ZFR: VBA macros (yes, I worked a lot with those :( ) Often repairing others' "code".

The Microsoft employee who thought it would be a good idea to include "Record macro" with Excel, should be forced to work fixing the code people create using this function.

Example of one of the worst offenders:

To sort a column:
_Select the column
_Copy
_Create new sheet
_Paste
_Sort (using Excel's built in sort)
_Select the sorted column
_Copy
_Select original column
_Paste

Lifthrasil: Can't quote it here since I deleted it. But when debugging a program that my predecessor at the institute I was working at wrote, I found it contained a block of about 2000 lines (C++) that did exactly nothing but wasting processing power. I could never figure out what this sub-routine was supposed to do. It was called quite often in the code and made lots of calculations that were then never used. So I deleted it and all references to it and voila, the program was much more stable and a lot faster too.

Wishbone: Then of course there's code that's not bad as such, in fact it's so good it's almost magical, but quite unintuitive and extremely difficult to understand. Example: The fast inverse square root algorithm.

Lifthrasil: Can't quote it here since I deleted it. But when debugging a program that my predecessor at the institute I was working at wrote, I found it contained a block of about 2000 lines (C++) that did exactly nothing but wasting processing power. I could never figure out what this sub-routine was supposed to do. It was called quite often in the code and made lots of calculations that were then never used. So I deleted it and all references to it and voila, the program was much more stable and a lot faster too.

rtcvb32: Good general explanation..

Error range of 0.17%, subtraction and multiplications used, but not divides (which is the slowest). It feels like the main advantage this would have is if you were doing software rendering and a lot of people still didn't have FPU's... Quake1 would hit that requirement, but not sure about Quake 2 and on...

Although for the code shown i'd have probably used a union instead of pointers to override type, although optimization might make that moot...

dtgreene: Actually, according to
http://blog.regehr.org/archives/959

the fastest way would be something like this (f float):
{
int i;
memcpy(&i, &f, 4);
}

(The memcpy() call is optimized into the lease assembly code.)

When it comes to optimization, sometimes compilers can surprise you, so be careful. (Of course, there are other surprises to watch out for, like CPUs and other hardware having unexpected performance characteristics.)