-
Notifications
You must be signed in to change notification settings - Fork 13.3k
/
Copy pathgeneric-arithmetic-pass.rs
145 lines (122 loc) · 5.95 KB
/
generic-arithmetic-pass.rs
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
140
141
142
143
144
145
//@ run-pass
#![allow(non_camel_case_types)]
#![feature(repr_simd, core_intrinsics)]
#[repr(simd)]
#[derive(Copy, Clone)]
struct i32x4(pub [i32; 4]);
impl i32x4 {
fn to_array(self) -> [i32; 4] { unsafe { std::mem::transmute(self) } }
}
#[repr(simd)]
#[derive(Copy, Clone)]
struct U32<const N: usize>([u32; N]);
impl<const N: usize> U32<N> {
fn to_array(self) -> [u32; N] { unsafe { std::intrinsics::transmute_unchecked(self) } }
}
#[repr(simd)]
#[derive(Copy, Clone)]
struct f32x4(pub [f32; 4]);
impl f32x4 {
fn to_array(self) -> [f32; 4] { unsafe { std::mem::transmute(self) } }
}
macro_rules! all_eq {
($a: expr, $b: expr) => {{
let a = $a;
let b = $b;
assert!(a.to_array() == b.to_array());
}};
}
use std::intrinsics::simd::*;
fn main() {
let x1 = i32x4([1, 2, 3, 4]);
let y1 = U32::<4>([1, 2, 3, 4]);
let z1 = f32x4([1.0, 2.0, 3.0, 4.0]);
let x2 = i32x4([2, 3, 4, 5]);
let y2 = U32::<4>([2, 3, 4, 5]);
let z2 = f32x4([2.0, 3.0, 4.0, 5.0]);
let x3 = i32x4([0, i32::MAX, i32::MIN, -1_i32]);
let y3 = U32::<4>([0, i32::MAX as _, i32::MIN as _, -1_i32 as _]);
unsafe {
all_eq!(simd_add(x1, x2), i32x4([3, 5, 7, 9]));
all_eq!(simd_add(x2, x1), i32x4([3, 5, 7, 9]));
all_eq!(simd_add(y1, y2), U32::<4>([3, 5, 7, 9]));
all_eq!(simd_add(y2, y1), U32::<4>([3, 5, 7, 9]));
all_eq!(simd_add(z1, z2), f32x4([3.0, 5.0, 7.0, 9.0]));
all_eq!(simd_add(z2, z1), f32x4([3.0, 5.0, 7.0, 9.0]));
all_eq!(simd_mul(x1, x2), i32x4([2, 6, 12, 20]));
all_eq!(simd_mul(x2, x1), i32x4([2, 6, 12, 20]));
all_eq!(simd_mul(y1, y2), U32::<4>([2, 6, 12, 20]));
all_eq!(simd_mul(y2, y1), U32::<4>([2, 6, 12, 20]));
all_eq!(simd_mul(z1, z2), f32x4([2.0, 6.0, 12.0, 20.0]));
all_eq!(simd_mul(z2, z1), f32x4([2.0, 6.0, 12.0, 20.0]));
all_eq!(simd_sub(x2, x1), i32x4([1, 1, 1, 1]));
all_eq!(simd_sub(x1, x2), i32x4([-1, -1, -1, -1]));
all_eq!(simd_sub(y2, y1), U32::<4>([1, 1, 1, 1]));
all_eq!(simd_sub(y1, y2), U32::<4>([!0, !0, !0, !0]));
all_eq!(simd_sub(z2, z1), f32x4([1.0, 1.0, 1.0, 1.0]));
all_eq!(simd_sub(z1, z2), f32x4([-1.0, -1.0, -1.0, -1.0]));
all_eq!(simd_div(x1, x1), i32x4([1, 1, 1, 1]));
all_eq!(simd_div(i32x4([2, 4, 6, 8]), i32x4([2, 2, 2, 2])), x1);
all_eq!(simd_div(y1, y1), U32::<4>([1, 1, 1, 1]));
all_eq!(simd_div(U32::<4>([2, 4, 6, 8]), U32::<4>([2, 2, 2, 2])), y1);
all_eq!(simd_div(z1, z1), f32x4([1.0, 1.0, 1.0, 1.0]));
all_eq!(simd_div(z1, z2), f32x4([1.0 / 2.0, 2.0 / 3.0, 3.0 / 4.0, 4.0 / 5.0]));
all_eq!(simd_div(z2, z1), f32x4([2.0 / 1.0, 3.0 / 2.0, 4.0 / 3.0, 5.0 / 4.0]));
all_eq!(simd_rem(x1, x1), i32x4([0, 0, 0, 0]));
