[Pw_forum] Possible bug(?) in q2r.x when using FFT routine in ESSL
Jiseok Kim
tube77 at gmail.com
Mon Sep 3 20:17:25 CEST 2012
Thank you very much.
The temporary workaround that I used is that re-compile the entire source
code with FFTW3 instead of ESSL.
Then I linked the 'q2r.x' only to the FFTW3 executable and the rest of the
executables are remained with ESSL.
So the problem is fixed.
For some reason, there is memory allocation problem with FFTW3 on AIX so
that the 'pw.x' crashes immediately when it calls the FFTW3 routines..
I tried many degrees of freedom when I compiled the FFTW3 (different
optimization levels, 32 or 64 bit and etc. ) and modified my user limit.
But none of them worked..
But interestingly, the FFTW3 does not complained when it's called from
'q2r.x' ..
Anyway, I agree that the best way to fix the problem is to modify the
'cfft3d' routine so that one can call 1d, 2d or 3d ESSL FFT routines
depending on the dimensionality of FFT grids..
Thank you.
On 9/1/12 2:37 AM, "pw_forum-request at pwscf.org"
<pw_forum-request at pwscf.org> wrote:
>Send Pw_forum mailing list submissions to
> pw_forum at pwscf.org
>
>To subscribe or unsubscribe via the World Wide Web, visit
> http://www.democritos.it/mailman/listinfo/pw_forum
>or, via email, send a message with subject or body 'help' to
> pw_forum-request at pwscf.org
>
>You can reach the person managing the list at
> pw_forum-owner at pwscf.org
>
>When replying, please edit your Subject line so it is more specific
>than "Re: Contents of Pw_forum digest..."
>
>
>Today's Topics:
>
> 1. Possible bug(?) in q2r.x when using FFT routine in ESSL
> (Jiseok Kim)
> 2. Re: Possible bug(?) in q2r.x when using FFT routine in ESSL
> (Paolo Giannozzi)
> 3. Re: Join to group , confirmation # (Yue-Wen Fang)
>
>
>----------------------------------------------------------------------
>
>Message: 1
>Date: Fri, 31 Aug 2012 03:19:59 -0500
>From: Jiseok Kim <tube77 at gmail.com>
>Subject: [Pw_forum] Possible bug(?) in q2r.x when using FFT routine in
> ESSL
>To: <pw_forum at pwscf.org>
>Message-ID: <CC65DDDF.2ED0%tube77 at gmail.com>
>Content-Type: text/plain; charset="us-ascii"
>
>Dear, QE users
>
>I have been struggling for a while to find out what's going on with q2r.x
>The problem was that I could not get rid of the error messages below when
>doing q2r.x.
>
>-------------------------------------------------
>DCFT3 : 2538-2030
>The transform length (ARG NO. 9) is not an allowed value. The next higher
>allowed value is (2).
>-------------------------------------------------
>
>I've compiled QE-v5.0.1. on AIX6.1(parallel machine) with 'mpxlf' with
>ESSL
>routines.
>So the 'q2r.x' calls 'cfft3d' and then the 'cfft3d' calls 'dcft3' in ESSL
>when performing FFT.
>
>The problem is this.
>The 'dcft3' in ESSL only accepts certain values of transforming length as
>shown in,
>
>-------------------------------------------------
>http://publib.boulder.ibm.com/infocenter/clresctr/vxrx/index.jsp?topic=%2F
>co
>m.ibm.cluster.essl.v5r1.essl100.doc%2Fam501_hscft3.htm
>-------------------------------------------------
>
>so the FFT dimensions (nx, ny, nz), which is read from outputs by ph.x,
>should be the values in the table.
>Otherwise, the 'dcft3' crashes.
>
>The FFT dimensions (nx, ny, nz) in q2r.x are equivalent to the (nq1, nq2,
>nq3) in ph.x so that the Monkhorst-pack grid for ph.x should be limited by
>the values in the table when using the ESSL.
>
>For example, the Monkhorst-pack grid (16x16x1) for q-points in the case of
>two-dimensional supercell, such as graphene, cannot be accepted by the
>'dcft3' so we have to avoid using the nq3=1.
>(For some reason, the 'dcft3' does not allow odd number of FFT
>dimensions..)
>
>So here is a question.
>I wonder if I can use the different Monkhorst-pack grid for pw.x and ph.x
>when calculating a phonon dispersion.
>For example, can I use the (16x16x1) k-points for pw.x and (16x16x2)
>q-points for ph.x. in the case of graphene or some two dimensional
>supercell
>?
>I guess it should not be matter but I just want to be sure ..
>
>Thank you.
>
>Sincerely,
>
>JISEOK KIM, Ph.D.
>Postdoctoral Research Associate
>Department of Materials Science and Engineering
>University of Texas at Dallas
>800 W. Campbell Rd. RL10
>Richardson, TX 75080
>(413)386-6285
>
>
>-------------- next part --------------
>An HTML attachment was scrubbed...
>URL:
>http://www.democritos.it/pipermail/pw_forum/attachments/20120831/9bf23f85/
>attachment.html
>
>------------------------------
>
>Message: 2
>Date: Fri, 31 Aug 2012 10:55:54 +0200
>From: Paolo Giannozzi <giannozz at democritos.it>
>Subject: Re: [Pw_forum] Possible bug(?) in q2r.x when using FFT
> routine in ESSL
>To: PWSCF Forum <pw_forum at pwscf.org>
>Message-ID: <1346403354.6068.21.camel at fe12lx.fisica.uniud.it>
>Content-Type: text/plain
>
>The simplest workaround is to move the *.dyn* files to another
>machine, compute there the force constants in reciprocal space.
