array operations consuming lower time

Hi, we have an array x, I would like to do such a operation on it. It can be possible to implement it another way that consumes lower time?

    y1y2 = x(2:n+2)+x(1:n+1) ;

Or is there any function to sum the two adjacent element of arrays? E.g., we want to add x(i) to x (i+1), for i=1 to n. and create the result as an array by the size of n-1. It means the i th element of new array is x(i)+x(i+1) .

Thanks in advance

Answer 1

[Daniel Shub]

On my computer I can do a little better with conv...

>> x = randn(1e7, 1);
>> tic, y1y2 = x(2:end)+x(1:(end-1)); toc
Elapsed time is 0.330432 seconds.
>> tic, z1z2 = conv(x, [1,1], 'valid'); toc
Elapsed time is 0.166114 seconds.
>> isequal(y1y2, z1z2)
ans =
   1

[Jan]

Under Matlab 2009a both methods have the same speed. (Win7/64, DualCore)

Answer 2

[Jan]

I'm curious about a speed comparison:

[EDITED, bugs fixed]

#include "mex.h"

void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[]) {
double *X, *Y, *XEnd, xi;
mwSize nX;

nX      = mxGetNumberOfElements(prhs[0]);
X       = mxGetPr(prhs[0]);
XEnd    = X + nX;
plhs[0] = mxCreateDoubleMatrix(1, nX - 1, mxREAL);
Y       = mxGetPr(plhs[0]);

xi = *X++;
while(X < XEnd) {
   *Y++ = xi + *X;
   xi   = *X++;
}

return;
}

[EDITED, add timings]:

Matlab 2009a/64, Win7, Core2Duo:

 tic, y1 = x(2:end)+x(1:(end-1)); toc
 tic, y2 = conv(x, [1,1], 'valid'); toc
 tic, y3 = addPairsMex(x); toc
 isequal(y1, y2, y3)

 Elapsed time is 0.276241 seconds.
 Elapsed time is 0.265500 seconds.
 Elapsed time is 0.095482 seconds.
 1