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, but
5; can't introduce runtime checks with a convergent call.
6;
7;   for (unsigned i = 0; i < 100; i++) {
8;     A[i+1] = convergent(B[i]) + 2;
9;     C[i] = A[i] * 2;
10;   }
11
12target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
13
14; CHECK-LABEL: @f_convergent(
15; CHECK: call i32 @llvm.convergent(
16; CHECK-NOT: call i32 @llvm.convergent(
17define void @f_convergent(i32* %A, i32* %B, i32* %C, i64 %N) #0 {
18
19entry:
20  br label %for.body
21
22for.body:                                         ; preds = %for.body, %entry
23  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
24  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
25
26  %Aidx_next = getelementptr inbounds i32, i32* %A, i64 %indvars.iv.next
27  %Bidx = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
28  %Cidx = getelementptr inbounds i32, i32* %C, i64 %indvars.iv
29  %Aidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
30
31  %b = load i32, i32* %Bidx, align 4
32  %convergent.b = call i32 @llvm.convergent(i32 %b)
33
34  %a_p1 = add i32 %convergent.b, 2
35  store i32 %a_p1, i32* %Aidx_next, align 4
36
37  %a = load i32, i32* %Aidx, align 1
38  %c = mul i32 %a, 2
39  store i32 %c, i32* %Cidx, align 4
40
41  %exitcond = icmp eq i64 %indvars.iv.next, %N
42  br i1 %exitcond, label %for.end, label %for.body
43
44for.end:                                          ; preds = %for.body
45  ret void
46}
47
48declare i32 @llvm.convergent(i32) #1
49
50attributes #0 = { nounwind convergent }
51attributes #1 = { nounwind readnone convergent }
52