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
//! SIMD compiler intrinsics.
//!
//! In this module, a "vector" is any `repr(simd)` type.

extern "platform-intrinsic" {
    /// Add two simd vectors elementwise.
    ///
    /// `T` must be a vector of integer or floating point primitive types.
    pub fn simd_add<T>(x: T, y: T) -> T;

    /// Subtract `rhs` from `lhs` elementwise.
    ///
    /// `T` must be a vector of integer or floating point primitive types.
    pub fn simd_sub<T>(lhs: T, rhs: T) -> T;

    /// Multiply two simd vectors elementwise.
    ///
    /// `T` must be a vector of integer or floating point primitive types.
    pub fn simd_mul<T>(x: T, y: T) -> T;

    /// Divide `lhs` by `rhs` elementwise.
    ///
    /// `T` must be a vector of integer or floating point primitive types.
    ///
    /// # Safety
    /// For integers, `rhs` must not contain any zero elements.
    /// Additionally for signed integers, `<int>::MIN / -1` is undefined behavior.
    pub fn simd_div<T>(lhs: T, rhs: T) -> T;

    /// Remainder of two vectors elementwise
    ///
    /// `T` must be a vector of integer or floating point primitive types.
    ///
    /// # Safety
    /// For integers, `rhs` must not contain any zero elements.
    /// Additionally for signed integers, `<int>::MIN / -1` is undefined behavior.
    pub fn simd_rem<T>(lhs: T, rhs: T) -> T;

    /// Elementwise vector left shift, with UB on overflow.
    ///
    /// Shift `lhs` left by `rhs`, shifting in sign bits for signed types.
    ///
    /// `T` must be a vector of integer primitive types.
    ///
    /// # Safety
    ///
    /// Each element of `rhs` must be less than `<int>::BITS`.
    pub fn simd_shl<T>(lhs: T, rhs: T) -> T;

    /// Elementwise vector right shift, with UB on overflow.
    ///
    /// `T` must be a vector of integer primitive types.
    ///
    /// Shift `lhs` right by `rhs`, shifting in sign bits for signed types.
    ///
    /// # Safety
    ///
    /// Each element of `rhs` must be less than `<int>::BITS`.
    pub fn simd_shr<T>(lhs: T, rhs: T) -> T;

    /// Elementwise vector "and".
    ///
    /// `T` must be a vector of integer primitive types.
    pub fn simd_and<T>(x: T, y: T) -> T;

    /// Elementwise vector "or".
    ///
    /// `T` must be a vector of integer primitive types.
    pub fn simd_or<T>(x: T, y: T) -> T;

    /// Elementwise vector "exclusive or".
    ///
    /// `T` must be a vector of integer primitive types.
    pub fn simd_xor<T>(x: T, y: T) -> T;

    /// Numerically cast a vector, elementwise.
    ///
    /// `T` and `U` must be vectors of integer or floating point primitive types, and must have the
    /// same length.
    ///
    /// When casting floats to integers, the result is truncated. Out-of-bounds result lead to UB.
    /// When casting integers to floats, the result is rounded.
    /// Otherwise, truncates or extends the value, maintaining the sign for signed integers.
    ///
    /// # Safety
    /// Casting from integer types is always safe.
    /// Casting between two float types is also always safe.
    ///
    /// Casting floats to integers truncates, following the same rules as `to_int_unchecked`.
    /// Specifically, each element must:
    /// * Not be `NaN`
    /// * Not be infinite
    /// * Be representable in the return type, after truncating off its fractional part
    pub fn simd_cast<T, U>(x: T) -> U;

    /// Numerically cast a vector, elementwise.
    ///
    /// `T` and `U` be a vectors of integer or floating point primitive types, and must have the
    /// same length.
    ///
    /// Like `simd_cast`, but saturates float-to-integer conversions (NaN becomes 0).
    /// This matches regular `as` and is always safe.
    ///
    /// When casting floats to integers, the result is truncated.
    /// When casting integers to floats, the result is rounded.
    /// Otherwise, truncates or extends the value, maintaining the sign for signed integers.
    pub fn simd_as<T, U>(x: T) -> U;

    /// Elementwise negation of a vector.
    ///
    /// `T` must be a vector of integer or floating-point primitive types.
    ///
    /// Rust panics for `-<int>::Min` due to overflow, but it is not UB with this intrinsic.
    pub fn simd_neg<T>(x: T) -> T;

    /// Elementwise absolute value of a vector.
    ///
    /// `T` must be a vector of floating-point primitive types.
    pub fn simd_fabs<T>(x: T) -> T;

