The support for switchable storage backends in ublas is really great. You
can e.g. use it for:
* Write containers that does not re-allocate memory when size is shrinking
(like std::vector). This is great for real-time applications where you want
to minimize the number of allocations.
* Write containers that allocate alligned data that can be used directly in
SIMD code without any copying/realignment.
If you are concerned about performance, you can extend ublas with your own
optimized linear algebra functions. I have good experience with my own SIMD
and OpenMP functions for small/medium-sized problems where cache effects is
not an issue. For larger problems, you can use the external bindings to
ATLAS or MKL.
-Fredrik
----- Original Message -----
From: "Andreas Kl?ckner" <lists at informa.tiker.net>
To: <pyublas at tiker.net>
Sent: Tuesday, January 27, 2009 4:05 PM
Subject: Re: [Pyublas] eigen
Post by Neal BeckerI've been reading good things about eigen (on the eigen ML, mostly). I
wonder if anyone has looked into possibly interfacing this with
python/numpy?
Doesn't seem to have switchable storage backends like Ublas does. I'm
guessing
that'd make it somewhat hard. (~re-implementation of their own "Matrix"
class)
Plus, I'm really not eager to abandon Ublas. It does the trick for me.
ANdreas