Engineering topics

Schreiber, Robert

Publications and source records attributed to Schreiber, Robert.

28 records · Page 2

NASA NTRS1990

Optimal parallel solution of sparse triangular systems

A method for the parallel solution of triangular sets of equations is described that is appropriate when there are many right-handed sides. By preprocessing, the method can reduce the number of parallel steps required to solve Lx = b compared to parallel forward or backsolve. Applications are to iterative solvers with triangular preconditioners, to structural analysis, or to power systems applications, where there may be many right-handed sides (not all available a priori). The inverse of L is represented as a product of sparse triangular factors. The problem is to find a factored representation of this inverse of L with the smallest number of factors (or partitions), subject to the requirement that no new nonzero elements be created in the formation of these inverse factors. A method from an earlier reference is shown to solve this problem. This method is improved upon by constructing a permutation of the rows and columns of L that preserves triangularity and allow for the best possible such partition. A number of practical examples and algorithmic details are presented. The parallelism attainable is illustrated by means of elimination trees and clique trees.

Alvarado, Fernando L.↗

NASA NTRS1990

Automatic blocking of nested loops

Blocked algorithms have much better properties of data locality and therefore can be much more efficient than ordinary algorithms when a memory hierarchy is involved. On the other hand, they are very difficult to write and to tune for particular machines. The reorganization is considered of nested loops through the use of known program transformations in order to create blocked algorithms automatically. The program transformations used are strip mining, loop interchange, and a variant of loop skewing in which invertible linear transformations (with integer coordinates) of the loop indices are allowed. Some problems are solved concerning the optimal application of these transformations. It is shown, in a very general setting, how to choose a nearly optimal set of transformed indices. It is then shown, in one particular but rather frequently occurring situation, how to choose an optimal set of block sizes.

Schreiber, Robert↗

NASA NTRS1990

Highly parallel sparse Cholesky factorization

Several fine grained parallel algorithms were developed and compared to compute the Cholesky factorization of a sparse matrix. The experimental implementations are on the Connection Machine, a distributed memory SIMD machine whose programming model conceptually supplies one processor per data element. In contrast to special purpose algorithms in which the matrix structure conforms to the connection structure of the machine, the focus is on matrices with arbitrary sparsity structure. The most promising algorithm is one whose inner loop performs several dense factorizations simultaneously on a 2-D grid of processors. Virtually any massively parallel dense factorization algorithm can be used as the key subroutine. The sparse code attains execution rates comparable to those of the dense subroutine. Although at present architectural limitations prevent the dense factorization from realizing its potential efficiency, it is concluded that a regular data parallel architecture can be used efficiently to solve arbitrarily structured sparse problems. A performance model is also presented and it is used to analyze the algorithms.

Gilbert, John R.↗

NASA NTRS1990

An assessment of the connection machine

The CM-2 is an example of a connection machine. The strengths and problems of this implementation are considered as well as important issues in the architecture and programming environment of connection machines in general. These are contrasted to the same issues in Multiple Instruction/Multiple Data (MIMD) microprocessors and multicomputers.

Schreiber, Robert↗

NASA NTRS1990

Optimal expression evaluation for data parallel architectures

A data parallel machine represents an array or other composite data structure by allocating one processor (at least conceptually) per data item. A pointwise operation can be performed between two such arrays in unit time, provided their corresponding elements are allocated in the same processors. If the arrays are not aligned in this fashion, the cost of moving one or both of them is part of the cost of the operation. The choice of where to perform the operation then affects this cost. If an expression with several operands is to be evaluated, there may be many choices of where to perform the intermediate operations. An efficient algorithm is given to find the minimum-cost way to evaluate an expression, for several different data parallel architectures. This algorithm applies to any architecture in which the metric describing the cost of moving an array is robust. This encompasses most of the common data parallel communication architectures, including meshes of arbitrary dimension and hypercubes. Remarks are made on several variations of the problem, some of which are solved and some of which remain open.

Gilbert, John R.↗

NASA NTRS1990

An improved Newton iteration for the generalized inverse of a matrix, with applications

The purpose here is to clarify and illustrate the potential for the use of variants of Newton's method of solving problems of practical interest on highly personal computers. The authors show how to accelerate the method substantially and how to modify it successfully to cope with ill-conditioned matrices. The authors conclude that Newton's method can be of value for some interesting computations, especially in parallel and other computing environments in which matrix products are especially easy to work with.

Pan, Victor↗

NASA NTRS1990

Solving unstructured grid problems on massively parallel computers

A highly parallel graph mapping technique that enables one to efficiently solve unstructured grid problems on massively parallel computers is presented. Many implicit and explicit methods for solving discretized partial differential equations require each point in the discretization to exchange data with its neighboring points every time step or iteration. The cost of this communication can negate the high performance promised by massively parallel computing. To eliminate this bottleneck, the graph of the irregular problem is mapped into the graph representing the interconnection topology of the computer such that the sum of the distances that the messages travel is minimized. It is shown that using the heuristic mapping algorithm significantly reduces the communication time compared to a naive assignment of processes to processors.

Hammond, Steven W.↗

NASA NTRS1989

Efficient ICCG on a shared memory multiprocessor

Different approaches are discussed for exploiting parallelism in the ICCG (Incomplete Cholesky Conjugate Gradient) method for solving large sparse symmetric positive definite systems of equations on a shared memory parallel computer. Techniques for efficiently solving triangular systems and computing sparse matrix-vector products are explored. Three methods for scheduling the tasks in solving triangular systems are implemented on the Sequent Balance 21000. Sample problems that are representative of a large class of problems solved using iterative methods are used. We show that a static analysis to determine data dependences in the triangular solve can greatly improve its parallel efficiency. We also show that ignoring symmetry and storing the whole matrix can reduce solution time substantially.

Hammond, Steven W.↗

NASA NTRS1989

A Storage-Efficient WY Representation for Products of Householder Transformations

A product Q=P1 ... P(sub r) of m x m Householder matrices can be written in the form Q = I + WY(sup T), where W and Y are each m x r. This is called the WY representation of Q. It is of interest when implementing Householder techniques in high-performance computing environments that are especially good at matrix-matrix multiplication. In this note a storage-efficient way to implement the WY representation is described. In particular, it is shown how the matrix Q can be expressed in the form Q = I + YTY(sup T). Usually r much less than m and so this 'compact' WY representation requires less storage. When compared with the recent block-reflector strategy the new technique still has a storage advantage and involves a comparable amount of work.

Schreiber, Robert↗

NASA NTRS1988

Bidiagonalization and symmetric tridiagonalization by systolic arrays

A systolic algorithm and an array for bidiagonalization of an nxn matrix in O(nlog(2)n) time, using O(n to the 2nd power) cells are given. Bandedness of the input matrix may be effectively exploited. If the matrix is banded, with p nonzero subdiagonals and q nonzero superdiagonals, then 4nln(p + q) + O(n) clocks and 2n(p + q) + O((p + q) to the 2nd power + n) cells are needed. This is faster than the best previously reported result by the factor log(2)e = 1.44.... Moreover, in contrast to earlier systolic designs, which require the matrix to be preloaded into the array and the result matrix extracted after bidiagonalization, the present arrays are pipelined.

Schreiber, Robert↗