x86 first scheduling pass revisited

[Jeffrey A Law]

  In message < 199906031913.MAA16508@netcom8.netcom.com >you write:
  > 
  > Is there any chance we can enable the x86 port's first instruction 
  > scheduling pass?
Someday.  There's still some correctness issues one has to resolve.

For example you have to make sure that you do not separate a call with the
insn which copies its return value from %eax to a pseudo.  Otherwise you'll
generate incorrect code.  There is a discussion of this issue somewhere in
the archives.

  > So, I'd like to propose the first instruction scheduling pass be enabled 
  > by default on the x86 port, except in the following circumstances:
Not until someone fixes the problem noted above.  There may be others lurking,
but we know that without fixing the bug above that enabling the first sched
pass will generate incorrect code.

jeff