1 | // Copyright 2018 Developers of the Rand project. |
2 | // |
3 | // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
4 | // https://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
5 | // <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your |
6 | // option. This file may not be copied, modified, or distributed |
7 | // except according to those terms. |
8 | |
9 | //! Helper functions for implementing `RngCore` functions. |
10 | //! |
11 | //! For cross-platform reproducibility, these functions all use Little Endian: |
12 | //! least-significant part first. For example, `next_u64_via_u32` takes `u32` |
13 | //! values `x, y`, then outputs `(y << 32) | x`. To implement `next_u32` |
14 | //! from `next_u64` in little-endian order, one should use `next_u64() as u32`. |
15 | //! |
16 | //! Byte-swapping (like the std `to_le` functions) is only needed to convert |
17 | //! to/from byte sequences, and since its purpose is reproducibility, |
18 | //! non-reproducible sources (e.g. `OsRng`) need not bother with it. |
19 | |
20 | use crate::RngCore; |
21 | use core::cmp::min; |
22 | |
23 | /// Implement `next_u64` via `next_u32`, little-endian order. |
24 | pub fn next_u64_via_u32<R: RngCore + ?Sized>(rng: &mut R) -> u64 { |
25 | // Use LE; we explicitly generate one value before the next. |
26 | let x = u64::from(rng.next_u32()); |
27 | let y = u64::from(rng.next_u32()); |
28 | (y << 32) | x |
29 | } |
30 | |
31 | /// Implement `fill_bytes` via `next_u64` and `next_u32`, little-endian order. |
32 | /// |
33 | /// The fastest way to fill a slice is usually to work as long as possible with |
34 | /// integers. That is why this method mostly uses `next_u64`, and only when |
35 | /// there are 4 or less bytes remaining at the end of the slice it uses |
36 | /// `next_u32` once. |
37 | pub fn fill_bytes_via_next<R: RngCore + ?Sized>(rng: &mut R, dest: &mut [u8]) { |
38 | let mut left = dest; |
39 | while left.len() >= 8 { |
40 | let (l, r) = { left }.split_at_mut(8); |
41 | left = r; |
42 | let chunk: [u8; 8] = rng.next_u64().to_le_bytes(); |
43 | l.copy_from_slice(&chunk); |
44 | } |
45 | let n = left.len(); |
46 | if n > 4 { |
47 | let chunk: [u8; 8] = rng.next_u64().to_le_bytes(); |
48 | left.copy_from_slice(&chunk[..n]); |
49 | } else if n > 0 { |
50 | let chunk: [u8; 4] = rng.next_u32().to_le_bytes(); |
51 | left.copy_from_slice(&chunk[..n]); |
52 | } |
53 | } |
54 | |
55 | trait Observable: Copy { |
56 | type Bytes: AsRef<[u8]>; |
57 | fn to_le_bytes(self) -> Self::Bytes; |
58 | |
59 | // Contract: observing self is memory-safe (implies no uninitialised padding) |
60 | fn as_byte_slice(x: &[Self]) -> &[u8]; |
61 | } |
62 | impl Observable for u32 { |
63 | type Bytes = [u8; 4]; |
64 | fn to_le_bytes(self) -> Self::Bytes { |
65 | self.to_le_bytes() |
66 | } |
67 | fn as_byte_slice(x: &[Self]) -> &[u8] { |
68 | let ptr = x.as_ptr() as *const u8; |
69 | let len = x.len() * core::mem::size_of::<Self>(); |
70 | unsafe { core::slice::from_raw_parts(ptr, len) } |
71 | } |
72 | } |
73 | impl Observable for u64 { |
74 | type Bytes = [u8; 8]; |
75 | fn to_le_bytes(self) -> Self::Bytes { |
76 | self.to_le_bytes() |
77 | } |
78 | fn as_byte_slice(x: &[Self]) -> &[u8] { |
79 | let ptr = x.as_ptr() as *const u8; |
80 | let len = x.len() * core::mem::size_of::<Self>(); |
81 | unsafe { core::slice::from_raw_parts(ptr, len) } |
82 | } |
83 | } |
84 | |
85 | fn fill_via_chunks<T: Observable>(src: &[T], dest: &mut [u8]) -> (usize, usize) { |
86 | let size = core::mem::size_of::<T>(); |
87 | let byte_len = min(src.len() * size, dest.len()); |
88 | let num_chunks = (byte_len + size - 1) / size; |
89 | |
90 | if cfg!(target_endian = "little" ) { |
91 | // On LE we can do a simple copy, which is 25-50% faster: |
92 | dest[..byte_len].copy_from_slice(&T::as_byte_slice(&src[..num_chunks])[..byte_len]); |
93 | } else { |
94 | // This code is valid on all arches, but slower than the above: |
95 | let mut i = 0; |
96 | let mut iter = dest[..byte_len].chunks_exact_mut(size); |
