Optimizing the Game of Life -- Michael Abrash

Date view Thread view Subject view Author view

From: Jeff Barr (jeff@vertexdev.com)
Date: Thu Jan 25 2001 - 23:01:32 PST

Best quote "The Ferraris are only Gravy, Honest!" -- John Carmack
Next best "Shipping software is an unnatural act."

This appeared in Gamasutra

Optimizing the Game of Life

Like most of you, I suspect, I just plain like writing software-I did it for
fun before I did it for a living-and one of the best things about game
programming is the opportunity to do challenging programming in a broad
range of areas, from graphics to physics to AI to you-name-it. Another is
that the game industry changes so fast and is so close to the leading edge
of consumer technology that it's a terrific opportunity to ride the wave of
the future. The rest of this paper presents a number of principles and
approaches to writing software and riding that wave that I've found to be
helpful and/or enlightening over the years. At the risk of sounding like
"Everybody's Free (To Wear Sunscreen)" (the alleged Kurt Vonnegut graduation
speech that became a novelty hit when recited over a musical background),
I'm going to present just the core points here, as succinctly as possible.
This is basically just an outline, so you'll have some idea in advance of
what to expect from the talk, and to jog your memory when you pick up the
proceedings later. The talk itself will expand upon and illustrate the
points, complete with anecdotes and examples ranging from working on Quake
to working on Windows NT to writing "the last great software rasterizer" to
speeding up the Game of Life.

Herewith, some thoughts on fast code, game programming, software
development, and more:

The three secrets to success are:
1) Hard work,
2) Hard work, and
3) Hard work.

Learn by doing. Thinking is great, but the only way you ever really learn
something thoroughly is by doing it-and particularly by finishing it, which
is harder than it sounds because shipping software is an unnatural act
(thanks to David Stafford for the phrase). Accept reality: no software is
ever going to be perfect, and perfection is a moving target anyway. Make the
necessary tradeoffs and get it out the door.

Success is more than just a big paycheck-or, as John Carmack put it, "The
Ferraris are only gravy, honest!" You're going to be financially solvent-or
better-anyway, so why not do what you love? If you do what you love and work
hard and passionately, you're building the foundation of a rewarding life,
and in the long term you'll succeed in every way that's important to you.

If you like it, game programming is about the best software job there is. It
touches on practically every area of programming, and it's one of the few
places where code can still be written clean instead of patched over or
pieced together from canned parts. It's also one of the few places where
optimization still really matters.

Aim high, think big. Right now is a particularly good time for ambitious
game programming, because so much more is possible now than ever before,
thanks to CPU performance and 3D accelerators.

Great software is more than just great code. First, a great design is
needed, before the code can be written. Don't write code until you have a
clear model in your head of what you're trying to accomplish. Take the time
to get this right, and don't jump on the first workable idea you come up

Design and optimize from the highest levels of your model on down, focusing
on the big architectural issues first. Premature optimization-including
premature low-level design--is the root of all evil.

Keep looking for new, better approaches, and be ready and willing to change
the model constantly as you learn more-or even throw it away entirely. Until
you start running out of time, think of whatever design you have as just a
jumping-off point for a better one.

Get past the abstractions and know what your code does. Profile in many
ways: profilers, NULL drivers, bracketing key code with timing calls,
breaking into the debugger at random and seeing where you are. If you don't
do this, you're guessing.

Knowing what to optimize matters as much as knowing how to optimize.
Otherwise, you'll optimize the wrong thing, and end up with really fast slow

When it seems you've made your code as good as it can be, and you still need
more from it, try to change the way you're seeing the problem. You may be
solving your abstraction of the problem rather than the problem itself.
Terje Mathisen says, "Almost all programming can be viewed as an exercise in
caching." David Stafford views almost all programming as a form of
compression. There are infinitely many programming perspectives available to
you-use them.

Check your assumptions.

Assume nothing. Always ask yourself questions and challenge your
assumptions, and just laugh and be glad for what you've learned when you
find you could have done something better.

Stand on the shoulders of giants. There are few brand-new ideas in the
world; read what others have done rather than trying to invent everything
yourself. Most of programming success lies in engineering and integrating
and finishing and being flexible, rather than in raw invention. Sure, it's
fun to reinvent things, but there's a huge amount you have to know now, so
it's hard to do it all on your own. Besides, there's plenty of cool stuff to
figure out after you learn what's already been solved.

Learn as much math as you can stand. Then learn some more. The industry is
moving from broad, seat-of-the-pants programming to software more dominated
by content and deep knowledge of areas like physics. Programming itself is
easy; it's expertise in knowledge-intensive areas that will increasingly set
programmers apart in the future.

Be patient. It takes one year to become competent at something new, three
years to become an expert.

Stretch until it hurts. Be flexible. Don't get attached to your designs or
code. Be willing to listen to new viewpoints, admit it when you're wrong,
and change your mind as appropriate.

Embrace change… or watch it run you over and leave you behind. Accept that
the goals, techniques, and rules will constantly change.

The only thing standing between you and great software ideas is you. Think
past the limits you've unconsciously set.

TANSTATFC: There ain't no such thing as the fastest code (with apologies to
Robert A. Heinlein and thanks to Terje Mathisen, who came up with the

Turn your right brain loose. Try lots of stuff, even wild stuff. Listen to
that little voice. Relax. Let yourself be creative. Accept that many ideas
will fail. This is the only way to learn and make real progress.

If your code is getting more complex, you're probably fine-tuning. Step back
and look at the bigger picture for more global solutions. Try to change the
rules so you can succeed, rather than straining to be mediocre under the
current constraints.

Don't get too comfortable. Aspire to excellence. Fight that satisfied
feeling about your code, and constantly ask how to improve. Fight code

Simplify. Try to do better with less, in a different way.

Programming is not a zero-sum game. Unless you have legal or employer
restrictions, share what you've learned, and learn from others. Having to
explain something forces you to know it cold. It also makes you known, and
leads to interesting opportunities.

Remember to have fun! It's about the best job around, but like anything
else, game programming can turn into a job if you let it. Don't let it.
You're one of the luckiest people in the history of the planet, getting
incredibly well rewarded for doing something challenging, fun, and just
plain cool. Remember that and count your blessings every day. Then get out
there and write some great software!


Jeff Barr - Vertex Development - (mailto:jeff@vertexdev.com)
  Address: 4610 191st Place NE. Redmond, WA 98074;
  Phone: Office: 425-868-4919 - Home: 425-836-5624
  Homepage: http://www.vertexdev.com/~jeff
  Weblog: http://jeffbarr.editthispage.com/
  Resume: http://www.vertexdev.com/~jeff/real_jb_resume.html

Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Fri Jan 26 2001 - 04:55:10 PST