17 Mar
2010
17 Mar
'10
8:47 a.m.
On 17/03/2010 00:17, Louis Wasserman wrote:
I tested, and this implementation actually performs better if the spine is maintained lazily, so we'll test that version.
May I request that, unless there's a significant speedup from using a strict spine, that you use a lazy spine where possible. The reason being that lazy data structures work much better in a parallel setting: a strict spine is a course-grained lock on the whole operation, whereas a lazy spine corresponds to a fine-grained locking strategy. Apart from this, as you were ;) Cheers, Simon