[prev in list] [next in list] [prev in thread] [next in thread] 

List:       gmp-devel
Subject:    Replacing "redc"
From:       tg () gmplib ! org (=?utf-8?Q?Torbj=C3=B6rn?= Granlund)
Date:       2017-07-20 13:56:12
Message-ID: 861spbfc03.fsf () shell ! gmplib ! org
[Download RAW message or body]

paul zimmermann <Paul.Zimmermann at inria.fr> writes:

  quite interesting. Is the API of the new interface fixed? I'd like to try it
  with GMP-ECM, where we currently use the redc_* internal functions. For example
  on my machine Intel(R) Core(TM) i5-4590 CPU @ 3.30GHz, running at 3499.804Mhz:
  
No internal interfaces are strictly fixed, I suppose.

We've had internal 2-adic *_bdiv_* functions for a while, with redc_*
being mostly redundant, and also less general.  Originally, the
*_bdiv_* functions and redc_* could generate remainders which where off
by d, d being the divisor.

Obsoleting redc_* is a cleanup, simplification, and generalisation as
they *_bdiv_* functions allow separately defined operand sizes.

I don't expect any speedup except as a result of better implementations
of newer assembly.  Currently, we only have one sbpi1_bdiv_r in assembly
(for AMD Zen).  I plan on converting the assembly redc_1 into
sbpi1_bdiv_r form.

-- 
Torbj?rn
Please encrypt, key id 0xC8601622

[prev in list] [next in list] [prev in thread] [next in thread]