arXiv subjects

Marc Casas

Publications and source records attributed to Marc Casas.

14 recordsLinked to original sources

arXiv2026

Enhancing Instruction Prefetching via Cache and TLB Management

Modern server workloads exhibit massive instruction footprints that heavily pressure the processor front-end, making L1 instruction (L1I) prefetching critical for sustaining performance. However, this paper shows that current L1I prefetchers fail to reach their full potential due to two key limitations. First, L1I prefetches crossing page boundaries require address translation before issuance, and translation latency reduces prefetch timeliness. Second, the reuse behavior of code lines fetched by L1I prefetches is highly heterogeneous: while some lines are reused many times, others are dead-on-arrival. This paper introduces Instruction Prefetch-Centric Cache and TLB Management (IP-CaT), the first microarchitectural framework jointly optimizing TLB and cache management for L1I prefetching. IP-CaT consists of two components: (i) the translation Prefetch Buffer (tPB), a small structure colocated with the second-level TLB (sTLB) that stores page table entries fetched by page-crossing L1I prefetches, reducing translation overheads; and (ii) the Trimodal Instruction Prefetch Replacement Policy (TIPRP), a decision-tree-based L2 cache replacement policy specialized for lines fetched by L1I prefetches. We evaluate IP-CaT with three state-of-the-art L1I prefetchers: EPI, FNL+MMA, and Barca. Across 105 contemporary server workloads, IP-CaT consistently improves performance. For example, IP-CaT+EPI achieves an 8.7% geomean speedup over EPI alone. We further show that IP-CaT outperforms state-of-the-art instruction TLB prefetching, advanced TLB replacement (CHiRP), and state-of-the-art code-aware, prefetch-aware, and general-purpose cache replacement policies, including Emissary, SHiP++, and Mockingjay.

cs.AR↗

arXiv2025

A Flexible Instruction Set Architecture for Efficient GEMMs

GEneral Matrix Multiplications (GEMMs) are recurrent in high-performance computing and deep learning workloads. Typically, high-end CPUs accelerate GEMM workloads with Single-Instruction Multiple Data (SIMD) or vector Instruction Set Architectures (ISAs). Since these ISAs face significant issues when running GEMM workloads, particularly when dealing with small, tall, or skinny matrices, matrix ISAs have been proposed and implemented by major hardware vendors in the last years. Although these matrix ISAs deliver larger throughput when running GEMMs than their SIMD/vector counterparts, they are rigid solutions unable to dynamically adapt themselves to application-specific aspects like the data format. This paper demonstrates that the state-of-the-art matrix ISAs deliver suboptimal performance when running the most commonly used convolution and transformer models. This paper proposes the Matrix Tile Extension (MTE), the first matrix ISA that completely decouples the instruction set architecture from the microarchitecture and seamlessly interacts with existing vector ISAs. MTE incurs minimal implementation overhead since it only requires a few additional instructions and a 64-bit Control Status Register (CSR) to keep its state. Specifically, MTE can i) vectorize GEMMs across the three dimensions M, N, and K; ii) leverage the capacity of the existing vector register file; and iii) decouple the tile shape from the underlying microarchitecture. MTE achieves speed-ups of 1.35x over the best state-of-the-art matrix ISA.

