20220527, 05:03  #639  
Jun 2003
2^{2}·7·193 Posts 
Quote:
Quote:
When p=k*2^n+1, 2p+1 = k*2^(n+1)+3, but is still provable, since, by definition, we have 100% factorization of N1. But you'll need PFGW with p=k*2^n+1 as a helper Last fiddled with by axn on 20220527 at 05:07 

20220527, 13:15  #640 
"Mark"
Apr 2003
Between here and the
2^{2}×3^{2}×11×17 Posts 
From the pfgw doc on newpgen formats:
Code:
99991:S:0:11:10 k*11^n1 & k*11^(n+1)1 (SG) (generalized SG since base is not 2) 99991:C:0:2:5 k*2^n+1 & k*2^(n+1)+1 (CC) 99991:B:0:2:15 k*2^n+1 & k*2^(n+1)+1 (BiTwin) 99991:Y:3:2:29 k*2^n+1 & k*2^(n1)+1 & k*2^(n+1)+1 (LP) 99991:Z:3:2:46 k*2^n+1 & k*2^(n1)1 & k*2^(n+1)1 (LM) 99991:J:3:2:11 k*2^n+1 & k*2^(n+1)1 (Twin/SG) 99991:K:3:2:7 k*2^n+1 & k*2^(n+1)+1 (Twin/CC) 99991:1:3:2:42 CC 1st kind len 3 k*2^(n1)1 & k*2^n1 & k*2^(n+1)1 99991:2:3:2:21 CC 2st kind len 3 k*2^(n1)+1 & k*2^n+1 & k*2^(n+1)+1 The difference between "C", "1", and "2" is that "C" always has length of 2 whereas "1" and "2" will have a length > 2. Note that sgsieve is limited to base 2. It should allow for other bases. At some point I will modify sgsieve to be more generic. Some day I will create a ccsieve to support all of the forms above. sgsieve might be rolled into ccsieve when that happens. It appears that the pfgw doc has CC 1st kind and CC 2nd kind reversed per the definition of Cunningham Chains (see MathWorld), but it is possible that newpgen is also reversed. There are a few possible places with a mistake: the pfgw doc, the pfgw code, the llr code, the newpgen code. I suspect that all of them are wrong. It would be easy enough in the pfgw doc to change the description of "1" and "2", but that could be misleading if anyone is using newpgen. 
20220529, 15:25  #641 
"Alexander"
Nov 2008
The Alamo City
2^{6}×13 Posts 
Regarding the ARM version, what ever happened to the ARM ASM routines I sent you? Those were supposed to be translations of the fpu_*.S files for x86. Did those not work?
Edit: They were attached to this post. Last fiddled with by Happy5214 on 20220529 at 15:33 Reason: Found post 
20220530, 00:11  #642  
"Mark"
Apr 2003
Between here and the
2^{2}·3^{2}·11·17 Posts 
Quote:


20220602, 16:01  #643  
"AMD YES!"
Jan 2020
Bellevue, WA
2×41 Posts 
Quote:
I found a CC length 2 of 2nd kind with it anyways... 

20220602, 16:35  #644  
"Mark"
Apr 2003
Between here and the
2^{2}·3^{2}·11·17 Posts 
Quote:
If a term output from sgsieve is prime/PRP, then it is both SophieGermain and CC of the first kind, length 2. I need to look at sgsieve and verify that it is doing what I think it is doing. I just haven't had time to do that. If pfgw or llr say that both terms are prime/PRP, then that is good enough. 

20220603, 01:38  #645  
"AMD YES!"
Jan 2020
Bellevue, WA
2×41 Posts 
Quote:
Oh that's fine. At least twinsieve works :) LLR does find the primes, but they don't find SGs, as I said above. When will the fix come out? 

20220603, 12:28  #646 
"Mark"
Apr 2003
Between here and the
6732_{10} Posts 
I do not know. I have yet to finish the work for the Apple Metal support. I'm close to getting that working, but I haven't had much time.

20220603, 16:02  #647 
"Mark"
Apr 2003
Between here and the
2^{2}×3^{2}×11×17 Posts 
I have just uploaded mtsieve 2.3.2 to sourceforge. Here are the changes:
Code:
Switched to llvmmingw compiler in Windows. I have tried three other compilers, I cannot use gdb with the latest msys2 or mingw64 compilers. gdb gives an error when I start the programs, but only if they are built with g. The latest cygwin compiler fails to link the code. Others have seen the same error I have seen with other sofwtware and nobody has provided a solution. There were some spurious reports of bugs with the 2.3.0/2.3.1 release. This bugs seem to have gone away now that I switched to llvmmingw. framework: Updated primesieve to 7.8. All remaining SSE code has been removed due to a bug resetting the rounding mode. kbbsieve was the only program using it and is now faster without it. Created abstract Device and Kernel classes and extend OpenCL and Metal classes from those. This means that none of the GpuWorker classes need to include or need to rely on logic specific to OpenCL or Metal as the implementation is hidden from those workers. kbbsieve: version 1.1 Switched to Montgomery logic. This is about 20% faster than the SSE logic. 
20220603, 16:37  #648  
"Mark"
Apr 2003
Between here and the
2^{2}×3^{2}×11×17 Posts 
Quote:
Last fiddled with by rogue on 20220603 at 16:39 

20220606, 15:09  #649  
"Alexander"
Nov 2008
The Alamo City
2^{6}×13 Posts 
Quote:

