1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
| ; RUN: llc -mtriple=aarch64-none-linux-gnu < %s | FileCheck %s
; CHECK-LABEL: @test1
; CHECK: sbfx {{x[0-9]+}}, x0, #23, #9
define i64 @test1(i32 %a) {
%tmp = ashr i32 %a, 23
%ext = sext i32 %tmp to i64
%res = add i64 %ext, 1
ret i64 %res
}
; CHECK-LABEL: @test2
; CHECK: sbfx w0, w0, #23, #8
define signext i8 @test2(i32 %a) {
%tmp = ashr i32 %a, 23
%res = trunc i32 %tmp to i8
ret i8 %res
}
; CHECK-LABEL: @test3
; CHECK: sbfx w0, w0, #23, #8
define signext i8 @test3(i32 %a) {
%tmp = lshr i32 %a, 23
%res = trunc i32 %tmp to i8
ret i8 %res
}
; CHECK-LABEL: @test4
; CHECK: sbfx w0, w0, #15, #16
define signext i16 @test4(i32 %a) {
%tmp = lshr i32 %a, 15
%res = trunc i32 %tmp to i16
ret i16 %res
}
; CHECK-LABEL: @test5
; CHECK: sbfx w0, w0, #16, #8
define signext i8 @test5(i64 %a) {
%tmp = lshr i64 %a, 16
%res = trunc i64 %tmp to i8
ret i8 %res
}
; CHECK-LABEL: @test6
; CHECK: sbfx x0, x0, #30, #8
define signext i8 @test6(i64 %a) {
%tmp = lshr i64 %a, 30
%res = trunc i64 %tmp to i8
ret i8 %res
}
; CHECK-LABEL: @test7
; CHECK: sbfx x0, x0, #23, #16
define signext i16 @test7(i64 %a) {
%tmp = lshr i64 %a, 23
%res = trunc i64 %tmp to i16
ret i16 %res
}
; CHECK-LABEL: @test8
; CHECK: asr w0, w0, #25
define signext i8 @test8(i32 %a) {
%tmp = ashr i32 %a, 25
%res = trunc i32 %tmp to i8
ret i8 %res
}
; CHECK-LABEL: @test9
; CHECK: lsr w0, w0, #25
define signext i8 @test9(i32 %a) {
%tmp = lshr i32 %a, 25
%res = trunc i32 %tmp to i8
ret i8 %res
}
; CHECK-LABEL: @test10
; CHECK: lsr x0, x0, #49
define signext i16 @test10(i64 %a) {
%tmp = lshr i64 %a, 49
%res = trunc i64 %tmp to i16
ret i16 %res
}
; SHR with multiple uses is fine as SXTH and SBFX are both aliases of SBFM.
; However, allowing the transformation means the SHR and SBFX can execute in
; parallel.
;
; CHECK-LABEL: @test11
; CHECK: lsr x1, x0, #23
; CHECK: sbfx x0, x0, #23, #16
define void @test11(i64 %a) {
%tmp = lshr i64 %a, 23
%res = trunc i64 %tmp to i16
call void @use(i16 %res, i64 %tmp)
ret void
}
declare void @use(i16 signext, i64)
; CHECK-LABEL: test_complex_node:
; CHECK: ldr d0, [x0], #8
; CHECK: ubfx x[[VAL:[0-9]+]], x0, #5, #27
; CHECK: str w[[VAL]], [x2]
define <2 x i32> @test_complex_node(<2 x i32>* %addr, <2 x i32>** %addr2, i32* %bf ) {
%vec = load <2 x i32>, <2 x i32>* %addr
%vec.next = getelementptr <2 x i32>, <2 x i32>* %addr, i32 1
store <2 x i32>* %vec.next, <2 x i32>** %addr2
%lo = ptrtoint <2 x i32>* %vec.next to i32
%val = lshr i32 %lo, 5
store i32 %val, i32* %bf
ret <2 x i32> %vec
}
|