Files
zstd-rs/rust/src/zstd_lazy.rs
T
ddidderr 7f16a07375 feat(rust): port lazy block matching
Move greedy, lazy, lazy2, and binary-tree block matchers into Rust. A narrow
C state projection preserves configuration-specific context layout while Rust
owns hash-chain, row-based, attached-dictionary, and external-dictionary
searches and sequence-store updates.

The component map now records the lazy family as migrated; the optimized
matcher and high-level contexts remain outside this commit.

Test Plan:
- cargo test --all-targets
- cargo test --target i686-unknown-linux-gnu --all-targets
- cargo clippy && cargo clippy --benches && cargo clippy --tests
- cargo +nightly fmt
- make -B -C tests -j2 fuzzer zstreamtest invalidDictionaries poolTests
- ./tests/fuzzer -s5346 -i1 --no-big-tests
- ./tests/zstreamtest -i3000 -s334462
- ./tests/invalidDictionaries
- timeout 20s stdbuf -oL ./tests/poolTests

Refs: rust/README.md
2026-07-11 08:06:30 +02:00

2624 lines
90 KiB
Rust

#![allow(non_camel_case_types)]
#![allow(non_snake_case)]
#![allow(clippy::missing_safety_doc)]
#![allow(clippy::too_many_arguments)]
#![allow(clippy::not_unsafe_ptr_arg_deref)]
//! Lazy, lazy2, greedy, and binary-tree block match finders.
//!
//! `ZSTD_MatchState_t` deliberately stays on the C side of the boundary. The
//! small representation below contains only the fields that the match finders
//! use, with mutable scalar fields represented by pointers back into the C
//! match state. This keeps the matching and sequence-generation loops in
//! Rust without making a private C layout part of the Rust ABI.
use crate::bits::ZSTD_highbit32;
use crate::mem::{
MEM_64bits, MEM_isLittleEndian, MEM_read16, MEM_read32, MEM_readLE32, MEM_readLE64, MEM_readST,
};
use std::ffi::c_void;
use std::mem::size_of;
use std::os::raw::c_int;
use std::ptr;
const ZSTD_REP_NUM: usize = 3;
const MINMATCH: usize = 3;
const HASH_READ_SIZE: usize = 8;
const K_LAZY_SKIPPING_STEP: usize = 8;
const K_SEARCH_STRENGTH: usize = 8;
const ZSTD_DUBT_UNSORTED_MARK: u32 = 1;
const ZSTD_LAZY_DDSS_BUCKET_LOG: u32 = 2;
const ZSTD_ROW_HASH_TAG_BITS: u32 = 8;
const ZSTD_ROW_HASH_TAG_MASK: u32 = (1 << ZSTD_ROW_HASH_TAG_BITS) - 1;
const ZSTD_ROW_HASH_CACHE_SIZE: usize = 8;
const ZSTD_ROW_HASH_MAX_ENTRIES: usize = 64;
const REPCODE1_TO_OFFBASE: u32 = 1;
const OFFSET_OFFBASE: u32 = ZSTD_REP_NUM as u32;
const SEARCH_HASH_CHAIN: c_int = 0;
const SEARCH_BINARY_TREE: c_int = 1;
const SEARCH_ROW_HASH: c_int = 2;
const DICT_NO_DICT: c_int = 0;
const DICT_EXT: c_int = 1;
const DICT_MATCH_STATE: c_int = 2;
const DICT_DEDICATED: c_int = 3;
#[repr(C)]
#[derive(Clone, Copy)]
struct SeqDef {
offBase: u32,
litLength: u16,
mlBase: u16,
}
/// The C shim verifies the layout of this leaf structure. The matcher only
/// needs its sequence and literal append cursors.
#[repr(C)]
struct SeqStore_t {
sequencesStart: *mut SeqDef,
sequences: *mut SeqDef,
litStart: *mut u8,
lit: *mut u8,
llCode: *mut u8,
mlCode: *mut u8,
ofCode: *mut u8,
maxNbSeq: usize,
maxNbLit: usize,
longLengthType: c_int,
longLengthPos: u32,
}
/// A field-level view of `ZSTD_MatchState_t` used by the lazy match finders.
///
/// All pointers remain owned by C. `dictMatchState` points at a second view
/// constructed by the C shim for calls that search an attached dictionary.
#[repr(C)]
pub struct ZSTD_RustLazyState {
hashTable: *mut u32,
chainTable: *mut u32,
tagTable: *mut u8,
hashCache: *mut u32,
base: *const u8,
dictBase: *const u8,
nextSrc: *const u8,
dictLimit: u32,
lowLimit: u32,
loadedDictEnd: u32,
nextToUpdate: *mut u32,
lazySkipping: *mut c_int,
hashLog: u32,
chainLog: u32,
minMatch: u32,
searchLog: u32,
windowLog: u32,
rowHashLog: u32,
hashSalt: u64,
hashSaltEntropy: *mut u32,
dictMatchState: *const ZSTD_RustLazyState,
}
#[inline]
fn ptr_lt(left: *const u8, right: *const u8) -> bool {
(left as usize) < (right as usize)
}
#[inline]
fn ptr_le(left: *const u8, right: *const u8) -> bool {
(left as usize) <= (right as usize)
}
#[inline]
fn ptr_gt(left: *const u8, right: *const u8) -> bool {
(left as usize) > (right as usize)
}
#[inline]
unsafe fn index_from(base: *const u8, value: *const u8) -> u32 {
(value as usize).wrapping_sub(base as usize) as u32
}
#[inline]
unsafe fn read32(value: *const u8) -> u32 {
unsafe { MEM_read32(value.cast::<c_void>()) }
}
#[inline]
unsafe fn table_get(table: *const u32, index: usize) -> u32 {
unsafe { *table.add(index) }
}
#[inline]
unsafe fn table_set(table: *mut u32, index: usize, value: u32) {
unsafe { *table.add(index) = value };
}
#[inline]
unsafe fn byte_get(table: *const u8, index: usize) -> u8 {
unsafe { *table.add(index) }
}
#[inline]
unsafe fn byte_set(table: *mut u8, index: usize, value: u8) {
unsafe { *table.add(index) = value };
}
#[inline]
unsafe fn next_to_update(state: &ZSTD_RustLazyState) -> u32 {
unsafe { *state.nextToUpdate }
}
#[inline]
unsafe fn set_next_to_update(state: &ZSTD_RustLazyState, value: u32) {
unsafe { *state.nextToUpdate = value };
}
#[inline]
unsafe fn lazy_skipping(state: &ZSTD_RustLazyState) -> bool {
unsafe { *state.lazySkipping != 0 }
}
#[inline]
unsafe fn set_lazy_skipping(state: &ZSTD_RustLazyState, value: bool) {
unsafe { *state.lazySkipping = c_int::from(value) };
}
#[inline]
unsafe fn add_hash_salt_entropy(state: &ZSTD_RustLazyState, value: u32) {
unsafe {
*state.hashSaltEntropy = (*state.hashSaltEntropy).wrapping_add(value);
}
}
#[inline]
unsafe fn dict_state<'a>(state: &ZSTD_RustLazyState) -> &'a ZSTD_RustLazyState {
debug_assert!(!state.dictMatchState.is_null());
unsafe { &*state.dictMatchState }
}
#[inline]
fn bounded(low: u32, value: u32, high: u32) -> u32 {
value.clamp(low, high)
}
#[inline]
fn hash_shift32(value: u32, hbits: u32) -> usize {
if hbits == 0 {
0
} else {
(value >> (32 - hbits)) as usize
}
}
#[inline]
fn hash_shift64(value: u64, hbits: u32) -> usize {
if hbits == 0 {
0
} else {
(value >> (64 - hbits)) as usize
}
}
#[inline]
unsafe fn hash_ptr(ptr: *const u8, hbits: u32, mls: u32) -> usize {
match mls {
5 => {
let value = unsafe { MEM_readLE64(ptr.cast::<c_void>()) };
hash_shift64(value.wrapping_shl(24).wrapping_mul(889_523_592_379), hbits)
}
6 => {
let value = unsafe { MEM_readLE64(ptr.cast::<c_void>()) };
hash_shift64(
value.wrapping_shl(16).wrapping_mul(227_718_039_650_203),
hbits,
)
}
7 => {
let value = unsafe { MEM_readLE64(ptr.cast::<c_void>()) };
hash_shift64(
value.wrapping_shl(8).wrapping_mul(58_295_818_150_454_627),
hbits,
)
}
8 => {
let value = unsafe { MEM_readLE64(ptr.cast::<c_void>()) };
hash_shift64(value.wrapping_mul(0xCF1B_BCDC_B7A5_6463), hbits)
}
_ => {
let value = unsafe { MEM_readLE32(ptr.cast::<c_void>()) };
hash_shift32(value.wrapping_mul(2_654_435_761), hbits)
}
}
}
#[inline]
unsafe fn hash_ptr_salted(ptr: *const u8, hbits: u32, mls: u32, salt: u64) -> usize {
match mls {
5 => {
let value = unsafe { MEM_readLE64(ptr.cast::<c_void>()) };
hash_shift64(
value.wrapping_shl(24).wrapping_mul(889_523_592_379) ^ salt,
hbits,
)
}
6 => {
let value = unsafe { MEM_readLE64(ptr.cast::<c_void>()) };
hash_shift64(
value.wrapping_shl(16).wrapping_mul(227_718_039_650_203) ^ salt,
hbits,
)
}
7 => {
let value = unsafe { MEM_readLE64(ptr.cast::<c_void>()) };
hash_shift64(
value.wrapping_shl(8).wrapping_mul(58_295_818_150_454_627) ^ salt,
hbits,
)
}
8 => {
let value = unsafe { MEM_readLE64(ptr.cast::<c_void>()) };
hash_shift64(value.wrapping_mul(0xCF1B_BCDC_B7A5_6463) ^ salt, hbits)
}
_ => {
let value = unsafe { MEM_readLE32(ptr.cast::<c_void>()) };
hash_shift32(value.wrapping_mul(2_654_435_761) ^ salt as u32, hbits)
}
}
}
#[inline]
fn common_bytes(word: usize) -> usize {
let zeros = if MEM_isLittleEndian() {
word.trailing_zeros()
} else {
word.leading_zeros()
};
(zeros / 8) as usize
}
/// Equivalent to C's `ZSTD_count()`.
