[deepnotderp]

1. Yes ,I've seen that, and I think that an OpenMP-esque annotation system specified by the programmer would be optimal, but I don't see any reason that a "sufficiently smart compiler" [ ;) ] can't produce an approximation. Given the SpMT-like "thread" (task) squashing, it shouldn't cause any semantic issues.

2. Absolutely, SpMT has been proposed as a general purpose technique, that's why I found it odd that the benchmarks weren't on a general purpose benchmark like SpecINT, do you have any benchmarks on SpecINT or any plan to do so?

[suvinay]

2: We are currently working on a compiler to automatically delineate tasks, assign domains etc. given sequential code. As part of that effort, we are looking at SPEC workloads, yes.