Files
test2/source/blender/blenkernel/intern/node_runtime.cc

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

433 lines
16 KiB
C++
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0-or-later */
#include "BKE_node.h"
#include "BKE_node_runtime.hh"
#include "DNA_node_types.h"
#include "BLI_function_ref.hh"
#include "BLI_stack.hh"
#include "BLI_task.hh"
#include "BLI_timeit.hh"
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
#include "NOD_geometry_nodes_lazy_function.hh"
namespace blender::bke::node_tree_runtime {
void preprocess_geometry_node_tree_for_evaluation(bNodeTree &tree_cow)
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
{
BLI_assert(tree_cow.type == NTREE_GEOMETRY);
/* Rebuild geometry nodes lazy function graph. */
tree_cow.runtime->geometry_nodes_lazy_function_graph_info.reset();
blender::nodes::ensure_geometry_nodes_lazy_function_graph(tree_cow);
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
}
static void double_checked_lock(std::mutex &mutex, bool &data_is_dirty, FunctionRef<void()> fn)
{
if (!data_is_dirty) {
return;
}
std::lock_guard lock{mutex};
if (!data_is_dirty) {
return;
}
fn();
data_is_dirty = false;
}
static void double_checked_lock_with_task_isolation(std::mutex &mutex,
bool &data_is_dirty,
FunctionRef<void()> fn)
{
double_checked_lock(mutex, data_is_dirty, [&]() { threading::isolate_task(fn); });
}
static void update_node_vector(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
tree_runtime.nodes.clear();
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
tree_runtime.group_nodes.clear();
tree_runtime.has_undefined_nodes_or_sockets = false;
LISTBASE_FOREACH (bNode *, node, &ntree.nodes) {
node->runtime->index_in_tree = tree_runtime.nodes.append_and_get_index(node);
node->runtime->owner_tree = const_cast<bNodeTree *>(&ntree);
tree_runtime.has_undefined_nodes_or_sockets |= node->typeinfo == &NodeTypeUndefined;
Geometry Nodes: new evaluation system This refactors the geometry nodes evaluation system. No changes for the user are expected. At a high level the goals are: * Support using geometry nodes outside of the geometry nodes modifier. * Support using the evaluator infrastructure for other purposes like field evaluation. * Support more nodes, especially when many of them are disabled behind switch nodes. * Support doing preprocessing on node groups. For more details see T98492. There are fairly detailed comments in the code, but here is a high level overview for how it works now: * There is a new "lazy-function" system. It is similar in spirit to the multi-function system but with different goals. Instead of optimizing throughput for highly parallelizable work, this system is designed to compute only the data that is actually necessary. What data is necessary can be determined dynamically during evaluation. Many lazy-functions can be composed in a graph to form a new lazy-function, which can again be used in a graph etc. * Each geometry node group is converted into a lazy-function graph prior to evaluation. To evaluate geometry nodes, one then just has to evaluate that graph. Node groups are no longer inlined into their parents. Next steps for the evaluation system is to reduce the use of threads in some situations to avoid overhead. Many small node groups don't benefit from multi-threading at all. This is much easier to do now because not everything has to be inlined in one huge node tree anymore. Differential Revision: https://developer.blender.org/D15914
2022-09-13 08:44:26 +02:00
if (node->is_group()) {
tree_runtime.group_nodes.append(node);
}
}
}
static void update_link_vector(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
