double/float and speed

if you need high precision (in a game probably you don''t) use doubles. otherwise use floats. calculations with doubles take more CPU (FPU) cycles. its slower.


"Knowledge is no more expensive than ignorance, and at least as satisfying." -Barrin

functions:

void floatFunc()
{
	float x = 1;
	for (int i = 0; i < 200000000; i++)
	{
		x *= ((x + 1.0f) / 2.0f);
	}
	std::cout << x << ''\n'';
}

void doubleFunc()
{
	double x = 1;
	for (int i = 0; i < 200000000; i++)
	{
		x *= ((x + 1.0) / 2.0);
	}
	std::cout << x << ''\n'';
} 

Output from gprof (profiler):

  %   cumulative   self              self     total           
 time   seconds   seconds    calls   s/call   s/call  name    
 49.83      3.01     3.01        1     3.01     3.01  floatFunc()
 49.83      6.02     3.01        1     3.01     3.01  doubleFunc()

So there is no difference in execution time for floats vs doubles. Doubles may be slower to fetch/transfer though, due to their size.

Enigma

Enigma, thanks, that was very helpfull !

If you are looking at doing precise physics or calculations, I would use doubles. I''m not sure if this is correct but I think that floats can have a variation in their decimal values at like the 8th decimal number (.#######1). I made a gravity simulation and used floats to store position and acceleration etc., and a system the should have been stable would eventually become unstable becasue of the tiny variations in the float. I used doubles and the problem went away.

That''s correct, i first used code to draw a quad using many tiny triangles, but the last row of vertices wasn''t drawn. I finally found out (when i decided to place all data in a text file to) that the value, when it should have been something like 16.0 it was 16.000001, so the >=16.0f test failed. I use floats now for all calculations except drawing planes in pieces (is that tesselating or is tesselating something else, sorry if it comes to names of tecniques i am a n00b ).