Re: Gimple dump

["Richard Warburton" <richard dot warburton at gmail dot com>]

Thanks for the response.  The documentation currently availible for
the LTO seems somewhat spare on the implementation front.  For
example, neither the wiki entry, nor the documents that it links to
explain where any of the source code is availible.  Nor how different
functional areas decompose into the source.

It is also somewhat unclear from documentation to what extent the IR
pickling is complete.  For example, does stating that "Languages other
than C, C++, Objective-C and Fortran. " mean that any gimple generated
for C or C++ will definitely work with with this pickling system?

Any help on these matters would be greatly appreciated.

 Richard

On 11/21/06, Diego Novillo <dnovillo@redhat.com> wrote:
> Richard Warburton wrote on 11/21/06 10:35:
>
> > 1. how much information is lost from the Gimple IR in the dumping
> > process?
> >
> A good chunk.  Enough that you would have to spend significant effort
> reconstructing it.  You might be interested in the link time
> optimization project (LTO).  One of the goals of the project is to
> stream the IL on disk.  See http://gcc.gnu.org/wiki/LinkTimeOptimization
>
> > 2. Is there any current implementation that takes the dumped form of
> >  gimple, and can recreate the IR.
> >
> No, but I believe that the LTO branch is now at the point where the IR
> is stream to disk and read back.