Problem installing GPU version

[milton...@gmail.com]

Hello! I'm trying to install the GPU version of Vampire with "make gcc-cuda", but I get the following error:

In file included from ./hdr/cusplibrary-0.5.1/cusp/system/cuda/detail/multiply/coo_flat_spmv.h:21,
                 from ./hdr/cusplibrary-0.5.1/cusp/system/cuda/detail/multiply.h:21,
                 from ./hdr/cusplibrary-0.5.1/cusp/system/cuda/execution_policy.h:50,
                 from ./hdr/cusplibrary-0.5.1/cusp/execution_policy.h:34,
                 from ./hdr/cusplibrary-0.5.1/cusp/convert.h:25,
                 from ./hdr/cusplibrary-0.5.1/cusp/detail/array2d.inl:322,
                 from ./hdr/cusplibrary-0.5.1/cusp/array2d.h:515,
                 from ./hdr/cusplibrary-0.5.1/cusp/detail/format_utils.inl:18,
                 from ./hdr/cusplibrary-0.5.1/cusp/format_utils.h:169,
                 from ./hdr/cusplibrary-0.5.1/cusp/detail/csr_matrix.inl:17,
                 from ./hdr/cusplibrary-0.5.1/cusp/csr_matrix.h:525,
                 from src/cuda/typedefs.hpp:25,
                 from src/cuda/data.hpp:12,
                 from src/cuda/config.cu:18:
./hdr/cusplibrary-0.5.1/cusp/system/cuda/arch.h:24:10: fatal error: thrust/system/cuda/detail/detail/launch_calculator.h: No such file or directory
 #include <thrust/system/cuda/detail/detail/launch_calculator.h>
          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
make: *** [src/cuda/makefile:47: obj/cuda/config.o] Error 1

Has anyone else encountered this problem? I tried installing cuda 9.0 following this guide. It did not help, however. Should also say that I'm using Debian, stable version.

Any help would be appreciated!

[Richard Evans]

Hi Milton,

Apologies - we are working to fix the CUDA implementation over the next few months to work with CUDA 10, as well as add MPI+CUDA support. Matt Ellis has very kindly fixed the CUSP error in the CUDA branch. In the meantime the cuda branch should work with some minor tweaks:

1) comment out #include <cub-1.5.2/cub/cub.cuh> in cuda_utils.hpp (this is now included internally I think)
2) remove old and unsupported hardware targets in the makefile (mainly 3xx series).

All the best,

Richard 

[milton...@gmail.com]

Hello Richard, and thanks!

I didn't really understand what you meant with the second point, but it works now after installing cuda 9.0 and gcc 6 and g++ 6. Strangely though, it only worked on the Vampire "version" a colleague had installed on a Ubuntu machine (I use Debian and downloaded Vampire with the git command in the manual). In that "version" the file arch.h is different. Could it be that he installed the developer branch? Where can I see what the exact version of the Vampire installation is?

The first file below is the one that gives the above error (downloaded with the git command in the manual) and the second arch.h file is the one that works.

Hopefully this information is helpful somehow!
/Milton

[Richard Evans]

So if you clone a repo, then 

git branch

will give a list of branches, and a star next to the active one. 

git log

will give a specific commit (at the top) which is on that branch (depending on when the code was downloaded). Assuming no other changes to the code have been made, it is always possible to revert to any commit in the entire repo (across all branches) with

git checkout #gitcommithash

where #gitcommithash is usually some random thing for each commit in the list from git status, like 0b505a0d71b66319ffb5d33b67537b0a495ce8d3. However, this will delete any modified directories in the folder.

All the best,

Richard