unsafe fn count(mut input: *const u8, mut matched: *const u8, input_limit: *const u8) -> usize {
let input_start = input;
let word_size = size_of::<usize>();
while (input_limit as usize).wrapping_sub(input as usize) >= word_size {
let diff =
unsafe { MEM_readST(matched.cast::<c_void>()) ^ MEM_readST(input.cast::<c_void>()) };
if diff != 0 {
return (input as usize).wrapping_sub(input_start as usize) + common_bytes(diff);
}
input = input.wrapping_add(word_size);
matched = matched.wrapping_add(word_size);
}
if MEM_64bits()
&& (input_limit as usize).wrapping_sub(input as usize) >= 4
&& unsafe { MEM_read32(matched.cast::<c_void>()) == MEM_read32(input.cast::<c_void>()) }
{
input = input.wrapping_add(4);
matched = matched.wrapping_add(4);
}
if (input_limit as usize).wrapping_sub(input as usize) >= 2
&& unsafe { MEM_read16(matched.cast::<c_void>()) == MEM_read16(input.cast::<c_void>()) }
{
input = input.wrapping_add(2);
matched = matched.wrapping_add(2);
}
if ptr_lt(input, input_limit) && unsafe { *matched == *input } {
input = input.wrapping_add(1);
}
(input as usize).wrapping_sub(input_start as usize)
}
unsafe fn count_2segments(
input: *const u8,
matched: *const u8,
input_end: *const u8,
match_end: *const u8,
input_start: *const u8,
) -> usize {
let match_remaining = (match_end as usize).wrapping_sub(matched as usize);
let input_remaining = (input_end as usize).wrapping_sub(input as usize);
let first_end = input.wrapping_add(match_remaining.min(input_remaining));
let first_count = unsafe { count(input, matched, first_end) };
if matched.wrapping_add(first_count) != match_end {
return first_count;
}
first_count + unsafe { count(input.wrapping_add(first_count), input_start, input_end) }
}
#[inline]
fn lowest_prefix_index(dict_limit: u32, loaded_dict_end: u32, curr: u32, window_log: u32) -> u32 {
let max_distance = 1u32.wrapping_shl(window_log);
let within_window = if curr.wrapping_sub(dict_limit) > max_distance {
curr.wrapping_sub(max_distance)
} else {
dict_limit
};
if loaded_dict_end != 0 {
dict_limit
} else {
within_window
}
}
#[inline]
fn lowest_match_index(low_limit: u32, loaded_dict_end: u32, curr: u32, window_log: u32) -> u32 {
let max_distance = 1u32.wrapping_shl(window_log);
let within_window = if curr.wrapping_sub(low_limit) > max_distance {
curr.wrapping_sub(max_distance)
} else {
low_limit
};
if loaded_dict_end != 0 {
low_limit
} else {
within_window
}
}
#[inline]
fn index_overlap_check(prefix_lowest_index: u32, rep_index: u32) -> bool {
prefix_lowest_index.wrapping_sub(1).wrapping_sub(rep_index) >= 3
}
/// Stores exactly the observable bytes that C's `ZSTD_storeSeq()` writes.
unsafe fn store_seq(
seq_store: *mut SeqStore_t,
lit_length: usize,
literals: *const u8,
_lit_limit: *const u8,
off_base: u32,
match_length: usize,
) {
let seq_store = unsafe { &mut *seq_store };
let sequence = seq_store.sequences;
debug_assert!(
(sequence as usize).wrapping_sub(seq_store.sequencesStart as usize) / size_of::<SeqDef>()
< seq_store.maxNbSeq
);
debug_assert!(lit_length <= seq_store.maxNbLit);
debug_assert!(match_length >= MINMATCH);
if lit_length != 0 {
unsafe { ptr::copy_nonoverlapping(literals, seq_store.lit, lit_length) };
}
seq_store.lit = seq_store.lit.wrapping_add(lit_length);
let sequence_index =
(sequence as usize).wrapping_sub(seq_store.sequencesStart as usize) / size_of::<SeqDef>();
if lit_length > u16::MAX as usize {
debug_assert_eq!(seq_store.longLengthType, 0);
seq_store.longLengthType = 1;
seq_store.longLengthPos = sequence_index as u32;
}
unsafe { (*sequence).litLength = lit_length as u16 };
unsafe { (*sequence).offBase = off_base };
let match_base = match_length - MINMATCH;
if match_base > u16::MAX as usize {
debug_assert_eq!(seq_store.longLengthType, 0);
seq_store.longLengthType = 2;
seq_store.longLengthPos = sequence_index as u32;
}
unsafe { (*sequence).mlBase = match_base as u16 };
seq_store.sequences = sequence.wrapping_add(1);
}
#[inline]
fn offset_to_offbase(offset: u32) -> u32 {
offset.wrapping_add(OFFSET_OFFBASE)
}
#[inline]
fn offbase_is_offset(offbase: usize) -> bool {
offbase > OFFSET_OFFBASE as usize
}
#[inline]
fn offbase_to_offset(offbase: usize) -> u32 {
(offbase as u32).wrapping_sub(OFFSET_OFFBASE)
}
#[inline]
fn match_improves(
match_length: usize,
best_length: usize,
curr: u32,
match_index: u32,
old_offbase: usize,
old_offbase_plus_one: bool,
) -> bool {
let lhs = (match_length.wrapping_sub(best_length) as i32).wrapping_mul(4);
let new_hb = ZSTD_highbit32(curr.wrapping_sub(match_index).wrapping_add(1)) as i32;
let old_value = (old_offbase as u32).wrapping_add(u32::from(old_offbase_plus_one));
let old_hb = ZSTD_highbit32(old_value) as i32;
lhs > new_hb.wrapping_sub(old_hb)
}
/* ------------------------------------------------------------------------- */
/* Binary tree search */
/* ------------------------------------------------------------------------- */
unsafe fn update_dubt(state: &ZSTD_RustLazyState, ip: *const u8, _iend: *const u8, mls: u32) {
let target = unsafe { index_from(state.base, ip) };
let mut idx = unsafe { next_to_update(state) };
let bt_log = state.chainLog - 1;
let bt_mask = 1u32.wrapping_shl(bt_log).wrapping_sub(1);
while idx < target {
let hash = unsafe { hash_ptr(state.base.wrapping_add(idx as usize), state.hashLog, mls) };
let match_index = unsafe { table_get(state.hashTable, hash) };
let cell = 2 * (idx & bt_mask) as usize;
unsafe {
table_set(state.hashTable, hash, idx);
table_set(state.chainTable, cell, match_index);
table_set(state.chainTable, cell + 1, ZSTD_DUBT_UNSORTED_MARK);
}
idx = idx.wrapping_add(1);
}
unsafe { set_next_to_update(state, target) };
}
unsafe fn insert_dubt1(
state: &ZSTD_RustLazyState,
curr: u32,
input_end: *const u8,
mut nb_compares: u32,
bt_low: u32,
dict_mode: c_int,
) {
let bt_log = state.chainLog - 1;
let bt_mask = 1u32.wrapping_shl(bt_log).wrapping_sub(1);
let mut common_smaller = 0usize;
let mut common_larger = 0usize;
let ip = if curr >= state.dictLimit {
state.base.wrapping_add(curr as usize)
} else {
state.dictBase.wrapping_add(curr as usize)
};
let iend = if curr >= state.dictLimit {
input_end
} else {
state.dictBase.wrapping_add(state.dictLimit as usize)
};
let dict_end = state.dictBase.wrapping_add(state.dictLimit as usize);
let prefix_start = state.base.wrapping_add(state.dictLimit as usize);
let window_valid = state.lowLimit;
let max_distance = 1u32.wrapping_shl(state.windowLog);
let window_low = if curr.wrapping_sub(window_valid) > max_distance {
curr.wrapping_sub(max_distance)
} else {
window_valid
};
let mut smaller_slot = 2 * (curr & bt_mask) as usize;
let mut larger_slot = smaller_slot + 1;
let mut smaller_real = true;
let mut larger_real = true;
let mut match_index = unsafe { table_get(state.chainTable, smaller_slot) };
while nb_compares != 0 && match_index > window_low {
let next_cell = 2 * (match_index & bt_mask) as usize;
let mut match_length = common_smaller.min(common_larger);
let mut matched;
if dict_mode != DICT_EXT || match_index.wrapping_add(match_length as u32) >= state.dictLimit
{
matched = state.base.wrapping_add(match_index as usize);
match_length += unsafe {
count(
ip.wrapping_add(match_length),
matched.wrapping_add(match_length),
iend,
)
};
} else {
matched = state.dictBase.wrapping_add(match_index as usize);
match_length += unsafe {
count_2segments(
ip.wrapping_add(match_length),
matched.wrapping_add(match_length),
iend,
dict_end,
prefix_start,
)
};
if match_index.wrapping_add(match_length as u32) >= state.dictLimit {
matched = state.base.wrapping_add(match_index as usize);
}
}
if ip.wrapping_add(match_length) == iend {
break;
}
if unsafe { *matched.wrapping_add(match_length) < *ip.wrapping_add(match_length) } {
if smaller_real {
unsafe { table_set(state.chainTable, smaller_slot, match_index) };
}
common_smaller = match_length;
if match_index <= bt_low {
smaller_real = false;
break;
}
smaller_slot = next_cell + 1;
match_index = unsafe { table_get(state.chainTable, next_cell + 1) };
} else {
if larger_real {
unsafe { table_set(state.chainTable, larger_slot, match_index) };
}
common_larger = match_length;
if match_index <= bt_low {
larger_real = false;
