ispc

ispc represents an alternative to C programming language and features several extensions to SPMD or single program, multiple data. Based on the principles of the later, developers can write apps that appear to be regular serial programs in spite of the fact that the execution model consists of a number of program instances that are executed parallel on the hardware and more specifically, the CPU and GPU.

The direct advantage of the parallelization across devices is that the programs can help improve the overall performance. Then again, developers should bear in mind that the performance achieved via the newly created program is scalable with the number of cores of the CPU and the vector unit size. The program can create an abstraction layer between the hardware and programmer, where the latter benefits from some sort of execution and data model and can therefore, map the source program to the compiled assembly and hardware.

A further noteworthy feature of the program is that it enables developers to harness the computation power of SIMD vector units more conveniently. Simply put, developers do not need to spend too much effort to write intrinsics directly. The convenience of the tool also stems from the fact that it supports C syntax, idioms and it can adopt existing software systems.