DigitalUM :: Browsing by browse.metadata.contributordepartment "Ingeniería y Tecnología de Computadores"

Browsing by browse.metadata.contributordepartment "Ingeniería y Tecnología de Computadores"

Now showing 1 - 20 of 167

Open Access
A Comparative Analysis of Machine Learning and Deep Learning Approaches for Multiclass Nucleus Classifcation in Histological Images
(Wiley, 2026-01-30) Sánchez-Torres, Antonio Luis; García-Salmerón, Jesús ; González Férez, María Pilar; Bernabé García, Gregorio; García Carrasco, José Manuel; Ingeniería y Tecnología de Computadores; Kalapraveen Bagadi; Facultades de la UMU::Facultad de Informática
Precisely classifying cells in histological images is critical for early cancer diagnosis and tumor assessment. Traditional manual methods are time-consuming and labor-intensive for histopathologists, driving the development of automated approaches using machine learning (ML) and deep learning (DL). Convolutional neural networks (CNNs) and, more recently, vision transformers (ViTs) have demonstrated signifcant potential in addressing the challenges of cell classifcation by leveraging their ability to automatically extract and learn complex features from histological images. In this work, we evaluate multiple classifcation architectures applied to stained histological images to determine their efectiveness in identifying cancerous cells. We compare traditional ML models, which rely on manually extracted features such as shape and texture, against two DL-based classifers: a CNN-based model (ResNet50) and a ViT-based model. To optimize ML models, we apply principal component analysis (PCA) to refne feature selection. Meanwhile, DL models are trained on cropped cell images using two preprocessing strategies: one that includes additional surrounding cellular context and another that uses only the cell pixels. Additionally, we investigate class balancing strategies, including downsampling and oversampling through data augmentation, to mitigate the efects of dataset imbalance. Experimental results highlight the clear advantage of DL models over traditional ML approaches. ResNet50 consistently delivers robust and reliable performance across diferent preprocessing strategies, confrming its efectiveness for histopathological classifcation tasks. Meanwhile, ViTs achieve results that are comparable to those of CNNs while demonstrating a distinct advantage in classifying underrepresented nucleus classes, likely due to their ability to capture long-range dependencies. Furthermore, incorporating the surrounding cellular environment signifcantly improves classifcation accuracy, underscoring the importance of contextual information in distinguishing between diferent types of nuclei.
Open Access
A complexity-effective local delta prefetcher
(Institute of Electrical and Electronics Engineers, 2025-01-31) Navarro-Torres, Agustín; Panda, Biswabandan; Alastruey-Benedé, Jesús; Ibáñez, Pablo; Viñals-Yúfera, Víctor; Ros Bardisa, Alberto; Ingeniería y Tecnología de Computadores
Data prefetching is crucial for performance in modern processors by effectively masking long-latency memory accesses. Over the past decades, numerous data prefetching mechanisms have been proposed, which have continuously reduced the access latency to the memory hierarchy. Several state-of-the-art prefetchers, namely Instruction Pointer Classifier Prefetcher (IPCP) and Berti, target the first-level data cache, and thus, they are able to completely hide the miss latency for timely prefetched cache lines. Berti exploits timely local deltas to achieve high accuracy and performance. This paper extends Berti with a larger evaluation and with extra optimizations on top of the previous conference paper. The result is a complexity-effective version of Berti that outperforms it for a large amount of workloads and simplifies its control logic. The key for those advancements is a simple mechanism for learning timely deltas without the need to track the fetch latency of each cache miss. Our experiments conducted with a wide range of workloads (CVP traces by Qualcomm, SPEC CPU2017, and GAP) show performance improvements by 4.0% over a mainstream stride prefetcher, and by a non-negligible 1.4% over the previously published version of Berti requiring similar storage.