break;
}
larger_slot = next_cell;
match_index = unsafe { table_get(state.chainTable, next_cell) };
}
nb_compares = nb_compares.wrapping_sub(1);
}
if smaller_real {
unsafe { table_set(state.chainTable, smaller_slot, 0) };
}
if larger_real {
unsafe { table_set(state.chainTable, larger_slot, 0) };
}
}
unsafe fn dubt_find_better_dict_match(
state: &ZSTD_RustLazyState,
ip: *const u8,
iend: *const u8,
offbase: &mut usize,
mut best_length: usize,
mut nb_compares: u32,
mls: u32,
) -> usize {
let dms = unsafe { dict_state(state) };
let dict_hash = unsafe { hash_ptr(ip, dms.hashLog, mls) };
let mut dict_match_index = unsafe { table_get(dms.hashTable, dict_hash) };
let prefix_start = state.base.wrapping_add(state.dictLimit as usize);
let curr = unsafe { index_from(state.base, ip) };
let dict_base = dms.base;
let dict_end = dms.nextSrc;
let dict_high_limit = unsafe { index_from(dms.base, dms.nextSrc) };
let dict_low_limit = dms.lowLimit;
let dict_index_delta = state.lowLimit.wrapping_sub(dict_high_limit);
let bt_log = dms.chainLog - 1;
let bt_mask = 1u32.wrapping_shl(bt_log).wrapping_sub(1);
let bt_low = if bt_mask >= dict_high_limit.wrapping_sub(dict_low_limit) {
dict_low_limit
} else {
dict_high_limit.wrapping_sub(bt_mask)
};
let mut common_smaller = 0usize;
let mut common_larger = 0usize;
while nb_compares != 0 && dict_match_index > dict_low_limit {
let next_cell = 2 * (dict_match_index & bt_mask) as usize;
let mut match_length = common_smaller.min(common_larger);
let mut matched = dict_base.wrapping_add(dict_match_index as usize);
match_length += unsafe {
count_2segments(
ip.wrapping_add(match_length),
matched.wrapping_add(match_length),
iend,
dict_end,
prefix_start,
)
};
if dict_match_index.wrapping_add(match_length as u32) >= dict_high_limit {
matched = state
.base
.wrapping_add(dict_match_index.wrapping_add(dict_index_delta) as usize);
}
if match_length > best_length {
let match_index = dict_match_index.wrapping_add(dict_index_delta);
if match_improves(match_length, best_length, curr, match_index, *offbase, true) {
best_length = match_length;
*offbase = offset_to_offbase(curr.wrapping_sub(match_index)) as usize;
}
if ip.wrapping_add(match_length) == iend {
break;
}
}
if unsafe { *matched.wrapping_add(match_length) < *ip.wrapping_add(match_length) } {
if dict_match_index <= bt_low {
break;
}
common_smaller = match_length;
dict_match_index = unsafe { table_get(dms.chainTable, next_cell + 1) };
} else {
if dict_match_index <= bt_low {
break;
}
common_larger = match_length;
dict_match_index = unsafe { table_get(dms.chainTable, next_cell) };
}
nb_compares = nb_compares.wrapping_sub(1);
}
best_length
}
unsafe fn dubt_find_best_match(
state: &ZSTD_RustLazyState,
ip: *const u8,
iend: *const u8,
offbase: &mut usize,
mls: u32,
dict_mode: c_int,
) -> usize {
let hash = unsafe { hash_ptr(ip, state.hashLog, mls) };
let mut match_index = unsafe { table_get(state.hashTable, hash) };
let curr = unsafe { index_from(state.base, ip) };
let window_low = lowest_match_index(state.lowLimit, state.loadedDictEnd, curr, state.windowLog);
let bt_log = state.chainLog - 1;
let bt_mask = 1u32.wrapping_shl(bt_log).wrapping_sub(1);
let bt_low = if bt_mask >= curr {
0
} else {
curr.wrapping_sub(bt_mask)
};
let unsort_limit = bt_low.max(window_low);
let mut next_candidate = 2 * (match_index & bt_mask) as usize;
let mut unsorted_mark = next_candidate + 1;
let mut nb_compares = 1u32.wrapping_shl(state.searchLog);
let mut nb_candidates = nb_compares;
let mut previous_candidate = 0u32;
while match_index > unsort_limit
&& unsafe { table_get(state.chainTable, unsorted_mark) == ZSTD_DUBT_UNSORTED_MARK }
&& nb_candidates > 1
{
unsafe { table_set(state.chainTable, unsorted_mark, previous_candidate) };
previous_candidate = match_index;
match_index = unsafe { table_get(state.chainTable, next_candidate) };
next_candidate = 2 * (match_index & bt_mask) as usize;
unsorted_mark = next_candidate + 1;
nb_candidates = nb_candidates.wrapping_sub(1);
}
if match_index > unsort_limit
&& unsafe { table_get(state.chainTable, unsorted_mark) == ZSTD_DUBT_UNSORTED_MARK }
{
unsafe {
table_set(state.chainTable, next_candidate, 0);
table_set(state.chainTable, unsorted_mark, 0);
}
}
match_index = previous_candidate;
while match_index != 0 {
let next_candidate_idx =
unsafe { table_get(state.chainTable, 2 * (match_index & bt_mask) as usize + 1) };
unsafe {
insert_dubt1(
state,
match_index,
iend,
nb_candidates,
unsort_limit,
dict_mode,
)
};
match_index = next_candidate_idx;
nb_candidates = nb_candidates.wrapping_add(1);
}
let mut common_smaller = 0usize;
let mut common_larger = 0usize;
let dict_base = state.dictBase;
let dict_limit = state.dictLimit;
let dict_end = dict_base.wrapping_add(dict_limit as usize);
let prefix_start = state.base.wrapping_add(dict_limit as usize);
let mut smaller_slot = 2 * (curr & bt_mask) as usize;
let mut larger_slot = smaller_slot + 1;
let mut smaller_real = true;
let mut larger_real = true;
let mut match_end_idx = curr.wrapping_add(HASH_READ_SIZE as u32 + 1);
let mut best_length = 0usize;
match_index = unsafe { table_get(state.hashTable, hash) };
unsafe { table_set(state.hashTable, hash, curr) };
while nb_compares != 0 && match_index > window_low {
let next_cell = 2 * (match_index & bt_mask) as usize;
let mut match_length = common_smaller.min(common_larger);
let mut matched;
if dict_mode != DICT_EXT || match_index.wrapping_add(match_length as u32) >= dict_limit {
matched = state.base.wrapping_add(match_index as usize);
match_length += unsafe {
count(
ip.wrapping_add(match_length),
matched.wrapping_add(match_length),
iend,
)
};
} else {
matched = dict_base.wrapping_add(match_index as usize);
match_length += unsafe {
count_2segments(
ip.wrapping_add(match_length),
matched.wrapping_add(match_length),
iend,
dict_end,
prefix_start,
)
};
if match_index.wrapping_add(match_length as u32) >= dict_limit {
matched = state.base.wrapping_add(match_index as usize);
}
}
if match_length > best_length {
if match_length > match_end_idx.wrapping_sub(match_index) as usize {
match_end_idx = match_index.wrapping_add(match_length as u32);
}
if match_improves(
match_length,
best_length,
curr,
match_index,
*offbase,
false,
) {
best_length = match_length;
*offbase = offset_to_offbase(curr.wrapping_sub(match_index)) as usize;
}
if ip.wrapping_add(match_length) == iend {
if dict_mode == DICT_MATCH_STATE {
nb_compares = 0;
}
break;
}
}
if unsafe { *matched.wrapping_add(match_length) < *ip.wrapping_add(match_length) } {
if smaller_real {
unsafe { table_set(state.chainTable, smaller_slot, match_index) };
}
common_smaller = match_length;
if match_index <= bt_low {
smaller_real = false;
break;
}
smaller_slot = next_cell + 1;
match_index = unsafe { table_get(state.chainTable, next_cell + 1) };
} else {
if larger_real {
unsafe { table_set(state.chainTable, larger_slot, match_index) };
}
common_larger = match_length;
if match_index <= bt_low {
larger_real = false;
break;
}
larger_slot = next_cell;
match_index = unsafe { table_get(state.chainTable, next_cell) };
}
nb_compares = nb_compares.wrapping_sub(1);
}
if smaller_real {
unsafe { table_set(state.chainTable, smaller_slot, 0) };
}
if larger_real {
unsafe { table_set(state.chainTable, larger_slot, 0) };
}
if dict_mode == DICT_MATCH_STATE && nb_compares != 0 {
best_length = unsafe {
dubt_find_better_dict_match(state, ip, iend, offbase, best_length, nb_compares, mls)
};
}
unsafe { set_next_to_update(state, match_end_idx.wrapping_sub(HASH_READ_SIZE as u32)) };
best_length
}
unsafe fn bt_find_best_match(
state: &ZSTD_RustLazyState,
ip: *const u8,
iend: *const u8,
offbase: &mut usize,
mls: u32,
dict_mode: c_int,
) -> usize {
if ptr_lt(
ip,
state
.base
.wrapping_add(unsafe { next_to_update(state) } as usize),
) {
return 0;
}
unsafe { update_dubt(state, ip, iend, mls) };
unsafe { dubt_find_best_match(state, ip, iend, offbase, mls, dict_mode) }
}
/* ------------------------------------------------------------------------- */
/* Dedicated dictionary search */
/* ------------------------------------------------------------------------- */
unsafe fn dedicated_dict_search_load_dictionary(state: &ZSTD_RustLazyState, ip: *const u8) {
