Source file src/math/rand/v2/rand.go
1 // Copyright 2009 The Go Authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style 3 // license that can be found in the LICENSE file. 4 5 // Package rand implements pseudo-random number generators suitable for tasks 6 // such as simulation, but it should not be used for security-sensitive work. 7 // 8 // Random numbers are generated by a [Source], usually wrapped in a [Rand]. 9 // Both types should be used by a single goroutine at a time: sharing among 10 // multiple goroutines requires some kind of synchronization. 11 // 12 // Top-level functions, such as [Float64] and [Int], 13 // are safe for concurrent use by multiple goroutines. 14 // 15 // This package's outputs might be easily predictable regardless of how it's 16 // seeded. For random numbers suitable for security-sensitive work, see the 17 // [crypto/rand] package. 18 package rand 19 20 import ( 21 "math/bits" 22 _ "unsafe" // for go:linkname 23 ) 24 25 // A Source is a source of uniformly-distributed 26 // pseudo-random uint64 values in the range [0, 1<<64). 27 // 28 // A Source is not safe for concurrent use by multiple goroutines. 29 type Source interface { 30 Uint64() uint64 31 } 32 33 // A Rand is a source of random numbers. 34 type Rand struct { 35 src Source 36 } 37 38 // New returns a new Rand that uses random values from src 39 // to generate other random values. 40 func New(src Source) *Rand { 41 return &Rand{src: src} 42 } 43 44 // Int64 returns a non-negative pseudo-random 63-bit integer as an int64. 45 func (r *Rand) Int64() int64 { return int64(r.src.Uint64() &^ (1 << 63)) } 46 47 // Uint32 returns a pseudo-random 32-bit value as a uint32. 48 func (r *Rand) Uint32() uint32 { return uint32(r.src.Uint64() >> 32) } 49 50 // Uint64 returns a pseudo-random 64-bit value as a uint64. 51 func (r *Rand) Uint64() uint64 { return r.src.Uint64() } 52 53 // Int32 returns a non-negative pseudo-random 31-bit integer as an int32. 54 func (r *Rand) Int32() int32 { return int32(r.src.Uint64() >> 33) } 55 56 // Int returns a non-negative pseudo-random int. 57 func (r *Rand) Int() int { return int(uint(r.src.Uint64()) << 1 >> 1) } 58 59 // Uint returns a pseudo-random uint. 60 func (r *Rand) Uint() uint { return uint(r.src.Uint64()) } 61 62 // Int64N returns, as an int64, a non-negative pseudo-random number in the half-open interval [0,n). 63 // It panics if n <= 0. 64 func (r *Rand) Int64N(n int64) int64 { 65 if n <= 0 { 66 panic("invalid argument to Int64N") 67 } 68 return int64(r.uint64n(uint64(n))) 69 } 70 71 // Uint64N returns, as a uint64, a non-negative pseudo-random number in the half-open interval [0,n). 72 // It panics if n == 0. 73 func (r *Rand) Uint64N(n uint64) uint64 { 74 if n == 0 { 75 panic("invalid argument to Uint64N") 76 } 77 return r.uint64n(n) 78 } 79 80 // uint64n is the no-bounds-checks version of Uint64N. 81 func (r *Rand) uint64n(n uint64) uint64 { 82 if is32bit && uint64(uint32(n)) == n { 83 return uint64(r.uint32n(uint32(n))) 84 } 85 if n&(n-1) == 0 { // n is power of two, can mask 86 return r.Uint64() & (n - 1) 87 } 88 89 // Suppose we have a uint64 x uniform in the range [0,2⁶⁴) 90 // and want to reduce it to the range [0,n) preserving exact uniformity. 91 // We can simulate a scaling arbitrary precision x * (n/2⁶⁴) by 92 // the high bits of a double-width multiply of x*n, meaning (x*n)/2⁶⁴. 93 // Since there are 2⁶⁴ possible inputs x and only n possible outputs, 94 // the output is necessarily biased if n does not divide 2⁶⁴. 95 // In general (x*n)/2⁶⁴ = k for x*n in [k*2⁶⁴,(k+1)*2⁶⁴). 