26 Oct
2010
26 Oct
'10
11:39 p.m.
On 2010-10-27 06:20, William Donzelli<wdonzelli at gmail.com> wrote:
I honestly don't know.
The PDP-11 have been attributed with the common I/O and memory bus
(Unibus), with memory mapped I/O as well as the concept of condition
codes. Also the general registers with a nice set of addressing modes to
use on them. And we should probably not forget having the PC as just a
general register (although few, if any, picked that one up). So the
PDP-11 can be used as a accumulator-based machine, a memory-memory based
machine, or a stack based machine. It's possible to implement all
concepts found in architectures at that time easily on the PDP-11.
The basic PDP-11 architecture was deigned so that if an instruction took
an argument, any kind of argument was equally valid.
But as usual, the question is: was really the PDP-11 first with these
things, or can you find earlier examples?
Johnny
--
Johnny Billquist || "I'm on a bus
|| on a psychedelic trip
email: bqt at softjar.se || Reading murder books
pdp is alive! || tryin' to stay hip" - B. Idol
> 1.>
The PDP-11 was in architectural ways more important than the VAX, if
> ?> nothing else than just because the VAX was basically just extending the
> ?> PDP-11.
Just to throw another question into the fire:
Just how important was the PDP-11 or VAX-11/780 hardware architecture
in the grand scheme of things? Did either machine really bring
anything new to the table? 
27 Oct
27 Oct
1:27 p.m.
-----Original Message-----
From: cctalk-bounces at classiccmp.org [mailto:cctalk-
bounces at classiccmp.org] On Behalf Of Johnny Billquist
Sent: Tuesday, October 26, 2010 11:40 PM
To: cctalk at classiccmp.org
Subject: Re: Happy Birthday VAX 11/780 (influence of)
On 2010-10-27 06:20, William Donzelli<wdonzelli at gmail.com> wrote:
I honestly don't know.
The PDP-11 have been attributed with the common I/O and memory bus
(Unibus), with memory mapped I/O as well as the concept of condition
codes. Also the general registers with a nice set of addressing modes
to
use on them. And we should probably not forget having the PC as just a
general register (although few, if any, picked that one up). So the
PDP-11 can be used as a accumulator-based machine, a memory-memory
based
machine, or a stack based machine. It's possible to implement all
concepts found in architectures at that time easily on the PDP-11.
The basic PDP-11 architecture was deigned so that if an instruction
took
an argument, any kind of argument was equally valid.
But as usual, the question is: was really the PDP-11 first with these
things, or can you find earlier examples?
I do believe the PDP-11's use of memory-mapped I/O was original - at least I can't
think of any earlier examples. -- Ian
> > 1.> The PDP-11 was in architectural
ways more important than the
VAX, if
> > ?> nothing else than just because
the VAX was basically just
extending the
>
?> PDP-11.
Just to throw another question into the fire:
Just how important was the PDP-11 or VAX-11/780 hardware architecture
in the grand scheme of things? Did either machine really bring
anything new to the table? 
1:39 p.m.
On 10/27/10, Ian King <IanK at vulcan.com> wrote:
It's the earliest one I am aware of, but my knowledge of non-DEC stuff
before 1970 is admittedly full of gaps.
I started with the 6502 (in 1977), so I thought memory-mapped I/O was
"normal". The 1802 was the first processor I used that had I/O
instructions.
So I guess since any processor that has gaps in its memory map _could_
have those gaps filled with memory-mapped I/O, the question is _did_
any implement it prior to 1969/1970?
-ethan
But as usual,
the question is: was really the PDP-11 first with these
things, or can you find earlier examples?
I do believe the PDP-11's use of memory-mapped I/O was original - at least I
can't think of any earlier examples. -- Ian 
1:57 p.m.
On 27 Oct 2010 at 16:39, Ethan Dicks wrote:
It's the earliest one I am aware of, but my
knowledge of non-DEC stuff
before 1970 is admittedly full of gaps.
Consider, for example, the CDC 7600. Each peripheral processor (or
"channel" in alternative terminology) is hard-assigned a buffer in
SCM. The CPU has no way to talk its channels other than by writing
into these memory-mapped buffers. There are no CPU I/O ports or any
other way for the CPU to reach the outside world. The 6000 series
behaves similarly, but the PPUs get to read any memory address.
