XKCD Hash SKein1024 brute forcing.

Goal is to find input strings such that skein1024(input) is as 'close' as possible to the given value.

This is for XKCD's April 1st comic: http://xkcd.com/1193/

Primary UIUC implementation.

$ make
$ ./hash <num_threads>

If unspecified, num_threads is set to the number of available processors.

This should work on most 32/64bit machines, and has been reported to work well on both Linux and Mac machines.

On Linux, we use assembly versions of the most compute-intensive part of the skein hash calculation, but are unable to use this on Mac OS X due to incompatibility with the gas style assembly.

Porting this to windows should be straightforward, although things like the way we seed the PRNG will need to be changed.

The Skein implementation used in this project is from the most recent NIST submission, making use of the assembly implementation when possible. As a byproduct of this competition, I ended up fixing a few minor bugs that caused their implementation to generate incorrect hashes or crash. These changes may be use beyond this project.

Our string generation is accomplished by periodically filling the majority of the string with random characters from our character set, and exhaustively trying all possibilities for the remaining characters.

Generating random strings is expensive, requiring many calls to random(). The exhaustive search is done to amortize this cost, while still mostly generating random strings.

The repeated random generation is an attempt to prevent two identical starting states from causing hosts to compute purely redundant calculations. This way, happening to output the same 128 integers is not taken to be a reflection of having the same internal PRNG state.

Strings of length 128 are used to match the 1024bit block size, which means we can hash them faster.

For fun, we use a fixed vanity prefix and suffix string in the generated input strings. This also aids in demonstrating who was responsible for generating a particular string.