Octoproths

[smh]

I think i found a new record size octoproth:

n=239, kmin=1, kmax=1000000000000000
237534540236595 239
689532548527515 239
988916004684825 239
n=240, kmin=1, kmax=1000000000000000
n=244, kmin=1, kmax=10000000000000
n=245, kmin=1, kmax=10000000000000
n=246, kmin=1, kmax=10000000000000
n=247, kmin=1, kmax=100000000000000

[fetofs]

Originally Posted by R. Gerbicz

where is fetofs?

Fetofs was at home, waiting for his internet connection to work again, expecting octo2.0 (and now I get octo3!)

P.S: The executable works fine! Thanks!

[Kosmaj]

Found the new largest octo with 116 digits in k*2^n+-1 bitwins and 102 digits in 2^n+-k, 2^(n+1)+-k members. All 8 numbers certified primes.
157715286549765 336 [160T]

The following n's are checked to 100T, no octos: 332-335, 337-348.
The following checked to 50T, no octos: 349-350.
Also, n=232 checked to 400T, no octos.
n=158-160 in progress, will post results soon.

[Kosmaj]

BTW, I couldn't verify that ver. 3.0 is 40% faster than ver. 2.0, on the contrary for n=336, kmin=100T, and kmax=111T, ver. 2.0 is faster on my AhtlonMP-2000+ (1.67GHz) compiled by gcc without any optimization options and linked against libgmp.dll.a
ver. 2.0: 302 sec
ver. 3.0: 306 sec (145880 prp tests, x=8)

While I found the n=336 octo using modified ver. 2.0 with prp block similar to ver. 3.0 but without counting the number of prp tests. That one takes 299 sec.

Finally, ver. 3.0 modified to do prp tests in increasing order of 8 octo members took 301 sec, or about 2% speed up.

[smh]

My v3 with cpu optimizations was about 40% faster compared to v2 with no optimizations.

[R. Gerbicz]

Originally Posted by Kosmaj

ver. 2.0 is faster on my AhtlonMP-2000+ (1.67GHz) compiled by gcc without any optimization options and linked against libgmp.dll.a

Try with gcc -g -O2 -fomit-frame-pointer -mcpu=athlon-mp -march=athlon-mp octo_3_0.c -o octo_3_0 .libs/libgmp.a
Why you didn't use flags? These give you additional speedups.

Originally Posted by Kosmaj

While I found the n=336 octo using modified ver. 2.0 with prp block similar to ver. 3.0 but without counting the number of prp tests. That one takes 299 sec.

Really? How many times have you tried this?
An mpz_add_ui instruction takes constant number of clock cycles, say 5 then 145880 such additions takes about 1 million cycles Your computer' speed say about 5*10^9 cycles/sec so this computation cost is less than 0.001 sec. Compare this to the runtime=300 sec. Believe it or not.

Note that I've tested octo 3.0 only on Windows and on that I've gotten about 40% speedup. This is P4 Celeron 1.7 GHz and L2 cache=128 Kbyte ,this is also very important for octo, because the current version is using about 550 Kbyte of memory, so this is not fit on my fast L2 cache and also Celeron isn't the best choice.

Originally Posted by Kosmaj

n=158-160 in progress, will post results soon.

160 has done by rogue:see

Originally Posted by rogue

801902642350095 160
939298231941615 160
642731976956685 160
10628676454605 160
237329407435575 160
637494677259825 160
655785200172285 160
170160450178095 160
708011709049455 160
346058710723935 160
687793712441745 160
596547394620555 160
70736688101295 160
585905207678175 160
636116523125355 160

I cannot get any version to run, on XP pro SP2 without cygwin.
This is a no go for me then, darn..

[R. Gerbicz]

Originally Posted by TTn

I cannot get any version to run, on XP pro SP2 without cygwin.
This is a no go for me then, darn..

See my previous posts! I've created exe's for Windows, you don't need Cygwin!
You can download from my homepage the newest version for windows:

http://www.robertgerbicz.tar.hu/octo_3_5.exe

Or see my attachment for the c source code.

I've made a very small change. Version 3.0 is correct but this new version uses less prp tests, because there was some missed composites (I haven't deleted some numbers in the sieve which has got small prime factors, it means that there is no missed octoproths! for version 3.0 ) in the sieve ( this happens only in 3.0 ). You get small improvement, because sieving is much faster, but it is suggested to download the new version.

[nibble4bits]

Version 2.0:
"Faulting application octo2.exe, version 0.0.0.0, faulting module octo2.exe, version 0.0.0.0, fault address 0x00009144."

"The application, C:\Documents and Settings\JoeDesktop\Desktop\octo2.exe, generated an application error The error occurred on 01/06/2006 @ 22:08:44.078 The exception generated was c000001d at address 00409144 (octo2)"



Current version:
Faulting application octo_3_5.exe, version 0.0.0.0, faulting module octo_3_5.exe, version 0.0.0.0, fault address 0x00008994.


Are sure this program works on Athlon XP-Mobile processors? Those errors sound like unsupported features of the instruction set. Exception c000001d is generally illegal instructions according to my documentation. I'll try it on my P3 laptop but I'm pretty sure it will work on it. This is most certainly your version of GMP being compiled for another CPU. :)

[robert44444uk]

n=181, kmin=1, kmax=200000000000000, version=3.0
Starting the sieve...
Using the first 9 primes to reduce the size of the sieve array
152614043635305 181
The sieving is complete.

n=182, kmin=1, kmax=200000000000000, version=3.0
Starting the sieve...
Using the first 9 primes to reduce the size of the sieve array
127732547041365 182
The sieving is complete.

n=183, kmin=1, kmax=180000000000000, version=3.0
Starting the sieve...
Using the first 9 primes to reduce the size of the sieve array
The sieving is complete.

n=184, kmin=1, kmax=500000000000000, version=3.0
Starting the sieve...
Using the first 9 primes to reduce the size of the sieve array
276221124938745 184
212311049145495 184
342826651623555 184
381356034701265 184
The sieving is complete.

Will take on 185-190 next

Regards

Robert Smith

[robert44444uk]

pops forgot to show:

n=185, kmin=1, kmax=500000000000000, version=3.0
Starting the sieve...
Using the first 9 primes to reduce the size of the sieve array
229673680116405 185
77793510170775 185
The sieving is complete.

Robert Smith