[LLVMbugs] [Bug 6394] New: Optimizer swaps bitcast and getelementptr in an invalid way

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=6394

           Summary: Optimizer swaps bitcast and getelementptr in an
                    invalid way
           Product: libraries
           Version: 2.6
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P5
         Component: Scalar Optimizations
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: llvm at henning-thielemann.de
                CC: llvmbugs at cs.uiuc.edu
   Estimated Hours: 0.0


I have written a custom malloc and free for vector aligned data, because as far
as I can see, the C interface to LLVM does not provide an alignment parameter
for malloc and without that, malloc does not choose 16 byte alignment for
vectors.

Anyway the following issue is certainly a bug.
Consider the custom free routine that I have written:

%wrap = type { float, <4 x float>* }

define void @_myfree(float*) {
_L1:
  %1 = bitcast float* %0 to %wrap*
  %2 = getelementptr %wrap* %1, i32 0, i32 1
  %3 = load <4 x float>** %2
  free <4 x float>* %3
  ret void
}

It is compiled to

    movl    8(%esp), %eax
    movl    4(%eax), %eax
    movl    %eax, (%esp)
    call    free

However, if I run optimization (opt -O1) I obtain

define void @_myfree(float* nocapture) nounwind {
_L1:
  %1 = getelementptr float* %0, i64 2             ; <float*> [#uses=1]
  %2 = bitcast float* %1 to <4 x float>**         ; <<4 x float>**> [#uses=1]
  %3 = load <4 x float>** %2                      ; <<4 x float>*> [#uses=1]
  free <4 x float>* %3
  ret void
}

which is compiled to

    movl    8(%esp), %eax
    movl    8(%eax), %eax
    movl    %eax, (%esp)
    call    free

which is certainly wrong. (See the differing displacement for (%eax).)

-- 
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.