0ct0pu5/ladybird
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 2
ladybird/Userland/Libraries/LibAudio/FlacLoader.cpp
kleines Filmröllchen 264cc76ab4 LibAudio: Move audio stream buffering into the loader 
Before, some loader plugins implemented their own buffering (FLAC&MP3),
some didn't require any (WAV), and some didn't buffer at all (QOA). This
meant that in practice, while you could load arbitrary amounts of
samples from some loader plugins, you couldn't do that with some others.
Also, it was ill-defined how many samples you would actually get back
from a get_more_samples call.

This commit fixes that by introducing a layer of abstraction between the
loader and its plugins (because that's the whole point of having the
extra class!). The plugins now only implement a load_chunks() function,
which is much simpler to implement and allows plugins to play fast and
loose with what they actually return. Basically, they can return many
chunks of samples, where one chunk is simply a convenient block of
samples to load. In fact, some loaders such as FLAC and QOA have
separate internal functions for loading exactly one chunk. The loaders
*should* load as many chunks as necessary for the sample count to be
reached or surpassed (the latter simplifies loading loops in the
implementations, since you don't need to know how large your next chunk
is going to be; a problem for e.g. FLAC). If a plugin has no problems
returning data of arbitrary size (currently WAV), it can return a single
chunk that exactly (or roughly) matches the requested sample count. If a
plugin is at the stream end, it can also return less samples than was
requested! The loader can handle all of these cases and may call into
load_chunk multiple times. If the plugin returns an empty chunk list (or
only empty chunks; again, they can play fast and loose), the loader
takes that as a stream end signal. Otherwise, the loader will always
return exactly as many samples as the user requested. Buffering is
handled by the loader, allowing any underlying plugin to deal with any
weird sample count requirement the user throws at it (looking at you,
SoundPlayer!).

This (not accidentally!) makes QOA work in SoundPlayer.
2023-03-13 13:25:42 +01:00

914 lines
39 KiB
C++
Raw Blame History

/*
* Copyright (c) 2021, kleines Filmröllchen <filmroellchen@serenityos.org>
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#include <AK/Debug.h>
#include <AK/DeprecatedFlyString.h>
#include <AK/DeprecatedString.h>
#include <AK/FixedArray.h>
#include <AK/Format.h>
#include <AK/IntegralMath.h>
#include <AK/Math.h>
#include <AK/MemoryStream.h>
#include <AK/ScopeGuard.h>
#include <AK/StdLibExtras.h>
#include <AK/Try.h>
#include <AK/TypedTransfer.h>
#include <AK/UFixedBigInt.h>
#include <LibAudio/FlacLoader.h>
#include <LibAudio/FlacTypes.h>
#include <LibAudio/LoaderError.h>
#include <LibAudio/Resampler.h>
#include <LibCore/File.h>
namespace Audio {
FlacLoaderPlugin::FlacLoaderPlugin(NonnullOwnPtr<SeekableStream> stream)
: LoaderPlugin(move(stream))
{
}
Result<NonnullOwnPtr<FlacLoaderPlugin>, LoaderError> FlacLoaderPlugin::create(StringView path)
{
auto stream = LOADER_TRY(Core::BufferedFile::create(LOADER_TRY(Core::File::open(path, Core::File::OpenMode::Read))));
auto loader = make<FlacLoaderPlugin>(move(stream));
LOADER_TRY(loader->initialize());
return loader;
}
Result<NonnullOwnPtr<FlacLoaderPlugin>, LoaderError> FlacLoaderPlugin::create(Bytes buffer)
{
auto stream = LOADER_TRY(try_make<FixedMemoryStream>(buffer));
auto loader = make<FlacLoaderPlugin>(move(stream));
LOADER_TRY(loader->initialize());
return loader;
}
MaybeLoaderError FlacLoaderPlugin::initialize()
{
TRY(parse_header());
TRY(reset());
return {};
}
// 11.5 STREAM
MaybeLoaderError FlacLoaderPlugin::parse_header()
{
BigEndianInputBitStream bit_input { MaybeOwned<Stream>(*m_stream) };
// A mixture of VERIFY and the non-crashing TRY().
#define FLAC_VERIFY(check, category, msg) \
do { \
if (!(check)) { \
return LoaderError { category, static_cast<size_t>(m_data_start_location), DeprecatedString::formatted("FLAC header: {}", msg) }; \
} \
} while (0)
// Magic number
u32 flac = LOADER_TRY(bit_input.read_bits<u32>(32));
m_data_start_location += 4;
FLAC_VERIFY(flac == 0x664C6143, LoaderError::Category::Format, "Magic number must be 'flaC'"); // "flaC"
// Receive the streaminfo block
auto streaminfo = TRY(next_meta_block(bit_input));
FLAC_VERIFY(streaminfo.type == FlacMetadataBlockType::STREAMINFO, LoaderError::Category::Format, "First block must be STREAMINFO");
FixedMemoryStream streaminfo_data_memory { streaminfo.data.bytes() };
BigEndianInputBitStream streaminfo_data { MaybeOwned<Stream>(streaminfo_data_memory) };
// 11.10 METADATA_BLOCK_STREAMINFO
m_min_block_size = LOADER_TRY(streaminfo_data.read_bits<u16>(16));
FLAC_VERIFY(m_min_block_size >= 16, LoaderError::Category::Format, "Minimum block size must be 16");
m_max_block_size = LOADER_TRY(streaminfo_data.read_bits<u16>(16));
FLAC_VERIFY(m_max_block_size >= 16, LoaderError::Category::Format, "Maximum block size");
m_min_frame_size = LOADER_TRY(streaminfo_data.read_bits<u32>(24));
m_max_frame_size = LOADER_TRY(streaminfo_data.read_bits<u32>(24));
m_sample_rate = LOADER_TRY(streaminfo_data.read_bits<u32>(20));
FLAC_VERIFY(m_sample_rate <= 655350, LoaderError::Category::Format, "Sample rate");
m_num_channels = LOADER_TRY(streaminfo_data.read_bits<u8>(3)) + 1; // 0 = one channel
u8 bits_per_sample = LOADER_TRY(streaminfo_data.read_bits<u8>(5)) + 1;
if (bits_per_sample == 8) {
// FIXME: Signed/Unsigned issues?
