1 | use super::addition::__add2; |
2 | use super::{cmp_slice, BigUint}; |
3 | |
4 | use crate::big_digit::{self, BigDigit, DoubleBigDigit}; |
5 | use crate::UsizePromotion; |
6 | |
7 | use core::cmp::Ordering::{Equal, Greater, Less}; |
8 | use core::mem; |
9 | use core::ops::{Div, DivAssign, Rem, RemAssign}; |
10 | use num_integer::Integer; |
11 | use num_traits::{CheckedDiv, CheckedEuclid, Euclid, One, ToPrimitive, Zero}; |
12 | |
13 | pub(super) const FAST_DIV_WIDE: bool = cfg!(any(target_arch = "x86" , target_arch = "x86_64" )); |
14 | |
15 | /// Divide a two digit numerator by a one digit divisor, returns quotient and remainder: |
16 | /// |
17 | /// Note: the caller must ensure that both the quotient and remainder will fit into a single digit. |
18 | /// This is _not_ true for an arbitrary numerator/denominator. |
19 | /// |
20 | /// (This function also matches what the x86 divide instruction does). |
21 | #[cfg (any(miri, not(any(target_arch = "x86" , target_arch = "x86_64" ))))] |
22 | #[inline ] |
23 | fn div_wide(hi: BigDigit, lo: BigDigit, divisor: BigDigit) -> (BigDigit, BigDigit) { |
24 | debug_assert!(hi < divisor); |
25 | |
26 | let lhs = big_digit::to_doublebigdigit(hi, lo); |
27 | let rhs = DoubleBigDigit::from(divisor); |
28 | ((lhs / rhs) as BigDigit, (lhs % rhs) as BigDigit) |
29 | } |
30 | |
31 | /// x86 and x86_64 can use a real `div` instruction. |
32 | #[cfg (all(not(miri), any(target_arch = "x86" , target_arch = "x86_64" )))] |
33 | #[inline ] |
34 | fn div_wide(hi: BigDigit, lo: BigDigit, divisor: BigDigit) -> (BigDigit, BigDigit) { |
35 | // This debug assertion covers the potential #DE for divisor==0 or a quotient too large for one |
36 | // register, otherwise in release mode it will become a target-specific fault like SIGFPE. |
37 | // This should never occur with the inputs from our few `div_wide` callers. |
38 | debug_assert!(hi < divisor); |
39 | |
40 | // SAFETY: The `div` instruction only affects registers, reading the explicit operand as the |
41 | // divisor, and implicitly reading RDX:RAX or EDX:EAX as the dividend. The result is implicitly |
42 | // written back to RAX or EAX for the quotient and RDX or EDX for the remainder. No memory is |
43 | // used, and flags are not preserved. |
44 | unsafe { |
45 | let (div, rem); |
46 | |
47 | cfg_digit!( |
48 | macro_rules! div { |
49 | () => { |
50 | "div {0:e}" |
51 | }; |
52 | } |
53 | macro_rules! div { |
54 | () => { |
55 | "div {0:r}" |
56 | }; |
57 | } |
58 | ); |
59 | |
60 | core::arch::asm!( |
61 | div!(), |
62 | in(reg) divisor, |
63 | inout("dx" ) hi => rem, |
64 | inout("ax" ) lo => div, |
65 | options(pure, nomem, nostack), |
66 | ); |
67 | |
68 | (div, rem) |
69 | } |
70 | } |
71 | |
72 | /// For small divisors, we can divide without promoting to `DoubleBigDigit` by |
73 | /// using half-size pieces of digit, like long-division. |
74 | #[inline ] |
75 | fn div_half(rem: BigDigit, digit: BigDigit, divisor: BigDigit) -> (BigDigit, BigDigit) { |
76 | use crate::big_digit::{HALF, HALF_BITS}; |
77 | |
78 | debug_assert!(rem < divisor && divisor <= HALF); |
79 | let (hi: u64, rem: u64) = ((rem << HALF_BITS) | (digit >> HALF_BITS)).div_rem(&divisor); |
80 | let (lo: u64, rem: u64) = ((rem << HALF_BITS) | (digit & HALF)).div_rem(&divisor); |
