dolfin team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Sun, Jun 01, 2008 at 12:21:37AM +0200, Johan Hake wrote:
> On Saturday 31 May 2008 23:10:05 DOLFIN wrote:
> > One or more new changesets pushed to the primary dolfin repository.
> > A short summary of the last three changesets is included below.
> >
> > changeset:   4271:8431a991a0be07ead18512308cf293a309d4b2a5
> > tag:         tip
> > user:        Anders Logg <logg@xxxxxxxxx>
> > date:        Sat May 31 23:09:58 2008 +0200
> > files:       demo/mesh/ale/cpp/main.cpp demo/mesh/ale/python/demo.py
> > dolfin/parameter/DefaultParameters.h site-packages/dolfin/assemble.py
> > description:
> > Several Python assembly optimizations and new options:
> > The following options are now available:
> >
> >   dolfin_set("optimize form", True)              # Use optimization -O2
> > when compiling generated code dolfin_set("optimize use dof map cache",
> > True) # Store dof maps in cache for reuse dolfin_set("optimize use tensor
> > cache", True)  # Store tensors in cache for reuse
> >
> >   dolfin_set("optimize", True)                   # All of the above
> >
> > Each one of these may make a significant difference for the
> > speed of assembly from Python, in particular for repeated
> > assembly.
> >
> > With these options set, the difference in speed between assembly
> > in C++ and Python using the JIT compiler is negligible. (For forms
> > which don't depend on functions that are evaluated in Python.)
> 
> Nice features!
> 
> It would be nice to be able to use inlined and just in time compiled c++ 
> functions, avoiding the time consuming python callback functions.
> 
> Johan

This is possible (Martin does it), but I don't know how.

Martin, is it be easy to do this directly from DOLFIN (with FFC) or
does it depend on SyFi?

-- 
Anders