LibVideo/VP9: Move the left non-zero tokens context to TileContext

The array containing the vertical line of bools indicating whether non-
zero tokens were decoded in each sub-block is moved to TileContext, and
a span of the valid range for a block to read and write to is created
when we construct a BlockContext.

Userland/Libraries/LibVideo/VP9/Context.h

@@ -9,6 +9,7 @@
 
 #include <AK/Array.h>
 #include <AK/Error.h>
+#include <AK/FixedArray.h>
 #include <LibGfx/Size.h>
 #include <LibVideo/Color/CodingIndependentCodePoints.h>
 
@@ -139,16 +140,49 @@ private:
     Vector2D<FrameBlockContext>& m_block_contexts;
 };
 
+static ErrorOr<NonZeroTokens> create_non_zero_tokens(u32 size_in_sub_blocks, bool subsampling)
+{
+    return NonZeroTokens {
+        TRY(FixedArray<bool>::try_create(size_in_sub_blocks)),
+        TRY(FixedArray<bool>::try_create(size_in_sub_blocks >>= subsampling)),
+        TRY(FixedArray<bool>::try_create(size_in_sub_blocks)),
+    };
+}
+
+template<typename T>
+static Span<T> safe_slice(FixedArray<T>& array, u32 start, u32 size)
+{
+    return array.span().slice(start, min(size, array.size() - start));
+}
+
+static NonZeroTokensView create_non_zero_tokens_view(NonZeroTokens& non_zero_tokens, u32 start_in_sub_blocks, u32 size_in_sub_blocks, bool subsampling)
+{
+    NonZeroTokensView result;
+    // Y plane
+    result[0] = safe_slice(non_zero_tokens[0], start_in_sub_blocks, size_in_sub_blocks);
+    // UV planes
+    start_in_sub_blocks >>= subsampling;
+    size_in_sub_blocks >>= subsampling;
+    result[1] = safe_slice(non_zero_tokens[1], start_in_sub_blocks, size_in_sub_blocks);
+    result[2] = safe_slice(non_zero_tokens[2], start_in_sub_blocks, size_in_sub_blocks);
+    return result;
+}
+
 struct TileContext {
 public:
-    TileContext(FrameContext& frame_context, u32 rows_start, u32 rows_end, u32 columns_start, u32 columns_end)
-        : frame_context(frame_context)
-        , rows_start(rows_start)
-        , rows_end(rows_end)
-        , columns_start(columns_start)
-        , columns_end(columns_end)
-        , block_contexts_view(frame_context.m_block_contexts.view(rows_start, columns_start, rows_end - rows_start, columns_end - columns_start))
+    static ErrorOr<TileContext> try_create(FrameContext& frame_context, u32 rows_start, u32 rows_end, u32 columns_start, u32 columns_end)
     {
+        auto context_view = frame_context.m_block_contexts.view(rows_start, columns_start, rows_end - rows_start, columns_end - columns_start);
+
+        return TileContext {
+            frame_context,
+            rows_start,
+            rows_end,
+            columns_start,
+            columns_end,
+            context_view,
+            TRY(create_non_zero_tokens(blocks_to_sub_blocks(rows_end - rows_start), frame_context.color_config.subsampling_y)),
+        };
     }
 
     Vector2D<FrameBlockContext> const& frame_block_contexts() const { return frame_context.block_contexts(); }
@@ -158,20 +192,33 @@ public:
     u32 rows_end { 0 };
     u32 columns_start { 0 };
     u32 columns_end { 0 };
+    u32 rows() const { return rows_end - rows_start; }
+    u32 columns() const { return columns_end - columns_start; }
     Vector2DView<FrameBlockContext> block_contexts_view;
+
+    NonZeroTokens left_non_zero_tokens;
 };
 