let target = unsafe { index_from(state.base, ip) };
let chain_size = 1u32.wrapping_shl(state.chainLog);
let mut idx = unsafe { next_to_update(state) };
let min_chain = if chain_size < target.wrapping_sub(idx) {
target.wrapping_sub(chain_size)
} else {
idx
};
let bucket_size = 1u32 << ZSTD_LAZY_DDSS_BUCKET_LOG;
let cache_size = bucket_size - 1;
let chain_attempts = (1u32 << state.searchLog).wrapping_sub(cache_size);
let chain_limit = chain_attempts.min(255);
let hash_log = state.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
let tmp_hash_table = state.hashTable;
let tmp_chain_table = state
.hashTable
.wrapping_add((1usize).wrapping_shl(hash_log));
let tmp_chain_size = ((1u32 << ZSTD_LAZY_DDSS_BUCKET_LOG) - 1).wrapping_shl(hash_log);
let tmp_min_chain = if tmp_chain_size < target {
target.wrapping_sub(tmp_chain_size)
} else {
idx
};
while idx < target {
let hash = unsafe {
hash_ptr(
state.base.wrapping_add(idx as usize),
hash_log,
state.minMatch,
)
};
if idx >= tmp_min_chain {
let previous = unsafe { table_get(state.hashTable, hash) };
unsafe {
table_set(
tmp_chain_table,
idx.wrapping_sub(tmp_min_chain) as usize,
previous,
)
};
}
unsafe { table_set(tmp_hash_table, hash, idx) };
idx = idx.wrapping_add(1);
}
let mut chain_pos = 0u32;
let mut hash_idx = 0u32;
while hash_idx < (1u32 << hash_log) {
let mut count = 0u32;
let mut count_beyond_min_chain = 0u32;
let mut chain_index = unsafe { table_get(tmp_hash_table, hash_idx as usize) };
while chain_index >= tmp_min_chain && count < cache_size {
if chain_index < min_chain {
count_beyond_min_chain = count_beyond_min_chain.wrapping_add(1);
}
chain_index = unsafe {
table_get(
tmp_chain_table,
chain_index.wrapping_sub(tmp_min_chain) as usize,
)
};
count = count.wrapping_add(1);
}
if count == cache_size {
count = 0;
while count < chain_limit {
if chain_index < min_chain {
if chain_index == 0 || count_beyond_min_chain.wrapping_add(1) > cache_size {
break;
}
count_beyond_min_chain = count_beyond_min_chain.wrapping_add(1);
}
unsafe { table_set(state.chainTable, chain_pos as usize, chain_index) };
chain_pos = chain_pos.wrapping_add(1);
count = count.wrapping_add(1);
if chain_index < tmp_min_chain {
break;
}
chain_index = unsafe {
table_get(
tmp_chain_table,
chain_index.wrapping_sub(tmp_min_chain) as usize,
)
};
}
} else {
count = 0;
}
if count != 0 {
unsafe {
table_set(
tmp_hash_table,
hash_idx as usize,
((chain_pos.wrapping_sub(count)) << 8).wrapping_add(count),
)
};
} else {
unsafe { table_set(tmp_hash_table, hash_idx as usize, 0) };
}
hash_idx = hash_idx.wrapping_add(1);
}
hash_idx = 1u32 << hash_log;
while hash_idx != 0 {
hash_idx = hash_idx.wrapping_sub(1);
let bucket_idx = hash_idx << ZSTD_LAZY_DDSS_BUCKET_LOG;
let packed = unsafe { table_get(tmp_hash_table, hash_idx as usize) };
let mut i = 0u32;
while i < cache_size {
unsafe { table_set(state.hashTable, bucket_idx.wrapping_add(i) as usize, 0) };
i = i.wrapping_add(1);
}
unsafe {
table_set(
state.hashTable,
bucket_idx.wrapping_add(bucket_size).wrapping_sub(1) as usize,
packed,
)
};
}
idx = unsafe { next_to_update(state) };
while idx < target {
let hash = unsafe {
hash_ptr(
state.base.wrapping_add(idx as usize),
hash_log,
state.minMatch,
)
} << ZSTD_LAZY_DDSS_BUCKET_LOG;
let mut i = cache_size.wrapping_sub(1);
while i != 0 {
let previous =
unsafe { table_get(state.hashTable, hash.wrapping_add((i - 1) as usize)) };
unsafe { table_set(state.hashTable, hash.wrapping_add(i as usize), previous) };
i = i.wrapping_sub(1);
}
unsafe { table_set(state.hashTable, hash, idx) };
idx = idx.wrapping_add(1);
}
unsafe { set_next_to_update(state, target) };
}
unsafe fn dedicated_dict_search(
offset: &mut usize,
mut ml: usize,
nb_attempts: u32,
dms: &ZSTD_RustLazyState,
ip: *const u8,
ilimit: *const u8,
prefix_start: *const u8,
curr: u32,
dict_limit: u32,
dds_idx: usize,
) -> usize {
let dds_lowest_index = dms.dictLimit;
let dds_base = dms.base;
let dds_end = dms.nextSrc;
let dds_size = unsafe { index_from(dds_base, dds_end) };
let dds_index_delta = dict_limit.wrapping_sub(dds_size);
let bucket_size = 1u32 << ZSTD_LAZY_DDSS_BUCKET_LOG;
let bucket_limit = nb_attempts.min(bucket_size - 1);
let mut dds_attempt = 0u32;
while dds_attempt < bucket_limit {
let match_index = unsafe { table_get(dms.hashTable, dds_idx + dds_attempt as usize) };
if match_index == 0 {
return ml;
}
debug_assert!(match_index >= dds_lowest_index);
let matched = dds_base.wrapping_add(match_index as usize);
let mut current_ml = 0usize;
if unsafe { read32(matched) == read32(ip) } {
current_ml = unsafe {
count_2segments(
ip.wrapping_add(4),
matched.wrapping_add(4),
ilimit,
dds_end,
prefix_start,
) + 4
};
}
if current_ml > ml {
ml = current_ml;
*offset =
offset_to_offbase(curr.wrapping_sub(match_index.wrapping_add(dds_index_delta)))
as usize;
if ip.wrapping_add(current_ml) == ilimit {
return ml;
}
}
dds_attempt = dds_attempt.wrapping_add(1);
}
let packed = unsafe { table_get(dms.hashTable, dds_idx + (bucket_size - 1) as usize) };
let mut chain_index = packed >> 8;
let chain_length = packed & 0xff;
let chain_attempts = nb_attempts.wrapping_sub(dds_attempt);
let chain_limit = chain_attempts.min(chain_length);
let mut chain_attempt = 0u32;
while chain_attempt < chain_limit {
let match_index = unsafe { table_get(dms.chainTable, chain_index as usize) };
let matched = dds_base.wrapping_add(match_index as usize);
let mut current_ml = 0usize;
if unsafe { read32(matched) == read32(ip) } {
current_ml = unsafe {
count_2segments(
ip.wrapping_add(4),
matched.wrapping_add(4),
ilimit,
dds_end,
prefix_start,
) + 4
};
}
if current_ml > ml {
ml = current_ml;
*offset =
offset_to_offbase(curr.wrapping_sub(match_index.wrapping_add(dds_index_delta)))
as usize;
if ip.wrapping_add(current_ml) == ilimit {
break;
}
}
chain_index = chain_index.wrapping_add(1);
chain_attempt = chain_attempt.wrapping_add(1);
}
ml
}
/* ------------------------------------------------------------------------- */
/* Hash-chain search */
/* ------------------------------------------------------------------------- */
unsafe fn insert_and_find_first_index_internal(
state: &ZSTD_RustLazyState,
ip: *const u8,
mls: u32,
lazy_skip: bool,
) -> u32 {
let chain_mask = 1u32.wrapping_shl(state.chainLog).wrapping_sub(1);
let target = unsafe { index_from(state.base, ip) };
let mut idx = unsafe { next_to_update(state) };
while idx < target {
let hash = unsafe { hash_ptr(state.base.wrapping_add(idx as usize), state.hashLog, mls) };
let previous = unsafe { table_get(state.hashTable, hash) };
unsafe {
table_set(state.chainTable, (idx & chain_mask) as usize, previous);
table_set(state.hashTable, hash, idx);
}
idx = idx.wrapping_add(1);
if lazy_skip {
break;
}
}
unsafe { set_next_to_update(state, target) };
let hash = unsafe { hash_ptr(ip, state.hashLog, mls) };
unsafe { table_get(state.hashTable, hash) }
}
unsafe fn hc_find_best_match(
state: &ZSTD_RustLazyState,
ip: *const u8,
ilimit: *const u8,
offset: &mut usize,
mls: u32,
dict_mode: c_int,
) -> usize {
let chain_size = 1u32.wrapping_shl(state.chainLog);
let chain_mask = chain_size.wrapping_sub(1);
let base = state.base;
let dict_base = state.dictBase;
let dict_limit = state.dictLimit;
let prefix_start = base.wrapping_add(dict_limit as usize);
let dict_end = dict_base.wrapping_add(dict_limit as usize);
let curr = unsafe { index_from(base, ip) };
let max_distance = 1u32.wrapping_shl(state.windowLog);
let lowest_valid = state.lowLimit;
let within_max_distance = if curr.wrapping_sub(lowest_valid) > max_distance {
curr.wrapping_sub(max_distance)
} else {
lowest_valid
};
let low_limit = if state.loadedDictEnd != 0 {
lowest_valid
} else {
within_max_distance
};
let min_chain = if curr > chain_size {
curr.wrapping_sub(chain_size)
} else {
0
};
let mut nb_attempts = 1u32.wrapping_shl(state.searchLog);
let mut ml = 3usize;
let dds_hash_log = if dict_mode == DICT_DEDICATED {
unsafe { dict_state(state) }.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG
} else {
0
};
let dds_idx = if dict_mode == DICT_DEDICATED {
(unsafe { hash_ptr(ip, dds_hash_log, mls) }) << ZSTD_LAZY_DDSS_BUCKET_LOG
} else {
