Hierarchical composite regular parallel architecture

The design space of emerging heterogenous multi-core architectures with re-configurability element makes it feasible to design mixed fine-grained and coarse-grained parallel architectures. This paper presents a hierarchical composite array design which extends the curret design space of regular array design by combining a sequence of transformations. This technique is applied to derive a new design of a pipelined parallel regular array with different dataflow between phases of computation.

Compositional technique for synthesising multi-phase regular arrays

We describe a high-level design method to synthesize multi-phase regular arrays. The method is based on deriving component designs using classical regular (or systolic) array synthesis techniques and composing these separately evolved component design into a unified global design. Similarity transformations ar e applied to component designs in the composition stage in order to align data ow between the phases of the computations. Three transformations are considered: rotation, re ection and translation. The technique is aimed at the design of hardware components for high-throughput embedded systems applications and we demonstrate this by deriving a multi-phase regular array for the 2-D DCT algorithm which is widely used in many vide ocommunications applications.

Dispersion of a point-source release of a passive scalar through an urban-like array for different wind directions

The dispersion of a point-source release of a passive scalar in a regular array of cubical, urban-like, obstacles is investigated by means of direct numerical simulations. The simulations are conducted under conditions of neutral stability and fully rough turbulent flow, at a roughness Reynolds number of Reτ = 500. The Navier–Stokes and scalar equations are integrated assuming a constant rate release from a point source close to the ground within the array. We focus on short-range dispersion, when most of the material is still within the building canopy. Mean and fluctuating concentrations are computed for three different pressure gradient directions (0◦ , 30◦ , 45◦). The results agree well with available experimental data measured in a water channel for a flow angle of 0◦ . Profiles of mean concentration and the three-dimensional structure of the dispersion pattern are compared for the different forcing angles. A number of processes affecting the plume structure are identified and discussed, including: (i) advection or channelling of scalar down ‘streets’, (ii) lateral dispersion by turbulent fluctuations and topological dispersion induced by dividing streamlines around buildings, (iii) skewing of the plume due to flow turning with height, (iv) detrainment by turbulent dispersion or mean recirculation, (v) entrainment and release of scalar in building wakes, giving rise to ‘secondary sources’, (vi) plume meandering due to unsteady turbulent fluctuations. Finally, results on relative concentration fluctuations are presented and compared with the literature for point source dispersion over flat terrain and urban arrays. Keywords Direct numerical simulation · Dispersion modelling · Urban array

Fine-grained multi-phase array designs

Hybrid multiprocessor architectures which combine re-configurable computing and multiprocessors on a chip are being proposed to transcend the performance of standard multi-core parallel systems. Both fine-grained and coarse-grained parallel algorithm implementations are feasible in such hybrid frameworks. A compositional strategy for designing fine-grained multi-phase regular processor arrays to target hybrid architectures is presented in this paper. The method is based on deriving component designs using classical regular array techniques and composing the components into a unified global design. Effective designs with phase-changes and data routing at run-time are characteristics of these designs. In order to describe the data transfer between phases, the concept of communication domain is introduced so that the producer–consumer relationship arising from multi-phase computation can be treated in a unified way as a data routing phase. This technique is applied to derive new designs of multi-phase regular arrays with different dataflow between phases of computation.

Spatial and temporal variability of the concentration field from localized releases in a regular building array

Spatial and temporal fluctuations in the concentration field from an ensemble of continuous point-source releases in a regular building array are analyzed from data generated by direct numerical simulations. The release is of a passive scalar under conditions of neutral stability. Results are related to the underlying flow structure by contrasting data for an imposed wind direction of 0 deg and 45 deg relative to the buildings. Furthermore, the effects of distance from the source and vicinity to the plume centreline on the spatial and temporal variability are documented. The general picture that emerges is that this particular geometry splits the flow domain into segments (e.g. “streets” and “intersections”) in each of which the air is, to a first approximation, well mixed. Notable exceptions to this general rule include regions close to the source, near the plume edge, and in unobstructed channels when the flow is aligned. In the oblique (45 deg) case the strongly three-dimensional nature of the flow enhances mixing of a scalar within the canopy leading to reduced temporal and spatial concentration fluctuations within the plume core. These fluctuations are in general larger for the parallel flow (0 deg) case, especially so in the long unobstructed channels. Due to the more complex flow structure in the canyon-type streets behind buildings, fluctuations are lower than in the open channels, though still substantially larger than for oblique flow. These results are relevant to the formulation of simple models for dispersion in urban areas and to the quantification of the uncertainties in their predictions.

