[AVX-512] Unpack support in new shuffle lowering

[Adam Nemet]

> On Jan 9, 2015, at 12:47 AM, Chandler Carruth <chandlerc at google.com> wrote:
> 
> 
> On Thu, Jan 8, 2015 at 10:11 PM, Adam Nemet <anemet at apple.com <mailto:anemet at apple.com>> wrote:
> The change should be obvious.  For the test, I decided to start resurrecting
> (and extending) avx512-shuffle.ll.  This was removed when the new shuffle
> lowering was made default.
> 
> We could also add these to vector-shuffle-512-v8.ll but I think those are
> somehow generated with Chandler's fuzzer, so I decided against it.
> 
> Chandler, is this reasonable?
> 
> My inclination would be to follow the testing structure of vector-shuffle-512-v8.ll and extend it. Those aren't generated by my fuzzer directly. I have a number of scripts that help generate these style of test cases some I have mailed out (shufle_test_gen.py) but never really got consensus on submitting (its terrible python code I'm afraid). I also have a script that takes an existing test of the form of these shuffle tests and re-computes the file-check portions. I can mail that out if I haven't already.

OK, updated version is committed in r225837 and r225838.

Adam

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150113/8f48898c/attachment.html>