0
};
let mut match_index =
unsafe { insert_and_find_first_index_internal(state, ip, mls, lazy_skipping(state)) };
while match_index >= low_limit && nb_attempts != 0 {
let mut current_ml = 0usize;
if dict_mode != DICT_EXT || match_index >= dict_limit {
let matched = base.wrapping_add(match_index as usize);
if unsafe { read32(matched.wrapping_add(ml - 3)) == read32(ip.wrapping_add(ml - 3)) } {
current_ml = unsafe { count(ip, matched, ilimit) };
}
} else {
let matched = dict_base.wrapping_add(match_index as usize);
if unsafe { read32(matched) == read32(ip) } {
current_ml = unsafe {
count_2segments(
ip.wrapping_add(4),
matched.wrapping_add(4),
ilimit,
dict_end,
prefix_start,
) + 4
};
}
}
if current_ml > ml {
ml = current_ml;
*offset = offset_to_offbase(curr.wrapping_sub(match_index)) as usize;
if ip.wrapping_add(current_ml) == ilimit {
break;
}
}
if match_index <= min_chain {
break;
}
match_index = unsafe { table_get(state.chainTable, (match_index & chain_mask) as usize) };
nb_attempts = nb_attempts.wrapping_sub(1);
}
if dict_mode == DICT_DEDICATED {
ml = unsafe {
dedicated_dict_search(
offset,
ml,
nb_attempts,
dict_state(state),
ip,
ilimit,
prefix_start,
curr,
dict_limit,
dds_idx,
)
};
} else if dict_mode == DICT_MATCH_STATE {
let dms = unsafe { dict_state(state) };
let dms_chain_size = 1u32.wrapping_shl(dms.chainLog);
let dms_chain_mask = dms_chain_size.wrapping_sub(1);
let dms_lowest_index = dms.dictLimit;
let dms_base = dms.base;
let dms_end = dms.nextSrc;
let dms_size = unsafe { index_from(dms_base, dms_end) };
let dms_index_delta = dict_limit.wrapping_sub(dms_size);
let dms_min_chain = if dms_size > dms_chain_size {
dms_size.wrapping_sub(dms_chain_size)
} else {
0
};
match_index = unsafe { table_get(dms.hashTable, hash_ptr(ip, dms.hashLog, mls)) };
while match_index >= dms_lowest_index && nb_attempts != 0 {
let matched = dms_base.wrapping_add(match_index as usize);
let mut current_ml = 0usize;
if unsafe { read32(matched) == read32(ip) } {
current_ml = unsafe {
count_2segments(
ip.wrapping_add(4),
matched.wrapping_add(4),
ilimit,
dms_end,
prefix_start,
) + 4
};
}
if current_ml > ml {
ml = current_ml;
*offset =
offset_to_offbase(curr.wrapping_sub(match_index.wrapping_add(dms_index_delta)))
as usize;
if ip.wrapping_add(current_ml) == ilimit {
break;
}
}
if match_index <= dms_min_chain {
break;
}
match_index =
unsafe { table_get(dms.chainTable, (match_index & dms_chain_mask) as usize) };
nb_attempts = nb_attempts.wrapping_sub(1);
}
}
ml
}
/* ------------------------------------------------------------------------- */
/* Row hash search */
/* ------------------------------------------------------------------------- */
#[inline]
unsafe fn row_next_index(tag_row: *mut u8, row_mask: u32) -> u32 {
let mut next = unsafe { ((*tag_row).wrapping_sub(1) as u32) & row_mask };
if next == 0 {
next = next.wrapping_add(row_mask);
}
unsafe { *tag_row = next as u8 };
next
}
unsafe fn row_fill_hash_cache(
state: &ZSTD_RustLazyState,
base: *const u8,
_row_log: u32,
mls: u32,
mut idx: u32,
ilimit: *const u8,
) {
let hash_log = state.rowHashLog;
let at = base.wrapping_add(idx as usize);
let max_to_fill = if ptr_gt(at, ilimit) {
0
} else {
((ilimit as usize).wrapping_sub(at as usize)).wrapping_add(1) as u32
};
let limit = idx.wrapping_add((ZSTD_ROW_HASH_CACHE_SIZE as u32).min(max_to_fill));
while idx < limit {
let hash = unsafe {
hash_ptr_salted(
base.wrapping_add(idx as usize),
hash_log + ZSTD_ROW_HASH_TAG_BITS,
mls,
state.hashSalt,
) as u32
};
unsafe {
table_set(
state.hashCache,
(idx as usize) & (ZSTD_ROW_HASH_CACHE_SIZE - 1),
hash,
)
};
idx = idx.wrapping_add(1);
}
}
unsafe fn row_next_cached_hash(
cache: *mut u32,
_hash_table: *const u32,
_tag_table: *const u8,
base: *const u8,
idx: u32,
hash_log: u32,
_row_log: u32,
mls: u32,
hash_salt: u64,
) -> u32 {
let new_hash = unsafe {
hash_ptr_salted(
base.wrapping_add(idx as usize + ZSTD_ROW_HASH_CACHE_SIZE),
hash_log + ZSTD_ROW_HASH_TAG_BITS,
mls,
hash_salt,
) as u32
};
let slot = (idx as usize) & (ZSTD_ROW_HASH_CACHE_SIZE - 1);
let hash = unsafe { table_get(cache, slot) };
unsafe { table_set(cache, slot, new_hash) };
hash
}
unsafe fn row_update_internal_impl(
state: &ZSTD_RustLazyState,
mut update_start: u32,
update_end: u32,
mls: u32,
row_log: u32,
row_mask: u32,
use_cache: bool,
) {
while update_start < update_end {
let hash = if use_cache {
unsafe {
row_next_cached_hash(
state.hashCache,
state.hashTable,
state.tagTable,
state.base,
update_start,
state.rowHashLog,
row_log,
mls,
state.hashSalt,
)
}
} else {
unsafe {
hash_ptr_salted(
state.base.wrapping_add(update_start as usize),
state.rowHashLog + ZSTD_ROW_HASH_TAG_BITS,
mls,
state.hashSalt,
) as u32
}
};
let rel_row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << row_log;
let row = state.hashTable.wrapping_add(rel_row as usize);
let tag_row = state.tagTable.wrapping_add(rel_row as usize);
let pos = unsafe { row_next_index(tag_row, row_mask) };
unsafe {
byte_set(tag_row, pos as usize, (hash & ZSTD_ROW_HASH_TAG_MASK) as u8);
table_set(row, pos as usize, update_start);
}
update_start = update_start.wrapping_add(1);
}
}
unsafe fn row_update_internal(
state: &ZSTD_RustLazyState,
ip: *const u8,
mls: u32,
row_log: u32,
row_mask: u32,
use_cache: bool,
) {
let mut idx = unsafe { next_to_update(state) };
let target = unsafe { index_from(state.base, ip) };
const SKIP_THRESHOLD: u32 = 384;
const MAX_START: u32 = 96;
const MAX_END: u32 = 32;
if use_cache && target.wrapping_sub(idx) > SKIP_THRESHOLD {
let bound = idx.wrapping_add(MAX_START);
unsafe { row_update_internal_impl(state, idx, bound, mls, row_log, row_mask, use_cache) };
idx = target.wrapping_sub(MAX_END);
unsafe { row_fill_hash_cache(state, state.base, row_log, mls, idx, ip.wrapping_add(1)) };
}
unsafe { row_update_internal_impl(state, idx, target, mls, row_log, row_mask, use_cache) };
unsafe { set_next_to_update(state, target) };
}
/// Returns matching tag positions in the same circular-buffer order as the
/// C SSE/SWAR mask iterator. The scalar version avoids architecture-specific
/// vector intrinsics while preserving the selected candidate order.
unsafe fn row_matching_positions(
tag_row: *const u8,
tag: u8,
head: u32,
row_mask: u32,
row_entries: u32,
out: &mut [u32; ZSTD_ROW_HASH_MAX_ENTRIES],
) -> usize {
let mut len = 0usize;
let mut relative = 0u32;
while relative < row_entries {
let pos = head.wrapping_add(relative) & row_mask;
if unsafe { byte_get(tag_row, pos as usize) } == tag {
out[len] = pos;
len += 1;
}
relative = relative.wrapping_add(1);
}
len
}
unsafe fn row_find_best_match(
state: &ZSTD_RustLazyState,
ip: *const u8,
ilimit: *const u8,
offset: &mut usize,
mls: u32,
dict_mode: c_int,
row_log: u32,
) -> usize {
let base = state.base;
let dict_base = state.dictBase;
let dict_limit = state.dictLimit;
let prefix_start = base.wrapping_add(dict_limit as usize);
let dict_end = dict_base.wrapping_add(dict_limit as usize);
let curr = unsafe { index_from(base, ip) };
let max_distance = 1u32.wrapping_shl(state.windowLog);
let lowest_valid = state.lowLimit;
let within_max_distance = if curr.wrapping_sub(lowest_valid) > max_distance {
curr.wrapping_sub(max_distance)
} else {
lowest_valid
};
let low_limit = if state.loadedDictEnd != 0 {
lowest_valid
} else {
within_max_distance
};
let row_entries = 1u32 << row_log;
let row_mask = row_entries - 1;
let capped_search_log = state.searchLog.min(row_log);
let mut nb_attempts = 1u32 << capped_search_log;
let mut ml = 3usize;
let (mut dds_idx, mut dds_extra_attempts) = (0usize, 0u32);
if dict_mode == DICT_DEDICATED {
let dms = unsafe { dict_state(state) };
let dds_hash_log = dms.hashLog - ZSTD_LAZY_DDSS_BUCKET_LOG;
dds_idx = unsafe { hash_ptr(ip, dds_hash_log, mls) } << ZSTD_LAZY_DDSS_BUCKET_LOG;
dds_extra_attempts = if state.searchLog > row_log {
1u32 << (state.searchLog - row_log)
} else {
0
};
}
let mut dms_tag = 0u32;
let mut dms_row = ptr::null_mut::<u32>();
let mut dms_tag_row = ptr::null_mut::<u8>();
if dict_mode == DICT_MATCH_STATE {
let dms = unsafe { dict_state(state) };
let dms_hash = unsafe { hash_ptr(ip, dms.rowHashLog + ZSTD_ROW_HASH_TAG_BITS, mls) as u32 };
