Request from Phil

[michaf]

I just read on the primenumbers forum the following request:
Can we provide him with such a list?

Cheers, Micha

I'm trying to resuscitate some old code for testing proth-like
numbers (in part so that people with non-x86 architectures can
participate in such searches), and would very much like a _complete_
list of all of the primes found in the Riesel/Sierpinski-5 searches
so that I can test my code very thoroughly.

(Mark R has already been testing old versions of the code and and
not-so-merrily finding several unpleasant, but in retrospect doomed
to fail, error cases on his G5 - many many thanks Mark!)

The Mersenne forum pages do seem to contain most of the information,
but it's all smooshed out across multiple threads and bogillions of
posts, and alas the most promising posts contained dead links, which
is a tad inconvenient.

Who's actually running the project? I assumed it was Robert as I
thought it was his braichild, but the forum posts imply that perhaps
he's not, but maybe I misunderstand.

Phil

[axn]

http://geocities.com/base5_sierpinski_riesel/

The "primes.txt" under Riesel & Sierpinski directories has all the project primes. (It is not yet updated with the latest 5 or 6 primes).

[fatphil]

I don't intend to test with all of the largest primes, as I don't believe size is the troublesome issue, so I'll test with tens of thousands of small primes first. If that all goes well, I'll get back to you. Mark has found a failure case on PPC, but so far everything works OK on x86. Does anyone else apart from Mark and I have non-x86 machines?

[fatphil]

Oh bollocks:

"The GeoCities web site you were trying to view has temporarily exceeded its data transfer limit. Please try again later."

I think I grabbed the whole riesel file, not didn't get the sierpinski one. Fortunately, I'm only looking at riesel ones currently.

[axn]

Originally Posted by fatphil

Oh bollocks:

"The GeoCities web site you were trying to view has temporarily exceeded its data transfer limit. Please try again later."

I think I grabbed the whole riesel file, not didn't get the sierpinski one. Fortunately, I'm only looking at riesel ones currently.

PM me your email-id and I can send you the both the files.

[axn]

Originally Posted by fatphil

I think I grabbed the whole riesel file, not didn't get the sierpinski one. Fortunately, I'm only looking at riesel ones currently.

YGM

[fatphil]

I'm reticent to hit asm as I'd like to keep my code portable.
Can anyone tell me roughly what times a 2GHz AthlonXP takes for the following, for example. I don't believe more than a couple of percent improvement would be possible on a register starved architecture like the x86. I guess I could actually download LLR at some point...

Actually testing 194196250*1953125^11117-1 (11119/24576 limbs)
310714*5^100057-1 [] is PRP. (t=364.70s)
Actually testing 5718593750*1953125^11264-1 (11266/24576 limbs)
73198*5^101383-1 [] is PRP. (t=376.42s)
Actually testing 34058000*1953125^11296-1 (11298/24576 limbs)
272464*5^101667-1 [] is PRP. (t=379.82s)
Actually testing 97943750*1953125^11394-1 (11396/24576 limbs)
31342*5^102551-1 [] is PRP. (t=384.77s)
Actually testing 172598750*1953125^11440-1 (11442/24576 limbs)
276158*5^102964-1 [] is PRP. (t=369.83s)
Actually testing 1876900*1953125^11509-1 (11510/24576 limbs)
75076*5^103583-1 [] is PRP. (t=377.25s)
Actually testing 197946250*1953125^11871-1 (11873/24576 limbs)
316714*5^106843-1 [] is PRP. (t=395.33s)
Actually testing 41354500*1953125^11925-1 (11927/24576 limbs)
330836*5^107328-1 [] is PRP. (t=399.21s)

(No erronious results on x86 - from about 15000 tests - looks good!)

Phil

[axn]

Originally Posted by fatphil

Can anyone tell me roughly what times a 2GHz AthlonXP takes for the following, for example.

Don't have any data on Athlons, but on my 2.8GHz Pentium D, tests of this size were taking appr. 460 sec. So I would guess that an Athlon XP 2GHz might take something in the vicinity of 800 sec?

[tnerual]

i will start a test on a athlon xp 2500+ @1980mhz.

is it good for your benchmark ?

i can't get closer

edit:
310714*5^100057-1 is a probable prime. Time: 1068.439 sec.

[michaf]

With an Athlon 64 X2 3800+ running at 2260 MHz I get about 3.78 ms per bit on number of size 5^138000-ish

[fatphil]

My stake in the ground (hmmmm, steak...) is 360s for 5^100000 at 2.083GHz, say. And the responses vary from:

* 1068s on 1980MHz -> 1015s
* ~460s on 2.8GHz P4 -> ~800s
* bigger number on Ath64 -> ~620s

I, so far, have only built my code for win/P3 and linux/ath. In windows, I use cygwin. The code is GPL, and so people can build their own (and are invited to, tweak the compiler options - tweak the code - or anything to make it better!). However, it requires building YEAFFT (from GLucas), and last time I looked, SSE2 didn't work in YEAFFT on cygwin. (I patched it last year, but the patches never got into the main line.)

I will try to build for other architectures when I can. I hope that linux/P4 and linux/Ath64 will make good use of SSE2.

So, those who are familiar with the concept of how I code - stuff's _always_ work in progress (ask Mark - he's had 5 code drops in as many days!) - are invited to volunteer to help me test this on real-world candidates. Please state chip type and OS, so we can work out the best way to get an executable to you (I build or you build).