Tools for Cross-Platform Software Development and Performance Projection

ABSTRACT: The past decade has seen a proliferation of computing devices. A landscape once dominated by single core microprocessors has given way to a plethora of competing technologies including multicore x86, ARM, and Power-family processors, graphics processing units (GPUs), field-programmable gate arrays (FPGAs), Xeon Phis, DSPs, and TILE architectures, among others. Further complicating matters, each of these architectures often requires its own toolchain to work with. This makes it difficult to not only develop software that works across different devices, but to even determine what hardware would be best suited for a particular application. EM Photonics is currently developing a toolchain to address the first issue, namely the creation of programs that can be written once and targeted to different architectures. In this project, we propose expanding on this technology to allow the developer to work with a wider range of hardware devices as well as provide complementary analysis tools to help determine which hardware devices would be best suited for their applications. We will also create a new Matlab-like frontend for our toolchain to ease the burden on programmers already familiar with that paradigm.; BENEFIT: The tools developed in this project will be complementary to our effort in building tools that make it easier for the programmer to access the wide array of hardware that is now available. The work in this project will result in three specific benefits: 1) Make our cross-platform toolchain accessible to developers accustomed to working in Matlab, 2) In addition to current hardware options such as x86, ARM, GPU, etc., allow users programs to target more exotic architectures such as TILE, and 3) Provide additional tools that help the developer chose which hardware option is best suited for their application. Our target market for this work is developers looking to access the power of the multitude of available processing devices without having to learn and rewrite their software for each one. Instead, we are providing a toolchain that allows a program to be written one time in either a Matlab-based or C/C++-style language, use analysis tools to project how it will perform on various devices, and build binaries for the platforms of interest. These tools will be marketed to the same user community we are approaching with the other development tools we are providing.