let dms_rel_row = (dms_hash >> ZSTD_ROW_HASH_TAG_BITS) << row_log;
dms_tag = dms_hash & ZSTD_ROW_HASH_TAG_MASK;
dms_tag_row = dms.tagTable.wrapping_add(dms_rel_row as usize);
dms_row = dms.hashTable.wrapping_add(dms_rel_row as usize);
}
let hash = if !unsafe { lazy_skipping(state) } {
unsafe { row_update_internal(state, ip, mls, row_log, row_mask, true) };
unsafe {
row_next_cached_hash(
state.hashCache,
state.hashTable,
state.tagTable,
base,
curr,
state.rowHashLog,
row_log,
mls,
state.hashSalt,
)
}
} else {
let h = unsafe {
hash_ptr_salted(
ip,
state.rowHashLog + ZSTD_ROW_HASH_TAG_BITS,
mls,
state.hashSalt,
) as u32
};
unsafe { set_next_to_update(state, curr) };
h
};
unsafe { add_hash_salt_entropy(state, hash) };
let rel_row = (hash >> ZSTD_ROW_HASH_TAG_BITS) << row_log;
let tag = (hash & ZSTD_ROW_HASH_TAG_MASK) as u8;
let row = state.hashTable.wrapping_add(rel_row as usize);
let tag_row = state.tagTable.wrapping_add(rel_row as usize);
let head = unsafe { byte_get(tag_row, 0) as u32 } & row_mask;
let mut positions = [0u32; ZSTD_ROW_HASH_MAX_ENTRIES];
let position_len = unsafe {
row_matching_positions(tag_row, tag, head, row_mask, row_entries, &mut positions)
};
let mut match_buffer = [0u32; ZSTD_ROW_HASH_MAX_ENTRIES];
let mut num_matches = 0usize;
for pos in positions[..position_len].iter().copied() {
if nb_attempts == 0 {
break;
}
if pos == 0 {
continue;
}
let match_index = unsafe { table_get(row, pos as usize) };
if match_index < low_limit {
break;
}
match_buffer[num_matches] = match_index;
num_matches += 1;
nb_attempts = nb_attempts.wrapping_sub(1);
}
let insert_pos = unsafe { row_next_index(tag_row, row_mask) };
unsafe {
byte_set(tag_row, insert_pos as usize, tag);
let index = next_to_update(state);
table_set(row, insert_pos as usize, index);
set_next_to_update(state, index.wrapping_add(1));
}
for match_index in match_buffer[..num_matches].iter().copied() {
let mut current_ml = 0usize;
if dict_mode != DICT_EXT || match_index >= dict_limit {
let matched = base.wrapping_add(match_index as usize);
if unsafe { read32(matched.wrapping_add(ml - 3)) == read32(ip.wrapping_add(ml - 3)) } {
current_ml = unsafe { count(ip, matched, ilimit) };
}
} else {
let matched = dict_base.wrapping_add(match_index as usize);
if unsafe { read32(matched) == read32(ip) } {
current_ml = unsafe {
count_2segments(
ip.wrapping_add(4),
matched.wrapping_add(4),
ilimit,
dict_end,
prefix_start,
) + 4
};
}
}
if current_ml > ml {
ml = current_ml;
*offset = offset_to_offbase(curr.wrapping_sub(match_index)) as usize;
if ip.wrapping_add(current_ml) == ilimit {
break;
}
}
}
if dict_mode == DICT_DEDICATED {
ml = unsafe {
dedicated_dict_search(
offset,
ml,
nb_attempts.wrapping_add(dds_extra_attempts),
dict_state(state),
ip,
ilimit,
prefix_start,
curr,
dict_limit,
dds_idx,
)
};
} else if dict_mode == DICT_MATCH_STATE {
let dms = unsafe { dict_state(state) };
let dms_lowest_index = dms.dictLimit;
let dms_base = dms.base;
let dms_end = dms.nextSrc;
let dms_size = unsafe { index_from(dms_base, dms_end) };
let dms_index_delta = dict_limit.wrapping_sub(dms_size);
let dms_head = unsafe { byte_get(dms_tag_row, 0) as u32 } & row_mask;
let mut dms_positions = [0u32; ZSTD_ROW_HASH_MAX_ENTRIES];
let dms_position_len = unsafe {
row_matching_positions(
dms_tag_row,
dms_tag as u8,
dms_head,
row_mask,
row_entries,
&mut dms_positions,
)
};
let mut dms_matches = [0u32; ZSTD_ROW_HASH_MAX_ENTRIES];
let mut dms_count = 0usize;
for pos in dms_positions[..dms_position_len].iter().copied() {
if nb_attempts == 0 {
break;
}
if pos == 0 {
continue;
}
let match_index = unsafe { table_get(dms_row, pos as usize) };
if match_index < dms_lowest_index {
break;
}
dms_matches[dms_count] = match_index;
dms_count += 1;
nb_attempts = nb_attempts.wrapping_sub(1);
}
for match_index in dms_matches[..dms_count].iter().copied() {
let matched = dms_base.wrapping_add(match_index as usize);
let mut current_ml = 0usize;
if unsafe { read32(matched) == read32(ip) } {
current_ml = unsafe {
count_2segments(
ip.wrapping_add(4),
matched.wrapping_add(4),
ilimit,
dms_end,
prefix_start,
) + 4
};
}
if current_ml > ml {
ml = current_ml;
*offset =
offset_to_offbase(curr.wrapping_sub(match_index.wrapping_add(dms_index_delta)))
as usize;
if ip.wrapping_add(current_ml) == ilimit {
break;
}
}
}
}
ml
}
unsafe fn search_max(
state: &ZSTD_RustLazyState,
ip: *const u8,
iend: *const u8,
offset: &mut usize,
mls: u32,
row_log: u32,
search_method: c_int,
dict_mode: c_int,
) -> usize {
match search_method {
SEARCH_HASH_CHAIN => unsafe { hc_find_best_match(state, ip, iend, offset, mls, dict_mode) },
SEARCH_BINARY_TREE => unsafe {
bt_find_best_match(state, ip, iend, offset, mls, dict_mode)
},
SEARCH_ROW_HASH => unsafe {
row_find_best_match(state, ip, iend, offset, mls, dict_mode, row_log)
},
_ => 0,
}
}
#[inline]
fn score(length: usize, multiplier: usize, offbase: usize, adjustment: usize) -> i32 {
let raw = length
.wrapping_mul(multiplier)
.wrapping_sub(ZSTD_highbit32(offbase as u32) as usize)
.wrapping_add(adjustment);
raw as i32
}
/* ------------------------------------------------------------------------- */
/* Lazy parser for no dictionary, dictionary match state, and DDS */
/* ------------------------------------------------------------------------- */
unsafe fn compress_block_lazy_generic(
state: &ZSTD_RustLazyState,
seq_store: *mut SeqStore_t,
reps: *mut u32,
src: *const u8,
src_size: usize,
search_method: c_int,
depth: u32,
dict_mode: c_int,
) -> usize {
let istart = src;
let mut ip = istart;
let mut anchor = istart;
let iend = istart.wrapping_add(src_size);
let required = if search_method == SEARCH_ROW_HASH {
HASH_READ_SIZE + ZSTD_ROW_HASH_CACHE_SIZE
} else {
HASH_READ_SIZE
};
if src_size < required {
return src_size;
}
let ilimit = iend.wrapping_sub(required);
let base = state.base;
let prefix_lowest_index = state.dictLimit;
let prefix_lowest = base.wrapping_add(prefix_lowest_index as usize);
let mls = bounded(4, state.minMatch, 6);
let row_log = bounded(4, state.searchLog, 6);
let mut offset_1 = unsafe { *reps };
let mut offset_2 = unsafe { *reps.add(1) };
let mut offset_saved_1 = 0u32;
let mut offset_saved_2 = 0u32;
let is_dms = dict_mode == DICT_MATCH_STATE;
let is_dds = dict_mode == DICT_DEDICATED;
let is_dxs = is_dms || is_dds;
let dms = if is_dxs {
Some(unsafe { dict_state(state) })
} else {
None
};
let dict_lowest_index = dms.map_or(0, |value| value.dictLimit);
let dict_base = dms.map_or(ptr::null(), |value| value.base);
let dict_lowest = if is_dxs {
dict_base.wrapping_add(dict_lowest_index as usize)
} else {
ptr::null()
};
let dict_end = dms.map_or(ptr::null(), |value| value.nextSrc);
let dict_index_delta = if is_dxs {
prefix_lowest_index.wrapping_sub(unsafe { index_from(dict_base, dict_end) })
} else {
0
};
let dict_and_prefix_length = if is_dxs {
(ip as usize)
.wrapping_sub(prefix_lowest as usize)
.wrapping_add((dict_end as usize).wrapping_sub(dict_lowest as usize)) as u32
} else {
(ip as usize).wrapping_sub(prefix_lowest as usize) as u32
};
if dict_and_prefix_length == 0 {
ip = ip.wrapping_add(1);
}
if dict_mode == DICT_NO_DICT {
let curr = unsafe { index_from(base, ip) };
let window_low =
lowest_prefix_index(state.dictLimit, state.loadedDictEnd, curr, state.windowLog);
let max_rep = curr.wrapping_sub(window_low);
if offset_2 > max_rep {
offset_saved_2 = offset_2;
offset_2 = 0;
}
if offset_1 > max_rep {
offset_saved_1 = offset_1;
offset_1 = 0;
}
}
unsafe { set_lazy_skipping(state, false) };
if search_method == SEARCH_ROW_HASH {
unsafe { row_fill_hash_cache(state, base, row_log, mls, next_to_update(state), ilimit) };
}
while ptr_lt(ip, ilimit) {
let mut match_length = 0usize;
let mut offbase = REPCODE1_TO_OFFBASE as usize;
let mut start = ip.wrapping_add(1);
let mut store_directly = false;
if is_dxs {
let rep_index = unsafe { index_from(base, ip) }
.wrapping_add(1)
.wrapping_sub(offset_1);
let rep_match = if rep_index < prefix_lowest_index {
dict_base.wrapping_add(rep_index.wrapping_sub(dict_index_delta) as usize)
} else {
base.wrapping_add(rep_index as usize)
};
if index_overlap_check(prefix_lowest_index, rep_index)
&& unsafe { read32(rep_match) == read32(ip.wrapping_add(1)) }
