RL1) “Topology-aware optimization of MPI collective communication primitives”. This RL was the main topic of my Juan de la Cierva grant. The most novel and relevant contribution is the application of pipelining to collective communication primitives in MPI, as demonstrated in the journal paper "Accelerating distributed deep neural network training with pipelined MPI Allreduce". This RL was directly bound to the WP3 from the EU H2020 RED-SEA Distributed Cluster Network Topology Research and Optimization Project. I continue the project in collaboration with the AccelCom research group at BSC, where a novel offloading MPI technique is being developed.
RL2) “HPC numerical libraries for DNNs on low-power processors”. This RL addresses the lack of specialized scientific libraries for low-power devices. The work done in this RL has resulted in implementing matrix multiplication and convolution operators, which are key computational elements in DNNs, for ARM and MCUs boards. This RL is aligned with the PID2023-146569NB-C21 project. Collaborations with Universidad de Córdoba, Carnegie Mellon University, and The University of Texas at Austin,
RL3) “Code auto-generation for high-performance computing.” This RL is aligned with my APOSTD fellowship and my PAID-06-2024 from the UPV project. It includes the analysis, design, and implementation of automatic solutions for high performance computing libraries with Exo or Apache TVM. This RL has established collaborations with UCM, Berkeley’s Computer Science Division, Massachusetts Institute of Technology, the University of Glasgow, the University of Washington, and Photonicsense Inc.
“Integración y producción de apiCAM para el mercado de dispositivos móviles (INTECAM)”. (INNTA3/2023/12) PI: Enrique S. Quintana-Ortí.
“Inteligencia Sostenible en el Borde-UPV”. (PID2023-146569NB-C21) Agencia Estatal de Investigación. From 1/09/2024 to 31/08/2027 PI: E.S. Quintana-Ortí.
“Digital Autonomy for RISC-V in Europe (Specific Grant Agreement 1) (DARE SGA 1)” (EU HORIZON-EUROHPC-JU-2024-DARE-SGA-04 101202459). European Union and AEI. From 1/01/2025 to 31/12/2027. PI: M. Valero (BSC).
“Reducción de Consumo de Energía en Dispositivos IoT mediante la Generación Automática de Códigos Optimizados para Redes Neuronales” (PAID-06-24) UPV From 01/01/25 to 31/12/25 PI: A. Castelló.
EU H2020-JTI-EuroHPC-2019-1 955776 "RED-SEA Distributed Cluster Network Topology Research and Optimization Project". From 1/04/2021 to 1/03/2024. European Union PI E. Walker (ATOS)
PID2020-113656RB-C22 "Computacion y Comunicaciones de Altas Prestaciones Conscientes del Consumo Energetico. Aplicaciones al Aprendizaje Profundo Computacional - Upv" From: 01/09/21 to 30/04/23. Agencia Estatal de Investigación. PIs Enrique S. Quintana-Ortí and Alonso-Jordá, Pedro.
TED2021-129334B-I00 Aceleración de algoritmos de aprendizaje automático en microcontroladores y microprocesadores de consumo ultra-reducido”. From: 1/12/2022 to 30/11/2024. Agencia Estatal de Investigación. PI: E.S. Quintana-Ortí.
PyDTNN - Python Distributed Training of Neural Networks Develop and maintain the GPU version of the framework
European Distributed Deep Learning (EDDL) library. A general-purpose library initially developed to cover deep learning needs in healthcare use cases within the DeepHealth project. Responsibles: Jon Ander and Roberto Paredes. Dates: January-April 2021. Main tasks: cuDNN development.
H2020-FETPROACT- 2016-2017 732631 "OPRECOMP - Open Transprecision Computing Analyzing and developing libraries and tools for reduced precision computations (UJI)
PROYECTO UJI-B2017-46 "Desarrollo de una librería paralela para el manejo de matrices jerárquicas sobre procesadores multinúcleo y aceleradores."