m_sample_format = PcmSampleFormat::Uint8;
} else if (bits_per_sample == 16) {
m_sample_format = PcmSampleFormat::Int16;
} else if (bits_per_sample == 24) {
m_sample_format = PcmSampleFormat::Int24;
} else if (bits_per_sample == 32) {
m_sample_format = PcmSampleFormat::Int32;
} else {
FLAC_VERIFY(false, LoaderError::Category::Format, "Sample bit depth invalid");
}
m_total_samples = LOADER_TRY(streaminfo_data.read_bits<u64>(36));
FLAC_VERIFY(m_total_samples > 0, LoaderError::Category::Format, "Number of samples is zero");
// Parse checksum into a buffer first
[[maybe_unused]] u128 md5_checksum;
VERIFY(streaminfo_data.is_aligned_to_byte_boundary());
auto md5_bytes_read = LOADER_TRY(streaminfo_data.read(md5_checksum.bytes()));
FLAC_VERIFY(md5_bytes_read.size() == sizeof(md5_checksum), LoaderError::Category::IO, "MD5 Checksum size");
md5_checksum.bytes().copy_to({ m_md5_checksum, sizeof(m_md5_checksum) });
// Parse other blocks
[[maybe_unused]] u16 meta_blocks_parsed = 1;
[[maybe_unused]] u16 total_meta_blocks = meta_blocks_parsed;
FlacRawMetadataBlock block = streaminfo;
while (!block.is_last_block) {
block = TRY(next_meta_block(bit_input));
switch (block.type) {
case (FlacMetadataBlockType::SEEKTABLE):
TRY(load_seektable(block));
break;
case FlacMetadataBlockType::PICTURE:
TRY(load_picture(block));
break;
case FlacMetadataBlockType::APPLICATION:
// Note: Third-party library can encode specific data in this.
dbgln("Unknown 'Application' metadata block encountered.");
[[fallthrough]];
case FlacMetadataBlockType::PADDING:
// Note: A padding block is empty and does not need any treatment.
default:
// TODO: Parse the remaining metadata block types.
break;
}
++total_meta_blocks;
}
dbgln_if(AFLACLOADER_DEBUG, "Parsed FLAC header: blocksize {}-{}{}, framesize {}-{}, {}Hz, {}bit, {} channels, {} samples total ({:.2f}s), MD5 {}, data start at {:x} bytes, {} headers total (skipped {})", m_min_block_size, m_max_block_size, is_fixed_blocksize_stream() ? " (constant)" : "", m_min_frame_size, m_max_frame_size, m_sample_rate, pcm_bits_per_sample(m_sample_format), m_num_channels, m_total_samples, static_cast<float>(m_total_samples) / static_cast<float>(m_sample_rate), md5_checksum, m_data_start_location, total_meta_blocks, total_meta_blocks - meta_blocks_parsed);
return {};
}
// 11.19. METADATA_BLOCK_PICTURE
MaybeLoaderError FlacLoaderPlugin::load_picture(FlacRawMetadataBlock& block)
{
FixedMemoryStream memory_stream { block.data.bytes() };
BigEndianInputBitStream picture_block_bytes { MaybeOwned<Stream>(memory_stream) };
PictureData picture {};
picture.type = static_cast<ID3PictureType>(LOADER_TRY(picture_block_bytes.read_bits(32)));
auto const mime_string_length = LOADER_TRY(picture_block_bytes.read_bits(32));
// Note: We are seeking before reading the value to ensure that we stayed inside buffer's size.
