[QUOTE=geoff;87053]I have made an improvement to the way these sequences are handled in sr5sieve version 1.2.1, so now if the odd terms are QR and the even ones are not then only the odd terms are included in the bsgs iteration.[/QUOTE]

For some reason, I am getting a 2% slowdown with 1.2.1. Anybody else getting a similar result?

[QUOTE=axn1;87215]For some reason, I am getting a 2% slowdown with 1.2.1. Anybody else getting a similar result?[/QUOTE]

I'm getting a 1% slowdown here; from 51kp/s to 50.5kp/s
_mutter...5 minute waiting time...mutter..._

[QUOTE=axn1;87215]For some reason, I am getting a 2% slowdown with 1.2.1. Anybody else getting a similar result?[/QUOTE]

This only affected GCC 3.4, which I use to compile the mingw32 binaries. For some reason this version of GCC doesn't like variable sized local arrays. Fixed in sr5sieve version 1.2.2

I have also compiled a version in the sr5sieve/tests directory that works in base 2 and reads Sob.dat/riesel.dat files. This can be used to benchmark against Proth sieve etc.

[QUOTE=geoff;87355]This only affected GCC 3.4, which I use to compile the mingw32 binaries. For some reason this version of GCC doesn't like variable sized local arrays. Fixed in sr5sieve version 1.2.2[/QUOTE]

1% speedup!!

Went from v.1.2.0 to v.1.2.2
from 302 to 301 candidates
and the speed went from 51kp/s to 53.5kp/s!
+- 5% speedup :grin:

srsieve 0.5.1
 

This version of srsieve includes a similar quadratic residue filter mechanism to sr5sieve 1.2.0 (extended to allow for sequences k*b^n+c where |c| > 1) and should result in a similar speedup (30% or so) for sieves with a large number of sequences.

sr5sieve 1.2.3
 

All possible Legendre symbols for each sequence are now calculated during initialisation, so checking whether a sequence consists of quadratic residues only involved a table lookup.

The speedup depends a lot on machine type, my Coppermine P3 is about 12% faster, but my P4/Celeron is only about 3% faster than version 1.2.2. Memory footprint is larger, the current 300-k data file uses about 31MB RSS (Linux32).

I don't know what the effects will be for the PPC, the bitmap code may need to be changed as test_bit() is now heavily used.

[B]sr5sieve 1.2.3 is always crashing here.

AppName: sr5sieve.exe AppVer: 0.0.0.0 ModName: sr5sieve.exe
ModVer: 0.0.0.0 Offset: 0000600d

back to 1.2.2
[/B]

On athlon 64X2 3800 so far so good...
from 53.5kp/s to 58kp/s

oh... and nice to have the speed updated everytime there is a new factor :)
keeps the clutter off-screen :)

Is it possible (or even sensible) to let sr5sieve make a save-file for the legendre-symbols?

I tend to stop/start the sieve quite a lot...

[QUOTE=em99010pepe;87881][B]sr5sieve 1.2.3 is always crashing here.

AppName: sr5sieve.exe AppVer: 0.0.0.0 ModName: sr5sieve.exe
ModVer: 0.0.0.0 Offset: 0000600d

back to 1.2.2
[/B][/QUOTE]

I will compile a binary with debugging information for you to run so I can get some more details, but it will probably be tomorrow before I can post it.

Has anyone else had any problems with this version?