mirrors/ladybird
1
0
Fork 0
mirror of https://github.com/LadybirdBrowser/ladybird.git synced 2025-01-23 09:46:04 -05:00
Code Issues Projects Releases Packages Wiki Activity
ladybird/Userland/Libraries/LibVideo/VP9/Parser.cpp

1782 lines
85 KiB
C++
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

/*
* Copyright (c) 2021, Hunter Salyer <thefalsehonesty@gmail.com>
* Copyright (c) 2022, Gregory Bertilson <zaggy1024@gmail.com>
*
* SPDX-License-Identifier: BSD-2-Clause
*/
#include <AK/MemoryStream.h>
#include <LibGfx/Point.h>
#include <LibGfx/Size.h>
#include <LibThreading/WorkerThread.h>
#include "Context.h"
#include "Decoder.h"
#include "Parser.h"
#include "Utilities.h"
#if defined(AK_COMPILER_GCC)
# pragma GCC optimize("O3")
#endif
// Beware, threading is unstable in Serenity with smp=on, and performs worse than with it off.
#define VP9_TILE_THREADING
namespace Video::VP9 {
#define TRY_READ(expression) DECODER_TRY(DecoderErrorCategory::Corrupted, expression)
Parser::Parser(Decoder& decoder)
: m_decoder(decoder)
{
}
Parser::~Parser()
{
}
Vector<size_t> Parser::parse_superframe_sizes(ReadonlyBytes frame_data)
{
if (frame_data.size() < 1)
return {};
// The decoder determines the presence of a superframe by:
// 1. parsing the final byte of the chunk and checking that the superframe_marker equals 0b110,
// If the checks in steps 1 and 3 both pass, then the chunk is determined to contain a superframe and each
// frame in the superframe is passed to the decoding process in turn.
// Otherwise, the chunk is determined to not contain a superframe, and the whole chunk is passed to the
// decoding process.
// NOTE: Reading from span data will be quicker than spinning up a BitStream.
u8 superframe_byte = frame_data[frame_data.size() - 1];
// NOTE: We have to read out of the byte from the little end first, hence the padding bits in the masks below.
u8 superframe_marker = superframe_byte & 0b1110'0000;
if (superframe_marker == 0b1100'0000) {
u8 bytes_per_framesize = ((superframe_byte >> 3) & 0b11) + 1;
u8 frames_in_superframe = (superframe_byte & 0b111) + 1;
// 2. setting the total size of the superframe_index SzIndex equal to 2 + NumFrames * SzBytes,
size_t index_size = 2 + bytes_per_framesize * frames_in_superframe;
if (index_size > frame_data.size())
return {};
auto superframe_header_data = frame_data.data() + frame_data.size() - index_size;
u8 start_superframe_byte = *(superframe_header_data++);
// 3. checking that the first byte of the superframe_index matches the final byte.
if (superframe_byte != start_superframe_byte)
return {};
Vector<size_t> result;
for (u8 i = 0; i < frames_in_superframe; i++) {
size_t frame_size = 0;
for (u8 j = 0; j < bytes_per_framesize; j++)
frame_size |= (static_cast<size_t>(*(superframe_header_data++)) << (j * 8));
result.append(frame_size);
}
return result;
}
return {};
}
/* (6.1) */
DecoderErrorOr<FrameContext> Parser::parse_frame(ReadonlyBytes frame_data)
{
if (!m_probability_tables)
m_probability_tables = DECODER_TRY_ALLOC(try_make<ProbabilityTables>());
// NOTE: m_reusable_frame_block_contexts does not need to retain any data between frame decodes.
// This is only stored so that we don't need to allocate a frame's block contexts on each
// call to this function, since it will rarely change sizes.
auto frame_context = DECODER_TRY_ALLOC(FrameContext::create(frame_data, m_reusable_frame_block_contexts));
TRY(uncompressed_header(frame_context));
// FIXME: This should not be an error. Spec says that we consume padding bits until the end of the sample.
if (frame_context.header_size_in_bytes == 0)
return DecoderError::corrupted("Frame header is zero-sized"sv);
m_probability_tables->load_probs(frame_context.probability_context_index);
m_probability_tables->load_probs2(frame_context.probability_context_index);
TRY(compressed_header(frame_context));
TRY(m_decoder.allocate_buffers(frame_context));
TRY(decode_tiles(frame_context));
TRY(refresh_probs(frame_context));
m_previous_frame_type = frame_context.type;
m_previous_frame_size = frame_context.size();
m_previous_show_frame = frame_context.shows_a_frame();
m_previous_color_config = frame_context.color_config;
m_previous_loop_filter_ref_deltas = frame_context.loop_filter_reference_deltas;
m_previous_loop_filter_mode_deltas = frame_context.loop_filter_mode_deltas;
if (frame_context.segmentation_enabled) {
m_previous_should_use_absolute_segment_base_quantizer = frame_context.should_use_absolute_segment_base_quantizer;
m_previous_segmentation_features = frame_context.segmentation_features;
}
return frame_context;
}
DecoderErrorOr<void> Parser::refresh_probs(FrameContext const& frame_context)
{
if (!frame_context.error_resilient_mode && !frame_context.parallel_decoding_mode) {
m_probability_tables->load_probs(frame_context.probability_context_index);
TRY(m_decoder.adapt_coef_probs(frame_context));
if (frame_context.is_inter_predicted()) {
m_probability_tables->load_probs2(frame_context.probability_context_index);
TRY(m_decoder.adapt_non_coef_probs(frame_context));
}
}
if (frame_context.should_replace_probability_context)
m_probability_tables->save_probs(frame_context.probability_context_index);
return {};
}
DecoderErrorOr<VideoFullRangeFlag> Parser::read_video_full_range_flag(BigEndianInputBitStream& bit_stream)
{
if (TRY_READ(bit_stream.read_bit()))
return VideoFullRangeFlag::Full;
return VideoFullRangeFlag::Studio;
}
template<Signed T = i8>
static ErrorOr<T> read_signed(BigEndianInputBitStream& bit_stream, u8 bits)
{
auto value_unsigned = static_cast<T>(TRY(bit_stream.read_bits(bits)));
if (TRY(bit_stream.read_bit()))
return -value_unsigned;
return value_unsigned;
}
static DecoderErrorOr<i8> read_delta_q(BigEndianInputBitStream& bit_stream)
{
if (TRY_READ(bit_stream.read_bit()))
return TRY_READ(read_signed(bit_stream, 4));
return 0;
}
struct QuantizationParameters {
u8 base_quantizer_index { 0 };
i8 y_dc_quantizer_index_delta { 0 };
i8 uv_dc_quantizer_index_delta { 0 };
i8 uv_ac_quantizer_index_delta { 0 };
};
static DecoderErrorOr<QuantizationParameters> quantization_params(BigEndianInputBitStream& bit_stream)
{
QuantizationParameters result;
result.base_quantizer_index = TRY_READ(bit_stream.read_bits(8));
result.y_dc_quantizer_index_delta = TRY(read_delta_q(bit_stream));
result.uv_dc_quantizer_index_delta = TRY(read_delta_q(bit_stream));
result.uv_ac_quantizer_index_delta = TRY(read_delta_q(bit_stream));
return result;
}
/* (6.2) */
DecoderErrorOr<void> Parser::uncompressed_header(FrameContext& frame_context)
{
frame_context.color_config = m_previous_color_config;
auto frame_marker = TRY_READ(frame_context.bit_stream.read_bits(2));
if (frame_marker != 2)
return DecoderError::corrupted("uncompressed_header: Frame marker must be 2"sv);
auto profile_low_bit = TRY_READ(frame_context.bit_stream.read_bit());
auto profile_high_bit = TRY_READ(frame_context.bit_stream.read_bit());
frame_context.profile = (profile_high_bit << 1u) + profile_low_bit;
if (frame_context.profile == 3 && TRY_READ(frame_context.bit_stream.read_bit()))
return DecoderError::corrupted("uncompressed_header: Profile 3 reserved bit was non-zero"sv);
if (TRY_READ(frame_context.bit_stream.read_bit())) {
frame_context.set_existing_frame_to_show(TRY_READ(frame_context.bit_stream.read_bits(3)));
return {};
}
bool is_keyframe = !TRY_READ(frame_context.bit_stream.read_bit());
if (!TRY_READ(frame_context.bit_stream.read_bit()))
frame_context.set_frame_hidden();
frame_context.error_resilient_mode = TRY_READ(frame_context.bit_stream.read_bit());
FrameType type;
Gfx::Size<u32> frame_size;
Gfx::Size<u32> render_size;
u8 reference_frames_to_update_flags = 0xFF; // Save frame to all reference indices by default.
enum class ResetProbabilities : u8 {
No = 0,
// 1 also means No here, but we don't need to do anything with the No case.
