> On 07/11/2009 05:59 AM, Jan Hubicka wrote: > >Well, we can either teach inlinable_call_p to handle your new indirect > >calls as "for sure uninlinable", make it conservative and consider all > >calls inlinable or we can stop doing the early removal of MUST_NOT_THROW > >receivers. > > I think this last option would probably be best. Something to work on > next week...
It is easilly doable by making inlinable_call_p to return always true pre-inlining (even for non-calls). I can drop it to C++ testers so we can see how much memory/compile time overhad does this imply. I would like to bring more of EH lowering to tree level (i.e. instead of relying on RTL to lower RESX instructions into gotos/calls/jumptables do this at gimple and keep to RTL world only job of constructing landing pads). Then it might be nice to consider option to generate these in middle-end instead of preserving whatever frontend builds. The receivers are different for C++ and Java EH (calling std::terminate or abort) but otherwise there is nothing backend can't do. Currently we tend to end up with many std::terminate recerivers in single function that we should unify anyway. Honza > > > r~
