by Alan Gatherer
This is the second of a multi-part 2020 Vision series outlining what the future may hold, as viewed by technologists within Texas Instruments. Click here for part 1.
Predicting the future is primarily an act of the imagination. However, digital signal processors are showing some strong trends and I think it is possible to predict what will happen in the next few years as we move towards the next order of magnitude increase in computational efficiency.
2010″2015: Component-based software. The number of cores on a device is still fairly modest, and individual software components are developed for a single computational cluster by “component developers” and then “assembled” onto a multi-core system. Development tools for this methodology improve steadily as virtualization of hardware through middleware is driven by efforts such as the SCA (Software Communications Architecture) for SDR (software-defined radio). Auto generation of glue code between components becomes the norm.
2015″2020: Single program multiple data (SPMD). The component-based approach begins to fail as the number of cores reaches 32. Turning to techniques used in high-performance computing (HPC), the embedded software community develops the SPMD approach where a program can be compiled to run over multiple cores. While initially requiring explicit description of the communication flow, pragmas are now employed to enable the parallel nature of algorithms to be exploited by a variety of multi-core devices.