cs.AR↗

arXiv2024

An Open-Source Framework for Efficient Numerically-Tailored Computations

We present a versatile open-source framework designed to facilitate efficient, numerically-tailored Matrix-Matrix Multiplications (MMMs). The framework offers two primary contributions: first, a fine-tuned, automated pipeline for arithmetic datapath generation, enabling highly customizable systolic MMM kernels; second, seamless integration of the generated kernels into user code, irrespective of the programming language employed, without necessitating modifications. The framework demonstrates a systematic enhancement in accuracy per energy cost across diverse High Performance Computing (HPC) workloads displaying a variety of numerical requirements, such as Artificial Intelligence (AI) inference and Sea Surface Height (SSH) computation. For AI inference, we consider a set of state-of-the-art neural network models, namely ResNet18, ResNet34, ResNet50, DenseNet121, DenseNet161, DenseNet169, and VGG11, in conjunction with two datasets, two computer formats, and 27 distinct intermediate arithmetic datapaths. Our approach consistently reduces energy consumption across all cases, with a notable example being the reduction by factors of $3.3\times$ for IEEE754-32 and $1.4\times$ for Bfloat16 during ImageNet inference with ResNet50. This is accomplished while maintaining accuracies of $82.3\%$ and $86\%$, comparable to those achieved with conventional Floating-Point Units (FPUs). In the context of SSH computation, our method achieves fully-reproducible results using double-precision words, surpassing the accuracy of conventional double- and quad-precision arithmetic in FPUs. Our approach enhances SSH computation accuracy by a minimum of $5\times$ and $27\times$ compared to IEEE754-64 and IEEE754-128, respectively, resulting in $5.6\times$ and $15.1\times$ improvements in accuracy per power cost.

cs.MS↗

arXiv2024

A Two Level Neural Approach Combining Off-Chip Prediction with Adaptive Prefetch Filtering

To alleviate the performance and energy overheads of contemporary applications with large data footprints, we propose the Two Level Perceptron (TLP) predictor, a neural mechanism that effectively combines predicting whether an access will be off-chip with adaptive prefetch filtering at the first-level data cache (L1D). TLP is composed of two connected microarchitectural perceptron predictors, named First Level Predictor (FLP) and Second Level Predictor (SLP). FLP performs accurate off-chip prediction by using several program features based on virtual addresses and a novel selective delay component. The novelty of SLP relies on leveraging off-chip prediction to drive L1D prefetch filtering by using physical addresses and the FLP prediction as features. TLP constitutes the first hardware proposal targeting both off-chip prediction and prefetch filtering using a multi-level perceptron hardware approach. TLP only requires 7KB of storage. To demonstrate the benefits of TLP we compare its performance with state-of-the-art approaches using off-chip prediction and prefetch filtering on a wide range of single-core and multi-core workloads. Our experiments show that TLP reduces the average DRAM transactions by 30.7% and 17.7%, as compared to a baseline using state-of-the-art cache prefetchers but no off-chip prediction mechanism, across the single-core and multi-core workloads, respectively, while recent work significantly increases DRAM transactions. As a result, TLP achieves geometric mean performance speedups of 6.2% and 11.8% across single-core and multi-core workloads, respectively. In addition, our evaluation demonstrates that TLP is effective independently of the L1D prefetching logic.

cs.AR↗

arXiv2023

Compressed Real Numbers for AI: a case-study using a RISC-V CPU

As recently demonstrated, Deep Neural Networks (DNN), usually trained using single precision IEEE 754 floating point numbers (binary32), can also work using lower precision. Therefore, 16-bit and 8-bit compressed format have attracted considerable attention. In this paper, we focused on two families of formats that have already achieved interesting results in compressing binary32 numbers in machine learning applications, without sensible degradation of the accuracy: bfloat and posit. Even if 16-bit and 8-bit bfloat/posit are routinely used for reducing the storage of the weights/biases of trained DNNs, the inference still often happens on the 32-bit FPU of the CPU (especially if GPUs are not available). In this paper we propose a way to decompress a tensor of bfloat/posits just before computations, i.e., after the compressed operands have been loaded within the vector registers of a vector capable CPU, in order to save bandwidth usage and increase cache efficiency. Finally, we show the architectural parameters and considerations under which this solution is advantageous with respect to the uncompressed one.

