In the last post I described a circuit in which the MC68008 CPU executed a free run; basically, it executed an unending series of nop instructions. An LED connected to A19 (the high address line) should blink on and off once each time the CPU cycles through its 1 MB address space.
Since a nop takes 8 cycles to execute, the CPU should execute 1,000,000 nop instructions per second with an 8 MHz system clock. With 1,048,576 instructions in the address space, this should have caused the LED to blink about once per second. However, I observed two blinks per second, which would imply a nop executing every 4 cycles, not 8.
I thought of a possible explanation: the CPU could be prefetching the next instruction while the current instruction is executing. An MC68008 read cycle requires 4 clock cycles, so each pair of successive nop instructions would overlap by 4 cycles, resulting in the effective execution of one nop every 4 cycles. One web page I found does seem to indicate that the MC68000 (and its sibling, the ‘08) has a two-word prefetch queue.
I’ll know for sure when I observe the low address lines with a logic analyzer.
blog comments powered by Disqus