MATLAB: Need to maintain two matlab codebases for generating both 32 and 64 bit C or C++ code

MATLAB Compiler

On the mathwork webpage for "Upgrading MEX-Files to Use 64-Bit API",under the heading "Update Arguments Used to Call Functions in the 64-Bit API," it shows that I might need to replace my current value with mw types (the example shows ints being replaced by mwSize).

If I wanted be able to compile to both 32 and 64 bit, would I have to maintain two matlab code versions (using the example, one with int values and the other with mwSize), or would the 64 bit data structures be "downcast" to the 32 bit types, so I would need to maintain only one codebase?

Best Answer

When compiling with a 64 bit version of Matlab, mwSize is a 64 bit integer used for addressing. And in a 32 bit version, mwSize is automatically defined as 32 bit integer. So using mwSize, mwIndex and mwSignedIndex is enough to support both platforms.

In addition you can use uint32_T etc. whenever you need an specific integer type. This reduces the dependencies to the underlying platform and the effort needed for testing and debugging. These specific integer types are defined in "tmwtypes.h", which is included from "mex.h" automatically.

Related Solutions

MATLAB: Do I get an assertion violation when I port the 32-bit FORTRAN MEX source to a 64-bit MATLAB 7.6 (R2008a)

Some legacy FORTRAN code use the following syntax to define the MEXFUNCTION:

SUBROUTINE MEXFUNCTION(NLHS, PLHS, NRHS, PRHS)
INTEGER PLHS(*), PRHS(*)
INTEGER NLHS, NRHS

This might compile and execute on a 32-bit platform, however, the recommended syntax is to use MWPOINTER as shown below (for any platform).

This is essential for the MEX file to execute successfully on a 64-bit platform.

SUBROUTINE MEXFUNCTION(NLHS, PLHS, NRHS, PRHS)
MWPOINTER PLHS(*), PRHS(*)
INTEGER NLHS, NRHS

MATLAB: Is comparing type mwSize to an int valid

If int_T is a 32-bit integer and mwSize is a 64-bit integer (typically size_t, unsigned) then yes it is physically possible to have aproblem.

E.g., if n happens to be bigger than the largest value for an int_T then you would have an infinite loop because the i++ would typically wrap around to a negative value when it overflowed (not required for the compiler to do this, but typically that is what happens). Whether mxGetNumberOfElements could ever return a number this big is another question, and may be dependent on the particular system you are running on.

The i < n test itself may have a problem if int_T is a signed 64-bit and mwSize is an unsigned 64-bit and n happens to be larger than the largest int_T value. Again you would have an infinite loop for the same reason (i < n is always true). This case is not a practical issue, however, because the massive memory that such a value would imply would not be physically possible.

So, I would guess your code will probably work as long as the variable sizes are not too large.

Best Answer

Related Solutions

MATLAB: Do I get an assertion violation when I port the 32-bit FORTRAN MEX source to a 64-bit MATLAB 7.6 (R2008a)

MATLAB: Is comparing type mwSize to an int valid

Related Question