Synthesis Lectures on Computer Architecture – serie

Most emerging applications in imaging and machine learning must perform immense amounts of computation while holding to strict limits on energy and power. To meet these goals, architects are building increasingly specialized compute engines tailored for these specific tasks. The resulting computer systems are heterogeneous, containing multiple processing cores with wildly different execution models. Unfortunately, the cost of producing this specialized hardware—and the software to control it—is astronomical. Moreover, the task of porting algorithms to these heterogeneous machines typically requires that the algorithm be partitioned across the machine and rewritten for each specific architecture, which is time consuming and prone to error.Over the last several years, the authors have approached this problem using domain-specific languages (DSLs): high-level programming languages customized for specific domains, such as database manipulation, machine learning, or image processing. By giving up generality, these languages are able to provide high-level abstractions to the developer while producing high-performance output. The purpose of this book is to spur the adoption and the creation of domain-specific languages, especially for the task of creating hardware designs.In the first chapter, a short historical journey explains the forces driving computer architecture today. Chapter 2 describes the various methods for producing designs for accelerators, outlining the push for more abstraction and the tools that enable designers to work at a higher conceptual level. From there, Chapter 3 provides a brief introduction to image processing algorithms and hardware design patterns for implementing them. Chapters 4 and 5 describe and compare Darkroom and Halide, two domain-specific languages created for image processing that produce high-performance designs for both FPGAs and CPUs from the same source code, enabling rapid design cycles and quick porting of algorithms. The final section describes how the DSL approach also simplifies the problem of interfacing between application code and the accelerator by generating the driver stack in addition to the accelerator configuration.This book should serve as a useful introduction to domain-specialized computing for computer architecture students and as a primer on domain-specific languages and image processing hardware for those with more experience in the field.

Chip multiprocessors - also called multi-core microprocessors or CMPs for short - are now the only way to build high-performance microprocessors, for a variety of reasons. Large uniprocessors are no longer scaling in performance, because it is only possible to extract a limited amount of parallelism from a typical instruction stream using conventional superscalar instruction issue techniques. In addition, one cannot simply ratchet up the clock speed on today's processors, or the power dissipation will become prohibitive in all but water-cooled systems. Compounding these problems is the simple fact that with the immense numbers of transistors available on today's microprocessor chips, it is too costly to design and debug ever-larger processors every year or two. CMPs avoid these problems by filling up a processor die with multiple, relatively simpler processor cores instead of just one huge core. The exact size of a CMP's cores can vary from very simple pipelines to moderately complex superscalar processors, but once a core has been selected the CMP's performance can easily scale across silicon process generations simply by stamping down more copies of the hard-to-design, high-speed processor core in each successive chip generation. In addition, parallel code execution, obtained by spreading multiple threads of execution across the various cores, can achieve significantly higher performance than would be possible using only a single core. While parallel threads are already common in many useful workloads, there are still important workloads that are hard to divide into parallel threads. The low inter-processor communication latency between the cores in a CMP helps make a much wider range of applications viable candidates for parallel execution than was possible with conventional, multi-chip multiprocessors; nevertheless, limited parallelism in key applications is the main factor limiting acceptance of CMPs in some types of systems. After a discussion of the basic pros and cons of CMPs when they are compared with conventional uniprocessors, this book examines how CMPs can best be designed to handle two radically different kinds of workloads that are likely to be used with a CMP: highly parallel, throughput-sensitive applications at one end of the spectrum, and less parallel, latency-sensitive applications at the other. Throughput-sensitive applications, such as server workloads that handle many independent transactions at once, require careful balancing of all parts of a CMP that can limit throughput, such as the individual cores, on-chip cache memory, and off-chip memory interfaces. Several studies and example systems, such as the Sun Niagara, that examine the necessary tradeoffs are presented here. In contrast, latency-sensitive applications - many desktop applications fall into this category - require a focus on reducing inter-core communication latency and applying techniques to help programmers divide their programs into multiple threads as easily as possible. This book discusses many techniques that can be used in CMPs to simplify parallel programming, with an emphasis on research directions proposed at Stanford University. To illustrate the advantages possible with a CMP using a couple of solid examples, extra focus is given to thread-level speculation (TLS), a way to automatically break up nominally sequential applications into parallel threads on a CMP, and transactional memory. This model can greatly simplify manual parallel programming by using hardware - instead of conventional software locks - to enforce atomic code execution of blocks of instructions, a technique that makes parallel coding much less error-prone. Contents: The Case for CMPs / Improving Throughput / Improving Latency Automatically / Improving Latency using Manual Parallel Programming / A Multicore World: The Future of CMPs

