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 vpx_codec_err_t decoder_init(vpx_codec_ctx_t *ctx,
vpx_codec_priv_enc_mr_cfg_t *data) {
// This function only allocates space for the vpx_codec_alg_priv_t
// structure. More memory may be required at the time the stream
// information becomes known.
if (!ctx->priv) {
vpx_codec_alg_priv_t *alg_priv = vpx_memalign(32, sizeof(*alg_priv));
if (alg_priv == NULL)
return VPX_CODEC_MEM_ERROR;
vp9_zero(*alg_priv);
ctx->priv = (vpx_codec_priv_t *)alg_priv;
ctx->priv->sz = sizeof(*ctx->priv);
ctx->priv->iface = ctx->iface;
ctx->priv->alg_priv = alg_priv;
ctx->priv->alg_priv->si.sz = sizeof(ctx->priv->alg_priv->si);
ctx->priv->init_flags = ctx->init_flags;
if (ctx->config.dec) {
// Update the reference to the config structure to an internal copy.
ctx->priv->alg_priv->cfg = *ctx->config.dec;
ctx->config.dec = &ctx->priv->alg_priv->cfg;
}
}
return VPX_CODEC_OK;
}
// Deallocate lf synchronization related mutex and data
void vp9_loop_filter_dealloc(VP9LfSync *lf_sync, int rows) {
#if !CONFIG_MULTITHREAD
(void)rows;
#endif // !CONFIG_MULTITHREAD
if (lf_sync != NULL) {
#if CONFIG_MULTITHREAD
int i;
if (lf_sync->mutex_ != NULL) {
for (i = 0; i < rows; ++i) {
pthread_mutex_destroy(&lf_sync->mutex_[i]);
}
vpx_free(lf_sync->mutex_);
}
if (lf_sync->cond_ != NULL) {
for (i = 0; i < rows; ++i) {
pthread_cond_destroy(&lf_sync->cond_[i]);
}
vpx_free(lf_sync->cond_);
}
#endif // CONFIG_MULTITHREAD
vpx_free(lf_sync->cur_sb_col);
// clear the structure as the source of this call may be a resize in which
// case this call will be followed by an _alloc() which may fail.
vp9_zero(*lf_sync);
}
}
// Deallocate row based multi-threading synchronization related mutex and data
void vp9_row_mt_sync_mem_dealloc(VP9RowMTSync *row_mt_sync) {
if (row_mt_sync != NULL) {
#if CONFIG_MULTITHREAD
int i;
if (row_mt_sync->mutex_ != NULL) {
for (i = 0; i < row_mt_sync->rows; ++i) {
pthread_mutex_destroy(&row_mt_sync->mutex_[i]);
}
vpx_free(row_mt_sync->mutex_);
}
if (row_mt_sync->cond_ != NULL) {
for (i = 0; i < row_mt_sync->rows; ++i) {
pthread_cond_destroy(&row_mt_sync->cond_[i]);
}
vpx_free(row_mt_sync->cond_);
}
#endif // CONFIG_MULTITHREAD
vpx_free(row_mt_sync->cur_col);
// clear the structure as the source of this call may be dynamic change
// in tiles in which case this call will be followed by an _alloc()
// which may fail.
vp9_zero(*row_mt_sync);
}
}
void vp9_setup_past_independence(VP9_COMMON *cm) {
// Reset the segment feature data to the default stats:
// Features disabled, 0, with delta coding (Default state).
struct loopfilter *const lf = &cm->lf;
int i;
vp9_clearall_segfeatures(&cm->seg);
cm->seg.abs_delta = SEGMENT_DELTADATA;
if (cm->last_frame_seg_map && !cm->frame_parallel_decode)
memset(cm->last_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols));
if (cm->current_frame_seg_map)
memset(cm->current_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols));
// Reset the mode ref deltas for loop filter
vp9_zero(lf->last_ref_deltas);
vp9_zero(lf->last_mode_deltas);
set_default_lf_deltas(lf);
// To force update of the sharpness
lf->last_sharpness_level = -1;
vp9_default_coef_probs(cm);
init_mode_probs(cm->fc);
vp9_init_mv_probs(cm);
cm->fc->initialized = 1;
if (cm->frame_type == KEY_FRAME ||
cm->error_resilient_mode || cm->reset_frame_context == 3) {
// Reset all frame contexts.
for (i = 0; i < FRAME_CONTEXTS; ++i)
cm->frame_contexts[i] = *cm->fc;
} else if (cm->reset_frame_context == 2) {
// Reset only the frame context specified in the frame header.
cm->frame_contexts[cm->frame_context_idx] = *cm->fc;
}
// prev_mip will only be allocated in encoder.
