[RFC/patch] Callgraph based inlining heuristics

[Jan Hubicka]

> > Jan Hubicka wrote:
> > 
> > >Hi,
> > >I finally got C++ unit-at-a-time into shape so it don't consume all my free
> > >time and got into implementing first try of my proposed inlining heuristics
> > >that uses callgraph to decide on inlining top-down while inline using 
> > >current
> > >code bottom-up hopefully getting better from the both worlds.
> > >
> > >I am attaching somewhat raw implementation of the idea.  At the moment I am
> > >using cgraph edges to flag on whether particular inlining will happen so
> > >function can be inlinined into once caller but not inlined into another, 
> > >but
> > >the decisions must be transitive.
> > >
> > I am not sure what exactly you're saying in the last sentence here.
> 
> Assume that function A calls B that calls C.
> If we decide to make A leaf, we must also inline all calls to C into B
> in the other incarnations of B as there is only one edge in between A
> and B.
actually between B and C.

Honza