[Omp] Error: The PPD directives stack limit has been exceeded.

[Faustin Roman]

On 4/18/06, Meadows, Lawrence F <lawrence.f.meadows at intel.com> wrote:
> I can't give you a solution. This is an internal error of some
> sort. Do you object to sending me a test case?
>
> >-----Original Message-----
> >From: Faustin Roman [mailto:faustinroman at gmail.com]
> >Sent: Monday, April 17, 2006 11:55 PM
> >To: Meadows, Lawrence F
> >Subject: Re: [Omp] Error: The PPD directives stack limit has
> >been exceeded.
> >
> >On 4/18/06, Meadows, Lawrence F <lawrence.f.meadows at intel.com> wrote:
> >> This is a compile-time error message, not a runtime error message.
> >> It doesn't have anything to do with the OpenMP stack.
> >> If you have intel support you can submit a problem report.
> >> Also feel free to send me or your favorite Intel contact
> >> a failing test case and we'll make sure it gets to the right
> >> person.
> >>
> >> Regards,
> >>
> >> Larry Meadows
> >>
> >
> >Thank you for the swift response.
> >
> >Although I understand that my problem is a compile-time error
> >can you please indicate me a possible solution?
> >
> >I don't think I have Intel support. I'm using the
> >non-comercial version of
> >Intel fortran compiler for linux.
> >
> >Regarding the test case, so far I am still trying to get it working
> >by reducing the number of directives, but thank you for the kind offer.
> >
> >Regards,
> >Faustin Roman
> >
> >> >-----Original Message-----
> >> >From: omp-bounces at openmp.org [mailto:omp-bounces at openmp.org]
> >> >On Behalf Of Faustin Roman
> >> >Sent: Monday, April 17, 2006 4:23 PM
> >> >To: omp at openmp.org
> >> >Subject: [Omp] Error: The PPD directives stack limit has
> >been exceeded.
> >> >
> >> >Hi everyone!
> >> >
> >> >I get this message when trying to compile a fortran code with Intel
> >> >Fortran v.9.0.021:
> >> >
> >> >fortcom: Error: mpi_prog.for, line 315: The PPD directives
> >stack limit
> >> >has been exceeded.
> >> >
> >> >Please tell me if I should increase the stack limit (currently
> >> >at 8192) or the
> >> >KMP_SET_STACKSIZE_S()
> >> >
> >> >I have tried both ways but nothing seems to work...
> >> >
> >> >(I don't know if the version of glibc is built with the
> >> >FLOATING_STACKS parameter defined
> >> >as required by Intel workaround - see below)
> >> >
> >> >Any ideeas?
> >> >
> >> >Thanks,
> >> >Faustin Roman
> >> >
> >> >PS: If it helps someone, here is what I found so far...
> >> >
> >> >Intel(r) Fortran Compiler 9.0 for Linux Release Notes:
> >> >

Thank you for reply.

I've managed to reduce the number of directives involved in my program
and now everything works except that I had the following message:

OMP warning: Cannot form a team with 2 threads, using 1 instead.
Consider setting KMP_ALL_THREADS to a value larger than 32.

Increasing KMP_ALL_THREADS to 64 resolved this issue.

I don't have any objections sending you the whole code except that
there are some implementations of nuclear models that are not public
yet and making a test case without them is pointless and so instead
I tried finding another approach of the problem.

In my code I had constructions like this:

!$OMP PARALLEL IF(ind1.ne.1)
!$OMP DO SCHEDULE(STATIC) PRIVATE(ind1)
      do ind1=1,nd(1)
        a(1)=a1(1)+(ind1-1)*epsa(1)
!$OMP PARALLEL IF(ind2.ne.1)
!$OMP DO SCHEDULE(STATIC) PRIVATE(ind2)
        do ind2=1,nd(2)
	  a(2)=a1(2)+(ind2-1)*epsa(2)
!$OMP PARALLEL IF(ind3.ne.1)
!$OMP DO SCHEDULE(STATIC) PRIVATE(ind3)
	  do ind3=1,nd(3)
	    a(3)=a1(3)+(ind3-1)*epsa(3)

[...]

Instead I did :

!$OMP PARALLEL IF(ind1.ne.1)
!$OMP DO SCHEDULE(STATIC) PRIVATE(ind1)
      do ind1=1,nd(1)
        a(1)=a1(1)+(ind1-1)*epsa(1)
!$OMP PARALLEL DO SCHEDULE(STATIC) PRIVATE(ind2)
        do ind2=1,nd(2)
	  a(2)=a1(2)+(ind2-1)*epsa(2)
!$OMP PARALLEL DO SCHEDULE(STATIC) PRIVATE(ind3)
	  do ind3=1,nd(3)
	    a(3)=a1(3)+(ind3-1)*epsa(3)

and so on...

Thank you again for the offer,

Regards,
Faustin Roman

--
----------------------------------------------------------------------
Faustin Laurentiu Roman,                          Assistant Researcher
Nuclear Physics Department
"Horia Hulubei" National Institute for Physics and Nuclear Engineering
(IFIN-HH), P.O. Box MG-6, 76900 Bucharest-Magurele, Romania
Tel: 40-21-4042300/ ext. 4315, Fax: 40-21-4574432, Mobile:+40721244414
E-mail: froman_at_tandem.nipne.ro, faustinroman_at_yahoo.com
URL: http://tandem.nipne.ro/~vavrig/preeq.html
----------------------------------------------------------------------