>The ultimate solution is to go into Modules/fft_scalar.f90,
>edit routine cfft3d in such a way that it can cope with
>a dimension of length 1. You need to modify line 1238:
> IF( isign /= 0 ) CALL dcft3( f(1), ldx,ldx*ldy, f(1), ldx,ldx*ldy, &
> nx,ny,nz, idir, tscale, work(1), lwork)
>and call dcft2 instead. You will also need to add initialization of
>dft2 in line 1159, definition of needed arrays somewhere around
>line 1094.
>
>Of course you can use any grid of phonon wave-vectors, but it
>is a waste of time.
>
>P.
>
>On Fri, 2012-08-31 at 03:19 -0500, Jiseok Kim wrote:
>> Dear, QE users
>>
>>
>> I have been struggling for a while to find out what's going on with
>> q2r.x
>> The problem was that I could not get rid of the error messages below
>> when doing q2r.x.
>>
>>
>> -------------------------------------------------
>> DCFT3 : 2538-2030
>> The transform length (ARG NO. 9) is not an allowed value. The next
>> higher
>> allowed value is (2).
>> -------------------------------------------------
>>
>>
>> I've compiled QE-v5.0.1. on AIX6.1(parallel machine) with 'mpxlf' with
>> ESSL routines.
>> So the 'q2r.x' calls 'cfft3d' and then the 'cfft3d' calls 'dcft3' in
>> ESSL when performing FFT.
>>
>>
>> The problem is this.
>> The 'dcft3' in ESSL only accepts certain values of transforming length
>> as shown in,
>>
>>
>> -------------------------------------------------
>>
>>http://publib.boulder.ibm.com/infocenter/clresctr/vxrx/index.jsp?topic=%2
>>Fcom.ibm.cluster.essl.v5r1.essl100.doc%2Fam501_hscft3.htm
>> -------------------------------------------------
>>
>>
>> so the FFT dimensions (nx, ny, nz), which is read from outputs by
>> ph.x, should be the values in the table.
>> Otherwise, the 'dcft3' crashes.
>>
>>
>> The FFT dimensions (nx, ny, nz) in q2r.x are equivalent to the (nq1,
>> nq2, nq3) in ph.x so that the Monkhorst-pack grid for ph.x should be
>> limited by the values in the table when using the ESSL.
>>
>>
>> For example, the Monkhorst-pack grid (16x16x1) for q-points in the
>> case of two-dimensional supercell, such as graphene, cannot be
>> accepted by the 'dcft3' so we have to avoid using the nq3=1.
>> (For some reason, the 'dcft3' does not allow odd number of FFT
>> dimensions..)
>>
>>
>> So here is a question.
>> I wonder if I can use the different Monkhorst-pack grid for pw.x and
>> ph.x when calculating a phonon dispersion.
>> For example, can I use the (16x16x1) k-points for pw.x and (16x16x2)
>> q-points for ph.x. in the case of graphene or some two dimensional
>> supercell ?
>> I guess it should not be matter but I just want to be sure ..
>>
>>
>> Thank you.
>>
>>
>> Sincerely,
>>
>>
>> JISEOK KIM, Ph.D.
>> Postdoctoral Research Associate
>> Department of Materials Science and Engineering
>> University of Texas at Dallas
>> 800 W. Campbell Rd. RL10
>> Richardson, TX 75080
>> (413)386-6285
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://www.democritos.it/mailman/listinfo/pw_forum
>--
>Paolo Giannozzi, IOM-Democritos and University of Udine, Italy
>
>
>
>
>------------------------------
>
>Message: 3
>Date: Fri, 31 Aug 2012 20:03:34 +0800
>From: Yue-Wen Fang <yuewen.fang at gmail.com>
>Subject: Re: [Pw_forum] Join to group , confirmation #
>To: PWSCF Forum <pw_forum at pwscf.org>
>Message-ID:
> <CAMUwqQfX9gFb7E+kYPhQFzj8MF8WJ2HDgQuDK8fB0GRaznO9nw at mail.gmail.com>
>Content-Type: text/plain; charset="iso-8859-1"
>
>Hello, you have succeeded in subscribing this mail list and you can post
>your problems by sending emails to pw_forum at pwscf.org
>
>Best Regards
>Yuewen
>East China Normal Univ
>
>2012/8/29 Saeedeh S Ravandi <sxr109320 at utdallas.edu>
>
>> Hi ,
>>
>> --
>> Best Regards,
>> Saeedeh Ravandi
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://www.democritos.it/mailman/listinfo/pw_forum
>>
>
>
>
>--
>----
>Yue-Wen Fang
>-------------- next part --------------
>An HTML attachment was scrubbed...
>URL:
>http://www.democritos.it/pipermail/pw_forum/attachments/20120831/b3ac7846/
>attachment-0001.htm
>
>------------------------------
>
>_______________________________________________
>Pw_forum mailing list
>Pw_forum at pwscf.org
>http://www.democritos.it/mailman/listinfo/pw_forum
>
>
>End of Pw_forum Digest, Vol 63, Issue 1
>***************************************
More information about the users
mailing list