[PATCH] MemoryDependenceAnalysis always depends on DominatorTree

[Bill Wendling]

On Mar 31, 2013, at 10:52 AM, Evan Cheng <evan.cheng at apple.com> wrote:

> On Mar 29, 2013, at 3:54 PM, Daniel Berlin <dberlin at dberlin.org> wrote:
> 
>> On Fri, Mar 29, 2013 at 3:39 PM, Evan Cheng <evan.cheng at apple.com> wrote:
>>> 
>>> On Mar 27, 2013, at 5:44 PM, Matt Arsenault <Matthew.Arsenault at amd.com> wrote:
>>> 
>>>> Hi nlewycky,
>>>> 
>>>> The DominatorTree was optional, but the uses of the tree aren't checked, causing it to crash when not available.
>>>> 
>>>> I've made DominatorTree required, although I'm not sure if it would be better to put the checks before its uses instead.
>>> 
>>> Adding the checks before DT is used is probably the better fix. Can you try that? Also, please include a test case.
>>> 
>> If done, this seems like it would just cause the bug fixed in 166490
>> to reoccur, unless that got fixed some other way.  (This is the reason
>> the DT checks were added)
> 
> It could be fixed by having the DT check return conservatively safe answer when DT is not available. Bill, please comment.
> 
It's important to make sure that we aren't trying to run GVN on constructs that r166490 is trying to prevent. If DT isn't available, then a conservative answer should work just fine in its place.

-bw