Yeah I remember that, it came out only a day or two before FloppusMaximus announced he factored a key on United-TI. It was a major revolution when BrandonW released this, even thought it was quickly obsoleted

I'm glad to hear this, I've been checking the discussion on and off in #omnimaga/OmnomIRC in the past two days. I wish you good luck for the networked version. I would love to help factoring that key.

Well, I don't know what your exact problem was, but the "0x" in "mpz_init_set_str(tap, "0xa48...");" was definitely a problem
(I got myself bit too several days ago by the exact same problem, in Java ).


I'm attaching a new version of the C(++) LFSR-based program. It's much faster than the Perl script indeed, I thought the overhead of Perl would be lower. The LFSR is seeded with the current time; for better seeding, on POSIX platforms, we could use gettimeofday().
Some performance analysis using Valgrind on my Core 2 reveals that:
* it's much faster to mindlessly divide by all odd integers than to use any form of primality testing. And we can do slightly better than mindlessly dividing, by adding several trial divisions by the smallest primes;
* mpz_divisible_p is slightly faster than mpz_mod + mpz_cmp_si.

Enjoy

NOTE1: you probably want to rate-limit the program's output further, e.g. write data every 1048576 iterations rather than every 65536 iterations
NOTE2: for best results, compile GMP yourself from the sources (`./configure --prefix=$PREFIX --enable-cxx; make; make check; make install`): GMP tailors itself to your processor architecture, which results in a noticeable speedup for me, compared to stock Debian x86_64 binaries.

[EDIT: attaching a new version, with resuming support and automatic "0x" removal.]

[q u o t e=<poster name>]...[/q u o t e]

With GCC, I'd say yes. With compilers that don't support 'long long', no
(well, MSVC doesn't embed stdint.h, unless you're using the 2010 version. I don't know whether that one supports 'long long')

I don't know what any of that stuff means, being a TI-BASIC/Axe coder mostly, but good luck on this project

We don't have access to the only practical method (or rather, the only not-completely-impractical method) for factoring 1024-bit integers.

My 4 cores (4-year-old models, Core 2 Duo T7200 & T7600) perform between 10^6 and 2*10^6 per second each, which translates to <250K seconds (~3 days full-time) for 10^12 operations. Even if we used a million equivalent computers for a year (which is wildly unrealistic), we wouldn't reach much above 10^20 operations, hence my figure above.
TF-ing these keys is no different in principle from playing at the lottery - the odds are extremely lower, but no money (my computers are loaded 100% with grid computing anyway, this is just a different kind of load) is involved.