Programming Many-Cores on Different Levels of Abstraction

Many-core hardware is targeted specifically at obtaining high-performance. However, obtaining high-performance is often challenging because hardware-specific details have to be taken into account. This leads to low-level code that is difficult to read, maintain, and port to other architectures. This is a well-recognized problem and there are many programming models that try to balance a high-level programming model with high-performance. We propose an entirely different, novel approach that allows programmers to define and choose their own level of abstraction: a high level of abstraction for readability and portability, and user-defined, lower levels to incorporate more hardware-specific details to obtain higher performance.