[djmips]

Very nice! Good work.

Some more info on the digi player on the ST. It used a timer interrupt to service the PCM sample but if you used just the interrupt there was significant noise because there was significant variability of the timing on the interrupt. To get the timing tighter the interrupt timing was changed to hit the routine on every video line just prior to the actual hsync and then hsync was polled to get very precise timing.

The PCM was just a linearization by combining three logarithmic volumes of the three PSG voices.

During the title sequence, a special version of the code was running where several 68000 registers were reserved globally for the digi player. So those did not need to be saved / restored in the interrupt routine!

The SWAR was involved when advancing the four wrapping 8 bit indices into the each of the four voice's 256 entry sample tables. This was part of a monumental effort to get the interrupt routine to be as quick as possible.

Your video decoding reminds me of when I worked at a video card company in the nineties we had a competitive advantage by using a commodity part in an unusual way. This video decoder hardware was commonly used to take composite video and decode it. We supported that but we also did a bunch of advanced features by using a seldom used mode where it could be used in a raw mode where it took the composite signal and stored the analog to digital conversion in memory. We had high speed assembly code that could decode the video better than the hardware and supported some cool additional features. Anyway... memories a bit hazy. Been a while but I remember it being very cool.