Home | History | Annotate | Download | only in LoopDistribute
      1 ; RUN: opt -basicaa -loop-distribute -loop-vectorize -force-vector-width=4 \
      2 ; RUN:   -verify-loop-info -verify-dom-info -S < %s | FileCheck %s
      3 
      4 ; If only A and B can alias here, we don't need memchecks to distribute since
      5 ; A and B are in the same partition.  This used to cause a crash in the
      6 ; memcheck generation.
      7 ;
      8 ;   for (i = 0; i < n; i++) {
      9 ;     A[i + 1] = A[i] * B[i];
     10 ; ------------------------------
     11 ;     C[i] = D[i] * E[i];
     12 ;   }
     13 
     14 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
     15 target triple = "x86_64-apple-macosx10.10.0"
     16 
     17 define void @f(i32*  %a,
     18                i32*  %b,
     19                i32* noalias %c,
     20                i32* noalias %d,
     21                i32* noalias %e) {
     22 entry:
     23   br label %for.body
     24 
     25 ; CHECK-NOT: memcheck:
     26 ; CHECK: mul <4 x i32>
     27 
     28 for.body:                                         ; preds = %for.body, %entry
     29   %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
     30 
     31   %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind
     32   %loadA = load i32, i32* %arrayidxA, align 4
     33 
     34   %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind
     35   %loadB = load i32, i32* %arrayidxB, align 4
     36 
     37   %mulA = mul i32 %loadB, %loadA
     38 
     39   %add = add nuw nsw i64 %ind, 1
     40   %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add
     41   store i32 %mulA, i32* %arrayidxA_plus_4, align 4
     42 
     43   %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind
     44   %loadD = load i32, i32* %arrayidxD, align 4
     45 
     46   %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind
     47   %loadE = load i32, i32* %arrayidxE, align 4
     48 
     49   %mulC = mul i32 %loadD, %loadE
     50 
     51   %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind
     52   store i32 %mulC, i32* %arrayidxC, align 4
     53 
     54   %exitcond = icmp eq i64 %add, 20
     55   br i1 %exitcond, label %for.end, label %for.body
     56 
     57 for.end:                                          ; preds = %for.body
     58   ret void
     59 }
     60