Spatial variability of flow statistics within regular building arrays

Turbulence statistics obtained by direct numerical simulations are analysed to investigate spatial heterogeneity within regular arrays of building-like cubical obstacles. Two different array layouts are studied, staggered and square, both at a packing density of λp=0.25 . The flow statistics analysed are mean streamwise velocity ( u− ), shear stress ( u′w′−−−− ), turbulent kinetic energy (k) and dispersive stress fraction ( u˜w˜ ). The spatial flow patterns and spatial distribution of these statistics in the two arrays are found to be very different. Local regions of high spatial variability are identified. The overall spatial variances of the statistics are shown to be generally very significant in comparison with their spatial averages within the arrays. Above the arrays the spatial variances as well as dispersive stresses decay rapidly to zero. The heterogeneity is explored further by separately considering six different flow regimes identified within the arrays, described here as: channelling region, constricted region, intersection region, building wake region, canyon region and front-recirculation region. It is found that the flow in the first three regions is relatively homogeneous, but that spatial variances in the latter three regions are large, especially in the building wake and canyon regions. The implication is that, in general, the flow immediately behind (and, to a lesser extent, in front of) a building is much more heterogeneous than elsewhere, even in the relatively dense arrays considered here. Most of the dispersive stress is concentrated in these regions. Considering the experimental difficulties of obtaining enough point measurements to form a representative spatial average, the error incurred by degrading the sampling resolution is investigated. It is found that a good estimate for both area and line averages can be obtained using a relatively small number of strategically located sampling points.

Precise analysis of array usage in scientific programs

The automatic transformation of sequential programs for efficient execution on parallel computers involves a number of analyses and restructurings of the input. Some of these analyses are based on computing array sections, a compact description of a range of array elements. Array sections describe the set of array elements that are either read or written by program statements. These sections can be compactly represented using shape descriptors such as regular sections, simple sections, or generalized convex regions. However, binary operations such as Union performed on these representations do not satisfy a straightforward closure property, e.g., if the operands to Union are convex, the result may be nonconvex. Approximations are resorted to in order to satisfy this closure property. These approximations introduce imprecision in the analyses and, furthermore, the imprecisions resulting from successive operations have a cumulative effect. Delayed merging is a technique suggested and used in some of the existing analyses to minimize the effects of approximation. However, this technique does not guarantee an exact solution in a general setting. This article presents a generalized technique to precisely compute Union which can overcome these imprecisions.

Development of multi-electrode array screening for anticonvulsants in acute rat brain slices

The acute hippocampal brain slice preparation is an important in vitro screening tool for potential anticonvulsants. Application of 4-aminopyridine (4-AP) or removal of external Mg2+ ions induces epileptiform bursting in slices which is analogous to electrical brain activity seen in status epilepticus states. We have developed these epileptiform models for use with multi-electrode arrays (MEAs), allowing recording across the hippocampal slice surface from 59 points. We present validation of this novel approach and analyses using two anticonvulsants, felbamate and phenobarbital, the effects of which have already been assessed in these models using conventional extracellular recordings. In addition to assessing drug effects on commonly described parameters (duration, amplitude and frequency), we describe novel methods using the MEA to assess burst propagation speeds and the underlying frequencies that contribute to the epileptiform activity seen. Contour plots are also used as a method of illustrating burst activity. Finally, we describe hitherto unreported properties of epileptiform bursting induced by 100M4-AP or removal of external Mg2+ ions. Specifically, we observed decreases over time in burst amplitude and increase over time in burst frequency in the absence of additional pharmacological interventions. These MEA methods enhance the depth, quality and range of data that can be derived from the hippocampal slice preparation compared to conventional extracellular recordings. It may also uncover additional modes of action that contribute to anti-epileptiform drug effects

Generic systolic array for genetic algorithms

The authors present a systolic design for a simple GA mechanism which provides high throughput and unidirectional pipelining by exploiting the inherent parallelism in the genetic operators. The design computes in O(N+G) time steps using O(N2) cells where N is the population size and G is the chromosome length. The area of the device is independent of the chromosome length and so can be easily scaled by replicating the arrays or by employing fine-grain migration. The array is generic in the sense that it does not rely on the fitness function and can be used as an accelerator for any GA application using uniform crossover between pairs of chromosomes. The design can also be used in hybrid systems as an add-on to complement existing designs and methods for fitness function acceleration and island-style population management

Synthesis of a systolic array genetic algorithm

The paper presents a design for a hardware genetic algorithm which uses a pipeline of systolic arrays. These arrays have been designed using systolic synthesis techniques which involve expressing the algorithm as a set of uniform recurrence relations. The final design divorces the fitness function evaluation from the hardware and can process chromosomes of different lengths, giving the design a generic quality. The paper demonstrates the design methodology by progressively re-writing a simple genetic algorithm, expressed in C code, into a form from which systolic structures can be deduced. This paper extends previous work by introducing a simplification to a previous systolic design for the genetic algorithm. The simplification results in the removal of 2N 2 + 4N cells and reduces the time complexity by 3N + 1 cycles.

The systolic array genetic algorithm, an example of systolic arrays as a reconfigurable design methodology

We advocate the use of systolic design techniques to create custom hardware for Custom Computing Machines. We have developed a hardware genetic algorithm based on systolic arrays to illustrate the feasibility of the approach. The architecture is independent of the lengths of chromosomes used and can be scaled in size to accommodate different population sizes. An FPGA prototype design can process 16 million genes per second.