Thread: Combined Sieve View Single Post
2020-12-11, 23:23   #11
SethTro

"Seth"
Apr 2019

26910 Posts

Quote:
 Originally Posted by henryzz For sanity's sake you will want to make sure that you are on a Ubuntu 20.04 version not 18.04 as that will have issues as mentioned above. Any other issues would be useful to record so they can be added to the instructions.
Please do comment on issues, I'll write this (sqlite3 > 3.18 + suggestion of Ubuntu 20.04) up in README.md#setup later today. Does this mean you got some/all of it running? under WSL?

Quote:
 Originally Posted by henryzz @Seth If I use gap_test.py to test the best 1% of a file, is there a way to test the next 1%? Is this the point of the database? Do I just run again?
Yes, the gap_test.py is smart, just run it again with what every total percent you want tested.

If you make the first call with --top-prp-percent=1 you can quit half way through and all progress is recorded so when you run again it will skip finished tests. If you run again and change to --top-prp-percent=5 it will test any record in the top 5% not yet tested (so 4% more if --top-prp-percent=1 had already finished).

gap_stats should have told you an approximately optimal number which is generally 10-40%.

Quote:
 Originally Posted by henryzz Is there an easy way of extracting the results from the log for submission to the online database?
Yes ./misc/record_check.py will print all the records that improve over your local gaps.db file (occasionally you should cd into prime-gap-list and git pull). The output can be pasted into primegaps.cloudygo.com/

Quote:
 Originally Posted by henryzz How many tests does pfgw run per call? There would be a lot less overhead if multiple are done in one call. I would imagine that this is more of an issue in WSL as there is extra overhead for disk access.
The code uses gmp for numbers with < 8000 bits and pfgw for larger numbers. Sadly it's currently making one call per prime (e.g. hundreds per interval) I haven't found any easy way to make multiple calls (maybe I can write a list of all the offsets to check like Robert did in https://www.mersenneforum.org/showpo...&postcount=203 but I wasn't sure if pfgw stopped after the first prime.

If the overhead is bad in the shortterm I'd recommend adjust the threshold (in gap_utils.py) up a bit.