From: Parry Husbands (pjrhusbands_at_lbl_dot_gov)
Date: Fri Jan 07 2005 - 09:19:52 PST
Venelin Mitov wrote: (snip) > The problem is that t0..tn-1 is a subrange of 0..THREADS-1, so > upc_barrier cannot be used. > Can you propose me an effective solution (maybe something with > upc_locks which works like "upc_barrier for threads t0..tn-1")? > Unfortunately UPC does not have a built-in facility for synchronizing subsets of threads. There are two solutions: - If the algorithm is synchronous enough, you might be able to get by with upc_barrier. - You can use the global memory to write the necessary synchronization primitives yourself. I've got such a library written. I can try to clean it up and send it to you if you like. Parry