mfaktc: a CUDA program for Mersenne prefactoring

[Uncwilly]

Originally Posted by TheJudger

Uncwilly: M3312xxxxxx from 2^1 to 2^71 needs ~5m 35s on my system :)

If you or anyone else gets bored, here is a lovely link to some work that might keep you busy for a little while.
http://v5www.mersenne.org/report_fac...=1&B1=Get+Data

[moebius]

Originally Posted by BigBrother

I don't know if MSVC is the first choice for CUDA on Windows, it's the only one i've got here. I'm using a student license for MSVC.

MSVC 8.0 or 9.0 are default for CUDA

I have now installed MSVC 9.0, however I can not compile the sourcecode for version 0.03 because the librays <sys/time.h> <unistd.h> don't exist
in VisualC++.

Please tell me the trick.

[BigBrother]

Originally Posted by moebius

MSVC 8.0 or 9.0 are default for CUDA

I have now installed MSVC 9.0, however I can not compile the sourcecode for version 0.03 because the librays <sys/time.h> <unistd.h> don't exist
in VisualC++.

Please tell me the trick.

I removed those two lines and included <timeval.h>, which is available here.

You'll also have to replace every %Lu by %llu in printf statements, and maybe some other minor things as well. Just ask if you run into more troubles.

[ET_]

Originally Posted by Uncwilly

70->71 on a single core ~8h on my laptop. 0.72 GHz Days credit.

Did you test it with Factor5? AFAIK, Prime95 can't work with such exponents...

Luigi

[TheJudger]

Originally Posted by Uncwilly

70->71 on a single core ~8h on my laptop. 0.72 GHz Days credit.

That is for M332xxxxxx from your link the the primenet server?
I was talking about M3312xxxxxx.
Otherwise the speedup seems to be way to much!

Oliver

[Uncwilly]

Originally Posted by TheJudger

That is for M332xxxxxx from your link the the primenet server?
I was talking about M3312xxxxxx.
Otherwise the speedup seems to be way to much!

Since you called me out, I assumed that you were talking about a 100M digit. I didn't count the xxx's. I was quoting based upon 332,2xx,xxx.

[moebius]

Originally Posted by BigBrother

Just ask if you run into more troubles.

This has so basically works, but I had to compile with the /TP option because of the error message
referring to unresolved external "_sieve_candidates" in ......


Now, still enters the following "syntax error" in mfaktc.cu.


E:\Programme\Microsoft Visual Studio 9.0\VC\bin>nvcc --compiler-options "/TP" E:\Programme\mfaktc\mfaktc-0.03\mfaktc.cu


mfaktc.cu
cl : Befehlszeile warning D9025 : "/TC" wird durch "/TP" überschrieben
tmpxft_00002ecc_00000000-3_mfaktc.cudafe1.gpu
C:\DOKUME~1\avatara\LOKALE~1\Temp/tmpxft_00002ecc_00000000-3_mfaktc.cudafe1.gpu(
5): error: expected an identifier
C:\DOKUME~1\avatara\LOKALE~1\Temp/tmpxft_00002ecc_00000000-3_mfaktc.cudafe1.gpu(
5): error: expected a ";"
e:\programme\cudatoolkit\include\cuda_texture_types.h(60): warning: parsing restarts here after previous syntax error
..
..

[TheJudger]

Hi,

Originally Posted by Uncwilly

Since you called me out, I assumed that you were talking about a 100M digit. I didn't count the xxx's. I was quoting based upon 332,2xx,xxx.

sorry, my fault. I've mixed up some stuff. I remembered that you've asked "can it handle <some size> exponents" before. Actually it was 100+M digits while I was did some 1G digits exponents in the past days.

-----
Luigi: Let me first have atleast one code which is "known to work". Expanding it to a bigger range is not that difficult. I'll do it later (no timeline yet).
When you can't resist and want to try yourself to implement bigger factors: wait at least for the next version. I expect some changes in the function mod_144_72(). I hope for some speedups there.


Oliver

[henryzz]

Does the latest version support trial factoring of exponents less than 1M? When will it?
When my pc is fixed i will try the latest version.

[TheJudger]

Hi David,

No, the current version doesn't support exponents <1M. But this limit is artificial. Actually I haven't even tried such low exponents.


Is there any usage for TF on such low exponents?


Oliver

[ET_]

Originally Posted by TheJudger

Luigi: Let me first have atleast one code which is "known to work". Expanding it to a bigger range is not that difficult. I'll do it later (no timeline yet).
When you can't resist and want to try yourself to implement bigger factors: wait at least for the next version. I expect some changes in the function mod_144_72(). I hope for some speedups there.


Oliver

Thanks Oliver, you are doing great on this field.

I'll be watching the thread and waiting for some news, thank you for your effort.

Luigi