From 3b47f335c6792cb96b2f97cf1f24cdf91e2b59bf Mon Sep 17 00:00:00 2001 From: Brecht Van Lommel Date: Thu, 23 Apr 2020 15:15:05 +0200 Subject: BLI: remove TaskParallelRangePool This is not currently used and will take some work to support with TBB, so remove it until we have a new implementation based on TBB. Fixes T76005, parallel range pool tests failing. Ref D7475 --- source/blender/blenlib/BLI_task.h | 12 -- source/blender/blenlib/intern/task_iterator.c | 204 -------------------------- 2 files changed, 216 deletions(-) (limited to 'source/blender/blenlib') diff --git a/source/blender/blenlib/BLI_task.h b/source/blender/blenlib/BLI_task.h index 2b5964d091c..42dd47266dc 100644 --- a/source/blender/blenlib/BLI_task.h +++ b/source/blender/blenlib/BLI_task.h @@ -201,18 +201,6 @@ void BLI_task_parallel_range(const int start, TaskParallelRangeFunc func, TaskParallelSettings *settings); -typedef struct TaskParallelRangePool TaskParallelRangePool; -struct TaskParallelRangePool *BLI_task_parallel_range_pool_init( - const struct TaskParallelSettings *settings); -void BLI_task_parallel_range_pool_push(struct TaskParallelRangePool *range_pool, - const int start, - const int stop, - void *userdata, - TaskParallelRangeFunc func, - const struct TaskParallelSettings *settings); -void BLI_task_parallel_range_pool_work_and_wait(struct TaskParallelRangePool *range_pool); -void BLI_task_parallel_range_pool_free(struct TaskParallelRangePool *range_pool); - /* This data is shared between all tasks, its access needs thread lock or similar protection. */ typedef struct TaskParallelIteratorStateShared { diff --git a/source/blender/blenlib/intern/task_iterator.c b/source/blender/blenlib/intern/task_iterator.c index 8a596879b2c..1189ec0d0c0 100644 --- a/source/blender/blenlib/intern/task_iterator.c +++ b/source/blender/blenlib/intern/task_iterator.c @@ -382,210 +382,6 @@ void BLI_task_parallel_range(const int start, } } -/** - * Initialize a task pool to parallelize several for loops at the same time. - * - * See public API doc of ParallelRangeSettings for description of all settings. - * Note that loop-specific settings (like 'tls' data or reduce/free functions) must be left NULL - * here. Only settings controlling how iteration is parallelized must be defined, as those will - * affect all loops added to that pool. - */ -TaskParallelRangePool *BLI_task_parallel_range_pool_init(const TaskParallelSettings *settings) -{ - TaskParallelRangePool *range_pool = MEM_callocN(sizeof(*range_pool), __func__); - - BLI_assert(settings->userdata_chunk == NULL); - BLI_assert(settings->func_reduce == NULL); - BLI_assert(settings->func_free == NULL); - range_pool->settings = MEM_mallocN(sizeof(*range_pool->settings), __func__); - *range_pool->settings = *settings; - - return range_pool; -} - -/** - * Add a loop task to the pool. It does not execute it at all. - * - * See public API doc of ParallelRangeSettings for description of all settings. - * Note that only 'tls'-related data are used here. - */ -void BLI_task_parallel_range_pool_push(TaskParallelRangePool *range_pool, - const int start, - const int stop, - void *userdata, - TaskParallelRangeFunc func, - const TaskParallelSettings *settings) -{ - BLI_assert(range_pool->pool == NULL); - - if (start == stop) { - return; - } - - BLI_assert(start < stop); - if (settings->userdata_chunk_size != 0) { - BLI_assert(settings->userdata_chunk != NULL); - } - - TaskParallelRangeState *state = MEM_callocN(sizeof(*state), __func__); - state->start = start; - state->stop = stop; - state->userdata_shared = userdata; - state->func = func; - state->iter_value = start; - state->initial_tls_memory = settings->userdata_chunk; - state->tls_data_size = settings->userdata_chunk_size; - state->func_reduce = settings->func_reduce; - state->func_free = settings->func_free; - - state->next = range_pool->parallel_range_states; - range_pool->parallel_range_states = state; -} - -static void parallel_range_func_finalize(TaskPool *__restrict pool, - void *v_state, - int UNUSED(thread_id)) -{ - TaskParallelRangePool *__restrict range_pool = BLI_task_pool_user_data(pool); - TaskParallelRangeState *state = v_state; - - for (int i = 0; i < range_pool->num_tasks; i++) { - void *tls_data = (char *)state->flatten_tls_storage + (state->tls_data_size * (size_t)i); - if (state->func_reduce != NULL) { - state->func_reduce(state->userdata_shared, state->initial_tls_memory, tls_data); - } - if (state->func_free != NULL) { - /* `func_free` should only free data that was created during execution of `func`. */ - state->func_free(state->userdata_shared, tls_data); - } - } -} - -/** - * Run all tasks pushed to the range_pool. - * - * Note that the range pool is re-usable (you may push new tasks into it and call this function - * again). - */ -void BLI_task_parallel_range_pool_work_and_wait(TaskParallelRangePool *range_pool) -{ - BLI_assert(range_pool->pool == NULL); - - /* If it's not enough data to be crunched, don't bother with tasks at all, - * do everything from the current thread. - */ - if (!range_pool->settings->use_threading) { - parallel_range_single_thread(range_pool); - return; - } - - TaskScheduler *task_scheduler = BLI_task_scheduler_get(); - const int num_threads = BLI_task_scheduler_num_threads(task_scheduler); - - /* The idea here is to prevent creating task for each of the loop iterations - * and instead have tasks which are evenly distributed across CPU cores and - * pull next iter to be crunched using the queue. - */ - int num_tasks = num_threads + 2; - range_pool->num_tasks = num_tasks; - - task_parallel_range_calc_chunk_size(range_pool); - range_pool->num_tasks = num_tasks = min_ii( - num_tasks, max_ii(1, range_pool->num_total_iters / range_pool->chunk_size)); - - if (num_tasks == 1) { - parallel_range_single_thread(range_pool); - return; - } - - /* We create all 'tls' data here in a single loop. */ - for (TaskParallelRangeState *state = range_pool->parallel_range_states; state != NULL; - state = state->next) { - void *userdata_chunk = state->initial_tls_memory; - const size_t userdata_chunk_size = state->tls_data_size; - if (userdata_chunk_size == 0) { - BLI_assert(userdata_chunk == NULL); - continue; - } - - void *userdata_chunk_array = NULL; - state->flatten_tls_storage = userdata_chunk_array = MALLOCA(userdata_chunk_size * - (size_t)num_tasks); - for (int i = 0; i < num_tasks; i++) { - void *userdata_chunk_local = (char *)userdata_chunk_array + - (userdata_chunk_size * (size_t)i); - memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size); - } - } - - TaskPool *task_pool = range_pool->pool = BLI_task_pool_create_suspended( - task_scheduler, range_pool, TASK_PRIORITY_HIGH); - - range_pool->current_state = range_pool->parallel_range_states; - const int thread_id = BLI_task_pool_creator_thread_id(task_pool); - for (int i = 0; i < num_tasks; i++) { - BLI_task_pool_push_from_thread( - task_pool, parallel_range_func, POINTER_FROM_INT(i), false, NULL, thread_id); - } - - BLI_task_pool_work_and_wait(task_pool); - - BLI_assert(atomic_cas_ptr((void **)&range_pool->current_state, NULL, NULL) == NULL); - - /* Finalize all tasks. */ - for (TaskParallelRangeState *state = range_pool->parallel_range_states; state != NULL; - state = state->next) { - const size_t userdata_chunk_size = state->tls_data_size; - void *userdata_chunk_array = state->flatten_tls_storage; - UNUSED_VARS_NDEBUG(userdata_chunk_array); - if (userdata_chunk_size == 0) { - BLI_assert(userdata_chunk_array == NULL); - continue; - } - - if (state->func_reduce != NULL || state->func_free != NULL) { - BLI_task_pool_push_from_thread( - task_pool, parallel_range_func_finalize, state, false, NULL, thread_id); - } - } - - BLI_task_pool_work_and_wait(task_pool); - BLI_task_pool_free(task_pool); - range_pool->pool = NULL; - - /* Cleanup all tasks. */ - TaskParallelRangeState *state_next; - for (TaskParallelRangeState *state = range_pool->parallel_range_states; state != NULL; - state = state_next) { - state_next = state->next; - - const size_t userdata_chunk_size = state->tls_data_size; - void *userdata_chunk_array = state->flatten_tls_storage; - if (userdata_chunk_size != 0) { - BLI_assert(userdata_chunk_array != NULL); - MALLOCA_FREE(userdata_chunk_array, userdata_chunk_size * (size_t)num_tasks); - } - - MEM_freeN(state); - } - range_pool->parallel_range_states = NULL; -} - -/** - * Clear/free given \a range_pool. - */ -void BLI_task_parallel_range_pool_free(TaskParallelRangePool *range_pool) -{ - TaskParallelRangeState *state_next = NULL; - for (TaskParallelRangeState *state = range_pool->parallel_range_states; state != NULL; - state = state_next) { - state_next = state->next; - MEM_freeN(state); - } - MEM_freeN(range_pool->settings); - MEM_freeN(range_pool); -} - typedef struct TaskParallelIteratorState { void *userdata; TaskParallelIteratorIterFunc iter_func; -- cgit v1.2.3