Pminus1 - Bug report and feature request

[kruoli]

After removing the save files, I ran the same test again, this time with B1=200,000,000 while keeping MaxStage0Prime=100000000. Stage 0 ran about 10 % slower! Is this because 1277 is such a small exponent?

[axn]

Originally Posted by kruoli

After removing the save files, I ran the same test again, this time with B1=200,000,000 while keeping MaxStage0Prime=100000000. Stage 0 ran about 10 % slower! Is this because 1277 is such a small exponent?

How much time did it take for 0-100M part of the run (i.e., reaching 50% of stage 1) and how much time it took to reach from 100M-200M? We expect the first half to take x seconds, and second half to take about 1.5x seconds.

Or do you mean when you ran only B1=100M, stage 1 completed in x seconds, and when you ran B1=200M, stage 0 (0-100M) ran in x+10% seconds? That is kind of expected (but not sure how much it should be). This is because we are including all primes and prime powers < B1. Primes between sqrt(B1) and B1 are included only once. For B1=100M, that means primes between 10^4 and 100M. For B1=200M, that means primes between 2*10^4 and 100M (since stage 0 is capped at 100M). Therefore the primes between 10^4 and 2*10^4 are additional inclusions in stage 0 in the case of B1=200M. Similarly, some smaller primes get included more times in B2=200M case than B1=100M case, so the stage 0 will be slightly bigger, even though your max prime limit is the same.

EDIT:- Just did some checking. Whatever I said above shouldn't make any difference at all. It is all very negligible.

EDIT2:- sqrt(2e8) is 1.4e4, not 2e4 :-(

[kruoli]

Hopefully the full log is somewhat self-explanatory:

[Work thread Jun 8 11:41] P-1 on M1277 with B1=200000000
[Work thread Jun 8 11:41] Using AVX FFT length 64
[Work thread Jun 8 11:41] M1277 stage 1 is 3.46% complete. Time: 4.331 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 6.93% complete. Time: 4.350 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 10.39% complete. Time: 4.351 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 13.86% complete. Time: 4.360 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 17.32% complete. Time: 4.351 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 20.79% complete. Time: 4.343 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 24.25% complete. Time: 4.366 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 27.72% complete. Time: 4.342 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 31.19% complete. Time: 4.346 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 34.65% complete. Time: 4.346 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 38.12% complete. Time: 4.349 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 41.58% complete. Time: 4.352 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 45.05% complete. Time: 4.349 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 48.51% complete. Time: 4.344 sec.
[Work thread Jun 8 11:42] M1277 stage 1 is 51.98% complete. Time: 4.348 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 54.63% complete. Time: 4.129 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 56.97% complete. Time: 4.041 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 59.28% complete. Time: 4.040 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 61.62% complete. Time: 4.047 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 63.92% complete. Time: 4.048 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 66.20% complete. Time: 4.035 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 68.56% complete. Time: 4.035 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 70.95% complete. Time: 4.025 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 73.32% complete. Time: 4.026 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 75.65% complete. Time: 4.018 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 78.03% complete. Time: 4.015 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 80.36% complete. Time: 4.014 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 82.68% complete. Time: 4.036 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 85.01% complete. Time: 4.028 sec.
[Work thread Jun 8 11:43] M1277 stage 1 is 87.38% complete. Time: 4.010 sec.
[Work thread Jun 8 11:44] M1277 stage 1 is 89.72% complete. Time: 4.082 sec.
[Work thread Jun 8 11:44] M1277 stage 1 is 92.02% complete. Time: 4.064 sec.
[Work thread Jun 8 11:44] M1277 stage 1 is 94.35% complete. Time: 4.068 sec.
[Work thread Jun 8 11:44] M1277 stage 1 is 96.66% complete. Time: 4.089 sec.
[Work thread Jun 8 11:44] M1277 stage 1 is 98.96% complete. Time: 4.098 sec.
[Work thread Jun 8 11:44] M1277 stage 1 complete. 709217288 transforms. Time: 156.362 sec.
[Work thread Jun 8 11:44] Stage 1 GCD complete. Time: 0.000 sec.
[Work thread Jun 8 11:44] M1277 completed P-1, B1=200000000, Wh8: 17C7840E
[Work thread Jun 8 11:44] No work to do at the present time.  Waiting.

@George: Could you have a print out for how long initalization of stage 0 took? It doesn't seem to report that, currently.

[axn]

It looks like stage 0 took about 63 seconds, stage 1 took about 85 second and init took about 8 seconds.

EDIT:- What happens if you change the stage 0 max prime to 200M and rerun the B1=200M? Will it complete faster than 156 seconds?

[kruoli]

Yes, now it's 144 s. I forgot that stage 0 will print out in larger steps (in my example about 3.5 percent-point) vs stage 1 (about 2.3 percent points). First; I only looked at the times. Sorry for the trouble.

[axn]

Originally Posted by kruoli

Yes, now it's 144 s. I forgot that stage 0 will print out in larger steps (in my example about 3.5 percent-point) vs stage 1 (about 2.3 percent points). First; I only looked at the times. Sorry for the trouble.

Cool. I think this is a must-have feature for anyone doing super large P-1s.

[kruoli]

Originally Posted by kruoli

Same for me here, Windows 10, 64 bit. The process is visible for a second or two in task manager and then vanishes..

Windows log says:
Faulty module: libhwloc-15.dll
Error code: 0xc0000005
Error offset: 0x0000000000008d4e

Interesting...

[kruoli]

New workaround: Use the hwloc from 29.8b6, then 29.8b8 works!

Edit:
But it crashes when accessing libgmp now.
@George: Have you built all of the DLLs yourself and maybe it's using CPU instructions that are not supported on my processor (AVX512 etc.)? I ran the test on a i5-2400.

[Prime95]

I'm supposed to build a DLL that works on all machines. I may have screwed up.

[ixfd64]

Originally Posted by ixfd64

Dumb question: what does P-1 stage 0 do?

After reading some posts more closely, it seems "stage 0" in this context refers to what is normally called stage 1 elsewhere.

[Prime95]

Originally Posted by ixfd64

After reading some posts more closely, it seems "stage 0" in this context refers to what is normally called stage 1 elsewhere.

Yes, stage 0 is part of stage 1. Maybe I should have called them stage 1a and stage 1b.