This is n adapted version of the [original CUDA libFES implementation](http://www.polycephaly.org/projects/forcemq/index.shtml)
by Hsieh-Chung (Kevin) Chen, Tung (Tony) Chou, and Ruben Niederhagen.

Its in a way adapted, that each thread of the CUDA kernel can find multiple
solutions instead of only one. This is managed via shared memory. Thus each
thread can find all solutions and keep them in the GPU memory. Therefore a
second check function is provided, wich checks for each partial solutions if its
a ful solutions or not, while completly working on the GPU.

USAGE:
=====

An example of how to use the software is in `bench.sh` and `bensh_unroll.sh`


ORIGINAL README:
================

This software solves MQ systems on CUDA-capable GPUs.

This implementation is based on the source code from Hsieh-Chung (Kevin) Chen,
Tung (Tony) Chou, and Ruben Niederhagen (see AUTHORS file) from the paper "Fast
Exhaustive Search for Polynomial Systems in F2" by Charles Bouillaguet,
Hsieh-Chung Chen, Chen-Mou Cheng, Tony Chou, Ruben Niederhagen, Adi Shamir, and
Bo-Yin Yang (CHES 2010, Lecture Notes in Computer Science, Vol. 6225, pp.
203—218. Springer, 2010).

The input file format is the same as used in the Fukuoka MQ challenges [1].

The script 'gen_sys.py' can be used to generate a solvable random system for a
given number of variables and equations.

The script 'fix.py' can be used to fix some variables in an input system.

See the Makefile for examples, e.g., 'make test-small' and 'make test-large'.


Requirements:

- CUDA release 9.0
- Python 3.5.2