When is the stack pointer unreliable?

[Jeffrey A Law]

  In message < 199809060409.AAA02189@jwlab.FEITH.COM >you write:
  > I noticed that i386 backend reloads the stack pointer in
  > situations where it shouldn't be necessary.  In i386.c
  > (ix86_epilogue) is the comment:
  > 
  >   /* sp is often  unreliable so we must go off the frame pointer.
  > 
  > Since reloading the stack pointer causes a stall on some processors
  > it should be avoided when possible.  Under what situations is sp
  > unreliable?
alloca calls, variable sized objects in the stack come immediately
to mind.

You might also need to check if the compiler has optimized away
popping of args after a function call if we unconditionally reach the
epilogue after performing the call.

jeff