OnlyCurrent = 2,
All = 3,
};
ResetProbabilities reset_frame_context = ResetProbabilities::All;
if (is_keyframe) {
type = FrameType::KeyFrame;
TRY(frame_sync_code(frame_context.bit_stream));
frame_context.color_config = TRY(parse_color_config(frame_context.bit_stream, frame_context.profile));
frame_size = TRY(parse_frame_size(frame_context.bit_stream));
render_size = TRY(parse_render_size(frame_context.bit_stream, frame_size));
} else {
if (!frame_context.shows_a_frame() && TRY_READ(frame_context.bit_stream.read_bit())) {
type = FrameType::IntraOnlyFrame;
} else {
type = FrameType::InterFrame;
reset_frame_context = ResetProbabilities::No;
}
if (!frame_context.error_resilient_mode)
reset_frame_context = static_cast<ResetProbabilities>(TRY_READ(frame_context.bit_stream.read_bits(2)));
if (type == FrameType::IntraOnlyFrame) {
TRY(frame_sync_code(frame_context.bit_stream));
if (frame_context.profile == 0) {
frame_context.color_config = ColorConfig();
} else {
frame_context.color_config = TRY(parse_color_config(frame_context.bit_stream, frame_context.profile));
}
reference_frames_to_update_flags = TRY_READ(frame_context.bit_stream.read_bits(8));
frame_size = TRY(parse_frame_size(frame_context.bit_stream));
render_size = TRY(parse_render_size(frame_context.bit_stream, frame_size));
} else {
reference_frames_to_update_flags = TRY_READ(frame_context.bit_stream.read_bits(NUM_REF_FRAMES));
for (auto i = 0; i < REFS_PER_FRAME; i++) {
frame_context.reference_frame_indices[i] = TRY_READ(frame_context.bit_stream.read_bits(LOG2_OF_NUM_REF_FRAMES));
frame_context.reference_frame_sign_biases[ReferenceFrameType::LastFrame + i] = TRY_READ(frame_context.bit_stream.read_bit());
}
frame_size = TRY(parse_frame_size_with_refs(frame_context.bit_stream, frame_context.reference_frame_indices));
render_size = TRY(parse_render_size(frame_context.bit_stream, frame_size));
frame_context.high_precision_motion_vectors_allowed = TRY_READ(frame_context.bit_stream.read_bit());
frame_context.interpolation_filter = TRY(read_interpolation_filter(frame_context.bit_stream));
for (auto i = 0; i < REFS_PER_FRAME; i++) {
TRY(m_decoder.prepare_referenced_frame(frame_size, frame_context.reference_frame_indices[i]));
}
}
}
bool should_replace_probability_context = false;
bool parallel_decoding_mode = true;
if (!frame_context.error_resilient_mode) {
should_replace_probability_context = TRY_READ(frame_context.bit_stream.read_bit());
parallel_decoding_mode = TRY_READ(frame_context.bit_stream.read_bit());
}
u8 probability_context_index = TRY_READ(frame_context.bit_stream.read_bits(2));
switch (reset_frame_context) {
case ResetProbabilities::All:
setup_past_independence();
for (auto i = 0; i < 4; i++) {
m_probability_tables->save_probs(i);
}
probability_context_index = 0;
break;
case ResetProbabilities::OnlyCurrent:
setup_past_independence();
m_probability_tables->save_probs(probability_context_index);
probability_context_index = 0;
break;
default:
break;
}
frame_context.type = type;
DECODER_TRY_ALLOC(frame_context.set_size(frame_size));
frame_context.render_size = render_size;
TRY(compute_image_size(frame_context));
frame_context.reference_frames_to_update_flags = reference_frames_to_update_flags;
frame_context.parallel_decoding_mode = parallel_decoding_mode;
frame_context.should_replace_probability_context = should_replace_probability_context;
frame_context.probability_context_index = probability_context_index;
TRY(loop_filter_params(frame_context));
auto quantization_parameters = TRY(quantization_params(frame_context.bit_stream));
TRY(segmentation_params(frame_context));
precalculate_quantizers(frame_context, quantization_parameters);
TRY(parse_tile_counts(frame_context));
frame_context.header_size_in_bytes = TRY_READ(frame_context.bit_stream.read_bits(16));
frame_context.bit_stream.align_to_byte_boundary();
return {};
}
DecoderErrorOr<void> Parser::frame_sync_code(BigEndianInputBitStream& bit_stream)
{
if (TRY_READ(bit_stream.read_bits(24)) != 0x498342) {
return DecoderError::corrupted("frame sync code was not 0x498342."sv);
}
return {};
}
DecoderErrorOr<ColorConfig> Parser::parse_color_config(BigEndianInputBitStream& bit_stream, u8 profile)
{
// (6.2.2) color_config( )
u8 bit_depth;
if (profile >= 2) {
bit_depth = TRY_READ(bit_stream.read_bit()) ? 12 : 10;
} else {
bit_depth = 8;
}
auto color_space = static_cast<ColorSpace>(TRY_READ(bit_stream.read_bits(3)));
if (color_space == ColorSpace::Reserved)
return DecoderError::corrupted("color_config: Color space reserved value was set"sv);
VERIFY(color_space <= ColorSpace::RGB);
VideoFullRangeFlag video_full_range_flag;
bool subsampling_x, subsampling_y;
if (color_space != ColorSpace::RGB) {
video_full_range_flag = TRY(read_video_full_range_flag(bit_stream));
if (profile == 1 || profile == 3) {
subsampling_x = TRY_READ(bit_stream.read_bit());
subsampling_y = TRY_READ(bit_stream.read_bit());
if (TRY_READ(bit_stream.read_bit()))
return DecoderError::corrupted("color_config: Subsampling reserved zero was set"sv);
} else {
subsampling_x = true;
subsampling_y = true;
}
} else {
video_full_range_flag = VideoFullRangeFlag::Full;
if (profile == 1 || profile == 3) {
subsampling_x = false;
subsampling_y = false;
if (TRY_READ(bit_stream.read_bit()))
return DecoderError::corrupted("color_config: RGB reserved zero was set"sv);
} else {
// FIXME: Spec does not specify the subsampling value here. Is this an error or should we set a default?
return DecoderError::corrupted("color_config: Invalid subsampling value for profile 0 or 2"sv);
}
}
return ColorConfig { bit_depth, color_space, video_full_range_flag, subsampling_x, subsampling_y };
}
DecoderErrorOr<Gfx::Size<u32>> Parser::parse_frame_size(BigEndianInputBitStream& bit_stream)
{
return Gfx::Size<u32> { TRY_READ(bit_stream.read_bits(16)) + 1, TRY_READ(bit_stream.read_bits(16)) + 1 };
}
DecoderErrorOr<Gfx::Size<u32>> Parser::parse_render_size(BigEndianInputBitStream& bit_stream, Gfx::Size<u32> frame_size)
{
// FIXME: This function should save this bit as a value in the FrameContext. The bit can be
// used in files where the pixel aspect ratio changes between samples in the video.
// If the bit is set, the pixel aspect ratio should be recalculated, whereas if only
// the frame size has changed and the render size is unadjusted, then the pixel aspect
// ratio should be retained and the new render size determined based on that.
// See the Firefox source code here:
// https://searchfox.org/mozilla-central/source/dom/media/platforms/wrappers/MediaChangeMonitor.cpp#268-276
if (!TRY_READ(bit_stream.read_bit()))
return frame_size;
return Gfx::Size<u32> { TRY_READ(bit_stream.read_bits(16)) + 1, TRY_READ(bit_stream.read_bits(16)) + 1 };
}
DecoderErrorOr<Gfx::Size<u32>> Parser::parse_frame_size_with_refs(BigEndianInputBitStream& bit_stream, Array<u8, 3> const& reference_indices)
{
Optional<Gfx::Size<u32>> size;
for (auto frame_index : reference_indices) {
if (TRY_READ(bit_stream.read_bit())) {
if (!m_reference_frames[frame_index].is_valid())
return DecoderError::corrupted("Frame size referenced a frame that does not exist"sv);
size.emplace(m_reference_frames[frame_index].size);
break;
}
}
if (size.has_value())
return size.value();
return TRY(parse_frame_size(bit_stream));
}
DecoderErrorOr<void> Parser::compute_image_size(FrameContext& frame_context)
{
// 7.2.6 Compute image size semantics
// When compute_image_size is invoked, the following ordered steps occur:
// 1. If this is the first time compute_image_size is invoked, or if either FrameWidth or FrameHeight have
// changed in value compared to the previous time this function was invoked, then the segmentation map is
// cleared to all zeros by setting SegmentId[ row ][ col ] equal to 0 for row = 0..MiRows-1 and col =
// 0..MiCols-1.
// FIXME: What does this mean? SegmentIds is scoped to one frame, so it will not contain values here. It's
// also suspicious that spec refers to this as SegmentId rather than SegmentIds (plural). Is this
// supposed to refer to PrevSegmentIds?
bool first_invoke = m_is_first_compute_image_size_invoke;
m_is_first_compute_image_size_invoke = false;
bool same_size = m_previous_frame_size == frame_context.size();
// 2. The variable UsePrevFrameMvs is set equal to 1 if all of the following conditions are true:
// a. This is not the first time compute_image_size is invoked.
// b. Both FrameWidth and FrameHeight have the same value compared to the previous time this function
// was invoked.
// c. show_frame was equal to 1 the previous time this function was invoked.
// d. error_resilient_mode is equal to 0.
// e. FrameIsIntra is equal to 0.
// Otherwise, UsePrevFrameMvs is set equal to 0.
frame_context.use_previous_frame_motion_vectors = !first_invoke && same_size && m_previous_show_frame && !frame_context.error_resilient_mode && frame_context.is_inter_predicted();
return {};
}
DecoderErrorOr<InterpolationFilter> Parser::read_interpolation_filter(BigEndianInputBitStream& bit_stream)
{
if (TRY_READ(bit_stream.read_bit())) {
return InterpolationFilter::Switchable;
}
return literal_to_type[TRY_READ(bit_stream.read_bits(2))];
}
DecoderErrorOr<void> Parser::loop_filter_params(FrameContext& frame_context)
{
// FIXME: These should be moved to their own struct to return here.
frame_context.loop_filter_level = TRY_READ(frame_context.bit_stream.read_bits(6));
frame_context.loop_filter_sharpness = TRY_READ(frame_context.bit_stream.read_bits(3));
frame_context.loop_filter_delta_enabled = TRY_READ(frame_context.bit_stream.read_bit());
auto reference_deltas = m_previous_loop_filter_ref_deltas;
auto mode_deltas = m_previous_loop_filter_mode_deltas;
if (frame_context.loop_filter_delta_enabled && TRY_READ(frame_context.bit_stream.read_bit())) {
for (auto& loop_filter_ref_delta : reference_deltas) {
if (TRY_READ(frame_context.bit_stream.read_bit()))
loop_filter_ref_delta = TRY_READ(read_signed(frame_context.bit_stream, 6));
}
for (auto& loop_filter_mode_delta : mode_deltas) {
if (TRY_READ(frame_context.bit_stream.read_bit()))
loop_filter_mode_delta = TRY_READ(read_signed(frame_context.bit_stream, 6));
}
}
frame_context.loop_filter_reference_deltas = reference_deltas;
frame_context.loop_filter_mode_deltas = mode_deltas;
return {};
}
DecoderErrorOr<void> Parser::segmentation_params(FrameContext& frame_context)
{
frame_context.segmentation_enabled = TRY_READ(frame_context.bit_stream.read_bit());
if (!frame_context.segmentation_enabled)
return {};
frame_context.should_use_absolute_segment_base_quantizer = m_previous_should_use_absolute_segment_base_quantizer;
frame_context.segmentation_features = m_previous_segmentation_features;
if (TRY_READ(frame_context.bit_stream.read_bit())) {
frame_context.use_full_segment_id_tree = true;
for (auto& segmentation_tree_prob : frame_context.full_segment_id_tree_probabilities)
segmentation_tree_prob = TRY(read_prob(frame_context.bit_stream));
if (TRY_READ(frame_context.bit_stream.read_bit())) {
frame_context.use_predicted_segment_id_tree = true;
for (auto& segmentation_pred_prob : frame_context.predicted_segment_id_tree_probabilities)
segmentation_pred_prob = TRY(read_prob(frame_context.bit_stream));
}
}
auto segmentation_update_data = (TRY_READ(frame_context.bit_stream.read_bit()));
if (!segmentation_update_data)
return {};
frame_context.should_use_absolute_segment_base_quantizer = TRY_READ(frame_context.bit_stream.read_bit());
for (auto segment_id = 0; segment_id < MAX_SEGMENTS; segment_id++) {
for (auto feature_id = 0; feature_id < to_underlying(SegmentFeature::Sentinel); feature_id++) {
auto& feature = frame_context.segmentation_features[segment_id][feature_id];
feature.enabled = TRY_READ(frame_context.bit_stream.read_bit());
if (feature.enabled) {
auto bits_to_read = segmentation_feature_bits[feature_id];
feature.value = TRY_READ(frame_context.bit_stream.read_bits(bits_to_read));
if (segmentation_feature_signed[feature_id]) {
if (TRY_READ(frame_context.bit_stream.read_bit()))
feature.value = -feature.value;
}
}
}
}
return {};
}
DecoderErrorOr<u8> Parser::read_prob(BigEndianInputBitStream& bit_stream)
{
if (TRY_READ(bit_stream.read_bit()))
return TRY_READ(bit_stream.read_bits(8));
return 255;
}
void Parser::precalculate_quantizers(FrameContext& frame_context, QuantizationParameters quantization_parameters)
{
frame_context.lossless = quantization_parameters.base_quantizer_index == 0
&& quantization_parameters.y_dc_quantizer_index_delta == 0
&& quantization_parameters.uv_dc_quantizer_index_delta == 0
&& quantization_parameters.uv_ac_quantizer_index_delta == 0;
// Pre-calculate the quantizers so that the decoder doesn't have to do it repeatedly.
