Browse Prior Art Database

Dynamic Optimization Technique for Inter-Processor Communication on Massively Parallel Processors

IP.com Disclosure Number: IPCOM000114923D
Original Publication Date: 1995-Feb-01
Included in the Prior Art Database: 2005-Mar-30
Document File: 4 page(s) / 82K

Publishing Venue

IBM

Related People

Ishizaki, K: AUTHOR [+2]

Abstract

Disclosed is a optimization technique whereby the executable code analyzes the region of inter-processor communication in program loops at runtime. This new method is especially when either the loop bound or the array size in the program is decided at runtime, this new method is effective.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 54% of the total text.

Dynamic Optimization Technique for Inter-Processor Communication
on Massively Parallel Processors

      Disclosed is a optimization technique whereby the executable
code analyzes the region of inter-processor communication in program
loops at runtime.  This new method is especially when either the loop
bound or the array size in the program is decided at runtime, this
new method is effective.

      If there are any parameters with unknown values at the compile
time for analyzing the region of inter-processor communication in a
program, the compiler cannot proceed.  The parameters needed for this
purpose are as follows:
  1.  the lower and upper of loop bounds
  2.  the array size that are accessed in the loop
  3.  the loop invariant value in the array subscript expression

The disclosed method is effective when parameters 1 and 2 are
unknown.

      If the communication region is not decidable, the compiler
cannot vectorize inter-processor communication.  In the conventional
method, if the communication is not vectorized, a check is performed
to determine whether there is communication between processors in
each loop iteration.  As a result, the execution of the program is
too slow.

      In the new method, when the values of unknown parameters of the
loop bounds and the array size are decided at runtime, the executable
code calls the module for analyzing the region of inter-processor
communication.

      A particular advantage of this metho...