[gmx-developers] Gromacs-2016: Win64 quasi-portable

Mark Abraham mark.j.abraham at gmail.com
Mon Oct 3 08:59:40 CEST 2016


Hi,

On Mon, Oct 3, 2016 at 12:01 AM Justin Lemkul <jalemkul at vt.edu> wrote:

>
>
> On 10/2/16 5:42 PM, Boris Timofeev wrote:
> >
> > Good afternoon to all!
> >
> > I did a great job on compilation of the "portable" Gromacs-2016 version
> on the
> > Win64  platform without cygwin with OpenCL support and  AVX_256/AVX2_256
> > expansions. The project is builded on VS-2015. By small changes in
> CMakeLists, I
> > managed to achieve that both the "embedded" tests, and regressiontests,
> are
> > executed after assembly immediately from VS IDE. It was tested on
> Windows7/
> > Windows10  with IntelCore i3, i5, i7  processors and video cards from
> Nvidia and
> > AMD. So far it was not succeeded to win  - against the video card from
> Intel -
> > Intel OpenCL compiler preprocessor is left unfinished, does not recognize
> > directive "-I" and have buggy concatention (##) implementation.
> > It was necessary to realize a primitive preprocessor,  became
> successful, but
> > all the same calculations (tests and regressiontests) are wrong as it was
> > already metioned here (https://bugs.freedesktop.org/show_bug.cgi?
> > id=94265#add_comment).
>

Intel isn't supported - see
http://manual.gromacs.org/documentation/2016/user-guide/mdrun-performance.html#running-the-opencl-version-of-mdrun



> > If it is interesting to community, I am ready to report about some
> necessary
> > changes in progect and to provide the main CMake-file.
>

Suggestions are indeed welcome. :-) If you want to apply a fix to our git
branch called release-2016, you can upload that to our code review site
https://gerrit.gromacs.org/#/q/status:open. Instructions at
http://www.gromacs.org/Developer_Zone/Git/Gerrit. Or if it's a simple
thing, just describe it here.


> > There are several questions to developers.
> >
> > 1. The nbnxn_ocl_kernel_nvidia.clh, nbnxn_ocl_kernel_nowarp.clh and
> > nbnxn_ocl_kernel_amd.clh files, if to compare their with kdiff, differ
> only in
> > comments and lack of unroll pragma for Nvidia. Why not to unite them in
> one?
> > Where the promised optimization?
>

Nobody has had the resources to look into that more closely, but with
GROMACS being an open source project, such contributions are welcome!
NVIDIA isn't a priority because there the native CUDA port does better (and
NVIDIA's OpenCL seems to have a bug that means the non-blocking offloading
calls that GROMACS needs run blocking style).

> 2. Why in the Gromacs'a code so many paths to files are embedded up? To
> start
> > tests without development environment on other computer it is necessary
> to copy
> > practically all project.
>

I'm not sure what you're referring to. The JIT compilation that is required
for OpenCL requires that some source code is available. The regression
tests run a lot of simulations, and those require to be able to read the
databases of how to describe simulation systems. Nor have we targeted being
able to run the tests with neither a development environment nor a full
installation available, since we've never encountered a need for that.

Mark


> > 3. A question "on science". When performing an example of
> >
> http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin/gmx-tutorials/lysozyme/01_pdb2gmx.html
> ,
> >  pdb2gmx for the aminoacid residue  HIS ("HIZE-branch" on stdout ),
> unlike all
> > others residue,  gives a nonintegral charge 0.883,  inexplicable with
> round-off
> > errors.
> > Whether there is no program mistake here?
> >
>
> I can't speak to points 1 and 2, but here: what is HIZE-branch?  I've never
> heard of that.  Note, too, that if you're working with a single amino acid
> with
> OPLS-AA, you can't rely on the default terminus selection.  You need to
> choose
> the zwitterion termini, otherwise the charges will be junk because OPLS-AA
> makes
> changes to CA depending on whether the residue is a zwitterion or in a
> polypeptide.
>
> -Justin
>
> --
> ==================================================
>
> Justin A. Lemkul, Ph.D.
> Ruth L. Kirschstein NRSA Postdoctoral Fellow
>
> Department of Pharmaceutical Sciences
> School of Pharmacy
> Health Sciences Facility II, Room 629
> University of Maryland, Baltimore
> 20 Penn St.
> Baltimore, MD 21201
>
> jalemkul at outerbanks.umaryland.edu | (410) 706-7441
> http://mackerell.umaryland.edu/~jalemkul
>
> ==================================================
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
> or send a mail to gmx-developers-request at gromacs.org.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20161003/e849b9eb/attachment.html>


More information about the gromacs.org_gmx-developers mailing list