for (u8 segment_id = 0; segment_id < MAX_SEGMENTS; segment_id++) {
auto alternative_quantizer_feature = frame_context.get_segment_feature(segment_id, SegmentFeature::AlternativeQuantizerBase);
auto base = Decoder::get_base_quantizer_index(alternative_quantizer_feature, frame_context.should_use_absolute_segment_base_quantizer, quantization_parameters.base_quantizer_index);
// The function get_ac_quant( plane ) returns the quantizer value for the ac coefficient for a particular plane and
// is derived as follows:
// − If plane is equal to 0, return ac_q( get_qindex( ) ).
// − Otherwise, return ac_q( get_qindex( ) + delta_q_uv_ac ).
auto& current_quantizers = frame_context.segment_quantizers[segment_id];
current_quantizers.y_ac_quantizer = Decoder::get_ac_quantizer(frame_context.color_config.bit_depth, base, 0);
current_quantizers.uv_ac_quantizer = Decoder::get_ac_quantizer(frame_context.color_config.bit_depth, base, quantization_parameters.uv_ac_quantizer_index_delta);
// The function get_dc_quant( plane ) returns the quantizer value for the dc coefficient for a particular plane and
// is derived as follows:
// − If plane is equal to 0, return dc_q( get_qindex( ) + delta_q_y_dc ).
// − Otherwise, return dc_q( get_qindex( ) + delta_q_uv_dc ).
current_quantizers.y_dc_quantizer = Decoder::get_dc_quantizer(frame_context.color_config.bit_depth, base, quantization_parameters.y_dc_quantizer_index_delta);
current_quantizers.uv_dc_quantizer = Decoder::get_dc_quantizer(frame_context.color_config.bit_depth, base, quantization_parameters.uv_dc_quantizer_index_delta);
}
}
static u16 calc_min_log2_of_tile_columns(u32 superblock_columns)
{
auto min_log_2 = 0u;
while ((u32)(MAX_TILE_WIDTH_B64 << min_log_2) < superblock_columns)
min_log_2++;
return min_log_2;
}
static u16 calc_max_log2_tile_cols(u32 superblock_columns)
{
u16 max_log_2 = 1;
while ((superblock_columns >> max_log_2) >= MIN_TILE_WIDTH_B64)
max_log_2++;
return max_log_2 - 1;
}
DecoderErrorOr<void> Parser::parse_tile_counts(FrameContext& frame_context)
{
auto superblock_columns = frame_context.superblock_columns();
auto log2_of_tile_columns = calc_min_log2_of_tile_columns(superblock_columns);
auto log2_of_tile_columns_maximum = calc_max_log2_tile_cols(superblock_columns);
while (log2_of_tile_columns < log2_of_tile_columns_maximum) {
if (TRY_READ(frame_context.bit_stream.read_bit()))
log2_of_tile_columns++;
else
break;
}
u16 log2_of_tile_rows = TRY_READ(frame_context.bit_stream.read_bit());
if (log2_of_tile_rows > 0) {
log2_of_tile_rows += TRY_READ(frame_context.bit_stream.read_bit());
}
frame_context.log2_of_tile_counts = Gfx::Size<u16>(log2_of_tile_columns, log2_of_tile_rows);
return {};
}
void Parser::setup_past_independence()
{
m_previous_block_contexts.reset();
m_previous_loop_filter_ref_deltas[ReferenceFrameType::None] = 1;
m_previous_loop_filter_ref_deltas[ReferenceFrameType::LastFrame] = 0;
m_previous_loop_filter_ref_deltas[ReferenceFrameType::GoldenFrame] = -1;
m_previous_loop_filter_ref_deltas[ReferenceFrameType::AltRefFrame] = -1;
m_previous_loop_filter_mode_deltas.fill(0);
m_previous_should_use_absolute_segment_base_quantizer = false;
for (auto& segment_levels : m_previous_segmentation_features)
segment_levels.fill({ false, 0 });
m_probability_tables->reset_probs();
}
DecoderErrorOr<void> Parser::compressed_header(FrameContext& frame_context)
{
auto decoder = TRY(frame_context.create_range_decoder(frame_context.header_size_in_bytes));
frame_context.transform_mode = read_tx_mode(decoder, frame_context);
if (frame_context.transform_mode == TransformMode::Select)
tx_mode_probs(decoder);
read_coef_probs(decoder, frame_context.transform_mode);
read_skip_prob(decoder);
if (frame_context.is_inter_predicted()) {
read_inter_mode_probs(decoder);
if (frame_context.interpolation_filter == Switchable)
read_interp_filter_probs(decoder);
read_is_inter_probs(decoder);
frame_reference_mode(frame_context, decoder);
frame_reference_mode_probs(decoder, frame_context);
read_y_mode_probs(decoder);
read_partition_probs(decoder);
mv_probs(decoder, frame_context);
}
TRY_READ(decoder.finish_decode());
return {};
}
TransformMode Parser::read_tx_mode(BooleanDecoder& decoder, FrameContext const& frame_context)
{
if (frame_context.lossless) {
return TransformMode::Only_4x4;
}
auto tx_mode = decoder.read_literal(2);
if (tx_mode == to_underlying(TransformMode::Allow_32x32))
tx_mode += decoder.read_literal(1);
return static_cast<TransformMode>(tx_mode);
}
void Parser::tx_mode_probs(BooleanDecoder& decoder)
{
auto& tx_probs = m_probability_tables->tx_probs();
for (auto i = 0; i < TX_SIZE_CONTEXTS; i++) {
for (auto j = 0; j < TX_SIZES - 3; j++)
tx_probs[Transform_8x8][i][j] = diff_update_prob(decoder, tx_probs[Transform_8x8][i][j]);
}
for (auto i = 0; i < TX_SIZE_CONTEXTS; i++) {
for (auto j = 0; j < TX_SIZES - 2; j++)
tx_probs[Transform_16x16][i][j] = diff_update_prob(decoder, tx_probs[Transform_16x16][i][j]);
}
for (auto i = 0; i < TX_SIZE_CONTEXTS; i++) {
for (auto j = 0; j < TX_SIZES - 1; j++)
tx_probs[Transform_32x32][i][j] = diff_update_prob(decoder, tx_probs[Transform_32x32][i][j]);
}
}
u8 Parser::diff_update_prob(BooleanDecoder& decoder, u8 prob)
{
auto update_prob = decoder.read_bool(252);
if (update_prob) {
auto delta_prob = decode_term_subexp(decoder);
prob = inv_remap_prob(delta_prob, prob);
}
return prob;
}
u8 Parser::decode_term_subexp(BooleanDecoder& decoder)
{
if (decoder.read_literal(1) == 0)
return decoder.read_literal(4);
if (decoder.read_literal(1) == 0)
return decoder.read_literal(4) + 16;
if (decoder.read_literal(1) == 0)
return decoder.read_literal(5) + 32;
auto v = decoder.read_literal(7);
if (v < 65)
return v + 64;
return (v << 1u) - 1 + decoder.read_literal(1);
}
u8 Parser::inv_remap_prob(u8 delta_prob, u8 prob)
{
u8 m = prob - 1;
auto v = inv_map_table[delta_prob];
if ((m << 1u) <= 255)
return 1 + inv_recenter_nonneg(v, m);
return 255 - inv_recenter_nonneg(v, 254 - m);
}
u8 Parser::inv_recenter_nonneg(u8 v, u8 m)
{
if (v > 2 * m)
return v;
if (v & 1u)
return m - ((v + 1u) >> 1u);
return m + (v >> 1u);
}
void Parser::read_coef_probs(BooleanDecoder& decoder, TransformMode transform_mode)
{
auto max_tx_size = tx_mode_to_biggest_tx_size[to_underlying(transform_mode)];
for (u8 transform_size = 0; transform_size <= max_tx_size; transform_size++) {
auto update_probs = decoder.read_literal(1);
if (update_probs == 1) {
for (auto i = 0; i < 2; i++) {
for (auto j = 0; j < 2; j++) {
for (auto k = 0; k < 6; k++) {
auto max_l = (k == 0) ? 3 : 6;
for (auto l = 0; l < max_l; l++) {
for (auto m = 0; m < 3; m++) {
auto& prob = m_probability_tables->coef_probs()[transform_size][i][j][k][l][m];
prob = diff_update_prob(decoder, prob);
}
}
}
}
}
}
}
}
void Parser::read_skip_prob(BooleanDecoder& decoder)
{
for (auto i = 0; i < SKIP_CONTEXTS; i++)
m_probability_tables->skip_prob()[i] = diff_update_prob(decoder, m_probability_tables->skip_prob()[i]);
}
void Parser::read_inter_mode_probs(BooleanDecoder& decoder)
{
for (auto i = 0; i < INTER_MODE_CONTEXTS; i++) {
for (auto j = 0; j < INTER_MODES - 1; j++)
m_probability_tables->inter_mode_probs()[i][j] = diff_update_prob(decoder, m_probability_tables->inter_mode_probs()[i][j]);
}
}
void Parser::read_interp_filter_probs(BooleanDecoder& decoder)
{
for (auto i = 0; i < INTERP_FILTER_CONTEXTS; i++) {
for (auto j = 0; j < SWITCHABLE_FILTERS - 1; j++)
m_probability_tables->interp_filter_probs()[i][j] = diff_update_prob(decoder, m_probability_tables->interp_filter_probs()[i][j]);
}
}
void Parser::read_is_inter_probs(BooleanDecoder& decoder)
{
for (auto i = 0; i < IS_INTER_CONTEXTS; i++)
m_probability_tables->is_inter_prob()[i] = diff_update_prob(decoder, m_probability_tables->is_inter_prob()[i]);
}
static void setup_compound_reference_mode(FrameContext& frame_context)
{
ReferenceFrameType fixed_reference;
ReferenceFramePair variable_references;
if (frame_context.reference_frame_sign_biases[ReferenceFrameType::LastFrame] == frame_context.reference_frame_sign_biases[ReferenceFrameType::GoldenFrame]) {
fixed_reference = ReferenceFrameType::AltRefFrame;
variable_references = { ReferenceFrameType::LastFrame, ReferenceFrameType::GoldenFrame };
} else if (frame_context.reference_frame_sign_biases[ReferenceFrameType::LastFrame] == frame_context.reference_frame_sign_biases[ReferenceFrameType::AltRefFrame]) {
fixed_reference = ReferenceFrameType::GoldenFrame;
