A multiple k/c sieve for Sierpinski/Riesel problems

[geoff]

Originally Posted by Xentar

Just found version 1.3.3 on the server.
Could you please tell us, what´s new in this version?

The only change in version 1.3.3 is that the number of found factors is written to the log file when the range is finished.

[tnerual]

cool, no need to open the file with excel now ...

notepad is more memory friendly ... and less blue screener

[geoff]

A small change to the hashtable code makes it 3% faster for P2/P3, 1% faster for P4.

[Xentar]

1.3.4 with newest .dat file:
83k p/s
It´s getting faster and faster

[geoff]

These versions fix a serious bug in the i386 assembler 64-bit modular arithmetic functions that could cause some factors larger than 2^46 to be missed. All previous versions were affected by this bug. (But not the PPC or x86-64 variants). Luckily the distributed sieve has not yet reached 2^42, so no work needs to be repeated.

I have completely re-written the mulmod64() and powmod64() functions and tested them against GMP for random primes up to 2^62 (2^52 for non-assembler versions). For those who compile from source and have the GMP library installed, there is a 'srtest' program that will do these tests.

[Xentar]

And the new version is 5 - 7% faster, nice

[geoff]

Originally Posted by Xentar

And the new version is 5 - 7% faster, nice

What type of machine is that result for?

On my P3 there is not much difference in speed between 1.3.4 and 1.4.0, 1% faster at most, but I found my P4/Celeron is about 7% faster.

[Xentar]

Using a Core 2 Duo.
At the moment, it runs with 87k p/s (per core).

[axn]

Originally Posted by geoff

What type of machine is that result for?

On my P3 there is not much difference in speed between 1.3.4 and 1.4.0, 1% faster at most, but I found my P4/Celeron is about 7% faster.

Around 7% improvement in my P-M (went from 44.5 to 47.5)

[michaf]

On Athlon 64-x2 around 0.5-1% faster

[geoff]

This version is about 10% faster on my P3 and 15% faster on my P4 Celeron compared to 1.4.0, mainly from improving the FPU instruction scheduling in the new mulmod/powmod code.