C28x compiler 6.1.2: intrinsic declaration is incompatible

StephanS said:

So how can I define these intrinsics correctly?

I recommend ...

#ifndef __TMS320C2000__                      // line 1                 
extern void __mswapf(float a , float b );    // line 2
#endif                                       // line 3

Lines 1 and 3 mean line 2 will not be seen by the TI C2000 compiler.  Only your static analysis tool will see it.  The symbol __TMS320C2000__ is predefined by the C2000 compiler.

The problem is the semantics (what it actually does) of the __mswapf intrinsic cannot be represented in C.  It borrows the reference concept from C++.  For example, "float &" is a reference to a float variable.  This intrinsic swaps the contents of a and b without using any pointers.  You cannot represent that in C.

StephanS said:

Same problem for __f32_max_idx, __f32_min_idx

Yes, it is.  Though these are not CLA intrinsics, but regular compiler intrinsics.

StephanS said:

- The __fmax and __fmin definitions are not
float         __fmax( float x, float y ) as in spru514e.pdf, but
double         __fmax( double x, double y )
Is this a compiler or documentation error?

These intrinsics also use references, which means these declarations must also not be seen by the C2000 compiler.  Since float and double are both 32-bits, you can define them however you like.

Thanks and regards,

-George