Multithreaded refinement

[tdaede]

Medium-grained parallelism may be possible in the refinement stages. The idea would be to do the refinement in an interlaced pattern. Locking would only be needed near the end of a group of rows or columns.

[danesh-d]

Why don't you use OpenMP? It is supported by gcc as well. Moreover, should the decoder allocate all available processors in a machine or should it be configurable? Some people, specially when they are running the decoder on a notebook, might not like to allocate all their processors to a process.