Re: Proposed major change to combine

[kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)]

    However, if we could avoid unsharing RTL in advance that might
    alleviate some of the problems,
 
It's certainly true that the main reason for not sharing is combine,
but not the only, so I don't think we can do that, though perhaps
removing the *other* places that change RTL in place and not unsharing
might work.  But I don't think very much is shared, so I don't see that as
a major memory issue.
 
    and if we could run GC (if needed) after each BB was processed then
    the costs might not be too bad.
 
My thinking was to run it after every N insns, where N might be something
like 500.  Of course, the call won't do anything if not enough memory
is allocated.