Floating Point Precision Loss

Posted on 16th Feb 2014 by admin

I'm writing a program to draw a three-dimensional cube (with a corner cut off) without using any 3D graphics APIs. The only API call I make is win32's SetPixel()

Everything is going great. I've drawn lines, faces, and even the entire cube. Then rotation came along.

For some reason, when I rotate my cube, it begins to shrink. I became confused, then I realised something. Floating point numbers are truncated.

Here's some code that is fired everytime the user presses 'X' on the keyboard.

Code: p->setY( p->getY() * cos(5.0/180.0*PI) - p->getZ() * sin(5.0/180.0*PI)); p->setZ( p->getY() * sin(5.0/180.0*PI) + p->getZ() * cos(5.0/180.0*PI));

p is a Point object that I created. Here the Y value and the Z value of the Point are being set. The problem is everytime X is pressed, and this code is fired, a small amount of precision is lost. As a result, the cube shrinks.

I'm really completely stumped. I remember when studying assembler that floating point numbers can be truncated or rounded. I think rounding would solve my problem. Since 50% of the time the value would round up, and 50% of the time the value would round down.

Whereas with truncation values are always lost, and never gained.

Other forums