[RFC] Merge criteria for tree-ssa

[Jan Hubicka]

> On Fri, Feb 06, 2004 at 12:24:02AM +0100, Jan Hubicka wrote:
> > THat means that tree-ssa happens to inline fewer
> > functions so you get kind of apples-to-orranges comparsion.
> 
> No, it isn't.  The metric is "time".  There is, fortunately, a
> widely accepted standard for this metric.
> 
> The user doesn't care about inlining, limits or otherwise.
> The user cares about the end results.

The item I commented on was about memory usage, not times.  What I
worried about is that for compile time, we do have bootstrap times that
are more or less unaffected by inlining decisions, but for memory there
are only tests that are very affected by this factor.

But this all does really not matter I guess, we have to work out how to
fix both.
> 
> If tree-ssa has a side effect that causes some internal widget
> to change that affects one of the external visible metrics of
> "compile time" or "run time", then that's what we have to deal with.
> 
> Yes, the internal widget that changed makes it more difficult
> for us to analyze what went wrong, or what needs changing, or
> that the bottlenecks are.  But that's just too bad.

Sure :)
It may me slightly easier to attack individual components at least
partly component wise, but we have to get it all working together
anyway.

Honza
> 
> 
> r~