Automatic Compilation of MATLAB Programs for Synergistic Execution on Heterogeneous Processors

MATLAB is an array language, initially popular for rapid prototyping, but is now being increasingly used to develop production code for numerical and scientific applications. Typical MATLAB programs have abundant data parallelism. These programs also have control flow dominated scalar regions that have an impact on the program's execution time. Today's computer systems have tremendous computing power in the form of traditional CPU cores and throughput oriented accelerators such as graphics processing units(GPUs). Thus, an approach that maps the control flow dominated regions to the CPU and the data parallel regions to the GPU can significantly improve program performance. In this paper, we present the design and implementation of MEGHA, a compiler that automatically compiles MATLAB programs to enable synergistic execution on heterogeneous processors. Our solution is fully automated and does not require programmer input for identifying data parallel regions. We propose a set of compiler optimizations tailored for MATLAB. Our compiler identifies data parallel regions of the program and composes them into kernels. The problem of combining statements into kernels is formulated as a constrained graph clustering problem. Heuristics are presented to map identified kernels to either the CPU or GPU so that kernel execution on the CPU and the GPU happens synergistically and the amount of data transfer needed is minimized. In order to ensure required data movement for dependencies across basic blocks, we propose a data flow analysis and edge splitting strategy. Thus our compiler automatically handles composition of kernels, mapping of kernels to CPU and GPU, scheduling and insertion of required data transfer. The proposed compiler was implemented and experimental evaluation using a set of MATLAB benchmarks shows that our approach achieves a geometric mean speedup of 19.8X for data parallel benchmarks over native execution of MATLAB.

Multi-agent search using?sensors with heterogeneous capabilities

In this thesis we address the problem of multi-agent search. We formulate two deploy and search strategies based on optimal deployment of agents in search space so as to maximize the search effectiveness in a single step. We show that a variation of centroidal Voronoi configuration is the optimal deployment. When the agents have sensors with different capabilities, the problem will be heterogeneous in nature. We introduce a new concept namely, generalized Voronoi partition in order to formulate and solve the heterogeneous multi-agent search problem. We address a few theoretical issues such as optimality of deployment, convergence and spatial distributedness of the control law and the search strategies. Simulation experiments are carried out to compare performances of the proposed strategies with a few simple search strategies.

Bounds on Defect Level and Fault Coverage in Linear Analog Circuit Testing

Transfer function coefficients (TFC) are widely used to test linear analog circuits for parametric and catastrophic faults. This paper presents closed form expressions for an upper bound on the defect level (DL) and a lower bound on fault coverage (FC) achievable in TFC based test method. The computed bounds have been tested and validated on several benchmark circuits. Further, application of these bounds to scalable RC ladder networks reveal a number of interesting characteristics. The approach adopted here is general and can be extended to find bounds of DL and FC of other parametric test methods for linear and non-linear circuits.

Streaming FFT on REDEFINE-v2: An Application-Architecture Design Space Exploration

In this paper we explore an implementation of a high-throughput, streaming application on REDEFINE-v2, which is an enhancement of REDEFINE. REDEFINE is a polymorphic ASIC combining the flexibility of a programmable solution with the execution speed of an ASIC. In REDEFINE Compute Elements are arranged in an 8x8 grid connected via a Network on Chip (NoC) called RECONNECT, to realize the various macrofunctional blocks of an equivalent ASIC. For a 1024-FFT we carry out an application-architecture design space exploration by examining the various characterizations of Compute Elements in terms of the size of the instruction store. We further study the impact by using application specific, vectorized FUs. By setting up different partitions of the FFT algorithm for persistent execution on REDEFINE-v2, we derive the benefits of setting up pipelined execution for higher performance. The impact of the REDEFINE-v2 micro-architecture for any arbitrary N-point FFT (N > 4096) FFT is also analyzed. We report the various algorithm-architecture tradeoffs in terms of area and execution speed with that of an ASIC implementation. In addition we compare the performance gain with respect to a GPP.

Performance analysis of a fluid queue with random service rate in discrete time

We consider a fluid queue in discrete time with random service rate. Such a queue has been used in several recent studies on wireless networks where the packets can be arbitrarily fragmented. We provide conditions on finiteness of moments of stationary delay, its Laplace-Stieltjes transform and various approximations under heavy traffic. Results are extended to the case where the wireless link can transmit in only a few slots during a frame.

