[llvm-bugs] [Bug 26881] New: Non-vectorised code slower on ARM

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=26881

            Bug ID: 26881
           Summary: Non-vectorised code slower on ARM
           Product: libraries
           Version: 3.8
          Hardware: Other
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: ARM
          Assignee: unassignedbugs at nondot.org
          Reporter: tulipawn at gmail.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

So I thought I'd see how well code from issue #26837 was doing on ARM and
here's the result on Cortex-A5:

running 2 tests

wo/NEON

test folds1 ... bench:       3,107 ns/iter (+/- 59)
test folds2 ... bench:       2,490 ns/iter (+/- 34)

w/NEON

test folds1 ... bench:       1,293 ns/iter (+/- 25)
test folds2 ... bench:       2,493 ns/iter (+/- 28)

Judging from those x86 results, there should be no difference between the two
versions in the absence of vector instructions' availability and yet the first
ARM result seems inverted. Probably shouldn't happen.

Flags used:
 -C target-cpu=cortex-a5 -C target-feature=+vfp4,-neon -C
llvm-args=-force-target-max-vector-interleave=4

-- 
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/20160308/ed219f89/attachment.html>