Hierarchical Directory-Based Shared Memory Architecture

We describe the design of a directory-based shared memory architecture on a hierarchical network of hypercubes. The distributed directory scheme comprises two separate hierarchical networks for handling cache requests and transfers. Further, the scheme assumes a single address space and each processing element views the entire network as contiguous memory space. The size of individual directories stored at each node of the network remains constant throughout the network. Although the size of the directory increases with the network size, the architecture is scalable. The results of the analytical studies demonstrate superior performance characteristics of our scheme compared with those of other schemes.

SHAMP: An experimental shared memory multimicroprocessor system for performance evaluation of parallel algorithms

In this paper, the design and implementation of a single shared bus, shared memory multiprocessing system using Intel's single board computers is presented. The hardware configuration and the operating system developed to execute the parallel algorithms are discussed. The performance evaluation studies carried out on Image are outlined.

Design and evaluation of a dual-microcomputer shared memory system with a shared I/O bus

The design, implementation and evaluation are described of a dual-microcomputer system based on the concept of shared memory. Shared memory is useful for passing large blocks of data and it also provides a means to hold and work with shared data. In addition to the shared memory, a separate bus between the I/O ports of the microcomputers is provided. This bus is utilized for interprocessor synchronization. Software routines helpful in applying the dual-microcomputer system to realistic problems are presented. Performance evaluation of the system is carried out using benchmarks.

CUDA-for-clusters: a system for efficient execution of CUDA kernels on multi-core clusters

Rapid advancements in multi-core processor architectures coupled with low-cost, low-latency, high-bandwidth interconnects have made clusters of multi-core machines a common computing resource. Unfortunately, writing good parallel programs that efficiently utilize all the resources in such a cluster is still a major challenge. Various programming languages have been proposed as a solution to this problem, but are yet to be adopted widely to run performance-critical code mainly due to the relatively immature software framework and the effort involved in re-writing existing code in the new language. In this paper, we motivate and describe our initial study in exploring CUDA as a programming language for a cluster of multi-cores. We develop CUDA-For-Clusters (CFC), a framework that transparently orchestrates execution of CUDA kernels on a cluster of multi-core machines. The well-structured nature of a CUDA kernel, the growing popularity, support and stability of the CUDA software stack collectively make CUDA a good candidate to be considered as a programming language for a cluster. CFC uses a mixture of source-to-source compiler transformations, a work distribution runtime and a light-weight software distributed shared memory to manage parallel executions. Initial results on running several standard CUDA benchmark programs achieve impressive speedups of up to 7.5X on a cluster with 8 nodes, thereby opening up an interesting direction of research for further investigation.

Variable Granularity Access Tracking Scheme for Improving the Performance of Software Transactional Memory

Software transactional memory (STM) has been proposed as a promising programming paradigm for shared memory multi-threaded programs as an alternative to conventional lock based synchronization primitives. Typical STM implementations employ a conflict detection scheme, which works with uniform access granularity, tracking shared data accesses either at word/cache line or at object level. It is well known that a single fixed access tracking granularity cannot meet the conflicting goals of reducing false conflicts without impacting concurrency adversely. A fine grained granularity while improving concurrency can have an adverse impact on performance due to lock aliasing, lock validation overheads, and additional cache pressure. On the other hand, a coarse grained granularity can impact performance due to reduced concurrency. Thus, in general, a fixed or uniform granularity access tracking (UGAT) scheme is application-unaware and rarely matches the access patterns of individual application or parts of an application, leading to sub-optimal performance for different parts of the application(s). In order to mitigate the disadvantages associated with UGAT scheme, we propose a Variable Granularity Access Tracking (VGAT) scheme in this paper. We propose a compiler based approach wherein the compiler uses inter-procedural whole program static analysis to select the access tracking granularity for different shared data structures of the application based on the application's data access pattern. We describe our prototype VGAT scheme, using TL2 as our STM implementation. Our experimental results reveal that VGAT-STM scheme can improve the application performance of STAMP benchmarks from 1.87% to up to 21.2%.

WOMM: A Weak Operational Memory Model

Memory models of shared memory concurrent programs define the values a read of a shared memory location is allowed to see. Such memory models are typically weaker than the intuitive sequential consistency semantics to allow efficient execution. In this paper, we present WOMM (abbreviation for Weak Operational Memory Model) that formally unifies two sources of weak behavior in hardware memory models: reordering of instructions and weakly consistent memory. We show that a large number of optimizations are allowed by WOMM. We also show that WOMM is weaker than a number of hardware memory models. Consequently, if a program behaves correctly under WOMM, it will be correct with respect to those hardware memory models. Hence, WOMM can be used as a formally specified abstraction of the hardware memory models. Moreover; unlike most weak memory models, WOMM is described using operational semantics, making it easy to integrate into a model checker for concurrent programs. We further show that WOMM has an important property - it has sequential consistency semantics for datarace-free programs.

