[Pw_forum] Errors in parallel 3D FFT

[O. Baris Malcioglu]

Hi Iain,

Please note that this kind of errors may also depend on the
compilation environment. I have recently encountered a similar error
of large drifts using IBM compilers in SP6 where my zeroes were not
zero enough. Apparently IBM compilers tend to use float precision in
places you do not intend for the sake of speed if you let the compiler
do the optimization. This can be turned off via some combination of
flags. If this is the case, please contact your administrator
regarding how to deal with the problem in your specific setup.

Best,
Baris



2010/1/20 Iain Bethune <ibethune at staffmail.ed.ac.uk>:
> Thanks for the suggestions - I have looked a bit closer at the code, and it
> appears that all the grid elements (x,y,z) where these large errors appear
> are those where dffts%iplw(x) == 1 .AND. dffts%iplw(y) == 1
>
> However, there are some grid elements which meet the above criteria that do
> not have errors, so maybe I am not understanding something about how the FFT
> works?
>
> I can't think of anything else in the fft_dlay_descriptor data structure
> that helps narrow down further where these errors occur - any ideas?
>
> Cheers
>
> - Iain
>
> --
>
> Iain Bethune
> Applications Consultant, EPCC
>
> Email: ibethune at epcc.ed.ac.uk
> Tel/Fax: +44 (0)131 650 5201/6555
> Mob: +44 (0)7598317015
> Addr: 2404 JCMB, King's Buildings, Mayfield Road, Edinburgh, EH9 3JZ
>
> The University of Edinburgh is a charitable body, registered in Scotland,
> with registration number SC005336.
>
>
>> -----Original Message-----
>> From: Paolo Giannozzi [mailto:giannozz at democritos.it]
>> Sent: 15 January 2010 17:41
>> To: ibethune at epcc.ed.ac.uk; PWSCF Forum
>> Subject: Re: [Pw_forum] Errors in parallel 3D FFT
>>
>>
>> On Jan 14, 2010, at 11:41 , Iain Bethune wrote:
>>
>> > So most of the elements have small errors (which is as expected),
>> > but there
>> > are a large number which have very large errors (>1e-06), and I don't
>> > understand why this should happen.
>>
>> nor do I, but it seems to me rather unlikely that the FFT used in QE
>> is wrong.
>> Try first with a simpler system in serial execution and then try more
>> complex
>> cases. Note that not all elements of FFTs arrays are "physical": for
>> reasons
>> of increased speed, it is sometimes convenient to increase the first
>> dimension of arrays by 1
>>
>> P.
>> ---
>> Paolo Giannozzi, Dept of Physics, University of Udine
>> via delle Scienze 208, 33100 Udine, Italy
>> Phone +39-0432-558216, fax +39-0432-558222
>>
>>
>>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>