Interprocedural analysis in Parallelizing compiler

hi!


I am looking for a paper/method which does the foll:- In Fortran,
one can specify in a call the dimensions of an array on which the routine
will act. If we assume that the subroutine will act on the entire array,
then this leads to a loss of potential parallelism. This loss is to be
eleminated by evaluating the strip of array passed. An example follows:-


program xyz
integer A(100,100)
integer l1,l2,l3,l4


...
l1=..
l2=..
l3=..
l4=..
use/modify A
call abc(A,l1,l2,l3,l4)
use/modify A
stop
end


subroutine abc(a, l1,l2,l3,l4)
integer a(l1;l2,l3;l4)
..
modify array A
return
end


The routine will evaluate the portion of array bounded by the values of
the parameters. Any statement in the calling routine that modifies/uses
portions of array A not being passed to abc can be executed in parallel
with call statement.
.
Please mail your replies to:
kamal@cse.iitb.ernet.in (or)
cdacgrp@cse.iitb.ernet.in


Regards
-Kamal
--

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