Complementing User-Level Coarse-Grain Parallelism With Implicit Speculative Parallelism