[ODE] ODE Digest, Vol 49, Issue 31

[Oleh Derevenko]

Why do you think operations with float are faster than operations with 
double? This is correct for emulations only. In FPU everything is converted 
to 80-bit and there is no difference in source float format you load numbers 
from. Modern CPU cache lines are much larger than both 4 and 8 bytes, so 
there is no difference in load/store time from/to memory if the data is 
aligned. The only difference is more cache pollution with double and larger 
memory demands.

Oleh Derevenko
-- ICQ: 36361783

----- Original Message ----- 
From: "Ruud van Gaal" <ruud at racer.nl>
To: <ode at ode.org>
Sent: 28 вересня 2007 р. 14:08
Subject: Re: [ODE] ODE Digest, Vol 49, Issue 31


> OPCODE probably stores its data as floats and that should probably remain
> for performance's sake, but the warnings on type conversion should 
> indicate
> where a float should be changed into dReal.