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.

Dynamics of a class of social interaction systems

A mathematical model of social interaction in the form of two coupler! first-order non-linear differential equations, forms the topic of this study. This non-conservative model io representative of such varied social interaction problems as coexisting sub-populations of two different species, arms race between two rival countries and the like. Differential transformation techniques developed elsewhere in the literature are seen to be effective tools of dynamic analysis of this non-linear non-conservative mode! of social interaction process.

Integral treatment for the representation of thermodynamic properties in multicomponent systems using interaction parameters

The partial thermodynamic functions of the solvent component of a ternary system have been deduced in terms of the interaction parameters by integration of several series which emerge from the Maclaurin infinite series based on the integral property of the system and subjected to appropriate boundary conditions. The series integration shows that the resulting partial functions are suitable for interpreting the thermodynamic properties of the system and are independent of compositional paths. In the present analysis, the higher order terms of these series are found to make insignificant contributions.

Representation of thermodynamic properties of quaternary systems using interaction parameters

Integral excess free energy of a quaternary system has been expressed in terms of the MacLaurin infinite series. The series is subjected to appropriate boundary conditions and each of the derivatives correlated to the corresponding interaction coefficients. The derivation of the partial functions involves extensive summation of various infinite series pertaining to the first order and quaternary parameters to remove any truncational error. The thermodynamic consistency of the derived partials has been established based on the Gibbs-Duhem relations. The equations are used to interpret the thermodynamic properties of the Fe-Cr-Ni-N system.

MAX: A Multi Objective Memory Architecture eXploration Framework for Embedded Systems-on-Chip

Today's feature-rich multimedia products require embedded system solution with complex System-on-Chip (SoC) to meet market expectations of high performance at a low cost and lower energy consumption. The memory architecture of the embedded system strongly influences these parameters. Hence the embedded system designer performs a complete memory architecture exploration. This problem is a multi-objective optimization problem and can be tackled as a two-level optimization problem. The outer level explores various memory architecture while the inner level explores placement of data sections (data layout problem) to minimize memory stalls. Further, the designer would be interested in multiple optimal design points to address various market segments. However, tight time-to-market constraints enforces short design cycle time. In this paper we address the multi-level multi-objective memory architecture exploration problem through a combination of Multi-objective Genetic Algorithm (Memory Architecture exploration) and an efficient heuristic data placement algorithm. At the outer level the memory architecture exploration is done by picking memory modules directly from a ASIC memory Library. This helps in performing the memory architecture exploration in a integrated framework, where the memory allocation, memory exploration and data layout works in a tightly coupled way to yield optimal design points with respect to area, power and performance. We experimented our approach for 3 embedded applications and our approach explores several thousand memory architecture for each application, yielding a few hundred optimal design points in a few hours of computation time on a standard desktop.

Interference in Multiprocessor Systems with Localized Memory Access Probabilities

Past studies of memory interference in multiprocessor systems have generally assumed that the references of each processor are uniformly distributed among the memory modules. In this paper we develop a model with local referencing, which reflects more closely the behavior of real-life programs. This model is analyzed using Markov chain techniques and expressions are derived for the multiprocessor performance. New expressions are also obtained for the performance in the traditional uniform reference model and are compared with other expressions-available in the literature. Results of a simulation study are given to show the accuracy of the expressions for both models.

Memory effect in SrRu(1-x)O3 (0.01 < x < 0.07)

SrRuO3 is widely known to be an itinerant ferromagnet with a T-C similar to 160 K. It is well known that glassy materials exhibit time dependent phenomena such as memory effect due to their generic slow dynamics. However, for the first time, we have observed memory effect in SrRu(1-x)O3 (0.01memory etc. arise in disordered glassy systems. Thus the observation of memory effect in case of an itinerant ferromagnetic system like SrRuO3 is quite strange. The emergence of such unusual magnetic response is strongly believed to be connected with a cryptic interactions arises in the low temperature. Our effort on neutron diffraction study has been able to trace the cause of such hidden magnetic interaction responsible for bringing glassiness in a ferromagnet.

Energy Efficient Memory Decoder Design for Ultra-Low Voltage Systems

This paper presents a low energy memory decoder architecture for ultra-low-voltage systems containing multiple voltage domains. Due to limitations in scalability of memory supply voltages, these systems typically contain a core operating at subthreshold voltages and memories operating at a higher voltage. This difference in voltage provides a timing slack on the memory path as the core supply is scaled. The paper analyzes the feasibility and trade-offs in utilizing this timing slack to operate a greater section of memory decoder circuitry at the lower supply. A 256x16-bit SRAM interface has been designed in UMC 65nm low-leakage process to evaluate the above technique with the core and memory operating at 280 mV and 500 mV respectively. The technique provides a reduction of up to 20% in energy/cycle of the row decoder without any penalty in area and system-delay.