auto offset_before_seeking = memory_stream.offset();
LOADER_TRY(memory_stream.seek(mime_string_length, SeekMode::FromCurrentPosition));
picture.mime_string = { block.data.bytes().data() + offset_before_seeking, (size_t)mime_string_length };
auto const description_string_length = LOADER_TRY(picture_block_bytes.read_bits(32));
offset_before_seeking = memory_stream.offset();
LOADER_TRY(memory_stream.seek(description_string_length, SeekMode::FromCurrentPosition));
picture.description_string = Vector<u32> { Span<u32> { reinterpret_cast<u32*>(block.data.bytes().data() + offset_before_seeking), (size_t)description_string_length } };
picture.width = LOADER_TRY(picture_block_bytes.read_bits(32));
picture.height = LOADER_TRY(picture_block_bytes.read_bits(32));
picture.color_depth = LOADER_TRY(picture_block_bytes.read_bits(32));
picture.colors = LOADER_TRY(picture_block_bytes.read_bits(32));
auto const picture_size = LOADER_TRY(picture_block_bytes.read_bits(32));
offset_before_seeking = memory_stream.offset();
LOADER_TRY(memory_stream.seek(picture_size, SeekMode::FromCurrentPosition));
picture.data = Vector<u8> { Span<u8> { block.data.bytes().data() + offset_before_seeking, (size_t)picture_size } };
m_pictures.append(move(picture));
return {};
}
// 11.13. METADATA_BLOCK_SEEKTABLE
MaybeLoaderError FlacLoaderPlugin::load_seektable(FlacRawMetadataBlock& block)
{
FixedMemoryStream memory_stream { block.data.bytes() };
BigEndianInputBitStream seektable_bytes { MaybeOwned<Stream>(memory_stream) };
for (size_t i = 0; i < block.length / 18; ++i) {
// 11.14. SEEKPOINT
FlacSeekPoint seekpoint {
.sample_index = LOADER_TRY(seektable_bytes.read_bits<u64>(64)),
.byte_offset = LOADER_TRY(seektable_bytes.read_bits<u64>(64)),
.num_samples = LOADER_TRY(seektable_bytes.read_bits<u16>(16))
};
m_seektable.append(seekpoint);
}
dbgln_if(AFLACLOADER_DEBUG, "Loaded seektable of size {}", m_seektable.size());
return {};
}
// 11.6 METADATA_BLOCK
ErrorOr<FlacRawMetadataBlock, LoaderError> FlacLoaderPlugin::next_meta_block(BigEndianInputBitStream& bit_input)
{
// 11.7 METADATA_BLOCK_HEADER
bool is_last_block = LOADER_TRY(bit_input.read_bit());
// The block type enum constants agree with the specification
FlacMetadataBlockType type = (FlacMetadataBlockType)LOADER_TRY(bit_input.read_bits<u8>(7));
m_data_start_location += 1;
FLAC_VERIFY(type != FlacMetadataBlockType::INVALID, LoaderError::Category::Format, "Invalid metadata block");
u32 block_length = LOADER_TRY(bit_input.read_bits<u32>(24));
m_data_start_location += 3;
// Blocks can be zero-sized, which would trip up the raw data reader below.
if (block_length == 0)
return FlacRawMetadataBlock {
.is_last_block = is_last_block,
.type = type,
.length = 0,
.data = LOADER_TRY(ByteBuffer::create_uninitialized(0))
};
auto block_data_result = ByteBuffer::create_uninitialized(block_length);
FLAC_VERIFY(!block_data_result.is_error(), LoaderError::Category::IO, "Out of memory");
auto block_data = block_data_result.release_value();
// Blocks might exceed our buffer size.
auto bytes_left_to_read = block_data.bytes();
while (bytes_left_to_read.size()) {
auto read_bytes = LOADER_TRY(bit_input.read(bytes_left_to_read));
bytes_left_to_read = bytes_left_to_read.slice(read_bytes.size());
}
m_data_start_location += block_length;
return FlacRawMetadataBlock {
is_last_block,
type,
block_length,
block_data,
};
}
#undef FLAC_VERIFY
MaybeLoaderError FlacLoaderPlugin::reset()
{
TRY(seek(0));
m_current_frame.clear();
return {};
}
MaybeLoaderError FlacLoaderPlugin::seek(int int_sample_index)
{
auto sample_index = static_cast<size_t>(int_sample_index);
if (sample_index == m_loaded_samples)
return {};
auto maybe_target_seekpoint = m_seektable.last_matching([sample_index](auto& seekpoint) { return seekpoint.sample_index <= sample_index; });
// No seektable or no fitting entry: Perform normal forward read
if (!maybe_target_seekpoint.has_value()) {
if (sample_index < m_loaded_samples) {
LOADER_TRY(m_stream->seek(m_data_start_location, SeekMode::SetPosition));
m_loaded_samples = 0;
}
auto to_read = sample_index - m_loaded_samples;
if (to_read == 0)
return {};
dbgln_if(AFLACLOADER_DEBUG, "Seeking {} samples manually", to_read);
(void)TRY(load_chunks(to_read));
} else {
auto target_seekpoint = maybe_target_seekpoint.release_value();
// When a small seek happens, we may already be closer to the target than the seekpoint.
