dolfin team mailing list archive
-
dolfin team
-
Mailing list archive
-
Message #17352
Re: New handling of form compiler parameters
On Mon, Feb 01, 2010 at 09:30:16AM +0000, Garth N. Wells wrote:
>
>
> Anders Logg wrote:
> > I have simplified the access to form compiler parameters from DOLFIN.
> > Here's a summary.
> >
> > 1. 'options' --> 'parameters' in FFC
> >
> > 2. DOLFIN stores form compiler parameters in parameters["form_compiler"]
> > and the values are easily accessible by
> >
> > info(parameters, True)
> >
> > or
> >
> > info(parameters["form_compiler"])
> >
> > 3. DOLFIN reads the default values at startup by calling
> >
> > form_compiler.default_parameters()
> >
> > 4. The 'options' parameter to jit, assemble and Form have has been
> > renamed 'form_compiler_parameters'.
> >
> > 5. Before calling the form compiler, DOLFIN first reads the global
> > parameters from parameters["form_compiler"], then overrides the values
> > with any local values provided by the form_compiler_parameters
> > argument.
> >
> > 6. To see more output from the form compiler, *decrease* the log level
> > of the form compiler by form example setting it to INFO:
> >
> > parameters["form_compiler"]["log_level"] = INFO
> >
> > This is the default log level in FFC, but DOLFIN changes it to INFO + 5
> > at startup to hide messages from the form compiler by default.
> >
> > Have a go and see if it works as expected.
> >
>
> How can I have FFC report its parameters? I've set the logging level to
> 'INFO', but I can't tell whether or not FFC is using optimisation for
> the generated code (the -O flag).
>
> Garth
We could let FFC print the parameters it gets, but that's not much use
if they are ignored in the code generation anyway. Instead, FFC should
print a message when it actually uses an option (like it does for
representation). So the quadrature optimizations need to write some
message that it's generating optimized code. (Something for Kristian
to add where he thinks it appropriate.)
That aside, there seems to be a bug in passing the parameters to FFC
so they are in fact ignored. I'm looking into it.
--
Anders
Attachment:
signature.asc
Description: Digital signature
References