Re: [boost] Request to allow contribution to Boost.Multiprecision cpp_bin_float
Thanks for your replies.
I would like to contribute to the development of Boost.Multiprecision
cpp_bin_float library. Can I work on it? If yes, then how can I start?
Hi Jyotesh,
Paul Bristow, Chris Kormanyos and I have been discussing this off-list. There is a prototype implementation in the sandbox under the multiprecision.cpp_bin_float directory.? The current code is pretty much complete, but under-tested.? The binary-decimal and decimal-binary conversion routines used are also brain-dead and do not round correctly (I have a mostly working version that does this correctly on my hard drive, but it's not quite ready for commit yet).
Yes. But remember, Jyotesh, John has taken a different direction than my original code in the sandbox. His implementation is in a different directory, and it's likely that we will stick with John's and leave my prototyping to rest.
May I know which directory is this?
So one of the things we need is for folks to try this out and see what breaks (or rather to see what they can break!).
Such as try some simple math functions or to use it in conjunction with boost::math::constants for, say, computing a few hundred digits of pi or log(2).
OK. I will try these.
Some of the std lib function test cases in the sandbox also fail currently as their error rates are too high.? Many are also much slower than say MPFR.? Chris, would have a better idea which most need working on.
Always important are square root, sine, cosine, and log.
And finally, the implementation uses all integral arithmetic internally (via cpp_int), so one big improvement we need is for someone to write Karatsuba and FFT based multiplication routines for cpp_int (Karatsuba is probably the higher priority).? Oh and we have no docs yet.
Yes, very important, as this is a huge limitation of cpp_dec_float --- at least for the digit hunters among us. :-)
I will implement Karatsuba algorithm and let you know when I am finished. Thanks and Regards, Jyotesh Choudhari
Anything there look like something you'd like to work on? Thanks for the interest, John.
Thanks for the interest.
Sincerely, Chris.
Thanks for your replies.
I would like to contribute to the development of Boost.Multiprecision cpp_bin_float library. Can I work on it? If yes, then how can I start?
Hi Jyotesh,
Yes. But remember, Jyotesh, John has taken a different direction than my original code in the sandbox. His implementation is in a different directory, and it's likely that we will stick with John's and leave my prototyping to rest.
May I know which directory is this?
The cpp_bin_float.hpp prototype from John Maddock is here: https://svn.boost.org/svn/boost/sandbox/multiprecision.cpp_bin_float/boost/m... There is detailed documentation in the "libs" directory here: https://svn.boost.org/svn/boost/sandbox/multiprecision.cpp_bin_float/libs
So one of the things we need is for folks to try this out and see what breaks (or rather to see what they can break!).
Such as try some simple math functions or to use it in conjunction with boost::math::constants for, say, computing a few hundred digits of pi or log(2).
OK. I will try these.
This should be a straight forward exercise. But be aware that cpp_bin_float.hpp requires the SVN trunk of Boost in order to be used.
And finally, the implementation uses all integral arithmetic internally (via cpp_int), so one big improvement we need is for someone to write Karatsuba and FFT based multiplication routines for cpp_int (Karatsuba is probably the higher priority).? Oh and we have no docs yet.
Yes, very important, as this is a huge limitation of cpp_dec_float --- at least for the digit hunters among us. :-)
I will implement Karatsuba algorithm and let you know when I am finished. Thanks and Regards,
Jyotesh Choudhari
OK, give it a stab. Good luck, and get back to us if you have any questions. Sincerely, Chris.
participants (2)
-
Christopher Kormanyos -
Jyotesh Choudhari