Alternative LLR tester implementation

[paulunderwood]

I had a quick look at the source and I think you need 50 or so rounds of "safe" multiplication before and after each of the main loops. This will solve any FFT entropy problems.

[paulunderwood]

Is that BSD licence alright? You are using George Woltman's library and GMP. Should they not be attributed as well?

[kuratkull]

In the file llr.zig lines 103:106 I do gwsquare2_carefully for the first and last 30 rounds. Without that even the selftest fails on my system.


Well the thing with the license is that my code is BSD licensed. Linking to LGPL GMP is okay. But the weird license from gwnum says that if you find a Mersenne prime you have to share the prize with GIMPS.

I probably have to at least add both licenses to my project. LLR64 doesn't seem to have a copyright license, so it has reserved all rights, which I am pretty sure is not the correct way to do this...

[Happy5214]

LLR having no license for its own code (i.e. not the gwnum code) is a problem. Someone should clarify that with Jean.

I'm not a lawyer, so take this advice with a grain of salt. As long as your LLR code wouldn't be considered a derivative work of Jean's LLR, which may be an issue, you could probably get away with the BSD license for your code. It's compatible with the LGPL, and linking with gwnum should be OK as long as it's clear that the library has its own license. If you distribute gwnum with your code, include its license in that folder. Keep in mind that the combined program would inherit gwnum's restrictions on Mersenne primes. Unless you're bundling GMP, I don't think you'll need to include the LGPL text in your code. As Paul suggested, you should consider crediting these libraries in a README or similar file.

[kuratkull]

I am not bundling any third party code and I have not copied code from LLR64, so RPT is completely mine. I think only the compiled binary is suhject to the licenses, because the executable contains statically linked GMP and gwnum. But yes, other than that I agree with your conclusions @Happy5214

[Happy5214]

Originally Posted by kuratkull

I am not bundling any third party code and I have not copied code from LLR64, so RPT is completely mine. I think only the compiled binary is suhject to the licenses, because the executable contains statically linked GMP and gwnum. But yes, other than that I agree with your conclusions @Happy5214

Keep in mind that translations (i.e. ports) also count as derivative works, at least according to the armchair lawyers. I don't know how much you consulted Jean's C code when writing your Zig code, and TBH I don't know your or his code well enough to compare them. As you probably know, algorithms can't be copyrighted. I still want clarity from him, if only for the general question of his licensing, so I'll go ahead and PM him about it.

Since the executable is statically linked, it probably would have to be distributed under the most restrictive terms of the combination (e.g. gwnum's Mersenne prime rule, GMP's rules on providing the license and an offer of the source code). Since you're really distributing the source code, however, I think you're right that those static link rules don't apply as such to that source distribution.

[kuratkull]

I didn't translate nor copy his code in any even barely attributable capacity, so this is not an issue.

[LaurV]

Originally Posted by kuratkull

But the weird license from gwnum says that if you find a Mersenne prime you have to share the prize with GIMPS.

In my layman opinion, there is nothing wrong with it, as long as you used a lot of information from PrimeNet, like for example their base of exponents for which the corespondent mersenne numbers were already testes or factored. Otherwise, starting from scratch by yourself, it would take you centuries. Of course, when it comes to take the EFF's money, you can claim that you never consulted this data base and you just picked a random exponent, and got lucky. But everybody around you will know that's not true, and you may look like an asshole in the eyes of the community. Not that I would care about that, if that would happen to me ever , but pragmatically speaking, your chance is much-MUCH higher to find first a smaller than 100M digits prime, for which EFF gives no money, but GIMPS does, so accepting that prize money, you are automatically bonded to GIMPS condition. Not counting the efforts you have to go through, to GET those money from EFF, i.e. a lot of paperwork and nerves, about which you have no experience, but GIMPS does, as they did it for other primes in the past.

If I would be to find a 100M-digits prime in the future, using my own program, completely written by myself and without consulting any library (I do not have such a program, this is only for the sake of the argument), or, by the way, cudaLucas or gpuOwl (to my knowledge, they are not bounded - yet - by GIMPS condition), then for sure I would let GIMPS handle it, speed up verification, do the paper work, and I would be totally satisfied with the $50k and the short-term fame.

Or maybe, I just got an idea! In my greed, , and especially if I would have a "handpicked" exponent which would be difficult to guess, I would try to "negotiate" how the prize is split, before telling the exponent. Haha. The blackmailer in me would go like: "You can split the prize my way, or wait another 7 years till somebody finds a prime". I would not think especially to get more money for myself, but to give some small chunks to the people who contributed to the development of the tools we use today (like msft, Oliver, Bdot, flash, etc), in detriment of GIMPS charity, or whatever crazy ideas I would have at that time. All dogs are good till there is a bone to share, hehe...

But to go for the prize by myself, that would be too much of an effort and arguments, and I am kinda too lazy for it

[retina]

Originally Posted by kuratkull

... prize ...

Forget about the prize. The chances of finding a prime to claim any money from anyone (EFF or GIMPS) is too low to make any practical difference. And if you are doing this to claim some prize then you have better odds buying lottery tickets.

[LaurV]

Originally Posted by retina

Forget about the prize. The chances of finding a prime to claim any money from anyone (EFF or GIMPS) is too low to make any practical difference. And if you are doing this to claim some prize then you have better odds buying lottery tickets.

That for sure! Hunting for primes and generally, any math related computations, have to be treated like a hobby, and source of fun and entertainment. Participating in forums like this, meeting clever people (and less clever too :P, there are a lot of cranks wandering around), finding a prime or a large factor from time to time, that is priceless. But on the other hand, you will have to "invest" a lot of money and/or time and/or resources to stay "in touch". We personally spent over $100k in hardware, electricity, nerves, etc, along the ~20 years, and no prize money could cover for it, but we learned a lot of things meantime, and had a lot of fun... Unless you are tremendously lucky, your path will be the same. You will not get rich from hunting for primes. If you think you are tremendously lucky, go hunting for bitcoins, you have about the same chances to lock a block by yourself, running a couple of good GPUs, as you have to find a mersenne prime with the same hardware, and a block today, even after the HUGE CRASH from yesterday, still worth $70k, which is better than the $50k you get from EFF after GIMPS split.

[kuratkull]

Originally Posted by retina

Forget about the prize. The chances of finding a prime to claim any money from anyone (EFF or GIMPS) is too low to make any practical difference. And if you are doing this to claim some prize then you have better odds buying lottery tickets.

I would have written the same if the license said that you have to perform an interpretive dance in public when you find a prime.They have a strange license because they do not ask me to notify the users of binary files at all of the license conditions. So if someone downloads the software and doesn't know about that license point then they and GIMPS will have a totally different understanding of the situation. From a license perspective I have to considered the issue as if every user finds a new Mersenne prime.