Hello!

Abstract:

0. http://forum.dlang.org/post/luknet$o2h$1@digitalmars.com

1. I have made PR
https://github.com/D-Programming-Language/phobos/pull/2513
with full precision summation function fsum.

The Precise (full precision), Naive, Reverse, Pairwise, Kahan, KBN and KB2 summation algorithms are available for floating point numbers and std.complex.Complex.

Naive, Reverse, Pairwise and Kahan algorithms also available for user defined types.

fsum works with dmd 64-bit and LDC both 64-bit and 32-bit.
(dmd 32-bit compiler has an optimization bug)

Both std.numeric.fsum and std.algorithm.sum use new std.internal.math.summation.


Improvements or critics are welcome ;)



Questions:

0. How can I generate Phobos documentation?

1. What the better organization for other numeric algorithms: normalize, dotProduct, entropy, kullbackLeiblerDivergence, jensenShannonDivergence?

2. Can the same concept (like in fsum, but with naive summation algorithm by default) be added for this numeric algorithms?


Best Regards,
Ilya

> 2. Can the same concept (like in fsum, but with naive summation algorithm by default) be added for this numeric algorithms?

 Or Precise by default?