forked from mirror/brotli
175 lines
5.0 KiB
Go
175 lines
5.0 KiB
Go
package brotli
|
|
|
|
/* NOTE: this hasher does not search in the dictionary. It is used as
|
|
backup-hasher, the main hasher already searches in it. */
|
|
/* NOLINT(build/header_guard) */
|
|
/* Copyright 2018 Google Inc. All Rights Reserved.
|
|
|
|
Distributed under MIT license.
|
|
See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
|
|
*/
|
|
|
|
/* Rolling hash for long distance long string matches. Stores one position
|
|
per bucket, bucket key is computed over a long region. */
|
|
var kRollingHashMul32HROLLING uint32 = 69069
|
|
|
|
var kInvalidPosHROLLING uint32 = 0xffffffff
|
|
|
|
/* This hasher uses a longer forward length, but returning a higher value here
|
|
will hurt compression by the main hasher when combined with a composite
|
|
hasher. The hasher tests for forward itself instead. */
|
|
func HashTypeLengthHROLLING() uint {
|
|
return 4
|
|
}
|
|
|
|
func StoreLookaheadHROLLING() uint {
|
|
return 4
|
|
}
|
|
|
|
/* Computes a code from a single byte. A lookup table of 256 values could be
|
|
used, but simply adding 1 works about as good. */
|
|
func HashByteHROLLING(byte byte) uint32 {
|
|
return uint32(byte) + 1
|
|
}
|
|
|
|
func HashRollingFunctionInitialHROLLING(state uint32, add byte, factor uint32) uint32 {
|
|
return uint32(factor*state + HashByteHROLLING(add))
|
|
}
|
|
|
|
func HashRollingFunctionHROLLING(state uint32, add byte, rem byte, factor uint32, factor_remove uint32) uint32 {
|
|
return uint32(factor*state + HashByteHROLLING(add) - factor_remove*HashByteHROLLING(rem))
|
|
}
|
|
|
|
type HROLLING struct {
|
|
HasherCommon
|
|
state uint32
|
|
table []uint32
|
|
next_ix uint
|
|
chunk_len uint32
|
|
factor uint32
|
|
factor_remove uint32
|
|
}
|
|
|
|
func SelfHROLLING(handle HasherHandle) *HROLLING {
|
|
return handle.(*HROLLING)
|
|
}
|
|
|
|
func (h *HROLLING) Initialize(params *BrotliEncoderParams) {
|
|
var i uint
|
|
h.state = 0
|
|
h.next_ix = 0
|
|
|
|
h.factor = kRollingHashMul32HROLLING
|
|
|
|
/* Compute the factor of the oldest byte to remove: factor**steps modulo
|
|
0xffffffff (the multiplications rely on 32-bit overflow) */
|
|
h.factor_remove = 1
|
|
|
|
for i = 0; i < 32; i += 1 {
|
|
h.factor_remove *= h.factor
|
|
}
|
|
|
|
h.table = make([]uint32, 16777216)
|
|
for i = 0; i < 16777216; i++ {
|
|
h.table[i] = kInvalidPosHROLLING
|
|
}
|
|
}
|
|
|
|
func (h *HROLLING) Prepare(one_shot bool, input_size uint, data []byte) {
|
|
var i uint
|
|
|
|
/* Too small size, cannot use this hasher. */
|
|
if input_size < 32 {
|
|
return
|
|
}
|
|
h.state = 0
|
|
for i = 0; i < 32; i += 1 {
|
|
h.state = HashRollingFunctionInitialHROLLING(h.state, data[i], h.factor)
|
|
}
|
|
}
|
|
|
|
func StoreHROLLING(handle HasherHandle, data []byte, mask uint, ix uint) {
|
|
}
|
|
|
|
func StoreRangeHROLLING(handle HasherHandle, data []byte, mask uint, ix_start uint, ix_end uint) {
|
|
}
|
|
|
|
func (h *HROLLING) StitchToPreviousBlock(num_bytes uint, position uint, ringbuffer []byte, ring_buffer_mask uint) {
|
|
var position_masked uint
|
|
/* In this case we must re-initialize the hasher from scratch from the
|
|
current position. */
|
|
|
|
var available uint = num_bytes
|
|
if position&(1-1) != 0 {
|
|
var diff uint = 1 - (position & (1 - 1))
|
|
if diff > available {
|
|
available = 0
|
|
} else {
|
|
available = available - diff
|
|
}
|
|
position += diff
|
|
}
|
|
|
|
position_masked = position & ring_buffer_mask
|
|
|
|
/* wrapping around ringbuffer not handled. */
|
|
if available > ring_buffer_mask-position_masked {
|
|
available = ring_buffer_mask - position_masked
|
|
}
|
|
|
|
h.Prepare(false, available, ringbuffer[position&ring_buffer_mask:])
|
|
h.next_ix = position
|
|
}
|
|
|
|
func PrepareDistanceCacheHROLLING(handle HasherHandle, distance_cache *int) {
|
|
}
|
|
|
|
func FindLongestMatchHROLLING(handle HasherHandle, dictionary *BrotliEncoderDictionary, data []byte, ring_buffer_mask uint, distance_cache *int, cur_ix uint, max_length uint, max_backward uint, gap uint, max_distance uint, out *HasherSearchResult) {
|
|
var self *HROLLING = SelfHROLLING(handle)
|
|
var cur_ix_masked uint = cur_ix & ring_buffer_mask
|
|
var pos uint = self.next_ix
|
|
|
|
if cur_ix&(1-1) != 0 {
|
|
return
|
|
}
|
|
|
|
/* Not enough lookahead */
|
|
if max_length < 32 {
|
|
return
|
|
}
|
|
|
|
for pos = self.next_ix; pos <= cur_ix; pos += 1 {
|
|
var code uint32 = self.state & ((16777216 * 64) - 1)
|
|
var rem byte = data[pos&ring_buffer_mask]
|
|
var add byte = data[(pos+32)&ring_buffer_mask]
|
|
var found_ix uint = uint(kInvalidPosHROLLING)
|
|
|
|
self.state = HashRollingFunctionHROLLING(self.state, add, rem, self.factor, self.factor_remove)
|
|
|
|
if code < 16777216 {
|
|
found_ix = uint(self.table[code])
|
|
self.table[code] = uint32(pos)
|
|
if pos == cur_ix && uint32(found_ix) != kInvalidPosHROLLING {
|
|
/* The cast to 32-bit makes backward distances up to 4GB work even
|
|
if cur_ix is above 4GB, despite using 32-bit values in the table. */
|
|
var backward uint = uint(uint32(cur_ix - found_ix))
|
|
if backward <= max_backward {
|
|
var found_ix_masked uint = found_ix & ring_buffer_mask
|
|
var len uint = FindMatchLengthWithLimit(data[found_ix_masked:], data[cur_ix_masked:], max_length)
|
|
if len >= 4 && len > out.len {
|
|
var score uint = BackwardReferenceScore(uint(len), backward)
|
|
if score > out.score {
|
|
out.len = uint(len)
|
|
out.distance = backward
|
|
out.score = score
|
|
out.len_code_delta = 0
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
self.next_ix = cur_ix + 1
|
|
}
|