Yesterday I ran a Ruby version of my Crystal implementation of the Miller-Rabin primality test. I was surprised to find Ruby is significantly faster for the same code with the same inputs. Here’s the Ruby code. def prime? (n, k = 4) primes = [2,3,5,7,11,13,17,19,23,29,31,37,41,43,47,53,59,61,…

Crystal bigint is slow. I tried to brute force one advent of code problem last year (the modulo problem) with bigint and it ran like forever.

mpz_powm_sec is intended to be slow (the sec stands for “secure”) in order to be resilient against timing attacks. Did you try mpz_powm?

Please look at this from the point of view of a user doing numerical algorithms. I used the resources that are available in the standard documentation for both languages. I coded the same routine exactly the same (per implementation details). And Ruby, to my surprise with 3.2.2, creamed Crystal 1…

I don’t understand why you see this as a problem in Crystal’s stdlib. The algorithm you’ve analyzed is a custom binding to mpz_powm_sec, a function implemented by libgmp and currently not exposed in Crystal’s standard library. @Hertzdevil already suggested to try using mpz_powm instead. That’s wha…

You’re missing the points that I’m emphasizing. Right off the bat the Ruby implementation is significantly faster because its numerical implementation is (has become) apparently inherently faster. It automatically adjusts for integer sizes (and whatever that physically entails) without the user hav…

On my M2 I’m getting 0.856s with mpz_powm and 2.293s with mpz_powm_sec. The Ruby snippet takes 33 seconds to run for the same value of n. I also tried the 20,008-digit value and it took 46.3s and 359s on Crystal with the two funs respectively. Did you pass any special command-line arguments to Ruby? …

I ran it on my own computer. The time for both Crystal and Ruby was almost the same. The results were almost the same with or without the ruby --yjit option. I checked if yjit is enabled with `RubyVM::YJIT.enabled? Prime Digits | 3,731 | 5,003 | 20,008 | --------------|-------|--------|------- …

@kojix2 those Crystal numbers must be wrong. Do $ crystal build --release yourfile then $ ./yourfile and see what you get. @HertzDevil I didn’t pass anything to Ruby or Crystal not shown. When I changed mpz_powm_sec to mpz_powm I get no change. But the code spends most of the time in the until..…

I built the executable with the `crystal build --release" command. I am sure of that. And I just checked it on another PC. The time is still almost the same. I had a similar situation before. At that time, most of the time was consumed by the bound library. Let’s take a profile to investigate the…

Okay, thanks for clarifying. Makes sense as a general observation. It’s just that the specific example in this thread is not a good one to make that point, that confused me. We can certainly use more tools for scientific computing in Crystal. I don’t think we’ll have much of it in stdlib, so it’s …

Ruby out performs Crystal significantly on this numerical algorithm

Help & Support

zw963 April 29, 2023, 6:05am 19

Check following thread.

Topic		Replies	Views
BigInt is really really slow when run on linux, Or just Boehm GC issue? Help & Support	31	1042	September 26, 2022
BigInt performance	13	1327	October 16, 2020
On the Infinity of Twin Primes and other K-tuples	22	1263	November 27, 2019
Optimized Mersenne Prime Search	18	1006	November 30, 2024
Improve Crystal's profile on Rosetta Code Community	100	6379	February 28, 2024

Ruby out performs Crystal significantly on this numerical algorithm

Related topics