Mexing s-function with compiler flag -fshort-double fails....

Hello,

for numerical tests of a filter algorithm implemented in C++ (testing/simulation environment is a larger simulink project) I would like to switch to single precision variables. For the gcc/g++ the option -fshort-double seems to make switching to single precision very comfortable, because it basically shrinks doubles to the size of floats.

However I get the following warning when mexing the algorithm:

/opt/matlab/v24/simulink/include/simulink.c: In function ‘int_T ssWriteRTWParamSettings(SimStruct*, int_T, ...)’: /opt/matlab/v24/simulink/include/simulink.c:1538: warning: ‘real_T’ is promoted to ‘double’ when passed through ‘...’ /opt/matlab/v24/simulink/include/simulink.c:1538: note: (so you should pass ‘double’ not ‘real_T’ to ‘va_arg’) /opt/matlab/v24/simulink/include/simulink.c:1538: note: if this code is reached, the program will abort

As stated in the warning, the simulation doesn't run, but gives the error: Invalid sizes vector returned by MEX S-function

Can this be resolved or does anybody have an alternative way of testing an S-function with single precision.

Regards, Sebastian

[Titus Edelhofer]

Hi Sebastian,

I understand you are writing a Simulink S-function? I would not use the flag for the compiler but tell the S-function to take single precision inputs/outputs and parameters...

Titus

[Kaustubha Govind]

I agree with Titus. What you are doing will cause problems because the Simulink engine allocates and reads from signals and parameters that are supposedly double, but when it dispatched a call to MEX, it is really accessing single variables.

Answer 1

[Wendy Fullam]

Answered in comments by Titus

Hi Sebastian,

I understand you are writing a Simulink S-function? I would not use the flag for the compiler but tell the S-function to take single precision inputs/outputs and parameters...

Titus