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
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
//! Masks that take up full SIMD vector registers.

use crate::simd::intrinsics;
use crate::simd::{LaneCount, MaskElement, Simd, SupportedLaneCount};

#[repr(transparent)]
pub struct Mask<T, const N: usize>(Simd<T, N>)
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount;

impl<T, const N: usize> Copy for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
}

impl<T, const N: usize> Clone for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    fn clone(&self) -> Self {
        *self
    }
}

impl<T, const N: usize> PartialEq for Mask<T, N>
where
    T: MaskElement + PartialEq,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn eq(&self, other: &Self) -> bool {
        self.0.eq(&other.0)
    }
}

impl<T, const N: usize> PartialOrd for Mask<T, N>
where
    T: MaskElement + PartialOrd,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn partial_cmp(&self, other: &Self) -> Option<core::cmp::Ordering> {
        self.0.partial_cmp(&other.0)
    }
}

impl<T, const N: usize> Eq for Mask<T, N>
where
    T: MaskElement + Eq,
    LaneCount<N>: SupportedLaneCount,
{
}

impl<T, const N: usize> Ord for Mask<T, N>
where
    T: MaskElement + Ord,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn cmp(&self, other: &Self) -> core::cmp::Ordering {
        self.0.cmp(&other.0)
    }
}

// Used for bitmask bit order workaround
pub(crate) trait ReverseBits {
    // Reverse the least significant `n` bits of `self`.
    // (Remaining bits must be 0.)
    fn reverse_bits(self, n: usize) -> Self;
}

macro_rules! impl_reverse_bits {
    { $($int:ty),* } => {
        $(
        impl ReverseBits for $int {
            #[inline(always)]
            fn reverse_bits(self, n: usize) -> Self {
                let rev = <$int>::reverse_bits(self);
                let bitsize = core::mem::size_of::<$int>() * 8;
                if n < bitsize {
                    // Shift things back to the right
                    rev >> (bitsize - n)
                } else {
                    rev
                }
            }
        }
        )*
    }
}

impl_reverse_bits! { u8, u16, u32, u64 }

impl<T, const N: usize> Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    pub fn splat(value: bool) -> Self {
        Self(Simd::splat(if value { T::TRUE } else { T::FALSE }))
    }

    #[inline]
    #[must_use = "method returns a new bool and does not mutate the original value"]
    pub unsafe fn test_unchecked(&self, lane: usize) -> bool {
        T::eq(self.0[lane], T::TRUE)
    }

    #[inline]
    pub unsafe fn set_unchecked(&mut self, lane: usize, value: bool) {
        self.0[lane] = if value { T::TRUE } else { T::FALSE }
    }

    #[inline]
    #[must_use = "method returns a new vector and does not mutate the original value"]
    pub fn to_int(self) -> Simd<T, N> {
        self.0
    }

    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    pub unsafe fn from_int_unchecked(value: Simd<T, N>) -> Self {
        Self(value)
    }

    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    pub fn convert<U>(self) -> Mask<U, N>
    where
        U: MaskElement,
    {
        // Safety: masks are simply integer vectors of 0 and -1, and we can cast the element type.
        unsafe { Mask(intrinsics::simd_cast(self.0)) }
    }

    #[inline]
    #[must_use = "method returns a new vector and does not mutate the original value"]
    pub fn to_bitmask_vector(self) -> Simd<u8, N> {
        let mut bitmask = Simd::splat(0);

        // Safety: Bytes is the right size array
        unsafe {
            // Compute the bitmask
            let mut bytes: <LaneCount<N> as SupportedLaneCount>::BitMask =
                intrinsics::simd_bitmask(self.0);

            // LLVM assumes bit order should match endianness
            if cfg!(target_endian = "big") {
                for x in bytes.as_mut() {
                    *x = x.reverse_bits()
                }
            }

            bitmask.as_mut_array()[..bytes.as_ref().len()].copy_from_slice(bytes.as_ref());
        }

        bitmask
    }

    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    pub fn from_bitmask_vector(bitmask: Simd<u8, N>) -> Self {
        let mut bytes = <LaneCount<N> as SupportedLaneCount>::BitMask::default();

        // Safety: Bytes is the right size array
        unsafe {
            let len = bytes.as_ref().len();
            bytes.as_mut().copy_from_slice(&bitmask.as_array()[..len]);

            // LLVM assumes bit order should match endianness
            if cfg!(target_endian = "big") {
                for x in bytes.as_mut() {
                    *x = x.reverse_bits();
                }
            }

            // Compute the regular mask
            Self::from_int_unchecked(intrinsics::simd_select_bitmask(
                bytes,
                Self::splat(true).to_int(),
                Self::splat(false).to_int(),
            ))
        }
    }