Typically, a "mailbox" is set up for PPU communication.
So, did Seymour invent memory-mapped I/O?
I suspect not.
--Chuck
5:28 p.m.
Chuck Guzis wrote:
Consider, for example, the CDC 7600. Each peripheral
processor (or
"channel" in alternative terminology) is hard-assigned a buffer in
SCM. The CPU has no way to talk its channels other than by writing
into these memory-mapped buffers. There are no CPU I/O ports or any
other way for the CPU to reach the outside world. The 6000 series
behaves similarly, but the PPUs get to read any memory address.
Typically, a "mailbox" is set up for PPU communication.
So, did Seymour invent memory-mapped I/O?
I suspect not.
I'd consider a channel using a shared memory area as a type of
I/O
distinct from either port-mapped or memory-mapped. And the CDC 7600
(and 6600) certainly didn't invent the idea of a channel.
2:20 p.m.
> I do believe the PDP-11's use of
memory-mapped I/O was original - at least I
> can't think of any earlier examples. -- Ian
On Wed, 27 Oct 2010, Ethan Dicks wrote:
It's the earliest one I am aware of, but my
knowledge of non-DEC stuff
before 1970 is admittedly full of gaps.
I started with the 6502 (in 1977), so I thought memory-mapped I/O was
"normal". The 1802 was the first processor I used that had I/O
instructions.
So I guess since any processor that has gaps in its memory map _could_
have those gaps filled with memory-mapped I/O, the question is _did_
any implement it prior to 1969/1970?
My memories from ~1968 are not very clear, and definitions of
memory-mapped I/O may vary, . . .
Didn't the 1401 sort-of have it? Not fully automatic memory-mapped I/O
such as the TRS80 and PC video RAM, more like the CP/M and PC's PSP DTAs.
I seem to recall there being a command to read a card into a fixed buffer
in memory, and another to write a card from anbother fixed buffer. I
remember "cheating" and using those buffer spaces when I ran short of
space writing short programs on the 1401 emulator on the 1620.
Strange, . . .
I remember more about that than I do about PDQ FORTRAN on the 1620, nor
about "141 SPS"? on the 1401.
--
Grumpy Ol' Fred cisin at xenosoft.com
2:41 p.m.
New subject: 1401 "memory-mapped I/O" [was RE: Happy Birthday VAX 11/780 (influence of)]
From: Fred Cisin
Sent: Wednesday, October 27, 2010 2:21 PM
>> I do believe the PDP-11's use of
memory-mapped I/O was original - at least I
>> can't think of any earlier examples. -- Ian
On Wed, 27 Oct 2010, Ethan Dicks wrote:
> It's the earliest one I am aware of, but my knowledge of non-DEC stuff
> before 1970 is admittedly full of gaps.
> I started with the 6502 (in 1977), so I thought memory-mapped I/O was
> "normal". The 1802 was the first processor I used that had I/O
> instructions.
> So I guess since any processor that has gaps in its memory map _could_
> have those gaps filled with memory-mapped I/O, the question is _did_
> any implement it prior to 1969/1970?
My memories from ~1968 are not very clear, and
definitions of
memory-mapped I/O may vary, . . .
Didn't the 1401 sort-of have it? Not fully automatic memory-mapped I/O
such as the TRS80 and PC video RAM, more like the CP/M and PC's PSP DTAs.
I seem to recall there being a command to read a card into a fixed buffer
in memory, and another to write a card from anbother fixed buffer. I
remember "cheating" and using those buffer spaces when I ran short of
space writing short programs on the 1401 emulator on the 1620.
The way 1401 addressing works, you address the high-numbered character and
everything is operated on towards the first lower numbered character with
the Word Mark bit set. Locations 80, 180, and 332 are special: A MOVE
instruction pointed at location 80 as the source reads a card from the 1402,
a MOVE with location 180 as the destination punches a card on the 1402, and
a MOVE with 332 as the destination prints a line on the 1403. Word marks
are set in locations 1, 101, and 201.
I don't know how the disks (or tapes, if the system had them) handled things.
I saw Autocoder as a losing proposition by then.
