Trial division with CUDA (mmff) -- used, but runs like new! - Page 27

Batalov · 2012-12-04, 22:52

CUDA is not supported with GCC >= 4.6, but it doesn't mean that it doesn't work. It means that if you write to them with a bug report, they will not take it. I use gcc version 4.7.1 20120723 [gcc-4_7-branch revision 189773] (SUSE Linux) and one odd thing that I had to add was -lstd++ (where something is expected to be defined); linker spat out this recommendation without me even asking. Weird, but the binaries work.

(Of course, something might not work, if they rely on some optimizations in some old way. More likely, they just don't want to be bothered.)

Prime95 · 2012-12-13, 00:25

Minor update -- v 0.27:

What's new:

1) Bug in testing 187-bit factors of MM127 fixed.
2) -lm makefile bug fixed (I hope).
3) With Batalov's help, the next set of 32 n values in k*2^n+1 Fermat factor testing is available.

As always previous savefiles wiln not work with 0.27 unless the -nocheck argument is used.

My dual boot box is running Windows right now, so I don't have a Linux executable right now.

Batalov · 2012-12-13, 00:54

Under linux, I get some errors in or around

Code:

gpusieve.cu(1991): error: expected a declaration
 
that's around innocuously looking 
        else {
                if (gpusieve_initialized) return;
        }

I'll diff to my branch when I get home. (it's ugly to figure out in vim over ssh.)

Prime95 · 2012-12-13, 02:03

Quote:

Originally Posted by Batalov

Under linux, I get some errors

Try again. I haven't compiled this code in well over a month. I minor tweak had a typo.

Batalov · 2012-12-13, 02:31

Looks good.
Test passed:

Code:

got assignment: k*2^41+1, k range 2864929972000000 to 2864929973000000 (93-bit factors)
Starting trial factoring of k*2^41+1 in k range: 2864929972M to 2864929973M (93-bit factors)
 k_min = 2864929972000000
 k_max = 2864929973000000
Using GPU kernel "mfaktc_barrett96_F32_63gs"
    class | candidates |    time |    ETA | raw  rate | SievePrimes | CPU wait
2471/4620 |      0.01M |  0.275s |  2m03s |   0.03M/s |      210485
F39 has a factor: 6300047635658008393597059073

Dubslow · 2012-12-13, 02:35

Quote:

Originally Posted by Prime95

I minor tweak had a typo.

*snirk*

RichD · 2012-12-13, 02:59

Quote:

Originally Posted by Prime95

Minor update -- v 0.27:

As always previous savefiles wiln not work with 0.27 unless the -nocheck argument is used.

And resumes with the same level savefile -- at least with Linux v.0.26 and below.
(Haven't check 0.27 yet.)

RichD · 2012-12-13, 03:04

Quote:

Originally Posted by Batalov

CUDA is not supported with GCC >= 4.6, but it doesn't mean that it doesn't work. It means that if you write to them with a bug report, they will not take it. I use gcc version 4.7.1 20120723 [gcc-4_7-branch revision 189773] (SUSE Linux) and ...

Exactly my setup. I remember using a compiler override switch during install. Seems to work but (as you say) don't bother with a trouble report.

Batalov · 2012-12-13, 06:20

Thank you, we will!

And when I will PM you a link to the mmff-gfn source, would you repeat the previous exercise? TIA! I will then post to the mmff-gfn folder. 5 binaries, like last time.

flashjh · 2012-12-13, 06:21

Quote:

Originally Posted by Batalov

Thank you, we will!

And when I will PM you a link to the mmff-gfn source, would you repeat the previous exercise? TIA! I will then post to the mmff-gfn folder. 5 binaries, like last time.

I'm ready ;)

Edit:

Just did some testing, .27 fails with this error:

Code:

got assignment: k*2^101+1, k range 3334000000000 to 3335000000000 (143-bit factors)
Starting trial factoring of k*2^101+1 in k range: 3334G to 3335G (143-bit factors)
 k_min = 3334000000000
 k_max = 3335000000000
Using GPU kernel "mfaktc_barrett152_F96_127gs"
ERROR: Exponentiation failure

With this still left in the worktodo file:

Code:

FermatFactor=101,3334e9,3335e9
FermatFactor=111,141,142
FermatFactor=120,3e9,4e9
FermatFactor=135,880e8,881e8
FermatFactor=148,173,174
FermatFactor=149,175,176

.26 works

Edit2: It's only the mfaktc_barrett152_F96_127gs kernel that fails, the rest work fine so far. These lines won't complete:

Code:

FermatFactor=101,3334e9,3335e9
FermatFactor=111,141,142
FermatFactor=120,3e9,4e9

ET_ · 2012-12-13, 08:04

Quote:

Originally Posted by Prime95

Minor update -- v 0.27:

3) With Batalov's help, the next set of 32 n values in k*2^n+1 Fermat factor testing is available.

...and now the max range is...?

2012-12-13, 00:54	#289
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3⁶×13 Posts	Under linux, I get some errors in or around Code: gpusieve.cu(1991): error: expected a declaration that's around innocuously looking else { if (gpusieve_initialized) return; } I'll diff to my branch when I get home. (it's ugly to figure out in vim over ssh.)

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Mersenne trial division implementation	mathPuzzles	Math	8	2017-04-21 07:21
trial division over a factor base	Peter Hackman	Factoring	7	2009-10-26 18:27
P95 trial division strategy	SPWorley	Math	8	2009-08-24 23:26
Trial division software for Mersenne	SPWorley	Factoring	7	2009-08-16 00:23
Need GMP trial-division timings	ewmayer	Factoring	7	2008-12-11 22:12

2012-12-04, 22:52	#287
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3⁶·13 Posts	CUDA is not supported with GCC >= 4.6, but it doesn't mean that it doesn't work. It means that if you write to them with a bug report, they will not take it. I use gcc version 4.7.1 20120723 [gcc-4_7-branch revision 189773] (SUSE Linux) and one odd thing that I had to add was -lstd++ (where something is expected to be defined); linker spat out this recommendation without me even asking. Weird, but the binaries work. (Of course, something might not work, if they rely on some optimizations in some old way. More likely, they just don't want to be bothered.)

2012-12-13, 06:20	#295
Batalov "Serge" Mar 2008 Phi(4,2^7658614+1)/2 3⁶·13 Posts	Thank you, we will! And when I will PM you a link to the mmff-gfn source, would you repeat the previous exercise? TIA! I will then post to the mmff-gfn folder. 5 binaries, like last time.