On Wed, Mar 17, 2010 at 12:30 PM, Barry Smith <bsmith at mcs.anl.gov> wrote:
> > Some of the PETSc examples are crashing under some circumstances (like > with complex) using PLAPACK and valgrind is showing uninitialized variables > etc in PLAPACK code. I don't want to debug PLAPACK code, so will let it go > for this release. > > BUT, is there something else we can move to for petsc-dev in the future. > Should we throw away the use of PLAPACK and use FLAMELIB? Should we switch > to ScaLAPACK (just kidding?). I hate to bag dense parallell matrices from > PETSc completely but don't want to spend a lot of time on whatever we do > support. Perhaps there should be a map of Robert's software empire. However, these things are disjoint: - FLAME: Serial LA algorithms generated from a worksheet specification - PLAPACK: Distributed parallel LA - FLASH: Multicore parallel LA generated from worksheet specification - Jack's code: Update of PLAPACK which has much better scalability Of course, I want to switch to the later. I am going to talk to Jack about this when I am in Austin. He is now the grad student of Lexing Ying, who if you don't know him, you really should check him out. Matt > > Barry > -- What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. -- Norbert Wiener -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20100317/1bbf8f07/attachment.html>
