Fastest C++ Rational Number library

Question

I'm using GMPXX wrapper of GMP and it is not fast enough. Is it possible to find some comparison of rational number libraries' speed?

During my calculation a very big rational number will appear with 10^100 denominator and same size numerator.

Do you know something faster than GMP?

score 5 · Answer 1 · answered Oct 04 '12 at 19:22

5

Do you know something faster than GMP?

It appears the Haskell folks faced a similar problem as yours. Here are their notes:

answered Oct 04 '12 at 19:22

Robᵩ

163,533
20
239
308

at this point, if you can consider a switch to another language consider Python. – axis Oct 04 '12 at 19:28
2

@axis, Python has a built-in Rational class but there's no reason to believe it will be speedier than C++. – Mark Ransom Oct 04 '12 at 19:44

score -5 · Answer 2 · edited May 23 '17 at 10:29

-5

A rational is a float/double, the problem with that is basically the base 2 ( used by computers ) vs base 10 ( used by humans in the classic math calculus ), in the end getting a good representation of a generic rational number is a good challenge, considering a value with a magnitude of 10^100 this becomes an extremely good challenge.

I think that you should pause for a minute and think about this because a float generated by 10^100/10^100 can be really gigantic and doing this with a computer is something for a really advanced lab in my opinion, you can't expect a library to solve this kind of problems with efficiency and most importantly with accuracy with a magnitude this big.

further reading

edited May 23 '17 at 10:29

Community

1
1

answered Oct 04 '12 at 18:59

axis

874
2
7
13

3

I fear that you have missed the OP's point. He isn't using `float` or `double`. In fact, he isn't using floating-point representation at all. He is using a library called GMP which can store arbitrarily large numbers (beyond 32-bit or 64-bit). Yes, doing the math in software instead of in hardware **is** slower, and he is asking who publishes the fastest such software. – Robᵩ Oct 04 '12 at 19:04
@Robᵩ I'm supposed to suggest assembly or a random optimized assembly instruction set ? For what i know the problem base 2 vs base 10 still remains; store a number that i large or small but this problem is still there. – axis Oct 04 '12 at 19:07
2

@axis, rational numbers are represented as two integers, a numerator and a denominator. If the denominator is a power of 10 then you can precisely represent any decimal number you want. The base2/base10 problem doesn't occur with integers. – Mark Ransom Oct 04 '12 at 19:25
@MarkRansom ok, and what are you doing with this numbers ? just using random portion of memory for sport ? If you are going to use a rational it will be a float or worst. – axis Oct 04 '12 at 19:30

Fastest C++ Rational Number library

2 Answers2