diff options
Diffstat (limited to 'vp9/decoder/vp9_decodeframe.c')
-rw-r--r-- | vp9/decoder/vp9_decodeframe.c | 204 |
1 files changed, 103 insertions, 101 deletions
diff --git a/vp9/decoder/vp9_decodeframe.c b/vp9/decoder/vp9_decodeframe.c index 9dc0cf1a3..de58939fc 100644 --- a/vp9/decoder/vp9_decodeframe.c +++ b/vp9/decoder/vp9_decodeframe.c @@ -316,7 +316,7 @@ static MB_MODE_INFO *set_offsets(VP9_COMMON *const cm, MACROBLOCKD *const xd, // as they are always compared to values that are in 1/8th pel units set_mi_row_col(xd, tile, mi_row, bh, mi_col, bw, cm->mi_rows, cm->mi_cols); - vp9_setup_dst_planes(xd, get_frame_new_buffer(cm), mi_row, mi_col); + vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col); return &xd->mi[0]->mbmi; } @@ -675,64 +675,6 @@ static void setup_frame_size_with_refs(VP9_COMMON *cm, setup_display_size(cm, rb); } -static void decode_tile(VP9Decoder *pbi, const TileInfo *const tile, - int do_loopfilter_inline, vp9_reader *r) { - const int num_threads = pbi->max_threads; - VP9_COMMON *const cm = &pbi->common; - int mi_row, mi_col; - MACROBLOCKD *xd = &pbi->mb; - - if (do_loopfilter_inline) { - LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; - lf_data->frame_buffer = get_frame_new_buffer(cm); - lf_data->cm = cm; - lf_data->xd = pbi->mb; - lf_data->stop = 0; - lf_data->y_only = 0; - vp9_loop_filter_frame_init(cm, cm->lf.filter_level); - } - - for (mi_row = tile->mi_row_start; mi_row < tile->mi_row_end; - mi_row += MI_BLOCK_SIZE) { - // For a SB there are 2 left contexts, each pertaining to a MB row within - vp9_zero(xd->left_context); - vp9_zero(xd->left_seg_context); - for (mi_col = tile->mi_col_start; mi_col < tile->mi_col_end; - mi_col += MI_BLOCK_SIZE) { - decode_partition(cm, xd, tile, mi_row, mi_col, r, BLOCK_64X64); - } - - if (do_loopfilter_inline) { - const int lf_start = mi_row - MI_BLOCK_SIZE; - LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; - - // delay the loopfilter by 1 macroblock row. - if (lf_start < 0) continue; - - // decoding has completed: finish up the loop filter in this thread. - if (mi_row + MI_BLOCK_SIZE >= tile->mi_row_end) continue; - - vp9_worker_sync(&pbi->lf_worker); - lf_data->start = lf_start; - lf_data->stop = mi_row; - if (num_threads > 1) { - vp9_worker_launch(&pbi->lf_worker); - } else { - vp9_worker_execute(&pbi->lf_worker); - } - } - } - - if (do_loopfilter_inline) { - LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; - - vp9_worker_sync(&pbi->lf_worker); - lf_data->start = lf_data->stop; - lf_data->stop = cm->mi_rows; - vp9_worker_execute(&pbi->lf_worker); - } -} - static void setup_tile_info(VP9_COMMON *cm, struct vp9_read_bit_buffer *rb) { int min_log2_tile_cols, max_log2_tile_cols, max_ones; vp9_get_tile_n_bits(cm->mi_cols, &min_log2_tile_cols, &max_log2_tile_cols); @@ -811,16 +753,35 @@ static void get_tile_buffers(VP9Decoder *pbi, static const uint8_t *decode_tiles(VP9Decoder *pbi, const uint8_t *data, - const uint8_t *data_end, - int do_loopfilter_inline) { + const uint8_t *data_end) { VP9_COMMON *const cm = &pbi->common; const int aligned_cols = mi_cols_aligned_to_sb(cm->mi_cols); const int tile_cols = 1 << cm->log2_tile_cols; const int tile_rows = 1 << cm->log2_tile_rows; TileBuffer