物件導向軟體之架構(Object-Oriented Software Construction)探討

Approximate computing trades off computation quality with effort expended, and as rising performance demands confront plateauing resource budgets, approximate computing has become not merely attractive, but even imperative. In this article, we present a survey of techniques for approximate computing (AC). We discuss strategies for finding approximable program portions and monitoring output quality, techniques for using AC in different processing units (e.g., CPU, GPU, and FPGA), processor components, memory technologies, and so forth, as well as programming frameworks for AC. We classify these techniques based on several key characteristics to emphasize their similarities and differences. The aim of this article is to provide insights to researchers into working of AC techniques and inspire more efforts in this area to make AC the mainstream computing approach in future systems.

/pdf/a-survey-of-techniques-for-approximate-computing-3va5vv4jnf.pdf

A Survey of Techniques for Approximate Computing

The continued success of Deep Neural Networks (DNNs) in classification tasks has sparked a trend of accelerating their execution with specialized hardware. While published designs easily give an order of magnitude improvement over general-purpose hardware, few look beyond an initial implementation. This paper presents Minerva, a highly automated co-design approach across the algorithm, architecture, and circuit levels to optimize DNN hardware accelerators. Compared to an established fixed-point accelerator baseline, we show that fine-grained, heterogeneous datatype optimization reduces power by 1.5×; aggressive, inline predication and pruning of small activity values further reduces power by 2.0×; and active hardware fault detection coupled with domain-aware error mitigation eliminates an additional 2.7× through lowering SRAM voltages. Across five datasets, these optimizations provide a collective average of 8.1× power reduction over an accelerator baseline without compromising DNN model accuracy. Minerva enables highly accurate, ultra-low power DNN accelerators (in the range of tens of milliwatts), making it feasible to deploy DNNs in power-constrained IoT and mobile devices.

Minerva: enabling low-power, highly-accurate deep neural network accelerators

This paper describes a learning-based approach to the acceleration of approximate programs. We describe the \emph{Parrot transformation}, a program transformation that selects and trains a neural network to mimic a region of imperative code. After the learning phase, the compiler replaces the original code with an invocation of a low-power accelerator called a \emph{neural processing unit} (NPU). The NPU is tightly coupled to the processor pipeline to accelerate small code regions. Since neural networks produce inherently approximate results, we define a programming model that allows programmers to identify approximable code regions -- code that can produce imprecise but acceptable results. Offloading approximable code regions to NPUs is faster and more energy efficient than executing the original code. For a set of diverse applications, NPU acceleration provides whole-application speedup of 2.3x and energy savings of 3.0x on average with quality loss of at most 9.6%.

/pdf/neural-acceleration-for-general-purpose-approximate-programs-chuxql7jbp.pdf

Neural Acceleration for General-Purpose Approximate Programs

Disciplined approximate programming lets programmers declare which parts of a program can be computed approximately and consequently at a lower energy cost. The compiler proves statically that all approximate computation is properly isolated from precise computation. The hardware is then free to selectively apply approximate storage and approximate computation with no need to perform dynamic correctness checks.In this paper, we propose an efficient mapping of disciplined approximate programming onto hardware. We describe an ISA extension that provides approximate operations and storage, which give the hardware freedom to save energy at the cost of accuracy. We then propose Truffle, a microarchitecture design that efficiently supports the ISA extensions. The basis of our design is dual-voltage operation, with a high voltage for precise operations and a low voltage for approximate operations. The key aspect of the microarchitecture is its dependence on the instruction stream to determine when to use the low voltage. We evaluate the power savings potential of in-order and out-of-order Truffle configurations and explore the resulting quality of service degradation. We evaluate several applications and demonstrate energy savings up to 43%.

