[llvm-bugs] [Bug 34361] New: Slowdown on x86 due to r310260

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=34361

            Bug ID: 34361
           Summary: Slowdown on x86 due to r310260
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: saher.odeh at intel.com
                CC: llvm-bugs at lists.llvm.org

After the above commit, the compiler now decides to vectorize additional code.
However, in spec2017/525 (avx2-x86_64 Linux), the vectorization of
x264_pixel_satd_8x4() causes a ~12% slow-down - presumably due to heavy use of
port 5 (shuffles?).

the following is the breakdown of the resulting instructions and their count:
  Count "Instruction
--------------------
     50 "vmovd
     32 "vpmovzxdq
     32 "vpmovzxbd
     32 "movzxw
     30 "vpsubq
     17 "vpaddq
     14 "vpextrd
     13 "add
     12 "sub
     11 "vpinsrd
     10 "vpshufd
     10 "leal
      8 "vpsllq
      8 "vpblendd
      6 "pushq
      6 "popq
      5 "vpaddd
      4 "leaq
      2 "vpxor
      2 "vpsrld
      2 "vpmulld
      2 "vpbroadcastdl
      2 "vpand

please let me know if more information is needed.

-- 
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/20170829/81d0f534/attachment.html>