That's not exactly true: Treasure Hunter G has a DMA queue which allows some DMA transfers be delayed to next NMI if there is no more VBlank time left. If DMA could execute faster than the original SNES, then we could measure improves in all games than implements such a queue (I guess THG isn't the only one).jwdonal wrote:Performance enhancements would certainly be possible since the hardware is fully programmable but the real question is what games would use them? The answer is none - at least not any official ones anyway. So while it's a cool thing to think about, there would really be no point in doing it.
And the same with FIFO-buffered accesses to $2140/41/42 to send samples to APU. Most games use this scheme to send samples, so buffering acceses could reduce the latency to the most: each time SNES CPU puts a byte into $2140, it is buffered so it is acknowledged inmediately from the APU (FIFO) side. This, again, reduces CPU load and increased performance could be measured.
Besides, it could be used for new homebrew development, which would be nice