/pdf/architecture-support-for-disciplined-approximate-programming-rnoarszkkg.pdf

Architecture support for disciplined approximate programming

Energy is increasingly a first-order concern in computer systems. Exploiting energy-accuracy trade-offs is an attractive choice in applications that can tolerate inaccuracies. Recent work has explored exposing this trade-off in programming models. A key challenge, though, is how to isolate parts of the program that must be precise from those that can be approximated so that a program functions correctly even as quality of service degrades.We propose using type qualifiers to declare data that may be subject to approximate computation. Using these types, the system automatically maps approximate variables to low-power storage, uses low-power operations, and even applies more energy-efficient algorithms provided by the programmer. In addition, the system can statically guarantee isolation of the precise program component from the approximate component. This allows a programmer to control explicitly how information flows from approximate data to precise data. Importantly, employing static analysis eliminates the need for dynamic checks, further improving energy savings. As a proof of concept, we develop EnerJ, an extension to Java that adds approximate data types. We also propose a hardware architecture that offers explicit approximate storage and computation. We port several applications to EnerJ and show that our extensions are expressive and effective; a small number of annotations lead to significant potential energy savings (10%-50%) at very little accuracy cost.

/pdf/enerj-approximate-data-types-for-safe-and-general-low-power-45zkh0umzv.pdf

EnerJ: approximate data types for safe and general low-power computation

Object-oriented programs with arbitrary object structures are difficult to understand, to maintain, and to reason about. Ownership has been applied successfully to structure the object store and to restrict how references can be passed and used. We describe how ownership relations can be expressed in the Java Modeling Language, JML. These ownership specifications can be checked by standard verification techniques, runtime assertion checking, ownership type systems, or combinations of these techniques. We show that the combination of the lightweight Universe type system and JML specifications is flexible enough to handle interesting implementations while keeping the annotation and checking overhead small. The Universe type system has been implemented in the JML compiler. This integration enables the application of ownership-based verification techniques to programs specified in JML.

/pdf/universes-lightweight-ownership-for-jml-2haavm9j5z.pdf

Universes: Lightweight Ownership for JML.

Ownership is a powerful concept to structure the object store and to control aliasing and modifications of objects. This paper presents an ownership type system for a Java-like programming language with generic types. Like our earlier Universe type system, Generic Universe Types enforce the owner-as-modifier discipline. This discipline does not restrict aliasing, but requires modifications of an object to be initiated by its owner. This allows owner objects to control state changes of owned objects, for instance, to maintain invariants. Generic Universe Types require a small annotation overhead and provide strong static guarantees. They are the first type system that combines the owner-as-modifier discipline with type genericity.

/pdf/generic-universe-types-3ozjhdmdn2.pdf

Generic universe types

This paper describes practical experience building and using pluggable type-checkers. A pluggable type-checker refines (strengthens) the built-in type system of a programming language. This permits programmers to detect and prevent, at compile time, defects that would otherwise have been manifested as run-time errors. The prevented defects may be generally applicable to all programs, such as null pointer dereferences. Or, an application-specific pluggable type system may be designed for a single application. We built a series of pluggable type checkers using the Checker Framework, and evaluated them on 2 million lines of code, finding hundreds of bugs in the process. We also observed 28 first-year computer science students use a checker to eliminate null pointer errors in their course projects. Along with describing the checkers and characterizing the bugs we found, we report the insights we had throughout the process. Overall, we found that the type checkers were easy to write, easy for novices to productively use, and effective in finding real bugs and verifying program properties, even for widely tested and used open source projects.

/pdf/building-and-using-pluggable-type-checkers-34f3mx8k18.pdf

Building and using pluggable type-checkers

Implicit or indirect control flow is a transfer of control between procedures using some mechanism other than an explicit procedure call. Implicit control flow is a staple design pattern that adds flexibility to system design. However, it is challenging for a static analysis to compute or verify properties about a system that uses implicit control flow. This paper presents static analyses for two types of implicit control flow that frequently appear in Android apps: Java reflection and Android intents. Our analyses help to resolve where control flows and what data is passed. This information improves the precision of downstream analyses, which no longer need to make conservative assumptions about implicit control flow. We have implemented our techniques for Java. We enhanced an existing security analysis with a more precise treatment of reflection and intents. In a case study involving ten real-world Android apps that use both intents and reflection, the precision of the security analysis was increased on average by two orders of magnitude. The precision of two other downstream analyses was also improved.

/pdf/static-analysis-of-implicit-control-flow-resolving-java-3tnsiody1q.pdf

Werner Dietl

Papers

EnerJ: approximate data types for safe and general low-power computation

Universes: Lightweight Ownership for JML.

Generic universe types

Building and using pluggable type-checkers

Static Analysis of Implicit Control Flow: Resolving Java Reflection and Android Intents (T)