PFGW latest well-tested version
 

Hi all,

A couple of days ago, PFGW 3.2.0 was [URL="http://www.mersenneforum.org/showthread.php?t=12214"]released[/URL]. This version fixes many of the problems in 3.1.0 including the residue bug. The roundoff errors were not fully fixed with this release. Versions 3.3.0 and later correct all roundoff errors. Versions 3.3.6 and later correct all issues related to the proof of smaller PRPs. Here are links to the latest well-tested versions of PFGW:

[URL="http://sourceforge.net/projects/openpfgw/files/pfgw_win_3.4.1_20100927.zip/download"]PFGW 3.4.1 for Windows[/URL]
[URL="http://sourceforge.net/projects/openpfgw/files/pfgw_linux_3.4.0_20100925.zip/download"]PFGW 3.4.0 for Linux[/URL]
[URL="http://sourceforge.net/projects/openpfgw/files/pfgw_mac_3.4.0_20100925.zip/download"]PFGW 3.4.0 for Mac[/URL]

PFGW 3.4.0 & 3.4.1 are now the most highly recommended programs for testing at CRUS.

PRPnet 3.3.5, which likewise fixes all known bugs in PRPnet, utilizes PFGW for non-power-of-2 PRP tests if available. If you're using PRPnet, we highly recommend downloading this latest version. See [URL="http://www.mersenneforum.org/showthread.php?t=12225"]this thread[/URL] for more information.

Max :smile:

[quote=mdettweiler;183154]Hi all,

A couple of days ago, PFGW 3.2.0 was [URL="http://www.mersenneforum.org/showthread.php?t=12214"]released[/URL]. This version fixes all of the problems in 3.1.0: the residue bug, as well as the FFT roundoff errors.

Now that the residue bug in PFGW has been remedied, PFGW and Prime95 are equally recommended for non-power-of-2 PRP testing. Actually, at this time, PFGW is more highly recomended since it's fixed the FFT roundoff errors, though Prime95 v25.12 should also be released soon to fix these as well.

PRPnet 2.2.3, which likewise fixes all known bugs in PRPnet, utilizes PFGW for non-power-of-2 PRP tests if available. If you're using PRPnet, I highly recommend downloading this latest version. See [URL="http://www.mersenneforum.org/showthread.php?t=12225"]this thread[/URL] for more information. The client packages downloadable from there include the latest version of PFGW.

Max :smile:[/quote]
why doesnt it use pfgw for base 2?
ll tests have been made a little faster since gwnum 24.14 and also has speed ups for 64-bit
i suspect that those speedups will also help llr

[QUOTE=henryzz;183248]why doesnt it use pfgw for base 2?
ll tests have been made a little faster since gwnum 24.14 and also has speed ups for 64-bit
i suspect that those speedups will also help llr[/QUOTE]

Because LLR does a primality test for base 2 (both +1 and -1), not a PRP test. For the +1 side, PFGW would have to be modified to dynamically choose a base that can be used for a Proth test. By doing that the PRP test turns into a primality test. This is what LLR and phrot do. For the -1 side, PFGW would need to implement the LLR algorithm (which is in PFGW 1.4, which we don't have source for). I could "borrow" the code from LLR for the -1 test, but all we really need is for Jean to upgrade LLR. Finally, when PFGW does primality tests, it doesn't not give a residue if the result is composite.

BTW, I have not heard from Jean in a while, so I don't what's going on with him. Since LLR source is available, it would be relatively easy to upgrade it, but I would prefer to have his permission before doing so.

As for a 64-bit build, I'm working with Geoff Reynolds as there is a single asm module in PFGW that will not compile as 64-bit. I'm not an x86 asm guru, so I'm hoping that he can come up with a single module that works for both 32-bit and 64-bit environments.

I'm getting an error on PFGW with my new Sierp base 52 reservation. Here it is:

[code]
-----------------------------------------------------------------------
Error occuring in PFGW at Tue Aug 04 05:07:35 2009
Expr = 9243*52^2046+1
Failed at bit 1877 of 11676
Msg = Detected in MAXERR>0.45 (round off check) in prp_using_gwnum
Iteration: 1877/11676 ERROR: ROUND OFF 0.46875>0.45
[/code]

I'm running on a 32-bit 1.6 Ghz HP dual-core laptop under Windows Vista.

Its running without sieving up to n=5000 so I have the -f100 switch set for trial factoring. A quick check using Syd's DB found no factors up to P=10M but it is composite. So no problems caused at this point but it could cause a problem on a real prime in the future.

Can someone look into this for me?


