swap(x,y)

[Herman Rubin]

In article <10790 at riks.csl.sony.co.jp>, diamond at csl.sony.co.jp (Norman Diamond) writes:
> In article <1545 at l.cc.purdue.edu> cik at l.cc.purdue.edu (Herman Rubin) writes:
> 
> >If one adds a pseudo-op, say use (I will not object to a more appropriate
> >term) so that (use int)x means treat x as type int no matter what it was,
> >if possible, then it
>  [the XOR trick]
> >will work on float and pointer, and even swap between,
> >say, float and pointer.  If this is modified to allow multiple-word types,
> >it could handle double and struct.
> >This is another example of keeping the tools from the programmer.
> 
> The way to write this pseudo-op in C is:
> 
>    *(int*)&x

What if x is in a register?  Much of the time things will be in registers.
Indeed, on some machines, arithmetic operations can only be done using
registers.  One might not even have memory to memory moves.  On such machines,
there is no point in attempting to swap two items in memory by other than
moving both to registers and storing the registers, or memory and register
by other than using a spare register.  But the situation with few spare
registers and a register-register swap wanted is much more common.

> It still doesn't work if x is longer than an int.  Most hardware does
> not supply tools for performing XOR on floating-point registers.  When
> Mr. Rubin designs his language, he will have to design a machine to
> run it on too.

Quite a few machines do not have separate floating-point registers.  On
those, the swap is appropriate for XOR-sized items.  And on a double
length operand, two XORs might be the best way to do it, etc.  This
is again especially important if registers are scarce.

This problem is a good example of the need to consider that the way to
perform operations on quantities may depend on not only the types of the
arguments, but also upon where they are stored.

The machine language programmer has no problems with this.  The hardware
limitations are machine dependent.
-- 
Herman Rubin, Dept. of Statistics, Purdue Univ., West Lafayette IN47907
Phone: (317)494-6054
hrubin at l.cc.purdue.edu (Internet, bitnet, UUCP)