static void alloc_mode_context(VP9_COMMON *cm, int num_4x4_blk,
PICK_MODE_CONTEXT *ctx) {
const int num_blk = (num_4x4_blk < 4 ? 4 : num_4x4_blk);
const int num_pix = num_blk << 4;
int i, k;
ctx->num_4x4_blk = num_blk;
CHECK_MEM_ERROR(cm, ctx->zcoeff_blk,
vpx_calloc(num_4x4_blk, sizeof(uint8_t)));
for (i = 0; i < MAX_MB_PLANE; ++i) {
for (k = 0; k < 3; ++k) {
CHECK_MEM_ERROR(cm, ctx->coeff[i][k],
vpx_memalign(16, num_pix * sizeof(int16_t)));
CHECK_MEM_ERROR(cm, ctx->qcoeff[i][k],
vpx_memalign(16, num_pix * sizeof(int16_t)));
CHECK_MEM_ERROR(cm, ctx->dqcoeff[i][k],
vpx_memalign(16, num_pix * sizeof(int16_t)));
CHECK_MEM_ERROR(cm, ctx->eobs[i][k],
vpx_memalign(16, num_pix * sizeof(uint16_t)));
ctx->coeff_pbuf[i][k] = ctx->coeff[i][k];
ctx->qcoeff_pbuf[i][k] = ctx->qcoeff[i][k];
ctx->dqcoeff_pbuf[i][k] = ctx->dqcoeff[i][k];
ctx->eobs_pbuf[i][k] = ctx->eobs[i][k];
}
}
}
void vp9_dec_alloc_row_mt_mem(RowMTWorkerData *row_mt_worker_data,
VP9_COMMON *cm, int num_sbs, int max_threads,
int num_jobs) {
int plane;
const size_t dqcoeff_size = (num_sbs << DQCOEFFS_PER_SB_LOG2) *
sizeof(*row_mt_worker_data->dqcoeff[0]);
row_mt_worker_data->num_jobs = num_jobs;
#if CONFIG_MULTITHREAD
{
int i;
CHECK_MEM_ERROR(
cm, row_mt_worker_data->recon_sync_mutex,
vpx_malloc(sizeof(*row_mt_worker_data->recon_sync_mutex) * num_jobs));
if (row_mt_worker_data->recon_sync_mutex) {
for (i = 0; i < num_jobs; ++i) {
pthread_mutex_init(&row_mt_worker_data->recon_sync_mutex[i], NULL);
}
}
CHECK_MEM_ERROR(
cm, row_mt_worker_data->recon_sync_cond,
vpx_malloc(sizeof(*row_mt_worker_data->recon_sync_cond) * num_jobs));
if (row_mt_worker_data->recon_sync_cond) {
for (i = 0; i < num_jobs; ++i) {
pthread_cond_init(&row_mt_worker_data->recon_sync_cond[i], NULL);
}
}
}
#endif
row_mt_worker_data->num_sbs = num_sbs;
for (plane = 0; plane < 3; ++plane) {
CHECK_MEM_ERROR(cm, row_mt_worker_data->dqcoeff[plane],
vpx_memalign(16, dqcoeff_size));
memset(row_mt_worker_data->dqcoeff[plane], 0, dqcoeff_size);
CHECK_MEM_ERROR(cm, row_mt_worker_data->eob[plane],
vpx_calloc(num_sbs << EOBS_PER_SB_LOG2,
sizeof(*row_mt_worker_data->eob[plane])));
}
CHECK_MEM_ERROR(cm, row_mt_worker_data->partition,
vpx_calloc(num_sbs * PARTITIONS_PER_SB,
sizeof(*row_mt_worker_data->partition)));
CHECK_MEM_ERROR(cm, row_mt_worker_data->recon_map,
vpx_calloc(num_sbs, sizeof(*row_mt_worker_data->recon_map)));
// allocate memory for thread_data
if (row_mt_worker_data->thread_data == NULL) {
const size_t thread_size =
max_threads * sizeof(*row_mt_worker_data->thread_data);
CHECK_MEM_ERROR(cm, row_mt_worker_data->thread_data,
vpx_memalign(32, thread_size));
}
}
int vp8_yv12_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
int width, int height, int border) {
if (ybf) {
int aligned_width = (width + 15) & ~15;
int aligned_height = (height + 15) & ~15;
int y_stride = ((aligned_width + 2 * border) + 31) & ~31;
int yplane_size = (aligned_height + 2 * border) * y_stride;
int uv_width = aligned_width >> 1;
int uv_height = aligned_height >> 1;
/** There is currently a bunch of code which assumes
* uv_stride == y_stride/2, so enforce this here. */
int uv_stride = y_stride >> 1;
int uvplane_size = (uv_height + border) * uv_stride;
const int frame_size = yplane_size + 2 * uvplane_size;
if (!ybf->buffer_alloc) {
ybf->buffer_alloc = (uint8_t *)vpx_memalign(32, frame_size);
ybf->buffer_alloc_sz = frame_size;
}
if (!ybf->buffer_alloc || ybf->buffer_alloc_sz < frame_size)
return -1;
/* Only support allocating buffers that have a border that's a multiple
* of 32. The border restriction is required to get 16-byte alignment of
