Spurred by the recent query about multiplying
a 15000 x 15000 matrix, I came across the
FLAME project with its libflame:

  http://www.cs.utexas.edu/~flame/web/

This is a rearchitected linear algebra approach
based on the idea of blockwise decomposition
rather than elementwise computation.  The
performance is competitive with the best LAPACK
implementations and it naturally supports
out-of-core and distributed computations.

They have FLAME versions of BLAS and many
of the key LAPACK routines implemented and
eigenvalue solver is in the works.  This could
be an ideal library to provide future basic
matrix computations for PDL-2.x and PDL3.

An interesting question is how such a FLASH
hierarchical decomposition for matrix ops
might be of use for PDL threading.

--Chris

_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl

Reply via email to