1; RUN: opt < %s -S -debug-only=loop-unroll -loop-unroll -unroll-allow-peeling 2>&1 | FileCheck %s
2; REQUIRES: asserts
3
4; Make sure we use the profile information correctly to peel-off 3 iterations
5; from the loop, and update the branch weights for the peeled loop properly.
6
7; CHECK: Loop Unroll: F[basic]
8; CHECK: PEELING loop %for.body with iteration count 3!
9; CHECK: Loop Unroll: F[optsize]
10; CHECK-NOT: PEELING
11
12; CHECK-LABEL: @basic
13; CHECK: br i1 %{{.*}}, label %[[NEXT0:.*]], label %for.cond.for.end_crit_edge, !prof !1
14; CHECK: [[NEXT0]]:
15; CHECK: br i1 %{{.*}}, label %[[NEXT1:.*]], label %for.cond.for.end_crit_edge, !prof !2
16; CHECK: [[NEXT1]]:
17; CHECK: br i1 %{{.*}}, label %[[NEXT2:.*]], label %for.cond.for.end_crit_edge, !prof !3
18; CHECK: [[NEXT2]]:
19; CHECK: br i1 %{{.*}}, label %for.body, label %{{.*}}, !prof !4
20
21define void @basic(i32* %p, i32 %k) #0 !prof !0 {
22entry:
23  %cmp3 = icmp slt i32 0, %k
24  br i1 %cmp3, label %for.body.lr.ph, label %for.end
25
26for.body.lr.ph:                                   ; preds = %entry
27  br label %for.body
28
29for.body:                                         ; preds = %for.body.lr.ph, %for.body
30  %i.05 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
31  %p.addr.04 = phi i32* [ %p, %for.body.lr.ph ], [ %incdec.ptr, %for.body ]
32  %incdec.ptr = getelementptr inbounds i32, i32* %p.addr.04, i32 1
33  store i32 %i.05, i32* %p.addr.04, align 4
34  %inc = add nsw i32 %i.05, 1
35  %cmp = icmp slt i32 %inc, %k
36  br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge, !prof !1
37
38for.cond.for.end_crit_edge:                       ; preds = %for.body
39  br label %for.end
40
41for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry
42  ret void
43}
44
45; We don't want to peel loops when optimizing for size.
46; CHECK-LABEL: @optsize
47; CHECK: for.body.lr.ph:
48; CHECK-NEXT: br label %for.body
49; CHECK: for.body:
50; CHECK-NOT: br
51; CHECK: br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge
52define void @optsize(i32* %p, i32 %k) #1 !prof !0 {
53entry:
54  %cmp3 = icmp slt i32 0, %k
55  br i1 %cmp3, label %for.body.lr.ph, label %for.end
56
57for.body.lr.ph:                                   ; preds = %entry
58  br label %for.body
59
60for.body:                                         ; preds = %for.body.lr.ph, %for.body
61  %i.05 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
62  %p.addr.04 = phi i32* [ %p, %for.body.lr.ph ], [ %incdec.ptr, %for.body ]
63  %incdec.ptr = getelementptr inbounds i32, i32* %p.addr.04, i32 1
64  store i32 %i.05, i32* %p.addr.04, align 4
65  %inc = add nsw i32 %i.05, 1
66  %cmp = icmp slt i32 %inc, %k
67  br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge, !prof !1
68
69for.cond.for.end_crit_edge:                       ; preds = %for.body
70  br label %for.end
71
72for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry
73  ret void
74}
75
76attributes #0 = { nounwind }
77attributes #1 = { nounwind optsize }
78
79!0 = !{!"function_entry_count", i64 1}
80!1 = !{!"branch_weights", i32 3001, i32 1001}
81
82;CHECK: !1 = !{!"branch_weights", i32 900, i32 101}
83;CHECK: !2 = !{!"branch_weights", i32 540, i32 360}
84;CHECK: !3 = !{!"branch_weights", i32 162, i32 378}
85;CHECK: !4 = !{!"branch_weights", i32 1399, i32 162}
86
87