Unification of Lightweight Thread Solutions and their Application in High-Performance Programming Models. PhD Dissertation advised by Rafael Mayo Gual (UJI) and Antonio J. Peña (BSC) and supported by Generalitat Valenciana Vali+D 2015 program. 2015-2018.
BOLT: A Lightning-Fast OpenMP Implementation. OpenMP over Argobots implementation leaded by Sangmin Seo and Pavan Balaji (ANL). Dates: June 2016 - 2019. Main tasks: Development of OpenMP standard over Argobots.
rCUDA development HPC&A research group at Universitat Jaume I de Castelló, Spain, Responsible: Enrique S. Quintana-Ortí. Dates: Sep 2011 - Jul 2015. Main tasks: rCUDA technology development (www.hpca.uji.es/rCUDA)
H2020-FETHPC-2014 671602 “INTERTWinE. Programming Model Interoperability Towards Exascale” Responsible: Enrique S. Quintana-Ortí. Dates: Sep 2015 Dec 2015 and May 2018 - Sep 2018
Grups d’Investigació d’Alt Rendiment Segons el PPF Grup 201. HPC&A research group at Universitat Jaume I de Castelló, Spain, Responsible: Enrique S. Quintana-Ortí. Dates: Sep - Nov 2011, Aug - Oct 2012 and Oct - Dec 2015. Main tasks: rCUDA technology development (www.hpca.uji.es/rCUDA) and Lightweight threads libraries analysis.
FP7 318793. EXA2GREEN: Energy-Aware Sustainable Computing on Future Technology - Paving the Road to Exascale Computing. HPC&A research group at Universitat Jaume I de Castelló, Spain, Responsible: Enrique S. Quintana-Ortí, Dates: May - Jul 2014, Jan - May 2015 and Sep - Oct 2015. Main tasks: Study the use of virtualized remote accelerators for energy savings in computation based on GPU accelerators.
Desarrollo de Librerías para Gestionar el Acceso d Dispositivos Remotos Compartidos en Servidores de Altas Prestaciones. GAP research group at Universitat Politècnica de València, Spain, Responsible: José Duato, Dates: Oct 2013 - Jan 2014. Main tasks: rCUDA technology libraries development (www.hpca.uji.es/rCUDA).
Power-Aware HPC. HPC&A research group at Universitat Jaume I de Castelló, Spain, Responsible: Enrique S. Quintana-Ortí, Dates: Apr - Sep 2013. Main tasks: Adapting Remote GPU Virtualization Framework to ARM processors
Conveni de Col·laboració entre la Universitat Jaume I i Mellanox Technologies INC. HPC&A research group at Universitat Jaume I de Castelló, Spain, Responsible: Enrique S. Quintana-Ortí, Dates: Jan - Mar 2013 and Feb - Apr 2014. Main tasks: rCUDA technology development (www.hpca.uji.es/rCUDA).
Real Data Center Cloud Services and Environment. HPC&A research group at Universitat Jaume I de Castelló, Spain, Responsible: Rafael Mayo Gual, Dates: Nov - Dec 2012. Main tasks: Analyze the Use of Remote GPUs in Cloud Computing Environments.
MONICA. Sistema de Monitorización y Control con Gestión Inteligente de la Eficiencia Energética para Centros de Procesos de Datos Ultradensos Orientados a HPC y Cloud Computing. HPC&A research group at Universitat Jaume I de Castelló, Spain, Responsible: Rafael Mayo Gual, Dates: Apr - Oct 2012. Main tasks: Explore the Viability of the Use of Remote GPUs in Cloud Computing Environments.
rOpenCL. Sistema de Acceso Remoto a GPUs para Cálculo de Propósito General. HPC&A research group at Universitat Jaume I de Castelló, Spain, Responsible: Rafael Mayo Gual, Dates: Jan - Mar 2012. Main tasks: Explore the Viability of the OpenCL Code Execution using Remote GPUs.