Uploading audio at the end of retro_run() with
retro_audio_sample_batch_t instead of using retro_audio_sample_t to
submit samples as they are generated has peformance benefits when using
low audio latency settings. Frontends that implement audio sync and DRC
(like RetroArch) will block for a long time when submitting audio. This
can cause video frame presentation to be delayed too long. Low audio
latency settings make it worse, as DRC and audio sync is performed more
often, creating a bottleneck.
Simply accumulating all generated audio samples and uploading them at
the end with retro_audio_sample_batch_t improves things condiderably.
9e8f918b27