| 1 | use crate::big_digit::{self, BigDigit}; |
| 2 | |
| 3 | use alloc::string::String; |
| 4 | use alloc::vec::Vec; |
| 5 | use core::cmp; |
| 6 | use core::cmp::Ordering; |
| 7 | use core::default::Default; |
| 8 | use core::fmt; |
| 9 | use core::hash; |
| 10 | use core::mem; |
| 11 | use core::str; |
| 12 | |
| 13 | use num_integer::{Integer, Roots}; |
| 14 | use num_traits::{ConstZero, Num, One, Pow, ToPrimitive, Unsigned, Zero}; |
| 15 | |
| 16 | mod addition; |
| 17 | mod division; |
| 18 | mod multiplication; |
| 19 | mod subtraction; |
| 20 | |
| 21 | mod arbitrary; |
| 22 | mod bits; |
| 23 | mod convert; |
| 24 | mod iter; |
| 25 | mod monty; |
| 26 | mod power; |
| 27 | mod serde; |
| 28 | mod shift; |
| 29 | |
| 30 | pub(crate) use self::convert::to_str_radix_reversed; |
| 31 | pub use self::iter::{U32Digits, U64Digits}; |
| 32 | |
| 33 | /// A big unsigned integer type. |
| 34 | pub struct BigUint { |
| 35 | data: Vec<BigDigit>, |
| 36 | } |
| 37 | |
| 38 | // Note: derived `Clone` doesn't specialize `clone_from`, |
| 39 | // but we want to keep the allocation in `data`. |
| 40 | impl Clone for BigUint { |
| 41 | #[inline ] |
| 42 | fn clone(&self) -> Self { |
| 43 | BigUint { |
| 44 | data: self.data.clone(), |
| 45 | } |
| 46 | } |
| 47 | |
| 48 | #[inline ] |
| 49 | fn clone_from(&mut self, other: &Self) { |
| 50 | self.data.clone_from(&other.data); |
| 51 | } |
| 52 | } |
| 53 | |
| 54 | impl hash::Hash for BigUint { |
| 55 | #[inline ] |
| 56 | fn hash<H: hash::Hasher>(&self, state: &mut H) { |
| 57 | debug_assert!(self.data.last() != Some(&0)); |
| 58 | self.data.hash(state); |
| 59 | } |
| 60 | } |
| 61 | |
| 62 | impl PartialEq for BigUint { |
| 63 | #[inline ] |
| 64 | fn eq(&self, other: &BigUint) -> bool { |
| 65 | debug_assert!(self.data.last() != Some(&0)); |
| 66 | debug_assert!(other.data.last() != Some(&0)); |
| 67 | self.data == other.data |
| 68 | } |
| 69 | } |
| 70 | impl Eq for BigUint {} |
| 71 | |
| 72 | impl PartialOrd for BigUint { |
| 73 | #[inline ] |
| 74 | fn partial_cmp(&self, other: &BigUint) -> Option<Ordering> { |
| 75 | Some(self.cmp(other)) |
| 76 | } |
| 77 | } |
| 78 | |
| 79 | impl Ord for BigUint { |
| 80 | #[inline ] |
| 81 | fn cmp(&self, other: &BigUint) -> Ordering { |
| 82 | cmp_slice(&self.data[..], &other.data[..]) |
| 83 | } |
| 84 | } |
| 85 | |
| 86 | #[inline ] |
| 87 | fn cmp_slice(a: &[BigDigit], b: &[BigDigit]) -> Ordering { |
| 88 | debug_assert!(a.last() != Some(&0)); |
| 89 | debug_assert!(b.last() != Some(&0)); |
| 90 | |
| 91 | match Ord::cmp(&a.len(), &b.len()) { |
| 92 | Ordering::Equal => Iterator::cmp(self:a.iter().rev(), other:b.iter().rev()), |
| 93 | other: Ordering => other, |
| 94 | } |
| 95 | } |
| 96 | |
| 97 | impl Default for BigUint { |
| 98 | #[inline ] |
| 99 | fn default() -> BigUint { |
| 100 | Self::ZERO |
| 101 | } |
| 102 | } |
| 103 | |
| 104 | impl fmt::Debug for BigUint { |
| 105 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
| 106 | fmt::Display::fmt(self, f) |
| 107 | } |
| 108 | } |
| 109 | |
| 110 | impl fmt::Display for BigUint { |
| 111 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
| 112 | f.pad_integral(is_nonnegative:true, prefix:"" , &self.to_str_radix(10)) |
| 113 | } |
| 114 | } |
| 115 | |
| 116 | impl fmt::LowerHex for BigUint { |
| 117 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
| 118 | f.pad_integral(is_nonnegative:true, prefix:"0x" , &self.to_str_radix(16)) |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | impl fmt::UpperHex for BigUint { |
| 123 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
| 124 | let mut s: String = self.to_str_radix(16); |
| 125 | s.make_ascii_uppercase(); |
| 126 | f.pad_integral(is_nonnegative:true, prefix:"0x" , &s) |
| 127 | } |
| 128 | } |
| 129 | |
| 130 | impl fmt::Binary for BigUint { |
| 131 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
| 132 | f.pad_integral(is_nonnegative:true, prefix:"0b" , &self.to_str_radix(2)) |
| 133 | } |
| 134 | } |
| 135 | |
| 136 | impl fmt::Octal for BigUint { |
| 137 | fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { |
| 138 | f.pad_integral(is_nonnegative:true, prefix:"0o" , &self.to_str_radix(8)) |
| 139 | } |
| 140 | } |
| 141 | |
| 142 | impl Zero for BigUint { |
| 143 | #[inline ] |
| 144 | fn zero() -> BigUint { |
| 145 | Self::ZERO |
| 146 | } |
| 147 | |
| 148 | #[inline ] |
| 149 | fn set_zero(&mut self) { |
| 150 | self.data.clear(); |
| 151 | } |
| 152 | |
| 153 | #[inline ] |
| 154 | fn is_zero(&self) -> bool { |
| 155 | self.data.is_empty() |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | impl ConstZero for BigUint { |
| 160 | // forward to the inherent const |
| 161 | const ZERO: Self = Self::ZERO; // BigUint { data: Vec::new() }; |
| 162 | } |
| 163 | |
| 164 | impl One for BigUint { |
| 165 | #[inline ] |
| 166 | fn one() -> BigUint { |
| 167 | BigUint { data: vec![1] } |
| 168 | } |
| 169 | |
| 170 | #[inline ] |
| 171 | fn set_one(&mut self) { |
| 172 | self.data.clear(); |
| 173 | self.data.push(1); |
| 174 | } |
| 175 | |
| 176 | #[inline ] |
