PerTrackEffect.cpp

Go to the documentation of this file.

/**********************************************************************
 
  Audacity: A Digital Audio Editor
 
  PerTrackEffect.cpp
 
  Dominic Mazzoni
  Vaughan Johnson
  Martyn Shaw
 
  Paul Licameli split from Effect.cpp
 
*******************************************************************//*******************************************************************/
 
 
#include "PerTrackEffect.h"
#include "EffectOutputTracks.h"
 
#include "AudioGraphBuffers.h"
#include "AudioGraphTask.h"
#include "EffectStage.h"
#include "SyncLock.h"
#include "TimeWarper.h"
#include "ViewInfo.h"
#include "WaveTrack.h"
#include "WaveTrackSink.h"
#include "WideSampleSource.h"
 
PerTrackEffect::Instance::~Instance() = default;
 
bool PerTrackEffect::Instance::Process(EffectSettings &settings)
{
   return mProcessor.Process(*this, settings);
}
 
bool PerTrackEffect::Instance::ProcessInitialize(EffectSettings &,
   double, ChannelNames)
{
   return true;
}
 
bool PerTrackEffect::Instance::ProcessFinalize() noexcept
{
   return true;
}
 
PerTrackEffect::~PerTrackEffect() = default;
 
bool PerTrackEffect::DoPass1() const
{
   return true;
}
 
bool PerTrackEffect::DoPass2() const
{
   return false;
}
 
bool PerTrackEffect::Process(
   EffectInstance &instance, EffectSettings &settings) const
{
   auto pThis = const_cast<PerTrackEffect *>(this);
 
   // Destroy any pre-formed output tracks when done
   auto pOutputs = mpOutputTracks.get();
 
   std::optional<EffectOutputTracks> outputs;
   if (!pOutputs)
      pOutputs = &outputs.emplace(*mTracks, GetType(),
         EffectOutputTracks::TimeInterval{ mT0, mT1 }, true);
 
   bool bGoodResult = true;
   // mPass = 1;
   if (DoPass1()) {
      auto &myInstance = dynamic_cast<Instance&>(instance);
      bGoodResult = pThis->ProcessPass(pOutputs->Get(), myInstance, settings);
      // mPass = 2;
      if (bGoodResult && DoPass2())
         bGoodResult = pThis->ProcessPass(pOutputs->Get(), myInstance, settings);
   }
   if (bGoodResult)
      pOutputs->Commit();
   DestroyOutputTracks();
   return bGoodResult;
}
 
bool PerTrackEffect::ProcessPass(TrackList &outputs,
   Instance &instance, EffectSettings &settings)
{
   const auto duration = settings.extra.GetDuration();
   bool bGoodResult = true;
   bool isGenerator = GetType() == EffectTypeGenerate;
   bool isProcessor = GetType() == EffectTypeProcess;
 
   Buffers inBuffers, outBuffers;
   ChannelName map[3];
   size_t prevBufferSize = 0;
   int count = 0;
   bool clear = false;
 
   // It's possible that the number of channels the effect expects changed based on
   // the parameters (the Audacity Reverb effect does when the stereo width is 0).
   const auto numAudioIn = instance.GetAudioInCount();
   const auto numAudioOut = instance.GetAudioOutCount();
   if (numAudioOut < 1)
      return false;
 
   // Instances that can be reused in each loop pass
   std::vector<std::shared_ptr<EffectInstance>> recycledInstances{
      // First one is the given one; any others pushed onto here are
      // discarded when we exit
      std::dynamic_pointer_cast<EffectInstanceEx>(instance.shared_from_this())
   };
 
   const bool multichannel = numAudioIn > 1;
   int iChannel = 0;
   TrackListHolder results;
   const auto waveTrackVisitor =
      [&](WaveTrack &wt, WaveChannel &chan, bool isFirst) {
         if (isFirst)
            iChannel = 0;
 
         sampleCount len = 0;
         sampleCount start = 0;
         WaveChannel *pRight{};
 
         const int channel = (multichannel ? -1 : iChannel++);
         const auto numChannels = MakeChannelMap(wt.NChannels(), channel, map);
         if (multichannel) {
            assert(numAudioIn > 1);
            if (numChannels == 2) {
               // TODO: more-than-two-channels
               pRight = (*wt.Channels().rbegin()).get();
               clear = false;
            }
         }
 
         if (!isGenerator) {
            GetBounds(wt, &start, &len);
            mSampleCnt = len;
            if (len > 0 && numAudioIn < 1) {
               bGoodResult = false;
               return;
            }
         }
         else
            mSampleCnt = wt.TimeToLongSamples(duration);
 
         const auto sampleRate = wt.GetRate();
 
