Patch fixing 3.3 bug 9745 =?ISO-8859-1?Q?=2F=C2?= 10021

[Mark Mitchell]

> In the current situation, the bug can be avoided by using
> -fno-strict-aliasing.  If this patch was applied, the performance
> regression could not be avoided.  I therefore think that the current
> situation is least worst of these two choices.

How does -fstrict-aliasing come into play?  Is that because of the
fixed-scalar-varying-struct optimization?  

If so, might that not have to do with Jim's observation about MEMs not
having the MEM_IN_STRUCT_P bit set correctly for ARRAY_REFs?

Jim, that part of your patch seems non-controversial.  Can you check
that part in on the mainline and on the branch if it fixes any of the
outstanding branch PRs?

Thanks,

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com