xref: /llvm-project/llvm/test/Transforms/LoopStrengthReduce/X86/sibling-loops.ll (revision 3d7242f05e6ebae394fe47039abb0568083209a0)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -loop-reduce -S < %s | FileCheck %s
3; We find it is very bad to allow LSR formula containing SCEVAddRecExpr Reg
4; from siblings of current loop. When one loop is LSR optimized, it can
5; insert lsr.iv for other sibling loops, which sometimes leads to many extra
6; lsr.iv inserted for loops.
7
8target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
9
10@cond = common local_unnamed_addr global i64 0, align 8
11
12; Check there is no extra lsr.iv generated in foo.
13define void @foo(i64 %N) local_unnamed_addr {
14; CHECK-LABEL: @foo(
15; CHECK-NEXT:  entry:
16; CHECK-NEXT:    br label [[DO_BODY:%.*]]
17; CHECK:       do.body:
18; CHECK-NEXT:    [[I_0:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[DO_BODY]] ]
19; CHECK-NEXT:    tail call void @goo(i64 [[I_0]], i64 [[I_0]])
20; CHECK-NEXT:    [[INC]] = add nuw i64 [[I_0]], 1
21; CHECK-NEXT:    [[T0:%.*]] = load i64, ptr @cond, align 8
22; CHECK-NEXT:    [[TOBOOL:%.*]] = icmp eq i64 [[T0]], 0
23; CHECK-NEXT:    br i1 [[TOBOOL]], label [[DO_BODY2_PREHEADER:%.*]], label [[DO_BODY]]
24; CHECK:       do.body2.preheader:
25; CHECK-NEXT:    br label [[DO_BODY2:%.*]]
26; CHECK:       do.body2:
27; CHECK-NEXT:    [[I_1:%.*]] = phi i64 [ [[INC3:%.*]], [[DO_BODY2]] ], [ 0, [[DO_BODY2_PREHEADER]] ]
28; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INC]], [[I_1]]
29; CHECK-NEXT:    tail call void @goo(i64 [[I_1]], i64 [[TMP0]])
30; CHECK-NEXT:    [[INC3]] = add nuw i64 [[I_1]], 1
31; CHECK-NEXT:    [[T1:%.*]] = load i64, ptr @cond, align 8
32; CHECK-NEXT:    [[TOBOOL6:%.*]] = icmp eq i64 [[T1]], 0
33; CHECK-NEXT:    br i1 [[TOBOOL6]], label [[DO_BODY8_PREHEADER:%.*]], label [[DO_BODY2]]
34; CHECK:       do.body8.preheader:
35; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INC]], [[INC3]]
36; CHECK-NEXT:    br label [[DO_BODY8:%.*]]
37; CHECK:       do.body8:
38; CHECK-NEXT:    [[I_2:%.*]] = phi i64 [ [[INC9:%.*]], [[DO_BODY8]] ], [ 0, [[DO_BODY8_PREHEADER]] ]
39; CHECK-NEXT:    [[J_2:%.*]] = phi i64 [ [[INC10:%.*]], [[DO_BODY8]] ], [ [[TMP1]], [[DO_BODY8_PREHEADER]] ]
40; CHECK-NEXT:    tail call void @goo(i64 [[I_2]], i64 [[J_2]])
41; CHECK-NEXT:    [[INC9]] = add nuw nsw i64 [[I_2]], 1
42; CHECK-NEXT:    [[INC10]] = add i64 [[J_2]], 1
43; CHECK-NEXT:    [[T2:%.*]] = load i64, ptr @cond, align 8
44; CHECK-NEXT:    [[TOBOOL12:%.*]] = icmp eq i64 [[T2]], 0
45; CHECK-NEXT:    br i1 [[TOBOOL12]], label [[DO_BODY14_PREHEADER:%.*]], label [[DO_BODY8]]
46; CHECK:       do.body14.preheader:
47; CHECK-NEXT:    [[INC10_LCSSA:%.*]] = phi i64 [ [[INC10]], [[DO_BODY8]] ]
48; CHECK-NEXT:    br label [[DO_BODY14:%.*]]
49; CHECK:       do.body14:
50; CHECK-NEXT:    [[I_3:%.*]] = phi i64 [ [[INC15:%.*]], [[DO_BODY14]] ], [ 0, [[DO_BODY14_PREHEADER]] ]
