alanm@cognos.UUCP (Alan Myrvold) (10/12/89)
There's been another (well deserved) round of bashing
in comp.lang.c for the way "Numerical Recipes" dynamically
allocates matrices.
It is kinda too bad that they picked such an ugly (and illegal)
way to handle matrices and vectors. I have their book (in FORTRAN)
and find it a nice readable introduction to many numerical
methods. Had they chosen a reasonable style, it might have become
commonplace.
So my question is, what C style SHOULD become commonplace for
allocating matrices :
1) Encapsulate size and data into a struct :
typedef struct { int row,col; double **x; } matrix;
2) Allocate a matrix as a vector and do arithmetic in subscript
calculations (like Fortran, but the C compiler probably
won't optimize the loops as well as Fortran).
typedef double *matrix;
x = malloc(row*col*sizeof(double));
x[j*row+i] = 3.0;
3) K&R style 2-d matrices -- number of rows required to free matrix
typedef double **matrix;
matrix mat_alloc(row,col)
int row,col;
{
matrix x;
x = (double **) malloc(row*sizeof(double *));
for (i = 0; i < row; i++)
x[i] = (double *) malloc(col*sizeof(double));
return x;
}
void free_mat(x,row)
matrix x;
int row;
{
for (i = row-1; i >= 0; i--) free(x[i]);
free(x);
}
4) 2-d matrices -- number of rows NOT required to free matrix
typedef double **matrix;
matrix mat_alloc(row,col)
int row,col;
{
matrix x;
x = (double **) malloc((row+1)*sizeof(double *));
for (i = 0; i < row; i++)
x[i] = (double *) malloc(col*sizeof(double));
x[row] = NULL;
return x;
}
void free_mat(x)
matrix x;
{
matrix y;
for (y = x; *y; y++) free(*y);
free(x);
}
5) Are there any other styles commonly used for
non-sparse matrices in C ???
---
Alan Myrvold 3755 Riverside Dr. uunet!mitel!sce!cognos!alanm
Cognos Incorporated P.O. Box 9707 alanm@cognos.uucp
(613) 738-1440 x5530 Ottawa, Ontario
CANADA K1G 3Z4 dave@dsi.COM (Dave Rifkind) (10/16/89)
In article <7229@cognos.UUCP> alanm@cognos.UUCP (Alan Myrvold) writes: >So my question is, what C style SHOULD become commonplace for >allocating matrices.... Here's one that doesn't require any special action to free up the matrix: double **mat_alloc(row, col) { int i; double *dp, **dpp; dpp = malloc(row * sizeof(double *) + row * col * sizeof(double)); if (dpp != NULL) { dp = dpp + row; for (i = 0; i < row; i++) dpp[i] = dp + i * col; } return dpp; } This is off the top of my head and may not make sense. The idea is to put the pointer array and all of the row arrays in a single allocated block. To get rid of it, you just free it.
harish@csl.ncsu.edu (Harish Hiriyannaiah) (10/17/89)
In article <278@dsi.COM> dave@dsi.UUCP (Dave Rifkind) writes:
-Here's one that doesn't require any special action to free up the
-matrix:
-
- double **mat_alloc(row, col)
- {
- int i;
- double *dp, **dpp;
-
- dpp = malloc(row * sizeof(double *) + row * col * sizeof(double));
- if (dpp != NULL) {
- dp = dpp + row;
- for (i = 0; i < row; i++)
- dpp[i] = dp + i * col;
- }
- return dpp;
- }
Here you are assuming that the pointers are word aligned, which
can be pretty dangerous.....
Consider, for example short pointers to floats. If we have 7 rows
(or any odd number), then the allocated space for the first float
will fall on a 16-bit boundary.....
harish pu. hi. harish@ecelet.ncsu.edu
harish@ecebucolix.ncsu.edu