oak-editor/app/render/renderprocessor.cpp

/***

  Olive - Non-Linear Video Editor
  Copyright (C) 2021 Olive Team

  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 3 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, see <http://www.gnu.org/licenses/>.

***/

#include "renderprocessor.h"

#include <QOpenGLContext>
#include <QVector2D>
#include <QVector3D>
#include <QVector4D>

#include "audio/packedprocessor.h"
#include "audio/planarprocessor.h"
#include "audio/tempoprocessor.h"
#include "node/block/clip/clip.h"
#include "node/block/transition/transition.h"
#include "node/project/project.h"
#include "rendermanager.h"

namespace olive {

#define super NodeTraverser

RenderProcessor::RenderProcessor(RenderTicketPtr ticket, Renderer *render_ctx, DecoderCache* decoder_cache, ShaderCache *shader_cache, QVariant default_shader) :
  ticket_(ticket),
  render_ctx_(render_ctx),
  decoder_cache_(decoder_cache),
  shader_cache_(shader_cache),
  default_shader_(default_shader)
{
}

TexturePtr RenderProcessor::GenerateTexture(const rational &time, const rational &frame_length)
{
  ViewerOutput* viewer = Node::ValueToPtr<ViewerOutput>(ticket_->property("viewer"));

  NodeValueTable table;
  if (Node *texture_output = viewer->GetConnectedTextureOutput()) {
    table = GenerateTable(texture_output, viewer->GetValueHintForInput(ViewerOutput::kTextureInput), TimeRange(time, time + frame_length));
  }

  return table.Get(NodeValue::kTexture).value<TexturePtr>();
}

FramePtr RenderProcessor::GenerateFrame(TexturePtr texture, const rational& time)
{
  // Set up output frame parameters
  VideoParams frame_params = GetCacheVideoParams();

  QSize frame_size = ticket_->property("size").value<QSize>();
  if (!frame_size.isNull()) {
    frame_params.set_width(frame_size.width());
    frame_params.set_height(frame_size.height());
  }

  VideoParams::Format frame_format = static_cast<VideoParams::Format>(ticket_->property("format").toInt());
  if (frame_format != VideoParams::kFormatInvalid) {
    frame_params.set_format(frame_format);
  }

  frame_params.set_channel_count(texture ? texture->channel_count() : VideoParams::kRGBChannelCount);

  FramePtr frame = Frame::Create();
  frame->set_timestamp(time);
  frame->set_video_params(frame_params);
  frame->allocate();

  if (!texture) {
    // Blank frame out
    memset(frame->data(), 0, frame->allocated_size());
  } else {
    // Dump texture contents to frame
    ColorProcessorPtr output_color_transform = ticket_->property("coloroutput").value<ColorProcessorPtr>();
    const VideoParams& tex_params = texture->params();

    if (tex_params.effective_width() != frame_params.effective_width()
        || tex_params.effective_height() != frame_params.effective_height()
        || tex_params.format() != frame_params.format()
        || output_color_transform) {
      TexturePtr blit_tex = render_ctx_->CreateTexture(frame_params);

      QMatrix4x4 matrix = ticket_->property("matrix").value<QMatrix4x4>();

      if (output_color_transform) {
        // Yes color transform, blit color managed
        render_ctx_->BlitColorManaged(output_color_transform, texture,
                                      Config::Current()[QStringLiteral("ReassocLinToNonLin")].toBool() ? Renderer::kAlphaAssociated : Renderer::kAlphaNone,
                                      blit_tex.get(), true, matrix);
      } else {
        // No color transform, just blit
        ShaderJob job;
        job.InsertValue(QStringLiteral("ove_maintex"), NodeValue(NodeValue::kTexture, QVariant::fromValue(texture)));
        job.InsertValue(QStringLiteral("ove_mvpmat"), NodeValue(NodeValue::kMatrix, matrix));

        render_ctx_->BlitToTexture(default_shader_, job, blit_tex.get());
      }

      // Replace texture that we're going to download in the next step
      texture = blit_tex;
    }

    render_ctx_->DownloadFromTexture(texture.get(), frame->data(), frame->linesize_pixels());
  }

  return frame;
}

void RenderProcessor::Run()
{
  // Depending on the render ticket type, start a job
  RenderManager::TicketType type = ticket_->property("type").value<RenderManager::TicketType>();