Open Access
A Cost-Effective Entangling Prefetcher for Instructions
(2021-06) Ros, Alberto; Jimborean, Alexandra; Ingeniería y Tecnología de Computadores
Prefetching instructions in the instruction cache is a fundamental technique for designing high-performance computers. There are three key properties to consider when designing an efficient and effective prefetcher: timeliness, coverage, and accuracy. Timeliness is essential, as bringing instructions too early increases the risk of the instructions being evicted from the cache before their use and requesting them too late can lead to the instructions arriving after they are demanded. Coverage is important to reduce the number of instruction cache misses and accuracy to ensure that the prefetcher does not pollute the cache or interacts negatively with the other hardware mechanisms. This paper presents the Entangling Prefetcher for Instructions that entangles instructions to maximize timeliness. The prefetcher works by finding which instruction should trigger the prefetch for a subsequent instruction, accounting for the latency of each cache miss. The prefetcher is carefully adjusted to account for both coverage and accuracy. Our evaluation shows that with 40KB of storage, Entangling can increase performance up to 23%, outperforming state-of-the-art prefetchers.
Open Access
A Data-Centric Algorithmic Pipeline for Enhancing Cardiac MRI Segmentation Using ViTUNeT and Quality-Aware Filtering
(MDPI, 2026-03-06) Salvador de Haro; Jesús Cámara; González Férez, María Pilar; José Manuel García; García Carrasco, José Manuel; Bernabé García, Gregorio; Ingeniería y Tecnología de Computadores; Facultades de la UMU::Facultad de Informática
The performance of deep-learning-based segmentation models is strongly dependent on the quality of the input data, which is frequently heterogeneous or degraded in real-world medical imaging scenarios. This work presents a data-centric algorithmic pipeline designed to improve cardiac MRI segmentation accuracy through systematic image enhancement and automatic slice-quality filtering. The proposed method is formalized as deterministic algorithm that combines image processing and supervised learning components. The approach integrates a contrast- and structure-preserving enhancement stage, based on bilateral filtering and adaptive histogram equalization, with a quality-aware selection algorithm. Slice quality is assessed using anatomical attributes extracted via YOLOv11sbased localization and a supervised classification model trained to identify diagnostically reliable images. When applied to transformer-based segmentation architectures such as ViTUNeT, the pipeline yields consistent improvements across all evaluation metrics without increasing model complexity or training cost. These findings emphasize the importance of algorithmic data curation as an effective strategy for enhancing robustness and stability in deep-learning segmentation pipelines and demonstrate the broader applicability of the proposed approach to computer-vision tasks involving heterogeneous or low-quality image datasets.
Open Access
A Dual-Consistency Cache Coherence Protocol
(IEEE, 2015-05-25) Jimborean, Alexandra; Ros Bardisa, Alberto; Ingeniería y Tecnología de Computadores
Weak memory consistency models can maximize system performance by enabling hardware and compiler optimizations, but increase programming complexity since they do not match programmers’ intuition. The design of an efficient system with an intuitive memory model is an open challenge. This paper proposes SPEL, a dual-consistency cache coherence protocol which simultaneously guarantees the strongest memory consistency model provided by the hardware and yields improvements in both performance and energy consumption. The design of the protocol exploits a compile-time identification of code regions which can be executed under a less restrictive, thus optimized protocol, without harming correctness. Outside these regions, code is executed under a more restrictive protocol which enforces sequential consistency. Compared to a standard directory protocol, we show improvements in performance of 24% and reductions in energy consumption of 32%, on average, for a 64-core chip multiprocessor.