cs.LG↗

arXiv2023

Optimization of SpGEMM with Risc-V vector instructions

The Sparse GEneral Matrix-Matrix multiplication (SpGEMM) $C = A \times B$ is a fundamental routine extensively used in domains like machine learning or graph analytics. Despite its relevance, the efficient execution of SpGEMM on vector architectures is a relatively unexplored topic. The most recent algorithm to run SpGEMM on these architectures is based on the SParse Accumulator (SPA) approach, and it is relatively efficient for sparse matrices featuring several tens of non-zero coefficients per column as it computes C columns one by one. However, when dealing with matrices containing just a few non-zero coefficients per column, the state-of-the-art algorithm is not able to fully exploit long vector architectures when computing the SpGEMM kernel. To overcome this issue we propose the SPA paRallel with Sorting (SPARS) algorithm, which computes in parallel several C columns among other optimizations, and the HASH algorithm, which uses dynamically sized hash tables to store intermediate output values. To combine the efficiency of SPA for relatively dense matrix blocks with the high performance that SPARS and HASH deliver for very sparse matrix blocks we propose H-SPA(t) and H-HASH(t), which dynamically switch between different algorithms. H-SPA(t) and H-HASH(t) obtain 1.24$\times$ and 1.57$\times$ average speed-ups with respect to SPA respectively, over a set of 40 sparse matrices obtained from the SuiteSparse Matrix Collection. For the 22 most sparse matrices, H-SPA(t) and H-HASH(t) deliver 1.42$\times$ and 1.99$\times$ average speed-ups respectively.

cs.DC↗

arXiv2023

Open-Source GEMM Hardware Kernels Generator: Toward Numerically-Tailored Computations

Many scientific computing problems can be reduced to Matrix-Matrix Multiplications (MMM), making the General Matrix Multiply (GEMM) kernels in the Basic Linear Algebra Subroutine (BLAS) of interest to the high-performance computing community. However, these workloads have a wide range of numerical requirements. Ill-conditioned linear systems require high-precision arithmetic to ensure correct and reproducible results. In contrast, emerging workloads such as deep neural networks, which can have millions up to billions of parameters, have shown resilience to arithmetic tinkering and precision lowering.

cs.AR↗

arXiv2023

Characterizing the impact of last-level cache replacement policies on big-data workloads

In recent years, graph-processing has become an essential class of workloads with applications in a rapidly growing number of fields. Graph-processing typically uses large input sets, often in multi-gigabyte scale, and data-dependent graph traversal methods exhibiting irregular memory access patterns. Recent work demonstrates that, due to the highly irregular memory access patterns of data-dependent graph traversals, state-of-the-art graph-processing workloads spend up to 80 % of the total execution time waiting for memory accesses to be served by the DRAM. The vast disparity between the Last Level Cache (LLC) and main memory latencies is a problem that has been addressed for years in computer architecture. One of the prevailing approaches when it comes to mitigating this performance gap between modern CPUs and DRAM is cache replacement policies. In this work, we characterize the challenges drawn by graph-processing workloads and evaluate the most relevant cache replacement policies.

cs.AR↗

arXiv2022

Optimization of the Sparse Multi-Threaded Cholesky Factorization for A64FX

Sparse linear algebra routines are fundamental building blocks of a large variety of scientific applications. Direct solvers, which are methods for solving linear systems via the factorization of matrices into products of triangular matrices, are commonly used in many contexts. The Cholesky factorization is the fastest direct method for symmetric and definite positive matrices. This paper presents selective nesting, a method to determine the optimal task granularity for the parallel Cholesky factorization based on the structure of sparse matrices. We propose the OPT-D-COST algorithm, which automatically and dynamically applies selective nesting. OPT-D-COST leverages matrix sparsity to drive complex task-based parallel workloads in the context of direct solvers. We run an extensive evaluation campaign considering a heterogeneous set of 60 sparse matrices and a parallel machine featuring the A64FX processor. OPT-D-COST delivers an average performance speedup of 1.46$\times$ with respect to the best state-of-the-art parallel method to run direct solvers.

cs.DC↗

arXiv2021

Generating Efficient DNN-Ensembles with Evolutionary Computation

In this work, we leverage ensemble learning as a tool for the creation of faster, smaller, and more accurate deep learning models. We demonstrate that we can jointly optimize for accuracy, inference time, and the number of parameters by combining DNN classifiers. To achieve this, we combine multiple ensemble strategies: bagging, boosting, and an ordered chain of classifiers. To reduce the number of DNN ensemble evaluations during the search, we propose EARN, an evolutionary approach that optimizes the ensemble according to three objectives regarding the constraints specified by the user. We run EARN on 10 image classification datasets with an initial pool of 32 state-of-the-art DCNN on both CPU and GPU platforms, and we generate models with speedups up to $7.60\times$, reductions of parameters by $10\times$, or increases in accuracy up to $6.01\%$ regarding the best DNN in the pool. In addition, our method generates models that are $5.6\times$ faster than the state-of-the-art methods for automatic model generation.

