summaryrefslogtreecommitdiff
path: root/vp9/decoder/vp9_decodeframe.c
diff options
context:
space:
mode:
Diffstat (limited to 'vp9/decoder/vp9_decodeframe.c')
-rw-r--r--vp9/decoder/vp9_decodeframe.c165
1 files changed, 49 insertions, 116 deletions
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c
index 728d92dfb..2f15bfb52 100644
--- a/vp9/decoder/vp9_decodeframe.c
+++ b/vp9/decoder/vp9_decodeframe.c
@@ -187,53 +187,13 @@ static void setup_plane_dequants(VP9_COMMON *cm, MACROBLOCKD *xd, int q_index) {
xd->plane[i].dequant = cm->uv_dequant[q_index];
}
-// Allocate storage for each tile column.
-// TODO(jzern): when max_threads <= 1 the same storage could be used for each
-// tile.
-static void alloc_tile_storage(VP9D_COMP *pbi, int tile_rows, int tile_cols) {
- VP9_COMMON *const cm = &pbi->common;
- const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
- int i, tile_row, tile_col;
-
- CHECK_MEM_ERROR(cm, cm->mi_streams,
- vpx_realloc(cm->mi_streams, tile_rows * tile_cols *
- sizeof(*cm->mi_streams)));
- for (tile_row = 0; tile_row < tile_rows; ++tile_row) {
- for (tile_col = 0; tile_col < tile_cols; ++tile_col) {
- TileInfo tile;
- vp9_tile_init(&tile, cm, tile_row, tile_col);
- cm->mi_streams[tile_row * tile_cols + tile_col] =
- &cm->mi[tile.mi_row_start * cm->mode_info_stride + tile.mi_col_start];
- }
- }
-
- // 2 contexts per 'mi unit', so that we have one context per 4x4 txfm
- // block where mi unit size is 8x8.
- CHECK_MEM_ERROR(cm, pbi->above_context[0],
- vpx_realloc(pbi->above_context[0],
- sizeof(*pbi->above_context[0]) * MAX_MB_PLANE *
- 2 * aligned_mi_cols));
- for (i = 1; i < MAX_MB_PLANE; ++i) {
- pbi->above_context[i] = pbi->above_context[0] +
- i * sizeof(*pbi->above_context[0]) *
- 2 * aligned_mi_cols;
- }
-
- // This is sized based on the entire frame. Each tile operates within its
- // column bounds.
- CHECK_MEM_ERROR(cm, pbi->above_seg_context,
- vpx_realloc(pbi->above_seg_context,
- sizeof(*pbi->above_seg_context) *
- aligned_mi_cols));
-}
-
static void inverse_transform_block(MACROBLOCKD* xd, int plane, int block,
TX_SIZE tx_size, uint8_t *dst, int stride,
int eob) {
struct macroblockd_plane *const pd = &xd->plane[plane];
if (eob > 0) {
TX_TYPE tx_type;
- const int plane_type = pd->plane_type;
+ const PLANE_TYPE plane_type = pd->plane_type;
int16_t *const dqcoeff = BLOCK_OFFSET(pd->dqcoeff, block);
switch (tx_size) {
case TX_4X4:
@@ -281,7 +241,7 @@ struct intra_args {
static void predict_and_reconstruct_intra_block(int plane, int block,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
- struct intra_args *const args = arg;
+ struct intra_args *const args = (struct intra_args *)arg;
VP9_COMMON *const cm = args->cm;
MACROBLOCKD *const xd = args->xd;
struct macroblockd_plane *const pd = &xd->plane[plane];