* the start of the chroma rows without introducing an arbitrary gap
* between planes, which would break the semantics of things like
* vpx_img_set_rect(). */
if (border & 0x1f)
return -3;
ybf->y_crop_width = width;
ybf->y_crop_height = height;
ybf->y_width = aligned_width;
ybf->y_height = aligned_height;
ybf->y_stride = y_stride;
ybf->uv_crop_width = (width + 1) / 2;
ybf->uv_crop_height = (height + 1) / 2;
ybf->uv_width = uv_width;
ybf->uv_height = uv_height;
ybf->uv_stride = uv_stride;
ybf->alpha_width = 0;
ybf->alpha_height = 0;
ybf->alpha_stride = 0;
ybf->border = border;
ybf->frame_size = frame_size;
ybf->y_buffer = ybf->buffer_alloc + (border * y_stride) + border;
ybf->u_buffer = ybf->buffer_alloc + yplane_size + (border / 2 * uv_stride) + border / 2;
ybf->v_buffer = ybf->buffer_alloc + yplane_size + uvplane_size + (border / 2 * uv_stride) + border / 2;
ybf->alpha_buffer = NULL;
ybf->corrupted = 0; /* assume not currupted by errors */
return 0;
}
return -2;
}
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;
}
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);
}
void vp8cx_create_encoder_threads(VP8_COMP *cpi)
{
cpi->b_multi_threaded = 0;
cpi->processor_core_count = 32; //vp8_get_proc_core_count();
CHECK_MEM_ERROR(cpi->tplist, vpx_malloc(sizeof(TOKENLIST) * cpi->common.mb_rows));
#if CONFIG_MULTITHREAD
if (cpi->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
{
int ithread;
if (cpi->oxcf.multi_threaded > cpi->processor_core_count)
cpi->encoding_thread_count = cpi->processor_core_count - 1;
else
cpi->encoding_thread_count = cpi->oxcf.multi_threaded - 1;
CHECK_MEM_ERROR(cpi->h_encoding_thread, vpx_malloc(sizeof(pthread_t) * cpi->encoding_thread_count));
CHECK_MEM_ERROR(cpi->h_event_mbrencoding, vpx_malloc(sizeof(sem_t) * cpi->encoding_thread_count));
CHECK_MEM_ERROR(cpi->mb_row_ei, vpx_memalign(32, sizeof(MB_ROW_COMP) * cpi->encoding_thread_count));
vpx_memset(cpi->mb_row_ei, 0, sizeof(MB_ROW_COMP) * cpi->encoding_thread_count);
CHECK_MEM_ERROR(cpi->en_thread_data, vpx_malloc(sizeof(ENCODETHREAD_DATA) * cpi->encoding_thread_count));
//cpi->h_event_main = CreateEvent(NULL, FALSE, FALSE, NULL);
sem_init(&cpi->h_event_main, 0, 0);
cpi->b_multi_threaded = 1;
//printf("[VP8:] multi_threaded encoding is enabled with %d threads\n\n", (cpi->encoding_thread_count +1));
for (ithread = 0; ithread < cpi->encoding_thread_count; ithread++)
{
//cpi->h_event_mbrencoding[ithread] = CreateEvent(NULL, FALSE, FALSE, NULL);
sem_init(&cpi->h_event_mbrencoding[ithread], 0, 0);
cpi->en_thread_data[ithread].ithread = ithread;
cpi->en_thread_data[ithread].ptr1 = (void *)cpi;
cpi->en_thread_data[ithread].ptr2 = (void *)&cpi->mb_row_ei[ithread];
//printf(" call begin thread %d \n", ithread);
//cpi->h_encoding_thread[ithread] = (HANDLE)_beginthreadex(
// NULL, // security
// 0, // stksize
// thread_encoding_proc,
// (&cpi->en_thread_data[ithread]), // Thread data
// 0,
// NULL);
pthread_create(&cpi->h_encoding_thread[ithread], 0, thread_encoding_proc, (&cpi->en_thread_data[ithread]));
}
}
#endif
}
static struct VP8D_COMP * create_decompressor(VP8D_CONFIG *oxcf)
{
VP8D_COMP *pbi = vpx_memalign(32, sizeof(VP8D_COMP));
if (!pbi)
return NULL;
memset(pbi, 0, sizeof(VP8D_COMP));
if (setjmp(pbi->common.error.jmp))
{
pbi->common.error.setjmp = 0;
remove_decompressor(pbi);
return 0;
}
pbi->common.error.setjmp = 1;
vp8_create_common(&pbi->common);
pbi->common.current_video_frame = 0;
pbi->ready_for_new_data = 1;
/* vp8cx_init_de_quantizer() is first called here. Add check in frame_init_dequantizer() to avoid
* unnecessary calling of vp8cx_init_de_quantizer() for every frame.
