NaN in Dissipation/test_ah.par test case

Create issue
Issue #1731 resolved
Erik Schnetter created an issue

I see nans returned from the elliptic solver in the test case Dissipation/test_ah.par when I switch off optimization and enable debug information. Things are fine with optimization. I do not know where exactly the nans are generated; the input to the solver looks fine, the output does not.

This is on OS X, with gcc 4.9.2.

Keyword:

Comments (3)

  1. Roland Haas

    I cannot reproduce this error right now, but from the thorns involved and the description I suspect this may have been fixed in git hash 194dc127 "IDAxiBrillBH: clear full arrays to avoid unitialized data access" of einsteininitialdata which contain:

    diff --git a/IDAxiBrillBH/src/mg59p.F b/IDAxiBrillBH/src/mg59p.F
    index 09e5ff10..8d33cf7d 100644
    --- a/IDAxiBrillBH/src/mg59p.F
    +++ b/IDAxiBrillBH/src/mg59p.F
    @@ -789,7 +789,15 @@ c   Now call the multigrid routine
    
           integer iscale,itry
    
    -
    +c clear all arrays
    +      ac = 0.
    +      an = 0.
    +      as = 0.
    +      aw = 0.
    +      ae = 0.
    +      q = 0.
    +      f = 0.
    +      gam = 0.
    
           ier=0
     *
    diff --git a/IDAxiBrillBH/src/shmgp.F77 b/IDAxiBrillBH/src/shmgp.F77
    index 7b863d09..046037aa 100644
    --- a/IDAxiBrillBH/src/shmgp.F77
    +++ b/IDAxiBrillBH/src/shmgp.F77
    @@ -258,6 +258,10 @@ c .....................................................................
           common /io/ linp,lout
           CCTK_REAL one
           parameter (one=1)
    +c clear arrays
    +      resid = 0.
    +      confac = 0.
    +      np2 = -1
     c
     c-time           tsu0=second()
              if(iskip.eq.0) then
    @@ -909,8 +913,8 @@ c
    
     c                                                              define pc
              if(ipc.ge.1) then
    -      do 20 j=2,jm1
    -      do 20 i=2,im1
    +      do 20 j=1,jm
    +      do 20 i=1,im
           fw(i,j)=0.
        20 qw(i,j)=1.
     c
    

    since otherwise some array elements were uninitialized. The commit is from Tue Jul 4 2017 ie after the ticket was created.

    Unless there are objections I will close this ticket with (probably) fixed.

  2. Log in to comment