Testing new Ranges for Sierpinski/Riesel - Page 24

rogue · 2017-08-24, 20:10

Quote:

Originally Posted by wombatman

If you have Windows 10, it will build in the Ubuntu shell with

Code:

g++ nsieve64.cpp -o nsieve64

And testing with

Code:

./nsieve64 -s 6209,800000000,800000001 -N 6299 -o nsieve64_testoutput.txt -p 2,1000000000000

seems to be working. One note: you cannot currently use abbreviated inputs like 1e12; they must be typed out fully.

I recommend building with this:

g++ -O3 -m64 nsieve64.cpp -lstdc++ -o nsieve

It is currently slower than fermfact, but I found one optimization that seems to make it faster for multiple n, but I need to run some side-by-side comparisons with fermfact. If someone can build and compare with newpgen (for a single n), that would be great. Note that newpgen stops automatically when p > sqrt(maxk*2^n+1). nsieve does not stop automatically.

I also intend to modify nsieve to be multi-threaded, but I'm concerned about the performance it could take.

rogue · 2017-08-24, 20:26

I just ran one test. nsieve and newpgen do not produce the same outputs. I assume that nsieve must have some bugs in it. That will take time to resolve.

wombatman · 2017-08-24, 20:30

Yeah, I was just about to post the same thing. Using n=30 with k=8e8-800000500, I get 12 candidates remaining from newpgen. Nsieve's last output on screen shows 8 remaining, and the output file has the proper header and then nothing else.

pepi37 · 2017-08-24, 20:31

I got empty output from your build Rogue

rogue · 2017-08-24, 20:36

Quote:

Originally Posted by pepi37

I got empty output from your build Rogue

It isn't my code. I'm trying to find contact information for Ken Brazier as he wrote it although I might be able to debug without his help.

rogue · 2017-08-26, 17:21

It appears to be a miscompile on Windows with mingw. The outputs on my Mac are different and appear to be correct, although I haven't run any comparisons.

wombatman · 2017-08-26, 20:07

Are you able to compile it on a Linux system at all and confirm whether it mismatches?

rogue · 2017-08-29, 00:29

Quote:

Originally Posted by rogue

It appears to be a miscompile on Windows with mingw. The outputs on my Mac are different and appear to be correct, although I haven't run any comparisons.

The first test I ran shows non-matching output on my Mac. I don't think this is a miscompile. It is either bad assembly code or the prime sieve is missing primes.

rogue · 2017-08-30, 23:27

nsieve is only for base 2, so it isn't too helpful for this project.

rogue · 2017-12-15, 16:48

I ran into an issue with srbsieve. When it breaks up large ranges for PRP testing it is skipping some of those ranges. I don't how that is happening, but I'm looking into it.

This means that if you did a large range of k with srbsieve it is likely that some of the k in pl_remain.txt have a prime for n < 25000. This has a definite impact on R3 and S3, but I can't speak for other bases. If you have used srbsieve then you should run from n=1 to 25000 for the k in pl_remain.txt to find any primes that it missed. There are no issues with the numbers in pl_prime.txt. They will be prime. Unfortunately this is going to create a lot of work for me due to the work I have done on S3.

Hopefully I will have time between Christmas and New Year's to track down the problem and fix it.

KEP · 2017-12-15, 18:42

Does this problem concern all versions of srbsieve or only those that did checkpoint and allowed to resume? I used for my R3 ranges, the version that did not support stopping and starting again - in other words no resume was allowed.

2017-08-24, 20:26	#255
rogue "Mark" Apr 2003 Between here and the 18D0₁₆ Posts	I just ran one test. nsieve and newpgen do not produce the same outputs. I assume that nsieve must have some bugs in it. That will take time to resolve. Last fiddled with by rogue on 2017-08-24 at 20:26

2017-12-15, 16:48	#263
rogue "Mark" Apr 2003 Between here and the 14320₈ Posts	I ran into an issue with srbsieve. When it breaks up large ranges for PRP testing it is skipping some of those ranges. I don't how that is happening, but I'm looking into it. This means that if you did a large range of k with srbsieve it is likely that some of the k in pl_remain.txt have a prime for n < 25000. This has a definite impact on R3 and S3, but I can't speak for other bases. If you have used srbsieve then you should run from n=1 to 25000 for the k in pl_remain.txt to find any primes that it missed. There are no issues with the numbers in pl_prime.txt. They will be prime. Unfortunately this is going to create a lot of work for me due to the work I have done on S3. Hopefully I will have time between Christmas and New Year's to track down the problem and fix it. Last fiddled with by rogue on 2017-12-15 at 16:49

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Very Prime Riesel and Sierpinski k	robert44444uk	Open Projects	587	2016-11-13 15:26
Sierpinski/ Riesel bases 6 to 18	robert44444uk	Conjectures 'R Us	139	2007-12-17 05:17
Sierpinski/Riesel Base 10	rogue	Conjectures 'R Us	11	2007-12-17 05:08
Sierpinski / Riesel - Base 23	michaf	Conjectures 'R Us	2	2007-12-17 05:04
Sierpinski / Riesel - Base 22	michaf	Conjectures 'R Us	49	2007-12-17 05:03

2017-08-24, 20:30	#256
wombatman I moo ablest echo power! May 2013 2²×5×89 Posts	Yeah, I was just about to post the same thing. Using n=30 with k=8e8-800000500, I get 12 candidates remaining from newpgen. Nsieve's last output on screen shows 8 remaining, and the output file has the proper header and then nothing else.

2017-08-24, 20:31	#257
pepi37 Dec 2011 After milion nines:) 2³·181 Posts	I got empty output from your build Rogue

2017-08-26, 17:21	#259
rogue "Mark" Apr 2003 Between here and the 2⁴×397 Posts	It appears to be a miscompile on Windows with mingw. The outputs on my Mac are different and appear to be correct, although I haven't run any comparisons.

2017-08-26, 20:07	#260
wombatman I moo ablest echo power! May 2013 2²×5×89 Posts	Are you able to compile it on a Linux system at all and confirm whether it mismatches?

2017-08-30, 23:27	#262
rogue "Mark" Apr 2003 Between here and the 14320₈ Posts	nsieve is only for base 2, so it isn't too helpful for this project.

2017-12-15, 18:42	#264
KEP Quasi Admin Thing May 2005 2×3×7×23 Posts	Does this problem concern all versions of srbsieve or only those that did checkpoint and allowed to resume? I used for my R3 ranges, the version that did not support stopping and starting again - in other words no resume was allowed.