serenity/AK/CircularBuffer.cpp
Tim Schumacher b1136ba357 AK: Efficiently resize CircularBuffer seekback copy distance
Previously, if we copied the last byte for a length of 100, we'd
recalculate the read span 100 times and memmove one byte 100 times,
which resulted in a lot of overhead.

Now, if we know that we have two consecutive copies of the data, we just
extend the distance to cover both copies, which halves the number of
times that we recalculate the span and actually call memmove.

This takes the running time of the attached benchmark case from 150ms
down to 15ms.
2023-04-14 10:03:42 +02:00

236 lines
6.5 KiB
C++

/*
* Copyright (c) 2022, Lucas Chollet <lucas.chollet@free.fr>
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#include <AK/CircularBuffer.h>
#include <AK/MemMem.h>
#include <AK/Stream.h>
namespace AK {
CircularBuffer::CircularBuffer(ByteBuffer buffer)
: m_buffer(move(buffer))
{
}
ErrorOr<CircularBuffer> CircularBuffer::create_empty(size_t size)
{
auto temporary_buffer = TRY(ByteBuffer::create_uninitialized(size));
CircularBuffer circular_buffer { move(temporary_buffer) };
return circular_buffer;
}
ErrorOr<CircularBuffer> CircularBuffer::create_initialized(ByteBuffer buffer)
{
CircularBuffer circular_buffer { move(buffer) };
circular_buffer.m_used_space = circular_buffer.m_buffer.size();
return circular_buffer;
}
size_t CircularBuffer::empty_space() const
{
return capacity() - m_used_space;
}
size_t CircularBuffer::used_space() const
{
return m_used_space;
}
size_t CircularBuffer::capacity() const
{
return m_buffer.size();
}
size_t CircularBuffer::seekback_limit() const
{
return m_seekback_limit;
}
bool CircularBuffer::is_wrapping_around() const
{
return capacity() <= m_reading_head + m_used_space;
}
Optional<size_t> CircularBuffer::offset_of(StringView needle, Optional<size_t> from, Optional<size_t> until) const
{
auto const read_from = from.value_or(0);
auto const read_until = until.value_or(m_used_space);
VERIFY(read_from <= read_until);
Array<ReadonlyBytes, 2> spans {};
spans[0] = next_read_span();
auto const original_span_0_size = spans[0].size();
if (read_from > 0)
spans[0] = spans[0].slice(min(spans[0].size(), read_from));
if (spans[0].size() + read_from > read_until)
spans[0] = spans[0].trim(read_until - read_from);
else if (is_wrapping_around())
spans[1] = m_buffer.span().slice(max(original_span_0_size, read_from) - original_span_0_size, min(read_until, m_used_space) - original_span_0_size);
auto maybe_found = AK::memmem(spans.begin(), spans.end(), needle.bytes());
if (maybe_found.has_value())
*maybe_found += read_from;
return maybe_found;
}
void CircularBuffer::clear()
{
m_reading_head = 0;
m_used_space = 0;
m_seekback_limit = 0;
}
Bytes CircularBuffer::next_write_span()
{
if (is_wrapping_around())
return m_buffer.span().slice(m_reading_head + m_used_space - capacity(), capacity() - m_used_space);
return m_buffer.span().slice(m_reading_head + m_used_space, capacity() - (m_reading_head + m_used_space));
}
ReadonlyBytes CircularBuffer::next_read_span() const
{
return m_buffer.span().slice(m_reading_head, min(capacity() - m_reading_head, m_used_space));
}
ReadonlyBytes CircularBuffer::next_read_span_with_seekback(size_t distance) const
{
VERIFY(m_seekback_limit <= capacity());
VERIFY(distance <= m_seekback_limit);
// Note: We are adding the capacity once here to ensure that we can wrap around the negative space by using modulo.
auto read_offset = (capacity() + m_reading_head + m_used_space - distance) % capacity();
return m_buffer.span().slice(read_offset, min(capacity() - read_offset, distance));
}
size_t CircularBuffer::write(ReadonlyBytes bytes)
{
auto remaining = bytes.size();
while (remaining > 0) {
auto const next_span = next_write_span();
if (next_span.size() == 0)
break;
auto const written_bytes = bytes.slice(bytes.size() - remaining).copy_trimmed_to(next_span);
m_used_space += written_bytes;
m_seekback_limit += written_bytes;
if (m_seekback_limit > capacity())
m_seekback_limit = capacity();
remaining -= written_bytes;
}
return bytes.size() - remaining;
}
Bytes CircularBuffer::read(Bytes bytes)
{
auto remaining = bytes.size();
while (remaining > 0) {
auto const next_span = next_read_span();
if (next_span.size() == 0)
break;
auto written_bytes = next_span.copy_trimmed_to(bytes.slice(bytes.size() - remaining));
m_used_space -= written_bytes;
m_reading_head += written_bytes;
if (m_reading_head >= capacity())
m_reading_head -= capacity();
remaining -= written_bytes;
}
return bytes.trim(bytes.size() - remaining);
}
ErrorOr<Bytes> CircularBuffer::read_with_seekback(Bytes bytes, size_t distance)
{
if (distance > m_seekback_limit)
return Error::from_string_literal("Tried a seekback read beyond the seekback limit");
auto remaining = bytes.size();
while (remaining > 0) {
auto const next_span = next_read_span_with_seekback(distance);
if (next_span.size() == 0)
break;
auto written_bytes = next_span.copy_trimmed_to(bytes.slice(bytes.size() - remaining));
distance -= written_bytes;
remaining -= written_bytes;
}
return bytes.trim(bytes.size() - remaining);
}
ErrorOr<void> CircularBuffer::discard(size_t discarding_size)
{
if (m_used_space < discarding_size)
return Error::from_string_literal("Can not discard more data than what the buffer contains");
m_used_space -= discarding_size;
m_reading_head = (m_reading_head + discarding_size) % capacity();
return {};
}
ErrorOr<size_t> CircularBuffer::fill_from_stream(Stream& stream)
{
auto next_span = next_write_span();
if (next_span.size() == 0)
return 0;
auto bytes = TRY(stream.read_some(next_span));
m_used_space += bytes.size();
m_seekback_limit += bytes.size();
if (m_seekback_limit > capacity())
m_seekback_limit = capacity();
return bytes.size();
}
ErrorOr<size_t> CircularBuffer::copy_from_seekback(size_t distance, size_t length)
{
if (distance > m_seekback_limit)
return Error::from_string_literal("Tried a seekback copy beyond the seekback limit");
auto remaining_length = length;
while (remaining_length > 0) {
if (empty_space() == 0)
break;
auto next_span = next_read_span_with_seekback(distance);
if (next_span.size() == 0)
break;
auto length_written = write(next_span.trim(remaining_length));
remaining_length -= length_written;
// If we copied right from the end of the seekback area (i.e. our length is larger than the distance)
// and the last copy was one complete "chunk", we can now double the distance to copy twice as much data in one go.
if (remaining_length > distance && length_written == distance)
distance *= 2;
}
return length - remaining_length;
}
}