variable_references = { ReferenceFrameType::LastFrame, ReferenceFrameType::AltRefFrame };
} else {
fixed_reference = ReferenceFrameType::LastFrame;
variable_references = { ReferenceFrameType::GoldenFrame, ReferenceFrameType::AltRefFrame };
}
frame_context.fixed_reference_type = fixed_reference;
frame_context.variable_reference_types = variable_references;
}
void Parser::frame_reference_mode(FrameContext& frame_context, BooleanDecoder& decoder)
{
auto compound_reference_allowed = false;
for (size_t i = 2; i <= REFS_PER_FRAME; i++) {
if (frame_context.reference_frame_sign_biases[i] != frame_context.reference_frame_sign_biases[1])
compound_reference_allowed = true;
}
ReferenceMode reference_mode;
if (compound_reference_allowed) {
auto non_single_reference = decoder.read_literal(1);
if (non_single_reference == 0) {
reference_mode = SingleReference;
} else {
auto reference_select = decoder.read_literal(1);
if (reference_select == 0)
reference_mode = CompoundReference;
else
reference_mode = ReferenceModeSelect;
}
} else {
reference_mode = SingleReference;
}
frame_context.reference_mode = reference_mode;
if (reference_mode != SingleReference)
setup_compound_reference_mode(frame_context);
}
void Parser::frame_reference_mode_probs(BooleanDecoder& decoder, FrameContext const& frame_context)
{
if (frame_context.reference_mode == ReferenceModeSelect) {
for (auto i = 0; i < COMP_MODE_CONTEXTS; i++) {
auto& comp_mode_prob = m_probability_tables->comp_mode_prob();
comp_mode_prob[i] = diff_update_prob(decoder, comp_mode_prob[i]);
}
}
if (frame_context.reference_mode != CompoundReference) {
for (auto i = 0; i < REF_CONTEXTS; i++) {
auto& single_ref_prob = m_probability_tables->single_ref_prob();
single_ref_prob[i][0] = diff_update_prob(decoder, single_ref_prob[i][0]);
single_ref_prob[i][1] = diff_update_prob(decoder, single_ref_prob[i][1]);
}
}
if (frame_context.reference_mode != SingleReference) {
for (auto i = 0; i < REF_CONTEXTS; i++) {
auto& comp_ref_prob = m_probability_tables->comp_ref_prob();
comp_ref_prob[i] = diff_update_prob(decoder, comp_ref_prob[i]);
}
}
}
void Parser::read_y_mode_probs(BooleanDecoder& decoder)
{
for (auto i = 0; i < BLOCK_SIZE_GROUPS; i++) {
for (auto j = 0; j < INTRA_MODES - 1; j++) {
auto& y_mode_probs = m_probability_tables->y_mode_probs();
y_mode_probs[i][j] = diff_update_prob(decoder, y_mode_probs[i][j]);
}
}
}
void Parser::read_partition_probs(BooleanDecoder& decoder)
{
for (auto i = 0; i < PARTITION_CONTEXTS; i++) {
for (auto j = 0; j < PARTITION_TYPES - 1; j++) {
auto& partition_probs = m_probability_tables->partition_probs();
partition_probs[i][j] = diff_update_prob(decoder, partition_probs[i][j]);
}
}
}
void Parser::mv_probs(BooleanDecoder& decoder, FrameContext const& frame_context)
{
for (auto j = 0; j < MV_JOINTS - 1; j++) {
auto& mv_joint_probs = m_probability_tables->mv_joint_probs();
mv_joint_probs[j] = update_mv_prob(decoder, mv_joint_probs[j]);
}
for (auto i = 0; i < 2; i++) {
auto& mv_sign_prob = m_probability_tables->mv_sign_prob();
mv_sign_prob[i] = update_mv_prob(decoder, mv_sign_prob[i]);
for (auto j = 0; j < MV_CLASSES - 1; j++) {
auto& mv_class_probs = m_probability_tables->mv_class_probs();
mv_class_probs[i][j] = update_mv_prob(decoder, mv_class_probs[i][j]);
}
auto& mv_class0_bit_prob = m_probability_tables->mv_class0_bit_prob();
mv_class0_bit_prob[i] = update_mv_prob(decoder, mv_class0_bit_prob[i]);
for (auto j = 0; j < MV_OFFSET_BITS; j++) {
auto& mv_bits_prob = m_probability_tables->mv_bits_prob();
mv_bits_prob[i][j] = update_mv_prob(decoder, mv_bits_prob[i][j]);
}
}
for (auto i = 0; i < 2; i++) {
for (auto j = 0; j < CLASS0_SIZE; j++) {
for (auto k = 0; k < MV_FR_SIZE - 1; k++) {
auto& mv_class0_fr_probs = m_probability_tables->mv_class0_fr_probs();
mv_class0_fr_probs[i][j][k] = update_mv_prob(decoder, mv_class0_fr_probs[i][j][k]);
}
}
for (auto k = 0; k < MV_FR_SIZE - 1; k++) {
auto& mv_fr_probs = m_probability_tables->mv_fr_probs();
mv_fr_probs[i][k] = update_mv_prob(decoder, mv_fr_probs[i][k]);
}
}
if (frame_context.high_precision_motion_vectors_allowed) {
for (auto i = 0; i < 2; i++) {
auto& mv_class0_hp_prob = m_probability_tables->mv_class0_hp_prob();
auto& mv_hp_prob = m_probability_tables->mv_hp_prob();
mv_class0_hp_prob[i] = update_mv_prob(decoder, mv_class0_hp_prob[i]);
mv_hp_prob[i] = update_mv_prob(decoder, mv_hp_prob[i]);
}
}
}
u8 Parser::update_mv_prob(BooleanDecoder& decoder, u8 prob)
{
if (decoder.read_bool(252)) {
return (decoder.read_literal(7) << 1u) | 1u;
}
return prob;
}
static u32 get_tile_offset(u32 tile_start, u32 frame_size_in_blocks, u32 tile_size_log2)
{
u32 superblocks = blocks_ceiled_to_superblocks(frame_size_in_blocks);
u32 offset = superblocks_to_blocks((tile_start * superblocks) >> tile_size_log2);
return min(offset, frame_size_in_blocks);
}
DecoderErrorOr<void> Parser::decode_tiles(FrameContext& frame_context)
{
auto log2_dimensions = frame_context.log2_of_tile_counts;
auto tile_cols = 1u << log2_dimensions.width();
auto tile_rows = 1u << log2_dimensions.height();
PartitionContext above_partition_context = DECODER_TRY_ALLOC(PartitionContext::create(superblocks_to_blocks(frame_context.superblock_columns())));
NonZeroTokens above_non_zero_tokens = DECODER_TRY_ALLOC(create_non_zero_tokens(blocks_to_sub_blocks(frame_context.columns()), frame_context.color_config.subsampling_x));
SegmentationPredictionContext above_segmentation_ids = DECODER_TRY_ALLOC(SegmentationPredictionContext::create(frame_context.columns()));
// FIXME: To implement tiled decoding, we'll need to pre-parse the tile positions and sizes into a 2D vector of ReadonlyBytes,
// then run through each column of tiles in top to bottom order afterward. Each column can be sent to a worker thread
// for execution. Each worker thread will want to create a set of above contexts sized to its tile width, then provide
// those to each tile as it decodes them.
Vector<Vector<TileContext, 1>, 4> tile_workloads;
DECODER_TRY_ALLOC(tile_workloads.try_ensure_capacity(tile_cols));
for (auto tile_col = 0u; tile_col < tile_cols; tile_col++) {
tile_workloads.append({});
DECODER_TRY_ALLOC(tile_workloads[tile_col].try_ensure_capacity(tile_rows));
}
for (auto tile_row = 0u; tile_row < tile_rows; tile_row++) {
for (auto tile_col = 0u; tile_col < tile_cols; tile_col++) {
auto last_tile = (tile_row == tile_rows - 1) && (tile_col == tile_cols - 1);
size_t tile_size;
if (last_tile)
tile_size = frame_context.stream->remaining();
else
tile_size = TRY_READ(frame_context.bit_stream.read_bits(32));
auto rows_start = get_tile_offset(tile_row, frame_context.rows(), log2_dimensions.height());
auto rows_end = get_tile_offset(tile_row + 1, frame_context.rows(), log2_dimensions.height());
auto columns_start = get_tile_offset(tile_col, frame_context.columns(), log2_dimensions.width());
auto columns_end = get_tile_offset(tile_col + 1, frame_context.columns(), log2_dimensions.width());
auto width = columns_end - columns_start;
auto above_partition_context_for_tile = above_partition_context.span().slice(columns_start, superblocks_to_blocks(blocks_ceiled_to_superblocks(width)));
auto above_non_zero_tokens_view = create_non_zero_tokens_view(above_non_zero_tokens, blocks_to_sub_blocks(columns_start), blocks_to_sub_blocks(columns_end - columns_start), frame_context.color_config.subsampling_x);
auto above_segmentation_ids_for_tile = safe_slice(above_segmentation_ids.span(), columns_start, columns_end - columns_start);
tile_workloads[tile_col].append(TRY(TileContext::try_create(frame_context, tile_size, rows_start, rows_end, columns_start, columns_end, above_partition_context_for_tile, above_non_zero_tokens_view, above_segmentation_ids_for_tile)));
}
}
auto decode_tile_column = [this, tile_rows](auto& column_workloads) -> DecoderErrorOr<void> {
VERIFY(column_workloads.size() == tile_rows);
for (auto tile_row = 0u; tile_row < tile_rows; tile_row++)
TRY(decode_tile(column_workloads[tile_row]));
return {};
};
#ifdef VP9_TILE_THREADING
auto const worker_count = tile_cols - 1;
if (m_worker_threads.size() < worker_count) {
m_worker_threads.clear();
m_worker_threads.ensure_capacity(worker_count);
for (auto i = 0u; i < worker_count; i++)
m_worker_threads.append(DECODER_TRY_ALLOC(Threading::WorkerThread<DecoderError>::create("Decoder Worker"sv)));
}
VERIFY(m_worker_threads.size() >= worker_count);
// Start tile column decoding tasks in thread workers starting from the second column.
