2 Sep
2015
2 Sep
'15
11:05 p.m.
This isn't quite the way I remember the CMU instructions working. Nor
is it exactly how I've implemented them in my emulator.
Another oddball thing was then then-new Cyber 72/73
CMU. An interesting
beast, but not present on the 74.
Presumably because an instack loop could move data faster. I always
wondered, though, if you could put a CMU in the serial CPU on a 74-2x.
Probably too messy.
It was possible initially to write code with the 46xxx
CMU instruction
in first 2 parcels of an instruction word. All Cyber 73 CMU
instructions were 60 bits,
No, the IM instruction is 30 bits. It's supposed to be forced upper. I
don't recall whether the 72/73 were that picky and it would work in the
first 3 parcels, or work only in the first one (thus being 60-bit
equivalent) and pass in the others (with the rest of the instructions
becoming arbitrary).
you could pack a call to a subroutine to do
the equivalent thing in the lower 30 bits for the 74. Worked pretty
cool
Looks like it would. Aren't there some variant OSes that would put the
monitor call parameters in the lower 30 bits of an XJ instruction? You
could probably also pack a parameter in the B register, and another in
the address, since those would mode out anyway (or be ignored in user
mode)--you'll get to CPUMTR no matter what. ECS instructions would
execute the lower 30 bits if the transfer failed; this all in user mode
with no monitor intervention.
You could probably do the same thing with a conventional XJ instruction.
6000s lacking ECS hardware and CEJ/MEJ treated all 01x instructions as
010 (RJ). Probably can't get away with that on a 70.
until the 170. There, different models supported
different subsets
of CMU instructions (or not at all)
It was all-or-nothing, no subsets.
--and attempting to execute one not
implemented was greeted with an error stop. The 170 people really
screwed that one up.
464xx and up were illegal on machines that did not (171) or could not
(175) have the CMU option, or in other parcels on any 170. 46000-46377
were pass instructions. I don't know why they did this either, since
there were no 170s with unmatched CPUs (like the 6700 or 74-2x). Also,
some dual-CP machines (including the 72/73) could have 1 or 2 CMUs.
Needed some special code in CPUMTR to avoid using the CMU for storage
moves if there were fewer CMUs than CPUs, lest it be busy in the other
CPU and we had to wait for it in monitor mode.
It's possible that on 180s, 46100-46377 were illegal instructions,
though the design goal for 170 mode was a 173. Perhaps the
175/740/750/760 did that too. Note also on 180s with CMUs (emulated),
the instruction was limited to data in the first 262K; in particular
CPUMTR couldn't use it for storage moves unless it played games with
RA/FL in program mode (and it would still have to drag the instruction
around into each segment). Too much of a kludge, I expect.
--
Jeff Woolsey
jlw at jlw.com
3 Sep
3 Sep
9:28 p.m.
On 09/02/2015 11:05 PM, Jeff Woolsey wrote:
This isn't quite the way I remember the CMU
instructions working.
Nor is it exactly how I've implemented them in my emulator.
Sorry that my explanation was scrambled with a few blown synapses. I'll
try again.
CMU was implemented in the lower Cyber 70 (72,73) and 170 (
720, 730) as instructions 464xx through 467xx. However, what happens
with the upper Cybers varies--the 74 and 176 (as well as the 7600 and
6600) treat them as no-ops (46xxx). However, the 750 and 760 treat
464...467 as illegal instructions and error out.
According to my Cyber 72/73 reference, all CMU instructions must be
located in the first parcel of an instruction word. The Cyber 170
restates this in at least two places in the instruction reference. So
no putting an IM in the second or third parcel. The reference for the
lower 170s states that the machine does not execute any instructions in
the lower 2 parcels of a word.
Of course, the other CMU instructions 465-567 occupy a fullword, so no
problems with alignment.
One of the early "tricks' was to take the IM instruction 464 B KKKKKK
instruction, set up a B register and offset, such that the offset would
be encoded as 46000 - a plain old no-op. So, if your B register was 2,
the first 2 parcels of an instruction word looked like:
46420 46000
Two no-ops to a Cyber 74, 7600, 176 so an RJ to the regular move routine
could be put in the lower half of the word. Worked like a dream.
But when the upper Cyber 170 systems (750, 760) came onto the scene,
that "clever" code quit working. Maybe NOS/NOS-BE recognized the
illegal instruction exception and implemented the IM and other CMU
instructions in software, but I was off onto non-Cyber stuff by then.
Regardless, I do recall some grumbling at the time.
I never knew if the CMU ever amounted to much in the real world other
than for the IM instruction. You pretty much had to build the
instructions for CC CU and DM, which involved a fair amount of setup
code. I'm sure that Don Nelson had a directive from on high to put them
into COBOL, but I never saw any benchmark results--I was off working on
64 bit hardware and 6-bit characters weren't part of my world.
Such is my limited exposure to the late Cyber 70/early Cyber 170
machines re CMU. I have no knowledge of how such stuff worked in 60 bit
mode on the 180s. It's been 40 years after all.
--Chuck
3180
days inactive
3181
days old
1 comments
2 participants
participants (2)
-
cclist＠sydex.com -
jlw＠jlw.com