vpx_codec_err_t vp9_copy_reference_dec(VP9D_PTR ptr,
VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
VP9D_COMP *pbi = (VP9D_COMP *) ptr;
VP9_COMMON *cm = &pbi->common;
/* TODO(jkoleszar): The decoder doesn't have any real knowledge of what the
* encoder is using the frame buffers for. This is just a stub to keep the
* vpxenc --test-decode functionality working, and will be replaced in a
* later commit that adds VP9-specific controls for this functionality.
*/
if (ref_frame_flag == VP9_LAST_FLAG) {
YV12_BUFFER_CONFIG *cfg = &cm->yv12_fb[cm->ref_frame_map[0]];
if (!equal_dimensions(cfg, sd))
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Incorrect buffer dimensions");
else
vp8_yv12_copy_frame(cfg, sd);
} else {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Invalid reference frame");
}
return cm->error.error_code;
}
vpx_codec_err_t vp8dx_get_reference(VP8D_COMP *pbi, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd)
{
VP8_COMMON *cm = &pbi->common;
int ref_fb_idx;
if (ref_frame_flag == VP8_LAST_FRAME)
ref_fb_idx = cm->lst_fb_idx;
else if (ref_frame_flag == VP8_GOLD_FRAME)
ref_fb_idx = cm->gld_fb_idx;
else if (ref_frame_flag == VP8_ALTR_FRAME)
ref_fb_idx = cm->alt_fb_idx;
else{
vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
"Invalid reference frame");
return pbi->common.error.error_code;
}
if(cm->yv12_fb[ref_fb_idx].y_height != sd->y_height ||
cm->yv12_fb[ref_fb_idx].y_width != sd->y_width ||
cm->yv12_fb[ref_fb_idx].uv_height != sd->uv_height ||
cm->yv12_fb[ref_fb_idx].uv_width != sd->uv_width){
vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
"Incorrect buffer dimensions");
}
else
vp8_yv12_copy_frame(&cm->yv12_fb[ref_fb_idx], sd);
return pbi->common.error.error_code;
}
static void create_enc_workers(VP9_COMP *cpi, int num_workers) {
VP9_COMMON *const cm = &cpi->common;
const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
int i;
// Only run once to create threads and allocate thread data.
if (cpi->num_workers == 0) {
int allocated_workers = num_workers;
// While using SVC, we need to allocate threads according to the highest
// resolution. When row based multithreading is enabled, it is OK to
// allocate more threads than the number of max tile columns.
if (cpi->use_svc && !cpi->row_mt) {
int max_tile_cols = get_max_tile_cols(cpi);
allocated_workers = VPXMIN(cpi->oxcf.max_threads, max_tile_cols);
}
CHECK_MEM_ERROR(cm, cpi->workers,
vpx_malloc(allocated_workers * sizeof(*cpi->workers)));
CHECK_MEM_ERROR(cm, cpi->tile_thr_data,
vpx_calloc(allocated_workers, sizeof(*cpi->tile_thr_data)));
for (i = 0; i < allocated_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *thread_data = &cpi->tile_thr_data[i];
++cpi->num_workers;
winterface->init(worker);
if (i < allocated_workers - 1) {
thread_data->cpi = cpi;
// Allocate thread data.
CHECK_MEM_ERROR(cm, thread_data->td,
vpx_memalign(32, sizeof(*thread_data->td)));
vp9_zero(*thread_data->td);
// Set up pc_tree.
thread_data->td->leaf_tree = NULL;
thread_data->td->pc_tree = NULL;
vp9_setup_pc_tree(cm, thread_data->td);
// Allocate frame counters in thread data.
CHECK_MEM_ERROR(cm, thread_data->td->counts,
vpx_calloc(1, sizeof(*thread_data->td->counts)));
// Create threads
if (!winterface->reset(worker))
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Tile encoder thread creation failed");
} else {
// Main thread acts as a worker and uses the thread data in cpi.
thread_data->cpi = cpi;
thread_data->td = &cpi->td;
}
winterface->sync(worker);
}
}
}
static void init_buffer_callbacks(vpx_codec_alg_priv_t *ctx) {
int i;
for (i = 0; i < ctx->num_frame_workers; ++i) {
VPxWorker *const worker = &ctx->frame_workers[i];
FrameWorkerData *const frame_worker_data = (FrameWorkerData *)worker->data1;
VP9_COMMON *const cm = &frame_worker_data->pbi->common;
BufferPool *const pool = cm->buffer_pool;
cm->new_fb_idx = INVALID_IDX;
cm->byte_alignment = ctx->byte_alignment;
cm->skip_loop_filter = ctx->skip_loop_filter;
if (ctx->get_ext_fb_cb != NULL && ctx->release_ext_fb_cb != NULL) {
pool->get_fb_cb = ctx->get_ext_fb_cb;
pool->release_fb_cb = ctx->release_ext_fb_cb;
pool->cb_priv = ctx->ext_priv;
} else {
pool->get_fb_cb = vp9_get_frame_buffer;
pool->release_fb_cb = vp9_release_frame_buffer;
if (vp9_alloc_internal_frame_buffers(&pool->int_frame_buffers))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to initialize internal frame buffers");
pool->cb_priv = &pool->int_frame_buffers;
}
}
}
void vp9_create_encoding_threads(VP9_COMP *cpi) {
VP9_COMMON * const cm = &cpi->common;
const VP9WorkerInterface * const winterface = vp9_get_worker_interface();
int i;
CHECK_MEM_ERROR(cm, cpi->enc_thread_hndl,
vpx_malloc(sizeof(*cpi->enc_thread_hndl) * cpi->max_threads));
for (i = 0; i < cpi->max_threads; ++i) {
VP9Worker * const worker = &cpi->enc_thread_hndl[i];
winterface->init(worker);
CHECK_MEM_ERROR(cm, worker->data1,
vpx_memalign(32, sizeof(thread_context)));
worker->data2 = NULL;
if (i < cpi->max_threads - 1 && !winterface->reset(worker)) {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Tile decoder thread creation failed");
}
}
// set row encoding hook
for (i = 0; i < cpi->max_threads; ++i) {
winterface->sync(&cpi->enc_thread_hndl[i]);
cpi->enc_thread_hndl[i].hook = (VP9WorkerHook) encoding_thread_process;
}
CHECK_MEM_ERROR(cm, cpi->cur_sb_col,
vpx_malloc(sizeof(*cpi->cur_sb_col) * cm->sb_rows));
// init cur sb col
vpx_memset(cpi->cur_sb_col, -1, (sizeof(*cpi->cur_sb_col) * cm->sb_rows));
// set up nsync (currently unused).
cpi->sync_range = get_sync_range(cpi->oxcf.width);
}
static unsigned int read_available_partition_size(
VP8D_COMP *pbi,
const unsigned char *token_part_sizes,
const unsigned char *fragment_start,
const unsigned char *first_fragment_end,
const unsigned char *fragment_end,
int i,
int num_part)
{
VP8_COMMON* pc = &pbi->common;
const unsigned char *partition_size_ptr = token_part_sizes + i * 3;
unsigned int partition_size = 0;
ptrdiff_t bytes_left = fragment_end - fragment_start;
/* Calculate the length of this partition. The last partition
* size is implicit. If the partition size can't be read, then
* either use the remaining data in the buffer (for EC mode)
* or throw an error.
*/
if (i < num_part - 1)
{
if (read_is_valid(partition_size_ptr, 3, first_fragment_end))
partition_size = read_partition_size(pbi, partition_size_ptr);
else if (pbi->ec_active)
partition_size = (unsigned int)bytes_left;
else
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Truncated partition size data");
}
else
partition_size = (unsigned int)bytes_left;
/* Validate the calculated partition length. If the buffer
* described by the partition can't be fully read, then restrict
* it to the portion that can be (for EC mode) or throw an error.
*/
if (!read_is_valid(fragment_start, partition_size, fragment_end))
{
if (pbi->ec_active)
partition_size = (unsigned int)bytes_left;
else
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Truncated packet or corrupt partition "
"%d length", i + 1);
}
return partition_size;
}
vpx_codec_err_t vp10_set_reference_dec(VP10_COMMON *cm,
VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
RefBuffer *ref_buf = NULL;
RefCntBuffer *const frame_bufs = cm->buffer_pool->frame_bufs;
// TODO(jkoleszar): The decoder doesn't have any real knowledge of what the
// encoder is using the frame buffers for. This is just a stub to keep the
// vpxenc --test-decode functionality working, and will be replaced in a
// later commit that adds VP9-specific controls for this functionality.
if (ref_frame_flag == VP9_LAST_FLAG) {
ref_buf = &cm->frame_refs[0];
} else if (ref_frame_flag == VP9_GOLD_FLAG) {
ref_buf = &cm->frame_refs[1];
} else if (ref_frame_flag == VP9_ALT_FLAG) {
ref_buf = &cm->frame_refs[2];
} else {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Invalid reference frame");
return cm->error.error_code;
}
if (!equal_dimensions(ref_buf->buf, sd)) {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Incorrect buffer dimensions");
} else {
int *ref_fb_ptr = &ref_buf->idx;
// Find an empty frame buffer.
const int free_fb = get_free_fb(cm);
if (cm->new_fb_idx == INVALID_IDX)
return VPX_CODEC_MEM_ERROR;
// Decrease ref_count since it will be increased again in
// ref_cnt_fb() below.
