Merge pull request #1416 from afonso360/aarch64-intrinsics-1 · rust-lang/rust@9f426ce (original) (raw)
``
1
`+
// Most of these tests are copied from https://github.com/japaric/stdsimd/blob/0f4413d01c4f0c3ffbc5a69e9a37fbc7235b31a9/coresimd/arm/neon.rs
`
``
2
+
``
3
`+
#![feature(portable_simd)]
`
``
4
+
``
5
`+
#[cfg(target_arch = "aarch64")]
`
``
6
`+
use std::arch::aarch64::*;
`
``
7
`+
use std::mem::transmute;
`
``
8
`+
use std::simd::*;
`
``
9
+
``
10
`+
#[cfg(target_arch = "aarch64")]
`
``
11
`+
unsafe fn test_vpmin_s8() {
`
``
12
`+
let a = i8x8::from([1, -2, 3, -4, 5, 6, 7, 8]);
`
``
13
`+
let b = i8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
`
``
14
`+
let e = i8x8::from([-2, -4, 5, 7, 0, 2, 4, 6]);
`
``
15
`+
let r: i8x8 = transmute(vpmin_s8(transmute(a), transmute(b)));
`
``
16
`+
assert_eq!(r, e);
`
``
17
`+
}
`
``
18
+
``
19
`+
#[cfg(target_arch = "aarch64")]
`
``
20
`+
unsafe fn test_vpmin_s16() {
`
``
21
`+
let a = i16x4::from([1, 2, 3, -4]);
`
``
22
`+
let b = i16x4::from([0, 3, 2, 5]);
`
``
23
`+
let e = i16x4::from([1, -4, 0, 2]);
`
``
24
`+
let r: i16x4 = transmute(vpmin_s16(transmute(a), transmute(b)));
`
``
25
`+
assert_eq!(r, e);
`
``
26
`+
}
`
``
27
+
``
28
`+
#[cfg(target_arch = "aarch64")]
`
``
29
`+
unsafe fn test_vpmin_s32() {
`
``
30
`+
let a = i32x2::from([1, -2]);
`
``
31
`+
let b = i32x2::from([0, 3]);
`
``
32
`+
let e = i32x2::from([-2, 0]);
`
``
33
`+
let r: i32x2 = transmute(vpmin_s32(transmute(a), transmute(b)));
`
``
34
`+
assert_eq!(r, e);
`
``
35
`+
}
`
``
36
+
``
37
`+
#[cfg(target_arch = "aarch64")]
`
``
38
`+
unsafe fn test_vpmin_u8() {
`
``
39
`+
let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
`
``
40
`+
let b = u8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
`
``
41
`+
let e = u8x8::from([1, 3, 5, 7, 0, 2, 4, 6]);
`
``
42
`+
let r: u8x8 = transmute(vpmin_u8(transmute(a), transmute(b)));
`
``
43
`+
assert_eq!(r, e);
`
``
44
`+
}
`
``
45
+
``
46
`+
#[cfg(target_arch = "aarch64")]
`
``
47
`+
unsafe fn test_vpmin_u16() {
`
``
48
`+
let a = u16x4::from([1, 2, 3, 4]);
`
``
49
`+
let b = u16x4::from([0, 3, 2, 5]);
`
``
50
`+
let e = u16x4::from([1, 3, 0, 2]);
`
``
51
`+
let r: u16x4 = transmute(vpmin_u16(transmute(a), transmute(b)));
`
``
52
`+
assert_eq!(r, e);
`
``
53
`+
}
`
``
54
+
``
55
`+
#[cfg(target_arch = "aarch64")]
`
``
56
`+
unsafe fn test_vpmin_u32() {
`
``
57
`+
let a = u32x2::from([1, 2]);
`
``
58
`+
let b = u32x2::from([0, 3]);
`
``
59
`+
let e = u32x2::from([1, 0]);
`
``
60
`+
let r: u32x2 = transmute(vpmin_u32(transmute(a), transmute(b)));
`
``
61
`+
assert_eq!(r, e);
`
``
62
`+
}
`
``
63
+
``
64
`+
#[cfg(target_arch = "aarch64")]
`
``
65
`+
unsafe fn test_vpmin_f32() {
`
``
66
`+
let a = f32x2::from([1., -2.]);
`
``
67
`+
let b = f32x2::from([0., 3.]);
`
``
68
`+
let e = f32x2::from([-2., 0.]);
`
``
69
`+
let r: f32x2 = transmute(vpmin_f32(transmute(a), transmute(b)));
`
``
70
`+
assert_eq!(r, e);
`
``
71
`+
}
`
``
72
+
``
73
`+
#[cfg(target_arch = "aarch64")]
`
``
74
`+
unsafe fn test_vpmax_s8() {
`
``
75
`+
let a = i8x8::from([1, -2, 3, -4, 5, 6, 7, 8]);
`
``
76
`+
