Files
oak-editor/app/render/backend/renderbackend.cpp
T
itsmattkc 7f796bd99f revised video renderer's invalidate cache to use ranges rather than discrete
frames

Previously, when the video renderer received a dirty cache signal, it would
proceed to extract all frames from the range and queue them. However, this could
be extremely slow for long ranges since it had to iterate through the entire
range and calculate the individual frames it contained. Now, we use the same
range combining system as audio and automatically calculate the next frame
within the range only when necessary. Essentially the same work, but split up
over time and done only when needed leading to no discernible UI pause when
invalidating cache.
2020-01-03 15:50:43 +11:00

385 lines
7.9 KiB
C++

#include "renderbackend.h"
#include <QDateTime>
#include <QThread>
RenderBackend::RenderBackend(QObject *parent) :
QObject(parent),
compiled_(false),
started_(false),
viewer_node_(nullptr),
copied_viewer_node_(nullptr),
recompile_queued_(false),
input_update_queued_(false)
{
}
bool RenderBackend::Init()
{
if (started_) {
return true;
}
threads_.resize(QThread::idealThreadCount());
for (int i=0;i<threads_.size();i++) {
QThread* thread = new QThread(this);
threads_.replace(i, thread);
// We use low priority to keep the app responsive at all times (GUI thread should always prioritize over this one)
thread->start(QThread::LowPriority);
}
started_ = InitInternal();
// Connects workers and moves them to their respective threads
InitWorkers();
if (!started_) {
Close();
}
return started_;
}
void RenderBackend::Close()
{
if (!started_) {
return;
}
started_ = false;
Decompile();
CloseInternal();
for (int i=0;i<processors_.size();i++) {
// Invoke close and quit signals on processor and thred
QMetaObject::invokeMethod(processors_.at(i),
"Close",
Qt::QueuedConnection);
threads_.at(i)->quit();
}
for (int i=0;i<processors_.size();i++) {
threads_.at(i)->wait(); // FIXME: Maximum time in case a thread is stuck?
delete threads_.at(i);
delete processors_.at(i);
}
threads_.clear();
processors_.clear();
}
const QString &RenderBackend::GetError() const
{
return error_;
}
void RenderBackend::SetViewerNode(ViewerOutput *viewer_node)
{
if (viewer_node_ != nullptr) {
DisconnectViewer(viewer_node_);
Decompile();
}
viewer_node_ = viewer_node;
if (viewer_node_ != nullptr) {
ConnectViewer(viewer_node_);
RegenerateCacheID();
}
}
bool RenderBackend::IsInitiated()
{
return started_;
}
void RenderBackend::InvalidateCache(const rational &start_range, const rational &end_range)
{
if (!CanRender()) {
return;
}
// Adjust range to min/max values
rational start_range_adj = qMax(rational(0), start_range);
rational end_range_adj = qMin(GetSequenceLength(), end_range);
qDebug() << "Cache invalidated between"
<< start_range_adj.toDouble()
<< "and"
<< end_range_adj.toDouble();
// Add the range to the list
cache_queue_.InsertTimeRange(TimeRange(start_range_adj, end_range_adj));
// Queue value update
QueueValueUpdate();
CacheNext();
}
bool RenderBackend::Compile()
{
if (compiled_) {
return true;
}
// Get dependencies of viewer node
source_node_list_.append(viewer_node_);
source_node_list_.append(viewer_node_->GetDependencies());
// Copy all dependencies into graph
foreach (Node* n, source_node_list_) {
Node* copy = n->copy();
Node::CopyInputs(n, copy, false);
copied_graph_.AddNode(copy);
}
// We just copied the inputs, so if an input update is queued, it's unnecessary
input_update_queued_ = false;
// We know that the first node will be the viewer node since we appended that first in the copy
copied_viewer_node_ = static_cast<ViewerOutput*>(copied_graph_.nodes().first());
// Copy connections
Node::DuplicateConnectionsBetweenLists(source_node_list_, copied_graph_.nodes());
compiled_ = CompileInternal();
if (!compiled_) {
Decompile();
}
return compiled_;
}
void RenderBackend::Decompile()
{
if (!compiled_) {
return;
}
DecompileInternal();
copied_graph_.Clear();
