Re: Duplicate code (update_flow_info et al.) in schedulers

[Jeffrey A Law <law at cygnus dot com>]

  In message <Pine.SOL.4.10.9906051438340.25850-100000@jennifer.informatik.rwth
-aachen.de>you write:
  > Both schedulers contain code that is called from recog.c to update register
  > life information after splitting instructions.  This patch moves these four
  > functions to flow.c, where they fit somewhat better and aren't duplicated
  > anymore.
Given that we have zero interest in working on the scheduler, does it still
make sense to install this patch?  In fact, we should put the old scheduler
on the "kill as soon as possible list".  It has really lingered on much longer
than it should have.

  > There are minimal differences in the two versions, none of which seem to
  > be very important.  The haifa version uses alloc_EXPR_LIST calls, while
  > the sched.c version uses rtx_alloc.  In haifa, there is one place where
  > there's an additional check that an insn has rtx class 'i' before attemptin
The alloc_EXPR_LIST stuff is actually rather important in keeping haifa's
memory consuption down to something that is bearable.  Though I believe the
case where we ate up tons of memory was for dependency info that is stored
in EXPR_LIST and INSN_LIST nodes.