|Definition of Data Dependence Distance email@example.com (1991-12-06)|
|Re: Definition of Data Dependence Distance maydan@Neon.Stanford.EDU (1991-12-06)|
|Re: Definition of Data Dependence Distance frazier@CS.UCLA.EDU (1991-12-06)|
|Re: Definition of Data Dependence Distance grover@brahmand.Eng.Sun.COM (1991-12-10)|
|From:||grover@brahmand.Eng.Sun.COM (Vinod Grover)|
|Organization:||Sun Microsystems, Mt. View, Ca.|
|Date:||10 Dec 1991 06:51:44 GMT|
In article 91-12-041 firstname.lastname@example.org (Bill Pugh) writes:
> Does dependence distance represent the difference in the values of
> the loop variables, or the difference in the loop trip counts?
I think the answer depends on why the compiler needs to know the dependence
distance. If one is interested in doing software pipelining or loop
unrolling, then you would care about the "iteration distance", becuase you
are interested in scheduling as many iterations as possible without
encountering a dependence.
If, on the other and, you are interested in doing a data-locality
transformation, you care about the "index-variable distance". Essentially
this requires knowing the distance between memory references.
Many compilers just normalize the loop and punt the whole issue.
Return to the
Search the comp.compilers archives again.