Bootstrap comparison failures on i586 (mainline)

[Graham Stott]

Neil

Neil Booth wrote:
> 
> Am I the only one seeing these?  They've persisted for quite a
> while now:-
> 
> Bootstrap comparison failure!
> cxxmain.o differs
> fold-const.o differs
> real.o differs
> tree.o differs
> f/com.o differs
> 
> Neil.

I believe this is caused this change by Bernds 

        * cselib.c (hash_rtx): Don't do tail recursion elimination by hand.

I reverted this change in my local tree and the bootstrap got passed the
comparison stage. I was hoping to look at what's going wrong over the weekend
but wasn't able to do so.

Graham