Bulding CP2K with CUDA (Ubuntu 16.04)

Henrique Junior henri... at gmail.com
Tue Sep 5 01:25:57 UTC 2017



Dear friends, I’m “refactoring” the .deb packages available in Ubuntu 16.04 
to enable CUDA in our cluster. Everything looks fine during the build 
process but, so far, none of our tests seem to be taking advantage of our 
GPU (that was tested with gpu-burn 0.7).

As you may know, the default CP2K in Ubuntu 16.04 is the 2.6.2. I have 
added the CUDA flags[1] in my ARCH files using a patch (attached):


+NVCC = /usr/local/cuda-8.0/bin/nvcc

+DFLAGS += -D__ACC -D__DBCSR_ACC -D__PW_CUDA -D__CUDA_PROFILING

+LIBS += -lcudart -lcublas -lcufft -lrt

 

I have the latest CUDA Toolkit and the related CUDA packages installed in 
my system.

Can someone point me in the right direction here?

 

Thanks in advance

 

[1] - https://www.cp2k.org/howto:compile_with_cuda

 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170904/a850bdfb/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: makeflags.patch
Type: text/x-diff
Size: 3327 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170904/a850bdfb/attachment.patch>


More information about the CP2K-user mailing list