if (sample_index - target_seekpoint.sample_index > sample_index - m_loaded_samples) {
dbgln_if(AFLACLOADER_DEBUG, "Close enough to target: seeking {} samples manually", sample_index - m_loaded_samples);
(void)TRY(load_chunks(sample_index - m_loaded_samples));
return {};
}
dbgln_if(AFLACLOADER_DEBUG, "Seeking to seektable: sample index {}, byte offset {}, sample count {}", target_seekpoint.sample_index, target_seekpoint.byte_offset, target_seekpoint.num_samples);
auto position = target_seekpoint.byte_offset + m_data_start_location;
if (m_stream->seek(static_cast<i64>(position), SeekMode::SetPosition).is_error())
return LoaderError { LoaderError::Category::IO, m_loaded_samples, DeprecatedString::formatted("Invalid seek position {}", position) };
auto remaining_samples_after_seekpoint = sample_index - m_data_start_location;
if (remaining_samples_after_seekpoint > 0)
(void)TRY(load_chunks(remaining_samples_after_seekpoint));
m_loaded_samples = target_seekpoint.sample_index;
}
return {};
}
ErrorOr<Vector<FixedArray<Sample>>, LoaderError> FlacLoaderPlugin::load_chunks(size_t samples_to_read_from_input)
{
ssize_t remaining_samples = static_cast<ssize_t>(m_total_samples - m_loaded_samples);
if (remaining_samples <= 0)
return Vector<FixedArray<Sample>> {};
size_t samples_to_read = min(samples_to_read_from_input, remaining_samples);
Vector<FixedArray<Sample>> frames;
size_t sample_index = 0;
while (sample_index < samples_to_read) {
TRY(frames.try_append(TRY(next_frame())));
sample_index += m_current_frame->sample_count;
}
m_loaded_samples += sample_index;
return frames;
}
// 11.21. FRAME
LoaderSamples FlacLoaderPlugin::next_frame()
{
#define FLAC_VERIFY(check, category, msg) \
do { \
if (!(check)) { \
return LoaderError { category, static_cast<size_t>(m_current_sample_or_frame), DeprecatedString::formatted("FLAC header: {}", msg) }; \
} \
} while (0)
BigEndianInputBitStream bit_stream { MaybeOwned<Stream>(*m_stream) };
// TODO: Check the CRC-16 checksum (and others) by keeping track of read data
// 11.22. FRAME_HEADER
u16 sync_code = LOADER_TRY(bit_stream.read_bits<u16>(14));
FLAC_VERIFY(sync_code == 0b11111111111110, LoaderError::Category::Format, "Sync code");
bool reserved_bit = LOADER_TRY(bit_stream.read_bit());
FLAC_VERIFY(reserved_bit == 0, LoaderError::Category::Format, "Reserved frame header bit");
// 11.22.2. BLOCKING STRATEGY
[[maybe_unused]] bool blocking_strategy = LOADER_TRY(bit_stream.read_bit());
u32 sample_count = TRY(convert_sample_count_code(LOADER_TRY(bit_stream.read_bits<u8>(4))));
u32 frame_sample_rate = TRY(convert_sample_rate_code(LOADER_TRY(bit_stream.read_bits<u8>(4))));
u8 channel_type_num = LOADER_TRY(bit_stream.read_bits<u8>(4));
FLAC_VERIFY(channel_type_num < 0b1011, LoaderError::Category::Format, "Channel assignment");
FlacFrameChannelType channel_type = (FlacFrameChannelType)channel_type_num;
PcmSampleFormat bit_depth = TRY(convert_bit_depth_code(LOADER_TRY(bit_stream.read_bits<u8>(3))));
reserved_bit = LOADER_TRY(bit_stream.read_bit());
FLAC_VERIFY(reserved_bit == 0, LoaderError::Category::Format, "Reserved frame header end bit");
// 11.22.8. CODED NUMBER
// FIXME: sample number can be 8-56 bits, frame number can be 8-48 bits
m_current_sample_or_frame = LOADER_TRY(read_utf8_char(bit_stream));
// Conditional header variables
// 11.22.9. BLOCK SIZE INT
if (sample_count == FLAC_BLOCKSIZE_AT_END_OF_HEADER_8) {
sample_count = LOADER_TRY(bit_stream.read_bits<u32>(8)) + 1;
} else if (sample_count == FLAC_BLOCKSIZE_AT_END_OF_HEADER_16) {
sample_count = LOADER_TRY(bit_stream.read_bits<u32>(16)) + 1;
}
// 11.22.10. SAMPLE RATE INT
if (frame_sample_rate == FLAC_SAMPLERATE_AT_END_OF_HEADER_8) {
frame_sample_rate = LOADER_TRY(bit_stream.read_bits<u32>(8)) * 1000;
} else if (frame_sample_rate == FLAC_SAMPLERATE_AT_END_OF_HEADER_16) {
frame_sample_rate = LOADER_TRY(bit_stream.read_bits<u32>(16));
} else if (frame_sample_rate == FLAC_SAMPLERATE_AT_END_OF_HEADER_16X10) {
frame_sample_rate = LOADER_TRY(bit_stream.read_bits<u32>(16)) * 10;
}
// 11.22.11. FRAME CRC
// TODO: check header checksum, see above
[[maybe_unused]] u8 checksum = LOADER_TRY(bit_stream.read_bits<u8>(8));
dbgln_if(AFLACLOADER_DEBUG, "Frame: {} samples, {}bit {}Hz, channeltype {:x}, {} number {}, header checksum {}", sample_count, pcm_bits_per_sample(bit_depth), frame_sample_rate, channel_type_num, blocking_strategy ? "sample" : "frame", m_current_sample_or_frame, checksum);
m_current_frame = FlacFrameHeader {
sample_count,
frame_sample_rate,
channel_type,
bit_depth,
};
u8 subframe_count = frame_channel_type_to_channel_count(channel_type);
Vector<Vector<i32>> current_subframes;
current_subframes.ensure_capacity(subframe_count);
for (u8 i = 0; i < subframe_count; ++i) {
FlacSubframeHeader new_subframe = TRY(next_subframe_header(bit_stream, i));
Vector<i32> subframe_samples = TRY(parse_subframe(new_subframe, bit_stream));
VERIFY(subframe_samples.size() == m_current_frame->sample_count);
current_subframes.unchecked_append(move(subframe_samples));
}
// 11.2. Overview ("The audio data is composed of...")
bit_stream.align_to_byte_boundary();
// 11.23. FRAME_FOOTER
// TODO: check checksum, see above
[[maybe_unused]] u16 footer_checksum = LOADER_TRY(bit_stream.read_bits<u16>(16));
dbgln_if(AFLACLOADER_DEBUG, "Subframe footer checksum: {}", footer_checksum);
float sample_rescale = 1 / static_cast<float>(1 << (pcm_bits_per_sample(m_current_frame->bit_depth) - 1));
dbgln_if(AFLACLOADER_DEBUG, "Sample rescaled from {} bits: factor {:.1f}", pcm_bits_per_sample(m_current_frame->bit_depth), sample_rescale);
FixedArray<Sample> samples = TRY(FixedArray<Sample>::create(m_current_frame->sample_count));
switch (channel_type) {
case FlacFrameChannelType::Mono:
for (size_t i = 0; i < m_current_frame->sample_count; ++i)
samples[i] = Sample { static_cast<float>(current_subframes[0][i]) * sample_rescale };
break;
case FlacFrameChannelType::Stereo:
// TODO mix together surround channels on each side?
