nodes: minor overhaul to functionality
The nodes now have more control over how their accelerated shaders/sample functions are run, as well as how items are popped off the value tables. This allows for various optimizations that we didn't have access to before.
This commit is contained in:
itsmattkc
+114
-151
@@ -47,8 +47,19 @@ void RenderWorker::RenderFrame(RenderTicketPtr ticket, ViewerOutput* viewer, con
|
||||
|
||||
QVariant texture = table.Get(NodeParam::kTexture);
|
||||
|
||||
PixelFormat::Format output_format;
|
||||
if (!texture.isNull() && TextureHasAlpha(texture)) {
|
||||
output_format = PixelFormat::GetFormatWithAlphaChannel(video_params_.format());
|
||||
} else {
|
||||
output_format = PixelFormat::GetFormatWithoutAlphaChannel(video_params_.format());
|
||||
}
|
||||
|
||||
FramePtr frame = Frame::Create();
|
||||
frame->set_video_params(video_params_);
|
||||
frame->set_video_params(VideoParams(video_params_.width(),
|
||||
video_params_.height(),
|
||||
video_params_.time_base(),
|
||||
output_format,
|
||||
video_params_.divider()));
|
||||
frame->set_timestamp(time);
|
||||
frame->allocate();
|
||||
|
||||
@@ -172,10 +183,11 @@ NodeValueTable RenderWorker::GenerateBlockTable(const TrackOutput *track, const
|
||||
}
|
||||
}
|
||||
|
||||
QVariant RenderWorker::ProcessSamples(const Node *node, const TimeRange &range, NodeValueDatabase &input_params_in)
|
||||
QVariant RenderWorker::ProcessSamples(const Node *node, const TimeRange &range, const SampleJob& job)
|
||||
{
|
||||
// Copy database so we can make some temporary modifications to it
|
||||
NodeValueDatabase input_params = input_params_in;
|
||||
// FIX THIS CODE:
|
||||
return QVariant();
|
||||
/*
|
||||
NodeInput* sample_input = node->ProcessesSamplesFrom(input_params);
|
||||
|
||||
// Try to find the sample buffer in the table
|
||||
@@ -225,134 +237,20 @@ QVariant RenderWorker::ProcessSamples(const Node *node, const TimeRange &range,
|
||||
}
|
||||
|
||||
return QVariant::fromValue(output_buffer);
|
||||
*/
|
||||
}
|
||||
|
||||
void RenderWorker::ProcessNodeEvent(const Node *node, const TimeRange &range, NodeValueDatabase &input_params_in, NodeValueTable &output_params)
|
||||
QVariant RenderWorker::GetCachedFrame(const Node* node, const rational& time)
|
||||
{
|
||||
// Convert footage to image/sample buffers
|
||||
if (node->id() == QStringLiteral("org.olivevideoeditor.Olive.videoinput")) {
|
||||
QByteArray hash = RenderBackend::HashNode(node, video_params(), range.in());
|
||||
QByteArray hash = RenderBackend::HashNode(node, video_params(), time);
|
||||
|
||||
QString fn = FrameHashCache::CachePathName(hash);
|
||||
|
||||
if (QFileInfo::exists(fn)) {
|
||||
FramePtr f = FrameHashCache::LoadCacheFrame(hash);
|
||||
|
||||
QVariant cached = CachedFrameToTexture(f);
|
||||
|
||||
if (!cached.isNull()) {
|
||||
output_params.Push(NodeParam::kTexture, cached, node);
|
||||
|
||||
// No more to do here
|
||||
return;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
QList<NodeInput*> inputs = node->GetInputsIncludingArrays();
|
||||
foreach (NodeInput* input, inputs) {
|
||||
if (input->data_type() == NodeParam::kFootage) {
|
||||
TimeRange input_time = node->InputTimeAdjustment(input, range);
|
||||
|
||||
StreamPtr stream = ResolveStreamFromInput(input);
|
||||
|
||||
if (stream) {
|
||||
QVariant v = ProcessFootage(stream, input_time);
|
||||
|
||||
if (!v.isNull()) {
|
||||
if (stream->type() == Stream::kVideo || stream->type() == Stream::kImage) {
|
||||
output_params.Push(NodeParam::kTexture, v, node);
|
||||
} else if (stream->type() == Stream::kAudio) {
|
||||
output_params.Push(NodeParam::kSamples, v, node);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Check if node has a shader
|
||||
if (node->GetCapabilities(input_params_in) & Node::kShader) {
|
||||
QVariant v = ProcessShader(node, range, input_params_in);
|
||||
|
||||
if (!v.isNull()) {
|
||||