for (auto tile_col = 1u; tile_col < tile_cols; tile_col++) {
auto& column_workload = tile_workloads[tile_col];
m_worker_threads[tile_col - 1]->start_task([&decode_tile_column, &column_workload]() -> DecoderErrorOr<void> {
return decode_tile_column(column_workload);
});
}
// Decode the first column in this thread.
auto result = decode_tile_column(tile_workloads[0]);
for (auto& worker_thread : m_worker_threads) {
auto task_result = worker_thread->wait_until_task_is_finished();
if (!result.is_error() && task_result.is_error())
result = move(task_result);
}
if (result.is_error())
return result;
#else
for (auto& column_workloads : tile_workloads)
TRY(decode_tile_column(column_workloads));
#endif
// Sum up all tile contexts' syntax element counters after all decodes have finished.
for (auto& tile_contexts : tile_workloads) {
for (auto& tile_context : tile_contexts) {
*frame_context.counter += *tile_context.counter;
}
}
return {};
}
DecoderErrorOr<void> Parser::decode_tile(TileContext& tile_context)
{
for (auto row = tile_context.rows_start; row < tile_context.rows_end; row += 8) {
clear_left_context(tile_context);
for (auto col = tile_context.columns_start; col < tile_context.columns_end; col += 8) {
TRY(decode_partition(tile_context, row, col, Block_64x64));
}
}
TRY_READ(tile_context.decoder.finish_decode());
return {};
}
void Parser::clear_left_context(TileContext& tile_context)
{
for (auto& context_for_plane : tile_context.left_non_zero_tokens)
context_for_plane.fill_with(false);
tile_context.left_segmentation_ids.fill_with(0);
tile_context.left_partition_context.fill_with(0);
}
DecoderErrorOr<void> Parser::decode_partition(TileContext& tile_context, u32 row, u32 column, BlockSubsize subsize)
{
if (row >= tile_context.frame_context.rows() || column >= tile_context.frame_context.columns())
return {};
u8 num_8x8 = num_8x8_blocks_wide_lookup[subsize];
auto half_block_8x8 = num_8x8 >> 1;
bool has_rows = (row + half_block_8x8) < tile_context.frame_context.rows();
bool has_cols = (column + half_block_8x8) < tile_context.frame_context.columns();
u32 row_in_tile = row - tile_context.rows_start;
u32 column_in_tile = column - tile_context.columns_start;
auto partition = TreeParser::parse_partition(tile_context.decoder, *m_probability_tables, *tile_context.counter, has_rows, has_cols, subsize, num_8x8, tile_context.above_partition_context, tile_context.left_partition_context.span(), row_in_tile, column_in_tile, !tile_context.frame_context.is_inter_predicted());
auto child_subsize = subsize_lookup[partition][subsize];
if (child_subsize < Block_8x8 || partition == PartitionNone) {
TRY(decode_block(tile_context, row, column, child_subsize));
} else if (partition == PartitionHorizontal) {
TRY(decode_block(tile_context, row, column, child_subsize));
if (has_rows)
TRY(decode_block(tile_context, row + half_block_8x8, column, child_subsize));
} else if (partition == PartitionVertical) {
TRY(decode_block(tile_context, row, column, child_subsize));
if (has_cols)
TRY(decode_block(tile_context, row, column + half_block_8x8, child_subsize));
} else {
TRY(decode_partition(tile_context, row, column, child_subsize));
TRY(decode_partition(tile_context, row, column + half_block_8x8, child_subsize));
TRY(decode_partition(tile_context, row + half_block_8x8, column, child_subsize));
TRY(decode_partition(tile_context, row + half_block_8x8, column + half_block_8x8, child_subsize));
}
if (subsize == Block_8x8 || partition != PartitionSplit) {
auto above_context = 15 >> b_width_log2_lookup[child_subsize];
auto left_context = 15 >> b_height_log2_lookup[child_subsize];
for (size_t i = 0; i < num_8x8; i++) {
tile_context.above_partition_context[column_in_tile + i] = above_context;
tile_context.left_partition_context[row_in_tile + i] = left_context;
}
}
return {};
}
size_t Parser::get_image_index(FrameContext const& frame_context, u32 row, u32 column) const
{
VERIFY(row < frame_context.rows() && column < frame_context.columns());
return row * frame_context.columns() + column;
}
DecoderErrorOr<void> Parser::decode_block(TileContext& tile_context, u32 row, u32 column, BlockSubsize subsize)
{
auto above_context = row > 0 ? tile_context.frame_block_contexts().at(row - 1, column) : FrameBlockContext();
auto left_context = column > tile_context.columns_start ? tile_context.frame_block_contexts().at(row, column - 1) : FrameBlockContext();
auto block_context = BlockContext::create(tile_context, row, column, subsize);
mode_info(block_context, above_context, left_context);
auto had_residual_tokens = TRY(residual(block_context, above_context.is_available, left_context.is_available));
if (block_context.is_inter_predicted() && subsize >= Block_8x8 && !had_residual_tokens)
block_context.should_skip_residuals = true;
for (size_t y = 0; y < block_context.contexts_view.height(); y++) {
for (size_t x = 0; x < block_context.contexts_view.width(); x++) {
auto sub_block_context = FrameBlockContext { true, block_context.should_skip_residuals, block_context.transform_size, block_context.y_prediction_mode(), block_context.sub_block_prediction_modes, block_context.interpolation_filter, block_context.reference_frame_types, block_context.sub_block_motion_vectors, block_context.segment_id };
block_context.contexts_view.at(y, x) = sub_block_context;
VERIFY(block_context.frame_block_contexts().at(row + y, column + x).transform_size == sub_block_context.transform_size);
}
}
return {};
}
void Parser::mode_info(BlockContext& block_context, FrameBlockContext above_context, FrameBlockContext left_context)
{
if (block_context.frame_context.is_inter_predicted())
inter_frame_mode_info(block_context, above_context, left_context);
else
intra_frame_mode_info(block_context, above_context, left_context);
}
void Parser::intra_frame_mode_info(BlockContext& block_context, FrameBlockContext above_context, FrameBlockContext left_context)
{
block_context.reference_frame_types = { ReferenceFrameType::None, ReferenceFrameType::None };
VERIFY(!block_context.is_inter_predicted());
set_intra_segment_id(block_context);
block_context.should_skip_residuals = read_should_skip_residuals(block_context, above_context, left_context);
block_context.transform_size = read_tx_size(block_context, above_context, left_context, true);
// FIXME: This if statement is also present in parse_default_intra_mode. The selection of parameters for
// the probability table lookup should be inlined here.
if (block_context.size >= Block_8x8) {
auto mode = TreeParser::parse_default_intra_mode(block_context.decoder, *m_probability_tables, block_context.size, above_context, left_context, block_context.sub_block_prediction_modes, 0, 0);
for (auto& block_sub_mode : block_context.sub_block_prediction_modes)
block_sub_mode = mode;
} else {
auto size_in_sub_blocks = block_context.get_size_in_sub_blocks();
for (auto idy = 0; idy < 2; idy += size_in_sub_blocks.height()) {
for (auto idx = 0; idx < 2; idx += size_in_sub_blocks.width()) {
auto sub_mode = TreeParser::parse_default_intra_mode(block_context.decoder, *m_probability_tables, block_context.size, above_context, left_context, block_context.sub_block_prediction_modes, idx, idy);
for (auto y = 0; y < size_in_sub_blocks.height(); y++) {
for (auto x = 0; x < size_in_sub_blocks.width(); x++) {
auto index = (idy + y) * 2 + idx + x;
block_context.sub_block_prediction_modes[index] = sub_mode;
}
}
}
}
}
block_context.uv_prediction_mode = TreeParser::parse_default_uv_mode(block_context.decoder, *m_probability_tables, block_context.y_prediction_mode());
}
void Parser::set_intra_segment_id(BlockContext& block_context)
{
if (block_context.frame_context.segmentation_enabled && block_context.frame_context.use_full_segment_id_tree)
block_context.segment_id = TreeParser::parse_segment_id(block_context.decoder, block_context.frame_context.full_segment_id_tree_probabilities);
else
block_context.segment_id = 0;
}
bool Parser::read_should_skip_residuals(BlockContext& block_context, FrameBlockContext above_context, FrameBlockContext left_context)
{
if (block_context.get_segment_feature(SegmentFeature::SkipResidualsOverride).enabled)
return true;
return TreeParser::parse_skip(block_context.decoder, *m_probability_tables, block_context.counter, above_context, left_context);
}
TransformSize Parser::read_tx_size(BlockContext& block_context, FrameBlockContext above_context, FrameBlockContext left_context, bool allow_select)
{
auto max_tx_size = max_txsize_lookup[block_context.size];
if (allow_select && block_context.frame_context.transform_mode == TransformMode::Select && block_context.size >= Block_8x8)
return (TreeParser::parse_tx_size(block_context.decoder, *m_probability_tables, block_context.counter, max_tx_size, above_context, left_context));
return min(max_tx_size, tx_mode_to_biggest_tx_size[to_underlying(block_context.frame_context.transform_mode)]);
}
void Parser::inter_frame_mode_info(BlockContext& block_context, FrameBlockContext above_context, FrameBlockContext left_context)
{
set_inter_segment_id(block_context);
block_context.should_skip_residuals = read_should_skip_residuals(block_context, above_context, left_context);
auto is_inter = read_is_inter(block_context, above_context, left_context);
block_context.transform_size = read_tx_size(block_context, above_context, left_context, !block_context.should_skip_residuals || !is_inter);
if (is_inter) {
inter_block_mode_info(block_context, above_context, left_context);
} else {
intra_block_mode_info(block_context);
}
}
void Parser::set_inter_segment_id(BlockContext& block_context)
{
if (!block_context.frame_context.segmentation_enabled) {
block_context.segment_id = 0;
return;
}
auto predicted_segment_id = get_segment_id(block_context);
if (!block_context.frame_context.use_full_segment_id_tree) {
block_context.segment_id = predicted_segment_id;
return;
}
if (!block_context.frame_context.use_predicted_segment_id_tree) {
block_context.segment_id = TreeParser::parse_segment_id(block_context.decoder, block_context.frame_context.full_segment_id_tree_probabilities);
return;
}
auto above_segmentation_id = block_context.tile_context.above_segmentation_ids[block_context.row - block_context.tile_context.rows_start];
auto left_segmentation_id = block_context.tile_context.left_segmentation_ids[block_context.column - block_context.tile_context.columns_start];
auto seg_id_predicted = TreeParser::parse_segment_id_predicted(block_context.decoder, block_context.frame_context.predicted_segment_id_tree_probabilities, above_segmentation_id, left_segmentation_id);
if (seg_id_predicted)
block_context.segment_id = predicted_segment_id;
else
block_context.segment_id = TreeParser::parse_segment_id(block_context.decoder, block_context.frame_context.full_segment_id_tree_probabilities);
// (7.4.1) AboveSegPredContext[ i ] only needs to be set to 0 for i = 0..MiCols-1.
