overhauled render classes to support video AND audio

Major refactoring work to try sharing as much code as possible between the
video renderers and audio renderers, as well as make them as
platform-independent as possible.

app/render/backend/audio/audiobackend.cpp

@@ -1,13 +1,34 @@
 #include "audiobackend.h"
 
+#include "audioworker.h"
+
 AudioBackend::AudioBackend(QObject *parent) :
   AudioRenderBackend(parent)
 {
 }
 
+AudioBackend::~AudioBackend()
+{
+  Close();
+}
+
+QIODevice *AudioBackend::GetAudioPullDevice()
+{
+  pull_device_.setFileName(CachePathName());
+
+  return &pull_device_;
+}
+
 bool AudioBackend::InitInternal()
 {
-  // This backend doesn't init anything yet
+  // Initiate one thread per CPU core
+  for (int i=0;i<threads().size();i++) {
+    // Create one processor object for each thread
+    AudioWorker* processor = new AudioWorker(decoder_cache());
+    processor->SetParameters(params());
+    processors_.append(processor);
+  }
+
   return true;
 }
 
@@ -26,3 +47,34 @@ void AudioBackend::DecompileInternal()
 {
   // This backend doesn't compile anything yet
 }
+
+void AudioBackend::ConnectWorkerToThis(RenderWorker *worker)
+{
+  connect(worker, SIGNAL(CompletedCache(NodeDependency)), this, SLOT(ThreadCompletedCache(NodeDependency)));
+}
+
+void AudioBackend::ThreadCompletedCache(NodeDependency dep)
+{
+  caching_ = false;
+
+  QByteArray cached_samples = dep.node()->get_cached_value(dep.range()).toByteArray();
+
+  int offset = params().time_to_bytes(dep.in());
+  int length = params().time_to_bytes(dep.range().length());
+  int out_point = offset + length;
+
+  if (pcm_data_.size() < out_point) {
+    pcm_data_.resize(out_point);
+  }
+
+  // Replace data with this data
+  memcpy(pcm_data_.data() + offset, cached_samples.data(), static_cast<size_t>(length));
+
+  QFile f(CachePathName());
+  if (f.open(QFile::WriteOnly)) {
+    f.write(pcm_data_);
+    f.close();
+  }
+
+  CacheNext();
+}