81 | ((hi << HALF_BITS) | lo, rem) |
82 | } |
83 | |
84 | #[inline ] |
85 | pub(super) fn div_rem_digit(mut a: BigUint, b: BigDigit) -> (BigUint, BigDigit) { |
86 | if b == 0 { |
87 | panic!("attempt to divide by zero" ) |
88 | } |
89 | |
90 | let mut rem: u64 = 0; |
91 | |
92 | if !FAST_DIV_WIDE && b <= big_digit::HALF { |
93 | for d: &mut u64 in a.data.iter_mut().rev() { |
94 | let (q: u64, r: u64) = div_half(rem, *d, divisor:b); |
95 | *d = q; |
96 | rem = r; |
97 | } |
98 | } else { |
99 | for d: &mut u64 in a.data.iter_mut().rev() { |
100 | let (q: u64, r: u64) = div_wide(hi:rem, *d, divisor:b); |
101 | *d = q; |
102 | rem = r; |
103 | } |
104 | } |
105 | |
106 | (a.normalized(), rem) |
107 | } |
108 | |
109 | #[inline ] |
110 | fn rem_digit(a: &BigUint, b: BigDigit) -> BigDigit { |
111 | if b == 0 { |
112 | panic!("attempt to divide by zero" ) |
113 | } |
114 | |
115 | let mut rem: u64 = 0; |
116 | |
117 | if !FAST_DIV_WIDE && b <= big_digit::HALF { |
118 | for &digit: u64 in a.data.iter().rev() { |
119 | let (_, r: u64) = div_half(rem, digit, divisor:b); |
120 | rem = r; |
121 | } |
122 | } else { |
123 | for &digit: u64 in a.data.iter().rev() { |
124 | let (_, r: u64) = div_wide(hi:rem, lo:digit, divisor:b); |
125 | rem = r; |
126 | } |
127 | } |
128 | |
129 | rem |
130 | } |
131 | |
132 | /// Subtract a multiple. |
133 | /// a -= b * c |
134 | /// Returns a borrow (if a < b then borrow > 0). |
135 | fn sub_mul_digit_same_len(a: &mut [BigDigit], b: &[BigDigit], c: BigDigit) -> BigDigit { |
136 | debug_assert!(a.len() == b.len()); |
137 | |
138 | // carry is between -big_digit::MAX and 0, so to avoid overflow we store |
139 | // offset_carry = carry + big_digit::MAX |
140 | let mut offset_carry = big_digit::MAX; |
141 | |
142 | for (x, y) in a.iter_mut().zip(b) { |
143 | // We want to calculate sum = x - y * c + carry. |
144 | // sum >= -(big_digit::MAX * big_digit::MAX) - big_digit::MAX |
145 | // sum <= big_digit::MAX |
146 | // Offsetting sum by (big_digit::MAX << big_digit::BITS) puts it in DoubleBigDigit range. |
147 | let offset_sum = big_digit::to_doublebigdigit(big_digit::MAX, *x) |
148 | - big_digit::MAX as DoubleBigDigit |
149 | + offset_carry as DoubleBigDigit |
150 | - *y as DoubleBigDigit * c as DoubleBigDigit; |
151 | |
152 | let (new_offset_carry, new_x) = big_digit::from_doublebigdigit(offset_sum); |
153 | offset_carry = new_offset_carry; |
154 | *x = new_x; |
155 | } |
156 | |
157 | // Return the borrow. |
158 | big_digit::MAX - offset_carry |
159 | } |
160 | |
161 | fn div_rem(mut u: BigUint, mut d: BigUint) -> (BigUint, BigUint) { |
162 | if d.is_zero() { |
163 | panic!("attempt to divide by zero" ) |
164 | } |
165 | if u.is_zero() { |
166 | return (BigUint::ZERO, BigUint::ZERO); |
167 | } |
168 | |
169 | if d.data.len() == 1 { |
170 | if d.data == [1] { |
171 | return (u, BigUint::ZERO); |
172 | } |
173 | let (div, rem) = div_rem_digit(u, d.data[0]); |
174 | // reuse d |
175 | d.data.clear(); |
176 | d += rem; |
177 | return (div, d); |
178 | } |
179 | |
180 | // Required or the q_len calculation below can underflow: |
181 | match u.cmp(&d) { |
182 | Less => return (BigUint::ZERO, u), |
183 | Equal => { |
184 | u.set_one(); |
185 | return (u, BigUint::ZERO); |
186 | } |
187 | Greater => {} // Do nothing |
188 | } |
189 | |
