This repository has been archived on 2023-10-09. You can view files and clone it. You cannot open issues or pull requests or push a commit.
Files
blender-archive/source/blender/depsgraph/intern/eval/deg_eval.cc
Brecht Van Lommel d8a3f3595a Task: Use TBB as Task Scheduler
This patch enables TBB as the default task scheduler. TBB stands for Threading Building Blocks and is developed by Intel. The library contains several threading patters. This patch maps blenders BLI_task_* function to their counterpart. After this patch we can add more patterns. A promising one is TBB:graph that can be used for depsgraph, draw manager and compositor.

Performance changes depends on the actual hardware. It was tested on different hardwares from laptops to workstations and we didn't detected any downgrade of the performance.
* Linux Xeon E5-2699 v4 got FPS boost from 12 to 17 using Spring's 04_010_A.anim.blend.
* AMD Ryzen Threadripper 2990WX 32-Core Animation playback goes from 9.5-10.5 FPS to 13.0-14.0 FPS on Agent 327 , 10_03_B.anim.blend.

Reviewed By: brecht, sergey

Differential Revision: https://developer.blender.org/D7475
2020-04-30 08:09:21 +02:00

420 lines
13 KiB
C++

/*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software Foundation,
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*
* The Original Code is Copyright (C) 2013 Blender Foundation.
* All rights reserved.
*/
/** \file
* \ingroup depsgraph
*
* Evaluation engine entrypoints for Depsgraph Engine.
*/
#include "intern/eval/deg_eval.h"
#include "PIL_time.h"
#include "BLI_compiler_attrs.h"
#include "BLI_gsqueue.h"
#include "BLI_task.h"
#include "BLI_utildefines.h"
#include "BKE_global.h"
#include "DNA_node_types.h"
#include "DNA_object_types.h"
#include "DNA_scene_types.h"
#include "DEG_depsgraph.h"
#include "DEG_depsgraph_query.h"
#include "atomic_ops.h"
#include "intern/depsgraph.h"
#include "intern/depsgraph_relation.h"
#include "intern/eval/deg_eval_copy_on_write.h"
#include "intern/eval/deg_eval_flush.h"
#include "intern/eval/deg_eval_stats.h"
#include "intern/node/deg_node.h"
#include "intern/node/deg_node_component.h"
#include "intern/node/deg_node_id.h"
#include "intern/node/deg_node_operation.h"
#include "intern/node/deg_node_time.h"
namespace DEG {
namespace {
struct DepsgraphEvalState;
void deg_task_run_func(TaskPool *pool, void *taskdata);
template<typename ScheduleFunction, typename... ScheduleFunctionArgs>
void schedule_children(DepsgraphEvalState *state,
OperationNode *node,
ScheduleFunction *schedule_function,
ScheduleFunctionArgs... schedule_function_args);
void schedule_node_to_pool(OperationNode *node, const int UNUSED(thread_id), TaskPool *pool)
{
BLI_task_pool_push(pool, deg_task_run_func, node, false, NULL);
}
/* Denotes which part of dependency graph is being evaluated. */
enum class EvaluationStage {
/* Stage 1: Only Copy-on-Write operations are to be evaluated, prior to anything else.
* This allows other operations to access its dependencies when there is a dependency cycle
* involved. */
COPY_ON_WRITE,
/* Threaded evaluation of all possible operations. */
THREADED_EVALUATION,
/* Workaround for areas which can not be evaluated in threads.
*
* For example, metaballs, which are iterating over all bases and are requesting dupli-lists
* to see whether there are metaballs inside. */
SINGLE_THREADED_WORKAROUND,
};
struct DepsgraphEvalState {
Depsgraph *graph;
bool do_stats;
EvaluationStage stage;
bool need_single_thread_pass;
};
void evaluate_node(const DepsgraphEvalState *state, OperationNode *operation_node)
{
::Depsgraph *depsgraph = reinterpret_cast<::Depsgraph *>(state->graph);
/* Sanity checks. */
BLI_assert(!operation_node->is_noop() && "NOOP nodes should not actually be scheduled");
/* Perform operation. */
if (state->do_stats) {
const double start_time = PIL_check_seconds_timer();
operation_node->evaluate(depsgraph);
operation_node->stats.current_time += PIL_check_seconds_timer() - start_time;
}
else {
operation_node->evaluate(depsgraph);
}
}
void deg_task_run_func(TaskPool *pool, void *taskdata)
{
void *userdata_v = BLI_task_pool_user_data(pool);
DepsgraphEvalState *state = (DepsgraphEvalState *)userdata_v;
/* Evaluate node. */
OperationNode *operation_node = reinterpret_cast<OperationNode *>(taskdata);
evaluate_node(state, operation_node);
/* Schedule children. */
schedule_children(state, operation_node, schedule_node_to_pool, pool);
