25 November 2019 in IEEE Access
This paper presents kNN STreaming Unit For Fpgas (kNN-STUFF), a modular, scalable and efficient Hardware/Software implementation of k-Nearest Neighbors (kNN) classifier targeting System on Chip (SoC) devices. It takes advantage of custom accelerators, implemented on the reconfigurable fabric of the SoC device, to perform most of the classifier’s workload, whereas the processor coordinates the accelerators and runs the remaining workload of the kNN algorithm. kNN-STUFF offers a highly flexible framework, where the designer has the possibility to define the number of parallel instances of the classifier and the parallelism within each instance. This capability allows creating the most suitable implementation for a target device of any size. Results show that kNN-STUFF, with 24 accelerators, attains performance improvements up to 67.4×, when compared to an optimized (-O3) software-only implementation of the kNN running on a single core of the ARM Cortex-A9 CPU. Furthermore, its energy efficiency improvements are as high as 50.6×.
1 November 2023 in IEEE Computer Architecture Letters
Attaining the performance and efficiency levels required by modern applications often requires the use of application-specific accelerators. However, writing synthesizable Register-Transfer Level (RTL) code for such accelerators is a complex, expensive, and time-consuming process, which is cumbersome for early architecture development phases. To tackle this issue, a pre-synthesis simulation toolchain is herein proposed that facilitates the early architectural evaluation of complex accelerators aggregated to multi-level memory hierarchies. To demonstrate its usefulness, the proposed toolchain is used to model a tensor accelerator based on Gemmini, showing that gem5-accel can successfully anticipate the results of complex hardware accelerators executing deep Neural Network (NN) models.
11 January 2024 in IEEE Access
The accurate simulation and performance assessment of Near-Data Accelerators (NDAccs) is a complex challenge as it must consider the operation of the entire processing system, the impact of the Operating System (OS) overheads, and the memory contention caused by concurrent processes. While recent proposals have attempted to repurpose and extend existing tools, the offered support for the development and evaluation of NDAccs is limited and full-system simulation is rarely provided. To mitigate this problem, the NDPmulator simulation framework, based on the widely established gem5 architectural simulator, is herein proposed. NDPmulator provides System Emulation (SE) and Full System (FS) support for the development and evaluation of novel NDAccs deployed at multiple levels of the memory hierarchy. To demonstrate its versatility and performance-efficiency, the proposed NDPmulator is used to model three existing NDAccs, showing that it can accurately estimate and anticipate the results of the evaluation performed by the original authors while requiring a significantly smaller implementation effort and a fraction of the simulation time. Furthermore, NDPmulator offers the possibility to conduct complex experiments where the NDAcc is coupled to a real system featuring an OS. Hence it allows modeling all overheads related to the NDAcc device driver, the OS, and the contention caused by concurrent and background processes.
12 April 2021 in Journal of Signal Processing Systems