Re: Memory requirement regressions

[Richard Guenther <rguenth at tat dot physik dot uni-tuebingen dot de>]

Andrew Pinski wrote:
> > > We consume a lot more time in 'integration' than I expected...
> >
> > Indeed - whatever exactly integration is.
>
> integration is the inliner, it was the name for the old RTL inliner.
> We just kept the old time name.

Ah, ok.  Then may the excessive time spent in it due to we don't re-use 
inlined function bodies (do we?) like in:

inline void foo() { }
inline void bar() { foo(); }
void foobar()
{
   bar();
   bar();
}

Do we actually create the body of bar with inlined foo once, for 
inlining the same body twice into foobar or do we repeat this inlining
every time we inline bar into foobar (uh, that's a interwinded 
sentence)?  Do we / can we share the memory for the trees of the two 
inlined bars inside foobar?

Richard.