[LLVMbugs] [Bug 10709] New: Miscompilation of checksumming code

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=10709

           Summary: Miscompilation of checksumming code
           Product: libraries
           Version: trunk
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Loop Optimizer
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: bhelf at flitterfly.whirpon.com
                CC: llvmbugs at cs.uiuc.edu


Created an attachment (id=7105)
 --> (http://llvm.org/bugs/attachment.cgi?id=7105)
Test case that shows bug

I've got some code that does a (weak) checksum calculation of a data file,
which appears to be miscompiled by clang/llvm at -O3.  When it's compiled at
-O0 the result matches the output of gcc/msvc/suncc/mipspro, and when it's -O3,
it's coming up with a different result.  

I've got a simplified test case that demonstrates this, attached below.

The line marked with // $$$ crazy loop is important.  If you take out that for
statement (even though it's just a 1 iteration loop) the problem goes away.  I
have no idea what that implies, to be honest.

Everything is using unsigned math, so I believe that it's not a problem with
undefined signed math behavior.  I believe that the hash_string code is working
correctly, since the "namesum" value it outputs is consistent between the -O3
and -O0 runs.  Using a simplified calculation for that string hash makes the
problem go away as well, it appears, so it's still there.

This is on linux x86-64, with the current clang and llvm svn trunk checkouts as
of 19 Aug 2011.

-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.