[Wien] IFORT 11.1 and symmetry (2x missing)

Sun Jun 13 16:19:01 CEST 2010

Dear Laurence
thanks for youre suggestions,
the problem is resolved if using the "new" switch -fltconsistency
instead of the combination -mp1 -prec-div etc.

such that the FOPT looks like
-FR -w -fltconsistency -xHost -O3 -m64 -ip
instead of
-FR -mp1 -w -prec_div -pc80 -pad -DINTEL_VML -traceback -xHost -m64 -ip

(note -pc80 does not help as it is the default anyway and I also did not check whether -pad is always a good idea).

The problem of matrot.f is indeed only in INVA, i used it separately and
changed rather every loop and floating point operation without resolving the problem with -O3,
usually I don't find that insistent problems.

The somewhat cleaned version  of matrot is attached,
The matrix inversion was not used anyway, but only the determinant was checked by INVA 

During the next week I will check
if some of the -fp options are a better choice for the overall performance
otherwise I would suggest to use 
-FR -w -fltconsistency -DINTEL_VML -traceback
or similar as default for siteconfig

Ciao
Gerhard

====================================
Dr. Gerhard H. Fecher
Institut of Inorganic and Analytical Chemistry
Johannes Gutenberg - University
55099 Mainz
________________________________________
Von: wien-bounces at zeus.theochem.tuwien.ac.at [wien-bounces at zeus.theochem.tuwien.ac.at]&quot; im Auftrag von &quot;Laurence Marks [L-marks at northwestern.edu]
Gesendet: Sonntag, 13. Juni 2010 15:03
An: A Mailing list for WIEN2k users
Betreff: Re: [Wien] IFORT 11.1 and symmetry (2x missing)

Dear Gerhard,

Also, it could matter whether you place these options before or after -O3.

Can you send a test case perhaps to my private email? As I said
earlier I suspect that ifort 11.X may be doing somethings a bit odd,
perhaps switching to fast, less accurate algorithms (for speed).

