/* * Copyright (c) 2016, Alliance for Open Media. All rights reserved. * * This source code is subject to the terms of the BSD 2 Clause License and * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License * was not distributed with this source code in the LICENSE file, you can * obtain it at www.aomedia.org/license/software. If the Alliance for Open * Media Patent License 1.0 was not distributed with this source code in the * PATENTS file, you can obtain it at www.aomedia.org/license/patent. */ #include "aom/aom_image.h" #include "config/aom_config.h" #include "config/aom_scale_rtcd.h" #include "aom_dsp/aom_dsp_common.h" #include "aom_dsp/txfm_common.h" #include "aom_mem/aom_mem.h" #include "aom_util/aom_pthread.h" #include "aom_util/aom_thread.h" #include "av1/common/av1_loopfilter.h" #include "av1/common/blockd.h" #include "av1/common/cdef.h" #include "av1/common/entropymode.h" #include "av1/common/enums.h" #include "av1/common/thread_common.h" #include "av1/common/reconinter.h" #include "av1/common/reconintra.h" #include "av1/common/restoration.h" // Set up nsync by width. static inline int get_sync_range(int width) { // nsync numbers are picked by testing. For example, for 4k // video, using 4 gives best performance. if (width < 640) return 1; else if (width <= 1280) return 2; else if (width <= 4096) return 4; else return 8; } #if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER static inline int get_lr_sync_range(int width) { #if 0 // nsync numbers are picked by testing. For example, for 4k // video, using 4 gives best performance. if (width < 640) return 1; else if (width <= 1280) return 2; else if (width <= 4096) return 4; else return 8; #else (void)width; return 1; #endif } #endif // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER // Allocate memory for lf row synchronization void av1_loop_filter_alloc(AV1LfSync *lf_sync, AV1_COMMON *cm, int rows, int width, int num_workers) { lf_sync->rows = rows; #if CONFIG_MULTITHREAD { int i, j; for (j = 0; j < MAX_MB_PLANE; j++) { CHECK_MEM_ERROR(cm, lf_sync->mutex_[j], aom_malloc(sizeof(*(lf_sync->mutex_[j])) * rows)); if (lf_sync->mutex_[j]) { for (i = 0; i < rows; ++i) { pthread_mutex_init(&lf_sync->mutex_[j][i], NULL); } } CHECK_MEM_ERROR(cm, lf_sync->cond_[j], aom_malloc(sizeof(*(lf_sync->cond_[j])) * rows)); if (lf_sync->cond_[j]) { for (i = 0; i < rows; ++i) { pthread_cond_init(&lf_sync->cond_[j][i], NULL); } } } CHECK_MEM_ERROR(cm, lf_sync->job_mutex, aom_malloc(sizeof(*(lf_sync->job_mutex)))); if (lf_sync->job_mutex) { pthread_mutex_init(lf_sync->job_mutex, NULL); } } #endif // CONFIG_MULTITHREAD CHECK_MEM_ERROR(cm, lf_sync->lfdata, aom_malloc(num_workers * sizeof(*(lf_sync->lfdata)))); lf_sync->num_workers = num_workers; for (int j = 0; j < MAX_MB_PLANE; j++) { CHECK_MEM_ERROR(cm, lf_sync->cur_sb_col[j], aom_malloc(sizeof(*(lf_sync->cur_sb_col[j])) * rows)); } CHECK_MEM_ERROR( cm, lf_sync->job_queue, aom_malloc(sizeof(*(lf_sync->job_queue)) * rows * MAX_MB_PLANE * 2)); // Set up nsync. lf_sync->sync_range = get_sync_range(width); } // Deallocate lf synchronization related mutex and data void av1_loop_filter_dealloc(AV1LfSync *lf_sync) { if (lf_sync != NULL) { int j; #if CONFIG_MULTITHREAD int i; for (j = 0; j < MAX_MB_PLANE; j++) { if (lf_sync->mutex_[j] != NULL) { for (i = 0; i < lf_sync->rows; ++i) { pthread_mutex_destroy(&lf_sync->mutex_[j][i]); } aom_free(lf_sync->mutex_[j]); } if (lf_sync->cond_[j] != NULL) { for (i = 0; i < lf_sync->rows; ++i) { pthread_cond_destroy(&lf_sync->cond_[j][i]); } aom_free(lf_sync->cond_[j]); } } if (lf_sync->job_mutex != NULL) { pthread_mutex_destroy(lf_sync->job_mutex); aom_free(lf_sync->job_mutex); } #endif // CONFIG_MULTITHREAD aom_free(lf_sync->lfdata); for (j = 0; j < MAX_MB_PLANE; j++) { aom_free(lf_sync->cur_sb_col[j]); } aom_free(lf_sync->job_queue); // 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. av1_zero(*lf_sync); } } void av1_alloc_cdef_sync(AV1_COMMON *const cm, AV1CdefSync *cdef_sync, int num_workers) { if (num_workers < 1) return; #if CONFIG_MULTITHREAD if (cdef_sync->mutex_ == NULL) { CHECK_MEM_ERROR(cm, cdef_sync->mutex_, aom_malloc(sizeof(*(cdef_sync->mutex_)))); if (cdef_sync->mutex_) pthread_mutex_init(cdef_sync->mutex_, NULL); } #else (void)cm; (void)cdef_sync; #endif // CONFIG_MULTITHREAD } void av1_free_cdef_sync(AV1CdefSync *cdef_sync) { if (cdef_sync == NULL) return; #if CONFIG_MULTITHREAD if (cdef_sync->mutex_ != NULL) { pthread_mutex_destroy(cdef_sync->mutex_); aom_free(cdef_sync->mutex_); } #endif // CONFIG_MULTITHREAD } static inline void cdef_row_mt_sync_read(AV1CdefSync *const cdef_sync, int row) { if (!row) return; #if CONFIG_MULTITHREAD AV1CdefRowSync *const cdef_row_mt = cdef_sync->cdef_row_mt; pthread_mutex_lock(cdef_row_mt[row - 1].row_mutex_); while (cdef_row_mt[row - 1].is_row_done != 1) pthread_cond_wait(cdef_row_mt[row - 1].row_cond_, cdef_row_mt[row - 1].row_mutex_); cdef_row_mt[row - 1].is_row_done = 0; pthread_mutex_unlock(cdef_row_mt[row - 1].row_mutex_); #else (void)cdef_sync; #endif // CONFIG_MULTITHREAD } static inline void cdef_row_mt_sync_write(AV1CdefSync *const cdef_sync, int row) { #if CONFIG_MULTITHREAD AV1CdefRowSync *const cdef_row_mt = cdef_sync->cdef_row_mt; pthread_mutex_lock(cdef_row_mt[row].row_mutex_); pthread_cond_signal(cdef_row_mt[row].row_cond_); cdef_row_mt[row].is_row_done = 1; pthread_mutex_unlock(cdef_row_mt[row].row_mutex_); #else (void)cdef_sync; (void)row; #endif // CONFIG_MULTITHREAD } static inline void sync_read(AV1LfSync *const lf_sync, int r, int c, int plane) { #if CONFIG_MULTITHREAD const int nsync = lf_sync->sync_range; if (r && !(c & (nsync - 1))) { pthread_mutex_t *const mutex = &lf_sync->mutex_[plane][r - 1]; pthread_mutex_lock(mutex); while (c > lf_sync->cur_sb_col[plane][r - 1] - nsync) { pthread_cond_wait(&lf_sync->cond_[plane][r - 1], mutex); } pthread_mutex_unlock(mutex); } #else (void)lf_sync; (void)r; (void)c; (void)plane; #endif // CONFIG_MULTITHREAD } static inline void sync_write(AV1LfSync *const lf_sync, int r, int c, const int sb_cols, int plane) { #if CONFIG_MULTITHREAD const int nsync = lf_sync->sync_range; int cur; // Only signal when there are enough filtered SB for next row to run. int sig = 1; if (c < sb_cols - 1) { cur = c; if (c % nsync) sig = 0; } else { cur = sb_cols + nsync; } if (sig) { pthread_mutex_lock(&lf_sync->mutex_[plane][r]); // When a thread encounters an error, cur_sb_col[plane][r] is set to maximum // column number. In this case, the AOMMAX operation here ensures that // cur_sb_col[plane][r] is not overwritten with a smaller value thus // preventing the infinite waiting of threads in the relevant sync_read() // function. lf_sync->cur_sb_col[plane][r] = AOMMAX(lf_sync->cur_sb_col[plane][r], cur); pthread_cond_broadcast(&lf_sync->cond_[plane][r]); pthread_mutex_unlock(&lf_sync->mutex_[plane][r]); } #else (void)lf_sync; (void)r; (void)c; (void)sb_cols; (void)plane; #endif // CONFIG_MULTITHREAD } // One job of row loopfiltering. void av1_thread_loop_filter_rows( const YV12_BUFFER_CONFIG *const frame_buffer, AV1_COMMON *const cm, struct macroblockd_plane *planes, MACROBLOCKD *xd, int mi_row, int plane, int dir, int lpf_opt_level, AV1LfSync *const lf_sync, struct aom_internal_error_info *error_info, AV1_DEBLOCKING_PARAMETERS *params_buf, TX_SIZE *tx_buf, int num_mis_in_lpf_unit_height_log2) { // TODO(aomedia:3276): Pass error_info to the low-level functions as required // in future to handle error propagation. (void)error_info; const int sb_cols = CEIL_POWER_OF_TWO(cm->mi_params.mi_cols, MAX_MIB_SIZE_LOG2); const int r = mi_row >> num_mis_in_lpf_unit_height_log2; int mi_col, c; const bool joint_filter_chroma = (lpf_opt_level == 2) && plane > AOM_PLANE_Y; const int num_planes = joint_filter_chroma ? 2 : 1; assert(IMPLIES(joint_filter_chroma, plane == AOM_PLANE_U)); if (dir == 0) { for (mi_col = 0; mi_col < cm->mi_params.mi_cols; mi_col += MAX_MIB_SIZE) { c = mi_col >> MAX_MIB_SIZE_LOG2; av1_setup_dst_planes(planes, cm->seq_params->sb_size, frame_buffer, mi_row, mi_col, plane, plane + num_planes); if (lpf_opt_level) { if (plane == AOM_PLANE_Y) { av1_filter_block_plane_vert_opt(cm, xd, &planes[plane], mi_row, mi_col, params_buf, tx_buf, num_mis_in_lpf_unit_height_log2); } else { av1_filter_block_plane_vert_opt_chroma( cm, xd, &planes[plane], mi_row, mi_col, params_buf, tx_buf, plane, joint_filter_chroma, num_mis_in_lpf_unit_height_log2); } } else { av1_filter_block_plane_vert(cm, xd, plane, &planes[plane], mi_row, mi_col); } if (lf_sync != NULL) { sync_write(lf_sync, r, c, sb_cols, plane); } } } else if (dir == 1) { for (mi_col = 0; mi_col < cm->mi_params.mi_cols; mi_col += MAX_MIB_SIZE) { c = mi_col >> MAX_MIB_SIZE_LOG2; if (lf_sync != NULL) { // Wait for vertical edge filtering of the top-right block to be // completed sync_read(lf_sync, r, c, plane); // Wait for vertical edge filtering of the right block to be completed sync_read(lf_sync, r + 1, c, plane); } #if CONFIG_MULTITHREAD if (lf_sync && lf_sync->num_workers > 1) { pthread_mutex_lock(lf_sync->job_mutex); const bool lf_mt_exit = lf_sync->lf_mt_exit; pthread_mutex_unlock(lf_sync->job_mutex); // Exit in case any worker has encountered an error. if (lf_mt_exit) return; } #endif av1_setup_dst_planes(planes, cm->seq_params->sb_size, frame_buffer, mi_row, mi_col, plane, plane + num_planes); if (lpf_opt_level) { if (plane == AOM_PLANE_Y) { av1_filter_block_plane_horz_opt(cm, xd, &planes[plane], mi_row, mi_col, params_buf, tx_buf, num_mis_in_lpf_unit_height_log2); } else { av1_filter_block_plane_horz_opt_chroma( cm, xd, &planes[plane], mi_row, mi_col, params_buf, tx_buf, plane, joint_filter_chroma, num_mis_in_lpf_unit_height_log2); } } else { av1_filter_block_plane_horz(cm, xd, plane, &planes[plane], mi_row, mi_col); } } } } void