*/
vp8cx_init_de_quantizer(pbi);
vp8_loop_filter_init(&pbi->common);
pbi->common.error.setjmp = 0;
#if CONFIG_ERROR_CONCEALMENT
pbi->ec_enabled = oxcf->error_concealment;
pbi->overlaps = NULL;
#else
(void)oxcf;
pbi->ec_enabled = 0;
#endif
/* Error concealment is activated after a key frame has been
* decoded without errors when error concealment is enabled.
*/
pbi->ec_active = 0;
pbi->decoded_key_frame = 0;
/* Independent partitions is activated when a frame updates the
* token probability table to have equal probabilities over the
* PREV_COEF context.
*/
pbi->independent_partitions = 0;
vp8_setup_block_dptrs(&pbi->mb);
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;
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;
}
VP10Decoder *vp10_decoder_create(BufferPool *const pool) {
VP10Decoder *volatile const pbi = vpx_memalign(32, sizeof(*pbi));
VP10_COMMON *volatile const cm = pbi ? &pbi->common : NULL;
if (!cm)
return NULL;
vp10_zero(*pbi);
if (setjmp(cm->error.jmp)) {
cm->error.setjmp = 0;
vp10_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 = vp10_dec_alloc_mi;
cm->free_mi = vp10_dec_free_mi;
cm->setup_mi = vp10_dec_setup_mi;
vp10_loop_filter_init(cm);
cm->error.setjmp = 0;
vpx_get_worker_interface()->init(&pbi->lf_worker);
return pbi;
}
VP8D_PTR vp8dx_create_decompressor(VP8D_CONFIG *oxcf)
{
VP8D_COMP *pbi = vpx_memalign(32, sizeof(VP8D_COMP));
if (!pbi)
return NULL;
vpx_memset(pbi, 0, sizeof(VP8D_COMP));
if (setjmp(pbi->common.error.jmp))
{
pbi->common.error.setjmp = 0;
vp8dx_remove_decompressor(pbi);
return 0;
}
pbi->common.error.setjmp = 1;
vp8dx_initialize();
vp8_create_common(&pbi->common);
vp8_dmachine_specific_config(pbi);
pbi->common.current_video_frame = 0;
pbi->ready_for_new_data = 1;
pbi->CPUFreq = 0; /*vp8_get_processor_freq();*/
#if CONFIG_MULTITHREAD
pbi->max_threads = oxcf->max_threads;
vp8_decoder_create_threads(pbi);
#endif
/* vp8cx_init_de_quantizer() is first called here. Add check in frame_init_dequantizer() to avoid
* unnecessary calling of vp8cx_init_de_quantizer() for every frame.
*/
vp8cx_init_de_quantizer(pbi);
{
VP8_COMMON *cm = &pbi->common;
vp8_init_loop_filter(cm);
cm->last_frame_type = KEY_FRAME;
cm->last_filter_type = cm->filter_type;
cm->last_sharpness_level = cm->sharpness_level;
}
pbi->common.error.setjmp = 0;
return (VP8D_PTR) pbi;
}
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;
}
void vp8cx_create_encoder_threads(VP8_COMP *cpi)
{
const VP8_COMMON * cm = &cpi->common;
cpi->b_multi_threaded = 0;
cpi->encoding_thread_count = 0;
if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
{
int ithread;
int th_count = cpi->oxcf.multi_threaded - 1;
/* don't allocate more threads than cores available */
if (cpi->oxcf.multi_threaded > cm->processor_core_count)
th_count = cm->processor_core_count - 1;
/* we have th_count + 1 (main) threads processing one row each */
/* no point to have more threads than the sync range allows */
if(th_count > ((cm->mb_cols / cpi->mt_sync_range) - 1))
{
th_count = (cm->mb_cols / cpi->mt_sync_range) - 1;
}
if(th_count == 0)
return;
CHECK_MEM_ERROR(cpi->h_encoding_thread, vpx_malloc(sizeof(pthread_t) * th_count));
CHECK_MEM_ERROR(cpi->h_event_start_encoding, vpx_malloc(sizeof(sem_t) * th_count));
CHECK_MEM_ERROR(cpi->mb_row_ei, vpx_memalign(32, sizeof(MB_ROW_COMP) * th_count));
vpx_memset(cpi->mb_row_ei, 0, sizeof(MB_ROW_COMP) * th_count);
CHECK_MEM_ERROR(cpi->en_thread_data,
vpx_malloc(sizeof(ENCODETHREAD_DATA) * th_count));
CHECK_MEM_ERROR(cpi->mt_current_mb_col,
vpx_malloc(sizeof(*cpi->mt_current_mb_col) * cm->mb_rows));
sem_init(&cpi->h_event_end_encoding, 0, 0);
