This paper presents a high performance frequency domain filter implementation for a moving window-type processing. The computational structure consists of three stages: a sliding discrete Fourier transform (SDFT) for ...
详细信息
This paper presents a high performance frequency domain filter implementation for a moving window-type processing. The computational structure consists of three stages: a sliding discrete Fourier transform (SDFT) for a vectorized updating of the DFT; a frequency domain filter; and a one-point inverse discrete Fourier transform (IDFT). The total computation required for generating one filtered output point is 2/spl times/N multiplications (N is the frequency window length) and 3/spl times/N additions compared to 2/spl times/N/spl times/log/sub 2/N multiplications and additions if using FFT and IFFT. The proposed structure also has the advantage of being parallel in nature and can be used in various real-time frequency processing, continuous dataflow, single or multiple channel applications.
The ability to dynamically change data input to a computation is a key feature enabling simulation to be used in many applications. In this study, computation of geophysical mass flow is updated on the fly by changing...
详细信息
The ability to dynamically change data input to a computation is a key feature enabling simulation to be used in many applications. In this study, computation of geophysical mass flow is updated on the fly by changing terrain data. Accommodating such changes in a parallel environment entails new developments in parallel data management and gridding. Adaptivity, and in particular unrefinement, is critical for maintaining parallel efficiency. The application under study in this work is the result of a multidisciplinary collaboration between engineers, mathematicians, geologists, and hazard assessment personnel. In addition, adaptive gridding enables efficient use of computational resources, allowing for run-time determination of optimal computing resources. Combining these attributes allows run time conditions to inform calculations, which in turn provide up-to-date information to hazard management personnel.
data now (DF) architecture resulting from the DF program representation by the functional language being developed at the Department of Computers and Informatics is proposed in the paper. The contribution is concentra...
详细信息
data now (DF) architecture resulting from the DF program representation by the functional language being developed at the Department of Computers and Informatics is proposed in the paper. The contribution is concentrated to the efficient implementation of functional language in the DF architecture supporting the idea of the programming parallel problems by specification and transformation using mathematical methods. The combination of a local control flow with a global dataflow model in the proposed DF architecture is connected with the direct operand matching mechanism using matching vectors. The concepts of the dataflow graph representation using the functional language, the approaches to the parallel execution of the functional program in the data now implementational environment and the DF architecture layout are presented in the paper. It is supposed that the hardware implementation of the architecture can be used as a specialised accelerator in problem-oriented computer systems with high requirements on the operation speed.
The radio frequency spectrum, a scarce resource in mobile communications, has to be efficiently utilized with the objective of increasing the network capacity and minimizing the interference. A variety of channel assi...
详细信息
The radio frequency spectrum, a scarce resource in mobile communications, has to be efficiently utilized with the objective of increasing the network capacity and minimizing the interference. A variety of channel assignment strategies have been developed to achieve these objectives. Not only the mobility of the wireless user but also different types of wireless multimedia services create uncertainty in demand and non-optimum use of radio resources. As the cell sizes get smaller, there is a greater need for efficient channel assignment algorithms which are desired to dynamically balance the load of the system by performing reassignments when needed. Static schemes are no longer desirable for small cell systems under heavy and non-uniform multi-rate traffic. We propose a dynamic multi-channel assignment algorithm where the assignment decision is assisted by the mobiles. Our algorithm is based on the concept of network flows, which provide us with the framework to uniformly handle all the events occurring in the system including possible degradation and improvement of the existing assignments. This model enables fast local computations and avoids global reconfiguration which prevent other algorithms being practical.
The arithmetic pipeline analysis package (APAP), an interactive PC-based software package for simulating dataflow within an arithmetic pipeline, is described. The package is designed for use by undergraduate and grad...
详细信息
The arithmetic pipeline analysis package (APAP), an interactive PC-based software package for simulating dataflow within an arithmetic pipeline, is described. The package is designed for use by undergraduate and graduate students with background in computer architecture and parallel processing. It allows the student to design hardware and analyze its functionality without building the circuit. In addition, APAP shows detailed dataflow which makes it easy for the student to envision exactly how the pipeline works. The main simulation process used follows five major steps: construction of the arithmetic pipeline, specification of pipeline connections, definition of the reservation table, definition of the input data, and calculation and presentation of dataflow. In order to provide the student with a detailed dataflow to further understand the entire operating condition, APAP allows the student to choose between a flow diagram, flow table, or flow analysis trace to analyze the results.< >
The authors present a systematic procedure to find the exact iteration bound for an arbitrary nonterminating dataflow graph (DFG). The complexity of the procedure to identify the critical loop and the iteration bound...
详细信息
The authors present a systematic procedure to find the exact iteration bound for an arbitrary nonterminating dataflow graph (DFG). The complexity of the procedure to identify the critical loop and the iteration bound is O(n/sup 3/ log n), where n is the number of nodes in the DFG. Lookahead is another technique to enhance concurrency. It inserts delays in the critical loop of the DFG, resulting in better performance than the iteration bound. To achieve the maximum degree of improvement it is further necessary to identify the next-critical loop. A method to identify the next-critical loop of a nonterminating DFG is also presented.< >
The authors present a technique for analyzing the time cost behavior of data-flow programs that are loop free or contain non tested loops that meet certain constraints. It is assumed that the underlying computer syste...
详细信息
The authors present a technique for analyzing the time cost behavior of data-flow programs that are loop free or contain non tested loops that meet certain constraints. It is assumed that the underlying computer system is a dynamic data-flow computer which has sufficient processors to support the execution of data-flow programs. The token model is modified to describe both the execution and time cost behavior. The technique is similar to some flow-analysis and graph-reduction techniques. In analyzing the time cost of a data-flow program, the technique first determines the number of tokens that pass through each edge in the program. All nodes that are neither input nor output nodes are then deleted. Every time a node or an edge is deleted, the time cost associated with it is properly distributed within the program. Since the resultant program only contains edges that connect input and output nodes, the time cost of the program can be easily obtained. Two examples are given to show the use of the technique.< >
The paper evaluates three improved methods for static allocation of program modules into processing elements of a macro data-flow computer model (MDFCM). A vertically layered algorithm is presented first. We have impr...
详细信息
The paper evaluates three improved methods for static allocation of program modules into processing elements of a macro data-flow computer model (MDFCM). A vertically layered algorithm is presented first. We have improved it with the determination of operands that are triggered during execution of modules. The second algorithm is based on execution levels of modules. We have used E1 and L1 execution levels. The static allocation algorithm gives priority of allocation of program modules on ascending order of L1 or E1 level (depending on the algorithm). Evaluation is done by the simulation of MDFCM. The results have shown that the first algorithm is the best one according to program execution time.< >
data-intensive flowcomputing allows efficient processing of large volumes of data otherwise unapproachable. This paper introduces a new semantic-driven data-intensive flow infrastructure which: (1) provides a robust ...
详细信息
data-intensive flowcomputing allows efficient processing of large volumes of data otherwise unapproachable. This paper introduces a new semantic-driven data-intensive flow infrastructure which: (1) provides a robust and transparent scalable solution from a laptop to large-scale clusters,(2) creates an unified solution for batch and interactive tasks in high-performance computing environments, and (3) encourages reusing and sharing components. Banking on virtualization and cloud computing techniques the Meandre infrastructure is able to create and dispose Meandre clusters on demand, being transparent to the final user. This paper also presents a prototype of such clustered infrastructure and some results obtained using it.
暂无评论