ScaLAPACK Archives

[Scalapack] QR factorization for 1-D block layout


I was wondering if ScaLAPACK's QR factorization does any
special optimizations for the special case of a 1-D block layout
of an n x k matrix, where k << n (n/p x k blocks).  In particular, it
seems that pdlarfg.f computes the 2-norm of an entire column
(line 229 -- PDNRM2 call), which suggests that a standard sort
of Householder QR factorization is used -- please correct me if
I'm wrong.

I'm asking because I have a need for a Q-less QR factorization
for that particular layout, and I want to keep the number of
messages as small as possible.  I have some ideas how to do
it, but I want to check first if someone else has written software
to do just that.

Mark Hoemmen
UC Berkeley

<Prev in Thread] Current Thread [Next in Thread>
  • [Scalapack] QR factorization for 1-D block layout, Mark Hoemmen <=

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