SR2sieve

[Joe O]

PRS52 v0.102b for processor type SSE2


First of all, you are running a very old version of JJsieve. The newer versions were much faster, but SR2sieve is faster than even the last released version.
Head on over to: http://www.geocities.com/g_w_reynolds/sr2sieve/ and pick up the latest version of SR2sieve.
Any references to the url http://www.geocities.com/g_w_reynolds/ should be replaced with the url http://sites.google.com/site/geoffreywalterreynolds/programs/.
There are Linux (32bit and 64bit), Windows (32bit and 64bit), and OSX versions available. READ the readme.
There is also now a way to drop SR2sieve in as an almost compatible replacement for JJsieve:


sr2sieve 1.6.10 has a new -j switch that will make it behave in a similar way to JJsieve or proth_sieve, for compatibility with Sobistrator. Checkpoints are written to SoBStatus.dat, ranges are read from nextrange.txt.

It should be possible to start a range with JJsieve and continue it with `sr2sieve -j -s', and vice versa.

I haven't actually tested it with Sobistrator though, so let me know if there are any problems.

Let us know what your speed improvement is after you upgrade.

Since .gz files cannot be attached, the .txt extension was added to two of the following files. Remove the .txt extension before trying to use them. These files are here for convenience only. The latest versions can always be found at http://www.geocities.com/g_w_reynolds/sr2sieve/.

Any references to the url http://www.geocities.com/g_w_reynolds/ should be replaced with the url http://sites.google.com/site/geoffreywalterreynolds/programs/.

[vjs]

Joe,

II tried the drop method into a folder currently running jjsieve and come up with the following error.

Line 1: unparsed range in file 'SoBStatus.dat'


also with a core quad should i be using the --sse2 switch or does it autodetect?

[Joe O]

Joe,

II tried the drop method into a folder currently running jjsieve and come up with the following error.

Line 1: unparsed range in file 'SoBStatus.dat'


also with a core quad should i be using the --sse2 switch or does it autodetect?
It should autodetect. The switch just forces it if the autodetect fails. Add the -vv switch to see more info.

[vjs]

O.k. I was pretty sure it was auto detecting...

no dice on the drag and drop solution to getting it started.

Easiest way was to simply get it working, add a new range from where you currently are in Sobistrator to the end point.

Then delete sobstatus.dat.

Restart using the sr2 and the client switch.. seems to be working like a champ.

[Sloth151]

In the sobstatus.dat file jjsieve starts as
pmax=
pmin=

sr2sieve has them in the other order. If you open sobstatus and put the pmax line below the pmin line then things will take off from there.
Found this with the latest sr2sieve using the -s -j command.

[Sloth151]

Someone was looking for speeds changing from jjsieve to sr2sieve

the jjsieve software was jjsievesse2 - not sure which version but the file on the systems are dated about a year ago. The sr2sieve is 1.6.11 using the -s -j arguments

p4 6300 1.86gHz, 1gb ram, xp pro
jjsieve 367kp/s per core
sr2sieve 603kp/s per core 18k dat file
sr2sieve 630kp/s per core 17k dat

p4 4400 2.0gHz, 1gb ram, xp pro
jjsieve 397kp/s per core
sr2sieve 654kp/s per core 18k dat
sr2sieve 676kp/s per core 17k dat

p4 6300 1.86gHz, 2gb ram, vista ultimate 64 bit
sr2sieve 64 bit 865kp/s per core 17k dat file
changing the dat file added about 3% to this machines speed. I dont have the number written down what it was before but I did check it when I updated the dat.

xp boxes are running as a service on boot up. Vista box had it running in a couple of dos windows

[geoff]

In the sobstatus.dat file jjsieve starts as
pmax=
pmin=

sr2sieve has them in the other order. If you open sobstatus and put the pmax line below the pmin line then things will take off from there.
Found this with the latest sr2sieve using the -s -j command.

Thanks, I will try to fix that for the next sr2sieve version. (I actually used Proth sieve 0.42 for Linux for testing, which could be why this problem didn't show up for me).

[Joe O]

Thanks, I will try to fix that for the next sr2sieve version. (I actually used Proth sieve 0.42 for Linux for testing, which could be why this problem didn't show up for me).
Actually Proth_sieve and JJsieve can read them in either order. It was the previous program that required the pmax line first, so of course everyone has continued to have it first.

[vjs]

geoff,

give it a few itterations it seems to correct itself.

[geoff]

Actually Proth_sieve and JJsieve can read them in either order. It was the previous program that required the pmax line first, so of course everyone has continued to have it first.

