Thread: Fastest float to int conversion

Salem, still a happy camper I see.

This is a personal project, so I am hardly going to hire somebody. Also, if someone asks a question like "how do I step through all the letters in my name in the quickest time, its quite long you see" - I'm not going to say - "Why the hell do you have a long name in the first place dumbass".

My code in this instance involves moving objects around at arbitrary angles. This involves computing a dx and dy step based on the sin and cos of the angle (LUT), and then stepping a sprite by that amount every 'frame'. Floats are used to keep the movement accurate.

The float->int conversion is needed when the sprite coordinates are transformed to screen space, as obviously I can't place something at x=10.5, y=8.756 !

floats are also used in various other parts of my code, and although in some instances I can probably make use of integers<<5 (for example a resolution of 1/32 pixels) it would make the code messy, and I can't do this in all instances.

So in the instances that I DO have to convert a float to an int, which method would be the fastest?

Yup. I'd use a power of 2, say 1024 to do like quzah says. Then you only need to shift & mask to get the integer part. A lot faster than casting from float!

@Jez - Yip, that was one of my theories. (However, quazah doesn't mention a power of 2, or bit shifting, but rather division - which is hell slow).

@pianorain - This is because say the step size was 1.3, and x started at 0, I would want something like this:

0: x = 0
1: x = 1.3, round = 1
2: x = 2.6, round = 3
3: x = 3.9, round = 4
4: x = 5.2, round = 5
5: x = 6.5, round = 7
6: x = 7.8, round = 8 etc

As you see, I DONT throw away the floating point, I keep it in to keep the precision of the base arithmetic, but just loose the precision when I render to the screen.

AND back to the original Q
If I WAS to convert from float to int, what is the fastest way?

Surely if the only answers are 'don't use floats' then why would compilers even support them?

But you asked how to do something specific with floats as fast as possible.

My point exactly