"Nobody programs in machine language" (was: Modern

From: Sean 'Captain Napalm' Conner <spc_at_conman.org>
Date: Tue Jun 22 00:53:09 2004

It was thus said that the Great Fred Cisin once stated:
>
> And, in response to your STRAWMAN argument,
> NO, you can NOT "excel in driving" without
> an understanding of how a car works.
> You can get by as an adequate driver,
> but you will NEVER "excel" if you don't understand the engine.
> [Yes, I CAN build an engine from scratch. But I have always
> built them from existing auto parts. I have extensive
> professional automotive experience.]

  I would expect that an expert driver would need to have an understanding
of how the engine relates to the rest of the car, in addition to
understanding how the transmission and breaks relate. Yes, understanding
how the car works overall helps, but the ability to rebuild an engine? I
have to be sold on that one ...

> I will concede a total lack of experience with MODERN games.

  In 1992 when Doom came out, it only had two (2) routines hand coded in
assembler---the rest was in C.

> In the "vast array of criteria", which kinds of optimization
> do you think are not helped by an understanding of machine language?

  Algorithmic optimizations, which is where you now get the most benefit.
An extreme example would be replacing a bubble sort with quick sort (for a
sufficient amount of items to be sorted). I remember several years ago
working on a project (in C) that wasn't as fast as the previous
implementation (also in C). Instead of reaching for assembly (which I could
have done) I instead profiled the program---finding out which parts of the
program where the actual hot spots instead of my guessing where the hot
spots were. I basically converted seven small routines into macros (inline
the code, instead of having a function call overhead) and easily doubled the
speed, bringing it into line with the previous implementation (which had
fewer features by the way).

  Was I helped with my knowledge of assembly? Possibly in writing the
intial code. With the optimization (for speed)? I don't think so. Do I
regret learning assembly? Not at all, I think it made me a better
programmer, but that's me.

  I think the reason assembly isn't taught anymore is that industry (or
academia) don't see it worth the time and effort to. Right now, the most
expensive component in programming is the programmer, and anything to make
the programmer more productive is seen as a good thing, and the one thing
that makes a programmer productive is working with higher and higher level
languages.

  Another example. A project (play project really) I worked on I started
with Perl (one to learn the language). Text munging language and it took
perhaps eight hours total to get it to the point I like it. But its
sluggish on my development system (166MHz AMD 486). I then spent perhaps
twice the time (16 hours, maybe 20) rewriting it in C (because I can). Vast
improvement in speed---two, three orders of magnitude but part of that comes
from over a decade of experience in C, and two, over a decade of experience
in Unix systems programming, and using a trick or two (like mapping a file
directly into memory instead of reading each line, for one thing). But ...
my programming expertise isn't exactly cheap and in a company setting,
unless there were extenuating circumstances, the Perl version would be fast
enough on modern hardware (where on modern hardware, it's the hardware
itself that gives the two, three orders of magnitude). And the money paid
me to write it in C can instead be put towards faster hardware.

  I would contend that it's cheaper to buy the faster hardware than to keep
me on staff (or rather, it's more cost effective to have me doing other
things than translating Perl into C). And now a days, who can afford to
have a C programmer on staff? A friend of mine works at Sportsline (2nd
only to ESPN for sports related news on the web). They have a Perl program
to generate stats from the web server logs and to process a single day it
takes nearly 22 hours. Sure, they could hire a programmer to rewrite their
log processor in C and (depending upon the programmer and the complexity of
the processing) reduce the runtime of the program. But then they have to
keep someone around that knows C (or C++ or ... ) and that's an expense they
probably don't want to incure. Perl programmers are cheaper, as is
hardware.

> What do you think "optimization" means?

  Trade-offs. Right now, the optimization being done is developer time and
expertise. Faster with less skilled programmers. Or in other words:

        Good. Fast. Cheap. Pick two.

  -spc ("Premature optimzation is the root of all evil."
                ---Donald Knuth)
Received on Tue Jun 22 2004 - 00:53:09 BST