primo question

[CRGreathouse]

Originally Posted by fivemack

So, thanks very much for the timings so far, but I'm afraid I won't have comparable pari timings for several days, maybe not until next weekend

It's OK, Primo is still running on 10^3000+1027. I'm at 9241/9966 bits on phase 2.

[schickel]

Originally Posted by fivemack

So, thanks very much for the timings so far, but I'm afraid I won't have comparable pari timings for several days, maybe not until next weekend

Don't worry, I'm not going anywhere. My biggest lesson so far is that Celerons really suck at this.

After 64 hours, I'm at 5871/9966 and still on pahse 1......extrapolating with a WAG, I'd say at least another 18 hours or so, depending on how much acceleration there is in the last stages.

Oh, make that 5830/9966, it managed another 41 bits while I was composing this.

[CRGreathouse]

Primo 3.0.2 on one core of a Pentium D 915 @ 2.8 GHz under Windows Vista Business (32-bit):
10^2000-9297: 12h08m07s
10^3000+1027: 114h54m18s

[fivemack]

ECPP 6.4.5 on one core of K8/2400: 10^1000+453 in 8821 seconds

currently running 10^1500+2329 and 10^2000+4561. Not anticipating answers any time soon.

[CRGreathouse]

Primo 3.0.2 on one core of a Pentium D 915 @ 2.8 GHz under Windows Vista Business (32-bit):
10^1000+453: 39m24s (3144s)
10^2000-9297: 12h08m07s (43687s)
10^3000+1027: 114h54m18s (413658s)

[schickel]

Well, I probably won't be doing any more of these until I get some more horsepower:

[Running Times]
Initialization=17.95s
1stPhase=71h 10mn 45s
2ndPhase=23h 34mn 55s
Total=94h 45mn 58s

[Candidate]
File=C:\Program Files\Primo\work\FiveMackSecond.in
Expression=10^3000+1027

Stats again: Primo 3.0.6, 3.06 GHz Celeron, WinXP Home SP3, max RAM the several times I peaked in were <9 MB Phase 1, <20 MB Phase 2.

Oh, and since I didn't post the Phase 1/2 split from the first, here it is:

[Running Times]
Initialization=6.48s
1stPhase=9h 19mn 23s
2ndPhase=1h 34mn 11s
Total=10h 53mn 40s

[Candidate]
File=C:\Program Files\Primo\work\FiveMackFirst.in
Expression=10^2000-9297

[schickel]

Originally Posted by fivemack

pari-gp probably does count as a careful implementation of APRCL with unimpeded source code (it's a step-by-step implementation of the algorithm as given in Cohen's book, down to the function names)

Is there a source of the articles about APRT/APRT-CL freely available on the web or would I have to spring for the book (or article from jstor) to read them? I've looked at the ubasic implementation but looking at code with no underlying explanation isn't very helpful. The pari-gp version suffers the same problem: code with no explanation doesn't really help.

[wpolly]

Originally Posted by schickel

Is there a source of the articles about APRT/APRT-CL freely available on the web or would I have to spring for the book (or article from jstor) to read them? I've looked at the ubasic implementation but looking at code with no underlying explanation isn't very helpful. The pari-gp version suffers the same problem: code with no explanation doesn't really help.

Look at the source code of Tony Forbes' VFYPR.

[ldesnogu]

Originally Posted by wpolly

Look at the source code of Tony Forbes' VFYPR.

It lacks at least one very good source of information: Bosma and van der Hulst PhD thesis.
Wieb Bosma, Marc-Paul van der Hulst, Primality proving with cyclotomy (Ph.D. thesis), Universiteit van Amsterdam, 1990.

[FactorEyes]

Maybe now that there's less belief that Cryptography is the way to endless wealth, ECPP and elliptic-curve point counting algorithms can return to the realm of mathematical software.

Man, you are correct.

The lack of public-domain ECPP source code is bizarre.

[CRGreathouse]

Agreed.