190 | // This algorithm is from Knuth, TAOCP vol 2 section 4.3, algorithm D: |
191 | // |
192 | // First, normalize the arguments so the highest bit in the highest digit of the divisor is |
193 | // set: the main loop uses the highest digit of the divisor for generating guesses, so we |
194 | // want it to be the largest number we can efficiently divide by. |
195 | // |
196 | let shift = d.data.last().unwrap().leading_zeros() as usize; |
197 | |
198 | if shift == 0 { |
199 | // no need to clone d |
200 | div_rem_core(u, &d.data) |
201 | } else { |
202 | let (q, r) = div_rem_core(u << shift, &(d << shift).data); |
203 | // renormalize the remainder |
204 | (q, r >> shift) |
205 | } |
206 | } |
207 | |
208 | pub(super) fn div_rem_ref(u: &BigUint, d: &BigUint) -> (BigUint, BigUint) { |
209 | if d.is_zero() { |
210 | panic!("attempt to divide by zero" ) |
211 | } |
212 | if u.is_zero() { |
213 | return (BigUint::ZERO, BigUint::ZERO); |
214 | } |
215 | |
216 | if d.data.len() == 1 { |
217 | if d.data == [1] { |
218 | return (u.clone(), BigUint::ZERO); |
219 | } |
220 | |
221 | let (div, rem) = div_rem_digit(u.clone(), d.data[0]); |
222 | return (div, rem.into()); |
223 | } |
224 | |
225 | // Required or the q_len calculation below can underflow: |
226 | match u.cmp(d) { |
227 | Less => return (BigUint::ZERO, u.clone()), |
228 | Equal => return (One::one(), BigUint::ZERO), |
229 | Greater => {} // Do nothing |
230 | } |
231 | |
232 | // This algorithm is from Knuth, TAOCP vol 2 section 4.3, algorithm D: |
233 | // |
234 | // First, normalize the arguments so the highest bit in the highest digit of the divisor is |
235 | // set: the main loop uses the highest digit of the divisor for generating guesses, so we |
236 | // want it to be the largest number we can efficiently divide by. |
237 | // |
238 | let shift = d.data.last().unwrap().leading_zeros() as usize; |
239 | |
240 | if shift == 0 { |
241 | // no need to clone d |
242 | div_rem_core(u.clone(), &d.data) |
243 | } else { |
244 | let (q, r) = div_rem_core(u << shift, &(d << shift).data); |
245 | // renormalize the remainder |
246 | (q, r >> shift) |
247 | } |
248 | } |
249 | |
250 | /// An implementation of the base division algorithm. |
251 | /// Knuth, TAOCP vol 2 section 4.3.1, algorithm D, with an improvement from exercises 19-21. |
252 | fn div_rem_core(mut a: BigUint, b: &[BigDigit]) -> (BigUint, BigUint) { |
253 | debug_assert!(a.data.len() >= b.len() && b.len() > 1); |
254 | debug_assert!(b.last().unwrap().leading_zeros() == 0); |
255 | |
256 | // The algorithm works by incrementally calculating "guesses", q0, for the next digit of the |
257 | // quotient. Once we have any number q0 such that (q0 << j) * b <= a, we can set |
258 | // |
259 | // q += q0 << j |
260 | // a -= (q0 << j) * b |
261 | // |
262 | // and then iterate until a < b. Then, (q, a) will be our desired quotient and remainder. |
263 | // |
264 | // q0, our guess, is calculated by dividing the last three digits of a by the last two digits of |
265 | // b - this will give us a guess that is close to the actual quotient, but is possibly greater. |
266 | // It can only be greater by 1 and only in rare cases, with probability at most |
267 | // 2^-(big_digit::BITS-1) for random a, see TAOCP 4.3.1 exercise 21. |
268 | // |
