OpenCL "virtual" functions

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

Does OpenCL support some kind of C++ virtual functions or C function pointers as kernel arguments?

The idea is to call functions compiled from other .cl file.

For example, I have a file called MyCustomAdd.cl

__device int MyCustomAdd ( const int a, const int b)

{

   return a+b;

}

and I have this kernel.cl file

__device int (*deviceFunctionPointer) ( const int a, const int b );

__kernel __device void MyOpenCLKernel ( deviceFunctionPointer *p )

{

   //Do some things...

//Now call the user-defined function pointer

   const int result = deviceFunctionPointer(1,2);

}

in my CPU code I could do this:

void main ()

{

   //init OpenCL, etc...

//load the user-defined shader function/interface

   clModule *mod = clCompileModule("MyCustomAdd.cl");

   void *MyCustomAddPtr = clGetDeviceFunction(mod,"MyCustomAdd");

//load my kernel

   clModule *krnMod = clCompileModule("kernel.cl");

//fire my kernel

   array<void*> krnParams;

   krnParams.push_back(MyCustomAdd);

clExecuteKernel ( krnMod, "MyOpenCLKernel", 256_threads_per_block, krnParams );

}

Thanks

2

No, OpenCL doesn’t support function pointers in device code (most current hardware doesn’t support jumping to computed addresses).