Thinking out loud about getting under 20M unfactored exponents - Page 22

kruoli · 2020-12-17, 21:33

I'd be willing to implement a script for finding known factors of mostly Mersenne numbers, at least (if they were found by P-1 or ECM).

@James et al., could you compile a list of those factors? Not all of them, but some?

Edit 1: The script should create worktodo.txt entries and then I'll be running them.

@George, do we have to test different machines or is it sufficient to test on a single machine, when assuming the big number arithmetic itself is rock solid?

Edit 2: I may have forgotten; how to specify Sigma when executing ECM with Prime95 or mprime?

masser · 2020-12-17, 22:49

Quote:

Originally Posted by Prime95

Any interest? Please specify Windows or Linux.

I can test a linux version. Might not get to it until the weekend, though.

nordi · 2020-12-17, 22:54

Quote:

Originally Posted by Prime95

I'm looking for volunteers to test a pre-release of version 30.4. 30.4 is a nearly complete rewrite of the ECM and P-1 code.

I'm interested in trying it out on Linux. Roughly when will the testing begin, and when do you need the results?

petrw1 · 2020-12-17, 22:55

Quote:

Originally Posted by Prime95

I'm looking for volunteers to test a pre-release of version 30.4. 30.4 is a nearly complete rewrite of the ECM and P-1 code.

Is this for PFACTOR only or also for PMINUS1?

Prime95 · 2020-12-17, 23:18

Quote:

Originally Posted by petrw1

Is this for PFACTOR only or also for PMINUS1?

Pfactor and P-1 are the same (in both existing and 30.4 versions). The only difference is Pfactor computes the P-1 bounds based on expected #LL tests saved, Pminus1 uses user specified bounds.

30.4 will prefer you use this format for your P-1 work:
Pminus1=1,2,n,-1,B1,B2_which_will_be_ignored,TF_sieve_depth
This is different than what is displayed on the James' "poorly factored" page.

Quote:

Originally Posted by kruoli

@George, do we have to test different machines or is it sufficient to test on a single machine, when assuming the big number arithmetic itself is rock solid?

The FFT code changed a little, but I'm more worried about the ECM and P-1 C code.
The change to FFT code is an assembly implementation of (a+b)*c and (a-b)*c using less memory bandwidth. I only implemented this for AVX, FMA3, and AVX-512 -- for older machines that operation is emulated.

Quote:

Originally Posted by nordi

I'm interested in trying it out on Linux. Roughly when will the testing begin, and when do you need the results?

Let me make a few more tweaks and tests, then I'll build a Windows and Linux executable.

Uncwilly · 2020-12-17, 23:24

I am willing to run it on a couple of Win64 machines.

LaurV · 2020-12-18, 04:36

Quote:

Originally Posted by Prime95

I'm looking for volunteers to test a pre-release of version 30.4.

Put me on the list and PM link if Win7/i7-6950X is what you look for as a OS/CPU.
I'm in holiday next week, and not going anywhere (moving the rubbish around the house from here to there and back, etc.)

ET_ · 2020-12-18, 12:59

Quote:

Originally Posted by Prime95

I'm looking for volunteers to test a pre-release of version 30.4. 30.4 is a nearly complete rewrite of the ECM and P-1 code.

Any interest? Please specify Windows or Linux.

If you still need Linux testes, I'm here to help.

Luigi

firejuggler · 2020-12-18, 13:58

Might I suggest 32148013 as one of the PM1 test?
B1=451 519 /B2= 2 846 113
This ought to find a composite 172 bit factor (2)

lycorn · 2020-12-18, 15:37

@george: count me in. Win 10 64-bit on a kaby lake 16 GB machine, and Linux on several Colab instances.

Prime95 · 2020-12-18, 22:48

Warnings:
1) Resuming stage 2 P-1 or ECM from a v30.3 save file will not work. Stage 2 will start again from scratch.
2) I've not tested resuming from a P-1 v30.3 save file. It is supposed to work, but may not due to substantial changes in save files. Report a bug if you encounter an error.

I recommend installing v30.4 in a new directory and copying over your prime.txt, local.txt, worktodo.txt and save files. Thanks for your help in locating bugs or suggestions for more improvements.

Win64 version:

https://www.dropbox.com/s/ilw7k3x3om...win64.zip?dl=0

Linux64 soon.

P.S. Calculating optimal B2 for P-1 and ECM is based upon estimates of the work involved in B1 and B2. If there are bugs in the cost estimating code, wrong optimal B2 will be calculated. Thus, if you find a "head scratcher" such as "B2 is faster with 6GB of memory than with 8GB" please report your observations.

2020-12-17, 21:33	#232
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 439 Posts	I'd be willing to implement a script for finding known factors of mostly Mersenne numbers, at least (if they were found by P-1 or ECM). @James et al., could you compile a list of those factors? Not all of them, but some? Edit 1: The script should create worktodo.txt entries and then I'll be running them. @George, do we have to test different machines or is it sufficient to test on a single machine, when assuming the big number arithmetic itself is rock solid? Edit 2: I may have forgotten; how to specify Sigma when executing ECM with Prime95 or mprime? Last fiddled with by kruoli on 2020-12-17 at 21:38 Reason: Additional information.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Thinking of Joining GPU to 72	jschwar313	GPU to 72	3	2016-01-31 00:50
Thinking about lasieve5	Batalov	Factoring	6	2011-12-27 22:40
Thinking about buying a panda	jasong	jasong	1	2008-11-11 09:43
Loud thinking on irregular primes	devarajkandadai	Math	4	2007-07-25 03:01
Question on unfactored numbers...	WraithX	GMP-ECM	1	2006-03-19 22:16

2020-12-17, 23:24	#237
Uncwilly 6809 > 6502 """"""""""""""""""" Aug 2003 101×103 Posts 2·3²·521 Posts	I am willing to run it on a couple of Win64 machines.

2020-12-18, 13:58	#240
firejuggler Apr 2010 Over the rainbow 2×5×11×23 Posts	Might I suggest 32148013 as one of the PM1 test? B1=451 519 /B2= 2 846 113 This ought to find a composite 172 bit factor (2)

2020-12-18, 15:37	#241
lycorn Sep 2002 Oeiras, Portugal 2³×181 Posts	@george: count me in. Win 10 64-bit on a kaby lake 16 GB machine, and Linux on several Colab instances.

2020-12-18, 22:48	#242
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 17×433 Posts	Warnings: 1) Resuming stage 2 P-1 or ECM from a v30.3 save file will not work. Stage 2 will start again from scratch. 2) I've not tested resuming from a P-1 v30.3 save file. It is supposed to work, but may not due to substantial changes in save files. Report a bug if you encounter an error. I recommend installing v30.4 in a new directory and copying over your prime.txt, local.txt, worktodo.txt and save files. Thanks for your help in locating bugs or suggestions for more improvements. Win64 version: https://www.dropbox.com/s/ilw7k3x3om...win64.zip?dl=0 Linux64 soon. P.S. Calculating optimal B2 for P-1 and ECM is based upon estimates of the work involved in B1 and B2. If there are bugs in the cost estimating code, wrong optimal B2 will be calculated. Thus, if you find a "head scratcher" such as "B2 is faster with 6GB of memory than with 8GB" please report your observations.