    /// Elementwise minimum of a vector.
    ///
    /// `T` must be a vector of floating-point primitive types.
    ///
    /// Follows IEEE-754 `minNum` semantics.
    pub fn simd_fmin<T>(x: T, y: T) -> T;

    /// Elementwise maximum of a vector.
    ///
    /// `T` must be a vector of floating-point primitive types.
    ///
    /// Follows IEEE-754 `maxNum` semantics.
    pub fn simd_fmax<T>(x: T, y: T) -> T;

    /// Tests elementwise equality of two vectors.
    ///
    /// `T` must be a vector of floating-point primitive types.
    ///
    /// `U` must be a vector of integers with the same number of elements and element size as `T`.
    ///
    /// Returns `0` for false and `!0` for true.
    pub fn simd_eq<T, U>(x: T, y: T) -> U;

    /// Tests elementwise inequality equality of two vectors.
    ///
    /// `T` must be a vector of floating-point primitive types.
    ///
    /// `U` must be a vector of integers with the same number of elements and element size as `T`.
    ///
    /// Returns `0` for false and `!0` for true.
    pub fn simd_ne<T, U>(x: T, y: T) -> U;

    /// Tests if `x` is less than `y`, elementwise.
    ///
    /// `T` must be a vector of floating-point primitive types.
    ///
    /// `U` must be a vector of integers with the same number of elements and element size as `T`.
    ///
    /// Returns `0` for false and `!0` for true.
    pub fn simd_lt<T, U>(x: T, y: T) -> U;

    /// Tests if `x` is less than or equal to `y`, elementwise.
    ///
    /// `T` must be a vector of floating-point primitive types.
    ///
    /// `U` must be a vector of integers with the same number of elements and element size as `T`.
    ///
    /// Returns `0` for false and `!0` for true.
    pub fn simd_le<T, U>(x: T, y: T) -> U;

    /// Tests if `x` is greater than `y`, elementwise.
    ///
    /// `T` must be a vector of floating-point primitive types.
    ///
    /// `U` must be a vector of integers with the same number of elements and element size as `T`.
    ///
    /// Returns `0` for false and `!0` for true.
    pub fn simd_gt<T, U>(x: T, y: T) -> U;

    /// Tests if `x` is greater than or equal to `y`, elementwise.
    ///
    /// `T` must be a vector of floating-point primitive types.
    ///
    /// `U` must be a vector of integers with the same number of elements and element size as `T`.
    ///
    /// Returns `0` for false and `!0` for true.
    pub fn simd_ge<T, U>(x: T, y: T) -> U;

    /// Shuffle two vectors by const indices.
    ///
    /// `T` must be a vector.
    ///
    /// `U` must be a const array of `i32`s.
    ///
    /// `V` must be a vector with the same element type as `T` and the same length as `U`.
    ///
    /// Concatenates `x` and `y`, then returns a new vector such that each element is selected from
    /// the concatenation by the matching index in `idx`.
    pub fn simd_shuffle<T, U, V>(x: T, y: T, idx: U) -> V;

    /// Read a vector of pointers.
    ///
    /// `T` must be a vector.
    ///
    /// `U` must be a vector of pointers to the element type of `T`, with the same length as `T`.
    ///
    /// `V` must be a vector of integers with the same length as `T` (but any element size).
    ///
    /// `idx` must be a constant: either naming a constant item, or an inline
    /// `const {}` expression.
    ///
    /// For each pointer in `ptr`, if the corresponding value in `mask` is `!0`, read the pointer.
    /// Otherwise if the corresponding value in `mask` is `0`, return the corresponding value from
    /// `val`.
    ///
    /// # Safety
    /// Unmasked values in `T` must be readable as if by `<ptr>::read` (e.g. aligned to the element
    /// type).
    ///
    /// `mask` must only contain `0` or `!0` values.
    pub fn simd_gather<T, U, V>(val: T, ptr: U, mask: V) -> T;

    /// Write to a vector of pointers.
    ///
    /// `T` must be a vector.
    ///
    /// `U` must be a vector of pointers to the element type of `T`, with the same length as `T`.
    ///
    /// `V` must be a vector of integers with the same length as `T` (but any element size).
    ///
    /// For each pointer in `ptr`, if the corresponding value in `mask` is `!0`, write the
    /// corresponding value in `val` to the pointer.
    /// Otherwise if the corresponding value in `mask` is `0`, do nothing.
    ///
    /// # Safety
    /// Unmasked values in `T` must be writeable as if by `<ptr>::write` (e.g. aligned to the element
    /// type).
    ///
    /// `mask` must only contain `0` or `!0` values.
    pub fn simd_scatter<T, U, V>(val: T, ptr: U, mask: V);