}
bool check_operation_node_visible(OperationNode *op_node)
{
const ComponentNode *comp_node = op_node->owner;
/* Special exception, copy on write component is to be always evaluated,
* to keep copied "database" in a consistent state. */
if (comp_node->type == NodeType::COPY_ON_WRITE) {
return true;
}
return comp_node->affects_directly_visible;
}
void calculate_pending_parents_for_node(OperationNode *node)
{
/* Update counters, applies for both visible and invisible IDs. */
node->num_links_pending = 0;
node->scheduled = false;
/* Invisible IDs requires no pending operations. */
if (!check_operation_node_visible(node)) {
return;
}
/* No need to bother with anything if node is not tagged for update. */
if ((node->flag & DEPSOP_FLAG_NEEDS_UPDATE) == 0) {
return;
}
for (Relation *rel : node->inlinks) {
if (rel->from->type == NodeType::OPERATION && (rel->flag & RELATION_FLAG_CYCLIC) == 0) {
OperationNode *from = (OperationNode *)rel->from;
/* TODO(sergey): This is how old layer system was checking for the
* calculation, but how is it possible that visible object depends
* on an invisible? This is something what is prohibited after
* deg_graph_build_flush_layers(). */
if (!check_operation_node_visible(from)) {
continue;
}
/* No need to wait for operation which is up to date. */
if ((from->flag & DEPSOP_FLAG_NEEDS_UPDATE) == 0) {
continue;
}
++node->num_links_pending;
}
}
}
void calculate_pending_parents(Depsgraph *graph)
{
for (OperationNode *node : graph->operations) {
calculate_pending_parents_for_node(node);
}
}
void initialize_execution(DepsgraphEvalState *state, Depsgraph *graph)
{
const bool do_stats = state->do_stats;
calculate_pending_parents(graph);
/* Clear tags and other things which needs to be clear. */
for (OperationNode *node : graph->operations) {
if (do_stats) {
node->stats.reset_current();
}
}
}
bool is_metaball_object_operation(const OperationNode *operation_node)
{
const ComponentNode *component_node = operation_node->owner;
const IDNode *id_node = component_node->owner;
if (GS(id_node->id_cow->name) != ID_OB) {
return false;
}
const Object *object = reinterpret_cast<const Object *>(id_node->id_cow);
return object->type == OB_MBALL;
}
bool need_evaluate_operation_at_stage(DepsgraphEvalState *state,
const OperationNode *operation_node)
{
const ComponentNode *component_node = operation_node->owner;
switch (state->stage) {
case EvaluationStage::COPY_ON_WRITE:
return (component_node->type == NodeType::COPY_ON_WRITE);
case EvaluationStage::THREADED_EVALUATION:
/* Sanity check: copy-on-write node should be evaluated already. This will be indicated by
* scheduled flag (we assume that scheduled operations have been actually handled by previous
* stage). */
BLI_assert(operation_node->scheduled || component_node->type != NodeType::COPY_ON_WRITE);
if (is_metaball_object_operation(operation_node)) {
state->need_single_thread_pass = true;
return false;
}
return true;
case EvaluationStage::SINGLE_THREADED_WORKAROUND:
return true;
}
BLI_assert(!"Unhandled evaluation stage, should never happen.");
return false;
}
/* Schedule a node if it needs evaluation.
* dec_parents: Decrement pending parents count, true when child nodes are
* scheduled after a task has been completed.
*/
template<typename ScheduleFunction, typename... ScheduleFunctionArgs>
void schedule_node(DepsgraphEvalState *state,
OperationNode *node,
bool dec_parents,
ScheduleFunction *schedule_function,
ScheduleFunctionArgs... schedule_function_args)
{
/* No need to schedule nodes of invisible ID. */
if (!check_operation_node_visible(node)) {
return;
}
/* No need to schedule operations which are not tagged for update, they are
* considered to be up to date. */
if ((node->flag & DEPSOP_FLAG_NEEDS_UPDATE) == 0) {
return;
}
/* TODO(sergey): This is not strictly speaking safe to read
* num_links_pending. */
if (dec_parents) {
BLI_assert(node->num_links_pending > 0);
atomic_sub_and_fetch_uint32(&node->num_links_pending, 1);
}
/* Cal not schedule operation while its dependencies are not yet
* evaluated. */
if (node->num_links_pending != 0) {
return;
}
/* During the COW stage only schedule COW nodes. */
if (!need_evaluate_operation_at_stage(state, node)) {
return;
}
/* Actually schedule the node. */
bool is_scheduled = atomic_fetch_and_or_uint8((uint8_t *)&node->scheduled, (uint8_t) true);
if (!is_scheduled) {
if (node->is_noop()) {