av1_set_vert_loop_filter_done(AV1_COMMON *cm, AV1LfSync *lf_sync, int num_mis_in_lpf_unit_height_log2) { int plane, sb_row; const int sb_cols = CEIL_POWER_OF_TWO(cm->mi_params.mi_cols, num_mis_in_lpf_unit_height_log2); const int sb_rows = CEIL_POWER_OF_TWO(cm->mi_params.mi_rows, num_mis_in_lpf_unit_height_log2); // In case of loopfilter row-multithreading, the worker on an SB row waits for // the vertical edge filtering of the right and top-right SBs. Hence, in case // a thread (main/worker) encounters an error, update that vertical // loopfiltering of every SB row in the frame is complete in order to avoid // dependent workers waiting indefinitely. for (sb_row = 0; sb_row < sb_rows; ++sb_row) for (plane = 0; plane < MAX_MB_PLANE; ++plane) sync_write(lf_sync, sb_row, sb_cols - 1, sb_cols, plane); } static inline void sync_lf_workers(AVxWorker *const workers, AV1_COMMON *const cm, int num_workers) { const AVxWorkerInterface *const winterface = aom_get_worker_interface(); int had_error = workers[0].had_error; struct aom_internal_error_info error_info; // Read the error_info of main thread. if (had_error) { AVxWorker *const worker = &workers[0]; error_info = ((LFWorkerData *)worker->data2)->error_info; } // Wait till all rows are finished. for (int i = num_workers - 1; i > 0; --i) { AVxWorker *const worker = &workers[i]; if (!winterface->sync(worker)) { had_error = 1; error_info = ((LFWorkerData *)worker->data2)->error_info; } } if (had_error) aom_internal_error_copy(cm->error, &error_info); } // Row-based multi-threaded loopfilter hook static int loop_filter_row_worker(void *arg1, void *arg2) { AV1LfSync *const lf_sync = (AV1LfSync *)arg1; LFWorkerData *const lf_data = (LFWorkerData *)arg2; AV1LfMTInfo *cur_job_info; #if CONFIG_MULTITHREAD pthread_mutex_t *job_mutex_ = lf_sync->job_mutex; #endif struct aom_internal_error_info *const error_info = &lf_data->error_info; // The jmp_buf is valid only for the duration of the function that calls // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 // before it returns. if (setjmp(error_info->jmp)) { error_info->setjmp = 0; #if CONFIG_MULTITHREAD pthread_mutex_lock(job_mutex_); lf_sync->lf_mt_exit = true; pthread_mutex_unlock(job_mutex_); #endif av1_set_vert_loop_filter_done(lf_data->cm, lf_sync, MAX_MIB_SIZE_LOG2); return 0; } error_info->setjmp = 1; while ((cur_job_info = get_lf_job_info(lf_sync)) != NULL) { const int lpf_opt_level = cur_job_info->lpf_opt_level; av1_thread_loop_filter_rows( lf_data->frame_buffer, lf_data->cm, lf_data->planes, lf_data->xd, cur_job_info->mi_row, cur_job_info->plane, cur_job_info->dir, lpf_opt_level, lf_sync, error_info, lf_data->params_buf, lf_data->tx_buf, MAX_MIB_SIZE_LOG2); } error_info->setjmp = 0; return 1; } static void loop_filter_rows_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm, MACROBLOCKD *xd, int start, int stop, const int planes_to_lf[MAX_MB_PLANE], AVxWorker *workers, int num_workers, AV1LfSync *lf_sync, int lpf_opt_level) { const AVxWorkerInterface *const winterface = aom_get_worker_interface(); int i; loop_filter_frame_mt_init(cm, start, stop, planes_to_lf, num_workers, lf_sync, lpf_opt_level, MAX_MIB_SIZE_LOG2); // Set up loopfilter thread data. for (i = num_workers - 1; i >= 0; --i) { AVxWorker *const worker = &workers[i]; LFWorkerData *const lf_data = &lf_sync->lfdata[i]; worker->hook = loop_filter_row_worker; worker->data1 = lf_sync; worker->data2 = lf_data; // Loopfilter data loop_filter_data_reset(lf_data, frame, cm, xd); // Start loopfiltering worker->had_error = 0; if (i == 0) { winterface->execute(worker); } else { winterface->launch(worker); } } sync_lf_workers(workers, cm, num_workers); } static void loop_filter_rows(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm, MACROBLOCKD *xd, int start, int stop, const int planes_to_lf[MAX_MB_PLANE], int lpf_opt_level) { // Filter top rows of all planes first, in case the output can be partially // reconstructed row by row. int mi_row, plane, dir; AV1_DEBLOCKING_PARAMETERS params_buf[MAX_MIB_SIZE]; TX_SIZE tx_buf[MAX_MIB_SIZE]; for (mi_row = start; mi_row < stop; mi_row += MAX_MIB_SIZE) { for (plane = 0; plane < MAX_MB_PLANE; ++plane) { if (skip_loop_filter_plane(planes_to_lf, plane, lpf_opt_level)) { continue; } for (dir = 0; dir < 2; ++dir) { av1_thread_loop_filter_rows(frame, cm, xd->plane, xd, mi_row, plane, dir, lpf_opt_level, /*lf_sync=*/NULL, xd->error_info, params_buf, tx_buf, MAX_MIB_SIZE_LOG2); } } } } void av1_loop_filter_frame_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm, MACROBLOCKD *xd, int plane_start, int plane_end, int partial_frame, AVxWorker *workers, int num_workers, AV1LfSync *lf_sync, int lpf_opt_level) { int start_mi_row, end_mi_row, mi_rows_to_filter; int planes_to_lf[MAX_MB_PLANE]; if (!check_planes_to_loop_filter(&cm->lf, planes_to_lf, plane_start, plane_end)) return; start_mi_row = 0; mi_rows_to_filter = cm->mi_params.mi_rows; if (partial_frame && cm->mi_params.mi_rows > 8) { start_mi_row = cm->mi_params.mi_rows >> 1; start_mi_row &= 0xfffffff8; mi_rows_to_filter = AOMMAX(cm->mi_params.mi_rows / 8, 8); } end_mi_row = start_mi_row + mi_rows_to_filter; av1_loop_filter_frame_init(cm, plane_start, plane_end); if (num_workers > 1) { // Enqueue and execute loopfiltering jobs. loop_filter_rows_mt(frame, cm, xd, start_mi_row, end_mi_row, planes_to_lf, workers, num_workers, lf_sync, lpf_opt_level); } else { // Directly filter in the main thread. loop_filter_rows(frame, cm, xd, start_mi_row, end_mi_row, planes_to_lf, lpf_opt_level); } } #if !