Right Perfect Prime Numbers - Page 3

Batalov · 2016-01-27, 17:30

Much further. Of course.

Re: PFGW: Yes, one can try smaller members simply to see that the FFT size tends to be ~3.75-4 times larger. For P74207281, after long initialization the process takes off:

Code:

No factoring at all, not even trivial division
Generic modular reduction using generic reduction FMA3 FFT length 16000K, Pass1=640, Pass2=25K on A 148414563-bit number
PRP: 2^148414561-2^74207280+1 35000/148414560

ET_ · 2016-01-27, 17:44

Quote:

Originally Posted by Batalov

Much further. Of course.

Re: PFGW: Yes, one can try smaller members simply to see that the FFT size tends to be ~3.75-4 times larger. For P74207281, after long initialization the process takes off:

Code:

No factoring at all, not even trivial division
Generic modular reduction using generic reduction FMA3 FFT length 16000K, Pass1=640, Pass2=25K on A 148414563-bit number
PRP: 2^148414561-2^74207280+1 35000/148414560

What command line did you use?

wombatman · 2016-01-27, 17:56

Quote:

Originally Posted by Batalov

Much further. Of course.

Re: PFGW: Yes, one can try smaller members simply to see that the FFT size tends to be ~3.75-4 times larger. For P74207281, after long initialization the process takes off:

Code:

No factoring at all, not even trivial division
Generic modular reduction using generic reduction FMA3 FFT length 16000K, Pass1=640, Pass2=25K on A 148414563-bit number
PRP: 2^148414561-2^74207280+1 35000/148414560

Interesting. PFGW crashes for me trying to run that same expression. Tried both 3.7.7 and 3.7.10.

Batalov · 2016-01-27, 21:29

You'd need quite a bit of free memory (perhaps for the first, "giants", step; it can also be the reason for a very low start; "giants" object is created first and then converted to a gwnum object and may be used for the mod step; for a faster implementation, this needs to be custom written inside the code), maybe this is why it is crashes for you guys.

I've used three ways, and all worked (all on linux, can't vouch for W!ndows).
One (common with trying LLR*):
File a.abc

Code:

ABC $a^$b-$a^$c+1
2 148414561 74207280

And then ./pfgw -f0 -V -N -l a.abc and/or ./sllr -d a.abc.
Another, ./pfgw -f0 -V -N -q"2^148414561-2^74207280+1" (works, too).
Third one, ./pfgw -f0 -V -N -t -hhelper -q"2^148414561-2^74207280+1" (where helper file has 2^74207281-1 in it)

________________
*LLR turns out to be not immediately helpful for the purpose of the N-1 proof because it implements a >50% proof only (a k*2^n+1 approach with desired k < 2^n).

wombatman · 2016-01-27, 22:09

I'll try those. I tried "pfgw64.exe -f0 -q"2^148414561-2^74207280+1" on Windows 7 boxes with 32GB and 16GB respectively. Watching Task Manager, the process never pulled more than about 1 GB of RAM for use.

It's not a big deal, obviously, but now I'm just curious.

Edit: Yeah, none of those worked for me (pfgw or llr). I guess I'll stick to slightly lower numbers then.

Batalov · 2016-01-28, 01:25

It does segfault on Windows. What a terribly written program!!

Quote:

...unfortunate, but I suppose state of the art ~~factoring~~ primality software is unlikely to win any awards for user friendliness.

wombatman · 2016-01-28, 02:10

It's all good. This is the first issue I've run into with it after a good amount of use, so I'd say it's doing alright.

Batalov · 2016-01-28, 02:17

Of course! Mark is also very helpful,; the only wrinkle I thought of was that he doesn't read every thread (not that I blame him), so I PM'd him with a link, too. Undoubtedly this will get fixed.

rogue · 2016-01-28, 12:50

I finally read Batalov's PM. I'll see what I can do.

As for pfgw being "terribly written", I would agree. The code base is very convoluted. I have considered rewriting the program, but that would probably take months.

henryzz · 2016-01-28, 15:50

Quote:

Originally Posted by rogue

I finally read Batalov's PM. I'll see what I can do.