96 // There are either floor(2⁶⁴/n) or ceil(2⁶⁴/n) possible products 97 // in that range, depending on k. 98 // But suppose we reject the sample and try again when 99 // x*n is in [k*2⁶⁴, k*2⁶⁴+(2⁶⁴%n)), meaning rejecting fewer than n possible 100 // outcomes out of the 2⁶⁴. 101 // Now there are exactly floor(2⁶⁴/n) possible ways to produce 102 // each output value k, so we've restored uniformity. 103 // To get valid uint64 math, 2⁶⁴ % n = (2⁶⁴ - n) % n = -n % n, 104 // so the direct implementation of this algorithm would be: 105 // 106 // hi, lo := bits.Mul64(r.Uint64(), n) 107 // thresh := -n % n 108 // for lo < thresh { 109 // hi, lo = bits.Mul64(r.Uint64(), n) 110 // } 111 // 112 // That still leaves an expensive 64-bit division that we would rather avoid. 113 // We know that thresh < n, and n is usually much less than 2⁶⁴, so we can 114 // avoid the last four lines unless lo < n. 115 // 116 // See also: 117 // https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction 118 // https://lemire.me/blog/2016/06/30/fast-random-shuffling 119 hi, lo := bits.Mul64(r.Uint64(), n) 120 if lo < n { 121 thresh := -n % n 122 for lo < thresh { 123 hi, lo = bits.Mul64(r.Uint64(), n) 124 } 125 } 126 return hi 127 } 128 129 // uint32n is an identical computation to uint64n 130 // but optimized for 32-bit systems. 131 func (r *Rand) uint32n(n uint32) uint32 { 132 if n&(n-1) == 0 { // n is power of two, can mask 133 return uint32(r.Uint64()) & (n - 1) 134 } 135 // On 64-bit systems we still use the uint64 code below because 136 // the probability of a random uint64 lo being < a uint32 n is near zero, 137 // meaning the unbiasing loop almost never runs. 138 // On 32-bit systems, here we need to implement that same logic in 32-bit math, 139 // both to preserve the exact output sequence observed on 64-bit machines 140 // and to preserve the optimization that the unbiasing loop almost never runs. 141 // 142 // We want to compute 143 // hi, lo := bits.Mul64(r.Uint64(), n) 144 // In terms of 32-bit halves, this is: 145 // x1:x0 := r.Uint64() 146 // 0:hi, lo1:lo0 := bits.Mul64(x1:x0, 0:n) 147 // Writing out the multiplication in terms of bits.Mul32 allows 148 // using direct hardware instructions and avoiding 149 // the computations involving these zeros. 150 x := r.Uint64() 151 lo1a, lo0 := bits.Mul32(uint32(x), n) 152 hi, lo1b := bits.Mul32(uint32(x>>32), n) 153 lo1, c := bits.Add32(lo1a, lo1b, 0) 154 hi += c 155 if lo1 == 0 && lo0 < uint32(n) { 156 n64 := uint64(n) 157 thresh := uint32(-n64 % n64) 158 for lo1 == 0 && lo0 < thresh { 159 x := r.Uint64() 160 lo1a, lo0 = bits.Mul32(uint32(x), n) 161 hi, lo1b = bits.Mul32(uint32(x>>32), n) 162 lo1, c = bits.Add32(lo1a, lo1b, 0) 163 hi += c 164 } 165 } 166 return hi 167 } 168 169 // Int32N returns, as an int32, a non-negative pseudo-random number in the half-open interval [0,n). 170 // It panics if n <= 0. 171 func (r *Rand) Int32N(n int32) int32 { 172 if n <= 0 { 173 panic("invalid argument to Int32N") 174 } 175 return int32(r.uint64n(uint64(n))) 176 } 177 178 // Uint32N returns, as a uint32, a non-negative pseudo-random number in the half-open interval [0,n). 179 // It panics if n == 0. 180 func (r *Rand) Uint32N(n uint32) uint32 { 181 if n == 0 { 182 panic("invalid argument to Uint32N") 183 } 184 return uint32(r.uint64n(uint64(n))) 185 } 186 187 const is32bit = ^uint(0)>>32 == 0 188 189 // IntN returns, as an int, a non-negative pseudo-random number in the half-open interval [0,n). 190 // It panics if n <= 0. 