[llvm] r214011 - [x86] Fix PR20355 (for real). There are many layers to this bug.

[Quentin Colombet]

On Jul 28, 2014, at 2:21 PM, Quentin Colombet <qcolombet at apple.com> wrote:

> 
> On Jul 28, 2014, at 12:58 PM, Chandler Carruth <chandlerc at gmail.com> wrote:
> 
>> 
>> On Mon, Jul 28, 2014 at 12:46 PM, Quentin Colombet <qcolombet at apple.com> wrote:
>> I think we still have a bug in the current implementation.
>> See my inline comments.
>> 
>> Seems likely. Can you craft a test case to demonstrate it?
> 
> I will see if I can come up with something.

I haven’t found anything yet. The case I know that was using this path (div -> mul) just uses the high part of the result. Thus, the problem does not show up.

CC’ed Benjamin, he may have an idea on how to test that.

Cheers,
-Quentin
> 
>>  
>> 
>> Also, I think it would be nice to update the comments on the masks value instead of removing them.
>> When I reversed engineered this code, it took me ages to figure out what were these magic values.
>> 
>> I'm not sure that's going to help. The problem with comments is that they can get out of sync. I also don't think that little endian offsets are surprising or need lots of clarification here? The 1 instead of 0 thing you point out screams wrong to me at a glance, and I don't think a comment would really make it better (or worse, unless it got out of date).
> 
> Fair enough!
> 
> -Quentin
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140729/e6ca3214/attachment.html>