 struct BlockContext {
-    BlockContext(TileContext& tile_context, u32 row, u32 column, BlockSubsize size)
-        : frame_context(tile_context.frame_context)
-        , tile_context(tile_context)
-        , row(row)
-        , column(column)
-        , size(size)
-        , contexts_view(tile_context.block_contexts_view.view(row - tile_context.rows_start, column - tile_context.columns_start,
-              min<u32>(num_8x8_blocks_high_lookup[size], tile_context.frame_context.rows() - row),
-              min<u32>(num_8x8_blocks_wide_lookup[size], tile_context.frame_context.columns() - column)))
+    static BlockContext create(TileContext& tile_context, u32 row, u32 column, BlockSubsize size)
     {
+        auto contexts_view = tile_context.block_contexts_view.view(
+            row - tile_context.rows_start,
+            column - tile_context.columns_start,
+            min<u32>(num_8x8_blocks_high_lookup[size], tile_context.frame_context.rows() - row),
+            min<u32>(num_8x8_blocks_wide_lookup[size], tile_context.frame_context.columns() - column));
+
+        auto size_in_sub_blocks = block_size_to_sub_blocks(get_subsampled_block_size(size, false, false));
+
+        return BlockContext {
+            .frame_context = tile_context.frame_context,
+            .tile_context = tile_context,
+            .row = row,
+            .column = column,
+            .size = size,
+            .contexts_view = contexts_view,
+            .left_non_zero_tokens = create_non_zero_tokens_view(tile_context.left_non_zero_tokens, blocks_to_sub_blocks(row - tile_context.rows_start), size_in_sub_blocks.height(), tile_context.frame_context.color_config.subsampling_y),
+        };
     }
 
     Vector2D<FrameBlockContext> const& frame_block_contexts() const { return frame_context.block_contexts(); }
@@ -193,22 +240,24 @@ struct BlockContext {
 
     TransformSize transform_size { Transform_4x4 };
 
-    ReferenceFramePair reference_frame_types;
+    ReferenceFramePair reference_frame_types {};
     bool is_inter_predicted() const { return reference_frame_types.primary != ReferenceFrameType::None; }
     bool is_compound() const { return reference_frame_types.secondary != ReferenceFrameType::None; }
 
-    Array<PredictionMode, 4> sub_block_prediction_modes;
+    Array<PredictionMode, 4> sub_block_prediction_modes {};
     PredictionMode y_prediction_mode() const { return sub_block_prediction_modes.last(); }
     PredictionMode& y_prediction_mode() { return sub_block_prediction_modes.last(); }
     PredictionMode uv_prediction_mode { 0 };
 
     InterpolationFilter interpolation_filter { EightTap };
-    Array<MotionVectorPair, 4> sub_block_motion_vectors;
+    Array<MotionVectorPair, 4> sub_block_motion_vectors {};
 
-    Array<i32, 1024> residual_tokens;
+    Array<i32, 1024> residual_tokens {};
 
     // Indexed by ReferenceFrame enum.
-    Array<u8, 4> mode_context;
+    Array<u8, 4> mode_context {};
+
+    NonZeroTokensView left_non_zero_tokens;
 };
 
 struct BlockMotionVectorCandidateSet {

Userland/Libraries/LibVideo/VP9/ContextStorage.h

@@ -248,4 +248,7 @@ struct ColorConfig {
 struct BlockMotionVectorCandidateSet;
 using BlockMotionVectorCandidates = ReferencePair<BlockMotionVectorCandidateSet>;
 
+using NonZeroTokens = Array<FixedArray<bool>, 3>;
+using NonZeroTokensView = Array<Span<bool>, 3>;
+
 }

Userland/Libraries/LibVideo/VP9/Parser.cpp

@@ -864,7 +864,7 @@ DecoderErrorOr<void> Parser::decode_tiles(FrameContext& frame_context)
             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 tile_context = TileContext(frame_context, rows_start, rows_end, columns_start, columns_end);
+            auto tile_context = DECODER_TRY_ALLOC(TileContext::try_create(frame_context, rows_start, rows_end, columns_start, columns_end));
 
