Wrap-Up and Revision
(2021)

A Weird One

So, this "lecture" is going to be a bit unusual...

...it's partially a "grab bag" of bits and pieces that I didn't/couldn't squeeze in elsewhere and, as such,
it will be a bit less structured and somewhat
more "random" than usual.

Optimisation

What's that?

Well, technically, it means "making optimal"
i.e. making a thing as good as it can possibly be.

That's a bit of a stretch, really!

In practice, what we actually mean by it is this:
"To take something that works... and make it work better."

For "some value of better" e.g. faster, smaller, cheaper.

Faster

Often (but not always) we are trying to "optimise"
to make things faster i.e. optimising for time.

In the "olden days" this was relatively simple.
The procedure was roughly as follows:

  1. Find a better algorithm
  2. Implement it
  3. Look for redundant code
  4. Remove it

But, on modern systems, it's no longer as easy as that...

Cache is King


Caches are a hugely important part of modern computer systems, and you really ought to be aware of them.

The simple version is that a cache is just a smaller,
higher-speed area of memory in which important data 
can be temporarily stored (yes, I'm simplifying!)

In fact, it's "Caches All The Way Down!"

Caches All The Way Down


Your Point Being?


It turns out that the issue of whether your data
(and, indeed, your code) is in one of these cache layers
has a huge impact on runtime performance.

Main memory access speeds can actually be hundreds
of times slower than our CPU clock rates these days!

Caching often dominates any other performance factors
(including algorithmic ones!)

But caches are strange and unpredictable beasts...

Taming The Beast

The fact that caches are (largely) automatic,
history-dependent, and not directly under your
control, makes them difficult to reason about...

So, today, the advice is simple: "Always Measure"

If you aren't actually measuring your performance,
you are working blind, and will make invalid choices.

Many techniques which were once "optimisations"
(e.g. loop unrolling) can sometimes be "pessimisations"
in the presence of caches! So be careful.

Measure Twice, Cut Once?


In the land of Hard Cache, a good strategy is to start by implementing a simple, and obviously correct, version
of something and then get some measurements on it.

Then (and only then), if it isn't as fast as it needs to be,
look at algorithmic changes which might improve it
e.g. moving from O(N^2) to O(N), and redundancies
which could be eliminated e.g. stuff that isn't changing.

The "simple" version can then serve as a perf. baseline,
and a correctness check, for the "complex" one.

Profiling Tools


You can actually do decent performance analysis in the browser these days: modern browsers now have quite fancy "call-graph tracing" facilities which you can use to explore what's really going on in your JavaScript code.

In other contexts (e.g. C++ stuff) you often have to write your own call-tracking infrastructure (my preferred approach),
or use 3rd party tools which do something similar
(usually with higher overheads though).

It can be a pain (PC timers are crap!), but it's always worth it.

Smaller


You might also have to find yourself optimising for
memory usage. Often, the strategy here is one of avoiding redundancy (don't store the same data in multiple places,
and consider whether it's better to recompute than to store!)

There are various ways of packing data into more compact formats too, but I won't go into them here.

I'll just tell you one evil trick, which some people use...

Prepared Heroism

In game development, memory is usually at a premium,
and is always being stretched to the limit by the demands
of the "content" people. Being forced to cut "completed" features due to lack of memory is painful for all involved.

Having seen this once too often, some (slightly evil) programmers have adopted the following tactic:

On day one of a new project, preallocate a secret, unused, Megabyte (or 2?) somewhere in your personal diagnostic routines. Later, when "all is lost", take it out again, sit back,
and savour the accolades for your heroic "optimisations".

Object Orientation


This whole course has taken a broadly OO approach,
but that's not the only way to go...

I like OO, but some detractors point out that it is
somewhat prone to some inefficiencies, and that a
"Data Oriented" way of thinking has advantages.

Parallelism


I haven't really addressed parallelism in this course,
but modern hardware is increasingly multi-core,
so high-end games have to think about how to
utilise such hardware efficiently.

It's another huge and complex problem,
which is a bit too hard for an introductory course...

But here's a talk about it (look at 30:00 and 35:50):
Parallelizing the Naughty Dog Engine

What's in the Exam?

I'm not telling you.
That would be cheating.

