How to wrap on overflow when transfer a double to integer

I would like to transfer a double to an unsigned integer, specific 16-bit integer, with WRAP on overflow in Matlab coding.

I currently use, but it SATURATEs on overflow

y = uint16(x);

Does anyone have solution?

Answer 1

[James Tursa]

You could use:

y = mod(x,double(intmax('uint16'))+1);

But, if x is too large so that eps(x) > 1 the result might be somewhat meaningless.

[Alex]

Thanks James Tursa for your suggestion.

I also would like to avod use MOD function for DOUBLE format.

I have tried to transfer double uint32 format, then use the MOD function. However, I would like to find a better solution.

[James Tursa]

Why don't you like MOD? That's essentially the functionality you are trying to reproduce, so why avoid it? I don't understand this restriction. I suppose you could call a mex function and have the C compiler do a natural conversion, but this seems unnecessary in this case.

[Alex]

Hi James Tursa,

I am going to implement code for Real Time application. I have tested and the MOD function take alot of time to execute. I need a solution which should be fast for Real Time.

Thank you for your suggestion. I am going to have a look on MEX file function.

Kind Regarcds

[James Tursa]

The guts of the mex code is simply

double d;
unsigned short u; /* or whatever type your compiler will support for 16-bit unsigned integer */
  :
u = d;

Maybe let the coder generate the saturation code that you don't want, and then go in and manually change it to the above one-liner.

[Alex]

Thank you for your advice.

But there is an error when I did. What was I wrong?

The 'Double2Uint.c' file is

double d;
unsigned short u;

u = d;

In Matlab, I type

mex Double2Uint.c

Then it showed errors.

>> mex Double2Uint.c
Building with 'MinGW64 Compiler (C)'.
Error using mex
C:\---\Documents\MATLAB\Double2Uint.c:4:1: warning: data
definition has no type or storage class
 u = d;
 ^
C:\---\Documents\MATLAB\Double2Uint.c:4:1: warning: type defaults
to 'int' in declaration of 'u' [-Wimplicit-int]
C:\---\Documents\MATLAB\Double2Uint.c:4:1: error: conflicting
types for 'u'
C:\---\Documents\MATLAB\Double2Uint.c:2:16: note: previous
declaration of 'u' was here
 unsigned short u;
                ^
C:\---\Documents\MATLAB\Double2Uint.c:4:5: error: initializer
element is not constant
 u = d;
     ^

Could you give me more detail.

Thanks

[James Tursa]

Sorry. I just gave you the gist of the code, not the complete code. A complete mex file would be:

(EDIT: Updated code to account for mwSize possibly being different size than size_t)

/* Convert a full real double input to a uint16 output with natural mod conversion */
#include "mex.h"
#include <stdint.h>
void mexFunction(int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[])
{
    double *d;
    uint16_t *u;
    size_t n, ndims;
    size_t *sdims;
    mwSize *mdims;
    
    if( nrhs != 1 || !mxIsDouble(prhs[0]) || mxIsComplex(prhs[0]) || mxIsSparse(prhs[0]) ) {
        mexErrMsgTxt("Need exactly one full real double input");
    }
    if( nlhs > 1 ) {
        mexErrMsgTxt("Too many outputs");
    }
    ndims = mxGetNumberOfDimensions(prhs[0]);
    mdims = (mwSize *) mxGetDimensions(prhs[0]);
    sdims = (size_t *) mxMalloc(ndims*sizeof(*sdims));
    for( n = 0; n<ndims; n++ ) {
        sdims[n] = mdims[n];
    }
    plhs[0] = mxCreateUninitNumericArray(ndims, sdims, mxUINT16_CLASS, mxREAL);
    mxFree(sdims);
    d = (double *) mxGetData(prhs[0]);
    u = (uint16_t *) mxGetData(plhs[0]);
    n = mxGetNumberOfElements(prhs[0]);
    while( n-- ) {
        *u++ = *d++;
    }
}

[Alex]

Dear James Tursa,

Thank you for your help. It works.

Regards