1; RUN: opt -loop-load-elim -S < %s | FileCheck %s
2; RUN: opt -passes=loop-load-elim -S < %s | FileCheck %s
3
4; Simple st->ld forwarding derived from a lexical forward dep.
5;
6;   for (unsigned i = 0; i < 100; i++) {
7;     A[i+1] = B[i] + 2;
8;     C[i] = A[i] * 2;
9;   }
10
11target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
12
13define void @f(i32* %A, i32* %B, i32* %C, i64 %N) {
14
15; CHECK:   for.body.lver.check:
16; CHECK:     %found.conflict{{.*}} =
17; CHECK-NOT: %found.conflict{{.*}} =
18
19entry:
20; Make sure the hoisted load keeps the alignment
21; CHECK: %load_initial = load i32, i32* %A, align 1
22  br label %for.body
23
24for.body:                                         ; preds = %for.body, %entry
25; CHECK: %store_forwarded = phi i32 [ %load_initial, %for.body.ph ], [ %a_p1, %for.body ]
26  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
27  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
28
29  %Aidx_next = getelementptr inbounds i32, i32* %A, i64 %indvars.iv.next
30  %Bidx = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
31  %Cidx = getelementptr inbounds i32, i32* %C, i64 %indvars.iv
32  %Aidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
33
34  %b = load i32, i32* %Bidx, align 4
35  %a_p1 = add i32 %b, 2
36  store i32 %a_p1, i32* %Aidx_next, align 4
37
38  %a = load i32, i32* %Aidx, align 1
39; CHECK: %c = mul i32 %store_forwarded, 2
40  %c = mul i32 %a, 2
41  store i32 %c, i32* %Cidx, align 4
42
43  %exitcond = icmp eq i64 %indvars.iv.next, %N
44  br i1 %exitcond, label %for.end, label %for.body
45
46for.end:                                          ; preds = %for.body
47  ret void
48}
49