A total of 14 key exploitable results generated by the project were identified and they have been uploaded to the Horizon Results Platform organized in the following categories:
- Energy-aware profiling tool: Energy-profiling tool for monitoring energy consumption of HPC systems. It can make runtime decisions in terms of which task to schedule given an objective, either highest performance or lowest power consumption.
- PARALELLWARE Analyzer: This is the first static code analyzer that is designed specifically to boost the performance of C/C++/Fortran HPC codes. It comes along with an open catalog of rules for performance optimization.
- ArgoDSM: A modern page-based distributed shared memory system that offers a novel coherence protocol and a novel hierarchical approach to synchronization based on delegation. This permits to exploit the synergy of the transparent nature of caching and the scheduling flexibility of a task-based system to effectively manage locality in a distributed system toward the exascale transition.
- COMPREX (GASPI/GPI): Communication library built on top of GPI and GaspiCxx. It lowers the impact of the communication overhead by reducing the volume of data that needs to be communicated in the allreduce operation.
- JACC: Just In Time (JIT) OpenACC framework in which pragma directives are replaced with runtime routines, allowing further analysis and optimization possibilities at runtime compared to at compile time for the OpenACC programming model.
- OmpSs: Parallel programming model based on pragma annotations and task decomposition, implementing programming productivity for heterogeneous computing using directives, providing asynchronicity, multi-device and automatic inter-node distribution. EPEEC features include OmpSs@OpenACC, OmpSs@OpenMP, OmpSs@ArgoDSM, and tasking within accelerators.
- ecoHMEM: Software framework for automatic data placement in heterogeneous memory systems. First released by EPEEC.
- System support for data-intensive applications on heterogeneous memory architectures (BWAP/AMBIX): New system libraries for optimizing the data placement of unmodified applications, exploiting the latest memory technologies provided in Intel-based high-end server architectures.
- AVBP+GPU solver: Massively parallel unstructured mesh code with GPU acceleration using OpenACC for CFD software. It scales on latest hardware and accelerator-based exascale machines.
- Quantum Espresso FFT miniapp, OmpSs+GASPI enabled: New communication algorithm based on GASPI over OmpSs which improves on the MPI_Alltoall communication in MPI. Tests have shown under some circumstances it can be faster than standard MPI implementations.
- SMURFF: Tthe only available high-performance Bayesian matrix factorization package. SMURFF combines both user-friendliness and high performance. EPEEC provided the development of SMURFF into a fully-featured package while freeing the developers from the burden of detailed performance engineering improving moreover the overall performance and energy efficiency of the code
- TREEPART Library: Topology-aware massively parallel unstructured mesh partitioning and load-balancing tool and library, first in its kind. It implements an on-the-fly solver agnostic interface available for coupling with existing CFD solvers.
- Maxw3D_DGTD / Simplified DGTD solver extracted from DIOGENeS: Maxw3D_DGTD is a domain application code for nano-optics used to numerically solve the system of time-domain Maxwell equations in three-dimensional space using a high order finite element type method.
- ZPIC miniapp - Asynchronous Tasking with Data Dependencies: This miniapp has been developed for shared-memory Particle-In-Cell codes. It implements a fully asynchronous design based on OmpSs-2 tasking with data-dependencies on a spatial decomposition.