The TEXTAROSSA results are summarized in the above Figure: all the components developed or improved in this project can be sketched as an integrated HW/SW layered architecture.
- Hardware. At the bottom layer we have the two types of accelerated prototype heterogeneous nodes (IDV-A and IDV-E), equipped with last generation NVIDIA GPUs and Xilinx Alveo FPGAs and the two-phase cooling system developed in this project.
- Basic Environment. On the top of the above infrastructure, we have a basic toolchain including standard compilers and libraries for parallel programming as well as run-time supports and tools by vendors which, where it was needed for efficiency and scalability motivations, were directly used by some higher-level software tools and applications.
- Project Toolchain. In the middle layer we put all the new tools proposed in this project, including structured programming environments, application-specific workflow management tools, automatic code generation tools for exploitation of mixed-precision variable storage and computations, tools and libraries for modeling and measurements of energy consumption.
- Applications. Finally, on the top of this infrastructure, we put the applications, which in turn can be classified as general-purpose mathematical libraries, domain-specific codes and mini applications. All the above applications are representatives of different computational needs and were developed as significant benchmarks of some of the components of the above architecture. Their development and testing provided useful feedback for the development and setup of all the basic components of the TEXTAROSSA architecture and significant guidelines, not only for the tools developed in this project, but more generally, for efficient and scalable exploitations of current heterogeneous computing nodes in relevant application domains.