all_eq!(simd_rem(x2, x1), i32x4([0, 1, 1, 1]));
all_eq!(simd_rem(y1, y1), U32::<4>([0, 0, 0, 0]));
all_eq!(simd_rem(y2, y1), U32::<4>([0, 1, 1, 1]));
all_eq!(simd_rem(z1, z1), f32x4([0.0, 0.0, 0.0, 0.0]));
all_eq!(simd_rem(z1, z2), z1);
all_eq!(simd_rem(z2, z1), f32x4([0.0, 1.0, 1.0, 1.0]));
all_eq!(simd_shl(x1, x2), i32x4([1 << 2, 2 << 3, 3 << 4, 4 << 5]));
all_eq!(simd_shl(x2, x1), i32x4([2 << 1, 3 << 2, 4 << 3, 5 << 4]));
all_eq!(simd_shl(y1, y2), U32::<4>([1 << 2, 2 << 3, 3 << 4, 4 << 5]));
all_eq!(simd_shl(y2, y1), U32::<4>([2 << 1, 3 << 2, 4 << 3, 5 << 4]));
// test right-shift by assuming left-shift is correct
all_eq!(simd_shr(simd_shl(x1, x2), x2), x1);
all_eq!(simd_shr(simd_shl(x2, x1), x1), x2);
all_eq!(simd_shr(simd_shl(y1, y2), y2), y1);
all_eq!(simd_shr(simd_shl(y2, y1), y1), y2);
// ensure we get logical vs. arithmetic shifts correct
let (a, b, c, d) = (-12, -123, -1234, -12345);
all_eq!(simd_shr(i32x4([a, b, c, d]), x1), i32x4([a >> 1, b >> 2, c >> 3, d >> 4]));
all_eq!(
simd_shr(U32::<4>([a as u32, b as u32, c as u32, d as u32]), y1),
U32::<4>([(a as u32) >> 1, (b as u32) >> 2, (c as u32) >> 3, (d as u32) >> 4])
);
all_eq!(simd_and(x1, x2), i32x4([0, 2, 0, 4]));
all_eq!(simd_and(x2, x1), i32x4([0, 2, 0, 4]));
all_eq!(simd_and(y1, y2), U32::<4>([0, 2, 0, 4]));
all_eq!(simd_and(y2, y1), U32::<4>([0, 2, 0, 4]));
all_eq!(simd_or(x1, x2), i32x4([3, 3, 7, 5]));
all_eq!(simd_or(x2, x1), i32x4([3, 3, 7, 5]));
all_eq!(simd_or(y1, y2), U32::<4>([3, 3, 7, 5]));
all_eq!(simd_or(y2, y1), U32::<4>([3, 3, 7, 5]));
all_eq!(simd_xor(x1, x2), i32x4([3, 1, 7, 1]));
all_eq!(simd_xor(x2, x1), i32x4([3, 1, 7, 1]));
all_eq!(simd_xor(y1, y2), U32::<4>([3, 1, 7, 1]));
all_eq!(simd_xor(y2, y1), U32::<4>([3, 1, 7, 1]));
all_eq!(simd_neg(x1), i32x4([-1, -2, -3, -4]));
all_eq!(simd_neg(x2), i32x4([-2, -3, -4, -5]));
all_eq!(simd_neg(z1), f32x4([-1.0, -2.0, -3.0, -4.0]));
all_eq!(simd_neg(z2), f32x4([-2.0, -3.0, -4.0, -5.0]));
all_eq!(simd_bswap(x1), i32x4([0x01000000, 0x02000000, 0x03000000, 0x04000000]));
all_eq!(simd_bswap(y1), U32::<4>([0x01000000, 0x02000000, 0x03000000, 0x04000000]));
all_eq!(
simd_bitreverse(x1),
i32x4([0x80000000u32 as i32, 0x40000000, 0xc0000000u32 as i32, 0x20000000])
);
all_eq!(simd_bitreverse(y1), U32::<4>([0x80000000, 0x40000000, 0xc0000000, 0x20000000]));
all_eq!(simd_ctlz(x1), i32x4([31, 30, 30, 29]));
all_eq!(simd_ctlz(y1), U32::<4>([31, 30, 30, 29]));
all_eq!(simd_ctpop(x1), i32x4([1, 1, 2, 1]));
all_eq!(simd_ctpop(y1), U32::<4>([1, 1, 2, 1]));
all_eq!(simd_ctpop(x2), i32x4([1, 2, 1, 2]));
all_eq!(simd_ctpop(y2), U32::<4>([1, 2, 1, 2]));
all_eq!(simd_ctpop(x3), i32x4([0, 31, 1, 32]));
all_eq!(simd_ctpop(y3), U32::<4>([0, 31, 1, 32]));
all_eq!(simd_cttz(x1), i32x4([0, 1, 0, 2]));
all_eq!(simd_cttz(y1), U32::<4>([0, 1, 0, 2]));
}
}