Minor Change Proposal for Class 'Numeric'

[Wolfgang Nádasi-Donner]

Hi!

I didn't find there a discussion about this here, so first a description of this
(very) minor change proposal.

I miss the "Sign function" in class "Numeric".

It can simply be added by something like...

class Numeric
def sign
self<0?-1:self>0?1:0
end
end

....but it should be much faster if be defined in the original class.

This method will be used in several algorithms, so it should be available from
my point of view.

Wolfgang Nádasi-Donner

 

[Vincent Fourmond]

Hi!

I didn't find there a discussion about this here, so first a description
of this (very) minor change proposal.

I miss the "Sign function" in class "Numeric".

It can simply be added by something like...

class Numeric
def sign
self<0?-1:self>0?1:0
end
end

...but it should be much faster if be defined in the original class.

This method will be used in several algorithms, so it should be
available from my point of view.

Seconded.

Vince

 

[Phrogz]

class Numeric
def sign
self<0?-1:self>0?1:0
end
end

...but it should be much faster if be defined in the original class.

a) Why do you need it to be so fast? Two comparisons seems pretty
lightweight to me!
b) Why do you think it's generally useful? I think such a method would
have helped me only once or twice in all my programming years.

c) You can implement that more compactly (if not more efficiently) as:

irb(main):001:0> class Numeric; def sign; self <=> 0; end; end
=> nil
irb(main):002:0> -1.sign
=> -1
irb(main):003:0> -5.sign
=> -1
irb(main):004:0> 0.sign
=> 0
irb(main):005:0> 3.sign
=> 1

 

[Vincent Fourmond]

a) Why do you need it to be so fast? Two comparisons seems pretty
lightweight to me!

You forget function lookup: there is three function lookup for this
code. Hardcoded in C is way faster (with only one function lookup).

Vince

 

[Eric Hodel]

You forget function lookup: there is three function lookup for this
code. Hardcoded in C is way faster (with only one function lookup).

So you've benchmarked it and found that calling Numeric#sign takes up =20=

a significant part of your runtime?

Note that you can write this in C for about 5 extra lines using =20
RubyInline.

--=20
Eric Hodel - (e-mail address removed) - http://blog.segment7.net

I LIT YOUR GEM ON FIRE!

 

[Vincent Fourmond]

So you've benchmarked it and found that calling Numeric#sign takes up a
significant part of your runtime?

No ;-) I just point out that a C implementation will be faster than a
pure Ruby one. Then, if you rely heavily on it, well, it would count,
wouldn't it ?

But I mainly raised the point to provoke, I obviously got what I
wanted, didn't I ;-)

Note that you can write this in C for about 5 extra lines using

RubyInline.

That definitely is a good idea !

Cheers,

Vince

 

[Phrogz]

You forget function lookup: there is three function lookup for this
code. Hardcoded in C is way faster (with only one function lookup).

I actually hadn't forgotten it, but (mistakenly) thought that < and >
didn't have normal method overhead. So, the spaceship operator is
actually more terse and faster:

class Numeric
def sign1; self < 1 ? -1 : self > 1 ? 1 : 0; end
def sign2; self<=>0; end
def sign3; 0; end
end

require 'benchmark'

N = 1_000_000
Benchmark.bmbm{ |x|
x.report{
N.times{ -5.sign1; 5.sign1; 0.sign1 }
}
x.report{
N.times{ -5.sign2; 5.sign2; 0.sign2 }
}
x.report{
N.times{ -5.sign3; 5.sign3; 0.sign3 }
}
x.report{
N.times{ -5.abs; 5.abs; 0.abs }
}
}

Rehearsal ------------------------------------
3.391000 0.000000 3.391000 ( 3.421000)
2.296000 0.000000 2.296000 ( 2.313000)
1.610000 0.000000 1.610000 ( 1.609000)
0.969000 0.000000 0.969000 ( 0.969000)
--------------------------- total: 8.266000sec

user system total real
2.781000 0.000000 2.781000 ( 2.797000)
2.250000 0.000000 2.250000 ( 2.249000)
1.547000 0.000000 1.547000 ( 1.547000)
0.922000 0.000000 0.922000 ( 0.922000)

So, if we assume that it would be about as fast as #abs, then you're
talking about saving about 1.3 MICRO seconds per call (on a 3GHz P4). I
would be surprised (and interested) if saving that amount of time made
a difference in someone's application, for this particular method.

To put it in perspective:
Suppose your Ruby program were controlling a game or simulation.
Suppose you were getting 30fps. Suppose you were calling this method
10,000 times PER FRAME. The difference between having this method
rewritten in the core and using the spaceship operator gets you all the
way to...30.9fps.

 

[Wolfgang Nádasi-Donner]

b) Why do you think it's generally useful? I think such a method would
have helped me only once or twice in all my programming years.

There is only one reason: It is a standard function used in mathematics and
nearly everybody knows a function names "sign" or "sgn" from other programming
languages.

I think that the code for this method is still there in the kernel.

c) You can implement that more compactly (if not more efficiently) as:

irb(main):001:0> class Numeric; def sign; self <=> 0; end; end

It was only an example to describe what I mean. I have no suggestions for an
implementation in the Ruby kernel.

I only think, that a mathematical standard function should be available.

Wolfgang Nádasi-Donner