Architectural implications of locate mode I/O

After our recent silly arguments about locate vs move mode I/O, I got
to thinking about what a computer needs for locate mode even to be
interesting.

Early computers had tiny memories and rudimentary I/O.  When doing
an I/O operation the CPU was too busy servicing the I/O device to
do much computing.  Hence the normal approach was to do the read
or write directly into the memory where the program used it,
no buffering needed.  OS/360 still had that with BSAM in which
you did a READ or WRITE macro to

The 709 introduced data channels in 1958 which allowed the CPU to do
other stuff while the channel did the I/O. Wikipedia says the first
I/O interrupt was on the NBS DYSEAC in 1954 but it's hard to see how
an I/O interrupt would be of much use before channels. Once you had a
channel, I/O buffering made sense, have the channel read or write one
area while you're working on the other.

The other thing you need to make locate mode useful is index
registers, since without them, it's about as hard to change the
instructions to point to the buffer as to move the data so you might
as well move the data.

The IBM 7070 was a short lived machine with 10 digit fixed length
decimal words. It had channels and index registers in locations in low
memory which it called index words. Its IOCS could run several devices
at the same time, e.g. two tape drives. Once you had your tape file
open, you used GET and PUT macros. Each had a locate form that set an
index word to point to the record, and a move form that copied the
data to or from your own work area.  PUT also had some other
options like writing a record just read from one file onto
another.

https://bitsavers.org/pdf/ibm/7080/C28-6237_7080_IOCS80_1962.pdf

The 7090 was a 36 bit binary machine with index registers and indirect
addressing and channels and interrupts. Its IOCS had read and write
calls that patched the addreses of the record areas into words
following the calls. You could them use those as indirect addresses or
load them into index registers. That makes sense since there were only
three index registers but you usually had more than three tapes going.

https://bitsavers.org/pdf/ibm/7090/C28-6100-2_7090_IOCS.pdf

So it looks like as soon as the machine architectures made it
practical to have overlapped I/O and efficient ways to do indirect or
indexed addressing, I/O systems took advantage of it by passing buffer
pointers to user code.

--
Regards,
John Levine, johnl@taugh.com, Primary Perpetrator of "The Internet for Dummies",
Please consider the environment before reading this e-mail. https://jl.ly