[TUHS] Re: early unix rand

[Clem Cole <clemc@ccc.com>]

[-- Attachment #1: Type: text/plain, Size: 5046 bytes --]

Prof Kahan's Floating Point Test Program - the original from his and his
students in his computer arithmetic seminar wrote during my days at UCB:
https://www.netlib.org/paranoia/
Kahan was always miffed at how bad the different floating point units were
- (Seymour was notorious for being fast but not very precise on most of his
FP units).
Here is an updated FORTRAN 90 version:
https://people.math.sc.edu/Burkardt/f_src/paranoia/paranoia.html
ᐧ

On Wed, Mar 13, 2024 at 4:25 PM Rob Pike <robpike@gmail.com> wrote:

> Norm Schryer wrote a (nearly?) exhaustive floating-point tester that he
> ran when a new CPU arrived, always with wrong results. Doug McIlroy
> probably knows more about it than I do, who only observed it from afar.
>
> -rob
>
>
> On Thu, Mar 14, 2024 at 4:18 AM ron minnich <rminnich@gmail.com> wrote:
>
>> Got the name wrong: Computer Engineering: A DEC View of Hardware Systems
>> Design
>>
>> On Wed, Mar 13, 2024 at 9:41 AM ron minnich <rminnich@gmail.com> wrote:
>>
>>> by the way, I realize that random number urban legend sounds ridiculous,
>>> in light of how hardware design is done today, but those of you who did
>>> hardware design in those days (guilty!), and had access to -11
>>> schematics and boards, might wonder if it's not possible. There was a
>>> habit, in those days,  for performance reasons, of subbing transparent
>>> latches for flip-flops to gain a little time. An engineer I knew at Amdahl
>>> said that was a pretty hot topic there. Certainly, the technique of design
>>> for testability was not really in wide use in the -11 days. Gordon Bell's
>>> book "Computer Design" is particularly instructive.
>>>
>>> E.g., how did you verify the floating point on your new machine? Put an
>>> older machine next to a new machine, do lots of computation, see if there
>>> is disagreement, you've found a bug in the new machine, right? Maybe.
>>> Sometimes,  you discover the older machine had a bug the newer one did not
>>> ... happened more than once, including on the 360 to 370 transition.
>>>
>>> On Tue, Mar 12, 2024 at 6:09 PM ron minnich <rminnich@gmail.com> wrote:
>>>
>>>> There used to be an urban legend about multiply overflow and the PDP 11.
>>>>
>>>> This would’ve been circa 1976. Someone from DEC told us that on a
>>>> multiply overflow, the contents of the destination register would be “kind
>>>> of” random. I was never able to verify that claim. But that might explain
>>>> this code.
>>>>
>>>> On Tue, Mar 12, 2024 at 16:05 Jonathan Gray <jsg@jsg.id.au> wrote:
>>>>
>>>>> On Tue, Mar 12, 2024 at 08:55:02AM -0400, Russ Cox wrote:
>>>>> > Hi all (and TUHS),
>>>>> >
>>>>> > The Third Edition rand(III) page [1] ends with
>>>>> >
>>>>> > WARNING  The author of this routine has been writing
>>>>> >     random-number generators for many years and has
>>>>> >     never been known to write one that worked.
>>>>> >
>>>>> > My understanding is that Ken wrote the rand implementation.
>>>>> > But I'm curious about the origin of this warning.
>>>>> > I had assumed that Ken wrote it as a combination warning+joke,
>>>>> > but Rob suggested that to him it didn't sound like Ken and
>>>>> > perhaps Doug or Dennis had written it. Does anyone remember?
>>>>> >
>>>>> > Separately, I am trying to find out what the very first
>>>>> > Unix rand implementation was. In the TUHS archives,
>>>>> > the incomplete V2 sources contain a reference to srand
>>>>> > in cmd/bas0.s [2], but there is no definition in the tree.
>>>>> > The V3 man pages list it, but as far as I can tell full
>>>>> > library sources do not appear in the TUHS archives
>>>>> > until the V6 snapshot. The V6 rand [3] is:
>>>>> >
>>>>> > rand:
>>>>> >     mov r1,-(sp)
>>>>> >     mov ranx,r1
>>>>> >     mpy $13077.,r1
>>>>> >     add $6925.,r1
>>>>> >     mov r1,r0
>>>>> >     mov r0,ranx
>>>>> >     bic $100000,r0
>>>>> >     mov (sp)+,r1
>>>>> >     rts pc
>>>>>
>>>>> matches V5:
>>>>> https://www.tuhs.org/cgi-bin/utree.pl?file=V5/usr/source/s3/rand.s
>>>>> Distributions/Research/Dennis_v5/v5root.tar.gz
>>>>> <https://www.tuhs.org/cgi-bin/utree.pl?file=V5/usr/source/s3/rand.sDistributions/Research/Dennis_v5/v5root.tar.gz>
>>>>>
>>>>> >
>>>>> > Perhaps this is the original rand as well? It is hard to imagine
>>>>> > a much simpler one, other than perhaps removing the addition,
>>>>> > but doing so would create a sequence of only odd numbers.
>>>>> > >From the man page description it sounds like this has to be the
>>>>> > original generator, perhaps with different constants.
>>>>> >
>>>>> > Thanks!
>>>>> >
>>>>> > Best,
>>>>> > Russ
>>>>> >
>>>>> > [1]
>>>>> >
>>>>> https://github.com/dspinellis/unix-history-repo/blob/Research-V3/man/man3/rand.3
>>>>> > [2]
>>>>> >
>>>>> https://github.com/dspinellis/unix-history-repo/blob/Research-V2/cmd/bas0.s
>>>>> > [3]
>>>>> >
>>>>> https://github.com/dspinellis/unix-history-repo/blob/Research-V6/usr/source/s3/rand.s
>>>>>
>>>>

[-- Attachment #2: Type: text/html, Size: 7852 bytes --]