--frame_bufs[free_fb].ref_count;
// Manage the reference counters and copy image.
ref_cnt_fb(frame_bufs, ref_fb_ptr, free_fb);
ref_buf->buf = &frame_bufs[*ref_fb_ptr].buf;
vp8_yv12_copy_frame(sd, ref_buf->buf);
}
return cm->error.error_code;
}
vpx_codec_err_t vp9_set_reference_dec(VP9_COMMON *cm,
VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
int idx;
YV12_BUFFER_CONFIG *ref_buf = NULL;
// TODO(jkoleszar): The decoder doesn't have any real knowledge of what the
// encoder is using the frame buffers for. This is just a stub to keep the
// vpxenc --test-decode functionality working, and will be replaced in a
// later commit that adds VP9-specific controls for this functionality.
// (Yunqing) The set_reference control depends on the following setting in
// encoder.
// cpi->lst_fb_idx = 0;
// cpi->gld_fb_idx = 1;
// cpi->alt_fb_idx = 2;
if (ref_frame_flag == VP9_LAST_FLAG) {
idx = cm->ref_frame_map[0];
} else if (ref_frame_flag == VP9_GOLD_FLAG) {
idx = cm->ref_frame_map[1];
} else if (ref_frame_flag == VP9_ALT_FLAG) {
idx = cm->ref_frame_map[2];
} else {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR, "Invalid reference frame");
return cm->error.error_code;
}
if (idx < 0 || idx >= FRAME_BUFFERS) {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Invalid reference frame map");
return cm->error.error_code;
}
// Get the destination reference buffer.
ref_buf = &cm->buffer_pool->frame_bufs[idx].buf;
if (!equal_dimensions(ref_buf, sd)) {
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Incorrect buffer dimensions");
} else {
// Overwrite the reference frame buffer.
vpx_yv12_copy_frame(sd, ref_buf);
}
return cm->error.error_code;
}
vpx_codec_err_t vp9_set_reference_dec(VP9D_PTR ptr, VP9_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd) {
VP9D_COMP *pbi = (VP9D_COMP *) ptr;
VP9_COMMON *cm = &pbi->common;
int *ref_fb_ptr = NULL;
/* TODO(jkoleszar): The decoder doesn't have any real knowledge of what the
* encoder is using the frame buffers for. This is just a stub to keep the
* vpxenc --test-decode functionality working, and will be replaced in a
* later commit that adds VP9-specific controls for this functionality.
*/
if (ref_frame_flag == VP9_LAST_FLAG) {
ref_fb_ptr = &pbi->common.active_ref_idx[0];
} else if (ref_frame_flag == VP9_GOLD_FLAG) {
ref_fb_ptr = &pbi->common.active_ref_idx[1];
} else if (ref_frame_flag == VP9_ALT_FLAG) {
ref_fb_ptr = &pbi->common.active_ref_idx[2];
} else {
vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
"Invalid reference frame");
return pbi->common.error.error_code;
}
if (!equal_dimensions(&cm->yv12_fb[*ref_fb_ptr], sd)) {
vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
"Incorrect buffer dimensions");
} else {
// Find an empty frame buffer.
const int free_fb = get_free_fb(cm);
// Decrease fb_idx_ref_cnt since it will be increased again in
// ref_cnt_fb() below.
cm->fb_idx_ref_cnt[free_fb]--;
// Manage the reference counters and copy image.
ref_cnt_fb(cm->fb_idx_ref_cnt, ref_fb_ptr, free_fb);
vp8_yv12_copy_frame(sd, &cm->yv12_fb[*ref_fb_ptr]);
}
return pbi->common.error.error_code;
}
vpx_codec_err_t vp8dx_set_reference(VP8D_COMP *pbi, enum vpx_ref_frame_type ref_frame_flag, YV12_BUFFER_CONFIG *sd)
{
VP8_COMMON *cm = &pbi->common;
int *ref_fb_ptr = NULL;
int free_fb;
if (ref_frame_flag == VP8_LAST_FRAME)
ref_fb_ptr = &cm->lst_fb_idx;
else if (ref_frame_flag == VP8_GOLD_FRAME)
ref_fb_ptr = &cm->gld_fb_idx;
else if (ref_frame_flag == VP8_ALTR_FRAME)
ref_fb_ptr = &cm->alt_fb_idx;
else{
vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
"Invalid reference frame");
return pbi->common.error.error_code;
}
if(cm->yv12_fb[*ref_fb_ptr].y_height != sd->y_height ||
cm->yv12_fb[*ref_fb_ptr].y_width != sd->y_width ||
cm->yv12_fb[*ref_fb_ptr].uv_height != sd->uv_height ||
cm->yv12_fb[*ref_fb_ptr].uv_width != sd->uv_width){
vpx_internal_error(&pbi->common.error, VPX_CODEC_ERROR,
"Incorrect buffer dimensions");
}
else{
/* Find an empty frame buffer. */
free_fb = get_free_fb(cm);
/* Decrease fb_idx_ref_cnt since it will be increased again in
* ref_cnt_fb() below. */
cm->fb_idx_ref_cnt[free_fb]--;
/* Manage the reference counters and copy image. */
ref_cnt_fb (cm->fb_idx_ref_cnt, ref_fb_ptr, free_fb);
vp8_yv12_copy_frame(sd, &cm->yv12_fb[*ref_fb_ptr]);
}
return pbi->common.error.error_code;
}
// TODO(hkuang): Remove worker parameter as it is only used in debug code.
void vp10_frameworker_wait(VPxWorker *const worker, RefCntBuffer *const ref_buf,
int row) {
#if CONFIG_MULTITHREAD
if (!ref_buf) return;
#ifndef BUILDING_WITH_TSAN
// The following line of code will get harmless tsan error but it is the key
// to get best performance.
if (ref_buf->row >= row && ref_buf->buf.corrupted != 1) return;
#endif
{
// Find the worker thread that owns the reference frame. If the reference
// frame has been fully decoded, it may not have owner.
