[LLVMbugs] [Bug 22589] New: 22% performance degradation in SciMark2's Sparse matmult benchmark

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=22589

            Bug ID: 22589
           Summary: 22% performance degradation in SciMark2's Sparse
                    matmult benchmark
           Product: clang
           Version: 3.6
          Hardware: Macintosh
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: -New Bugs
          Assignee: unassignedclangbugs at nondot.org
          Reporter: howarth.mailing.lists at gmail.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

There is a serious 22% performance degradation in the code generated for the
Sparse matmult benchmark of SciMark2 in the 3.6 release branch clang compiler
as well as the current svn trunk clang compiler.

Building the sources from http://math.nist.gov/scimark2/scimark2_1c.zip with...

make CFLAGS="-O3 -march=native"

on x86_64-apple-darwin14 reveals timings of...

 998.439+/-0.482 for the clang 3.5.1 release compilers

 1217.363+/-1.1004 for the clang 3.6 release branch compilers

This problem not to exist in current llvm/clang svn trunk.

 989.315+/-0.686

I'll try to do a inverse-regression hunt in current trunk and see at what
recent commit the performance regression in the Sparse matmult benchmark
disappeared.

-- 
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/20150214/08c8eaef/attachment.html>