> Can you say what you mean by "still works well"? Something like this:
>
> ./configure cray (any other options needed? can I just add -mpi
> or -openmp flags if desired?)
>
> make install (or did you have to explicitly mention a
> target like "cray_serial"?)
'./configure cray' does not work out of the box, but configure will
guide the user as to the necessary flags. When all is said and done,
the final configure line looks like:
./configure -nofftw3 --skip-python --with-netcdf $NETCDF_DIR -noX11
[{-openmp | -mpi --with-pnetcdf $PNETCDF_DIR}] cray
Note that the '--skip-python' flag isn't strictly required for 'cray'
but I would probably recommend it in any help text. An external FFTW3
is currently required for pmemd. The original reason was that there
were errors in the built-in FFT (perhaps since fixed?) when compiled
with cray compilers on BW (GNU too at one point). Configure checks
that there is an external fftw library available. An external NetCDF
library must also be specified since the bundled one has lots of
trouble building with cray compilers. I've found that the X11 stuff is
incompatible with cray compilers (at least on BW), so configure
requires -noX11.
> If/when things work like the above, it seems to me that we should
> add "cray" to the allowed list of compilers in the --help message.
I can do this along with some hints.
> Also, as pointed out earlier, an explanation of how "cray" differs from
> "crayxt5": I would expect most users will not understand the existing error
> message without more hand-holding:
>
> Error: '-crayxt5' flag not compatible with 'cray' compiler target.
I can improve the error message.
-Dan
On Tue, Jan 17, 2017 at 12:07 PM, Daniel Roe <daniel.r.roe.gmail.com> wrote:
> (Something strange happened with my gmail, so I'm sending this again.
> Apologies if you end up receiving it twice.)
>
> The 'cray' compiler target still works well on Blue Waters for serial
> and openmp. The mpi target initially only built for AmberTools but not
> for pmemd. Only took some minor fixing to get it to work though -
> seems like some C-style comments made it into the Fortran source code
> (.Devs, please don't do that!).
>
> There are a bunch of flags that end up being required for the 'cray'
> target, but the checks in configure point them out. The only remaining
> issue is that the compiler version check function now has a
> hard-coded call to the compiler with a '--version' flag, which is not
> supported by cray compilers (probably others as well) and which
> generates a spurious error message. The comment in configure2 seems
> to indicate it may only be needed for OSX so it may be better to only
> perform this call when on an OSX platform.
>
> I can add documentation to configure about cray compilers if no one
> has objections.
>
> -Dan
>
> On Tue, Jan 17, 2017 at 9:04 AM, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>> On Tue, Jan 17, 2017 at 8:54 AM, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>>> On Tue, Jan 17, 2017 at 12:09 AM, Thomas Cheatham <tec3.utah.edu> wrote:
>>>>
>>>> p.s. Dan if you can look at this on BW next week, I'll look at the CPPTRAJ
>>>> || paper :-)
>>>
>>> Deal - I'll check it out today.
>>>
>>> -Dan
>>
>> Unfortunately Blue Waters appears to be moving about as fast as
>> molasses today, so it might not get done until tomorrow.
>>
>> -Dan
>
>
>
> --
> -------------------------
> Daniel R. Roe
> Laboratory of Computational Biology
> National Institutes of Health, NHLBI
> 5635 Fishers Ln, Rm T900
> Rockville MD, 20852
> https://www.lobos.nih.gov/lcb
--
-------------------------
Daniel R. Roe
Laboratory of Computational Biology
National Institutes of Health, NHLBI
5635 Fishers Ln, Rm T900
Rockville MD, 20852
https://www.lobos.nih.gov/lcb
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Jan 17 2017 - 10:30:03 PST
