Universitat Oberta de Catalunya
The shift towards on-chip parallelism brings forth an effort to design intuitive parallel programming models that can be used by common programmers. In that context, dataflow programming models show promise for their simplicity and potential performance gains. However, dataflow models require complex runtime support as it is infeasible to statically identify all data dependencies at compile time. In this paper, the author's overview a hardware based out-of-order, pipeline that man-ages coarse-grain data dependencies at runtime.