Hybrid aggregated-vector algorithm for efficient parallelization of fast multipole method

An efficient parallelization algorithm for the Fast Multipole Method which aims to alleviate the parallelization bottleneck arising from lower job-count closer to root levels is presented. An electrostatic problem of 12 million non-uniformly distributed mesh elements is solved with 80-85% parallel efficiency in matrix setup and matrix-vector product using 60GB and 16 threads on shared memory architecture.

A new parallel algorithm for parsing arithmetic infix expressions

A new parallel algorithm for transforming an arithmetic infix expression into a par se tree is presented. The technique is based on a result due to Fischer (1980) which enables the construction of the parse tree, by appropriately scanning the vector of precedence values associated with the elements of the expression. The algorithm presented here is suitable for execution on a shared memory model of an SIMD machine with no read/write conflicts permitted. It uses O(n) processors and has a time complexity of O(log2n) where n is the expression length. Parallel algorithms for generating code for an SIMD machine are also presented.

Analyzing Cache Performance Bottlenecks of STM Applications and addressing them with Compiler's help

Software transactional memory (STM) is a promising programming paradigm for shared memory multithreaded programs as an alternative to traditional lock based synchronization. However adoption of STM in mainstream software has been quite low due to its considerable overheads and its poor cache/memory performance. In this paper, we perform a detailed study of the cache behavior of STM applications and quantify the impact of different STM factors on the cache misses experienced by the applications. Based on our analysis, we propose a compiler driven Lock-Data Colocation (LDC), targeted at reducing the cache overheads on STM. We show that LDC is effective in improving the cache behavior of STM applications by reducing the dcache miss latency and improving execution time performance.

Believe it or Not! Multicore CPUs can Match GPUs for FLOP-intensive Applications

In this work, we evaluate performance of a real-world image processing application that uses a cross-correlation algorithm to compare a given image with a reference one. The algorithm processes individual images represented as 2-dimensional matrices of single-precision floating-point values using O(n4) operations involving dot-products and additions. We implement this algorithm on a nVidia GTX 285 GPU using CUDA, and also parallelize it for the Intel Xeon (Nehalem) and IBM Power7 processors, using both manual and automatic techniques. Pthreads and OpenMP with SSE and VSX vector intrinsics are used for the manually parallelized version, while a state-of-the-art optimization framework based on the polyhedral model is used for automatic compiler parallelization and optimization. The performance of this algorithm on the nVidia GPU suffers from: (1) a smaller shared memory, (2) unaligned device memory access patterns, (3) expensive atomic operations, and (4) weaker single-thread performance. On commodity multi-core processors, the application dataset is small enough to fit in caches, and when parallelized using a combination of task and short-vector data parallelism (via SSE/VSX) or through fully automatic optimization from the compiler, the application matches or beats the performance of the GPU version. The primary reasons for better multi-core performance include larger and faster caches, higher clock frequency, higher on-chip memory bandwidth, and better compiler optimization and support for parallelization. The best performing versions on the Power7, Nehalem, and GTX 285 run in 1.02s, 1.82s, and 1.75s, respectively. These results conclusively demonstrate that, under certain conditions, it is possible for a FLOP-intensive structured application running on a multi-core processor to match or even beat the performance of an equivalent GPU version.

Parallel Computation of 3D Morse-Smale Complexes

The Morse-Smale complex is a topological structure that captures the behavior of the gradient of a scalar function on a manifold. This paper discusses scalable techniques to compute the Morse-Smale complex of scalar functions defined on large three-dimensional structured grids. Computing the Morse-Smale complex of three-dimensional domains is challenging as compared to two-dimensional domains because of the non-trivial structure introduced by the two types of saddle criticalities. We present a parallel shared-memory algorithm to compute the Morse-Smale complex based on Forman's discrete Morse theory. The algorithm achieves scalability via synergistic use of the CPU and the GPU. We first prove that the discrete gradient on the domain can be computed independently for each cell and hence can be implemented on the GPU. Second, we describe a two-step graph traversal algorithm to compute the 1-saddle-2-saddle connections efficiently and in parallel on the CPU. Simultaneously, the extremasaddle connections are computed using a tree traversal algorithm on the GPU.