| 177 | fn is_one(&self) -> bool { |
| 178 | self.data[..] == [1] |
| 179 | } |
| 180 | } |
| 181 | |
| 182 | impl Unsigned for BigUint {} |
| 183 | |
| 184 | impl Integer for BigUint { |
| 185 | #[inline ] |
| 186 | fn div_rem(&self, other: &BigUint) -> (BigUint, BigUint) { |
| 187 | division::div_rem_ref(self, other) |
| 188 | } |
| 189 | |
| 190 | #[inline ] |
| 191 | fn div_floor(&self, other: &BigUint) -> BigUint { |
| 192 | let (d, _) = division::div_rem_ref(self, other); |
| 193 | d |
| 194 | } |
| 195 | |
| 196 | #[inline ] |
| 197 | fn mod_floor(&self, other: &BigUint) -> BigUint { |
| 198 | let (_, m) = division::div_rem_ref(self, other); |
| 199 | m |
| 200 | } |
| 201 | |
| 202 | #[inline ] |
| 203 | fn div_mod_floor(&self, other: &BigUint) -> (BigUint, BigUint) { |
| 204 | division::div_rem_ref(self, other) |
| 205 | } |
| 206 | |
| 207 | #[inline ] |
| 208 | fn div_ceil(&self, other: &BigUint) -> BigUint { |
| 209 | let (d, m) = division::div_rem_ref(self, other); |
| 210 | if m.is_zero() { |
| 211 | d |
| 212 | } else { |
| 213 | d + 1u32 |
| 214 | } |
| 215 | } |
| 216 | |
| 217 | /// Calculates the Greatest Common Divisor (GCD) of the number and `other`. |
| 218 | /// |
| 219 | /// The result is always positive. |
| 220 | #[inline ] |
| 221 | fn gcd(&self, other: &Self) -> Self { |
| 222 | #[inline ] |
| 223 | fn twos(x: &BigUint) -> u64 { |
| 224 | x.trailing_zeros().unwrap_or(0) |
| 225 | } |
| 226 | |
| 227 | // Stein's algorithm |
| 228 | if self.is_zero() { |
| 229 | return other.clone(); |
| 230 | } |
| 231 | if other.is_zero() { |
| 232 | return self.clone(); |
| 233 | } |
| 234 | let mut m = self.clone(); |
| 235 | let mut n = other.clone(); |
| 236 | |
| 237 | // find common factors of 2 |
| 238 | let shift = cmp::min(twos(&n), twos(&m)); |
| 239 | |
| 240 | // divide m and n by 2 until odd |
| 241 | // m inside loop |
| 242 | n >>= twos(&n); |
| 243 | |
| 244 | while !m.is_zero() { |
| 245 | m >>= twos(&m); |
| 246 | if n > m { |
| 247 | mem::swap(&mut n, &mut m) |
| 248 | } |
| 249 | m -= &n; |
| 250 | } |
| 251 | |
| 252 | n << shift |
| 253 | } |
| 254 | |
| 255 | /// Calculates the Lowest Common Multiple (LCM) of the number and `other`. |
| 256 | #[inline ] |
| 257 | fn lcm(&self, other: &BigUint) -> BigUint { |
| 258 | if self.is_zero() && other.is_zero() { |
| 259 | Self::ZERO |
| 260 | } else { |
| 261 | self / self.gcd(other) * other |
| 262 | } |
| 263 | } |
| 264 | |
| 265 | /// Calculates the Greatest Common Divisor (GCD) and |
| 266 | /// Lowest Common Multiple (LCM) together. |
| 267 | #[inline ] |
| 268 | fn gcd_lcm(&self, other: &Self) -> (Self, Self) { |
| 269 | let gcd = self.gcd(other); |
| 270 | let lcm = if gcd.is_zero() { |
| 271 | Self::ZERO |
| 272 | } else { |
| 273 | self / &gcd * other |
| 274 | }; |
| 275 | (gcd, lcm) |
| 276 | } |
| 277 | |
| 278 | /// Deprecated, use `is_multiple_of` instead. |
| 279 | #[inline ] |
| 280 | fn divides(&self, other: &BigUint) -> bool { |
| 281 | self.is_multiple_of(other) |
| 282 | } |
| 283 | |
| 284 | /// Returns `true` if the number is a multiple of `other`. |
| 285 | #[inline ] |
| 286 | fn is_multiple_of(&self, other: &BigUint) -> bool { |
| 287 | if other.is_zero() { |
| 288 | return self.is_zero(); |
| 289 | } |
| 290 | (self % other).is_zero() |
| 291 | } |
| 292 | |
| 293 | /// Returns `true` if the number is divisible by `2`. |
| 294 | #[inline ] |
| 295 | fn is_even(&self) -> bool { |
| 296 | // Considering only the last digit. |
| 297 | match self.data.first() { |
| 298 | Some(x) => x.is_even(), |
| 299 | None => true, |
| 300 | } |
| 301 | } |
| 302 | |
| 303 | /// Returns `true` if the number is not divisible by `2`. |
| 304 | #[inline ] |
| 305 | fn is_odd(&self) -> bool { |
| 306 | !self.is_even() |
| 307 | } |
| 308 | |
| 309 | /// Rounds up to nearest multiple of argument. |
| 310 | #[inline ] |
| 311 | fn next_multiple_of(&self, other: &Self) -> Self { |
| 312 | let m = self.mod_floor(other); |
| 313 | if m.is_zero() { |
| 314 | self.clone() |
| 315 | } else { |
| 316 | self + (other - m) |
| 317 | } |
| 318 | } |
| 319 | /// Rounds down to nearest multiple of argument. |
| 320 | #[inline ] |
| 321 | fn prev_multiple_of(&self, other: &Self) -> Self { |
| 322 | self - self.mod_floor(other) |
| 323 | } |
| 324 | |
| 325 | fn dec(&mut self) { |
| 326 | *self -= 1u32; |
| 327 | } |
| 328 | |
| 329 | fn inc(&mut self) { |
| 330 | *self += 1u32; |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | #[inline ] |
| 335 | fn fixpoint<F>(mut x: BigUint, max_bits: u64, f: F) -> BigUint |
| 336 | where |
| 337 | F: Fn(&BigUint) -> BigUint, |
| 338 | { |
| 339 | let mut xn: BigUint = f(&x); |
| 340 | |
| 341 | // If the value increased, then the initial guess must have been low. |
| 342 | // Repeat until we reverse course. |
| 343 | while x < xn { |
| 344 | // Sometimes an increase will go way too far, especially with large |
| 345 | // powers, and then take a long time to walk back. We know an upper |
| 346 | // bound based on bit size, so saturate on that. |
| 347 | x = if xn.bits() > max_bits { |
| 348 | BigUint::one() << max_bits |
| 349 | } else { |
| 350 | xn |
| 351 | }; |
| 352 | xn = f(&x); |
| 353 | } |
| 354 | |
| 355 | // Now keep repeating while the estimate is decreasing. |
| 356 | while x > xn { |
| 357 | x = xn; |
| 358 | xn = f(&x); |
| 359 | } |
| 360 | x |
| 361 | } |
| 362 | |
| 363 | impl Roots for BigUint { |
| 364 | // nth_root, sqrt and cbrt use Newton's method to compute |
| 365 | // principal root of a given degree for a given integer. |
| 366 | |
| 367 | // Reference: |
| 368 | // Brent & Zimmermann, Modern Computer Arithmetic, v0.5.9, Algorithm 1.14 |
| 369 | fn nth_root(&self, n: u32) -> Self { |
| 370 | assert!(n > 0, "root degree n must be at least 1" ); |
| 371 | |
| 372 | if self.is_zero() || self.is_one() { |
| 373 | return self.clone(); |
| 374 | } |
| 375 | |
| 376 | match n { |
| 377 | // Optimize for small n |
| 378 | 1 => return self.clone(), |
| 379 | 2 => return self.sqrt(), |
| 380 | 3 => return self.cbrt(), |
| 381 | _ => (), |
| 382 | } |
| 383 | |
| 384 | // The root of non-zero values less than 2ⁿ can only be 1. |
| 385 | let bits = self.bits(); |
| 386 | let n64 = u64::from(n); |
| 387 | if bits <= n64 { |
| 388 | return BigUint::one(); |
| 389 | } |
| 390 | |
| 391 | // If we fit in `u64`, compute the root that way. |
| 392 | if let Some(x) = self.to_u64() { |
| 393 | return x.nth_root(n).into(); |
| 394 | } |
| 395 | |
| 396 | let max_bits = bits / n64 + 1; |
| 397 | |
| 398 | #[cfg (feature = "std" )] |
| 399 | let guess = match self.to_f64() { |
| 400 | Some(f) if f.is_finite() => { |
| 401 | use num_traits::FromPrimitive; |
| 402 | |
| 403 | // We fit in `f64` (lossy), so get a better initial guess from that. |
| 404 | BigUint::from_f64((f.ln() / f64::from(n)).exp()).unwrap() |
| 405 | } |
| 406 | _ => { |
| 407 | // Try to guess by scaling down such that it does fit in `f64`. |
| 408 | // With some (x * 2ⁿᵏ), its nth root ≈ (ⁿ√x * 2ᵏ) |
| 409 | let extra_bits = bits - (f64::MAX_EXP as u64 - 1); |
| 410 | let root_scale = Integer::div_ceil(&extra_bits, &n64); |
| 411 | let scale = root_scale * n64; |
| 412 | if scale < bits && bits - scale > n64 { |
| 413 | (self >> scale).nth_root(n) << root_scale |
| 414 | } else { |
| 415 | BigUint::one() << max_bits |
| 416 | } |
| 417 | } |
| 418 | }; |
| 419 | |
| 420 | #[cfg (not(feature = "std" ))] |
| 421 | let guess = BigUint::one() << max_bits; |
| 422 | |
| 423 | let n_min_1 = n - 1; |
| 424 | fixpoint(guess, max_bits, move |s| { |
| 425 | let q = self / s.pow(n_min_1); |
| 426 | let t = n_min_1 * s + q; |
| 427 | t / n |
| 428 | }) |
| 429 | } |
| 430 | |
| 431 | // Reference: |
| 432 | // Brent & Zimmermann, Modern Computer Arithmetic, v0.5.9, Algorithm 1.13 |
| 433 | fn sqrt(&self) -> Self { |
| 434 | if self.is_zero() || self.is_one() { |
| 435 | return self.clone(); |
| 436 | } |
| 437 | |
| 438 | // If we fit in `u64`, compute the root that way. |
| 439 | if let Some(x) = self.to_u64() { |
| 440 | return x.sqrt().into(); |
| 441 | } |
| 442 | |
| 443 | let bits = self.bits(); |
| 444 | let max_bits = bits / 2 + 1; |
| 445 | |
| 446 | #[cfg (feature = "std" )] |
| 447 | let guess = match self.to_f64() { |
| 448 | Some(f) if f.is_finite() => { |
| 449 | use num_traits::FromPrimitive; |
| 450 | |
| 451 | // We fit in `f64` (lossy), so get a better initial guess from that. |
| 452 | BigUint::from_f64(f.sqrt()).unwrap() |
| 453 | } |
| 454 | _ => { |
| 455 | // Try to guess by scaling down such that it does fit in `f64`. |
| 456 | // With some (x * 2²ᵏ), its sqrt ≈ (√x * 2ᵏ) |
| 457 | let extra_bits = bits - (f64::MAX_EXP as u64 - 1); |
| 458 | let root_scale = (extra_bits + 1) / 2; |
| 459 | let scale = root_scale * 2; |
| 460 | (self >> scale).sqrt() << root_scale |
| 461 | } |
| 462 | }; |
| 463 | |
| 464 | #[cfg (not(feature = "std" ))] |
| 465 | let guess = BigUint::one() << max_bits; |
| 466 | |
| 467 | fixpoint(guess, max_bits, move |s| { |
| 468 | let q = self / s; |
| 469 | let t = s + q; |
| 470 | t >> 1 |
| 471 | }) |
| 472 | } |
| 473 | |
| 474 | fn cbrt(&self) -> Self { |
| 475 | if self.is_zero() || self.is_one() { |
| 476 | return self.clone(); |
| 477 | } |
| 478 | |
| 479 | // If we fit in `u64`, compute the root that way. |
| 480 | if let Some(x) = self.to_u64() { |
| 481 | return x.cbrt().into(); |
| 482 | } |
| 483 | |
| 484 | let bits = self.bits(); |
| 485 | let max_bits = bits / 3 + 1; |
| 486 | |
| 487 | #[cfg (feature = "std" )] |
| 488 | let guess = match self.to_f64() { |
| 489 | Some(f) if f.is_finite() => { |
| 490 | use num_traits::FromPrimitive; |
| 491 | |
| 492 | // We fit in `f64` (lossy), so get a better initial guess from that. |
| 493 | BigUint::from_f64(f.cbrt()).unwrap() |
| 494 | } |
| 495 | _ => { |
| 496 | // Try to guess by scaling down such that it does fit in `f64`. |
| 497 | // With some (x * 2³ᵏ), its cbrt ≈ (∛x * 2ᵏ) |
| 498 | let extra_bits = bits - (f64::MAX_EXP as u64 - 1); |
| 499 | let root_scale = (extra_bits + 2) / 3; |
| 500 | let scale = root_scale * 3; |
| 501 | (self >> scale).cbrt() << root_scale |
| 502 | } |
| 503 | }; |
| 504 | |
| 505 | #[cfg (not(feature = "std" ))] |
| 506 | let guess = BigUint::one() << max_bits; |
| 507 | |
| 508 | fixpoint(guess, max_bits, move |s| { |
| 509 | let q = self / (s * s); |
| 510 | let t = (s << 1) + q; |
| 511 | t / 3u32 |
| 512 | }) |
| 513 | } |
| 514 | } |
| 515 | |
| 516 | /// A generic trait for converting a value to a [`BigUint`]. |
| 517 | pub trait ToBigUint { |
| 518 | /// Converts the value of `self` to a [`BigUint`]. |
| 519 | fn to_biguint(&self) -> Option<BigUint>; |
| 520 | } |
| 521 | |
| 522 | /// Creates and initializes a [`BigUint`]. |
| 523 | /// |
| 524 | /// The digits are in little-endian base matching `BigDigit`. |
| 525 | #[inline ] |
| 526 | pub(crate) fn biguint_from_vec(digits: Vec<BigDigit>) -> BigUint { |
| 527 | BigUint { data: digits }.normalized() |
| 528 | } |
| 529 | |
| 530 | impl BigUint { |
| 531 | /// A constant `BigUint` with value 0, useful for static initialization. |
| 532 | pub const ZERO: Self = BigUint { data: Vec::new() }; |
| 533 | |
| 534 | /// Creates and initializes a [`BigUint`]. |
| 535 | /// |
| 536 | /// The base 2<sup>32</sup> digits are ordered least significant digit first. |
| 537 | #[inline ] |
| 538 | pub fn new(digits: Vec<u32>) -> BigUint { |
| 539 | let mut big = Self::ZERO; |
| 540 | |
| 541 | cfg_digit_expr!( |
| 542 | { |
| 543 | big.data = digits; |
| 544 | big.normalize(); |
| 545 | }, |
| 546 | big.assign_from_slice(&digits) |
| 547 | ); |
| 548 | |
| 549 | big |
| 550 | } |
| 551 | |
| 552 | /// Creates and initializes a [`BigUint`]. |
| 553 | /// |
| 554 | /// The base 2<sup>32</sup> digits are ordered least significant digit first. |
| 555 | #[inline ] |
| 556 | pub fn from_slice(slice: &[u32]) -> BigUint { |
| 557 | let mut big = Self::ZERO; |
| 558 | big.assign_from_slice(slice); |
| 559 | big |
| 560 | } |
| 561 | |
| 562 | /// Assign a value to a [`BigUint`]. |
| 563 | /// |
| 564 | /// The base 2<sup>32</sup> digits are ordered least significant digit first. |
| 565 | #[inline ] |
| 566 | pub fn assign_from_slice(&mut self, slice: &[u32]) { |
| 567 | self.data.clear(); |
| 568 | |
| 569 | cfg_digit_expr!( |
| 570 | self.data.extend_from_slice(slice), |
| 571 | self.data.extend(slice.chunks(2).map(u32_chunk_to_u64)) |
| 572 | ); |
| 573 | |
| 574 | self.normalize(); |
| 575 | } |
| 576 | |
| 577 | /// Creates and initializes a [`BigUint`]. |
| 578 | /// |
| 579 | /// The bytes are in big-endian byte order. |
| 580 | /// |
| 581 | /// # Examples |
| 582 | /// |
| 583 | /// ``` |
| 584 | /// use num_bigint::BigUint; |
| 585 | /// |
| 586 | /// assert_eq!(BigUint::from_bytes_be(b"A" ), |
| 587 | /// BigUint::parse_bytes(b"65" , 10).unwrap()); |
| 588 | /// assert_eq!(BigUint::from_bytes_be(b"AA" ), |
| 589 | /// BigUint::parse_bytes(b"16705" , 10).unwrap()); |
| 590 | /// assert_eq!(BigUint::from_bytes_be(b"AB" ), |
| 591 | /// BigUint::parse_bytes(b"16706" , 10).unwrap()); |
| 592 | /// assert_eq!(BigUint::from_bytes_be(b"Hello world!" ), |
| 593 | /// BigUint::parse_bytes(b"22405534230753963835153736737" , 10).unwrap()); |
| 594 | /// ``` |
| 595 | #[inline ] |
| 596 | pub fn from_bytes_be(bytes: &[u8]) -> BigUint { |
| 597 | if bytes.is_empty() { |
| 598 | Self::ZERO |
| 599 | } else { |
| 600 | let mut v = bytes.to_vec(); |
| 601 | v.reverse(); |
| 602 | BigUint::from_bytes_le(&v) |
| 603 | } |
| 604 | } |
| 605 | |
| 606 | /// Creates and initializes a [`BigUint`]. |
| 607 | /// |
| 608 | /// The bytes are in little-endian byte order. |
| 609 | #[inline ] |
| 610 | pub fn from_bytes_le(bytes: &[u8]) -> BigUint { |
| 611 | if bytes.is_empty() { |
| 612 | Self::ZERO |
| 613 | } else { |
| 614 | convert::from_bitwise_digits_le(bytes, 8) |
| 615 | } |
| 616 | } |
| 617 | |
| 618 | /// Creates and initializes a [`BigUint`]. The input slice must contain |
| 619 | /// ascii/utf8 characters in [0-9a-zA-Z]. |
| 620 | /// `radix` must be in the range `2...36`. |
| 621 | /// |
| 622 | /// The function `from_str_radix` from the `Num` trait provides the same logic |
| 623 | /// for `&str` buffers. |
| 624 | /// |
| 625 | /// # Examples |
| 626 | /// |
| 627 | /// ``` |
| 628 | /// use num_bigint::{BigUint, ToBigUint}; |
| 629 | /// |
| 630 | /// assert_eq!(BigUint::parse_bytes(b"1234" , 10), ToBigUint::to_biguint(&1234)); |
| 631 | /// assert_eq!(BigUint::parse_bytes(b"ABCD" , 16), ToBigUint::to_biguint(&0xABCD)); |
| 632 | /// assert_eq!(BigUint::parse_bytes(b"G" , 16), None); |
| 633 | /// ``` |
| 634 | #[inline ] |
| 635 | pub fn parse_bytes(buf: &[u8], radix: u32) -> Option<BigUint> { |
| 636 | let s = str::from_utf8(buf).ok()?; |
| 637 | BigUint::from_str_radix(s, radix).ok() |
| 638 | } |
| 639 | |
| 640 | /// Creates and initializes a [`BigUint`]. Each `u8` of the input slice is |
| 641 | /// interpreted as one digit of the number |
| 642 | /// and must therefore be less than `radix`. |
| 643 | /// |
| 644 | /// The bytes are in big-endian byte order. |
| 645 | /// `radix` must be in the range `2...256`. |
| 646 | /// |
| 647 | /// # Examples |
| 648 | /// |
| 649 | /// ``` |
| 650 | /// use num_bigint::{BigUint}; |
| 651 | /// |
| 652 | /// let inbase190 = &[15, 33, 125, 12, 14]; |
| 653 | /// let a = BigUint::from_radix_be(inbase190, 190).unwrap(); |
| 654 | /// assert_eq!(a.to_radix_be(190), inbase190); |
| 655 | /// ``` |
| 656 | pub fn from_radix_be(buf: &[u8], radix: u32) -> Option<BigUint> { |
| 657 | convert::from_radix_be(buf, radix) |
| 658 | } |
| 659 | |
| 660 | /// Creates and initializes a [`BigUint`]. Each `u8` of the input slice is |
| 661 | /// interpreted as one digit of the number |
| 662 | /// and must therefore be less than `radix`. |
| 663 | /// |
| 664 | /// The bytes are in little-endian byte order. |
| 665 | /// `radix` must be in the range `2...256`. |
| 666 | /// |
| 667 | /// # Examples |
| 668 | /// |
| 669 | /// ``` |
| 670 | /// use num_bigint::{BigUint}; |
| 671 | /// |
| 672 | /// let inbase190 = &[14, 12, 125, 33, 15]; |
| 673 | /// let a = BigUint::from_radix_be(inbase190, 190).unwrap(); |
| 674 | /// assert_eq!(a.to_radix_be(190), inbase190); |
| 675 | /// ``` |
| 676 | pub fn from_radix_le(buf: &[u8], radix: u32) -> Option<BigUint> { |
| 677 | convert::from_radix_le(buf, radix) |
| 678 | } |
| 679 | |
| 680 | /// Returns the byte representation of the [`BigUint`] in big-endian byte order. |
| 681 | /// |
| 682 | /// # Examples |
| 683 | /// |
| 684 | /// ``` |
| 685 | /// use num_bigint::BigUint; |
| 686 | /// |
| 687 | /// let i = BigUint::parse_bytes(b"1125" , 10).unwrap(); |
| 688 | /// assert_eq!(i.to_bytes_be(), vec![4, 101]); |
| 689 | /// ``` |
| 690 | #[inline ] |
| 691 | pub fn to_bytes_be(&self) -> Vec<u8> { |
| 692 | let mut v = self.to_bytes_le(); |
| 693 | v.reverse(); |
| 694 | v |
| 695 | } |
| 696 | |
| 697 | /// Returns the byte representation of the [`BigUint`] in little-endian byte order. |
| 698 | /// |
| 699 | /// # Examples |
| 700 | /// |
| 701 | /// ``` |
| 702 | /// use num_bigint::BigUint; |
| 703 | /// |
| 704 | /// let i = BigUint::parse_bytes(b"1125" , 10).unwrap(); |
| 705 | /// assert_eq!(i.to_bytes_le(), vec![101, 4]); |
| 706 | /// ``` |
| 707 | #[inline ] |
| 708 | pub fn to_bytes_le(&self) -> Vec<u8> { |
| 709 | if self.is_zero() { |
| 710 | vec![0] |
| 711 | } else { |
| 712 | convert::to_bitwise_digits_le(self, 8) |
| 713 | } |
| 714 | } |
| 715 | |
| 716 | /// Returns the `u32` digits representation of the [`BigUint`] ordered least significant digit |
| 717 | /// first. |
| 718 | /// |
| 719 | /// # Examples |
| 720 | /// |
| 721 | /// ``` |
| 722 | /// use num_bigint::BigUint; |
| 723 | /// |
| 724 | /// assert_eq!(BigUint::from(1125u32).to_u32_digits(), vec![1125]); |
| 725 | /// assert_eq!(BigUint::from(4294967295u32).to_u32_digits(), vec![4294967295]); |
| 726 | /// assert_eq!(BigUint::from(4294967296u64).to_u32_digits(), vec![0, 1]); |
| 727 | /// assert_eq!(BigUint::from(112500000000u64).to_u32_digits(), vec![830850304, 26]); |
| 728 | /// ``` |
| 729 | #[inline ] |
| 730 | pub fn to_u32_digits(&self) -> Vec<u32> { |
| 731 | self.iter_u32_digits().collect() |
| 732 | } |
| 733 | |
| 734 | /// Returns the `u64` digits representation of the [`BigUint`] ordered least significant digit |
| 735 | /// first. |
| 736 | /// |
| 737 | /// # Examples |
| 738 | /// |
| 739 | /// ``` |
| 740 | /// use num_bigint::BigUint; |
| 741 | /// |
| 742 | /// assert_eq!(BigUint::from(1125u32).to_u64_digits(), vec![1125]); |
| 743 | /// assert_eq!(BigUint::from(4294967295u32).to_u64_digits(), vec![4294967295]); |
| 744 | /// assert_eq!(BigUint::from(4294967296u64).to_u64_digits(), vec![4294967296]); |
| 745 | /// assert_eq!(BigUint::from(112500000000u64).to_u64_digits(), vec![112500000000]); |
| 746 | /// assert_eq!(BigUint::from(1u128 << 64).to_u64_digits(), vec![0, 1]); |
| 747 | /// ``` |
| 748 | #[inline ] |
| 749 | pub fn to_u64_digits(&self) -> Vec<u64> { |
| 750 | self.iter_u64_digits().collect() |
| 751 | } |
| 752 | |
| 753 | /// Returns an iterator of `u32` digits representation of the [`BigUint`] ordered least |
| 754 | /// significant digit first. |
| 755 | /// |
| 756 | /// # Examples |
| 757 | /// |
| 758 | /// ``` |
| 759 | /// use num_bigint::BigUint; |
| 760 | /// |
| 761 | /// assert_eq!(BigUint::from(1125u32).iter_u32_digits().collect::<Vec<u32>>(), vec![1125]); |
| 762 | /// assert_eq!(BigUint::from(4294967295u32).iter_u32_digits().collect::<Vec<u32>>(), vec![4294967295]); |
| 763 | /// assert_eq!(BigUint::from(4294967296u64).iter_u32_digits().collect::<Vec<u32>>(), vec![0, 1]); |
| 764 | /// assert_eq!(BigUint::from(112500000000u64).iter_u32_digits().collect::<Vec<u32>>(), vec![830850304, 26]); |
| 765 | /// ``` |
| 766 | #[inline ] |
| 767 | pub fn iter_u32_digits(&self) -> U32Digits<'_> { |
| 768 | U32Digits::new(self.data.as_slice()) |
| 769 | } |
| 770 | |
| 771 | /// Returns an iterator of `u64` digits representation of the [`BigUint`] ordered least |
| 772 | /// significant digit first. |
| 773 | /// |
| 774 | /// # Examples |
| 775 | /// |
| 776 | /// ``` |
| 777 | /// use num_bigint::BigUint; |
| 778 | /// |
| 779 | /// assert_eq!(BigUint::from(1125u32).iter_u64_digits().collect::<Vec<u64>>(), vec![1125]); |
| 780 | /// assert_eq!(BigUint::from(4294967295u32).iter_u64_digits().collect::<Vec<u64>>(), vec![4294967295]); |
| 781 | /// assert_eq!(BigUint::from(4294967296u64).iter_u64_digits().collect::<Vec<u64>>(), vec![4294967296]); |
| 782 | /// assert_eq!(BigUint::from(112500000000u64).iter_u64_digits().collect::<Vec<u64>>(), vec![112500000000]); |
| 783 | /// assert_eq!(BigUint::from(1u128 << 64).iter_u64_digits().collect::<Vec<u64>>(), vec![0, 1]); |
| 784 | /// ``` |
| 785 | #[inline ] |
| 786 | pub fn iter_u64_digits(&self) -> U64Digits<'_> { |
| 787 | U64Digits::new(self.data.as_slice()) |
| 788 | } |
| 789 | |
| 790 | /// Returns the integer formatted as a string in the given radix. |
| 791 | /// `radix` must be in the range `2...36`. |
| 792 | /// |
| 793 | /// # Examples |
| 794 | /// |
| 795 | /// ``` |
| 796 | /// use num_bigint::BigUint; |
| 797 | /// |
| 798 | /// let i = BigUint::parse_bytes(b"ff" , 16).unwrap(); |
| 799 | /// assert_eq!(i.to_str_radix(16), "ff" ); |
| 800 | /// ``` |
| 801 | #[inline ] |
| 802 | pub fn to_str_radix(&self, radix: u32) -> String { |
| 803 | let mut v = to_str_radix_reversed(self, radix); |
| 804 | v.reverse(); |
| 805 | unsafe { String::from_utf8_unchecked(v) } |
| 806 | } |
| 807 | |
| 808 | /// Returns the integer in the requested base in big-endian digit order. |
| 809 | /// The output is not given in a human readable alphabet but as a zero |
| 810 | /// based `u8` number. |
| 811 | /// `radix` must be in the range `2...256`. |
| 812 | /// |
| 813 | /// # Examples |
| 814 | /// |
| 815 | /// ``` |
| 816 | /// use num_bigint::BigUint; |
| 817 | /// |
| 818 | /// assert_eq!(BigUint::from(0xFFFFu64).to_radix_be(159), |
| 819 | /// vec![2, 94, 27]); |
| 820 | /// // 0xFFFF = 65535 = 2*(159^2) + 94*159 + 27 |
| 821 | /// ``` |
| 822 | #[inline ] |
| 823 | pub fn to_radix_be(&self, radix: u32) -> Vec<u8> { |
| 824 | let mut v = convert::to_radix_le(self, radix); |
| 825 | v.reverse(); |
| 826 | v |
| 827 | } |
| 828 | |
| 829 | /// Returns the integer in the requested base in little-endian digit order. |
| 830 | /// The output is not given in a human readable alphabet but as a zero |
| 831 | /// based u8 number. |
| 832 | /// `radix` must be in the range `2...256`. |
| 833 | /// |
| 834 | /// # Examples |
| 835 | /// |
| 836 | /// ``` |
| 837 | /// use num_bigint::BigUint; |
| 838 | /// |
| 839 | /// assert_eq!(BigUint::from(0xFFFFu64).to_radix_le(159), |
| 840 | /// vec![27, 94, 2]); |
| 841 | /// // 0xFFFF = 65535 = 27 + 94*159 + 2*(159^2) |
| 842 | /// ``` |
| 843 | #[inline ] |
| 844 | pub fn to_radix_le(&self, radix: u32) -> Vec<u8> { |
| 845 | convert::to_radix_le(self, radix) |
| 846 | } |
| 847 | |
| 848 | /// Determines the fewest bits necessary to express the [`BigUint`]. |
| 849 | #[inline ] |
| 850 | pub fn bits(&self) -> u64 { |
| 851 | if self.is_zero() { |
| 852 | return 0; |
| 853 | } |
| 854 | let zeros: u64 = self.data.last().unwrap().leading_zeros().into(); |
| 855 | self.data.len() as u64 * u64::from(big_digit::BITS) - zeros |
| 856 | } |
| 857 | |
| 858 | /// Strips off trailing zero bigdigits - comparisons require the last element in the vector to |
| 859 | /// be nonzero. |
| 860 | #[inline ] |
| 861 | fn normalize(&mut self) { |
| 862 | if let Some(&0) = self.data.last() { |
| 863 | let len = self.data.iter().rposition(|&d| d != 0).map_or(0, |i| i + 1); |
| 864 | self.data.truncate(len); |
| 865 | } |
| 866 | if self.data.len() < self.data.capacity() / 4 { |
| 867 | self.data.shrink_to_fit(); |
| 868 | } |
| 869 | } |
| 870 | |
| 871 | /// Returns a normalized [`BigUint`]. |
| 872 | #[inline ] |
| 873 | fn normalized(mut self) -> BigUint { |
| 874 | self.normalize(); |
| 875 | self |
| 876 | } |
| 877 | |
| 878 | /// Returns `self ^ exponent`. |
| 879 | pub fn pow(&self, exponent: u32) -> Self { |
| 880 | Pow::pow(self, exponent) |
| 881 | } |
| 882 | |
| 883 | /// Returns `(self ^ exponent) % modulus`. |
| 884 | /// |
| 885 | /// Panics if the modulus is zero. |
| 886 | pub fn modpow(&self, exponent: &Self, modulus: &Self) -> Self { |
| 887 | power::modpow(self, exponent, modulus) |
| 888 | } |
| 889 | |
| 890 | /// Returns the modular multiplicative inverse if it exists, otherwise `None`. |
| 891 | /// |
| 892 | /// This solves for `x` in the interval `[0, modulus)` such that `self * x ≡ 1 (mod modulus)`. |
| 893 | /// The solution exists if and only if `gcd(self, modulus) == 1`. |
| 894 | /// |
| 895 | /// ``` |
| 896 | /// use num_bigint::BigUint; |
| 897 | /// use num_traits::{One, Zero}; |
| 898 | /// |
| 899 | /// let m = BigUint::from(383_u32); |
| 900 | /// |
| 901 | /// // Trivial cases |
| 902 | /// assert_eq!(BigUint::zero().modinv(&m), None); |
| 903 | /// assert_eq!(BigUint::one().modinv(&m), Some(BigUint::one())); |
| 904 | /// let neg1 = &m - 1u32; |
| 905 | /// assert_eq!(neg1.modinv(&m), Some(neg1)); |
| 906 | /// |
| 907 | /// let a = BigUint::from(271_u32); |
| 908 | /// let x = a.modinv(&m).unwrap(); |
| 909 | /// assert_eq!(x, BigUint::from(106_u32)); |
| 910 | /// assert_eq!(x.modinv(&m).unwrap(), a); |
| 911 | /// assert!((a * x % m).is_one()); |
| 912 | /// ``` |
| 913 | pub fn modinv(&self, modulus: &Self) -> Option<Self> { |
| 914 | // Based on the inverse pseudocode listed here: |
| 915 | // https://en.wikipedia.org/wiki/Extended_Euclidean_algorithm#Modular_integers |
| 916 | // TODO: consider Binary or Lehmer's GCD algorithms for optimization. |
| 917 | |
| 918 | assert!( |
| 919 | !modulus.is_zero(), |
| 920 | "attempt to calculate with zero modulus!" |
| 921 | ); |
| 922 | if modulus.is_one() { |
| 923 | return Some(Self::zero()); |
| 924 | } |
| 925 | |
| 926 | let mut r0; // = modulus.clone(); |
| 927 | let mut r1 = self % modulus; |
| 928 | let mut t0; // = Self::zero(); |
| 929 | let mut t1; // = Self::one(); |
| 930 | |
| 931 | // Lift and simplify the first iteration to avoid some initial allocations. |
| 932 | if r1.is_zero() { |
| 933 | return None; |
| 934 | } else if r1.is_one() { |
| 935 | return Some(r1); |
| 936 | } else { |
| 937 | let (q, r2) = modulus.div_rem(&r1); |
| 938 | if r2.is_zero() { |
| 939 | return None; |
| 940 | } |
| 941 | r0 = r1; |
| 942 | r1 = r2; |
| 943 | t0 = Self::one(); |
| 944 | t1 = modulus - q; |
| 945 | } |
| 946 | |
| 947 | while !r1.is_zero() { |
| 948 | let (q, r2) = r0.div_rem(&r1); |
| 949 | r0 = r1; |
| 950 | r1 = r2; |
| 951 | |
| 952 | // let t2 = (t0 - q * t1) % modulus; |
| 953 | let qt1 = q * &t1 % modulus; |
| 954 | let t2 = if t0 < qt1 { |
| 955 | t0 + (modulus - qt1) |
| 956 | } else { |
| 957 | t0 - qt1 |
| 958 | }; |
| 959 | t0 = t1; |
| 960 | t1 = t2; |
| 961 | } |
| 962 | |
| 963 | if r0.is_one() { |
| 964 | Some(t0) |
| 965 | } else { |
| 966 | None |
| 967 | } |
| 968 | } |
| 969 | |
| 970 | /// Returns the truncated principal square root of `self` -- |
| 971 | /// see [Roots::sqrt](https://docs.rs/num-integer/0.1/num_integer/trait.Roots.html#method.sqrt) |
| 972 | pub fn sqrt(&self) -> Self { |
| 973 | Roots::sqrt(self) |
| 974 | } |
| 975 | |
| 976 | /// Returns the truncated principal cube root of `self` -- |
| 977 | /// see [Roots::cbrt](https://docs.rs/num-integer/0.1/num_integer/trait.Roots.html#method.cbrt). |
| 978 | pub fn cbrt(&self) -> Self { |
| 979 | Roots::cbrt(self) |
| 980 | } |
| 981 | |
| 982 | /// Returns the truncated principal `n`th root of `self` -- |
| 983 | /// see [Roots::nth_root](https://docs.rs/num-integer/0.1/num_integer/trait.Roots.html#tymethod.nth_root). |
| 984 | pub fn nth_root(&self, n: u32) -> Self { |
| 985 | Roots::nth_root(self, n) |
| 986 | } |
| 987 | |
| 988 | /// Returns the number of least-significant bits that are zero, |
| 989 | /// or `None` if the entire number is zero. |
| 990 | pub fn trailing_zeros(&self) -> Option<u64> { |
| 991 | let i = self.data.iter().position(|&digit| digit != 0)?; |
| 992 | let zeros: u64 = self.data[i].trailing_zeros().into(); |
| 993 | Some(i as u64 * u64::from(big_digit::BITS) + zeros) |
| 994 | } |
| 995 | |
| 996 | /// Returns the number of least-significant bits that are ones. |
| 997 | pub fn trailing_ones(&self) -> u64 { |
| 998 | if let Some(i) = self.data.iter().position(|&digit| !digit != 0) { |
| 999 | let ones: u64 = self.data[i].trailing_ones().into(); |
| 1000 | i as u64 * u64::from(big_digit::BITS) + ones |
| 1001 | } else { |
| 1002 | self.data.len() as u64 * u64::from(big_digit::BITS) |
| 1003 | } |
| 1004 | } |
| 1005 | |
| 1006 | /// Returns the number of one bits. |
| 1007 | pub fn count_ones(&self) -> u64 { |
| 1008 | self.data.iter().map(|&d| u64::from(d.count_ones())).sum() |
| 1009 | } |
| 1010 | |
| 1011 | /// Returns whether the bit in the given position is set |
| 1012 | pub fn bit(&self, bit: u64) -> bool { |
| 1013 | let bits_per_digit = u64::from(big_digit::BITS); |
| 1014 | if let Some(digit_index) = (bit / bits_per_digit).to_usize() { |
| 1015 | if let Some(digit) = self.data.get(digit_index) { |
| 1016 | let bit_mask = (1 as BigDigit) << (bit % bits_per_digit); |
| 1017 | return (digit & bit_mask) != 0; |
| 1018 | } |
| 1019 | } |
| 1020 | false |
| 1021 | } |
| 1022 | |
| 1023 | /// Sets or clears the bit in the given position |
| 1024 | /// |
| 1025 | /// Note that setting a bit greater than the current bit length, a reallocation may be needed |
| 1026 | /// to store the new digits |
| 1027 | pub fn set_bit(&mut self, bit: u64, value: bool) { |
| 1028 | // Note: we're saturating `digit_index` and `new_len` -- any such case is guaranteed to |
| 1029 | // fail allocation, and that's more consistent than adding our own overflow panics. |
| 1030 | let bits_per_digit = u64::from(big_digit::BITS); |
| 1031 | let digit_index = (bit / bits_per_digit).to_usize().unwrap_or(usize::MAX); |
| 1032 | let bit_mask = (1 as BigDigit) << (bit % bits_per_digit); |
| 1033 | if value { |
| 1034 | if digit_index >= self.data.len() { |
| 1035 | let new_len = digit_index.saturating_add(1); |
| 1036 | self.data.resize(new_len, 0); |
| 1037 | } |
| 1038 | self.data[digit_index] |= bit_mask; |
| 1039 | } else if digit_index < self.data.len() { |
| 1040 | self.data[digit_index] &= !bit_mask; |
| 1041 | // the top bit may have been cleared, so normalize |
| 1042 | self.normalize(); |
| 1043 | } |
| 1044 | } |
| 1045 | } |
| 1046 | |
| 1047 | impl num_traits::FromBytes for BigUint { |
| 1048 | type Bytes = [u8]; |
| 1049 | |
| 1050 | fn from_be_bytes(bytes: &Self::Bytes) -> Self { |
| 1051 | Self::from_bytes_be(bytes) |
| 1052 | } |
| 1053 | |
| 1054 | fn from_le_bytes(bytes: &Self::Bytes) -> Self { |
| 1055 | Self::from_bytes_le(bytes) |
| 1056 | } |
| 1057 | } |
| 1058 | |
| 1059 | impl num_traits::ToBytes for BigUint { |
| 1060 | type Bytes = Vec<u8>; |
| 1061 | |
| 1062 | fn to_be_bytes(&self) -> Self::Bytes { |
| 1063 | self.to_bytes_be() |
| 1064 | } |
| 1065 | |
| 1066 | fn to_le_bytes(&self) -> Self::Bytes { |
| 1067 | self.to_bytes_le() |
| 1068 | } |
| 1069 | } |
| 1070 | |
| 1071 | pub(crate) trait IntDigits { |
| 1072 | fn digits(&self) -> &[BigDigit]; |
| 1073 | fn digits_mut(&mut self) -> &mut Vec<BigDigit>; |
| 1074 | fn normalize(&mut self); |
| 1075 | fn capacity(&self) -> usize; |
| 1076 | fn len(&self) -> usize; |
| 1077 | } |
| 1078 | |
| 1079 | impl IntDigits for BigUint { |
| 1080 | #[inline ] |
| 1081 | fn digits(&self) -> &[BigDigit] { |
| 1082 | &self.data |
| 1083 | } |
| 1084 | #[inline ] |
| 1085 | fn digits_mut(&mut self) -> &mut Vec<BigDigit> { |
| 1086 | &mut self.data |
| 1087 | } |
| 1088 | #[inline ] |
| 1089 | fn normalize(&mut self) { |
| 1090 | self.normalize(); |
| 1091 | } |
| 1092 | #[inline ] |
| 1093 | fn capacity(&self) -> usize { |
| 1094 | self.data.capacity() |
| 1095 | } |
| 1096 | #[inline ] |
| 1097 | fn len(&self) -> usize { |
| 1098 | self.data.len() |
| 1099 | } |
| 1100 | } |
| 1101 | |
| 1102 | /// Convert a `u32` chunk (len is either 1 or 2) to a single `u64` digit |
| 1103 | #[inline ] |
| 1104 | fn u32_chunk_to_u64(chunk: &[u32]) -> u64 { |
| 1105 | // raw could have odd length |
| 1106 | let mut digit: u64 = chunk[0] as u64; |
| 1107 | if let Some(&hi: u32) = chunk.get(index:1) { |
| 1108 | digit |= (hi as u64) << 32; |
| 1109 | } |
| 1110 | digit |
| 1111 | } |
| 1112 | |
| 1113 | cfg_32_or_test!( |
| 1114 | /// Combine four `u32`s into a single `u128`. |
| 1115 | #[inline ] |
| 1116 | fn u32_to_u128(a: u32, b: u32, c: u32, d: u32) -> u128 { |
| 1117 | u128::from(d) | (u128::from(c) << 32) | (u128::from(b) << 64) | (u128::from(a) << 96) |
| 1118 | } |
| 1119 | ); |
| 1120 | |
| 1121 | cfg_32_or_test!( |
| 1122 | /// Split a single `u128` into four `u32`. |
| 1123 | #[inline ] |
| 1124 | fn u32_from_u128(n: u128) -> (u32, u32, u32, u32) { |
| 1125 | ( |
| 1126 | (n >> 96) as u32, |
| 1127 | (n >> 64) as u32, |
| 1128 | (n >> 32) as u32, |
| 1129 | n as u32, |
| 1130 | ) |
| 1131 | } |
| 1132 | ); |
| 1133 | |
| 1134 | cfg_digit!( |
| 1135 | #[test ] |
| 1136 | fn test_from_slice() { |
| 1137 | fn check(slice: &[u32], data: &[BigDigit]) { |
| 1138 | assert_eq!(BigUint::from_slice(slice).data, data); |
| 1139 | } |
| 1140 | check(&[1], &[1]); |
| 1141 | check(&[0, 0, 0], &[]); |
| 1142 | check(&[1, 2, 0, 0], &[1, 2]); |
| 1143 | check(&[0, 0, 1, 2], &[0, 0, 1, 2]); |
| 1144 | check(&[0, 0, 1, 2, 0, 0], &[0, 0, 1, 2]); |
| 1145 | check(&[-1i32 as u32], &[-1i32 as BigDigit]); |
| 1146 | } |
| 1147 | |
| 1148 | #[test ] |
| 1149 | fn test_from_slice() { |
| 1150 | fn check(slice: &[u32], data: &[BigDigit]) { |
| 1151 | assert_eq!( |
| 1152 | BigUint::from_slice(slice).data, |
| 1153 | data, |
| 1154 | "from {:?}, to {:?}" , |
| 1155 | slice, |
| 1156 | data |
| 1157 | ); |
| 1158 | } |
| 1159 | check(&[1], &[1]); |
| 1160 | check(&[0, 0, 0], &[]); |
| 1161 | check(&[1, 2], &[8_589_934_593]); |
| 1162 | check(&[1, 2, 0, 0], &[8_589_934_593]); |
| 1163 | check(&[0, 0, 1, 2], &[0, 8_589_934_593]); |
| 1164 | check(&[0, 0, 1, 2, 0, 0], &[0, 8_589_934_593]); |
| 1165 | check(&[-1i32 as u32], &[(-1i32 as u32) as BigDigit]); |
| 1166 | } |
| 1167 | ); |
| 1168 | |
| 1169 | #[test ] |
| 1170 | fn test_u32_u128() { |
| 1171 | assert_eq!(u32_from_u128(0u128), (0, 0, 0, 0)); |
| 1172 | assert_eq!( |
| 1173 | u32_from_u128(u128::MAX), |
| 1174 | (u32::MAX, u32::MAX, u32::MAX, u32::MAX) |
| 1175 | ); |
| 1176 | |
| 1177 | assert_eq!(u32_from_u128(u32::MAX as u128), (0, 0, 0, u32::MAX)); |
| 1178 | |
| 1179 | assert_eq!(u32_from_u128(u64::MAX as u128), (0, 0, u32::MAX, u32::MAX)); |
| 1180 | |
| 1181 | assert_eq!( |
| 1182 | u32_from_u128((u64::MAX as u128) + u32::MAX as u128), |
| 1183 | (0, 1, 0, u32::MAX - 1) |
| 1184 | ); |
| 1185 | |
| 1186 | assert_eq!(u32_from_u128(36_893_488_151_714_070_528), (0, 2, 1, 0)); |
| 1187 | } |
| 1188 | |
| 1189 | #[test ] |
| 1190 | fn test_u128_u32_roundtrip() { |
| 1191 | // roundtrips |
| 1192 | let values = vec![ |
| 1193 | 0u128, |
| 1194 | 1u128, |
| 1195 | u64::MAX as u128 * 3, |
| 1196 | u32::MAX as u128, |
| 1197 | u64::MAX as u128, |
| 1198 | (u64::MAX as u128) + u32::MAX as u128, |
| 1199 | u128::MAX, |
| 1200 | ]; |
| 1201 | |
| 1202 | for val in &values { |
| 1203 | let (a, b, c, d) = u32_from_u128(*val); |
| 1204 | assert_eq!(u32_to_u128(a, b, c, d), *val); |
| 1205 | } |
| 1206 | } |
| 1207 | |