SAN FRANCISCO With the rise of multicore CPUs the push for parallel software has become job number 1 for Intel's microprocessor R&D labs. As part of its effort, the 250 person group is researching domain-specific languages, as well as parallel versions of existing languages and applications.
"In the past software was just looking to run at ever higher frequencies, but that formula is about to change and the free ride is over," said Shekhar Y. Borkar, an Intel fellow and director of the company's microprocessor technology lab in Hillsboro, Oregon.
Driving more parallelism into software "is the top priority for my group. It's imperative that software follow Moore's Law to double parallelism every two years or so," said Borkar in a group press interview here Friday (May 25).
Borkar said as many as 200 of the 250 people in his lab are working on projects in parallel code. "If you look at our Oregon site about three quarters of what we are working on are software issues," he said.
The projects span a variety of areas including new and enhanced languages and applications. For example, "we are researching how to put data-parallel constructs into interpreted languages such as Ruby," he said.
One of the most promising new research areas is in what Borkar called domain-specific languages that are meant to bring explicit parallelism into software in sectors such as graphics, networking, debuggers and libraries. For example, Intel is now in the process of creating a domain-specific language to handle processing TCP/IP in an explicitly parallel manner.
The company is also prototyping applications in areas known for being highly parallel such as recognition and game graphics. "This is all about solving matrices, and we know how to do that in parallel from our supercomputing days," said Borkar who worked for several years in a now-defunct Intel's supercomputing division.
Borkar used the press meeting to call for developers to accelerate their efforts in parallel software across application, task and basic coding levels. "There is an abundance of parallelism out there to be harvested, and we are evangelizing universities and industry to do it, but this is not going to happen in a year or two," he said.
Borkar said the move to teaching parallel programming techniques at the undergraduate level is getting real traction at a variety of top universities including the Berkeley, Stanford, MIT and the University of Illinois. "This is really happening now," he said