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
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
//! Types and traits associated with masking elements of vectors.
//! Types representing
#![allow(non_camel_case_types)]

#[cfg_attr(
    not(all(target_arch = "x86_64", target_feature = "avx512f")),
    path = "masks/full_masks.rs"
)]
#[cfg_attr(
    all(target_arch = "x86_64", target_feature = "avx512f"),
    path = "masks/bitmask.rs"
)]
mod mask_impl;

use crate::simd::{LaneCount, Simd, SimdCast, SimdElement, SupportedLaneCount};
use core::cmp::Ordering;
use core::{fmt, mem};

mod sealed {
    use super::*;

    /// Not only does this seal the `MaskElement` trait, but these functions prevent other traits
    /// from bleeding into the parent bounds.
    ///
    /// For example, `eq` could be provided by requiring `MaskElement: PartialEq`, but that would
    /// prevent us from ever removing that bound, or from implementing `MaskElement` on
    /// non-`PartialEq` types in the future.
    pub trait Sealed {
        fn valid<const N: usize>(values: Simd<Self, N>) -> bool
        where
            LaneCount<N>: SupportedLaneCount,
            Self: SimdElement;

        fn eq(self, other: Self) -> bool;

        fn to_usize(self) -> usize;
        fn max_unsigned() -> u64;

        type Unsigned: SimdElement;

        const TRUE: Self;

        const FALSE: Self;
    }
}
use sealed::Sealed;

/// Marker trait for types that may be used as SIMD mask elements.
///
/// # Safety
/// Type must be a signed integer.
pub unsafe trait MaskElement: SimdElement<Mask = Self> + SimdCast + Sealed {}

macro_rules! impl_element {
    { $ty:ty, $unsigned:ty } => {
        impl Sealed for $ty {
            #[inline]
            fn valid<const N: usize>(value: Simd<Self, N>) -> bool
            where
                LaneCount<N>: SupportedLaneCount,
            {
                // We can't use `Simd` directly, because `Simd`'s functions call this function and
                // we will end up with an infinite loop.
                // Safety: `value` is an integer vector
                unsafe {
                    use core::intrinsics::simd;
                    let falses: Simd<Self, N> = simd::simd_eq(value, Simd::splat(0 as _));
                    let trues: Simd<Self, N> = simd::simd_eq(value, Simd::splat(-1 as _));
                    let valid: Simd<Self, N> = simd::simd_or(falses, trues);
                    simd::simd_reduce_all(valid)
                }
            }

            #[inline]
            fn eq(self, other: Self) -> bool { self == other }

            #[inline]
            fn to_usize(self) -> usize {
                self as usize
            }

            #[inline]
            fn max_unsigned() -> u64 {
                <$unsigned>::MAX as u64
            }

            type Unsigned = $unsigned;

            const TRUE: Self = -1;
            const FALSE: Self = 0;
        }

        // Safety: this is a valid mask element type
        unsafe impl MaskElement for $ty {}
    }
}

impl_element! { i8, u8 }
impl_element! { i16, u16 }
impl_element! { i32, u32 }
impl_element! { i64, u64 }
impl_element! { isize, usize }

/// A SIMD vector mask for `N` elements of width specified by `Element`.
///
/// Masks represent boolean inclusion/exclusion on a per-element basis.
///
/// The layout of this type is unspecified, and may change between platforms
/// and/or Rust versions, and code should not assume that it is equivalent to
/// `[T; N]`.
#[repr(transparent)]
pub struct Mask<T, const N: usize>(mask_impl::Mask<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]
    fn clone(&self) -> Self {
        *self
    }
}

impl<T, const N: usize> Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    /// Construct a mask by setting all elements to the given value.
    #[inline]
    pub fn splat(value: bool) -> Self {
        Self(mask_impl::Mask::splat(value))
    }

    /// Converts an array of bools to a SIMD mask.
    #[inline]
    pub fn from_array(array: [bool; N]) -> Self {
        // SAFETY: Rust's bool has a layout of 1 byte (u8) with a value of
        //     true:    0b_0000_0001
        //     false:   0b_0000_0000
        // Thus, an array of bools is also a valid array of bytes: [u8; N]
        // This would be hypothetically valid as an "in-place" transmute,
        // but these are "dependently-sized" types, so copy elision it is!
        unsafe {
            let bytes: [u8; N] = mem::transmute_copy(&array);
            let bools: Simd<i8, N> =
                core::intrinsics::simd::simd_ne(Simd::from_array(bytes), Simd::splat(0u8));
            Mask::from_int_unchecked(core::intrinsics::simd::simd_cast(bools))
        }
    }

    /// Converts a SIMD mask to an array of bools.
    #[inline]
    pub fn to_array(self) -> [bool; N] {
        // This follows mostly the same logic as from_array.
        // SAFETY: Rust's bool has a layout of 1 byte (u8) with a value of
        //     true:    0b_0000_0001
        //     false:   0b_0000_0000
        // Thus, an array of bools is also a valid array of bytes: [u8; N]
        // Since our masks are equal to integers where all bits are set,
        // we can simply convert them to i8s, and then bitand them by the
        // bitpattern for Rust's "true" bool.
        // This would be hypothetically valid as an "in-place" transmute,
        // but these are "dependently-sized" types, so copy elision it is!
        unsafe {
            let mut bytes: Simd<i8, N> = core::intrinsics::simd::simd_cast(self.to_int());
            bytes &= Simd::splat(1i8);
            mem::transmute_copy(&bytes)
        }
    }

