Presentations: When Light Speed Isn't Fast Enough

Originally held at assembly '05. Video version available here.

As the name suggests, it focuses on optimization techniques. I'm talking about things "mildly" architecture-specific, that is things common to nearly all modern architectures, without harping on specific details too much (so no assembly-language hacking :).

Slides

You can download the slides here.

References / Further Reading

Topics in order:

• Memory/Cache optimization
  • I don't know good general articles on this, but both Intels and AMDs optimization manuals are helpful guides if you're coding on x86 Platforms; those are links for the 32-bit parts. And Intel seems to be fond of changing their website directory structure every 2 months or so, so if the link dies, just drop me a mail.
• Fast rot-zooming
  • Well, there's always pascal's original source code. As already hinted at in the seminar, take the description with a grain of salt (current caches are much bigger and better), but this is still a classic.
• AGP Writes
  • The ultimate reference on different types of memory, their characteristics and how to use them properly is the IA-32 Intel® Architecture Software Developer's Manual, Volume 3 (Chapter 10).
• Radix sort
  • Submissive's Radix-Sort tutorial is a nice introduction to the topic.
  • Zappy's Radix Sort Revisited discusses how to do Radix Sort with floats (although there is a simpler way to do it).
  • Michael Herf's article on the topic describes ways to make it simpler (and even faster!)