Re: [TUHS] Bryan Cantrill on bfs & ta

Greg the 68000 and 68010 were 16 bit internals. ie a 16 bit barrel shifter and it took 2 ticks to perform 32 bit ops. a natural int was indeed 16 bits in the base registers although they also could be used as 32 bit registers (2 ticks ) so some 68k compilers defined int as 16 others as 32 (more in a minute). it technically was LP32 not ILP32 the external logic (ie pins) supported 24 bits of address. moto fortunately passed all 32 bits along on the first chip and onto storage (thank you Les & Nick) so when later versions had a full 32 bit shifter everything just worked. this meant that both chips supported 32bit addressing. So while the machine operated 16 bits at a time the programmer had a full 32 bit view (and eventually a 32 bit virtual address space) even though the chip only supported 24 bits of address and operated 16 bits of data at a time. They're so even thought they were 16 bit internally. I'd never think of either chop as a 16 bit machine as for 16/32 bit argument - the hacked C compiler that Glasser and I cobbled together in 1979 at Tektronix used 16 bit. IIRC the rts guys (jack test et al) used 32 bits at day one even though the natural int was 16 bits. there were advantages to both solutions and Jeff Mogul and i used argue about. looking back on the 68000 in hind sight MIT got it right and while it worked mildly faster a 16 bit int was a bad idea Clem PS but I agree. SunOS never ran on a pure 16 not machine. on fact they started with compiler derived from Test et al. PPS. we relived this whole argument with 64 bits and it was interesting that we generally came to think LP64 made more sense for chips like Alpha On Oct 18, 2012, at 6:40 PM, Greg 'groggy' Lehey <grog(a)lemis.com> wrote: