gcds 1.0.0

Efficient implementations of gcd algorithms
Documentation
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
// The following doc comment is kept in sync with the `README.md` file.
// Please run `cargo sync-readme` after modifying its contents.
//! This crate implements several algorithms for finding the greatest
//! common divisor of two single-precision numbers.
//!
//! The _greatest common divisor_ $\gcd(u,v)$ of two integers $u$ and $v$,
//! not both zero, is the largest integer that evenly divides them both.
//! This definition does not apply when $u$ and $v$ are both zero, since
//! every number divides zero; for convenience, all the algorithms adhere
//! to the convention that $\gcd(0,0)=0$.

/// Computes the greatest common divisor of $u$ and $v$ using the modern
/// version of the Euclidean algorithm, as described in Algorithm 4.5.2A
/// of _TAOCP_.
///
/// # Examples
/// ```
/// use gcds::euclid;
///
/// assert_eq!(euclid(0, 0), 0);
/// assert_eq!(euclid(2, 0), 2);
/// assert_eq!(euclid(1, 1), 1);
/// assert_eq!(euclid(1, 13), 1);
/// assert_eq!(euclid(1769, 551), 29);
/// assert_eq!(euclid(7000, 4400), 200);
/// assert_eq!(euclid(40902, 24140), 34);
/// assert_eq!(euclid(u64::MAX, u64::MAX - 1), 1);
/// assert_eq!(euclid(u64::MAX, u64::MAX), u64::MAX);
///
/// // Find the gcd of two signed integers.
/// assert_eq!(euclid((-9i64).unsigned_abs(), 3), 3);
/// ```
#[must_use]
pub const fn euclid(mut u: u64, mut v: u64) -> u64 {
  // If $v=0$, we have $\gcd(u,0)=u$.
  while v != 0 {
    // Note that $\gcd(u,v)=\gcd(v,u\bmod v)$, because
    // 1. there is an integer $q$ such that $u\bmod v=u-qv$,
    // 2. any number that divides $u$ and $v$ also divides $u-qv$, and
    // 3. any number that divides $v$ and $u-qv$ also divides $u$.
    (u, v) = (v, u % v);
  }
  u
}

/// Computes the greatest common divisor of $u$ and $v$ using the binary
/// gcd algorithm of J. Stein [["Computational Problems Associated with Racah Algebra,"][stein]
/// _J. Comp. Phys._ **1** (1967), 397--405].
///
/// A summary of what was known up to 1999 about the average running time
/// of this method appears in R. P. Brent's [_"Further analysis of the Binary Euclidean algorithm,"_][brent]
/// Technical Report PRG TR-7--99 (Oxford Univ. Computing Laboratory, 1999).
/// See also Section 4.5.2 of _TAOCP_.
///
/// # Examples
/// ```
/// use gcds::binary_stein;
///
/// assert_eq!(binary_stein(0, 0), 0);
/// assert_eq!(binary_stein(0, 3), 3);
/// assert_eq!(binary_stein(1, 1), 1);
/// assert_eq!(binary_stein(5, 1), 1);
/// assert_eq!(binary_stein(114, 551), 19);
/// assert_eq!(binary_stein(1992, 581), 83);
/// assert_eq!(binary_stein(51041, 18017), 43);
/// assert_eq!(binary_stein(u64::MAX, u64::MAX - 1), 1);
/// assert_eq!(binary_stein(u64::MAX, u64::MAX), u64::MAX);
///
/// // Find the gcd of two signed numbers.
/// assert_eq!(binary_stein(36, (-48i64).unsigned_abs()), 12);
/// ```
///
/// [stein]: https://doi.org/10.1016/0021-9991(67)90047-2
/// [brent]: https://maths-people.anu.edu.au/~brent/pd/rpb183tr.pdf
#[must_use]
pub fn binary_stein(mut u: u64, mut v: u64) -> u64 {
  if u == 0 {
    return v;
  }
  if v == 0 {
    return u;
  }
  // If $u$ and $v$ are both even, then $\gcd(u,v)=2\gcd(u/2,v/2)$.
  // It follows by induction that $\gcd(u,v)=2^k\gcd(u/2^k,v/2^k)$,
  // where $2^k$ is a common divisor of $u$ and $v$. Also, if $u$
  // is even and $v$ is odd, then $\gcd(u,v)=\gcd(u/2,v)$. More
  // generally, $\gcd(u,v)=\gcd(u/2^{k_u},v)$ if $u$ is a multiple
  // of $2^{k_u}$ and $v$ is odd. Combine these results to prove
  // that $\gcd(u,v)=2^k\gcd(u/2^{k_u},v/2^{k_v})$, whenever $u$
  // is a multiple of $2^{k_u}$, $v$ is a multiple of $2^{k_v}$,
  // and $k=\min(k_u,k_v)$.
  let k = (u | v).trailing_zeros();
  // SAFETY: Since $u$ and $v$ are nonzero, $k_u$ and $k_v$ are
  //         less than `u64::BITS`.
  u >>= u.trailing_zeros();
  v >>= v.trailing_zeros();
  loop {
    // At the beginning of each iteration of this loop, we have
    // the invariant condition that both $u$ and $v$ are odd.
    // Ensure that $u\le v$; note that $\gcd(u,v)=\gcd(v,u)$.
    if u > v {
      (u, v) = (v, u);
    }
    // As in the Euclidean algorithm, $\gcd(u,v)=\gcd(u,v-u)$.
    v -= u;
    if v == 0 {
      return u << k; // Terminate with $u\cdot2^k$.
    }
    // Now $v$ is even, and $u$ remains odd. Therefore we can apply the
    // identity $\gcd(u,v)=\gcd(u,v/2^{k_v})$ again. After this step, the
    // assertion that $u$ and $v$ are both odd at the start of the next
    // iteration is true.
    // SAFETY: $v$ is nonzero, so it has at least one 1 bit.
    v >>= v.trailing_zeros();
  }
}

