Archives Discussions

[tomschi]

i have the following code that yields an compile-time error (using OpenCL compiler, AMD APP 1445.5 on Intel Core i7, using static C++ extensions):

namespace myNS

{

  using ::fmin;

}

void test()

{

  float f = myNS::fmin(1.0f, 2.0f);

}

results in

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

Error: "fmin" has already been declared in the current scope

"OpenCL-CPP-JIT-Source", line 9: error: a value of type "double16" cannot be

          used to initialize an entity of type "float"

    float f = myNS::fmin(1.0f, 2.0f);

i can get rid of the error if i insert the following declaration at the begin of the code snippet:

float fmin(float, float);

This is not specific to fmin, but i assume for all built-in, overloaded (?) functions (i've only tested a few: fabs, min, max...)

Why do i have to declare the built-in function in this case first?

Have i missed something or is this a compiler problem ?

thanks and regards,

  tomschi

[dipak]

Hi,

I checked it and my observation was same as yours. I've asked someone from compiler team and will get back to you as soon as I get any reply.

Regards,

[dipak]

Hi,

As I came to know that the compilation errors are due to conflict with the built-in functions defined in our implementation. You can use those built-in functions directly from your own code as follows:

void test()

{

  float f = ::fmin(1.0f, 2.0f);

}

Regards

[tomschi]

thanks for the reply,

i'm aware of the fact that my code compiles using the built-in from the global namespace,

but why can't a built-in function not be referenced in a using declaration without declaring it first?

i don't know any internals of the OpenCL compiler, but i always thought that built-ins ressemble

already defined functions (just as a theoretical model) and thus a function declaration in some sense (float fmin(float, float))

should be already present  (and i don't have to write it 🙂

my use case is a wrapper library (OpenCL/CUDA/Host C++), where i would like to have a "fmin" in a dedicated namespace.

i also would like to avoid implementing a dummy fmin in the namespace that simply calls the ::fmin (avoid any redudant piece of code, also the declaration if possible).

thanks and regards,

  tomschi