Some games have very tight scheduling requirements for their audio which can't really be matched on the host, adding a constant to the reported value helps to provide some leeway.
Waiting on the host side is inaccurate and leads to desyncs in the event of the sink missing a deadline that require stalls to fix. By waiting for the sink to have space before even starting rendering such desyncs can be avoided.
This avoids the need to stall if the host sink sporadically misses the deadline, in such a case the previous implementation would report them samples as being played on-time, causing the guest to send more samples and leading to a gradual buildup.
The existing implementation only supports 64 invoc-per-subgroup GPUs, and misbehaves on adreno when invocations need to be split into 4 emulated subgroups.
The current arguments worked by happenstance as games only ever submit
one syncpoint and request one fence back, if a game were to do something
other than this then the arguments would've been parsed entirely wrong.
The puller register array is made up of u32s however the `NUM_REGS` value is the size in bytes, so switch it to avoid making the struct unnecessary large. Also fix a small typo in a comment.
Billy Laws