[llvm] r214318 - Rename llvm-uselistorder => verify-uselistorder

Sean Silva chisophugis at gmail.com
Thu Jul 31 13:04:00 PDT 2014


On Thu, Jul 31, 2014 at 12:59 PM, Duncan P. N. Exon Smith <
dexonsmith at apple.com> wrote:

>
> > On 2014-Jul-31, at 11:39, Sean Silva <chisophugis at gmail.com> wrote:
> >
> > Ok, I guess that's not a problem. I guess it's sort of weird to see an
> option like -num-shuffles being used like this. Why not just test 1
> shuffle? or 10? Testing 5 "random" shuffles just so that it tests the same
> 5 shuffles every time doesn't seem very productive. If it's going to test 5
> deterministic shuffles then it should at least check 5 specific shuffles
> like reverse, pipe organ, interleave, etc.
>
> A given use-list isn't guaranteed to be changed.  Shuffling more times
> makes it more likely use-lists get changed.
>
> > At the very least, the number "5" should not be on every RUN line if it
> is just some arbitrary default number of shuffles to try and has no reason
> to vary from test to test.
>
> Larger testcases should shuffle more times, so that it's statistically
> likely that every use-list actually gets changed at least once.
>

Why not just change it so that it ensures that every use list is changed?
E.g. reject identity permutations. It seems silly to use a policy of "try
multiple times and hope that a non-identetity permutation comes out".

-- Sean Silva


>
> However, 1 is a pretty bad default (since every other 2-element list
> won't et changed).  Since all the testcases are pretty small, I've
> changed the default to 5 in r214419.
>
> > This would be great to add to the top-of-file comment in the .cpp file.
>
> Also r214419.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140731/8be6caf9/attachment.html>


More information about the llvm-commits mailing list