1; REQUIRES: asserts
2; RUN: opt < %s -S -debug-only=loop-unroll -loop-unroll -unroll-runtime -unroll-peel-multi-deopt-exit 2>&1 | FileCheck %s
3; RUN: opt < %s -S -debug-only=loop-unroll -unroll-peel-multi-deopt-exit -passes='require<profile-summary>,function(require<opt-remark-emit>,loop-unroll)' 2>&1 | FileCheck %s
4; RUN: opt < %s -S -debug-only=loop-unroll -unroll-peel-multi-deopt-exit -passes='require<profile-summary>,function(require<opt-remark-emit>,loop-unroll<no-profile-peeling>)' 2>&1 | FileCheck %s --check-prefixes=CHECK-NO-PEEL
5
6; Make sure we use the profile information correctly to peel-off 3 iterations
7; from the loop, and update the branch weights for the peeled loop properly.
8; All side exits to deopt does not change weigths.
9
10; CHECK: Loop Unroll: F[basic]
11; CHECK: PEELING loop %for.body with iteration count 4!
12; CHECK-NO-PEEL-NOT: PEELING loop %for.body
13; CHECK-LABEL: @basic
14; CHECK: br i1 %c, label %{{.*}}, label %side_exit, !prof !15
15; CHECK: br i1 %{{.*}}, label %[[NEXT0:.*]], label %for.cond.for.end_crit_edge, !prof !16
16; CHECK: [[NEXT0]]:
17; CHECK: br i1 %c, label %{{.*}}, label %side_exit, !prof !15
18; CHECK: br i1 %{{.*}}, label %[[NEXT1:.*]], label %for.cond.for.end_crit_edge, !prof !17
19; CHECK: [[NEXT1]]:
20; CHECK: br i1 %c, label %{{.*}}, label %side_exit, !prof !15
21; CHECK: br i1 %{{.*}}, label %[[NEXT2:.*]], label %for.cond.for.end_crit_edge, !prof !18
22; CHECK: [[NEXT2]]:
23; CHECK: br i1 %c, label %{{.*}}, label %side_exit.loopexit, !prof !15
24; CHECK: br i1 %{{.*}}, label %for.body, label %{{.*}}, !prof !19
25
26define i32 @basic(i32* %p, i32 %k, i1 %c) #0 !prof !15 {
27entry:
28  %cmp3 = icmp slt i32 0, %k
29  br i1 %cmp3, label %for.body.lr.ph, label %for.end
30
31for.body.lr.ph:                                   ; preds = %entry
32  br label %for.body
33
34for.body:                                         ; preds = %for.body.lr.ph, %for.body
35  %i.05 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %continue ]
36  %p.addr.04 = phi i32* [ %p, %for.body.lr.ph ], [ %incdec.ptr, %continue ]
37  %incdec.ptr = getelementptr inbounds i32, i32* %p.addr.04, i32 1
38  store i32 %i.05, i32* %p.addr.04, align 4
39  %inc = add nsw i32 %i.05, 1
40  %cmp = icmp slt i32 %inc, %k
41  br i1 %c, label %continue, label %side_exit, !prof !17
42
43continue:
44  br i1 %cmp, label %for.body, label %for.cond.for.end_crit_edge, !prof !16
45
46for.cond.for.end_crit_edge:                       ; preds = %for.body
47  br label %for.end
48
49for.end:                                          ; preds = %for.cond.for.end_crit_edge, %entry
50  %res = phi i32 [ 0, %entry ], [ %inc, %for.cond.for.end_crit_edge ]
51  ret i32 %res
52
53side_exit:
54  %rval = call i32(...) @llvm.experimental.deoptimize.i32() [ "deopt"(i32 %inc) ]
55  ret i32 %rval
56}
57
58declare i32 @llvm.experimental.deoptimize.i32(...)
59
60attributes #0 = { nounwind }
61attributes #1 = { nounwind optsize }
62
63!llvm.module.flags = !{!1}
64
65!1 = !{i32 1, !"ProfileSummary", !2}
66!2 = !{!3, !4, !5, !6, !7, !8, !9, !10}
67!3 = !{!"ProfileFormat", !"InstrProf"}
68!4 = !{!"TotalCount", i64 10}
69!5 = !{!"MaxCount", i64 3}
70!6 = !{!"MaxInternalCount", i64 1}
71!7 = !{!"MaxFunctionCount", i64 3}
72!8 = !{!"NumCounts", i64 2}
73!9 = !{!"NumFunctions", i64 2}
74!10 = !{!"DetailedSummary", !11}
75!11 = !{!12, !13, !14}
76!12 = !{i32 10000, i64 3, i32 2}
77!13 = !{i32 999000, i64 1, i32 10}
78!14 = !{i32 999999, i64 1, i32 10}
79!15 = !{!"function_entry_count", i64 1}
80!16 = !{!"branch_weights", i32 3001, i32 1001}
81!17 = !{!"branch_weights", i32 1, i32 0}
82
83; This is a weights of deopt side-exit.
84;CHECK: !15 = !{!"branch_weights", i32 1, i32 0}
85; This is a weights of latch and its copies.
86;CHECK: !16 = !{!"branch_weights", i32 3001, i32 1001}
87;CHECK: !17 = !{!"branch_weights", i32 2000, i32 1001}
88;CHECK: !18 = !{!"branch_weights", i32 999, i32 1001}
89;CHECK: !19 = !{!"branch_weights", i32 1, i32 1001}
90
91