  SetCancelPointer(&ticket_->IsCancelled());

  switch (type) {
  case RenderManager::kTypeVideo:
  {
    SetCacheVideoParams(ticket_->property("vparam").value<VideoParams>());
    rational time = ticket_->property("time").value<rational>();

    rational frame_length = GetCacheVideoParams().frame_rate_as_time_base();
    if (GetCacheVideoParams().interlacing() != VideoParams::kInterlaceNone) {
      frame_length /= 2;
    }

    TexturePtr texture = GenerateTexture(time, frame_length);

    if (GetCacheVideoParams().interlacing() != VideoParams::kInterlaceNone) {
      // Get next between frame and interlace it
      TexturePtr top = texture;
      TexturePtr bottom = GenerateTexture(time + frame_length, frame_length);

      if (GetCacheVideoParams().interlacing() == VideoParams::kInterlacedBottomFirst) {
        std::swap(top, bottom);
      }

      texture = render_ctx_->InterlaceTexture(top, bottom, GetCacheVideoParams());
    }

    if (ticket_->IsCancelled()) {
      // Finish cancelled ticket with nothing since we can't guarantee the frame we generated
      // is actually "complete
      ticket_->Finish();
    } else if (ticket_->property("textureonly").toBool()) {
      // Return GPU texture
      if (!texture) {
        texture = render_ctx_->CreateTexture(GetCacheVideoParams());
        render_ctx_->ClearDestination(texture.get());
      }

      render_ctx_->Flush();

      ticket_->Finish(QVariant::fromValue(texture));
    } else {
      // Convert to CPU frame
      FramePtr frame = GenerateFrame(texture, time);

      ticket_->Finish(QVariant::fromValue(frame));
    }
    break;
  }
  case RenderManager::kTypeAudio:
  {
    ViewerOutput* viewer = Node::ValueToPtr<ViewerOutput>(ticket_->property("viewer"));
    TimeRange time = ticket_->property("time").value<TimeRange>();

    NodeValueTable table;
    if (Node *texture_output = viewer->GetConnectedSampleOutput()) {
      table = GenerateTable(texture_output, viewer->GetValueHintForInput(ViewerOutput::kSamplesInput),time);
    }

    QVariant sample_variant = table.Get(NodeValue::kSamples);
    SampleBufferPtr samples = sample_variant.value<SampleBufferPtr>();
    if (samples && ticket_->property("enablewaveforms").toBool()) {
      AudioVisualWaveform vis;
      vis.set_channel_count(samples->audio_params().channel_count());
      vis.OverwriteSamples(samples, samples->audio_params().sample_rate());
      ticket_->setProperty("waveform", QVariant::fromValue(vis));
    }

    if (ticket_->IsCancelled()) {
      ticket_->Finish();
    } else {
      ticket_->Finish(sample_variant);
    }
    break;
  }
  case RenderManager::kTypeVideoDownload:
  {
    QString cache = ticket_->property("cache").toString();
    FramePtr frame = ticket_->property("frame").value<FramePtr>();
    QByteArray hash = ticket_->property("hash").toByteArray();

    ticket_->Finish(FrameHashCache::SaveCacheFrame(cache, hash, frame));
    break;
  }
  default:
    // Fail
    ticket_->Finish();
  }
}

DecoderPtr RenderProcessor::ResolveDecoderFromInput(const QString& decoder_id, const Decoder::CodecStream &stream)
{
  if (!stream.IsValid()) {
    qWarning() << "Attempted to resolve the decoder of a null stream";
    return nullptr;
  }

  QMutexLocker locker(decoder_cache_->mutex());