@@ -317,7 +277,7 @@ struct inter_args {
static void reconstruct_inter_block(int plane, int block,
BLOCK_SIZE plane_bsize,
TX_SIZE tx_size, void *arg) {
- struct inter_args *args = arg;
+ struct inter_args *args = (struct inter_args *)arg;
VP9_COMMON *const cm = args->cm;
MACROBLOCKD *const xd = args->xd;
struct macroblockd_plane *const pd = &xd->plane[plane];
@@ -339,13 +299,10 @@ static void set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd,
const int x_mis = MIN(bw, cm->mi_cols - mi_col);
const int y_mis = MIN(bh, cm->mi_rows - mi_row);
const int offset = mi_row * cm->mode_info_stride + mi_col;
- const int tile_offset = tile->mi_row_start * cm->mode_info_stride +
- tile->mi_col_start;
int x, y;
xd->mi_8x8 = cm->mi_grid_visible + offset;
- xd->prev_mi_8x8 = cm->prev_mi_grid_visible + offset;
- xd->mi_8x8[0] = xd->mi_stream + offset - tile_offset;
+ xd->mi_8x8[0] = &cm->mi[offset];
xd->mi_8x8[0]->mbmi.sb_type = bsize;
for (y = 0; y < y_mis; ++y)
for (x = !y; x < x_mis; ++x)
@@ -428,16 +385,14 @@ static void decode_block(VP9_COMMON *const cm, MACROBLOCKD *const xd,
static PARTITION_TYPE read_partition(VP9_COMMON *cm, MACROBLOCKD *xd, int hbs,
int mi_row, int mi_col, BLOCK_SIZE bsize,
vp9_reader *r) {
- const int ctx = partition_plane_context(xd->above_seg_context,
- xd->left_seg_context,
- mi_row, mi_col, bsize);
+ const int ctx = partition_plane_context(xd, mi_row, mi_col, bsize);
const vp9_prob *const probs = get_partition_probs(cm, ctx);
const int has_rows = (mi_row + hbs) < cm->mi_rows;
const int has_cols = (mi_col + hbs) < cm->mi_cols;
PARTITION_TYPE p;
if (has_rows && has_cols)
- p = vp9_read_tree(r, vp9_partition_tree, probs);
+ p = (PARTITION_TYPE)vp9_read_tree(r, vp9_partition_tree, probs);
else if (!has_rows && has_cols)
p = vp9_read(r, probs[1]) ? PARTITION_SPLIT : PARTITION_HORZ;
else if (has_rows && !has_cols)
@@ -495,8 +450,7 @@ static void decode_partition(VP9_COMMON *const cm, MACROBLOCKD *const xd,
// update partition context
if (bsize >= BLOCK_8X8 &&
(bsize == BLOCK_8X8 || partition != PARTITION_SPLIT))
- update_partition_context(xd->above_seg_context, xd->left_seg_context,
- mi_row, mi_col, subsize, bsize);
+ update_partition_context(xd, mi_row, mi_col, subsize, bsize);
}
static void setup_token_decoder(const uint8_t *data,
@@ -664,9 +618,7 @@ static void setup_display_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
read_frame_size(rb, &cm->display_width, &cm->display_height);
}
-static void apply_frame_size(VP9D_COMP *pbi, int width, int height) {
- VP9_COMMON *cm = &pbi->common;
-
+static void apply_frame_size(VP9_COMMON *cm, int width, int height) {
if (cm->width != width || cm->height != height) {
// Change in frame size.
// TODO(agrange) Don't test width/height, check overall size.