    #[inline]
    unsafe fn to_bitmask_impl<U: ReverseBits, const M: usize>(self) -> U
    where
        LaneCount<M>: SupportedLaneCount,
    {
        let resized = self.to_int().resize::<M>(T::FALSE);

        // Safety: `resized` is an integer vector with length M, which must match T
        let bitmask: U = unsafe { intrinsics::simd_bitmask(resized) };

        // LLVM assumes bit order should match endianness
        if cfg!(target_endian = "big") {
            bitmask.reverse_bits(M)
        } else {
            bitmask
        }
    }

    #[inline]
    unsafe fn from_bitmask_impl<U: ReverseBits, const M: usize>(bitmask: U) -> Self
    where
        LaneCount<M>: SupportedLaneCount,
    {
        // LLVM assumes bit order should match endianness
        let bitmask = if cfg!(target_endian = "big") {
            bitmask.reverse_bits(M)
        } else {
            bitmask
        };

        // SAFETY: `mask` is the correct bitmask type for a u64 bitmask
        let mask: Simd<T, M> = unsafe {
            intrinsics::simd_select_bitmask(
                bitmask,
                Simd::<T, M>::splat(T::TRUE),
                Simd::<T, M>::splat(T::FALSE),
            )
        };

        // SAFETY: `mask` only contains `T::TRUE` or `T::FALSE`
        unsafe { Self::from_int_unchecked(mask.resize::<N>(T::FALSE)) }
    }

    #[inline]
    pub(crate) fn to_bitmask_integer(self) -> u64 {
        // TODO modify simd_bitmask to zero-extend output, making this unnecessary
        if N <= 8 {
            // Safety: bitmask matches length
            unsafe { self.to_bitmask_impl::<u8, 8>() as u64 }
        } else if N <= 16 {
            // Safety: bitmask matches length
            unsafe { self.to_bitmask_impl::<u16, 16>() as u64 }
        } else if N <= 32 {
            // Safety: bitmask matches length
            unsafe { self.to_bitmask_impl::<u32, 32>() as u64 }
        } else {
            // Safety: bitmask matches length
            unsafe { self.to_bitmask_impl::<u64, 64>() }
        }
    }

    #[inline]
    pub(crate) fn from_bitmask_integer(bitmask: u64) -> Self {
        // TODO modify simd_bitmask_select to truncate input, making this unnecessary
        if N <= 8 {
            // Safety: bitmask matches length
            unsafe { Self::from_bitmask_impl::<u8, 8>(bitmask as u8) }
        } else if N <= 16 {
            // Safety: bitmask matches length
            unsafe { Self::from_bitmask_impl::<u16, 16>(bitmask as u16) }
        } else if N <= 32 {
            // Safety: bitmask matches length
            unsafe { Self::from_bitmask_impl::<u32, 32>(bitmask as u32) }
        } else {
            // Safety: bitmask matches length
            unsafe { Self::from_bitmask_impl::<u64, 64>(bitmask) }
        }
    }

    #[inline]
    #[must_use = "method returns a new bool and does not mutate the original value"]
    pub fn any(self) -> bool {
        // Safety: use `self` as an integer vector
        unsafe { intrinsics::simd_reduce_any(self.to_int()) }
    }

    #[inline]
    #[must_use = "method returns a new vector and does not mutate the original value"]
    pub fn all(self) -> bool {
        // Safety: use `self` as an integer vector
        unsafe { intrinsics::simd_reduce_all(self.to_int()) }
    }
}

impl<T, const N: usize> From<Mask<T, N>> for Simd<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn from(value: Mask<T, N>) -> Self {
        value.0
    }
}

impl<T, const N: usize> core::ops::BitAnd for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    type Output = Self;
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    fn bitand(self, rhs: Self) -> Self {
        // Safety: `self` is an integer vector
        unsafe { Self(intrinsics::simd_and(self.0, rhs.0)) }
    }
}

impl<T, const N: usize> core::ops::BitOr for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    type Output = Self;
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    fn bitor(self, rhs: Self) -> Self {
        // Safety: `self` is an integer vector
        unsafe { Self(intrinsics::simd_or(self.0, rhs.0)) }
    }
}

impl<T, const N: usize> core::ops::BitXor for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    type Output = Self;
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    fn bitxor(self, rhs: Self) -> Self {
        // Safety: `self` is an integer vector
        unsafe { Self(intrinsics::simd_xor(self.0, rhs.0)) }
    }
}

impl<T, const N: usize> core::ops::Not for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    type Output = Self;
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    fn not(self) -> Self::Output {
        Self::splat(true) ^ self
    }
}