PROCESSOR-SDK-J784S4: cfloat operator*(float, cfloat)

Please understand that, while the C++ standards do not define cfloat as a primitive built-in type, the C7x compiler does.  An operator function cannot be defined for a primitive type.  For more details, please see this FAQ (not from TI).  

Thierry BERNIER said:

This overloading works with HostEmu. It fails with CGT 4.1.0.

In HE (host emulation), cfloat is modeled as a C++ class, which means you can define an operator function.  This is a limitation of using HE.

Thanks and regards,

-George

Yes, that was also my understanding, as (poorly) said in my ticket : cfloat is a CGT extension of C/C++ langage.

My request is that this extension shall be (more) complete, i.e. provides the same operations as the STL does with std::complex<>.

Most notably the external math product between real and complex values.

Thierry BERNIER said:

My request is that this extension shall be (more) complete, i.e. provides the same operations as the STL does with std::complex<>.

Because std::complex is not a built-in type, it is possible to define the meaning of the multiply operator * for it.  Because cfloat is a built-in type, the meaning of the multiply operator * is pre-defined, and there is no method for changing it.  I suppose we could define an intrinsic for multiplying a cfloat and a float.  But there is no meaningful difference between that and you writing a function that does the same thing.

Thanks and regards,

-George

I understand the built-in type limitations, and this is my request.

As TI extended the C langage providing cfloat as a std::complex<float> imitation, it would be fair for the imitation to be complete.
The same C++ rules forbids to overload operator* as cfloat is built-in. So I shall write a "static inline" function (and EXT_EP-11846 may apply).

- it's less readable

- it's potentialy weaker for the optimizer

- what about vector/SIMD case ?

Think of :

cfloat z1, z2, z3;
float f1, f2, f3

// usual and std portable
z1 = f2 * z2 + f3 * z3;

// not in the CppCoreGuidelines spirit, not overloadable => distinct name, etc
z1 = mystd::multfz(f2, z2) + mystd::multfz(f3, z3);