@@ -693,18 +645,15 @@ static void apply_frame_size(VP9D_COMP *pbi, int width, int height) {
}
}
-static void setup_frame_size(VP9D_COMP *pbi,
- struct vp9_read_bit_buffer *rb) {
+static void setup_frame_size(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) {
int width, height;
read_frame_size(rb, &width, &height);
- apply_frame_size(pbi, width, height);
- setup_display_size(&pbi->common, rb);
+ apply_frame_size(cm, width, height);
+ setup_display_size(cm, rb);
}
-static void setup_frame_size_with_refs(VP9D_COMP *pbi,
+static void setup_frame_size_with_refs(VP9_COMMON *cm,
struct vp9_read_bit_buffer *rb) {
- VP9_COMMON *const cm = &pbi->common;
-
int width, height;
int found = 0, i;
for (i = 0; i < REFS_PER_FRAME; ++i) {
@@ -724,23 +673,19 @@ static void setup_frame_size_with_refs(VP9D_COMP *pbi,
vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME,
"Referenced frame with invalid size");
- apply_frame_size(pbi, width, height);
+ apply_frame_size(cm, width, height);
setup_display_size(cm, rb);
}
-static void setup_tile_context(VP9D_COMP *const pbi, MACROBLOCKD *const xd,
+static void setup_tile_context(VP9_COMMON *cm, MACROBLOCKD *const xd,
int tile_row, int tile_col) {
- VP9_COMMON *const cm = &pbi->common;
- const int tile_cols = 1 << cm->log2_tile_cols;
int i;
- xd->mi_stream = cm->mi_streams[tile_row * tile_cols + tile_col];
+ for (i = 0; i < MAX_MB_PLANE; ++i)
+ xd->above_context[i] = cm->above_context +
+ i * sizeof(*cm->above_context) * 2 * mi_cols_aligned_to_sb(cm->mi_cols);
- for (i = 0; i < MAX_MB_PLANE; ++i) {
- xd->above_context[i] = pbi->above_context[i];
- }
- // see note in alloc_tile_storage().
- xd->above_seg_context = pbi->above_seg_context;
+ xd->above_seg_context = cm->above_seg_context;
}
static void decode_tile(VP9D_COMP *pbi, const TileInfo *const tile,
@@ -848,7 +793,9 @@ typedef struct TileBuffer {
int col; // only used with multi-threaded decoding
} TileBuffer;
-static const uint8_t *decode_tiles(VP9D_COMP *pbi, const uint8_t *data) {
+static const uint8_t *decode_tiles(VP9D_COMP *pbi,
+ const uint8_t *data,
+ const uint8_t *data_end) {
VP9_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &pbi->mb;
const int aligned_cols = mi_cols_aligned_to_sb(cm->mi_cols);
@@ -856,7 +803,6 @@ static const uint8_t *decode_tiles(VP9D_COMP *pbi, const uint8_t *data) {
const int tile_rows = 1 << cm->log2_tile_rows;
TileBuffer tile_buffers[4][1 << 6];
int tile_row, tile_col;
- const uint8_t *const data_end = pbi->source + pbi->source_sz;
const uint8_t *end = NULL;
vp9_reader r;
@@ -865,11 +811,11 @@ static const uint8_t *decode_tiles(VP9D_COMP *pbi, const uint8_t *data) {
// Note: this memset assumes above_context[0], [1] and [2]
// are allocated as part of the same buffer.
- vpx_memset(pbi->above_context[0], 0,
- sizeof(*pbi->above_context[0]) * MAX_MB_PLANE * 2 * aligned_cols);
+ vpx_memset(cm->above_context, 0,
+ sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_cols);
- vpx_memset(pbi->above_seg_context, 0,
- sizeof(*pbi->above_seg_context) * aligned_cols);
+ vpx_memset(cm->above_seg_context, 0,
+ sizeof(*cm->above_seg_context) * aligned_cols);
// Load tile data into tile_buffers