tree_runtime.links.clear();
LISTBASE_FOREACH (bNodeLink *, link, &ntree.links) {
tree_runtime.links.append(link);
}
}
static void update_internal_links(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
for (bNode *node : tree_runtime.nodes) {
node->runtime->internal_links.clear();
for (bNodeSocket *socket : node->runtime->outputs) {
socket->runtime->internal_link_input = nullptr;
}
LISTBASE_FOREACH (bNodeLink *, link, &node->internal_links) {
node->runtime->internal_links.append(link);
link->tosock->runtime->internal_link_input = link->fromsock;
}
}
}
static void update_socket_vectors_and_owner_node(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
tree_runtime.sockets.clear();
tree_runtime.input_sockets.clear();
tree_runtime.output_sockets.clear();
for (bNode *node : tree_runtime.nodes) {
bNodeRuntime &node_runtime = *node->runtime;
node_runtime.inputs.clear();
node_runtime.outputs.clear();
LISTBASE_FOREACH (bNodeSocket *, socket, &node->inputs) {
socket->runtime->index_in_node = node_runtime.inputs.append_and_get_index(socket);
socket->runtime->index_in_all_sockets = tree_runtime.sockets.append_and_get_index(socket);
socket->runtime->index_in_inout_sockets = tree_runtime.input_sockets.append_and_get_index(
socket);
socket->runtime->owner_node = node;
tree_runtime.has_undefined_nodes_or_sockets |= socket->typeinfo == &NodeSocketTypeUndefined;
}
LISTBASE_FOREACH (bNodeSocket *, socket, &node->outputs) {
socket->runtime->index_in_node = node_runtime.outputs.append_and_get_index(socket);
socket->runtime->index_in_all_sockets = tree_runtime.sockets.append_and_get_index(socket);
socket->runtime->index_in_inout_sockets = tree_runtime.output_sockets.append_and_get_index(
socket);
socket->runtime->owner_node = node;
tree_runtime.has_undefined_nodes_or_sockets |= socket->typeinfo == &NodeSocketTypeUndefined;
}
}
}
static void update_directly_linked_links_and_sockets(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
for (bNode *node : tree_runtime.nodes) {
for (bNodeSocket *socket : node->runtime->inputs) {
socket->runtime->directly_linked_links.clear();
socket->runtime->directly_linked_sockets.clear();
}
for (bNodeSocket *socket : node->runtime->outputs) {
socket->runtime->directly_linked_links.clear();
socket->runtime->directly_linked_sockets.clear();
}
node->runtime->has_available_linked_inputs = false;
node->runtime->has_available_linked_outputs = false;
}
for (bNodeLink *link : tree_runtime.links) {
link->fromsock->runtime->directly_linked_links.append(link);
link->fromsock->runtime->directly_linked_sockets.append(link->tosock);
link->tosock->runtime->directly_linked_links.append(link);
if (link->is_available()) {
link->fromnode->runtime->has_available_linked_outputs = true;
link->tonode->runtime->has_available_linked_inputs = true;
}
}
for (bNodeSocket *socket : tree_runtime.input_sockets) {
if (socket->flag & SOCK_MULTI_INPUT) {
std::sort(socket->runtime->directly_linked_links.begin(),
socket->runtime->directly_linked_links.end(),
[&](const bNodeLink *a, const bNodeLink *b) {
return a->multi_input_socket_index > b->multi_input_socket_index;
});
}
}
for (bNodeSocket *socket : tree_runtime.input_sockets) {
for (bNodeLink *link : socket->runtime->directly_linked_links) {
/* Do this after sorting the input links. */
socket->runtime->directly_linked_sockets.append(link->fromsock);
}
}
}
static void find_logical_origins_for_socket_recursive(
bNodeSocket &input_socket,
bool only_follow_first_input_link,
Vector<bNodeSocket *, 16> &sockets_in_current_chain,
Vector<bNodeSocket *> &r_logical_origins,
Vector<bNodeSocket *> &r_skipped_origins)
{