51; CHECK-NEXT:    [[J_3:%.*]] = phi i64 [ [[INC16:%.*]], [[DO_BODY14]] ], [ [[INC10_LCSSA]], [[DO_BODY14_PREHEADER]] ]
52; CHECK-NEXT:    tail call void @goo(i64 [[I_3]], i64 [[J_3]])
53; CHECK-NEXT:    [[INC15]] = add nuw nsw i64 [[I_3]], 1
54; CHECK-NEXT:    [[INC16]] = add i64 [[J_3]], 1
55; CHECK-NEXT:    [[T3:%.*]] = load i64, ptr @cond, align 8
56; CHECK-NEXT:    [[TOBOOL18:%.*]] = icmp eq i64 [[T3]], 0
57; CHECK-NEXT:    br i1 [[TOBOOL18]], label [[DO_BODY20_PREHEADER:%.*]], label [[DO_BODY14]]
58; CHECK:       do.body20.preheader:
59; CHECK-NEXT:    [[INC16_LCSSA:%.*]] = phi i64 [ [[INC16]], [[DO_BODY14]] ]
60; CHECK-NEXT:    br label [[DO_BODY20:%.*]]
61; CHECK:       do.body20:
62; CHECK-NEXT:    [[I_4:%.*]] = phi i64 [ [[INC21:%.*]], [[DO_BODY20]] ], [ 0, [[DO_BODY20_PREHEADER]] ]
63; CHECK-NEXT:    [[J_4:%.*]] = phi i64 [ [[INC22:%.*]], [[DO_BODY20]] ], [ [[INC16_LCSSA]], [[DO_BODY20_PREHEADER]] ]
64; CHECK-NEXT:    tail call void @goo(i64 [[I_4]], i64 [[J_4]])
65; CHECK-NEXT:    [[INC21]] = add nuw nsw i64 [[I_4]], 1
66; CHECK-NEXT:    [[INC22]] = add i64 [[J_4]], 1
67; CHECK-NEXT:    [[T4:%.*]] = load i64, ptr @cond, align 8
68; CHECK-NEXT:    [[TOBOOL24:%.*]] = icmp eq i64 [[T4]], 0
69; CHECK-NEXT:    br i1 [[TOBOOL24]], label [[DO_BODY26_PREHEADER:%.*]], label [[DO_BODY20]]
70; CHECK:       do.body26.preheader:
71; CHECK-NEXT:    [[INC22_LCSSA:%.*]] = phi i64 [ [[INC22]], [[DO_BODY20]] ]
72; CHECK-NEXT:    br label [[DO_BODY26:%.*]]
73; CHECK:       do.body26:
74; CHECK-NEXT:    [[I_5:%.*]] = phi i64 [ [[INC27:%.*]], [[DO_BODY26]] ], [ 0, [[DO_BODY26_PREHEADER]] ]
75; CHECK-NEXT:    [[J_5:%.*]] = phi i64 [ [[INC28:%.*]], [[DO_BODY26]] ], [ [[INC22_LCSSA]], [[DO_BODY26_PREHEADER]] ]
76; CHECK-NEXT:    tail call void @goo(i64 [[I_5]], i64 [[J_5]])
77; CHECK-NEXT:    [[INC27]] = add nuw nsw i64 [[I_5]], 1
78; CHECK-NEXT:    [[INC28]] = add nsw i64 [[J_5]], 1
79; CHECK-NEXT:    [[T5:%.*]] = load i64, ptr @cond, align 8
80; CHECK-NEXT:    [[TOBOOL30:%.*]] = icmp eq i64 [[T5]], 0
81; CHECK-NEXT:    br i1 [[TOBOOL30]], label [[DO_END31:%.*]], label [[DO_BODY26]]
82; CHECK:       do.end31:
83; CHECK-NEXT:    ret void
84;
85entry:
86  br label %do.body
87
88do.body:                                          ; preds = %do.body, %entry
89  %i.0 = phi i64 [ 0, %entry ], [ %inc, %do.body ]
90  tail call void @goo(i64 %i.0, i64 %i.0)
91  %inc = add nuw nsw i64 %i.0, 1
92  %t0 = load i64, ptr @cond, align 8
93  %tobool = icmp eq i64 %t0, 0
94  br i1 %tobool, label %do.body2.preheader, label %do.body
95
96do.body2.preheader:                               ; preds = %do.body
97  br label %do.body2
98
99do.body2:                                         ; preds = %do.body2.preheader, %do.body2
100  %i.1 = phi i64 [ %inc3, %do.body2 ], [ 0, %do.body2.preheader ]
