IP: More on " The SWAC"

[Dave Farber <farber () cis upenn edu>]

Date: Thu, 07 Jan 1999 21:44:28 -0500
To: farber () cis upenn edu
From: Steve Crocker <crocker () mbl edu>

Dave,

You might add that the SWAC was NOT retired in 1957.  I started hanging
around UCLA in 1960 and was fortunate enough to learn how to program the
SWAC.  I don't remember exactly when it was turned off and dismantled, but
I'd guess it was closer to 1967.

The SWAC had 256 words of high speed memory implemented with Williams
tubes.  Each word was 37 bits -- 36 bits of data plus a sign -- and I think
there were some spare bits.  I think it came to forty tubes.  The bank of
tubes was well above six feet high and several feet wide.  Memory cycle
time was pretty fast for the era, but I can't remember whether it was 16
usec or 64 usec.  Instructions used 36 of the 37 bits.  Four bits for the
op code and four eight-bit addresses.  The add instruction was "add alpha
to beta, store the result in gamma and transfer (jump, branch) to delta if
the addition overflowed.  The operations were addition, subtraction,
multipication and a couple of logical operations.  No division, no floating
point, no index registers, no indirect addressing, no subroutine calls, no
stacks or list processing.  Certainly qualifies as one of the early "RISC"
machines, eh?  Programming was done in hexadecimal (0..9, u..z).  No
assembler.  No compiler.  Debugging was done by single stepping the machine
and reading the contents of memory in lights at the console.

256 words was obviously pretty tight.  There was also an 8K drum by the
time I got there.  With the limited instruction set, very small memory size
and extremely primitive tools, the programming style was considerably
different from today's object-oriented productions.

UCLA eventually got an IBM 709, which was succeeded by a 7090, 7094, and
360/91.  An awful lot of code was built for the SWAC, so the UCLA folks
built a SWAC simulator for the 709.  For a while, the 7094 was in the next
room down from the SWAC and they were both in use.  Most of the pioneer
programming for the machine had been done before I got there and I'm not
fully up to speed on it, but I think there was important work on factor
analysis (statistics), numerical analysis, and computational number theory.

Steve

----------------------------------
Steve Crocker Associates, LLC                   Bus:   +1 301 654 4569
5110 Edgemoor Lane                              Fax:   +1 202 478 0458
Bethesda, MD 20814                              crocker () mbl edu