xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/RISCV/getpointerschaincost.ll (revision 38fffa630ee80163dc65e759392ad29798905679)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
2; RUN: opt -S -mtriple=riscv64 -mattr=+v -riscv-v-slp-max-vf=0 -passes=slp-vectorizer -pass-remarks-output=%t < %s | FileCheck %s
3; RUN: FileCheck --input-file=%t --check-prefix=YAML %s
4
5; Because all of these addresses are foldable, the scalar cost should be 0 when
6; computing the pointers chain cost.
7;
8; TODO: These are currently costed as free the indices are all constants, but we
9; should check if the constants are actually foldable
10define void @f(ptr %dest, i64 %i) {
11; CHECK-LABEL: define void @f
12; CHECK-SAME: (ptr [[DEST:%.*]], i64 [[I:%.*]]) #[[ATTR0:[0-9]+]] {
13; CHECK-NEXT:  entry:
14; CHECK-NEXT:    [[P1:%.*]] = getelementptr i32, ptr [[DEST]], i32 0
15; CHECK-NEXT:    store <4 x i32> splat (i32 1), ptr [[P1]], align 4
16; CHECK-NEXT:    ret void
17;
18entry:
19; YAML:      Pass:            slp-vectorizer
20; YAML-NEXT: Name:            StoresVectorized
21; YAML-NEXT: Function:        f
22; YAML-NEXT: Args:
23; YAML-NEXT:   - String:          'Stores SLP vectorized with cost '
24; YAML-NEXT:   - Cost:            '-2'
25; YAML-NEXT:   - String:          ' and with tree size '
26; YAML-NEXT:   - TreeSize:        '2'
27  %p1 = getelementptr i32, ptr %dest, i32 0
28  store i32 1, ptr %p1
29  %p2 = getelementptr i32, ptr %dest, i32 1
30  store i32 1, ptr %p2
31  %p3 = getelementptr i32, ptr %dest, i32 2
32  store i32 1, ptr %p3
33  %p4 = getelementptr i32, ptr %dest, i32 3
34  store i32 1, ptr %p4
35  ret void
36}
37
38; When computing the scalar pointers chain cost here, there is a cost of 1 for
39; the base pointer, and the rest can be folded in, so the scalar cost should be
40; 1.
41;
42; TODO: These are currently costed as free the indices are all constants, but we
43; should check if the constants are actually foldable
44define void @g(ptr %dest, i64 %i) {
45; CHECK-LABEL: define void @g
46; CHECK-SAME: (ptr [[DEST:%.*]], i64 [[I:%.*]]) #[[ATTR0]] {
47; CHECK-NEXT:  entry:
48; CHECK-NEXT:    [[P1:%.*]] = getelementptr i32, ptr [[DEST]], i32 2048
49; CHECK-NEXT:    store <4 x i32> splat (i32 1), ptr [[P1]], align 4
50; CHECK-NEXT:    ret void
51;
52entry:
53; YAML:      Pass:            slp-vectorizer
54; YAML-NEXT: Name:            StoresVectorized
55; YAML-NEXT: Function:        g
56; YAML-NEXT: Args:
57; YAML-NEXT:   - String:          'Stores SLP vectorized with cost '
58; YAML-NEXT:   - Cost:            '-2'
59; YAML-NEXT:   - String:          ' and with tree size '
60; YAML-NEXT:   - TreeSize:        '2'
61  %p1 = getelementptr i32, ptr %dest, i32 2048
62  store i32 1, ptr %p1
63  %p2 = getelementptr i32, ptr %dest, i32 2049
64  store i32 1, ptr %p2
65  %p3 = getelementptr i32, ptr %dest, i32 2050
66  store i32 1, ptr %p3
67  %p4 = getelementptr i32, ptr %dest, i32 2051
68  store i32 1, ptr %p4
69  ret void
70}
71
72; When computing the scalar pointers chain cost here, there is a cost of
73; 1 for the base pointer, and the rest can be folded in, so the scalar cost
74; should be 1.
75define void @h(ptr %dest, i32 %i) {
76; CHECK-LABEL: define void @h
77; CHECK-SAME: (ptr [[DEST:%.*]], i32 [[I:%.*]]) #[[ATTR0]] {
78; CHECK-NEXT:  entry:
79; CHECK-NEXT:    [[P1:%.*]] = getelementptr [4 x i32], ptr [[DEST]], i32 [[I]], i32 0
80; CHECK-NEXT:    store <4 x i32> splat (i32 1), ptr [[P1]], align 4
81; CHECK-NEXT:    ret void
82;
83entry:
84; YAML:      Pass:            slp-vectorizer
85; YAML-NEXT: Name:            StoresVectorized
86; YAML-NEXT: Function:        h
87; YAML-NEXT: Args:
88; YAML-NEXT:   - String:          'Stores SLP vectorized with cost '
89; YAML-NEXT:   - Cost:            '-2'
90; YAML-NEXT:   - String:          ' and with tree size '
91; YAML-NEXT:   - TreeSize:        '2'
92  %p1 = getelementptr [4 x i32], ptr %dest, i32 %i, i32 0
93  store i32 1, ptr %p1
94  %p2 = getelementptr [4 x i32], ptr %dest, i32 %i, i32 1
95  store i32 1, ptr %p2
96  %p3 = getelementptr [4 x i32], ptr %dest, i32 %i, i32 2
97  store i32 1, ptr %p3
98  %p4 = getelementptr [4 x i32], ptr %dest, i32 %i, i32 3
99  store i32 1, ptr %p4
100  ret void
101}
102