Yafu

[Andi_HB]

Looks like the new Version don`t find my msieve.
The msieve.exe is in the gnfs + yafu directory but yafu 1.22.2 crashs.
I have tested with 1.22.1 - no problem there.

[bsquared]

Originally Posted by Andi_HB

Looks like the new Version don`t find my msieve.
The msieve.exe is in the gnfs + yafu directory but yafu 1.22.2 crashs.
I have tested with 1.22.1 - no problem there.

Thanks for the report - I see the crashes too.

Slight clarification: yafu doesn't need the msieve binary - I incorporate portions of msieve directly into the yafu binary. Which is increasingly looking like a bad decision because of issues like this - the API of msieve was not necessarily designed to be used in this way. For 1.22.2, I rebuilt msieve from the latest SVN, and now we see this crash. I have no idea why it would do that. x64 versions of 1.22.2 work just fine...

For now I'll rebuild using a downrev'ed msieve and repost the binaries. Going forward I'll have to think about restructuring yafu's NFS to rely on an external msieve.exe.

Thanks for your patience.

- ben.

[bsquared]

Ok, I've looked at the msieve changelog and I think I've fixed the problem (in yafu).

Updated binaries are available for download (only Win32 were affected by this change).

I'll try to do a better job keeping up with msieve API changes for future builds (or at least a better job testing!).

[Andi_HB]

Thank you ben for updating the binaries.
I think you do a very good job!

Best Regards Andi_HB

[jasonp]

Was it the change in the size of an mp_t in the last rev? I removed mp.h from the external interface in the latest SVN because it's just too dangerous to leave it there.

[bsquared]

Originally Posted by jasonp

Was it the change in the size of an mp_t in the last rev? I removed mp.h from the external interface in the latest SVN because it's just too dangerous to leave it there.

Did the size change? MAX_MP_WORDS is still 32. Anyway, I saw that mp_sprintf_buf changed from being statically to dynamically allocated in the msieve_obj structure. When I changed yafu to match - that fixed the problem.

I just neglected to check for msieve interface changes prior to rebuilding everything.

[lorgix]

What's this?

[bsquared]

Originally Posted by lorgix

What's this?

Errrrm, a bug? :)

It works if you specify the second argument to nextprime (1 for next bigger, 0 for previous smaller).

factor(nextprime(rand(17),1)*nextprime(rand(18),1))

[lorgix]

Darn, I had my money on mammal...

[em99010pepe]

Ben,

Did you manage to fix the "lost cores" issue during yafu running?

Carlos

[bsquared]

Originally Posted by em99010pepe

Ben,

Did you manage to fix the "lost cores" issue during yafu running?

Carlos

Sorry, no I haven't. The only experiments I can think of to try are pretty invasive. For example a complete reorganization of the threading architecture or windows API calls to enforce thread affinity to a core. As such I've been putting off trying them out...

- ben.