Luis Gerardo León-Vega

Luis Gerardo León-Vega

Auto-generated profile, based on publicly available information

Luis Gerardo León-Vega

Università degli Studi di Trieste

Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
24 July 2023 in ACM Transactions on Embedded Computing Systems
Verified
Journal Article
Open access
LLLuis G. León-Vega
Luis G. León-Vega
ESEduardo Salazar-Villalobos
Eduardo Salazar-Villalobos
ARAlejandro Rodriguez-Figueroa
Alejandro Rodriguez-Figueroa
JCJorge Castro-Godínez
Jorge Castro-Godínez
Low-power consumption and scarce computational resources limit the computation at the edge. Besides, the approximate computing paradigm reports promising techniques for designing accelerators to deal with inherent limitations of the edge, and high-level synthesis with C++ opens the opportunity to use meta-programming for specialisable generic design. This work proposes a framework for automatically generating synthesis-time configurable processing elements (PEs) for matrix multiplication-addition (GEMMA) and convolution. To evaluate our work, we perform a design exploration after varying data bit-width, operand sizes, and kernel sizes. Our analyses include resource consumption scaling, clocks-to-solution, design efficiency, and error distribution, presenting a comprehensive view of how the parameters affect the properties of our generic implementations. The GEMMA presented a trade-off between granularity vs efficiency , where large PEs with short data widths are favoured by the design efficiency, achieving, theoretically, up to 75 GMAC/s on a Xilinx XC7Z020 @ 100 MHz with an efficiency of 27%. For design efficiency, we propose a figure of merit to evaluate operations per second and resource utilisation with respect to the maximum achievable by the FPGA. Regarding the convolution PEs, we implemented two algorithms: a window-based spatial convolution and Winograd. The former is the best in terms of performance with 150 GMAC/s, reaching up to 47% of efficiency. Winograd also outperformed numerically using a 3× 3 kernel filter, presenting a mean error of 11.01% in 4-bits operands with a PSNR=16.28 dB, compared to the spatial convolution with 38.2% of mean error and PSNR=5.89 dB. Finally, we discuss how the error is mostly dependent on the PE’s parameters. In the GEMMA, the error depends on the matrix size, causing limitations in the PE scaling but still applicable to accelerators. The PEs developed during this research will lead to further granular approximate accelerator research.


29 July 2024
Verified
Conference Paper
Closed access
LLLuis G. Leon-Vega
Luis G. Leon-Vega
EOErick Obregon-Fonseca
Erick Obregon-Fonseca
JCJorge Castro-Godinez
Jorge Castro-Godinez
The introduction of FPGAs in High-Performance Embedded Computing and Artificial Intelligence still faces chal-enges regarding the difficulty of getting started. It requires hardware knowledge, familiarity with multiple tooling, libraries and frameworks and long synthesis times. To encourage the usage of FPGAs, this work proposes an ecosystem that includes a library with a set of pre-built accelerators for common Digital Signal Processing and Artificial Intelligence workloads, an engine for runtime arbitrary-precision quantisation and an agnostic API, allowing the development of FPGA-accelerated user applications while abstracting the details about the FPGA design and implementation. Our approach is based on hardware reuse, introducing software resource management of a series of pre-built IP cores, allowing low-end FPGAs to be used as hardware accelerators and multiple applications to share resources. Our work is better than managed FPGA standalone applications with Vitis HLS-based quantisation, accelerating 1.22 x, thanks to our quantisation engine, which accelerates 5.12 x the quantisation and 13.30 x the de-quantisation, while keeping close the accelerator execution times.


14 February 2025
Verified
Book Chapter
Closed access
PDF file
LLLuis G. León-Vega
Luis G. León-Vega
NTNiccolò Tosato
Niccolò Tosato
SCStefano Cozzini
Stefano Cozzini

18 November 2024
Verified
Conference Paper
Closed access
LLLuis G. León-Vega
Luis G. León-Vega
DÁDiego Ávila-Torres
Diego Ávila-Torres
ILIndra León-Vega
Indra León-Vega
JCJorge Castro-Godínez
Jorge Castro-Godínez
FPGAs for hardware acceleration face several issues in their adoption by the emerging AI embedded software commu-nity, mainly due to the difficulty of getting started in the field. To address these issues, FPGA vendors have tried to simplify and combine workflows to make them user-friendly. This has also impacted the runtime libraries used on Linux to allow user space applications to interact with FPGA programmable logic. However, this approach can result in suboptimal results, given the added overhead in single-purpose applications in an attempt to generalise the runtime libraries. This work presents CYNQ Runtime Library, a simplistic and API-agnostic C/C++ runtime library with the ease of PYNQ and the efficiency of C++ and XRT. In comparison to most popular runtime libraries for AMD FPGAs, our work outperforms PYNQ and XRT with more than 5.34 × speedup in execution latency and more than 1.20 × in large workloads than XRT, which poses a better scenario for high-performance single-purpose applications.