Open Access
A few-shot learning methodology for improving safety in industrial scenarios through universal self-supervised visual features and dense optical flow
(Elsevier, 2024-12) López de Teruel Alcolea, Pedro Enrique; Ruiz García, Alberto; Losada del Olmo, Juan Jesús; Perales Gómez, Ángel Luis; Ingeniería y Tecnología de Computadores
Industrial safety aims to prevent and mitigate workplace accidents and property damage. One common approach to identifying and analyzing potentially risky situations involves the use of static cameras to capture images or videos of facilities and production processes. However, current state-of-the-art deep learning-based solutions require extensive labeled datasets and substantial computational power to detect these dangerous situations. To address these limitations, this paper presents DINOFSAFE, a methodology that combines dense optical flow and the DINOv2 model, a vision transformer that learns universal visual features without supervision. Our methodology demonstrates dual efficacy by both minimizing the manual labeling efforts necessary for model training and ensuring computational efficiency. Optical flow estimates the apparent motion of objects in the input video streams, while the DINOv2 model generates high-dimensional universal representations capturing their visual properties. Using these representations, we train simple linear classifiers to identify moving objects and categorize them. This information aids in identifying and preventing hazardous conditions in industrial settings, such as pedestrians crossing paths with forklifts, forklifts approaching dangerous areas, loads falling from forklifts, and similar situations. We tested our solution on real videos sourced from industrial environments, resulting in promising outcomes. Furthermore, we compiled a comprehensive dataset consisting of approximately 6 500 images, which we have made publicly available for research and development purposes.
Restricted
A first exploration of fine-grain coherence for integrity metadata
(IEEE Computer Society, 2024-11-05) Ekemark, Per; Ros Bardisa, Alberto; Sagonas, Konstantinos; Kaxiras, Stefanos; Ingeniería y Tecnología de Computadores
Memory integrity protection is intended for secure execution, and it is typically associated with programs running on a single core. However, with the emergence of multi-processor systems-on-chip and chiplets, extending memory integrity protection to cache-coherent multiprocessors becomes essential. In this work, we explore for the first time the design space for maintaining coherence in fine-grain integrity metadata at the block level. We discuss various policies for updating the integrity tree using the underlying coherence protocol, and examine how these policies affect coherence traffic. We introduce the concepts of proactive and reactive update initiation, and discuss their implications for data and integrity-tree blocks. We also investigate the trade-offs between eager and lazy update propagation policies, focusing on coherence transactions such as invalidations and downgrades to analyse the pros and cons of different approaches. What we observe is that for some benchmarks the choice between the eager and the lazy update initiation policy does not make much difference, while for many other benchmarks one policy is better over the other, depending on how the benchmark shares its data.
Open Access
A Framework for Dynamic Configuration of TLS Connections Based on Standards
(Springer, 2022-01-20) Pastor-Galindo, Javier; López-Millán, Gabriel; Marín-López, Rafael; Cánovas, Óscar; Pereñíguez García, Fernando; Ingeniería y Tecnología de Computadores
The Transport Layer Security (TLS) protocol is widely used for protecting end-to-end communications between network peers (applications or nodes). However, the administrators usually have to configure parameters (e.g., cryptography algorithms or authentication credentials) to establish TLS connections manually. However, this way of managing security connections becomes infeasible when the number of network peers is high. This paper proposes a TLS management framework that configures and manages TLS connections in a dynamic and autonomous manner. The solution is based on well-known standardized protocols and models that allow providing the necessary configuration parameters to establish a TLS connection between two network nodes. Nowadays, this is required in several application scenarios such as virtual private networks, virtualized network functions, or service function chains. Our framework is based on standard elements of the Software Defined Networking paradigm, widely adopted to provide flexibility to network management, such as for the scenarios aforementioned. The proposed framework has been implemented in a proof of concept to validate the suitability of the proposed solution to manage the dynamic configuration of TLS connections. The experimental results confirm that the implementation of this framework enables an operable and flexible procedure to manage TLS connections between network nodes in different scenarios.