{
let rep_end = if rep_index < prefix_lowest_index {
dict_end
} else {
iend
};
match_length = unsafe {
count_2segments(
ip.wrapping_add(5),
rep_match.wrapping_add(4),
iend,
rep_end,
prefix_lowest,
) + 4
};
if depth == 0 {
store_directly = true;
}
}
}
if dict_mode == DICT_NO_DICT
&& offset_1 > 0
&& unsafe {
read32(ip.wrapping_add(1).wrapping_sub(offset_1 as usize))
== read32(ip.wrapping_add(1))
}
{
match_length = unsafe {
count(
ip.wrapping_add(5),
ip.wrapping_add(5).wrapping_sub(offset_1 as usize),
iend,
) + 4
};
if depth == 0 {
store_directly = true;
}
}
if !store_directly {
let mut candidate = 999_999_999usize;
let found = unsafe {
search_max(
state,
ip,
iend,
&mut candidate,
mls,
row_log,
search_method,
dict_mode,
)
};
if found > match_length {
match_length = found;
start = ip;
offbase = candidate;
}
if match_length < 4 {
let step = (ip as usize).wrapping_sub(anchor as usize) >> K_SEARCH_STRENGTH;
ip = ip.wrapping_add(step.wrapping_add(1));
unsafe { set_lazy_skipping(state, step.wrapping_add(1) > K_LAZY_SKIPPING_STEP) };
continue;
}
if depth >= 1 {
while ptr_lt(ip, ilimit) {
ip = ip.wrapping_add(1);
if dict_mode == DICT_NO_DICT
&& offbase != 0
&& offset_1 > 0
&& unsafe { read32(ip) == read32(ip.wrapping_sub(offset_1 as usize)) }
{
let ml_rep = unsafe {
count(
ip.wrapping_add(4),
ip.wrapping_add(4).wrapping_sub(offset_1 as usize),
iend,
) + 4
};
if ml_rep >= 4
&& score(ml_rep, 3, REPCODE1_TO_OFFBASE as usize, 0)
> score(match_length, 3, offbase, 1)
{
match_length = ml_rep;
offbase = REPCODE1_TO_OFFBASE as usize;
start = ip;
}
}
if is_dxs {
let rep_index = unsafe { index_from(base, ip) }.wrapping_sub(offset_1);
let rep_match = if rep_index < prefix_lowest_index {
dict_base
.wrapping_add(rep_index.wrapping_sub(dict_index_delta) as usize)
} else {
base.wrapping_add(rep_index as usize)
};
if index_overlap_check(prefix_lowest_index, rep_index)
&& unsafe { read32(rep_match) == read32(ip) }
{
let rep_end = if rep_index < prefix_lowest_index {
dict_end
} else {
iend
};
let ml_rep = unsafe {
count_2segments(
ip.wrapping_add(4),
rep_match.wrapping_add(4),
iend,
rep_end,
prefix_lowest,
) + 4
};
if ml_rep >= 4
&& score(ml_rep, 3, REPCODE1_TO_OFFBASE as usize, 0)
> score(match_length, 3, offbase, 1)
{
match_length = ml_rep;
offbase = REPCODE1_TO_OFFBASE as usize;
start = ip;
}
}
}
let mut candidate = 999_999_999usize;
let found = unsafe {
search_max(
state,
ip,
iend,
&mut candidate,
mls,
row_log,
search_method,
dict_mode,
)
};
if found >= 4
&& score(found, 4, candidate, 0) > score(match_length, 4, offbase, 4)
{
match_length = found;
offbase = candidate;
start = ip;
continue;
}
if depth == 2 && ptr_lt(ip, ilimit) {
ip = ip.wrapping_add(1);
if dict_mode == DICT_NO_DICT
&& offbase != 0
&& offset_1 > 0
&& unsafe { read32(ip) == read32(ip.wrapping_sub(offset_1 as usize)) }
{
let ml_rep = unsafe {
count(
ip.wrapping_add(4),
ip.wrapping_add(4).wrapping_sub(offset_1 as usize),
iend,
) + 4
};
if ml_rep >= 4
&& score(ml_rep, 4, REPCODE1_TO_OFFBASE as usize, 0)
> score(match_length, 4, offbase, 1)
{
match_length = ml_rep;
offbase = REPCODE1_TO_OFFBASE as usize;
start = ip;
}
}
if is_dxs {
let rep_index = unsafe { index_from(base, ip) }.wrapping_sub(offset_1);
let rep_match = if rep_index < prefix_lowest_index {
dict_base
.wrapping_add(rep_index.wrapping_sub(dict_index_delta) as usize)
} else {
base.wrapping_add(rep_index as usize)
};
if index_overlap_check(prefix_lowest_index, rep_index)
&& unsafe { read32(rep_match) == read32(ip) }
{
let rep_end = if rep_index < prefix_lowest_index {
dict_end
} else {
iend
};
let ml_rep = unsafe {
count_2segments(
ip.wrapping_add(4),
rep_match.wrapping_add(4),
iend,
rep_end,
prefix_lowest,
) + 4
};
if ml_rep >= 4
&& score(ml_rep, 4, REPCODE1_TO_OFFBASE as usize, 0)
> score(match_length, 4, offbase, 1)
{
match_length = ml_rep;
offbase = REPCODE1_TO_OFFBASE as usize;
start = ip;
}
}
}
let mut second_candidate = 999_999_999usize;
let second_found = unsafe {
search_max(
state,
ip,
iend,
&mut second_candidate,
mls,
row_log,
search_method,
dict_mode,
)
};
if second_found >= 4
&& score(second_found, 4, second_candidate, 0)
> score(match_length, 4, offbase, 7)
{
match_length = second_found;
offbase = second_candidate;
start = ip;
continue;
}
}
break;
}
}
if offbase_is_offset(offbase) {
let offset = offbase_to_offset(offbase);
if dict_mode == DICT_NO_DICT {
while ptr_gt(start, anchor)
&& ptr_gt(start.wrapping_sub(offset as usize), prefix_lowest)
&& unsafe {
*start.wrapping_sub(1) == *start.wrapping_sub(offset as usize + 1)
}
{
start = start.wrapping_sub(1);
match_length += 1;
}
}
if is_dxs {
let match_index = unsafe { index_from(base, start) }.wrapping_sub(offset);
let mut matched = if match_index < prefix_lowest_index {
dict_base.wrapping_add(match_index.wrapping_sub(dict_index_delta) as usize)
} else {
base.wrapping_add(match_index as usize)
};
let match_start = if match_index < prefix_lowest_index {
dict_lowest
} else {
prefix_lowest
};
while ptr_gt(start, anchor)
&& ptr_gt(matched, match_start)
&& unsafe { *start.wrapping_sub(1) == *matched.wrapping_sub(1) }
{
start = start.wrapping_sub(1);
matched = matched.wrapping_sub(1);
match_length += 1;
}
}
offset_2 = offset_1;
offset_1 = offset;
}
}
let lit_length = (start as usize).wrapping_sub(anchor as usize);
unsafe {
store_seq(
seq_store,
lit_length,
anchor,
iend,
offbase as u32,
match_length,
)
};
anchor = start.wrapping_add(match_length);
ip = anchor;
if unsafe { lazy_skipping(state) } {
if search_method == SEARCH_ROW_HASH {
unsafe {
row_fill_hash_cache(state, base, row_log, mls, next_to_update(state), ilimit)
};
}
unsafe { set_lazy_skipping(state, false) };
}
if is_dxs {
while ptr_le(ip, ilimit) {
let current = unsafe { index_from(base, ip) };
let rep_index = current.wrapping_sub(offset_2);
let rep_match = if rep_index < prefix_lowest_index {
dict_base.wrapping_add(rep_index.wrapping_sub(dict_index_delta) as usize)
} else {
base.wrapping_add(rep_index as usize)
};
if index_overlap_check(prefix_lowest_index, rep_index)
&& unsafe { read32(rep_match) == read32(ip) }
{
let rep_end = if rep_index < prefix_lowest_index {
dict_end
} else {
iend
};
match_length = unsafe {
count_2segments(
ip.wrapping_add(4),
rep_match.wrapping_add(4),
iend,
rep_end,
prefix_lowest,
) + 4
};
std::mem::swap(&mut offset_2, &mut offset_1);
unsafe {
store_seq(
seq_store,
0,
anchor,
iend,
REPCODE1_TO_OFFBASE,
match_length,
)
};
ip = ip.wrapping_add(match_length);
anchor = ip;
continue;
}
break;
}
}
if dict_mode == DICT_NO_DICT {
while ptr_le(ip, ilimit)
&& offset_2 > 0
&& unsafe { read32(ip) == read32(ip.wrapping_sub(offset_2 as usize)) }
{
match_length = unsafe {
count(
ip.wrapping_add(4),
ip.wrapping_add(4).wrapping_sub(offset_2 as usize),
iend,
) + 4
};
std::mem::swap(&mut offset_2, &mut offset_1);
unsafe {
store_seq(
seq_store,
0,
anchor,
iend,
REPCODE1_TO_OFFBASE,
match_length,
)
};
ip = ip.wrapping_add(match_length);
anchor = ip;
}
}
}
if offset_saved_1 != 0 && offset_1 != 0 {
offset_saved_2 = offset_saved_1;
}
unsafe {
*reps = if offset_1 != 0 {
offset_1
} else {
offset_saved_1
};
*reps.add(1) = if offset_2 != 0 {
offset_2
} else {
offset_saved_2
};
}
(iend as usize).wrapping_sub(anchor as usize)
}
/* ------------------------------------------------------------------------- */
/* Lazy parser for external dictionaries */
/* ------------------------------------------------------------------------- */
unsafe fn compress_block_lazy_ext_dict_generic(
state: &ZSTD_RustLazyState,
seq_store: *mut SeqStore_t,
reps: *mut u32,
src: *const u8,
src_size: usize,
search_method: c_int,
depth: u32,
) -> usize {
let istart = src;
let mut ip = istart;
let mut anchor = istart;
let iend = istart.wrapping_add(src_size);
let required = if search_method == SEARCH_ROW_HASH {
HASH_READ_SIZE + ZSTD_ROW_HASH_CACHE_SIZE
} else {
HASH_READ_SIZE
};
if src_size < required {
return src_size;
}
let ilimit = iend.wrapping_sub(required);
let base = state.base;
let dict_limit = state.dictLimit;
let prefix_start = base.wrapping_add(dict_limit as usize);
let dict_base = state.dictBase;
let dict_end = dict_base.wrapping_add(dict_limit as usize);
