1 ; RUN: opt -basicaa -loop-accesses -analyze < %s | FileCheck %s
8 ; A[i][j] = A[i-1][j] * B[j]
9 ; B[j+1] = 2 // backward dep between this and the previous
12 ; is transformed by Load-PRE to stash away A[i] for the next iteration of the
15 ; Curr = A[0]; // Prev_0
17 ; Prev = Curr; // Prev = PHI (Prev_0, Curr)
20 ; Curr[j] = Prev[j] * B[j]
21 ; B[j+1] = 2 // backward dep between this and the previous
25 ; Since A[i] and A[i-1] are likely to be independent, getUnderlyingObjects
26 ; should not assume that Curr and Prev share the same underlying object.
28 ; If it did we would try to dependence-analyze Curr and Prev and the analysis
29 ; would fail with non-constant distance.
31 ; To illustrate one of the negative consequences of this, if the loop has a
32 ; backward dependence we won't detect this but instead fully fall back on
33 ; memchecks (that is what LAA does after encountering a case of non-constant
36 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
37 target triple = "x86_64-apple-macosx10.10.0"
40 ; CHECK-NEXT: Report: unsafe dependent memory operations in loop
41 ; CHECK-NEXT: Dependences:
42 ; CHECK-NEXT: Backward:
43 ; CHECK-NEXT: %loadB = load i8, i8* %gepB, align 1 ->
44 ; CHECK-NEXT: store i8 2, i8* %gepB_plus_one, align 1
46 define void @f(i8** noalias %A, i8* noalias %B, i64 %N) {
48 %prev_0 = load i8*, i8** %A, align 8
52 %i = phi i64 [1, %for_i.preheader], [%i.1, %for_j.end]
53 %prev = phi i8* [%prev_0, %for_i.preheader], [%curr, %for_j.end]
54 %gep = getelementptr inbounds i8*, i8** %A, i64 %i
55 %curr = load i8*, i8** %gep, align 8
56 br label %for_j.preheader
62 %j = phi i64 [0, %for_j.preheader], [%j.1, %for_j.body]
64 %gepPrev = getelementptr inbounds i8, i8* %prev, i64 %j
65 %gepCurr = getelementptr inbounds i8, i8* %curr, i64 %j
66 %gepB = getelementptr inbounds i8, i8* %B, i64 %j
68 %loadPrev = load i8, i8* %gepPrev, align 1
69 %loadB = load i8, i8* %gepB, align 1
71 %mul = mul i8 %loadPrev, %loadB
73 store i8 %mul, i8* %gepCurr, align 1
75 %gepB_plus_one = getelementptr inbounds i8, i8* %gepB, i64 1
76 store i8 2, i8* %gepB_plus_one, align 1
78 %j.1 = add nuw i64 %j, 1
79 %exitcondj = icmp eq i64 %j.1, %N
80 br i1 %exitcondj, label %for_j.end, label %for_j.body
84 %i.1 = add nuw i64 %i, 1
85 %exitcond = icmp eq i64 %i.1, %N
86 br i1 %exitcond, label %for_i.end, label %for_i.body