if (frame_is_intra_only(cm) && cm->prev_mip && !cm->frame_parallel_decode)
memset(cm->prev_mip, 0,
cm->mi_stride * (cm->mi_rows + 1) * sizeof(*cm->prev_mip));
vp9_zero(cm->ref_frame_sign_bias);
cm->frame_context_idx = 0;
}
void vp9_setup_past_independence(VP9_COMMON *cm) {
// Reset the segment feature data to the default stats:
// Features disabled, 0, with delta coding (Default state).
struct loopfilter *const lf = &cm->lf;
int i;
vp9_clearall_segfeatures(&cm->seg);
cm->seg.abs_delta = SEGMENT_DELTADATA;
if (cm->last_frame_seg_map)
vpx_memset(cm->last_frame_seg_map, 0, (cm->mi_rows * cm->mi_cols));
// Reset the mode ref deltas for loop filter
vp9_zero(lf->last_ref_deltas);
vp9_zero(lf->last_mode_deltas);
set_default_lf_deltas(lf);
// To force update of the sharpness
lf->last_sharpness_level = -1;
vp9_default_coef_probs(cm);
vp9_init_mbmode_probs(cm);
vp9_init_mv_probs(cm);
vp9_copy(cm->fc.inter_mode_probs, default_inter_mode_probs);
if (cm->frame_type == KEY_FRAME ||
cm->error_resilient_mode || cm->reset_frame_context == 3) {
// Reset all frame contexts.
for (i = 0; i < NUM_FRAME_CONTEXTS; ++i)
cm->frame_contexts[i] = cm->fc;
} else if (cm->reset_frame_context == 2) {
// Reset only the frame context specified in the frame header.
cm->frame_contexts[cm->frame_context_idx] = cm->fc;
}
vpx_memset(cm->prev_mip, 0,
cm->mode_info_stride * (cm->mi_rows + 1) * sizeof(MODE_INFO));
vpx_memset(cm->mip, 0,
cm->mode_info_stride * (cm->mi_rows + 1) * sizeof(MODE_INFO));
vp9_update_mode_info_border(cm, cm->mip);
vp9_update_mode_info_border(cm, cm->prev_mip);
vp9_zero(cm->ref_frame_sign_bias);
cm->frame_context_idx = 0;
}
VP9Decoder *vp9_decoder_create() {
VP9Decoder *const pbi = vpx_memalign(32, sizeof(*pbi));
VP9_COMMON *const cm = pbi ? &pbi->common : NULL;
if (!cm)
return NULL;
vp9_zero(*pbi);
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
vp9_decoder_remove(pbi);
return NULL;
}
cm->error.setjmp = 1;
CHECK_MEM_ERROR(cm, cm->fc,
(FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
CHECK_MEM_ERROR(cm, cm->frame_contexts,
(FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
sizeof(*cm->frame_contexts)));
pbi->need_resync = 1;
initialize_dec();
// Initialize the references to not point to any frame buffers.
vpx_memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
cm->current_video_frame = 0;
pbi->ready_for_new_data = 1;
cm->bit_depth = VPX_BITS_8;
cm->dequant_bit_depth = VPX_BITS_8;
cm->alloc_mi = vp9_dec_alloc_mi;
cm->free_mi = vp9_dec_free_mi;
cm->setup_mi = vp9_dec_setup_mi;
// vp9_init_dequantizer() is first called here. Add check in
// frame_init_dequantizer() to avoid unnecessary calling of
// vp9_init_dequantizer() for every frame.
vp9_init_dequantizer(cm);
vp9_loop_filter_init(cm);
cm->error.setjmp = 0;
vp9_get_worker_interface()->init(&pbi->lf_worker);
return pbi;
}
VP9Decoder *vp9_decoder_create(BufferPool *const pool) {
VP9Decoder *volatile const pbi = vpx_memalign(32, sizeof(*pbi));
VP9_COMMON *volatile const cm = pbi ? &pbi->common : NULL;
if (!cm)
return NULL;
vp9_zero(*pbi);
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
vp9_decoder_remove(pbi);
return NULL;
}
cm->error.setjmp = 1;
CHECK_MEM_ERROR(cm, cm->fc,
(FRAME_CONTEXT *)vpx_calloc(1, sizeof(*cm->fc)));
CHECK_MEM_ERROR(cm, cm->frame_contexts,
(FRAME_CONTEXT *)vpx_calloc(FRAME_CONTEXTS,
sizeof(*cm->frame_contexts)));
pbi->need_resync = 1;
once(initialize_dec);
// Initialize the references to not point to any frame buffers.
memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
memset(&cm->next_ref_frame_map, -1, sizeof(cm->next_ref_frame_map));
cm->current_video_frame = 0;
pbi->ready_for_new_data = 1;
pbi->common.buffer_pool = pool;
cm->bit_depth = VPX_BITS_8;
cm->dequant_bit_depth = VPX_BITS_8;
cm->alloc_mi = vp9_dec_alloc_mi;
cm->free_mi = vp9_dec_free_mi;
cm->setup_mi = vp9_dec_setup_mi;
vp9_loop_filter_init(cm);
cm->error.setjmp = 0;
vpx_get_worker_interface()->init(&pbi->lf_worker);
return pbi;
}
void init_context_counters(void) {
FILE *f = fopen("context.bin", "rb");
if (!f) {
vp9_zero(context_counters);
} else {
fread(context_counters, sizeof(context_counters), 1, f);
fclose(f);
}
f = fopen("treeupdate.bin", "rb");
if (!f) {
vpx_memset(tree_update_hist, 0, sizeof(tree_update_hist));
} else {
fread(tree_update_hist, sizeof(tree_update_hist), 1, f);
fclose(f);
}
}
static void write_modes(VP9_COMP *cpi,
const TileInfo *const tile, vpx_writer *w,
TOKENEXTRA **tok, const TOKENEXTRA *const tok_end) {
const VP9_COMMON *const cm = &cpi->common;
MACROBLOCKD *const xd = &cpi->td.mb.e_mbd;
int mi_row, mi_col;
set_partition_probs(cm, xd);
for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end;
mi_row += MI_BLOCK_SIZE) {
vp9_zero(xd->left_seg_context);
for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end;
mi_col += MI_BLOCK_SIZE)
write_modes_sb(cpi, tile, w, tok, tok_end, mi_row, mi_col,
BLOCK_64X64);
}
}
static int first_pass_worker_hook(void *arg1, void *arg2) {
EncWorkerData *const thread_data = (EncWorkerData *)arg1;
MultiThreadHandle *multi_thread_ctxt = (MultiThreadHandle *)arg2;
VP9_COMP *const cpi = thread_data->cpi;
const VP9_COMMON *const cm = &cpi->common;
const int tile_cols = 1 << cm->log2_tile_cols;
int tile_row, tile_col;
TileDataEnc *this_tile;
int end_of_frame;
int thread_id = thread_data->thread_id;
int cur_tile_id = multi_thread_ctxt->thread_id_to_tile_id[thread_id];
JobNode *proc_job = NULL;
FIRSTPASS_DATA fp_acc_data;
MV zero_mv = { 0, 0 };
MV best_ref_mv;
int mb_row;
end_of_frame = 0;
while (0 == end_of_frame) {
// Get the next job in the queue
proc_job =
(JobNode *)vp9_enc_grp_get_next_job(multi_thread_ctxt, cur_tile_id);
if (NULL == proc_job) {
// Query for the status of other tiles
end_of_frame = vp9_get_tiles_proc_status(
multi_thread_ctxt, thread_data->tile_completion_status, &cur_tile_id,
tile_cols);
} else {
tile_col = proc_job->tile_col_id;
tile_row = proc_job->tile_row_id;
this_tile = &cpi->tile_data[tile_row * tile_cols + tile_col];
mb_row = proc_job->vert_unit_row_num;
best_ref_mv = zero_mv;
vp9_zero(fp_acc_data);
fp_acc_data.image_data_start_row = INVALID_ROW;
vp9_first_pass_encode_tile_mb_row(cpi, thread_data->td, &fp_acc_data,
this_tile, &best_ref_mv, mb_row);
}
}
return 0;
}
VP9D_PTR vp9_create_decompressor(VP9D_CONFIG *oxcf) {
VP9D_COMP *const pbi = vpx_memalign(32, sizeof(VP9D_COMP));
VP9_COMMON *const cm = pbi ? &pbi->common : NULL;
if (!cm)
return NULL;
vp9_zero(*pbi);
// Initialize the references to not point to any frame buffers.
memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
vp9_remove_decompressor(pbi);
return NULL;
}
cm->error.setjmp = 1;
vp9_initialize_dec();
vp9_create_common(cm);
pbi->oxcf = *oxcf;
pbi->ready_for_new_data = 1;
cm->current_video_frame = 0;
// vp9_init_dequantizer() is first called here. Add check in
// frame_init_dequantizer() to avoid unnecessary calling of
// vp9_init_dequantizer() for every frame.
vp9_init_dequantizer(cm);
vp9_loop_filter_init(cm);
cm->error.setjmp = 0;
pbi->decoded_key_frame = 0;
init_macroblockd(pbi);
vp9_worker_init(&pbi->lf_worker);
return pbi;
}
VP9Decoder *vp9_decoder_create() {
VP9Decoder *const pbi = vpx_memalign(32, sizeof(*pbi));
VP9_COMMON *const cm = pbi ? &pbi->common : NULL;
if (!cm)
return NULL;
vp9_zero(*pbi);
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
vp9_decoder_remove(pbi);
return NULL;
}
cm->error.setjmp = 1;
initialize_dec();
vp9_rtcd();
// Initialize the references to not point to any frame buffers.
