warning: backslash and newline separated by space [enabled by default]
>> ar: creating 
>> /home/Sebastian/programme/cp2k/cp2k-2.5.1_cuda/makefiles/../lib/Linux-x86-64-gfortran-lukas/ssmp/libcp2k_lib.a
>> ar: creating 
>> /home/Sebastian/programme/cp2k/cp2k-2.5.1_cuda/makefiles/../lib/Linux-x86-64-gfortran-lukas/ssmp/libcp2k_dbcsr_lib.a
>> ar: creating 
>> /home/Sebastian/programme/cp2k/cp2k-2.5.1_cuda/makefiles/../lib/Linux-x86-64-gfortran-lukas/ssmp/libcp2k_fft_lib.a
>> ar: creating 
>> /home/Sebastian/programme/cp2k/cp2k-2.5.1_cuda/makefiles/../lib/Linux-x86-64-gfortran-lukas/ssmp/libcp2k_base_lib.a
>> ar: creating 
>> /home/Sebastian/programme/cp2k/cp2k-2.5.1_cuda/makefiles/../lib/Linux-x86-64-gfortran-lukas/ssmp/libcp2k_ma_lib.a
>> Command-line error: invalid option: --allow_managed
>>
>> 1 catastrophic error detected in this compilation.
>> Compilation terminated.
>> make[1]: *** [cuda_nvtx_cu.o] Fehler 4
>> make: *** [build] Fehler 2
>>
>>
> It seems like this --allow_managed option results in the error, but the 
> last command on stdout was
>
> nvcc -c -O2 -arch=sm_30 -I/usr/local/cuda/lib64/../include 
>> -L/usr/local/cuda/lib64 --dont-use-profile -ldir 
>> /usr/local/cuda/nvvm/libdevice 
>> /home/Sebastian/programme/cp2k/cp2k-2.5.1_cuda/makefiles/../src/cuda/
>> cuda_nvtx_cu.cu
>>
>
> and there is no option like this. Without the CUDA support I was able to 
> build the program without errors and it is working fine.
> I attached the full output at stderr and stdout of the make command and 
> the corresponding arch file.
>
> Could anyone tell me what I am doning wrong.
>
> Best regards
> Sebastian
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140630/2074b404/attachment.htm>