Project

General

Profile

Actions

Feature #15161

closed

making gcd faster for 3x3

Added by jzakiya (Jabari Zakiya) over 5 years ago. Updated over 5 years ago.

Status:
Closed
Assignee:
-
Target version:
-
[ruby-core:89165]

Description

With the goal of making Ruby as fast as possible for 3x3 I would like to propose
a faster implementation of the gcd function. I use gcd a lot in my
primes-utils gem, and in cryptography and Number Theory problems.

The current implementation

https://ruby-doc.org/core-2.5.1/Integer.html#method-i-gcd

uses a recursive implementation.

I propose using the binary (Stein's) algorithm, which I believe has been proposed/discussed before.

https://en.wikipedia.org/wiki/Binary_GCD_algorithm

However, I recently raised an issues with Crystal's implementation (also recursive)
and suggested using Stein's algorithm, which they approved.

https://github.com/crystal-lang/crystal/issues/6683

However, the default (iterative) wikipedia implementation is nowhere near as fast as possible.

https://lemire.me/blog/2013/12/26/fastest-way-to-compute-the-greatest-common-divisor/

The author provides benchmarks of different implmentations (including recursive)

https://github.com/lemire/Code-used-on-Daniel-Lemire-s-blog/blob/master/2013/12/26/gcd.cpp

and the version below is the fastest, and also the simplest.

unsigned int gcdwikipedia2fastswap(unsigned int u, unsigned int v)
{
    int shift;
    if (u == 0) return v;
    if (v == 0) return u;
    shift = __builtin_ctz(u | v);
    u >>= __builtin_ctz(u);
    do {
        v >>= __builtin_ctz(v);
        if(u>v) std::swap(u,v);
        v = v - u;
    } while (v != 0);
    return u << shift;
}

Thank you for considering this.


Related issues 1 (0 open1 closed)

Is duplicate of Ruby master - Bug #13503: Improve performance of some Time & Rational methodsClosedActions

Updated by shevegen (Robert A. Heiler) over 5 years ago

I think a benchmark may be useful to the issue here.

I don't think the ruby team minds any speed improvement but what
is usually done is to verify any potential gain through benchmarks
(e. g. optcarrot for, I think, ruby 3x3 goals).

Updated by mame (Yusuke Endoh) over 5 years ago

  • Status changed from Open to Closed

It already uses the variant of the binary (Stein's) algorithm.

https://github.com/ruby/ruby/blob/3abbaab1a7a97d18f481164c7dc48749b86d7f39/rational.c#L285-L307

See #13503.

Actions #3

Updated by mame (Yusuke Endoh) over 5 years ago

  • Is duplicate of Bug #13503: Improve performance of some Time & Rational methods added
Actions

Also available in: Atom PDF

Like0
Like0Like0Like0