Table of Contents: Introduction / Modeling, Simulation, and Measurement / Using Voltage and Frequency Adjustments to Manage Dynamic Power / Optimizing Capacitance and Switching Activity to Reduce Dynamic Power / Managing Static (Leakage) Power / Conclusions

For many years, most computer architects have pursued one primary goal: performance. Table of Contents: Introduction / Error Detection / Error Recovery / Diagnosis / Self-Repair / The Future

Today, computer-system optimization, at both the hardware and software levels, must consider the details of the memory system in its analysis; failing to do so yields systems that are increasingly inefficient as those systems become more complex. This lecture seeks to introduce the reader to the most important details of the memory system; it targets both computer scientists and computer engineers in industry and in academia. Roughly speaking, computer scientists are the users of the memory system and computer engineers are the designers of the memory system. Both can benefit tremendously from a basic understanding of how the memory system really works: the computer scientist will be better equipped to create algorithms that perform well and the computer engineer will be better equipped to design systems that approach the optimal, given the resource limitations. Currently, there is consensus among architecture researchers that the memory system is "the bottleneck," and this consensus has held for over a decade. Somewhat inexplicably, most of the research in the field is still directed toward improving the CPU to better tolerate a slow memory system, as opposed to addressing the weaknesses of the memory system directly. This lecture should get the bulk of the computer science and computer engineering population up the steep part of the learning curve. Not every CS/CE researcher/developer needs to do work in the memory system, but, just as a carpenter can do his job more efficiently if he knows a little of architecture, and an architect can do his job more efficiently if he knows a little of carpentry, giving the CS/CE worlds better intuition about the memory system should help them build better systems, both software and hardware. Table of Contents: Primers / It Must Be Modeled Accurately / ...\ and It Will Change Soon

Table of Contents: FPGA Technology / Reconfigurable Supercomputing / Algorithmic Considerations / FPGA Programming Languages / Case Study: Sorting / Alternative Technologies and Concluding Remarks

Table of Contents: Introduction / Performance Metrics / Workload Design / Analytical Performance Modeling / Simulation / Sampled Simulation / Statistical Simulation / Parallel Simulation and Hardware Acceleration / Concluding Remarks

The advent of multicore processors has renewed interest in the idea of incorporating transactions into the programming model used to write parallel programs. Table of Contents: Introduction / Basic Transactions / Building on Basic Transactions / Software Transactional Memory / Hardware-Supported Transactional Memory / Conclusions

Table of Contents: Introduction / Caches / The Instruction Fetch Unit / Decode / Allocation / The Issue Stage / Execute / The Commit Stage / References / Author Biographies

Table of Contents: Introduction / Background / Topology Basics / High-Radix Topologies / Routing / Scalable Switch Microarchitecture / System Packaging / Case Studies / Closing Remarks

Quantum computers can (in theory) solve certain problems far faster than a classical computer running any known classical algorithm. While existing technologies for building quantum computers are in their infancy, it is not too early to consider their scalability and reliability in the context of the design of large-scale quantum computers. To architect such systems, one must understand what it takes to design and model a balanced, fault-tolerant quantum computer architecture. The goal of this lecture is to provide architectural abstractions for the design of a quantum computer and to explore the systems-level challenges in achieving scalable, fault-tolerant quantum computation. In this lecture, we provide an engineering-oriented introduction to quantum computation with an overview of the theory behind key quantum algorithms. Next, we look at architectural case studies based upon experimental data and future projections for quantum computation implemented using trapped ions. While we focus here on architectures targeted for realization using trapped ions, the techniques for quantum computer architecture design, quantum fault-tolerance, and compilation described in this lecture are applicable to many other physical technologies that may be viable candidates for building a large-scale quantum computing system. We also discuss general issues involved with programming a quantum computer as well as a discussion of work on quantum architectures based on quantum teleportation. Finally, we consider some of the open issues remaining in the design of quantum computers. Table of Contents: Introduction / Basic Elements for Quantum Computation / Key Quantum Algorithms / Building Reliable and Scalable Quantum Architectures / Simulation of Quantum Computation / Architectural Elements / Case Study: The Quantum Logic Array Architecture / Programming the Quantum Architecture / Using the QLA for Quantum Simulation: The Transverse Ising Model / Teleportation-Based Quantum Architectures/ Concluding Remarks

Table of Contents: Dynamic Binary Modification: Overview / Using a Dynamic Binary Modifier / Program Analysis and Debugging / Active Program Modification / Architectural Exploration / Advanced System Internals / Historical Perspectives / Summary and Observations

Table of Contents: Basic Elements of Large Cache Design / Organizing Data in CMP Last Level Caches / Policies Impacting Cache Hit Rates / Interconnection Networks within Large Caches / Technology / Concluding Remarks

