[time-nuts] time-nuts Digest, Vol 43, Issue 53

[christopher hoover]

Chuck Harris wrote:
> I liked programming on Motorola 68020's.  It was easy, and I never felt
> like I had to work at all to solve a problem.  But at the same time, it
> was glacially slow.  We got significantly better performance with the
> same programs running under the 386 and 486 processors of the day.  The
> compilers made smaller code for the intel processors than they did for the
> Motorola, and the Motorola lost all of the benchmarks we ran. 

I'm not taking sides here; just relating my experience, esp. on the
workstation side, vis a vis language implementation.

I worked on early Suns (68K) and early DEC MIPS boxes (e.g. PMAX).  These
came on the scene about the same time that Stallman started the GNU and
kicked it off with the gcc project.  gcc produced decent code for the 68K
and MIPS from early on, but the first backend for x86 didn't show up for a
long while and was horrible to boot!  IIRC, it took a major re-work of gcc's
internals to get decent x86 code out and that didn't happen for many years.
I'm sure there were much better commercial compilers, but I never used any
of those.

I had a hand in writing back ends for an industrial strength compiler for
Common Lisp for both 68K, MIPS and IBM's ROMP (descendant of the 801).
Registers are awfully nice things to have for a compiler -- especially ones
based on Wulf's TN-BIND idea and especially for Common Lisp, where we like
to pin a couple of registers for the runtime for various purposes.

But, at least for workstations, it was the RISC folks that put the bullet in
the head of the 68K.  Motorola couldn't get the clock rate up fast enough.
MIPS in particular was blowing them away.  

-ch