2020-06-01 22:13:56 +03:00
|
|
|
// Copyright (c) 2019-2020 Alexander Medvednikov. All rights reserved.
|
|
|
|
// Use of this source code is governed by an MIT license
|
|
|
|
// that can be found in the LICENSE file.
|
|
|
|
module rand
|
|
|
|
|
|
|
|
import math.bits
|
|
|
|
|
|
|
|
// Implementation note:
|
|
|
|
// ====================
|
|
|
|
// C.rand() is okay to use within its defined range of C.RAND_MAX.
|
|
|
|
// (See: https://web.archive.org/web/20180801210127/http://eternallyconfuzzled.com/arts/jsw_art_rand.aspx)
|
|
|
|
// The problem is, this value varies with the libc implementation. On windows,
|
|
|
|
// for example, RAND_MAX is usually a measly 32767, whereas on (newer) linux it's generaly
|
|
|
|
// 2147483647. The repetition period also varies wildly. In order to provide more entropy
|
|
|
|
// without altering the underlying algorithm too much, this implementation simply
|
|
|
|
// requests for more random bits until the necessary width for the integers is achieved.
|
|
|
|
const (
|
|
|
|
rand_limit = u64(C.RAND_MAX)
|
|
|
|
rand_bitsize = bits.len_64(rand_limit)
|
|
|
|
u32_iter_count = calculate_iterations_for(32)
|
|
|
|
u64_iter_count = calculate_iterations_for(64)
|
|
|
|
)
|
|
|
|
|
|
|
|
fn calculate_iterations_for(bits int) int {
|
|
|
|
base := bits / rand_bitsize
|
|
|
|
extra := if bits % rand_bitsize == 0 { 0 } else { 1 }
|
|
|
|
return base + extra
|
|
|
|
}
|
|
|
|
|
|
|
|
// Size constants to avoid importing the entire math module
|
|
|
|
const (
|
|
|
|
max_u32 = 0xFFFFFFFF
|
|
|
|
max_u64 = 0xFFFFFFFFFFFFFFFF
|
|
|
|
max_u32_as_f32 = f32(max_u32)
|
|
|
|
max_u64_as_f64 = f64(max_u64)
|
|
|
|
)
|
|
|
|
|
|
|
|
// Masks for fast modular division
|
|
|
|
const (
|
|
|
|
u31_mask = u32(0x7FFFFFFF)
|
|
|
|
u63_mask = u64(0x7FFFFFFFFFFFFFFF)
|
|
|
|
)
|
|
|
|
|
|
|
|
// C.rand returns a pseudorandom integer from 0 (inclusive) to C.RAND_MAX (exclusive)
|
|
|
|
fn C.rand() int
|
|
|
|
|
|
|
|
// C.srand seeds the internal PRNG with the given int seed.
|
|
|
|
// fn C.srand(seed int)
|
|
|
|
// SysRNG is the PRNG provided by default in the libc implementiation that V uses.
|
|
|
|
pub struct SysRNG {
|
|
|
|
mut:
|
|
|
|
seed u32 = time_seed_32()
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.seed() sets the seed of the accepting SysRNG to the given data.
|
|
|
|
pub fn (mut r SysRNG) seed(seed_data []u32) {
|
|
|
|
if seed_data.len != 1 {
|
|
|
|
eprintln('SysRNG needs one 32-bit unsigned integer as the seed.')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
r.seed = seed_data[0]
|
|
|
|
C.srand(int(r.seed))
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.default_rand() exposes the default behavior of the system's RNG
|
|
|
|
// (equivalent to calling C.rand()). Recommended for testing/comparison
|
|
|
|
// b/w V and other languages using libc and not for regular use.
|
|
|
|
// This is also a one-off feature of SysRNG, similar to the global seed
|
|
|
|
// situation. Other generators will not have this.
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) default_rand() int {
|
|
|
|
return C.rand()
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.u32() returns a pseudorandom u32 value less than 2^32
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) u32() u32 {
|
|
|
|
mut result := u32(C.rand())
|
|
|
|
for i in 1 .. u32_iter_count {
|
|
|
|
result = result ^ (u32(C.rand()) << (rand_bitsize * i))
|
|
|
|
}
|
|
|
|
return result
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.u64() returns a pseudorandom u64 value less than 2^64
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) u64() u64 {
|
|
|
|
mut result := u64(C.rand())
|
|
|
|
for i in 1 .. u64_iter_count {
|
|
|
|
result = result ^ (u64(C.rand()) << (rand_bitsize * i))
|
|
|
|
}
|
|
|
|
return result
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.u32n(max) returns a pseudorandom u32 value that is guaranteed to be less than max
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) u32n(max u32) u32 {
|
|
|
|
if max == 0 {
|
|
|
|
eprintln('max must be positive integer')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
// Owing to the pigeon-hole principle, we can't simply do
|
|
|
|
// val := rng.u32() % max.
