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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
#![allow(non_camel_case_types)]
use core::ops::{Add, AddAssign, BitAnd, BitOr, BitXor, BitXorAssign, Not};

pub trait AndNot {
    type Output;
    fn andnot(self, rhs: Self) -> Self::Output;
}
pub trait BSwap {
    fn bswap(self) -> Self;
}
/// Ops that depend on word size
pub trait ArithOps: Add<Output = Self> + AddAssign + Sized + Copy + Clone + BSwap {}
/// Ops that are independent of word size and endian
pub trait BitOps0:
    BitAnd<Output = Self>
    + BitOr<Output = Self>
    + BitXor<Output = Self>
    + BitXorAssign
    + Not<Output = Self>
    + AndNot<Output = Self>
    + Sized
    + Copy
    + Clone
{
}

pub trait BitOps32: BitOps0 + RotateEachWord32 {}
pub trait BitOps64: BitOps32 + RotateEachWord64 {}
pub trait BitOps128: BitOps64 + RotateEachWord128 {}

pub trait RotateEachWord32 {
    fn rotate_each_word_right7(self) -> Self;
    fn rotate_each_word_right8(self) -> Self;
    fn rotate_each_word_right11(self) -> Self;
    fn rotate_each_word_right12(self) -> Self;
    fn rotate_each_word_right16(self) -> Self;
    fn rotate_each_word_right20(self) -> Self;
    fn rotate_each_word_right24(self) -> Self;
    fn rotate_each_word_right25(self) -> Self;
}

pub trait RotateEachWord64 {
    fn rotate_each_word_right32(self) -> Self;
}

pub trait RotateEachWord128 {}

// Vector type naming scheme:
// uN[xP]xL
// Unsigned; N-bit words * P bits per lane * L lanes
//
// A lane is always 128-bits, chosen because common SIMD architectures treat 128-bit units of
// wide vectors specially (supporting e.g. intra-lane shuffles), and tend to have limited and
// slow inter-lane operations.

use crate::arch::{vec128_storage, vec256_storage, vec512_storage};

#[allow(clippy::missing_safety_doc)]
pub trait UnsafeFrom<T> {
    unsafe fn unsafe_from(t: T) -> Self;
}

/// A vector composed of two elements, which may be words or themselves vectors.
pub trait Vec2<W> {
    fn extract(self, i: u32) -> W;
    fn insert(self, w: W, i: u32) -> Self;
}

/// A vector composed of four elements, which may be words or themselves vectors.
pub trait Vec4<W> {
    fn extract(self, i: u32) -> W;
    fn insert(self, w: W, i: u32) -> Self;
}

// TODO: multiples of 4 should inherit this
/// A vector composed of four words; depending on their size, operations may cross lanes.
pub trait Words4 {
    fn shuffle1230(self) -> Self;
    fn shuffle2301(self) -> Self;
    fn shuffle3012(self) -> Self;
}

/// A vector composed one or more lanes each composed of four words.
pub trait LaneWords4 {
    fn shuffle_lane_words1230(self) -> Self;
    fn shuffle_lane_words2301(self) -> Self;
    fn shuffle_lane_words3012(self) -> Self;
}

// TODO: make this a part of BitOps
/// Exchange neigboring ranges of bits of the specified size
pub trait Swap64 {
    fn swap1(self) -> Self;
    fn swap2(self) -> Self;
    fn swap4(self) -> Self;
    fn swap8(self) -> Self;
    fn swap16(self) -> Self;
    fn swap32(self) -> Self;
    fn swap64(self) -> Self;
}

pub trait u32x4<M: Machine>:
    BitOps32
    + Store<vec128_storage>
    + ArithOps
    + Vec4<u32>
    + Words4
    + LaneWords4
    + StoreBytes
    + MultiLane<[u32; 4]>
    + Into<vec128_storage>
{
}
pub trait u64x2<M: Machine>:
    BitOps64
    + Store<vec128_storage>
    + ArithOps
    + Vec2<u64>
    + MultiLane<[u64; 2]>
    + Into<vec128_storage>
{
}
pub trait u128x1<M: Machine>:
    BitOps128 + Store<vec128_storage> + Swap64 + MultiLane<[u128; 1]> + Into<vec128_storage>
{
}

