Implement min/max neon intrisics · rust-lang/rust@1f09bae (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
`+
use std::arch::aarch64::*;
`
``
5
`+
use std::mem::transmute;
`
``
6
`+
use std::simd::*;
`
``
7
+
``
8
`+
#[cfg(target_arch = "aarch64")]
`
``
9
`+
unsafe fn test_vpmin_s8() {
`
``
10
`+
let a = i8x8::from([1, -2, 3, -4, 5, 6, 7, 8]);
`
``
11
`+
let b = i8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
`
``
12
`+
let e = i8x8::from([-2, -4, 5, 7, 0, 2, 4, 6]);
`
``
13
`+
let r: i8x8 = transmute(vpmin_s8(transmute(a), transmute(b)));
`
``
14
`+
assert_eq!(r, e);
`
``
15
`+
}
`
``
16
+
``
17
`+
#[cfg(target_arch = "aarch64")]
`
``
18
`+
unsafe fn test_vpmin_s16() {
`
``
19
`+
let a = i16x4::from([1, 2, 3, -4]);
`
``
20
`+
let b = i16x4::from([0, 3, 2, 5]);
`
``
21
`+
let e = i16x4::from([1, -4, 0, 2]);
`
``
22
`+
let r: i16x4 = transmute(vpmin_s16(transmute(a), transmute(b)));
`
``
23
`+
assert_eq!(r, e);
`
``
24
`+
}
`
``
25
+
``
26
`+
#[cfg(target_arch = "aarch64")]
`
``
27
`+
unsafe fn test_vpmin_s32() {
`
``
28
`+
let a = i32x2::from([1, -2]);
`
``
29
`+
let b = i32x2::from([0, 3]);
`
``
30
`+
let e = i32x2::from([-2, 0]);
`
``
31
`+
let r: i32x2 = transmute(vpmin_s32(transmute(a), transmute(b)));
`
``
32
`+
assert_eq!(r, e);
`
``
33
`+
}
`
``
34
+
``
35
`+
#[cfg(target_arch = "aarch64")]
`
``
36
`+
unsafe fn test_vpmin_u8() {
`
``
37
`+
let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
`
``
38
`+
let b = u8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
`
``
39
`+
let e = u8x8::from([1, 3, 5, 7, 0, 2, 4, 6]);
`
``
40
`+
let r: u8x8 = transmute(vpmin_u8(transmute(a), transmute(b)));
`
``
41
`+
assert_eq!(r, e);
`
``
42
`+
}
`
``
43
+
``
44
`+
#[cfg(target_arch = "aarch64")]
`
``
45
`+
unsafe fn test_vpmin_u16() {
`
``
46
`+
let a = u16x4::from([1, 2, 3, 4]);
`
``
47
`+
let b = u16x4::from([0, 3, 2, 5]);
`
``
48
`+
let e = u16x4::from([1, 3, 0, 2]);
`
``
49
`+
let r: u16x4 = transmute(vpmin_u16(transmute(a), transmute(b)));
`
``
50
`+
assert_eq!(r, e);
`
``
51
`+
}
`
``
52
+
``
53
`+
#[cfg(target_arch = "aarch64")]
`
``
54
`+
unsafe fn test_vpmin_u32() {
`
``
55
`+
let a = u32x2::from([1, 2]);
`
``
56
`+
let b = u32x2::from([0, 3]);
`
``
57
`+
let e = u32x2::from([1, 0]);
`
``
58
`+
let r: u32x2 = transmute(vpmin_u32(transmute(a), transmute(b)));
`
``
59
`+
assert_eq!(r, e);
`
``
60
`+
}
`
``
61
+
``
62
`+
#[cfg(target_arch = "aarch64")]
`
``
63
`+
unsafe fn test_vpmin_f32() {
`
``
64
`+
let a = f32x2::from([1., -2.]);
`
``
65
`+
let b = f32x2::from([0., 3.]);
`
``
66
`+
let e = f32x2::from([-2., 0.]);
`
``
67
`+
let r: f32x2 = transmute(vpmin_f32(transmute(a), transmute(b)));
`
``
68
`+
assert_eq!(r, e);
`
``
69
`+
}
`
``
70
+
``
71
`+
#[cfg(target_arch = "aarch64")]
`
``
72
`+
unsafe fn test_vpmax_s8() {
`
``
73
`+
let a = i8x8::from([1, -2, 3, -4, 5, 6, 7, 8]);
`
``
74
`+
let b = i8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
