fix 23714

[Mark Mitchell]

Ulrich Weigand wrote:
> Andrew Pinski wrote:
> 
> 
>>	* tree-cfg.c (mark_array_ref_addressable_1): If the array
>>	reference is in bounds, don't do anything.
> 
> 
> I can confirm that this patch gets code generation for the
> affected SPECint function back to what we had before.

So, what do we think of this approach?

Andrew's patch reduces the number of arrays forced to memory.

I suggested an alternative method, which would be to notice in the back
end when we try to generate the address of a register in this way, and
simply return zero at that point; we know that executing that code must
be undefined memory.

I'm not sure which approach is safer, but I'd like to see about
recovering that performance.

RTH, what say you?

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304