Ah, thanks for the clarification. From the post I gathered that the atomic region locked the entire data structure, and that any thread would wait, I didn't realize it was a specific data region.
That sounds pretty nice, then. I'm looking forward to see the sort of speedups possible with this.
Well take a look at any of the copious literature on the subject:
This paper shows some example speedups, but generally doesn't give anything to compare to. Find papers that reference this paper, to see more examples.
C. C. Minh, J. Chung, C. Kozyrakis, and K. Olukotun. STAMP: Stanford transactional applications for multi-processing. In IEEE International Symposium on Workload Characterization, 2008.
That sounds pretty nice, then. I'm looking forward to see the sort of speedups possible with this.