Hardware Benchmark Jest Thread for 100M exponents - Page 24

retina · 2019-02-04, 17:53

Quote:

Originally Posted by kriesel

Memory requirements will prevent running this test on my 4K SRAM 1Mhz 6502 Synertek SYM-1 board. (Unless there's a very compact program that could use cassette or open reel tape for large storage...)

OMG, I would love to see that complete just one iteration and get the correct residue. I doubt that I will live long enough to see that though.

kriesel · 2019-02-04, 17:58

Quote:

Originally Posted by retina

OMG, I would love to see that complete just one iteration and get the correct residue. I doubt that I will live long enough to see that though.

Actually, the first two iterations should be pretty easy, and fast too. I could program it to run variable length grammar school math. 4x4-2=14. 14x14-2=194. 8-bit unsigned ints. No mul instruction means build it from shifts and adds. Also, build a lookup table of square values as you go; 256 values, 2 bytes each. Or preload a precomputed table, 512 bytes. Off diagonal products, 4x4 table, and use that with Karatsuba or grammar school by nybbles. The hundredth iteration though, is big and slow.

retina · 2019-02-04, 18:05

Quote:

Originally Posted by kriesel

Actually, the first two iterations should be pretty easy, and fast too. I could program it to run variable length grammar school math. 4x4-2=14. 14x14-2=194. 8-bit unsigned ints.

No, we gotta have the shifted start value also, or spin through the first easy iterations before the real results start to matter. No easy cheating allowed. A proper complete iteration of the full 100M value.

kriesel · 2019-02-04, 18:28

Quote:

Originally Posted by retina

No, we gotta have the shifted start value also, or spin through the first easy iterations before the real results start to matter. No easy cheating allowed. A proper complete iteration of the full 100M value.

There you go again, making rules during the game.

retina · 2019-02-04, 18:33

Quote:

Originally Posted by kriesel

There you go again, making rules during the game.

Of course.

Someone has to set the rules. Might as well be me. Any objections? No. Okay, I'll grudgingly do it. You're welcome.

kriesel · 2019-02-04, 18:40

Quote:

Originally Posted by retina

Of course.

Someone has to set the rules. Might as well be me. Any objections? No. Okay, I'll grudgingly do it. You're welcome.

Do you by chance work with tax law? (Those darn taxpayers, following the rules to save money. Looks like we better change the rules during the game again, they're getting away with keeping too much of their own money.)
And to answer your question, actually, yes.

ldesnogu · 2019-02-06, 16:28

This reminds me of my first implementation of LL. It was in assembly language on an Atari ST (68000 CPU). It was using Karatsuba mul and the data was put in video RAM to watch the computation. Quite funny

That was almost 10 years before prime95 was released. Make me feel old...

kriesel · 2019-02-06, 16:52

Quote:

Originally Posted by retina

Of course.

Someone has to set the rules. Might as well be me. Any objections? No. Okay, I'll grudgingly do it. You're welcome.

One of the rules of any credible game (and some not so credible) is the referees don't get to play, or to gamble on the outcome.

retina · 2019-02-06, 17:12

Quote:

Originally Posted by kriesel

Do you by chance work with tax law? (Those darn taxpayers, following the rules to save money. Looks like we better change the rules during the game again, they're getting away with keeping too much of their own money.)
And to answer your question, actually, yes.

There ain't no money at stake here. Feel free to post all sorts of results with sub-par code. It won't mean anything though.

Quote:

Originally Posted by kriesel

One of the rules of any credible game (and some not so credible) is the referees don't get to play, or to gamble on the outcome.

Referees only enforce the rules. That doesn't necessarily mean they make them. Oh, I see, you would like me to be a referee also, okay, sure. You're welcome.

kriesel · 2019-02-06, 17:49

Quote:

Originally Posted by retina

There ain't no money at stake here. Feel free to post all sorts of results with sub-par code. It won't mean anything though.Referees only enforce the rules. That doesn't necessarily mean they make them. Oh, I see, you would like me to be a referee also, okay, sure. You're welcome.

Even the NFL commissioner has to stick with one set of rules for a whole season. Unless it interferes with selecting a patsy for the Patriots, that is. And you know that referee was an analogy, not an appointment. (Your eyesight may be too good to qualify, anyway.)

ewmayer · 2019-02-07, 02:46

Quote:

Originally Posted by retina

No, we gotta have the shifted start value also, or spin through the first easy iterations before the real results start to matter. No easy cheating allowed. A proper complete iteration of the full 100M value.

OK, fine - let initial shift = s, thus initial LL-test seed = 4*2^s.

Iteration 1 gives (4^2-2)*2^2s (mod 2^p-1) = 14*2^(2s % p).

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Perpetual benchmark thread...	Xyzzy	Hardware	897	2023-06-15 13:46
Sieve Benchmark Thread	Historian	Twin Prime Search	105	2013-02-05 01:35
LLR benchmark thread	Oddball	Riesel Prime Search	5	2010-08-02 00:11
sr5sieve Benchmark thread	axn	Sierpinski/Riesel Base 5	25	2010-05-28 23:57
Old Hardware Thread	E_tron	Hardware	0	2004-06-18 03:32

2019-02-06, 16:28	#260
ldesnogu Jan 2008 France 3×199 Posts	This reminds me of my first implementation of LL. It was in assembly language on an Atari ST (68000 CPU). It was using Karatsuba mul and the data was put in video RAM to watch the computation. Quite funny That was almost 10 years before prime95 was released. Make me feel old...