Re: [tree-ssa] CCP inefficiencies

[law at redhat dot com]

In message <20030214212222.GB16166@redhat.com>, Richard Henderson writes:
 >On Fri, Feb 14, 2003 at 10:03:32AM -0700, law@redhat.com wrote:
 >> Again, the number of cases we need to handle is drastically smaller
 >> than what's dealt with in fold.  I really don't think it's going to 
 >> be terribly difficult.
 >
 >Nevertheless, it's code duplication.
 >
 >Any chance you could pull out the bits you need from the
 >existing fold, and then have fold call into them?  Probably
 >not a bad idea anyway, since the fold function itself is
 >too damn big....
Some of the code can easily be factored and shared (in fact, I had
already started down this path when I ran into some of CCP's sillyness
that I decided to fix before fixing fold :-)

If you look at fold, conceptually the code we're going to care about
is everything where "wins" is true.  But you'll also find that "wins"
code is a tiny tiny fraction of what happens inside fold.

A lot of what fold does is scramble nested expressions in the hopes
that it can be simplified -- we don't have that to deal with since
we're in GIMPLE form.

jeff