[RFC PATCH 0/5] Cgraph changes and various devirtualizations

[Martin Jambor]

Hi,

I have incorporated the comments I have received into my patch queue
for improved devirtualization and indirect inlining.  Furthermore, I
have also included a patch for capability I forgot about and that is
to do multiple indirect inlining, i.e. to inline indirect calls to
function which has itself been indirectly inlined.  This probably
isn't really necessary in case of normal indirect inlining but chained
virtual calls of small functions which should be inlined are more
probable.

Unfortunately the SPEC 2006 gains turned out to be noise (which
somehow occurs fairly often with 483.xalancbmk).  Nevertheless, I
believe that there are many applications out there which have a lot of
small virtual functions that will benefit, especially from the
inlining.

The patches are to be applied on top of fix for PR 42450:
http://gcc.gnu.org/ml/gcc-patches/2010-03/msg00270.html

The patches bootstrap and test without any regression on x86_64-linux.
I intend to commit something along the lines into 4.6 later and to
pretty-ipa branch rather soon.

Thanks,

Martin