tile_buffers[4][1 << 6]; int tile_row, tile_col; - const uint8_t *end = NULL; - vp9_reader r; + int mi_row, mi_col; + TileData *tile_data = NULL; + + if (cm->lf.filter_level && pbi->lf_worker.data1 == NULL) { + CHECK_MEM_ERROR(cm, pbi->lf_worker.data1, + vpx_memalign(32, sizeof(LFWorkerData))); + pbi->lf_worker.hook = (VP9WorkerHook)vp9_loop_filter_worker; + if (pbi->max_threads > 1 && !vp9_worker_reset(&pbi->lf_worker)) { + vpx_internal_error(&cm->error, VPX_CODEC_ERROR, + "Loop filter thread creation failed"); + } + } + + if (cm->lf.filter_level) { + LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; + lf_data->frame_buffer = get_frame_new_buffer(cm); + lf_data->cm = cm; + vp9_copy(lf_data->planes, pbi->mb.plane); + lf_data->stop = 0; + lf_data->y_only = 0; + vp9_loop_filter_frame_init(cm, cm->lf.filter_level); + } assert(tile_rows <= 4); assert(tile_cols <= (1 << 6)); @@ -835,26 +796,88 @@ static const uint8_t *decode_tiles(VP9Decoder *pbi, get_tile_buffers(pbi, data, data_end, tile_cols, tile_rows, tile_buffers); - // Decode tiles using data from tile_buffers + if (pbi->tile_data == NULL || + (tile_cols * tile_rows) != pbi->total_tiles) { + vpx_free(pbi->tile_data); + CHECK_MEM_ERROR( + cm, + pbi->tile_data, + vpx_malloc(tile_cols * tile_rows * (sizeof(*pbi->tile_data)))); + pbi->total_tiles = tile_rows * tile_cols; + } + + // Load all tile information into tile_data. for (tile_row = 0; tile_row < tile_rows; ++tile_row) { for (tile_col = 0; tile_col < tile_cols; ++tile_col) { - const int col = pbi->inv_tile_order ? tile_cols - tile_col - 1 : tile_col; - const int last_tile = tile_row == tile_rows - 1 && - col == tile_cols - 1; - const TileBuffer *const buf = &tile_buffers[tile_row][col]; TileInfo tile; + const TileBuffer *const buf = &tile_buffers[tile_row][tile_col]; + tile_data = pbi->tile_data + tile_cols * tile_row + tile_col; + tile_data->cm = cm; + tile_data->xd = pbi->mb; + tile_data->xd.corrupted = 0; + vp9_tile_init(&tile, tile_data->cm, tile_row, tile_col); + setup_token_decoder(buf->data, data_end, buf->size, &cm->error, + &tile_data->bit_reader, pbi->decrypt_cb, + pbi->decrypt_state); + init_macroblockd(cm, &tile_data->xd); + vp9_zero(tile_data->xd.dqcoeff); + } + } - vp9_tile_init(&tile, cm, tile_row, col); - setup_token_decoder(buf->data, data_end, buf->size, &cm->error, &r, - pbi->decrypt_cb, pbi->decrypt_state); - decode_tile(pbi, &tile, do_loopfilter_inline, &r); - - if (last_tile) - end = vp9_reader_find_end(&r); + for (tile_row = 0; tile_row < tile_rows; ++tile_row) { + TileInfo tile; + vp9_tile_set_row(&tile, cm, tile_row); + for (mi_row = tile.mi_row_start; mi_row < tile.mi_row_end; + mi_row += MI_BLOCK_SIZE) { + for (tile_col = 0; tile_col < tile_cols; ++tile_col) { + const int col = pbi->inv_tile_order ? + tile_cols - tile_col - 1 : tile_col; + tile_data = pbi->tile_data + tile_cols * tile_row + col; + vp9_tile_set_col(&tile, tile_data->cm, col); + vp9_zero(tile_data->xd.left_context); + vp9_zero(tile_data->xd.left_seg_context); + for (mi_col = tile.mi_col_start; mi_col < tile.mi_col_end; + mi_col += MI_BLOCK_SIZE) { + decode_partition(tile_data->cm, &tile_data->xd, &tile, mi_row, mi_col, + &tile_data->bit_reader, BLOCK_64X64); + } + } + // Loopfilter