[PATCH] D30262: [SLP] Fix for PR32036: Vectorized horizontal reduction returning wrong result

Michael Kuperstein via llvm-commits llvm-commits at lists.llvm.org
Wed Feb 22 15:40:26 PST 2017


The test change looks wrong (we really need the instruction that
disappeared), and, oddly enough, it wasn't in the review.
And I guess the original failure on the bots was masked by the build break
in r295867.

Alexey, I'm going to revert it for now.

Thanks,
  Michael

On Wed, Feb 22, 2017 at 2:40 PM, Zachary Turner <zturner at google.com> wrote:

> I'm getting some test failures with this.
>
> FAIL: LLVM :: Transforms/SLPVectorizer/X86/horizontal-list.ll (18687 of
> 19826)
> ******************** TEST 'LLVM :: Transforms/SLPVectorizer/X86/horizontal-list.ll'
> FAILED ********************
> Script:
> --
> D:/src/llvmbuild/ninja-mono/./bin\opt.EXE -slp-vectorizer
> -slp-vectorize-hor -slp-vectorize-hor-store -S < D:\src\llvm-mono\llvm\test\
> Transforms\SLPVectorizer\X86\horizontal-list.ll
> -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 |
> D:/src/llvmbuild/ninja-mono/./bin\FileCheck.EXE
> D:\src\llvm-mono\llvm\test\Transforms\SLPVectorizer\X86\horizontal-list.ll
> D:/src/llvmbuild/ninja-mono/./bin\opt.EXE -slp-vectorizer
> -slp-vectorize-hor -slp-vectorize-hor-store -S < D:\src\llvm-mono\llvm\test\
> Transforms\SLPVectorizer\X86\horizontal-list.ll
> -mtriple=x86_64-unknown-linux-gnu -mcpu=bdver2 -slp-threshold=-10 |
> D:/src/llvmbuild/ninja-mono/./bin\FileCheck.EXE
> D:\src\llvm-mono\llvm\test\Transforms\SLPVectorizer\X86\horizontal-list.ll
> --check-prefix=THRESHOLD
> --
> Exit Code: 1
>
> Command Output (stdout):
> --
> $ "D:/src/llvmbuild/ninja-mono/./bin\opt.EXE" "-slp-vectorizer"
> "-slp-vectorize-hor" "-slp-vectorize-hor-store" "-S"
> "-mtriple=x86_64-unknown-linux-gnu" "-mcpu=bdver2"
> $ "D:/src/llvmbuild/ninja-mono/./bin\FileCheck.EXE"
> "D:\src\llvm-mono\llvm\test\Transforms\SLPVectorizer\X86\
> horizontal-list.ll"
> # command stderr:
> D:\src\llvm-mono\llvm\test\Transforms\SLPVectorizer\X86\horizontal-list.ll:1683:15:
> error: expected string not found in input
> ; CHECK-NEXT: [[BIN_EXTRA3:%.*]] = add i32 [[BIN_EXTRA]], [[TMP9]]
>               ^
> <stdin>:574:2: note: scanning from here
>  %r5 = add i32 %r4, undef
>  ^
> <stdin>:574:2: note: with variable "BIN_EXTRA" equal to "%bin.extra"
>  %r5 = add i32 %r4, undef
>  ^
> <stdin>:574:2: note: with variable "TMP9" equal to "%9"
>  %r5 = add i32 %r4, undef
>  ^
>
> error: command failed with exit status: 1
>
> --
>
> ********************
> Testing: 0 .. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90..
> Testing Time: 255.91s
> ********************
> Failing Tests (1):
>     LLVM :: Transforms/SLPVectorizer/X86/horizontal-list.ll
>
>   Expected Passes    : 19126
>   Expected Failures  : 186
>   Unsupported Tests  : 513
>   Unexpected Failures: 1
> FAILED: cmd.exe /C "cd /D D:\src\llvmbuild\ninja-mono\test &&
> C:\Python27\python.exe D:/src/llvm-mono/llvm/utils/lit/lit.py -sv --param
> llvm_site_config=D:/src/llvmbuild/ninja-mono/test/lit.site.cfg --param
> llvm_unit_site_config=D:/src/llvmbuild/ninja-mono/test/Unit/lit.site.cfg
> D:/src/llvmbuild/ninja-mono/test"
> ninja: build stopped: subcommand failed.
>
>
> Could you please take a look?
>
> On Wed, Feb 22, 2017 at 12:18 PM Alexey Bataev via Phabricator via
> llvm-commits <llvm-commits at lists.llvm.org> wrote:
>
>> This revision was automatically updated to reflect the committed changes.
>> Closed by commit rL295868: [SLP] Fix for PR32036: Vectorized horizontal
>> reduction returning wrong result (authored by ABataev).
>>
>> Changed prior to commit:
>>   https://reviews.llvm.org/D30262?vs=89378&id=89401#toc
>>
>> Repository:
>>   rL LLVM
>>
>> https://reviews.llvm.org/D30262
>>
>> Files:
>>   llvm/trunk/lib/Transforms/Vectorize/SLPVectorizer.cpp
>>   llvm/trunk/test/Transforms/SLPVectorizer/X86/horizontal-list.ll
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170222/20deecb6/attachment.html>


More information about the llvm-commits mailing list