case FlacFrameChannelType::StereoCenter:
case FlacFrameChannelType::Surround4p0:
case FlacFrameChannelType::Surround5p0:
case FlacFrameChannelType::Surround5p1:
case FlacFrameChannelType::Surround6p1:
case FlacFrameChannelType::Surround7p1:
for (size_t i = 0; i < m_current_frame->sample_count; ++i)
samples[i] = { static_cast<float>(current_subframes[0][i]) * sample_rescale, static_cast<float>(current_subframes[1][i]) * sample_rescale };
break;
case FlacFrameChannelType::LeftSideStereo:
// channels are left (0) and side (1)
for (size_t i = 0; i < m_current_frame->sample_count; ++i) {
// right = left - side
samples[i] = { static_cast<float>(current_subframes[0][i]) * sample_rescale,
static_cast<float>(current_subframes[0][i] - current_subframes[1][i]) * sample_rescale };
}
break;
case FlacFrameChannelType::RightSideStereo:
// channels are side (0) and right (1)
for (size_t i = 0; i < m_current_frame->sample_count; ++i) {
// left = right + side
samples[i] = { static_cast<float>(current_subframes[1][i] + current_subframes[0][i]) * sample_rescale,
static_cast<float>(current_subframes[1][i]) * sample_rescale };
}
break;
case FlacFrameChannelType::MidSideStereo:
// channels are mid (0) and side (1)
for (size_t i = 0; i < current_subframes[0].size(); ++i) {
i64 mid = current_subframes[0][i];
i64 side = current_subframes[1][i];
mid *= 2;
// prevent integer division errors
samples[i] = { static_cast<float>((mid + side) * .5f) * sample_rescale,
static_cast<float>((mid - side) * .5f) * sample_rescale };
}
break;
}
return samples;
#undef FLAC_VERIFY
}
// 11.22.3. INTERCHANNEL SAMPLE BLOCK SIZE
ErrorOr<u32, LoaderError> FlacLoaderPlugin::convert_sample_count_code(u8 sample_count_code)
{
// single codes
switch (sample_count_code) {
case 0:
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), "Reserved block size" };
case 1:
return 192;
case 6:
return FLAC_BLOCKSIZE_AT_END_OF_HEADER_8;
case 7:
return FLAC_BLOCKSIZE_AT_END_OF_HEADER_16;
}
if (sample_count_code >= 2 && sample_count_code <= 5) {
return 576 * AK::exp2(sample_count_code - 2);
}
return 256 * AK::exp2(sample_count_code - 8);
}
// 11.22.4. SAMPLE RATE
ErrorOr<u32, LoaderError> FlacLoaderPlugin::convert_sample_rate_code(u8 sample_rate_code)
{
switch (sample_rate_code) {
case 0:
return m_sample_rate;
case 1:
return 88200;
case 2:
return 176400;
case 3:
return 192000;
case 4:
return 8000;
case 5:
return 16000;
case 6:
return 22050;
case 7:
return 24000;
case 8:
return 32000;
case 9:
return 44100;
case 10:
return 48000;
case 11:
return 96000;
case 12:
return FLAC_SAMPLERATE_AT_END_OF_HEADER_8;
case 13:
return FLAC_SAMPLERATE_AT_END_OF_HEADER_16;
case 14:
return FLAC_SAMPLERATE_AT_END_OF_HEADER_16X10;
default:
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), "Invalid sample rate code" };
}
}
// 11.22.6. SAMPLE SIZE
ErrorOr<PcmSampleFormat, LoaderError> FlacLoaderPlugin::convert_bit_depth_code(u8 bit_depth_code)
{
switch (bit_depth_code) {
case 0:
return m_sample_format;
case 1:
return PcmSampleFormat::Uint8;
case 4:
return PcmSampleFormat::Int16;
case 6:
return PcmSampleFormat::Int24;
case 3:
case 7:
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), "Reserved sample size" };
default:
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), DeprecatedString::formatted("Unsupported sample size {}", bit_depth_code) };
}
}
// 11.22.5. CHANNEL ASSIGNMENT
u8 frame_channel_type_to_channel_count(FlacFrameChannelType channel_type)
{
if (channel_type <= FlacFrameChannelType::Surround7p1)
return to_underlying(channel_type) + 1;
return 2;
}
// 11.25. SUBFRAME_HEADER
ErrorOr<FlacSubframeHeader, LoaderError> FlacLoaderPlugin::next_subframe_header(BigEndianInputBitStream& bit_stream, u8 channel_index)
{
u8 bits_per_sample = static_cast<u16>(pcm_bits_per_sample(m_current_frame->bit_depth));
// For inter-channel correlation, the side channel needs an extra bit for its samples
switch (m_current_frame->channels) {
case FlacFrameChannelType::LeftSideStereo:
case FlacFrameChannelType::MidSideStereo:
if (channel_index == 1) {
++bits_per_sample;
}
break;
case FlacFrameChannelType::RightSideStereo:
if (channel_index == 0) {
++bits_per_sample;
}
break;
// "normal" channel types
default:
break;
}
// zero-bit padding
if (LOADER_TRY(bit_stream.read_bit()) != 0)
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), "Zero bit padding" };
// 11.25.1. SUBFRAME TYPE