/// Computes the greatest common divisor of $u$ and $v$ using an optimized
/// variant of the [binary gcd algorithm] proposed by P. Bonzini, and studied
/// by D. Lemire [["Greatest common divisor, the extended Euclidean algorithm, and speed!"][lemire]
/// (April 2024)].
///
/// # Examples
/// ```
/// use gcds::binary_bonzini;
///
/// assert_eq!(binary_bonzini(0, 0), 0);
/// assert_eq!(binary_bonzini(2, 0), 2);
/// assert_eq!(binary_bonzini(1, 1), 1);
/// assert_eq!(binary_bonzini(1, 4), 1);
/// assert_eq!(binary_bonzini(792, 252), 36);
/// assert_eq!(binary_bonzini(954, 5883), 159);
/// assert_eq!(binary_bonzini(14003, 63194), 19);
/// assert_eq!(binary_bonzini(u64::MAX, u64::MAX - 1), 1);
/// assert_eq!(binary_bonzini(u64::MAX, u64::MAX), u64::MAX);
///
/// // Find the gcd of two signed numbers.
/// assert_eq!(binary_bonzini((-8i64).unsigned_abs(), 76), 4);
/// ```
///
/// [binary gcd algorithm]: binary_stein
/// [lemire]: https://lemire.me/blog/2024/04/13/greatest-common-divisor-the-extended-euclidean-algorithm-and-speed/
pub fn binary_bonzini(mut u: u64, mut v: u64) -> u64 {
  if u == 0 {
    return v;
  }
  if v == 0 {
    return u;
  }
  // As in the original binary gcd algorithm, we begin by applying the
  // formula $\gcd(u,v)=2^k\gcd(u/2^{k_u},v/2^{k_v})$, where $2^{k_u}$
  // and $2^{k_v}$ are the largest powers of 2 dividing $u$ and $v$,
  // respectively, and $k=\min(k_u,k_v)$. The only difference is that
  // the assignment "$v\gets v/2^{k_v}$" appears within the main loop,
  // consolidated with the same operation appearing at the end of the
  // original version of the loop.
  let k = (u | v).trailing_zeros();
  u >>= u.trailing_zeros();
  let mut k_v = v.trailing_zeros();
  loop {
    v >>= k_v;
    // In essence, the updating rule of the binary gcd algorithm is
    // $(u,v)\gets(\min(u,v),|u-v|)$. (As we have already observed,
    // this transformation leaves $\gcd(u,v)$ unchanged.) Perhaps
    // the most obvious way to find the new value of $v$ is to use
    // `u.abs_diff(v)` or the equivalent sequence of operations in
    // `binary_stein`. Here we start by computing $u-v$ instead:
    let (diff, neg_diff) = u.overflowing_sub(v);
    // Why? In the first place, the termination condition $|u-v|=0$
    // of the original algorithm holds if and only if $u-v=0$. The
    // latter quantity is easier to compute on most architectures.
    if diff == 0 {
      return u << k;
    }
    // In the second place, $|u-v|$ and $u-v$ have the same dyadic
    // valuation $k_v$ (equivalently, the greatest power of $2$
    // that divides each of them is the same, $2^{k_v}$). So we
    // can compute $k_v$ directly from $u-v$, without having to
    // wait for the result of $|u-v|$ to become available.
    k_v = diff.trailing_zeros();
    u = u.min(v);
    // Actually compute $|u-v|$ from `diff` and `neg_diff`. Note
    // that a smart compiler will lower the required conditional
    // branch to a `cmov` instruction on x86 and a `cneg` on ARM,
    // much like what it would produce for `u.abs_diff(v)`.
    v = if neg_diff { diff.wrapping_neg() } else { diff };
    // After this operation, we recover the invariant condition that
    // $u$ is odd and $v$ is even. The next iteration will reduce $v$
    // by applying the identity $\gcd(u,v)=2^{k_v}\gcd(u,v/2^{k_v})$.
  }
}