VPxWorker *const ref_worker = ref_buf->frame_worker_owner;
FrameWorkerData *const ref_worker_data =
(FrameWorkerData *)ref_worker->data1;
const VP10Decoder *const pbi = ref_worker_data->pbi;
#ifdef DEBUG_THREAD
{
FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
printf("%d %p worker is waiting for %d %p worker (%d) ref %d \r\n",
worker_data->worker_id, worker, ref_worker_data->worker_id,
ref_buf->frame_worker_owner, row, ref_buf->row);
}
#endif
vp10_frameworker_lock_stats(ref_worker);
while (ref_buf->row < row && pbi->cur_buf == ref_buf &&
ref_buf->buf.corrupted != 1) {
pthread_cond_wait(&ref_worker_data->stats_cond,
&ref_worker_data->stats_mutex);
}
if (ref_buf->buf.corrupted == 1) {
FrameWorkerData *const worker_data = (FrameWorkerData *)worker->data1;
vp10_frameworker_unlock_stats(ref_worker);
vpx_internal_error(&worker_data->pbi->common.error,
VPX_CODEC_CORRUPT_FRAME,
"Worker %p failed to decode frame", worker);
}
vp10_frameworker_unlock_stats(ref_worker);
}
#else
(void)worker;
(void)ref_buf;
(void)row;
(void)ref_buf;
#endif // CONFIG_MULTITHREAD
}
static void init_buffer_callbacks(vpx_codec_alg_priv_t *ctx) {
VP9_COMMON *const cm = &ctx->pbi->common;
cm->new_fb_idx = -1;
if (ctx->get_ext_fb_cb != NULL && ctx->release_ext_fb_cb != NULL) {
cm->get_fb_cb = ctx->get_ext_fb_cb;
cm->release_fb_cb = ctx->release_ext_fb_cb;
cm->cb_priv = ctx->ext_priv;
} else {
cm->get_fb_cb = vp9_get_frame_buffer;
cm->release_fb_cb = vp9_release_frame_buffer;
if (vp9_alloc_internal_frame_buffers(&cm->int_frame_buffers))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to initialize internal frame buffers");
cm->cb_priv = &cm->int_frame_buffers;
}
}
void vp9_setup_version(VP9_COMMON *cm) {
if (cm->version & 0x4) {
if (!CONFIG_EXPERIMENTAL)
vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
"Bitstream was created by an experimental "
"encoder");
cm->experimental = 1;
}
switch (cm->version & 0x3) {
case 0:
cm->no_lpf = 0;
cm->filter_type = NORMAL_LOOPFILTER;
cm->use_bilinear_mc_filter = 0;
cm->full_pixel = 0;
break;
case 1:
cm->no_lpf = 0;
cm->filter_type = SIMPLE_LOOPFILTER;
cm->use_bilinear_mc_filter = 1;
cm->full_pixel = 0;
break;
case 2:
case 3:
cm->no_lpf = 1;
cm->filter_type = NORMAL_LOOPFILTER;
cm->use_bilinear_mc_filter = 1;
cm->full_pixel = 0;
break;
// Full pel only code deprecated in experimental code base
// case 3:
// cm->no_lpf = 1;
// cm->filter_type = SIMPLE_LOOPFILTER;
// cm->use_bilinear_mc_filter = 1;
// cm->full_pixel = 1;
// break;
}
}
static void read_inter_block_mode_info(VP9_COMMON *const cm,
MACROBLOCKD *const xd,
const TileInfo *const tile,
MODE_INFO *const mi,
int mi_row, int mi_col, vp9_reader *r) {
MB_MODE_INFO *const mbmi = &mi->mbmi;
const BLOCK_SIZE bsize = mbmi->sb_type;
const int allow_hp = cm->allow_high_precision_mv;
int_mv nearest[2], nearmv[2], best[2];
int inter_mode_ctx, ref, is_compound;
read_ref_frames(cm, xd, r, mbmi->segment_id, mbmi->ref_frame);
is_compound = has_second_ref(mbmi);
for (ref = 0; ref < 1 + is_compound; ++ref) {
const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
vp9_find_mv_refs(cm, xd, tile, mi, xd->last_mi, frame, mbmi->ref_mvs[frame],
mi_row, mi_col);
}
inter_mode_ctx = mbmi->mode_context[mbmi->ref_frame[0]];
if (vp9_segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
mbmi->mode = ZEROMV;
if (bsize < BLOCK_8X8) {
vpx_internal_error(&cm->error, VPX_CODEC_UNSUP_BITSTREAM,
"Invalid usage of segement feature on small blocks");
return;
}
} else {
if (bsize >= BLOCK_8X8)
mbmi->mode = read_inter_mode(cm, r, inter_mode_ctx);
}
if (bsize < BLOCK_8X8 || mbmi->mode != ZEROMV) {
for (ref = 0; ref < 1 + is_compound; ++ref) {
vp9_find_best_ref_mvs(xd, allow_hp, mbmi->ref_mvs[mbmi->ref_frame[ref]],
&nearest[ref], &nearmv[ref]);
best[ref].as_int = nearest[ref].as_int;
}
}
mbmi->interp_filter = (cm->mcomp_filter_type == SWITCHABLE)
? read_switchable_filter_type(cm, xd, r)
: cm->mcomp_filter_type;
if (bsize < BLOCK_8X8) {
const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize]; // 1 or 2
const int num_4x4_h = num_4x4_blocks_high_lookup[bsize]; // 1 or 2
int idx, idy;
int b_mode;
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
int_mv block[2];
const int j = idy * 2 + idx;
b_mode = read_inter_mode(cm, r, inter_mode_ctx);
if (b_mode == NEARESTMV || b_mode == NEARMV)
for (ref = 0; ref < 1 + is_compound; ++ref)
vp9_append_sub8x8_mvs_for_idx(cm, xd, tile, j, ref, mi_row, mi_col,
&nearest[ref], &nearmv[ref]);
if (!assign_mv(cm, b_mode, block, best, nearest, nearmv,
is_compound, allow_hp, r)) {
xd->corrupted |= 1;
break;
};
mi->bmi[j].as_mv[0].as_int = block[0].as_int;
if (is_compound)
mi->bmi[j].as_mv[1].as_int = block[1].as_int;
if (num_4x4_h == 2)
mi->bmi[j + 2] = mi->bmi[j];
if (num_4x4_w == 2)
mi->bmi[j + 1] = mi->bmi[j];
}
}
mi->mbmi.mode = b_mode;
mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
} else {
xd->corrupted |= !assign_mv(cm, mbmi->mode, mbmi->mv,
best, nearest, nearmv,
is_compound, allow_hp, r);
}
}
void vp9_encode_tiles_mt(VP9_COMP *cpi) {
VP9_COMMON *const cm = &cpi->common;
const int tile_cols = 1 << cm->log2_tile_cols;
const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
const int num_workers = VPXMIN(cpi->oxcf.max_threads, tile_cols);
int i;
vp9_init_tile_data(cpi);
// Only run once to create threads and allocate thread data.
if (cpi->num_workers == 0) {
int allocated_workers = num_workers;
// While using SVC, we need to allocate threads according to the highest
// resolution.
if (cpi->use_svc) {
int max_tile_cols = get_max_tile_cols(cpi);
allocated_workers = VPXMIN(cpi->oxcf.max_threads, max_tile_cols);
}
CHECK_MEM_ERROR(cm, cpi->workers,
vpx_malloc(allocated_workers * sizeof(*cpi->workers)));
CHECK_MEM_ERROR(cm, cpi->tile_thr_data,
vpx_calloc(allocated_workers,
sizeof(*cpi->tile_thr_data)));
for (i = 0; i < allocated_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *thread_data = &cpi->tile_thr_data[i];
++cpi->num_workers;
winterface->init(worker);
if (i < allocated_workers - 1) {
thread_data->cpi = cpi;
// Allocate thread data.
CHECK_MEM_ERROR(cm, thread_data->td,
vpx_memalign(32, sizeof(*thread_data->td)));
vp9_zero(*thread_data->td);
// Set up pc_tree.
thread_data->td->leaf_tree = NULL;
thread_data->td->pc_tree = NULL;
vp9_setup_pc_tree(cm, thread_data->td);
// Allocate frame counters in thread data.
CHECK_MEM_ERROR(cm, thread_data->td->counts,
vpx_calloc(1, sizeof(*thread_data->td->counts)));
// Create threads
if (!winterface->reset(worker))
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Tile encoder thread creation failed");
} else {
// Main thread acts as a worker and uses the thread data in cpi.
thread_data->cpi = cpi;
thread_data->td = &cpi->td;
}
winterface->sync(worker);
}
}
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *thread_data;
worker->hook = (VPxWorkerHook)enc_worker_hook;
worker->data1 = &cpi->tile_thr_data[i];
worker->data2 = NULL;
thread_data = (EncWorkerData*)worker->data1;
// Before encoding a frame, copy the thread data from cpi.
if (thread_data->td != &cpi->td) {
thread_data->td->mb = cpi->td.mb;
thread_data->td->rd_counts = cpi->td.rd_counts;
}
if (thread_data->td->counts != &cpi->common.counts) {
memcpy(thread_data->td->counts, &cpi->common.counts,
sizeof(cpi->common.counts));
}
// Handle use_nonrd_pick_mode case.
if (cpi->sf.use_nonrd_pick_mode) {
MACROBLOCK *const x = &thread_data->td->mb;
MACROBLOCKD *const xd = &x->e_mbd;
struct macroblock_plane *const p = x->plane;
struct macroblockd_plane *const pd = xd->plane;
PICK_MODE_CONTEXT *ctx = &thread_data->td->pc_root->none;
int j;
for (j = 0; j < MAX_MB_PLANE; ++j) {
p[j].coeff = ctx->coeff_pbuf[j][0];
p[j].qcoeff = ctx->qcoeff_pbuf[j][0];
pd[j].dqcoeff = ctx->dqcoeff_pbuf[j][0];
p[j].eobs = ctx->eobs_pbuf[j][0];
}
}
}
// Encode a frame
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData*)worker->data1;
// Set the starting tile for each thread.
thread_data->start = i;
if (i == cpi->num_workers - 1)
winterface->execute(worker);
else
winterface->launch(worker);
}
// Encoding ends.
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
winterface->sync(worker);
}
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData*)worker->data1;
// Accumulate counters.
if (i < cpi->num_workers - 1) {
vp9_accumulate_frame_counts(cm, thread_data->td->counts, 0);
accumulate_rd_opt(&cpi->td, thread_data->td);
}
}
}
static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t *ctx,
const uint8_t *data,
unsigned int data_sz,
void *user_priv,
long deadline)
{
vpx_codec_err_t res = VPX_CODEC_OK;
unsigned int resolution_change = 0;
unsigned int w, h;
/* Determine the stream parameters. Note that we rely on peek_si to
* validate that we have a buffer that does not wrap around the top
* of the heap.
*/
w = ctx->si.w;
h = ctx->si.h;
res = ctx->base.iface->dec.peek_si(data, data_sz, &ctx->si);
if((res == VPX_CODEC_UNSUP_BITSTREAM) && !ctx->si.is_kf)
{
/* the peek function returns an error for non keyframes, however for
* this case, it is not an error */
res = VPX_CODEC_OK;
}
if(!ctx->decoder_init && !ctx->si.is_kf)
res = VPX_CODEC_UNSUP_BITSTREAM;
if ((ctx->si.h != h) || (ctx->si.w != w))
resolution_change = 1;
/* Perform deferred allocations, if required */
if (!res && ctx->defer_alloc)
{
int i;
for (i = 1; !res && i < NELEMENTS(ctx->mmaps); i++)
{
vpx_codec_dec_cfg_t cfg;
cfg.w = ctx->si.w;
cfg.h = ctx->si.h;
ctx->mmaps[i].id = vp8_mem_req_segs[i].id;
ctx->mmaps[i].sz = vp8_mem_req_segs[i].sz;
ctx->mmaps[i].align = vp8_mem_req_segs[i].align;
ctx->mmaps[i].flags = vp8_mem_req_segs[i].flags;
if (!ctx->mmaps[i].sz)
ctx->mmaps[i].sz = vp8_mem_req_segs[i].calc_sz(&cfg,
ctx->base.init_flags);
res = vp8_mmap_alloc(&ctx->mmaps[i]);
}
if (!res)
vp8_finalize_mmaps(ctx);
ctx->defer_alloc = 0;
}
/* Initialize the decoder instance on the first frame*/
if (!res && !ctx->decoder_init)
{
res = vp8_validate_mmaps(&ctx->si, ctx->mmaps, ctx->base.init_flags);
if (!res)
{
VP8D_CONFIG oxcf;
struct VP8D_COMP* optr;
oxcf.Width = ctx->si.w;
oxcf.Height = ctx->si.h;
oxcf.Version = 9;
oxcf.postprocess = 0;
oxcf.max_threads = ctx->cfg.threads;
oxcf.error_concealment =
(ctx->base.init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT);
oxcf.input_fragments =
(ctx->base.init_flags & VPX_CODEC_USE_INPUT_FRAGMENTS);
optr = vp8dx_create_decompressor(&oxcf);
/* If postprocessing was enabled by the application and a
* configuration has not been provided, default it.
*/
if (!ctx->postproc_cfg_set
&& (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC))
{
ctx->postproc_cfg.post_proc_flag =
VP8_DEBLOCK | VP8_DEMACROBLOCK | VP8_MFQE;
ctx->postproc_cfg.deblocking_level = 4;
ctx->postproc_cfg.noise_level = 0;
}
if (!optr)
res = VPX_CODEC_ERROR;
else
ctx->pbi = optr;
}
ctx->decoder_init = 1;
}
if (!res && ctx->pbi)
{
if(resolution_change)
{
VP8D_COMP *pbi = ctx->pbi;
VP8_COMMON *const pc = & pbi->common;
MACROBLOCKD *const xd = & pbi->mb;
#if CONFIG_MULTITHREAD
int i;
#endif
pc->Width = ctx->si.w;
pc->Height = ctx->si.h;
{
int prev_mb_rows = pc->mb_rows;
if (setjmp(pbi->common.error.jmp))
{
pbi->common.error.setjmp = 0;
/* same return value as used in vp8dx_receive_compressed_data */
return -1;
}
pbi->common.error.setjmp = 1;
if (pc->Width <= 0)
{
pc->Width = w;
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Invalid frame width");
}
if (pc->Height <= 0)
{
pc->Height = h;
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Invalid frame height");
}
if (vp8_alloc_frame_buffers(pc, pc->Width, pc->Height))
vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffers");
xd->pre = pc->yv12_fb[pc->lst_fb_idx];
xd->dst = pc->yv12_fb[pc->new_fb_idx];
#if CONFIG_MULTITHREAD
for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
{
pbi->mb_row_di[i].mbd.dst = pc->yv12_fb[pc->new_fb_idx];
vp8_build_block_doffsets(&pbi->mb_row_di[i].mbd);
}
#endif
vp8_build_block_doffsets(&pbi->mb);
/* allocate memory for last frame MODE_INFO array */
#if CONFIG_ERROR_CONCEALMENT
if (pbi->ec_enabled)
{
/* old prev_mip was released by vp8_de_alloc_frame_buffers()
* called in vp8_alloc_frame_buffers() */
pc->prev_mip = vpx_calloc(
(pc->mb_cols + 1) * (pc->mb_rows + 1),
sizeof(MODE_INFO));
if (!pc->prev_mip)
{
vp8_de_alloc_frame_buffers(pc);
vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate"
"last frame MODE_INFO array");
}
pc->prev_mi = pc->prev_mip + pc->mode_info_stride + 1;
if (vp8_alloc_overlap_lists(pbi))
vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate overlap lists "
"for error concealment");
}
#endif
#if CONFIG_MULTITHREAD
if (pbi->b_multithreaded_rd)
vp8mt_alloc_temp_buffers(pbi, pc->Width, prev_mb_rows);
#else
(void)prev_mb_rows;
#endif
}
pbi->common.error.setjmp = 0;
/* required to get past the first get_free_fb() call */
ctx->pbi->common.fb_idx_ref_cnt[0] = 0;
}
ctx->user_priv = user_priv;
if (vp8dx_receive_compressed_data(ctx->pbi, data_sz, data, deadline))
{
VP8D_COMP *pbi = (VP8D_COMP *)ctx->pbi;
res = update_error_state(ctx, &pbi->common.error);
}
}
return res;
}
static void read_inter_block_mode_info(VP9Decoder *const pbi,
MACROBLOCKD *const xd,
const TileInfo *const tile,
MODE_INFO *const mi,
int mi_row, int mi_col, vp9_reader *r) {
VP9_COMMON *const cm = &pbi->common;
MB_MODE_INFO *const mbmi = &mi->mbmi;
const BLOCK_SIZE bsize = mbmi->sb_type;
const int allow_hp = cm->allow_high_precision_mv;
int_mv nearestmv[2], nearmv[2];
int inter_mode_ctx, ref, is_compound;
read_ref_frames(cm, xd, r, mbmi->segment_id, mbmi->ref_frame);
is_compound = has_second_ref(mbmi);
for (ref = 0; ref < 1 + is_compound; ++ref) {
const MV_REFERENCE_FRAME frame = mbmi->ref_frame[ref];
RefBuffer *ref_buf = &cm->frame_refs[frame - LAST_FRAME];
xd->block_refs[ref] = ref_buf;
if ((!vp9_is_valid_scale(&ref_buf->sf)))
vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM,
"Reference frame has invalid dimensions");
vp9_setup_pre_planes(xd, ref, ref_buf->buf, mi_row, mi_col,
&ref_buf->sf);
vp9_find_mv_refs(cm, xd, tile, mi, frame, mbmi->ref_mvs[frame],
mi_row, mi_col, fpm_sync, (void *)pbi);
}
inter_mode_ctx = mbmi->mode_context[mbmi->ref_frame[0]];
if (vp9_segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
mbmi->mode = ZEROMV;
if (bsize < BLOCK_8X8) {
vpx_internal_error(xd->error_info, VPX_CODEC_UNSUP_BITSTREAM,
"Invalid usage of segement feature on small blocks");
return;
}
} else {
if (bsize >= BLOCK_8X8)
mbmi->mode = read_inter_mode(cm, xd, r, inter_mode_ctx);
}
if (bsize < BLOCK_8X8 || mbmi->mode != ZEROMV) {
for (ref = 0; ref < 1 + is_compound; ++ref) {
vp9_find_best_ref_mvs(xd, allow_hp, mbmi->ref_mvs[mbmi->ref_frame[ref]],
&nearestmv[ref], &nearmv[ref]);
}
}
mbmi->interp_filter = (cm->interp_filter == SWITCHABLE)
? read_switchable_interp_filter(cm, xd, r)
: cm->interp_filter;
if (bsize < BLOCK_8X8) {
const int num_4x4_w = num_4x4_blocks_wide_lookup[bsize]; // 1 or 2
const int num_4x4_h = num_4x4_blocks_high_lookup[bsize]; // 1 or 2
int idx, idy;
PREDICTION_MODE b_mode;
int_mv nearest_sub8x8[2], near_sub8x8[2];
for (idy = 0; idy < 2; idy += num_4x4_h) {
for (idx = 0; idx < 2; idx += num_4x4_w) {
int_mv block[2];
const int j = idy * 2 + idx;
b_mode = read_inter_mode(cm, xd, r, inter_mode_ctx);
if (b_mode == NEARESTMV || b_mode == NEARMV)
for (ref = 0; ref < 1 + is_compound; ++ref)
vp9_append_sub8x8_mvs_for_idx(cm, xd, tile, j, ref, mi_row, mi_col,
&nearest_sub8x8[ref],
&near_sub8x8[ref]);
if (!assign_mv(cm, xd, b_mode, block, nearestmv,
nearest_sub8x8, near_sub8x8,
is_compound, allow_hp, r)) {
xd->corrupted |= 1;
break;
};
mi->bmi[j].as_mv[0].as_int = block[0].as_int;
if (is_compound)
mi->bmi[j].as_mv[1].as_int = block[1].as_int;
if (num_4x4_h == 2)
mi->bmi[j + 2] = mi->bmi[j];
if (num_4x4_w == 2)
mi->bmi[j + 1] = mi->bmi[j];
}
}
mi->mbmi.mode = b_mode;
mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
} else {
xd->corrupted |= !assign_mv(cm, xd, mbmi->mode, mbmi->mv, nearestmv,
nearestmv, nearmv, is_compound, allow_hp, r);
}
}
static vpx_codec_err_t vp8_decode(vpx_codec_alg_priv_t *ctx,
const uint8_t *data,
unsigned int data_sz,
void *user_priv,
long deadline)
{
vpx_codec_err_t res = VPX_CODEC_OK;
unsigned int resolution_change = 0;
unsigned int w, h;
if (!ctx->fragments.enabled && (data == NULL && data_sz == 0))
{
return 0;
}
/* Update the input fragment data */
if(update_fragments(ctx, data, data_sz, &res) <= 0)
return res;
/* Determine the stream parameters. Note that we rely on peek_si to
* validate that we have a buffer that does not wrap around the top
* of the heap.