copied_viewer_node_ = nullptr;
source_node_list_.clear();
compiled_ = false;
}
void RenderBackend::RegenerateCacheID()
{
QCryptographicHash hash(QCryptographicHash::Sha1);
if (!viewer_node_
|| !GenerateCacheIDInternal(hash)) {
cache_id_.clear();
CacheIDChangedEvent(QString());
return;
}
hash.addData(viewer_node_->uuid().toByteArray());
QByteArray bytes = hash.result();
cache_id_ = bytes.toHex();
CacheIDChangedEvent(cache_id_);
}
bool RenderBackend::CanRender()
{
return true;
}
TimeRange RenderBackend::PopNextFrameFromQueue()
{
return cache_queue_.takeFirst();
}
rational RenderBackend::GetSequenceLength()
{
if (viewer_node_ == nullptr) {
return 0;
}
return viewer_node_->Length();
}
void RenderBackend::SetError(const QString &error)
{
error_ = error;
}
void RenderBackend::ConnectViewer(ViewerOutput *node)
{
Q_UNUSED(node)
}
void RenderBackend::DisconnectViewer(ViewerOutput *node)
{
Q_UNUSED(node)
}
void RenderBackend::CacheNext()
{
if (!Init()
|| !ViewerIsConnected()
|| !CanRender()) {
return;
}
if (cache_queue_.isEmpty()) {
emit QueueComplete();
return;
}
if ((input_update_queued_ || recompile_queued_) && !AllProcessorsAreAvailable()) {
return;
}
if (recompile_queued_) {
Decompile();
recompile_queued_ = false;
}
if (!compiled_ && !Compile()) {
return;
}
if (input_update_queued_) {
for (int i=0;i<source_node_list_.size();i++) {
Node* src = source_node_list_.at(i);
Node* dst = copied_graph_.nodes().at(i);
Node::CopyInputs(src, dst, false);
}
input_update_queued_ = false;
}
Node* node_connected_to_viewer = GetDependentInput()->get_connected_node();
if (!node_connected_to_viewer) {
return;
}
foreach (RenderWorker* worker, processors_) {
if (cache_queue_.isEmpty()) {
break;
}
if (!WorkerIsBusy(worker)) {
TimeRange cache_frame = PopNextFrameFromQueue();
NodeDependency dep = NodeDependency(node_connected_to_viewer,
cache_frame);
// Timestamp this render job
qint64 job_time = QDateTime::currentMSecsSinceEpoch();
// Ensure the job's time is unique (since that's the whole point)
// NOTE: This value will be 0 if it doesn't exist, which will never be the result of currentMSecsSinceEpoch so we
// can safely assume 0 means it doesn't exist.
qint64 existing_job_time = render_job_info_.value(cache_frame);
if (existing_job_time == job_time) {
job_time = existing_job_time + 1;
}
render_job_info_.insert(cache_frame, job_time);
SetWorkerBusyState(worker, true);
QMetaObject::invokeMethod(worker,
"Render",
Qt::QueuedConnection,
Q_ARG(NodeDependency, dep),
Q_ARG(qint64, job_time));
}
}
}
ViewerOutput *RenderBackend::viewer_node() const
{
return copied_viewer_node_;
}
bool RenderBackend::ViewerIsConnected() const
{
return viewer_node_ != nullptr;
}
const QString &RenderBackend::cache_id() const
{
return cache_id_;
}
void RenderBackend::QueueValueUpdate()
{
input_update_queued_ = true;
}
bool RenderBackend::WorkerIsBusy(RenderWorker *worker) const
{
return processor_busy_state_.at(processors_.indexOf(worker));
}
void RenderBackend::SetWorkerBusyState(RenderWorker *worker, bool busy)
{
processor_busy_state_.replace(processors_.indexOf(worker), busy);
}
bool RenderBackend::AllProcessorsAreAvailable() const
{
foreach (bool busy, processor_busy_state_) {
if (busy) {
return false;
}
}
return true;
}
const QVector<QThread *> &RenderBackend::threads()
{
return threads_;
}
void RenderBackend::CacheIDChangedEvent(const QString &id)
{
Q_UNUSED(id)
}
void RenderBackend::InitWorkers()
{
for (int i=0;i<processors_.size();i++) {
RenderWorker* processor = processors_.at(i);
QThread* thread = threads().at(i);
// Connect to it
ConnectWorkerToThis(processor);
// Finally, we can move it to its own thread
processor->moveToThread(thread);
// This function blocks the main thread intentionally. See the documentation for this function to see why.
processor->Init();
}
processor_busy_state_.resize(processors_.size());
processor_busy_state_.fill(false);
}
void RenderBackend::QueueRecompile()
{
recompile_queued_ = true;
}