  DecoderPair decoder = decoder_cache_->value(stream);

  qint64 file_last_modified = QFileInfo(stream.filename()).lastModified().toMSecsSinceEpoch();

  if (!decoder.decoder || decoder.last_modified != file_last_modified) {
    // No decoder
    decoder.decoder = Decoder::CreateFromID(decoder_id);
    decoder.last_modified = file_last_modified;

    if (decoder.decoder->Open(stream)) {
      decoder_cache_->insert(stream, decoder);
    } else {
      qWarning() << "Failed to open decoder for" << stream.filename()
                 << "::" << stream.stream();
      return nullptr;
    }
  }

  return decoder.decoder;
}

void RenderProcessor::Process(RenderTicketPtr ticket, Renderer *render_ctx, DecoderCache *decoder_cache, ShaderCache *shader_cache, QVariant default_shader)
{
  RenderProcessor p(ticket, render_ctx, decoder_cache, shader_cache, default_shader);
  p.Run();
}

NodeValueTable RenderProcessor::GenerateBlockTable(const Track *track, const TimeRange &range)
{
  if (track->type() == Track::kAudio) {

    const AudioParams& audio_params = ticket_->property("aparam").value<AudioParams>();

    QVector<Block*> active_blocks = track->BlocksAtTimeRange(range);

    // All these blocks will need to output to a buffer so we create one here
    SampleBufferPtr block_range_buffer = SampleBuffer::CreateAllocated(audio_params,
                                                                       audio_params.time_to_samples(range.length()));
    block_range_buffer->silence();

    NodeValueTable merged_table;

    // Loop through active blocks retrieving their audio
    foreach (Block* b, active_blocks) {
      if (dynamic_cast<ClipBlock*>(b) || dynamic_cast<TransitionBlock*>(b)) {
        TimeRange range_for_block(qMax(b->in(), range.in()),
                                  qMin(b->out(), range.out()));

        int destination_offset = audio_params.time_to_samples(range_for_block.in() - range.in());
        int max_dest_sz = audio_params.time_to_samples(range_for_block.length());

        // Destination buffer
        NodeValueTable table = GenerateTable(b, track->GetValueHintForInput(Track::kBlockInput, track->GetArrayIndexFromBlock(b)),Track::TransformRangeForBlock(b, range_for_block));
        SampleBufferPtr samples_from_this_block = table.Take(NodeValue::kSamples).value<SampleBufferPtr>();
        ClipBlock *clip_cast = dynamic_cast<ClipBlock*>(b);

        if (samples_from_this_block) {
          // If this is a clip, we might have extra speed/reverse information
          if (clip_cast) {
            double speed_value = clip_cast->speed();
            bool reversed = clip_cast->reverse();

            if (qIsNull(speed_value)) {
              // Just silence, don't think there's any other practical application of 0 speed audio
              samples_from_this_block->silence();
            } else if (!qFuzzyCompare(speed_value, 1.0)) {
              if (clip_cast->maintain_audio_pitch()) {
                PackedProcessor packer;
                packer.Open(samples_from_this_block->audio_params());

                QByteArray packed = packer.Convert(samples_from_this_block);

                if (!packed.isEmpty()) {
                  TempoProcessor tp;
                  tp.Open(samples_from_this_block->audio_params(), speed_value);

                  // FIXME: This is not the best way to do this, the TempoProcessor works best
                  //        when it's given a continuous stream of audio, which is challenging
                  //        in our current "modular" audio system. This should still work reasonably
                  //        well on export (assuming audio is all generated at once on export), but
                  //        users may hear clicks and pops in the audio during preview due to this
                  //        approach.
                  tp.Push(packed);
                  tp.Flush();
                  packed = tp.Pull();
                  tp.Close();

                  if (!packed.isEmpty()) {
                    PlanarProcessor planar;
                    planar.Open(samples_from_this_block->audio_params());
                    samples_from_this_block = planar.Convert(packed);
                  }
                }
              } else {
                // Multiply time
                samples_from_this_block->speed(speed_value);
              }
            }

            if (reversed) {
              samples_from_this_block->reverse();
            }
          }

          int copy_length = qMin(max_dest_sz, samples_from_this_block->sample_count());

          // Copy samples into destination buffer
          for (int i=0; i<samples_from_this_block->audio_params().channel_count(); i++) {
            block_range_buffer->set(i, samples_from_this_block->data(i), destination_offset, copy_length);
          }