Opportunistic scheduling of wireless links

We consider the problem of scheduling of a wireless channel (server) to several queues. Each queue has its own link (transmission) rate. The link rate of a queue can vary randomly from slot to slot. The queue lengths and channel states of all users are known at the beginning of each slot. We show the existence of an optimal policy that minimizes the long term (discounted) average sum of queue lengths. The optimal policy, in general needs to be computed numerically. Then we identify a greedy (one step optimal) policy, MAX-TRANS which is easy to implement and does not require the channel and traffic statistics. The cost of this policy is close to optimal and better than other well-known policies (when stable) although it is not throughput optimal for asymmetric systems. We (approximately) identify its stability region and obtain approximations for its mean queue lengths and mean delays. We also modify this policy to make it throughput optimal while retaining good performance.

Sequences for Phase-Encoded Optical CDMA

In phase-encoded optical CDMA (OCDMA) spreading is achieved by encoding the phase of signal spectrum. Here, a mathematical model for the output signal of a phase-encoded OCDMA system is first derived. This is shown to lead to a performance metric for the design of spreading sequences for asynchronous transmission. Generalized bent functions are used to construct a family of efficient phase-encoding sequences. It is shown how M-ary modulation of these spreading sequences is possible. The problem of designing efficient phaseencoded sequences is then related to the problem of minimizing PMEPR (peak-to-mean envelope power ratio) in an OFDM communication system.

On the Production of Anorexic Plan Diagrams

A "plan diagram" is a pictorial enumeration of the execution plan choices of a database query optimizer over the relational selectivity space. We have shown recently that, for industrial-strength database engines, these diagrams are often remarkably complex and dense, with a large number of plans covering the space. However, they can often be reduced to much simpler pictures, featuring significantly fewer plans, without materially affecting the query processing quality. Plan reduction has useful implications for the design and usage of query optimizers, including quantifying redundancy in the plan search space, enhancing useability of parametric query optimization, identifying error-resistant and least-expected-cost plans, and minimizing the overheads of multi-plan approaches. We investigate here the plan reduction issue from theoretical, statistical and empirical perspectives. Our analysis shows that optimal plan reduction, w.r.t. minimizing the number of plans, is an NP-hard problem in general, and remains so even for a storage-constrained variant. We then present a greedy reduction algorithm with tight and optimal performance guarantees, whose complexity scales linearly with the number of plans in the diagram for a given resolution. Next, we devise fast estimators for locating the best tradeoff between the reduction in plan cardinality and the impact on query processing quality. Finally, extensive experimentation with a suite of multi-dimensional TPCH-based query templates on industrial-strength optimizers demonstrates that complex plan diagrams easily reduce to "anorexic" (small absolute number of plans) levels incurring only marginal increases in the estimated query processing costs.

Efficient Algorithms for Computing All Low s-t Edge Connectivities and Related Problems

Given an undirected unweighted graph G = (V, E) and an integer k ≥ 1, we consider the problem of computing the edge connectivities of all those (s, t) vertex pairs, whose edge connectivity is at most k. We present an algorithm with expected running time Õ(m + nk3) for this problem, where |V| = n and |E| = m. Our output is a weighted tree T whose nodes are the sets V1, V2,..., V l of a partition of V, with the property that the edge connectivity in G between any two vertices s ε Vi and t ε Vj, for i ≠ j, is equal to the weight of the lightest edge on the path between Vi and Vj in T. Also, two vertices s and t belong to the same Vi for any i if and only if they have an edge connectivity greater than k. Currently, the best algorithm for this problem needs to compute all-pairs min-cuts in an O(nk) edge graph; this takes Õ(m + n5/2kmin{k1/2, n1/6}) time. Our algorithm is much faster for small values of k; in fact, it is faster whenever k is o(n5/6). Our algorithm yields the useful corollary that in Õ(m + nc3) time, where c is the size of the global min-cut, we can compute the edge connectivities of all those pairs of vertices whose edge connectivity is at most αc for some constant α. We also present an Õ(m + n) Monte Carlo algorithm for the approximate version of this problem. This algorithm is applicable to weighted graphs as well. Our algorithm, with some modifications, also solves another problem called the minimum T-cut problem. Given T ⊆ V of even cardinality, we present an Õ(m + nk3) algorithm to compute a minimum cut that splits T into two odd cardinality components, where k is the size of this cut.

A decidable temporal logic with repeating values

Various logical formalisms with the freeze quantifier have been recently considered to model computer systems even though this is a powerful mechanism that often leads to undecidability. In this paper, we study a linear-time temporal logic with past-time operators such that the freeze operator is only used to express that some value from an infinite set is repeated in the future or in the past. Such a restriction has been inspired by a recent work on spatio-temporal logics. We show decidability of finitary and infinitary satisfiability by reduction into the verification of temporal properties in Petri nets. This is a surprising result since the logic is closed under negation, contains future-time and past-time temporal operators and can express the nonce property and its negation. These ingredients are known to lead to undecidability with a more liberal use of the freeze quantifier.

