[LLVMbugs] [Bug 1741] New: BasicAA miscompilation

bugzilla-daemon at cs.uiuc.edu bugzilla-daemon at cs.uiuc.edu
Tue Oct 23 22:50:57 PDT 2007


http://llvm.org/bugs/show_bug.cgi?id=1741

           Summary: BasicAA miscompilation
           Product: libraries
           Version: 2.1
          Platform: PC
        OS/Version: All
            Status: NEW
          Severity: normal
          Priority: P2
         Component: Global Analyses
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: sabre at nondot.org
                CC: llvmbugs at cs.uiuc.edu


Consider this code:
struct A { int x ;};
struct B : A { };
B a;
int foo(A* b) {
  a.x = 1;
  b->x = 0;
  return a.x;
}
int main() {
   foo(&a);  
}

llvm-gcc incorrectly misoptimizes foo to return 1 directly.  If you reduce it
down, it is because GVN is transforming this code invalidly:

target datalayout =
"e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
target triple = "i686-apple-darwin8"
        %struct.A = type { i32 }
        %struct.B = type { %struct.A }
@a = global %struct.B zeroinitializer           ; <%struct.B*> [#uses=2]

define i32 @_Z3fooP1A(%struct.A* %b) {
entry:
        store i32 1, i32* getelementptr (%struct.B* @a, i32 0, i32 0, i32 0),
align 8
        %tmp4 = getelementptr %struct.A* %b, i32 0, i32 0               ;
<i32*> [#uses=1]
        store i32 0, i32* %tmp4, align 4
        %tmp7 = load i32* getelementptr (%struct.B* @a, i32 0, i32 0, i32 0),
align 8           ; <i32> [#uses=1]
        ret i32 %tmp7
}

It deletes the load and makes 'ret i32 1'.

This invalid transformation is happening because BasicAA is returning no-alias
for tmp4 and the @A gep constant expr:

$ llvm-as < t.ll | opt -count-aa -gvn -count-aa-print-all-queries | llvm-dis
No alias:       [4B] i32* %tmp4, [4B] i32* getelementptr (%struct.B* @a, i32 0,
i32 0, i32 0)
...

This should be investigated,

-Chris


-- 
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.



More information about the llvm-bugs mailing list