         // Get the block size the client wants to use
         auto max = wt.GetMaxBlockSize() * 2;
         const auto blockSize = instance.SetBlockSize(max);
         if (blockSize == 0) {
            bGoodResult = false;
            return;
         }
 
         // Calculate the buffer size to be at least the max rounded up to the clients
         // selected block size.
         const auto bufferSize =
            ((max + (blockSize - 1)) / blockSize) * blockSize;
         if (bufferSize == 0) {
            bGoodResult = false;
            return;
         }
 
         // Always create the number of input buffers the client expects even
         // if we don't have
         // the same number of channels.
         // (These resizes may do nothing after the first track)
 
         if (len > 0)
            assert(numAudioIn > 0); // checked above
         inBuffers.Reinit(
            // TODO fix this hack for making Generator progress work without
            // assertion violations.  Make a dummy Source class that doesn't
            // care about the buffers.
            std::max(1u, numAudioIn),
            blockSize,
            std::max<size_t>(1, bufferSize / blockSize));
         if (len > 0)
            // post of Reinit later satisfies pre of Source::Acquire()
            assert(inBuffers.Channels() > 0);
 
         if (prevBufferSize != bufferSize) {
            // Buffer size has changed
            // We won't be using more than the first 2 buffers,
            // so clear the rest (if any)
            for (size_t i = 2; i < numAudioIn; i++)
               inBuffers.ClearBuffer(i, bufferSize);
         }
         prevBufferSize = bufferSize;
 
         // Always create the number of output buffers the client expects
         // even if we don't have the same number of channels.
         // (These resizes may do nothing after the first track)
         // Output buffers get an extra blockSize worth to give extra room if
         // the plugin adds latency -- PRL:  actually not important to do
         assert(numAudioOut > 0); // checked above
         outBuffers.Reinit(numAudioOut, blockSize,
            (bufferSize / blockSize) + 1);
         // post of Reinit satisfies pre of ProcessTrack
         assert(outBuffers.Channels() > 0);
 
         // (Re)Set the input buffer positions
         inBuffers.Rewind();
 
         // Clear unused input buffers
         if (!pRight && !clear && numAudioIn > 1) {
            inBuffers.ClearBuffer(1, bufferSize);
            clear = true;
         }
 
         const auto genLength = [this, &settings, &wt, isGenerator](
         ) -> std::optional<sampleCount> {
            double genDur = 0;
            if (isGenerator) {
               const auto duration = settings.extra.GetDuration();
               if (IsPreviewing()) {
                  gPrefs->Read(wxT("/AudioIO/EffectsPreviewLen"), &genDur, 6.0);
                  genDur = std::min(duration, CalcPreviewInputLength(settings, genDur));
               }
               else
                  genDur = duration;
               // round to nearest sample
               return sampleCount{ (wt.GetRate() * genDur) + 0.5 };
            }
            else
               return {};
         }();
 
         const auto pollUser = [this, numChannels, count, start,
            length = (genLength ? *genLength : len).as_double()
         ](sampleCount inPos){
            if (numChannels > 1) {
               if (TrackGroupProgress(
                  count, (inPos - start).as_double() / length)
               )
                  return false;
            }
            else {
               if (TrackProgress(count, (inPos - start).as_double() / length))
                  return false;
            }
            return true;
         };
 
         // Assured above
         assert(len == 0 || inBuffers.Channels() > 0);
         // TODO fix this hack to make the time remaining of the generator
         // progress dialog correct
         if (len == 0 && genLength)
            len = *genLength;
         WideSampleSequence *pSeq = &chan;
         if (pRight)
            pSeq = &wt;
         WideSampleSource source{
            *pSeq, size_t(pRight ? 2 : 1), start, len, pollUser };
         // Assert source is safe to Acquire inBuffers
         assert(source.AcceptsBuffers(inBuffers));
         assert(source.AcceptsBlockSize(inBuffers.BlockSize()));
 
