The ABINIT Users Mailing List ( CLOSED )

["Javier Lopez" <sklnw@hotmail.com>]

I have done both Test_v3#4 and #84 with IFC7.1 abinis compiled binaries. I 
have used two different binaries:

1) The files marked as "isw" used a SSE2 optimized abinis, generated with 
the following flags:
FFLAGS=-FR -O3 -xW -tpp7 -ip
FFLAGS_Src_2psp=-FR -O0 -xW -tpp7 -ip
FFLAGS_Src_3xc=-FR -O0 -xW -tpp7 -ip
FFLAGS_Src_5common=-FR -O0 -xW -tpp7 -ip
FFLAGS_LIBS=-O3 -xW -tpp7 -ip
These tests were run in a PIV

2) The files marked as "isk" used SSE1 optimizations, with:
FFLAGS=-FR -O3 -xK -tpp7 -ip
FFLAGS_Src_2psp=-FR -O2 -xK -tpp7 -ip
FFLAGS_LIBS=-O3 -xK -tpp7 -ip
I used an AthlonMP to run these tests. -O2 in 2psp was required to complete 
the internal test v4 (or 2, sorry, I can't remember) though no error ocurred 
at compilation time. -tpp7 is faster for AthlonXP/MP.

I have included the results in the attached zips for anyone who wants to 
analize them (to complicate things further, I have used "rs" instead of 
"out", and "rl" instead of "log"). There are a few very noticeable 
differences within the reference out and "isw" files, though I'm don't know 
if they are critical or not. I hope someone can tell!

Javier


> From: Xavier Gonze <gonze@pcpm.ucl.ac.be>
> Reply-To: forum@abinit.org
> To: forum@abinit.org
> Subject: Re: [abinit-forum] Intel compiler v7.0
> Date: Mon, 28 Apr 2003 12:15:24 +0200
>
> Javier Lopez wrote:
> > Thanks for the advice. Just two questions:
> > i)Is the problem related to the use of the -xW flag? Ie, is the problem 
> > still present with only sse-1 optmizations, -xK?
>
> With IFCv7.0, I have tested different options (though not the -xK),
> but only -O0 gives automatically no problem. -O1 is equivalent
> to -O3, according to the manual, and its use is problematic for
> several directories (including those found by Michel Cote, but
> not restricted to them, for cases beyond ground-state, or
> including spin-orbit).
>
> This is my main
> > concern, since I've used IFC7.1 with -O3 -xK to compile abinit for some 
> > AthlonMP dual machines. (If you haven't tested it, maybe I can if you send 
> > me the .in and other necessary files)
>
> I haven't tested IFCv7.1, that might be free of any problem ...
>
> All the tests are present in the ABINIT package (Test_fast, Test_v1, 
> Test_v2,
> Test_v3, Test_v4). Problems were seen
> with Test_v3#04 (and some other response functions tests),
> as well as Test_v3#84 (and some other spin-orbit test cases).
> More important : the features causing the problems were correct FORTRAN 
> construct,
> present in numerous places in the code. So, even if one can obtain
> both sufficient speed and the correct execution of ABINIT for all
> built-in test case of the
> ABINIT package with non-optimized compiler options for selected 
> directories,
> such constructs are present in practically all directories, and thus
> potentially harmfull !
>
> > ii)Which version of the IFC is used to compile the binaries distributed at 
> > abinit.org?
>
> Most were IFCv5.0 using, some IFCv6.0, but in both case
> corresponding ABINIT binaries were extensively tested and worked properly.
>
> Good continuation,
> Xavier


_________________________________________________________________
Multiplica por cinco el tamaño de tu buzón de correo y envía adjuntos de 
hasta 2 Mb con MSN Almacenamiento Extra. http://join.msn.com/?pgmarket=es-es

Attachment: isk.zip
Description: Binary data

Attachment: isw.zip
Description: Binary data