It is fixed in sr2sieve 1.6.12: pmax is written first but reading can be in either order.

[Death]

11/06/07 18:01:16 ERROR: Failed to open input file `sr2data.txt'.

what's this???

[Death]

oh well it works only with -j -s order
not with -s -j

strange....

[Death]

sr2sieve 1.6.12 -- A sieve for multiple sequences k*b^n+/-1 and b^n+/-k.
Compiled on Nov 2 2007 with GCC version 3.4.5 (mingw special).
SSE2 code path, L1 data cache 16Kb (detected), L2 cache 2048Kb (detected).
Read 2135444 terms for 17 sequences from dat format file `SoB.dat'.
Split 17 base 2 sequences into 169 base 2^360 subsequences.
Continuing with range pmin=1272000000000000,pmax=1272210000000000 in `SoBStatus.
dat'.
Resuming from checkpoint pmin=1272103091251393 in `SoBStatus.dat'.
Using 16 Kb for the baby-steps giant-steps hashtable, maximum density 0.58.
Best time for baby step method sse2/2: 295856.
Best time for baby step method sse2/4: 292112.
Best time for baby step method sse2/8: 185584.
Best time for baby step method sse2/16: 168864.
Best time for baby step method gen/2: 364160.
Best time for baby step method gen/4: 238640.
Best time for baby step method gen/8: 183776.
Best time for giant step method sse2/2: 147264.
Best time for giant step method sse2/4: 118864.
Best time for giant step method sse2/8: 107760.
Best time for giant step method sse2/16: 104640.
Best time for giant step method gen/2: 175520.
Best time for giant step method gen/4: 167888.
Best time for giant step method gen/8: 136576.
Best time for ladder method sse2/2: 22576.
Best time for ladder method sse2/4: 17808.
Best time for ladder method sse2/8: 10640.
Best time for ladder method sse2/16: 9040.
Best time for ladder method gen/2: 29104.
Best time for ladder method gen/4: 16528.
Best time for ladder method gen/8: 11232.
Best time for ladder method add/1: 26240.
Baby step method sse2/16, giant step method sse2/16, ladder method sse2/16.
BSGS range: 384*362 - 4790*29.
Using 1024Kb for the Sieve of Eratosthenes bitmap.
Expecting to find factors for about 10.14 terms in this range.
sr2sieve started: 991 <= n <= 49999997, 1272103091251393 <= p <= 127221000000000
0
p=1272103110912889, 327435 p/sec, 0 factors, 95.5% cpu, ETA 11 Nov 10:00

should i use -D to search for duplicates?

[Death]

as you can see with P4 3GHz it goes from 200k/sec to 327p/sec
nice!

and with Using 1024Kb - maybe I should give it 2 or 3 Mb? will it go even faster?

[KriZp]

When using the -j switch you shouldn't need to specify the -D switch, as per http://mersenneforum.org/showpost.php?p=117010&postcount=394 factexcl.txt will be written unless -D is used.

[geoff]

oh well it works only with -j -s order not with -s -j
It should work in either order.


and with Using 1024Kb - maybe I should give it 2 or 3 Mb? will it go even faster?
1024Kb is the size of the bitmap used for the Sieve of Eratosthenes which takes only a small part of the program's time, increasing it will probably only have a very small effect, but it will probably slow down if you increase it too far. The default is based on half L2 cache size, and the optimum might depend on what other processes on your system need L2 cache: on multi-core systems is might even be better to reduce the size instead. (Use the -L switch to override the detected L2 cache size).

[Death]

they should, but in my case it starts only with js not sj =((

well, this is a work box, so when I'm at home it works only for sieveng, so I suppose it uses L2 completely.

[Death]

just want to drop a few waord about sr2sieve

now I'm using version 1.7.10 and it's very fast.

sorry that I'm doing rieselsieve now not sob+psp

That's all because there's no credits for team Ukraine in RS, so I should add some points to it. I believe that is temporary and in a week I came back.

Sorry not to writing to forum so long. Got my XP died and that's make me depressed so I don't crunch anything and don't write at forums. I even TURN MY BOXEN OFF AT NIGHT!!!

But now I'm back. )))

[Death]

geocities is dead now.

Any references to the url http://www.geocities.com/g_w_reynolds/ should be replaced with the url http://sites.google.com/site/geoffreywalterreynolds/programs/.

[enderak]

The newest versions of sr2sieve can be found at http://sites.google.com/site/sr2sieve/

Any references to the url http://www.geocities.com/g_w_reynolds/ should be replaced with the url http://sites.google.com/site/geoffreywalterreynolds/programs/.