Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: nyu notesfiles V1.1 4/1/84; site ur-msbvax.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!panda!talcott!harvard!seismo!rochester!ur-msbvax!therneau
From: therneau@ur-msbvax.UUCP
Newsgroups: net.lang.c
Subject: Re: help converting multi-dim arrays
Message-ID: <1000002@ur-msbvax.UUCP>
Date: Tue, 9-Apr-85 16:26:00 EST
Article-I.D.: ur-msbva.1000002
Posted: Tue Apr  9 16:26:00 1985
Date-Received: Thu, 11-Apr-85 01:51:31 EST
References: <569@utcs.UUCP>
Organization: University of Rochester: Math, Stat, Bio-Stat
Lines: 40
Nf-ID: #R:utcs:-56900:ur-msbvax:1000002:000:1083
Nf-From: ur-msbvax!therneau    Apr  9 16:26:00 1985


  A good solution is to use an array of vectors, i.e.,


	static double **data;
		.
		.
		.
		.

	data = calloc(nrows, sizeof(*data));  /*allocate a vector of pointers*/
	for (i=0; i<ncols; i++)
		data[i] = calloc(ncols, sizeof(**data));


The loop allocates storage for each row of the array, and initializes the
the vector of pointers to that storage.  Now the line

		whatever = data[i][j]

in the C program works just fine.  The operation done is two fetches from
memory rather than the (i*ncols + j) of an ordinary array reference; there
is some evidence that the (old) multiplication is slower than the (new)
double memory reference.  As an added bonus, 'data' can be passed to a
subroutine without dimensioning hassles.

    The allocation loop above could certainly be speeded up by doing
only two calloc calls--

	data = calloc(nrows, sizeof(*data));  /*allocate a vector of pointers*/
	dbl_ptr = calloc(nrows*ncols, sizeof(**data));
	for (i=0; i<ncols; i++)  {
		data[i] = dbl_ptr;
		dbl_ptr += ncols;
		}


			Terry Therneau
			Statistics Dept.
			University of Rochester