Tools for Cross-Platform Software Development and Performance Projection

Developers of modern software, particularly those targeting mobile or embedded platforms, are presented with a wide variety of hardware choices. Solutions can be built on microprocessors (e.g. x86, ARM, or Power architecture), GPUs (e.g. NVIDIAs Tegra), DSPs, FPGAs, or some combination of these technologies. Not only does this place a significant burden on programmers in understanding the intricacies of developing for each, it makes it difficult to decide at design time which approach is preferable. EM Photonics is pursuing a technology that will allow developers to write computational kernels once, then have optimized binaries built for each platform of interest. Using information gleaned by our toolchain, we will also be able to project the softwares performance across various processors thus helping project managers determine which devices they should use for a given application. In this project, we will extend the frontend and backend components of this toolchain to provide a more powerful programming interface (via addition of MATLAB-style syntax elements) and much wider hardware support. Additionally, we will build the tools capable of analyzing the users code and predicting computational performance.