*/
w = ctx->si.w;
h = ctx->si.h;
res = vp8_peek_si_internal(ctx->fragments.ptrs[0], ctx->fragments.sizes[0],
&ctx->si, ctx->decrypt_cb, ctx->decrypt_state);
if((res == VPX_CODEC_UNSUP_BITSTREAM) && !ctx->si.is_kf)
{
/* the peek function returns an error for non keyframes, however for
* this case, it is not an error */
res = VPX_CODEC_OK;
}
if(!ctx->decoder_init && !ctx->si.is_kf)
res = VPX_CODEC_UNSUP_BITSTREAM;
if ((ctx->si.h != h) || (ctx->si.w != w))
resolution_change = 1;
/* Initialize the decoder instance on the first frame*/
if (!res && !ctx->decoder_init)
{
VP8D_CONFIG oxcf;
oxcf.Width = ctx->si.w;
oxcf.Height = ctx->si.h;
oxcf.Version = 9;
oxcf.postprocess = 0;
oxcf.max_threads = ctx->cfg.threads;
oxcf.error_concealment =
(ctx->base.init_flags & VPX_CODEC_USE_ERROR_CONCEALMENT);
/* If postprocessing was enabled by the application and a
* configuration has not been provided, default it.
*/
if (!ctx->postproc_cfg_set
&& (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)) {
ctx->postproc_cfg.post_proc_flag =
VP8_DEBLOCK | VP8_DEMACROBLOCK | VP8_MFQE;
ctx->postproc_cfg.deblocking_level = 4;
ctx->postproc_cfg.noise_level = 0;
}
res = vp8_create_decoder_instances(&ctx->yv12_frame_buffers, &oxcf);
ctx->decoder_init = 1;
}
/* Set these even if already initialized. The caller may have changed the
* decrypt config between frames.
*/
if (ctx->decoder_init) {
ctx->yv12_frame_buffers.pbi[0]->decrypt_cb = ctx->decrypt_cb;
ctx->yv12_frame_buffers.pbi[0]->decrypt_state = ctx->decrypt_state;
}
if (!res)
{
VP8D_COMP *pbi = ctx->yv12_frame_buffers.pbi[0];
if (resolution_change)
{
VP8_COMMON *const pc = & pbi->common;
MACROBLOCKD *const xd = & pbi->mb;
#if CONFIG_MULTITHREAD
int i;
#endif
pc->Width = ctx->si.w;
pc->Height = ctx->si.h;
{
int prev_mb_rows = pc->mb_rows;
if (setjmp(pbi->common.error.jmp))
{
pbi->common.error.setjmp = 0;
vp8_clear_system_state();
/* same return value as used in vp8dx_receive_compressed_data */
return -1;
}
pbi->common.error.setjmp = 1;
if (pc->Width <= 0)
{
pc->Width = w;
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Invalid frame width");
}
if (pc->Height <= 0)
{
pc->Height = h;
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Invalid frame height");
}
if (vp8_alloc_frame_buffers(pc, pc->Width, pc->Height))
vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffers");
xd->pre = pc->yv12_fb[pc->lst_fb_idx];
xd->dst = pc->yv12_fb[pc->new_fb_idx];
#if CONFIG_MULTITHREAD
for (i = 0; i < pbi->allocated_decoding_thread_count; i++)
{
pbi->mb_row_di[i].mbd.dst = pc->yv12_fb[pc->new_fb_idx];
vp8_build_block_doffsets(&pbi->mb_row_di[i].mbd);
}
#endif
vp8_build_block_doffsets(&pbi->mb);
/* allocate memory for last frame MODE_INFO array */
#if CONFIG_ERROR_CONCEALMENT
if (pbi->ec_enabled)
{
/* old prev_mip was released by vp8_de_alloc_frame_buffers()
* called in vp8_alloc_frame_buffers() */
pc->prev_mip = vpx_calloc(
(pc->mb_cols + 1) * (pc->mb_rows + 1),
sizeof(MODE_INFO));
if (!pc->prev_mip)
{
vp8_de_alloc_frame_buffers(pc);
vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate"
"last frame MODE_INFO array");
}
pc->prev_mi = pc->prev_mip + pc->mode_info_stride + 1;
if (vp8_alloc_overlap_lists(pbi))
vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate overlap lists "
"for error concealment");
}
#endif
#if CONFIG_MULTITHREAD
if (pbi->b_multithreaded_rd)
vp8mt_alloc_temp_buffers(pbi, pc->Width, prev_mb_rows);
#else
(void)prev_mb_rows;
#endif
}
pbi->common.error.setjmp = 0;
/* required to get past the first get_free_fb() call */
pbi->common.fb_idx_ref_cnt[0] = 0;
}
/* update the pbi fragment data */
pbi->fragments = ctx->fragments;
ctx->user_priv = user_priv;
if (vp8dx_receive_compressed_data(pbi, data_sz, data, deadline))
{
res = update_error_state(ctx, &pbi->common.error);
}
/* get ready for the next series of fragments */
ctx->fragments.count = 0;
}
return res;
}
int vp8_decode_frame(VP8D_COMP *pbi)
{
vp8_reader *const bc = & pbi->bc;
VP8_COMMON *const pc = & pbi->common;
MACROBLOCKD *const xd = & pbi->mb;
const unsigned char *data = pbi->fragments[0];
const unsigned char *data_end = data + pbi->fragment_sizes[0];
ptrdiff_t first_partition_length_in_bytes;
int mb_row;
int i, j, k, l;
const int *const mb_feature_data_bits = vp8_mb_feature_data_bits;
int corrupt_tokens = 0;
int prev_independent_partitions = pbi->independent_partitions;
/* start with no corruption of current frame */
xd->corrupted = 0;
pc->yv12_fb[pc->new_fb_idx].corrupted = 0;
if (data_end - data < 3)
{
if (!pbi->ec_active)
{
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Truncated packet");
}
/* Declare the missing frame as an inter frame since it will
be handled as an inter frame when we have estimated its
motion vectors. */
pc->frame_type = INTER_FRAME;
pc->version = 0;
pc->show_frame = 1;
first_partition_length_in_bytes = 0;
}
else
{
pc->frame_type = (FRAME_TYPE)(data[0] & 1);
pc->version = (data[0] >> 1) & 7;
pc->show_frame = (data[0] >> 4) & 1;
first_partition_length_in_bytes =
(data[0] | (data[1] << 8) | (data[2] << 16)) >> 5;
if (!pbi->ec_active && (data + first_partition_length_in_bytes > data_end
|| data + first_partition_length_in_bytes < data))
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Truncated packet or corrupt partition 0 length");
data += 3;
vp8_setup_version(pc);
if (pc->frame_type == KEY_FRAME)
{
const int Width = pc->Width;
const int Height = pc->Height;
/* vet via sync code */
/* When error concealment is enabled we should only check the sync
* code if we have enough bits available
*/
if (!pbi->ec_active || data + 3 < data_end)
{
if (data[0] != 0x9d || data[1] != 0x01 || data[2] != 0x2a)
vpx_internal_error(&pc->error, VPX_CODEC_UNSUP_BITSTREAM,
"Invalid frame sync code");
}
/* If error concealment is enabled we should only parse the new size
* if we have enough data. Otherwise we will end up with the wrong
* size.
