mtsieve

[rogue]

I have tracked down and resolved the issue with Legendre tables in srsieve2. That being written I have released mtsieve 2.5.0. Here are the changes:

framework:
   Fix how max p tested is computed as it could be wrong when running many workers.

srsieve2/srsieve2cl: version 1.7.2
   Reverted part of the change for algebraic factorizations as it excluded too many.
   Fix issue when using Legendre tables where factors would be missed.

twinsieve: version 1.6.2
   Fixed issue with base 2 where even k are factored even though even k are not used.
   The same issue can occur with odd bases.

xyyxsieve/xyyxsievecl: version 2.1
   Added -Z to allow user to force sparse logic, which can be faster or slower
   depending upon the density of x and y.
   Fixed an issue when starting with -i that can lead to a crash.

[ryanp]

Originally Posted by rogue

Ryan, I have not been able to reproduce the xyyxsievecl issue. I did make that change that I specified when starting with an old file. If you still have issues, then I will need some more help from you to track down the cause.

Let me send you my input file in a PM. I can repro it pretty consistently.

[rogue]

Originally Posted by ryanp

Let me send you my input file in a PM. I can repro it pretty consistently.

Okay. Is this with 2.1? Can you include the command line arguments you use?

[ryanp]

Originally Posted by rogue

Okay. Is this with 2.1? Can you include the command line arguments you use?

I hadn't tried 2.1 yet; just did an svn up and I'm trying it now. So far, at least, the input file reading is fixed:

$ ./xyyxsievecl  -P 1e12 -g 108 -i ./xyyx_1.5M.txt -o out.txt
xyyxsieve v2.1, a program to find factors numbers of the form x^y+y^x or x^y-y^x
GPU primes per worker is 2985984
Sieve started: 2009105977 <= p <= 1e12 with 295337 terms (300000 <= x <= 302000, 150001 <= y <= 159999) (expecting 66376 factors)

We'll see what happens after an hour.

[pepi37]

It looks like you fixed srsieve2. Now we have increase in speed since we can use Legendre tables, and still all factors are found :)

Good work! and thanks!

[storm5510]

Originally Posted by rogue

I have tracked down and resolved the issue with Legendre tables in srsieve2. That being written I have released mtsieve 2.5.0.

Looking good here. I replaced the v1.6.9 I was using with my R78 sieve. 63 sequences with 315K terms. It will get a good workout. I deleted the Legendre tables the old one was using and had the new one recreate them.

There is somewhat of a speed difference. 1.6.9 was running 210K p/sec. 1.7.2 is running 160K p/sec. If this is a result of error corrections, then this is the way it will be. I have no problems with it.

Many thanks!

[ryanp]

With xyyxsievecl 2.1, both the file reading bug and the "Fatal Error" crash after 1 hour appear to be fixed.

  p=11285155907, 22.40K p/sec, 22114 factors found at 1.67 sec per factor (last 303 min), 0.9% done. ETC 2023-07-16 20:51

[gd_barnes]

Srsieve2 1.7.2 fails on algebraic factor removal similar to how versions 1.7.0 and prior did.

It is removing algebraic factors that it should not be when the k and base are different perfect powers. Therefore primes could be subsequently missed.

This was fixed in version 1.7.1 at the expense of having it not remove some algebraic factors that it should. Unfortunately it is now removing more than it should.

Examples:

9*8^n-1: It removes all terms. It should only remove terms divisible by 2.
27*32^n-1: It removes all terms. It should only remove terms divisible by 3.
32*125^n-1: It removes all terms. It should only remove terms divisible by 5.

As best as I can tell, this is only happening on the Riesel side at this time.

On the Sierpinski side, it is still missing some algebraic factors but that would have no affect on future primes found.

For anyone using 1.7.2 for the extra speed because it is no longer missing factors when using the Legendre tables, you should be OK if your base is not a power. Storm and Pepi, this means you will be OK. Your bases 78 and 773 are not powers so this error will not affect them.

I will Email Mark the test cases for this later this morning.

[pepi37]

Ok thanks Gary

[rogue]

Originally Posted by ryanp

With xyyxsievecl 2.1, both the file reading bug and the "Fatal Error" crash after 1 hour appear to be fixed.

  p=11285155907, 22.40K p/sec, 22114 factors found at 1.67 sec per factor (last 303 min), 0.9% done. ETC 2023-07-16 20:51

Awesome!

[rogue]

Originally Posted by gd_barnes

Srsieve2 1.7.2 fails on algebraic factor removal similar to how versions 1.7.0 and prior did.

It is removing algebraic factors that it should not be when the k and base are different perfect powers. Therefore primes could be subsequently missed.

This was fixed in version 1.7.1 at the expense of having it not remove some algebraic factors that it should. Unfortunately it is now removing more than it should.

Examples:

9*8^n-1: It removes all terms. It should only remove terms divisible by 2.
27*32^n-1: It removes all terms. It should only remove terms divisible by 3.
32*125^n-1: It removes all terms. It should only remove terms divisible by 5.

For these cases it can remove terms because k=x^f, but it can also try to remove terms because k=x^f and b=y^g, but it should only remove terms if gcd(f,g) > 0. It is likely failing that.

I hope to take a look at it this weekend.

The speed loss is due to the fixed code.