cpi->b_multi_threaded = 1;
cpi->encoding_thread_count = th_count;
/*
printf("[VP8:] multi_threaded encoding is enabled with %d threads\n\n",
(cpi->encoding_thread_count +1));
*/
for (ithread = 0; ithread < th_count; ithread++)
{
ENCODETHREAD_DATA * ethd = &cpi->en_thread_data[ithread];
sem_init(&cpi->h_event_start_encoding[ithread], 0, 0);
ethd->ithread = ithread;
ethd->ptr1 = (void *)cpi;
ethd->ptr2 = (void *)&cpi->mb_row_ei[ithread];
pthread_create(&cpi->h_encoding_thread[ithread], 0, thread_encoding_proc, ethd);
}
{
LPFTHREAD_DATA * lpfthd = &cpi->lpf_thread_data;
sem_init(&cpi->h_event_start_lpf, 0, 0);
sem_init(&cpi->h_event_end_lpf, 0, 0);
lpfthd->ptr1 = (void *)cpi;
pthread_create(&cpi->h_filter_thread, 0, loopfilter_thread, lpfthd);
}
}
}
int vp8_alloc_frame_buffers(VP8_COMMON *oci, int width, int height)
{
int i;
vp8_de_alloc_frame_buffers(oci);
/* our internal buffers are always multiples of 16 */
if ((width & 0xf) != 0)
width += 16 - (width & 0xf);
if ((height & 0xf) != 0)
height += 16 - (height & 0xf);
for (i = 0; i < NUM_YV12_BUFFERS; i++)
{
oci->fb_idx_ref_cnt[i] = 0;
oci->yv12_fb[i].flags = 0;
if (vp8_yv12_alloc_frame_buffer(&oci->yv12_fb[i], width, height, VP8BORDERINPIXELS) < 0)
goto allocation_fail;
}
oci->new_fb_idx = 0;
oci->lst_fb_idx = 1;
oci->gld_fb_idx = 2;
oci->alt_fb_idx = 3;
oci->fb_idx_ref_cnt[0] = 1;
oci->fb_idx_ref_cnt[1] = 1;
oci->fb_idx_ref_cnt[2] = 1;
oci->fb_idx_ref_cnt[3] = 1;
if (vp8_yv12_alloc_frame_buffer(&oci->temp_scale_frame, width, 16, VP8BORDERINPIXELS) < 0)
goto allocation_fail;
oci->mb_rows = height >> 4;
oci->mb_cols = width >> 4;
oci->MBs = oci->mb_rows * oci->mb_cols;
oci->mode_info_stride = oci->mb_cols + 1;
oci->mip = vpx_calloc((oci->mb_cols + 1) * (oci->mb_rows + 1), sizeof(MODE_INFO));
if (!oci->mip)
goto allocation_fail;
oci->mi = oci->mip + oci->mode_info_stride + 1;
/* Allocation of previous mode info will be done in vp8_decode_frame()
* as it is a decoder only data */
oci->above_context = vpx_calloc(sizeof(ENTROPY_CONTEXT_PLANES) * oci->mb_cols, 1);
if (!oci->above_context)
goto allocation_fail;
#if CONFIG_POSTPROC
if (vp8_yv12_alloc_frame_buffer(&oci->post_proc_buffer, width, height, VP8BORDERINPIXELS) < 0)
goto allocation_fail;
oci->post_proc_buffer_int_used = 0;
vpx_memset(&oci->postproc_state, 0, sizeof(oci->postproc_state));
vpx_memset(oci->post_proc_buffer.buffer_alloc, 128,
oci->post_proc_buffer.frame_size);
/* Allocate buffer to store post-processing filter coefficients.
*
* Note: Round up mb_cols to support SIMD reads
*/
oci->pp_limits_buffer = vpx_memalign(16, 24 * ((oci->mb_cols + 1) & ~1));
if (!oci->pp_limits_buffer)
goto allocation_fail;
#endif
return 0;
allocation_fail:
vp8_de_alloc_frame_buffers(oci);
return 1;
}
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);
}
}
}
int vp8_alloc_frame_buffers(VP8_COMMON *oci, int width, int height)
{
int i;
vp8_de_alloc_frame_buffers(oci);
if ((width & 0xf) != 0)
width += 16 - (width & 0xf);
if ((height & 0xf) != 0)
height += 16 - (height & 0xf);
for (i = 0; i < NUM_YV12_BUFFERS; i++)
{
oci->fb_idx_ref_cnt[i] = 0;
oci->yv12_fb[i].flags = 0;
if (vp8_yv12_alloc_frame_buffer(&oci->yv12_fb[i], width, height, VP8BORDERINPIXELS) < 0)
goto allocation_fail;
}
oci->new_fb_idx = 0;
oci->lst_fb_idx = 1;
oci->gld_fb_idx = 2;
oci->alt_fb_idx = 3;
oci->fb_idx_ref_cnt[0] = 1;
oci->fb_idx_ref_cnt[1] = 1;
oci->fb_idx_ref_cnt[2] = 1;
oci->fb_idx_ref_cnt[3] = 1;
if (vp8_yv12_alloc_frame_buffer(&oci->temp_scale_frame, width, 16, VP8BORDERINPIXELS) < 0)
goto allocation_fail;
oci->mb_rows = height >> 4;
oci->mb_cols = width >> 4;
oci->MBs = oci->mb_rows * oci->mb_cols;
