[LLVMbugs] [Bug 23067] New: Alias analysis fails to recognize non-zero variable offset

[bugzilla-daemon at llvm.org]

https://llvm.org/bugs/show_bug.cgi?id=23067

            Bug ID: 23067
           Summary: Alias analysis fails to recognize non-zero variable
                    offset
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Global Analyses
          Assignee: unassignedbugs at nondot.org
          Reporter: james.molloy at arm.com
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

In the following code, &a[i] cannot alias &a[i+m], as m is known to be nonzero.
However, *x is reloaded across the "*w >>= 1" statement.

void test(int n, int m, int * restrict a)                                       
{                                                                               
    for (int j = 0; j < m; ++j) {                                               
        for (int i = j; i < n; ++i) {                                           
            int *x = &a[i];                                                     
            int *w = &a[i+m];                                                   

            *x >>= 1;                                                           
            *w >>= 1;                                                           

            *x += 2;                                                            
        }                                                                       
    }                                                                           
}

... Produces:

for.body3:                                        ; preds =
%for.body3.preheader, %for.body3
  %indvars.iv23 = phi i64 [ %indvars.iv.next24, %for.body3 ], [ %indvars.iv,
%for.body3.preheader ]
  %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv23
  %1 = add nsw i64 %indvars.iv23, %0
  %arrayidx5 = getelementptr inbounds i32, i32* %a, i64 %1
  %2 = load i32, i32* %arrayidx, align 4, !tbaa !1
  %shr = ashr i32 %2, 1
  store i32 %shr, i32* %arrayidx, align 4, !tbaa !1          *** POINTLESS
STORE
  %3 = load i32, i32* %arrayidx5, align 4, !tbaa !1
  %shr6 = ashr i32 %3, 1
  store i32 %shr6, i32* %arrayidx5, align 4, !tbaa !1
  %4 = load i32, i32* %arrayidx, align 4, !tbaa !1           *** POINTLESS
RELOAD
  %add7 = add nsw i32 %4, 2
  store i32 %add7, i32* %arrayidx, align 4, !tbaa !1
  %indvars.iv.next24 = add nuw nsw i64 %indvars.iv23, 1
  %lftr.wideiv = trunc i64 %indvars.iv.next24 to i32
  %exitcond = icmp eq i32 %lftr.wideiv, %n
  br i1 %exitcond, label %for.inc8.loopexit, label %for.body3

-- 
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/20150330/36dd50c5/attachment.html>