mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Software (https://www.mersenneforum.org/forumdisplay.php?f=10)
-   -   mtsieve (https://www.mersenneforum.org/showthread.php?t=23042)

pepi37 2022-03-22 00:13

What is use of k1b2sieve? ( what kind of sieve files we got, or what type of prime we search)
Thanks

rogue 2022-03-22 12:20

[QUOTE=pepi37;602254]What is use of k1b2sieve? ( what kind of sieve files we got, or what type of prime we search)
Thanks[/QUOTE]

These are primes of the form 2^n+c for variable n and c. The output file is ABC, which can be used with pfgw. I do not know if its ABC format is compatible with llr. When abs(c) > 1 the numbers will be PRP, unless small enough to be proven prime using other means.

pepi37 2022-03-22 12:27

Is there any prime in that form exist?

rogue 2022-03-22 14:02

[QUOTE=pepi37;602286]Is there any prime in that form exist?[/QUOTE]

Mersenne Primes (when c = -1). Nothing for c = +1.

Almost all other PRPs cannot be proven prime with current hardware.

chris2be8 2022-03-22 16:41

[QUOTE=rogue;602295]Nothing for c = +1.
[/QUOTE]

Except for Fermat primes when n is a power of 2.

Most cases with abs(c) > 1 will be difficult to prove prime if they are large enough to be out of range of primo etc. Exceptions might be when c+/-1 is a power of 2 and the remainder after dividing out 2s has a lot of algebraic factors (or is a Mersenne prime).

rogue 2022-05-09 14:28

I am pleased to announce that mtsieve 2.3.1 is now available at sourceforge. Here is a summary of the changes:

[code]
framework:
Updated to msys2 gcc version 11.2.0, which required changing string to std:string
and vector to std::vector instead of using the std namespace.

Due to a compiler bug, now compiling xyyxsieve and afsieve with -O2 instead of -O3.

Refactored the OpenCL implementation so that Metal implementation can use the same
interface as OpenCL. As a result KernelArgument.cpp no longer exists. The Kernel
now manages all CPU and GPU memory used by the Kernel.

Removed the embedded ASM logic for NVIDIA from all GPU kernels. The OpenCL compiler
generates code that is about 10% faster. This is likely due to hard-coded register
usage in the ASM.

Added Metal support (for Apple hardware) since Apple has deprecated OpenCL on their
hardware. As a result, all sieves that can use OpenCL on Apple hardware can now use
Metal on that same hardware.

Added ARM support for sieves that do not require x86 ASM functions.

Updated factor validation of mfsieve, gcwsieve, etc. to no longer require x86 asm.

Moved cltoh.pl to the main directory. make will now build the xxx.gpu.h file used
by the GpuWorkers as part of compiling the GpuWorker object files.

afsieve, afsievecl: version 1.2:
The GPU code uses the Montgomery logic for the mulmod.
All factors, found by either CPU or GPU are now validated.

psieve, psievecl: version 1.5
Replaced x86 ASM FPU code with Montgomery logic for the mulmod as the x86 ASM FPU
code was missing factors. There is no speed difference between the two. The x86
AVX code worked and has not been changed.

srsieve2, srsieve2cl: version 1.6.2
Changed default value for -g from to to 16. GPUs tend to like powers of 2 due
to their architecture.

Added -C forsieving with the GPU. This reduces threading overhead which has a
noticeable impact on sieving speed when sieving a single sequence. Using -C5 can
improve speed by 50%.

Change -S to -K so that less guessing is needed when one requires multiple GPU
kernels due to having a lot of sequences. The program will now create groups of
sequences that are approximately the same size for each call to the kernel.
[/code]

This doesn't mean that the Metal code is working yet or that all of the sieves will build on Apple M1. Many of them will build and run on Apple M1 out of the box but others needs more work. The components for Metal support are there, but I haven't tried to compile any Metal code yet. I have three tasks remaining:
[LIST=1][*]Build and test sieves with no x86 ASM on Apple M1.[*]Modify all Open CL sieves to build with Metal on Apple Intel.[*]Build and test Metal sieves with no x86 ASM on Apple M1. These are the sieves for the first item that also have an OpenCL version.[/LIST]
What has been most annoying for me is that I cannot upgrade gdb on msys2 so I cannot use the debugger on Windows at all right now. I am getting an error trying to update gdb using pacman and none of the resolutions I have tried (per google) have worked.

pepi37 2022-05-14 00:35

e:\PRIME>srsieve2 -P 50000000000000 -W 1 -w1e6 -L legend.txt -i t17_b10_k99999998.npg -o t17_b10_k99999998.npg -O fact999991.txt -f B
srsieve2 v1.6.2, a program to find factors of k*b^n+c numbers for fixed b and variable k and n
Sieving with single sequence c=1 logic for p >= 110728654783
BASE_MULTIPLE = 30, POWER_RESIDUE_LCM = 720, LIMIT_BASE = 720
Split 1 base 10 sequence into 30 base 10^240 sequences.
Fatal Error: Could not open Legendre file legend.txt\b10_k99999998_c-1.leg9999998)

pepi37 2022-05-14 06:22

e:\PRIME\srsieve2 -P 50000000000000 -W 6 -w1e7 -i sr_10.abcd -O factors.txt
srsieve2 v1.6.2, a program to find factors of k*b^n+c numbers for fixed b and variable k and n
Sieving with multi-sequence c=1 logic for p >= 79917202699
BASE_MULTIPLE = 2, POWER_RESIDUE_LCM = 720, LIMIT_BASE = 720
Split 10 base 10 sequences into 27 base 10^144 sequences.
Legendre summary: Approximately 4740 B needed for Legendre tables
10 total sequences
1 are eligible for Legendre tables
9 are not eligible for Legendre tables
1 have Legendre tables in memory
9 cannot have Legendre tables in memory
0 have Legendre tables loaded from files
1 required building of the Legendre tables
864000 bytes used for congruent subseq indices
12000 bytes used for congruent subseqs
Sieve started: 79917202699 < p < 5e13 with 6902 terms (100005 < n < 999892, k*10^n+1) (expecting 1409 factors)

And just exit , no sieve process started
When I use old version of srsieve2 sieving process continue without any problem



e:\PRIME\srsieve2 -P 20000000000000 -W 6 -w 5e6 -i sr_10.abcd -O factors.txt
srsieve2 v1.5.3, a program to find factors of k*b^n+c numbers for fixed b and variable k and n
Cannot use Legendre tables because square-free part of k is too large
Must use generic sieving logic because there is more than one sequence
Sieving with generic logic for p >= 79917202700
Split 10 base 10 sequences into 14 base 10^96 sequences.
Sieve started: 79917202700 < p < 2e13 with 6897 terms (100005 < n < 999892, k*10^n+1) (expecting 1244 factors)

rogue 2022-05-14 13:09

Can you send me the file you are using?

Also, you can use -l0 to use the generic logic for multiple sequences. It will be faster.

pepi37 2022-05-14 14:44

[QUOTE=rogue;605867]Can you send me the file you are using?

Also, you can use -l0 to use the generic logic for multiple sequences. It will be faster.[/QUOTE]


[URL]https://www.dropbox.com/t/YD3zba2CafIGX80M[/URL]


bat file, exe file and sieve file inside zip file :)

rogue 2022-05-16 14:42

There are two different bugs. Unfortunately I need to update gdb in my environment as it doesn't work with the newer gcc, but I'm getting errors when trying to update gdb and nothing I've tried (per google searches) has worked yet. Ugh!


All times are UTC. The time now is 04:29.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2023, Jelsoft Enterprises Ltd.