mersenneforum.org - The Original 'Ode To Joy' thread

[QUOTE=cheesehead;243113].bu and .bu2 files?[/QUOTE]
When Prime95 saves more than one save file for a test, it names the extra ones [save file name].bu, .bu2, etc. .bu and .bu2 are the file extensions for the extra save files when prime95 is set to its default of 3 save files.
[QUOTE=cheesehead;243113]So, it doesn't happen (so far as we know) when the B1 is increased before resuming an interrupted test, but only when it's decreased.[/QUOTE]
Right. As a test, I did the reverse, (i.e. starting B1 at 1700 and changing it to 21000) but without running it to completion of stage 1, and it seemed to silently ignore the save file's B1 and use the larger B1 value. It was able to use the old save files to continue from where that was (not in percentage, of course, but in apparent number of iterations complete). I'm not entirely sure if that means the end result is still correct or not.
Edit: Then I changed the B1 back to 1700, and it picked up where it left off yet again! (i.e. it went something like this: I started at 0% with B1=1700, ran to >8%, switched to B1=21000 which for that was ~0.7%, ran that to >1%, then switched back to B1=1700 where the current progress made it ~19.5%, ran that to ~24%, switched back which is now 2.13%, and am now continuing until I get about 10% of B1=21000, where I'll be well over B1=1700's 100% point, then I'll switch back to that and see what happens) Hmm, I wonder what would happen if I put it past B1=1700's 100% point by running it with B1=21000 and then put it back to B1=1700...
Edit 2: I think I confused Prime95... When it got to the 8.33% point (for B1=21000), which for B1=1700 is 100%, it made some weird and inconsistent output:[code][Dec 23 14:51] Worker starting
[Dec 23 14:51] Setting affinity to run worker on any logical CPU.
[Dec 23 14:51] P-1 on M53035711 with B1=21000
[Dec 23 14:51] Using Core2 type-3 FFT length 2880K, Pass1=640, Pass2=4608
[Dec 23 14:51] M53035711 stage 1 is 2.132337% complete.
[Dec 23 14:51] M53035711 stage 1 is 2.458382% complete. Time: 5.693 sec.
[Dec 23 14:51] M53035711 stage 1 is 2.784428% complete. Time: 5.716 sec.
[Dec 23 14:51] M53035711 stage 1 is 3.110473% complete. Time: 5.661 sec.
[Dec 23 14:51] M53035711 stage 1 is 3.436519% complete. Time: 5.733 sec.
[Dec 23 14:51] M53035711 stage 1 is 3.762564% complete. Time: 5.671 sec.
[Dec 23 14:51] M53035711 stage 1 is 4.088610% complete. Time: 5.703 sec.
[Dec 23 14:51] M53035711 stage 1 is 4.414655% complete. Time: 5.633 sec.
[Dec 23 14:51] M53035711 stage 1 is 4.740701% complete. Time: 5.640 sec.
[Dec 23 14:52] M53035711 stage 1 is 5.066746% complete. Time: 5.648 sec.
[Dec 23 14:52] M53035711 stage 1 is 5.392792% complete. Time: 5.640 sec.
[Dec 23 14:52] M53035711 stage 1 is 5.718837% complete. Time: 5.638 sec.
[Dec 23 14:52] M53035711 stage 1 is 6.044883% complete. Time: 5.726 sec.
[Dec 23 14:52] M53035711 stage 1 is 6.370928% complete. Time: 5.681 sec.
[Dec 23 14:52] M53035711 stage 1 is 6.696974% complete. Time: 5.713 sec.
[Dec 23 14:52] M53035711 stage 1 is 7.023019% complete. Time: 5.754 sec.
[Dec 23 14:52] M53035711 stage 1 is 7.349065% complete. Time: 5.650 sec.
[Dec 23 14:52] M53035711 stage 1 is 7.675110% complete. Time: 5.707 sec.
[Dec 23 14:52] M53035711 stage 1 is 8.001156% complete. Time: 5.775 sec.
[Dec 23 14:52] M53035711 stage 1 is 0.204761% complete. Time: 6.313 sec.
[Dec 23 14:53] M53035711 stage 1 is 0.480952% complete. Time: 6.156 sec.
[Dec 23 14:53] M53035711 stage 1 is 8.195238% complete. Time: 6.612 sec.
[Dec 23 14:53] M53035711 stage 1 is 8.461904% complete. Time: 6.632 sec.
[Dec 23 14:53] M53035711 stage 1 is 8.680952% complete. Time: 5.803 sec.
[Dec 23 14:53] M53035711 stage 1 is 8.938095% complete. Time: 6.415 sec.
[Dec 23 14:53] M53035711 stage 1 is 9.204761% complete. Time: 6.616 sec.
[Dec 23 14:53] M53035711 stage 1 is 9.490476% complete. Time: 5.860 sec.
[Dec 23 14:53] M53035711 stage 1 is 9.652380% complete. Time: 6.004 sec.
[Dec 23 14:53] M53035711 stage 1 is 9.919047% complete. Time: 6.075 sec.
[Dec 23 14:53] Worker stopped.
[/code]
Up thru 8.00%, the percentage increase on each line was a consistent 0.3260%, after that the percentage went down to 0.20%, then 0.48%, then up to 8.20% and continued on, but not consistently. It averaged about 0.25% per line, but had a standard deviation of about 0.04%. The time per line also slowed, from about 5.69 sec to 6.2 sec.
Then I switched the B1 back to 1700. It immediately ran the GCD for stage 1 (no factor, as expected) and went to stage 2.
So, all in all, very odd, but not very bad handling of it when you increase the B1. But I'd definitely say the bigger bug is the previously reported one, that it might silently report a different B1 (and B2?) than what's really being run.
[QUOTE=cheesehead;243113]I'm still studying the code in ecm.c because I'm not sure I understand all the cases, but I think that adjusting B1=1700 to the old higher B1 was [I]not[/I] an error (nor is a result of misplaced code). The error was that there was no display message along the lines of "New B1 is less than old B1; continuing with old B1".[/QUOTE]
Yes, I agree. :smile:
[QUOTE=cheesehead;243113](... and this post, along with all the preceding posts about this bug, really ought to be moved out of this thread, to a separate thread in the Software subforum.)[/QUOTE]
Yes, it should.