[eXeler0]

Originally Posted by coder76

Well, obviously we're not going to get to a level close to the original virtua fighter, a 68030/50 MHz has the performance of about a 486/25 MHz SX machine (no FPU). The external 68030 FPU is too slow to be usable. We would somehow need to get the performance of at least a 486/50 MHz or better to get somewhere.

The c2p idea was bad, when doing flatshaded polygons, it can be done efficiently with CPU directly to chip ram, with some predefined routines based on linelength. Then you can do fills with long writes, where possible or even movem's. This eliminates the c2p pass and you can also reduce filling by not using all planes. With 4 bitplanes, when only 2 is used for filling, then you still get 10 out of 16 colors and fill workload is reduced by around 50%. These are the optimizations that can be done with planar graphics.

Then, the blitter can be used to clear previous frame for free in parallell with CPU (even when CPU is accessing chip ram). With these sorts of optimizations we can get better performance than what the CPU speed alone would indicate.

I measured that my 68030/50 MHz does a memory read at 24 MB/sec to register and writes at around 32 MB/sec with cached code (a loop) in fast fam. Table lookup reads (long word) are around 3 times faster than doing a muls.w (around 25 cycles) and 6 times faster than divs.w.

It's time to start a new thread, "Could an Amiga with an 040 do Virtua Fighter?" ;-)

[Spoiler: No]