Multicore Programming is One Tough Nut
The new Parallel Computing Lab at the University of California at Berkeley is funded by Intel and Microsoft. UCB beat out MIT, Stanford, and the University of Illinois, for the honors of cracking this tough nut.
The EETimes posting Multicore puts screws to parallel-programming models gives some of the details.
I’ve seen several schemes over the years to harness Symmetric multiprocessing, or SMP, computing cycles. Some highly hand tweaked programs showed what could be done. I don’t think any compiler has been very successful turning generic source code into screaming performers.
The Parallel Computing Laboratory (“Par Lab”) approach is
We decided on a fresh approach: to start top-down from applications; to innovate across disciplinary boundaries by creating a culture that encourages interaction and cooperation; and to create prototypes that can be quickly adapted to reflect multidisciplinary innovation.
I’m looking forward to the prototypes, and seeing how they crack the share-cache and share-nothing multicore nuts.
…John