vpx_memset(&cm->ref_frame_map, -1, sizeof(cm->ref_frame_map));
cm->current_video_frame = 0;
pbi->ready_for_new_data = 1;
// vp9_init_dequantizer() is first called here. Add check in
// frame_init_dequantizer() to avoid unnecessary calling of
// vp9_init_dequantizer() for every frame.
vp9_init_dequantizer(cm);
vp9_loop_filter_init(cm);
cm->error.setjmp = 0;
vp9_get_worker_interface()->init(&pbi->lf_worker);
return pbi;
}
static vpx_codec_err_t decode_one_recon_ex(vpx_codec_alg_priv_t *ctx,
const uint8_t **data, unsigned int data_sz,
void *user_priv, int64_t deadline, void *texture) {
vpx_codec_err_t res = VPX_CODEC_OK;
VP9D_COMP *pbi;
VP9D_COMP *pbi_storage;
VP9D_COMP *my_pbi;
static int flag = 0;
int i_is_last_frame = 0;
int ret = -1;
struct vpx_usec_timer timer;
unsigned long yuv2rgb_time = 0;
unsigned long decode_time = 0;
// ctx->img_avail = 0;
vpx_usec_timer_start(&timer);
if (data_sz == 0) {
pbi = (VP9D_COMP *)ctx->pbi;
if (!pbi->l_bufpool_flag_output) {
return 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);
/* 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 = vp9_mem_req_segs[i].id;
ctx->mmaps[i].sz = vp9_mem_req_segs[i].sz;
ctx->mmaps[i].align = vp9_mem_req_segs[i].align;
ctx->mmaps[i].flags = vp9_mem_req_segs[i].flags;
if (!ctx->mmaps[i].sz)
ctx->mmaps[i].sz = vp9_mem_req_segs[i].calc_sz(&cfg,
ctx->base.init_flags);
res = vpx_mmap_alloc(&ctx->mmaps[i]);
}
if (!res)
vp9_finalize_mmaps(ctx);
ctx->defer_alloc = 0;
}
/* Initialize the decoder instance on the first frame*/
if (!res && !ctx->decoder_init) {
res = vpx_validate_mmaps(&ctx->si, ctx->mmaps,
vp9_mem_req_segs, NELEMENTS(vp9_mem_req_segs),
ctx->base.init_flags);
if (!res) {
VP9D_CONFIG oxcf;
VP9D_PTR optr;
VP9D_COMP *const new_pbi = vpx_memalign(32, sizeof(VP9D_COMP));
VP9D_COMP *const new_pbi_two = vpx_memalign(32, sizeof(VP9D_COMP));
vp9_initialize_dec();
oxcf.width = ctx->si.w;
oxcf.height = ctx->si.h;
oxcf.version = 9;
oxcf.postprocess = 0;
oxcf.max_threads = ctx->cfg.threads;
oxcf.inv_tile_order = ctx->invert_tile_order;
optr = vp9_create_decompressor_recon(&oxcf);
vp9_zero(*new_pbi);
vp9_zero(*new_pbi_two);
// 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;
VP9_COMMON *const cm0 = &new_pbi->common;
VP9_COMMON *const cm1 = &new_pbi_two->common;
if (ctx->fb_list != NULL && ctx->realloc_fb_cb != NULL &&
ctx->fb_count > 0) {
cm->fb_list = ctx->fb_list;
cm->fb_count = ctx->fb_count;
cm->realloc_fb_cb = ctx->realloc_fb_cb;
cm->user_priv = ctx->user_priv;
CpuFlag = 1;
} else {
CpuFlag = 0;
cm->fb_count = FRAME_BUFFERS;
}
cm->fb_lru = ctx->fb_lru;
CHECK_MEM_ERROR(cm, cm->yv12_fb,
vpx_calloc(cm->fb_count, sizeof(*cm->yv12_fb)));
CHECK_MEM_ERROR(cm, cm->fb_idx_ref_cnt,
vpx_calloc(cm->fb_count, sizeof(*cm->fb_idx_ref_cnt)));
if (cm->fb_lru) {
CHECK_MEM_ERROR(cm, cm->fb_idx_ref_lru,
vpx_calloc(cm->fb_count,
sizeof(*cm->fb_idx_ref_lru)));
}
ctx->pbi = optr;
ctx->storage_pbi[0] = new_pbi;
ctx->storage_pbi[1] = new_pbi_two;
// cm 0
if (ctx->fb_list != NULL && ctx->realloc_fb_cb != NULL &&
ctx->fb_count > 0) {
cm0->fb_list = ctx->fb_list;
cm0->fb_count = ctx->fb_count;
cm0->realloc_fb_cb = ctx->realloc_fb_cb;
cm0->user_priv = ctx->user_priv;
} else {
cm0->fb_count = FRAME_BUFFERS;
}
