Path: utzoo!utgpu!water!watmath!clyde!rutgers!rochester!ritcv!cci632!ccicpg!puivax!ian
From: ian@puivax.UUCP (Ian Wilson)
Newsgroups: comp.lang.c
Subject: Re: comma operator (really about xor)
Message-ID: <131@puivax.UUCP>
Date: 27 Jan 88 21:45:08 GMT
References: <3819@sigi.Colorado.EDU> <5080013@hpfcdc.HP.COM> <7120@brl-smoke.ARPA> <3476@ihlpf.ATT.COM> <564@cresswell.quintus.UUCP>
Reply-To: ian@puivax.UUCP (Ian Wilson)
Organization: Philips Ultrasound International, Santa Ana, CA.
Lines: 31

In article <564@cresswell.quintus.UUCP> ok@quintus.UUCP (Richard A. O'Keefe) writes:
>> >> In article <3887@sigi.Colorado.EDU> swarbric@tramp.Colorado.EDU (SWARBRICK FRANCIS JOHN) writes:
>> >> >#define swap(a,b) ((a) = ((b) = ((a) = (a) ^ (b)) ^ (b)) ^ (a))
>
>	swap(a, a);
>sets a to 0, whatever its previous value was.  If you do
>	swap(*p, *q);
>it can easily happen that p==q, in which case you get this problem.

I am posting this because I once had a *useful* application of
this way of swapping two pieces of storage without using any
temporary storage. Using an image processor with an asymmetric
ALU, ie. it expected operand A to come from image memory A and
operand B from B, it enabled me to swap memories in 3 operations:

	1. A ^= B
	2. B ^= A
	3. A ^= B

(At least the ALU knew how to xor!!).

Each image memory was 1/4Mb, so doing it this way (90ms) was OK,
compared with the prospect of moving 2x1/4M off to mass storage
and back again.....

The transformation works whatever the values of A & B happen to
be - exclusive or being the only information-preserving logical
operation. So the original `swap' macro may or may not be very
useful but the principle is correct.

 ---- ccicpg!puivax!ian