20140707, 15:17  #45  
"Mark"
Apr 2003
Between here and the
19×367 Posts 
Quote:
As for two "f" options, I modified factorsinrate to use F. If anyone can provide some speed comparisons with MultiSieve that would be nice. I'm also working on a change to pfgw so that the factors output from sieving programs can be easily verified. I though I had done something like that already, but maybe I never released that code. Last fiddled with by rogue on 20140707 at 15:43 

20140707, 17:38  #46 
"Mark"
Apr 2003
Between here and the
19·367 Posts 
The "Out of resources" seems to have been caused by having too many small factors. I'll do two things to address that. First, I'll increase the buffer size so that more factors can be collected. (I've also added a check so the program can tell you if you missed factors due to too many being found.) Second, I've added a "small primes" check to eliminate all terms that are factored by p < 100 before I start the GPU sieve. That can eliminate 75% of the terms up front.

20140707, 18:36  #47 
I moo ablest echo power!
May 2013
2^{2}×463 Posts 
Excellent!

20140707, 19:55  #48 
"Mark"
Apr 2003
Between here and the
1101100111101_{2} Posts 

20140707, 19:56  #49 
"Mark"
Apr 2003
Between here and the
19×367 Posts 

20140707, 20:12  #50 
I moo ablest echo power!
May 2013
2^{2}×463 Posts 
Why not both?

20140707, 20:47  #51 
"Mark"
Apr 2003
Between here and the
1B3D_{16} Posts 
Attached is the latest code. Here is what the output looks like now. This is a rather extreme range to do. It still fails if the range gets larger. I know why, but fixing is a bigger challenge. It is probably better to take smaller ranges and paste together one file and restart from there, although I haven't test the ability to restart the sieve from the output file.
Code:
xyyxsievecl64.exe x2 X2500 y3 Y7500 P1e6 t2 s2000 S+ oxyyx.pfgw xyyxsievecl v1.0.1, a GPU program to find factors numbers of the form x^y+y^x Quick elimination of terms info (in order of check): 7807501 because the term is even 1478788 because x and y have a common divisor 4472597 because the term is divisible by a prime < 100 Sieve started: (cmdline) 0 <= p < 1000000 with 1857365 terms Thread 2 has completed 1818380 of 1862363 iterations CTRLC accepted. Please wait for threads to completed.s/factor, 0.15 CPU cores, 25.6% done. ETA 07 Jul 15:53 Thread 2 has completed 383059 of 390048 iterations Sieve interrupted: 3 <= p < 1000000 26624 primes tested Clock time: 539.27 seconds at at 49 p/sec. Factors found: 1472315 Processor time: 92.09 sec. (8.53 init + 83.55 sieve). Seconds spent in CPU and GPU: 65.70 (cpu), 1044.80 (gpu) Percent of time spent in CPU vs. GPU: 5.92 (cpu), 94.08 (gpu) CPU/GPU utilization: 0.17 (cores), 1.00 (devices) Percent of GPU time waiting for GPU: 46.87 Started with 1857365 terms and sieved to 281719. 385050 remaining terms written to xyyx.pfgw 
20140707, 20:56  #52 
"Mark"
Apr 2003
Between here and the
19·367 Posts 
I think I just found another "Out of resources" trigger. Working on it.

20140707, 22:05  #53 
"Mark"
Apr 2003
Between here and the
1101100111101_{2} Posts 
I fixed that one, but I still have a problem elsewhere. It is only an issue when trying to do obscenely large search spaces. I tried to implement different logic to collect factors, but my video card does not support the OpenCL functcion atom_min.

20140707, 23:44  #54 
I moo ablest echo power!
May 2013
11100111100_{2} Posts 
No problem. Like you said, smaller ranges work fine. I just wanted to point it out in case it was a larger issue than that. Very impressive work!

20140708, 12:04  #55 
"Mark"
Apr 2003
Between here and the
1B3D_{16} Posts 
Thanks. This was a very difficult program to optimize. If the range of x is large you think of optimizing in one direction. If the range of y is large, then you think of optimizing in a different direction. If both are large, then you think of a way to get the best of both worlds. Unfortunately memory and latency quickly get in the way of most potential optimizations.
Due to the bug in MultiSieve, I'm going to sieve x^y+y^x for x and y < 10000. I will also update PRPNet to support an x^y+y^x search. I don't know if I'll be able to run every term (as there will probably be close to a million after sieving), but since all of them are small it might be doable. 
Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Distribution of Mersenne primes before and after couples of primes found  emily  Math  35  20221221 16:32 
Mersenne Primes p which are in a set of twin primes is finite?  carpetpool  Miscellaneous Math  4  20220714 02:29 
Leyland Primes: ECPP proofs  Batalov  XYYXF Project  57  20220630 17:24 
On Leyland Primes  davar55  Puzzles  9  20160315 20:55 
possible primes (real primes & poss.prime products)  troels munkner  Miscellaneous Math  4  20060602 08:35 