Related articles |
---|
Maximum Register Pressure? ammalik@cs.uwaterloo.ca (2006-03-12) |
Re: Maximum Register Pressure? Sid-pasdespam.Touati-nospam@inria-nospam.fr (Sid Touati) (2006-03-14) |
Re: Maximum Register Pressure? Sid.Touati@inria-nospam.fr (Sid Touati) (2006-03-14) |
From: | Sid Touati <Sid-pasdespam.Touati-nospam@inria-nospam.fr> |
Newsgroups: | comp.compilers |
Date: | 14 Mar 2006 00:50:18 -0500 |
Organization: | I.N.R.I.A Rocquencourt |
References: | 06-03-039 |
Keywords: | registers |
Posted-Date: | 14 Mar 2006 00:50:18 EST |
ammalik@cs.uwaterloo.ca a écrit :
> I have a question. Does an optimal schedule of a basic block also
> gives maximum register pressure for that basic block?
>
> Regards
> Abid
No;
here is a DAG represented with data dependences :
a -> b : flow arc latency = 20
c -> d : flow arc latency = 1
e -> f : flow arc latency = 1
A time-optimal schedule is of latency 20 (assuming enough resources).
Assuming no exit values, ie. the operations b d f do not write into
registers, you can use only two registers in a time-optimal schedule
(schedule the whains a->d and e-> in serial), while the maximal register
pressure is 3 registers (a c e simultaneously alive)
S
Return to the
comp.compilers page.
Search the
comp.compilers archives again.