Clr.l	-(a0)	is 25% faster than
Clr.l	(a0)+

move.l (a0,d0.l)	is faster than
move.l (a0,d0.w)
&	same with adda.w <-> adda.l etc.

move.b d0,-(a7)

will decrease a7 with 2!

Never use those silly #xx,([a0]) new modes, they are slower than

move.l (a0),a0

move.l #xx,(a0)

..unless you have no extra spare registers, but... that's rare.

time optimization on the Falcon is more like sparing BUS than CPU. The st-ram is just too slow..... Adjust your writes to st-ram. Adjusting memory reads will not speed up, unless you get rid of a whole instruction :).

The interrupts are exceedingly slow. A simple movem*2 Trap and RTE will take an enormous amount of cycles.

movem.l all,-(sp)
movem.l (sp)+,all
rte

will take about 80 Cycles... ok, hehe...why use all regs.... ;-)

English version not yet available.