Haifa problems

[David Edelsohn]

>>>>> Jakub Jelinek writes:

Jakub> Wouldn't it be better to have certain treshold beyond which gcc would stop
Jakub> adding these notes and just stop doing some optimizations?
Jakub> At least egcs 1.1.2 had no problems with these lame functions.

	This is a suggestion that I have made on a few occassions.  The
IBM XL compilers precompute the amount of memory required for an
optimization pass.  If the size is beyond a threshold, the optimization is
skipped or re-attempted on a smaller scale (e.g., try applying to next
inner loop).  The IBM XL compiler warns that an optimization was not
performed, so that the user could invoke the compiler with a larger memory
limit. 

	I wish that GCC backend would approach optimizations using this or
some similar structure instead of all of these independent heuristics
thrown around the compiler.

David