some (useful) remarks about particleemitter
Posted: 19 Aug 2008, 02:13
in fact about randomness in general...
The use of rand() should be avoided. Not because of poor randomness, of course you don't need better quality, but because of performance. The overhead caused by repeated function calls can be significant for particles generation.
I would recommend to replace rand with a more efficient custom function.
you might use the glibc rand definition (according to http://en.wikipedia.org/wiki/Linear_con ... _generator)
(putting this inside a class could be better)
this gives a 10x faster rand
This is really easy to do (mainly change only minmax.h) and can bring better performance for large particles systems.
It is also really clear and easy to use...
If you find some reason not to use this... I would be glad to hear them!
The use of rand() should be avoided. Not because of poor randomness, of course you don't need better quality, but because of performance. The overhead caused by repeated function calls can be significant for particles generation.
I would recommend to replace rand with a more efficient custom function.
you might use the glibc rand definition (according to http://en.wikipedia.org/wiki/Linear_con ... _generator)
Code: Select all
unsigned int state=0x19082008; //might be seeded somewhere
inline int myrand(){
state = (state*1103515245)+12345;
return state & RAND_MAX;
}
this gives a 10x faster rand
This is really easy to do (mainly change only minmax.h) and can bring better performance for large particles systems.
It is also really clear and easy to use...
If you find some reason not to use this... I would be glad to hear them!