269 | // If the quotient turns out to be too large, we adjust it by 1: |
270 | // q -= 1 << j |
271 | // a += b << j |
272 | |
273 | // a0 stores an additional extra most significant digit of the dividend, not stored in a. |
274 | let mut a0 = 0; |
275 | |
276 | // [b1, b0] are the two most significant digits of the divisor. They never change. |
277 | let b0 = b[b.len() - 1]; |
278 | let b1 = b[b.len() - 2]; |
279 | |
280 | let q_len = a.data.len() - b.len() + 1; |
281 | let mut q = BigUint { |
282 | data: vec![0; q_len], |
283 | }; |
284 | |
285 | for j in (0..q_len).rev() { |
286 | debug_assert!(a.data.len() == b.len() + j); |
287 | |
288 | let a1 = *a.data.last().unwrap(); |
289 | let a2 = a.data[a.data.len() - 2]; |
290 | |
291 | // The first q0 estimate is [a1,a0] / b0. It will never be too small, it may be too large |
292 | // by at most 2. |
293 | let (mut q0, mut r) = if a0 < b0 { |
294 | let (q0, r) = div_wide(a0, a1, b0); |
295 | (q0, r as DoubleBigDigit) |
296 | } else { |
297 | debug_assert!(a0 == b0); |
298 | // Avoid overflowing q0, we know the quotient fits in BigDigit. |
299 | // [a1,a0] = b0 * (1<<BITS - 1) + (a0 + a1) |
300 | (big_digit::MAX, a0 as DoubleBigDigit + a1 as DoubleBigDigit) |
301 | }; |
302 | |
303 | // r = [a1,a0] - q0 * b0 |
304 | // |
305 | // Now we want to compute a more precise estimate [a2,a1,a0] / [b1,b0] which can only be |
306 | // less or equal to the current q0. |
307 | // |
308 | // q0 is too large if: |
309 | // [a2,a1,a0] < q0 * [b1,b0] |
310 | // (r << BITS) + a2 < q0 * b1 |
311 | while r <= big_digit::MAX as DoubleBigDigit |
312 | && big_digit::to_doublebigdigit(r as BigDigit, a2) |
313 | < q0 as DoubleBigDigit * b1 as DoubleBigDigit |
314 | { |
315 | q0 -= 1; |
316 | r += b0 as DoubleBigDigit; |
317 | } |
318 | |
319 | // q0 is now either the correct quotient digit, or in rare cases 1 too large. |
320 | // Subtract (q0 << j) from a. This may overflow, in which case we will have to correct. |
321 | |
322 | let mut borrow = sub_mul_digit_same_len(&mut a.data[j..], b, q0); |
323 | if borrow > a0 { |
324 | // q0 is too large. We need to add back one multiple of b. |
325 | q0 -= 1; |
326 | borrow -= __add2(&mut a.data[j..], b); |
327 | } |
328 | // The top digit of a, stored in a0, has now been zeroed. |
329 | debug_assert!(borrow == a0); |
330 | |
331 | q.data[j] = q0; |
332 | |
333 | // Pop off the next top digit of a. |
334 | a0 = a.data.pop().unwrap(); |
335 | } |
336 | |
337 | a.data.push(a0); |
338 | a.normalize(); |
339 | |
340 | debug_assert_eq!(cmp_slice(&a.data, b), Less); |
341 | |
342 | (q.normalized(), a) |
343 | } |
344 | |
345 | forward_val_ref_binop!(impl Div for BigUint, div); |
346 | forward_ref_val_binop!(impl Div for BigUint, div); |
347 | forward_val_assign!(impl DivAssign for BigUint, div_assign); |
348 | |
349 | impl Div<BigUint> for BigUint { |
350 | type Output = BigUint; |
351 | |
352 | #[inline ] |
353 | fn div(self, other: BigUint) -> BigUint { |
354 | let (q: BigUint, _) = div_rem(self, d:other); |
355 | q |
356 | } |
357 | } |
358 | |
359 | impl Div<&BigUint> for &BigUint { |
360 | type Output = BigUint; |
361 | |
362 | #[inline ] |
363 | fn div(self, other: &BigUint) -> BigUint { |
364 | let (q: BigUint, _) = self.div_rem(other); |
365 | q |
366 | } |
367 | } |