*/
if (!pbi->ec_active || data + 6 < data_end)
{
pc->Width = (data[3] | (data[4] << 8)) & 0x3fff;
pc->horiz_scale = data[4] >> 6;
pc->Height = (data[5] | (data[6] << 8)) & 0x3fff;
pc->vert_scale = data[6] >> 6;
}
data += 7;
if (Width != pc->Width || Height != pc->Height)
{
int prev_mb_rows = pc->mb_rows;
if (pc->Width <= 0)
{
pc->Width = Width;
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Invalid frame width");
}
if (pc->Height <= 0)
{
pc->Height = Height;
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Invalid frame height");
}
if (vp8_alloc_frame_buffers(pc, pc->Width, pc->Height))
vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate frame buffers");
#if CONFIG_ERROR_CONCEALMENT
pbi->overlaps = NULL;
if (pbi->ec_enabled)
{
if (vp8_alloc_overlap_lists(pbi))
vpx_internal_error(&pc->error, VPX_CODEC_MEM_ERROR,
"Failed to allocate overlap lists "
"for error concealment");
}
#endif
#if CONFIG_MULTITHREAD
if (pbi->b_multithreaded_rd)
vp8mt_alloc_temp_buffers(pbi, pc->Width, prev_mb_rows);
#endif
}
}
}
void vp10_encode_tiles_mt(VP10_COMP *cpi) {
VP10_COMMON *const cm = &cpi->common;
const int tile_cols = 1 << cm->log2_tile_cols;
const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
const int num_workers = VPXMIN(cpi->oxcf.max_threads, tile_cols);
int i;
vp10_init_tile_data(cpi);
// Only run once to create threads and allocate thread data.
if (cpi->num_workers == 0) {
int allocated_workers = num_workers;
CHECK_MEM_ERROR(cm, cpi->workers,
vpx_malloc(allocated_workers * sizeof(*cpi->workers)));
CHECK_MEM_ERROR(cm, cpi->tile_thr_data,
vpx_calloc(allocated_workers,
sizeof(*cpi->tile_thr_data)));
for (i = 0; i < allocated_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *thread_data = &cpi->tile_thr_data[i];
++cpi->num_workers;
winterface->init(worker);
if (i < allocated_workers - 1) {
thread_data->cpi = cpi;
// Allocate thread data.
CHECK_MEM_ERROR(cm, thread_data->td,
vpx_memalign(32, sizeof(*thread_data->td)));
vp10_zero(*thread_data->td);
// Set up pc_tree.
thread_data->td->leaf_tree = NULL;
thread_data->td->pc_tree = NULL;
vp10_setup_pc_tree(cm, thread_data->td);
// Allocate frame counters in thread data.
CHECK_MEM_ERROR(cm, thread_data->td->counts,
vpx_calloc(1, sizeof(*thread_data->td->counts)));
// Create threads
if (!winterface->reset(worker))
vpx_internal_error(&cm->error, VPX_CODEC_ERROR,
"Tile encoder thread creation failed");
} else {
// Main thread acts as a worker and uses the thread data in cpi.
thread_data->cpi = cpi;
thread_data->td = &cpi->td;
}
winterface->sync(worker);
}
}
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *thread_data;
worker->hook = (VPxWorkerHook)enc_worker_hook;
worker->data1 = &cpi->tile_thr_data[i];
worker->data2 = NULL;
thread_data = (EncWorkerData*)worker->data1;
// Before encoding a frame, copy the thread data from cpi.
if (thread_data->td != &cpi->td) {
thread_data->td->mb = cpi->td.mb;
thread_data->td->rd_counts = cpi->td.rd_counts;
}
if (thread_data->td->counts != &cpi->common.counts) {
memcpy(thread_data->td->counts, &cpi->common.counts,
sizeof(cpi->common.counts));
}
}
// Encode a frame
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData*)worker->data1;
// Set the starting tile for each thread.
thread_data->start = i;
if (i == cpi->num_workers - 1)
winterface->execute(worker);
else
winterface->launch(worker);
}
// Encoding ends.
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
winterface->sync(worker);
}
for (i = 0; i < num_workers; i++) {
VPxWorker *const worker = &cpi->workers[i];
EncWorkerData *const thread_data = (EncWorkerData*)worker->data1;
// Accumulate counters.
if (i < cpi->num_workers - 1) {
vp10_accumulate_frame_counts(cm, thread_data->td->counts, 0);
accumulate_rd_opt(&cpi->td, thread_data->td);
}
}
}
int vp8_decode_frame(VP8D_COMP *pbi)
{
vp8_reader *const bc = &pbi->mbc[8];
VP8_COMMON *const pc = &pbi->common;
MACROBLOCKD *const xd = &pbi->mb;
const unsigned char *data = pbi->fragments.ptrs[0];
const unsigned char *data_end = data + pbi->fragments.sizes[0];
ptrdiff_t first_partition_length_in_bytes;
int i, j, k, l;
const int *const mb_feature_data_bits = vp8_mb_feature_data_bits;
int corrupt_tokens = 0;
int prev_independent_partitions = pbi->independent_partitions;
YV12_BUFFER_CONFIG *yv12_fb_new = pbi->dec_fb_ref[INTRA_FRAME];
/* start with no corruption of current frame */
xd->corrupted = 0;
yv12_fb_new->corrupted = 0;
if (data_end - data < 3)
{
if (!pbi->ec_active)
{
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Truncated packet");
}
/* Declare the missing frame as an inter frame since it will
be handled as an inter frame when we have estimated its
motion vectors. */
pc->frame_type = INTER_FRAME;
pc->version = 0;
pc->show_frame = 1;
first_partition_length_in_bytes = 0;
}
else
{
unsigned char clear_buffer[10];
const unsigned char *clear = data;
if (pbi->decrypt_cb)
{
int n = (int)MIN(sizeof(clear_buffer), data_end - data);
pbi->decrypt_cb(pbi->decrypt_state, data, clear_buffer, n);
clear = clear_buffer;
}
pc->frame_type = (FRAME_TYPE)(clear[0] & 1);
pc->version = (clear[0] >> 1) & 7;
pc->show_frame = (clear[0] >> 4) & 1;
first_partition_length_in_bytes =
(clear[0] | (clear[1] << 8) | (clear[2] << 16)) >> 5;
if (!pbi->ec_active &&
(data + first_partition_length_in_bytes > data_end
|| data + first_partition_length_in_bytes < data))
vpx_internal_error(&pc->error, VPX_CODEC_CORRUPT_FRAME,
"Truncated packet or corrupt partition 0 length");
data += 3;
clear += 3;
vp8_setup_version(pc);
if (pc->frame_type == KEY_FRAME)
{
/* vet via sync code */
/* When error concealment is enabled we should only check the sync
* code if we have enough bits available
*/
if (!pbi->ec_active || data + 3 < data_end)
{
if (clear[0] != 0x9d || clear[1] != 0x01 || clear[2] != 0x2a)
vpx_internal_error(&pc->error, VPX_CODEC_UNSUP_BITSTREAM,
"Invalid frame sync code");
}
/* If error concealment is enabled we should only parse the new size
* if we have enough data. Otherwise we will end up with the wrong
* size.