Making STMs Cache Friendly with Compiler Transformations

Software transactional memory (STM) is a promising programming paradigm for shared memory multithreaded programs. In order for STMs to be adopted widely for performance critical software, understanding and improving the cache performance of applications running on STM becomes increasingly crucial, as the performance gap between processor and memory continues to grow. In this paper, we present the most detailed experimental evaluation to date, of the cache behavior of STM applications and quantify the impact of the different STM factors on the cache misses experienced by the applications. We find that STMs are not cache friendly, with the data cache stall cycles contributing to more than 50% of the execution cycles in a majority of the benchmarks. We find that on an average, misses occurring inside the STM account for 62% of total data cache miss latency cycles experienced by the applications and the cache performance is impacted adversely due to certain inherent characteristics of the STM itself. The above observations motivate us to propose a set of specific compiler transformations targeted at making the STMs cache friendly. We find that STM's fine grained and application unaware locking is a major contributor to its poor cache behavior. Hence we propose selective Lock Data co-location (LDC) and Redundant Lock Access Removal (RLAR) to address the lock access misses. We find that even transactions that are completely disjoint access parallel, suffer from costly coherence misses caused by the centralized global time stamp updates and hence we propose the Selective Per-Partition Time Stamp (SPTS) transformation to address this. We show that our transformations are effective in improving the cache behavior of STM applications by reducing the data cache miss latency by 20.15% to 37.14% and improving execution time by 18.32% to 33.12% in five of the 8 STAMP applications.

Dataflow analysis for data-race free programs

Memory models for shared-memory concurrent programming languages typically guarantee sequential consistency (SC) semantics for datarace-free (DRF) programs, while providing very weak or no guarantees for non-DRF programs. In effect programmers are expected to write only DRF programs, which are then executed with SC semantics. With this in mind, we propose a novel scalable solution for dataflow analysis of concurrent programs, which is proved to be sound for DRF programs with SC semantics. We use the synchronization structure of the program to propagate dataflow information among threads without requiring to consider all interleavings explicitly. Given a dataflow analysis that is sound for sequential programs and meets certain criteria, our technique automatically converts it to an analysis for concurrent programs.

Reconciling transactional conflicts with compiler's help

Software transactional memory(STM) is a promising programming paradigm for shared memory multithreaded programs. While STM offers the promise of being less error-prone and more programmer friendly compared to traditional lock-based synchronization, it also needs to be competitive in performance in order for it to be adopted in mainstream software. A major source of performance overheads in STM is transactional aborts. Conflict resolution and aborting a transaction typically happens at the transaction level which has the advantage that it is automatic and application agnostic. However it has a substantial disadvantage in that STM declares the entire transaction as conflicting and hence aborts it and re-executes it fully, instead of partially re-executing only those part(s) of the transaction, which have been affected due to the conflict. This "Re-execute Everything" approach has a significant adverse impact on STM performance. In order to mitigate the abort overheads, we propose a compiler aided Selective Reconciliation STM (SR-STM) scheme, wherein certain transactional conflicts can be reconciled by performing partial re-execution of the transaction. Ours is a selective hybrid approach which uses compiler analysis to identify those data accesses which are legal and profitable candidates for reconciliation and applies partial re-execution only to these candidates selectively while other conflicting data accesses are handled by the default STM approach of abort and full re-execution. We describe the compiler analysis and code transformations required for supporting selective reconciliation. We find that SR-STM is effective in reducing the transactional abort overheads by improving the performance for a set of five STAMP benchmarks by 12.58% on an average and up to 22.34%.

Generating Efficient Data Movement Code for Heterogeneous Architectures with Distributed-Memory

Programming for parallel architectures that do not have a shared address space is extremely difficult due to the need for explicit communication between memories of different compute devices. A heterogeneous system with CPUs and multiple GPUs, or a distributed-memory cluster are examples of such systems. Past works that try to automate data movement for distributed-memory architectures can lead to excessive redundant communication. In this paper, we propose an automatic data movement scheme that minimizes the volume of communication between compute devices in heterogeneous and distributed-memory systems. We show that by partitioning data dependences in a particular non-trivial way, one can generate data movement code that results in the minimum volume for a vast majority of cases. The techniques are applicable to any sequence of affine loop nests and works on top of any choice of loop transformations, parallelization, and computation placement. The data movement code generated minimizes the volume of communication for a particular configuration of these. We use a combination of powerful static analyses relying on the polyhedral compiler framework and lightweight runtime routines they generate, to build a source-to-source transformation tool that automatically generates communication code. We demonstrate that the tool is scalable and leads to substantial gains in efficiency. On a heterogeneous system, the communication volume is reduced by a factor of 11X to 83X over state-of-the-art, translating into a mean execution time speedup of 1.53X. On a distributed-memory cluster, our scheme reduces the communication volume by a factor of 1.4X to 63.5X over state-of-the-art, resulting in a mean speedup of 1.55X. In addition, our scheme yields a mean speedup of 2.19X over hand-optimized UPC codes.