for (tile_row = 0; tile_row < tile_rows; ++tile_row) {
@@ -896,7 +842,7 @@ static const uint8_t *decode_tiles(VP9D_COMP *pbi, const uint8_t *data) {
vp9_tile_init(&tile, cm, tile_row, col);
setup_token_decoder(buf->data, data_end, buf->size, &cm->error, &r);
- setup_tile_context(pbi, xd, tile_row, col);
+ setup_tile_context(cm, xd, tile_row, col);
decode_tile(pbi, &tile, &r);
if (last_tile)
@@ -907,17 +853,6 @@ static const uint8_t *decode_tiles(VP9D_COMP *pbi, const uint8_t *data) {
return end;
}
-static void setup_tile_macroblockd(TileWorkerData *const tile_data) {
- MACROBLOCKD *xd = &tile_data->xd;
- struct macroblockd_plane *const pd = xd->plane;
- int i;
-
- for (i = 0; i < MAX_MB_PLANE; ++i) {
- pd[i].dqcoeff = tile_data->dqcoeff[i];
- vpx_memset(xd->plane[i].dqcoeff, 0, 64 * 64 * sizeof(int16_t));
- }
-}
-
static int tile_worker_hook(void *arg1, void *arg2) {
TileWorkerData *const tile_data = (TileWorkerData*)arg1;
const TileInfo *const tile = (TileInfo*)arg2;
@@ -949,10 +884,11 @@ static int compare_tile_buffers(const void *a, const void *b) {
}
}
-static const uint8_t *decode_tiles_mt(VP9D_COMP *pbi, const uint8_t *data) {
+static const uint8_t *decode_tiles_mt(VP9D_COMP *pbi,
+ const uint8_t *data,
+ const uint8_t *data_end) {
VP9_COMMON *const cm = &pbi->common;
const uint8_t *bit_reader_end = NULL;
- const uint8_t *const data_end = pbi->source + pbi->source_sz;
const int aligned_mi_cols = mi_cols_aligned_to_sb(cm->mi_cols);
const int tile_cols = 1 << cm->log2_tile_cols;
const int tile_rows = 1 << cm->log2_tile_rows;
@@ -992,11 +928,10 @@ static const uint8_t *decode_tiles_mt(VP9D_COMP *pbi, const uint8_t *data) {
// Note: this memset assumes above_context[0], [1] and [2]
// are allocated as part of the same buffer.
- vpx_memset(pbi->above_context[0], 0,
- sizeof(*pbi->above_context[0]) * MAX_MB_PLANE *
- 2 * aligned_mi_cols);
- vpx_memset(pbi->above_seg_context, 0,
- sizeof(*pbi->above_seg_context) * aligned_mi_cols);
+ vpx_memset(cm->above_context, 0,
+ sizeof(*cm->above_context) * MAX_MB_PLANE * 2 * aligned_mi_cols);
+ vpx_memset(cm->above_seg_context, 0,
+ sizeof(*cm->above_seg_context) * aligned_mi_cols);
// Load tile data into tile_buffers
for (n = 0; n < tile_cols; ++n) {
@@ -1044,8 +979,10 @@ static const uint8_t *decode_tiles_mt(VP9D_COMP *pbi, const uint8_t *data) {
setup_token_decoder(buf->data, data_end, buf->size, &cm->error,
&tile_data->bit_reader);
- setup_tile_context(pbi, &tile_data->xd, 0, buf->col);
- setup_tile_macroblockd(tile_data);
+
+ setup_tile_context(cm, &tile_data->xd, 0, buf->col);
+ init_macroblockd(&tile_data->xd);
+ vp9_zero(tile_data->xd.dqcoeff);
worker->had_error = 0;
if (i == num_workers - 1 || n == tile_cols - 1) {
@@ -1134,7 +1071,7 @@ static size_t read_uncompressed_header(VP9D_COMP *pbi,
if (cm->frame_type == KEY_FRAME) {
check_sync_code(cm, rb);
- cm->color_space = vp9_rb_read_literal(rb, 3); // colorspace
+ cm->color_space = (COLOR_SPACE)vp9_rb_read_literal(rb, 3);
if (cm->color_space != SRGB) {
vp9_rb_read_bit(rb); // [16,235] (including xvycc) vs [0,255] range
if (cm->version == 1) {
@@ -1161,7 +1098,7 @@ static size_t read_uncompressed_header(VP9D_COMP *pbi,