`
``
75
`+
let e = i8x8::from([1, 3, 6, 8, 3, 5, 7, 9]);
`
``
76
`+
let r: i8x8 = transmute(vpmax_s8(transmute(a), transmute(b)));
`
``
77
`+
assert_eq!(r, e);
`
``
78
`+
}
`
``
79
+
``
80
`+
#[cfg(target_arch = "aarch64")]
`
``
81
`+
unsafe fn test_vpmax_s16() {
`
``
82
`+
let a = i16x4::from([1, 2, 3, -4]);
`
``
83
`+
let b = i16x4::from([0, 3, 2, 5]);
`
``
84
`+
let e = i16x4::from([2, 3, 3, 5]);
`
``
85
`+
let r: i16x4 = transmute(vpmax_s16(transmute(a), transmute(b)));
`
``
86
`+
assert_eq!(r, e);
`
``
87
`+
}
`
``
88
+
``
89
`+
#[cfg(target_arch = "aarch64")]
`
``
90
`+
unsafe fn test_vpmax_s32() {
`
``
91
`+
let a = i32x2::from([1, -2]);
`
``
92
`+
let b = i32x2::from([0, 3]);
`
``
93
`+
let e = i32x2::from([1, 3]);
`
``
94
`+
let r: i32x2 = transmute(vpmax_s32(transmute(a), transmute(b)));
`
``
95
`+
assert_eq!(r, e);
`
``
96
`+
}
`
``
97
+
``
98
`+
#[cfg(target_arch = "aarch64")]
`
``
99
`+
unsafe fn test_vpmax_u8() {
`
``
100
`+
let a = u8x8::from([1, 2, 3, 4, 5, 6, 7, 8]);
`
``
101
`+
let b = u8x8::from([0, 3, 2, 5, 4, 7, 6, 9]);
`
``
102
`+
let e = u8x8::from([2, 4, 6, 8, 3, 5, 7, 9]);
`
``
103
`+
let r: u8x8 = transmute(vpmax_u8(transmute(a), transmute(b)));
`
``
104
`+
assert_eq!(r, e);
`
``
105
`+
}
`
``
106
+
``
107
`+
#[cfg(target_arch = "aarch64")]
`
``
108
`+
unsafe fn test_vpmax_u16() {
`
``
109
`+
let a = u16x4::from([1, 2, 3, 4]);
`
``
110
`+
let b = u16x4::from([0, 3, 2, 5]);
`
``
111
`+
let e = u16x4::from([2, 4, 3, 5]);
`
``
112
`+
let r: u16x4 = transmute(vpmax_u16(transmute(a), transmute(b)));
`
``
113
`+
assert_eq!(r, e);
`
``
114
`+
}
`
``
115
+
``
116
`+
#[cfg(target_arch = "aarch64")]
`
``
117
`+
unsafe fn test_vpmax_u32() {
`
``
118
`+
let a = u32x2::from([1, 2]);
`
``
119
`+
let b = u32x2::from([0, 3]);
`
``
120
`+
let e = u32x2::from([2, 3]);
`
``
121
`+
let r: u32x2 = transmute(vpmax_u32(transmute(a), transmute(b)));
`
``
122
`+
assert_eq!(r, e);
`
``
123
`+
}
`
``
124
+
``
125
`+
#[cfg(target_arch = "aarch64")]
`
``
126
`+
unsafe fn test_vpmax_f32() {
`
``
127
`+
let a = f32x2::from([1., -2.]);
`
``
128
`+
let b = f32x2::from([0., 3.]);
`
``
129
`+
let e = f32x2::from([1., 3.]);
`
``
130
`+
let r: f32x2 = transmute(vpmax_f32(transmute(a), transmute(b)));
`
``
131
`+
assert_eq!(r, e);
`
``
132
`+
}
`
``
133
+
``
134
`+
#[cfg(target_arch = "aarch64")]
`
``
135
`+
fn main() {
`
``
136
`+
unsafe {
`
``
137
`+
test_vpmin_s8();
`
``
138
`+
test_vpmin_s16();
`
``
139
`+
test_vpmin_s32();
`
``
140
`+
test_vpmin_u8();
`
``
141
`+
test_vpmin_u16();
`
``
142
`+
test_vpmin_u32();
`
``
143
`+
test_vpmin_f32();
`
``
144
`+
test_vpmax_s8();
`
``
145
`+
test_vpmax_s16();
`
``
146
`+
test_vpmax_s32();
`
``
147
`+
test_vpmax_u8();
`
``
148
`+
test_vpmax_u16();
`
``
149
`+
test_vpmax_u32();
`
``
150
`+
test_vpmax_f32();
`
``
151
`+
}
`
``
152
`+
}
`
``
153
+
``
154
`+
#[cfg(target_arch = "x86_64")]
`
``
155
`+
fn main() {}
`