Heterogeneous multicore systems for signal processing

This thesis explores the capabilities of heterogeneous multi-core systems, based on multiple Graphics Processing Units (GPUs) in a standard desktop framework. Multi-GPU accelerated desk side computers are an appealing alternative to other high performance computing (HPC) systems: being composed of commodity hardware components fabricated in large quantities, their price-performance ratio is unparalleled in the world of high performance computing. Essentially bringing “supercomputing to the masses”, this opens up new possibilities for application fields where investing in HPC resources had been considered unfeasible before. One of these is the field of bioelectrical imaging, a class of medical imaging technologies that occupy a low-cost niche next to million-dollar systems like functional Magnetic Resonance Imaging (fMRI). In the scope of this work, several computational challenges encountered in bioelectrical imaging are tackled with this new kind of computing resource, striving to help these methods approach their true potential. Specifically, the following main contributions were made: Firstly, a novel dual-GPU implementation of parallel triangular matrix inversion (TMI) is presented, addressing an crucial kernel in computation of multi-mesh head models of encephalographic (EEG) source localization. This includes not only a highly efficient implementation of the routine itself achieving excellent speedups versus an optimized CPU implementation, but also a novel GPU-friendly compressed storage scheme for triangular matrices. Secondly, a scalable multi-GPU solver for non-hermitian linear systems was implemented. It is integrated into a simulation environment for electrical impedance tomography (EIT) that requires frequent solution of complex systems with millions of unknowns, a task that this solution can perform within seconds. In terms of computational throughput, it outperforms not only an highly optimized multi-CPU reference, but related GPU-based work as well. Finally, a GPU-accelerated graphical EEG real-time source localization software was implemented. Thanks to acceleration, it can meet real-time requirements in unpreceeded anatomical detail running more complex localization algorithms. Additionally, a novel implementation to extract anatomical priors from static Magnetic Resonance (MR) scansions has been included.

Variability-tolerant High-reliability Multicore Platforms

Next generation electronic devices have to guarantee high performance while being less power-consuming and highly reliable for several application domains ranging from the entertainment to the business. In this context, multicore platforms have proven the most efficient design choice but new challenges have to be faced. The ever-increasing miniaturization of the components produces unexpected variations on technological parameters and wear-out characterized by soft and hard errors. Even though hardware techniques, which lend themselves to be applied at design time, have been studied with the objective to mitigate these effects, they are not sufficient; thus software adaptive techniques are necessary. In this thesis we focus on multicore task allocation strategies to minimize the energy consumption while meeting performance constraints. We firstly devise a technique based on an Integer Linear Problem formulation which provides the optimal solution but cannot be applied on-line since the algorithm it needs is time-demanding; then we propose a sub-optimal technique based on two steps which can be applied on-line. We demonstrate the effectiveness of the latter solution through an exhaustive comparison against the optimal solution, state-of-the-art policies, and variability-agnostic task allocations by running multimedia applications on the virtual prototype of a next generation industrial multicore platform. We also face the problem of the performance and lifetime degradation. We firstly focus on embedded multicore platforms and propose an idleness distribution policy that increases core expected lifetimes by duty cycling their activity; then, we investigate the use of micro thermoelectrical coolers in general-purpose multicore processors to control the temperature of the cores at runtime with the objective of meeting lifetime constraints without performance loss.

