🚀 go-pugleaf

In article <5047@garfield.MUN.EDU> john13@garfield.MUN.EDU (John Russell) writes:
>Manx-generated code uses the form
>
>addq.w #(4*N),sp
>     ^
>
>to reset the stack pointer after calling a function with N (long) args on the 
>stack.  Examples which adhere to the syntax of other Amiga assemblers use
>
>addq.l #(4*N),sp
>     ^
>
>Is the Manx format dangerous if the stack is being incremented past a
>64k boundary? Is there a reason to use one and not the other?
>
	It turns out that, for ADDQ, specifying .w over .l doesn't buy you
anything; both are 8 clocks.  However, ADDQ only works for immediate values
between 1-8.  Outside that range, you would be inclined to use ADDA.  To use
your notation:

	adda.l	#(4*N),sp	; 16 clocks, 6 bytes

	However, 32K of local variables in a single context is highly
unusual (unless you've got a lot of buffers hanging around, in which case
you should probably be using AllocMem() to get them).  Thus, it is (usually)
safe to say:

	adda.w	#(4*N),sp	; 12 clocks, 4 bytes

	A smart compiler will recognize when the desired offset is greater
than 32K (the offset is signed, remember), and use the ADDA.L when needed.

	It further turns out that you can get even better performance than
that by saying:

	lea	4*N(sp),sp	; 8 clocks, 4 bytes

	This does exactly what the ADDA.W construct above does, only
cheaper.

	The 68000 is full of little non-orthogonal goodies like this.

[Timing information source: MC68000 programming quick-reference card,
available from Motorola.]
_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_
Leo L. Schwab -- The Guy in The Cape	INET: well!ewhac@ucbvax.Berkeley.EDU
 \_ -_		Recumbent Bikes:	UUCP: pacbell > !{well,unicom}!ewhac
O----^o	      The Only Way To Fly.	      hplabs / (pronounced "AE-wack")
"Work FOR?  I don't work FOR anybody!  I'm just having fun."  -- The Doctor