*/
if (!pbi->ec_active || data + 6 < data_end)
{
pc->Width = (clear[3] | (clear[4] << 8)) & 0x3fff;
pc->horiz_scale = clear[4] >> 6;
pc->Height = (clear[5] | (clear[6] << 8)) & 0x3fff;
pc->vert_scale = clear[6] >> 6;
}
data += 7;
clear += 7;
}
else
{
static void setup_token_decoder(VP8D_COMP *pbi,
const unsigned char* token_part_sizes)
{
vp8_reader *bool_decoder = &pbi->mbc[0];
unsigned int partition_idx;
unsigned int fragment_idx;
unsigned int num_token_partitions;
const unsigned char *first_fragment_end = pbi->fragments.ptrs[0] +
pbi->fragments.sizes[0];
TOKEN_PARTITION multi_token_partition =
(TOKEN_PARTITION)vp8_read_literal(&pbi->mbc[8], 2);
if (!vp8dx_bool_error(&pbi->mbc[8]))
pbi->common.multi_token_partition = multi_token_partition;
num_token_partitions = 1 << pbi->common.multi_token_partition;
/* Check for partitions within the fragments and unpack the fragments
* so that each fragment pointer points to its corresponding partition. */
for (fragment_idx = 0; fragment_idx < pbi->fragments.count; ++fragment_idx)
{
unsigned int fragment_size = pbi->fragments.sizes[fragment_idx];
const unsigned char *fragment_end = pbi->fragments.ptrs[fragment_idx] +
fragment_size;
/* Special case for handling the first partition since we have already
* read its size. */
if (fragment_idx == 0)
{
/* Size of first partition + token partition sizes element */
ptrdiff_t ext_first_part_size = token_part_sizes -
pbi->fragments.ptrs[0] + 3 * (num_token_partitions - 1);
fragment_size -= (unsigned int)ext_first_part_size;
if (fragment_size > 0)
{
pbi->fragments.sizes[0] = (unsigned int)ext_first_part_size;
/* The fragment contains an additional partition. Move to
* next. */
fragment_idx++;
pbi->fragments.ptrs[fragment_idx] = pbi->fragments.ptrs[0] +
pbi->fragments.sizes[0];
}
}
/* Split the chunk into partitions read from the bitstream */
while (fragment_size > 0)
{
ptrdiff_t partition_size = read_available_partition_size(
pbi,
token_part_sizes,
pbi->fragments.ptrs[fragment_idx],
first_fragment_end,
fragment_end,
fragment_idx - 1,
num_token_partitions);
pbi->fragments.sizes[fragment_idx] = (unsigned int)partition_size;
fragment_size -= (unsigned int)partition_size;
assert(fragment_idx <= num_token_partitions);
if (fragment_size > 0)
{
/* The fragment contains an additional partition.
* Move to next. */
fragment_idx++;
pbi->fragments.ptrs[fragment_idx] =
pbi->fragments.ptrs[fragment_idx - 1] + partition_size;
}
}
}
pbi->fragments.count = num_token_partitions + 1;
for (partition_idx = 1; partition_idx < pbi->fragments.count; ++partition_idx)
{
if (vp8dx_start_decode(bool_decoder,
pbi->fragments.ptrs[partition_idx],
pbi->fragments.sizes[partition_idx],
pbi->decrypt_cb, pbi->decrypt_state))
vpx_internal_error(&pbi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate bool decoder %d",
partition_idx);
bool_decoder++;
}
#if CONFIG_MULTITHREAD
/* Clamp number of decoder threads */
if (pbi->decoding_thread_count > num_token_partitions - 1)
pbi->decoding_thread_count = num_token_partitions - 1;
#endif
}
static vpx_codec_err_t decode_one(vpx_codec_alg_priv_t *ctx,
const uint8_t **data, unsigned int data_sz,
void *user_priv, int64_t deadline) {
vpx_codec_err_t res = VPX_CODEC_OK;
ctx->img_avail = 0;
// Determine the stream parameters. Note that we rely on peek_si to
// validate that we have a buffer that does not wrap around the top
// of the heap.
if (!ctx->si.h)
res = ctx->base.iface->dec.peek_si(*data, data_sz, &ctx->si);
/* Initialize the decoder instance on the first frame*/
if (!res && !ctx->decoder_init) {
VP9D_CONFIG oxcf;
struct VP9Decompressor *optr;
vp9_initialize_dec();
oxcf.width = ctx->si.w;
oxcf.height = ctx->si.h;
oxcf.version = 9;
oxcf.max_threads = ctx->cfg.threads;
oxcf.inv_tile_order = ctx->invert_tile_order;
optr = vp9_create_decompressor(&oxcf);
// If postprocessing was enabled by the application and a
// configuration has not been provided, default it.
if (!ctx->postproc_cfg_set &&
(ctx->base.init_flags & VPX_CODEC_USE_POSTPROC)) {
ctx->postproc_cfg.post_proc_flag = VP8_DEBLOCK | VP8_DEMACROBLOCK;
ctx->postproc_cfg.deblocking_level = 4;
ctx->postproc_cfg.noise_level = 0;
}
if (!optr) {
res = VPX_CODEC_ERROR;
} else {
VP9D_COMP *const pbi = (VP9D_COMP*)optr;
VP9_COMMON *const cm = &pbi->common;
// Set index to not initialized.
cm->new_fb_idx = -1;
if (ctx->get_ext_fb_cb != NULL && ctx->release_ext_fb_cb != NULL) {
cm->get_fb_cb = ctx->get_ext_fb_cb;
cm->release_fb_cb = ctx->release_ext_fb_cb;
cm->cb_priv = ctx->ext_priv;
} else {
cm->get_fb_cb = vp9_get_frame_buffer;
cm->release_fb_cb = vp9_release_frame_buffer;
if (vp9_alloc_internal_frame_buffers(&cm->int_frame_buffers))
vpx_internal_error(&cm->error, VPX_CODEC_MEM_ERROR,
"Failed to initialize internal frame buffers");
cm->cb_priv = &cm->int_frame_buffers;
}
ctx->pbi = optr;
}
ctx->decoder_init = 1;
}
if (!res && ctx->pbi) {
VP9D_COMP *const pbi = ctx->pbi;
VP9_COMMON *const cm = &pbi->common;
YV12_BUFFER_CONFIG sd;
int64_t time_stamp = 0, time_end_stamp = 0;
vp9_ppflags_t flags = {0};
if (ctx->base.init_flags & VPX_CODEC_USE_POSTPROC) {
flags.post_proc_flag =
#if CONFIG_POSTPROC_VISUALIZER
(ctx->dbg_color_ref_frame_flag ? VP9D_DEBUG_CLR_FRM_REF_BLKS : 0) |
(ctx->dbg_color_mb_modes_flag ? VP9D_DEBUG_CLR_BLK_MODES : 0) |
(ctx->dbg_color_b_modes_flag ? VP9D_DEBUG_CLR_BLK_MODES : 0) |
(ctx->dbg_display_mv_flag ? VP9D_DEBUG_DRAW_MV : 0) |
#endif
ctx->postproc_cfg.post_proc_flag;
flags.deblocking_level = ctx->postproc_cfg.deblocking_level;
flags.noise_level = ctx->postproc_cfg.noise_level;
#if CONFIG_POSTPROC_VISUALIZER
flags.display_ref_frame_flag = ctx->dbg_color_ref_frame_flag;
flags.display_mb_modes_flag = ctx->dbg_color_mb_modes_flag;
flags.display_b_modes_flag = ctx->dbg_color_b_modes_flag;
flags.display_mv_flag = ctx->dbg_display_mv_flag;
#endif
}
if (vp9_receive_compressed_data(pbi, data_sz, data, deadline))
res = update_error_state(ctx, &cm->error);
if (!res && 0 == vp9_get_raw_frame(pbi, &sd, &time_stamp,
&time_end_stamp, &flags)) {
yuvconfig2image(&ctx->img, &sd, user_priv);
ctx->img.fb_priv = cm->frame_bufs[cm->new_fb_idx].raw_frame_buffer.priv;
ctx->img_avail = 1;
}
}
return res;
}
void vp9_init_layer_context(VP9_COMP *const cpi) {
SVC *const svc = &cpi->svc;
const VP9EncoderConfig *const oxcf = &cpi->oxcf;
int layer;
int layer_end;
int alt_ref_idx = svc->number_spatial_layers;
svc->spatial_layer_id = 0;
svc->temporal_layer_id = 0;
if (svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR) {