u8 subframe_code = LOADER_TRY(bit_stream.read_bits<u8>(6));
if ((subframe_code >= 0b000010 && subframe_code <= 0b000111) || (subframe_code > 0b001100 && subframe_code < 0b100000))
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), "Subframe type" };
FlacSubframeType subframe_type;
u8 order = 0;
// LPC has the highest bit set
if ((subframe_code & 0b100000) > 0) {
subframe_type = FlacSubframeType::LPC;
order = (subframe_code & 0b011111) + 1;
} else if ((subframe_code & 0b001000) > 0) {
// Fixed has the third-highest bit set
subframe_type = FlacSubframeType::Fixed;
order = (subframe_code & 0b000111);
} else {
subframe_type = (FlacSubframeType)subframe_code;
}
// 11.25.2. WASTED BITS PER SAMPLE FLAG
bool has_wasted_bits = LOADER_TRY(bit_stream.read_bit());
u8 k = 0;
if (has_wasted_bits) {
bool current_k_bit = 0;
do {
current_k_bit = LOADER_TRY(bit_stream.read_bit());
++k;
} while (current_k_bit != 1);
}
return FlacSubframeHeader {
subframe_type,
order,
k,
bits_per_sample
};
}
ErrorOr<Vector<i32>, LoaderError> FlacLoaderPlugin::parse_subframe(FlacSubframeHeader& subframe_header, BigEndianInputBitStream& bit_input)
{
Vector<i32> samples;
switch (subframe_header.type) {
case FlacSubframeType::Constant: {
// 11.26. SUBFRAME_CONSTANT
u64 constant_value = LOADER_TRY(bit_input.read_bits<u64>(subframe_header.bits_per_sample - subframe_header.wasted_bits_per_sample));
dbgln_if(AFLACLOADER_DEBUG, "Constant subframe: {}", constant_value);
samples.ensure_capacity(m_current_frame->sample_count);
VERIFY(subframe_header.bits_per_sample - subframe_header.wasted_bits_per_sample != 0);
i32 constant = sign_extend(static_cast<u32>(constant_value), subframe_header.bits_per_sample - subframe_header.wasted_bits_per_sample);
for (u32 i = 0; i < m_current_frame->sample_count; ++i) {
samples.unchecked_append(constant);
}
break;
}
case FlacSubframeType::Fixed: {
dbgln_if(AFLACLOADER_DEBUG, "Fixed LPC subframe order {}", subframe_header.order);
samples = TRY(decode_fixed_lpc(subframe_header, bit_input));
break;
}
case FlacSubframeType::Verbatim: {
dbgln_if(AFLACLOADER_DEBUG, "Verbatim subframe");
samples = TRY(decode_verbatim(subframe_header, bit_input));
break;
}
case FlacSubframeType::LPC: {
dbgln_if(AFLACLOADER_DEBUG, "Custom LPC subframe order {}", subframe_header.order);
samples = TRY(decode_custom_lpc(subframe_header, bit_input));
break;
}
default:
return LoaderError { LoaderError::Category::Unimplemented, static_cast<size_t>(m_current_sample_or_frame), "Unhandled FLAC subframe type" };
}
for (size_t i = 0; i < samples.size(); ++i) {
samples[i] <<= subframe_header.wasted_bits_per_sample;
}
ResampleHelper<i32> resampler(m_current_frame->sample_rate, m_sample_rate);
return resampler.resample(samples);
}
// 11.29. SUBFRAME_VERBATIM
// Decode a subframe that isn't actually encoded, usually seen in random data
ErrorOr<Vector<i32>, LoaderError> FlacLoaderPlugin::decode_verbatim(FlacSubframeHeader& subframe, BigEndianInputBitStream& bit_input)
{
Vector<i32> decoded;
decoded.ensure_capacity(m_current_frame->sample_count);
VERIFY(subframe.bits_per_sample - subframe.wasted_bits_per_sample != 0);
for (size_t i = 0; i < m_current_frame->sample_count; ++i) {
decoded.unchecked_append(sign_extend(
LOADER_TRY(bit_input.read_bits<u32>(subframe.bits_per_sample - subframe.wasted_bits_per_sample)),
subframe.bits_per_sample - subframe.wasted_bits_per_sample));
}
return decoded;
}
// 11.28. SUBFRAME_LPC
// Decode a subframe encoded with a custom linear predictor coding, i.e. the subframe provides the polynomial order and coefficients
ErrorOr<Vector<i32>, LoaderError> FlacLoaderPlugin::decode_custom_lpc(FlacSubframeHeader& subframe, BigEndianInputBitStream& bit_input)
{
Vector<i32> decoded;
decoded.ensure_capacity(m_current_frame->sample_count);
VERIFY(subframe.bits_per_sample - subframe.wasted_bits_per_sample != 0);
// warm-up samples
for (auto i = 0; i < subframe.order; ++i) {
decoded.unchecked_append(sign_extend(
LOADER_TRY(bit_input.read_bits<u32>(subframe.bits_per_sample - subframe.wasted_bits_per_sample)),
subframe.bits_per_sample - subframe.wasted_bits_per_sample));
}
// precision of the coefficients
u8 lpc_precision = LOADER_TRY(bit_input.read_bits<u8>(4));
if (lpc_precision == 0b1111)
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), "Invalid linear predictor coefficient precision" };
lpc_precision += 1;
// shift needed on the data (signed!)
