I’ve been programming a bit lately in C++ Boost for two reasons (1) I find the template metaprogramming aspects intriguing and the generic programming aspects super-convenient. Yet, when I’m really concerned with knowing precisely what’s going on to avoid wasting CPU cycles and memory, I go back to C. But then of course C is doing all sorts of things for me as well (just see the difference in speed in your code with the -O3 optimization flag).
One trend I’ve noticed after reading a lot of code is that bit-twiddlers will often over-optimize. I’m more and more convinced that this not only affects the readability of the code, but it may also be indicative of a lack of trust in the compiler’s optimization. The more control you have, the better you feel, right?
On the other end of things, some people trust objects and functions as black boxes and could care less what’s going on underneath. This post is not for them.
I hate to pull the big CS “language” and “compiler” words out of my pocket again , but how can I trust you language and compiler? When do I know I’m over-twiddling and it’s better that the compiler do the work for me? While there are guidelines, I think what it comes down to is a decent understanding of the compiler you’re dealing with and then, basic experimentation.
Most of my questions about whether certain optimizations are actually occuring are typically answered by doing timing tests, creating little “experiment” code bits to test a simple hypothesis, or look directly at the assembly code (e.g. the “-S” flag in gcc).
This did get me thinking, though, with so many different compilers and architectures–even with standards and benchmarks–a lot of these little questions aren’t answered. This process of creating precise tests and getting the answers seems to be the best approach .
 But I do believe these topics have some of the most interesting research opportunities from a CS perspective. As I foray into scientific computing and its applications to the sciences, I hope I will have the chance to tackle a CSee problem or two along the way.
 Wouldn’t it be nice if there was a nice repository of contributed tests for a variety of architectures, compilers, and languages, and that these tests can be tagged with all this information? In the spirit of my last post, I say let’s start a massive conglomerate blog where people can post these tests. And then there could be another one where people just bitch about the goodness and badness of the tests. It’ll be like a somewhat moderated coding newsgroup for today’s age. Oh how I’ve blurred the line between honest desire and sarcasm here.