Fast matrix multiplication with diagonal matrices

22 views (last 30 days)
Let Wbe a large, sparse matrix. Let and be diagonal matrices of the same size. I would like to calculate . However, these matrices are large enough that matrix multiplication is very expensive. I would like to speed up the calculation of L.
I know that computing L can be sped up by utilizing the fact that and are diagonal. For example, I know that I can compute as follows.
diagD1 = diag(D1); % diagonal of the matrix D1.
D1W = W.*diadD1; % Equivalent to multiplying the ith row of W by D(i,i). Yields D1*W.
My question is whether there is a similar exploitation of the diagonality of that will allow me to avoid matrix multiplication to compute L.
Thank you.
  6 Comments
Matt J
Matt J on 26 Feb 2021
Edited: Matt J on 26 Feb 2021
It doesn't appear to be beneficial to avoid matrix multiplication, though I am surprised the latter is so slow (in R2020b).
N=1e5;
W=sprand(N,N,100/N);
d=rand(N,1);
D=spdiags(d,0,N,N);
tic
L1=D*W*D;
toc
Elapsed time is 0.572682 seconds.
tic;
L2=(d.*W).*d.';
toc
Elapsed time is 13.663614 seconds.
saskia leary
saskia leary on 20 Mar 2022
(diag(D))'.*A works for right mulipltication - i.e.
A*D=(diag(D)).*A

Sign in to comment.

Accepted Answer

James Tursa
James Tursa on 25 Feb 2021
Edited: James Tursa on 25 Feb 2021
Here is a mex routine to do this calculation. It relies on inputting the diagonal matrices as full vectors of the diagonal elements. It does not check for underflow to 0 for the calculations. A robust production version of this code would check for this and clean the sparse result of 0 entries, but I did not include that code here. It also does not check for inf or NaN entries. This could be made faster with parallel code such as OpenMP, but I didn't do that either.
/* File: spdmd.c */
/* Compile: mex spdmd.c */
/* Syntax C = spdmd(D1,M,D2) */
/* Does C = D1 * M * D2 */
/* where M = double real sparse NxN matrix */
/* D1 = double real N element full vector representing diagonal NxN matrix */
/* D2 = double real N element full vector representing diagonal NxN matrix */
/* C = double real sparse NxN matrix */
/* Programmer: James Tursa */
/* Date: 2/24/2021 */
/* Includes ----------------------------------------------------------- */
#include "mex.h"
#include <string.h>
/* Gateway ------------------------------------------------------------ */
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
mwSize m, n, j, nrow;
double *Mpr, *D1pr, *D2pr, *Cpr;
mwIndex *Mir, *Mjc, *Cir, *Cjc;
/* Argument checks */
if( nlhs > 1 ) {
mexErrMsgTxt("Too many outputs");
}
if( nrhs != 3 ) {
mexErrMsgTxt("Need exactly three inputs");
}
if (!mxIsDouble(prhs[1]) || !mxIsSparse(prhs[1]) || mxIsComplex(prhs[1])) {
mexErrMsgTxt("2nd argument must be real double sparse matrix");
}
if( !mxIsDouble(prhs[0]) || mxIsSparse(prhs[0]) || mxIsComplex(prhs[0]) ||
mxGetNumberOfDimensions(prhs[0]) != 2 || (mxGetM(prhs[0]) != 1 && mxGetN(prhs[0]) != 1)) {
mexErrMsgTxt("1st argument must be real double full vector");
}
if (!mxIsDouble(prhs[2]) || mxIsSparse(prhs[2]) || mxIsComplex(prhs[2]) ||
mxGetNumberOfDimensions(prhs[2]) != 2 || (mxGetM(prhs[2]) != 1 && mxGetN(prhs[2]) != 1)) {
mexErrMsgTxt("3rd argument must be real double full vector");
}
m = mxGetM(prhs[1]);
n = mxGetN(prhs[1]);
if (m != n || mxGetNumberOfElements(prhs[0]) != n || mxGetNumberOfElements(prhs[2]) != n) {
mexErrMsgTxt("Matrix dimensions must agree.");
}
/* Sparse info */
Mir = mxGetIr(prhs[1]);
Mjc = mxGetJc(prhs[1]);
/* Create output */
plhs[0] = mxCreateSparse( m, n, Mjc[n], mxREAL);
/* Get data pointers */
Mpr = (double *) mxGetData(prhs[1]);
D1pr = (double *) mxGetData(prhs[0]);
D2pr = (double *) mxGetData(prhs[2]);
Cpr = (double *) mxGetData(plhs[0]);
Cir = mxGetIr(plhs[0]);
Cjc = mxGetJc(plhs[0]);
/* Fill in sparse indexing */
memcpy(Cjc, Mjc, (n+1) * sizeof(mwIndex));
memcpy(Cir, Mir, Cjc[n] * sizeof(mwIndex));
/* Calculate result */
for( j=0; j<n; j++ ) {
nrow = Mjc[j+1] - Mjc[j]; /* Number of row elements for this column */
while( nrow-- ) {
*Cpr++ = *Mpr++ * (D2pr[j] * D1pr[*Cir++]);
}
}
}
  3 Comments
James Tursa
James Tursa on 25 Feb 2021
Edited: James Tursa on 25 Feb 2021
Fixed the include. Thanks. The speed gain, if any, will depend greatly on the actual sizes and sparsity involved.

Sign in to comment.

More Answers (0)

Products

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!