how to read and write half precision arrays via mexFunction

#include "mex.h"

#include "rtwhalf.h"

#define IN_X prhs[0]

#define IN_Y prhs[1]

#define OUT_Z plhs[0]

void mexFunction( int nlhs, mxArray *plhs[], int nrhs, const mxArray *prhs[] ) {

real16_T *X, *Y, *Z;

size_t ndim=3;

size_t *dims;

dims[0] = 128; dims[1] = 128; dims[2] = 128;

X = (real16_T*)(mxGetPr(IN_X));

Y = (real16_T*)(mxGetPr(IN_Y));

OUT_Z = mxCreateNumericArray(ndim, dims, mxSINGLE_CLASS, mxREAL);

Z = (real16_T*)(mxGetPr(OUT_Z));

}

https://www.mathworks.com/help/coder/ug/edge-detection-with-sobel-method-in-half-precision.html

By referring the link, I wanted to use half-precision float arrays in a MEX file. However, it failed, though the code worked for single/double precision float arrays. Is there any way to use half-precision float arrays in the MEX file? Thank you.

Answer 1

[James Tursa]

Very unfortunately, MATLAB has implemented the half precision data type as an opaque classdef type instead of a simple numeric type. The half precision data is hidden and cannot be accessed directly in a mex routine. You would have to use helper routines at the m-file level to get at the data and then pass that into the mex routine. See these related discussions:

https://www.mathworks.com/matlabcentral/answers/520582-why-is-the-new-half-data-type-an-opaque-class?s_tid=srchtitle

https://www.mathworks.com/matlabcentral/answers/520544-how-to-convert-half-precision-number-into-hexadecimal-representation?s_tid=answers_rc1-1_p1_BOTH