I loved this book and gutted to have lost the real life version I used to own. But even so someone's converted the entire thing to Markdown and also made it available as HTML, a format a lot easier to read than the PDFs originally released.
I find maths a bit tricky so an understanding how rotation matricies actually work is something I'll always be greatful to this book for
It's best to view the majority of the content as a historical document these days. There's not a lot of real need for ninja tricks with VGA hardware, cycle shaving on old CPUs or how you'd write an efficient 3d renderer but it is fascinating stuff explained by a bright and lucid guy at the peak of his powers
Yeah I loved this book too. It correct a LOT of bad habbits I had picked up from reading a couple earlier books by Christopher Lampten and Andre lamothe (not the books fault - just my desire to run before i could walk based on a hunch).
I really like his little anecdotes at the start of each chapter. I like the story about Mode X with the lady at the book signing coming up to him and saying "Are you the man who invented Mode X...." "Yes...?" "No you didnt - my husband did!".
Technology has moved on - few people use VGA, 486s or Pentiums anymore. But it's still a cracking read, and it's still worth reading it to learn about how to think about optimisation.
It's really great to have it available again in a nice format instead of a pile of crappy scans of photocopies.
"I am a very considerate person. If I want to go out all night on my miniature golf course and shoot lizards, well, no reason to keep the neighbours awake..."
I don't think I've read it (were sections published in DDJ back in the day?), but I still marvel at how many people writing assembly code need to read it or something like it.
OTOH, optimisation can be a whole lot more rewarding when you don't test it.
Scanning through this I see something specific which, while true at the time, has become a bit of a misperception in modern times.
Loop unrolling and function inlining. They're discussed in terms of the handful of machine instructions that they remove from unoptimised code.
In the modern world with modern compilers and modern CPUs the far, far greater benefit of unrolling and inlining relates to the propagation of constants and the reordering of memory operations and register colouring and shit like that.
These are optimisations that can be a real pig when done by hand in assembly. It's a case where the compiler can often beat a human simply by being harder to wear out.
Does Abrash, or some other authority to whom I can refer people, discuss this in later publications?
gravy wrote:Yeah it's out of date to the point of practical uselessness as mentioned in the OP, but it's still interesting
Yeah, nostalgic; but looking back now I think Abrash had a big influence and I _think_ I can see where these ideas I want to fix might have entered the public consciousness.
And, of course, when it was said it was correct. Just need someone as influential to say the new correct thing today.