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER static inline void lr_sync_read(void *const lr_sync, int r, int c, int plane) { #if CONFIG_MULTITHREAD AV1LrSync *const loop_res_sync = (AV1LrSync *)lr_sync; const int nsync = loop_res_sync->sync_range; if (r && !(c & (nsync - 1))) { pthread_mutex_t *const mutex = &loop_res_sync->mutex_[plane][r - 1]; pthread_mutex_lock(mutex); while (c > loop_res_sync->cur_sb_col[plane][r - 1] - nsync) { pthread_cond_wait(&loop_res_sync->cond_[plane][r - 1], mutex); } pthread_mutex_unlock(mutex); } #else (void)lr_sync; (void)r; (void)c; (void)plane; #endif // CONFIG_MULTITHREAD } static inline void lr_sync_write(void *const lr_sync, int r, int c, const int sb_cols, int plane) { #if CONFIG_MULTITHREAD AV1LrSync *const loop_res_sync = (AV1LrSync *)lr_sync; const int nsync = loop_res_sync->sync_range; int cur; // Only signal when there are enough filtered SB for next row to run. int sig = 1; if (c < sb_cols - 1) { cur = c; if (c % nsync) sig = 0; } else { cur = sb_cols + nsync; } if (sig) { pthread_mutex_lock(&loop_res_sync->mutex_[plane][r]); // When a thread encounters an error, cur_sb_col[plane][r] is set to maximum // column number. In this case, the AOMMAX operation here ensures that // cur_sb_col[plane][r] is not overwritten with a smaller value thus // preventing the infinite waiting of threads in the relevant sync_read() // function. loop_res_sync->cur_sb_col[plane][r] = AOMMAX(loop_res_sync->cur_sb_col[plane][r], cur); pthread_cond_broadcast(&loop_res_sync->cond_[plane][r]); pthread_mutex_unlock(&loop_res_sync->mutex_[plane][r]); } #else (void)lr_sync; (void)r; (void)c; (void)sb_cols; (void)plane; #endif // CONFIG_MULTITHREAD } // Allocate memory for loop restoration row synchronization void av1_loop_restoration_alloc(AV1LrSync *lr_sync, AV1_COMMON *cm, int num_workers, int num_rows_lr, int num_planes, int width) { lr_sync->rows = num_rows_lr; lr_sync->num_planes = num_planes; #if CONFIG_MULTITHREAD { int i, j; for (j = 0; j < num_planes; j++) { CHECK_MEM_ERROR(cm, lr_sync->mutex_[j], aom_malloc(sizeof(*(lr_sync->mutex_[j])) * num_rows_lr)); if (lr_sync->mutex_[j]) { for (i = 0; i < num_rows_lr; ++i) { pthread_mutex_init(&lr_sync->mutex_[j][i], NULL); } } CHECK_MEM_ERROR(cm, lr_sync->cond_[j], aom_malloc(sizeof(*(lr_sync->cond_[j])) * num_rows_lr)); if (lr_sync->cond_[j]) { for (i = 0; i < num_rows_lr; ++i) { pthread_cond_init(&lr_sync->cond_[j][i], NULL); } } } CHECK_MEM_ERROR(cm, lr_sync->job_mutex, aom_malloc(sizeof(*(lr_sync->job_mutex)))); if (lr_sync->job_mutex) { pthread_mutex_init(lr_sync->job_mutex, NULL); } } #endif // CONFIG_MULTITHREAD CHECK_MEM_ERROR(cm, lr_sync->lrworkerdata, aom_calloc(num_workers, sizeof(*(lr_sync->lrworkerdata)))); lr_sync->num_workers = num_workers; for (int worker_idx = 0; worker_idx < num_workers; ++worker_idx) { if (worker_idx < num_workers - 1) { CHECK_MEM_ERROR(cm, lr_sync->lrworkerdata[worker_idx].rst_tmpbuf, (int32_t *)aom_memalign(16, RESTORATION_TMPBUF_SIZE)); CHECK_MEM_ERROR(cm, lr_sync->lrworkerdata[worker_idx].rlbs, aom_malloc(sizeof(RestorationLineBuffers))); } else { lr_sync->lrworkerdata[worker_idx].rst_tmpbuf = cm->rst_tmpbuf; lr_sync->lrworkerdata[worker_idx].rlbs = cm->rlbs; } } for (int j = 0; j < num_planes; j++) { CHECK_MEM_ERROR( cm, lr_sync->cur_sb_col[j], aom_malloc(sizeof(*(lr_sync->cur_sb_col[j])) * num_rows_lr)); } CHECK_MEM_ERROR( cm, lr_sync->job_queue, aom_malloc(sizeof(*(lr_sync->job_queue)) * num_rows_lr * num_planes)); // Set up nsync. lr_sync->sync_range = get_lr_sync_range(width); } // Deallocate loop restoration synchronization related mutex and data void av1_loop_restoration_dealloc(AV1LrSync *lr_sync) { if (lr_sync != NULL) { int j; #if CONFIG_MULTITHREAD int i; for (j = 0; j < MAX_MB_PLANE; j++) { if (lr_sync->mutex_[j] != NULL) { for (i = 0; i < lr_sync->rows; ++i) { pthread_mutex_destroy(&lr_sync->mutex_[j][i]); } aom_free(lr_sync->mutex_[j]); } if (lr_sync->cond_[j] != NULL) { for (i = 0; i < lr_sync->rows; ++i) { pthread_cond_destroy(&lr_sync->cond_[j][i]); } aom_free(lr_sync->cond_[j]); } } if (lr_sync->job_mutex != NULL) { pthread_mutex_destroy(lr_sync->job_mutex); aom_free(lr_sync->job_mutex); } #endif // CONFIG_MULTITHREAD for (j = 0; j < MAX_MB_PLANE; j++) { aom_free(lr_sync->cur_sb_col[j]); } aom_free(lr_sync->job_queue); if (lr_sync->lrworkerdata) { for (int worker_idx = 0; worker_idx < lr_sync->num_workers - 1; worker_idx++) { LRWorkerData *const workerdata_data = lr_sync->lrworkerdata + worker_idx; aom_free(workerdata_data->rst_tmpbuf); aom_free(workerdata_data->rlbs); } aom_free(lr_sync->lrworkerdata); } // 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. av1_zero(*lr_sync); } } static void enqueue_lr_jobs(AV1LrSync *lr_sync, AV1LrStruct *lr_ctxt, AV1_COMMON *cm) { FilterFrameCtxt *ctxt = lr_ctxt->ctxt; const int num_planes = av1_num_planes(cm); AV1LrMTInfo *lr_job_queue = lr_sync->job_queue; int32_t lr_job_counter[2], num_even_lr_jobs = 0; lr_sync->jobs_enqueued = 0; lr_sync->jobs_dequeued = 0; for (int plane = 0; plane < num_planes; plane++) { if (cm->rst_info[plane].frame_restoration_type == RESTORE_NONE) continue; num_even_lr_jobs = num_even_lr_jobs + ((ctxt[plane].rsi->vert_units + 1) >> 1); } lr_job_counter[0] = 0; lr_job_counter[1] = num_even_lr_jobs; for (int plane = 0; plane < num_planes; plane++) { if (cm->rst_info[plane].frame_restoration_type == RESTORE_NONE) continue; const int is_uv = plane > 0; const int ss_y = is_uv && cm->seq_params->subsampling_y; const int unit_size = ctxt[plane].rsi->restoration_unit_size; const int plane_h = ctxt[plane].plane_h; const int ext_size = unit_size * 3 / 2; int y0 = 0, i = 0; while (y0 < plane_h) { int remaining_h = plane_h - y0; int h = (remaining_h < ext_size) ? remaining_h : unit_size; RestorationTileLimits limits; limits.v_start = y0; limits.v_end = y0 + h; assert(limits.v_end <= plane_h); // Offset upwards to align with the restoration processing stripe const int voffset = RESTORATION_UNIT_OFFSET >> ss_y; limits.v_start = AOMMAX(0, limits.v_start - voffset); if (limits.v_end < plane_h) limits.v_end -= voffset; assert(lr_job_counter[0] <= num_even_lr_jobs); lr_job_queue[lr_job_counter[i & 1]].lr_unit_row = i; lr_job_queue[lr_job_counter[i & 1]].plane = plane; lr_job_queue[lr_job_counter[i & 1]].v_start = limits.v_start; lr_job_queue[lr_job_counter[i & 1]].v_end = limits.v_end; lr_job_queue[lr_job_counter[i & 1]].sync_mode = i & 1; if ((i & 1) == 0) { lr_job_queue[lr_job_counter[i & 1]].v_copy_start = limits.v_start + RESTORATION_BORDER; lr_job_queue[lr_job_counter[i & 1]].v_copy_end = limits.v_end - RESTORATION_BORDER; if (i == 0) { assert(limits.v_start == 0); lr_job_queue[lr_job_counter[i & 1]].v_copy_start = 0; } if (i == (ctxt[plane].rsi->vert_units - 1)) { assert(limits.v_end == plane_h); lr_job_queue[lr_job_counter[i & 1]].v_copy_end = plane_h; } } else { lr_job_queue[lr_job_counter[i & 1]].v_copy_start = AOMMAX(limits.v_start - RESTORATION_BORDER, 0); lr_job_queue[lr_job_counter[i & 1]].v_copy_end = AOMMIN(limits.v_end + RESTORATION_BORDER, plane_h); } lr_job_counter[i & 1]++; lr_sync->jobs_enqueued++; y0 += h; ++i; } } } static AV1LrMTInfo *get_lr_job_info(AV1LrSync *lr_sync) { AV1LrMTInfo *cur_job_info = NULL; #if CONFIG_MULTITHREAD pthread_mutex_lock(lr_sync->job_mutex); if (!lr_sync->lr_mt_exit && lr_sync->jobs_dequeued < lr_sync->jobs_enqueued) { cur_job_info = lr_sync->job_queue + lr_sync->jobs_dequeued; lr_sync->jobs_dequeued++; } pthread_mutex_unlock(lr_sync->job_mutex); #else (void)lr_sync; #endif return cur_job_info; } static void set_loop_restoration_done(AV1LrSync *const lr_sync, FilterFrameCtxt *const ctxt) { for (int plane = 0; plane < MAX_MB_PLANE; ++plane) { if (ctxt[plane].rsi->frame_restoration_type == RESTORE_NONE) continue; int y0 = 0, row_number = 0; const int unit_size = ctxt[plane].rsi->restoration_unit_size; const int plane_h = ctxt[plane].plane_h; const int ext_size = unit_size * 3 / 2; const int hnum_rest_units = ctxt[plane].rsi->horz_units; while (y0 < plane_h) { const int remaining_h = plane_h - y0; const int h = (remaining_h < ext_size) ? remaining_h : unit_size; lr_sync_write(lr_sync, row_number, hnum_rest_units - 1, hnum_rest_units, plane); y0 += h; ++row_number; } } } // Implement row loop restoration for each thread. static int loop_restoration_row_worker(void *arg1, void *arg2) { AV1LrSync *const lr_sync = (AV1LrSync *)arg1; LRWorkerData *lrworkerdata = (LRWorkerData *)arg2; AV1LrStruct *lr_ctxt = (AV1LrStruct *)lrworkerdata->lr_ctxt; FilterFrameCtxt *ctxt = lr_ctxt->ctxt; int lr_unit_row; int plane; int plane_w; #if CONFIG_MULTITHREAD pthread_mutex_t *job_mutex_ = lr_sync->job_mutex; #endif struct aom_internal_error_info *const error_info = &lrworkerdata->error_info; // The jmp_buf is valid only for the duration of the function that calls // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 // before it returns. if (setjmp(error_info->jmp)) { error_info->setjmp = 0; #if CONFIG_MULTITHREAD pthread_mutex_lock(job_mutex_); lr_sync->lr_mt_exit = true; pthread_mutex_unlock(job_mutex_); #endif // In case of loop restoration multithreading, the worker on an even lr // block row waits for the completion of the filtering of the top-right and // bottom-right blocks. Hence, in case a thread (main/worker) encounters an // error, update that filtering of every row in the frame is complete in // order to avoid the dependent workers from waiting indefinitely. set_loop_restoration_done(lr_sync, lr_ctxt->ctxt); return 0; } error_info->setjmp = 1; typedef void (*copy_fun)(const YV12_BUFFER_CONFIG *src_ybc, YV12_BUFFER_CONFIG *dst_ybc, int hstart, int hend, int vstart, int vend); static const copy_fun copy_funs[MAX_MB_PLANE] = { aom_yv12_partial_coloc_copy_y, aom_yv12_partial_coloc_copy_u, aom_yv12_partial_coloc_copy_v }; while (1) { AV1LrMTInfo *cur_job_info = get_lr_job_info(lr_sync); if (cur_job_info != NULL) { RestorationTileLimits limits; sync_read_fn_t on_sync_read; sync_write_fn_t on_sync_write; limits.v_start = cur_job_info->v_start; limits.v_end = cur_job_info->v_end; lr_unit_row = cur_job_info->lr_unit_row; plane = cur_job_info->plane; plane_w = ctxt[plane].plane_w; // sync_mode == 1 implies only sync read is required in LR Multi-threading // sync_mode == 0 implies only sync write is required. on_sync_read = cur_job_info->sync_mode == 1 ? lr_sync_read : av1_lr_sync_read_dummy; on_sync_write = cur_job_info->sync_mode == 0 ? lr_sync_write : av1_lr_sync_write_dummy; av1_foreach_rest_unit_in_row( &limits, plane_w, lr_ctxt->on_rest_unit, lr_unit_row, ctxt[plane].rsi->restoration_unit_size, ctxt[plane].rsi->horz_units, ctxt[plane].rsi->vert_units, plane, &ctxt[plane], lrworkerdata->rst_tmpbuf, lrworkerdata->rlbs, on_sync_read, on_sync_write, lr_sync, error_info); copy_funs[plane](lr_ctxt->dst, lr_ctxt->frame, 0, plane_w, cur_job_info->v_copy_start, cur_job_info->v_copy_end); if (lrworkerdata->do_extend_border) { aom_extend_frame_borders_plane_row(lr_ctxt->frame, plane, cur_job_info->v_copy_start, cur_job_info->v_copy_end); } } else { break; } } error_info->setjmp = 0; return 1; } static inline void sync_lr_workers(AVxWorker *const workers, AV1_COMMON *const cm, int num_workers) { const AVxWorkerInterface *const winterface = aom_get_worker_interface(); int had_error = workers[0].had_error; struct aom_internal_error_info error_info; // Read the error_info of main thread. if (had_error) { AVxWorker *const worker = &workers[0]; error_info = ((LRWorkerData *)worker->data2)->error_info; } // Wait till all rows are finished. for (int i = num_workers - 1; i > 0; --i) { AVxWorker *const worker = &workers[i]; if (!winterface->sync(worker)) { had_error = 1; error_info = ((LRWorkerData *)worker->data2)->error_info; } } if (had_error) aom_internal_error_copy(cm->error, &error_info); } static void foreach_rest_unit_in_planes_mt(AV1LrStruct *lr_ctxt, AVxWorker *workers, int num_workers, AV1LrSync *lr_sync, AV1_COMMON *cm, int do_extend_border) { FilterFrameCtxt *ctxt = lr_ctxt->ctxt; const int num_planes = av1_num_planes(cm); const AVxWorkerInterface *const winterface = aom_get_worker_interface(); int num_rows_lr = 0; for (int plane = 0; plane < num_planes; plane++) { if (cm->rst_info[plane].frame_restoration_type == RESTORE_NONE) continue; const int plane_h = ctxt[plane].plane_h; const int unit_size = cm->rst_info[plane].restoration_unit_size; num_rows_lr = AOMMAX(num_rows_lr, av1_lr_count_units(unit_size, plane_h)); } int i; assert(MAX_MB_PLANE == 3); if (!lr_sync->sync_range || num_rows_lr > lr_sync->rows || num_workers > lr_sync->num_workers || num_planes > lr_sync->num_planes) { av1_loop_restoration_dealloc(lr_sync); av1_loop_restoration_alloc(lr_sync, cm, num_workers, num_rows_lr, num_planes, cm->width); } lr_sync->lr_mt_exit = false; // Initialize cur_sb_col to -1 for all SB rows. for (i = 0; i < num_planes; i++) { memset(lr_sync->cur_sb_col[i], -1, sizeof(*(lr_sync->cur_sb_col[i])) * num_rows_lr); } enqueue_lr_jobs(lr_sync, lr_ctxt, cm); // Set up looprestoration thread data. for (i = num_workers - 1; i >= 0; --i) { AVxWorker *const worker = &workers[i]; lr_sync->lrworkerdata[i].lr_ctxt = (void *)lr_ctxt; lr_sync->lrworkerdata[i].do_extend_border = do_extend_border; worker->hook = loop_restoration_row_worker; worker->data1 = lr_sync; worker->data2 = &lr_sync->lrworkerdata[i]; // Start loop restoration worker->had_error = 0; if (i == 0) { winterface->execute(worker); } else { winterface->launch(worker); } } sync_lr_workers(workers, cm, num_workers); } void av1_loop_restoration_filter_frame_mt(YV12_BUFFER_CONFIG *frame, AV1_COMMON *cm, int optimized_lr, AVxWorker *workers, int num_workers, AV1LrSync *lr_sync, void *lr_ctxt, int do_extend_border) { assert(!cm->features.all_lossless); const int num_planes = av1_num_planes(cm); AV1LrStruct *loop_rest_ctxt = (AV1LrStruct *)lr_ctxt; av1_loop_restoration_filter_frame_init(loop_rest_ctxt, frame, cm, optimized_lr, num_planes); foreach_rest_unit_in_planes_mt(loop_rest_ctxt, workers, num_workers, lr_sync, cm, do_extend_border); } #endif // !CONFIG_REALTIME_ONLY || CONFIG_AV1_DECODER // Initializes cdef_sync parameters. static inline void reset_cdef_job_info(AV1CdefSync *const cdef_sync) { cdef_sync->end_of_frame = 0; cdef_sync->fbr = 0; cdef_sync->fbc = 0; cdef_sync->cdef_mt_exit = false; } static inline void launch_cdef_workers(AVxWorker *const workers, int num_workers) { const AVxWorkerInterface *const winterface = aom_get_worker_interface(); for (int i = num_workers - 1; i >= 0; i--) { AVxWorker *const worker = &workers[i]; worker->had_error = 0; if (i == 0) winterface->execute(worker); else winterface->launch(worker); } } static inline void sync_cdef_workers(AVxWorker *const workers, AV1_COMMON *const cm, int num_workers) { const AVxWorkerInterface *const winterface = aom_get_worker_interface(); int had_error = workers[0].had_error; struct aom_internal_error_info error_info; // Read the error_info of main thread. if (had_error) { AVxWorker *const worker = &workers[0]; error_info = ((AV1CdefWorkerData *)worker->data2)->error_info; } // Wait till all rows are finished. for (int i = num_workers - 1; i > 0; --i) { AVxWorker *const worker = &workers[i]; if (!winterface->sync(worker)) { had_error = 1; error_info = ((AV1CdefWorkerData *)worker->data2)->error_info; } } if (had_error) aom_internal_error_copy(cm->error, &error_info); } // Updates the row index of the next job to be processed. // Also updates end_of_frame flag when the processing of all rows is complete. static void update_cdef_row_next_job_info(AV1CdefSync *const cdef_sync, const int nvfb) { cdef_sync->fbr++; if (cdef_sync->fbr == nvfb) { cdef_sync->end_of_frame = 1; } } // Checks if a job is available. If job is available, // populates next job information and returns 1, else returns 0. static inline int get_cdef_row_next_job(AV1CdefSync *const cdef_sync, volatile int *cur_fbr, const int nvfb) { #if CONFIG_MULTITHREAD pthread_mutex_lock(cdef_sync->mutex_); #endif // CONFIG_MULTITHREAD int do_next_row = 0; // Populates information needed for current job and update the row // index of the next row to be processed. if (!cdef_sync->cdef_mt_exit && cdef_sync->end_of_frame == 0) { do_next_row = 1; *cur_fbr = cdef_sync->fbr; update_cdef_row_next_job_info(cdef_sync, nvfb); } #if CONFIG_MULTITHREAD pthread_mutex_unlock(cdef_sync->mutex_); #endif // CONFIG_MULTITHREAD return do_next_row; } static void set_cdef_init_fb_row_done(AV1CdefSync *const cdef_sync, int nvfb) { for (int fbr = 0; fbr < nvfb; fbr++) cdef_row_mt_sync_write(cdef_sync, fbr); } // Hook function for each thread in CDEF multi-threading. static int cdef_sb_row_worker_hook(void *arg1, void *arg2) { AV1CdefSync *const cdef_sync = (AV1CdefSync *)arg1; AV1CdefWorkerData *const cdef_worker = (AV1CdefWorkerData *)arg2; AV1_COMMON *cm = cdef_worker->cm; const int nvfb = (cm->mi_params.mi_rows + MI_SIZE_64X64 - 1) / MI_SIZE_64X64; #if CONFIG_MULTITHREAD pthread_mutex_t *job_mutex_ = cdef_sync->mutex_; #endif struct aom_internal_error_info *const error_info = &cdef_worker->error_info; // The jmp_buf is valid only for the duration of the function that calls // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 // before it returns. if (setjmp(error_info->jmp)) { error_info->setjmp = 0; #if CONFIG_MULTITHREAD pthread_mutex_lock(job_mutex_); cdef_sync->cdef_mt_exit = true; pthread_mutex_unlock(job_mutex_); #endif // In case of cdef row-multithreading, the worker on a filter block row // (fbr) waits for the line buffers (top and bottom) copy of the above row. // Hence, in case a thread (main/worker) encounters an error before copying // of the line buffers, update that line buffer copy is complete in order to // avoid dependent workers waiting indefinitely. set_cdef_init_fb_row_done(cdef_sync, nvfb); return 0; } error_info->setjmp = 1; volatile int cur_fbr; const int num_planes = av1_num_planes(cm); while (get_cdef_row_next_job(cdef_sync, &cur_fbr, nvfb)) { MACROBLOCKD *xd = cdef_worker->xd; av1_cdef_fb_row(cm, xd, cdef_worker->linebuf, cdef_worker->colbuf, cdef_worker->srcbuf, cur_fbr, cdef_worker->cdef_init_fb_row_fn, cdef_sync, error_info); if (cdef_worker->do_extend_border) { for (int plane = 0; plane < num_planes; ++plane) { const YV12_BUFFER_CONFIG *ybf = &cm->cur_frame->buf; const int is_uv = plane > 0; const int mi_high = MI_SIZE_LOG2 - xd->plane[plane].subsampling_y; const int unit_height = MI_SIZE_64X64 << mi_high; const int v_start = cur_fbr * unit_height; const int v_end = AOMMIN(v_start + unit_height, ybf->crop_heights[is_uv]); aom_extend_frame_borders_plane_row(ybf, plane, v_start, v_end); } } } error_info->setjmp = 0; return 1; } // Assigns CDEF hook function and thread data to each worker. static void prepare_cdef_frame_workers( AV1_COMMON *const cm, MACROBLOCKD *xd, AV1CdefWorkerData *const cdef_worker, AVxWorkerHook hook, AVxWorker *const workers, AV1CdefSync *const cdef_sync, int num_workers, cdef_init_fb_row_t cdef_init_fb_row_fn, int do_extend_border) { const int num_planes = av1_num_planes(cm); cdef_worker[0].srcbuf = cm->cdef_info.srcbuf; for (int plane = 0; plane < num_planes; plane++) cdef_worker[0].colbuf[plane] = cm->cdef_info.colbuf[plane]; for (int i = num_workers - 1; i >= 0; i--) { AVxWorker *const worker = &workers[i]; cdef_worker[i].cm = cm; cdef_worker[i].xd = xd; cdef_worker[i].cdef_init_fb_row_fn = cdef_init_fb_row_fn; cdef_worker[i].do_extend_border = do_extend_border; for (int plane = 0; plane < num_planes; plane++) cdef_worker[i].linebuf[plane] = cm->cdef_info.linebuf[plane]; worker->hook = hook; worker->data1 = cdef_sync; worker->data2 = &cdef_worker[i]; } } // Initializes row-level parameters for CDEF frame. void av1_cdef_init_fb_row_mt(const AV1_COMMON *const cm, const MACROBLOCKD *const xd, CdefBlockInfo *const fb_info, uint16_t **const linebuf, uint16_t *const src, struct AV1CdefSyncData *const cdef_sync, int fbr) { const int num_planes = av1_num_planes(cm); const int nvfb = (cm->mi_params.mi_rows + MI_SIZE_64X64 - 1) / MI_SIZE_64X64; const int luma_stride = ALIGN_POWER_OF_TWO(cm->mi_params.mi_cols << MI_SIZE_LOG2, 4); // for the current filter block, it's top left corner mi structure (mi_tl) // is first accessed to check whether the top and left boundaries are // frame boundaries. Then bottom-left and top-right mi structures are // accessed to check whether the bottom and right boundaries // (respectively) are frame boundaries. // // Note that we can't just check the bottom-right mi structure - eg. if // we're at the right-hand edge of the frame but not the bottom, then // the bottom-right mi is NULL but the bottom-left is not. fb_info->frame_boundary[TOP] = (MI_SIZE_64X64 * fbr == 0) ? 1 : 0; if (fbr != nvfb - 1) fb_info->frame_boundary[BOTTOM] = (MI_SIZE_64X64 * (fbr + 1) == cm->mi_params.mi_rows) ? 1 : 0; else fb_info->frame_boundary[BOTTOM] = 1; fb_info->src = src; fb_info->damping = cm->cdef_info.cdef_damping; fb_info->coeff_shift = AOMMAX(cm->seq_params->bit_depth - 8, 0); av1_zero(fb_info->dir); av1_zero(fb_info->var); for (int plane = 0; plane < num_planes; plane++) { const int stride = luma_stride >> xd->plane[plane].subsampling_x; uint16_t *top_linebuf = &linebuf[plane][0]; uint16_t *bot_linebuf = &linebuf[plane][nvfb * CDEF_VBORDER * stride]; { const int mi_high_l2 = MI_SIZE_LOG2 - xd->plane[plane].subsampling_y; const int top_offset = MI_SIZE_64X64 * (fbr + 1) << mi_high_l2; const int bot_offset = MI_SIZE_64X64 * (fbr + 1) << mi_high_l2; if (fbr != nvfb - 1) // if (fbr != 0) // top line buffer copy av1_cdef_copy_sb8_16( cm, &top_linebuf[(fbr + 1) * CDEF_VBORDER * stride], stride, xd->plane[plane].dst.buf, top_offset - CDEF_VBORDER, 0, xd->plane[plane].dst.stride, CDEF_VBORDER, stride); if (fbr != nvfb - 1) // bottom line buffer copy av1_cdef_copy_sb8_16(cm, &bot_linebuf[fbr * CDEF_VBORDER * stride], stride, xd->plane[plane].dst.buf, bot_offset, 0, xd->plane[plane].dst.stride, CDEF_VBORDER, stride); } fb_info->top_linebuf[plane] = &linebuf[plane][fbr * CDEF_VBORDER * stride]; fb_info->bot_linebuf[plane] = &linebuf[plane] [nvfb * CDEF_VBORDER * stride + (fbr * CDEF_VBORDER * stride)]; } cdef_row_mt_sync_write(cdef_sync, fbr); cdef_row_mt_sync_read(cdef_sync, fbr); } // Implements multi-threading for CDEF. // Perform CDEF on input frame. // Inputs: // frame: Pointer to input frame buffer. // cm: Pointer to common structure. // xd: Pointer to common current coding block structure. // Returns: // Nothing will be returned. void av1_cdef_frame_mt(AV1_COMMON *const cm, MACROBLOCKD *const xd, AV1CdefWorkerData *const cdef_worker, AVxWorker *const workers, AV1CdefSync *const cdef_sync, int num_workers, cdef_init_fb_row_t cdef_init_fb_row_fn, int do_extend_border) { YV12_BUFFER_CONFIG *frame = &cm->cur_frame->buf; const int num_planes = av1_num_planes(cm); av1_setup_dst_planes(xd->plane, cm->seq_params->sb_size, frame, 0, 0, 0, num_planes); reset_cdef_job_info(cdef_sync); prepare_cdef_frame_workers(cm, xd, cdef_worker, cdef_sb_row_worker_hook, workers, cdef_sync, num_workers, cdef_init_fb_row_fn, do_extend_border); launch_cdef_workers(workers, num_workers); sync_cdef_workers(workers, cm, num_workers); } int av1_get_intrabc_extra_top_right_sb_delay(const AV1_COMMON *cm) { // No additional top-right delay when intraBC tool is not enabled. if (!av1_allow_intrabc(cm)) return 0; // Due to the hardware constraints on processing the intraBC tool with row // multithreading, a top-right delay of 3 superblocks of size 128x128 or 5 // superblocks of size 64x64 is mandated. However, a minimum top-right delay // of 1 superblock is assured with 'sync_range'. Hence return only the // additional superblock delay when the intraBC tool is enabled. return cm->seq_params->sb_size == BLOCK_128X128 ? 2 : 4; }