oci->mode_info_stride = oci->mb_cols + 1;
oci->mip = vpx_calloc((oci->mb_cols + 1) * (oci->mb_rows + 1), sizeof(MODE_INFO));
if (!oci->mip)
goto allocation_fail;
oci->mi = oci->mip + oci->mode_info_stride + 1;
oci->above_context = vpx_calloc(sizeof(ENTROPY_CONTEXT_PLANES) * oci->mb_cols, 1);
if (!oci->above_context)
goto allocation_fail;
#if CONFIG_POSTPROC
if (vp8_yv12_alloc_frame_buffer(&oci->post_proc_buffer, width, height, VP8BORDERINPIXELS) < 0)
goto allocation_fail;
oci->post_proc_buffer_int_used = 0;
vpx_memset(&oci->postproc_state, 0, sizeof(oci->postproc_state));
vpx_memset(oci->post_proc_buffer.buffer_alloc, 128,
oci->post_proc_buffer.frame_size);
oci->pp_limits_buffer = vpx_memalign(16, 24 * ((oci->mb_cols + 1) & ~1));
if (!oci->pp_limits_buffer)
goto allocation_fail;
#endif
return 0;
allocation_fail:
vp8_de_alloc_frame_buffers(oci);
return 1;
}
int vpx_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
int width, int height,
int ss_x, int ss_y,
#if CONFIG_VP9_HIGHBITDEPTH
int use_highbitdepth,
#endif
int border,
int byte_alignment,
vpx_codec_frame_buffer_t *fb,
vpx_get_frame_buffer_cb_fn_t cb,
void *cb_priv) {
if (ybf) {
const int vp9_byte_align = (byte_alignment == 0) ? 1 : byte_alignment;
const int aligned_width = (width + 7) & ~7;
const int aligned_height = (height + 7) & ~7;
const int y_stride = ((aligned_width + 2 * border) + 31) & ~31;
const uint64_t yplane_size = (aligned_height + 2 * border) *
(uint64_t)y_stride + byte_alignment;
const int uv_width = aligned_width >> ss_x;
const int uv_height = aligned_height >> ss_y;
const int uv_stride = y_stride >> ss_x;
const int uv_border_w = border >> ss_x;
const int uv_border_h = border >> ss_y;
const uint64_t uvplane_size = (uv_height + 2 * uv_border_h) *
(uint64_t)uv_stride + byte_alignment;
#if CONFIG_ALPHA
const int alpha_width = aligned_width;
const int alpha_height = aligned_height;
const int alpha_stride = y_stride;
const int alpha_border_w = border;
const int alpha_border_h = border;
const uint64_t alpha_plane_size = (alpha_height + 2 * alpha_border_h) *
(uint64_t)alpha_stride + byte_alignment;
#if CONFIG_VP9_HIGHBITDEPTH
const uint64_t frame_size = (1 + use_highbitdepth) *
(yplane_size + 2 * uvplane_size + alpha_plane_size);
#else
const uint64_t frame_size = yplane_size + 2 * uvplane_size +
alpha_plane_size;
#endif // CONFIG_VP9_HIGHBITDEPTH
#else
#if CONFIG_VP9_HIGHBITDEPTH
const uint64_t frame_size =
(1 + use_highbitdepth) * (yplane_size + 2 * uvplane_size);
#else
const uint64_t frame_size = yplane_size + 2 * uvplane_size;
#endif // CONFIG_VP9_HIGHBITDEPTH
#endif // CONFIG_ALPHA
uint8_t *buf = NULL;
if (cb != NULL) {
const int align_addr_extra_size = 31;
const uint64_t external_frame_size = frame_size + align_addr_extra_size;
assert(fb != NULL);
if (external_frame_size != (size_t)external_frame_size)
return -1;
// Allocation to hold larger frame, or first allocation.
if (cb(cb_priv, (size_t)external_frame_size, fb) < 0)
return -1;
if (fb->data == NULL || fb->size < external_frame_size)
return -1;
ybf->buffer_alloc = (uint8_t *)yv12_align_addr(fb->data, 32);
} else if (frame_size > (size_t)ybf->buffer_alloc_sz) {
// Allocation to hold larger frame, or first allocation.
vpx_free(ybf->buffer_alloc);
ybf->buffer_alloc = NULL;
if (frame_size != (size_t)frame_size)
return -1;
ybf->buffer_alloc = (uint8_t *)vpx_memalign(32, (size_t)frame_size);
if (!ybf->buffer_alloc)
return -1;
ybf->buffer_alloc_sz = (int)frame_size;
// This memset is needed for fixing valgrind error from C loop filter
// due to access uninitialized memory in frame border. It could be
// removed if border is totally removed.