cm0->fb_lru = ctx->fb_lru;
// CHECK_MEM_ERROR(cm, cm->yv12_fb,
// vpx_calloc(cm->fb_count, sizeof(*cm->yv12_fb)));
CHECK_MEM_ERROR(cm0, cm0->fb_idx_ref_cnt,
vpx_calloc(cm0->fb_count, sizeof(*cm0->fb_idx_ref_cnt)));
if (cm0->fb_lru) {
CHECK_MEM_ERROR(cm0, cm0->fb_idx_ref_lru,
vpx_calloc(cm0->fb_count,
sizeof(*cm0->fb_idx_ref_lru)));
}
// cm 1
if (ctx->fb_list != NULL && ctx->realloc_fb_cb != NULL &&
ctx->fb_count > 0) {
cm1->fb_list = ctx->fb_list;
cm1->fb_count = ctx->fb_count;
cm1->realloc_fb_cb = ctx->realloc_fb_cb;
cm1->user_priv = ctx->user_priv;
} else {
cm1->fb_count = FRAME_BUFFERS;
}
cm1->fb_lru = ctx->fb_lru;
// CHECK_MEM_ERROR(cm, cm->yv12_fb,
// vpx_calloc(cm->fb_count, sizeof(*cm->yv12_fb)));
CHECK_MEM_ERROR(cm1, cm1->fb_idx_ref_cnt,
vpx_calloc(cm1->fb_count, sizeof(*cm1->fb_idx_ref_cnt)));
if (cm1->fb_lru) {
CHECK_MEM_ERROR(cm1, cm1->fb_idx_ref_lru,
vpx_calloc(cm1->fb_count,
sizeof(*cm1->fb_idx_ref_lru)));
}
}
}
ctx->decoder_init = 1;
}
if (!res && ctx->pbi) {
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 0
if (vp9_receive_compressed_data(ctx->pbi, data_sz, data, deadline)) {
VP9D_COMP *pbi = (VP9D_COMP*)ctx->pbi;
res = update_error_state(ctx, &pbi->common.error);
}
if (!res && 0 == vp9_get_raw_frame(ctx->pbi, &sd, &time_stamp,
&time_end_stamp, &flags)) {
yuvconfig2image(&ctx->img, &sd, user_priv);
ctx->img_avail = 1;
}
#endif
if (data_sz == 0) {
i_is_last_frame = 1;
}
if (vp9_receive_compressed_data_recon(ctx->pbi, ctx->storage_pbi, data_sz, data,
deadline, i_is_last_frame)) {
pbi = (VP9D_COMP *)ctx->pbi;
if (pbi->l_bufpool_flag_output == 0)
pbi_storage = (VP9D_COMP *)ctx->storage_pbi[1];
else
pbi_storage = (VP9D_COMP *)ctx->storage_pbi[pbi->l_bufpool_flag_output & 1];
res = update_error_state(ctx, &pbi_storage->common.error);
}
vpx_usec_timer_mark(&timer);
decode_time = (unsigned int)vpx_usec_timer_elapsed(&timer);
if (ctx->pbi) {
pbi = (VP9D_COMP *)ctx->pbi;
if (pbi->l_bufpool_flag_output) {
ret = vp9_get_raw_frame(ctx->storage_pbi[pbi->l_bufpool_flag_output & 1],
&sd, &time_stamp, &time_end_stamp, &flags);
if (!pbi->res && 0 == ret ) {
//for render
my_pbi = (VP9D_COMP *)(ctx->storage_pbi[pbi->l_bufpool_flag_output & 1]);
yuv2rgba_ocl_obj.y_plane_offset = my_pbi->common.frame_to_show->y_buffer -
inter_ocl_obj.buffer_pool_map_ptr;
yuv2rgba_ocl_obj.u_plane_offset = my_pbi->common.frame_to_show->u_buffer -
inter_ocl_obj.buffer_pool_map_ptr;
yuv2rgba_ocl_obj.v_plane_offset = my_pbi->common.frame_to_show->v_buffer -
inter_ocl_obj.buffer_pool_map_ptr;
yuv2rgba_ocl_obj.Y_stride = my_pbi->common.frame_to_show->y_stride;
yuv2rgba_ocl_obj.UV_stride = my_pbi->common.frame_to_show->uv_stride;
yuv2rgba_ocl_obj.globalThreads[0] = my_pbi->common.width >> 1;
yuv2rgba_ocl_obj.globalThreads[1] = my_pbi->common.height >> 1;
vpx_usec_timer_start(&timer);
vp9_yuv2rgba(&yuv2rgba_ocl_obj, texture);
vpx_usec_timer_mark(&timer);
yuv2rgb_time = (unsigned int)vpx_usec_timer_elapsed(&timer);
fprintf(pLog, "decode one frame time(without YUV to RGB): %lu us\n"
"the whole time of YUV to RGB: %lu us\n", decode_time, yuv2rgb_time);
// for render end
yuvconfig2image(&ctx->img, &sd, user_priv);
ctx->img_avail = 1;
}
void init_tx_count_stats() {
vp9_zero(tx_count_32x32p_stats);