cm->frame_refs[i].buf = get_frame_new_buffer(cm);
}
- setup_frame_size(pbi, rb);
+ setup_frame_size(cm, rb);
} else {
cm->intra_only = cm->show_frame ? 0 : vp9_rb_read_bit(rb);
@@ -1172,7 +1109,7 @@ static size_t read_uncompressed_header(VP9D_COMP *pbi,
check_sync_code(cm, rb);
pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
- setup_frame_size(pbi, rb);
+ setup_frame_size(cm, rb);
} else {
pbi->refresh_frame_flags = vp9_rb_read_literal(rb, REF_FRAMES);
@@ -1184,7 +1121,7 @@ static size_t read_uncompressed_header(VP9D_COMP *pbi,
cm->ref_frame_sign_bias[LAST_FRAME + i] = vp9_rb_read_bit(rb);
}
- setup_frame_size_with_refs(pbi, rb);
+ setup_frame_size_with_refs(cm, rb);
cm->allow_high_precision_mv = vp9_rb_read_bit(rb);
cm->interp_filter = read_interp_filter(rb);
@@ -1332,14 +1269,12 @@ static void debug_check_frame_counts(const VP9_COMMON *const cm) {
}
#endif // NDEBUG
-int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
- int i;
+int vp9_decode_frame(VP9D_COMP *pbi,
+ const uint8_t *data, const uint8_t *data_end,
+ const uint8_t **p_data_end) {
VP9_COMMON *const cm = &pbi->common;
MACROBLOCKD *const xd = &pbi->mb;
- const uint8_t *data = pbi->source;
- const uint8_t *const data_end = pbi->source + pbi->source_sz;
-
struct vp9_read_bit_buffer rb = { data, data_end, 0, cm, error_handler };
const size_t first_partition_size = read_uncompressed_header(pbi, &rb);
const int keyframe = cm->frame_type == KEY_FRAME;
@@ -1365,7 +1300,8 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
pbi->do_loopfilter_inline =
(cm->log2_tile_rows | cm->log2_tile_cols) == 0 && cm->lf.filter_level;
if (pbi->do_loopfilter_inline && pbi->lf_worker.data1 == NULL) {
- CHECK_MEM_ERROR(cm, pbi->lf_worker.data1, vpx_malloc(sizeof(LFWorkerData)));
+ CHECK_MEM_ERROR(cm, pbi->lf_worker.data1,
+ vpx_memalign(32, sizeof(LFWorkerData)));
pbi->lf_worker.hook = (VP9WorkerHook)vp9_loop_filter_worker;
if (pbi->oxcf.max_threads > 1 && !vp9_worker_reset(&pbi->lf_worker)) {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
@@ -1373,8 +1309,6 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
}
}
- alloc_tile_storage(pbi, tile_rows, tile_cols);
-
xd->mode_info_stride = cm->mode_info_stride;
if (cm->coding_use_prev_mi)
set_prev_mi(cm);
@@ -1386,8 +1320,7 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
cm->fc = cm->frame_contexts[cm->frame_context_idx];
vp9_zero(cm->counts);
- for (i = 0; i < MAX_MB_PLANE; ++i)
- vpx_memset(xd->plane[i].dqcoeff, 0, 64 * 64 * sizeof(int16_t));
+ vp9_zero(xd->dqcoeff);
xd->corrupted = 0;
new_fb->corrupted = read_compressed_header(pbi, data, first_partition_size);
@@ -1396,9 +1329,9 @@ int vp9_decode_frame(VP9D_COMP *pbi, const uint8_t **p_data_end) {
// single-frame tile decoding.
if (pbi->oxcf.max_threads > 1 && tile_rows == 1 && tile_cols > 1 &&
cm->frame_parallel_decoding_mode) {
- *p_data_end = decode_tiles_mt(pbi, data + first_partition_size);
+ *p_data_end = decode_tiles_mt(pbi, data + first_partition_size, data_end);
} else {
- *p_data_end = decode_tiles(pbi, data + first_partition_size);
+ *p_data_end = decode_tiles(pbi, data + first_partition_size, data_end);
}
new_fb->corrupted |= xd->corrupted;