RFA: autoincrement patches for gcc 4 - updated patch

[Ian Lance Taylor]

Joern RENNECKE <joern.rennecke@st.com> writes:

> Ian Lance Taylor wrote:
> 
> >  That is good to hear.  But I think it might be easier to understand
> > if
> >it were written as two separate passes, one for this optimization, and
> >for the auto-increment detection, rather than one complex pass (which
> >is part of another complex pass).
> >
> You can't calculate the number of instructions used properly without
> combining
> these two things.  As a matter of fact, with web, it also makes sense
> to add analysis &
> generation of reg+offset addressing.
> 
> >  Sure.  This is something to fix.  If you queue it up with changes
> > like
> >the above, then I think it will be much easier to detect the
> >auto-increment opportunities.
> >
> But this can degrade scheduling.

It is probable that we can get the best code by writing a very
complicated pass which considers all different aspects of any
particular code change.  But I think we've discovered that the pass
winds up being unmaintainable and degrades over time.  We need to keep
our eyes on both maintainability and good code generation.  This is
unquestionably a hard problem.

Ian