Hi. This is my second approach to handling polyphony. I'm not submitting it into bugzilla, since I'm not sure if that's the way it should be resolved. It works for me, but it's not thoroughly tested even for my needs.
These are my concerns. 1. It can't be implemented in AbstractEngineChannel, because it needs an access to voice pool, so it's in template EngineChannelBase. 2. It implements self-masking for sfz and gig (not sure how the sustain pedal issue should be resolved, so I left it intact) and handles group conflict in a different way (turned on by default only for sfz). If you want to test it with gig and sf2 (I haven't), just change the default value of NewAlgorithm to true. 3. It creates an interface for Voice classes to return KeyGroup (or off_by for sfz with an argument set to true). I don't like this solution. Maybe it should have two separate functions like GetKeyGroup, GetOffBy returning same values for gig and sf2 and different for sfz. 4. [premature optimization alert] It creates a separate Event for each voice and processGroupEvents iterates over all events. I think the complexity of this solution is O(n^2). It would be better with one event containing a std::set of VoiceIDs to check against, but it's not possible inside Event's union. 5. The function is huge. I don't like it. 6. signed/unsigned mismatch for sfz opcodes These are declared as int, even though most of them (according to http://www.sfzformat.com/legacy/) are unsigned with values from 0 to 4 Gb (4294967296), and the "end" opcode can even take values from -1 to 4 Gb. This spec is weird, since 32 bit ULONG_MAX is 4294967295. It's not such a big deal, but the patch depends on off_by default value to be -1, since AbstractVoice requires iKeyGroup to be non-zero to processGroupEvents() 7. My reference sfz engine is ARIA 1.933 with sforzando VST plugin. Unfortunately it behaves differently with note_polyphony opcode. My patch releases oldest voices first. ARIA releases most recent ones. I think it's a bug, not a feature, because if there are two regions triggered at the same time the first one is suppressed and only the second one is played. I've attached an example file to reproduce this ARIA bug (you must provide a Crash.wav file which is long enough to test polyphony). I've sent it here to get your opinion, but I suppose I'm not going to develop my concerns further, only bugfixing.
polyphony2.patch
Description: Binary data
test.sfz
Description: Binary data
_______________________________________________ Linuxsampler-devel mailing list Linuxsampler-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linuxsampler-devel
