Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site brl-tgr.ARPA
Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!panda!talcott!harvard!seismo!brl-tgr!tgr!gwyn@Brl-Vld.ARPA
From: gwyn@Brl-Vld.ARPA (VLD/VMB)
Newsgroups: net.lang.c
Subject: Re:  help converting multi-dim arrays
Message-ID: <9880@brl-tgr.ARPA>
Date: Thu, 11-Apr-85 11:44:04 EST
Article-I.D.: brl-tgr.9880
Posted: Thu Apr 11 11:44:04 1985
Date-Received: Sat, 13-Apr-85 04:35:31 EST
Sender: news@brl-tgr.ARPA
Lines: 38

Re: large matrix use in C

The closest approach meeting your stated requirements is to vector
matrix rows:

	static	*datap[MAXROWS];	/* init NULL */
	...
	/* access like */ datap[row][col]

With this approach, when setting up the array you have to test
for a null datap[row] and when necessary malloc() a whole row:

	if ( datap[row] == NULL )
		if ( (datap[row] = (double *)malloc( MAXCOLS *
				sizeof(double) )) == NULL )
			/* punt */;
		else
			for ( col = 0; col < MAXCOLS; ++col )
				datap[row][col] = 0.0;

Clearly the only space gain here is if a sizeable fraction of
the rows are never used.  However, there is an enormous speed
gain in some cases by using vectoring instead of having code
generated for index arithmetic (a multiply and add).

To do better with more general sparse arrays, you have to give
up the requirement that array element references look like

	data[row][col]

and instead use macros or functions:

	init_data();	/* sets up structures and fills with 0.0 */
	put_data( row, col, value );	/* stores value in "data" */
	value = get_data( row, col );	/* retrieves value */

Most data structures textbooks discuss algorithms for sparse
arrays.