Citra: Implement Looped Buffers

[WishfulThinking]

Many games seem to use looped audio buffers and produce the following log message.

Audio.DSP <Error> audio_core/hle/source.cpp:DequeueBuffer:249: Looped buffers are unimplemented at the moment

We should implement them.

This is an issue to track the progress of the implementation.

[MerryMage]

The work would likely require either writing a few tests to confirm how looped buffers work or reverse engineering the DSP binary blob.

A buffer is a looped buffer if Buffer::is_looped is set. SourceConfiguration::loop_related indicates where the first sample position of the loop is. There is a second loop boolean SourceConfiguration::looping.

My own work on this is waiting on some DSP RE I'm doing; but I don't mind someone jumping in in the meantime.

[jmerdich]

Chiming in here after doing a bunch of fiddling based on @MerryMage 's test suite on my o3ds and a few games that use loops in the emulator...

• Simple loops are the only thing I've seen in a commercial game. Pretty much, having a buffer in the queue flagged as looping the entire, usually for a reverb effect (found this in Zelda OOT's background music, Mario Kart's button noises, and Mario 3D Land's "?" blocks). A patch to support this would be trivial, and I can clean it up and toss one out sometime tomorrow.
• I haven't been able to get loop_related to do anything on my hardware. The play_position register affects the start of the first playback of the embedded buffer. All following loops start from the beginning of the buffer regardless of the value of loop_related. Commercial games seem to be fond of setting the dirty bit for play_position=0 when looping, though that register does not seem to be updated to any other value by either the game or the DSP engine. Further testing is definitely necessary.
• Complex loops don't seem to work with queued buffers, only the embedded ones. This might just be a timing or caching issue though (Mario Kart 7 always invalidates the cache to the DSP command buffer when playing loops, so there might be some errata).

I'd be happy to make a PR for any of this, and any additional information can help speed the latter 2 bullet points.

[MerryMage]

It's been a while since I've taken a look at audio. The above statements were based on what I remembered of my previous work before I started writing everything down; apologies for what appears to be my poor memory and sending you on what seems to be a wild goose chase.

Feel free to submit a PR for what you do have; I'm currently (very slowly) reverse engineering the binary blob.

[MerryMage]

Mostly implemented afaik in #2422.

[jmerdich]

I implemented everything mentioned here in that PR. The only thing left is the FIFO like behavior in 3ds sound, as noted in the other pr.

[B3n30]

Is this issue solved?

[MerryMage]

Closing as there's no value in keeping this open; everything mentioned has been implemented.