Re: [AMBER] Not matching test AMBER14 with CUDA

From: Jason Swails <jason.swails.gmail.com>
Date: Tue, 1 Dec 2015 07:55:29 -0500

On Mon, Nov 30, 2015 at 11:20 PM, Mijiddorj Batsaikhan <
b.mijiddorj.gmail.com> wrote:

> Here I attached my last log file.
>
> my cofiguration line and last several lines are here:
>
> *./configure -cuda_DPFP gnu*
>
> ######
> If you have not already done so, you may need to add
> /usr/local/cuda-7.5/lib
> and/or /usr/local/cuda-7.5/lib64 to LD_LIBRARY_PATH using the command
> export LD_LIBRARY_PATH="/usr/local/cuda-7.5/lib:${LD_LIBRARY_PATH}"
> (bash, sh)
> setenv LD_LIBRARY_PATH "/usr/local/cuda-7.5/lib:${LD_LIBRARY_PATH}"
> (tcsh, csh)
>
> --------------------------------------------------------------------------------
>
> The next step is to type 'make install'
>
> Cleaning the src directories. This may take a few moments.
> Configure complete.
> ######
>
> then
>
>
> *export LD_LIBRARY_PATH="/usr/local/cuda-7.5/lib:${LD_LIBRARY_PATH}"*
>
>
> *make install*
>
> last rows of installation
>
> #####
>
> ./cuda/cuda.a -L/usr/local/cuda-7.5/lib64 -L/usr/local/cuda-7.5/lib
> -lcurand -lcufft -lcudart -lstdc++ -L/home/mijiddorj/installs/amber14/lib
> /home/mijiddorj/installs/amber14/lib/libnetcdff.a
> /home/mijiddorj/installs/amber14/lib/libnetcdf.a
> /home/mijiddorj/installs/amber14/lib/libemil.a -lstdc++
> make[3]: Leaving directory `/home/mijiddorj/installs/amber14/src/pmemd/src'
> Installation of pmemd.cuda complete
> make[2]: Leaving directory `/home/mijiddorj/installs/amber14/src/pmemd'
> make[1]: Leaving directory `/home/mijiddorj/installs/amber14/src'
>
> #####
>
>
> Finally,
>
> *make test*
>
> results same as following lines
>
> ....
>
> ./Run.cellulose_nvt: Program error
> make[3]: [test.pmemd.cuda.pme.big] Error 1 (ignored)
> cd cellulose/ && ./Run.cellulose_npt SPFP
> /home/mijiddorj/installs/amber14/include/netcdf.mod
> ../../../bin/pmemd.cuda_SPFP: error while loading shared libraries:
> libcurand.so.7.5: cannot open shared object file: No such file or directory
> ./Run.cellulose_npt: Program error
> make[3]: [test.pmemd.cuda.pme.big] Error 1 (ignored)
> make[3]: Leaving directory `/home/mijiddorj/installs/amber14/test/cuda'
> make[2]: Leaving directory `/home/mijiddorj/installs/amber14/test'
> 0 file comparisons passed
> 0 file comparisons failed
> 327 tests experienced errors
> Test log file saved as
>
> /home/mijiddorj/installs/amber14/logs/test_amber_cuda/2015-11-30_23-15-10.log
> No test diffs to save!
> make[1]: Leaving directory `/home/mijiddorj/installs/amber14/test'
>

​This indicates that it is still failing to find libcurand.so.7.5 at
runtime. This should have been fixed by adding
/usr/local/cuda-7.5/lib
​ to LD_LIBRARY_PATH, but it wasn't. What does the following command print
to the screen:

env | grep LD_LIBRARY_PATH

? Do you see a file named libcurand.so.7.5 in
/usr/local/cuda-7.5/lib
​? What about
/usr/local/cuda-7.5/lib
​64? What shell do you use? Is it bash? Or some c-shell derivative?

This error means one of two things:

1. You did not set the environment variable correctly. This could happen
if you use bash syntax for csh (i.e., export instead of setenv) or
vice-versa. It could also happen if you set the environment variable in a
different shell than you use to run "make test". Adding that command to
your resource file (e.g., ~/.bashrc) will usually ensure that it is set in
every ​


​shell that you open (just remember that modifying .bashrc does not affect
any shells that are already open at the time).

2. The requested library, libcurand.so.7.5, is not actually there.

​As many times as I've seen this error on Linux -- and it has been a lot --
the problem has *always* been one of these.

HTH,
Jason

-- 
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Dec 01 2015 - 05:00:02 PST
Custom Search