i8 lpc_shift = sign_extend(LOADER_TRY(bit_input.read_bits<u8>(5)), 5);
Vector<i32> coefficients;
coefficients.ensure_capacity(subframe.order);
// read coefficients
for (auto i = 0; i < subframe.order; ++i) {
u32 raw_coefficient = LOADER_TRY(bit_input.read_bits<u32>(lpc_precision));
i32 coefficient = static_cast<i32>(sign_extend(raw_coefficient, lpc_precision));
coefficients.unchecked_append(coefficient);
}
dbgln_if(AFLACLOADER_DEBUG, "{}-bit {} shift coefficients: {}", lpc_precision, lpc_shift, coefficients);
TRY(decode_residual(decoded, subframe, bit_input));
// approximate the waveform with the predictor
for (size_t i = subframe.order; i < m_current_frame->sample_count; ++i) {
// (see below)
i64 sample = 0;
for (size_t t = 0; t < subframe.order; ++t) {
// It's really important that we compute in 64-bit land here.
// Even though FLAC operates at a maximum bit depth of 32 bits, modern encoders use super-large coefficients for maximum compression.
// These will easily overflow 32 bits and cause strange white noise that abruptly stops intermittently (at the end of a frame).
// The simple fix of course is to do intermediate computations in 64 bits.
// These considerations are not in the original FLAC spec, but have been added to the IETF standard: https://datatracker.ietf.org/doc/html/draft-ietf-cellar-flac-03#appendix-A.3
sample += static_cast<i64>(coefficients[t]) * static_cast<i64>(decoded[i - t - 1]);
}
decoded[i] += sample >> lpc_shift;
}
return decoded;
}
// 11.27. SUBFRAME_FIXED
// Decode a subframe encoded with one of the fixed linear predictor codings
ErrorOr<Vector<i32>, LoaderError> FlacLoaderPlugin::decode_fixed_lpc(FlacSubframeHeader& subframe, BigEndianInputBitStream& bit_input)
{
Vector<i32> decoded;
decoded.ensure_capacity(m_current_frame->sample_count);
VERIFY(subframe.bits_per_sample - subframe.wasted_bits_per_sample != 0);
// warm-up samples
for (auto i = 0; i < subframe.order; ++i) {
decoded.unchecked_append(sign_extend(
LOADER_TRY(bit_input.read_bits<u32>(subframe.bits_per_sample - subframe.wasted_bits_per_sample)),
subframe.bits_per_sample - subframe.wasted_bits_per_sample));
}
TRY(decode_residual(decoded, subframe, bit_input));
dbgln_if(AFLACLOADER_DEBUG, "decoded length {}, {} order predictor", decoded.size(), subframe.order);
// Skip these comments if you don't care about the neat math behind fixed LPC :^)
// These coefficients for the recursive prediction formula are the only ones that can be resolved to polynomial predictor functions.
// The order equals the degree of the polynomial - 1, so the second-order predictor has an underlying polynomial of degree 1, a straight line.
// More specifically, the closest approximation to a polynomial is used, and the degree depends on how many previous values are available.
// This makes use of a very neat property of polynomials, which is that they are entirely characterized by their finitely many derivatives.
// (Mathematically speaking, the infinite Taylor series of any polynomial equals the polynomial itself.)
// Now remember that derivation is just the slope of the function, which is the same as the difference of two close-by values.
// Therefore, with two samples we can calculate the first derivative at a sample via the difference, which gives us a polynomial of degree 1.
// With three samples, we can do the same but also calculate the second derivative via the difference in the first derivatives.
// This gives us a polynomial of degree 2, as it has two "proper" (non-constant) derivatives.
// This can be continued for higher-order derivatives when we have more coefficients, giving us higher-order polynomials.
// In essence, it's akin to a Lagrangian polynomial interpolation for every sample (but already pre-solved).
// The coefficients for orders 0-3 originate from the SHORTEN codec:
// http://mi.eng.cam.ac.uk/reports/svr-ftp/auto-pdf/robinson_tr156.pdf page 4
// The coefficients for order 4 are undocumented in the original FLAC specification(s), but can now be found in
// https://datatracker.ietf.org/doc/html/draft-ietf-cellar-flac-03#section-10.2.5
switch (subframe.order) {
case 0:
// s_0(t) = 0
for (u32 i = subframe.order; i < m_current_frame->sample_count; ++i)
decoded[i] += 0;
break;
case 1:
// s_1(t) = s(t-1)
for (u32 i = subframe.order; i < m_current_frame->sample_count; ++i)
decoded[i] += decoded[i - 1];
break;
case 2:
// s_2(t) = 2s(t-1) - s(t-2)
for (u32 i = subframe.order; i < m_current_frame->sample_count; ++i)
decoded[i] += 2 * decoded[i - 1] - decoded[i - 2];
break;
case 3:
// s_3(t) = 3s(t-1) - 3s(t-2) + s(t-3)
for (u32 i = subframe.order; i < m_current_frame->sample_count; ++i)
decoded[i] += 3 * decoded[i - 1] - 3 * decoded[i - 2] + decoded[i - 3];
break;
case 4:
// s_4(t) = 4s(t-1) - 6s(t-2) + 4s(t-3) - s(t-4)
