Hi all,

There is a new version of Genefer ready for testing and eventual release. This contains a new implementation of the 'x87' transform based on the same Z-transform algorithm as for others. It is slightly (up to 10%, in my tests) faster than the previous implementation, and also has smaller round-off errors, so higher B limit.

Since this release changes only the x87 transform, the amount of testing is somewhat reduced compared to the last release. We will only check the new transform and 'sanity check' the CUDA and OpenCL binaries to make sure nothing was broken in the build process.

This release is only relevant for the PRPNet GFN searches, so testing with BOINC is not needed.

To participate in testing, please visit the page below and reserve tests:

https://docs.google.com/spreadsheets/d/1Xi8LOg7Zb5b8T20lcqVlQozwKGibjPhQISVNfVxjOZk

Binaries are available from SVN:

Mac: https://www.assembla.com/code/genefer/subversion/nodes/800/trunk/bin/mac
Windows: https://www.assembla.com/code/genefer/subversion/nodes/801/trunk/bin/windows
Linux: https://www.assembla.com/code/genefer/subversion/nodes/802/trunk/bin/linux

Any questions or problems, please post in this thread.

Thanks in advance for the help!

- Iain
____________
Twitter: IainBethune
Proud member of team "Aggie The Pew". Go Aggie!
3073428256125*2^1290000-1 is Prime!

Can you make the google doc public, please? I'll throw a few idle Sandy cores at it.
____________
Eating more cheese on Thursdays.

Hi Iain,

desided to give it a try on GFN32768.
(EDIT: same for GFN65536)

First time it slipper from FMA3 to x87 as expected but then right to PFGW64.
So I did commented out PFGW64 in PRPClient.ini.
Next time, it slipper to LLR64.
If I comment out LLR64 as well, it goes through all Genefer transformation with maxErr and return result to the server and all over again.
Can't make PRPNet client to finish a test using Genefer64.

Any ideas?

c:\_PG\PRPNet.1>prpclient.exe
[2015-03-24 09:41:56 SE(e] PRPNet Client application v5.3.2 started
[2015-03-24 09:41:56 SE(e] User name Honza at email address is x@x.cz
[2015-03-24 09:41:57 SE(e] GFN32768: Getting work from server prpnet.primegrid.com at port 12005
[2015-03-24 09:41:58 SE(e] GFN32768: PRPNet server is version 5.3.0
Generalized Fermat Number Prime Search N=32768
genefer 3.2.7-dev (Windows/CPU/64-bit)
Supported transform implementations: fma3 avx-intel sse4 sse2 default x87
Copyright 2001-2015, Yves Gallot
Copyright 2009, Mark Rodenkirch, David Underbakke
Copyright 2010-2012, Shoichiro Yamada, Ken Brazier
Copyright 2011-2015, Iain Bethune, Michael Goetz, Ronald Schneider

Command line: genefer64.exe work_GFN32768.in

Priority change succeeded.
Start test of file 'work_GFN32768.in' - 09:41:58

Testing 9005940^32768+1...
Using FMA3 transform
Starting initialization...
Initialization complete (0.022 seconds).
Testing 9005940^32768+1... 757019 steps to go
maxErr exceeded for 9005940^32768+1, 0.5000 > 0.4500
maxErr exceeded while using FMA3; switching to AVX (Intel).

Testing 9005940^32768+1...
Using AVX (Intel) transform
Resuming 9005940^32768+1 from a checkpoint (757019 iterations left)

maxErr exceeded for 9005940^32768+1, 0.5000 > 0.4500
maxErr exceeded while using AVX (Intel); switching to SSE4.

Testing 9005940^32768+1...
Using SSE4 transform
Resuming 9005940^32768+1 from a checkpoint (757019 iterations left)

maxErr exceeded for 9005940^32768+1, 0.5000 > 0.4500
maxErr exceeded while using SSE4; switching to SSE2.

Testing 9005940^32768+1...
Using SSE2 transform
Resuming 9005940^32768+1 from a checkpoint (757019 iterations left)

maxErr exceeded for 9005940^32768+1, 0.5000 > 0.4500
maxErr exceeded while using SSE2; switching to Default.

Testing 9005940^32768+1...
Using Default transform
Resuming 9005940^32768+1 from a checkpoint (757019 iterations left)

maxErr exceeded for 9005940^32768+1, 0.5000 > 0.4500
maxErr exceeded while using Default; switching to x87 (80-bit).

Testing 9005940^32768+1...
Using x87 (80-bit) transform
Resuming 9005940^32768+1 from a checkpoint (757019 iterations left)

maxErr exceeded for 9005940^32768+1, 4.0000 > 0.4500
maxErr exceeded by all available transform implementations
Base factorized as : 2^2*3^2*5*50033
Base prime factor(s) taken : 50033
Starting N-1 prime test of 9005940^32768+1
Using generic reduction FMA3 FFT length 80K, Pass1=320, Pass2=256, a = 3
9005940^32768+1, bit: 40000 / 756998 [5.28%]. Time per bit: 1.277 ms.
____________
My stats

OK, the googledoc is now editable by everyone - sorry about that!

