From: peter@pmoylan.org
On 13/11/25 18:11, Mike Spencer wrote:
> kludge@panix.com (Scott Dorsey) writes:
>
>> Lawrence =?iso-8859-13?q?D=FFOliveiro?= wrote:
>>
>>> Funny how, in the early days (e.g. 1940s/50s/60s), many Comp Sci
>>> disciplines started within electrical/electronic engineering
>>> departments.
>>
>> In the seventies and eighties there were two kinds of CS programs,
>> the ones that came out of EE departments and the ones that came out
>> of math departments. They were often dramatically different in
>> their approach and it wasn't until the ACM curriculum of the
>> mid-1980s that this really changed and some degree of uniformity
>> appeared.
>
> 1964, UMass, a course called "computer programming" was offered.
> The instructor was seconded from industry, not regular faculty. He
> proposed to teach a math course in "numerical methods". Most of the
> class announced that they'd drop the course if he didn't teach how
> to actually program a computer rather than a math subject. His
> protests that you couldn't write a useful program if you didn't
> understand numerical methods were to no avail.
>
> So he bit the bullet and taught us (some version of?) Fortran which
> we happily tried out with punch cards on the school's IBM 1620.
>
> I had no further encounters with computers until I got a chance to
> dick around with BASIC on a friend's Apple ][ many years later.
My introduction to Fortran came at the end of 1967, which was at the end
of my third year of university. I got a vacation job at a research
organisation, and my project there (a problem in heat transfer) required
a lot of numerical calculation. My boss just said "Look in the library,
there should be a book on computer programming there." In his view (and
mine too) learning Fortran was just a matter of looking up the manual.
We didn't have punched cards; I had to use punched paper tape. That made
editing harder.
At the beginning of 1968, before university resumed, I was supposed to
do a 3-week Fortran intensive course. I asked for an exemption on the
grounds that I was already experienced in Fortran. The person who
assessed my program listings nearly refused the exemption, on the
grounds that I was using an inefficient way of computing Bessel
functions. (Not surprising. Before that vacation job I had never heard
of Bessel functions, and I couldnt't just google it.) In the end he did
concede that I knew how to use Fortran.
In the middle of 1968, one of our programming assignments was to work
out what value of a capacitance was needed to give a certain filter
circuit a specified bandwidth. An elegant solution to the problem was to
use a successive approximation method of solving an equation of the form
f(x)=0. That would have been easy, except that it turned out that
computing f(x) required solving another equation g(y)=0. That would have
been straightforward in a modern programming language, but Fortran IV
did not permit recursive subroutine calls. My contribution to elegance
was to figure out a way to emulate recursion. I don't know what the
person grading the assignments thought of that, but I did get the right
answer.
By the way, the way we submitted work was to put our card deck into a
box, where it was transported to the computer centre. The next day our
deck would come back, accompanied by a printout that was the result of
running our program. From that we had to figure out what we had done
wrong, and update the card deck. (That was a stock-standard approach
back when everything had to be done, one job at a time, by the computer
operator.) After a while I became aware that somebody else was copying
my card deck. So I completed my assignment solution -- a different
assignment from the one above -- and then saved that card deck. Then I
introduced some errors into my program, and continued to submit the
faulty program. The day before the result was due, too late for the
cheater to recover, I restored the correct solution.
Ah, memory. That same year I had to write up a minor thesis on the Fast
Fourier Transform. Today everyone knows what an FFT is, but back then it
was new research. I could meet the requirements by writing up the
history of the development of the FFT, and its major variants, but for a
good thesis I felt I should implement all those variants, and do things
like timing tests. But to implement the algorithms I needed a couple of
small functions coded in assembly language [1]. That was a radical idea
back when assembly language was felt to be the property of the high
priests. I had to have an interview with the head of the CS department,
where I had to explain what those functions were, why I needed them, and
how I proposed to implement them. I think he was upset that a mere
electrical engineering student would presume to do advanced programming,
but in the end he have me permission to use assembly language on his
sacred machine.
[1] That was then, this is now. Now I know how to do the job without
assembly language. (And without Fortran.) But I was a mere beginner in
those days.
History repeats itself. Just a few years ago, after my retirement from
the university, I was given the job of designing a smart high-voltage
protection relay. (The application was in coal mining, where the
electrical noise is at a level you wouldn't believe.) To extract the
higher-frequency components of the noise, I needed an FFT with an
execution time in the millisecond range. In the end I designed an FFT
implementation so efficient that I can't read my own code[2]. But it
worked, and the relay got an award.
[2] Method:
(a) Do an FFT implementation in a high-level language (I used Modula-2).
(b) Do some code optimisations, like unwrapping the outermost loop.
(c) Continue doing testing and tweaks to improve the code.
(d) Translate the result into C, which is what the customer wants.
The result is not very readable, but it runs like the clappers.
--
Peter Moylan peter@pmoylan.org http://www.pmoylan.org
Newcastle, NSW
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