97 | for chunk in &mut iter { |
98 | chunk.copy_from_slice(src[i].to_le_bytes().as_ref()); |
99 | i += 1; |
100 | } |
101 | let chunk = iter.into_remainder(); |
102 | if !chunk.is_empty() { |
103 | chunk.copy_from_slice(&src[i].to_le_bytes().as_ref()[..chunk.len()]); |
104 | } |
105 | } |
106 | |
107 | (num_chunks, byte_len) |
108 | } |
109 | |
110 | /// Implement `fill_bytes` by reading chunks from the output buffer of a block |
111 | /// based RNG. |
112 | /// |
113 | /// The return values are `(consumed_u32, filled_u8)`. |
114 | /// |
115 | /// `filled_u8` is the number of filled bytes in `dest`, which may be less than |
116 | /// the length of `dest`. |
117 | /// `consumed_u32` is the number of words consumed from `src`, which is the same |
118 | /// as `filled_u8 / 4` rounded up. |
119 | /// |
120 | /// # Example |
121 | /// (from `IsaacRng`) |
122 | /// |
123 | /// ```ignore |
124 | /// fn fill_bytes(&mut self, dest: &mut [u8]) { |
125 | /// let mut read_len = 0; |
126 | /// while read_len < dest.len() { |
127 | /// if self.index >= self.rsl.len() { |
128 | /// self.isaac(); |
129 | /// } |
130 | /// |
131 | /// let (consumed_u32, filled_u8) = |
132 | /// impls::fill_via_u32_chunks(&mut self.rsl[self.index..], |
133 | /// &mut dest[read_len..]); |
134 | /// |
135 | /// self.index += consumed_u32; |
136 | /// read_len += filled_u8; |
137 | /// } |
138 | /// } |
139 | /// ``` |
140 | pub fn fill_via_u32_chunks(src: &[u32], dest: &mut [u8]) -> (usize, usize) { |
141 | fill_via_chunks(src, dest) |
142 | } |
143 | |
144 | /// Implement `fill_bytes` by reading chunks from the output buffer of a block |
145 | /// based RNG. |
146 | /// |
147 | /// The return values are `(consumed_u64, filled_u8)`. |
148 | /// `filled_u8` is the number of filled bytes in `dest`, which may be less than |
149 | /// the length of `dest`. |
150 | /// `consumed_u64` is the number of words consumed from `src`, which is the same |
151 | /// as `filled_u8 / 8` rounded up. |
152 | /// |
153 | /// See `fill_via_u32_chunks` for an example. |
154 | pub fn fill_via_u64_chunks(src: &[u64], dest: &mut [u8]) -> (usize, usize) { |
155 | fill_via_chunks(src, dest) |
156 | } |
157 | |
158 | /// Implement `next_u32` via `fill_bytes`, little-endian order. |
159 | pub fn next_u32_via_fill<R: RngCore + ?Sized>(rng: &mut R) -> u32 { |
160 | let mut buf = [0; 4]; |
161 | rng.fill_bytes(&mut buf); |
162 | u32::from_le_bytes(buf) |
163 | } |
164 | |
165 | /// Implement `next_u64` via `fill_bytes`, little-endian order. |
166 | pub fn next_u64_via_fill<R: RngCore + ?Sized>(rng: &mut R) -> u64 { |
167 | let mut buf = [0; 8]; |
168 | rng.fill_bytes(&mut buf); |
169 | u64::from_le_bytes(buf) |
170 | } |
171 | |
172 | #[cfg (test)] |
173 | mod test { |
174 | use super::*; |
175 | |
176 | #[test] |
177 | fn test_fill_via_u32_chunks() { |
178 | let src = [1, 2, 3]; |
179 | let mut dst = [0u8; 11]; |
180 | assert_eq!(fill_via_u32_chunks(&src, &mut dst), (3, 11)); |
181 | assert_eq!(dst, [1, 0, 0, 0, 2, 0, 0, 0, 3, 0, 0]); |
182 | |
183 | let mut dst = [0u8; 13]; |
184 | assert_eq!(fill_via_u32_chunks(&src, &mut dst), (3, 12)); |
185 | assert_eq!(dst, [1, 0, 0, 0, 2, 0, 0, 0, 3, 0, 0, 0, 0]); |
186 | |
187 | let mut dst = [0u8; 5]; |
188 | assert_eq!(fill_via_u32_chunks(&src, &mut dst), (2, 5)); |
189 | assert_eq!(dst, [1, 0, 0, 0, 2]); |
190 | } |
191 | |
192 | #[test] |
193 | fn test_fill_via_u64_chunks() { |
194 | let src = [1, 2]; |
195 | let mut dst = [0u8; 11]; |
196 | assert_eq!(fill_via_u64_chunks(&src, &mut dst), (2, 11)); |
197 | assert_eq!(dst, [1, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0]); |
198 | |
199 | let mut dst = [0u8; 17]; |
200 | assert_eq!(fill_via_u64_chunks(&src, &mut dst), (2, 16)); |
201 | assert_eq!(dst, [1, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0]); |
202 | |
203 | let mut dst = [0u8; 5]; |
204 | assert_eq!(fill_via_u64_chunks(&src, &mut dst), (1, 5)); |
205 | assert_eq!(dst, [1, 0, 0, 0, 0]); |
206 | } |
207 | } |
208 | |