CPUs are in the 3 GHz range now. Saving a cycle means saving one three-billionth (~0.000000000033) of a second. When computers's speed were measured in the KHz or <100 MHz, saving a few cycles might help in your innermost loop (which in those days was called 64k times per frame to put something interesting in each pixel on the screen, maybe alpha blend using a palette, etc). These days, don't count on it because processors are faster, code is run less often, and the GPU takes care of pixel processing. Not only that, but processors are more complicated these days and can optimize the machine code as the execute it (pipelining and instruction reordering etc). Unless you're an expert, the compiler knows more than you about how to make your code run faster in relation to microoptimizations. The only real code-level optimizations you can make these days are converting things to vector instructions, and even that is already tackled to some degree by certain compilers.

The kind of optimizations you should worry about are which algorithm fits your needs, which data structure has characteristics that match your use of the data structure, etc.

Quote:Original post by Demosthenes
Great answers all around, thanks.

iMalc, I made this post because of a question "Give me a fast way of multiplying a number by 7" in a "Frequently Asked Questions in Interviews" Doc.

From the answers here, I'm guessing that question is a bit old, from a time when processors were simpler and the optimizations made by compilers didn't cover shifting.

Oh, that makes sense then. Yeah it's probably the answer they want, but I would first suggest "Profile to see it there is any point in performing any optimisation on it. Then perhaps remove the need to perform the multiplication at all, or hoist it out of a loop if possible." Or maybe ask what processor and what data type first as it depends upon the processor. If it were a verbal test, they'd probably be blown away with this kind of answer.