memset(ybf->buffer_alloc, 0, ybf->buffer_alloc_sz);
}
/* Only support allocating buffers that have a border that's a multiple
* of 32. The border restriction is required to get 16-byte alignment of
* the start of the chroma rows without introducing an arbitrary gap
* between planes, which would break the semantics of things like
* vpx_img_set_rect(). */
if (border & 0x1f)
return -3;
ybf->y_crop_width = width;
ybf->y_crop_height = height;
ybf->y_width = aligned_width;
ybf->y_height = aligned_height;
ybf->y_stride = y_stride;
ybf->uv_crop_width = (width + ss_x) >> ss_x;
ybf->uv_crop_height = (height + ss_y) >> ss_y;
ybf->uv_width = uv_width;
ybf->uv_height = uv_height;
ybf->uv_stride = uv_stride;
ybf->border = border;
ybf->frame_size = (int)frame_size;
ybf->subsampling_x = ss_x;
ybf->subsampling_y = ss_y;
buf = ybf->buffer_alloc;
#if CONFIG_VP9_HIGHBITDEPTH
if (use_highbitdepth) {
// Store uint16 addresses when using 16bit framebuffers
buf = CONVERT_TO_BYTEPTR(ybf->buffer_alloc);
ybf->flags = YV12_FLAG_HIGHBITDEPTH;
} else {
ybf->flags = 0;
}
#endif // CONFIG_VP9_HIGHBITDEPTH
ybf->y_buffer = (uint8_t *)yv12_align_addr(
buf + (border * y_stride) + border, vp9_byte_align);
ybf->u_buffer = (uint8_t *)yv12_align_addr(
buf + yplane_size + (uv_border_h * uv_stride) + uv_border_w,
vp9_byte_align);
ybf->v_buffer = (uint8_t *)yv12_align_addr(
buf + yplane_size + uvplane_size + (uv_border_h * uv_stride) +
uv_border_w, vp9_byte_align);
#if CONFIG_ALPHA
ybf->alpha_width = alpha_width;
ybf->alpha_height = alpha_height;
ybf->alpha_stride = alpha_stride;
ybf->alpha_buffer = (uint8_t *)yv12_align_addr(
buf + yplane_size + 2 * uvplane_size +
(alpha_border_h * alpha_stride) + alpha_border_w, vp9_byte_align);
#endif
ybf->corrupted = 0; /* assume not corrupted by errors */
return 0;
}
return -2;
}
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 vp9_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf,
int width, int height,
int ss_x, int ss_y, int border,
vpx_codec_frame_buffer_t *fb,
vpx_get_frame_buffer_cb_fn_t cb,
void *cb_priv) {
if (ybf) {
const int aligned_width = (width + 7) & ~7;
const int aligned_height = (height + 7) & ~7;
const int y_stride = ((aligned_width + 2 * border) + 31) & ~31;
const int yplane_size = (aligned_height + 2 * border) * y_stride;
const int uv_width = aligned_width >> ss_x;
const int uv_height = aligned_height >> ss_y;
const int uv_stride = y_stride >> ss_x;
const int uv_border_w = border >> ss_x;
const int uv_border_h = border >> ss_y;
const int uvplane_size = (uv_height + 2 * uv_border_h) * uv_stride;
#if CONFIG_ALPHA
const int alpha_width = aligned_width;
const int alpha_height = aligned_height;
const int alpha_stride = y_stride;
const int alpha_border_w = border;
const int alpha_border_h = border;
const int alpha_plane_size = (alpha_height + 2 * alpha_border_h) *
alpha_stride;
const int frame_size = yplane_size + 2 * uvplane_size +
alpha_plane_size;
#else
const int frame_size = yplane_size + 2 * uvplane_size;
#endif
if (cb != NULL) {
const int align_addr_extra_size = 31;
const size_t external_frame_size = frame_size + align_addr_extra_size;
assert(fb != NULL);
// Allocation to hold larger frame, or first allocation.
if (cb(cb_priv, external_frame_size, fb) < 0)
return -1;
if (fb->data == NULL || fb->size < external_frame_size)
return -1;
// This memset is needed for fixing valgrind error from C loop filter
// due to access uninitialized memory in frame border. It could be
// removed if border is totally removed.
vpx_memset(fb->data, 0, fb->size);
ybf->buffer_alloc = (uint8_t *)yv12_align_addr(fb->data, 32);
} else if (frame_size > ybf->buffer_alloc_sz) {
// Allocation to hold larger frame, or first allocation.
if (ybf->buffer_alloc)
vpx_free(ybf->buffer_alloc);
ybf->buffer_alloc = (uint8_t *)vpx_memalign(32, frame_size);
if (!ybf->buffer_alloc)
return -1;
ybf->buffer_alloc_sz = frame_size;
// This memset is needed for fixing valgrind error from C loop filter
// due to access uninitialized memory in frame border. It could be
// removed if border is totally removed.
