open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

I obtain SIGSEGV errors by checking the Telemac3D installation in parallel using validation cases from validation_v6p1.zip available from this site: 009_Wesel and 310_NonLinearWave (the problem will repeat probably in all other cases). The problem occurs by partitioning of reference files with partel (also produced by myself) and for restart files - in 3D case (and not in 2D for telemac2d).

Separating partel to a directory and doing something like:

jaj@service0:~/partel> ifort -warn all -g -traceback -check all partel.f libmetis.a -o partel 2> compile.log

jaj@service0:~/partel> export F_UFMTENDIAN=big

jaj@service0:~/partel> cat partel.par
T3DPRE
T3DCLI
4
1
0

jaj@service0:~/partel> ./partel < partel.par > exec.log

...I obtain interesting outputs *.log included hereafter.

The results indicate that KNOLG index JJ is out of range, in a relatively fresh code indicated by "CD". Maybe the programmer is available to investigate what is happening in the case 3D restart/reference files? (Hint: 2D files are treated seemingly OK):

CD SECOND STEP
CD EACH RECORDING IS READ AND ONLY THE LOCAL VARIABLES ARE STORED
CD INTO THE PARALLEL GEO FILE
DO
READ(NINP, END=1111, ERR=300) TIMES
WRITE(NOUT) TIMES
DO K=3,NVAR+2
READ(NINP, END=300, ERR=300) (F(J,K), J=1,NPOIN)
DO JJ=1,NPOIN
IF (KNOGL(JJ,I) .NE. 0) THEN !!! this is the line 1556 !!
CD IF KNOGL(JJ,I) > 0 THE VARIABLE HAVING GLOBAL NUMBER
CD JJ BELONGS TO THE SUBDOMAIN I AND ITS LOCAL NUMBER IS
CD KNOGL(JJ,I)
F_P(KNOGL(JJ,I),K,I)=F(JJ,K)
END IF
END DO
END DO
DO K=3,NVAR+2
IF(NPLAN.EQ.0) THEN
WRITE(NOUT) (F_P(J,K,I),J=1,NPOIN_P(I))
ELSE
WRITE(NOUT) ((F(KNOLG(J,I)+(L-1)*NPOIN2,K),
& J=1,NPOIN_P(I)),
& L=1,NPLAN)
ENDIF
END DO
END DO
1111 CLOSE (NINP)
CLOSE (NOUT)
END DO
CD

---

CD END OF THE MODIFICATION TO PUT ALL THE
CD RECORDINGS IN PARALLEL GEO FILE 08/06/2011
CD

---

Looking forward for your answer,
Best regards,
Jacek

Hello Jacek,

This has been corrected and sent by svn yesterday. The subroutine changed is partel.f in library parallel.

Note that now files are partitioned with all their records, whereas in previous versions only the last time step was kept. This raised problems for Sisyphe and Tomawac and for this reason partel.f was changed (and this triggered the problem...).

With best regards,

Jean-Michel Hervouet

I'm getting the same problem. Do I need special access to be able to download the update that you've sent via svn? It doesn't appear to be in the main v6p1 tag. Dev code appears to be restricted?

Thanks,
Graham

Hello,

Here is the corrected partel.f to put in the parallel library, in its folder sources (you need then to recompile).

With best regards,

Jean-Michel Hervouet

P.S. I cannot attach the file for some reason, please send your E-mail to:

This email address is being protected from spambots. You need JavaScript enabled to view it. and I'll mail the corrected version.

Other possibility, take the section below and replace the old corresponding section in partel.f

CD
CD

---

CD MODIFICATION TO PUT ALL THE RECORDINGS IN PARALLEL
CD GEO FILE 08/06/2011
CD

---

CD FIRST STEP : CLOSE/REOPEN/REWIND THE FILE AND READ ALL THE RECORDINGS UNTIL
CD THOSE CONCERNING THE TIME-DEPENDENT VARIABLES
CD
CLOSE(NINP)
OPEN(NINP,FILE=NAMEINP,STATUS='OLD',FORM='UNFORMATTED')
REWIND(NINP)
READ (NINP) TITLE
READ (NINP) II, JJ
NVAR = II + JJ
DO II=1,NVAR
READ(NINP) VARI
ENDDO
READ (NINP) (II, JJ=1,10)
IF(II.EQ.1) THEN
READ(NINP) DATE_TMP(1), DATE_TMP(2), DATE_TMP(3),
& TIME_TMP(1), TIME_TMP(2), TIME_TMP(3)
ENDIF
READ(NINP) II,II,II,II
READ(NINP) ((II,JJ=1,NDP),K=1,NELEM)
READ(NINP) (II,JJ=1,NPOIN)
READ(NINP) (TMP,JJ=1,NPOIN)
READ(NINP) (TMP,JJ=1,NPOIN)
CD SECOND STEP
CD EACH RECORDING IS READ AND ONLY THE LOCAL VARIABLES ARE STORED
CD INTO THE PARALLEL GEO FILE
DO
READ(NINP, END=1111, ERR=300) TIMES
WRITE(NOUT) TIMES
DO K=3,NVAR+2
READ(NINP, END=300, ERR=300) (F(J,K), J=1,NPOIN)
!
! CORRECTION JMH 05/09/2011 : F_P IS NOT DIMENSIONED
! FOR 3D AND IS NOT USED IN 3D
IF(NPLAN.EQ.0) THEN
DO JJ=1,NPOIN
IF(KNOGL(JJ,I).GT.0) THEN
CD IF KNOGL(JJ,I) > 0 THE VARIABLE HAVING GLOBAL NUMBER
CD JJ BELONGS TO THE SUBDOMAIN I AND ITS LOCAL NUMBER IS
CD KNOGL(JJ,I)
F_P(KNOGL(JJ,I),K,I)=F(JJ,K)
ENDIF
ENDDO
ENDIF
! END OF CORRECTION JMH
ENDDO
DO K=3,NVAR+2
IF(NPLAN.EQ.0) THEN
WRITE(NOUT) (F_P(J,K,I),J=1,NPOIN_P(I))
ELSE
WRITE(NOUT) ((F(KNOLG(J,I)+(L-1)*NPOIN2,K),
& J=1,NPOIN_P(I)),L=1,NPLAN)
ENDIF
ENDDO
ENDDO
1111 CONTINUE
CLOSE (NINP)
CLOSE (NOUT)
ENDDO
CD

---

CD END OF THE MODIFICATION TO PUT ALL THE
CD RECORDINGS IN PARALLEL GEO FILE 08/06/2011
CD

---

Dear Graham,

please find included the file partel.f.zip I've received from JMH, note he has sent in Windows-ASCII format. Yes, this is a bug truly critical for the execution (!!!) and the improvement should probably better find its way into the official distribution and not to some obscure password-secured SVN collection.

Best regards,
Jacek