368 | impl DivAssign<&BigUint> for BigUint { |
369 | #[inline ] |
370 | fn div_assign(&mut self, other: &BigUint) { |
371 | *self = &*self / other; |
372 | } |
373 | } |
374 | |
375 | promote_unsigned_scalars!(impl Div for BigUint, div); |
376 | promote_unsigned_scalars_assign!(impl DivAssign for BigUint, div_assign); |
377 | forward_all_scalar_binop_to_val_val!(impl Div<u32> for BigUint, div); |
378 | forward_all_scalar_binop_to_val_val!(impl Div<u64> for BigUint, div); |
379 | forward_all_scalar_binop_to_val_val!(impl Div<u128> for BigUint, div); |
380 | |
381 | impl Div<u32> for BigUint { |
382 | type Output = BigUint; |
383 | |
384 | #[inline ] |
385 | fn div(self, other: u32) -> BigUint { |
386 | let (q: BigUint, _) = div_rem_digit(self, b:other as BigDigit); |
387 | q |
388 | } |
389 | } |
390 | impl DivAssign<u32> for BigUint { |
391 | #[inline ] |
392 | fn div_assign(&mut self, other: u32) { |
393 | *self = &*self / other; |
394 | } |
395 | } |
396 | |
397 | impl Div<BigUint> for u32 { |
398 | type Output = BigUint; |
399 | |
400 | #[inline ] |
401 | fn div(self, other: BigUint) -> BigUint { |
402 | match other.data.len() { |
403 | 0 => panic!("attempt to divide by zero" ), |
404 | 1 => From::from(self as BigDigit / other.data[0]), |
405 | _ => BigUint::ZERO, |
406 | } |
407 | } |
408 | } |
409 | |
410 | impl Div<u64> for BigUint { |
411 | type Output = BigUint; |
412 | |
413 | #[inline ] |
414 | fn div(self, other: u64) -> BigUint { |
415 | let (q: BigUint, _) = div_rem(self, d:From::from(other)); |
416 | q |
417 | } |
418 | } |
419 | impl DivAssign<u64> for BigUint { |
420 | #[inline ] |
421 | fn div_assign(&mut self, other: u64) { |
422 | // a vec of size 0 does not allocate, so this is fairly cheap |
423 | let temp: BigUint = mem::replace(self, Self::ZERO); |
424 | *self = temp / other; |
425 | } |
426 | } |
427 | |
428 | impl Div<BigUint> for u64 { |
429 | type Output = BigUint; |
430 | |
431 | cfg_digit!( |
432 | #[inline ] |
433 | fn div(self, other: BigUint) -> BigUint { |
434 | match other.data.len() { |
435 | 0 => panic!("attempt to divide by zero" ), |
436 | 1 => From::from(self / u64::from(other.data[0])), |
437 | 2 => From::from(self / big_digit::to_doublebigdigit(other.data[1], other.data[0])), |
438 | _ => BigUint::ZERO, |
439 | } |
440 | } |
441 | |
442 | #[inline ] |
443 | fn div(self, other: BigUint) -> BigUint { |
444 | match other.data.len() { |
445 | 0 => panic!("attempt to divide by zero" ), |
446 | 1 => From::from(self / other.data[0]), |
447 | _ => BigUint::ZERO, |
448 | } |
449 | } |
450 | ); |
451 | } |
452 | |
453 | impl Div<u128> for BigUint { |
454 | type Output = BigUint; |
455 | |
456 | #[inline ] |
457 | fn div(self, other: u128) -> BigUint { |
458 | let (q: BigUint, _) = div_rem(self, d:From::from(other)); |
459 | q |
460 | } |
461 | } |
462 | |
463 | impl DivAssign<u128> for BigUint { |
464 | #[inline ] |
465 | fn div_assign(&mut self, other: u128) { |
466 | *self = &*self / other; |
467 | } |
468 | } |
469 | |
470 | impl Div<BigUint> for u128 { |
471 | type Output = BigUint; |
472 | |
473 | cfg_digit!( |
474 | #[inline ] |
475 | fn div(self, other: BigUint) -> BigUint { |
476 | use super::u32_to_u128; |
477 | match other.data.len() { |
478 | 0 => panic!("attempt to divide by zero" ), |
