The WWDC videos like the new "Boost your graphics performance with the M5 and A19 GPUs" contain extremely valuable information and tips on how to discover, diagnose and remedy performance issues. They seem to serve as quick reminders and distilled summaries of more comprehensive documentation that I assume can be found somewhere. Where do we find the underlying comprehensive documentation that explains Apple Silicon GPU architecture? How can I learn to understand the basis of the data presented by the Xcode Metal Debugger? Any hints at external literature and resources are welcome.

Am I correct in my understanding that the new MusicUnderstanding is not intended to be able to support analysis of Apple Music streams via MusicKit or in other ways?

Is there a way to attach a MusicUnderstandingSession to a MusicKit stream or audiofile?

Certain apps (eg. DJay) at least seem to have access to audio data from Apple Music streams, either directly or via an indirect dataset for visualization purposes either generated from audio data on the device or delivered from a remote service. Is there any framework provided by Apple or special agreement with Apple that gives access to Apple Music audio data, or sets of visualization meta data, or the ability to run processing of audio data on device or remotely (either on-demand or via preprocessing)?

Certain apps (eg. DJay) seem to have direct or indirect access to the audio data from Apple Music, either on the device or remotely preprocessed. Is that feature available for other third party apps? Am I correct in my understanding that the new framework MusicUnderstanding (which looks quite interesting, judging from the docs so far) is not in any way usable for streams or local files from Apple Music / MusicKit since it relies entirely on AVReadOnlyAudioPCMBuffer? Is there eg. a way to get the same kind of music understanding data (or some other meta data suitable for real time visualization) provided directly form Apple Music without going via a readable audio buffer?