I love these 'compile to hardware' projects. Pity they didn't manage to get the speedup they hoped for from the parallel approach, curious how the next version benchmarks against this one.