             TRY_READ(m_bit_stream->init_bool(tile_size));
             TRY(decode_tile(tile_context));
@@ -918,8 +918,8 @@ DecoderErrorOr<void> Parser::decode_tile(TileContext& tile_context)
 
 void Parser::clear_left_context(TileContext& tile_context)
 {
-    for (auto i = 0u; i < m_left_nonzero_context.size(); i++)
-        clear_context(m_left_nonzero_context[i], 2 * tile_context.frame_context.rows());
+    for (auto& context_for_plane : tile_context.left_non_zero_tokens)
+        context_for_plane.fill_with(false);
     clear_context(m_left_seg_pred_context, tile_context.frame_context.rows());
     clear_context(m_left_partition_context, tile_context.frame_context.superblock_rows() * 8);
 }
@@ -972,7 +972,7 @@ DecoderErrorOr<void> Parser::decode_block(TileContext& tile_context, u32 row, u3
 {
     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(tile_context, row, column, subsize);
+    auto block_context = BlockContext::create(tile_context, row, column, subsize);
 
     TRY(mode_info(block_context, above_context, left_context));
     auto had_residual_tokens = TRY(residual(block_context, above_context.is_available, left_context.is_available));
@@ -1355,9 +1355,9 @@ DecoderErrorOr<bool> Parser::residual(BlockContext& block_context, bool has_bloc
     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 : 0;
-        auto plane_subsampling_y = (plane > 0) ? block_context.frame_context.color_config.subsampling_y : 0;
-        auto plane_size = ss_size_lookup[block_context.size < Block_8x8 ? Block_8x8 : block_context.size][plane_subsampling_x][plane_subsampling_y];
+        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);
         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);
@@ -1380,8 +1380,8 @@ DecoderErrorOr<bool> Parser::residual(BlockContext& block_context, bool has_bloc
         auto frame_bottom_in_pixels = (blocks_to_pixels(block_context.frame_context.rows())) >> plane_subsampling_y;
 
         auto sub_block_index = 0;
-        for (auto y = 0; y < block_size_in_sub_blocks.height(); y += transform_size_in_sub_blocks) {
-            for (auto x = 0; x < block_size_in_sub_blocks.width(); x += transform_size_in_sub_blocks) {
+        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);
 
@@ -1403,11 +1403,10 @@ DecoderErrorOr<bool> Parser::residual(BlockContext& block_context, bool has_bloc
                 for (; above_sub_block_context_index < above_sub_block_context_end; above_sub_block_context_index++)
                     above_sub_block_context[above_sub_block_context_index] = sub_block_had_non_zero_tokens;
 
-                auto& left_sub_block_context = m_left_nonzero_context[plane];
-                auto left_sub_block_context_index = pixels_to_sub_blocks(transform_y_in_px);
-                auto left_sub_block_context_end = min(left_sub_block_context_index + transform_size_in_sub_blocks, left_sub_block_context.size());
-                for (; left_sub_block_context_index < left_sub_block_context_end; left_sub_block_context_index++)
-                    left_sub_block_context[left_sub_block_context_index] = 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++;
             }
@@ -1459,7 +1458,7 @@ DecoderErrorOr<bool> Parser::tokens(BlockContext& block_context, size_t plane, u
         auto token_position = scan[coef_index];
         TokensContext tokens_context;
         if (coef_index == 0)
-            tokens_context = TreeParser::get_context_for_first_token(block_context, m_above_nonzero_context, m_left_nonzero_context, transform_size, plane, sub_block_column, sub_block_row, block_context.is_inter_predicted(), band);
+            tokens_context = TreeParser::get_context_for_first_token(block_context, m_above_nonzero_context, 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);
 

Userland/Libraries/LibVideo/VP9/Parser.h

@@ -148,7 +148,6 @@ private:
 
     // FIXME: Move above and left contexts to structs
     Array<Vector<bool>, 3> m_above_nonzero_context;
-    Array<Vector<bool>, 3> m_left_nonzero_context;
     Vector<u8> m_above_seg_pred_context;
     Vector<u8> m_left_seg_pred_context;
     Vector<u8> m_above_partition_context;

Userland/Libraries/LibVideo/VP9/TreeParser.cpp

@@ -626,11 +626,9 @@ ErrorOr<bool> TreeParser::parse_motion_vector_hp(BitStream& bit_stream, Probabil
     return value;
 }
 