if (sockets_in_current_chain.contains(&input_socket)) {
/* Protect against reroute recursions. */
return;
}
sockets_in_current_chain.append(&input_socket);
Span<bNodeLink *> links_to_check = input_socket.runtime->directly_linked_links;
if (only_follow_first_input_link) {
links_to_check = links_to_check.take_front(1);
}
for (bNodeLink *link : links_to_check) {
if (link->is_muted()) {
continue;
}
if (!link->is_available()) {
continue;
}
bNodeSocket &origin_socket = *link->fromsock;
bNode &origin_node = *link->fromnode;
if (!origin_socket.is_available()) {
/* Non available sockets are ignored. */
continue;
}
if (origin_node.type == NODE_REROUTE) {
bNodeSocket &reroute_input = *origin_node.runtime->inputs[0];
bNodeSocket &reroute_output = *origin_node.runtime->outputs[0];
r_skipped_origins.append(&reroute_input);
r_skipped_origins.append(&reroute_output);
find_logical_origins_for_socket_recursive(
reroute_input, false, sockets_in_current_chain, r_logical_origins, r_skipped_origins);
continue;
}
if (origin_node.is_muted()) {
if (bNodeSocket *mute_input = origin_socket.runtime->internal_link_input) {
r_skipped_origins.append(&origin_socket);
r_skipped_origins.append(mute_input);
find_logical_origins_for_socket_recursive(
*mute_input, true, sockets_in_current_chain, r_logical_origins, r_skipped_origins);
}
continue;
}
r_logical_origins.append(&origin_socket);
}
sockets_in_current_chain.pop_last();
}
static void update_logical_origins(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
threading::parallel_for(tree_runtime.nodes.index_range(), 128, [&](const IndexRange range) {
for (const int i : range) {
bNode &node = *tree_runtime.nodes[i];
for (bNodeSocket *socket : node.runtime->inputs) {
Vector<bNodeSocket *, 16> sockets_in_current_chain;
socket->runtime->logically_linked_sockets.clear();
socket->runtime->logically_linked_skipped_sockets.clear();
find_logical_origins_for_socket_recursive(
*socket,
false,
sockets_in_current_chain,
socket->runtime->logically_linked_sockets,
socket->runtime->logically_linked_skipped_sockets);
}
}
});
}
static void update_nodes_by_type(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
tree_runtime.nodes_by_type.clear();
for (bNode *node : tree_runtime.nodes) {
tree_runtime.nodes_by_type.add(node->typeinfo, node);
}
}
static void update_sockets_by_identifier(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
threading::parallel_for(tree_runtime.nodes.index_range(), 128, [&](const IndexRange range) {
for (bNode *node : tree_runtime.nodes.as_span().slice(range)) {
node->runtime->inputs_by_identifier.clear();
node->runtime->outputs_by_identifier.clear();
for (bNodeSocket *socket : node->runtime->inputs) {
node->runtime->inputs_by_identifier.add_new(socket->identifier, socket);
}
for (bNodeSocket *socket : node->runtime->outputs) {
node->runtime->outputs_by_identifier.add_new(socket->identifier, socket);
}
}
});
}
enum class ToposortDirection {
LeftToRight,
RightToLeft,
};
struct ToposortNodeState {
bool is_done = false;
bool is_in_stack = false;
};
static void toposort_from_start_node(const ToposortDirection direction,
bNode &start_node,
MutableSpan<ToposortNodeState> node_states,
Vector<bNode *> &r_sorted_nodes,
bool &r_cycle_detected)
{
struct Item {
bNode *node;
int socket_index = 0;
int link_index = 0;
};
Stack<Item, 64> nodes_to_check;
nodes_to_check.push({&start_node});
2022-09-08 12:53:43 +02:00
node_states[start_node.runtime->index_in_tree].is_in_stack = true;
while (!nodes_to_check.is_empty()) {
Item &item = nodes_to_check.peek();
bNode &node = *item.node;
const Span<bNodeSocket *> sockets = (direction == ToposortDirection::LeftToRight) ?