Architetture di interconnessione per sistemi su singolo chip e per sistemi ad integrazione tridimensionale

I continui sviluppi nel campo della fabbricazione dei circuiti integrati hanno comportato frequenti travolgimenti nel design, nell’implementazione e nella scalabilità dei device elettronici, così come nel modo di utilizzarli. Anche se la legge di Moore ha anticipato e caratterizzato questo trend nelle ultime decadi, essa stessa si trova a fronteggiare attualmente enormi limitazioni, superabili solo attraverso un diverso approccio nella produzione di chip, consistente in pratica nella sovrapposizione verticale di diversi strati collegati elettricamente attraverso speciali vias. Sul singolo strato, le network on chip sono state suggerite per ovviare le profonde limitazioni dovute allo scaling di strutture di comunicazione condivise. Questa tesi si colloca principalmente nel contesto delle nascenti piattaforme multicore ad alte prestazioni basate sulle 3D NoC, in cui la network on chip viene estesa nelle 3 direzioni. L’obiettivo di questo lavoro è quello di fornire una serie di strumenti e tecniche per poter costruire e aratterizzare una piattaforma tridimensionale, cosi come dimostrato nella realizzazione del testchip 3D NOC fabbricato presso la fonderia IMEC. Il primo contributo è costituito sia una accurata caratterizzazione delle interconnessioni verticali (TSVs) (ovvero delle speciali vias che attraversano l’intero substrato del die), sia dalla caratterizzazione dei router 3D (in cui una o più porte sono estese nella direzione verticale) ed infine dal setup di un design flow 3D utilizzando interamente CAD 2D. Questo primo step ci ha permesso di effettuare delle analisi dettagliate sia sul costo sia sulle varie implicazioni. Il secondo contributo è costituito dallo sviluppo di alcuni blocchi funzionali necessari per garantire il corretto funziomento della 3D NoC, in presenza sia di guasti nelle TSVs (fault tolerant links) che di deriva termica nei vari clock tree dei vari die (alberi di clock indipendenti). Questo secondo contributo è costituito dallo sviluppo delle seguenti soluzioni circuitali: 3D fault tolerant link, Look Up Table riconfigurabili e un sicnronizzatore mesocrono. Il primo è costituito fondamentalmente un bus verticale equipaggiato con delle TSV di riserva da utilizzare per rimpiazzare le vias guaste, più la logica di controllo per effettuare il test e la riconfigurazione. Il secondo è rappresentato da una Look Up Table riconfigurabile, ad alte prestazioni e dal costo contenuto, necesaria per bilanciare sia il traffico nella NoC che per bypassare link non riparabili. Infine la terza soluzione circuitale è rappresentata da un sincronizzatore mesocrono necessario per garantire la sincronizzazione nel trasferimento dati da un layer and un altro nelle 3D Noc. Il terzo contributo di questa tesi è dato dalla realizzazione di un interfaccia multicore per memorie 3D (stacked 3D DRAM) ad alte prestazioni, e dall’esplorazione architetturale dei benefici e del costo di questo nuovo sistema in cui il la memoria principale non è piu il collo di bottiglia dell’intero sistema. Il quarto ed ultimo contributo è rappresentato dalla realizzazione di un 3D NoC test chip presso la fonderia IMEC, e di un circuito full custom per la caratterizzazione della variability dei parametri RC delle interconnessioni verticali.

Hardware/Software Design of Dynamic Real-Time Schedulers for Embedded Multiprocessor Systems

The new generation of multicore processors opens new perspectives for the design of embedded systems. Multiprocessing, however, poses new challenges to the scheduling of real-time applications, in which the ever-increasing computational demands are constantly flanked by the need of meeting critical time constraints. Many research works have contributed to this field introducing new advanced scheduling algorithms. However, despite many of these works have solidly demonstrated their effectiveness, the actual support for multiprocessor real-time scheduling offered by current operating systems is still very limited. This dissertation deals with implementative aspects of real-time schedulers in modern embedded multiprocessor systems. The first contribution is represented by an open-source scheduling framework, which is capable of realizing complex multiprocessor scheduling policies, such as G-EDF, on conventional operating systems exploiting only their native scheduler from user-space. A set of experimental evaluations compare the proposed solution to other research projects that pursue the same goals by means of kernel modifications, highlighting comparable scheduling performances. The principles that underpin the operation of the framework, originally designed for symmetric multiprocessors, have been further extended first to asymmetric ones, which are subjected to major restrictions such as the lack of support for task migrations, and later to re-programmable hardware architectures (FPGAs). In the latter case, this work introduces a scheduling accelerator, which offloads most of the scheduling operations to the hardware and exhibits extremely low scheduling jitter. The realization of a portable scheduling framework presented many interesting software challenges. One of these has been represented by timekeeping. In this regard, a further contribution is represented by a novel data structure, called addressable binary heap (ABH). Such ABH, which is conceptually a pointer-based implementation of a binary heap, shows very interesting average and worst-case performances when addressing the problem of tick-less timekeeping of high-resolution timers.