Table of Contents: Next Generation Memory Technologies / Architecting PCM for Main Memories / Tolerating Slow Writes in PCM / Wear Leveling for Durability / Wear Leveling Under Adversarial Settings / Error Resilience in Phase Change Memories /Storage and System Design With Emerging Non-Volatile Memories

Table of Contents: Introduction and overview / Dependence analysis, dependence graphs and alias analysis / Program parallelization / Transformations to modify and eliminate dependences / Transformation of iterative and recursive constructs / Compiling for distributed memory machines / Solving Diophantine equations / A guide to further reading

General-purpose graphics processing units (GPGPU) have emerged as an important class of shared memory parallel processing architectures, with widespread deployment in every computer class from high-end supercomputers to embedded mobile platforms.

Table of Contents: Introduction / Multithreaded Execution Models / Coarse-Grain Multithreading / Fine-Grain Multithreading / Simultaneous Multithreading / Managing Contention / New Opportunities for Multithreaded Processors / Experimentation and Metrics / Implementations of Multithreaded Processors / Conclusion

Table of Contents: Introduction / Modeling Voltage Variation / Understanding the Characteristics of Voltage Variation / Traditional Solutions and Emerging Solution Forecast / Allowing and Tolerating Voltage Emergencies / Predicting and Avoiding Voltage Emergencies / Eliminiating Recurring Voltage Emergencies / Future Directions on Resiliency

This book offers a comprehensive survey of shared-memory synchronization, with an emphasis on “systems-level” issues. It includes sufficient coverage of architectural details to understand correctness and performance on modern multicore machines, and sufficient coverage of higher-level issues to understand how synchronization is embedded in modern programming languages.The primary intended audience for this book is “systems programmers”—the authors of operating systems, library packages, language run-time systems, concurrent data structures, and server and utility programs. Much of the discussion should also be of interest to application programmers who want to make good use of the synchronization mechanisms available to them, and to computer architects who want to understand the ramifications of their design decisions on systems-level code.

Design for security is an essential aspect of the design of future computers. What the book focuses on are the fundamental security concepts, across different security communities, that should be understood by any computer architect trying to design or evaluate security-awarecomputer architectures.

Since the 1970’s, microprocessor-based digital platforms have been riding Moore’s law, allowing for doubling of density for the same area roughly every two years. However, whereas microprocessor fabrication has focused on increasing instruction execution rate, memory fabrication technologies have focused primarily on an increase in capacity with negligible increase in speed. This divergent trend in performance between the processors and memory has led to a phenomenon referred to as the “Memory Wall.” To overcome the memory wall, designers have resorted to a hierarchy of cache memory levels, which rely on the principal of memory access locality to reduce the observed memory access time and the performance gap between processors and memory. Unfortunately, important workload classes exhibit adverse memory access patterns that baffle the simple policies built into modern cache hierarchies to move instructions and data across cache levels. As such, processors often spend much time idling upon a demand fetch of memory blocks that miss in higher cache levels. Prefetching—predicting future memory accesses and issuing requests for the corresponding memory blocks in advance of explicit accesses—is an effective approach to hide memory access latency. There have been a myriad of proposed prefetching techniques, and nearly every modern processor includes some hardware prefetching mechanisms targeting simple and regular memory access patterns. This primer offers an overview of the various classes of hardware prefetchers for instructions and data proposed in the research literature, and presents examples of techniques incorporated into modern microprocessors.

Table of Contents: Preface / Acknowledgments / Introduction / Simulator Background / Accelerating Computer System Simulators with FPGAs / Simulation Virtualization / Categorizing FPGA-based Simulators / Conclusion / Bibliography / Authors' Biographies

Table of Contents: Introduction / Voltage and Frequency Management / Heterogeneity and Specialization / Communication and Memory Systems / Conclusions / Bibliography / Authors' Biographies

Having hit power limitations to even more aggressive out-of-order execution in processor cores, many architects in the past decade have turned to single-instruction-multiple-data (SIMD) execution to increase single-threaded performance.

The emerging three-dimensional (3D) chip architectures, with their intrinsic capability of reducing the wire length, promise attractive solutions to reduce the delay of interconnects in future microprocessors.

Since the end of Dennard scaling in the early 2000s, improving the energy efficiency of computation has been the main concern of the research community and industry.

It is designed to serve as a guide for addressing various open problems in accelerating analytics workloads, e.g., new architectural features for supporting analytics workloads, impact on programming models and runtime systems, and designing analytics systems.

Envisioning future computing systems with a diverse set of general-purpose cores and accelerators, computer architects must add accelerator-related research infrastructures to their toolboxes to explore future heterogeneous systems.