// This is taken care of by the slicing in BlockContext.
block_context.above_segmentation_ids.fill(seg_id_predicted);
// (7.4.1) LeftSegPredContext[ i ] only needs to be set to 0 for i = 0..MiRows-1.
// This is taken care of by the slicing in BlockContext.
block_context.left_segmentation_ids.fill(seg_id_predicted);
}
u8 Parser::get_segment_id(BlockContext const& block_context)
{
auto bw = num_8x8_blocks_wide_lookup[block_context.size];
auto bh = num_8x8_blocks_high_lookup[block_context.size];
auto xmis = min(block_context.frame_context.columns() - block_context.column, (u32)bw);
auto ymis = min(block_context.frame_context.rows() - block_context.row, (u32)bh);
u8 segment = 7;
for (size_t y = 0; y < ymis; y++) {
for (size_t x = 0; x < xmis; x++) {
segment = min(segment, m_previous_block_contexts.index_at(block_context.row + y, block_context.column + x));
}
}
return segment;
}
bool Parser::read_is_inter(BlockContext& block_context, FrameBlockContext above_context, FrameBlockContext left_context)
{
auto reference_frame_override_feature = block_context.get_segment_feature(SegmentFeature::ReferenceFrameOverride);
if (reference_frame_override_feature.enabled)
return reference_frame_override_feature.value != ReferenceFrameType::None;
return TreeParser::parse_block_is_inter_predicted(block_context.decoder, *m_probability_tables, block_context.counter, above_context, left_context);
}
void Parser::intra_block_mode_info(BlockContext& block_context)
{
block_context.reference_frame_types = { ReferenceFrameType::None, ReferenceFrameType::None };
VERIFY(!block_context.is_inter_predicted());
auto& sub_modes = block_context.sub_block_prediction_modes;
if (block_context.size >= Block_8x8) {
auto mode = TreeParser::parse_intra_mode(block_context.decoder, *m_probability_tables, block_context.counter, block_context.size);
for (auto& block_sub_mode : sub_modes)
block_sub_mode = mode;
} else {
auto size_in_sub_blocks = block_context.get_size_in_sub_blocks();
for (auto idy = 0; idy < 2; idy += size_in_sub_blocks.height()) {
for (auto idx = 0; idx < 2; idx += size_in_sub_blocks.width()) {
auto sub_intra_mode = TreeParser::parse_sub_intra_mode(block_context.decoder, *m_probability_tables, block_context.counter);
for (auto y = 0; y < size_in_sub_blocks.height(); y++) {
for (auto x = 0; x < size_in_sub_blocks.width(); x++)
sub_modes[(idy + y) * 2 + idx + x] = sub_intra_mode;
}
}
}
}
block_context.uv_prediction_mode = TreeParser::parse_uv_mode(block_context.decoder, *m_probability_tables, block_context.counter, block_context.y_prediction_mode());
}
static void select_best_reference_motion_vectors(BlockContext& block_context, MotionVectorPair reference_motion_vectors, BlockMotionVectorCandidates& candidates, ReferenceIndex);
void Parser::inter_block_mode_info(BlockContext& block_context, FrameBlockContext above_context, FrameBlockContext left_context)
{
read_ref_frames(block_context, above_context, left_context);
VERIFY(block_context.is_inter_predicted());
BlockMotionVectorCandidates motion_vector_candidates;
auto reference_motion_vectors = find_reference_motion_vectors(block_context, block_context.reference_frame_types.primary, -1);
select_best_reference_motion_vectors(block_context, reference_motion_vectors, motion_vector_candidates, ReferenceIndex::Primary);
if (block_context.is_compound()) {
auto reference_motion_vectors = find_reference_motion_vectors(block_context, block_context.reference_frame_types.secondary, -1);
select_best_reference_motion_vectors(block_context, reference_motion_vectors, motion_vector_candidates, ReferenceIndex::Secondary);
}
if (block_context.get_segment_feature(SegmentFeature::SkipResidualsOverride).enabled) {
block_context.y_prediction_mode() = PredictionMode::ZeroMv;
} else if (block_context.size >= Block_8x8) {
block_context.y_prediction_mode() = TreeParser::parse_inter_mode(block_context.decoder, *m_probability_tables, block_context.counter, block_context.mode_context[block_context.reference_frame_types.primary]);
}
if (block_context.frame_context.interpolation_filter == Switchable)
block_context.interpolation_filter = TreeParser::parse_interpolation_filter(block_context.decoder, *m_probability_tables, block_context.counter, above_context, left_context);
else
block_context.interpolation_filter = block_context.frame_context.interpolation_filter;
if (block_context.size < Block_8x8) {
auto size_in_sub_blocks = block_context.get_size_in_sub_blocks();
for (auto idy = 0; idy < 2; idy += size_in_sub_blocks.height()) {
for (auto idx = 0; idx < 2; idx += size_in_sub_blocks.width()) {
block_context.y_prediction_mode() = TreeParser::parse_inter_mode(block_context.decoder, *m_probability_tables, block_context.counter, block_context.mode_context[block_context.reference_frame_types.primary]);
if (block_context.y_prediction_mode() == PredictionMode::NearestMv || block_context.y_prediction_mode() == PredictionMode::NearMv) {
select_best_sub_block_reference_motion_vectors(block_context, motion_vector_candidates, idy * 2 + idx, ReferenceIndex::Primary);
if (block_context.is_compound())
select_best_sub_block_reference_motion_vectors(block_context, motion_vector_candidates, idy * 2 + idx, ReferenceIndex::Secondary);
}
auto new_motion_vector_pair = get_motion_vector(block_context, motion_vector_candidates);
for (auto y = 0; y < size_in_sub_blocks.height(); y++) {
for (auto x = 0; x < size_in_sub_blocks.width(); x++) {
auto sub_block_index = (idy + y) * 2 + idx + x;
block_context.sub_block_motion_vectors[sub_block_index] = new_motion_vector_pair;
}
}
}
}
return;
}
auto new_motion_vector_pair = get_motion_vector(block_context, motion_vector_candidates);
for (auto block = 0; block < 4; block++)
block_context.sub_block_motion_vectors[block] = new_motion_vector_pair;
}
void Parser::read_ref_frames(BlockContext& block_context, FrameBlockContext above_context, FrameBlockContext left_context)
{
auto reference_frame_override_feature = block_context.get_segment_feature(SegmentFeature::ReferenceFrameOverride);
if (reference_frame_override_feature.enabled) {
block_context.reference_frame_types = { static_cast<ReferenceFrameType>(reference_frame_override_feature.value), ReferenceFrameType::None };
return;
}
ReferenceMode compound_mode = block_context.frame_context.reference_mode;
auto fixed_reference = block_context.frame_context.fixed_reference_type;
if (compound_mode == ReferenceModeSelect)
compound_mode = TreeParser::parse_comp_mode(block_context.decoder, *m_probability_tables, block_context.counter, fixed_reference, above_context, left_context);
if (compound_mode == CompoundReference) {
auto variable_references = block_context.frame_context.variable_reference_types;
auto fixed_reference_index = ReferenceIndex::Primary;
auto variable_reference_index = ReferenceIndex::Secondary;
if (block_context.frame_context.reference_frame_sign_biases[fixed_reference])
swap(fixed_reference_index, variable_reference_index);
auto variable_reference_selection = TreeParser::parse_comp_ref(block_context.decoder, *m_probability_tables, block_context.counter, fixed_reference, variable_references, variable_reference_index, above_context, left_context);
block_context.reference_frame_types[fixed_reference_index] = fixed_reference;
block_context.reference_frame_types[variable_reference_index] = variable_references[variable_reference_selection];
return;
}
// FIXME: Maybe consolidate this into a tree. Context is different between part 1 and 2 but still, it would look nice here.
ReferenceFrameType primary_type = ReferenceFrameType::LastFrame;
auto single_ref_p1 = TreeParser::parse_single_ref_part_1(block_context.decoder, *m_probability_tables, block_context.counter, above_context, left_context);
if (single_ref_p1) {
auto single_ref_p2 = TreeParser::parse_single_ref_part_2(block_context.decoder, *m_probability_tables, block_context.counter, above_context, left_context);
primary_type = single_ref_p2 ? ReferenceFrameType::AltRefFrame : ReferenceFrameType::GoldenFrame;
}
block_context.reference_frame_types = { primary_type, ReferenceFrameType::None };
}
// assign_mv( isCompound ) in the spec.
MotionVectorPair Parser::get_motion_vector(BlockContext const& block_context, BlockMotionVectorCandidates const& candidates)
{
MotionVectorPair result;
auto read_one = [&](ReferenceIndex index) -> void {
switch (block_context.y_prediction_mode()) {
case PredictionMode::NewMv:
result[index] = read_motion_vector(block_context, candidates, index);
break;
case PredictionMode::NearestMv:
result[index] = candidates[index].nearest_vector;
break;
case PredictionMode::NearMv:
result[index] = candidates[index].near_vector;
break;
default:
result[index] = {};
break;
}
return;
};
read_one(ReferenceIndex::Primary);
if (block_context.is_compound())
read_one(ReferenceIndex::Secondary);
return result;
}
// use_mv_hp( deltaMv ) in the spec.
static bool should_use_high_precision_motion_vector(MotionVector const& delta_vector)
{
return (abs(delta_vector.row()) >> 3) < COMPANDED_MVREF_THRESH && (abs(delta_vector.column()) >> 3) < COMPANDED_MVREF_THRESH;
}
// read_mv( ref ) in the spec.