(Lord, that's dredging up 40+ year old memories. I stuck to the 360/370
family once I graduated high school, until I encountered the DEC-20 eight
years later in grad school.)
Rich Alderson
Vintage Computing Sr. Server Engineer
Vulcan, Inc.
505 5th Avenue S, Suite 900
Seattle, WA 98104
mailto:RichA at vulcan.com
mailto:RichA at LivingComputerMuseum.org
http://www.PDPplanet.org/
http://www.LivingComputerMuseum.org/
1 Nov
1 Nov
10:31 p.m.
On 10/27/2010 4:20 PM, Fred Cisin wrote:
My memories from ~1968 are not very clear, and
definitions of
memory-mapped I/O may vary, . . .
Didn't the 1401 sort-of have it? Not fully automatic memory-mapped I/O
such as the TRS80 and PC video RAM, more like the CP/M and PC's PSP DTAs.
I seem to recall there being a command to read a card into a fixed buffer
in memory, and another to write a card from anbother fixed buffer. I
remember "cheating" and using those buffer spaces when I ran short of
space writing short programs on the 1401 emulator on the 1620.
If my dim memories are correct that is correct for the 1402 card reader
and punch connected to a 1401. I'm pretty sure the 1403 printer also
had a separate fixed address output buffer as well. I think mag tape
could be read to or written from anywhere. I think "read a card",
"punch a card" and "print a line" were pretty much single character
instructions, though I think there were additional modifier characters
in some cases for things like pocket selection, etc. I could go dig out
my programmer's reference card if anyone really cares, which I seriously
doubt. :-)
The earliest example of memory mapped I/O in a general purpose sense for
me was the Motorola 6800, but Wiki says that was introduced in 1974 and
I suppose that is too late for what we were discussing. ISTR in our
hardware design the ACIA's etc. were just decoded at some memory address
range in the 6800's memory address space, and one used an ordinary
"store register contents at memory address" instruction to write to the
peripheral registers for example.
Strange, . . .
I remember more about that than I do about PDQ FORTRAN on the 1620, nor
about "141 SPS"? on the 1401.
Ah yes, good old Symbolic Programming
System, I think. I don't recall
seeing specifically the terminology "141 SPS", but it has been quite a
while! SPS was just an assembler as I recall. I don't believe there
was a linker. I think every program just had an absolute start address
(usually above the printer buffer, though not required if you wished to
"cheat") and SPS punched an executable deck that could then be loaded
and executed. If your program never punched or printed, you could start
your code just above the card reader buffer which I think was at 0-79.
Later,
Charlie C.
--
Grumpy Ol' Fred cisin at xenosoft.com
2 Nov
2 Nov
6:10 a.m.
On 2 Nov 2010 at 0:31, Charlie Carothers wrote:
Ah yes, good old Symbolic Programming System, I think.
I don't recall
seeing specifically the terminology "1401 SPS", but it has been quite a
while! SPS was just an assembler as I recall. I don't believe there
was a linker.
SPS was the name of the assembler on a few other IBM systems; also
the 1620. Later versions had a linker so you could combine SPS with
FORTRAN. The 1311 disk added a lot of capability, including JCL of
sorts. (starting decks off with record-mark record-mark JOB, etc.).
The options control statements to SPS, I recall, were
uncharacteristically non-terse. e.g. "PUNCH SYMBOL TABLE".
No memory-mapped I/O on the 1620, unless you want to count pressing
the LOAD button that would (numerically) read a card or paper tape
record into 0-79. It's odd, but I can still remember most of the
machine opcodes and an I/O instruction or two (e.g. 34 00000 00107,
ISTR was the command to return the carriage on the console
typewriter). Now, I can't remember where I left my glasses two
minutes ago...
I suspect that much of the same talent that wrote the system software
for the 1401 also did it for the 1620.
--Chuck
5164
days inactive
5170
days old
8 comments
8 participants
participants (8)
-
bqt＠softjar.se -
cclist＠sydex.com -
cisin＠xenosoft.com -
csquared3＠tx.rr.com -
eric＠brouhaha.com -
ethan.dicks＠gmail.com -
IanK＠vulcan.com -
RichA＠vulcan.com