let b = i8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
`
``
77
`+
let e = i8x8::from([1, 3, 6, 8, 3, 5, 7, 9]);
`
``
78
`+
let r: i8x8 = transmute(vpmax_s8(transmute(a), transmute(b)));
`
``
79
`+
assert_eq!(r, e);
`
``
80
`+
}
`
``
81
+
``
82
`+
#[cfg(target_arch = "aarch64")]
`
``
83
`+
unsafe fn test_vpmax_s16() {
`
``
84
`+
let a = i16x4::from([1, 2, 3, -4]);
`
``
85
`+
let b = i16x4::from([0, 3, 2, 5]);
`
``
86
`+
let e = i16x4::from([2, 3, 3, 5]);
`
``
87
`+
let r: i16x4 = transmute(vpmax_s16(transmute(a), transmute(b)));
`
``
88
`+
assert_eq!(r, e);
`
``
89
`+
}
`
``
90
+
``
91
`+
#[cfg(target_arch = "aarch64")]
`
``
92
`+
unsafe fn test_vpmax_s32() {
`
``
93
`+
let a = i32x2::from([1, -2]);
`
``
94
`+
let b = i32x2::from([0, 3]);
`
``
95
`+
let e = i32x2::from([1, 3]);
`
``
96
`+
let r: i32x2 = transmute(vpmax_s32(transmute(a), transmute(b)));
`
``
97
`+
assert_eq!(r, e);
`
``
98
`+
}
`
``
99
+
``
100
`+
#[cfg(target_arch = "aarch64")]
`
``
101
`+
unsafe fn test_vpmax_u8() {
`
``
102
`+
let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
`
``
103
`+
let b = u8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
`
``
104
`+
let e = u8x8::from([2, 4, 6, 8, 3, 5, 7, 9]);
`
``
105
`+
let r: u8x8 = transmute(vpmax_u8(transmute(a), transmute(b)));
`
``
106
`+
assert_eq!(r, e);
`
``
107
`+
}
`
``
108
+
``
109
`+
#[cfg(target_arch = "aarch64")]
`
``
110
`+
unsafe fn test_vpmax_u16() {
`
``
111
`+
let a = u16x4::from([1, 2, 3, 4]);
`
``
112
`+
let b = u16x4::from([0, 3, 2, 5]);
`
``
113
`+
let e = u16x4::from([2, 4, 3, 5]);
`
``
114
`+
let r: u16x4 = transmute(vpmax_u16(transmute(a), transmute(b)));
`
``
115
`+
assert_eq!(r, e);
`
``
116
`+
}
`
``
117
+
``
118
`+
#[cfg(target_arch = "aarch64")]
`
``
119
`+
unsafe fn test_vpmax_u32() {
`
``
120
`+
let a = u32x2::from([1, 2]);
`
``
121
`+
let b = u32x2::from([0, 3]);
`
``
122
`+
let e = u32x2::from([2, 3]);
`
``
123
`+
let r: u32x2 = transmute(vpmax_u32(transmute(a), transmute(b)));
`
``
124
`+
assert_eq!(r, e);
`
``
125
`+
}
`
``
126
+
``
127
`+
#[cfg(target_arch = "aarch64")]
`
``
128
`+
unsafe fn test_vpmax_f32() {
`
``
129
`+
let a = f32x2::from([1., -2.]);
`
``
130
`+
let b = f32x2::from([0., 3.]);
`
``
131
`+
let e = f32x2::from([1., 3.]);
`
``
132
`+
let r: f32x2 = transmute(vpmax_f32(transmute(a), transmute(b)));
`
``
133
`+
assert_eq!(r, e);
`
``
134
`+
}
`
``
135
+
``
136
`+
#[cfg(target_arch = "aarch64")]
`
``
137
`+
unsafe fn test_vpadd_s16() {
`
``
138
`+
let a = i16x4::from([1, 2, 3, 4]);
`
``
139
`+
let b = i16x4::from([0, -1, -2, -3]);
`
``
140
`+
let r: i16x4 = transmute(vpadd_s16(transmute(a), transmute(b)));
`
``
141
`+
let e = i16x4::from([3, 7, -1, -5]);
`
``
142
`+
assert_eq!(r, e);
`
``
143
`+
}
`
``
144
`+
#[cfg(target_arch = "aarch64")]
`
``
145
`+
unsafe fn test_vpadd_s32() {
`
``
146
`+
let a = i32x2::from([1, 2]);
`
``
147
`+
let b = i32x2::from([0, -1]);
`
``
148
`+
let r: i32x2 = transmute(vpadd_s32(transmute(a), transmute(b)));
`
``
149
`+
let e = i32x2::from([3, -1]);
`
``
150
`+
assert_eq!(r, e);
`
``
151
`+
}
`
``
152
`+
#[cfg(target_arch = "aarch64")]
`
``
153
`+
unsafe fn test_vpadd_s8() {
`
``
154
`+
let a = i8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
`
``
155
`+
let b = i8x8::from([0, -1, -2, -3, -4, -5, -6, -7]);
`
``
156
`+
let r: i8x8 = transmute(vpadd_s8(transmute(a), transmute(b)));
`
``
157
`+
let e = i8x8::from([3, 7, 11, 15, -1, -5, -9, -13]);
`
``
158
`+
assert_eq!(r, e);
`
``
159
`+
}
`
``
160
`+
#[cfg(target_arch = "aarch64")]
`
``
161
`+
unsafe fn test_vpadd_u16() {
`
``
162
`+
let a = u16x4::from([1, 2, 3, 4]);
`
``
163
`+
let b = u16x4::from([30, 31, 32, 33]);
`
``
164
`+
let r: u16x4 = transmute(vpadd_u16(transmute(a), transmute(b)));
`
``
165
`+
let e = u16x4::from([3, 7, 61, 65]);
`
``
166
`+
assert_eq!(r, e);
`
``
167
`+
}
`
``
168
`+
#[cfg(target_arch = "aarch64")]
`
``
169
`+
unsafe fn test_vpadd_u32() {
`
``
170
`+
let a = u32x2::from([1, 2]);
`
``
171
`+
let b = u32x2::from([30, 31]);
`
``
172
`+
let r: u32x2 = transmute(vpadd_u32(transmute(a), transmute(b)));
`
``
173
`+
let e = u32x2::from([3, 61]);
`
``
174
`+
assert_eq!(r, e);
`
``
175
`+
}
`
``
176
`+
#[cfg(target_arch = "aarch64")]
`
``
177
`+
unsafe fn test_vpadd_u8() {
`
``
178
`+
let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
`
``
179
`+
let b = u8x8::from([30, 31, 32, 33, 34, 35, 36, 37]);
`
``
180
`+
let r: u8x8 = transmute(vpadd_u8(transmute(a), transmute(b)));
`
``
181
`+
let e = u8x8::from([3, 7, 11, 15, 61, 65, 69, 73]);
`
``
182
`+
assert_eq!(r, e);
`
``
183
`+
}
`
``
184
+
``
185
`+
#[cfg(target_arch = "aarch64")]
`
``
186
`+
unsafe fn test_vqsub_u8() {
`
``
187
`+
let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 0xff]);
`
``
188
`+
let b = u8x8::from([30, 1, 1, 1, 34, 0xff, 36, 37]);
`
``
189
`+
let r: u8x8 = transmute(vqsub_u8(transmute(a), transmute(b)));
`
``
190
`+
let e = u8x8::from([0, 1, 2, 3, 0, 0, 0, 218]);
`
``
191
`+
assert_eq!(r, e);
`
``
192
`+
}
`
``
193
+
``
194
`+
#[cfg(target_arch = "aarch64")]
`
``
195
`+
unsafe fn test_vqadd_u8() {
`
``
196
`+
let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 0xff]);
`
``
197
`+
let b = u8x8::from([30, 1, 1, 1, 34, 0xff, 36, 37]);
`
``
198
`+
let r: u8x8 = transmute(vqadd_u8(transmute(a), transmute(b)));
`
``
199
`+
let e = u8x8::from([31, 3, 4, 5, 39, 0xff, 43, 0xff]);
`
``
200
`+
assert_eq!(r, e);
`
``
201
`+
}
`
``
202
+
``
203
`+
#[cfg(target_arch = "aarch64")]
`
``
204
`+
fn main() {
`
``
205
`+
unsafe {
`
``
206
`+
test_vpmin_s8();
`
``
207
`+
test_vpmin_s16();
`
``
208
`+
test_vpmin_s32();
`
``
209
`+
test_vpmin_u8();
`
``
210
`+
test_vpmin_u16();
`
``
211
`+
test_vpmin_u32();
`
``
212
`+
test_vpmin_f32();
`
``
213
`+
test_vpmax_s8();
`
``
214
`+
test_vpmax_s16();
`
``
215
`+
test_vpmax_s32();
`
``
216
`+
test_vpmax_u8();
`
``
217
`+
test_vpmax_u16();
`
``
218
`+
test_vpmax_u32();
`
``
219
`+
test_vpmax_f32();
`
``
220
+
``
221
`+
test_vpadd_s16();
`
``
222
`+
test_vpadd_s32();
`
``
223
`+
test_vpadd_s8();
`
``
224
`+
test_vpadd_u16();
`
``
225
`+
test_vpadd_u32();
`
``
226
`+
test_vpadd_u8();
`
``
227
+
``
228
`+
test_vqsub_u8();
`
``
229
`+
test_vqadd_u8();
`
``
230
`+
}
`
``
231
`+
}
`
``
232
+
``
233
`+
#[cfg(not(target_arch = "aarch64"))]
`
``
234
`+
fn main() {}
`