A minor bug in PRP-24.14

Kosmaj · 2005-08-11, 12:11

I'm testing the latest PRP-24.14 and I found a minor bug.
On start-up when the output file is created (to write PRPs) the sieve limit is truncated to lower 32 bits. For example, the first line of my input file is:
837001000000:M:0:2:258

but the output file reads
3777344576:M:1:2:258

and Mod(837001000000, 2^32)=3777344576.

This is not important but I wonder how reliable is the PRP testing part. Last year there was a bug there so serious testing is required. OTOH, that's not so easy when new PRP versions are being released almost every month. So far I only tested 17 known k*2^n-1 primes and all were successfully detected.

Cruelty · 2005-08-29, 14:46

Have you tested 24.14 some more? I wonder if there is a possibility for PRP to "miss" a prime and what would be the probability of such event...
I am right now at ~110k with 736320585 and I am looking for a way (other than newpgen) to speed it up - PRP is faster than LLR - the question is: how reliable it is?

Kosmaj · 2005-08-29, 15:14

I did more tests on 24.14 but nothing substantial, just about 20 primes k*2^n-1, one half k<300 and the other "15k" like k's (10^8-10^9), n in the 200-300k range, and all were correctly found to be PRPs. I stopped tests because I found that on my Pentium-M laptop PRP-24.14 performs almost the same like the latest LLR. Maybe it's faster on P-4. To answer your question, with a high probability the PRP test part is bug-free because no bugs were reported since last year. But it's impossible to be certain until we have a stable version and more people interested in testing it.

BTW, you can go straight to n=200k, the small guys cannot be reported to Top-5000 and the probability to find a Sophie-Germain by chance at n=100k is miniscule. If you are interested in SG you have to sieve specifically for SG (or "lucky minus"), such options are available in NewPGen.

Cruelty · 2005-08-29, 20:28

I'm planning to make it a comprehensive search from 0 to 1M

2005-08-11, 12:11	#1
Kosmaj Nov 2003 E26₁₆ Posts	A minor bug in PRP-24.14 I'm testing the latest PRP-24.14 and I found a minor bug. On start-up when the output file is created (to write PRPs) the sieve limit is truncated to lower 32 bits. For example, the first line of my input file is: 837001000000:M:0:2:258 but the output file reads 3777344576:M:1:2:258 and Mod(837001000000, 2^32)=3777344576. This is not important but I wonder how reliable is the PRP testing part. Last year there was a bug there so serious testing is required. OTOH, that's not so easy when new PRP versions are being released almost every month. So far I only tested 17 known k*2^n-1 primes and all were successfully detected.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Minor Issue with Primenet	KyleAskine	PrimeNet	12	2012-02-12 01:37
A minor typo	Chuck	GPU to 72	1	2011-12-12 13:36
Minor changes to a lot of sequences	Greebley	Aliquot Sequences	18	2010-08-21 13:52
Minor GMP-ECM bug	jasonp	GMP-ECM	2	2007-11-25 18:40
Minor bug	PhilF	Software	1	2006-03-22 01:04

2005-08-29, 14:46	#2
Cruelty May 2005 2²·11·37 Posts	Have you tested 24.14 some more? I wonder if there is a possibility for PRP to "miss" a prime and what would be the probability of such event... I am right now at ~110k with 736320585 and I am looking for a way (other than newpgen) to speed it up - PRP is faster than LLR - the question is: how reliable it is?

2005-08-29, 15:14	#3
Kosmaj Nov 2003 E26₁₆ Posts	I did more tests on 24.14 but nothing substantial, just about 20 primes k2^n-1, one half k<300 and the other "15k" like k's (10^8-10^9), n in the 200-300k range, and all were correctly found to be PRPs. I stopped tests because I found that on my Pentium-M laptop PRP-24.14 performs almost the same like the latest LLR. Maybe it's faster on P-4. To answer your question, with a high probability* the PRP test part is bug-free because no bugs were reported since last year. But it's impossible to be certain until we have a stable version and more people interested in testing it. BTW, you can go straight to n=200k, the small guys cannot be reported to Top-5000 and the probability to find a Sophie-Germain by chance at n=100k is miniscule. If you are interested in SG you have to sieve specifically for SG (or "lucky minus"), such options are available in NewPGen.

2005-08-29, 20:28	#4
Cruelty May 2005 2²×11×37 Posts	I'm planning to make it a comprehensive search from 0 to 1M