          NodeValueTable::Merge({merged_table, table});
        }

        // Create block waveforms if requested
        if (ticket_->property("enablewaveforms").toBool() && clip_cast) {
          // Format information for use in the main thread
          RenderedWaveform waveform_info;
          waveform_info.block = clip_cast;
          waveform_info.range = range_for_block - b->in();

          if (!(waveform_info.silence = !samples_from_this_block.get())) {
            // Generate a visual waveform from the samples acquired from this block
            AudioVisualWaveform visual_waveform;
            visual_waveform.set_channel_count(audio_params.channel_count());
            visual_waveform.OverwriteSamples(samples_from_this_block, audio_params.sample_rate());
            waveform_info.waveform = visual_waveform;
          }

          QVector<RenderedWaveform> waveform_list = ticket_->property("waveforms").value< QVector<RenderedWaveform> >();
          waveform_list.append(waveform_info);
          ticket_->setProperty("waveforms", QVariant::fromValue(waveform_list));
        }
      }
    }

    merged_table.Push(NodeValue::kSamples, QVariant::fromValue(block_range_buffer), track);

    return merged_table;

  } else {
    return super::GenerateBlockTable(track, range);
  }
}

TexturePtr RenderProcessor::ProcessVideoFootage(const FootageJob &stream, const rational &input_time)
{
  if (ticket_->property("type").value<RenderManager::TicketType>() != RenderManager::kTypeVideo) {
    // Video cannot contribute to audio, so we do nothing here
    return super::ProcessVideoFootage(stream, input_time);
  }

  // Check the still frame cache. On large frames such as high resolution still images, uploading
  // and color managing them for every frame is a waste of time, so we implement a small cache here
  // to optimize such a situation
  const VideoParams& render_params = GetCacheVideoParams();
  VideoParams stream_data = stream.video_params();

  ColorManager* color_manager = Node::ValueToPtr<ColorManager>(ticket_->property("colormanager"));

  // See if we can make this divider larger (i.e. if the fooage is smaller)
  int footage_divider = render_params.divider();
  while (footage_divider > 1
         && VideoParams::GetScaledDimension(stream_data.width(), footage_divider-1) < render_params.effective_width()
         && VideoParams::GetScaledDimension(stream_data.height(), footage_divider-1) < render_params.effective_height()) {
    footage_divider--;
  }

  QString using_colorspace = stream_data.colorspace();

  if (using_colorspace.isEmpty()) {
    // FIXME:
    qWarning() << "HAVEN'T GOTTEN DEFAULT INPUT COLORSPACE";
  }

  Decoder::CodecStream default_codec_stream(stream.filename(), stream_data.stream_index());

  QString decoder_id = stream.decoder();

  DecoderPtr decoder = nullptr;

  if (stream_data.video_type() == VideoParams::kVideoTypeVideo) {
    decoder = ResolveDecoderFromInput(decoder_id, default_codec_stream);
  } else {
    // Since image sequences involve multiple files, we don't engage the decoder cache
    decoder = Decoder::CreateFromID(decoder_id);

    QString frame_filename;

    if (stream_data.video_type() == VideoParams::kVideoTypeImageSequence) {
      int64_t frame_number = stream_data.get_time_in_timebase_units(input_time);
      frame_filename = Decoder::TransformImageSequenceFileName(stream.filename(), frame_number);
    } else {
      frame_filename = stream.filename();
    }

    // Decoder will close automatically since it's a stream_ptr
    decoder->Open(Decoder::CodecStream(frame_filename, stream_data.stream_index()));
  }

  if (decoder) {
    Decoder::RetrieveVideoParams p;
    p.divider = footage_divider;
    p.src_interlacing = stream_data.interlacing();
    p.dst_interlacing = GetCacheVideoParams().interlacing();

    if (!IsCancelled()) {
      FramePtr frame = decoder->RetrieveVideo((stream_data.video_type() == VideoParams::kVideoTypeVideo) ? input_time : Decoder::kAnyTimecode, p, GetCancelPointer());

      if (frame) {
        // Return a texture from the derived class
        TexturePtr unmanaged_texture = render_ctx_->CreateTexture(frame->video_params(),
                                                                  frame->data(),
                                                                  frame->linesize_pixels());