MotionVector Parser::read_motion_vector(BlockContext const& block_context, BlockMotionVectorCandidates const& candidates, ReferenceIndex reference_index)
{
auto use_high_precision = block_context.frame_context.high_precision_motion_vectors_allowed && should_use_high_precision_motion_vector(candidates[reference_index].best_vector);
MotionVector delta_vector;
auto joint = TreeParser::parse_motion_vector_joint(block_context.decoder, *m_probability_tables, block_context.counter);
if ((joint & MotionVectorNonZeroRow) != 0)
delta_vector.set_row(read_single_motion_vector_component(block_context.decoder, block_context.counter, 0, use_high_precision));
if ((joint & MotionVectorNonZeroColumn) != 0)
delta_vector.set_column(read_single_motion_vector_component(block_context.decoder, block_context.counter, 1, use_high_precision));
return candidates[reference_index].best_vector + delta_vector;
}
// read_mv_component( comp ) in the spec.
i32 Parser::read_single_motion_vector_component(BooleanDecoder& decoder, SyntaxElementCounter& counter, u8 component, bool use_high_precision)
{
auto mv_sign = TreeParser::parse_motion_vector_sign(decoder, *m_probability_tables, counter, component);
auto mv_class = TreeParser::parse_motion_vector_class(decoder, *m_probability_tables, counter, component);
u32 magnitude;
if (mv_class == MvClass0) {
auto mv_class0_bit = TreeParser::parse_motion_vector_class0_bit(decoder, *m_probability_tables, counter, component);
auto mv_class0_fr = TreeParser::parse_motion_vector_class0_fr(decoder, *m_probability_tables, counter, component, mv_class0_bit);
auto mv_class0_hp = TreeParser::parse_motion_vector_class0_hp(decoder, *m_probability_tables, counter, component, use_high_precision);
magnitude = ((mv_class0_bit << 3) | (mv_class0_fr << 1) | mv_class0_hp) + 1;
} else {
u32 bits = 0;
for (u8 i = 0; i < mv_class; i++) {
auto mv_bit = TreeParser::parse_motion_vector_bit(decoder, *m_probability_tables, counter, component, i);
bits |= mv_bit << i;
}
magnitude = CLASS0_SIZE << (mv_class + 2);
auto mv_fr = TreeParser::parse_motion_vector_fr(decoder, *m_probability_tables, counter, component);
auto mv_hp = TreeParser::parse_motion_vector_hp(decoder, *m_probability_tables, counter, component, use_high_precision);
magnitude += ((bits << 3) | (mv_fr << 1) | mv_hp) + 1;
}
return (mv_sign ? -1 : 1) * static_cast<i32>(magnitude);
}
static TransformSize get_uv_transform_size(TransformSize transform_size, BlockSubsize size_for_plane)
{
return min(transform_size, max_txsize_lookup[size_for_plane]);
}
static TransformSet select_transform_type(BlockContext const& block_context, u8 plane, TransformSize transform_size, u32 block_index)
{
if (plane > 0 || transform_size == Transform_32x32)
return TransformSet { TransformType::DCT, TransformType::DCT };
if (transform_size == Transform_4x4) {
if (block_context.frame_context.lossless || block_context.is_inter_predicted())
return TransformSet { TransformType::DCT, TransformType::DCT };
return mode_to_txfm_map[to_underlying(block_context.size < Block_8x8 ? block_context.sub_block_prediction_modes[block_index] : block_context.y_prediction_mode())];
}
return mode_to_txfm_map[to_underlying(block_context.y_prediction_mode())];
}
DecoderErrorOr<bool> Parser::residual(BlockContext& block_context, bool has_block_above, bool has_block_left)
{
bool block_had_non_zero_tokens = false;
Array<u8, 1024> token_cache;
for (u8 plane = 0; plane < 3; plane++) {
auto plane_subsampling_x = (plane > 0) ? block_context.frame_context.color_config.subsampling_x : false;
auto plane_subsampling_y = (plane > 0) ? block_context.frame_context.color_config.subsampling_y : false;
auto plane_size = get_subsampled_block_size(block_context.size, plane_subsampling_x, plane_subsampling_y);
if (plane_size == Block_Invalid) {
return DecoderError::corrupted("Invalid block size"sv);
}
auto transform_size = get_uv_transform_size(block_context.transform_size, plane_size);
auto transform_size_in_sub_blocks = transform_size_to_sub_blocks(transform_size);
auto block_size_in_sub_blocks = block_size_to_sub_blocks(plane_size);
auto base_x_in_pixels = (blocks_to_pixels(block_context.column)) >> plane_subsampling_x;
auto base_y_in_pixels = (blocks_to_pixels(block_context.row)) >> plane_subsampling_y;
if (block_context.is_inter_predicted()) {
if (block_context.size < Block_8x8) {
for (auto y = 0; y < block_size_in_sub_blocks.height(); y++) {
for (auto x = 0; x < block_size_in_sub_blocks.width(); x++) {
TRY(m_decoder.predict_inter(plane, block_context, base_x_in_pixels + sub_blocks_to_pixels(x), base_y_in_pixels + sub_blocks_to_pixels(y), sub_blocks_to_pixels(1), sub_blocks_to_pixels(1), (y * block_size_in_sub_blocks.width()) + x));
}
}
} else {
TRY(m_decoder.predict_inter(plane, block_context, base_x_in_pixels, base_y_in_pixels, sub_blocks_to_pixels(block_size_in_sub_blocks.width()), sub_blocks_to_pixels(block_size_in_sub_blocks.height()), 0));
}
}
auto frame_right_in_pixels = (blocks_to_pixels(block_context.frame_context.columns())) >> plane_subsampling_x;
auto frame_bottom_in_pixels = (blocks_to_pixels(block_context.frame_context.rows())) >> plane_subsampling_y;
auto sub_block_index = 0;
for (u32 y = 0; y < block_size_in_sub_blocks.height(); y += transform_size_in_sub_blocks) {
for (u32 x = 0; x < block_size_in_sub_blocks.width(); x += transform_size_in_sub_blocks) {
auto transform_x_in_px = base_x_in_pixels + sub_blocks_to_pixels(x);
auto transform_y_in_px = base_y_in_pixels + sub_blocks_to_pixels(y);
auto sub_block_had_non_zero_tokens = false;
if (transform_x_in_px < frame_right_in_pixels && transform_y_in_px < frame_bottom_in_pixels) {
if (!block_context.is_inter_predicted())
TRY(m_decoder.predict_intra(plane, block_context, transform_x_in_px, transform_y_in_px, has_block_left || x > 0, has_block_above || y > 0, (x + transform_size_in_sub_blocks) < block_size_in_sub_blocks.width(), transform_size, sub_block_index));
if (!block_context.should_skip_residuals) {
auto transform_set = select_transform_type(block_context, plane, transform_size, sub_block_index);
sub_block_had_non_zero_tokens = tokens(block_context, plane, x, y, transform_size, transform_set, token_cache);
block_had_non_zero_tokens = block_had_non_zero_tokens || sub_block_had_non_zero_tokens;
TRY(m_decoder.reconstruct(plane, block_context, transform_x_in_px, transform_y_in_px, transform_size, transform_set));
}
}
auto& above_sub_block_tokens = block_context.above_non_zero_tokens[plane];
auto transform_right_in_sub_blocks = min(x + transform_size_in_sub_blocks, above_sub_block_tokens.size());
for (size_t inside_x = x; inside_x < transform_right_in_sub_blocks; inside_x++)
above_sub_block_tokens[inside_x] = sub_block_had_non_zero_tokens;
auto& left_sub_block_context = block_context.left_non_zero_tokens[plane];
auto transform_bottom_in_sub_blocks = min(y + transform_size_in_sub_blocks, left_sub_block_context.size());
for (size_t inside_y = y; inside_y < transform_bottom_in_sub_blocks; inside_y++)
left_sub_block_context[inside_y] = sub_block_had_non_zero_tokens;
sub_block_index++;
}
}
}
return block_had_non_zero_tokens;
}
static u16 const* get_scan(TransformSize transform_size, TransformSet transform_set)
{
constexpr TransformSet adst_dct { TransformType::ADST, TransformType::DCT };
constexpr TransformSet dct_adst { TransformType::DCT, TransformType::ADST };
if (transform_size == Transform_4x4) {
if (transform_set == adst_dct)
return row_scan_4x4;
if (transform_set == dct_adst)
return col_scan_4x4;
return default_scan_4x4;
}
if (transform_size == Transform_8x8) {
if (transform_set == adst_dct)
return row_scan_8x8;
if (transform_set == dct_adst)
return col_scan_8x8;
return default_scan_8x8;
}
if (transform_size == Transform_16x16) {
if (transform_set == adst_dct)
return row_scan_16x16;
if (transform_set == dct_adst)
return col_scan_16x16;
return default_scan_16x16;
}
return default_scan_32x32;
}
bool Parser::tokens(BlockContext& block_context, size_t plane, u32 sub_block_column, u32 sub_block_row, TransformSize transform_size, TransformSet transform_set, Array<u8, 1024> token_cache)
{
block_context.residual_tokens.fill(0);
auto const* scan = get_scan(transform_size, transform_set);
auto check_for_more_coefficients = true;
u16 coef_index = 0;
u16 transform_pixel_count = 16 << (transform_size << 1);
for (; coef_index < transform_pixel_count; coef_index++) {
auto band = (transform_size == Transform_4x4) ? coefband_4x4[coef_index] : coefband_8x8plus[coef_index];
auto token_position = scan[coef_index];
TokensContext tokens_context;
if (coef_index == 0)
tokens_context = TreeParser::get_context_for_first_token(block_context.above_non_zero_tokens, block_context.left_non_zero_tokens, transform_size, plane, sub_block_column, sub_block_row, block_context.is_inter_predicted(), band);
else
tokens_context = TreeParser::get_context_for_other_tokens(token_cache, transform_size, transform_set, plane, token_position, block_context.is_inter_predicted(), band);
if (check_for_more_coefficients && !TreeParser::parse_more_coefficients(block_context.decoder, *m_probability_tables, block_context.counter, tokens_context))
break;
auto token = TreeParser::parse_token(block_context.decoder, *m_probability_tables, block_context.counter, tokens_context);
token_cache[token_position] = energy_class[token];
i32 coef;
if (token == ZeroToken) {
coef = 0;
check_for_more_coefficients = false;
} else {
coef = read_coef(block_context.decoder, block_context.frame_context.color_config.bit_depth, token);
check_for_more_coefficients = true;
}
block_context.residual_tokens[token_position] = coef;
}
return coef_index > 0;
}
i32 Parser::read_coef(BooleanDecoder& decoder, u8 bit_depth, Token token)
{
auto cat = extra_bits[token][0];
auto num_extra = extra_bits[token][1];
i32 coef = extra_bits[token][2];
if (token == DctValCat6) {
for (size_t e = 0; e < (u8)(bit_depth - 8); e++) {
auto high_bit = decoder.read_bool(255);
coef += high_bit << (5 + bit_depth - e);
}
}
for (size_t e = 0; e < num_extra; e++) {
auto coef_bit = decoder.read_bool(cat_probs[cat][e]);
coef += coef_bit << (num_extra - 1 - e);
}
bool sign_bit = decoder.read_literal(1);
coef = sign_bit ? -coef : coef;
return coef;
}
// is_inside( candidateR, candidateC ) in the spec.
static bool motion_vector_is_inside_tile(TileContext const& tile_context, MotionVector vector)
{
if (vector.row() < 0)
return false;
if (vector.column() < 0)
return false;
u32 row_positive = vector.row();
u32 column_positive = vector.column();
return row_positive < tile_context.frame_context.rows() && column_positive >= tile_context.columns_start && column_positive < tile_context.columns_end;
}
// add_mv_ref_list( refList ) in the spec.
static void add_motion_vector_to_list_deduped(MotionVector const& vector, Vector<MotionVector, 2>& list)
{
if (list.size() >= 2)
return;
if (list.size() == 1 && vector == list[0])
return;
list.append(vector);
}
// get_block_mv( candidateR, candidateC, refList, usePrev ) in the spec.