Open Access
A hybrid static-dynamic classification for dual-consistency cache coherence
(IEEE, 2016) Jimborean, Alexandra; Ros Bardisa, Alberto; Ingeniería y Tecnología de Computadores
Traditional cache coherence protocols manage all memory accesses equally and ensure the strongest memory model, namely, sequential consistency. Recent cache coherence protocols based on self-invalidation advocate for the model sequential consistency for data-race-free, which enables powerful optimizations for race-free code. However, for racy code these cache coherence protocols provide sub-optimal performance compared to traditional protocols. This paper proposes SPEL++, a dual-consistency cache coherence protocol that supports two execution modes: a traditional sequential-consistent protocol and a protocol that provides weak consistency (or sequential consistency for data-race-free). SPEL++ exploits a static-dynamic hybrid classification of memory accesses based on (i) a compile-time identification of extended data-race-free code regions for OpenMP applications and (ii) a runtime classification of accesses based on the operating system’s memory page management. By executing racy code under the sequential-consistent protocol and race-free code under the cache coherence protocol that provides sequential consistency for data-race-free, the end result is an efficient execution of the applications while still providing sequential consistency. Compared to a traditional protocol, we show improvements in performance from 19% to 38% and reductions in energy consumption from 47% to 53%, on average for different benchmark suites, on a 64-core chip multiprocessor
Open Access
A methodology for evaluating the robustness of anomaly detectors to adversarial attacks in industrial scenarios
(IEEE, 2022-11-28) Fernández Maimó, Lorenzo; García Clemente, Félix J.; Maroto Morales, Alejandro; Huertas Celdrán, Alberto; Bovet, Gérôme; Huertas Celdrán, Alberto; Perales Gómez, Ángel Luis; Ingeniería y Tecnología de Computadores
Anomaly Detection systems based on Machine and Deep learning are the most promising solutions to detect cyberattacks in the industry. However, these techniques are vulnerable to adversarial attacks that downgrade prediction performance. Several techniques have been proposed to measure the robustness of Anomaly Detection in the literature. However, they do not consider that, although a small perturbation in an anomalous sample belonging to an attack, i.e., Denial of Service, could cause it to be misclassified as normal while retaining its ability to damage, an excessive perturbation might also transform it into a truly normal sample, with no real impact on the industrial system. This paper presents a methodology to calculate the robustness of Anomaly Detection models in industrial scenarios. The methodology comprises four steps and uses a set of additional models called support models to determine if an adversarial sample remains anomalous. We carried out the validation using the Tennessee Eastman process, a simulated testbed of a chemical process. In such a scenario, we applied the methodology to both a Long-Short Term Memory (LSTM) neural network and 1-dimensional Convolutional Neural Network (1D-CNN) focused on detecting anomalies produced by different cyberattacks. The experiments showed that 1D-CNN is significantly more robust than LSTM for our testbed. Specifically, a perturbation of 60% (empirical robustness of 0.6) of the original sample is needed to generate adversarial samples for LSTM, whereas in 1D-CNN the perturbation required increases up to 111% (empirical robustness of 1.11).
Open Access
A Model-Driven Approach to Generate Schemas for Object-Document Mappers
(IEEE Access, 2019) Hernández Chillón, Alberto; Sevilla Ruiz, Diego; García Molina, Jesús J.; Feliciano Morales, Severino; Ingeniería y Tecnología de Computadores
Many actual NoSQL systems are schemaless, that is, the structure of the data is not defined beforehand in any schema, but it is implicit in the data itself. This characteristic is very convenient when the data structure suffers frequent changes. However, the agility and flexibility achieved is at the cost of losing some important benefits, such as 1) assuring that the data stored and retrieved fits the database schema; 2) some database utilities require to know the schema, and; 3) schema visualization helps developers to write better code. In previous work, we proposed a model-based reverse engineering approach to infer schema models from NoSQL data. Model-driven engineering (MDE) techniques can be used to take advantage of extracted models with different purposes, such as schema visualization or automatic code generation. Here, in this paper, we present an MDE solution to automate the usage of Object-NoSQL mappers when the database already exists. We will focus on mappers that are available for document systems (Object-Document mappers, ODMs), but the proposed approach is mapper-independent. These mappers are emerging to provide similar functionality to Object-Relational mappers: they are in charge of the mapping of objects into NoSQL data (documents in the case of ODMs) for object-oriented applications. We show how schemas and other artifacts (e.g. validators and indexes) for ODMs can be automatically generated from inferred schemas. The solution consists of a two-step model transformation chain, where an intermediate model is generated to ease the code generation. We have applied our approach for two popular ODMs: Mongoose and Morphia and validated it with the StackOverflow dataset.