    /// Read a vector of pointers.
    ///
    /// `T` must be a vector.
    ///
    /// `U` must be a vector of pointers to the element type of `T`, with the same length as `T`.
    ///
    /// `V` must be a vector of integers with the same length as `T` (but any element size).
    ///
    /// For each element, if the corresponding value in `mask` is `!0`, read the corresponding
    /// pointer from `ptr`.
    /// Otherwise if the corresponding value in `mask` is `0`, return the corresponding value from
    /// `val`.
    ///
    /// # Safety
    /// Unmasked values in `T` must be readable as if by `<ptr>::read` (e.g. aligned to the element
    /// type).
    ///
    /// `mask` must only contain `0` or `!0` values.
    pub fn simd_masked_load<V, U, T>(mask: V, ptr: U, val: T) -> T;

    /// Write to a vector of pointers.
    ///
    /// `T` must be a vector.
    ///
    /// `U` must be a vector of pointers to the element type of `T`, with the same length as `T`.
    ///
    /// `V` must be a vector of integers with the same length as `T` (but any element size).
    ///
    /// For each element, if the corresponding value in `mask` is `!0`, write the corresponding
    /// value in `val` to the pointer.
    /// Otherwise if the corresponding value in `mask` is `0`, do nothing.
    ///
    /// # Safety
    /// Unmasked values in `T` must be writeable as if by `<ptr>::write` (e.g. aligned to the element
    /// type).
    ///
    /// `mask` must only contain `0` or `!0` values.
    pub fn simd_masked_store<V, U, T>(mask: V, ptr: U, val: T);

    /// Add two simd vectors elementwise, with saturation.
    ///
    /// `T` must be a vector of integer primitive types.
    pub fn simd_saturating_add<T>(x: T, y: T) -> T;

    /// Subtract two simd vectors elementwise, with saturation.
    ///
    /// `T` must be a vector of integer primitive types.
    ///
    /// Subtract `rhs` from `lhs`.
    pub fn simd_saturating_sub<T>(lhs: T, rhs: T) -> T;

    /// Add elements within a vector from left to right.
    ///
    /// `T` must be a vector of integer or floating-point primitive types.
    ///
    /// `U` must be the element type of `T`.
    ///
    /// Starting with the value `y`, add the elements of `x` and accumulate.
    pub fn simd_reduce_add_ordered<T, U>(x: T, y: U) -> U;

    /// Multiply elements within a vector from left to right.
    ///
    /// `T` must be a vector of integer or floating-point primitive types.
    ///
    /// `U` must be the element type of `T`.
    ///
    /// Starting with the value `y`, multiply the elements of `x` and accumulate.
    pub fn simd_reduce_mul_ordered<T, U>(x: T, y: U) -> U;

    /// Check if all mask values are true.
    ///
    /// `T` must be a vector of integer primitive types.
    ///
    /// # Safety
    /// `x` must contain only `0` or `!0`.
    pub fn simd_reduce_all<T>(x: T) -> bool;

    /// Check if all mask values are true.
    ///
    /// `T` must be a vector of integer primitive types.
    ///
    /// # Safety
    /// `x` must contain only `0` or `!0`.
    pub fn simd_reduce_any<T>(x: T) -> bool;

    /// Return the maximum element of a vector.
    ///
    /// `T` must be a vector of integer or floating-point primitive types.
    ///
    /// `U` must be the element type of `T`.
    ///
    /// For floating-point values, uses IEEE-754 `maxNum`.
    pub fn simd_reduce_max<T, U>(x: T) -> U;

    /// Return the minimum element of a vector.
    ///
    /// `T` must be a vector of integer or floating-point primitive types.
    ///
    /// `U` must be the element type of `T`.
    ///
    /// For floating-point values, uses IEEE-754 `minNum`.
    pub fn simd_reduce_min<T, U>(x: T) -> U;

    /// Logical "and" all elements together.
    ///
    /// `T` must be a vector of integer or floating-point primitive types.
    ///
    /// `U` must be the element type of `T`.
    pub fn simd_reduce_and<T, U>(x: T) -> U;

    /// Logical "or" all elements together.
    ///
    /// `T` must be a vector of integer or floating-point primitive types.
    ///
    /// `U` must be the element type of `T`.
    pub fn simd_reduce_or<T, U>(x: T) -> U;