MotionVectorCandidate Parser::get_motion_vector_from_current_or_previous_frame(BlockContext const& block_context, MotionVector candidate_vector, ReferenceIndex reference_index, bool use_prev)
{
if (use_prev) {
auto const& prev_context = m_previous_block_contexts.at(candidate_vector.row(), candidate_vector.column());
return { prev_context.ref_frames[reference_index], prev_context.primary_motion_vector_pair[reference_index] };
}
auto const& current_context = block_context.frame_block_contexts().at(candidate_vector.row(), candidate_vector.column());
return { current_context.ref_frames[reference_index], current_context.primary_motion_vector_pair()[reference_index] };
}
// if_same_ref_frame_add_mv( candidateR, candidateC, refFrame, usePrev ) in the spec.
void Parser::add_motion_vector_if_reference_frame_type_is_same(BlockContext const& block_context, MotionVector candidate_vector, ReferenceFrameType ref_frame, Vector<MotionVector, 2>& list, bool use_prev)
{
for (auto i = 0u; i < 2; i++) {
auto candidate = get_motion_vector_from_current_or_previous_frame(block_context, candidate_vector, static_cast<ReferenceIndex>(i), use_prev);
if (candidate.type == ref_frame) {
add_motion_vector_to_list_deduped(candidate.vector, list);
return;
}
}
}
// scale_mv( refList, refFrame ) in the spec.
static void apply_sign_bias_to_motion_vector(FrameContext const& frame_context, MotionVectorCandidate& candidate, ReferenceFrameType ref_frame)
{
if (frame_context.reference_frame_sign_biases[candidate.type] != frame_context.reference_frame_sign_biases[ref_frame])
candidate.vector *= -1;
}
// if_diff_ref_frame_add_mv( candidateR, candidateC, refFrame, usePrev ) in the spec.
void Parser::add_motion_vector_if_reference_frame_type_is_different(BlockContext const& block_context, MotionVector candidate_vector, ReferenceFrameType ref_frame, Vector<MotionVector, 2>& list, bool use_prev)
{
auto first_candidate = get_motion_vector_from_current_or_previous_frame(block_context, candidate_vector, ReferenceIndex::Primary, use_prev);
if (first_candidate.type > ReferenceFrameType::None && first_candidate.type != ref_frame) {
apply_sign_bias_to_motion_vector(block_context.frame_context, first_candidate, ref_frame);
add_motion_vector_to_list_deduped(first_candidate.vector, list);
}
auto second_candidate = get_motion_vector_from_current_or_previous_frame(block_context, candidate_vector, ReferenceIndex::Secondary, use_prev);
auto mvs_are_same = first_candidate.vector == second_candidate.vector;
if (second_candidate.type > ReferenceFrameType::None && second_candidate.type != ref_frame && !mvs_are_same) {
apply_sign_bias_to_motion_vector(block_context.frame_context, second_candidate, ref_frame);
add_motion_vector_to_list_deduped(second_candidate.vector, list);
}
}
// This function handles both clamp_mv_row( mvec, border ) and clamp_mv_col( mvec, border ) in the spec.
static MotionVector clamp_motion_vector(BlockContext const& block_context, MotionVector vector, i32 border)
{
i32 blocks_high = num_8x8_blocks_high_lookup[block_context.size];
// Casts must be done here to prevent subtraction underflow from wrapping the values.
i32 mb_to_top_edge = -8 * (static_cast<i32>(block_context.row) * MI_SIZE);
i32 mb_to_bottom_edge = 8 * ((static_cast<i32>(block_context.frame_context.rows()) - blocks_high - static_cast<i32>(block_context.row)) * MI_SIZE);
i32 blocks_wide = num_8x8_blocks_wide_lookup[block_context.size];
i32 mb_to_left_edge = -8 * (static_cast<i32>(block_context.column) * MI_SIZE);
i32 mb_to_right_edge = 8 * ((static_cast<i32>(block_context.frame_context.columns()) - blocks_wide - static_cast<i32>(block_context.column)) * MI_SIZE);
return {
clip_3(mb_to_top_edge - border, mb_to_bottom_edge + border, vector.row()),
clip_3(mb_to_left_edge - border, mb_to_right_edge + border, vector.column())
};
}
// 6.5.1 Find MV refs syntax
// find_mv_refs( refFrame, block ) in the spec.
MotionVectorPair Parser::find_reference_motion_vectors(BlockContext& block_context, ReferenceFrameType reference_frame, i32 block)
{
// FIXME: We should be able to change behavior based on the reference motion vector that will be selected.
// If block_context.y_prediction_mode() != NearMv, then we only need the first motion vector that is added to our result.
// This behavior should combine this function with select_best_reference_motion_vectors(). When that is done, check whether
// the motion vector clamping in that function is always a larger area than in this function. If so, we can drop that call.
bool different_ref_found = false;
u8 context_counter = 0;
Vector<MotionVector, 2> list;
MotionVector base_coordinates = MotionVector(block_context.row, block_context.column);
for (auto i = 0u; i < 2; i++) {
auto offset_vector = mv_ref_blocks[block_context.size][i];
auto candidate = base_coordinates + offset_vector;
if (motion_vector_is_inside_tile(block_context.tile_context, candidate)) {
different_ref_found = true;
auto context = block_context.frame_block_contexts().at(candidate.row(), candidate.column());
context_counter += mode_2_counter[to_underlying(context.y_mode)];
for (auto i = 0u; i < 2; i++) {
auto reference_index = static_cast<ReferenceIndex>(i);
if (context.ref_frames[reference_index] == reference_frame) {
// This section up until add_mv_ref_list() is defined in spec as get_sub_block_mv().
constexpr u8 idx_n_column_to_subblock[4][2] = {
{ 1, 2 },
{ 1, 3 },
{ 3, 2 },
{ 3, 3 }
};
auto index = block >= 0 ? idx_n_column_to_subblock[block][offset_vector.column() == 0] : 3;
add_motion_vector_to_list_deduped(context.sub_block_motion_vectors[index][reference_index], list);
break;
}
}
}
}
block_context.mode_context[reference_frame] = counter_to_context[context_counter];
for (auto i = 2u; i < MVREF_NEIGHBORS; i++) {
MotionVector candidate = base_coordinates + mv_ref_blocks[block_context.size][i];
if (motion_vector_is_inside_tile(block_context.tile_context, candidate)) {
different_ref_found = true;
add_motion_vector_if_reference_frame_type_is_same(block_context, candidate, reference_frame, list, false);
}
}
if (block_context.frame_context.use_previous_frame_motion_vectors)
add_motion_vector_if_reference_frame_type_is_same(block_context, base_coordinates, reference_frame, list, true);
if (different_ref_found) {
for (auto i = 0u; i < MVREF_NEIGHBORS; i++) {
MotionVector candidate = base_coordinates + mv_ref_blocks[block_context.size][i];
if (motion_vector_is_inside_tile(block_context.tile_context, candidate))
add_motion_vector_if_reference_frame_type_is_different(block_context, candidate, reference_frame, list, false);
}
}
if (block_context.frame_context.use_previous_frame_motion_vectors)
add_motion_vector_if_reference_frame_type_is_different(block_context, base_coordinates, reference_frame, list, true);
for (auto i = 0u; i < list.size(); i++) {
// clamp_mv_ref( i ) in the spec.
list[i] = clamp_motion_vector(block_context, list[i], MV_BORDER);
}
MotionVectorPair result;
for (auto i = 0u; i < list.size(); i++)
result[static_cast<ReferenceIndex>(i)] = list[i];
result.primary = clamp_motion_vector(block_context, result.primary, MV_BORDER);
result.secondary = clamp_motion_vector(block_context, result.secondary, MV_BORDER);
return result;
}
// find_best_ref_mvs( refList ) in the spec.
static void select_best_reference_motion_vectors(BlockContext& block_context, MotionVectorPair reference_motion_vectors, BlockMotionVectorCandidates& candidates, ReferenceIndex reference_index)
{
auto adjust_and_clamp_vector = [&](MotionVector& vector) {
auto delta_row = vector.row();
auto delta_column = vector.column();
if (!block_context.frame_context.high_precision_motion_vectors_allowed || !should_use_high_precision_motion_vector(vector)) {
if ((delta_row & 1) != 0)
delta_row += delta_row > 0 ? -1 : 1;
if ((delta_column & 1) != 0)
delta_column += delta_column > 0 ? -1 : 1;
}
vector = { delta_row, delta_column };
vector = clamp_motion_vector(block_context, vector, (BORDERINPIXELS - INTERP_EXTEND) << 3);
};
adjust_and_clamp_vector(reference_motion_vectors.primary);
adjust_and_clamp_vector(reference_motion_vectors.secondary);
candidates[reference_index].nearest_vector = reference_motion_vectors.primary;
candidates[reference_index].near_vector = reference_motion_vectors.secondary;
candidates[reference_index].best_vector = reference_motion_vectors.primary;
}
// append_sub8x8_mvs( block, refList ) in the spec.
void Parser::select_best_sub_block_reference_motion_vectors(BlockContext& block_context, BlockMotionVectorCandidates& candidates, i32 block, ReferenceIndex reference_index)
{
Array<MotionVector, 2> sub_8x8_mvs;
MotionVectorPair reference_motion_vectors = find_reference_motion_vectors(block_context, block_context.reference_frame_types[reference_index], block);
auto destination_index = 0;
if (block == 0) {
sub_8x8_mvs[destination_index++] = reference_motion_vectors.primary;
sub_8x8_mvs[destination_index++] = reference_motion_vectors.secondary;
} else if (block <= 2) {
sub_8x8_mvs[destination_index++] = block_context.sub_block_motion_vectors[0][reference_index];
} else {
sub_8x8_mvs[destination_index++] = block_context.sub_block_motion_vectors[2][reference_index];
for (auto index = 1; index >= 0 && destination_index < 2; index--) {
auto block_vector = block_context.sub_block_motion_vectors[index][reference_index];
if (block_vector != sub_8x8_mvs[0])
sub_8x8_mvs[destination_index++] = block_vector;
}
}
for (auto n = 0u; n < 2 && destination_index < 2; n++) {
auto ref_list_vector = reference_motion_vectors[static_cast<ReferenceIndex>(n)];
if (ref_list_vector != sub_8x8_mvs[0])
sub_8x8_mvs[destination_index++] = ref_list_vector;
}
if (destination_index < 2)
sub_8x8_mvs[destination_index++] = {};
candidates[reference_index].nearest_vector = sub_8x8_mvs[0];
candidates[reference_index].near_vector = sub_8x8_mvs[1];
}
}
Reference in a new issue View git blame Copy permalink