Carol / Kynea search (Near-power primes) - Page 2

paulunderwood · 2016-02-10, 12:45

Okay, I think I can replace those routines with calls to gwtobinary32 and binary32togw and use some array arithmetic to achieve the mod calculation...

paulunderwood · 2016-02-12, 06:06

Quote:

real 3m44.806s
user 3m36.052s
sys 0m0.012s

For just squaring, multiplying by the base and converting back and forth to binary using the gwnum routines. 6.5 minutes with my first attempt at array arithmetic included. Is there a way to access the stored number directly? If so, how? And in what format is it?

rogue · 2016-02-16, 00:46

Quote:

Originally Posted by paulunderwood

For just squaring, multiplying by the base and converting back and forth to binary using the gwnum routines. 6.5 minutes with my first attempt at array arithmetic included. Is there a way to access the stored number directly? If so, how? And in what format is it?

You don't want to access it directly. In pfgw, you can convert between gnum and Integer types. The Integer class supports operator overloading.

rogue · 2016-02-16, 00:46

Completed both to n=400,000 and continuing.

Prime95 · 2016-02-16, 02:53

Quote:

Originally Posted by paulunderwood

converting back and forth to binary using the gwnum routines.

Conversion is a very slow process.

paulunderwood · 2016-02-16, 18:37

I have got the time down to 5:40 minutes (for a buggy program). This means the array arithmetic is taking 2:10 minutes, and I only have 30-40 seconds to play with. I really doubt I can beat Georges generic modular reduction.

rogue · 2016-02-29, 03:03

Add (2^442042+1)^2-2 as a new prime!

rogue · 2016-03-25, 23:37

Completed to 475,000. Nothing new, but continuing

rogue · 2016-04-07, 22:01

Completed to 505,000. Nothing new, but continuing

rogue · 2016-04-08, 22:10

I have a command line siever that works, but it slower than MultiSieve for base 2. I couple of reasons that it is slower is that I have generic logic for base 2 and poorly coded hash map code. Fortunately I think the non base 2 code is much faster the MultiSieve. The negative with MultiSieve is that it misses some factors, but that isn't a big enough issue to use the new code (yet).

rogue · 2016-04-12, 01:05

Here is a Windows build of my sieving code. I (ahem) borrowed parts of the discrete log code from srsieve so I won't release the source until I clean it up and remove references to that. This code runs about 10% faster than MultiSieve (for p < 1e9), unless you are logging factors. That slows it down a bit based upon my testing, but as all factors are double-checked, that shouldn't be a big deal. I've stated before MultiSieve misses 10% to 15% of the factors due to some bug, but as cksieve is faster there is no reason to continue using MultiSieve for sieving near-square primes. One core should be able to sieve about 150G for a range of 500,000 n per day. Unlike MultiSieve, this code is base agnostic, so sieving shouldn't be impacted when sieving a larger base. There is multi-threading code in the source, but I know it doesn't work so I don't know if I'm going to try to fix it or not.

2016-02-10, 12:45	#12
paulunderwood Sep 2002 Database er0rr 7233₈ Posts	Okay, I think I can replace those routines with calls to gwtobinary32 and binary32togw and use some array arithmetic to achieve the mod calculation... Last fiddled with by paulunderwood on 2016-02-10 at 12:54

2016-04-12, 01:05	#22
rogue "Mark" Apr 2003 Between here and the 11×577 Posts	Here is a Windows build of my sieving code. I (ahem) borrowed parts of the discrete log code from srsieve so I won't release the source until I clean it up and remove references to that. This code runs about 10% faster than MultiSieve (for p < 1e9), unless you are logging factors. That slows it down a bit based upon my testing, but as all factors are double-checked, that shouldn't be a big deal. I've stated before MultiSieve misses 10% to 15% of the factors due to some bug, but as cksieve is faster there is no reason to continue using MultiSieve for sieving near-square primes. One core should be able to sieve about 150G for a range of 500,000 n per day. Unlike MultiSieve, this code is base agnostic, so sieving shouldn't be impacted when sieving a larger base. There is multi-threading code in the source, but I know it doesn't work so I don't know if I'm going to try to fix it or not. Last fiddled with by rogue on 2020-09-24 at 19:47

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Carol / Kynea Coordinated Search - Reservations	rogue	And now for something completely different	293	2021-06-23 11:39
Carol / Kynea Primes	rogue	And now for something completely different	249	2021-05-19 12:14
Search primes of form 2*n^n ± 1	JeppeSN	And now for something completely different	27	2018-04-12 14:20
Factorial primes search?	flava	Open Projects	18	2010-12-04 05:24
Why Search for these Huge Primes?	Unregistered	Math	8	2005-04-27 00:55

2016-02-16, 00:46	#15
rogue "Mark" Apr 2003 Between here and the 11×577 Posts	Completed both to n=400,000 and continuing.

2016-02-16, 18:37	#17
paulunderwood Sep 2002 Database er0rr 3,739 Posts	I have got the time down to 5:40 minutes (for a buggy program). This means the array arithmetic is taking 2:10 minutes, and I only have 30-40 seconds to play with. I really doubt I can beat Georges generic modular reduction.

2016-02-29, 03:03	#18
rogue "Mark" Apr 2003 Between here and the 11·577 Posts	Add (2^442042+1)^2-2 as a new prime!

2016-03-25, 23:37	#19
rogue "Mark" Apr 2003 Between here and the 11×577 Posts	Completed to 475,000. Nothing new, but continuing

2016-04-07, 22:01	#20
rogue "Mark" Apr 2003 Between here and the 6347₁₀ Posts	Completed to 505,000. Nothing new, but continuing

2016-04-08, 22:10	#21
rogue "Mark" Apr 2003 Between here and the 14313₈ Posts	I have a command line siever that works, but it slower than MultiSieve for base 2. I couple of reasons that it is slower is that I have generic logic for base 2 and poorly coded hash map code. Fortunately I think the non base 2 code is much faster the MultiSieve. The negative with MultiSieve is that it misses some factors, but that isn't a big enough issue to use the new code (yet).