RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-07-27 01:07:36 +00:00
Code Issues Activity Actions

AK: Use hashing to accelerate searching a CircularBuffer

Browse source
This commit is contained in:
Tim Schumacher 2023-06-02 13:37:47 +02:00 • committed by Linus Groh
parent 4a10cf1506
commit 60ac254df6
4 changed files with 269 additions and 101 deletions
Download patch file Download diff file Expand all files Collapse all files

19
AK/CircularBuffer.h
View file

@ -8,6 +8,7 @@
#include <AK/ByteBuffer.h>
#include <AK/Error.h>
#include <AK/HashMap.h>
#include <AK/Noncopyable.h>
#include <AK/Vector.h>
@ -67,6 +68,11 @@ public:
[[nodiscard]] size_t search_limit() const;
// These functions update the read pointer, so we need to hash any data that we have processed.
ErrorOr<Bytes> read(Bytes bytes);
ErrorOr<void> discard(size_t discarded_bytes);
ErrorOr<size_t> flush_to_stream(Stream& stream);
struct Match {
size_t distance;
size_t length;
@ -74,15 +80,26 @@ public:
/// This searches the seekback buffer (between read head and limit) for occurrences where it matches the next `length` bytes from the read buffer.
/// Supplying any hints will only consider those distances, in case existing offsets need to be validated.
/// Note that, since we only start searching at the read head, the length between read head and write head is excluded from the distance.
ErrorOr<Vector<Match>> find_copy_in_seekback(size_t maximum_length, size_t minimum_length = 2) const;
Optional<Match> find_copy_in_seekback(size_t maximum_length, size_t minimum_length = 2);
Optional<Match> find_copy_in_seekback(ReadonlySpan<size_t> distances, size_t maximum_length, size_t minimum_length = 2) const;
// The chunk size for which the hash table holds hashes.
// This is nice for users to know, as picking a minimum match length that is
// equal or greater than this allows us to completely skip a slow memory search.
static constexpr size_t HASH_CHUNK_SIZE = 3;
private:
// Note: This function has a similar purpose as next_seekback_span, but they differ in their reference point.
// Seekback operations start counting their distance at the write head, while search operations start counting their distance at the read head.
[[nodiscard]] ReadonlyBytes next_search_span(size_t distance) const;
SearchableCircularBuffer(ByteBuffer);
HashMap<unsigned, size_t> m_hash_location_map;
HashMap<size_t, size_t> m_location_chain_map;
ErrorOr<void> insert_location_hash(ReadonlyBytes value, size_t raw_offset);
ErrorOr<void> hash_last_bytes(size_t count);
};
}