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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465
use crate::decoding::bit_reader::BitReader;
use crate::decoding::bit_reader_reverse::{BitReaderReversed, GetBitsError};
use alloc::vec::Vec;
/// FSE decoding involves a decoding table that describes the probabilities of
/// all literals from 0 to the highest present one
///
/// <https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#fse-table-description>
pub struct FSETable {
/// The maximum symbol in the table (inclusive). Limits the probabilities length to max_symbol + 1.
max_symbol: u8,
/// The actual table containing the decoded symbol and the compression data
/// connected to that symbol.
pub decode: Vec<Entry>, //used to decode symbols, and calculate the next state
/// The size of the table is stored in logarithm base 2 format,
/// with the **size of the table** being equal to `(1 << accuracy_log)`.
/// This value is used so that the decoder knows how many bits to read from the bitstream.
pub accuracy_log: u8,
/// In this context, probability refers to the likelihood that a symbol occurs in the given data.
/// Given this info, the encoder can assign shorter codes to symbols that appear more often,
/// and longer codes that appear less often, then the decoder can use the probability
/// to determine what code was assigned to what symbol.
///
/// The probability of a single symbol is a value representing the proportion of times the symbol
/// would fall within the data.
///
/// If a symbol probability is set to `-1`, it means that the probability of a symbol
/// occurring in the data is less than one.
pub symbol_probabilities: Vec<i32>, //used while building the decode Vector
/// The number of times each symbol occurs (The first entry being 0x0, the second being 0x1) and so on
/// up until the highest possible symbol (255).
symbol_counter: Vec<u32>,
}
#[derive(Debug)]
#[non_exhaustive]
pub enum FSETableError {
AccLogIsZero,
AccLogTooBig {
got: u8,
max: u8,
},
GetBitsError(GetBitsError),
ProbabilityCounterMismatch {
got: u32,
expected_sum: u32,
symbol_probabilities: Vec<i32>,
},
TooManySymbols {
got: usize,
},
}
#[cfg(feature = "std")]
impl std::error::Error for FSETableError {
fn source(&self) -> Option<&(dyn std::error::Error + 'static)> {
match self {
FSETableError::GetBitsError(source) => Some(source),
_ => None,
}
}
}
impl core::fmt::Display for FSETableError {
fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
match self {
FSETableError::AccLogIsZero => write!(f, "Acclog must be at least 1"),
FSETableError::AccLogTooBig { got, max } => {
write!(
f,
"Found FSE acc_log: {0} bigger than allowed maximum in this case: {1}",
got, max
)
}
FSETableError::GetBitsError(e) => write!(f, "{:?}", e),
FSETableError::ProbabilityCounterMismatch {
got,
expected_sum,
symbol_probabilities,
} => {
write!(f,
"The counter ({}) exceeded the expected sum: {}. This means an error or corrupted data \n {:?}",
got,
expected_sum,
symbol_probabilities,
)
}
FSETableError::TooManySymbols { got } => {
write!(
f,
"There are too many symbols in this distribution: {}. Max: 256",
got,
)
}
}
}
}
impl From<GetBitsError> for FSETableError {
fn from(val: GetBitsError) -> Self {
Self::GetBitsError(val)
}
}
pub struct FSEDecoder<'table> {
/// An FSE state value represents an index in the FSE table.
pub state: Entry,
/// A reference to the table used for decoding.
table: &'table FSETable,
}
#[derive(Debug)]
#[non_exhaustive]
pub enum FSEDecoderError {
GetBitsError(GetBitsError),
TableIsUninitialized,
}
#[cfg(feature = "std")]
impl std::error::Error for FSEDecoderError {
fn source(&self) -> Option<&(dyn std::error::Error + 'static)> {
match self {
FSEDecoderError::GetBitsError(source) => Some(source),
_ => None,
}
}
}
impl core::fmt::Display for FSEDecoderError {
fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
match self {
FSEDecoderError::GetBitsError(e) => write!(f, "{:?}", e),
FSEDecoderError::TableIsUninitialized => {
write!(f, "Tried to use an uninitialized table!")
}
}
}
}
impl From<GetBitsError> for FSEDecoderError {
fn from(val: GetBitsError) -> Self {
Self::GetBitsError(val)
}
}
/// A single entry in an FSE table.
