strange x86 generation of 2.95 compare to 1.1.2

[Jeffrey A Law]

  In message < 37698D00.C5C9FC79@sls.lcs.mit.edu >you write:
  > Jeffrey A Law wrote:
  > I see.  It still has two more %esp manipulations than it needs (just look
  > at the non-call path).
It's certainly not perfect.  But fixing the non-perfect problems isn't as
obvious and trivial as it might seem :-)

  > I just found that it is -mpreferred-stack-boundary=2.  BTW, the info page
  > mistakenly calls it -mpreferred_stack_boundary (with _ instead of -).
Thanks.  I'll take care of that.
jeff