Yafu

[bsquared]

Originally Posted by Andi_HB

O i have forgotten that i become Messages after the first sieving about the cat.exe.
But after the messages msieve filtering was running.

unxutils is a freeware package of unix utilities for windows. It contains things like head, tail, cat, wc, and so forth. If you are not a *nix user, "cat" is a program which concatenates two text files. It is really fast, and is useful for combining large files of relations into a single file for use in msieve. Windows doesn't have a native command to replace this one, so I wrote one quick in yafu - but its performance is nowhere near that of unxutils' cat.exe command. So I thought I'd add that message in there.

If enough people think its annoying, I'll remove it.

[xilman]

Originally Posted by bsquared

unxutils is a freeware package of unix utilities for windows. It contains things like head, tail, cat, wc, and so forth. If you are not a *nix user, "cat" is a program which concatenates two text files. It is really fast, and is useful for combining large files of relations into a single file for use in msieve. Windows doesn't have a native command to replace this one, so I wrote one quick in yafu - but its performance is nowhere near that of unxutils' cat.exe command. So I thought I'd add that message in there.

If enough people think its annoying, I'll remove it.

Funny, I thought that the good old MS-DOG command COPY did pretty much that:

C:\> COPY FILE1.TXT FILE2.TXT BOTH.TXT

is pretty much the same as

% cat file1 file2 > both

Or am I missing some requirement you have in mind and have failed to pick up?

Paul

[bsquared]

Originally Posted by xilman

Funny, I thought that the good old MS-DOG command COPY did pretty much that:

C:\> COPY FILE1.TXT FILE2.TXT BOTH.TXT

is pretty much the same as

% cat file1 file2 > both

Or am I missing some requirement you have in mind and have failed to pick up?

Paul

We often need to append one or more smallish files to a very large file. Doing this with copy is slow because it has to copy the large file as well as the smaller ones. cat with ">>" is much faster because only the smaller files need to be streamed.

[bchaffin]

Originally Posted by Karl M Johnson

...

Since I've already spoken, another feature suggestion: "parallelize polyselection".
From my point of view, it can be done like this.
When msieve's polyselection starts, it says "searching leading coefficients from 1 to 14167290".
Why not launch N msieve instances, each doing it's own work in that range ?

I agree that a multi-threaded polyselect phase would be a wonderful thing. In this thread there is discussion of adding this capability to msieve itself. If that's going to happen it would be a better place for it, so that YAFU does not have to change every time the polyselect algorithm is updated (as it is being right now).

[axn]

Originally Posted by bsquared

We often need to append one or more smallish files to a very large file. Doing this with copy is slow because it has to copy the large file as well as the smaller ones. cat with ">>" is much faster because only the smaller files need to be streamed.

Have you tried the "copy a+b" syntax? Or "type a >> b" syntax?

[Andi47]

Originally Posted by xilman

Funny, I thought that the good old MS-DOG command COPY did pretty much that:

C:\> COPY FILE1.TXT FILE2.TXT BOTH.TXT

is pretty much the same as

% cat file1 file2 > both

Or am I missing some requirement you have in mind and have failed to pick up?

Paul

it seems that copy doesn't work in a windows 7 "dos window" - at least it didn't work for me.

[mataje]

Hi Andi.
Do you try copy file1.txt+file2.txt both.txt?

[Andi47]

Originally Posted by mataje

Hi Andi.
Do you try copy file1.txt+file2.txt both.txt?

No, I did try something like copy *.out msieve.dat

where *.out are the outputfiles of the current NFS factorization. (manually adding a file named 0.out which contains the line N <my number>, the name "0.out should make sure that it goes to the very beginning of the msieve.dat file.)

Edit: This works in windows XP, but not in windows 7. (now I copied a cat.exe into the directory where I concatenate my outputfiles and use cat *.out >>msieve.dat)

[bsquared]

Originally Posted by axn

Have you tried the "copy a+b" syntax? Or "type a >> b" syntax?

I tried "type a >> b" first and got miserable performance. The fgets/fputs loop I wrote is faster than type. Yesterday was the first time I tried "copy a+b", simply because I had a mindset of _appending_ data, not copying it. But to append data I can't see a way around specifying the larger file in the input list to copy - which is wasted effort.

The whole issue isn't really a big deal. Appending data files is a negligible portion of the runtime of a nfs factorization. I'll remove the prompt for unxutils...

[bsquared]

Binaries and source code available here.

Changes include:
+ A new tune() function which measures the speed of several siqs and gnfs factorizations and computes regression curves for a particular computer. This info is used to estimate the amount of pretesting to do before siqs or nfs (ecm, pp1, pm1, etc.) as well as determine the optimal siqs/nfs crossover point. The yafu.ini file is automatically updated to track regression information for (optionally) multiple systems. Tune should take about a half hour to an hour to run, depending on your system.
+ new flags. ecm_qs_ratio and ecm_gnfs_ratio allow one to specify the target time ratio of pretesting to siqs or gnfs respectively. the noecm flag forces ecm to be skipped entirely.
+ tweaks to factoring via NFS. reporting to logfile, small prime divisibility and primalty checks, screen info, etc
+ bug fixes including a fix for this, and for an issue reported via email. Thanks lorgix and W. Ferguson!
+ gmp-ecm messages are now printed at a verbosity level equal to 2 less than the yafu verbosity level. for example, yafu started with options -v -v -v -v prints any gmp-ecm messages at verbosity level -v -v

Any issues, let me know.
- ben.

[lorgix]

Let's try it out!

sigma = 1940137128 completes the factorization of 198^198*4+1.