#[derive(Copy, Clone)]
pub struct Entry {
/// This value is used as an offset value, and it is added
/// to a value read from the stream to determine the next state value.
pub base_line: u32,
/// How many bits should be read from the stream when decoding this entry.
pub num_bits: u8,
/// The byte that should be put in the decode output when encountering this state.
pub symbol: u8,
}
/// This value is added to the first 4 bits of the stream to determine the
/// `Accuracy_Log`
const ACC_LOG_OFFSET: u8 = 5;
fn highest_bit_set(x: u32) -> u32 {
assert!(x > 0);
u32::BITS - x.leading_zeros()
}
impl<'t> FSEDecoder<'t> {
/// Initialize a new Finite State Entropy decoder.
pub fn new(table: &'t FSETable) -> FSEDecoder<'t> {
FSEDecoder {
state: table.decode.first().copied().unwrap_or(Entry {
base_line: 0,
num_bits: 0,
symbol: 0,
}),
table,
}
}
/// Returns the byte associated with the symbol the internal cursor is pointing at.
pub fn decode_symbol(&self) -> u8 {
self.state.symbol
}
/// Initialize internal state and prepare for decoding. After this, `decode_symbol` can be called
/// to read the first symbol and `update_state` can be called to prepare to read the next symbol.
pub fn init_state(&mut self, bits: &mut BitReaderReversed<'_>) -> Result<(), FSEDecoderError> {
if self.table.accuracy_log == 0 {
return Err(FSEDecoderError::TableIsUninitialized);
}
self.state = self.table.decode[bits.get_bits(self.table.accuracy_log) as usize];
Ok(())
}
/// Advance the internal state to decode the next symbol in the bitstream.
pub fn update_state(&mut self, bits: &mut BitReaderReversed<'_>) {
let num_bits = self.state.num_bits;
let add = bits.get_bits(num_bits);
let base_line = self.state.base_line;
let new_state = base_line + add as u32;
self.state = self.table.decode[new_state as usize];
//println!("Update: {}, {} -> {}", base_line, add, self.state);
}
}
impl FSETable {
/// Initialize a new empty Finite State Entropy decoding table.
pub fn new(max_symbol: u8) -> FSETable {
FSETable {
max_symbol,
symbol_probabilities: Vec::with_capacity(256), //will never be more than 256 symbols because u8
symbol_counter: Vec::with_capacity(256), //will never be more than 256 symbols because u8
decode: Vec::new(), //depending on acc_log.
accuracy_log: 0,
}
}
/// Reset `self` and update `self`'s state to mirror the provided table.
pub fn reinit_from(&mut self, other: &Self) {
self.reset();
self.symbol_counter.extend_from_slice(&other.symbol_counter);
self.symbol_probabilities
.extend_from_slice(&other.symbol_probabilities);
self.decode.extend_from_slice(&other.decode);
self.accuracy_log = other.accuracy_log;
}
/// Empty the table and clear all internal state.
pub fn reset(&mut self) {
self.symbol_counter.clear();
self.symbol_probabilities.clear();
self.decode.clear();
self.accuracy_log = 0;
}
/// returns how many BYTEs (not bits) were read while building the decoder
pub fn build_decoder(&mut self, source: &[u8], max_log: u8) -> Result<usize, FSETableError> {
self.accuracy_log = 0;
let bytes_read = self.read_probabilities(source, max_log)?;
self.build_decoding_table()?;
Ok(bytes_read)
}
/// Given the provided accuracy log, build a decoding table from that log.
pub fn build_from_probabilities(
&mut self,
acc_log: u8,
probs: &[i32],
) -> Result<(), FSETableError> {
if acc_log == 0 {
return Err(FSETableError::AccLogIsZero);
}
self.symbol_probabilities = probs.to_vec();
self.accuracy_log = acc_log;
self.build_decoding_table()
}
/// Build the actual decoding table after probabilities have been read into the table.
/// After this function is called, the decoding process can begin.
fn build_decoding_table(&mut self) -> Result<(), FSETableError> {
if self.symbol_probabilities.len() > self.max_symbol as usize + 1 {
return Err(FSETableError::TooManySymbols {
got: self.symbol_probabilities.len(),
});
}
self.decode.clear();
let table_size = 1 << self.accuracy_log;
if self.decode.len() < table_size {
self.decode.reserve(table_size - self.decode.len());
}
//fill with dummy entries
self.decode.resize(
table_size,
Entry {
base_line: 0,
num_bits: 0,
symbol: 0,
},
);
let mut negative_idx = table_size; //will point to the highest index with is already occupied by a negative-probability-symbol
//first scan for all -1 probabilities and place them at the top of the table
for symbol in 0..self.symbol_probabilities.len() {
if self.symbol_probabilities[symbol] == -1 {
negative_idx -= 1;
let entry = &mut self.decode[negative_idx];
entry.symbol = symbol as u8;
entry.base_line = 0;
entry.num_bits = self.accuracy_log;
}
}
//then place in a semi-random order all of the other symbols
let mut position = 0;
for idx in 0..self.symbol_probabilities.len() {
let symbol = idx as u8;
if self.symbol_probabilities[idx] <= 0 {
continue;
}
//for each probability point the symbol gets on slot
let prob = self.symbol_probabilities[idx];
for _ in 0..prob {
let entry = &mut self.decode[position];
entry.symbol = symbol;
position = next_position(position, table_size);
while position >= negative_idx {
position = next_position(position, table_size);
//everything above negative_idx is already taken
}
}
}
// baselines and num_bits can only be calculated when all symbols have been spread
self.symbol_counter.clear();
self.symbol_counter
.resize(self.symbol_probabilities.len(), 0);
for idx in 0..negative_idx {
let entry = &mut self.decode[idx];
let symbol = entry.symbol;
let prob = self.symbol_probabilities[symbol as usize];
let symbol_count = self.symbol_counter[symbol as usize];
let (bl, nb) = calc_baseline_and_numbits(table_size as u32, prob as u32, symbol_count);
//println!("symbol: {:2}, table: {}, prob: {:3}, count: {:3}, bl: {:3}, nb: {:2}", symbol, table_size, prob, symbol_count, bl, nb);
assert!(nb <= self.accuracy_log);
self.symbol_counter[symbol as usize] += 1;
entry.base_line = bl;
entry.num_bits = nb;
}
Ok(())
}
/// Read the accuracy log and the probability table from the source and return the number of bytes
/// read. If the size of the table is larger than the provided `max_log`, return an error.
