20200826, 19:10  #34 
"Tilman Neumann"
Jan 2016
Germany
2^{2}·3·5·7 Posts 
I have been provided with a Java implementation of this sieve a couple of months ago. You can find it here:
https://github.com/TilmanNeumann/jav...act/SSOZJ.java My impression was that it is very fast. Thanks Jabari and Pascal. 
20200826, 19:18  #35  
"Ben"
Feb 2007
3,361 Posts 
Quote:
Quote:
Sundaram's sieve was introduced by someone else, I can't find the actual paper online. I don't know whether Eratosthenes named his algorithm after himself or not, but I'd say one way or another he gets a pass... standards were probably quite different 2000 years ago. 

20200826, 19:37  #36 
Jul 2014
2^{5} Posts 
For some reason(s) you are hellbent on discrediting my work, and any and everybody else's independent use of it. If other people can find just a smidgen of new knowledge and benefit from it then it has value. Why you continue to try to deny that only you can answer, with others left to guess.

20200826, 20:03  #37  
"Ben"
Feb 2007
3,361 Posts 
Quote:
But there is a difference between good implementation and theoretical advancement, the latter of which is what the selfnaming promotes. In my opinion, the things you have written about sieving (not touching the twinprime conjecture stuff, here) are not the same caliber of work as the sieve of Eratosthenes itself, or Atkin's or Pritchard's work, and naming it as such strikes me as wrong. Were I to peerreview such claims as an article submission I would say the same and recommend rejection. Take it or leave it. 

20200826, 20:46  #38  
6809 > 6502
"""""""""""""""""""
Aug 2003
101×103 Posts
17·541 Posts 
Quote:
Misc Math is an area for things that really are not good math at all. Quote:
Quote:
Quote:
Quote:


20200827, 16:17  #39  
Aug 2006
3·1,987 Posts 
Quote:


20200827, 16:56  #40 
"Tilman Neumann"
Jan 2016
Germany
2^{2}·3·5·7 Posts 
Yes, one should not name stuff after himself.
You should have one or two friends that do that for you Apart from that: I didn't look at the paper but the implementation I got looks pretty fast. Unfortunately I have no other implementation to compare it to. So I am happy that bsquared appointed himself to investigate it. Btw. exploiting an unproven hypothesis (Schinzels H) in software wouldn't be such a low feat either, or not? 
20200827, 18:56  #41 
Jul 2014
20_{16} Posts 
I referenced all my implementations against primesieve, who I informed back in 2014 about my sieves using prime generators.
https://github.com/kimwalisch/primesieve The fastest twinprime sieve that runs on a common laptop is the Rust version of my twinprimes_ssoz, followed closely by the Nim version. Because I don't have access to a GPU I haven't done a CUDA/OpenCL version, which should fly with all of their independent cores. It's also equally amenable to implement in distributed/cloud networks, as it's inherently designed to be done in parallel processing systems. I haven't tried Pascal's Java version yet, so I can't tell you how it compares. I'm certain my implementations can be improved upon for various hardware environments. I would encourage people to do so and post their results. Also, primesieve is a very good program done in very optimized C++, that has been actively maintained and improved for a number of years. It has thousands of lines of code, spanning hundreds? of files, to implement. My Rust, Nim, D, Crystal, et al versions consist of one file of less than 300 lines of actual code. 
20200827, 19:11  #42 
"Ben"
Feb 2007
3,361 Posts 
I've just compared the java version of ssojz that Till provided with primesieve and yafu and I'm indeed impressed.
Counting twin primes to 10^11 with 16 threads: ssojz: 3.34 sec Counting twin (and triple... up to sextuple) primes to 10^11 with 16 threads: primesieve: 2.88 sec Counting primes (no twins or other tuples) to 10^11 with 16 threads: yafu: 1.99 sec Last fiddled with by bsquared on 20200827 at 19:40 Reason: corrected yafu time 
20200827, 19:30  #43 
Jul 2014
2^{5} Posts 
If you can, compile/test the Nim and Rust versions (Rust just released 1.46). They should be faster, especially with 16 cores.
My base I7 system only had 4C8T. I'm salivating on getting an AMD Threadrippper 12C24T system. 
20200827, 19:55  #44 
Aug 2006
5961_{10} Posts 
That is extremely impressive, especially considering it's in Java. I wonder how the Rust implementation will fare.

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Twin Prime Constellations  robert44444uk  Prime Gap Searches  21  20190609 12:51 
How do you efficiently sieve for prime 3/4tuples?  PuzzlePeter  Software  156  20190603 20:19 
find very easy twin prime in the infamy twin primes  hal1se  Miscellaneous Math  13  20181105 16:34 
Highest Prime is also a twin prime... NOT  hydeer  Lone Mersenne Hunters  9  20180403 22:54 
Twin Prime Days, Prime Day Clusters  cuBerBruce  Puzzles  3  20141201 18:15 