/// Computes the greatest common divisor of $u$ and $v$ using an alternative
/// formulation of the binary gcd algorithm proposed by R. P. Brent
/// [[_"Further analysis of the Binary Euclidean algorithm,"_][brent]
/// Technical Report PRG TR-7--99 (Oxford Univ. Computing Laboratory, 1999),
/// Section 5].
///
/// # Examples
/// ```
/// use gcds::binary_brent;
///
/// assert_eq!(binary_brent(0, 0), 0);
/// assert_eq!(binary_brent(0, 4), 4);
/// assert_eq!(binary_brent(1, 1), 1);
/// assert_eq!(binary_brent(1, 7), 1);
/// assert_eq!(binary_brent(306, 5049), 153);
/// assert_eq!(binary_brent(7684, 3400), 68);
/// assert_eq!(binary_brent(68825, 10150), 25);
/// assert_eq!(binary_brent(u64::MAX, u64::MAX - 1), 1);
/// assert_eq!(binary_brent(u64::MAX, u64::MAX), u64::MAX);
///
/// // Find the gcd of two signed integers.
/// assert_eq!(
///   binary_brent((-10i64).unsigned_abs(), (-5i64).unsigned_abs()),
///   5
/// );
/// ```
///
/// [brent]: https://maths-people.anu.edu.au/~brent/pd/rpb183tr.pdf
pub fn binary_brent(mut u: u64, mut v: u64) -> u64 {
  if u == 0 {
    return v;
  }
  if v == 0 {
    return u;
  }
  // A proof of correctness of this algorithm looks almost identical to
  // that in `binary_stein`, the original version of the binary method.
  // First recall that $\gcd(u,v)=2^k\gcd(u/2^{k_u},v/2^{k_v})$, where
  // $2^{k_u}$ and $2^{k_v}$ are the greatest powers of 2 that divide
  // $u$ and $v$, respectively, and $k=\min(k_u,k_v)$. (The integers
  // $k_u$ and $k_v$ are sometimes known as the _dyadic valuations_
  // or the _ruler functions_ of $u$ and $v$, respectively.)
  let k = (u | v).trailing_zeros();
  // SAFETY: Since $u$ and $v$ are nonzero, $k_u$ and $k_v$ are
  //         less than `u64::BITS`.
  u >>= u.trailing_zeros();
  v >>= v.trailing_zeros();
  // Ensure that $u\le v$ on entry to R. P. Brent's Algorithm V.
  if u > v {
    (u, v) = (v, u);
  }
  while u != v {
    while u < v {
      // Note: This inner loop satisfies the invariant condition that
      // $u$ and $v$ are both odd at the beginning of each iteration.
      // Apply the identity $\gcd(u,v)=\gcd(u,v-u)$.
      v -= u;
      // Now $u$ is odd and $v$ is even; we can use the fact that
      // $\gcd(u,v)=\gcd(u,v/2^{k_v})$, where $2^{k_v}$ divides $v$.
      // SAFETY: $v$ is nonzero, because it was greater than $u$ before
      //         the subtraction step.
      v >>= v.trailing_zeros();
      // At this point $v$ is odd, as required by the invariant.
    }
    (u, v) = (v, u);
  }
  // We have $\gcd(u,v)=u$; terminate with $u\cdot2^k$.
  u << k
}