191 func (r *Rand) IntN(n int) int { 192 if n <= 0 { 193 panic("invalid argument to IntN") 194 } 195 return int(r.uint64n(uint64(n))) 196 } 197 198 // UintN returns, as a uint, a non-negative pseudo-random number in the half-open interval [0,n). 199 // It panics if n == 0. 200 func (r *Rand) UintN(n uint) uint { 201 if n == 0 { 202 panic("invalid argument to UintN") 203 } 204 return uint(r.uint64n(uint64(n))) 205 } 206 207 // Float64 returns, as a float64, a pseudo-random number in the half-open interval [0.0,1.0). 208 func (r *Rand) Float64() float64 { 209 // There are exactly 1<<53 float64s in [0,1). Use Intn(1<<53) / (1<<53). 210 return float64(r.Uint64()<<11>>11) / (1 << 53) 211 } 212 213 // Float32 returns, as a float32, a pseudo-random number in the half-open interval [0.0,1.0). 214 func (r *Rand) Float32() float32 { 215 // There are exactly 1<<24 float32s in [0,1). Use Intn(1<<24) / (1<<24). 216 return float32(r.Uint32()<<8>>8) / (1 << 24) 217 } 218 219 // Perm returns, as a slice of n ints, a pseudo-random permutation of the integers 220 // in the half-open interval [0,n). 221 func (r *Rand) Perm(n int) []int { 222 p := make([]int, n) 223 for i := range p { 224 p[i] = i 225 } 226 r.Shuffle(len(p), func(i, j int) { p[i], p[j] = p[j], p[i] }) 227 return p 228 } 229 230 // Shuffle pseudo-randomizes the order of elements. 231 // n is the number of elements. Shuffle panics if n < 0. 232 // swap swaps the elements with indexes i and j. 233 func (r *Rand) Shuffle(n int, swap func(i, j int)) { 234 if n < 0 { 235 panic("invalid argument to Shuffle") 236 } 237 238 // Fisher-Yates shuffle: https://en.wikipedia.org/wiki/Fisher%E2%80%93Yates_shuffle 239 // Shuffle really ought not be called with n that doesn't fit in 32 bits. 240 // Not only will it take a very long time, but with 2³¹! possible permutations, 241 // there's no way that any PRNG can have a big enough internal state to 242 // generate even a minuscule percentage of the possible permutations. 243 // Nevertheless, the right API signature accepts an int n, so handle it as best we can. 244 for i := n - 1; i > 0; i-- { 245 j := int(r.uint64n(uint64(i + 1))) 246 swap(i, j) 247 } 248 } 249 250 /* 251 * Top-level convenience functions 252 */ 253 254 // globalRand is the source of random numbers for the top-level 255 // convenience functions. 256 var globalRand = &Rand{src: runtimeSource{}} 257 258 //go:linkname runtime_rand runtime.rand 259 func runtime_rand() uint64 260 261 // runtimeSource is a Source that uses the runtime fastrand functions. 262 type runtimeSource struct{} 263 264 func (runtimeSource) Uint64() uint64 { 265 return runtime_rand() 266 } 267 268 // Int64 returns a non-negative pseudo-random 63-bit integer as an int64 269 // from the default Source. 270 func Int64() int64 { return globalRand.Int64() } 271 272 // Uint32 returns a pseudo-random 32-bit value as a uint32 273 // from the default Source. 274 func Uint32() uint32 { return globalRand.Uint32() } 275 276 // Uint64N returns, as a uint64, a pseudo-random number in the half-open interval [0,n) 277 // from the default Source. 278 // It panics if n <= 0. 279 func Uint64N(n uint64) uint64 { return globalRand.Uint64N(n) } 280 281 // Uint32N returns, as a uint32, a pseudo-random number in the half-open interval [0,n) 282 // from the default Source. 283 // It panics if n <= 0. 284 func Uint32N(n uint32) uint32 { return globalRand.Uint32N(n) } 285 286 // Uint64 returns a pseudo-random 64-bit value as a uint64 287 // from the default Source. 