|
|
|
|
// It'll wreck the properties of the distribution unless
|
|
|
|
// max evenly divides 2^32. So we divide evenly to
|
|
|
|
// the closest power of two. Then we loop until we find
|
|
|
|
// an int in the required range
|
|
|
|
bit_len := bits.len_32(max)
|
|
|
|
if bit_len == 32 {
|
|
|
|
for {
|
|
|
|
value := r.u32()
|
|
|
|
if value < max {
|
|
|
|
return value
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
mask := (u32(1) << (bit_len + 1)) - 1
|
|
|
|
for {
|
|
|
|
value := r.u32() & mask
|
|
|
|
if value < max {
|
|
|
|
return value
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return u32(0)
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.u64n(max) returns a pseudorandom u64 value that is guaranteed to be less than max
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) u64n(max u64) u64 {
|
|
|
|
if max == 0 {
|
|
|
|
eprintln('max must be positive integer')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
// Similar procedure for u64s
|
|
|
|
bit_len := bits.len_64(max)
|
|
|
|
if bit_len == 64 {
|
|
|
|
for {
|
|
|
|
value := r.u64()
|
|
|
|
if value < max {
|
|
|
|
return value
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
mask := (u64(1) << (bit_len + 1)) - 1
|
|
|
|
for {
|
|
|
|
value := r.u64() & mask
|
|
|
|
if value < max {
|
|
|
|
return value
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return u64(0)
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.u32n(min, max) returns a pseudorandom u32 value that is guaranteed to be in [min, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) u32_in_range(min, max u32) u32 {
|
|
|
|
if max <= min {
|
|
|
|
eprintln('max must be greater than min')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
return min + r.u32n(max - min)
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.u64n(min, max) returns a pseudorandom u64 value that is guaranteed to be in [min, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) u64_in_range(min, max u64) u64 {
|
|
|
|
if max <= min {
|
|
|
|
eprintln('max must be greater than min')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
return min + r.u64n(max - min)
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.int() returns a pseudorandom 32-bit int (which may be negative)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) int() int {
|
|
|
|
return int(r.u32())
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.i64() returns a pseudorandom 64-bit i64 (which may be negative)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) i64() i64 {
|
|
|
|
return i64(r.u64())
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.int31() returns a pseudorandom 31-bit int which is non-negative
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) int31() int {
|
|
|
|
return int(r.u32() & u31_mask) // Set the 32nd bit to 0.
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.int63() returns a pseudorandom 63-bit int which is non-negative
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) int63() i64 {
|
|
|
|
return i64(r.u64() & u63_mask) // Set the 64th bit to 0.
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.intn(max) returns a pseudorandom int that lies in [0, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) intn(max int) int {
|
|
|
|
if max <= 0 {
|
|
|
|
eprintln('max has to be positive.')
|
|
|
|
exit(1)
|
|
|
|
}
|
2020-06-02 07:39:38 +03:00
|
|
|
return int(r.u32n(u32(max)))
|
2020-06-01 22:13:56 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// r.i64n(max) returns a pseudorandom i64 that lies in [0, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) i64n(max i64) i64 {
|
|
|
|
if max <= 0 {
|
|
|
|
eprintln('max has to be positive.')
|
|
|
|
exit(1)
|
|
|
|
}
|
2020-06-02 07:39:38 +03:00
|
|
|
return i64(r.u64n(u64(max)))
|
2020-06-01 22:13:56 +03:00
|
|
|
}
|
|
|
|
|
|
|
|
// r.int_in_range(min, max) returns a pseudorandom int that lies in [min, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) int_in_range(min, max int) int {
|
|
|
|
if max <= min {
|
|
|
|
eprintln('max must be greater than min')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
// This supports negative ranges like [-10, -5) because the difference is positive
|
|
|
|
return min + r.intn(max - min)
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.i64_in_range(min, max) returns a pseudorandom i64 that lies in [min, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) i64_in_range(min, max i64) i64 {
|
|
|
|
if max <= min {
|
|
|
|
eprintln('max must be greater than min')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
return min + r.i64n(max - min)
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.f32() returns a pseudorandom f32 value between 0.0 (inclusive) and 1.0 (exclusive) i.e [0, 1)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) f32() f32 {
|
|
|
|
return f32(r.u32()) / max_u32_as_f32
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.f64() returns a pseudorandom f64 value between 0.0 (inclusive) and 1.0 (exclusive) i.e [0, 1)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) f64() f64 {
|
|
|
|
return f64(r.u64()) / max_u64_as_f64
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.f32n() returns a pseudorandom f32 value in [0, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) f32n(max f32) f32 {
|
|
|
|
if max <= 0 {
|
|
|
|
eprintln('max has to be positive.')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
return r.f32() * max
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.f64n() returns a pseudorandom f64 value in [0, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) f64n(max f64) f64 {
|
|
|
|
if max <= 0 {
|
|
|
|
eprintln('max has to be positive.')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
return r.f64() * max
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.f32_in_range(min, max) returns a pseudorandom f32 that lies in [min, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) f32_in_range(min, max f32) f32 {
|
|
|
|
if max <= min {
|
|
|
|
eprintln('max must be greater than min')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
return min + r.f32n(max - min)
|
|
|
|
}
|
|
|
|
|
|
|
|
// r.i64_in_range(min, max) returns a pseudorandom i64 that lies in [min, max)
|
|
|
|
[inline]
|
|
|
|
pub fn (r SysRNG) f64_in_range(min, max f64) f64 {
|
|
|
|
if max <= min {
|
|
|
|
eprintln('max must be greater than min')
|
|
|
|
exit(1)
|
|
|
|
}
|
|
|
|
return min + r.f64n(max - min)
|
|
|
|
}
|