[llvm-bugs] [Bug 31515] New: test-suite/MultiSource/Benchmarks/mediabench/jpeg/jpeg-6a/cjpeg.execution_time fails for KNL target starting from r289653

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=31515

            Bug ID: 31515
           Summary: test-suite/MultiSource/Benchmarks/mediabench/jpeg/jpeg
                    -6a/cjpeg.execution_time fails for KNL target starting
                    from r289653
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: X86
          Assignee: unassignedbugs at nondot.org
          Reporter: elad2.cohen at intel.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

test-suite/MultiSource/Benchmarks/mediabench/jpeg/jpeg-6a/cjpeg.execution_time
fails for KNL target.
I ran a 'git bisect' and pinned down commit 289653.

A diff on the test artifacts after and before the commit shows the following
changes in MultiSource/Benchmarks/mediabench/jpeg/jpeg-6a/jcdctmgr.s:

before:
vpandn    %ymm0, %ymm1, %ymm0

after:
.LCPI2_0:
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
    .short    1                       # 0x1
...
vmovdqa    .LCPI2_0(%rip), %ymm9   # ymm9 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
...
vpxor    %ymm9, %ymm1, %ymm1
vpand    %ymm0, %ymm1, %ymm0


Looks like  r289653 creates a XOR with [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]
instead of "allones" in this specific case.

I'm working on uploading a fix.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20170103/2c42821d/attachment.html>