        // We convert to our rendering pixel format, since that will always be float-based which
        // is necessary for correct color conversion
        VideoParams managed_params = frame->video_params();
        managed_params.set_format(render_params.format());
        managed_params.set_pixel_aspect_ratio(stream_data.pixel_aspect_ratio());
        managed_params.set_interlacing(stream_data.interlacing());
        TexturePtr value = render_ctx_->CreateTexture(managed_params);

        ColorProcessorPtr processor = ColorProcessor::Create(color_manager,
                                                             using_colorspace,
                                                             color_manager->GetReferenceColorSpace());

        Renderer::AlphaAssociated alpha_assoc;
        if (stream_data.channel_count() != VideoParams::kRGBAChannelCount
            || stream_data.colorspace() == color_manager->GetReferenceColorSpace()) {
          alpha_assoc = Renderer::kAlphaNone;
        } else if (stream_data.premultiplied_alpha()) {
          alpha_assoc = Renderer::kAlphaAssociated;
        } else {
          alpha_assoc = Renderer::kAlphaUnassociated;
        }

        render_ctx_->BlitColorManaged(processor, unmanaged_texture,
                                      alpha_assoc,
                                      value.get());

        return value;
      }
    }
  }

  return super::ProcessVideoFootage(stream, input_time);
}

SampleBufferPtr RenderProcessor::ProcessAudioFootage(const FootageJob &stream, const TimeRange &input_time)
{
  DecoderPtr decoder = ResolveDecoderFromInput(stream.decoder(), Decoder::CodecStream(stream.filename(), stream.audio_params().stream_index()));

  if (decoder) {
    const AudioParams& audio_params = ticket_->property("aparam").value<AudioParams>();

    Decoder::RetrieveAudioData status = decoder->RetrieveAudio(input_time, audio_params,
                                                               stream.cache_path(),
                                                               stream.loop_mode(),
                                                               static_cast<RenderMode::Mode>(ticket_->property("mode").toInt()));

    if (status.status == Decoder::kOK && status.samples) {
      return status.samples;
    } else if (status.status == Decoder::kWaitingForConform) {
      ticket_->setProperty("incomplete", true);
    }
  }

  return super::ProcessAudioFootage(stream, input_time);
}

TexturePtr RenderProcessor::ProcessShader(const Node *node, const TimeRange &range, const ShaderJob &job)
{
  Q_UNUSED(range)

  QString full_shader_id = QStringLiteral("%1:%2").arg(node->id(), job.GetShaderID());

  QMutexLocker locker(shader_cache_->mutex());

  QVariant shader = shader_cache_->value(full_shader_id);

  if (shader.isNull()) {
    // FIXME: Reimplement as color job
    //if (job.UseOCIO()) {
    //  render_ctx_->ShaderJobInsertTextures(job.ColorProcessor(), &job, job.ShaderDesc());
    //}

    // Since we have shader code, compile it now
    shader = render_ctx_->CreateNativeShader(node->GetShaderCode(job.GetShaderID()));

    if (shader.isNull()) {
      // Couldn't find or build the shader required
      return super::ProcessShader(node, range, job);
    }
  }

  VideoParams tex_params = GetCacheVideoParams();

  tex_params.set_channel_count(GetChannelCountFromJob(job));

  TexturePtr destination = render_ctx_->CreateTexture(tex_params);

  // Run shader
  render_ctx_->BlitToTexture(shader, job, destination.get());

  return destination;
}

SampleBufferPtr RenderProcessor::ProcessSamples(const Node *node, const TimeRange &range, const SampleJob &job)
{
  if (!job.samples() || !job.samples()->is_allocated()) {
    return super::ProcessSamples(node, range, job);
  }

  SampleBufferPtr output_buffer = SampleBuffer::CreateAllocated(job.samples()->audio_params(), job.samples()->sample_count());
  NodeValueRow value_db;

  const AudioParams& audio_params = ticket_->property("aparam").value<AudioParams>();

  for (int i=0;i<job.samples()->sample_count();i++) {
    // Calculate the exact rational time at this sample
    double sample_to_second = static_cast<double>(i) / static_cast<double>(audio_params.sample_rate());

    rational this_sample_time = rational::fromDouble(range.in().toDouble() + sample_to_second);

