Could exponent of an array be inproved in speed ? Maybe mex file with exp function (reduced accuracy) for windows 64 bit.

Could this line of code be improved in speed ? MM is an array 512x512^2 elements.

MM =ones(512^2,512); tic; exp(1i*(MM)); toc

Elapsed time is 3.824361 seconds.

I am looking for mex file for 64bit windows of exp function with reduced accuracy ?

code from Exp function with reduced accuracy But I can not compile it.

#include <stddef.h>
#include <math.h>
#include "mex.h"

#ifndef mwSize
#define mwSize int
#endif

/* these 2 #define lines help make the later code more
readable */
/* Input Arguments */
#define PARAMETER_IN prhs[0]

/* Output Arguments */
#define RESULT_OUT plhs[0]

#define LITTLE_ENDIAN 1

static union
{
  double d;
  struct {
#ifdef LITTLE_ENDIAN
    int j,i;
#else
    int i,j;
#endif
  } n;
} _eco;

#define EXP_A (1048576/0.69314718055994530942)
#define EXP_C 60801
#define EXP(y) (_eco.n.i = EXP_A*(y) + (1072693248 - EXP_C),_eco.d)

void mexexp(double*y, double*yp, mwSize m) {
  while(m--) {
    *yp++ = EXP(*y++);
  }
}

void mexFunction( int nlhs, mxArray *plhs[],
                  int nrhs, const mxArray*prhs[] )
{
    mwSize numel;
    mwSize ndim;
    mwSize *dims;

      /* Check for proper number of arguments */
      if (nrhs != 1) {
          mexErrMsgTxt("One input argument required.");
      } else if (nlhs > 1) {
          mexErrMsgTxt("Too many output arguments.");
      } else if ( !mxIsDouble(PARAMETER_IN) ) {
          mexErrMsgTxt("Input must be double.");
      } else if ( mxIsComplex(PARAMETER_IN) ) {
          mexErrMsgTxt("Input cannot be complex.");
      }

      ndim = mxGetNumberOfDimensions(PARAMETER_IN);
      dims = mxGetDimensions(PARAMETER_IN);
      numel = mxGetNumberOfElements(PARAMETER_IN);

      /* Create a matrix for the return argument */
      RESULT_OUT = mxCreateNumericArray(ndim, dims,
  mxDOUBLE_CLASS, mxREAL);

      /* Do the actual computation*/
      mexexp(mxGetPr(PARAMETER_IN),mxGetPr(RESULT_OUT),numel);
      return;
  }

Answer 1

[James Tursa]

For your particular posed example all the elements are the same, so only compute the exp calculation once and don't do all of those multiplies either. E.g.,

NN = complex(zeros(512^2,512)); tic; NN(:) = exp(1i); toc

Is this the real problem, or is MM not really all 1's in your actual problem?

[Ole]

Thank you. The matrix MM is ones just for simplicity. I would like to take inverse dft of an irregularly sampled complex function in k-space, and MM is the kernel of a dft. I cannot find available function. It is exp(kx*X+ky*Y).

[Jan]

@Ole: Usually it is impossible to give suggestions for speed improvements, if only a roughly simplified version of the code is posted. So please post the relevant part of the code.