mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Software (https://www.mersenneforum.org/forumdisplay.php?f=10)
-   -   fpcsieve version 0.1.1 released (https://www.mersenneforum.org/showthread.php?t=20035)

TheCount 2015-02-03 12:23

fpcsieve version 0.1.1 released
 
I've taken Geoffrey Reynolds fpsieve from [URL]https://sites.google.com/site/geoffreywalterreynolds/programs/testing[/URL] and ported it to VS2012 and added Compositorial sieving support to create fpcsieve.
Windows version 0.1.1 binaries are available at:
[URL]https://github.com/RogerKarpin/fpcsieve/tree/master/bin[/URL]
GitHub is new to me but looks really good.

The only changes from fpsieve to include Compositorial sieving are additions to app.c and app.h.
If someone could compile the original code base with these changes and produce linux binaries that would be appreciated.

This is a CPU sieve that has multithreaded support.
Won't port well to GPU as needs to refer to list of Composites in memory (just like Primorial does to list of Primes).
See here for more info:
[URL]http://www.primegrid.com/forum_thread.php?id=5952[/URL]
I've started sieving n<1M. Currently up to 3.8G.
PM me if your interested in Compositorial prime searching.
This is one of the baseless prime searches like Factorial and Primorial.
Will need to reach n>80k to get in the top 5,000 list. Currently only searched to n=16k which is ridiculous for this day and age:
[URL]https://primes.utm.edu/primes/search.php?Number=200&Comment=orial[/URL]
This is my first new thread post on mersenneforum. Hopefully many more to come.

Batalov 2015-02-06 22:34

[URL="http://primes.utm.edu/primes/page.php?id=119245"]96835!/96835# + 1 is prime[/URL]

paulunderwood 2015-02-07 06:11

[QUOTE=Batalov;394753][URL="http://primes.utm.edu/primes/page.php?id=119245"]96835!/96835# + 1 is prime[/URL][/QUOTE]

Congrats :banana:

Are there any gaps in this table?

[code]
----- -------------------------------- ------- ----- ---- --------------
rank description digits who year comment
----- -------------------------------- ------- ----- ---- --------------
2209 96835!/96827#+1 398821 p381 2015 Compositorial (**)
[COLOR="Green"] ---- 41400!/41400#-1 155301 p381 2015 Compositorial[/COLOR]
[COLOR="Green"] ---- 33684!/33684#-1 123321 p381 2015 Compositorial[/COLOR]
40882 20943!/20939#-1 72401 p16 2000 Compositorial
42827 18160!/18149#-1 61635 p16 2000 Compositorial
44240 17258!/17257#+1 58207 p16 2000 Compositorial
47617 15250!/15241#-1 50623 p16 2000 Compositorial
48663 13917!/13913#+1 45631 p16 2000 Compositorial
49214 13724!/13723#-1 44919 p16 2000 Compositorial
51963 10977!/10973#-1 34876 p16 2000 Compositorial
56655 8711!/8707#-1 26816 p7 1999 Compositorial
59086 6851!/6841#+1 20374 p34 2000 Compositorial
60838 6187!/6173#+1 18135 p7 1999 Compositorial
68706 3692!/3691#+1 9996 p7 1999 Compositorial
----- -------------------------------- ------- ----- ---- --------------
[/code]

Batalov 2015-02-07 07:16

There must be!
...'cause I started from 90,200, really, and went up, only until one on either side... well, maybe I'll hunt for the minus side, as well.

[COLOR="Green"]Found in gaps:
Feb 9 18:39:34 PST: 33684!/33684#-1 is prime (123321 digits)
Feb 9 23:11:33 PST: 41400!/41400#-1 is prime (155301 digits)[/COLOR]

TheCount 2015-02-07 14:04

[QUOTE=paulunderwood;394780]Are there any gaps in this table?
[/QUOTE]
Was previously tested to 14000 on both the Plus and Minus sides according to OEIS.
[URL]https://oeis.org/A049420[/URL]

Before implementing the sieving program I double checked this initial range and continued on to:
- Compositorial Plus side: 21706
- Compositorial Minus side: 20621

I've now completed sieving n<1M up to 10G.
969528 terms remain (total of both sides).

