A Hybrid Parallel Algorithm for Computing and Tracking Level Set Topology

The contour tree is a topological abstraction of a scalar field that captures evolution in level set connectivity. It is an effective representation for visual exploration and analysis of scientific data. We describe a work-efficient, output sensitive, and scalable parallel algorithm for computing the contour tree of a scalar field defined on a domain that is represented using either an unstructured mesh or a structured grid. A hybrid implementation of the algorithm using the GPU and multi-core CPU can compute the contour tree of an input containing 16 million vertices in less than ten seconds with a speedup factor of upto 13. Experiments based on an implementation in a multi-core CPU environment show near-linear speedup for large data sets.

Sparsely Faceted Arrays: A Mechanism Supporting Parallel Allocation, Communication, and Garbage Collection

Conventional parallel computer architectures do not provide support for non-uniformly distributed objects. In this thesis, I introduce sparsely faceted arrays (SFAs), a new low-level mechanism for naming regions of memory, or facets, on different processors in a distributed, shared memory parallel processing system. Sparsely faceted arrays address the disconnect between the global distributed arrays provided by conventional architectures (e.g. the Cray T3 series), and the requirements of high-level parallel programming methods that wish to use objects that are distributed over only a subset of processing elements. A sparsely faceted array names a virtual globally-distributed array, but actual facets are lazily allocated. By providing simple semantics and making efficient use of memory, SFAs enable efficient implementation of a variety of non-uniformly distributed data structures and related algorithms. I present example applications which use SFAs, and describe and evaluate simple hardware mechanisms for implementing SFAs. Keeping track of which nodes have allocated facets for a particular SFA is an important task that suggests the need for automatic memory management, including garbage collection. To address this need, I first argue that conventional tracing techniques such as mark/sweep and copying GC are inherently unscalable in parallel systems. I then present a parallel memory-management strategy, based on reference-counting, that is capable of garbage collecting sparsely faceted arrays. I also discuss opportunities for hardware support of this garbage collection strategy. I have implemented a high-level hardware/OS simulator featuring hardware support for sparsely faceted arrays and automatic garbage collection. I describe the simulator and outline a few of the numerous details associated with a "real" implementation of SFAs and SFA-aware garbage collection. Simulation results are used throughout this thesis in the evaluation of hardware support mechanisms.

Cache-Integrated Network Interfaces: Flexible On-Chip Communication and Synchronization for Large-Scale CMPs

Per-core scratchpad memories (or local stores) allow direct inter-core communication, with latency and energy advantages over coherent cache-based communication, especially as CMP architectures become more distributed. We have designed cache-integrated network interfaces, appropriate for scalable multicores, that combine the best of two worlds – the flexibility of caches and the efficiency of scratchpad memories: on-chip SRAM is configurably shared among caching, scratchpad, and virtualized network interface (NI) functions. This paper presents our architecture, which provides local and remote scratchpad access, to either individual words or multiword blocks through RDMA copy. Furthermore, we introduce event responses, as a technique that enables software configurable communication and synchronization primitives. We present three event response mechanisms that expose NI functionality to software, for multiword transfer initiation, completion notifications for software selected sets of arbitrary size transfers, and multi-party synchronization queues. We implemented these mechanisms in a four-core FPGA prototype, and measure the logic overhead over a cache-only design for basic NI functionality to be less than 20%. We also evaluate the on-chip communication performance on the prototype, as well as the performance of synchronization functions with simulation of CMPs with up to 128 cores. We demonstrate efficient synchronization, low-overhead communication, and amortized-overhead bulk transfers, which allow parallelization gains for fine-grain tasks, and efficient exploitation of the hardware bandwidth.

