These are benchmarks from the FreeBench benchmark suite, version 1.03.

Freebench is a freely available and redistributable suite, with a web page here:
http://www.freebench.org/

In default mode, the LLVM harness runs the 'test' inputs, with 
LARGE_PROBLEM_SIZE, they run the full sized "ref" inputs.

Characteristics of the individual benchmarks are described here:
http://www.freebench.org/benchmarks.shtml