Originally Posted by Thorham
Indeed. You fill a bunch of chipmem and tell the chipset to start playing from that location at a specified sample rate, and the hardware plays the sample. There's no waveform generator, no nothing. Pure samples and sample manipulation, that's it.
I disagree - wavetable is a way how you use your sample hardware, not how you create samples - samples are usually or synthesized (mathematical modeling) or recorded (natural sources).
You can easily simulate for example granular synthesis by quick change between DMA start buffer and this have nothing to way how you fill your buffer.
As there is no difference between hardware sine implementation and software sine generation unless sine produced by both methods is exactly the same.
Originally Posted by Loedown
AM involves adjusting the value of the currently played sample value either up or down by amount x and is therefore computed, FM adjusts the speed register up or down by amount y and is therefore computed. PWM isn't used, volume control is just a value in a register which adjusts an internal amp to Paula I would suppose and mixing voices together would be done post DAC with a mixing amplifier. Not one single element of this is sound generation but sound manipulation, effects and nothing more. To have sound generation you need an oscillator and there aren't any.
Paula can perform AM modulation (complex by using one channel samples to modulate second channel), similar for FM, complex FM can be performed by Paula.
PWM is used to adjust level (it was confirmed by observing Paula output, also some block schematics shows partially presence of PWM and it is almost directly exposed to programmer as a bit 6 in each AUDxVOL to turn off PWM - for bits 0-5 set as 1's maximum audio level is 63/64 which is typical for PWM a 0 is no level at all).
Also HRM mentioned other methods for audio generation - CPU driven and as a PWM driven.