         // Make "wide" or "narrow" copy of the track if generating
         // Old generator code may still proceed "interval-major" and later
         // join mono into stereo
         auto wideTrack =
            (pRight && isGenerator) ? wt.EmptyCopy() : nullptr;
         auto narrowTrack =
            (!pRight && isGenerator) ? wt.EmptyCopy(1) : nullptr;
         const auto pGenerated = wideTrack
            ? wideTrack
            : narrowTrack;
 
         WaveTrackSink sink{ chan, pRight, pGenerated.get(), start, isProcessor,
            instance.NeedsDither() ? widestSampleFormat : narrowestSampleFormat
         };
         assert(sink.AcceptsBuffers(outBuffers));
 
         // Go process the track(s)
         const auto factory =
         [this, &recycledInstances, counter = 0]() mutable {
            auto index = counter++;
            if (index < recycledInstances.size())
               return recycledInstances[index];
            else
               return recycledInstances.emplace_back(MakeInstance());
         };
         bGoodResult = ProcessTrack(channel, factory, settings, source, sink,
            genLength, sampleRate, wt, inBuffers, outBuffers);
         if (bGoodResult) {
            sink.Flush(outBuffers);
            bGoodResult = sink.IsOk();
            if (bGoodResult && pGenerated) {
               if (!results)
                  results = TrackList::Temporary(nullptr, pGenerated);
               else {
                  results->Add(pGenerated);
                  if (!multichannel && !isFirst && narrowTrack) {
                     // Generated a stereo track, in channel-major fashion.
                     // Get the last track but one -- generated in the previous
                     // pass
                     const auto pLast =
                        static_cast<WaveTrack*>(*std::next(results->rbegin()));
                     pLast->ZipClips();
                  }
               }
            }
         }
         if (!bGoodResult)
            return;
         ++count;
      };
   const auto defaultTrackVisitor =
      [&](Track &t) {
         if (SyncLock::IsSyncLockSelected(t))
            t.SyncLockAdjust(mT1, mT0 + duration);
      };
 
   outputs.Any().VisitWhile(bGoodResult,
      [&](auto &&fallthrough){ return [&](WaveTrack &wt) {
         if (!wt.GetSelected())
            return fallthrough();
         const auto channels = wt.Channels();
         if (multichannel)
            waveTrackVisitor(wt, **channels.begin(), true);
         else {
            bool first = true;
            for (const auto pChannel : channels) {
               waveTrackVisitor(wt, *pChannel, first);
               first = false;
            }
         }
         if (results) {
            const auto t1 = ViewInfo::Get(*FindProject()).selectedRegion.t1();
            PasteTimeWarper warper { t1,
                                     mT0 + (*results->begin())->GetEndTime() };
            wt.ClearAndPaste(mT0, t1,
               static_cast<WaveTrack&>(*results->DetachFirst()),
               true, true, &warper);
            results.reset();
         }
      }; },
      defaultTrackVisitor
   );
 
   if (bGoodResult && GetType() == EffectTypeGenerate)
      mT1 = mT0 + duration;
 
   return bGoodResult;
}
 
bool PerTrackEffect::ProcessTrack(int channel, const Factory &factory,
   EffectSettings &settings,
   AudioGraph::Source &upstream, AudioGraph::Sink &sink,
   std::optional<sampleCount> genLength,
   const double sampleRate, const SampleTrack &wt,
   Buffers &inBuffers, Buffers &outBuffers)
{
   assert(upstream.AcceptsBuffers(inBuffers));
   assert(sink.AcceptsBuffers(outBuffers));
 
   const auto blockSize = inBuffers.BlockSize();
   assert(upstream.AcceptsBlockSize(blockSize));
   assert(blockSize == outBuffers.BlockSize());
 
   auto pSource = EffectStage::Create(
      channel, static_cast<const WideSampleSequence&>(wt).NChannels(), upstream,
      inBuffers, factory, settings, sampleRate, genLength);
   if (!pSource)
      return false;
   assert(pSource->AcceptsBlockSize(blockSize)); // post of ctor
   assert(pSource->AcceptsBuffers(outBuffers));
 
   AudioGraph::Task task{ *pSource, outBuffers, sink };
   return task.RunLoop();
}
 
std::shared_ptr<EffectOutputTracks> PerTrackEffect::MakeOutputTracks()
{
   return mpOutputTracks =
      std::make_shared<EffectOutputTracks>(*mTracks, GetType(),
         EffectOutputTracks::TimeInterval{ mT0, mT1 }, true);
}
 
void PerTrackEffect::DestroyOutputTracks() const
{
   mpOutputTracks.reset();
}