PMEMD has not been adapted to the AMD math libraries. It may be a trivial
fake-out, if all the interfaces are the same as MKL (just change some
library names). If not, one would need to go through, case by case,
modifying the code to support the different call interfaces. In the case of
pmemd (and at least to some extent sander - I don't remember the exact
file), the calls are encapsulated in the file veclib.fpp. IF there were
extensive demand for this, it could be done probably with a little effort in
pmemd. BUT also note, most pmemd code (gb being the exception) does not
really profit that much from use of the MKL. I think sander also uses a
bunch of other math library stuff that pmemd doesn't - linear algebra stuff
I believe, and there may be some nontrivial port issues there. Also note
that if amd is likely to go belly-up in the next couple of years, it may not
make that much sense. I know there are wonderful things about the opteron
architecture, but it has sounded to me like intel is beating amd to death in
the marketplace with its multicore chips (which I well and truly hate, but
the market is responding to different things than what we respond to).
Regards - Bob
----- Original Message -----
From: "Gustavo Seabra" <gustavo.seabra.gmail.com>
To: <amber-developers.scripps.edu>
Sent: Monday, October 06, 2008 1:41 PM
Subject: amber-developers: AMD Core Math Libraries?
> Hi Everyone,
>
> Has anyone here had any experience with the AMD Core Math Libraries?
> Can we use that instead of MKL to compile sander/pmemd?
>
> Thanks a lot,
> --
> Gustavo Seabra
> Postdoctoral Associate
> Quantum Theory Project - University of Florida
> Gainesville - Florida - USA
>
Received on Wed Oct 08 2008 - 05:08:09 PDT