layer_end = svc->number_temporal_layers;
} else {
layer_end = svc->number_spatial_layers;
if (cpi->oxcf.error_resilient_mode == 0 && cpi->oxcf.pass == 2) {
if (vp9_realloc_frame_buffer(&cpi->svc.empty_frame.img,
cpi->common.width, cpi->common.height,
cpi->common.subsampling_x,
cpi->common.subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cpi->common.use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS,
cpi->common.byte_alignment,
NULL, NULL, NULL))
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate empty frame for multiple frame "
"contexts");
vpx_memset(cpi->svc.empty_frame.img.buffer_alloc, 0x80,
cpi->svc.empty_frame.img.buffer_alloc_sz);
cpi->svc.empty_frame_width = cpi->common.width;
cpi->svc.empty_frame_height = cpi->common.height;
}
}
for (layer = 0; layer < layer_end; ++layer) {
LAYER_CONTEXT *const lc = &svc->layer_context[layer];
RATE_CONTROL *const lrc = &lc->rc;
int i;
lc->current_video_frame_in_layer = 0;
lc->layer_size = 0;
lc->frames_from_key_frame = 0;
lc->last_frame_type = FRAME_TYPES;
lrc->ni_av_qi = oxcf->worst_allowed_q;
lrc->total_actual_bits = 0;
lrc->total_target_vs_actual = 0;
lrc->ni_tot_qi = 0;
lrc->tot_q = 0.0;
lrc->avg_q = 0.0;
lrc->ni_frames = 0;
lrc->decimation_count = 0;
lrc->decimation_factor = 0;
for (i = 0; i < RATE_FACTOR_LEVELS; ++i) {
lrc->rate_correction_factors[i] = 1.0;
}
if (svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR) {
lc->target_bandwidth = oxcf->ts_target_bitrate[layer];
lrc->last_q[INTER_FRAME] = oxcf->worst_allowed_q;
lrc->avg_frame_qindex[INTER_FRAME] = oxcf->worst_allowed_q;
lrc->avg_frame_qindex[KEY_FRAME] = oxcf->worst_allowed_q;
} else {
lc->target_bandwidth = oxcf->ss_target_bitrate[layer];
lrc->last_q[KEY_FRAME] = oxcf->best_allowed_q;
lrc->last_q[INTER_FRAME] = oxcf->best_allowed_q;
lrc->avg_frame_qindex[KEY_FRAME] = (oxcf->worst_allowed_q +
oxcf->best_allowed_q) / 2;
lrc->avg_frame_qindex[INTER_FRAME] = (oxcf->worst_allowed_q +
oxcf->best_allowed_q) / 2;
if (oxcf->ss_enable_auto_arf[layer])
lc->alt_ref_idx = alt_ref_idx++;
else
lc->alt_ref_idx = INVALID_IDX;
lc->gold_ref_idx = INVALID_IDX;
}
lrc->buffer_level = oxcf->starting_buffer_level_ms *
lc->target_bandwidth / 1000;
lrc->bits_off_target = lrc->buffer_level;
}
// Still have extra buffer for base layer golden frame
if (!(svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR)
&& alt_ref_idx < REF_FRAMES)
svc->layer_context[0].gold_ref_idx = alt_ref_idx;
}
void vp9_init_layer_context(VP9_COMP *const cpi) {
SVC *const svc = &cpi->svc;
const VP9EncoderConfig *const oxcf = &cpi->oxcf;
int mi_rows = cpi->common.mi_rows;
int mi_cols = cpi->common.mi_cols;
int sl, tl, i;
int alt_ref_idx = svc->number_spatial_layers;
svc->spatial_layer_id = 0;
svc->temporal_layer_id = 0;
svc->first_spatial_layer_to_encode = 0;
svc->rc_drop_superframe = 0;
svc->force_zero_mode_spatial_ref = 0;
svc->use_base_mv = 0;
svc->scaled_temp_is_alloc = 0;
svc->scaled_one_half = 0;
svc->current_superframe = 0;
for (i = 0; i < REF_FRAMES; ++i)
svc->ref_frame_index[i] = -1;
for (sl = 0; sl < oxcf->ss_number_layers; ++sl) {
cpi->svc.ext_frame_flags[sl] = 0;
cpi->svc.ext_lst_fb_idx[sl] = 0;
cpi->svc.ext_gld_fb_idx[sl] = 1;
cpi->svc.ext_alt_fb_idx[sl] = 2;
}
if (cpi->oxcf.error_resilient_mode == 0 && cpi->oxcf.pass == 2) {
if (vpx_realloc_frame_buffer(&cpi->svc.empty_frame.img,
SMALL_FRAME_WIDTH, SMALL_FRAME_HEIGHT,
cpi->common.subsampling_x,
cpi->common.subsampling_y,
#if CONFIG_VP9_HIGHBITDEPTH
cpi->common.use_highbitdepth,
#endif
VP9_ENC_BORDER_IN_PIXELS,
cpi->common.byte_alignment,
NULL, NULL, NULL))
vpx_internal_error(&cpi->common.error, VPX_CODEC_MEM_ERROR,
"Failed to allocate empty frame for multiple frame "
"contexts");
memset(cpi->svc.empty_frame.img.buffer_alloc, 0x80,
cpi->svc.empty_frame.img.buffer_alloc_sz);
}
for (sl = 0; sl < oxcf->ss_number_layers; ++sl) {
for (tl = 0; tl < oxcf->ts_number_layers; ++tl) {
int layer = LAYER_IDS_TO_IDX(sl, tl, oxcf->ts_number_layers);
LAYER_CONTEXT *const lc = &svc->layer_context[layer];
RATE_CONTROL *const lrc = &lc->rc;
int i;
lc->current_video_frame_in_layer = 0;
lc->layer_size = 0;
lc->frames_from_key_frame = 0;
lc->last_frame_type = FRAME_TYPES;
lrc->ni_av_qi = oxcf->worst_allowed_q;
lrc->total_actual_bits = 0;
lrc->total_target_vs_actual = 0;
lrc->ni_tot_qi = 0;
lrc->tot_q = 0.0;
lrc->avg_q = 0.0;
lrc->ni_frames = 0;
lrc->decimation_count = 0;
lrc->decimation_factor = 0;
for (i = 0; i < RATE_FACTOR_LEVELS; ++i) {
lrc->rate_correction_factors[i] = 1.0;
}
if (cpi->oxcf.rc_mode == VPX_CBR) {
lc->target_bandwidth = oxcf->layer_target_bitrate[layer];
lrc->last_q[INTER_FRAME] = oxcf->worst_allowed_q;
lrc->avg_frame_qindex[INTER_FRAME] = oxcf->worst_allowed_q;
lrc->avg_frame_qindex[KEY_FRAME] = oxcf->worst_allowed_q;
} else {
lc->target_bandwidth = oxcf->layer_target_bitrate[layer];
lrc->last_q[KEY_FRAME] = oxcf->best_allowed_q;
lrc->last_q[INTER_FRAME] = oxcf->best_allowed_q;
lrc->avg_frame_qindex[KEY_FRAME] = (oxcf->worst_allowed_q +
oxcf->best_allowed_q) / 2;
lrc->avg_frame_qindex[INTER_FRAME] = (oxcf->worst_allowed_q +
oxcf->best_allowed_q) / 2;
if (oxcf->ss_enable_auto_arf[sl])
lc->alt_ref_idx = alt_ref_idx++;
else
lc->alt_ref_idx = INVALID_IDX;
lc->gold_ref_idx = INVALID_IDX;
}
lrc->buffer_level = oxcf->starting_buffer_level_ms *
lc->target_bandwidth / 1000;
lrc->bits_off_target = lrc->buffer_level;
// Initialize the cyclic refresh parameters. If spatial layers are used
// (i.e., ss_number_layers > 1), these need to be updated per spatial
// layer.
// Cyclic refresh is only applied on base temporal layer.
if (oxcf->ss_number_layers > 1 &&
tl == 0) {
size_t last_coded_q_map_size;
size_t consec_zero_mv_size;
VP9_COMMON *const cm = &cpi->common;
lc->sb_index = 0;
CHECK_MEM_ERROR(cm, lc->map,
vpx_malloc(mi_rows * mi_cols * sizeof(*lc->map)));
memset(lc->map, 0, mi_rows * mi_cols);
last_coded_q_map_size = mi_rows * mi_cols *
sizeof(*lc->last_coded_q_map);
CHECK_MEM_ERROR(cm, lc->last_coded_q_map,
vpx_malloc(last_coded_q_map_size));
assert(MAXQ <= 255);
memset(lc->last_coded_q_map, MAXQ, last_coded_q_map_size);
consec_zero_mv_size = mi_rows * mi_cols * sizeof(*lc->consec_zero_mv);
CHECK_MEM_ERROR(cm, lc->consec_zero_mv,
vpx_malloc(consec_zero_mv_size));
memset(lc->consec_zero_mv, 0, consec_zero_mv_size);
}
}
}
// Still have extra buffer for base layer golden frame
if (!(svc->number_temporal_layers > 1 && cpi->oxcf.rc_mode == VPX_CBR)
&& alt_ref_idx < REF_FRAMES)
svc->layer_context[0].gold_ref_idx = alt_ref_idx;
}
