[llvm-commits] [llvm] r81164 - /llvm/trunk/lib/Transforms/IPO/GlobalOpt.cpp

[Dan Gohman]

On Sep 8, 2009, at 4:19 AM, Duncan Sands wrote:

> Hi Dan,
>
>> Don't commit addresses of aggregate values. This avoids problems with
>> an aggregate store overlapping a different aggregate store, despite
>> the stores having distinct addresses.
>
> the same thing can happen for non-aggregates, for example storing an
> i32 and storing an i8 into the middle of it...


This code already rejects addressing that contains bitcasts and array
overindexing, so it shouldn't be possible to do an i32 store that  
overlaps with
an i8 store to a different address without having it detected.

Dan