[llvm-commits] ScalarEvolution fix for treeadd

[Nick Lewycky]

Dan Gohman wrote:
>>I'm not sure why, but it doesn't seem as promising as the previous test
>>I had. Could it be that when comparing the PDF I should ignore blocks
>>that are not part of the loop?
> 
> If there are branches from outside the loop to blocks inside the loop
> other than the loop header, then the exit branch may not be executed the
> same number of times as the loop header.
> 
> Can you identify a testcase that doesn't involve an infinite loop where
> the PDF test rejects a loop that the one of the other tests accepts?

No, it turns out that the previous test I had was Just Plain Buggy.

Dan, we should move this discussion over to PR817. I'll attach the patch
there.

Nick