479 | 1 => From::from(self / u128::from(other.data[0])), |
480 | 2 => From::from( |
481 | self / u128::from(big_digit::to_doublebigdigit(other.data[1], other.data[0])), |
482 | ), |
483 | 3 => From::from(self / u32_to_u128(0, other.data[2], other.data[1], other.data[0])), |
484 | 4 => From::from( |
485 | self / u32_to_u128(other.data[3], other.data[2], other.data[1], other.data[0]), |
486 | ), |
487 | _ => BigUint::ZERO, |
488 | } |
489 | } |
490 | |
491 | #[inline ] |
492 | fn div(self, other: BigUint) -> BigUint { |
493 | match other.data.len() { |
494 | 0 => panic!("attempt to divide by zero" ), |
495 | 1 => From::from(self / other.data[0] as u128), |
496 | 2 => From::from(self / big_digit::to_doublebigdigit(other.data[1], other.data[0])), |
497 | _ => BigUint::ZERO, |
498 | } |
499 | } |
500 | ); |
501 | } |
502 | |
503 | forward_val_ref_binop!(impl Rem for BigUint, rem); |
504 | forward_ref_val_binop!(impl Rem for BigUint, rem); |
505 | forward_val_assign!(impl RemAssign for BigUint, rem_assign); |
506 | |
507 | impl Rem<BigUint> for BigUint { |
508 | type Output = BigUint; |
509 | |
510 | #[inline ] |
511 | fn rem(self, other: BigUint) -> BigUint { |
512 | if let Some(other: u32) = other.to_u32() { |
513 | &self % other |
514 | } else { |
515 | let (_, r: BigUint) = div_rem(self, d:other); |
516 | r |
517 | } |
518 | } |
519 | } |
520 | |
521 | impl Rem<&BigUint> for &BigUint { |
522 | type Output = BigUint; |
523 | |
524 | #[inline ] |
525 | fn rem(self, other: &BigUint) -> BigUint { |
526 | if let Some(other: u32) = other.to_u32() { |
527 | self % other |
528 | } else { |
529 | let (_, r: BigUint) = self.div_rem(other); |
530 | r |
531 | } |
532 | } |
533 | } |
534 | impl RemAssign<&BigUint> for BigUint { |
535 | #[inline ] |
536 | fn rem_assign(&mut self, other: &BigUint) { |
537 | *self = &*self % other; |
538 | } |
539 | } |
540 | |
541 | promote_unsigned_scalars!(impl Rem for BigUint, rem); |
542 | promote_unsigned_scalars_assign!(impl RemAssign for BigUint, rem_assign); |
543 | forward_all_scalar_binop_to_ref_val!(impl Rem<u32> for BigUint, rem); |
544 | forward_all_scalar_binop_to_val_val!(impl Rem<u64> for BigUint, rem); |
545 | forward_all_scalar_binop_to_val_val!(impl Rem<u128> for BigUint, rem); |
546 | |
547 | impl Rem<u32> for &BigUint { |
548 | type Output = BigUint; |
549 | |
550 | #[inline ] |
551 | fn rem(self, other: u32) -> BigUint { |
552 | rem_digit(self, b:other as BigDigit).into() |
553 | } |
554 | } |
555 | impl RemAssign<u32> for BigUint { |
556 | #[inline ] |
557 | fn rem_assign(&mut self, other: u32) { |
558 | *self = &*self % other; |
559 | } |
560 | } |
561 | |
562 | impl Rem<&BigUint> for u32 { |
563 | type Output = BigUint; |
564 | |
565 | #[inline ] |
566 | fn rem(mut self, other: &BigUint) -> BigUint { |
567 | self %= other; |
568 | From::from(self) |
569 | } |
570 | } |
571 | |
572 | macro_rules! impl_rem_assign_scalar { |
573 | ($scalar:ty, $to_scalar:ident) => { |
574 | forward_val_assign_scalar!(impl RemAssign for BigUint, $scalar, rem_assign); |
575 | impl RemAssign<&BigUint> for $scalar { |
576 | #[inline] |
577 | fn rem_assign(&mut self, other: &BigUint) { |
578 | *self = match other.$to_scalar() { |
579 | None => *self, |
580 | Some(0) => panic!("attempt to divide by zero" ), |
581 | Some(v) => *self % v |
582 | }; |
583 | } |
584 | } |
585 | } |
586 | } |
587 | |
588 | // we can scalar %= BigUint for any scalar, including signed types |
589 | impl_rem_assign_scalar!(u128, to_u128); |
590 | impl_rem_assign_scalar!(usize, to_usize); |
591 | impl_rem_assign_scalar!(u64, to_u64); |
592 | impl_rem_assign_scalar!(u32, to_u32); |
593 | impl_rem_assign_scalar!(u16, to_u16); |
594 | impl_rem_assign_scalar!(u8, to_u8); |
595 | impl_rem_assign_scalar!(i128, to_i128); |
596 | impl_rem_assign_scalar!(isize, to_isize); |
597 | impl_rem_assign_scalar!(i64, to_i64); |
598 | impl_rem_assign_scalar!(i32, to_i32); |
599 | impl_rem_assign_scalar!(i16, to_i16); |
600 | impl_rem_assign_scalar!(i8, to_i8); |
601 | |
602 | impl Rem<u64> for BigUint { |
603 | type Output = BigUint; |
604 | |
605 | #[inline ] |
606 | fn rem(self, other: u64) -> BigUint { |
607 | let (_, r: BigUint) = div_rem(self, d:From::from(other)); |
608 | r |
609 | } |
610 | } |
611 | impl RemAssign<u64> for BigUint { |
612 | #[inline ] |
613 | fn rem_assign(&mut self, other: u64) { |
614 | *self = &*self % other; |
615 | } |
616 | } |
617 | |
618 | impl Rem<BigUint> for u64 { |
619 | type Output = BigUint; |
620 | |
621 | #[inline ] |
622 | fn rem(mut self, other: BigUint) -> BigUint { |
623 | self %= other; |
624 | From::from(self) |
625 | } |
626 | } |
627 | |
628 | impl Rem<u128> for BigUint { |
629 | type Output = BigUint; |
630 | |
631 | #[inline ] |
632 | fn rem(self, other: u128) -> BigUint { |
633 | let (_, r: BigUint) = div_rem(self, d:From::from(other)); |
634 | r |
635 | } |
636 | } |
637 | |
638 | impl RemAssign<u128> for BigUint { |
639 | #[inline ] |
640 | fn rem_assign(&mut self, other: u128) { |
641 | *self = &*self % other; |
642 | } |
643 | } |
644 | |
645 | impl Rem<BigUint> for u128 { |
646 | type Output = BigUint; |
647 | |
648 | #[inline ] |
649 | fn rem(mut self, other: BigUint) -> BigUint { |
650 | self %= other; |
651 | From::from(self) |
652 | } |
653 | } |
654 | |
655 | impl CheckedDiv for BigUint { |
656 | #[inline ] |
657 | fn checked_div(&self, v: &BigUint) -> Option<BigUint> { |
658 | if v.is_zero() { |
659 | return None; |
660 | } |
661 | Some(self.div(v)) |
662 | } |
663 | } |
664 | |
665 | impl CheckedEuclid for BigUint { |
666 | #[inline ] |
667 | fn checked_div_euclid(&self, v: &BigUint) -> Option<BigUint> { |
668 | if v.is_zero() { |
669 | return None; |
670 | } |
671 | Some(self.div_euclid(v)) |
672 | } |
673 | |
674 | #[inline ] |
675 | fn checked_rem_euclid(&self, v: &BigUint) -> Option<BigUint> { |
676 | if v.is_zero() { |
677 | return None; |
678 | } |
679 | Some(self.rem_euclid(v)) |
680 | } |
681 | |
682 | fn checked_div_rem_euclid(&self, v: &Self) -> Option<(Self, Self)> { |
683 | Some(self.div_rem_euclid(v)) |
684 | } |
685 | } |
686 | |
687 | impl Euclid for BigUint { |
688 | #[inline ] |
689 | fn div_euclid(&self, v: &BigUint) -> BigUint { |
690 | // trivially same as regular division |
691 | self / v |
692 | } |
693 | |
694 | #[inline ] |
695 | fn rem_euclid(&self, v: &BigUint) -> BigUint { |
696 | // trivially same as regular remainder |
697 | self % v |
698 | } |
699 | |
700 | fn div_rem_euclid(&self, v: &Self) -> (Self, Self) { |
701 | // trivially same as regular division and remainder |
702 | self.div_rem(v) |
703 | } |
704 | } |
705 | |