I’ve got a multi-threaded version of this algorithm working in D, Nim, and Rust, and am now trying to do a Crystal version. The serial section looks like this: cnts = [] of typeof(end_num) lastwins = [] of typeof(end_num) restwins.each_with_index do |r_hi, i| l, c = twins_sieve(r_hi, kmi…

[image] jzakiya: lastwins << l; cnts << c This part jumps out, you are changing a shared Array, without synchronization. The results can be weird, though I would not think it should break with exception, but possibly give wrong results in occasion. Can you share a compiling full code?

Here’s the single-threaded version, which produces correct outputs. https://gist.github.com/jzakiya/05b39b66ee6c1a25bb0dd39f83acbd64 twinprimes_ssoz_ser.cr # This Crystal source file is a single threaded implementation to perform an # extremely fast Segmented Sieve of Zakiya (SSoZ) to find …

cnts = Array(UInt64).new(restwins.size + 1) {0u64} lastwins = Array(UInt64).new(restwins.size + 1) {0u64} done = Channel(Nil).new() restwins.each_with_index do |r_hi, i| # sieve twinpair restracks spawn do l, c = twins_sieve(r_hi, kmin, kmax, kb, start_num, end_num, modpg, primes,…

Thanks @lribeiro , here’s the cleaned up code I used that’s works. It’s now in the twinprimes_ssoz.cr gist. cnts = Array(UInt64).new(pairscnt) {0u64} lastwins = Array(UInt64).new(pairscnt) {0u64} done = Channel(Nil).new() restwins.each_with_index do |r_hi, i| # sieve twinpair restracks …

BTW for those interested in the mathematical basis for the algorithm, see my paper below. [image] The Use of Prime Generators to Implement Fast Twin Primes Sieve of Zakiya... This paper describes the mathematical foundation of Prime Generators and their use in creating a fast Twin Primes …

This is an update. I tweaked the Crystal gist source code so that it now produces correct results over the full unsigned 64-bit integer space. (Some intermediary arithmetic was losing accuracy, had to increase integer types). Below is some output on my laptop (I7 8 threads, 16GB mem). The first …

How to Parallelize this?

Help & Support

jzakiya March 17, 2020, 3:01pm 6

BTW for those interested in the mathematical basis for the algorithm, see my paper below.

Timeline for multithreading support

Topic		Replies	Views
Perplexing spawn arithmetic overflow errors Help & Support	23	694	March 24, 2021
Using new parallel ExecutionContext Help & Support	21	404	August 11, 2026
Kudos to Crystal! News	12	2332	December 29, 2021
1.0 multi-threading memory use issues Crystal Contrib	32	2154	August 29, 2020
Wait-groups fiber pooling in 1.19	10	447	January 21, 2026

How to Parallelize this?

Related topics