Le 10/11/2011 15:43, [email protected] a écrit :
> Author: sebb
> Date: Thu Nov 10 14:43:44 2011
> New Revision: 1200350
>
> URL: http://svn.apache.org/viewvc?rev=1200350&view=rev
> Log:
> Unnecessary casts - angbd is double, so ensures the rest of the calculation
> uses double
>
> Modified:
>
> commons/proper/math/trunk/src/main/java/org/apache/commons/math/optimization/direct/BOBYQAOptimizer.java
>
> Modified:
> commons/proper/math/trunk/src/main/java/org/apache/commons/math/optimization/direct/BOBYQAOptimizer.java
> URL:
> http://svn.apache.org/viewvc/commons/proper/math/trunk/src/main/java/org/apache/commons/math/optimization/direct/BOBYQAOptimizer.java?rev=1200350&r1=1200349&r2=1200350&view=diff
> ==============================================================================
> ---
> commons/proper/math/trunk/src/main/java/org/apache/commons/math/optimization/direct/BOBYQAOptimizer.java
> (original)
> +++
> commons/proper/math/trunk/src/main/java/org/apache/commons/math/optimization/direct/BOBYQAOptimizer.java
> Thu Nov 10 14:43:44 2011
> @@ -2149,7 +2149,7 @@ public class BOBYQAOptimizer
> redsav = ZERO;
> iu = (int) (angbd * 17. + 3.1);
> for (int i = 0; i < iu; i++) {
> - angt = angbd * (double) i / (double) iu;
> + angt = angbd * i / iu;
I'm not sure about this. It depends on how the compiler performs
associativity. This multiplication can be compiled either as
(angbd * i) / iu
or as
angbd * (i / iu)
In the later case, the integer division is wrong.
Luv
> sth = (angt + angt) / (ONE + angt * angt);
> temp = shs + angt * (angt * dhd - dhs - dhs);
> rednew = sth * (angt * dredg - sredg - HALF * sth * temp);
> @@ -2171,7 +2171,7 @@ public class BOBYQAOptimizer
> }
> if (isav < iu) {
> temp = (rdnext - rdprev) / (redmax + redmax - rdprev -
> rdnext);
> - angt = angbd * ((double) isav + HALF * temp) / (double) iu;
> + angt = angbd * (isav + HALF * temp) / iu;
> }
> cth = (ONE - angt * angt) / (ONE + angt * angt);
> sth = (angt + angt) / (ONE + angt * angt);
>
>
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