/// Computes the greatest common divisor of $u$ and $v$ using the systolic
/// variant of the binary gcd algorithm by R. P. Brent and H. T. Kung [_IEEE
/// Symposium on Computer Arithmetic_ **7** (1985), [118--125][systolic]].
///
/// Exercise 4.5.2.40 of _TAOCP_ studies the worst-case running time of
/// this method.
///
/// # Examples
/// ```
/// use gcds::binary_brent_kung;
///
/// assert_eq!(binary_brent_kung(0, 0), 0);
/// assert_eq!(binary_brent_kung(-3, 0), 3);
/// assert_eq!(binary_brent_kung(1, 1), 1);
/// assert_eq!(binary_brent_kung(1, -6), 1);
/// assert_eq!(binary_brent_kung(-45, -63), 9);
/// assert_eq!(binary_brent_kung(4899, 690), 69);
/// assert_eq!(binary_brent_kung(-4515, 9765), 105);
/// assert_eq!(binary_brent_kung(75850, 20213), 41);
/// assert_eq!(binary_brent_kung(i64::MAX - 1, i64::MAX), 1);
/// assert_eq!(binary_brent_kung(i64::MAX, i64::MAX), i64::MAX as u64);
/// assert_eq!(binary_brent_kung(i64::MAX, -i64::MAX), i64::MAX as u64);
/// ```
///
/// [systolic]: https://maths-people.anu.edu.au/brent/pd/rpb077i.pdf
#[must_use]
pub fn binary_brent_kung(mut u: i64, mut v: i64) -> u64 {
  if u == 0 {
    return v.unsigned_abs();
  }
  // This variant of the binary gcd algorithm avoids the need to test
  // the sign of $u-v$ by maintaining (loose) upper bounds $|u|\le2^p$
  // and $|v|\le2^q$, and replacing the test ``$u>v$'' by the weaker
  // ``$p>q$.'' The important thing is that if $u$ and $v$ are $n$-bit
  // numbers, then $p$ and $q$ fit in at most $\floor{\lg n}+1$ bits.
  // Thus it is easier to test the sign of $p-q$ than of $u-v$. Also,
  // since no other step of the algorithm involves the variables $p$
  // and $q$, we can work directly with their difference. R. Brent and
  // H. T. Kung exploited these ideas to implement the new algorithm
  // on a _systolic array_, a collection of primitive data processing
  // units that work in parallel and communicate with their neighbors.

  // It should be noted that the original binary algorithm performs
  // at most $n+1$ iterations of its main loop (see exercise 4.5.2.37
  // of _TAOCP_), whereas its systolic counterpart might need up to
  // $2n+2$ (see exercise 4.5.2.40). Thus there is a tradeoff between
  // the number of iterations and their individual speed. (Of course
  // this tradeoff applies only when the test ``$p>q$'' is performed
  // efficiently; since the present routine operates on full computer
  // words, it is actually significantly slower than `binary_stein`.
  // We will content ourselves with this unoptimized implementation,
  // because it illustrates the salient features of the systolic
  // algorithm and the properties of the $\gcd$ function that it
  // relies on.)

  // Ensure that $u$ is odd by using the by-now-familiar identity
  // $\gcd(u,v)=2^k\gcd(u/2^k_u,v/2^k_v)$, where $k_u$ and $k_v$
  // are the dyadic valuations of $u$ and $v$, and $k=\min(k_u,k_v)$.
  // Unlike the other gcd methods, this routine accepts _signed_
  // integers as input. Rust works with two's complement notation,
  // so `x.trailing_zeros()` gives the value of $k_x$ whenever $x$
  // is a nonzero integer; the case $x=v=0$ is handled below.
  let (k_u, mut k_v) = (u.trailing_zeros(), v.trailing_zeros());
  u >>= k_u;
  let k = k_u.min(k_v);