vp9_zero(tx_count_16x16p_stats);
vp9_zero(tx_count_8x8p_stats);
}
void init_switchable_interp_stats() {
vp9_zero(switchable_interp_stats);
}
static void update_mbgraph_frame_stats(VP9_COMP *cpi,
MBGRAPH_FRAME_STATS *stats,
YV12_BUFFER_CONFIG *buf,
YV12_BUFFER_CONFIG *golden_ref,
YV12_BUFFER_CONFIG *alt_ref) {
MACROBLOCK *const x = &cpi->td.mb;
MACROBLOCKD *const xd = &x->e_mbd;
VP9_COMMON *const cm = &cpi->common;
int mb_col, mb_row, offset = 0;
int mb_y_offset = 0, arf_y_offset = 0, gld_y_offset = 0;
MV gld_top_mv = {0, 0};
MODE_INFO mi_local;
vp9_zero(mi_local);
// Set up limit values for motion vectors to prevent them extending outside
// the UMV borders.
x->mv_row_min = -BORDER_MV_PIXELS_B16;
x->mv_row_max = (cm->mb_rows - 1) * 8 + BORDER_MV_PIXELS_B16;
xd->up_available = 0;
xd->plane[0].dst.stride = buf->y_stride;
xd->plane[0].pre[0].stride = buf->y_stride;
xd->plane[1].dst.stride = buf->uv_stride;
xd->mi[0] = &mi_local;
mi_local.mbmi.sb_type = BLOCK_16X16;
mi_local.mbmi.ref_frame[0] = LAST_FRAME;
mi_local.mbmi.ref_frame[1] = NONE;
for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) {
MV gld_left_mv = gld_top_mv;
int mb_y_in_offset = mb_y_offset;
int arf_y_in_offset = arf_y_offset;
int gld_y_in_offset = gld_y_offset;
// Set up limit values for motion vectors to prevent them extending outside
// the UMV borders.
x->mv_col_min = -BORDER_MV_PIXELS_B16;
x->mv_col_max = (cm->mb_cols - 1) * 8 + BORDER_MV_PIXELS_B16;
xd->left_available = 0;
for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
MBGRAPH_MB_STATS *mb_stats = &stats->mb_stats[offset + mb_col];
update_mbgraph_mb_stats(cpi, mb_stats, buf, mb_y_in_offset,
golden_ref, &gld_left_mv, alt_ref,
mb_row, mb_col);
gld_left_mv = mb_stats->ref[GOLDEN_FRAME].m.mv.as_mv;
if (mb_col == 0) {
gld_top_mv = gld_left_mv;
}
xd->left_available = 1;
mb_y_in_offset += 16;
gld_y_in_offset += 16;
arf_y_in_offset += 16;
x->mv_col_min -= 16;
x->mv_col_max -= 16;
}
xd->up_available = 1;
mb_y_offset += buf->y_stride * 16;
gld_y_offset += golden_ref->y_stride * 16;
if (alt_ref)
arf_y_offset += alt_ref->y_stride * 16;
x->mv_row_min -= 16;
x->mv_row_max -= 16;
offset += cm->mb_cols;
}
}
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);
}
}
}
void vp9_mb_copy(VP9_COMP *cpi, MACROBLOCK *x_dst, MACROBLOCK *x_src) {
VP9_COMMON *cm = &cpi->common;
MACROBLOCKD *const xd_dst = &x_dst->e_mbd;
MACROBLOCKD *const xd_src = &x_src->e_mbd;
int i;
for (i = 0; i < MAX_MB_PLANE; ++i) {
x_dst->plane[i] = x_src->plane[i];
xd_dst->plane[i] = xd_src->plane[i];
}
xd_dst->mi_stride = xd_src->mi_stride;
xd_dst->mi = xd_src->mi;
xd_dst->mi[0] = xd_src->mi[0];
xd_dst->block_refs[0] = xd_src->block_refs[0];
xd_dst->block_refs[1] = xd_src->block_refs[1];
xd_dst->cur_buf = xd_src->cur_buf;
#if CONFIG_VP9_HIGHBITDEPTH
xd_dst->bd = xd_src->bd;
#endif
xd_dst->lossless = xd_src->lossless;
xd_dst->corrupted = 0;
for (i = 0; i < MAX_MB_PLANE; i++) {
xd_dst->above_context[i] = xd_src->above_context[i];
}
xd_dst->above_seg_context = xd_src->above_seg_context;
x_dst->skip_block = x_src->skip_block;
x_dst->select_tx_size = x_src->select_tx_size;
x_dst->skip_recode = x_src->skip_recode;
x_dst->skip_optimize = x_src->skip_optimize;
