Hi Bob,
> So maybe we should just wrap the stringizing stuff in an #ifdef CUDA
> (or
> whatever it is)? Sorry I am not more up-to-date on this stuff; I just
> have
> not been set up to play with this stuff yet. But if we did that, we
> would
> probably effectively be putting ourselves in a position where we can
> ignore
> old systems, since they are pretty unlikely to be used with cuda
> anyway...
It would definitely be inside the cuda directory and would only get built /
used if someone was building pmemd.cuda. I'm not sure how people got a
different idea here. Everything else works it is just the calling of c and
cuda routines from calls within the Fortran code that are within #ifdef CUDA
as it stands.
Right now ALL of the cuda stuff in no way impacts any of the standard pmemd
builds and it is only in building the cuda version of pmemd using Intel that
I run into problems.
The nvcc compiler is only specced to run with gnu and Intel as it is so
there is not much point trying to get pgi etc working with this. Plus there
is very little compute intensive stuff done on the cpu during a GPU run so
the payoff on tweaking the compiler for the Fortran side of the code in this
situation is minimal. I would just stick with GNU except that Intel
compilation is needed for Windows support and also because if you try to
compile AMBER with two different compilers, e.g. serial and MPI with Intel
and then switch to GNU for cuda you run into all sorts of problems with
NETCDF which could just be avoided if once builds all three targets with the
same compiler.
All the best
Ross
/\
\/
|\oss Walker
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org/ |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon Mar 29 2010 - 11:30:05 PDT
