Do not allocate MIDI buffers for signal analysis

At this point in time MIDI buffers are vastly over-sized.
They include VST and LV2 event structure. This added about a MB per
plugin for no benefit.

1 files changed, 15 insertions, 10 deletions

diff --git a/libs/ardour/plugin_insert.cc b/libs/ardour/plugin_insert.cc
index f2ba464665..9cb654bcb4 100644
--- a/libs/ardour/plugin_insert.cc
+++ b/libs/ardour/plugin_insert.cc
@@ -913,7 +913,7 @@ PluginInsert::connect_and_run (BufferSet& bufs, samplepos_t start, samplepos_t e
 		//std::cerr << "               streams " << internal_input_streams().n_audio() << std::endl;
 		//std::cerr << "filling buffer with " << collect_signal_nframes << " samples at " << _signal_analysis_collected_nframes << std::endl;
 
-		_signal_analysis_inputs.set_count(input_streams());
+		_signal_analysis_inputs.set_count (ChanCount (DataType::AUDIO, input_streams().n_audio()));
 
 		for (uint32_t i = 0; i < input_streams().n_audio(); ++i) {
 			_signal_analysis_inputs.get_audio(i).read_from (
@@ -1048,7 +1048,7 @@ PluginInsert::connect_and_run (BufferSet& bufs, samplepos_t start, samplepos_t e
 		//std::cerr << "       output, bufs " << bufs.count().n_audio() << " count,  " << bufs.available().n_audio() << " available" << std::endl;
 		//std::cerr << "               streams " << internal_output_streams().n_audio() << std::endl;
 
-		_signal_analysis_outputs.set_count(output_streams());
+		_signal_analysis_outputs.set_count (ChanCount (DataType::AUDIO, output_streams().n_audio()));
 
 		for (uint32_t i = 0; i < output_streams().n_audio(); ++i) {
 			_signal_analysis_outputs.get_audio(i).read_from(
@@ -2063,15 +2063,20 @@ PluginInsert::configure_io (ChanCount in, ChanCount out)
 	_delaybuffers.configure (_configured_out, _plugins.front ()->max_latency ());
 	_latency_changed = true;
 
-	// we don't know the analysis window size, so we must work with the
-	// current buffer size here. each request for data fills in these
-	// buffers and the analyser makes sure it gets enough data for the
-	// analysis window
-	session().ensure_buffer_set (_signal_analysis_inputs, in);
-	_signal_analysis_inputs.set_count (in);
+	/* we don't know the analysis window size, so we must work with the
+	 * current buffer size here. each request for data fills in these
+	 * buffers and the analyser makes sure it gets enough data for the
+	 * analysis window. We also only analyze audio, so we can ignore
+	 * MIDI buffers.
+	 */
+	ChanCount cc_analysis_in (DataType::AUDIO, in.n_audio());
+	ChanCount cc_analysis_out (DataType::AUDIO, out.n_audio());
+
+	session().ensure_buffer_set (_signal_analysis_inputs, cc_analysis_in);
+	_signal_analysis_inputs.set_count (cc_analysis_in);
 
-	session().ensure_buffer_set (_signal_analysis_outputs, out);
-	_signal_analysis_outputs.set_count (out);
+	session().ensure_buffer_set (_signal_analysis_outputs, cc_analysis_out);
+	_signal_analysis_outputs.set_count (cc_analysis_out);
 
 	// std::cerr << "set counts to i" << in.n_audio() << "/o" << out.n_audio() << std::endl;