Re: About Regmove !!

[Jeff Law <law at porcupine dot cygnus dot com>]

In message <Pine.LNX.4.21.0211121052080.12360-100000@mail.kloo.net>, tm writes:
 >On Tue, 12 Nov 2002, Michael Matz wrote:
 >
 >> Hi,
 >> 
 >> On Mon, 11 Nov 2002, tm wrote:
 >> 
 >> > Erm. Regmove does more than simple register-to-register copy elimination.
 >> 
 >> Yes.
 >> 
 >> > It also optimizes for address inheritance, which is critical for many
 >> > processors with short displacement addressing, such as the SH4.
 >> 
 >> Are you talking about the combine_stack_adjustment() which just happens to
 >> be implemented in regmove.c also, or about the transformation of
 >>   pX <= pA + N
 >>   ...
 >>   pX <= pA + M
 >> into
 >>   pX <= pA + N
 >>   ...
 >>   pX <= pX + (M - N)
 >> ?
 >> 
 >> I guess you meant the latter.  Interestingly it isn't done generally, but
 >> only applies when pX and pA need to match in an insn.  I.e. in that
 >> context it isn't used at all for shortening immediate constants.  If it's
 >> an important optimization it should be implemented more generally,
 >> although this might not be easy if one want to take register pressure into
 >> account.
 >
 >Yes, this latter optimization is address inheritance.
 >Code size can increase by a factor of two (or more, because of increased
 >spills) if this optimization is removed.
 >
 >> 
 >> > Does new-regalloc do address inheritance optimization?
 >> 
 >> No (t yet?).
 >> 
 >> > If not, can we retain the portion of regmove which implements this
 >> > optimization?
 >> 
 >> Of course.
 >
 >Okay. Please keep this in mind.
Note there are other ways to do this kind of optimization.   For example,
look at how folks have integrated strength reduction and gcse.  Better
yet, you get these optimizations on a global basis rather than local or
extended basic block.

jeff