2 ; There should be just a single copy of each loop when strictest mutiplier
3 ; candidates formula (unscaled candidates == 0) is enforced:
5 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
6 ; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=1 \
7 ; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
9 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
10 ; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=16 \
11 ; RUN: -passes='loop(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
13 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
14 ; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=1 \
15 ; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
17 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
18 ; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=16 \
19 ; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
21 ; When we relax the candidates part of a multiplier formula
22 ; (unscaled candidates == 4) we start getting some unswitches,
23 ; which leads to siblings multiplier kicking in.
25 ; The tests below also run licm, because it is needed to hoist out
26 ; loop-invariant freeze instructions, which otherwise may block further
29 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
30 ; RUN: -unswitch-num-initial-unscaled-candidates=4 -unswitch-siblings-toplevel-div=1 \
31 ; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
32 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE4-DIV1
34 ; NB: sort -b is essential here and below, otherwise blanks might lead to different
35 ; order depending on locale.
37 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
38 ; RUN: -unswitch-num-initial-unscaled-candidates=4 -unswitch-siblings-toplevel-div=2 \
39 ; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
40 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE4-DIV2
43 ; 2^(num conds) == 2^5 = 32
44 ; loop nests when cost multiplier is disabled:
46 ; RUN: opt < %s -enable-unswitch-cost-multiplier=false \
47 ; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
48 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP32
50 ; Single loop nest, not unswitched
51 ; LOOP1: Loop at depth 1 containing:
52 ; LOOP1: Loop at depth 2 containing:
53 ; LOOP1: Loop at depth 3 containing:
54 ; LOOP1-NOT: Loop at depth {{[0-9]+}} containing:
56 ; Half unswitched loop nests, with unscaled4 and div1 it gets less depth1 loops unswitched
57 ; since they have more cost.
58 ; LOOP-UNSCALE4-DIV1-COUNT-6: Loop at depth 1 containing:
59 ; LOOP-UNSCALE4-DIV1-COUNT-19: Loop at depth 2 containing:
60 ; LOOP-UNSCALE4-DIV1-COUNT-29: Loop at depth 3 containing:
61 ; LOOP-UNSCALE4-DIV1-NOT: Loop at depth {{[0-9]+}} containing:
63 ; Half unswitched loop nests, with unscaled4 and div2 it gets more depth1 loops unswitched
65 ; LOOP-UNSCALE4-DIV2-COUNT-11: Loop at depth 1 containing:
66 ; LOOP-UNSCALE4-DIV2-COUNT-22: Loop at depth 2 containing:
67 ; LOOP-UNSCALE4-DIV2-COUNT-29: Loop at depth 3 containing:
68 ; LOOP-UNSCALE4-DIV2-NOT: Loop at depth {{[0-9]+}} containing:
70 ; 32 loop nests, fully unswitched
71 ; LOOP32-COUNT-32: Loop at depth 1 containing:
72 ; LOOP32-COUNT-32: Loop at depth 2 containing:
73 ; LOOP32-COUNT-32: Loop at depth 3 containing:
74 ; LOOP32-NOT: Loop at depth {{[0-9]+}} containing:
78 define void @loop_nested3_conds5(ptr %addr, i1 %c1, i1 %c2, i1 %c3, i1 %c4, i1 %c5) {
80 %addr2 = getelementptr i32, ptr %addr, i64 1
81 %addr3 = getelementptr i32, ptr %addr, i64 2
84 %iv1 = phi i32 [0, %entry], [%iv1.next, %outer_latch]
85 %iv1.next = add i32 %iv1, 1
86 ;; skip nontrivial unswitch
90 %iv2 = phi i32 [0, %outer], [%iv2.next, %middle_latch]
91 %iv2.next = add i32 %iv2, 1
92 ;; skip nontrivial unswitch
96 %iv3 = phi i32 [0, %middle], [%iv3.next, %loop_latch]
97 %iv3.next = add i32 %iv3, 1
98 ;; skip nontrivial unswitch
100 br i1 %c1, label %loop_next1_left, label %loop_next1_right
107 br i1 %c2, label %loop_next2_left, label %loop_next2_right
114 br i1 %c3, label %loop_next3_left, label %loop_next3_right
121 br i1 %c4, label %loop_next4_left, label %loop_next4_right
128 br i1 %c5, label %loop_latch_left, label %loop_latch_right
135 store volatile i32 0, ptr %addr
136 %test_loop = icmp slt i32 %iv3, 50
137 br i1 %test_loop, label %loop, label %middle_latch
139 store volatile i32 0, ptr %addr2
140 %test_middle = icmp slt i32 %iv2, 50
141 br i1 %test_middle, label %middle, label %outer_latch
143 store volatile i32 0, ptr %addr3
144 %test_outer = icmp slt i32 %iv1, 50
145 br i1 %test_outer, label %outer, label %exit