288 func Uint64() uint64 { return globalRand.Uint64() } 289 290 // Int32 returns a non-negative pseudo-random 31-bit integer as an int32 291 // from the default Source. 292 func Int32() int32 { return globalRand.Int32() } 293 294 // Int returns a non-negative pseudo-random int from the default Source. 295 func Int() int { return globalRand.Int() } 296 297 // Uint returns a pseudo-random uint from the default Source. 298 func Uint() uint { return globalRand.Uint() } 299 300 // Int64N returns, as an int64, a pseudo-random number in the half-open interval [0,n) 301 // from the default Source. 302 // It panics if n <= 0. 303 func Int64N(n int64) int64 { return globalRand.Int64N(n) } 304 305 // Int32N returns, as an int32, a pseudo-random number in the half-open interval [0,n) 306 // from the default Source. 307 // It panics if n <= 0. 308 func Int32N(n int32) int32 { return globalRand.Int32N(n) } 309 310 // IntN returns, as an int, a pseudo-random number in the half-open interval [0,n) 311 // from the default Source. 312 // It panics if n <= 0. 313 func IntN(n int) int { return globalRand.IntN(n) } 314 315 // UintN returns, as a uint, a pseudo-random number in the half-open interval [0,n) 316 // from the default Source. 317 // It panics if n <= 0. 318 func UintN(n uint) uint { return globalRand.UintN(n) } 319 320 // N returns a pseudo-random number in the half-open interval [0,n) from the default Source. 321 // The type parameter Int can be any integer type. 322 // It panics if n <= 0. 323 func N[Int intType](n Int) Int { 324 if n <= 0 { 325 panic("invalid argument to N") 326 } 327 return Int(globalRand.uint64n(uint64(n))) 328 } 329 330 type intType interface { 331 ~int | ~int8 | ~int16 | ~int32 | ~int64 | 332 ~uint | ~uint8 | ~uint16 | ~uint32 | ~uint64 | ~uintptr 333 } 334 335 // Float64 returns, as a float64, a pseudo-random number in the half-open interval [0.0,1.0) 336 // from the default Source. 337 func Float64() float64 { return globalRand.Float64() } 338 339 // Float32 returns, as a float32, a pseudo-random number in the half-open interval [0.0,1.0) 340 // from the default Source. 341 func Float32() float32 { return globalRand.Float32() } 342 343 // Perm returns, as a slice of n ints, a pseudo-random permutation of the integers 344 // in the half-open interval [0,n) from the default Source. 345 func Perm(n int) []int { return globalRand.Perm(n) } 346 347 // Shuffle pseudo-randomizes the order of elements using the default Source. 348 // n is the number of elements. Shuffle panics if n < 0. 349 // swap swaps the elements with indexes i and j. 350 func Shuffle(n int, swap func(i, j int)) { globalRand.Shuffle(n, swap) } 351 352 // NormFloat64 returns a normally distributed float64 in the range 353 // [-math.MaxFloat64, +math.MaxFloat64] with 354 // standard normal distribution (mean = 0, stddev = 1) 355 // from the default Source. 356 // To produce a different normal distribution, callers can 357 // adjust the output using: 358 // 359 // sample = NormFloat64() * desiredStdDev + desiredMean 360 func NormFloat64() float64 { return globalRand.NormFloat64() } 361 362 // ExpFloat64 returns an exponentially distributed float64 in the range 363 // (0, +math.MaxFloat64] with an exponential distribution whose rate parameter 364 // (lambda) is 1 and whose mean is 1/lambda (1) from the default Source. 365 // To produce a distribution with a different rate parameter, 366 // callers can adjust the output using: 367 // 368 // sample = ExpFloat64() / desiredRateParameter 369 func ExpFloat64() float64 { return globalRand.ExpFloat64() } 370