Changes
/* Automatic feeding of sound generator */ - Detailed information on DMA pause and prescaler
The SAR must be loaded by the CPU with a physical RAM address between 0000h and FFFEh. This means that the most significant two bits select which pages 0 to 3 of the DRAM is used, and the remaining bits are the address relative to the page start. The DMA process is not affected by the RAM or ROM mapping registers, and will always fetch data from RAM and not ROM. Note that the least significant bit of the address is ignored, and the instructions are always fetched from word boundaries.
The pause prescaler counts N+1 scan lines (where N is the value written by the CPU), giving a minimum tick of 64us, and a maximum of 16.384ms. When set nonzero by a pause instruction, the pause counter for a particular channel is decremented every tick until it reaches zero. Therefore, if the PPR is set to a value N and a PAUSE M instruction is executed, the total delay time between the instruction before the PAUSE and that following the PAUSE will be M * (N+1) * 64us. Pauses of between 64us and 67s may thus be generated. If a DMA channel is executing a pause when the DMA channel is disabled or the SAR is changed, the pause counter will continue to decrement. It will not reset and when the DMA is again enabled the command at the current SAR address will not be executed until the pause is completed. Changing the prescale value mid-pause will vary the length of the pause (eg. If the prescale is set to 0 and a PAUSE 10 is executed and has already decremented 5 times, changing the prescale to 1 will cause the pause to have a duration of 15 scan lines).
The ASIC arbitrates accesses to the parallel interface device between the "DMA" channels and the CPU, allowing only one to access it at a time. CPU accesses to the 8255 could be held off by means of wait states for up to 8 microseconds if the "DMA" channel is currently executing a LOAD instruction. After a LOAD is executed, the ASIC must put the PSG address register back as it was before. To achieve this the 8255 parallel peripheral interface and the 74LS145 decoder have been integrated into the ASIC.
The exact timing is based on 1us cycles as follows. After the leading edge from HSYN from the 6845 there is one dead cycle followed by an instruction fetch cycle for each channel which is active (i.e. enabled and not paused). The execute cycles then follow for each active channel. All instructions execute in one cycle, except that LOAD requires at least 8 cycles. An extra cycle is added to a LOAD if the CPU is accessing the 8255, or two extra cycles if the CPU access was itself a PSG register write.
===Interrupt service (Vectored interrupts)===