Register Allocation and Optimal Spill code Scheduling in Software Pipelined Loops using 0-1 Integer Linear Programming Formulation

In achieving higher instruction level parallelism, software pipelining increases the register pressure in the loop. The usefulness of the generated schedule may be restricted to cases where the register pressure is less than the available number of registers. Spill instructions need to be introduced otherwise. But scheduling these spill instructions in the compact schedule is a difficult task. Several heuristics have been proposed to schedule spill code. These heuristics may generate more spill code than necessary, and scheduling them may necessitate increasing the initiation interval. We model the problem of register allocation with spill code generation and scheduling in software pipelined loops as a 0-1 integer linear program. The formulation minimizes the increase in initiation interval (II) by optimally placing spill code and simultaneously minimizes the amount of spill code produced. To the best of our knowledge, this is the first integrated formulation for register allocation, optimal spill code generation and scheduling for software pipelined loops. The proposed formulation performs better than the existing heuristics by preventing an increase in II in 11.11% of the loops and generating 18.48% less spill code on average among the loops extracted from Perfect Club and SPEC benchmarks with a moderate increase in compilation time.

Optimizing Multimedia Experience in a Thin Client Environment for a Resource Constrained Processor

In this paper, we study how TCP and UDP flows interact with each other when the end system is a CPU resource constrained thin client. The problem addressed is twofold, 1) the throughput of TCP flows degrades severely in the presence of heavily loaded UDP flows 2) fairness and minimum QoS requirements of UDP are not maintained. First, we identify the factors affecting the TCP throughput by providing an in-depth analysis of end to end delay and packet loss variations. The results obtained from the first part leads us to our second contribution. We propose and study the use of an algorithm that ensures fairness across flows. The algorithm improves the performance of TCP flows in the presence of multiple UDP flows admitted under an admission algorithm and maintains the minimum QoS requirements of the UDP flows. The advantage of the algorithm is that it requires no changes to TCP/IP stack and control is achieved through receiver window control.

Multiclass Core Vector Machine

Even though several techniques have been proposed in the literature for achieving multiclass classification using Support Vector Machine(SVM), the scalability aspect of these approaches to handle large data sets still needs much of exploration. Core Vector Machine(CVM) is a technique for scaling up a two class SVM to handle large data sets. In this paper we propose a Multiclass Core Vector Machine(MCVM). Here we formulate the multiclass SVM problem as a Quadratic Programming(QP) problem defining an SVM with vector valued output. This QP problem is then solved using the CVM technique to achieve scalability to handle large data sets. Experiments done with several large synthetic and real world data sets show that the proposed MCVM technique gives good generalization performance as that of SVM at a much lesser computational expense. Further, it is observed that MCVM scales well with the size of the data set.

WCET Estimation for Executables in the Presence of Data Caches

This paper describes techniques to estimate the worst case execution time of executable code on architectures with data caches. The underlying mechanism is Abstract Interpretation, which is used for the dual purposes of tracking address computations and cache behavior. A simultaneous numeric and pointer analysis using an abstraction for discrete sets of values computes safe approximations of access addresses which are then used to predict cache behavior using Must Analysis. A heuristic is also proposed which generates likely worst case estimates. It can be used in soft real time systems and also for reasoning about the tightness of the safe estimate. The analysis methods can handle programs with non-affine access patterns, for which conventional Presburger Arithmetic formulations or Cache Miss Equations do not apply. The precision of the estimates is user-controlled and can be traded off against analysis time. Executables are analyzed directly, which, apart from enhancing precision, renders the method language independent.

A Fast Linear Separability Test by Projection of Positive Points on Subspaces

A geometric and non parametric procedure for testing if two finite set of points are linearly separable is proposed. The Linear Separability Test is equivalent to a test that determines if a strictly positive point h > 0 exists in the range of a matrix A (related to the points in the two finite sets). The algorithm proposed in the paper iteratively checks if a strictly positive point exists in a subspace by projecting a strictly positive vector with equal co-ordinates (p), on the subspace. At the end of each iteration, the subspace is reduced to a lower dimensional subspace. The test is completed within r ≤ min(n, d + 1) steps, for both linearly separable and non separable problems (r is the rank of A, n is the number of points and d is the dimension of the space containing the points). The worst case time complexity of the algorithm is O(nr3) and space complexity of the algorithm is O(nd). A small review of some of the prominent algorithms and their time complexities is included. The worst case computational complexity of our algorithm is lower than the worst case computational complexity of Simplex, Perceptron, Support Vector Machine and Convex Hull Algorithms, if d