A multiple k/c sieve for Sierpinski/Riesel problems

[jasong]

Originally Posted by geoff

Are you running 64-bit Linux? If so it would be a great help if you could try out the linux64-k8 binary and see how speed compares to the linux32-i686 binary on the same machine.

That's a good question. I downloaded and installed a bunch of distributions before Xyzzy(sp?) finally explained that my lack of knowledge of the sudo command was the problem.

Is there an easy way to check for 64-bitness? I'm fairly certain it's a 64-bit cpu, I'm just not sure about the 64-bitness of the Linux distro, which is Ubuntu.

[Xyzzy]

cat /proc/cpuinfo

uname -a

[jasong]

Originally Posted by Xyzzy

cat /proc/cpuinfo

uname -a

yep, it's 64-bit. Now I just need to search the forum and figure out where to get the binary.

[geoff]

Originally Posted by jasong

yep, it's 64-bit. Now I just need to search the forum and figure out where to get the binary.

http://www.geocities.com/g_w_reynolds/sr5sieve/

The binary in sr5sieve-1.4.8-linux64-k8.tar.gz is compiled for 64 bit mode but doesn't use any assembly code for modular multiplication. It would be interesting to see how it compares to the binary in sr5sieve-1.4.8-linux32-i686.tar.gz which is compiled for 32 bit mode but has the benefit of the assembler modular multiplication.

I could try to write some 64 bit mode assembler if I have someone able to test it for me ;-)

[jasong]

Originally Posted by geoff

I could try to write some 64 bit mode assembler if I have someone able to test it for me ;-)

I´m just about to start testing it. :)

Edit: At least a 10% increase in speed. Unfortunately, I deleted a couple k´s between tries, so I can´t be more specific than that.

It WAS 85,500 with 254 k´s, Now it´s 106,200 with 252 k´s.

[axn]

Deleting two k's would have gained you at the most 1% speedup. So the rest is from the 64-bit goodness

[jasong]

Originally Posted by axn1

Deleting two k's would have gained you at the most 1% speedup. So the rest is from the 64-bit goodness

As my mom would say: ¨Kew-ell¨

[geoff]

Originally Posted by jasong

It WAS 85,500 with 254 k´s, Now it´s 106,200 with 252 k´s.

That is great, thanks for trying it out, you are the first that I know of :-)

I don't think the gain from hand-coded assembler will be huge, GCC seems to generate pretty tight code to start with on this platform, but I'll have a go and post some trial versions for you to try in the next week or so.

[jasong]

Originally Posted by geoff

That is great, thanks for trying it out, you are the first that I know of :-)

I don't think the gain from hand-coded assembler will be huge, GCC seems to generate pretty tight code to start with on this platform, but I'll have a go and post some trial versions for you to try in the next week or so.

Darn it, there's no thumbs-up sign in the forum. Anyway, I'm looking forward to it.

[geoff]

sr5sieve 1.4.11 adds inline assembler modular multiplication functions for x86-64 machines. They are based on working 32-bit prototypes, but the 64-bit versions are untested.

sr5sieve 1.4.12 extends the power residue checks to include octic residues. The gain from this for the base 5 sieve is marginal, but it should become more worthwhile as the number of k in the sieve drops. (The gain for the base sieve 2 with SoB.dat is about 5%).

PS: jasong, if possible can you test both the 1.4.11 and 1.4.12 linux64 binaries?

[geoff]

Sorry, version 1.4.12 had a bug which could have caused 1 factor in 240 to be missed. If anyone is using it then please upgrade to version 1.4.13.