dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

Nuno Lopes wrote:
> New question #95433 on DOLFIN:
> https://answers.launchpad.net/dolfin/+question/95433
> 
> Hi!
> 
> I've been updating my dolfwave code from dolfin 0.9.2 to 0.9.5.
> Apart from some CGAL problems it is working, and 
> I'm able to run  some of the demos now.
> I've noticed a major slowdown from 0.9.2 to 0.9.5 (in my codes)
> Is this a known issue?
>

Is the only substantial slow down in the cell assembly?

Are you using the '-O' flag with FFC? This was on by default, but now
seems to be off by default.

Garth

> I note that the  solutions are the same for both versions.
> Using UMFPACK with factorized_solve I have
> the following summary of timings:
> 
> dolfin 0.9.2:
> Summary of timings        |  Average time  Total time  Reps
> -----------------------------------------------------------
> Assemble cells                  |     0.0046482        6.87  1478
> Assemble exterior facets |         0.005           0.01     2
> Build sparsity                     |    9.4723e-05       0.14   1478
> Delete sparsity                  |    6.7659e-05        0.1    1478
> Init dof map                        |             0                   0      11
> Init tensor                           |    6.7659e-06        0.01  1478
> LU solver                           |          0.02             0.02     1
> X:DOLFWAVE:END        |          7.64             7.64      1
> 
> and for dolfin  0.9.5:
> Summary of timings          |  Average time  Total time  Reps
> -----------------------------------------------------------
> Assemble cells                   |      0.011779       17.41   1478
> Assemble exterior facets  |         0.005            0.01     2
> Build sparsity                      |    6.7659e-06        0.01  1478
> Delete sparsity                   |             0                 0         1478
> Init tensor                            |             0                 0          1478
> LU solver                             |          0.03              0.03     1
> X:DOLFWAVE:END           |         85.12            85.12   1
> 
> Any ideas?!
> 
> N.Lopes
>