Open Access
A Real Time Cardiomyopathy Detection Tool Using Ml Ensemble Models
(Wiley, Institution of Engineering and Technology (IET), 2025-07-29) de Haro Orenes, Salvador; Becerra Fernández, Estebán; González-Férez, Pilar; García Carrasco, José Manuel; Bernabé García, Gregorio; Ingeniería y Tecnología de Computadores
Open Access
A self-adaptive deep learning-based system for anomaly detection in 5G networks
(IEEE, 2018-02-07) Fernández Maimó, Lorenzo; García Clemente, Félix J.; Gil Pérez, Manuel; Martínez Pérez, Gregorio; Perales Gómez, Ángel Luis; Ingeniería y Tecnología de Computadores
The upcoming fifth-generation (5G) mobile technology, which includes advanced communication features, is posing new challenges on cybersecurity defense systems. Although innovative approaches have evolved in the last few years, 5G will make existing intrusion detection and defense procedures become obsolete, in case they are not adapted accordingly. In this sense, this paper proposes a novel 5G-oriented cyberdefense architecture to identify cyberthreats in 5G mobile networks efficient and quickly enough. For this, our architecture uses deep learning techniques to analyze network traffic by extracting features from network flows. Moreover, our proposal allows adapting, automatically, the configuration of the cyberdefense architecture in order to manage traffic fluctuation, aiming both to optimize the computing resources needed in each particular moment and to fine tune the behavior and the performance of analysis and detection processes. Experiments using a well-known botnet data set depict how a neural network model reaches a sufficient classification accuracy in our anomaly detection system. Extended experiments using diverse deep learning solutions analyze and determine their suitability and performance for different network traffic loads. The experimental results show how our architecture can self-adapt the anomaly detection system based on the volume of network flows gathered from 5G subscribers' user equipments in real-time and optimizing the resource consumption.
Open Access
A Trusted Approach for Decentralised and Privacy-Preserving Identity Management
(2021) Torres, R.; Garcia Rodriguez, J.; Bernal Bernabé, Jorge; Skarmeta Gomez, A. F.; Ingeniería y Tecnología de Computadores
Identity Management (IdM) systems have traditionally relied on a centralized model prone to privacy, trust, and security problems, like potential massive data breaches or identity spoofing. Identity providers accumulate excessive power that might allow them to become a big brother, analyzing and storing as much data as possible. Users should be able to trust identity providers and manage their personal information straightforwardly without compromising their privacy. The European OLYMPUS project introduces a distributed approach for IdM based on enhanced Attribute-Based Credentials (ABC) that splits the role of Identity Provider to limit their influence and chances to become a unique point of failure. However, the trust relationship between service providers, users, and identity providers is still a gap in those kinds of privacy-preserving ABC systems. Decentralized technologies are an opportunity to break away from the centralized model and propose systems that respect privacy while increasing users’ trust. This paper presents an evolution of the OLYMPUS architecture, maintaining all the privacy features and incorporating distributed ledger technologies to enhance trust and security in online transactions and IdM systems. The proposed system has been implemented, tested, and validated, showing its performance and feasibility to manage user’s identity in a fully privacy-preserving, distributed and reliable way.
