mabshoff
unread,Jan 6, 2009, 1:51:21 PM1/6/09Sign in to reply to author
Sign in to forward
You do not have permission to delete messages in this group
Either email addresses are anonymous for this group or you need the view member email addresses permission to view the original message
to linbox-use
Hi,
I have been playing with multi threaded ATLAS as well as GotoBLAS
(single and multi threaded) and LinBox, but the current blas detection
does not deal with that too well or not at all. It can be easily
monkey patched or the name of the needs libs can be supplied via
LDFLAGS, but I would prefer if we had a clean solution. There are
various ways to do this, but I am not sure which one is best, i.e.
should we default to threaded BLAS if it is available? Without
introducing some new --with-parallel-blas or something alike it seems
like a risky change since it would require all upstream code of LinBox
that doesn't use pkg-config for example to also fix their makefiles to
link against the appropriate threaded libraries.
For a more long term project I am also looking into using FLAME for
the BLAS and Lapack backend since it promises easy GPU integration and
potentially better parallel scalability. Once FFPack become GPU aware
this will likely be no longer a big incentive, but I am not sure what
the progress is on that front. I do now have access to a Linux box
with a Dual Tesla C1060, so if anyone wants account to work on FFPack
on top of CUDABLAS just ping me off list for accounts.
Cheers,
Michael