Home | History | Annotate | Download | only in LoopAccessAnalysis
      1 ; RUN: opt -loop-accesses -analyze < %s | FileCheck %s
      2 ; RUN: opt -passes='require<scalar-evolution>,require<aa>,loop(print-access-info)' -disable-output  < %s 2>&1 | FileCheck %s
      3 
      4 ; We give up analyzing the dependences in this loop due to non-constant
      5 ; distance between A[i+offset] and A[i] and add memchecks to prove
      6 ; independence.  Make sure that no dependences are reported in
      7 ; this case.
      8 ;
      9 ;   for (i = 0; i < n; i++)
     10 ;    A[i + offset] = A[i] * B[i] * C[i];
     11 
     12 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
     13 target triple = "x86_64-apple-macosx10.10.0"
     14 
     15 ; CHECK: Memory dependences are safe with run-time checks
     16 ; CHECK-NEXT: Dependences:
     17 ; CHECK-NEXT: Run-time memory checks:
     18 ; CHECK-NEXT: 0:
     19 ; CHECK-NEXT: Comparing group
     20 ; CHECK-NEXT:   %arrayidxA2 = getelementptr inbounds i16, i16* %a, i64 %idx
     21 ; CHECK-NEXT: Against group
     22 ; CHECK-NEXT:   %arrayidxA = getelementptr inbounds i16, i16* %a, i64 %indvar
     23 
     24 @B = common global i16* null, align 8
     25 @A = common global i16* null, align 8
     26 @C = common global i16* null, align 8
     27 
     28 define void @f(i64 %offset) {
     29 entry:
     30   %a = load i16*, i16** @A, align 8
     31   %b = load i16*, i16** @B, align 8
     32   %c = load i16*, i16** @C, align 8
     33   br label %for.body
     34 
     35 for.body:                                         ; preds = %for.body, %entry
     36   %indvar = phi i64 [ 0, %entry ], [ %add, %for.body ]
     37 
     38   %arrayidxA = getelementptr inbounds i16, i16* %a, i64 %indvar
     39   %loadA = load i16, i16* %arrayidxA, align 2
     40 
     41   %arrayidxB = getelementptr inbounds i16, i16* %b, i64 %indvar
     42   %loadB = load i16, i16* %arrayidxB, align 2
     43 
     44   %arrayidxC = getelementptr inbounds i16, i16* %c, i64 %indvar
     45   %loadC = load i16, i16* %arrayidxC, align 2
     46 
     47   %mul = mul i16 %loadB, %loadA
     48   %mul1 = mul i16 %mul, %loadC
     49 
     50   %idx = add i64 %indvar, %offset
     51   %arrayidxA2 = getelementptr inbounds i16, i16* %a, i64 %idx
     52   store i16 %mul1, i16* %arrayidxA2, align 2
     53 
     54   %add = add nuw nsw i64 %indvar, 1
     55   %exitcond = icmp eq i64 %add, 20
     56   br i1 %exitcond, label %for.end, label %for.body
     57 
     58 for.end:                                          ; preds = %for.body
     59   ret void
     60 }
     61