Ultrasonic attenuation in strongly disordered electronic systems

The attenuation of long-wavelength phonons due to their interaction with electronic excitations in disordered systems is investigated here. Lattice strain couples to electronic stress, and thus ultrasonic attenuation measures electronic viscosity. The enhancement and critical divergence of electronic viscosity due to localization effects is calculated for the first time. Experimental consequences for the anomalous increase of ultrasonic attenuation in disordered metals close to the metal-insulator transition are discussed. In the localized regime, the appropriate model is one of electronic two-level systems (TLS’s) coupled to phonons. The TLS consists of a pair of states with one localized state occupied and the other unoccupied. The density of such low-excitation-energy TLS’s is nonzero due to long-range Coulomb interactions. The question of whether these could be significant low-energy excitations in glasses is touched upon.

Constitutive modeling of shape memory alloy wire with non-local rate kinetics

A constitutive modeling approach for shape memory alloy (SMA) wire by taking into account the microstructural phase inhomogeneity and the associated solid-solid phase transformation kinetics is reported in this paper. The approach is applicable to general thermomechanical loading. Characterization of various scales in the non-local rate sensitive kinetics is the main focus of this paper. Design of SMA materials and actuators not only involve an optimal exploitation of the hysteresis loops during loading-unloading, but also accounts for fatigue and training cycle identifications. For a successful design of SMA integrated actuator systems, it is essential to include the microstructural inhomogeneity effects and the loading rate dependence of the martensitic evolution, since these factors play predominant role in fatigue. In the proposed formulation, the evolution of new phase is assumed according to Weibull distribution. Fourier transformation and finite difference methods are applied to arrive at the analytical form of two important scaling parameters. The ratio of these scaling parameters is of the order of 10(6) for stress-free temperature-induced transformation and 10(4) for stress-induced transformation. These scaling parameters are used in order to study the effect of microstructural variation on the thermo-mechanical force and interface driving force. It is observed that the interface driving force is significant during the evolution. Increase in the slopes of the transformation start and end regions in the stress-strain hysteresis loop is observed for mechanical loading with higher rates.

Sensing Shape Recovery Using Conductivity Noise in Thin Films of NiTi Shape Memory Alloys

Low frequency fluctuations in the electrical resistivity, or noise, have been used as a sensitive tool to probe into the temperature driven martensite transition in dc magnetron sputtered thin films of nickel titanium shape-memory alloys. Even in the equilibrium or static case, the noise magnitude was more than nine orders of magnitude larger than conventional metallic thin films and had a characteristic dependence on temperature. We observe that the noise while the temperature is being ramped is far larger as compared to the equilibrium noise indicating the sensitivity of electrical resistivity to the nucleation and propagation of domains during the shape recovery. Further, the higher order statistics suggests the existence of long range correlations during the transition. This new characterization is based on the kinetics of disorder in the system and separate from existing techniques and can be integrated to many device applications of shape memory alloys for in-situ shape recovery sensing.

Phased-Mission Analysis for Evaluating the Effectiveness of Aerospace Computing-Systems

Effectiveness evaluation of aerospace fault-tolerant computing systems used in a phased-mission environment is rather tricky and difficult because of the interaction of its several degraded performance levels with the multiple objectives of the mission and the use environment. Part I uses an approach based on multiobjective phased-mission analysis to evaluate the effectiveness of a distributed avionics architecture used in a transport aircraft. Part II views the computing system as a multistate s-coherent structure. Lower bounds on the probabilities of accomplishing various levels of performance are evaluated.

Identification of a class of non-linear systems

This paper considers the on-line identification of a non-linear system in terms of a Hammerstein model, with a zero-memory non-linear gain followed by a linear system. The linear part is represented by a Laguerre expansion of its impulse response and the non-linear part by a polynomial. The identification procedure involves determination of the coefficients of the Laguerre expansion of correlation functions and an iterative adjustment of the parameters of the non-linear gain by gradient methods. The method is applicable to situations involving a wide class of input signals. Even in the presence of additive correlated noise, satisfactory performance is achieved with the variance of the error converging to a value close to the variance of the noise. Digital computer simulation establishes the practicability of the scheme in different situations.

On a generalized dynamic model of bi-state social interaction processes

A non-linear model, construed as a generalized version of the models put forth earlier for the study of bi-state social interaction processes, is proposed in this study. The feasibility of deriving the dynamics of such processes is demonstrated by establishing equivalence between the non-linear model and a higher order linear model.