(In reply to comment #29)

> Hm, so actual problems were confirmed fixed with this? That seems rather
> strange, since that would mean the hardware counted the clock cycles wrong
> for the CTS value... Maybe the HW doesn't do it per-spec, instead
> calculating CTS in some other way?

Yes.  Using the sw cts/n values resulted in better results across asics
compared to the alternative fix (using the actual pll clock for the dto
calculation).  Using the hw cts/n values lead to playback speed issues on some
asics.