ffc team mailing list archive
-
ffc team
-
Mailing list archive
-
Message #00593
Re: FFC performance
-
To:
Discussion of FFC development <ffc-dev@xxxxxxxxxx>
-
From:
Anders Logg <logg@xxxxxxxxx>
-
Date:
Mon, 8 May 2006 15:21:33 +0200
-
In-reply-to:
<1147093735.22523.1.camel@localhost.localdomain>
-
Mail-followup-to:
Discussion of FFC development <ffc-dev@xxxxxxxxxx>
-
User-agent:
Mutt/1.5.11
On Mon, May 08, 2006 at 03:08:55PM +0200, Garth N. Wells wrote:
> Hi Anders,
>
> We're using FFC here to compile some pretty complicated mixed nonlinear
> problems, and the FFC compile time (around 1 hour) is becoming an issue.
> Are there any areas in the code that are ripe for some optimisations
> (data structures, etc) that we could help with?
>
> Garth
Yes.
Check with -d0 how long it takes to compute the reference
tensor(s). There might be several in your case. This should be fairly
quick (I hope) since this has been optimized.
I suspect the main issue may be the actual code generation going on in
elementtensor.py, that is, generating the list of Declarations that
are later written to file in dolfin.py. I haven't tried to optimize
this part so it should be ripe for optimization.
If you can find out what is taking so long, then we can think of how
to do the optimizsation.
Also check with -f blas and see if that speeds up the code
generation. It should.
/Anders
Follow ups
References