[LLVMbugs] [Bug 18526] New: LoopVectorizer: Mistakenly vectorize multi reduction value users

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=18526

            Bug ID: 18526
           Summary: LoopVectorizer: Mistakenly vectorize multi reduction
                    value users
           Product: tools
           Version: trunk
          Hardware: PC
                OS: All
            Status: NEW
          Severity: normal
          Priority: P
         Component: opt
          Assignee: unassignedbugs at nondot.org
          Reporter: aschwaighofer at apple.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

Created attachment 11889
  --> http://llvm.org/bugs/attachment.cgi?id=11889&action=edit
Test case

t3025 and t1519 from http://llvm.org/bugs/show_bug.cgi?id=16431#c22 are
misoptimized.

As described in
http://llvm.org/bugs/show_bug.cgi?id=16431#c23

"The diff tests are vectorizer bugs (look like the same bug): t3025 and t1519.
They disappear with -fno-vectorize.

Furthermore,

it seems to be the loops:

    for (k1 = 54; k1 > 2; k1--) {
        gp += gp - ((++n) + k1);

        // Test Loop Vectorization
        for (ju = 1; ju < 1; ++ju) {
            d5[ju] = ff[k1][k1];
        }

    }

and 

    for (k = 2; k < 61; ++k) {
        n += n - ((f + k7) + (nf++));
        xs5[k] = 13 + 5;
        xs5[k] += nt7--;
    }

We should not be vectorizing them. “gp” and “n” are not reductions. But we seem
to :(."

-- 
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/20140118/e362efbe/attachment.html>