    /// Converts a vector of integers to a mask, where 0 represents `false` and -1
    /// represents `true`.
    ///
    /// # Safety
    /// All elements must be either 0 or -1.
    #[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 {
        // Safety: the caller must confirm this invariant
        unsafe {
            core::intrinsics::assume(<T as Sealed>::valid(value));
            Self(mask_impl::Mask::from_int_unchecked(value))
        }
    }

    /// Converts a vector of integers to a mask, where 0 represents `false` and -1
    /// represents `true`.
    ///
    /// # Panics
    /// Panics if any element is not 0 or -1.
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    #[track_caller]
    pub fn from_int(value: Simd<T, N>) -> Self {
        assert!(T::valid(value), "all values must be either 0 or -1",);
        // Safety: the validity has been checked
        unsafe { Self::from_int_unchecked(value) }
    }

    /// Converts the mask to a vector of integers, where 0 represents `false` and -1
    /// represents `true`.
    #[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.to_int()
    }

    /// Converts the mask to a mask of any other element size.
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    pub fn cast<U: MaskElement>(self) -> Mask<U, N> {
        Mask(self.0.convert())
    }

    /// Tests the value of the specified element.
    ///
    /// # Safety
    /// `index` must be less than `self.len()`.
    #[inline]
    #[must_use = "method returns a new bool and does not mutate the original value"]
    pub unsafe fn test_unchecked(&self, index: usize) -> bool {
        // Safety: the caller must confirm this invariant
        unsafe { self.0.test_unchecked(index) }
    }

    /// Tests the value of the specified element.
    ///
    /// # Panics
    /// Panics if `index` is greater than or equal to the number of elements in the vector.
    #[inline]
    #[must_use = "method returns a new bool and does not mutate the original value"]
    #[track_caller]
    pub fn test(&self, index: usize) -> bool {
        assert!(index < N, "element index out of range");
        // Safety: the element index has been checked
        unsafe { self.test_unchecked(index) }
    }

    /// Sets the value of the specified element.
    ///
    /// # Safety
    /// `index` must be less than `self.len()`.
    #[inline]
    pub unsafe fn set_unchecked(&mut self, index: usize, value: bool) {
        // Safety: the caller must confirm this invariant
        unsafe {
            self.0.set_unchecked(index, value);
        }
    }

    /// Sets the value of the specified element.
    ///
    /// # Panics
    /// Panics if `index` is greater than or equal to the number of elements in the vector.
    #[inline]
    #[track_caller]
    pub fn set(&mut self, index: usize, value: bool) {
        assert!(index < N, "element index out of range");
        // Safety: the element index has been checked
        unsafe {
            self.set_unchecked(index, value);
        }
    }

    /// Returns true if any element is set, or false otherwise.
    #[inline]
    #[must_use = "method returns a new bool and does not mutate the original value"]
    pub fn any(self) -> bool {
        self.0.any()
    }

    /// Returns true if all elements are set, or false otherwise.
    #[inline]
    #[must_use = "method returns a new bool and does not mutate the original value"]
    pub fn all(self) -> bool {
        self.0.all()
    }

    /// Create a bitmask from a mask.
    ///
    /// Each bit is set if the corresponding element in the mask is `true`.
    /// If the mask contains more than 64 elements, the bitmask is truncated to the first 64.
    #[inline]
    #[must_use = "method returns a new integer and does not mutate the original value"]
    pub fn to_bitmask(self) -> u64 {
        self.0.to_bitmask_integer()
    }

    /// Create a mask from a bitmask.
    ///
    /// For each bit, if it is set, the corresponding element in the mask is set to `true`.
    /// If the mask contains more than 64 elements, the remainder are set to `false`.
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    pub fn from_bitmask(bitmask: u64) -> Self {
        Self(mask_impl::Mask::from_bitmask_integer(bitmask))
    }

    /// Find the index of the first set element.
    ///
    /// ```
    /// # #![feature(portable_simd)]
    /// # #[cfg(feature = "as_crate")] use core_simd::simd;
    /// # #[cfg(not(feature = "as_crate"))] use core::simd;
    /// # use simd::mask32x8;
    /// assert_eq!(mask32x8::splat(false).first_set(), None);
    /// assert_eq!(mask32x8::splat(true).first_set(), Some(0));
    ///
    /// let mask = mask32x8::from_array([false, true, false, false, true, false, false, true]);
    /// assert_eq!(mask.first_set(), Some(1));
    /// ```
    #[inline]
    #[must_use = "method returns the index and does not mutate the original value"]
    pub fn first_set(self) -> Option<usize> {
        // If bitmasks are efficient, using them is better
        if cfg!(target_feature = "sse") && N <= 64 {
            let tz = self.to_bitmask().trailing_zeros();
            return if tz == 64 { None } else { Some(tz as usize) };
        }