  // Variable $d$ stores the difference between the upper bounds $p$
  // and $q$ on $\lg|u|$ and $\lg|v|$, respectively. We have $p=q=n$
  // upon entry to the algorithm, and $p=n-k_u$ after dividing $u$ by
  // $2^{k_u}$. If $u$ and $v$ are expected to have wildly different
  // orders of magnitude, it is advantageous to initialize $d$ with
  // the exact difference $\ceil{\lg|u|}-\ceil{\lg|v|}$, but that
  // expression is more difficult to evaluate on a systolic array.
  let mut d = -(k_u as i32);
  while v != 0 {
    // Each iteration of this loop begins with $u$ odd and $v$ even.
    v >>= k_v;
    d += k_v as i32;
    // Swap $u\leftrightarrow v$ if the bound on $|u|$ is larger
    // than the bound on $|v|$.
    if d > 0 {
      (u, v) = (v, u);
      d = -d;
    }
    // At this point both $u$ and $v$ are odd, and it is likely that
    // $|u|\le|v|$. Unfortunately the latter condition is not strong
    // enough to guarantee that setting $v\gets v-u$ now (as in the
    // original binary algorithm) will make the algorithm converge.
    // Consider what happens if $u=-1$ and $v=1$, for example. One
    // solution suitable for use in a systolic array is to replace
    // $v$ by whichever of $w=(u+v)/2$ or $z=(u-v)/2$ is even. (Note
    // that $w$ and $z$ have different parity, because $z=w-v$ and
    // $v$ is odd.) To show that $\gcd(u,v)=\gcd(u,w)=\gcd(u,z)$,
    // observe that any (odd) common divisor of $u$ and $v$ is a
    // divisor of both $u+v$ and $u-v$, and hence of $w$ and $z$;
    // conversely, any common divisor of $u$ and $w$ (or $u$ and
    // $z$) is odd and must divide both $u$ and $v$. It remains to
    // prove that the algorithm always terminates: By the triangle
    // inequality and the fact that $d\le0$ (i.e., $p\le q$) after
    // the previous step, $w$ and $z$ are at most $(2^p+2^q)/2\le2^q$
    // in absolute value. The new value of $v$ is even, so the next
    // iteration of this loop will decrease $q$ by at least 1; it
    // follows by induction that $v$ eventually becomes zero.

    // Compute the average of $u$ and $v$ truncated towards zero,
    // using a popular formula from the book _Hacker's Delight_ by
    // H.~S.~Warren, Jr., 2nd edition (2012), Section 2.5. Since
    // $u$ and $v$ are odd, the correction term $(w\gg63)\&(u\oplus v)$
    // is zero; thus this is slightly faster than the built-in
    // `i64::midpoint` method.
    let w = (u & v) + ((u ^ v) >> 1);
    if w % 2 == 0 {
      v = w;
    } else {
      v = w - v; // $z$.
    }
    k_v = v.trailing_zeros();
  }
  // Now $v=0$, so the algorithm terminates with $|u|2^k$.
  u.unsigned_abs() << k
}

/// Computes the greatest common divisor of $u$ and $v$ using a cross
/// between [Euclid's algorithm](euclid) and the [binary gcd algorithm]
/// proposed by V. C. Harris \[_The Fibonacci Quarterly_ **8** (1970),
/// [102--103][harris]].
///
/// See Section 4.5.2 of _TAOCP_ for a short discussion of this method.
///
/// # Examples
/// ```
/// use gcds::harris;
///
/// assert_eq!(harris(0, 0), 0);
/// assert_eq!(harris(0, 3), 3);
/// assert_eq!(harris(1, 1), 1);
/// assert_eq!(harris(1, 4), 1);
/// assert_eq!(harris(45, 165), 15);
/// assert_eq!(harris(6119, 2175), 29);
/// assert_eq!(harris(69336, 82818), 1926);
/// assert_eq!(harris(u64::MAX, u64::MAX - 1), 1);
/// assert_eq!(harris(u64::MAX, u64::MAX), u64::MAX);
///
/// // Find the gcd of two signed integers.
/// assert_eq!(harris(18, (-27i64).unsigned_abs()), 9);
/// ```
///
/// [binary gcd algorithm]: binary_stein
/// [harris]: https://www.fq.math.ca/Scanned/8-1/harris1.pdf
#[must_use]
pub fn harris(mut u: u64, mut v: u64) -> u64 {
  if u == 0 {
    return v;
  }
  if v == 0 {
    return u;
  }
  // Like the binary gcd algorithm, the main loop of this scheme assumes
  // that $u$ and $v$ are both odd at the beginning of each iteration.
  // Apply the familiar identity $\gcd(u,v)=2^k\gcd(u/2^{k_u},v/2^{k_v})$,
  // where $k_u$ and $k_v$ are the dyadic valuations of $u$ and $v$,
  // and $k=\min(k_u,k_v)$. After this step, $u$ and $v$ become odd.
  let k = (u | v).trailing_zeros();
  u >>= u.trailing_zeros();
  v >>= v.trailing_zeros();
  // This scheme also requires that $0<v\le u$ upon entry to the main
  // loop. Obviously $\gcd(u,v)=\gcd(v,u)$.
  if u < v {
    (u, v) = (v, u);
  }
  loop {
    // The comment in `euclid` shows that $\gcd(u,v)=\gcd(v,u\bmod v)$,
    // so we may replace $u$ by its remainder modulo $v$.
    (u, v) = (v, u % v);
    if v == 0 {
      // Clearly $\gcd(u,0)=u$, and the answer is $u\cdot2^k$.
      return u << k;
    }
    // That was the "Euclidean" part of the algorithm; now let's turn
    // to the "binary" part. Since $0<v<u$ and $u$ is odd, either $v$
    // or $u-v$ is a positive even integer less than $u$. Now observe
    // that $\gcd(u,u-v)=\gcd(u,(u-v)\bmod u)=\gcd(u,v)$. Therefore
    // we can always replace $v$ by an even integer in $(0,u)$ that
    // has the same gcd with $u$.
    if v % 2 == 1 {
      v = u - v;
    }
    // SAFETY: See the previous comment.
    unsafe { core::hint::assert_unchecked(v > 0) };
    // At this point $u$ is odd and $v$ is even. If $k_v$ denotes the
    // dyadic valuation of $v$, then $\gcd(u,v)=2^{k_v}\gcd(u,v/2^{k_v})$.
    // Note that this step makes $v$ odd and preserves the inequality
    // $0<v\le u$.
    v >>= v.trailing_zeros();
  }
}