The Paralax Infrastructure: Automatic Parallelization With a Helping Hand

Speeding up sequential programs on multicores is a challenging problem that is in urgent need of a solution. Automatic parallelization of irregular pointer-intensive codes, exempli?ed by the SPECint codes, is a very hard problem. This paper shows that, with a helping hand, such auto-parallelization is possible and fruitful. This paper makes the following contributions: (i) A compiler framework for extracting pipeline-like parallelism from outer program loops is presented. (ii) Using a light-weight programming model based on annotations, the programmer helps the compiler to ?nd thread-level parallelism. Each of the annotations speci?es only a small piece of semantic information that compiler analysis misses, e.g. stating that a variable is dead at a certain program point. The annotations are designed such that correctness is easily veri?ed. Furthermore, we present a tool for suggesting annotations to the programmer. (iii) The methodology is applied to autoparallelize several SPECint benchmarks. For the benchmark with most parallelism (hmmer), we obtain a scalable 7-fold speedup on an AMD quad-core dual processor. The annotations constitute a parallel programming model that relies extensively on a sequential program representation. Hereby, the complexity of debugging is not increased and it does not obscure the source code. These properties could prove valuable to increase the ef?ciency of parallel programming.

Automatic Parallelization in the Paralax Compiler

The efficient development of multi-threaded software has, for many years, been an unsolved problem in computer science. Finding a solution to this problem has become urgent with the advent of multi-core processors. Furthermore, the problem has become more complicated because multi-cores are everywhere (desktop, laptop, embedded system). As such, they execute generic programs which exhibit very different characteristics than the scientific applications that have been the focus of parallel computing in the past.
Implicitly parallel programming is an approach to parallel pro- gramming that promises high productivity and efficiency and rules out synchronization errors and race conditions by design. There are two main ingredients to implicitly parallel programming: (i) a con- ventional sequential programming language that is extended with annotations that describe the semantics of the program and (ii) an automatic parallelizing compiler that uses the annotations to in- crease the degree of parallelization.
It is extremely important that the annotations and the automatic parallelizing compiler are designed with the target application do- main in mind. In this paper, we discuss the Paralax approach to im- plicitly parallel programming and we review how the annotations and the compiler design help to successfully parallelize generic programs. We evaluate Paralax on SPECint benchmarks, which are a model for such programs, and demonstrate scalable speedups, up to a factor of 6 on 8 cores.

Formal Analysis of MPI-based Parallel Programs

Most parallel computing applications in highperformance computing use the Message Passing Interface (MPI) API. Given the fundamental importance of parallel computing to science and engineering research, application correctness is paramount. MPI was originally developed around 1993 by the MPI Forum, a group of vendors, parallel programming researchers, and computational scientists. However, the document defining the standard is not issued by an official standards organization but has become a de facto standard © 2011 ACM.

Structured data access annotations for massively parallel computations

We describe an approach aimed at addressing the issue of joint exploitation of control (stream) and data parallelism in a skeleton based parallel programming environment, based on annotations and refactoring. Annotations drive efficient implementation of a parallel computation. Refactoring is used to transform the associated skeleton tree into a more efficient, functionally equivalent skeleton tree. In most cases, cost models are used to drive the refactoring process. We show how sample use case applications/kernels may be optimized and discuss preliminary experiments with FastFlow assessing the theoretical results. © 2013 Springer-Verlag.

A language-independent parallel refactoring framework

Recent trends towards increasingly parallel computers mean that there needs to be a seismic shift in programming practice. The time is rapidly approaching when most programming will be for parallel systems. However, most programming techniques in use today are geared towards sequential, or occasionally small-scale parallel, programming. While refactoring has so far mainly been applied to sequential programs, it is our contention that refactoring can play a key role in significantly improving the programmability of parallel systems, by allowing the programmer to apply a set of well-defined transformations in order to parallelise their programs. In this paper, we describe a new language-independent refactoring approach that helps introduce and tune parallelism through high-level design patterns targeting a set of well-specified parallel skeletons. We believe this new refactoring process is the key to allowing programmers to truly start thinking in parallel. © 2012 ACM.