one row. + if (cm->lf.filter_level) { + const int lf_start = mi_row - MI_BLOCK_SIZE; + LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; + + // delay the loopfilter by 1 macroblock row. + if (lf_start < 0) continue; + + // decoding has completed: finish up the loop filter in this thread. + if (mi_row + MI_BLOCK_SIZE >= cm->mi_rows) continue; + + vp9_worker_sync(&pbi->lf_worker); + lf_data->start = lf_start; + lf_data->stop = mi_row; + if (pbi->max_threads > 1) { + vp9_worker_launch(&pbi->lf_worker); + } else { + vp9_worker_execute(&pbi->lf_worker); + } + } } } - return end; + // Loopfilter remaining rows in the frame. + if (cm->lf.filter_level) { + LFWorkerData *const lf_data = (LFWorkerData*)pbi->lf_worker.data1; + vp9_worker_sync(&pbi->lf_worker); + lf_data->start = lf_data->stop; + lf_data->stop = cm->mi_rows; + vp9_worker_execute(&pbi->lf_worker); + } + + // Get last tile data. + tile_data = pbi->tile_data + tile_cols * tile_rows - 1; + + return vp9_reader_find_end(&tile_data->bit_reader); } static int tile_worker_hook(void *arg1, void *arg2) { @@ -1307,8 +1330,6 @@ int vp9_decode_frame(VP9Decoder *pbi, const int tile_rows = 1 << cm->log2_tile_rows; const int tile_cols = 1 << cm->log2_tile_cols; YV12_BUFFER_CONFIG *const new_fb = get_frame_new_buffer(cm); - const int do_loopfilter_inline = tile_rows == 1 && tile_cols == 1 && - cm->lf.filter_level; xd->cur_buf = new_fb; if (!first_partition_size) { @@ -1325,16 +1346,6 @@ int vp9_decode_frame(VP9Decoder *pbi, vpx_internal_error(&cm->error, VPX_CODEC_CORRUPT_FRAME, "Truncated packet or corrupt header length"); - if (do_loopfilter_inline && pbi->lf_worker.data1 == NULL) { - CHECK_MEM_ERROR(cm, pbi->lf_worker.data1, - vpx_memalign(32, sizeof(LFWorkerData))); - pbi->lf_worker.hook = (VP9WorkerHook)vp9_loop_filter_worker; - if (pbi->max_threads > 1 && !vp9_worker_reset(&pbi->lf_worker)) { - vpx_internal_error(&cm->error, VPX_CODEC_ERROR, - "Loop filter thread creation failed"); - } - } - init_macroblockd(cm, &pbi->mb); if (cm->coding_use_prev_mi) @@ -1357,9 +1368,11 @@ int vp9_decode_frame(VP9Decoder *pbi, if (pbi->max_threads > 1 && tile_rows == 1 && tile_cols > 1 && cm->frame_parallel_decoding_mode) { *p_data_end = decode_tiles_mt(pbi, data + first_partition_size, data_end); + // If multiple threads are used to decode tiles, then we use those threads + // to do parallel loopfiltering. + vp9_loop_filter_frame_mt(new_fb, pbi, cm, cm->lf.filter_level, 0); } else { - *p_data_end = decode_tiles(pbi, data + first_partition_size, data_end, - do_loopfilter_inline); + *p_data_end = decode_tiles(pbi, data + first_partition_size, data_end); } new_fb->corrupted |= xd->corrupted; @@ -1388,16 +1401,5 @@ int vp9_decode_frame(VP9Decoder *pbi, if (cm->refresh_frame_context) cm->frame_contexts[cm->frame_context_idx] = cm->fc; - // Loopfilter - if (!do_loopfilter_inline) { - // If multiple threads are used to decode tiles, then we use those threads - // to do parallel loopfiltering. - if (pbi->num_tile_workers) { - vp9_loop_filter_frame_mt(new_fb, pbi, cm, cm->lf.filter_level, 0, 0); - } else { - vp9_loop_filter_frame(new_fb, cm, &pbi->mb, cm->lf.filter_level, 0, 0); - } - } - return 0; } |