Use of shared memory in the context of embedded multi-core processors: exploration of the technology and its limits

Modern embedded systems embrace many-core shared-memory designs. Due to constrained power and area budgets, most of them feature software-managed scratchpad memories instead of data caches to increase the data locality. It is therefore programmers’ responsibility to explicitly manage the memory transfers, and this make programming these platform cumbersome. Moreover, complex modern applications must be adequately parallelized before they can the parallel potential of the platform into actual performance. To support this, programming languages were proposed, which work at a high level of abstraction, and rely on a runtime whose cost hinders performance, especially in embedded systems, where resources and power budget are constrained. This dissertation explores the applicability of the shared-memory paradigm on modern many-core systems, focusing on the ease-of-programming. It focuses on OpenMP, the de-facto standard for shared memory programming. In a first part, the cost of algorithms for synchronization and data partitioning are analyzed, and they are adapted to modern embedded many-cores. Then, the original design of an OpenMP runtime library is presented, which supports complex forms of parallelism such as multi-level and irregular parallelism. In the second part of the thesis, the focus is on heterogeneous systems, where hardware accelerators are coupled to (many-)cores to implement key functional kernels with orders-of-magnitude of speedup and energy efficiency compared to the “pure software” version. However, three main issues rise, namely i) platform design complexity, ii) architectural scalability and iii) programmability. To tackle them, a template for a generic hardware processing unit (HWPU) is proposed, which share the memory banks with cores, and the template for a scalable architecture is shown, which integrates them through the shared-memory system. Then, a full software stack and toolchain are developed to support platform design and to let programmers exploiting the accelerators of the platform. The OpenMP frontend is extended to interact with it.

Operating System Contribution to Composable Timing Behaviour in High-Integrity Real-Time Systems

The development of High-Integrity Real-Time Systems has a high footprint in terms of human, material and schedule costs. Factoring functional, reusable logic in the application favors incremental development and contains costs. Yet, achieving incrementality in the timing behavior is a much harder problem. Complex features at all levels of the execution stack, aimed to boost average-case performance, exhibit timing behavior highly dependent on execution history, which wrecks time composability and incrementaility with it. Our goal here is to restitute time composability to the execution stack, working bottom up across it. We first characterize time composability without making assumptions on the system architecture or the software deployment to it. Later, we focus on the role played by the real-time operating system in our pursuit. Initially we consider single-core processors and, becoming less permissive on the admissible hardware features, we devise solutions that restore a convincing degree of time composability. To show what can be done for real, we developed TiCOS, an ARINC-compliant kernel, and re-designed ORK+, a kernel for Ada Ravenscar runtimes. In that work, we added support for limited-preemption to ORK+, an absolute premiere in the landscape of real-word kernels. Our implementation allows resource sharing to co-exist with limited-preemptive scheduling, which extends state of the art. We then turn our attention to multicore architectures, first considering partitioned systems, for which we achieve results close to those obtained for single-core processors. Subsequently, we shy away from the over-provision of those systems and consider less restrictive uses of homogeneous multiprocessors, where the scheduling algorithm is key to high schedulable utilization. To that end we single out RUN, a promising baseline, and extend it to SPRINT, which supports sporadic task sets, hence matches real-world industrial needs better. To corroborate our results we present findings from real-world case studies from avionic industry.