[llvm-bugs] [Bug 32135] New: Loop vectorizer miscompile

via llvm-bugs llvm-bugs at lists.llvm.org
Fri Mar 3 15:53:20 PST 2017


            Bug ID: 32135
           Summary: Loop vectorizer miscompile
           Product: libraries
           Version: trunk
          Hardware: All
                OS: All
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Loop Optimizer
          Assignee: unassignedbugs at nondot.org
          Reporter: craig.topper at gmail.com
                CC: llvm-bugs at lists.llvm.org

Created attachment 18052
  --> http://bugs.llvm.org/attachment.cgi?id=18052&action=edit
Test case

The attached test case produces different output when compiled at -O1 and -O2.

The power of 2 trip count of the first loop seems to be significant. Values of
64, 128, and 256 all expose the problem. 16 does not.

It's also significant that the index for the second loop is a smaller type than
the first loop. If the same index type is used for both the loop is reduced to
a single store before the loop vectorizer even runs.

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/20170303/b65bb72b/attachment.html>

More information about the llvm-bugs mailing list