    /// Logical "exclusive or" all elements together.
    ///
    /// `T` must be a vector of integer or floating-point primitive types.
    ///
    /// `U` must be the element type of `T`.
    pub fn simd_reduce_xor<T, U>(x: T) -> U;

    /// Truncate an integer vector to a bitmask.
    ///
    /// `T` must be an integer vector.
    ///
    /// `U` must be either the smallest unsigned integer with at least as many bits as the length
    /// of `T`, or the smallest array of `u8` with as many bits as the length of `T`.
    ///
    /// Each element is truncated to a single bit and packed into the result.
    ///
    /// No matter whether the output is an array or an unsigned integer, it is treated as a single
    /// contiguous list of bits. The bitmask is always packed on the least-significant side of the
    /// output, and padded with 0s in the most-significant bits. The order of the bits depends on
    /// endianess:
    ///
    /// * On little endian, the least significant bit corresponds to the first vector element.
    /// * On big endian, the least significant bit corresponds to the last vector element.
    ///
    /// For example, `[!0, 0, !0, !0]` packs to `0b1101` on little endian and `0b1011` on big
    /// endian.
    ///
    /// To consider a larger example, `[!0, 0, 0, 0, 0, 0, 0, 0, !0, !0, 0, 0, 0, 0, !0, 0]` packs
    /// to `[0b00000001, 0b01000011]` or `0b0100001100000001` on little endian, and `[0b10000000,
    /// 0b11000010]` or `0b1000000011000010` on big endian.
    ///
    /// # Safety
    /// `x` must contain only `0` and `!0`.
    pub fn simd_bitmask<T, U>(x: T) -> U;

    /// Select elements from a mask.
    ///
    /// `M` must be an integer vector.
    ///
    /// `T` must be a vector with the same number of elements as `M`.
    ///
    /// For each element, if the corresponding value in `mask` is `!0`, select the element from
    /// `if_true`.  If the corresponding value in `mask` is `0`, select the element from
    /// `if_false`.
    ///
    /// # Safety
    /// `mask` must only contain `0` and `!0`.
    pub fn simd_select<M, T>(mask: M, if_true: T, if_false: T) -> T;

    /// Select elements from a bitmask.
    ///
    /// `M` must be an unsigned integer or array of `u8`, matching `simd_bitmask`.
    ///
    /// `T` must be a vector.
    ///
    /// For each element, if the bit in `mask` is `1`, select the element from
    /// `if_true`.  If the corresponding bit in `mask` is `0`, select the element from
    /// `if_false`.
    ///
    /// The bitmask bit order matches `simd_bitmask`.
    ///
    /// # Safety
    /// Padding bits must be all zero.
    pub fn simd_select_bitmask<M, T>(m: M, yes: T, no: T) -> T;

    /// Elementwise calculates the offset from a pointer vector, potentially wrapping.
    ///
    /// `T` must be a vector of pointers.
    ///
    /// `U` must be a vector of `isize` or `usize` with the same number of elements as `T`.
    ///
    /// Operates as if by `<ptr>::wrapping_offset`.
    pub fn simd_arith_offset<T, U>(ptr: T, offset: U) -> T;

    /// Cast a vector of pointers.
    ///
    /// `T` and `U` must be vectors of pointers with the same number of elements.
    pub fn simd_cast_ptr<T, U>(ptr: T) -> U;

    /// Expose a vector of pointers as a vector of addresses.
    ///
    /// `T` must be a vector of pointers.
    ///
    /// `U` must be a vector of `usize` with the same length as `T`.
    pub fn simd_expose_addr<T, U>(ptr: T) -> U;

    /// Create a vector of pointers from a vector of addresses.
    ///
    /// `T` must be a vector of `usize`.
    ///
    /// `U` must be a vector of pointers, with the same length as `T`.
    pub fn simd_from_exposed_addr<T, U>(addr: T) -> U;

    /// Swap bytes of each element.
    ///
    /// `T` must be a vector of integers.
    pub fn simd_bswap<T>(x: T) -> T;

    /// Reverse bits of each element.
    ///
    /// `T` must be a vector of integers.
    pub fn simd_bitreverse<T>(x: T) -> T;

    /// Count the leading zeros of each element.
    ///
    /// `T` must be a vector of integers.
    pub fn simd_ctlz<T>(x: T) -> T;

    /// Count the trailing zeros of each element.
    ///
    /// `T` must be a vector of integers.
    pub fn simd_cttz<T>(x: T) -> T;
}