node.runtime->inputs :
node.runtime->outputs;
while (true) {
if (item.socket_index == sockets.size()) {
/* All sockets have already been visited. */
break;
}
bNodeSocket &socket = *sockets[item.socket_index];
const Span<bNodeLink *> linked_links = socket.runtime->directly_linked_links;
if (item.link_index == linked_links.size()) {
/* All links connected to this socket have already been visited. */
item.socket_index++;
item.link_index = 0;
continue;
}
bNodeLink &link = *linked_links[item.link_index];
if (!link.is_available()) {
/* Ignore unavailable links. */
item.link_index++;
continue;
}
bNodeSocket &linked_socket = *socket.runtime->directly_linked_sockets[item.link_index];
bNode &linked_node = *linked_socket.runtime->owner_node;
ToposortNodeState &linked_node_state = node_states[linked_node.runtime->index_in_tree];
if (linked_node_state.is_done) {
/* The linked node has already been visited. */
item.link_index++;
continue;
}
if (linked_node_state.is_in_stack) {
r_cycle_detected = true;
}
else {
nodes_to_check.push({&linked_node});
linked_node_state.is_in_stack = true;
}
break;
}
/* If no other element has been pushed, the current node can be pushed to the sorted list. */
if (&item == &nodes_to_check.peek()) {
ToposortNodeState &node_state = node_states[node.runtime->index_in_tree];
node_state.is_done = true;
node_state.is_in_stack = false;
r_sorted_nodes.append(&node);
nodes_to_check.pop();
}
}
}
static void update_toposort(const bNodeTree &ntree,
const ToposortDirection direction,
Vector<bNode *> &r_sorted_nodes,
bool &r_cycle_detected)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
r_sorted_nodes.clear();
r_sorted_nodes.reserve(tree_runtime.nodes.size());
r_cycle_detected = false;
Array<ToposortNodeState> node_states(tree_runtime.nodes.size());
for (bNode *node : tree_runtime.nodes) {
if (node_states[node->runtime->index_in_tree].is_done) {
/* Ignore nodes that are done already. */
continue;
}
if ((direction == ToposortDirection::LeftToRight) ?
node->runtime->has_available_linked_outputs :
node->runtime->has_available_linked_inputs) {
/* Ignore non-start nodes. */
continue;
}
toposort_from_start_node(direction, *node, node_states, r_sorted_nodes, r_cycle_detected);
}
if (r_sorted_nodes.size() < tree_runtime.nodes.size()) {
r_cycle_detected = true;
for (bNode *node : tree_runtime.nodes) {
if (node_states[node->runtime->index_in_tree].is_done) {
/* Ignore nodes that are done already. */
continue;
}
/* Start toposort at this node which is somewhere in the middle of a loop. */
toposort_from_start_node(direction, *node, node_states, r_sorted_nodes, r_cycle_detected);
}
}
BLI_assert(tree_runtime.nodes.size() == r_sorted_nodes.size());
}
static void update_group_output_node(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
const bNodeType *node_type = nodeTypeFind("NodeGroupOutput");
const Span<bNode *> group_output_nodes = tree_runtime.nodes_by_type.lookup(node_type);
if (group_output_nodes.is_empty()) {
tree_runtime.group_output_node = nullptr;
}
else if (group_output_nodes.size() == 1) {
tree_runtime.group_output_node = group_output_nodes[0];
}
else {
for (bNode *group_output : group_output_nodes) {
if (group_output->flag & NODE_DO_OUTPUT) {
tree_runtime.group_output_node = group_output;
break;
}
}
}
}
static void ensure_topology_cache(const bNodeTree &ntree)
{
bNodeTreeRuntime &tree_runtime = *ntree.runtime;
double_checked_lock_with_task_isolation(
tree_runtime.topology_cache_mutex, tree_runtime.topology_cache_is_dirty, [&]() {
update_node_vector(ntree);
update_link_vector(ntree);
update_socket_vectors_and_owner_node(ntree);
update_internal_links(ntree);
update_directly_linked_links_and_sockets(ntree);
threading::parallel_invoke([&]() { update_logical_origins(ntree); },
[&]() { update_nodes_by_type(ntree); },
[&]() { update_sockets_by_identifier(ntree); },
[&]() {
update_toposort(ntree,
ToposortDirection::LeftToRight,
tree_runtime.toposort_left_to_right,
tree_runtime.has_available_link_cycle);
},
[&]() {
bool dummy;
update_toposort(ntree,
ToposortDirection::RightToLeft,
tree_runtime.toposort_right_to_left,
dummy);
});
update_group_output_node(ntree);
tree_runtime.topology_cache_exists = true;
});
}
} // namespace blender::bke::node_tree_runtime
void bNodeTree::ensure_topology_cache() const
{
blender::bke::node_tree_runtime::ensure_topology_cache(*this);
}