Pfgw crash

[nuggetprime]

pfgw just printed this error message when it was testing 9996*10^43139-1:

*** WARNING *** : An error was detected in the reciprocal routine.
If this occurred during normal use, contact bugs@primeform.net,
   including the number you were testing to produce this error,
   and disregard the results of tests on this number.
If this error occurs in optimization mode, you may ignore it.
PRP: 9996*10^43139-1 1/143317                                     
Iteration: 1/143317 ERROR: SUM(INPUTS) != SUM(OUTPUTS),
   9 != 9
  (Diff=0 max allowed=1209)
  (Test aborted, try again using the -a1 switch)

Signal SIGILL caught

What does this mean? My machine isn't overclocked so it's a software bug,or?

--nugget

[philmoore]

Which version of pfgw? It looks like a software bug resulting from using an FFT size which is too small. This should be easy to correct using the -a1 switch, except that I have found this switch broken with the development version 20050213. Perhaps someone at
http://tech.groups.yahoo.com/group/openpfgw/
will be able to help.

[Batalov]

PFGW Version 1.2.0 for Windows [FFT v23.8]
9996*10^43139-1 is composite: [15833EE5B6C6A04B] (94.5948s+0.0017s)
Done.

Doesn't crash here.
Anyway, "try again using the -a1 switch" is what it says. Try it, and check if it still crashes.

[nuggetprime]

Hmm,works with larger fftsize.
Will try to contact the guys at openpfgw.

Thanks,
nugget

[pacionet]

I'm using PFGW (PFGW 1.2.0 for Windows) too.
Why is so old ?
It isn't no more developed and improved ?
(I don't know the details, but maybe they can write some N-core version for the latest CPUs ?)

[rogue]

Originally Posted by nuggetprime

Hmm,works with larger fftsize.
Will try to contact the guys at openpfgw.

Thanks,
nugget

Good luck. PFGW hasn't been updated in years and is unlikely to fixed unless someone else takes it over.

[philmoore]

Originally Posted by rogue

Good luck. PFGW hasn't been updated in years and is unlikely to fixed unless someone else takes it over.

What would it take to get a new development version going? I think pfgw is a very useful program, especially with the capability of using SCRIPT files, but the "stable" version still uses the version 23 multiply routines. The last development version from over 3 years ago uses version 24 and is twice as fast on numbers of the form a*2^n +/- b, but some of the FFT crossover points were set a bit too aggressively. George has fixed this in version 25, successfully as far as I have checked, so it would be nice to incorporate the new multiply routines in a new pfgw version.

I think that Jim Fougeron compiled the last versions, but he seems to have lost interest in this project since then. I haven't been able to get the -a1 or -a2 flags to work in the last development version 20050213 (13 Feb 2005). Jim also said that the ability to perform strong probable prime tests was added, but I have not been able to get this to work either. I would be willing to look at some of the source code if someone knowledgeable with compiler experience would be able to help create a new version.

Phil

[Prime95]

Upgrading to version 25 should be fairly easy. To support multi-threading (elimination of global variables) a pointer to a gwnum structure is passed to every gwnum routine. The init routines are a little different to get this structure initialized.

[rogue]

Originally Posted by philmoore

What would it take to get a new development version going? I think pfgw is a very useful program, especially with the capability of using SCRIPT files, but the "stable" version still uses the version 23 multiply routines. The last development version from over 3 years ago uses version 24 and is twice as fast on numbers of the form a*2^n +/- b, but some of the FFT crossover points were set a bit too aggressively. George has fixed this in version 25, successfully as far as I have checked, so it would be nice to incorporate the new multiply routines in a new pfgw version.

I think that Jim Fougeron compiled the last versions, but he seems to have lost interest in this project since then. I haven't been able to get the -a1 or -a2 flags to work in the last development version 20050213 (13 Feb 2005). Jim also said that the ability to perform strong probable prime tests was added, but I have not been able to get this to work either. I would be willing to look at some of the source code if someone knowledgeable with compiler experience would be able to help create a new version.

Phil

I really don't know which compiler he used because gcc completely barfs at his use of templates. Not being a fan of templates I rarely use them and trying to work through the compiler issues caused by his use of them is not easy.

If someone wants to assist, I would be happy to help get it built using gcc (for Unix builds) and Visual Studio (for Windows builds). I suppose the best place to pose the question would be in the openpfgw group on Yahoo. I would not want to lead such a project, but would help anyone willing to takes the reigns.

[rogue]

Originally Posted by rogue

I really don't know which compiler he used because gcc completely barfs at his use of templates. Not being a fan of templates I rarely use them and trying to work through the compiler issues caused by his use of them is not easy.

If someone wants to assist, I would be happy to help get it built using gcc (for Unix builds) and Visual Studio (for Windows builds). I suppose the best place to pose the question would be in the openpfgw group on Yahoo. I would not want to lead such a project, but would help anyone willing to takes the reigns.

I spent a few minutes looking at the source again. I was able to get past the template issues (with a small number of code changes), but it is looking for gmpgwlib, which is not included in the openpfgw source d/l. I assume I can get this with the prime95 source, but I'll have to go digging for it.

[Harvey563]

I am very excited about the possibility of an updated PFGW!!!

Thanks!