Honza, that behaviour is odd - I will have a look, it may indicate a problem.

eXaPower, those error are about right - x87 or 128 should both be extremely accurate.

- Iain


____________
Twitter: IainBethune
Proud member of team "Aggie The Pew". Go Aggie!
3073428256125*2^1290000-1 is Prime!

Honza, that behaviour is odd - I will have a look, it may indicate a problem.

Thanks for the doc change Iain.

With the fixed binaries, do I need to rerun/restart the tests I've done?
____________
Eating more cheese on Thursdays.

Thanks for the doc change Iain.

With the fixed binaries, do I need to rerun/restart the tests I've done?

Looks fixed.

[2015-03-25 09:37:22 SE(e] Server: GFN32768, Candidate: 9014520^32768+1 Program: genefer64.exe Residue: 10f4553c5eaf5be6 Time: 954 seconds
[2015-03-25 09:53:21 SE(e] Server: GFN32768, Candidate: 9014586^32768+1 Program: genefer64.exe Residue: 486d57ec309d4e19 Time: 954 seconds
[2015-03-25 10:09:27 SE(e] Server: GFN32768, Candidate: 9014640^32768+1 Program: genefer64.exe Residue: dd976e10f340db0a Time: 962 seconds
[2015-03-25 10:25:28 SE(e] Server: GFN32768, Candidate: 9014784^32768+1 Program: genefer64.exe Residue: 5d360771dc6a08b1 Time: 957 seconds

...

Testing 9014914^32768+1...
Using Default transform
Resuming 9014914^32768+1 from a checkpoint (753663 iterations left)

maxErr exceeded for 9014914^32768+1, 1.0000 > 0.4500
maxErr exceeded while using Default; switching to x87 (80-bit).
Too many errors with Default; Calculation will proceed using only more accurate transforms.

Testing 9014914^32768+1...
Using x87 (80-bit) transform
Resuming 9014914^32768+1 from a checkpoint (753663 iterations left)
Estimated time remaining for 9014914^32768+1 is 0:15:54
Testing 9014914^32768+1... 393216 steps to go (0:08:18 remaining)
^C cau[2015-03-25 10:33:18 SE(e] GFN32768: No data in file [genefer.log]. Is genefer broken?

c:\_PG\PRPNet.1>prpclient.exe
[2015-03-25 10:33:19 SE(e] PRPNet Client application v5.3.2 started
[2015-03-25 10:33:19 SE(e] User name Honza at email address is x@x.cz
It appears that the PRPNet client was aborted without completing
the workunits asigned by server GFN32768. What do you want to do with them?
1 = Report completed and abort the rest, then get more work
2 = Report completed and abort the rest, then shut down
3 = Return completed, then continue
4 = Complete in-progress, abort the rest, report them, then get more work
5 = Complete in-progress, abort the rest, report them, then shut down
6 = Complete all work units, report them, then shut down
7 = Complete all work units, then shut down
9 = Continue from client left off when it was shut down

Choose option: 5
Huh? The test did not complete, yet you didn't terminate it.[2015-03-25 10:33:48 SE(e] Total Time: 0:00:29 Total Work
Units: 0 Special Results Found: 0
[2015-03-25 10:33:48 SE(e] Client shutdown complete

I ran it with the new version. I did run probably 5 or 6 with the old version that immediately errored out before I realized I had copied the wrong file over.
____________
Eating more cheese on Thursdays.

Note, linux binaries have been posted that fix the aforementioned bug. Link in the first post has been updated.
____________
Twitter: IainBethune
Proud member of team "Aggie The Pew". Go Aggie!
3073428256125*2^1290000-1 is Prime!

If anyone can help with testing on Linux, please head over to the googledoc. Any CPU is fine for these tests. If you need any help, just ask!

Thanks

- Iain
____________
Twitter: IainBethune
Proud member of team "Aggie The Pew". Go Aggie!
3073428256125*2^1290000-1 is Prime!

If anyone can help with testing on Linux, please head over to the googledoc. Any CPU is fine for these tests. If you need any help, just ask!

Thanks

- Iain

I'll get the 32 bit Linux tests started tomorrow afternoon (PST) when I get home from work. Though if anyone else gets to it first, go ahead.
____________
Eating more cheese on Thursdays.

Thanks everyone who contributed. The testing was successful and the new binaries will appear shortly in the SVN. These will be added to the next prpnet package release, but there is no need to release them to BOINC as we don't use the x87 transform on any of the current active ranges.

- Iain

____________
Twitter: IainBethune
Proud member of team "Aggie The Pew". Go Aggie!
3073428256125*2^1290000-1 is Prime!