Loop transformations: convexity, pruning, and optimization

High-level loop transformations are a key instrument in mapping computational kernels to effectively exploit the resources in modern processor architectures. Nevertheless, selecting required compositions of loop transformations to achieve this remains a significantly challenging task; current compilers may be off by orders of magnitude in performance compared to hand-optimized programs. To address this fundamental challenge, we first present a convex characterization of all distinct, semantics-preserving, multidimensional affine transformations. We then bring together algebraic, algorithmic, and performance analysis results to design a tractable optimization algorithm over this highly expressive space. Our framework has been implemented and validated experimentally on a representative set of benchmarks running on state-of-the-art multi-core platforms.

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.

Compiler Assisted Leakage Energy Optimization for Clustered VLIW Architectures

Miniaturization of devices and the ensuing decrease in the threshold voltage has led to a substantial increase in the leakage component of the total processor energy consumption. Relatively simpler issue logic and the presence of a large number of function units in the VLIW and the clustered VLIW architectures attribute a large fraction of this leakage energy consumption in the functional units. However, functional units are not fully utilized in the VLIW architectures because of the inherent variations in the ILP of the programs. This underutilization is even more pronounced in the context of clustered VLIW architectures because of the contentions for the limited number of slow intercluster communication channels which lead to many short idle cycles.In the past, some architectural schemes have been proposed to obtain leakage energy bene .ts by aggressively exploiting the idleness of functional units. However, presence of many short idle cycles cause frequent transitions from the active mode to the sleep mode and vice-versa and adversely a ffects the energy benefits of a purely hardware based scheme. In this paper, we propose and evaluate a compiler instruction scheduling algorithm that assist such a hardware based scheme in the context of VLIW and clustered VLIW architectures. The proposed scheme exploits the scheduling slacks of instructions to orchestrate the functional unit mapping with the objective of reducing the number of transitions in functional units thereby keeping them off for a longer duration. The proposed compiler-assisted scheme obtains a further 12% reduction of energy consumption of functional units with negligible performance degradation over a hardware-only scheme for a VLIW architecture. The benefits are 15% and 17% in the context of a 2-clustered and a 4-clustered VLIW architecture respectively. Our test bed uses the Trimaran compiler infrastructure.

Synthesis of configurable architectures for DSP algorithms

ASICs offer the best realization of DSP algorithms in terms of performance, but the cost is prohibitive, especially when the volumes involved are low. However, if the architecture synthesis trajectory for such algorithms is such that the target architecture can be identified as an interconnection of elementary parameterized computational structures, then it is possible to attain a close match, both in terms of performance and power with respect to an ASIC, for any algorithmic parameters of the given algorithm. Such an architecture is weakly programmable (configurable) and can be viewed as an application specific instruction-set processor (ASIP). In this work, we present a methodology to synthesize ASIPs for DSP algorithms.

Implementation of Art1 and Art2 Artificial Neural Networks on Ring and Mesh Architectures

The Artificial Neural Networks (ANNs) are being used to solve a variety of problems in pattern recognition, robotic control, VLSI CAD and other areas. In most of these applications, a speedy response from the ANNs is imperative. However, ANNs comprise a large number of artificial neurons, and a massive interconnection network among them. Hence, implementation of these ANNs involves execution of computer-intensive operations. The usage of multiprocessor systems therefore becomes necessary. In this article, we have presented the implementation of ART1 and ART2 ANNs on ring and mesh architectures. The overall system design and implementation aspects are presented. The performance of the algorithm on ring, 2-dimensional mesh and n-dimensional mesh topologies is presented. The parallel algorithm presented for implementation of ART1 is not specific to any particular architecture. The parallel algorithm for ARTE is more suitable for a ring architecture.

Simultaneous display of spectrum and image in a single plane, and the realization of a lensless coherent optical processor

A novel method, designated the holographic spectrum reconstruction (HSR) method, is proposed for achieving simultaneous display of the spectrum and image of an object in a single plane. A study of the scaling behaviour of both the spectrum and the image has been carried out and based on this study, it is demonstrated that a lensless coherent optical processor can be realized.

Compiling Techniques for Coarse Grained Runtime Reconfigurable Architectures

In this paper we develop compilation techniques for the realization of applications described in a High Level Language (HLL) onto a Runtime Reconfigurable Architecture. The compiler determines Hyper Operations (HyperOps) that are subgraphs of a data flow graph (of an application) and comprise elementary operations that have strong producer-consumer relationship. These HyperOps are hosted on computation structures that are provisioned on demand at runtime. We also report compiler optimizations that collectively reduce the overheads of data-driven computations in runtime reconfigurable architectures. On an average, HyperOps offer a 44% reduction in total execution time and a 18% reduction in management overheads as compared to using basic blocks as coarse grained operations. We show that HyperOps formed using our compiler are suitable to support data flow software pipelining.

