Yet another PREFERRED_STACK_BOUNDARY/nested calls fix

[Jeffrey A Law]

  In message < 20000322161541.B29524@atrey.karlin.mff.cuni.cz >you write:
  > Hi
  > This is my attempt to fix the problems with stack adjustments.
  > I've killed the arg_size_so_far and use stack_delta instead. It is updated
  > in lowlevel functions (emit_push_insn, adjust_stack and anti_adjust_stack)
  > and used by calls.c to generate proper alignment.
  > 
  > I expect there will be less problems with target macros such
  > as ACCUMULATE_OUTGOING_ARGS compared to arg_size_so_far.
  > 
  > The patch has some sanity checking and passes bootstrap/testsuite
  > on i386. With missalignment in builtins.c fixed.
  > 
  > I hope there will not be big problems on other machines as well.
  > I would welcome bootstrap results on other CPUs. Thanks.
I don't see any note about where this was bootstrapped or regression tested.

Given the multitude of problems with this stuff, I want to see it bootstrapped
and regression tested on x86 and at least one other platform before it is
installed.

jeff