I've been a four-star programmer a few times. Imagine a blocked symmetric matrix where the rows and columns are indexed by triples (u,v,w). The entries are zero whenever u != u' or v != v', and because of symmetry you only store entries with w <= w'. But the range of v depends on the value of u and the range of w on the value of v. So you do
double ****mat = calloc (UMAX, sizeof(*mat));
for (int u = 0; u < UMAX; ++u) {
mat[u] = calloc (u + 1, sizeof(**mat));
for (int v = 0; v <= u; ++v) {
mat[u][v] = calloc (v + 1, sizeof(***mat));
for (int w = 0; w <= v; ++w) {
mat[u][v][w] = calloc (w + 1, sizeof(****mat));
for (int ww = 0; ww <= w; ++ww)
mat[u][v][w][ww] = some_function (u, v, w, ww);
}
}
}
and weep a little. In reality, this gets a bit optimized by allocating a single chunk of memory and carving that up into the pointer and data arrays, so everything is reasonably close together in memory.
I've been a four-star programmer a few times. Imagine a blocked symmetric matrix where the rows and columns are indexed by triples (u,v,w). The entries are zero whenever u != u' or v != v', and because of symmetry you only store entries with w <= w'. But the range of v depends on the value of u and the range of w on the value of v. So you do
and weep a little. In reality, this gets a bit optimized by allocating a single chunk of memory and carving that up into the pointer and data arrays, so everything is reasonably close together in memory.