>>> In a message dated 11/17/2006 8:29:18 A.M. Eastern Standard  Time, 
_vincent@labri.fr_ (mailto:vincent@labri.fr)  
>>> writes:
>>> BTW, I saw that there are movel  %sp,%sp@- in a few places in
>>> libpthread. Could the behaviour of  this instruction be
>>> non-deterministic? I couldn't see anything  in the m68k manuals which
>>> would dictate the precise behaviour of  this instruction.

I expect that it is deterministic.  The Verdix host compiler for  sun3 uses 
this same instruction a lot.  I found that the TME sun3 emulator  did it 
incorrectly (ex: if sp starts at 100, it moves 96 to address 96 instead  of 100 to 
address 96).  If the instruction was nondeterminisitc I expect  that a lot of 
Verdix customers would have been upset back in the 1980's.   cmpm (sp)+,(sp)+ 
is also another weird one that performs identically each  execution.
/arthur