Phrot announcements

[rogue]

I've decided to move phrot related items so that they could be in the same sub-forum as PRPNet updates.

I am releasing phrot 0.62. Here are a list of changes:

• Address a gcc bug (on PPC only) in getResidue() that caused loop to terminate on second iteration, thus producing an incorrect residue. Oddly this only seems to happen for bases that are powers of 2.
• Can now perform a Proth primality test for bases that are powers of 2.
• Added dynamic selection of witness (if not specified on the command line) so that tests of Proth numbers can be primality tests instead of PRP tests. Phrot should now select the same witness that LLR would select so that the residues match.
• Add 1 to the residue of Proth primality tests so that they are compatible with LLR residues.
• Fixed issue with variable base Cullen/Woodall ABC files.
• Removed automatic turn on of error checking when the base is a power of 2 as testing has shown that the problem was resolved in version 0.52.
• Limit the max base for a test to 2^20 (1048576) because higher values are unlikely to have valid results.

Of note, phrot's Proth test is about half the speed of LLR when run on x86. phrot's Proth test on PPC is about one third the speed of an LLR Proth test on a Core 2 Duo of similar speed. I haven't compared them on P3 or P4, but I expect that LLR is still faster. I have included a CygWin build of phrot in the attached zip file.

If you are using phrot to do Proth tests with the PRPNet client, you must run version 1.0.2 of the client.

[Cruelty]

I have just noticed that phrot.ckpt is being created every minute - what is the purpose? I am using Goeff's v.0.62 under 64-bit linux.

[rogue]

Originally Posted by Cruelty

I have just noticed that phrot.ckpt is being created every minute - what is the purpose? I am using Goeff's v.0.62 under 64-bit linux.

It is written every 4096 iterations. It does this so tests can be stopped and restarted without having to start from the beginning. There is very little overhead in creating the file. LLR does far more I/O than phrot.

[rogue]

This release is fairly minor. There are no bug-fixes. Here is a highlight of the changes:

• Include a readme_phrot.txt file which gives the program history and some information on running it.
• Modified the makefile and sources so that they do not rely on any external headers or sources (except for yeafft of course).
• Raised max base limit to 1500000 as tests of bases up that value test appear to test fine without triggering maxerr.

You can d/l phrot 0.63 from here.This distribution includes a build with CygWin.

I intend to flush out more information in the readme file in the next release.

[kar_bon]

Originally Posted by rogue

You can d/l phrot 0.63 from here.This distribution includes a build with CygWin.

link is not working!

[rogue]

Originally Posted by kar_bon

link is not working!

Sorry, link fixed.

[rogue]

I've attached 0.64 to fix that and to fix a separate issue with Geoff's Windows build. Here are the fixes:

• Fixed a bug where the incorrect residue was calculated for Proth tests when the candidate is composite.
• Fixed the printing of the residue for MinGW builds.

Hopefully nobody on x86 is using phrot for base 2 as LLR would be twice as fast.

[rogue]

You can d/l phrot 0.65 from here.

There are a couple of very minor bug fixes in this release, but none that would adversely affect tests that have already been performed. This link provides a build done with CygWin. Most of the changes were to make it easier to build. Here is a summary of the changes:

• Removed requirements to modify any glucas source code before building phrot.
• Added more documentation to this readme that describe how glucas needs to be built before linking with phrot.
• Fixed some ABC file format strings so that those files could be processed.
• Start phrot with below normal authority so that it doesn't suck all of the CPU on a single-core/single-CPU computer.

It will be much easier to build now than it was before. Jean Penne discovered an issue that prompted me to address possible compatibility issues with more current distributions of glucas, most important of which that phrot cannot be built with the glucas switch of Y_USE_SSE2.

Eventually I will incorporate a build that doesn't require CygWin.

[rogue]

You can d/l phrot 0.66 from here. These are the changes for 0.65:

• First build with MinGW (thus no requirement for CygWin).
• Add code so that MinGW builds to not use ANSI escape sequences.
• Removed limit on max base limit as testing implies it is safe.
• Modified makefile to work with other OSs/CPUs.
• Updated readme with additional notes on building phrot.

phrot 0.66 should be faster than LLR for base 3 (phrot 0.65 wasn't). LLR overtakes phrot in speed when the base exceeds 40, although that is dependent upon the CPU in which it is run. For example, phrot is probably faster than LLR on AMD CPUs to higher bases. It is worthwhile to do some tests to determine which program is faster on your CPU.

To make it clear, a MinGW build is included in this distribution (called phrot.p3, not phrot.exe). No third party DLLs should be necessary to run phrot. If someone discovers that my assumption is incorrect, please let me know.

[rogue]

This build affects users who use phrot for numbers where k or b is a power of 2. Here is what has changed:

• Fixed a bug that where phrot did not check for an FPU overflow, thus giving results that were invalid.
• Restored automatic error checking when b or k is a power of 2. It is extremely rare than an overflow occurs under these conditions. The only known case where this happens is 2*3^229712-1. There are undoubtably others, but until the root cause is found, this additional checking will remain.

[rogue]

There is only one change for 0.67. I have added a -t command line option. By using this option, phrot will print only the current test and its results or any errors. It will not tell the user if error checking is turned on for base 2, that the end of the input file has been reached or the line that gives the input. If using an ABC file, this will cut the amount of output in half. It will cut the output even more when using it for a Proth test.

Of note, if you are using PRPNet, then I strongly suggest that you update to this build. I will be making a change to the PRPNet client to use this option. In essence, that change to PRPNet will require the use of phrot 0.67.