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 ; (unscaled candidates == 2) we start getting some unswitches in outer loops,
28 ; which leads to siblings multiplier kicking in.
30 ; The tests below also run licm, because it is needed to hoist out
31 ; loop-invariant freeze instructions, which otherwise may block further
34 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
35 ; RUN: -unswitch-num-initial-unscaled-candidates=3 -unswitch-siblings-toplevel-div=1 \
36 ; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
37 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE3-DIV1
39 ; NB: sort -b is essential here and below, otherwise blanks might lead to different
40 ; order depending on locale.
42 ; RUN: opt < %s -enable-unswitch-cost-multiplier=true \
43 ; RUN: -unswitch-num-initial-unscaled-candidates=3 -unswitch-siblings-toplevel-div=2 \
44 ; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
45 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-UNSCALE3-DIV2
47 ; With disabled cost-multiplier we get maximal possible amount of unswitches.
49 ; RUN: opt < %s -enable-unswitch-cost-multiplier=false \
50 ; RUN: -passes='loop-mssa(licm,simple-loop-unswitch<nontrivial>),print<loops>' -disable-output 2>&1 | \
51 ; RUN: sort -b -k 1 | FileCheck %s --check-prefixes=LOOP-MAX
53 ; Single loop nest, not unswitched
54 ; LOOP1: Loop at depth 1 containing:
55 ; LOOP1-NOT: Loop at depth 1 containing:
56 ; LOOP1: Loop at depth 2 containing:
57 ; LOOP1-NOT: Loop at depth 2 containing:
58 ; LOOP1: Loop at depth 3 containing:
59 ; LOOP1-NOT: Loop at depth 3 containing:
61 ; Half unswitched loop nests, with unscaled3 and div1 it gets less depth1 loops unswitched
62 ; since they have more cost.
63 ; LOOP-UNSCALE3-DIV1-COUNT-4: Loop at depth 1 containing:
64 ; LOOP-UNSCALE3-DIV1-NOT: Loop at depth 1 containing:
65 ; LOOP-UNSCALE3-DIV1-COUNT-1: Loop at depth 2 containing:
66 ; LOOP-UNSCALE3-DIV1-NOT: Loop at depth 2 containing:
67 ; LOOP-UNSCALE3-DIV1-COUNT-1: Loop at depth 3 containing:
68 ; LOOP-UNSCALE3-DIV1-NOT: Loop at depth 3 containing:
70 ; Half unswitched loop nests, with unscaled3 and div2 it gets more depth1 loops unswitched
72 ; LOOP-UNSCALE3-DIV2-COUNT-6: Loop at depth 1 containing:
73 ; LOOP-UNSCALE3-DIV2-NOT: Loop at depth 1 containing:
74 ; LOOP-UNSCALE3-DIV2-COUNT-1: Loop at depth 2 containing:
75 ; LOOP-UNSCALE3-DIV2-NOT: Loop at depth 2 containing:
76 ; LOOP-UNSCALE3-DIV2-COUNT-1: Loop at depth 3 containing:
77 ; LOOP-UNSCALE3-DIV2-NOT: Loop at depth 3 containing:
79 ; Maximally unswitched (copy of the outer loop per each condition)
80 ; LOOP-MAX-COUNT-6: Loop at depth 1 containing:
81 ; LOOP-MAX-NOT: Loop at depth 1 containing:
82 ; LOOP-MAX-COUNT-1: Loop at depth 2 containing:
83 ; LOOP-MAX-NOT: Loop at depth 2 containing:
84 ; LOOP-MAX-COUNT-1: Loop at depth 3 containing:
85 ; LOOP-MAX-NOT: Loop at depth 3 containing:
89 define void @loop_nested3_conds5(ptr %addr, i1 %c1, i1 %c2, i1 %c3, i1 %c4, i1 %c5) {
91 %addr2 = getelementptr i32, ptr %addr, i64 1
92 %addr3 = getelementptr i32, ptr %addr, i64 2
95 %iv1 = phi i32 [0, %entry], [%iv1.next, %outer_latch]
96 %iv1.next = add i32 %iv1, 1
97 ;; skip nontrivial unswitch
101 %iv2 = phi i32 [0, %outer], [%iv2.next, %middle_latch]
102 %iv2.next = add i32 %iv2, 1
103 ;; skip nontrivial unswitch
107 %iv3 = phi i32 [0, %middle], [%iv3.next, %loop_latch]
108 %iv3.next = add i32 %iv3, 1
109 ;; skip nontrivial unswitch
111 br i1 %c1, label %loop_next1_left, label %outer_latch
118 br i1 %c2, label %loop_next2_left, label %outer_latch
125 br i1 %c3, label %loop_next3_left, label %outer_latch
132 br i1 %c4, label %loop_next4_left, label %outer_latch
139 br i1 %c5, label %loop_latch_left, label %outer_latch
146 store volatile i32 0, ptr %addr
147 %test_loop = icmp slt i32 %iv3, 50
148 br i1 %test_loop, label %loop, label %middle_latch
150 store volatile i32 0, ptr %addr2
151 %test_middle = icmp slt i32 %iv2, 50
152 br i1 %test_middle, label %middle, label %outer_latch
154 store volatile i32 0, ptr %addr3
155 %test_outer = icmp slt i32 %iv1, 50
156 br i1 %test_outer, label %outer, label %exit