1; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z14 -prefetch-distance=50 \
2; RUN:   -loop-prefetch-writes -stop-after=loop-data-prefetch | FileCheck %s
3;
4; Check that prefetches are emitted in a position that is executed each
5; iteration for each targeted memory instruction. The two stores in %true and
6; %false are within one cache line in memory, so they should get a single
7; prefetch in %for.body.
8;
9; CHECK-LABEL: for.body
10; CHECK: call void @llvm.prefetch.p0i8(i8* {{.*}}, i32 0
11; CHECK: call void @llvm.prefetch.p0i8(i8* {{.*}}, i32 1
12; CHECK-LABEL: true
13; CHECK-LABEL: false
14; CHECK-LABEL: latch
15
16define void @fun(i32* nocapture %Src, i32* nocapture readonly %Dst) {
17entry:
18  br label %for.body
19
20for.body:
21  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next.9, %latch ]
22  %arrayidx = getelementptr inbounds i32, i32* %Dst, i64 %indvars.iv
23  %0 = load i32, i32* %arrayidx, align 4
24  %cmp = icmp sgt i32 %0, 0
25  br i1 %cmp, label %true, label %false
26
27true:
28  %arrayidx2 = getelementptr inbounds i32, i32* %Src, i64 %indvars.iv
29  store i32 %0, i32* %arrayidx2, align 4
30  br label %latch
31
32false:
33  %a = add i64 %indvars.iv, 8
34  %arrayidx3 = getelementptr inbounds i32, i32* %Src, i64 %a
35  store i32 %0, i32* %arrayidx3, align 4
36  br label %latch
37
38latch:
39  %indvars.iv.next.9 = add nuw nsw i64 %indvars.iv, 1600
40  %cmp.9 = icmp ult i64 %indvars.iv.next.9, 11200
41  br i1 %cmp.9, label %for.body, label %for.cond.cleanup
42
43for.cond.cleanup:
44  ret void
45}
46
47