As for pfgw being "terribly written", I would agree. The code base is very convoluted. I have considered rewriting the program, but that would probably take months.

What would it take to add gwnum multithreading for prp testing? Is it a case of using different gwnum functions?

rogue · 2016-01-29, 01:15

Quote:

Originally Posted by henryzz

What would it take to add gwnum multithreading for prp testing? Is it a case of using different gwnum functions?

I'm not familiar with that feature of gwnum.

I have good news and bad news. The good news is that I have been able to reproduce the problem on Windows. The bad news is that the bug is causing memory to be overwritten so the call stack is lost. It says the memcpy is to blame, but it is not one of the many calls to memcpy in the application. I suspect that an assignment is calling memcpy under the covers. This will take a lot more work to debug. Ugh!

2016-01-27, 17:30	#23
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2·4,663 Posts	Much further. Of course. Re: PFGW: Yes, one can try smaller members simply to see that the FFT size tends to be ~3.75-4 times larger. For P74207281, after long initialization the process takes off: Code: No factoring at all, not even trivial division Generic modular reduction using generic reduction FMA3 FFT length 16000K, Pass1=640, Pass2=25K on A 148414563-bit number PRP: 2^148414561-2^74207280+1 35000/148414560

2016-01-27, 21:29	#26
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2·4,663 Posts	You'd need quite a bit of free memory (perhaps for the first, "giants", step; it can also be the reason for a very low start; "giants" object is created first and then converted to a gwnum object and may be used for the mod step; for a faster implementation, this needs to be custom written inside the code), maybe this is why it is crashes for you guys. I've used three ways, and all worked (all on linux, can't vouch for W!ndows). One (common with trying LLR): File a.abc Code: ABC $a^$b-$a^$c+1 2 148414561 74207280 And then ./pfgw -f0 -V -N -l a.abc and/or ./sllr -d a.abc. Another, ./pfgw -f0 -V -N -q"2^148414561-2^74207280+1" (works, too). Third one, ./pfgw -f0 -V -N -t -hhelper -q"2^148414561-2^74207280+1" (where helper file has 2^74207281-1 in it) ________________ LLR turns out to be not immediately helpful for the purpose of the N-1 proof because it implements a >50% proof only (a k*2^n+1 approach with desired k < 2^n).

2016-01-27, 22:09	#27
wombatman I moo ablest echo power! May 2013 1,741 Posts	I'll try those. I tried "pfgw64.exe -f0 -q"2^148414561-2^74207280+1" on Windows 7 boxes with 32GB and 16GB respectively. Watching Task Manager, the process never pulled more than about 1 GB of RAM for use. It's not a big deal, obviously, but now I'm just curious. Edit: Yeah, none of those worked for me (pfgw or llr). I guess I'll stick to slightly lower numbers then. Last fiddled with by wombatman on 2016-01-27 at 22:39

2016-01-28, 12:50	#31
rogue "Mark" Apr 2003 Between here and the 2²×7×223 Posts	I finally read Batalov's PM. I'll see what I can do. As for pfgw being "terribly written", I would agree. The code base is very convoluted. I have considered rewriting the program, but that would probably take months. Last fiddled with by rogue on 2016-01-28 at 12:51

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Is this new formula for Perfect Numbers useful?	mahbel	Miscellaneous Math	20	2017-03-01 22:41
How to generate base10 representation of Mersenne-prime perfect numbers?	James Heinrich	Miscellaneous Math	10	2012-03-08 07:20
Odd Perfect Numbers	davar55	Miscellaneous Math	16	2011-01-29 01:53
Perfect Numbers	MajUSAFRet	Math	3	2003-12-13 03:55
Odd Perfect Numbers	Zeta-Flux	Math	1	2003-05-28 19:41

2016-01-28, 02:10	#29
wombatman I moo ablest echo power! May 2013 1,741 Posts	It's all good. This is the first issue I've run into with it after a good amount of use, so I'd say it's doing alright.

2016-01-28, 02:17	#30
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 2·4,663 Posts	Of course! Mark is also very helpful,; the only wrinkle I thought of was that he doesn't read every thread (not that I blame him), so I PM'd him with a link, too. Undoubtedly this will get fixed.