Jean-Pierre Flori on Thu, 14 Jan 2016 17:05:36 +0100


[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

Re: Faster exponentiation in some extensions of finite fields of small characteristics


Looking at what Peter did, I now see I should just wrap his code!
http://pari.math.u-bordeaux.fr/archives/pari-dev-1510/msg00004.html

2016-01-14 13:32 GMT+01:00 Jean-Pierre Flori <jpflori@gmail.com>:
> Should be better now.
>
> 2016-01-14 13:22 GMT+01:00 Jean-Pierre Flori <jpflori@gmail.com>:
>> It seems it was an old buggy version of the patch.
>> I'm trying to relocate the right commit.
>>
>> 2016-01-14 11:58 GMT+01:00 Jean-Pierre Flori <jpflori@gmail.com>:
>>> Dear all,
>>>
>>> Here is a preliminary patch to speed up exponentiation in some
>>> extensions of finite fields of small characteristics (< size of
>>> machine word/2).
>>>
>>> It packs more coefficients of the polynomial representation over the
>>> prime field into machine words when the finite field element is mapped
>>> to a multiprecision integer (i.e. KS).
>>> Two functions are added:
>>> * one function which packs 4 coeffs into a machine word,
>>> * one generic funciton which packs the coeffs as much as possible
>>> possibly crossing machine words boundaries.
>>>
>>> I did not take care of adding new tuning parameters or smartly
>>> choosing between the different functions, e.g. calling the
>>> 4-coeffs-in-1-word function when the (product) coeff size is
>>> BITS_IN_LONG/4-\epsilon might be more efficient than using the generic
>>> function which does more complicated packing when the polynomial
>>> degree is not large.
>>>
>>> I did not add (yet) optimal packing when the product coeffs are larger
>>> than a machine word.
>>>
>>> I did not really check it is bug free.
>>>
>>> Best,
>>> --
>>> Jean-Pierre Flori
>>
>>
>>
>> --
>> Jean-Pierre Flori
>
>
>
> --
> Jean-Pierre Flori



-- 
Jean-Pierre Flori