Re: papers on symbolic dependency analysis

Graham Jones <gxj@dcs.ed.ac.uk> writes:
|> In M. Haghighat and C. Polychronopoulos paper "Symbolic Dependence
|> Analysis for High Performance Parallelising Compilers, Advances in
|> Languages and Compilers for Parallel Processing, 1991" the authors
|> indicate the large number of arrays with symbolic subscripts. Classical
|> dependency tests fail in the presence of symbolic terms but can be
|> assisted by global constant propagation, forward and induction variable
|> substitution. Does anyone know of the number of cases that still remain
|> unanalysable after these optimisations have taken place ? Given that some
|> subsripts can still not be analysed, we need to develop symbolic
|> dependency tests to handle these cases. Does anyone know of any current
|> work and useful papers in this area ?


I don't know about " the number of cases that still remain unanalysable
after these optimisations have taken place" ; one useful source would be
"Effectiveness of data dep. analysis" by Maydan & Lam, or Maydan's PhD
thesis (ftp suif.stanford.edu, anonymous, etc).


Feautrier's PIP ("Parametric Integer Programming", 1988, or "Dataflow
analysis of array and scalar references", IJPP '91 no 1) runs (from the
beginning) in symbolic mode ; Pugh's work (Omega test and following - ftp
cs.umd.edu) also covers symbolic mode.


Then, you have Thomasset & Lichnewsky's "Introducing symbolic problem
solving techniques in the dependence testing phase of a vectorizer" in
Supercomp'88.




Hope this helps... Thanks for posting other info.




Laurent Challier.
--

Return to the comp.compilers page.
Search the comp.compilers archives again.