Also, it's sort of the wrong question.
The map is not the territory, and the exam is not the course.

Learning the subject is the important thing.
The exam is simply a crude instrument to assess that.

Concentrate on the former, and the latter will take
care of itself. Also, you will actually learn something.
...in fact, from what I can tell, you already have! Yay!

Chomsky On Exams




SpongeBob on Exams


Shape of the Exam

There are around 10 or 12 main areas in the course material, and I'll probably ask 4 or 5 questions about a subset of these.

I've decided to put more weight on the course work this year, so the exam will only count for 1/3 of your total mark.

There might be a little bit of coding involved,
but nothing too tricky, and I won't be pedantic about minor syntax errors, of the kind that would be caught easily if you were working on a real machine. If you've been able to do
the homeworks, you should be able to handle this too.

How To Revise

Read the slides, dummy!

In each main topic, there are a bunch of key principles,
and a cluster of related side-issues. You should learn these.

I don't expect you to memorise fine points of detail if they
are mere "factoids" that you could look-up when needed...

...but I do want you to understand orders of magnitude and rough chronologies e.g. knowing that Space Invaders was made in the 70s (not the 60s or the 80s) and that it ran on a 1MHz machine (not 100Hz; not 10MHz). You get the idea.

If...

If you've been paying attention and doing the homeworks, and if you spend some time going over the slides again,
then you should do just fine...

Any Questions?

After You Have Passed

...you may still want to keep up with the subject.

At least, I hope you will. Games provide a nice, interesting angle on Computer Science as a whole. They can sometimes be pretty "cutting edge" in fact, and are quite a fun way of staying in touch with developments across the field.

If you take the idea of being a programmer seriously, you'll want to go on learning new things and growing your skills.

The Internet is a godsend when it comes to this!
It's so easy these days, to learn almost anything...

A Way In...

In addition to just searching for things of interest, it can be useful to hook into some of the better online "communities" as a way of seeing what's going on, and discovering new things.

Twitter, for all its flaws, is actually pretty good for this.

Here's a decent list of people (and "entities") to follow.

It contains a mixture of the serious and the not-so-serious,
and I don't necessarily approve of, or agree with, all of them
(in fact, some of them are "old farts") --- but they are there for
a reason, and provide an interesting perspective on things.

Your Projects


I've been very pleased with the standard of the projects,
and the fact you've been able to make these things
in just three weeks, with just four people per team.

I hope you learned something from the exercise,
and that it was a relatively enjoyable thing to do.

Examples




More Examples!



Student Game Award?


If you want to try for it,
"Apply online before December 31st"

What Next?

If you're interesting in continuing to program computer
games (or anything else really), my main suggestion
at this point would be to "make another one".

If you tackle something a little bit different from what
you've already done in the course, it'll help you to extend
your capabilities and make you a bit more fluent.

Another thing worth trying would be to try revisiting some
of what you've done in the course in a few months, and to check that you can still do it e.g. try doing "Breakout" again.

Taking it to the Next Level

If you want to go even further, there are a couple of things
you could pursue: an obvious one would be to look at 3D.

I claimed at the start of the course that this was "easy",
and was really just a matter of "adding another D".

That may only be partially true. ;-)

3D is, of course, a big subject... but the core of it is not that complex, and you can take a course on it here, I believe.

You can even do it in JavaScript now, with WebGL.

Speaking of JavaScript...

How did you find learning/using it?

I think that, despite some odd warts here and there,
it's a pretty decent language, and its importance
on the Web can't be denied.

Having said that, if you really want to push things performance-wise, you might have to look elsewhere.

The main language for high-end games in still C++,
(although I think there's room for innovation there now).

Finally...This


King Solomon once played a trick on his aides by asking
them to locate a mythical (non-existent) "magic ring",
which had the mysterious property of being able to
"Make A Sad Man Happy, and A Happy Man Sad".

After many months of searching, an aide returned,
bearing what he claimed to be such an ornament.

It was plain and simple, and not supernatural at all,
but it was inscribed with the following remark...

"This Too Shall Pass"

...as, I hope, will you... in the exam. :-)

Wrap-Up and Revision (2021)

By Pat Kerr

Wrap-Up and Revision (2021)

The End.

  • 764