| Thread Previous • Date Previous • Date Next • Thread Next |
Anders Logg wrote:
On Thu, Aug 07, 2008 at 10:32:56PM +0100, Garth N. Wells wrote:Something I don't understand is the time for creating the sparsity pattern for MTL4 when it involves no operations?MTL4Factory still returns a sparsity pattern. I wasn't aware that MTL4 didn't need one. MT4Factory should return 0 in createPattern() and MTL4SparsityPattern can then be removed.
We're cheating a bit at the moment. MLT4 needs to know the maximum number on non-zeroes on any row. We're just guessing this for now.
Currently, MTLSparsityPattern::insert(..) is empty. If you look at my benchmark results, the cost of building the MTLSparsityPattern is negligible, but in yours and Ilmar's it's comparable to the cost of building the sparsity pattern for uBLAS and PETSc. I have no idea what's going on. I thought that the overhead was in SparsityPattern::insert(), which is what my results indicate, whereas yours indicate that the overhead is considerable in SparsityPatternBuilder.
I can fix this later today if you don't fix it before.
Let's leave it as id for now.
Why is uBLAS now faster than before (relative)?I didn't notice. I might have made an apt-get upgrade in between. I don't think anything has changed for uBlas in the code.
The only change I made was to SparsityPattern, which seems to have given you a bigger speed-up than it did me.
Garth
------------------------------------------------------------------------ _______________________________________________ DOLFIN-dev mailing list DOLFIN-dev@xxxxxxxxxx http://www.fenics.org/mailman/listinfo/dolfin-dev
| Thread Previous • Date Previous • Date Next • Thread Next |
