ScaLAPACK Archives

[Scalapack] PDGBTRF

I have a particular problem in using the Scalapack routine PDGBTRF/S .
The codes run ok on a single and multiple PE's for certain matrix sizes.

For a particular matrix size of order 2000 with BWL=BWU=59, the code
runs ok on a single PE, but there is segmentation violation in
PDGETRF near the lines
*               Factor the final 2 by 2 block matrix
     $                   IPIV( LN+1 ), INFO )
         END IF
      END IF
*        Last processor in an odd-sized NPACT skips to here

when entering DGETRF for 2 procs, 4 procs. This happens on a number of
different machines including  a Bull Itanium2 system , the Hector system
at etc. The hector system generates malloc errors, with
the NAG, pgi, and gnu compilers.

I am curious to know why this is happening and suspect a bug in the
way pdgbtrf works. If you want I can supply the codes and
data files for you try out.

Jitesh Gajjar

<Prev in Thread] Current Thread [Next in Thread>
  • [Scalapack] PDGBTRF, Jitesh Gajjar <=

For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or