Inthreads: a low granularity parallelization model

We present Inthreads, a computational model aimed at medium-level parallelism. The threads in our model, based on shared architectural registers and having extremely low overheads, are useful for parallelization at a level of a single function call.Inthreads parallelization provides significant speedups, from 35% in case of well-behaved programs to over 3.5x for benchmarks with problematic patterns of branches and memory accesses.