Open Access
A unified metamodel for NoSQL and relational databases
(Elsevier, 2021-09-25) Fernández Candel, Carlos; Sevilla Ruiz, Diego; García Molina, Jesús J.; Ingeniería y Tecnología de Computadores
The Database field is undergoing significant changes. Although relational systems are still predominant, the interest in NoSQL systems is continuously increasing. In this scenario, polyglot persistence is envisioned as the database architecture to be prevalent in the future. Therefore, database tools and systems are evolving to support several data models. Multi-model database tools normally use a generic or unified metamodel to represent schemas of the data model that they support. Such metamodels facilitate developing database utilities, as they can be built on a common representation. Also, the number of mappings required to migrate databases from a data model to another is reduced, and integrability is favored. In this paper, we present the U-Schema unified metamodel able to represent logical schemas for the four most popular NoSQL paradigms (columnar, document, key–value, and graph) as well as relational schemas. We will formally define the mappings between U-Schema and the data model defined for each database paradigm. How these mappings have been implemented and validated will be discussed, and some applications of U-Schema will be shown. To achieve flexibility to respond to data changes, most of NoSQL systems are “schema-on-read,” and the declaration of schemas is not required. Such an absence of schema declaration makes structural variability possible, i.e., stored data of the same entity type can have different structure. Moreover, data relationships supported by each data model are different; For example, document stores have aggregate objects but not relationship types, whereas graph stores offer the opposite. Through the paper, we will show how all these issues have been tackled in our approach. As far as we know, no proposal exists in the literature of a unified metamodel for relational and the NoSQL paradigms which describes how each individual data model is integrated and mapped. Our metamodel goes beyond the existing proposals by distinguishing entity types and relationship types, representing aggregation and reference relationships, and including the notion of structural variability. Our contributions also include developing schema extraction strategies for schemaless systems of each NoSQL data model, and tackling performance and scalability in the implementation for each store.
Open Access
A ViTUNeT-based model using YOLOv8 for efficient LVNC diagnosis and automatic cleaning of dataset
(De Gruyter, 2025-06-04) Haro Orenes, Salvador de; Bernabé García, Gregorio; García Carrasco, José Manuel; González Férez, Pilar; Ingeniería y Tecnología de Computadores
Left ventricular non-compaction is a cardiac condition marked by excessive trabeculae in the left ventricle’s inner wall. Although various methods exist to measure these structures, the medical community still lacks consensus on the best approach. Previously, we developed DL-LVTQ, a tool based on a UNet neural network, to quantify trabeculae in this region. In this study, we expand the dataset to include new patients with Titin cardiomyopathy and healthy individuals with fewer trabeculae, requiring retraining of our models to enhance predictions. We also propose ViTUNeT, a neural network architecture combining U-Net and Vision Transformers to segment the left ventricle more accurately. Additionally, we train a YOLOv8 model to detect the ventricle and integrate it with ViTUNeT model to focus on the region of interest. Results from ViTUNet and YOLOv8 are similar to DL-LVTQ, suggesting dataset quality limits further accuracy improvements. To test this, we analyze MRI images and develop a method using two YOLOv8 models to identify and remove problematic images, leading to better results. Combining YOLOv8 with deep learning networks offers a promising approach for improving cardiac image analysis and segmentation.
Open Access
Accelerating finite field arithmetic for homomorphic encryption on GPUs
(IEEE Computer Society, 2023-03-14) Livesay, Neal; Jonatan, Gilbert; Mora, Evelio; Shivdikar, Kaustubh; Agrawal, Rashmi; Joshi, Ajay; Kim, John; Kaeli, David; Abellán Miguel, José Luis; Ingeniería y Tecnología de Computadores; Facultad de Informática
Fully homomorphic encryption (FHE) is a rapidly developing technology that enables computation directly on encrypted data, making it a compelling solution for security in cloud-based systems. In addition, modern FHE schemes are believed to be resistant to quantum attacks. Although FHE offers unprecedented potential for security, current implementations suffer from prohibitively high latency. Finite field arithmetic operations, particularly the multiplication of high-degree polynomials, are key computational bottlenecks. The parallel processing capabilities provided by modern GPUs make them compelling candidates to target these highly parallelizable workloads. In this article, we discuss methods to accelerate polynomial multiplication with GPUs, with the goal of making FHE practical.
