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
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv8.1m.main-none-eabi -mattr=+mve,-vfp2 -o - %s | FileCheck %s --check-prefix=CHECK-NOFP
; RUN: llc -mtriple=thumbv8.1m.main-none-eabi -mattr=+mve.fp -o - %s | FileCheck --check-prefix=CHECK-FP %s
; This file tests tests that we expand floating point operations correctly,
; even if we do not have an fpu.
define arm_aapcs_vfpcc <8 x half> @vector_add_f16(<8 x half> %lhs, <8 x half> %rhs) {
; CHECK-NOFP-LABEL: vector_add_f16:
; CHECK-NOFP: @ %bb.0: @ %entry
; CHECK-NOFP-NEXT: .save {r4, lr}
; CHECK-NOFP-NEXT: push {r4, lr}
; CHECK-NOFP-NEXT: .vsave {d8, d9, d10, d11, d12, d13}
; CHECK-NOFP-NEXT: vpush {d8, d9, d10, d11, d12, d13}
; CHECK-NOFP-NEXT: vmov.u16 r0, q1[0]
; CHECK-NOFP-NEXT: vmov q5, q1
; CHECK-NOFP-NEXT: vmov q4, q0
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r4, r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q4[0]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r1, r4
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: bl __aeabi_f2h
; CHECK-NOFP-NEXT: vmov.16 q6[0], r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q5[1]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r4, r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q4[1]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r1, r4
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: bl __aeabi_f2h
; CHECK-NOFP-NEXT: vmov.16 q6[1], r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q5[2]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r4, r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q4[2]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r1, r4
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: bl __aeabi_f2h
; CHECK-NOFP-NEXT: vmov.16 q6[2], r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q5[3]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r4, r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q4[3]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r1, r4
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: bl __aeabi_f2h
; CHECK-NOFP-NEXT: vmov.16 q6[3], r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q5[4]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r4, r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q4[4]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r1, r4
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: bl __aeabi_f2h
; CHECK-NOFP-NEXT: vmov.16 q6[4], r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q5[5]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r4, r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q4[5]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r1, r4
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: bl __aeabi_f2h
; CHECK-NOFP-NEXT: vmov.16 q6[5], r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q5[6]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r4, r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q4[6]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r1, r4
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: bl __aeabi_f2h
; CHECK-NOFP-NEXT: vmov.16 q6[6], r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q5[7]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r4, r0
; CHECK-NOFP-NEXT: vmov.u16 r0, q4[7]
; CHECK-NOFP-NEXT: bl __aeabi_h2f
; CHECK-NOFP-NEXT: mov r1, r4
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: bl __aeabi_f2h
; CHECK-NOFP-NEXT: vmov.16 q6[7], r0
; CHECK-NOFP-NEXT: vmov q0, q6
; CHECK-NOFP-NEXT: vpop {d8, d9, d10, d11, d12, d13}
; CHECK-NOFP-NEXT: pop {r4, pc}
;
; CHECK-FP-LABEL: vector_add_f16:
; CHECK-FP: @ %bb.0: @ %entry
; CHECK-FP-NEXT: vadd.f16 q0, q0, q1
; CHECK-FP-NEXT: bx lr
entry:
%sum = fadd <8 x half> %lhs, %rhs
ret <8 x half> %sum
}
define arm_aapcs_vfpcc <4 x float> @vector_add_f32(<4 x float> %lhs, <4 x float> %rhs) {
; CHECK-NOFP-LABEL: vector_add_f32:
; CHECK-NOFP: @ %bb.0: @ %entry
; CHECK-NOFP-NEXT: .save {r4, r5, r7, lr}
; CHECK-NOFP-NEXT: push {r4, r5, r7, lr}
; CHECK-NOFP-NEXT: .vsave {d8, d9, d10, d11}
; CHECK-NOFP-NEXT: vpush {d8, d9, d10, d11}
; CHECK-NOFP-NEXT: vmov q4, q1
; CHECK-NOFP-NEXT: vmov q5, q0
; CHECK-NOFP-NEXT: vmov r4, r0, d11
; CHECK-NOFP-NEXT: vmov r5, r1, d9
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: vmov s19, r0
; CHECK-NOFP-NEXT: mov r0, r4
; CHECK-NOFP-NEXT: mov r1, r5
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: vmov s18, r0
; CHECK-NOFP-NEXT: vmov r4, r0, d10
; CHECK-NOFP-NEXT: vmov r5, r1, d8
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: vmov s17, r0
; CHECK-NOFP-NEXT: mov r0, r4
; CHECK-NOFP-NEXT: mov r1, r5
; CHECK-NOFP-NEXT: bl __aeabi_fadd
; CHECK-NOFP-NEXT: vmov s16, r0
; CHECK-NOFP-NEXT: vmov q0, q4
; CHECK-NOFP-NEXT: vpop {d8, d9, d10, d11}
; CHECK-NOFP-NEXT: pop {r4, r5, r7, pc}
;
; CHECK-FP-LABEL: vector_add_f32:
; CHECK-FP: @ %bb.0: @ %entry
; CHECK-FP-NEXT: vadd.f32 q0, q0, q1
; CHECK-FP-NEXT: bx lr
entry:
%sum = fadd <4 x float> %lhs, %rhs
ret <4 x float> %sum
}
|
