[LLVMbugs] [Bug 2341] New: memcmp() optimization is erroneous

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=2341

           Summary: memcmp() optimization is erroneous
           Product: new-bugs
           Version: unspecified
          Platform: PC
        OS/Version: MacOS X
            Status: NEW
          Severity: critical
          Priority: P2
         Component: new bugs
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: scottm at aero.org
                CC: llvmbugs at cs.uiuc.edu


Created an attachment (id=1651)
 --> (http://llvm.org/bugs/attachment.cgi?id=1651)
memcmp() optimization bug test case

After a long haitus on an internal work task, I'm (yet again) building llvm
from bootstrap (Release native compiled llvm -> llvm gcc 4.2 -> Release
llvm-gcc compiled llvm). Found a really good bug that I've been able to narrow
down to a test case.

The offending assembly code is:

        movw    _.str-"L4$pb"(%esi), %ax
        cmpw    %ax, _buf-"L4$pb"+4(%esi)

which is comparing two 16-bit words, not two 32-bit words. This allows things
like "code" and "cond" to be memcmp()-ed as equal. This is clearly erroneous.

This is TOT code.

To reproduce: llvm-c++ -O3 testcase.cpp -lstdc++
Output produced is: was_token(buf + 4, 4) = 1
Output should be: was_token(buf + 4, 4) = 0


-- 
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.