CRPC-TR94405


Title:    Performance Analysis of Data Parallel Programs

Authors:  Vikram S. Adve, Charles Koelbel, John M. Mellor-Crummey

Date:	  May 1994 


Of the authors listed above, please indicate which are:

  Minority authors:

  Female authors:    

  Student authors:   


Keywords (list up to 8):

  Data parallel, HPF, Performance analysis, Performance tuning,
  Automatic tuning, Program instrumentation, Parallelizing compiler

Abstract:
  Effective strategies for performance analysis and tuning will be essential
  for the success of data parallel languages such as High-Performance Fortran
  (HPF) and Fortran D.  Since compilers for these languages insert all
  communication, they have considerable knowledge about a program's dynamic
  structure and the relationship between its parallelism and communication.
  This paper explores how this compiler knowledge can be exploited to support
  performance evaluation and tuning.  First, the compiler itself can use
  parameterized models to tune the performance of individual program phases;
  this approach can be effective provided that the compiler can test and
  handle violations of the model assumptions.  Second, by exploiting compiler
  knowledge and introducing code transformations to improve monitorability,
  we can collect dynamic performance information that is far more compact
  than full communication traces, but well suited to the needs of tuning
  specific communication patterns.  Third, we discuss why an understanding of
  the compiler's capabilities can be important for effective performance
  tuning.  We use several Fortran 77D benchmark kernels to illustrate these
  points.  Finally, through our studies of these benchmarks, we identify the
  need for several generally applicable compiler optimizations that improve
  communication and computation overlap.


Publication History:

  Submitted to:	 SUPERCOMPUTING '94

  Published in: