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
//! We use the Mersenne prime 2^127-1 (i128::MAX) as the main modulo, which maximize the space of available hashing slots.
//! (The largest Mersenne prime under 2^64 is only 2^61-1, so we use u128 for hashing which is also future proof).
//!
//! The basic algorithm is similar to what is used in Python (see https://docs.python.org/3.8/library/stdtypes.html#hashing-of-numeric-types),
//! specifically if the numerically consistent hash function is denoted as num_hash, then:
//! - for an integer n: num_hash(n) = sgn(n) * (|n| % M127)
//! - for a rational number n/d (including floating numbers): sgn(n/d) * num_hash(|n|) * (num_hash(|d|)^-1 mod M127)
//! - for special values: num_hash(NaN) and num_hash(±∞) are specially chosen such that it won't overlap with normal numbers.
use crate::NumHash;
use core::hash::{Hash, Hasher};
use num_modular::{FixedMersenneInt, ModularAbs, ModularInteger};
// we use 2^127 - 1 (a Mersenne prime) as modulus
type MInt = FixedMersenneInt<127, 1>;
const M127: i128 = i128::MAX;
const M127U: u128 = M127 as u128;
const M127D: u128 = M127U + M127U;
const HASH_INF: i128 = i128::MAX; // 2^127 - 1
const HASH_NEGINF: i128 = i128::MIN + 1; // -(2^127 - 1)
const HASH_NAN: i128 = i128::MIN; // -2^127
#[cfg(feature = "num-complex")]
const PROOT: u128 = i32::MAX as u128; // a Mersenne prime
// TODO (v2.0): Use the coefficients of the characteristic polynomial to represent a number. By this way
// all algebraic numbers can be represented including complex and quadratic numbers.
// Case1: directly hash the i128 and u128 number (mod M127)
impl NumHash for i128 {
#[inline]
fn num_hash<H: Hasher>(&self, state: &mut H) {
const MINP1: i128 = i128::MIN + 1;
match *self {
i128::MAX | MINP1 => 0i128.hash(state),
i128::MIN => (-1i128).hash(state),
u => u.hash(state),
}
}
}
impl NumHash for u128 {
#[inline]
fn num_hash<H: Hasher>(&self, state: &mut H) {
match *self {
u128::MAX => 1i128.hash(state),
M127D => 0i128.hash(state),
u if u >= M127U => ((u - M127U) as i128).hash(state),
u => (u as i128).hash(state),
}
}
}
// Case2: convert other integers to 64 bit integer
macro_rules! impl_hash_for_small_int {
($($signed:ty)*) => ($(
impl NumHash for $signed {
#[inline]
fn num_hash<H: Hasher>(&self, state: &mut H) {
(&(*self as i128)).hash(state) // these integers are always smaller than M127
}
}
)*);
}
impl_hash_for_small_int! { i8 i16 i32 i64 u8 u16 u32 u64}
impl NumHash for usize {
#[inline]
fn num_hash<H: Hasher>(&self, state: &mut H) {
#[cfg(target_pointer_width = "32")]
return (&(*self as u32)).num_hash(state);
#[cfg(target_pointer_width = "64")]
return (&(*self as u64)).num_hash(state);
}
}
impl NumHash for isize {
#[inline]
fn num_hash<H: Hasher>(&self, state: &mut H) {
#[cfg(target_pointer_width = "32")]
return (&(*self as i32)).num_hash(state);
#[cfg(target_pointer_width = "64")]
return (&(*self as i64)).num_hash(state);
}
}
#[cfg(feature = "num-bigint")]
mod _num_bigint {
use super::*;
use num_bigint::{BigInt, BigUint};
use num_traits::ToPrimitive;
impl NumHash for BigUint {
fn num_hash<H: Hasher>(&self, state: &mut H) {
(self % BigUint::from(M127U)).to_i128().unwrap().hash(state)
}
}
impl NumHash for BigInt {
fn num_hash<H: Hasher>(&self, state: &mut H) {
(self % BigInt::from(M127)).to_i128().unwrap().hash(state)
}
}
}
// Case3: for rational(a, b) including floating numbers, the hash is `hash(a * b^-1 mod M127)` (b > 0)
trait FloatHash {
// Calculate mantissa * exponent^-1 mod M127
fn fhash(&self) -> i128;
}
impl FloatHash for f32 {
fn fhash(&self) -> i128 {
let bits = self.to_bits();
let sign_bit = bits >> 31;
let mantissa_bits = bits & 0x7fffff;
let mut exponent: i16 = ((bits >> 23) & 0xff) as i16;
if exponent == 0xff {
// deal with special floats
if mantissa_bits != 0 {
// nan
HASH_NAN
} else if sign_bit > 0 {
HASH_NEGINF // -inf
} else {
HASH_INF // inf
}
} else {
// then deal with normal floats
let mantissa = if exponent == 0 {
mantissa_bits << 1
} else {
mantissa_bits | 0x800000
};
exponent -= 0x7f + 23;
// calculate hash
let mantissa = MInt::new(mantissa as u128, &M127U);
// m * 2^e mod M127 = m * 2^(e mod 127) mod M127
let pow = mantissa.convert(1 << exponent.absm(&127));
let v = mantissa * pow;
v.residue() as i128 * if sign_bit == 0 { 1 } else { -1 }
}
}
}
impl NumHash for f32 {
fn num_hash<H: Hasher>(&self, state: &mut H) {
self.fhash().num_hash(state)
}
}
impl FloatHash for f64 {
fn fhash(&self) -> i128 {
let bits = self.to_bits();
let sign_bit = bits >> 63;
let mantissa_bits = bits & 0xfffffffffffff;
let mut exponent: i16 = ((bits >> 52) & 0x7ff) as i16;
if exponent == 0x7ff {
// deal with special floats
if mantissa_bits != 0 {
// nan
HASH_NAN
} else if sign_bit > 0 {
HASH_NEGINF // -inf
} else {
HASH_INF // inf
}
} else {
// deal with normal floats
let mantissa = if exponent == 0 {
mantissa_bits << 1
} else {
mantissa_bits | 0x10000000000000
};
// Exponent bias + mantissa shift
exponent -= 0x3ff + 52;
// calculate hash
let mantissa = MInt::new(mantissa as u128, &M127U);
// m * 2^e mod M127 = m * 2^(e mod 127) mod M127
let pow = mantissa.convert(1 << exponent.absm(&127));
let v = mantissa * pow;
v.residue() as i128 * if sign_bit == 0 { 1 } else { -1 }
}
}
}
impl NumHash for f64 {
fn num_hash<H: Hasher>(&self, state: &mut H) {
self.fhash().num_hash(state)
}
}
#[cfg(feature = "num-rational")]
mod _num_rational {
use super::*;
use core::ops::Neg;
use num_rational::Ratio;
macro_rules! impl_hash_for_ratio {
($($int:ty)*) => ($(
impl NumHash for Ratio<$int> {
fn num_hash<H: Hasher>(&self, state: &mut H) {
let ub = *self.denom() as u128; // denom is always positive in Ratio
let binv = if ub != M127U {
MInt::new(ub, &M127U).inv().unwrap()
} else {
// no modular inverse, use INF or NEGINF as the result
return if self.numer() > &0 { HASH_INF.num_hash(state) } else { HASH_NEGINF.num_hash(state) }
};
let ua = if self.numer() < &0 { (*self.numer() as u128).wrapping_neg() } else { *self.numer() as u128 }; // essentially calculate |self.numer()|
let ua = binv.convert(ua);
let ab = (ua * binv).residue() as i128;
if self.numer() >= &0 {
ab.num_hash(state)
} else {
ab.neg().num_hash(state)
}
}
}
)*);
}
impl_hash_for_ratio!(i8 i16 i32 i64 i128 isize);
#[cfg(feature = "num-bigint")]
mod _num_bigint {
use super::*;
use num_bigint::{BigInt, BigUint};
use num_traits::{Signed, ToPrimitive, Zero};
impl NumHash for Ratio<BigInt> {
fn num_hash<H: Hasher>(&self, state: &mut H) {
let ub = (self.denom().magnitude() % BigUint::from(M127U))
.to_u128()
.unwrap();
let binv = if !ub.is_zero() {
MInt::new(ub, &M127U).inv().unwrap()
} else {
// no modular inverse, use INF or NEGINF as the result
return if self.numer().is_negative() {
HASH_NEGINF.num_hash(state)
} else {
HASH_INF.num_hash(state)
};
};
let ua = (self.numer().magnitude() % BigUint::from(M127U))
.to_u128()
.unwrap();
let ua = binv.convert(ua);
let ab = (ua * binv).residue() as i128;
if self.numer().is_negative() {
ab.neg().num_hash(state)
} else {
ab.num_hash(state)
}
}
}
}
}
// Case4: for a + b*sqrt(r) where a, b are rational numbers, the hash is
// - `hash(a + PROOT^2*b^2*r)` if b > 0
// - `hash(a - PROOT^2*b^2*r)` if b < 0
// The generalized version is that, hash of (a + b*r^(1/k)) will be `hash(a + PROOT^k*b^k*r)`
// Some Caveats:
// 1. if r = 1, the hash is not consistent with normal integer, but r = 1 is forbidden in QuadraticSurd
// 2. a - b*sqrt(r) and a + b*sqrt(-r) has the same hash, which is usually not a problem
#[cfg(feature = "num-complex")]
mod _num_complex {
use super::*;
use num_complex::Complex;
macro_rules! impl_complex_hash_for_float {
($($float:ty)*) => ($(
impl NumHash for Complex<$float> {
fn num_hash<H: Hasher>(&self, state: &mut H) {
let a = self.re.fhash();
let b = self.im.fhash();
let bterm = if b >= 0 {
let pb = MInt::new(b as u128, &M127U) * PROOT;
-((pb * pb).residue() as i128)
} else {
let pb = MInt::new((-b) as u128, &M127U) * PROOT;
(pb * pb).residue() as i128
};
(a + bterm).num_hash(state)
}
}
)*);
}
impl_complex_hash_for_float!(f32 f64);
}