Fix PR 13987 (compile time regression), try 2

[Andrew Pinski]

On Aug 28, 2004, at 9:04 PM, Devang Patel wrote:

> Any numbers to indicate compile time improvements? Do you recover all 
> of fold-const.i regression you mentioned in PR/13987 ?


I recover all of the fold-const.i regression minus some (3 seconds) for 
the
new passes and new overhead with more time vars but "local alloc" is 
back down.

Also I had to add "--param ggc-min-expand=30 --param 
ggc-min-heapsize=4096" to
get a far comparison, without them 3.5.0 is faster than 3.3 was.

Thanks,
Andrew Pinski