pub trait u32x4x2<M: Machine>:
    BitOps32
    + Store<vec256_storage>
    + Vec2<M::u32x4>
    + MultiLane<[M::u32x4; 2]>
    + ArithOps
    + Into<vec256_storage>
{
}
pub trait u64x2x2<M: Machine>:
    BitOps64
    + Store<vec256_storage>
    + Vec2<M::u64x2>
    + MultiLane<[M::u64x2; 2]>
    + ArithOps
    + StoreBytes
    + Into<vec256_storage>
{
}
pub trait u64x4<M: Machine>:
    BitOps64
    + Store<vec256_storage>
    + Vec4<u64>
    + MultiLane<[u64; 4]>
    + ArithOps
    + Words4
    + StoreBytes
    + Into<vec256_storage>
{
}
pub trait u128x2<M: Machine>:
    BitOps128
    + Store<vec256_storage>
    + Vec2<M::u128x1>
    + MultiLane<[M::u128x1; 2]>
    + Swap64
    + Into<vec256_storage>
{
}

pub trait u32x4x4<M: Machine>:
    BitOps32
    + Store<vec512_storage>
    + Vec4<M::u32x4>
    + MultiLane<[M::u32x4; 4]>
    + ArithOps
    + LaneWords4
    + Into<vec512_storage>
{
}
pub trait u64x2x4<M: Machine>:
    BitOps64
    + Store<vec512_storage>
    + Vec4<M::u64x2>
    + MultiLane<[M::u64x2; 4]>
    + ArithOps
    + Into<vec512_storage>
{
}
// TODO: Words4
pub trait u128x4<M: Machine>:
    BitOps128
    + Store<vec512_storage>
    + Vec4<M::u128x1>
    + MultiLane<[M::u128x1; 4]>
    + Swap64
    + Into<vec512_storage>
{
}

/// A vector composed of multiple 128-bit lanes.
pub trait MultiLane<Lanes> {
    /// Split a multi-lane vector into single-lane vectors.
    fn to_lanes(self) -> Lanes;
    /// Build a multi-lane vector from individual lanes.
    fn from_lanes(lanes: Lanes) -> Self;
}

/// Combine single vectors into a multi-lane vector.
pub trait VZip<V> {
    fn vzip(self) -> V;
}

impl<V, T> VZip<V> for T
where
    V: MultiLane<T>,
{
    #[inline(always)]
    fn vzip(self) -> V {
        V::from_lanes(self)
    }
}

pub trait Machine: Sized + Copy {
    type u32x4: u32x4<Self>;
    type u64x2: u64x2<Self>;
    type u128x1: u128x1<Self>;

    type u32x4x2: u32x4x2<Self>;
    type u64x2x2: u64x2x2<Self>;
    type u64x4: u64x4<Self>;
    type u128x2: u128x2<Self>;

    type u32x4x4: u32x4x4<Self>;
    type u64x2x4: u64x2x4<Self>;
    type u128x4: u128x4<Self>;

    #[inline(always)]
    fn unpack<S, V: Store<S>>(self, s: S) -> V {
        unsafe { V::unpack(s) }
    }

    #[inline(always)]
    fn vec<V, A>(self, a: A) -> V
    where
        V: MultiLane<A>,
    {
        V::from_lanes(a)
    }

    #[inline(always)]
    fn read_le<V>(self, input: &[u8]) -> V
    where
        V: StoreBytes,
    {
        unsafe { V::unsafe_read_le(input) }
    }

    #[inline(always)]
    fn read_be<V>(self, input: &[u8]) -> V
    where
        V: StoreBytes,
    {
        unsafe { V::unsafe_read_be(input) }
    }

    /// # Safety
    /// Caller must ensure the type of Self is appropriate for the hardware of the execution
    /// environment.
    unsafe fn instance() -> Self;
}

pub trait Store<S> {
    /// # Safety
    /// Caller must ensure the type of Self is appropriate for the hardware of the execution
    /// environment.
    unsafe fn unpack(p: S) -> Self;
}

pub trait StoreBytes {
    /// # Safety
    /// Caller must ensure the type of Self is appropriate for the hardware of the execution
    /// environment.
    unsafe fn unsafe_read_le(input: &[u8]) -> Self;
    /// # Safety
    /// Caller must ensure the type of Self is appropriate for the hardware of the execution
    /// environment.
    unsafe fn unsafe_read_be(input: &[u8]) -> Self;
    fn write_le(self, out: &mut [u8]);
    fn write_be(self, out: &mut [u8]);
}