x_dst->q_index = x_src->q_index;
x_dst->errorperbit = x_src->errorperbit;
x_dst->sadperbit16 = x_src->sadperbit16;
x_dst->sadperbit4 = x_src->sadperbit4;
x_dst->rddiv = x_src->rddiv;
x_dst->rdmult = x_src->rdmult;
x_dst->mb_energy = x_src->mb_energy;
for (i = 0; i < MV_JOINTS; i++) {
x_dst->nmvjointcost[i] = x_src->nmvjointcost[i];
x_dst->nmvjointsadcost[i] = x_src->nmvjointsadcost[i];
}
x_dst->nmvcost[0] = x_src->nmvcost[0];
x_dst->nmvcost[1] = x_src->nmvcost[1];
x_dst->nmvcost_hp[0] = x_src->nmvcost_hp[0];
x_dst->nmvcost_hp[1] = x_src->nmvcost_hp[1];
x_dst->mvcost = x_src->mvcost;
x_dst->nmvsadcost[0] = x_src->nmvsadcost[0];
x_dst->nmvsadcost[1] = x_src->nmvsadcost[1];
x_dst->nmvsadcost_hp[0] = x_src->nmvsadcost_hp[0];
x_dst->nmvsadcost_hp[1] = x_src->nmvsadcost_hp[1];
x_dst->mvsadcost = x_src->mvsadcost;
x_dst->min_partition_size = x_src->min_partition_size;
x_dst->max_partition_size = x_src->max_partition_size;
vpx_memcpy(x_dst->token_costs, x_src->token_costs,
sizeof(x_src->token_costs));
vp9_zero(x_dst->counts);
vp9_zero(x_dst->coef_counts);
vpx_memcpy(x_dst->rd.threshes, cpi->rd.threshes, sizeof(cpi->rd.threshes));
// freq scaling factors initialization has to happen only for video frame 1.
// For all other frames, It self corrects itself while encoding.
if (cm->current_video_frame == 0)
vpx_memcpy(x_dst->rd.thresh_freq_fact, cpi->rd.thresh_freq_fact,
sizeof(cpi->rd.thresh_freq_fact));
vp9_zero(x_dst->rd.comp_pred_diff);
vp9_zero(x_dst->rd.tx_select_diff);
vp9_zero(x_dst->rd.tx_select_threshes);
vp9_zero(x_dst->rd.filter_diff);
x_dst->rd.RDMULT = cpi->rd.RDMULT;
x_dst->rd.RDDIV = cpi->rd.RDDIV;
x_dst->data_parallel_processing = 0;
x_dst->optimize = x_src->optimize;
x_dst->quant_fp = x_src->quant_fp;
vp9_zero(x_dst->skip_txfm);
vp9_zero(x_dst->bsse);
x_dst->fwd_txm4x4 = x_src->fwd_txm4x4;
x_dst->itxm_add = x_src->itxm_add;
#if CONFIG_VP9_HIGHBITDEPTH
x_dst->itxm_add = x_src->itxm_add;
#endif
}
int main(int argc, char **argv) {
FILE *infile = NULL;
// Encoder
vpx_codec_ctx_t ecodec;
vpx_codec_enc_cfg_t cfg;
unsigned int frame_in = 0;
vpx_image_t raw;
vpx_codec_err_t res;
VpxVideoInfo info;
VpxVideoWriter *writer = NULL;
const VpxInterface *encoder = NULL;
// Test encoder/decoder mismatch.
int test_decode = 1;
// Decoder
vpx_codec_ctx_t dcodec;
unsigned int frame_out = 0;
// The frame number to set reference frame on
unsigned int update_frame_num = 0;
int mismatch_seen = 0;
const int fps = 30;
const int bitrate = 500;
const char *width_arg = NULL;
const char *height_arg = NULL;
const char *infile_arg = NULL;
const char *outfile_arg = NULL;
const char *update_frame_num_arg = NULL;
unsigned int limit = 0;
vp9_zero(ecodec);
vp9_zero(cfg);
vp9_zero(info);
exec_name = argv[0];
if (argc < 6) die("Invalid number of arguments");
width_arg = argv[1];
height_arg = argv[2];
infile_arg = argv[3];
outfile_arg = argv[4];
update_frame_num_arg = argv[5];
encoder = get_vpx_encoder_by_name("vp9");
if (!encoder) die("Unsupported codec.");
update_frame_num = (unsigned int)strtoul(update_frame_num_arg, NULL, 0);
// In VP9, the reference buffers (cm->buffer_pool->frame_bufs[i].buf) are
// allocated while calling vpx_codec_encode(), thus, setting reference for
// 1st frame isn't supported.