Thanks,
Gary

[QUOTE=gd_barnes;183980]I'm getting an error on PFGW with my new Sierp base 52 reservation. Here it is:

[code]
-----------------------------------------------------------------------
Error occuring in PFGW at Tue Aug 04 05:07:35 2009
Expr = 9243*52^2046+1
Failed at bit 1877 of 11676
Msg = Detected in MAXERR>0.45 (round off check) in prp_using_gwnum
Iteration: 1877/11676 ERROR: ROUND OFF 0.46875>0.45
[/code]

I'm running on a 32-bit 1.6 Ghz HP dual-core laptop under Windows Vista.

Its running without sieving up to n=5000 so I have the -f100 switch set for trial factoring. A quick check using Syd's DB found no factors up to P=10M but it is composite. So no problems caused at this point but it could cause a problem on a real prime in the future.

Can someone look into this for me?


Thanks,
Gary[/QUOTE]

The problem is under investigation, since I got the same mistakes doing base 3 tests. In my case the problem was due to a mistake in Georges FFT tables (i think if I remember correctly). Also please be aware that PFGW skips the entire test if it fails in stead of checking it with different FFT lengths like Prime95 is. This is most likely going to be changed in the future according to words.

Regards

KEP

[QUOTE=gd_barnes;183980]I'm getting an error on PFGW with my new Sierp base 52 reservation. Here it is:

[code]
-----------------------------------------------------------------------
Error occuring in PFGW at Tue Aug 04 05:07:35 2009
Expr = 9243*52^2046+1
Failed at bit 1877 of 11676
Msg = Detected in MAXERR>0.45 (round off check) in prp_using_gwnum
Iteration: 1877/11676 ERROR: ROUND OFF 0.46875>0.45
[/code]

I'm running on a 32-bit 1.6 Ghz HP dual-core laptop under Windows Vista.

Its running without sieving up to n=5000 so I have the -f100 switch set for trial factoring. A quick check using Syd's DB found no factors up to P=10M but it is composite. So no problems caused at this point but it could cause a problem on a real prime in the future.

Can someone look into this for me?


Thanks,
Gary[/QUOTE]

This works in the upcoming release, so the problem has been resolved. It was most likely caused by gwnum being overly aggressive in choosing the FFT length. I hope to have a new release by the weekend.

While you're at it, the mouse-over pop-up for version 3.2 says version 3.1. :smile:

[quote=gd_barnes;183980]I'm getting an error on PFGW with my new Sierp base 52 reservation. Here it is:

[code]
-----------------------------------------------------------------------
Error occuring in PFGW at Tue Aug 04 05:07:35 2009
Expr = 9243*52^2046+1
Failed at bit 1877 of 11676
Msg = Detected in MAXERR>0.45 (round off check) in prp_using_gwnum
Iteration: 1877/11676 ERROR: ROUND OFF 0.46875>0.45
[/code]

I'm running on a 32-bit 1.6 Ghz HP dual-core laptop under Windows Vista.

Its running without sieving up to n=5000 so I have the -f100 switch set for trial factoring. A quick check using Syd's DB found no factors up to P=10M but it is composite. So no problems caused at this point but it could cause a problem on a real prime in the future.

Can someone look into this for me?


Thanks,
Gary[/quote]
As Mark said, it will be fixed in the next release; in the meantime, it's not a serious problem. It should work if you re-run that one number with the -a1 switch; that will force PFGW to use the next-higher FFT length which is really the correct one (and therefore it should be able to complete the test and give you a stable residual).

[QUOTE=Flatlander;183986]While you're at it, the mouse-over pop-up for version 3.2 says version 3.1. :smile:[/QUOTE]

Can you include a screen shot of that? I don't know what you are looking at.

Never mind, I think I found it.

I think it's because if you right-click and choose Properties/Version, it says version 3.1

In Windows btw.

[quote=mdettweiler;184001]As Mark said, it will be fixed in the next release; in the meantime, it's not a serious problem. It should work if you re-run that one number with the -a1 switch; that will force PFGW to use the next-higher FFT length which is really the correct one (and therefore it should be able to complete the test and give you a stable residual).[/quote]

Yeah, I figured it's something minor. I just verified that it was composite at Alpertron's site. In the mean time, I got another one:

[quote]
-----------------------------------------------------------------------
Error occuring in PFGW at Tue Aug 04 19:11:21 2009
Expr = 6385*52^3070+1
Failed at bit 40 of 17512
Msg = Detected in MAXERR>0.45 (round off check) in prp_using_gwnum
Iteration: 40/17512 ERROR: ROUND OFF 0.5>0.45
[/quote]

I also verified that this one is composite at Alpertron's site.

Nothing earth shattering but I just wanted you to know in case it helps you detect a pattern of them or their frequency. The test that found these 2 problems is for all k's that are not k==(2 mod 3) on Sierp base 52 up to k=28673 and n=5000 so the problem is very infrequent for a fairly large # of k/n pairs tested. No sieving ahead of time was done. PFGW is set to do trial factoring with the -f100 switch.


Gary