[llvm-commits] [LLVMdev] [polly] scev codegen (first step to remove the dependence on ivcanon pass)

[Andrew Trick]

On Dec 13, 2012, at 3:37 PM, Tobias Grosser <tobias at grosser.es> wrote:

>> As this functionality is only used in Polly right now, I don't know how to write
>> test cases in LLVM.
> 
> I was thinking of unittests/Analysis/ScalarEvolutionTest.cpp
> 
> However, it seems such tests are not very reliable:
> http://llvm.org/bugs/show_bug.cgi?id=11052

SCEV's canonicalization and uniquing doesn't generally find equivalence. The bug you referenced is an example of the tension between canonicalizing the expression and optimizing (factoring to reduce multiplication). Someone could in theory write a much better SCEV equivalence checker if they wanted. But I don't think this will stop you from unit testing the rewriter.

Incidentally, all of Sebastian's changes look great to me. I'm just sad there's no use for it in mainline.

-Andy
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20121219/d0c7885f/attachment.html>