if (update_frame_num <= 1) {
die("Couldn't parse frame number '%s'\n", update_frame_num_arg);
}
if (argc > 6) {
limit = (unsigned int)strtoul(argv[6], NULL, 0);
if (update_frame_num > limit)
die("Update frame number couldn't larger than limit\n");
}
info.codec_fourcc = encoder->fourcc;
info.frame_width = (int)strtol(width_arg, NULL, 0);
info.frame_height = (int)strtol(height_arg, NULL, 0);
info.time_base.numerator = 1;
info.time_base.denominator = fps;
if (info.frame_width <= 0 || info.frame_height <= 0 ||
(info.frame_width % 2) != 0 || (info.frame_height % 2) != 0) {
die("Invalid frame size: %dx%d", info.frame_width, info.frame_height);
}
if (!vpx_img_alloc(&raw, VPX_IMG_FMT_I420, info.frame_width,
info.frame_height, 1)) {
die("Failed to allocate image.");
}
printf("Using %s\n", vpx_codec_iface_name(encoder->codec_interface()));
res = vpx_codec_enc_config_default(encoder->codec_interface(), &cfg, 0);
if (res) die_codec(&ecodec, "Failed to get default codec config.");
cfg.g_w = info.frame_width;
cfg.g_h = info.frame_height;
cfg.g_timebase.num = info.time_base.numerator;
cfg.g_timebase.den = info.time_base.denominator;
cfg.rc_target_bitrate = bitrate;
cfg.g_lag_in_frames = 3;
writer = vpx_video_writer_open(outfile_arg, kContainerIVF, &info);
if (!writer) die("Failed to open %s for writing.", outfile_arg);
if (!(infile = fopen(infile_arg, "rb")))
die("Failed to open %s for reading.", infile_arg);
if (vpx_codec_enc_init(&ecodec, encoder->codec_interface(), &cfg, 0))
die_codec(&ecodec, "Failed to initialize encoder");
// Disable alt_ref.
if (vpx_codec_control(&ecodec, VP8E_SET_ENABLEAUTOALTREF, 0))
die_codec(&ecodec, "Failed to set enable auto alt ref");
if (test_decode) {
const VpxInterface *decoder = get_vpx_decoder_by_name("vp9");
if (vpx_codec_dec_init(&dcodec, decoder->codec_interface(), NULL, 0))
die_codec(&dcodec, "Failed to initialize decoder.");
}
// Encode frames.
while (vpx_img_read(&raw, infile)) {
if (limit && frame_in >= limit) break;
if (update_frame_num > 1 && frame_out + 1 == update_frame_num) {
vpx_ref_frame_t ref;
ref.frame_type = VP8_LAST_FRAME;
ref.img = raw;
// Set reference frame in encoder.
if (vpx_codec_control(&ecodec, VP8_SET_REFERENCE, &ref))
die_codec(&ecodec, "Failed to set reference frame");
printf(" <SET_REF>");
// If set_reference in decoder is commented out, the enc/dec mismatch
// would be seen.
if (test_decode) {
if (vpx_codec_control(&dcodec, VP8_SET_REFERENCE, &ref))
die_codec(&dcodec, "Failed to set reference frame");
}
}
encode_frame(&ecodec, &raw, frame_in, writer, test_decode, &dcodec,
&frame_out, &mismatch_seen);
frame_in++;
if (mismatch_seen) break;
}
// Flush encoder.
if (!mismatch_seen)
while (encode_frame(&ecodec, NULL, frame_in, writer, test_decode, &dcodec,
&frame_out, &mismatch_seen)) {
}
printf("\n");
fclose(infile);
printf("Processed %d frames.\n", frame_out);
if (test_decode) {
if (!mismatch_seen)
printf("Encoder/decoder results are matching.\n");
else
printf("Encoder/decoder results are NOT matching.\n");
}
if (test_decode)
if (vpx_codec_destroy(&dcodec))
die_codec(&dcodec, "Failed to destroy decoder");
vpx_img_free(&raw);
if (vpx_codec_destroy(&ecodec))
die_codec(&ecodec, "Failed to destroy encoder.");
vpx_video_writer_close(writer);
return EXIT_SUCCESS;
}