-TokensContext TreeParser::get_context_for_first_token(BlockContext const& block_context, Array<Vector<bool>, 3> const& above_non_zero_tokens, Array<Vector<bool>, 3> const& left_non_zero_tokens, TransformSize transform_size, u8 plane, u32 sub_block_column, u32 sub_block_row, bool is_inter, u8 band)
+TokensContext TreeParser::get_context_for_first_token(BlockContext const& block_context, Array<Vector<bool>, 3> const& above_non_zero_tokens, NonZeroTokensView left_non_zero_tokens_in_block, TransformSize transform_size, u8 plane, u32 sub_block_column, u32 sub_block_row, bool is_inter, u8 band)
 {
     auto subsampling_x = plane > 0 ? block_context.frame_context.color_config.subsampling_x : false;
-    auto subsampling_y = plane > 0 ? block_context.frame_context.color_config.subsampling_y : false;
-    auto transform_top_in_sub_blocks = (blocks_to_sub_blocks(block_context.row) >> subsampling_y) + sub_block_row;
     auto transform_left_in_sub_blocks = (blocks_to_sub_blocks(block_context.column) >> subsampling_x) + sub_block_column;
     u8 transform_size_in_sub_blocks = transform_size_to_sub_blocks(transform_size);
     bool above_has_non_zero_tokens = false;
@@ -641,8 +639,8 @@ TokensContext TreeParser::get_context_for_first_token(BlockContext const& block_
         }
     }
     bool left_has_non_zero_tokens = false;
-    for (u8 y = 0; y < transform_size_in_sub_blocks && y < left_non_zero_tokens[plane].size() - transform_top_in_sub_blocks; y++) {
-        if (left_non_zero_tokens[plane][transform_top_in_sub_blocks + y]) {
+    for (u8 y = 0; y < transform_size_in_sub_blocks && y < left_non_zero_tokens_in_block[plane].size() - sub_block_row; y++) {
+        if (left_non_zero_tokens_in_block[plane][sub_block_row + y]) {
             left_has_non_zero_tokens = true;
             break;
         }

Userland/Libraries/LibVideo/VP9/TreeParser.h

@@ -87,7 +87,7 @@ public:
     static ErrorOr<u8> parse_motion_vector_fr(BitStream&, ProbabilityTables const&, SyntaxElementCounter&, u8 component);
     static ErrorOr<bool> parse_motion_vector_hp(BitStream&, ProbabilityTables const&, SyntaxElementCounter&, u8 component, bool use_hp);
 
-    static TokensContext get_context_for_first_token(BlockContext const& block_context, Array<Vector<bool>, 3> const& above_non_zero_tokens, Array<Vector<bool>, 3> const& left_non_zero_tokens, TransformSize transform_size, u8 plane, u32 sub_block_column, u32 sub_block_row, bool is_inter, u8 band);
+    static TokensContext get_context_for_first_token(BlockContext const& block_context, Array<Vector<bool>, 3> const& above_non_zero_tokens, NonZeroTokensView left_non_zero_tokens, TransformSize transform_size, u8 plane, u32 sub_block_column, u32 sub_block_row, bool is_inter, u8 band);
     static TokensContext get_context_for_other_tokens(Array<u8, 1024> token_cache, TransformSize transform_size, TransformSet transform_set, u8 plane, u16 token_position, bool is_inter, u8 band);
     static ErrorOr<bool> parse_more_coefficients(BitStream&, ProbabilityTables const&, SyntaxElementCounter&, TokensContext const& context);
     static ErrorOr<Token> parse_token(BitStream&, ProbabilityTables const&, SyntaxElementCounter&, TokensContext const& context);

Userland/Libraries/LibVideo/VP9/Utilities.h

@@ -45,6 +45,11 @@ inline T brev(C bit_count, T value)
     return result;
 }
 
+inline BlockSubsize get_subsampled_block_size(BlockSubsize size, bool subsampling_x, bool subsampling_y)
+{
+    return ss_size_lookup[size < Block_8x8 ? Block_8x8 : size][subsampling_x][subsampling_y];
+}
+
 inline Gfx::Size<u8> block_size_to_sub_blocks(BlockSubsize size)
 {
     return Gfx::Size<u8>(num_4x4_blocks_wide_lookup[size], num_4x4_blocks_high_lookup[size]);