#[cfg(test)]
mod tests {
  use super::*;

  fn test_methods<I>(cases: I)
  where
    I: IntoIterator<Item = ((u64, u64), u64)>,
  {
    let unsigned_methods: [fn(u64, u64) -> u64; _] =
      [euclid, binary_stein, binary_bonzini, binary_brent, harris];
    let signed_methods: [fn(i64, i64) -> u64; _] = [binary_brent_kung];
    for ((u, v), expected) in cases {
      for method in unsigned_methods {
        assert_eq!(method(u, v), expected, "gcd({u}, {v})");
      }
      let Ok(u) = u.try_into() else { continue };
      let Ok(v) = v.try_into() else { continue };
      for method in signed_methods {
        assert_eq!(method(u, v), expected, "gcd({u}, {v})");
      }
    }
  }

  #[test]
  fn small_numbers() {
    test_methods([
      ((0, 0), 0),
      ((0, 1), 1),
      ((1, 2), 1),
      ((1, 5), 1),
      ((2, 4), 2),
      ((3, 8), 1),
      ((4, 8), 4),
      ((12, 54), 6),
      ((30, 70), 10),
      ((76, 95), 19),
      ((64, 128), 64),
      ((119, 544), 17),
      ((233, 377), 1),
      ((610, 2584), 2),
      ((512, 1024), 512),
      ((1989, 3003), 39),
      ((2166, 6099), 57),
      ((5046, 8004), 174),
      ((1 << 11, 1 << 13), 1 << 11),
    ]);
  }

  #[test]
  fn mersenne_numbers() {
    // Use the fact that $\gcd(2^u-1,2^v-1)=2^{\gcd(u,v)}-1$
    // for nonnegative integers $u$ and $v$.
    let exps = 0..u64::BITS;
    test_methods(exps.clone().zip(exps).map(|(u, v)| {
      let gcd_u_v = euclid(u as u64, v as u64);
      (((1 << u) - 1, (1 << v) - 1), (1 << gcd_u_v) - 1)
    }));
  }

  #[test]
  fn extreme_cases() {
    const MAX_SIGNED: u64 = i64::MAX as u64;
    test_methods([
      ((0, MAX_SIGNED), MAX_SIGNED),
      ((0, u64::MAX), u64::MAX),
      ((1, MAX_SIGNED), 1),
      ((1, u64::MAX), 1),
      ((MAX_SIGNED, MAX_SIGNED - 2), 1),
      ((u64::MAX, u64::MAX - 2), 1),
      ((MAX_SIGNED, MAX_SIGNED), MAX_SIGNED),
      ((u64::MAX, u64::MAX), u64::MAX),
    ]);
  }

  #[test]
  fn random_nonnegative_pairs() {
    use rand::prelude::*;
    let rng = SmallRng::seed_from_u64(2025);
    let pairs = rng
      .random_iter::<u128>()
      .take(10_000_000)
      .map(|x| (x as u64, (x >> u64::BITS) as u64));
    test_methods(pairs.map(|(u, v)| {
      let gcd = euclid(u, v);
      ((u, v), gcd)
    }));
  }
}