libcudacxx_force_include.h

// This is how we run libc++ tests on the GPU without modification.
// We force include this header into each test with `-include`.

__host__ __device__
int fake_main(int, char**);

__global__
void fake_main_kernel(int * ret)
{
    *ret = fake_main(0, NULL);
}

#define CUDA_CALL(err, ...) \
    do { \
        err = __VA_ARGS__; \
        if (err != cudaSuccess) \
        { \
            printf("CUDA ERROR: %s: %s\n", \
                   cudaGetErrorName(err), cudaGetErrorString(err)); \
            return err; \
        } \
    } while (false)

int main(int argc, char** argv)
{
    // Check if the CUDA driver/runtime are installed and working for sanity.
    cudaError_t err;
    CUDA_CALL(err, cudaDeviceSynchronize());

    list_devices();

    int ret = fake_main(argc, argv);
    if (ret != 0)
    {
        return ret;
    }

    int * cuda_ret = nullptr;
    CUDA_CALL(err, cudaMalloc(&cuda_ret, sizeof(int)));

    fake_main_kernel<<<1, 1>>>(cuda_ret);
    CUDA_CALL(err, cudaGetLastError());
    CUDA_CALL(err, cudaDeviceSynchronize());
    CUDA_CALL(err, cudaMemcpy(&ret, cuda_ret, sizeof(int), cudaMemcpyDeviceToHost));
    CUDA_CALL(err, cudaFree(cuda_ret));

    return ret;
}

#define main fake_main