1;
2; There should be just a single copy of each loop when strictest mutiplier
3; candidates formula (unscaled candidates == 0) is enforced:
4
5; RUN: opt < %s -enable-nontrivial-unswitch -enable-unswitch-cost-multiplier=true \
6; RUN:     -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=1 \
7; RUN:     -passes='loop(unswitch),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
8;
9; RUN: opt < %s -enable-nontrivial-unswitch -enable-unswitch-cost-multiplier=true \
10; RUN:     -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=16 \
11; RUN:     -passes='loop(unswitch),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
12;
13;
14; When we relax the candidates part of a multiplier formula
15; (unscaled candidates == 4) we start getting  some unswitches,
16; which leads to siblings multiplier kicking in.
17;
18; RUN: opt < %s -enable-nontrivial-unswitch -enable-unswitch-cost-multiplier=true \
19; RUN:     -unswitch-num-initial-unscaled-candidates=4 -unswitch-siblings-toplevel-div=1 \
20; RUN:     -passes='loop(unswitch),print<loops>' -disable-output 2>&1 | \
21; RUN:     sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE4-DIV1
22;
23; NB: sort -b is essential here and below, otherwise blanks might lead to different
24; order depending on locale.
25;
26; RUN: opt < %s -enable-nontrivial-unswitch -enable-unswitch-cost-multiplier=true \
27; RUN:     -unswitch-num-initial-unscaled-candidates=4 -unswitch-siblings-toplevel-div=2 \
28; RUN:     -passes='loop(unswitch),print<loops>' -disable-output 2>&1 | \
29; RUN:     sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE4-DIV2
30;
31;
32; Get
33;    2^(num conds) == 2^5 = 32
34; loop nests when cost multiplier is disabled:
35;
36; RUN: opt < %s -enable-nontrivial-unswitch -enable-unswitch-cost-multiplier=false \
37; RUN:     -passes='loop(unswitch),print<loops>' -disable-output 2>&1 | \
38; RUN:	   sort -b -k 1 | FileCheck %s --check-prefixes=LOOP32
39;
40; Single loop nest, not unswitched
41; LOOP1:     Loop at depth 1 containing:
42; LOOP1:     Loop at depth 2 containing:
43; LOOP1:     Loop at depth 3 containing:
44; LOOP1-NOT: Loop at depth {{[0-9]+}} containing:
45;
46; Half unswitched loop nests, with unscaled4 and div1 it gets less depth1 loops unswitched
47; since they have more cost.
48; LOOP-UNSCALE4-DIV1-COUNT-6: Loop at depth 1 containing:
49; LOOP-UNSCALE4-DIV1-COUNT-19: Loop at depth 2 containing:
50; LOOP-UNSCALE4-DIV1-COUNT-29: Loop at depth 3 containing:
51; LOOP-UNSCALE4-DIV1-NOT:      Loop at depth {{[0-9]+}} containing:
52;
53; Half unswitched loop nests, with unscaled4 and div2 it gets more depth1 loops unswitched
54; as div2 kicks in.
55; LOOP-UNSCALE4-DIV2-COUNT-11: Loop at depth 1 containing:
56; LOOP-UNSCALE4-DIV2-COUNT-22: Loop at depth 2 containing:
57; LOOP-UNSCALE4-DIV2-COUNT-29: Loop at depth 3 containing:
58; LOOP-UNSCALE4-DIV2-NOT:      Loop at depth {{[0-9]+}} containing:
59;
60; 32 loop nests, fully unswitched
61; LOOP32-COUNT-32: Loop at depth 1 containing:
62; LOOP32-COUNT-32: Loop at depth 2 containing:
63; LOOP32-COUNT-32: Loop at depth 3 containing:
64; LOOP32-NOT:      Loop at depth {{[0-9]+}} containing:
65
66declare void @bar()
67
68define void @loop_nested3_conds5(i32* %addr, i1 %c1, i1 %c2, i1 %c3, i1 %c4, i1 %c5) {
69entry:
70  %addr1 = getelementptr i32, i32* %addr, i64 0
71  %addr2 = getelementptr i32, i32* %addr, i64 1
72  %addr3 = getelementptr i32, i32* %addr, i64 2
73  br label %outer
74outer:
75  %iv1 = phi i32 [0, %entry], [%iv1.next, %outer_latch]
76  %iv1.next = add i32 %iv1, 1
77  ;; skip nontrivial unswitch
78  call void @bar()
79  br label %middle
80middle:
81  %iv2 = phi i32 [0, %outer], [%iv2.next, %middle_latch]
82  %iv2.next = add i32 %iv2, 1
83  ;; skip nontrivial unswitch
84  call void @bar()
85  br label %loop
86loop:
87  %iv3 = phi i32 [0, %middle], [%iv3.next, %loop_latch]
88  %iv3.next = add i32 %iv3, 1
89  ;; skip nontrivial unswitch
90  call void @bar()
91  br i1 %c1, label %loop_next1_left, label %loop_next1_right
92loop_next1_left:
93  br label %loop_next1
94loop_next1_right:
95  br label %loop_next1
96
97loop_next1:
98  br i1 %c2, label %loop_next2_left, label %loop_next2_right
99loop_next2_left:
100  br label %loop_next2
101loop_next2_right:
102  br label %loop_next2
103
104loop_next2:
105  br i1 %c3, label %loop_next3_left, label %loop_next3_right
106loop_next3_left:
107  br label %loop_next3
108loop_next3_right:
109  br label %loop_next3
110
111loop_next3:
112  br i1 %c4, label %loop_next4_left, label %loop_next4_right
113loop_next4_left:
114  br label %loop_next4
115loop_next4_right:
116  br label %loop_next4
117
118loop_next4:
119  br i1 %c5, label %loop_latch_left, label %loop_latch_right
120loop_latch_left:
121  br label %loop_latch
122loop_latch_right:
123  br label %loop_latch
124
125loop_latch:
126  store volatile i32 0, i32* %addr1
127  %test_loop = icmp slt i32 %iv3, 50
128  br i1 %test_loop, label %loop, label %middle_latch
129middle_latch:
130  store volatile i32 0, i32* %addr2
131  %test_middle = icmp slt i32 %iv2, 50
132  br i1 %test_middle, label %middle, label %outer_latch
133outer_latch:
134  store volatile i32 0, i32* %addr3
135  %test_outer = icmp slt i32 %iv1, 50
136  br i1 %test_outer, label %outer, label %exit
137exit:
138  ret void
139}
140