        // To find the first set index:
        // * create a vector 0..N
        // * replace unset mask elements in that vector with -1
        // * perform _unsigned_ reduce-min
        // * check if the result is -1 or an index

        let index = Simd::from_array(
            const {
                let mut index = [0; N];
                let mut i = 0;
                while i < N {
                    index[i] = i;
                    i += 1;
                }
                index
            },
        );

        // Safety: the input and output are integer vectors
        let index: Simd<T, N> = unsafe { core::intrinsics::simd::simd_cast(index) };

        let masked_index = self.select(index, Self::splat(true).to_int());

        // Safety: the input and output are integer vectors
        let masked_index: Simd<T::Unsigned, N> =
            unsafe { core::intrinsics::simd::simd_cast(masked_index) };

        // Safety: the input is an integer vector
        let min_index: T::Unsigned =
            unsafe { core::intrinsics::simd::simd_reduce_min(masked_index) };

        // Safety: the return value is the unsigned version of T
        let min_index: T = unsafe { core::mem::transmute_copy(&min_index) };

        if min_index.eq(T::TRUE) {
            None
        } else {
            Some(min_index.to_usize())
        }
    }
}

// vector/array conversion
impl<T, const N: usize> From<[bool; N]> for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn from(array: [bool; N]) -> Self {
        Self::from_array(array)
    }
}

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

impl<T, const N: usize> Default for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    #[must_use = "method returns a defaulted mask with all elements set to false (0)"]
    fn default() -> Self {
        Self::splat(false)
    }
}

impl<T, const N: usize> PartialEq for Mask<T, N>
where
    T: MaskElement + PartialEq,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    #[must_use = "method returns a new bool and does not mutate the original value"]
    fn eq(&self, other: &Self) -> bool {
        self.0 == other.0
    }
}

impl<T, const N: usize> PartialOrd for Mask<T, N>
where
    T: MaskElement + PartialOrd,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    #[must_use = "method returns a new Ordering and does not mutate the original value"]
    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
        self.0.partial_cmp(&other.0)
    }
}

impl<T, const N: usize> fmt::Debug for Mask<T, N>
where
    T: MaskElement + fmt::Debug,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_list()
            .entries((0..N).map(|i| self.test(i)))
            .finish()
    }
}

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 {
        Self(self.0 & rhs.0)
    }
}

impl<T, const N: usize> core::ops::BitAnd<bool> 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: bool) -> Self {
        self & Self::splat(rhs)
    }
}

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

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 {
        Self(self.0 | rhs.0)
    }
}

impl<T, const N: usize> core::ops::BitOr<bool> 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: bool) -> Self {
        self | Self::splat(rhs)
    }
}

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

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::Output {
        Self(self.0 ^ rhs.0)
    }
}

impl<T, const N: usize> core::ops::BitXor<bool> 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: bool) -> Self::Output {
        self ^ Self::splat(rhs)
    }
}

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

impl<T, const N: usize> core::ops::Not for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    type Output = Mask<T, N>;
    #[inline]
    #[must_use = "method returns a new mask and does not mutate the original value"]
    fn not(self) -> Self::Output {
        Self(!self.0)
    }
}

impl<T, const N: usize> core::ops::BitAndAssign for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn bitand_assign(&mut self, rhs: Self) {
        self.0 = self.0 & rhs.0;
    }
}

impl<T, const N: usize> core::ops::BitAndAssign<bool> for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn bitand_assign(&mut self, rhs: bool) {
        *self &= Self::splat(rhs);
    }
}

impl<T, const N: usize> core::ops::BitOrAssign for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn bitor_assign(&mut self, rhs: Self) {
        self.0 = self.0 | rhs.0;
    }
}

impl<T, const N: usize> core::ops::BitOrAssign<bool> for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn bitor_assign(&mut self, rhs: bool) {
        *self |= Self::splat(rhs);
    }
}

impl<T, const N: usize> core::ops::BitXorAssign for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn bitxor_assign(&mut self, rhs: Self) {
        self.0 = self.0 ^ rhs.0;
    }
}

impl<T, const N: usize> core::ops::BitXorAssign<bool> for Mask<T, N>
where
    T: MaskElement,
    LaneCount<N>: SupportedLaneCount,
{
    #[inline]
    fn bitxor_assign(&mut self, rhs: bool) {
        *self ^= Self::splat(rhs);
    }
}

macro_rules! impl_from {
    { $from:ty  => $($to:ty),* } => {
        $(
        impl<const N: usize> From<Mask<$from, N>> for Mask<$to, N>
        where
            LaneCount<N>: SupportedLaneCount,
        {
            #[inline]
            fn from(value: Mask<$from, N>) -> Self {
                value.cast()
            }
        }
        )*
    }
}
impl_from! { i8 => i16, i32, i64, isize }
impl_from! { i16 => i32, i64, isize, i8 }
impl_from! { i32 => i64, isize, i8, i16 }
impl_from! { i64 => isize, i8, i16, i32 }
impl_from! { isize => i8, i16, i32, i64 }