RE: [AMBER-Developers] configure script for pgi and gcc+g95/gcc+ifort issues

From: Volodymyr Babin <vbabin.ncsu.edu>
Date: Sun, 15 Nov 2009 11:17:06 -0500 (EST)

We also typically use a similar combination
(gcc+ifort, gcc+pgi, gcc+xlf90); sometimes it is not justified;
OTOH, we don't have xlc/gfortran/g95 on pwr5 machines so that
gcc+xlf90 is the only option there. OTOH2: I am perfectly
okay with editing the *.h by hands (or keeping several .h around)
so that IMHO the configure script is unneeded.

Have a nice day,

Volodymyr

On Sun, November 15, 2009 10:46, matthew_seetin.urmc.rochester.edu wrote:
> Our lab prefers gcc+ifort. In our testing, we actually got better
> performance from gcc/g++ than icc for a number of our lab's other software
> applications, so we saw no need to get icc. However, AMBER runs faster
> with
> ifort on our hardware, so that made sense for us to use over gfortran. If
> this pairing is all that unusual, I suppose I can keep hacking the script
> to
> suit us, but there's at least someone out there who appreciates the
> gcc+ifort combo.
>
> --Matt Seetin
>
> -----Original Message-----
> From: amber-developers-bounces.ambermd.org
> [mailto:amber-developers-bounces.ambermd.org] On Behalf Of Mengjuei Hsieh
> Sent: Friday, November 13, 2009 6:24 PM
> To: AMBER Developers Mailing List
> Subject: [AMBER-Developers] configure script for pgi and gcc+g95/gcc+ifort
> issues
>
> Hi,
>
> I do agree that extra compiler combinations are overkill, so I again
> removed the options for gcc+g95 and gcc+ifort. Besides, there is not
> so many counter arguments, either. At the same time, I checked-in a
> change that fixes a problem of netcdf that coughs up errors on pgi
> compiler. Just adding -DpgiFortran to the CPPFLAGS will do the trick.
>
> Sincerely,
> --
> Mengjuei Hsieh
>
> On Tue, Nov 10, 2009 at 03:13:25PM -0500, case wrote:
>> Mengjuei wrote:
>> > I just checked-in the modification for configure script:
>> > 1. Added support for combination of gcc + ifort
>> > 2. Added support for combination of gcc + g95
>> Aargh! Do we really need items 1 or 2? For item 1, are there really
> enough
>> people who have ifort but not icc? Or some other reason for wanting
>> this?
>> As for 2, does g95 have an advantage over gfortran?
>> The reason for restricting the number of platforms (and compiler
> combinations)
>> we support is to reduce the chance of corner cases of incompatibility.
> Plus,
>> it makes life a little easier for develops if "gnu" or "intel" are the
> main
>> platforms -- I can commit a change without having to go find a g95
> compiler to
>> make sure it works there as well.
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>


_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
http://www.rosswalker.co.uk/adsense_top10/
Received on Sun Nov 15 2009 - 08:30:03 PST
Custom Search