Open Access
ACTA: Automatic Configuration of the Tensor Memory Accelerator for High-End GPUs
(Association for Computing Machinery (ACM), 2025-03) Meseguer, Nicolás; Sun, Yifan; Pellauer, Michael; Abellán, José L.; Acacio, Manuel E.; Ingeniería y Tecnología de Computadores
Achieving peak GPU performance requires optimizing data locality and asynchronous execution to minimize memory access costs and overlap computation with transfers. While features like the Tensor Memory Accelerator (TMA) and warp specialization address these challenges, their complexity often limits programmers. In this work, we present ACTA (Automatic Configuration of the Tensor Memory Accelerator), a software library that simplifies and optimizes TMA usage. By leveraging the GPU Specification Table (GST), ACTA dynamically determines the optimal tile sizes and queue configurations for each kernel and architecture. Its algo- rithm ensures efficient overlap between memory and computation, drastically reducing programming complexity and eliminating the need for exhaustive design space exploration. Our evaluation across a diverse set of GPU kernels demonstrates that ACTA achieves performance within 2.78% of exhaustive tun-ing while requiring only a single configuration pass. This makes ACTA a practical and efficient solution for optimizing modern GPU workloads, combining near-optimal performance with significantly reduced programming effort.
Restricted
Alternate path µ-op cache prefetching
(IEEE Computer Society, 2024-08-01) Singh, Sawan; Perais, Arthur; Jimborean, Alexandra; Ros Bardisa, Alberto; Ingeniería y Tecnología de Computadores
Datacenter applications are well-known for their large code footprints. This has caused frontend design to evolve by implementing decoupled fetching and large prediction structures – branch predictors, Branch Target Buffers (BTBs) – to mitigate the stagnating size of the instruction cache by prefetching instructions well in advance. In addition, many designs feature a micro operation (µ-op) cache, which primarily provides power savings by bypassing the instruction cache and decoders once warmed up. However, this µ-op cache often has lower reach than the instruction cache, and it is not filled up speculatively using the decoupled fetcher. As a result, the µ-op cache is often over-subscribed by datacenter applications, up to the point of becoming a burden. This paper first shows that because of this pressure, blindly prefetching into the µ-op cache using state-of-the-art standalone prefetchers would not provide significant gains. As a consequence, this paper proposes to prefetch only critical µ-ops into the µop cache, by focusing on execution points where the µ-op cache provides the most gains: Pipeline refills. Concretely, we use hardto-predict conditional branches as indicators that a pipeline refill is likely to happen in the near future, and prefetch into the µ-op cache the µ-ops that belong to the path opposed to the predicted path, which we call alternate path. Identifying hard-to-predict branches requires no additional state if the branch predictor confidence is used to classify branches. Including extra alternate branch predictors with limited budget (8.95KB to 12.95KB), our proposal provides average speedups of 1.9% to 2% and as high as 12% on a subset of CVP-1 traces.
Open Access
An autotuning approach to select the inter-GPU communication library on heterogeneous systems
(Springer, 2024-12-12) Cámara, Jesús; Cuenca Muñoz, Antonio Javier; Cuenca, Javier; Boratto, Murilo; Vicente Jaén, Arturo; Galindo Garre, Víctor; Ingeniería y Tecnología de Computadores
In this work, an automatic optimisation approach for parallel routines on multi-GPU systems is presented. Several inter-GPU communication libraries (such as CUDA- Aware MPI or NCCL) are used with a set of routines to perform the numerical oper- ations among the GPUs located on the compute nodes. The main objective is the selection of the most appropriate communication library, the number of GPUs to be used and the workload to be distributed among them in order to reduce the cost of data movements, which represent a large percentage of the total execution time. To this end, a hierarchical modelling of the execution time of each routine to be opti- mised is proposed, combining experimental and theoretical approaches. The results show that near-optimal decisions are taken in all the scenarios analysed.