dolfin team mailing list archive

[Martin Sandve Alnæs <martinal@xxxxxxxxx>]

I agree that some serious profiling of dolfin would be the
natural place to start for further performance tweaking.

Note that with UFL (coming soon to a form compiler near you!),
the form compilation bottleneck for complicated equations
is as good as gone, and the work being done on the form
compilers now is more focused on functionality than speed.

Martin


On Mon, Mar 23, 2009 at 3:19 PM, Robert Kirby <robert.c.kirby@xxxxxxxxx> wrote:
> I agree with these statements, Martin.
> I would also recommend serious profiling on a range of problems before
> substantial reengineering.
>
>
> On Mon, Mar 23, 2009 at 8:18 AM, Martin Sandve Alnæs <martinal@xxxxxxxxx>
> wrote:
>>
>> On Mon, Mar 23, 2009 at 1:58 PM, A Navaei <axnavaei@xxxxxxxxxxxxxx> wrote:
>> > 2009/3/23 Robert Kirby <robert.c.kirby@xxxxxxxxx>:
>> >> Hi all, some thoughts:
>> >> 1.) In the current paradigm (build + apply), building the matrix is
>> >> typically not dominant in the overall run-time.
>> >
>> > Isn't this in contrast with what Kent said about matrix insertion
>> > being the bottleneck? Are there any references for these claims?
>>
>> Matrix insertion is the bottleneck during the
>> assembly of matrices from simple PDEs
>> (at least simple ones).
>>
>> Depending on the application, other parts like the
>> linear solver are often more costly than the assembly.
>>
>> Both these claims are easy to verify.
>>
>> Martin
>
>