Pathscale used to be pretty easy to use, and was a really solid compiler -
roughly comparable in performance with intel, but without the grief. The
problem, I would guess, is that intel has this "free for academic use" model
that gets them a broad foothold; pathscale may have also not done as good a
job of supporting clusters in terms of licensing issues (just based on a
user complaint that I fielded a while back). I don't know how good or bad
current experience with pgi has been - they had problems in the past, but
have hopefully improved. I think with gfortran coming along as well as it
has, and gcc widely available, making money on unix compilers is becoming a
more and more marginal enterprise. It is a shame for pathscale though, in
my opinion; when they first came out they were the cleanest thing available
for generic unix targets (ie., not the vendor's compiler). But if they have
for all intents and purposes failed, they are of course of no interest
moving forward. It is just a shame to have so few alternatives to intel,
which in my opinion at least, has become so painful to base development on
that I am almost tempted to start playing with gfortran.
Regards - Bob
----- Original Message -----
From: "Jason Swails" <jason.swails.gmail.com>
To: "AMBER Developers Mailing List" <amber-developers.ambermd.org>
Sent: Tuesday, March 09, 2010 2:45 PM
Subject: Re: [AMBER-Developers] PMEMD missing type?
On Tue, Mar 9, 2010 at 2:02 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
> Pathscale perhaps still have a pulse but it is a very weak one. I know no
> machines that have pathscale compilers available and I can't see how they
kraken has them available (module load pathscale/3.2). PGI is
default, though, and after looking at how much of a pain it would be
to support pathscale for amber, I quickly decided to remain using PGI.
:).
> can survive long term. If I recall right NVIDIA bought Pathscale, or a
> chunk
> of them and it became the basis for their NVCC cuda compiler. Although I
> may
> be confused on that so don't quote me.
>
> Gfortran is closing the performance gap fast though and if you hook it
> with
> MKL the difference for GB is minimal while PME stands around 10% or so
> last
> time I checked. Thus the value add in the pathscale compilers is really
> minimal considering the cost.
>
> Just my 3c.
>
> All the best
> Ross
>
>> -----Original Message-----
>> From: amber-developers-bounces.ambermd.org [mailto:amber-developers-
>> bounces.ambermd.org] On Behalf Of Robert Duke
>> Sent: Tuesday, March 09, 2010 10:55 AM
>> To: AMBER Developers Mailing List
>> Subject: Re: [AMBER-Developers] PMEMD missing type?
>>
>> So, just out of curiosity, does anyone know what is happening with
>> PathScale? I thought they had maybe died, but I can still find a
>> webpage.
>> Two or three years ago, they were pretty close on performance, with a
>> significant difference (positive) in usability. I had always figured
>> intel
>> would have an edge because they make the chips and have pretty deep
>> pockets,
>> but their usability factor has been just awful for some time now. When
>> I
>> first advocated using them (I think it was at least partially my fault,
>> though these days there are not a lot of other options anyway), things
>> were
>> not nearly this bad, except perhaps for Itanium...
>> Best Regards - Bob
>> ----- Original Message -----
>> From: "Ross Walker" <ross.rosswalker.co.uk>
>> To: "'AMBER Developers Mailing List'" <amber-developers.ambermd.org>
>> Sent: Tuesday, March 09, 2010 1:47 PM
>> Subject: RE: [AMBER-Developers] PMEMD missing type?
>>
>>
>> > 10.1.018 also works.
>> >
>> >> -----Original Message-----
>> >> From: amber-developers-bounces.ambermd.org [mailto:amber-developers-
>> >> bounces.ambermd.org] On Behalf Of case
>> >> Sent: Tuesday, March 09, 2010 10:03 AM
>> >> To: AMBER Developers Mailing List
>> >> Subject: Re: [AMBER-Developers] PMEMD missing type?
>> >>
>> >> On Tue, Mar 09, 2010, case wrote:
>> >> > >
>> >> > > > cpp -traditional -P -DBINTRAJ -DDIRFRC_EFS -DDIRFRC_COMTRANS
>> >> > > > -DDIRFRC_NOVEC -DFFTLOADBAL_2PROC -DPUBFFT pme_setup.fpp
>> >> pme_setup.f90
>> >> > > > ifort -fast -c pme_setup.f90
>> >> > > > pme_setup.f90(417): error #6404: This name does not have a
>> type,
>> >> and
>> >> > > > must
>> >> > > > have an explicit type. [NATOM]
>> >> > > > do n = 1, natom
>> >> > > > ------------^
>> >> > > >
>> >> >
>> >> > What Intel compiler are you using?
>> >>
>> >> One quick addtional point, then I have meetings all day:
>> >>
>> >> For me, it appears that ifort 10.1.025 works, but 11.1.069 fails.
>> >>
>> >> ...dac
>> >>
>> >>
>> >> _______________________________________________
>> >> 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
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
--
---------------------------------------
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
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
Received on Tue Mar 09 2010 - 12:30:04 PST