cs.NE↗

arXiv2020

Reducing Data Motion to Accelerate the Training of Deep Neural Networks

This paper reduces the cost of DNNs training by decreasing the amount of data movement across heterogeneous architectures composed of several GPUs and multicore CPU devices. In particular, this paper proposes an algorithm to dynamically adapt the data representation format of network weights during training. This algorithm drives a compression procedure that reduces data size before sending them over the parallel system. We run an extensive evaluation campaign considering several up-to-date deep neural network models and two high-end parallel architectures composed of multiple GPUs and CPU multicore chips. Our solution achieves average performance improvements from 6.18\% up to 11.91\%.

cs.DC↗

arXiv2018

Memory Vulnerability: A Case for Delaying Error Reporting

To face future reliability challenges, it is necessary to quantify the risk of error in any part of a computing system. To this goal, the Architectural Vulnerability Factor (AVF) has long been used for chips. However, this metric is used for offline characterisation, which is inappropriate for memory. We survey the literature and formalise one of the metrics used, the Memory Vulnerability Factor, and extend it to take into account false errors. These are reported errors which would have no impact on the program if they were ignored. We measure the False Error Aware MVF (FEA) and related metrics precisely in a cycle-accurate simulator, and compare them with the effects of injecting faults in a program's data, in native parallel runs. Our findings show that MVF and FEA are the only two metrics that are safe to use at runtime, as they both consistently give an upper bound on the probability of incorrect program outcome. FEA gives a tighter bound than MVF, and is the metric that correlates best with the incorrect outcome probability of all considered metrics.

cs.AR↗

arXiv2018

Low-Precision Floating-Point Schemes for Neural Network Training

The use of low-precision fixed-point arithmetic along with stochastic rounding has been proposed as a promising alternative to the commonly used 32-bit floating point arithmetic to enhance training neural networks training in terms of performance and energy efficiency. In the first part of this paper, the behaviour of the 12-bit fixed-point arithmetic when training a convolutional neural network with the CIFAR-10 dataset is analysed, showing that such arithmetic is not the most appropriate for the training phase. After that, the paper presents and evaluates, under the same conditions, alternative low-precision arithmetics, starting with the 12-bit floating-point arithmetic. These two representations are then leveraged using local scaling in order to increase accuracy and get closer to the baseline 32-bit floating-point arithmetic. Finally, the paper introduces a simplified model in which both the outputs and the gradients of the neural networks are constrained to power-of-two values, just using 7 bits for their representation. The evaluation demonstrates a minimal loss in accuracy for the proposed Power-of-Two neural network, avoiding the use of multiplications and divisions and thereby, significantly reducing the training time as well as the energy consumption and memory requirements during the training and inference phases.

cs.LG↗

arXiv2015

Adaptive and application dependent runtime guided hardware prefetcher reconfiguration on the IBM POWER7

Hardware data prefetcher engines have been extensively used to reduce the impact of memory latency. However, microprocessors' hardware prefetcher engines do not include any automatic hardware control able to dynamically tune their operation. This lacking architectural feature causes systems to operate with prefetchers in a fixed configuration, which in many cases harms performance and energy consumption. In this paper, a piece of software that solves the discussed problem in the context of the IBM POWER7 microprocessor is presented. The proposed solution involves using the runtime software as a bridge that is able to characterize user applications' workload and dynamically reconfigure the prefetcher engine. The proposed mechanisms has been deployed over OmpSs, a state-of-the-art task-based programming model. The paper shows significant performance improvements over a representative set of microbenchmarks and High Performance Computing (HPC) applications.

cs.DC↗