PFGW 3.8.3 (with gwnum v28.7) Released - Page 18

rogue · 2013-07-20, 21:57

I have posted pfgw 3.7.6 at sourceforge. The only change is to use gwnum v27.11 which fixes a bug with small AVX FFTs. This bug would cause numbers tested as PRP to be tested as composite with the primality test.

ET_ · 2013-07-22, 07:54

Quote:

Originally Posted by rogue

I have posted pfgw 3.7.6 at sourceforge. The only change is to use gwnum v27.11 which fixes a bug with small AVX FFTs. This bug would cause numbers tested as PRP to be tested as composite with the primality test.

Did the bug affect SandyBridge PCs with the -gx option for GFN primality? If so, I'll have to retest some ranges.

Luigi

rogue · 2013-07-22, 12:16

Quote:

Originally Posted by ET_

Did the bug affect SandyBridge PCs with the -gx option for GFN primality? If so, I'll have to retest some ranges.

I know if affected use of the -t switch. I do not know about -gx. If the FFT size changed from 3.7.5 to 3.7.6, (verify with -V) then you should retest.

ET_ · 2013-07-22, 13:43

Quote:

Originally Posted by rogue

I know if affected use of the -t switch. I do not know about -gx. If the FFT size changed from 3.7.5 to 3.7.6, (verify with -V) then you should retest.

Thanks for the hint. I will check.

Luigi

Prime95 · 2013-07-22, 15:38

Quote:

Originally Posted by ET_

Did the bug affect SandyBridge PCs with the -gx option for GFN primality? If so, I'll have to retest some ranges.

The bug affected small FFTs -- very likely only FFT size 32. I can't rule out that FFTs up to size 256 weren't affected, but it would take a combination of rare circumstances. FFT size 32 only needed a single rare circumstance.

The bug fix will not cause a different FFT size to be selected.

rogue · 2013-07-22, 16:53

Quote:

Originally Posted by Prime95

The bug affected small FFTs -- very likely only FFT size 32. I can't rule out that FFTs up to size 256 weren't affected, but it would take a combination of rare circumstances. FFT size 32 only needed a single rare circumstance.

The bug fix will not cause a different FFT size to be selected.

Really? I thought that the ones I reported were 32 in 3.7.5 and 64 in 3.7.6.

Prime95 · 2013-07-22, 21:05

The case I debugged was 10^107+57 which is a 356-bit number. I'm fairly certain prime95 was using a length 32 FFT. That doesn't mean one of the other test cases didn't use a length 64 FFT.

rogue · 2013-07-22, 22:41

Quote:

Originally Posted by Prime95

The case I debugged was 10^107+57 which is a 356-bit number. I'm fairly certain prime95 was using a length 32 FFT. That doesn't mean one of the other test cases didn't use a length 64 FFT.

That's quite possible.

rogue · 2013-07-23, 01:46

I have posted pfgw 3.7.7 at sourceforge. The only change is to fix a bug with modular factoring. The bug has been around for a long time, but first reported a few months ago. I finally had time tonight to dig into it and fix it.

pepi37 · 2013-07-23, 07:30

If you can remember from what version PFGW use AVX?
Thanks for reply

rogue · 2013-07-23, 12:47

Quote:

Originally Posted by pepi37

If you can remember from what version PFGW use AVX?
Thanks for reply

PFGW has supported AVX since version 3.6.2. It is almost always recommended to use the latest release.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
A possible bug in LLR/PFGW while using GWNUM (no bug in P95)	Batalov	Software	77	2015-04-14 09:01
PFGW 3.2.0 has been Released	rogue	Software	94	2010-09-14 21:39
PFGW 3.2.3 has been Released	rogue	Software	10	2009-10-28 07:07
PFGW 3.2.2 has been Released	rogue	Software	20	2009-08-23 12:14
PFGW 3.2.1 has been released	rogue	Software	5	2009-08-10 01:43

2013-07-20, 21:57	#188
rogue "Mark" Apr 2003 Between here and the 6354₁₀ Posts	I have posted pfgw 3.7.6 at sourceforge. The only change is to use gwnum v27.11 which fixes a bug with small AVX FFTs. This bug would cause numbers tested as PRP to be tested as composite with the primality test.

2013-07-22, 21:05	#194
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 1110101110001₂ Posts	The case I debugged was 10^107+57 which is a 356-bit number. I'm fairly certain prime95 was using a length 32 FFT. That doesn't mean one of the other test cases didn't use a length 64 FFT.

2013-07-23, 01:46	#196
rogue "Mark" Apr 2003 Between here and the 14322₈ Posts	I have posted pfgw 3.7.7 at sourceforge. The only change is to fix a bug with modular factoring. The bug has been around for a long time, but first reported a few months ago. I finally had time tonight to dig into it and fix it.

2013-07-23, 07:30	#197
pepi37 Dec 2011 After milion nines:) 2653₈ Posts	If you can remember from what version PFGW use AVX? Thanks for reply