Topic: vector processing

topics > computer science > Group: parallel processing

co-sequence operations
computer architecture
data parallel processing
massively parallel processors
parallel algorithms
parallel control statements
processing a sequence


Vector processing is a form of parallel processing. A vector processor performs the same operation on many vector elements at once, providing a high maximum throughput rate. Since many problems contain array computations, vector processing has many useful applications. Perrott's language Actus defines parallel arrays and parallel index variables for indicating possible vector operations. George's and Heyman's parametric expressions indicate function executions for all possible values in an operand set; thus producing an array of results. (cbb 5/80)
Subtopic: vectorized code up

Quote: in general, efficient vectorization is easier to achieve than efficient parallelism; should be attempted first [»thomCD11_1993]
Quote: Fortran guarantees that multiple array arguments do not overlap; allows spectacular savings on vector hardware [»stroB_1994]

Subtopic: vector processor up

Quote: add a full vector architecture to x86 and PowerPC; via their multimedia extensions; keeps the memory system busy [»gebiJ4_2007]
Quote: vector supercomputers are good for heavily used, vectorized codes that require high speed and large, primary memories [»thomCD11_1993]
QuoteRef: perrRH10_1979 ;;192 language called Actus for vector processors
QuoteRef: perrRH10_1979 ;;192 gives a number of syntactic devices to manipulate and specify vector operations

Subtopic: xector, one value per processor up

Quote: parallel operations in CmLisp via a xector--a set of processors with one value per processor; e.g., xector add [»hillWD_1985]

Subtopic: parallel array up

Quote: declare a parallel array variable by using '1:m' for one of its index ranges; then can manipulate m elements at a time [»perrRH10_1979]
Quote: a parallel variable represents a sequence of integer numbers; e.g., oddseq= 1:(2)31 [»perrRH10_1979]
Quote: an index set indicates members of a vector; can access in parallel or one at a time [»perrRH10_1979]

Subtopic: parametric expression up

QuoteRef: georJ12_1977 ;;43 parametric expression: expression (pl,...,pn) as pl,...,pn in (rangel,...rangen) results in array which is expression applied to cartesian product of ranges eg I as I in 0..8 is APL's .enumerate.8
QuoteRef: georJ12_1977 ;;45 in parametric expression. special symbols to combine by rows or columns

Subtopic: limits of vectorization up

Quote: vectorization is prevented by recursion, subroutine calls, I/O, loop exits, etc. [»thomCD11_1993]
Quote: the limiting factor for nonvectorized code is reading operands from memory; makes workstations comparable to supercomputers [»thomCD11_1993]
Quote: converting a bit-level implementation to word-level often requires algorithm-specific insights; vectorizing compilers are largely ineffective

Related Topics up

Topic: co-sequence operations (8 items)
Topic: computer architecture (46 items)
Topic: data parallel processing (14 items)
Topic: massively parallel processors (29 items)
Topic: parallel algorithms (15 items)
Topic: parallel control statements (12 items)
Topic: processing a sequence
(17 items)

Updated barberCB 3/05
Copyright © 2002-2008 by C. Bradford Barber. All rights reserved.
Thesa is a trademark of C. Bradford Barber.