    // Update all non-sample and non-footage inputs
    for (auto j=job.GetValues().constBegin(); j!=job.GetValues().constEnd(); j++) {
      NodeValueTable value = ProcessInput(node, j.key(), TimeRange(this_sample_time, this_sample_time));

      value_db.insert(j.key(), GenerateRowValue(node, j.key(), &value));
    }

    node->ProcessSamples(value_db,
                         job.samples(),
                         output_buffer,
                         i);
  }

  return output_buffer;
}

TexturePtr RenderProcessor::ProcessColorTransform(const Node *node, const ColorTransformJob &job)
{
  TexturePtr src = job.GetInputTexture();
  TexturePtr dest = render_ctx_->CreateTexture(src->params());

  render_ctx_->BlitColorManaged(job.GetColorProcessor(), src, Renderer::kAlphaAssociated, dest.get(), job.GetShaderPath());

  return dest;
}

TexturePtr RenderProcessor::ProcessFrameGeneration(const Node *node, const GenerateJob &job)
{
  FramePtr frame = Frame::Create();

  VideoParams frame_params = GetCacheVideoParams();
  frame_params.set_channel_count(GetChannelCountFromJob(job));
  if (job.GetRequestedFormat() != VideoParams::kFormatInvalid) {
    frame_params.set_format(job.GetRequestedFormat());
  }

  frame->set_video_params(frame_params);
  frame->allocate();

  node->GenerateFrame(frame, job);

  TexturePtr texture = render_ctx_->CreateTexture(frame->video_params(),
                                                  frame->data(),
                                                  frame->linesize_pixels());

  if (!job.GetColorspace().isEmpty()) {
    // Convert to reference space
    TexturePtr dest = render_ctx_->CreateTexture(GetCacheVideoParams());
    ColorManager* color_manager = Node::ValueToPtr<ColorManager>(ticket_->property("colormanager"));
    ColorProcessorPtr cp = ColorProcessor::Create(color_manager, job.GetColorspace(), color_manager->GetReferenceColorSpace());
    render_ctx_->BlitColorManaged(cp, texture, Renderer::kAlphaAssociated, dest.get());
    texture = dest;
  }

  return texture;
}

bool RenderProcessor::CanCacheFrames()
{
  return ticket_->property("type").value<RenderManager::TicketType>() == RenderManager::kTypeVideo;
}

TexturePtr RenderProcessor::GetCachedTexture(const QByteArray& hash)
{
  QString cache_dir = ticket_->property("cache").toString();
  if (cache_dir.isEmpty()) {
    return nullptr;
  }

  FramePtr f = FrameHashCache::LoadCacheFrame(cache_dir, hash);

  if (f) {
    TexturePtr texture = render_ctx_->CreateTexture(f->video_params(), f->data(), f->linesize_pixels());
    return texture;
  }

  return nullptr;
}

void RenderProcessor::SaveCachedTexture(const QByteArray &hash, TexturePtr tex_var)
{
  // FIXME: Temporarily disabled because I don't know how to ensure that the frame saved here is
  //        not the main frame. If it is, it'll be saved twice which will waste a lot of cycles.
  //        At least disabled, the frame will still save, and if nothing else alters the hash, it
  //        will pick up automatically from GetCachedTexture.
  /*if (!tex_var.isNull()) {
    QString cache_dir = ticket_->property("cache").toString();

    if (!cache_dir.isEmpty()) {
      TexturePtr texture = tex_var.value<TexturePtr>();
      FramePtr frame = Frame::Create();
      frame->set_video_params(texture->params());
      frame->allocate();
      render_ctx_->DownloadFromTexture(texture.get(), frame->data(), frame->linesize_pixels());
      FrameHashCache::SaveCacheFrame(cache_dir, hash, frame);
      qDebug() << "Saved mid-render frame to cache";
    }
  }*/
}

}