Batalov 2015-02-07 17:46

I've briefly searched google for "Daniel Heuer compositorial" and I think some traces (and possibly DH's limits) can be found in the archives of the primeform yahoo group. E.g. [URL="https://groups.yahoo.com/neo/groups/primeform/conversations/messages/3978"]this[/URL] (about C(20493)-1 which you will soon double-check) ...and this appears to be [URL="https://groups.yahoo.com/neo/groups/primeform/search/messages?advance=true&am=CONTAINS&at=email:heuer@&dm=IS_ANY&fs=false&count=1000"]all of his messages[/URL] (API of yahoo groups changed so many times over years; the new API is so inconvenient).

Daniel was active in many curious projects. I stumble over his decade old contributions in many areas.

Maybe Paul has Daniel's contacts? Paul "was always there", as they say.

Antonio 2015-02-07 17:58

There is a bug in the on screen status display. On my machine (Windows 7 64bit i5 3570k @ 4.4GHz) it displays: -

p=309370883, 18.19K p/sec, [COLOR=red]3.94x0MHz[/COLOR] CPU, 0.0% done

(This is with the 64bit version)

Question - wouldn't it sieve a lot faster if it didn't print the factors found file - fpfactors.txt?

paulunderwood 2015-02-07 19:18

[QUOTE=Batalov;394826]

Daniel was active in many curious projects. I stumble over his decade old contributions in many areas.

Maybe Paul has Daniel's contacts? Paul "was always there", as they say.[/QUOTE]

[url]http://primes.utm.edu/bios/page.php?id=223[/url] has his email address, but Daniel may have moved on.

rogue 2015-02-07 23:36

I see no reason why this can't be built upon fpsievecl, the OpenCL implementation of fpsieve. The reason is that this form only requires the exclusion of certain numbers from the multiplication of the factorial. It is possible that global memory access might cause slowdowns though.

TheCount 2015-02-08 08:03

[QUOTE=Antonio;394828]There is a bug in the on screen status display. On my machine (Windows 7 64bit i5 3570k @ 4.4GHz) it displays: -

p=309370883, 18.19K p/sec, [COLOR=red]3.94x0MHz[/COLOR] CPU, 0.0% done

(This is with the 64bit version)[/QUOTE]
I already posted this issue on GitHub:
[URL]https://github.com/RogerKarpin/fpcsieve/issues/2[/URL]
"Speed of CPU in periodic report not working for 64bit. Need separate asm file to read CPU speed as VS2012 does not compile C file with assembler inline."
Its a low priority issue.

[QUOTE]Question - wouldn't it sieve a lot faster if it didn't print the factors found file - fpfactors.txt?[/QUOTE]

You can use -c option to set the checkpoint period, default is every 5 minutes. Set to 0 if you don't want it to periodically checkpoint.
I wouldn't think writing a checkpoint file would affect performance much.

TheCount 2015-02-08 11:36

[QUOTE=rogue;394868]I see no reason why this can't be built upon fpsievecl, the OpenCL implementation of fpsieve. The reason is that this form only requires the exclusion of certain numbers from the multiplication of the factorial. It is possible that global memory access might cause slowdowns though.[/QUOTE]
Worth a try. I'd love to get a GPU sieve for Compositorial. I could have my GPU sieving while PRP testing on my CPU.

fpsievecl? I didn't know you'd extended it to Primorial. If you hadn't yet I would extend to Primorial while adding Compositorial. We can see if it's worthwhile after its implemented.

Is v1.0.8 the latest version? If so I already have it downloaded and know how to fix a bug:
[URL]http://www.primegrid.com/forum_thread.php?id=1163&nowrap=true#80600[/URL]

If Compositorial implemented would the new program be called fpcsievecl? I'd add a new GitHub repo.


All times are UTC. The time now is 14:54.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.