[PATCH][LoopStrengthReduce] Don't trim a formula that uses a subset of required registers

[Andrew Trick]

On Apr 28, 2014, at 5:49 PM, Andrew Trick <atrick at apple.com> wrote:

> 
> On Apr 28, 2014, at 5:05 PM, Chandler Carruth <chandlerc at google.com> wrote:
> 
>> I don't object to having the fix checked in but:
>> 
>> On Mon, Apr 28, 2014 at 4:51 PM, Andrew Trick <atrick at apple.com> wrote:
>> one of the major difficulties in getting it fixed was we just have absolutely no way to force it in llc
>> 
>> This is *exactly* why we should absolutely spend time isolating what it is about this test case that doesn't reproduce with llc. This is *critical* to the LLVM test infrastructure and project. It's not something we can or should paper over, for exactly the reason you described: it makes finding these bugs nearly impossible.
> 
> I figured out exactly why this specific LSR bug would not reproduce with llc. That was one year ago now. The bug has been open that long. It is well known that we have a fundamental problem testing individual passes because serialization changes the use lists, and we want to be able to write passes that depend on the use list order. Someone else was looking at fixing that problem at the time…. it must not be easy because it still isn’t fixed.

http://llvm.org/PR5680

reported in 2009.

> 
> -Andy
> _______________________________________________
> 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/20140428/77dc16e4/attachment.html>