[llvm] r288497 - [SLP] Fix for PR6246: vectorization for scalar ops on vector elements.

[Alexey Bataev via llvm-commits]

Hal, Renato,
I think there is enough data for the investigation. Will look at it on Monday

Best regards,
Alexey Bataev

> 2 дек. 2016 г., в 20:14, Hal Finkel <hfinkel at anl.gov> написал(а):
> 
> ----- Original Message -----
>> From: "Renato Golin via llvm-commits" <llvm-commits at lists.llvm.org>
>> To: "Alexey Bataev" <a.bataev at hotmail.com>
>> Cc: "LLVM Commits" <llvm-commits at lists.llvm.org>
>> Sent: Friday, December 2, 2016 11:07:31 AM
>> Subject: Re: [llvm] r288497 - [SLP] Fix for PR6246: vectorization for scalar ops on vector elements.
>> 
>> On 2 December 2016 at 12:20, Alexey Bataev via llvm-commits
>> <llvm-commits at lists.llvm.org> wrote:
>>> Author: abataev
>>> Date: Fri Dec  2 06:20:22 2016
>>> New Revision: 288497
>>> 
>>> URL: http://llvm.org/viewvc/llvm-project?rev=288497&view=rev
>>> Log:
>>> [SLP] Fix for PR6246: vectorization for scalar ops on vector
>>> elements.
>> 
>> Hi Alexey,
>> 
>> This commit broke the AArch64 build of the builtins (twice):
>> 
>> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/419
>> 
>> and
>> 
>> http://lab.llvm.org:8011/builders/clang-cmake-aarch64-full/builds/428
>> 
>> I have reverted it in r288497. We really should investigate this
>> offline.
> 
> Looks like it triggered a backend bug:
> 
> ExpandIntegerOperand Op #0: t143: f128 = select t142, t159, t8
> 
> Do not know how to expand this operator's operand!
> UNREACHABLE executed at /home/buildslave/buildslave/clang-cmake-aarch64-full/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp:2757!
> 
> we'll need to fix this before we can recommit the SLP change. It also looks like there's potentially a cost model problem here too (i.e. should the SLP vectorizer really be producing IR that the backend will need to expand?). Alexey, can you reproduce this or do you need someone to grab the preprocessed inputs for you?
> 
> -Hal
> 
>> 
>> cheers,
>> --renato
>> 
>> PS: You didn't get the email the second time because there was a test
>> failure on the build just before yours.
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>> 
> 
> -- 
> Hal Finkel
> Lead, Compiler Technology and Programming Languages
> Leadership Computing Facility
> Argonne National Laboratory