output_params.Push(NodeParam::kTexture, v, node);
|
||||
}
|
||||
}
|
||||
|
||||
// Check if node processes samples
|
||||
if (node->GetCapabilities(input_params_in) & Node::kSampleProcessor) {
|
||||
QVariant v = ProcessSamples(node, range, input_params_in);
|
||||
|
||||
if (!v.isNull()) {
|
||||
output_params.Push(NodeParam::kSamples, v, node);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
QVariant RenderWorker::GetDataFromStream(StreamPtr stream, const TimeRange &input_time)
|
||||
{
|
||||
DecoderPtr decoder = ResolveDecoderFromInput(stream);
|
||||
|
||||
if (decoder) {
|
||||
if (stream->type() == Stream::kVideo || stream->type() == Stream::kImage) {
|
||||
|
||||
FramePtr frame = decoder->RetrieveVideo(input_time.in(),
|
||||
video_params().divider());
|
||||
|
||||
if (frame) {
|
||||
// Return a texture from the derived class
|
||||
return FootageFrameToTexture(stream, frame);
|
||||
}
|
||||
|
||||
} else if (stream->type() == Stream::kAudio) {
|
||||
|
||||
// See if we have a conformed version of this audio
|
||||
if (!decoder->HasConformedVersion(audio_params())) {
|
||||
|
||||
// If not, check what audio mode we're in
|
||||
if (audio_mode_is_preview_) {
|
||||
|
||||
// For preview, we report the conform is missing and finish the render without it
|
||||
// temporarily. The backend that picks up this signal will recache this section once the
|
||||
// conform is available.
|
||||
emit AudioConformUnavailable(decoder->stream(),
|
||||
audio_render_time_,
|
||||
input_time.out(),
|
||||
audio_params());
|
||||
|
||||
} else {
|
||||
|
||||
// For online rendering/export, it's a waste of time to render the audio until we have
|
||||
// all we need, so we try to handle the conform ourselves
|
||||
AudioStreamPtr as = std::static_pointer_cast<AudioStream>(stream);
|
||||
|
||||
// Check if any other threads are conforming this audio
|
||||
if (as->try_start_conforming(audio_params())) {
|
||||
|
||||
// If not, conform it ourselves
|
||||
decoder->ConformAudio(&IsCancelled(), audio_params());
|
||||
|
||||
} else {
|
||||
|
||||
// If another thread is conforming already, hackily try to wait until it's done.
|
||||
do {
|
||||
QThread::msleep(1000);
|
||||
} while (!as->has_conformed_version(audio_params()) && !IsCancelled());
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
|
||||
if (decoder->HasConformedVersion(audio_params())) {
|
||||
SampleBufferPtr frame = decoder->RetrieveAudio(input_time.in(), input_time.length(),
|
||||
audio_params());
|
||||
|
||||
if (frame) {
|
||||
return QVariant::fromValue(frame);
|
||||
}
|
||||
}
|
||||
return CachedFrameToTexture(f);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -382,52 +280,117 @@ DecoderPtr RenderWorker::ResolveDecoderFromInput(StreamPtr stream)
|
||||
return decoder;
|
||||
}
|
||||
|
||||
QVariant RenderWorker::ProcessFootage(StreamPtr stream, const TimeRange &input_time)
|
||||
QVariant RenderWorker::ProcessVideoFootage(StreamPtr stream, const rational &input_time)
|
||||
{
|
||||
if (stream->type() == Stream::kVideo || stream->type() == Stream::kImage) {
|
||||
ImageStreamPtr video_stream = std::static_pointer_cast<ImageStream>(stream);
|
||||
rational time_match = (stream->type() == Stream::kImage) ? rational() : input_time;
|
||||
QString colorspace_match = video_stream->get_colorspace_match_string();
|
||||
|
||||
ImageStreamPtr video_stream = std::static_pointer_cast<ImageStream>(stream);
|
||||
rational time_match = (stream->type() == Stream::kImage) ? rational() : input_time.in();
|
||||
QString colorspace_match = video_stream->get_colorspace_match_string();
|
||||
QVariant value;
|
||||
bool found_cache = false;
|
||||
|
||||
QVariant value;
|
||||
bool found_cache = false;
|
||||
if (still_image_cache_.contains(stream.get())) {
|
||||
const CachedStill& cs = still_image_cache_[stream.get()];
|
||||
|
||||
if (still_image_cache_.contains(stream.get())) {
|
||||
