[M3commit] CVS Update: cm3

[Jay Krell]

CVSROOT:	/usr/cvs
Changes by:	jkrell at birch.	10/03/10 16:12:31

Modified files:
	cm3/m3-sys/m3back/src/: Stackx86.m3 

Log message:
	Now that I finally think enough, a nice little fix to the
	register allocator to handle when a multi register operand
	is partly in the right registers, but with the part mixed up.
	It yields us at the end of Long__Insert this:
	
	000003C6: 8B4D08              MOV ECX tv.65[_x]
	000003C9: 8B450C              MOV EAX tv.65[_x]+4
	000003CC: 23CB                AND ECX EBX
	000003CE: 23C2                AND EAX EDX
	000003D0: 0BCE                OR ECX ESI
	000003D2: 0BC7                OR EAX EDI
	exit_proc         Int.64
	000003D4: 91                  XCHG  ECX EAX
	000003D5: 8BD0                MOV EDX EAX
	000003D7: E900000000          JMP L.39
	end_procedure     p.24[_Long__Insert]
	
	again it'd be nice if we had ECX:EAX in the first place
	instead of EAX:ECX, or even EDX:EAX.
	
	And then, a roundabout but understandable way to eliminate
	the unnecessary stores in insert/extract.
	
	The extra checking under -debug forces to rearrange the
	virtual stack and discard(), rather than just say dealloc_reg.
	If "location" could be "nowhere" that might help, but
	we have to chose between immediate, register, float stack, memory_variable.
	And then Modula-3 WITH either because of how it behaves or
	because of my confusion -- I think actually for reasons related
	to "safety", we have to "reevaluate" a few times -- each time
	we discard anything from the stack. Leading to a bit extra code
	but it is fairly clear.