How do you efficiently sieve for prime 3/4-tuples?

[henryzz]

This is a binary that has the other problem as well? There is something odd about your compiles. I would try to compile a Linux binary for you myself but various things would need to be a later version on your system.

BTW you should be able to use the binary I posted above within wine.

[Puzzle-Peter]

Originally Posted by henryzz

This is a binary that has the other problem as well? There is something odd about your compiles. I would try to compile a Linux binary for you myself but various things would need to be a later version on your system.

BTW you should be able to use the binary I posted above within wine.

No, the other problem was gone after removing smallPrimes and bound_small_primes from the command line.

I know the system is outdated, but I am not the one to decide about updating it...

[Puzzle-Peter]

Updating do gcc 4.4.7 did not change anything.

When I try to replace the mulmod function with the assembler version, I get a lot of "line xxx: stray \240" messages. What's that?

[henryzz]

Originally Posted by Puzzle-Peter

Updating do gcc 4.4.7 did not change anything.

When I try to replace the mulmod function with the assembler version, I get a lot of "line xxx: stray \240" messages. What's that?

I go away on Sunday for a week. Hopefully after that I will sit down and try and compile it on linux myself and get to the bottom of this.

For the asm look at the following link. Search for 240 and read from there. I would type it in rather than copy it.
http://forums.whirlpool.net.au/archive/435288


Your linux is 64-bit right? 32-bits would be a possible reason for problems.

[Puzzle-Peter]

Originally Posted by henryzz

For the asm look at the following link. Search for 240 and read from there. I would type it in rather than copy it.
http://forums.whirlpool.net.au/archive/435288

Your linux is 64-bit right? 32-bits would be a possible reason for problems.

Typing instead of copying did the trick and there is some speedup.

Yes, it's 64-bit.

[henryzz]

I think I have fixed the bug in the gmp version. Replace mpz_tdiv_ui with mpz_fdiv_ui in two places. This was meaning I was getting for example 3 mod 7 rather than 4 mod 7. In some places this would be fine but not there apparently.
I would suggest testing this with any new polynomials you try for a while. I think the old code would work with positive coefficients of P(s) and Q(s) but not negative.

[henryzz]

I have also found the issue with the extra inputs I added.
Please change

scanf("%lld",&bound_small_primes);
and
scanf("%lld",&smallPrimes);

with

scanf("%d",&bound_small_primes);
and
scanf("%d",&smallPrimes);

I get off with this on windows using both mingw and cygwin. It is quite possibly what is causing you problems though.

Apologies for how long it took for me to find this. c isn't a language I am overly fluent in although I am getting better.

I have attached fixed source for both gmp and not. Please test before relying on it to work.

[henryzz]

Found three triples during testing
2238819531*((15*2^1246-1)^3+2*(15*2^1246-1)^2)-3*(15*2^1246-1)-6 -1/+1/+5 Only +5 proven.

509488329*((177*2^1032-1)^3+2*(177*2^1032-1)^2)-3*(177*2^1032-1)-6 -1/+1/+5 All proven

1914682779*((177*2^1032-1)^3+2*(177*2^1032-1)^2)-3*(177*2^1032-1)-6 -1/+1/+5 All proven

I have only tested ~21% of the second file so far so hopefully more will be found.

[henryzz]

Found the gmp bug was in stage 2 as well. Corrected source attached.

[Puzzle-Peter]

Originally Posted by henryzz

I think I have fixed the bug in the gmp version. Replace mpz_tdiv_ui with mpz_fdiv_ui in two places. This was meaning I was getting for example 3 mod 7 rather than 4 mod 7. In some places this would be fine but not there apparently.

That did the trick, thanks!

[henryzz]

Here is a binary for polysieve2 with all the latest fixes.
I have also included the final prime for each 10_tuple upto 1e16 of the form 0 2 6 12 14 20 24 26 30 32.