/* skip NOOP node, schedule children right away */
schedule_children(state, node, schedule_function, schedule_function_args...);
}
else {
/* children are scheduled once this task is completed */
schedule_function(node, 0, schedule_function_args...);
}
}
}
template<typename ScheduleFunction, typename... ScheduleFunctionArgs>
void schedule_graph(DepsgraphEvalState *state,
ScheduleFunction *schedule_function,
ScheduleFunctionArgs... schedule_function_args)
{
for (OperationNode *node : state->graph->operations) {
schedule_node(state, node, false, schedule_function, schedule_function_args...);
}
}
template<typename ScheduleFunction, typename... ScheduleFunctionArgs>
void schedule_children(DepsgraphEvalState *state,
OperationNode *node,
ScheduleFunction *schedule_function,
ScheduleFunctionArgs... schedule_function_args)
{
for (Relation *rel : node->outlinks) {
OperationNode *child = (OperationNode *)rel->to;
BLI_assert(child->type == NodeType::OPERATION);
if (child->scheduled) {
/* Happens when having cyclic dependencies. */
continue;
}
schedule_node(state,
child,
(rel->flag & RELATION_FLAG_CYCLIC) == 0,
schedule_function,
schedule_function_args...);
}
}
void schedule_node_to_queue(OperationNode *node,
const int /*thread_id*/,
GSQueue *evaluation_queue)
{
BLI_gsqueue_push(evaluation_queue, &node);
}
void evaluate_graph_single_threaded(DepsgraphEvalState *state)
{
GSQueue *evaluation_queue = BLI_gsqueue_new(sizeof(OperationNode *));
schedule_graph(state, schedule_node_to_queue, evaluation_queue);
while (!BLI_gsqueue_is_empty(evaluation_queue)) {
OperationNode *operation_node;
BLI_gsqueue_pop(evaluation_queue, &operation_node);
evaluate_node(state, operation_node);
schedule_children(state, operation_node, schedule_node_to_queue, evaluation_queue);
}
BLI_gsqueue_free(evaluation_queue);
}
void depsgraph_ensure_view_layer(Depsgraph *graph)
{
/* We update copy-on-write scene in the following cases:
* - It was not expanded yet.
* - It was tagged for update of CoW component.
* This allows us to have proper view layer pointer. */
Scene *scene_cow = graph->scene_cow;
if (deg_copy_on_write_is_expanded(&scene_cow->id) &&
(scene_cow->id.recalc & ID_RECALC_COPY_ON_WRITE) == 0) {
return;
}
const IDNode *scene_id_node = graph->find_id_node(&graph->scene->id);
deg_update_copy_on_write_datablock(graph, scene_id_node);
}
} // namespace
static TaskPool *deg_evaluate_task_pool_create(DepsgraphEvalState *state)
{
if (G.debug & G_DEBUG_DEPSGRAPH_NO_THREADS) {
return BLI_task_pool_create_no_threads(state);
}
else {
return BLI_task_pool_create_suspended(state, TASK_PRIORITY_HIGH);
}
}
/**
* Evaluate all nodes tagged for updating,
* \warning This is usually done as part of main loop, but may also be
* called from frame-change update.
*
* \note Time sources should be all valid!
*/
void deg_evaluate_on_refresh(Depsgraph *graph)
{
/* Nothing to update, early out. */
if (graph->entry_tags.is_empty()) {
return;
}
graph->debug.begin_graph_evaluation();
graph->is_evaluating = true;
depsgraph_ensure_view_layer(graph);
/* Set up evaluation state. */
DepsgraphEvalState state;
state.graph = graph;
state.do_stats = graph->debug.do_time_debug();
state.need_single_thread_pass = false;
/* Prepare all nodes for evaluation. */
initialize_execution(&state, graph);
/* Do actual evaluation now. */
/* First, process all Copy-On-Write nodes. */
state.stage = EvaluationStage::COPY_ON_WRITE;
TaskPool *task_pool = deg_evaluate_task_pool_create(&state);
schedule_graph(&state, schedule_node_to_pool, task_pool);
BLI_task_pool_work_and_wait(task_pool);
BLI_task_pool_free(task_pool);
/* After that, process all other nodes. */
state.stage = EvaluationStage::THREADED_EVALUATION;
task_pool = deg_evaluate_task_pool_create(&state);
schedule_graph(&state, schedule_node_to_pool, task_pool);
BLI_task_pool_work_and_wait(task_pool);
BLI_task_pool_free(task_pool);
if (state.need_single_thread_pass) {
state.stage = EvaluationStage::SINGLE_THREADED_WORKAROUND;
evaluate_graph_single_threaded(&state);
}
/* Finalize statistics gathering. This is because we only gather single
* operation timing here, without aggregating anything to avoid any extra
* synchronization. */
if (state.do_stats) {
deg_eval_stats_aggregate(graph);
}
/* Clear any uncleared tags - just in case. */
deg_graph_clear_tags(graph);
graph->is_evaluating = false;
graph->debug.end_graph_evaluation();
}
} // namespace DEG