Improving performance and productivity for software development on TI Multicore DSP platforms
Miguel Aguilar, Ronny Jimenez, Rainer Leupers, Gerd Ascheid · 2014
Complexity of modern applications, the performance requirements and the power constraints are the major driving forces that motivate the use of Multiprocessor Systems on Chip (MPSoCs). Programming these platforms is still a big challenge, posing a multitude of software design issues: What is the right MPSoC programming model to capture parallelism?, How to parallelize legacy C code?, How to achieve optimal utilization of processing elements?, How to minimize communication overhead?, How to explore the vast software mapping design space?. Traditional compiler technology does not solve these challenges, as it does not consider the architectural characteristics introduced by MPSoCs. Several research efforts have been directed to address these challenges. One example is the MAPS framework (MPSoC Application Programming Studio) that offers facilities for programming heterogeneous and homogeneous MPSoCs. In this paper, we focus on the applicability of this tool to the software development on the TI Keystone Multicore DSP platforms. The analysis considers both performance and productivity improvements achieved by MAPS.