for (u32 i = subframe.order; i < m_current_frame->sample_count; ++i)
decoded[i] += 4 * decoded[i - 1] - 6 * decoded[i - 2] + 4 * decoded[i - 3] - decoded[i - 4];
break;
default:
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), DeprecatedString::formatted("Unrecognized predictor order {}", subframe.order) };
}
return decoded;
}
// 11.30. RESIDUAL
// Decode the residual, the "error" between the function approximation and the actual audio data
MaybeLoaderError FlacLoaderPlugin::decode_residual(Vector<i32>& decoded, FlacSubframeHeader& subframe, BigEndianInputBitStream& bit_input)
{
// 11.30.1. RESIDUAL_CODING_METHOD
auto residual_mode = static_cast<FlacResidualMode>(LOADER_TRY(bit_input.read_bits<u8>(2)));
u8 partition_order = LOADER_TRY(bit_input.read_bits<u8>(4));
size_t partitions = 1 << partition_order;
if (residual_mode == FlacResidualMode::Rice4Bit) {
// 11.30.2. RESIDUAL_CODING_METHOD_PARTITIONED_EXP_GOLOMB
// decode a single Rice partition with four bits for the order k
for (size_t i = 0; i < partitions; ++i) {
auto rice_partition = TRY(decode_rice_partition(4, partitions, i, subframe, bit_input));
decoded.extend(move(rice_partition));
}
} else if (residual_mode == FlacResidualMode::Rice5Bit) {
// 11.30.3. RESIDUAL_CODING_METHOD_PARTITIONED_EXP_GOLOMB2
// five bits equivalent
for (size_t i = 0; i < partitions; ++i) {
auto rice_partition = TRY(decode_rice_partition(5, partitions, i, subframe, bit_input));
decoded.extend(move(rice_partition));
}
} else
return LoaderError { LoaderError::Category::Format, static_cast<size_t>(m_current_sample_or_frame), "Reserved residual coding method" };
return {};
}
// 11.30.2.1. EXP_GOLOMB_PARTITION and 11.30.3.1. EXP_GOLOMB2_PARTITION
// Decode a single Rice partition as part of the residual, every partition can have its own Rice parameter k
ALWAYS_INLINE ErrorOr<Vector<i32>, LoaderError> FlacLoaderPlugin::decode_rice_partition(u8 partition_type, u32 partitions, u32 partition_index, FlacSubframeHeader& subframe, BigEndianInputBitStream& bit_input)
{
// 11.30.2.2. EXP GOLOMB PARTITION ENCODING PARAMETER and 11.30.3.2. EXP-GOLOMB2 PARTITION ENCODING PARAMETER
u8 k = LOADER_TRY(bit_input.read_bits<u8>(partition_type));
u32 residual_sample_count;
if (partitions == 0)
residual_sample_count = m_current_frame->sample_count - subframe.order;
else
residual_sample_count = m_current_frame->sample_count / partitions;
if (partition_index == 0)
residual_sample_count -= subframe.order;
Vector<i32> rice_partition;
rice_partition.resize(residual_sample_count);
// escape code for unencoded binary partition
if (k == (1 << partition_type) - 1) {
u8 unencoded_bps = LOADER_TRY(bit_input.read_bits<u8>(5));
for (size_t r = 0; r < residual_sample_count; ++r) {
rice_partition[r] = LOADER_TRY(bit_input.read_bits<u8>(unencoded_bps));
}
} else {
for (size_t r = 0; r < residual_sample_count; ++r) {
rice_partition[r] = LOADER_TRY(decode_unsigned_exp_golomb(k, bit_input));
}
}
return rice_partition;
}
// Decode a single number encoded with Rice/Exponential-Golomb encoding (the unsigned variant)
ALWAYS_INLINE ErrorOr<i32> decode_unsigned_exp_golomb(u8 k, BigEndianInputBitStream& bit_input)
{
u8 q = 0;
while (TRY(bit_input.read_bit()) == 0)
++q;
// least significant bits (remainder)
u32 rem = TRY(bit_input.read_bits<u32>(k));
u32 value = q << k | rem;
return rice_to_signed(value);
}
ErrorOr<u64> read_utf8_char(BigEndianInputBitStream& input)
{
u64 character;
u8 buffer = 0;
Bytes buffer_bytes { &buffer, 1 };
TRY(input.read(buffer_bytes));
u8 start_byte = buffer_bytes[0];
// Signal byte is zero: ASCII character
if ((start_byte & 0b10000000) == 0) {
return start_byte;
} else if ((start_byte & 0b11000000) == 0b10000000) {
return Error::from_string_literal("Illegal continuation byte");
}
// This algorithm is too good and supports the theoretical max 0xFF start byte
u8 length = 1;
while (((start_byte << length) & 0b10000000) == 0b10000000)
++length;
u8 bits_from_start_byte = 8 - (length + 1);
u8 start_byte_bitmask = AK::exp2(bits_from_start_byte) - 1;
character = start_byte_bitmask & start_byte;
for (u8 i = length - 1; i > 0; --i) {
TRY(input.read(buffer_bytes));
u8 current_byte = buffer_bytes[0];
character = (character << 6) | (current_byte & 0b00111111);
}
return character;
}
i64 sign_extend(u32 n, u8 size)
{
// negative
if ((n & (1 << (size - 1))) > 0) {
return static_cast<i64>(n | (0xffffffff << size));
}
// positive
return n;
}
i32 rice_to_signed(u32 x)
{
// positive numbers are even, negative numbers are odd
// bitmask for conditionally inverting the entire number, thereby "negating" it
i32 sign = -static_cast<i32>(x & 1);
// copies the sign's sign onto the actual magnitude of x
return static_cast<i32>(sign ^ (x >> 1));
}
}
Reference in a new issue View git blame Copy permalink