CUDA suppress or otherwise get rid of "Calling host funciton from a host device functions is not allowed" warning

[Goubermouche]

I have code that looks something like this:

__host__ __device__ void Bar() {}

__host__ void Foo() {
    Bar();
}

This code produces the following warnings: #20014-D and #20011-D - how can I get around the compiler throwing these warnings at me? I’ve read that NVCC has its own pragma suppress, but when I tried it it didn’t work. I think separating the Bar function into a __host__ and __device__ implementation would probably work, but it doesn’t feel very clean. Any advice is very much appreciated.

[Robert_Crovella]

FWIW your example is not what would trigger the warning you indicate. (You are calling a host/device function from a host function which is entirely ok and would not trigger a warning. The warning, as indicated, comes from calling a host function from a host/device function.)

Otherwise this or this may be of interest.