On Sun, Jun 13, 2010 at 7:51 AM, Laurence Marks
<L-marks at northwestern.edu> wrote:
> Dear Gerhard,
>
> Looks very odd. Does adding -r8 -pc80 -fpconstant have any effect?
> What about -fltconsistency, -mieee-fp -fp-model strict (or
> alternatives)?
>
> On Sun, Jun 13, 2010 at 3:58 AM, Gerhard Fecher <fecher at uni-mainz.de> wrote:
>> Dear Laurence
>> it appears in matrot.f
>> if this is compiled with -O2 instead of -O3 the error vanishes.
>>
>> From a fast inspection it may be caused
>> by the CALL INVA(CB,CB2,IERR)
>> (subroutine is defined in matrot.f)
>>
>> I will separate it and do some more checks,
>> The subroutines have a little bit old-fashioned style,
>> probably they were never changed for a long time and maybe there is something confusing the compiler.
>> (I also don't like much IMPLICIT statements that are not immediately followed by NONE)
>>
>> I forgot yesterday to give a solution,
>> for those that look for a fast solution:
>> - Don't use -O3 in the FOPT
>> - or use -O2 (same as using no -Ox switch) for symmetry (change FOPT and recompile symmetry only)
>>
>>
>> Ciao
>> Gerhard
>>
>> ====================================
>> Dr. Gerhard H. Fecher
>> Institut of Inorganic and Analytical Chemistry
>> Johannes Gutenberg - University
>> 55099 Mainz
>> ________________________________________
>> Von: wien-bounces at zeus.theochem.tuwien.ac.at [wien-bounces at zeus.theochem.tuwien.ac.at]&quot; im Auftrag von &quot;Laurence Marks [L-marks at northwestern.edu]
>> Gesendet: Samstag, 12. Juni 2010 16:15
>> An: A Mailing list for WIEN2k users
>> Betreff: Re: [Wien] IFORT 11.1 and symmetry (2x missing)
>>
>> Can you try by hand compiling individual routines to see where it is?
>> I have a suspicion that some versions of ifort 11.1 are overoptimizing
>> but don't have a way to pin it down, this might.
>>
>> On Sat, Jun 12, 2010 at 6:14 AM, Gerhard Fecher <fecher at uni-mainz.de> wrote:
>>> IFORT 11.1 (.072) may lead to an overoptimization if using -O3 when compiling symmetry.
>>>
>>> For spacegroup 119
>>> I received a "2x missing" warning and the output shows that something goes wrong with the symmetry
>>> (and it produces an empty structure file)
>>> It appears in Wien2k versions 9.2 and 10.1 (I did not check for others or different small numers of ifort))
>>>
>>> This does not appear for several cubic structures (with or without center of inversion),
>>> therfore I wonder if there is something specific for the lower symmetries that might be the cause.
>>>
>>> The error does not appear for IFORT 11.0(.083) with -O3 (or older Versions)
>>>
>>> All other parts of Wien run smooth without problems if using -O3
>>> Note, usually the -mp1 -w -prec_div -pc80 -pad options usually prevent numerical nonsense if using -O3
>>>
>>> Has someone an idea whether there are some other new switches (combination of switches) that prevent
>>> other overoptimizations in the numerical parts ?
>>>
>>>
>>> Ciao
>>> Gerhard
>>>
>>> ====================================
>>> Dr. Gerhard H. Fecher
>>> Institut of Inorganic and Analytical Chemistry
>>> Johannes Gutenberg - University
>>> 55099 Mainz
>>> ________________________________________
>>>
>>> _______________________________________________
>>> Wien mailing list
>>> Wien at zeus.theochem.tuwien.ac.at
>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>>
>>
>>
>>
>> --
>> Laurence Marks
>> Department of Materials Science and Engineering
>> MSE Rm 2036 Cook Hall
>> 2220 N Campus Drive
>> Northwestern University
>> Evanston, IL 60208, USA
>> Tel: (847) 491-3996 Fax: (847) 491-7820
>> email: L-marks at northwestern dot edu
>> Web: www.numis.northwestern.edu
>> Chair, Commission on Electron Crystallography of IUCR
>> www.numis.northwestern.edu/
>> Electron crystallography is the branch of science that uses electron
>> scattering and imaging to study the structure of matter.
>> _______________________________________________
>> Wien mailing list
>> Wien at zeus.theochem.tuwien.ac.at
>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>> _______________________________________________
>> Wien mailing list
>> Wien at zeus.theochem.tuwien.ac.at
>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>
>
>
>
> --
> Laurence Marks
> Department of Materials Science and Engineering
> MSE Rm 2036 Cook Hall
> 2220 N Campus Drive
> Northwestern University
> Evanston, IL 60208, USA
> Tel: (847) 491-3996 Fax: (847) 491-7820
> email: L-marks at northwestern dot edu
> Web: www.numis.northwestern.edu
> Chair, Commission on Electron Crystallography of IUCR
> www.numis.northwestern.edu/
> Electron crystallography is the branch of science that uses electron
> scattering and imaging to study the structure of matter.
>

--
Laurence Marks
Department of Materials Science and Engineering
MSE Rm 2036 Cook Hall
2220 N Campus Drive
Northwestern University
Evanston, IL 60208, USA
Tel: (847) 491-3996 Fax: (847) 491-7820
email: L-marks at northwestern dot edu
Web: www.numis.northwestern.edu
Chair, Commission on Electron Crystallography of IUCR
www.numis.northwestern.edu/
Electron crystallography is the branch of science that uses electron
scattering and imaging to study the structure of matter.
_______________________________________________
Wien mailing list
Wien at zeus.theochem.tuwien.ac.at
http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
-------------- next part --------------
A non-text attachment was scrubbed...
Name: matrot_new.f
Type: text/x-fortran
Size: 6301 bytes
Desc: matrot_new.f
URL: <http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20100613/595a2d4a/attachment.f>