Re: [PATCH]: Fix PR tree-optimization/27755, add partial anticipation

[Jan Hubicka <hubicka at ucw dot cz>]

> >I am not sure about if we want to mix -O3 with compile time and code
> >growth implications.
> 
> O3 is supposed to be optimizing harder than O2, and make more of a
> space-speed tradeoff.
> That is what partial anticipation is doing.

Yep, what I have problem with is that -O3 now mix many concepts (we
mix together code growing optimizations, debugger unfritendly and
compile time unfriendly).  If there was separate knob for it, I would
enable it for -fprofile-use as people willing to wait twice as long for
compilation and desigining training run will probably wait few extra
percents of time just to get even slightly better code quality.
(ie I hope -fprofile-use -O2 to produce code as fast as -O3 without the
code growth implications because profile can guide the expansion)
But I don't have strong feelings here, at least until your new
transformation proves to be very effective on some important testcases
:))

Honza
> 
> > We already do have -fexpensive-optimizations
> >enabled at -O2, perhaps we can have separate knob for thise such as
> >-freally-expensive-optimizations ;))
> >
> 
> I seriously doubt even another knob is a good idea here.
> >Honza
> >