const CachedStill& cs = still_image_cache_[stream.get()];
|
||||
if (cs.colorspace == colorspace_match
|
||||
&& cs.alpha_is_associated == video_stream->premultiplied_alpha()
|
||||
&& cs.divider == video_params_.divider()
|
||||
&& cs.time == time_match) {
|
||||
value = cs.texture;
|
||||
found_cache = true;
|
||||
} else {
|
||||
still_image_cache_.remove(stream.get());
|
||||
}
|
||||
}
|
||||
|
||||
if (cs.colorspace == colorspace_match
|
||||
&& cs.alpha_is_associated == video_stream->premultiplied_alpha()
|
||||
&& cs.divider == video_params_.divider()
|
||||
&& cs.time == time_match) {
|
||||
value = cs.texture;
|
||||
found_cache = true;
|
||||
if (!found_cache) {
|
||||
|
||||
DecoderPtr decoder = ResolveDecoderFromInput(stream);
|
||||
|
||||
if (decoder) {
|
||||
FramePtr frame = decoder->RetrieveVideo(input_time,
|
||||
video_params().divider());
|
||||
|
||||
if (frame) {
|
||||
// Return a texture from the derived class
|
||||
value = FootageFrameToTexture(stream, frame);
|
||||
|
||||
if (value.isNull()) {
|
||||
qDebug() << "Texture from derivative was blank";
|
||||
} else {
|
||||
// Put this into the image cache instead
|
||||
still_image_cache_.insert(stream.get(), {value,
|
||||
colorspace_match,
|
||||
video_stream->premultiplied_alpha(),
|
||||
video_params_.divider(),
|
||||
time_match});
|
||||
}
|
||||
} else {
|
||||
still_image_cache_.remove(stream.get());
|
||||
qDebug() << "Frame from decoder was blank";
|
||||
}
|
||||
}
|
||||
|
||||
if (!found_cache) {
|
||||
}
|
||||
|
||||
value = GetDataFromStream(stream, input_time);
|
||||
return value;
|
||||
}
|
||||
|
||||
still_image_cache_.insert(stream.get(), {value,
|
||||
colorspace_match,
|
||||
video_stream->premultiplied_alpha(),
|
||||
video_params_.divider(),
|
||||
time_match});
|
||||
QVariant RenderWorker::ProcessAudioFootage(StreamPtr stream, const TimeRange &input_time)
|
||||
{
|
||||
QVariant value;
|
||||
|
||||
DecoderPtr decoder = ResolveDecoderFromInput(stream);
|
||||
|
||||
if (decoder) {
|
||||
// See if we have a conformed version of this audio
|
||||
if (!decoder->HasConformedVersion(audio_params())) {
|
||||
|
||||
// If not, check what audio mode we're in
|
||||
if (audio_mode_is_preview_) {
|
||||
|
||||
// For preview, we report the conform is missing and finish the render without it
|
||||
// temporarily. The backend that picks up this signal will recache this section once the
|
||||
// conform is available.
|
||||
emit AudioConformUnavailable(decoder->stream(),
|
||||
audio_render_time_,
|
||||
input_time.out(),
|
||||
audio_params());
|
||||
|
||||
} else {
|
||||
|
||||
// For online rendering/export, it's a waste of time to render the audio until we have
|
||||
// all we need, so we try to handle the conform ourselves
|
||||
AudioStreamPtr as = std::static_pointer_cast<AudioStream>(stream);
|
||||
|
||||
// Check if any other threads are conforming this audio
|
||||
if (as->try_start_conforming(audio_params())) {
|
||||
|
||||
// If not, conform it ourselves
|
||||
decoder->ConformAudio(&IsCancelled(), audio_params());
|
||||
|
||||
} else {
|
||||
|
||||
// If another thread is conforming already, hackily try to wait until it's done.
|
||||
do {
|
||||
QThread::msleep(1000);
|
||||
} while (!as->has_conformed_version(audio_params()) && !IsCancelled());
|
||||
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
return value;
|
||||
|
||||
} else if (stream->type() == Stream::kAudio) {
|
||||
|
||||
return GetDataFromStream(stream, input_time);
|
||||
if (decoder->HasConformedVersion(audio_params())) {
|
||||
SampleBufferPtr frame = decoder->RetrieveAudio(input_time.in(), input_time.length(),
|
||||
audio_params());
|
||||
|
||||
if (frame) {
|
||||
value = QVariant::fromValue(frame);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return QVariant();
|
||||
return value;
|
||||
}
|
||||
|
||||
OLIVE_NAMESPACE_EXIT
|
||||
|
||||
Reference in New Issue
Block a user