Intel, AMD Promote Parallel Programming

With everybody and your Aunt Sally making the move from single- to dual-core and quad-core CPUs, software developers face skyrocketing demand for multithreaded applications that can take advantage of the chips' simultaneous processing capabilities. This week, AMD and Intel vowed to lend them a hand.

AMD announced an initiative dubbed Hardware Extensions for Software Parallelism that will enlarge the x86 instruction set with extensions for real-time optimization of concurrent application tasks. The first of these make up a specification for Light-Weight Profiling, which enables code to make dynamic, on-the-fly decisions about how best to boost the performance of running tasks via memory organization, cache data relocation, and other techniques.

According to AMD, the additions will be particularly beneficial to multithreaded runtime environments such as Java and .NET. Naturally, while AMD will implement the extensions in its own CPUs (as it did with 3DNow! in 1998), Intel feels no obligation to do so.

For its part, Intel has released its Threading Building Blocks C++ library as an open-source (GNU General Public License v2) project to make parallelism more accessible for programmers. The library gives C++ developers generic code that implements tasks instead of threads, with an abstraction for parallelism that doesn't require the low-level programming of threading packages such as p-threads or Windows threads. Intel adds that the library is future-proof for quad-, octo-, and zillion-core processors, detecting and adapting to the number of cores on the hardware platform.