Parallel programs construction and optimization

The paper describes a compiler technology called index analysis which allows the presentation of information about the existence of cycled intra-operational depenencies referring to memory. Such information is used in many optimizations, especially the cyclic ones. The formation of this system takes several preparatory steps of searching for cycle nests and inductive variables, and the direct construction of dependency equations and inequality systems.

The article presents a description of key features of convex polyhedrons. The main content of the article is the building of Blaschke diagram. Achieving this includes solving numerous intermediate tasks. The total time of computation reduces to a minimum because of the optimal selection of algorithms. The article also contains a description of a software product created for performing all required computations.

The paper introduces an algorithm of visualization of internal representation of programs. Internal representation can be viewed as a hierarchical directed acyclic graph, each vertex of which contains two ordered sets of input and output points. The algorithm is based on the Sugiyama algorithm. The algorithm was used to create a visualization component. The paper shows what kind of images can be generated.

The article presents a description of image processing technology applied to images received from satellite. The algorithms of brightness recalculation and panorama stitching are suggested. At the end of the article details of program implementation and results are brought.

This article is dedicated to algorithms of image analysis for feature detection problem in order to diagnose plants diseases. Statistical and textural features are analyzed, which helps to define the disease. The data gathered is used in future analyses to detect similar images. Several algorithms of contrast objects contour and texture features detection are described in this article.

Sisal is the functional data flow single assignment language, developed at A.P. Ershov Institute of Informatics Systems. In a SISAL compiler, most optimizations are performed on the intermediate representation IR2, which is the hierarchical acyclic data flow graph of the program. IR2 consists of vertices and ports. Vertices correspond to operations in the source program. Ports are used to mark the beginning and end of the value transfer edge between two operations.

The paper is devoted to the functional programming system SFP that is under development at the Institute of Informatics Systems in Novosibirsk with financial support from RFBR. The SFP system is intended to support construction of high quality portable software for parallel computers on low cost sequential computers (PC).

The paper is an extended text of paper which was presented by the author in 28 October 2008 at the meeting N 696 of the United seminar "Program construction and optimization" of the Institute of Informatics Systems and the Novosibirsk State University, and was devoted to the 50th anniversary of the Programming Department of the Institute of Mathematics, SB AS USSR.

An experiment of function reconstruction from its integrals on the family of straight lines cross the segment [-1, 1] on the x-axis is described. The carrier of intergrable function of two variables f(x, y) is concentrated in the region -1 ≤ x ≤ 1. Two-dimensional direct and reverse Mellin transformations are used in the algorithm of reconstruction, and their computations are performed in parallel.

This article continues the approach described in [1] and extends the offered model, which allows solving several applied problems, particularly those related to building automatic taxonomy of application domain based on the set of publication.

The aim of these notes is to present a brief glance upon the seminars held by Zykov from the point of view of a student of the Novosibirsk State University, later a research intern of the Institute of Mathematics of the SB AS USSR. The article contains personal impressions of the author in 1964-1969. In a way, it's a view upon that period.

In the paper, the historical review of development of supercomputers is given, and statistics and analysis of tendencies in the area of high-performance computations are briefly considered.

This paper aims to help students in writing different kinds of documents, such as term papers and research works, accounts, program instructions, letters and so on. As a rule, students facing such tasks are those specializing in science, and not humanities. They may experience certain difficulties in creating such texts.

The paper consisits of two parts: theory and application. There are questions for self-control after each section. In the end of each part there is a list of literature. Exercises with explanations and references are given in the appendix of the second part.

This work is dedicated to the relevant problem of effective context-related text search in the Internet. The main goal is to develop new efficient algorithms of sentence comparison based on syntactic analysis schemes and consequently develop a search system based on this approach. Worth mentioning is that it suffices to stay on the level of syntax and grammar constructions while evaluating the relevance of templates in the text. Syntactic diagrams set up a primary structure in the text which allows to select and evaluate sentences and phrases from the text which have corresponding semantic links between keywords. As the result a search system which automates Internet searching was developed. Test have shown good result relevance.

CUDA (Compute Unified Device Architecture) is the technology, developed by the company Nvidia, which makes possible for programmers to realize in the programming language C the algorithms, feasible on the graphic processors GeForce of the eighth generation and older (GeForce of 8 Series, GeForce of 9 Series, GeForce of 200 Series), Nvidia Quadro and Tesla of the company Nvidia. The article presents an instant analysis of the CUDA technology; equipment and special features of programming are described.

The paper describes a generator that uses the formalism of parametric context-free grammar and allows generation of context-sensitive languages suitable for automatic construction of programs that can be compiled and have deterministic and finite execution. By analyzing differences in the output of these programs compiled with optimizations and without them, subtle errors in beta version of the industrial level compiler were found.

Generalizing clique trees by selecting other sorts of induced subgraphs allows certain concepts and results of chordal graphs to be transferred to other classes of graphs, such as strongly chordal graphs, totally balanced hypergraphs, chordal bipartite graphs etc.