2 ; Here all the branches we unswitch are exiting from the inner loop.
3 ; That means we should not be getting exponential behavior on inner-loop
4 ; unswitch. In fact there should be just a single version of inner-loop,
5 ; with possibly some outer loop copies.
7 ; There should be just a single copy of each loop when strictest mutiplier
8 ; candidates formula (unscaled candidates == 0) is enforced:
10 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
11 ; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=1 \
12 ; RUN: -passes='loop(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
14 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
15 ; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=16 \
16 ; RUN: -passes='loop(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
18 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
19 ; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=1 \
20 ; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
22 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
23 ; RUN: -unswitch-num-initial-unscaled-candidates=0 -unswitch-siblings-toplevel-div=16 \
24 ; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | FileCheck %s --check-prefixes=LOOP1
26 ; When we relax the candidates part of a multiplier formula
27 ; When we relax the candidates part of a multiplier formula
28 ; (unscaled candidates == 2) we start getting some unswitches in outer loops,
29 ; which leads to siblings multiplier kicking in.
31 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
32 ; RUN: -unswitch-num-initial-unscaled-candidates=3 -unswitch-siblings-toplevel-div=1 \
33 ; RUN: -passes='loop(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
34 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE3-DIV1
36 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
37 ; RUN: -unswitch-num-initial-unscaled-candidates=3 -unswitch-siblings-toplevel-div=1 \
38 ; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
39 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE3-DIV1
41 ; NB: sort -b is essential here and below, otherwise blanks might lead to different
42 ; order depending on locale.
44 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
45 ; RUN: -unswitch-num-initial-unscaled-candidates=3 -unswitch-siblings-toplevel-div=2 \
46 ; RUN: -passes='loop(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
47 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE3-DIV2
49 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
50 ; RUN: -unswitch-num-initial-unscaled-candidates=3 -unswitch-siblings-toplevel-div=2 \
51 ; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
52 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE3-DIV2
54 ; With disabled cost-multiplier we get maximal possible amount of unswitches.
56 ; RUN: opt < %s -enable-unswitch-cost-multiplier=false \
57 ; RUN: -passes='loop(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
58 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-MAX
60 ; RUN: opt < %s -enable-unswitch-cost-multiplier=false \
61 ; RUN: -passes='loop-mssa(simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
62 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-MAX
64 ; Single loop nest, not unswitched
65 ; LOOP1: Loop at depth 1 containing:
66 ; LOOP1-NOT: Loop at depth 1 containing:
67 ; LOOP1: Loop at depth 2 containing:
68 ; LOOP1-NOT: Loop at depth 2 containing:
69 ; LOOP1: Loop at depth 3 containing:
70 ; LOOP1-NOT: Loop at depth 3 containing:
72 ; Half unswitched loop nests, with unscaled3 and div1 it gets less depth1 loops unswitched
73 ; since they have more cost.
74 ; LOOP-UNSCALE3-DIV1-COUNT-4: Loop at depth 1 containing:
75 ; LOOP-UNSCALE3-DIV1-NOT: Loop at depth 1 containing:
76 ; LOOP-UNSCALE3-DIV1-COUNT-1: Loop at depth 2 containing:
77 ; LOOP-UNSCALE3-DIV1-NOT: Loop at depth 2 containing:
78 ; LOOP-UNSCALE3-DIV1-COUNT-1: Loop at depth 3 containing:
79 ; LOOP-UNSCALE3-DIV1-NOT: Loop at depth 3 containing:
81 ; Half unswitched loop nests, with unscaled3 and div2 it gets more depth1 loops unswitched
83 ; LOOP-UNSCALE3-DIV2-COUNT-6: Loop at depth 1 containing:
84 ; LOOP-UNSCALE3-DIV2-NOT: Loop at depth 1 containing:
85 ; LOOP-UNSCALE3-DIV2-COUNT-1: Loop at depth 2 containing:
86 ; LOOP-UNSCALE3-DIV2-NOT: Loop at depth 2 containing:
87 ; LOOP-UNSCALE3-DIV2-COUNT-1: Loop at depth 3 containing:
88 ; LOOP-UNSCALE3-DIV2-NOT: Loop at depth 3 containing:
90 ; Maximally unswitched (copy of the outer loop per each condition)
91 ; LOOP-MAX-COUNT-6: Loop at depth 1 containing:
92 ; LOOP-MAX-NOT: Loop at depth 1 containing:
93 ; LOOP-MAX-COUNT-1: Loop at depth 2 containing:
94 ; LOOP-MAX-NOT: Loop at depth 2 containing:
95 ; LOOP-MAX-COUNT-1: Loop at depth 3 containing:
96 ; LOOP-MAX-NOT: Loop at depth 3 containing:
100 define void @loop_nested3_conds5(i32* %addr, i1 %c1, i1 %c2, i1 %c3, i1 %c4, i1 %c5) {
102 %addr1 = getelementptr i32, i32* %addr, i64 0
103 %addr2 = getelementptr i32, i32* %addr, i64 1
104 %addr3 = getelementptr i32, i32* %addr, i64 2
107 %iv1 = phi i32 [0, %entry], [%iv1.next, %outer_latch]
108 %iv1.next = add i32 %iv1, 1
109 ;; skip nontrivial unswitch
113 %iv2 = phi i32 [0, %outer], [%iv2.next, %middle_latch]
114 %iv2.next = add i32 %iv2, 1
115 ;; skip nontrivial unswitch
119 %iv3 = phi i32 [0, %middle], [%iv3.next, %loop_latch]
120 %iv3.next = add i32 %iv3, 1
121 ;; skip nontrivial unswitch
123 br i1 %c1, label %loop_next1_left, label %outer_latch
130 br i1 %c2, label %loop_next2_left, label %outer_latch
137 br i1 %c3, label %loop_next3_left, label %outer_latch
144 br i1 %c4, label %loop_next4_left, label %outer_latch
151 br i1 %c5, label %loop_latch_left, label %outer_latch
158 store volatile i32 0, i32* %addr1
159 %test_loop = icmp slt i32 %iv3, 50
160 br i1 %test_loop, label %loop, label %middle_latch
162 store volatile i32 0, i32* %addr2
163 %test_middle = icmp slt i32 %iv2, 50
164 br i1 %test_middle, label %middle, label %outer_latch
166 store volatile i32 0, i32* %addr3
167 %test_outer = icmp slt i32 %iv1, 50
168 br i1 %test_outer, label %outer, label %exit