xref: /llvm-project/llvm/test/Transforms/LoopStrengthReduce/RISCV/lsr-cost-compare.ll (revision 27a62ec72aed3faf1388600f485552471b580e3b)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=loop-reduce,loop-term-fold -S | FileCheck %s
3
4target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n64-S128"
5target triple = "riscv64"
6
7; a[] = 1.0
8define void @test1(ptr %a) {
9; CHECK-LABEL: @test1(
10; CHECK-NEXT:  entry:
11; CHECK-NEXT:    [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 128000
12; CHECK-NEXT:    br label [[LOOP:%.*]]
13; CHECK:       loop:
14; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], [[LOOP]] ], [ [[A]], [[ENTRY:%.*]] ]
15; CHECK-NEXT:    store float 1.000000e+00, ptr [[LSR_IV1]], align 4
16; CHECK-NEXT:    [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i64 4
17; CHECK-NEXT:    [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND:%.*]] = icmp eq ptr [[SCEVGEP]], [[SCEVGEP2]]
18; CHECK-NEXT:    br i1 [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND]], label [[EXIT:%.*]], label [[LOOP]]
19; CHECK:       exit:
20; CHECK-NEXT:    ret void
21;
22entry:
23  br label %loop
24
25loop:                                             ; preds = %loop, %entry
26  %t15 = phi i64 [ 0, %entry ], [ %t20, %loop ]
27  %t19 = getelementptr inbounds [32000 x float], ptr %a, i64 0, i64 %t15
28  store float 1.0, ptr %t19, align 4
29  %t20 = add nuw nsw i64 %t15, 1
30  %t21 = icmp eq i64 %t20, 32000
31  br i1 %t21, label %exit, label %loop
32
33exit:                                             ; preds = %loop
34  ret void
35}
36
37; Same as test1, but with a use of a added outside the loop
38define void @test2(ptr %a) {
39; CHECK-LABEL: @test2(
40; CHECK-NEXT:  entry:
41; CHECK-NEXT:    [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A:%.*]], i64 128000
42; CHECK-NEXT:    br label [[LOOP:%.*]]
43; CHECK:       loop:
44; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], [[LOOP]] ], [ [[A]], [[ENTRY:%.*]] ]
45; CHECK-NEXT:    store float 1.000000e+00, ptr [[LSR_IV1]], align 4
46; CHECK-NEXT:    [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i64 4
47; CHECK-NEXT:    [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND:%.*]] = icmp eq ptr [[SCEVGEP]], [[SCEVGEP2]]
48; CHECK-NEXT:    br i1 [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND]], label [[EXIT:%.*]], label [[LOOP]]
49; CHECK:       exit:
50; CHECK-NEXT:    call void @use(ptr [[A]])
51; CHECK-NEXT:    ret void
52;
53entry:
54  br label %loop
55
56loop:                                             ; preds = %loop, %entry
57  %t15 = phi i64 [ 0, %entry ], [ %t20, %loop ]
58  %t19 = getelementptr inbounds [32000 x float], ptr %a, i64 0, i64 %t15
59  store float 1.0, ptr %t19, align 4
60  %t20 = add nuw nsw i64 %t15, 1
61  %t21 = icmp eq i64 %t20, 32000
62  br i1 %t21, label %exit, label %loop
63
64exit:                                             ; preds = %loop
65  call void @use(ptr %a)
66  ret void
67}
68
69; b[] = a[] + 1.0
70define void @test3(ptr %a, ptr %b) {
71; CHECK-LABEL: @test3(
72; CHECK-NEXT:  entry:
73; CHECK-NEXT:    [[SCEVGEP4:%.*]] = getelementptr i8, ptr [[B:%.*]], i64 128000
74; CHECK-NEXT:    br label [[LOOP:%.*]]
75; CHECK:       loop:
76; CHECK-NEXT:    [[LSR_IV2:%.*]] = phi ptr [ [[SCEVGEP3:%.*]], [[LOOP]] ], [ [[A:%.*]], [[ENTRY:%.*]] ]
77; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], [[LOOP]] ], [ [[B]], [[ENTRY]] ]
78; CHECK-NEXT:    [[T17:%.*]] = load float, ptr [[LSR_IV2]], align 4
79; CHECK-NEXT:    [[T18:%.*]] = fadd float [[T17]], 1.000000e+00
80; CHECK-NEXT:    store float [[T18]], ptr [[LSR_IV1]], align 4
81; CHECK-NEXT:    [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i64 4
82; CHECK-NEXT:    [[SCEVGEP3]] = getelementptr i8, ptr [[LSR_IV2]], i64 4
83; CHECK-NEXT:    [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND:%.*]] = icmp eq ptr [[SCEVGEP]], [[SCEVGEP4]]
84; CHECK-NEXT:    br i1 [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND]], label [[EXIT:%.*]], label [[LOOP]]
85; CHECK:       exit:
86; CHECK-NEXT:    ret void
87;
88entry:
89  br label %loop
90
91loop:                                             ; preds = %loop, %entry
92  %t15 = phi i64 [ 0, %entry ], [ %t20, %loop ]
93  %t16 = getelementptr inbounds [32000 x float], ptr %a, i64 0, i64 %t15
94  %t17 = load float, ptr %t16, align 4
95  %t18 = fadd float %t17, 1.000000e+00
96  %t19 = getelementptr inbounds [32000 x float], ptr %b, i64 0, i64 %t15
97  store float %t18, ptr %t19, align 4
98  %t20 = add nuw nsw i64 %t15, 1
99  %t21 = icmp eq i64 %t20, 32000
100  br i1 %t21, label %exit, label %loop
101
102exit:                                             ; preds = %loop
103  ret void
104}
105
106; Same as test3, but with a use of both a and b outside the loop
107define void @test4(ptr %a, ptr %b) {
108; CHECK-LABEL: @test4(
109; CHECK-NEXT:  entry:
110; CHECK-NEXT:    [[SCEVGEP4:%.*]] = getelementptr i8, ptr [[B:%.*]], i64 128000
111; CHECK-NEXT:    br label [[LOOP:%.*]]
112; CHECK:       loop:
113; CHECK-NEXT:    [[LSR_IV2:%.*]] = phi ptr [ [[SCEVGEP3:%.*]], [[LOOP]] ], [ [[A:%.*]], [[ENTRY:%.*]] ]
114; CHECK-NEXT:    [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], [[LOOP]] ], [ [[B]], [[ENTRY]] ]
115; CHECK-NEXT:    [[T17:%.*]] = load float, ptr [[LSR_IV2]], align 4
116; CHECK-NEXT:    [[T18:%.*]] = fadd float [[T17]], 1.000000e+00
117; CHECK-NEXT:    store float [[T18]], ptr [[LSR_IV1]], align 4
118; CHECK-NEXT:    [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i64 4
119; CHECK-NEXT:    [[SCEVGEP3]] = getelementptr i8, ptr [[LSR_IV2]], i64 4
120; CHECK-NEXT:    [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND:%.*]] = icmp eq ptr [[SCEVGEP]], [[SCEVGEP4]]
121; CHECK-NEXT:    br i1 [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND]], label [[EXIT:%.*]], label [[LOOP]]
122; CHECK:       exit:
123; CHECK-NEXT:    call void @use(ptr [[A]])
124; CHECK-NEXT:    call void @use(ptr [[B]])
125; CHECK-NEXT:    ret void
126;
127entry:
128  br label %loop
129
130loop:                                             ; preds = %loop, %entry
131  %t15 = phi i64 [ 0, %entry ], [ %t20, %loop ]
132  %t16 = getelementptr inbounds [32000 x float], ptr %a, i64 0, i64 %t15
133  %t17 = load float, ptr %t16, align 4
134  %t18 = fadd float %t17, 1.000000e+00
135  %t19 = getelementptr inbounds [32000 x float], ptr %b, i64 0, i64 %t15
136  store float %t18, ptr %t19, align 4
137  %t20 = add nuw nsw i64 %t15, 1
138  %t21 = icmp eq i64 %t20, 32000
139  br i1 %t21, label %exit, label %loop
140
141exit:                                             ; preds = %loop
142  call void @use(ptr %a)
143  call void @use(ptr %b)
144  ret void
145}
146
147declare void @use(ptr)
148
149