vpx_memset(ybf->buffer_alloc, 0, ybf->buffer_alloc_sz);
}
/* Only support allocating buffers that have a border that's a multiple
* of 32. The border restriction is required to get 16-byte alignment of
* the start of the chroma rows without introducing an arbitrary gap
* between planes, which would break the semantics of things like
* vpx_img_set_rect(). */
if (border & 0x1f)
return -3;
ybf->y_crop_width = width;
ybf->y_crop_height = height;
ybf->y_width = aligned_width;
ybf->y_height = aligned_height;
ybf->y_stride = y_stride;
ybf->uv_crop_width = (width + ss_x) >> ss_x;
ybf->uv_crop_height = (height + ss_y) >> ss_y;
ybf->uv_width = uv_width;
ybf->uv_height = uv_height;
ybf->uv_stride = uv_stride;
ybf->border = border;
ybf->frame_size = frame_size;
ybf->y_buffer = ybf->buffer_alloc + (border * y_stride) + border;
ybf->u_buffer = ybf->buffer_alloc + yplane_size +
(uv_border_h * uv_stride) + uv_border_w;
ybf->v_buffer = ybf->buffer_alloc + yplane_size + uvplane_size +
(uv_border_h * uv_stride) + uv_border_w;
#if CONFIG_ALPHA
ybf->alpha_width = alpha_width;
ybf->alpha_height = alpha_height;
ybf->alpha_stride = alpha_stride;
ybf->alpha_buffer = ybf->buffer_alloc + yplane_size + 2 * uvplane_size +
(alpha_border_h * alpha_stride) + alpha_border_w;
#endif
ybf->corrupted = 0; /* assume not corrupted by errors */
return 0;
}
return -2;
}
int vp8cx_create_encoder_threads(VP8_COMP *cpi)
{
const VP8_COMMON * cm = &cpi->common;
cpi->b_multi_threaded = 0;
cpi->encoding_thread_count = 0;
cpi->b_lpf_running = 0;
if (cm->processor_core_count > 1 && cpi->oxcf.multi_threaded > 1)
{
int ithread;
int th_count = cpi->oxcf.multi_threaded - 1;
int rc = 0;
/* don't allocate more threads than cores available */
if (cpi->oxcf.multi_threaded > cm->processor_core_count)
th_count = cm->processor_core_count - 1;
/* we have th_count + 1 (main) threads processing one row each */
/* no point to have more threads than the sync range allows */
if(th_count > ((cm->mb_cols / cpi->mt_sync_range) - 1))
{
th_count = (cm->mb_cols / cpi->mt_sync_range) - 1;
}
if(th_count == 0)
return 0;
CHECK_MEM_ERROR(cpi->h_encoding_thread,
vpx_malloc(sizeof(pthread_t) * th_count));
CHECK_MEM_ERROR(cpi->h_event_start_encoding,
vpx_malloc(sizeof(sem_t) * th_count));
CHECK_MEM_ERROR(cpi->mb_row_ei,
vpx_memalign(32, sizeof(MB_ROW_COMP) * th_count));
vpx_memset(cpi->mb_row_ei, 0, sizeof(MB_ROW_COMP) * th_count);
CHECK_MEM_ERROR(cpi->en_thread_data,
vpx_malloc(sizeof(ENCODETHREAD_DATA) * th_count));
sem_init(&cpi->h_event_end_encoding, 0, 0);
cpi->b_multi_threaded = 1;
cpi->encoding_thread_count = th_count;
/*
printf("[VP8:] multi_threaded encoding is enabled with %d threads\n\n",
(cpi->encoding_thread_count +1));
*/
for (ithread = 0; ithread < th_count; ithread++)
{
ENCODETHREAD_DATA *ethd = &cpi->en_thread_data[ithread];
/* Setup block ptrs and offsets */
vp8_setup_block_ptrs(&cpi->mb_row_ei[ithread].mb);
vp8_setup_block_dptrs(&cpi->mb_row_ei[ithread].mb.e_mbd);
sem_init(&cpi->h_event_start_encoding[ithread], 0, 0);
ethd->ithread = ithread;
ethd->ptr1 = (void *)cpi;
ethd->ptr2 = (void *)&cpi->mb_row_ei[ithread];
rc = pthread_create(&cpi->h_encoding_thread[ithread], 0,
thread_encoding_proc, ethd);
if(rc)
break;
}
if(rc)
{
/* shutdown other threads */
cpi->b_multi_threaded = 0;
for(--ithread; ithread >= 0; ithread--)
{
pthread_join(cpi->h_encoding_thread[ithread], 0);
sem_destroy(&cpi->h_event_start_encoding[ithread]);
}
sem_destroy(&cpi->h_event_end_encoding);
/* free thread related resources */
vpx_free(cpi->h_event_start_encoding);
vpx_free(cpi->h_encoding_thread);
vpx_free(cpi->mb_row_ei);
vpx_free(cpi->en_thread_data);
return -1;
}
{
LPFTHREAD_DATA * lpfthd = &cpi->lpf_thread_data;
sem_init(&cpi->h_event_start_lpf, 0, 0);
sem_init(&cpi->h_event_end_lpf, 0, 0);
lpfthd->ptr1 = (void *)cpi;
