Discussion related to cado-nfs

[paul zimmermann <Paul.Zimmermann@inria.fr>]

       Sam,

> From: "Gross, Samuel" <Samuel.Gross@noblis.org>
> Date: Wed, 1 Feb 2017 20:03:12 +0000
>
> Hello,

> I have been doing some heuristics for different configurations in bwc. I 
> haven’t found an efficient way of forgetting a previous run of bwc and 
> starting over. I have tried 2 things. It seems that discarding all bwc 
> files (leaving an empty $name.bwc directory), and all merge files, dep 
> files, character, and sqrt files, and use the allow_wipeout=True parameter, 
> I am able to start again, but I must repeat merging. Deleting any other 
> combination of files, attempting to preserve the merged matrix, results in 
> file not found errors.

> I have also tried just starting a new working directory, and importing my 
> relations. I have not been able to import the filtering results. Moreover, 
> as the ./scripts/cadofactor/README indicates:

>        “When importing relations, some statistics are gathered by parsing 
> the

>        relation files. This is a costly operation, especially when relations

>        from a large number of files are gathered. An expedient for this 
> problem

>        is to create, for each file cxx.sieving.xxx-yyy.zzzzzzz.gz, a file

>        cxx.sieving.xxx-yyy.zzzzzzz.gz.stats.txt, which contain the relevant

>        statistics. The format of this file should match the format of the

>        statistics files produced by the cado-nfs siever with the 
> -stats-stderr

>        option. Examples of such statistics files are found as the files

>        *.stderr0…”

> I have not figured out how to import these stats files as well, and thus 
> avoid regenerating them. I have attempted including them in the list @files 
> to import, using a separate list, as well as copying to the new work 
> directory under $name.upload. Both of these options fail as well for 
> re-using the data from the filtering stage. I am hoping that someone can 

> 1. Clarify the instructions in the README on importing relations and the 
> costly operation of re-gathering stats.


the README instructions were ok, but the file names do not contain any more
"sieving". I've updated the instructions in the git version, and tested them:
it works. Please confirm.

> 2. Suggest a method for restarting after merging and before linear algebra 
> begins, ignoring any previous runs.


the allow_wipeout=True command is indeed broken in the git version. A
workaround is to remove the cxx.bwc directory and run again the bwc.pl
command (1st line of cxx.bwc.stdout.1). However you'll then have to run
by hand the final commands (see cxx.cmd).

Best regards,
Paul Zimmermann