C++ AMP Provides Massive Parallelism
Microsoft wants to give C++ developers tools for writing parallel applications running on zillions of GPUs/cores locally or in the cloud.
Visual Studio C++ 2010 enables developers to write programs that take advantage of local parallel hardware such as multicore CPU using the Concurrency Runtime which has four major components: Parallel Patterns Library (PPL), Asynchronous Agents Library, Task Scheduler, and Resource Manager. But Microsoft wants to extend Visual C++ parallelism support even further. Herb Sutter, Microsoft Principal Architect, Native Languages, has announced a new C++ library called Accelerated Massive Parallelism (C++ AMP), at AMD Fusion Developer Summit. AMP extends parallelism support to local GPGPU and Accelerated Processing Unit (APU), and intends to cover it all the way to the cloud.
Microsoft wants to make C++ AMP available on all sorts of heterogeneous parallel hardware including multicore systems and clouds, aiming to support applications running on hundreds of thousands of cores/GPUs in the cloud.
C++ AMP introduces just one language extension, and makes use of DirectX 11’s DirectCompute API which offers general processing support on GPGPUs, “the architecture being such, that at this point can be thought of as an implementation detail that does not surface to the API layer”, according to Daniel Moth, Senior Program Manager at Microsoft.
C++ AMP will be an open specification, and Microsoft will provide a first implementation some time this year, but full support for this technology is to be included in the next version of Visual C++. C++ AMP will be basically an STL-like library “as part of the existing concurrency namespace and delivered in the new amp.h header file”, wrote Moth. Programming with AMP will benefit from all programming support available in VS: intellisense, building, debugging, profiling, etc.
Moth promised he will show C++ AMP code samples on his blog in the following months.