rc = pthread_create(&cpi->h_filter_thread, 0, thread_loopfilter,
lpfthd);
if(rc)
{
/* shutdown other threads */
cpi->b_multi_threaded = 0;
for(--ithread; ithread >= 0; ithread--)
{
sem_post(&cpi->h_event_start_encoding[ithread]);
pthread_join(cpi->h_encoding_thread[ithread], 0);
sem_destroy(&cpi->h_event_start_encoding[ithread]);
}
sem_destroy(&cpi->h_event_end_encoding);
sem_destroy(&cpi->h_event_end_lpf);
sem_destroy(&cpi->h_event_start_lpf);
/* free thread related resources */
vpx_free(cpi->h_event_start_encoding);
vpx_free(cpi->h_encoding_thread);
vpx_free(cpi->mb_row_ei);
vpx_free(cpi->en_thread_data);
return -2;
}
}
}
return 0;
}
static vpx_codec_err_t init_decoder(vpx_codec_alg_priv_t *ctx) {
int i;
const VPxWorkerInterface *const winterface = vpx_get_worker_interface();
ctx->last_show_frame = -1;
ctx->next_submit_worker_id = 0;
ctx->last_submit_worker_id = 0;
ctx->next_output_worker_id = 0;
ctx->frame_cache_read = 0;
ctx->frame_cache_write = 0;
ctx->num_cache_frames = 0;
ctx->need_resync = 1;
ctx->num_frame_workers =
(ctx->frame_parallel_decode == 1) ? ctx->cfg.threads: 1;
if (ctx->num_frame_workers > MAX_DECODE_THREADS)
ctx->num_frame_workers = MAX_DECODE_THREADS;
ctx->available_threads = ctx->num_frame_workers;
ctx->flushed = 0;
ctx->buffer_pool = (BufferPool *)vpx_calloc(1, sizeof(BufferPool));
if (ctx->buffer_pool == NULL)
return VPX_CODEC_MEM_ERROR;
#if CONFIG_MULTITHREAD
if (pthread_mutex_init(&ctx->buffer_pool->pool_mutex, NULL)) {
set_error_detail(ctx, "Failed to allocate buffer pool mutex");
return VPX_CODEC_MEM_ERROR;
}
#endif
ctx->frame_workers = (VPxWorker *)
vpx_malloc(ctx->num_frame_workers * sizeof(*ctx->frame_workers));
if (ctx->frame_workers == NULL) {
set_error_detail(ctx, "Failed to allocate frame_workers");
return VPX_CODEC_MEM_ERROR;
}
for (i = 0; i < ctx->num_frame_workers; ++i) {
VPxWorker *const worker = &ctx->frame_workers[i];
FrameWorkerData *frame_worker_data = NULL;
winterface->init(worker);
worker->data1 = vpx_memalign(32, sizeof(FrameWorkerData));
if (worker->data1 == NULL) {
set_error_detail(ctx, "Failed to allocate frame_worker_data");
return VPX_CODEC_MEM_ERROR;
}
frame_worker_data = (FrameWorkerData *)worker->data1;
frame_worker_data->pbi = vp9_decoder_create(ctx->buffer_pool);
if (frame_worker_data->pbi == NULL) {
set_error_detail(ctx, "Failed to allocate frame_worker_data");
return VPX_CODEC_MEM_ERROR;
}
frame_worker_data->pbi->frame_worker_owner = worker;
frame_worker_data->worker_id = i;
frame_worker_data->scratch_buffer = NULL;
frame_worker_data->scratch_buffer_size = 0;
frame_worker_data->frame_context_ready = 0;
frame_worker_data->received_frame = 0;
#if CONFIG_MULTITHREAD
if (pthread_mutex_init(&frame_worker_data->stats_mutex, NULL)) {
set_error_detail(ctx, "Failed to allocate frame_worker_data mutex");
return VPX_CODEC_MEM_ERROR;
}
if (pthread_cond_init(&frame_worker_data->stats_cond, NULL)) {
set_error_detail(ctx, "Failed to allocate frame_worker_data cond");
return VPX_CODEC_MEM_ERROR;
}
#endif
// If decoding in serial mode, FrameWorker thread could create tile worker
// thread or loopfilter thread.
frame_worker_data->pbi->max_threads =
(ctx->frame_parallel_decode == 0) ? ctx->cfg.threads : 0;
frame_worker_data->pbi->inv_tile_order = ctx->invert_tile_order;
frame_worker_data->pbi->frame_parallel_decode = ctx->frame_parallel_decode;
frame_worker_data->pbi->common.frame_parallel_decode =
ctx->frame_parallel_decode;
worker->hook = (VPxWorkerHook)frame_worker_hook;
if (!winterface->reset(worker)) {
set_error_detail(ctx, "Frame Worker thread creation failed");
return VPX_CODEC_MEM_ERROR;
}
}
// 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))
set_default_ppflags(&ctx->postproc_cfg);
init_buffer_callbacks(ctx);
return VPX_CODEC_OK;
}
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;
}