Parallel processor scheduling for minimizing total weighted tardiness using ant colony optimization

In the modern business environment, meeting due dates and avoiding delay penalties are very important goals that can be accomplished by minimizing total weighted tardiness. We consider a scheduling problem in a system of parallel processors with the objective of minimizing total weighted tardiness. Our aim in the present work is to develop an efficient algorithm for solving the parallel processor problem as compared to the available heuristics in the literature and we propose the ant colony optimization approach for this problem. An extensive experimentation is conducted to evaluate the performance of the ACO approach on different problem sizes with the varied tardiness factors. Our experimentation shows that the proposed ant colony optimization algorithm is giving promising results compared to the best of the available heuristics.

A Petri net model for evaluating packet buffering strategies in a network processor

Previous studies have shown that buffering packets in DRAM is a performance bottleneck. In order to understand the impediments in accessing the DRAM, we developed a detailed Petri net model of IP forwarding application on IXP2400 that models the different levels of the memory hierarchy. The cell based interface used to receive and transmit packets in a network processor leads to some small size DRAM accesses. Such narrow accesses to the DRAM expose the bank access latency, reducing the bandwidth that can be realized. With real traces up to 30% of the accesses are smaller than the cell size, resulting in 7.7% reduction in DRAM bandwidth. To overcome this problem, we propose buffering these small chunks of data in the on chip scratchpad memory. This scheme also exploits greater degree of parallelism between different levels of the memory hierarchy. Using real traces from the internet, we show that the transmit rate can be improved by an average of 21% over the base scheme without the use of additional hardware. Further, the impact of different traffic patterns on the network processor resources is studied. Under real traffic conditions, we show that the data bus which connects the off-chip packet buffer to the micro-engines, is the obstacle in achieving higher throughput.

Compiler-assisted instruction decoder energy optimization for clustered VLIW architectures

Traditionally, an instruction decoder is designed as a monolithic structure that inhibit the leakage energy optimization. In this paper, we consider a split instruction decoder that enable the leakage energy optimization. We also propose a compiler scheduling algorithm that exploits instruction slack to increase the simultaneous active and idle duration in instruction decoder. The proposed compiler-assisted scheme obtains a further 14.5% reduction of energy consumption of instruction decoder over a hardware-only scheme for a VLIW architecture. The benefits are 17.3% and 18.7% in the context of a 2-clustered and a 4-clustered VLIW architecture respectively.

Experimental implementation of a three qubit quantum game with corrupt source using nuclear magnetic resonance quantum information processor

In a three player quantum `Dilemma' game each player takes independent decisions to maximize his/her individual gain. The optimal strategy in the quantum version of this game has a higher payoff compared to its classical counterpart. However, this advantage is lost if the initial qubits provided to the players are from a noisy source. We have experimentally implemented the three player quantum version of the `Dilemma' game as described by Johnson, [N.F. Johnson, Phys. Rev. A 63 (2001) 020302(R)] using nuclear magnetic resonance quantum information processor and have experimentally verified that the payoff of the quantum game for various levels of corruption matches the theoretical payoff. (c) 2007 Elsevier Inc. All rights reserved.

High-Throughput Flexible Constraint Length Viterbi Decoders on De Bruijn, Shuffle-Exchange and Butterfly Connected Architectures

Flexible constraint length channel decoders are required for software defined radios. This paper presents a novel scalable scheme for realizing flexible constraint length Viterbi decoders on a de Bruijn interconnection network. Architectures for flexible decoders using the flattened butterfly and shuffle-exchange networks are also described. It is shown that these networks provide favourable substrates for realizing flexible convolutional decoders. Synthesis results for the three networks are provided and a comparison is performed. An architecture based on a 2D-mesh, which is a topology having a nominally lesser silicon area requirement, is also considered as a fourth point for comparison. It is found that of all the networks considered, the de Bruijn network offers the best tradeoff in terms of area versus throughput.

A real coded genetic algorithm for data partitioning and scheduling in networks with arbitrary processor release time

The problem of scheduling divisible loads in distributed computing systems, in presence of processor release time is considered. The objective is to find the optimal sequence of load distribution and the optimal load fractions assigned to each processor in the system such that the processing time of the entire processing load is a minimum. This is a difficult combinatorial optimization problem and hence genetic algorithms approach is presented for its solution.

A suprathreshold stochastic resonance based pre-processor for direction-of-arrival estimation

In this paper we propose a nonlinear preprocessor for enhancing the performance of processors used for direction-of-arrival (DOA) estimation in heavy-tailed non-Gaussian noise. The preprocessor based on the phenomenon of suprathreshold stochastic resonance (SSR), provides SNR gain. The preprocessed data is used for DOA estimation by the MUSIC algorithm. Simulation results are presented to show that the SSR preprocessor provides a significant improvement in the performance of MUSIC in heavy-tailed noise at low SNR.

Analysis of network architectures for zigbee sensor clusters

We provide a comparative performance analysis of network architectures for beacon enabled Zigbee sensor clusters using the CSMA/CA MAC defined in the IEEE 802.15.4 standard, and organised as (i) a star topology, and (ii) a two-hop topology. We provide analytical models for obtaining performance measures such as mean network delay, and mean node lifetime. We find that the star topology is substantially superior both in delay performance and lifetime performance than the two-hop topology.