dolfin team mailing list archive

[Johan Hake <747297@xxxxxxxxxxxxxxxxxx>]

On Friday April 1 2011 22:54:24 Joachim Haga wrote:
> I just realized: if it's cached *per object* and my other idea about vector
> re-use goes in, then most of the benefit will be there.

Hmmm, have to think about that. I am soon on vacation too ;)

Johan

> Den 2. apr. 2011 04.25 skrev "Johan Hake" <747297@xxxxxxxxxxxxxxxxxx>
> 
> følgende:
> > On Friday April 1 2011 14:12:00 Joachim Haga wrote:
> >> Aha, not so simple then. I don't quite get the multiple backend use
> >> case, but if it's supported then it's supported.
> > 
> > Well, it is not prohibited. It is probably not used by anyone. The point
> 
> is
> 
> > that Python is Python and any thing that is not prohibited is possible
> > and
> 
> we
> 
> > need to make sure it is not possible to multiply an EpetraMatrix with a
> > PETScVector.
> > 
> >> I didn't quite understand your suggestion. Do you mean to make
> >> down_cast() a method on Matrix et al instead of a global method? That
> >> sounds nice...
> > 
> > Yes something like that. But I am not sure we are able to get around a
> 
> check
> 
> > each time a Vector/Matrix is created. And then we are back to step one, I
> > guess.
> > 
> >> Anyhow: I'm away for six weeks, starting tomorrow, and don't know how
> >> much I'll be able to communicate when away. Feel free to leave this bug
> >> (and others, don't know if I have time now to report more) unresolved.
> >> Unless you want to fix them, of course :)
> > 
> > Have a good vacation!
> > 
> > Johan
> > 
> > --
> > You received this bug notification because you are a direct subscriber
> > of the bug.
> > https://bugs.launchpad.net/bugs/747297
> > 
> > Title:
> > Python: Cache tensor types
> > 
> > Status in DOLFIN:
> > New
> > 
> > Bug description:
> > In a matrix-multiplication heavy Python workload, I see something like
> > 5-10% of the time being spent in get_tensor_type(). The attached patch
> > memoizes the result, per type. Seems to work fine, but should be
> > sanity checked (is per-type result ok?).
> > 
> > To unsubscribe from this bug, go to:
> > https://bugs.launchpad.net/dolfin/+bug/747297/+subscribe

-- 
You received this bug notification because you are a member of DOLFIN
Team, which is subscribed to DOLFIN.
https://bugs.launchpad.net/bugs/747297

Title:
  Python: Cache tensor types

Status in DOLFIN:
  New

Bug description:
  In a matrix-multiplication heavy Python workload, I see something like
  5-10% of the time being spent in get_tensor_type(). The attached patch
  memoizes the result, per type. Seems to work fine, but should be
  sanity checked (is per-type result ok?).