let dict_start = dict_base.wrapping_add(state.lowLimit as usize);
let mls = bounded(4, state.minMatch, 6);
let row_log = bounded(4, state.searchLog, 6);
let mut offset_1 = unsafe { *reps };
let mut offset_2 = unsafe { *reps.add(1) };
if ip == prefix_start {
ip = ip.wrapping_add(1);
}
unsafe { set_lazy_skipping(state, false) };
if search_method == SEARCH_ROW_HASH {
unsafe { row_fill_hash_cache(state, base, row_log, mls, next_to_update(state), ilimit) };
}
while ptr_lt(ip, ilimit) {
let mut match_length = 0usize;
let mut offbase = REPCODE1_TO_OFFBASE as usize;
let mut start = ip.wrapping_add(1);
let mut curr = unsafe { index_from(base, ip) };
let mut store_directly = false;
let window_low = lowest_match_index(
state.lowLimit,
state.loadedDictEnd,
curr.wrapping_add(1),
state.windowLog,
);
let rep_index = curr.wrapping_add(1).wrapping_sub(offset_1);
let rep_base = if rep_index < dict_limit {
dict_base
} else {
base
};
let rep_match = rep_base.wrapping_add(rep_index as usize);
if index_overlap_check(dict_limit, rep_index)
&& offset_1 <= curr.wrapping_add(1).wrapping_sub(window_low)
&& unsafe { read32(ip.wrapping_add(1)) == read32(rep_match) }
{
let rep_end = if rep_index < dict_limit {
dict_end
} else {
iend
};
match_length = unsafe {
count_2segments(
ip.wrapping_add(5),
rep_match.wrapping_add(4),
iend,
rep_end,
prefix_start,
) + 4
};
if depth == 0 {
store_directly = true;
}
}
if !store_directly {
let mut candidate = 999_999_999usize;
let found = unsafe {
search_max(
state,
ip,
iend,
&mut candidate,
mls,
row_log,
search_method,
DICT_EXT,
)
};
if found > match_length {
match_length = found;
start = ip;
offbase = candidate;
}
if match_length < 4 {
let step = (ip as usize).wrapping_sub(anchor as usize) >> K_SEARCH_STRENGTH;
ip = ip.wrapping_add(step.wrapping_add(1));
unsafe { set_lazy_skipping(state, step > K_LAZY_SKIPPING_STEP) };
continue;
}
if depth >= 1 {
while ptr_lt(ip, ilimit) {
ip = ip.wrapping_add(1);
curr = curr.wrapping_add(1);
if offbase != 0 {
let window_low = lowest_match_index(
state.lowLimit,
state.loadedDictEnd,
curr,
state.windowLog,
);
let rep_index = curr.wrapping_sub(offset_1);
let rep_base = if rep_index < dict_limit {
dict_base
} else {
base
};
let rep_match = rep_base.wrapping_add(rep_index as usize);
if index_overlap_check(dict_limit, rep_index)
&& offset_1 <= curr.wrapping_sub(window_low)
&& unsafe { read32(ip) == read32(rep_match) }
{
let rep_end = if rep_index < dict_limit {
dict_end
} else {
iend
};
let rep_length = unsafe {
count_2segments(
ip.wrapping_add(4),
rep_match.wrapping_add(4),
iend,
rep_end,
prefix_start,
) + 4
};
if rep_length >= 4
&& score(rep_length, 3, REPCODE1_TO_OFFBASE as usize, 0)
> score(match_length, 3, offbase, 1)
{
match_length = rep_length;
offbase = REPCODE1_TO_OFFBASE as usize;
start = ip;
}
}
}
let mut candidate = 999_999_999usize;
let found = unsafe {
search_max(
state,
ip,
iend,
&mut candidate,
mls,
row_log,
search_method,
DICT_EXT,
)
};
if found >= 4
&& score(found, 4, candidate, 0) > score(match_length, 4, offbase, 4)
{
match_length = found;
offbase = candidate;
start = ip;
continue;
}
if depth == 2 && ptr_lt(ip, ilimit) {
ip = ip.wrapping_add(1);
curr = curr.wrapping_add(1);
if offbase != 0 {
let window_low = lowest_match_index(
state.lowLimit,
state.loadedDictEnd,
curr,
state.windowLog,
);
let rep_index = curr.wrapping_sub(offset_1);
let rep_base = if rep_index < dict_limit {
dict_base
} else {
base
};
let rep_match = rep_base.wrapping_add(rep_index as usize);
if index_overlap_check(dict_limit, rep_index)
&& offset_1 <= curr.wrapping_sub(window_low)
&& unsafe { read32(ip) == read32(rep_match) }
{
let rep_end = if rep_index < dict_limit {
dict_end
} else {
iend
};
let rep_length = unsafe {
count_2segments(
ip.wrapping_add(4),
rep_match.wrapping_add(4),
iend,
rep_end,
prefix_start,
) + 4
};
if rep_length >= 4
&& score(rep_length, 4, REPCODE1_TO_OFFBASE as usize, 0)
> score(match_length, 4, offbase, 1)
{
match_length = rep_length;
offbase = REPCODE1_TO_OFFBASE as usize;
start = ip;
}
}
}
let mut second_candidate = 999_999_999usize;
let second_found = unsafe {
search_max(
state,
ip,
iend,
&mut second_candidate,
mls,
row_log,
search_method,
DICT_EXT,
)
};
if second_found >= 4
&& score(second_found, 4, second_candidate, 0)
> score(match_length, 4, offbase, 7)
{
match_length = second_found;
offbase = second_candidate;
start = ip;
continue;
}
}
break;
}
}
if offbase_is_offset(offbase) {
let offset = offbase_to_offset(offbase);
let match_index = unsafe { index_from(base, start) }.wrapping_sub(offset);
let mut matched = if match_index < dict_limit {
dict_base.wrapping_add(match_index as usize)
} else {
base.wrapping_add(match_index as usize)
};
let match_start = if match_index < dict_limit {
dict_start
} else {
prefix_start
};
while ptr_gt(start, anchor)
&& ptr_gt(matched, match_start)
&& unsafe { *start.wrapping_sub(1) == *matched.wrapping_sub(1) }
{
start = start.wrapping_sub(1);
matched = matched.wrapping_sub(1);
match_length += 1;
}
offset_2 = offset_1;
offset_1 = offset;
}
}
let lit_length = (start as usize).wrapping_sub(anchor as usize);
unsafe {
store_seq(
seq_store,
lit_length,
anchor,
iend,
offbase as u32,
match_length,
)
};
anchor = start.wrapping_add(match_length);
ip = anchor;
if unsafe { lazy_skipping(state) } {
if search_method == SEARCH_ROW_HASH {
unsafe {
row_fill_hash_cache(state, base, row_log, mls, next_to_update(state), ilimit)
};
}
unsafe { set_lazy_skipping(state, false) };
}
while ptr_le(ip, ilimit) {
let rep_current = unsafe { index_from(base, ip) };
let window_low = lowest_match_index(
state.lowLimit,
state.loadedDictEnd,
rep_current,
state.windowLog,
);
let rep_index = rep_current.wrapping_sub(offset_2);
let rep_base = if rep_index < dict_limit {
dict_base
} else {
base
};
let rep_match = rep_base.wrapping_add(rep_index as usize);
if index_overlap_check(dict_limit, rep_index)
&& offset_2 <= rep_current.wrapping_sub(window_low)
&& unsafe { read32(ip) == read32(rep_match) }
{
let rep_end = if rep_index < dict_limit {
dict_end
} else {
iend
};
match_length = unsafe {
count_2segments(
ip.wrapping_add(4),
rep_match.wrapping_add(4),
iend,
rep_end,
prefix_start,
) + 4
};
std::mem::swap(&mut offset_2, &mut offset_1);
unsafe {
store_seq(
seq_store,
0,
anchor,
iend,
REPCODE1_TO_OFFBASE,
match_length,
)
};
ip = ip.wrapping_add(match_length);
anchor = ip;
continue;
}
break;
}
}
unsafe {
*reps = offset_1;
*reps.add(1) = offset_2;
}
(iend as usize).wrapping_sub(anchor as usize)
}
/* ------------------------------------------------------------------------- */
/* C ABI */
/* ------------------------------------------------------------------------- */
#[no_mangle]
pub unsafe extern "C" fn ZSTD_rust_lazy_insertAndFindFirstIndex(
state: *mut ZSTD_RustLazyState,
ip: *const c_void,
) -> u32 {
let state = unsafe { &*state };
unsafe { insert_and_find_first_index_internal(state, ip.cast::<u8>(), state.minMatch, false) }
}
#[no_mangle]
pub unsafe extern "C" fn ZSTD_rust_lazy_row_update(
state: *mut ZSTD_RustLazyState,
ip: *const c_void,
) {
let state = unsafe { &*state };
let row_log = bounded(4, state.searchLog, 6);
let row_mask = (1u32 << row_log) - 1;
let mls = state.minMatch.min(6);
unsafe { row_update_internal(state, ip.cast::<u8>(), mls, row_log, row_mask, false) };
}
#[no_mangle]
pub unsafe extern "C" fn ZSTD_rust_lazy_loadDedicatedDict(
state: *mut ZSTD_RustLazyState,
ip: *const c_void,
) {
let state = unsafe { &*state };
unsafe { dedicated_dict_search_load_dictionary(state, ip.cast::<u8>()) };
}
#[no_mangle]
pub unsafe extern "C" fn ZSTD_rust_compressBlock_lazy(
state: *mut ZSTD_RustLazyState,
seq_store: *mut c_void,
reps: *mut u32,
src: *const c_void,
src_size: usize,
search_method: c_int,
depth: u32,
dict_mode: c_int,
) -> usize {
let state = unsafe { &*state };
if dict_mode == DICT_EXT {
unsafe {
compress_block_lazy_ext_dict_generic(
state,
seq_store.cast::<SeqStore_t>(),
reps,
src.cast::<u8>(),
src_size,
search_method,
depth,
)
}
} else {
unsafe {
compress_block_lazy_generic(
state,
seq_store.cast::<SeqStore_t>(),
reps,
src.cast::<u8>(),
src_size,
search_method,
depth,
dict_mode,
)
}
}
}