fn read_probabilities(&mut self, source: &[u8], max_log: u8) -> Result<usize, FSETableError> {
self.symbol_probabilities.clear(); //just clear, we will fill a probability for each entry anyways. No need to force new allocs here
let mut br = BitReader::new(source);
self.accuracy_log = ACC_LOG_OFFSET + (br.get_bits(4)? as u8);
if self.accuracy_log > max_log {
return Err(FSETableError::AccLogTooBig {
got: self.accuracy_log,
max: max_log,
});
}
if self.accuracy_log == 0 {
return Err(FSETableError::AccLogIsZero);
}
let probability_sum = 1 << self.accuracy_log;
let mut probability_counter = 0;
while probability_counter < probability_sum {
let max_remaining_value = probability_sum - probability_counter + 1;
let bits_to_read = highest_bit_set(max_remaining_value);
let unchecked_value = br.get_bits(bits_to_read as usize)? as u32;
let low_threshold = ((1 << bits_to_read) - 1) - (max_remaining_value);
let mask = (1 << (bits_to_read - 1)) - 1;
let small_value = unchecked_value & mask;
let value = if small_value < low_threshold {
br.return_bits(1);
small_value
} else if unchecked_value > mask {
unchecked_value - low_threshold
} else {
unchecked_value
};
//println!("{}, {}, {}", self.symbol_probablilities.len(), unchecked_value, value);
let prob = (value as i32) - 1;
self.symbol_probabilities.push(prob);
if prob != 0 {
if prob > 0 {
probability_counter += prob as u32;
} else {
// probability -1 counts as 1
assert!(prob == -1);
probability_counter += 1;
}
} else {
//fast skip further zero probabilities
loop {
let skip_amount = br.get_bits(2)? as usize;
self.symbol_probabilities
.resize(self.symbol_probabilities.len() + skip_amount, 0);
if skip_amount != 3 {
break;
}
}
}
}
if probability_counter != probability_sum {
return Err(FSETableError::ProbabilityCounterMismatch {
got: probability_counter,
expected_sum: probability_sum,
symbol_probabilities: self.symbol_probabilities.clone(),
});
}
if self.symbol_probabilities.len() > self.max_symbol as usize + 1 {
return Err(FSETableError::TooManySymbols {
got: self.symbol_probabilities.len(),
});
}
let bytes_read = if br.bits_read() % 8 == 0 {
br.bits_read() / 8
} else {
(br.bits_read() / 8) + 1
};
Ok(bytes_read)
}
}
//utility functions for building the decoding table from probabilities
/// Calculate the position of the next entry of the table given the current
/// position and size of the table.
fn next_position(mut p: usize, table_size: usize) -> usize {
p += (table_size >> 1) + (table_size >> 3) + 3;
p &= table_size - 1;
p
}
fn calc_baseline_and_numbits(
num_states_total: u32,
num_states_symbol: u32,
state_number: u32,
) -> (u32, u8) {
let num_state_slices = if 1 << (highest_bit_set(num_states_symbol) - 1) == num_states_symbol {
num_states_symbol
} else {
1 << (highest_bit_set(num_states_symbol))
}; //always power of two
let num_double_width_state_slices = num_state_slices - num_states_symbol; //leftovers to the power of two need to be distributed
let num_single_width_state_slices = num_states_symbol - num_double_width_state_slices; //these will not receive a double width slice of states
let slice_width = num_states_total / num_state_slices; //size of a single width slice of states
let num_bits = highest_bit_set(slice_width) - 1; //number of bits needed to read for one slice
if state_number < num_double_width_state_slices {
let baseline = num_single_width_state_slices * slice_width + state_number * slice_width * 2;
(baseline, num_bits as u8 + 1)
} else {
let index_shifted = state_number - num_double_width_state_slices;
((index_shifted * slice_width), num_bits as u8)
}
}