Think about the dense containers in ArBB. To create those dense containers you either need to bind to an existing C++ array or use the range interface to populate it in the same style/syntax as a C++ array. You then define a kernel that takes as input/output ArBB dense containers, define the operations you would like done on those containers, and then use call() to invoke the ArBB function. So the dense container, a templatized C++ array that allows the runtime to evaluate only the ArBB types, is the essential building block for everything you do with the product.

So with Threading Building Blocks, you manage tasks (collections of threads) to solve a wide variety of parallelism problems, and with Array Building Blocks, you manage arrays to solve a wide variety of (mainly, but not restricted to) data parallel problems.

I will write a blog entry this week describing the logic behind the naming of all the Parallel Building Blocks (IntelTBB, IntelArBB, and IntelCilkPlus and link you to it once I am finished as well.

Thanks,

Noah