101  %j.1 = phi i64 [ %inc4, %do.body2 ], [ %inc, %do.body2.preheader ]
102  tail call void @goo(i64 %i.1, i64 %j.1)
103  %inc3 = add nuw nsw i64 %i.1, 1
104  %inc4 = add nsw i64 %j.1, 1
105  %t1 = load i64, ptr @cond, align 8
106  %tobool6 = icmp eq i64 %t1, 0
107  br i1 %tobool6, label %do.body8.preheader, label %do.body2
108
109do.body8.preheader:                               ; preds = %do.body2
110  br label %do.body8
111
112do.body8:                                         ; preds = %do.body8.preheader, %do.body8
113  %i.2 = phi i64 [ %inc9, %do.body8 ], [ 0, %do.body8.preheader ]
114  %j.2 = phi i64 [ %inc10, %do.body8 ], [ %inc4, %do.body8.preheader ]
115  tail call void @goo(i64 %i.2, i64 %j.2)
116  %inc9 = add nuw nsw i64 %i.2, 1
117  %inc10 = add nsw i64 %j.2, 1
118  %t2 = load i64, ptr @cond, align 8
119  %tobool12 = icmp eq i64 %t2, 0
120  br i1 %tobool12, label %do.body14.preheader, label %do.body8
121
122do.body14.preheader:                              ; preds = %do.body8
123  br label %do.body14
124
125do.body14:                                        ; preds = %do.body14.preheader, %do.body14
126  %i.3 = phi i64 [ %inc15, %do.body14 ], [ 0, %do.body14.preheader ]
127  %j.3 = phi i64 [ %inc16, %do.body14 ], [ %inc10, %do.body14.preheader ]
128  tail call void @goo(i64 %i.3, i64 %j.3)
129  %inc15 = add nuw nsw i64 %i.3, 1
130  %inc16 = add nsw i64 %j.3, 1
131  %t3 = load i64, ptr @cond, align 8
132  %tobool18 = icmp eq i64 %t3, 0
133  br i1 %tobool18, label %do.body20.preheader, label %do.body14
134
135do.body20.preheader:                              ; preds = %do.body14
136  br label %do.body20
137
138do.body20:                                        ; preds = %do.body20.preheader, %do.body20
139  %i.4 = phi i64 [ %inc21, %do.body20 ], [ 0, %do.body20.preheader ]
140  %j.4 = phi i64 [ %inc22, %do.body20 ], [ %inc16, %do.body20.preheader ]
141  tail call void @goo(i64 %i.4, i64 %j.4)
142  %inc21 = add nuw nsw i64 %i.4, 1
143  %inc22 = add nsw i64 %j.4, 1
144  %t4 = load i64, ptr @cond, align 8
145  %tobool24 = icmp eq i64 %t4, 0
146  br i1 %tobool24, label %do.body26.preheader, label %do.body20
147
148do.body26.preheader:                              ; preds = %do.body20
149  br label %do.body26
150
151do.body26:                                        ; preds = %do.body26.preheader, %do.body26
152  %i.5 = phi i64 [ %inc27, %do.body26 ], [ 0, %do.body26.preheader ]
153  %j.5 = phi i64 [ %inc28, %do.body26 ], [ %inc22, %do.body26.preheader ]
154  tail call void @goo(i64 %i.5, i64 %j.5)
155  %inc27 = add nuw nsw i64 %i.5, 1
156  %inc28 = add nsw i64 %j.5, 1
157  %t5 = load i64, ptr @cond, align 8
158  %tobool30 = icmp eq i64 %t5, 0
159  br i1 %tobool30, label %do.end31, label %do.body26
160
161do.end31:                                         ; preds = %do.body26
162  ret void
163}
164
165declare void @goo(i64, i64) local_unnamed_addr
166
167