ScaLAPACK-1.8.0 errata file updated on Mon September 07 2009 maintained by J. Langou, U. Colorado Denver LAPACK-3.2.1 errata file CLAPACK-3.2.1 errata file BLACS-1.1.patch3 errata file ScaLAPACK-1.8.0 errata file Wishlist Released in ScaLAPACK Installer 0.96 ==================================== (*) bug0018 :: (*) project :: scalapack installer o reported by Miska Le Louarn on Tue Mar 3 2009 o confirmed by Julie Langou on Tue Mar 3 2009 o see: emails Corrected in SVN ================ (*) bug :: workspace query not correct for IWORK in PDGECON when no local rows in process row (*) svn :: r68 o reported by Jill Reese (MathWorks) on Fri Aug 28 2009 o confirmed by Julien Langou on Fri Aug 28 2009 o fix from Jill and Julien o committed by Julien on Sat Sep 12 2009 (SVN: r68). (*) bug :: ScaLAPACK does not handle matrix larger than N > sqrt(2^31) (*) svn :: r69 o Reported several times o The following fix works fine. The fix has been provided by John Moyart from CNES, France. It works great for all who have tried it. replaced Mptr() definition: #define Mptr( a_, i_, j_, lda_, siz_ ) \ ( (a_) + ( ( (i_)+(j_)*(lda_) )*(siz_) ) ) by: #define Mptr( a_, i_, j_, lda_, siz_ ) \ ( (a_) + ( (off_t) ( (off_t)(i_)+(off_t)(j_)*(off_t)(lda_))*(off_t)(siz_) ) ) o Other fix: compile scalapack using 64-bit integers o committed by Julien on Sat Sep 12 2009 (SVN: r69). Confirmed problems -- bug needs to be fixed! ============================================ (*) bug :: sEncountering an infinite loop in psgels.f's use of pslascl.f for matrices with infinite ANRM (*) svn :: r71 o Reported by Jill Reese (MathWorks) on Sun Jun 06 2010 We've recently come across an issue with psgels.f and its use of pslascl.f. We are using the driver routine to solve a rectangular system of equations where the matrix is poorly conditioned (ANRM is inf). I realize that psgels.f assumes that the input matrix has full rank, and that we are using this routine in an instance where it isn't guaranteed to return the correct answer. Our issue is that the solve enters an infinite loop in the pslascl routine which is called by psgels on line 409. o Fixed by Jill Reese (MathWorks) on Mon Jun 14 2010 It turns out that the original LAPACK [cszd]lascl.f routines could encounter the same issue. These routines were patched about a year ago in LAPACK 3.2. Since these LAPACK routines are quite similar to their ScaLAPACK counterparts, our workaround for the time being is to apply the same patch to the ScaLAPACK routines. This has worked for us. o Patch apply by Julien Langou on Mon Jun 21 2010 (Followed Jill's comments.)