A Comparison of Programming Models for Multiprocessors with Explicitly Managed Memory Hierarchies

On multiprocessors with explicitly managed memory hierarchies (EMM), software has the responsibility of moving data in and out of fast local memories. This task can be complex and error-prone even for expert programmers. Before we can allow compilers to handle the complexity for us, we must identify the abstractions that are general enough to allow us to write applications with reasonable effort, yet speci?c enough to exploit the vast on-chip memory bandwidth of EMM multi-processors. To this end, we compare two programming models against hand-tuned codes on the STI Cell, paying attention to programmability and performance. The ?rst programming model, Sequoia, abstracts the memory hierarchy as private address spaces, each corresponding to a parallel task. The second, Cellgen, is a new framework which provides OpenMP-like semantics and the abstraction of a shared address spaces divided into private and shared data. We compare three applications programmed using these models against their hand-optimized counterparts in terms of abstractions, programming complexity, and performance.

Inference and Declaration of Independence in Task-Parallel Programs

The inherent difficulty of thread-based shared-memory programming has recently motivated research in high-level, task-parallel programming models. Recent advances of Task-Parallel models add implicit synchronization, where the system automatically detects and satisfies data dependencies among spawned tasks. However, dynamic dependence analysis incurs significant runtime overheads, because the runtime must track task resources and use this information to schedule tasks while avoiding conflicts and races.
We present SCOOP, a compiler that effectively integrates static and dynamic analysis in code generation. SCOOP combines context-sensitive points-to, control-flow, escape, and effect analyses to remove redundant dependence checks at runtime. Our static analysis can work in combination with existing dynamic analyses and task-parallel runtimes that use annotations to specify tasks and their memory footprints. We use our static dependence analysis to detect non-conflicting tasks and an existing dynamic analysis to handle the remaining dependencies. We evaluate the resulting hybrid dependence analysis on a set of task-parallel programs.

Analysis of Dependence Tracking Algorithms for Task Dataflow Execution

Processor architectures has taken a turn towards many-core processors, which integrate multiple processing cores on a single chip to increase overall performance, and there are no signs that this trend will stop in the near future. Many-core processors are harder to program than multi-core and single-core processors due to the need of writing parallel or concurrent programs with high degrees of parallelism. Moreover, many-cores have to operate in a mode of strong scaling because of memory bandwidth constraints. In strong scaling increasingly finer-grain parallelism must be extracted in order to keep all processing cores busy.

Task dataflow programming models have a high potential to simplify parallel program- ming because they alleviate the programmer from identifying precisely all inter-task de- pendences when writing programs. Instead, the task dataflow runtime system detects and enforces inter-task dependences during execution based on the description of memory each task accesses. The runtime constructs a task dataflow graph that captures all tasks and their dependences. Tasks are scheduled to execute in parallel taking into account dependences specified in the task graph.

Several papers report important overheads for task dataflow systems, which severely limits the scalability and usability of such systems. In this paper we study efficient schemes to manage task graphs and analyze their scalability. We assume a programming model that supports input, output and in/out annotations on task arguments, as well as commutative in/out and reductions. We analyze the structure of task graphs and identify versions and generations as key concepts for efficient management of task graphs. Then, we present three schemes to manage task graphs building on graph representations, hypergraphs and lists. We also consider a fourth edge-less scheme that synchronizes tasks using integers. Analysis using micro-benchmarks shows that the graph representation is not always scalable and that the edge-less scheme introduces least overhead in nearly all situations.

Deterministic Scale-Free Pipeline Parallelism with Hyperqueues

Ubiquitous parallel computing aims to make parallel programming accessible to a wide variety of programming areas using deterministic and scale-free programming models built on a task abstraction. However, it remains hard to reconcile these attributes with pipeline parallelism, where the number of pipeline stages is typically hard-coded in the program and defines the degree of parallelism.

This paper introduces hyperqueues, a programming abstraction that enables the construction of deterministic and scale-free pipeline parallel programs. Hyperqueues extend the concept of Cilk++ hyperobjects to provide thread-local views on a shared data structure. While hyperobjects are organized around private local views, hyperqueues require shared concurrent views on the underlying data structure. We define the semantics of hyperqueues and describe their implementation in a work-stealing scheduler. We demonstrate scalable performance on pipeline-parallel PARSEC benchmarks and find that hyperqueues provide comparable or up to 30% better performance than POSIX threads and Intel's Threading Building Blocks. The latter are highly tuned to the number of available processing cores, while programs using hyperqueues are scale-free.