I tried to make ROS package with CUDA. All is built and started, but CUDA doesn't work. I took a nvidia sample to realize:

using namespace std;
#include <iostream>

// CUDA headers
#include <cuda.h>
#include <cuda_runtime.h>

__global__ void add(int * a, int * b, int *c)
{
    int index = threadIdx.x + blockIdx.x * blockDim.x;
    c[index] = a[index] + b[index];
}

int main(int argc, char **argv)
{
    int *a, *b, *c;
    int *d_a, *d_b, *d_c;
    int length = 4;
    int N = length * length;
    size_t size = N * sizeof(int);

    a = (int *)malloc(size);
    for(size_t i = 0; i < N; ++i)
    {
        *(a+i) = i;
    }
    b = (int *)malloc(size);
    for(size_t i = 0; i < N; ++i)
    {
        *(b+i) = i;
    }
    c = (int *)malloc(size);

    cudaMalloc((void **)&d_a, size);
    cudaMalloc((void **)&d_b, size);
    cudaMalloc((void **)&d_c, size);

    cudaMemcpy(d_a, a, size, cudaMemcpyHostToDevice);
    cudaMemcpy(d_b, b, size, cudaMemcpyHostToDevice);

    add<<<length, length>>>(d_a, d_b, d_c);

    cudaMemcpy(c, d_c, size, cudaMemcpyDeviceToHost);
    cudaFree(d_a);
    cudaFree(d_b);
    cudaFree(d_c);


    for(size_t i = 0; i < N; ++i)
    {
        std::cout << *(a+i) << " + " << *(b + i) << " = " << *(c + i) << std::endl;
    }

    free(&a);
    free(&b);
    free(&c);
    return 0;
}

If I'm building without ROS and it`s working (CMakeLists.txt):

cmake_minimum_required(VERSION 2.8.3)
project(add_exec)

## Compile as C++11, supported in ROS Kinetic and newer
add_compile_options(-std=c++11)

find_package( CUDA REQUIRED )

set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -DMY_DEF=1")
set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DMY_DEF=1" )
set( CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DMY_DEF=1" )

## create an executable
cuda_add_executable( add_exec main.cu )
target_link_libraries( add_exec ${catkin_LIBRARIES} )

bur with ROS ...

cmake_minimum_required(VERSION 2.8.3)
project(ocv_gpu_test)

## Compile as C++11, supported in ROS Kinetic and newer
add_compile_options(-std=c++11)

## Find catkin and any catkin packages
find_package(catkin REQUIRED COMPONENTS roscpp rospy std_msgs genmsg)

## Generate added messages and services
generate_messages(DEPENDENCIES std_msgs)

## Declare a catkin package
catkin_package()

# Build add_exec and imclient
include_directories(include ${catkin_INCLUDE_DIRS})

find_package(CUDA REQUIRED)
find_package(VisionWorks NO_MODULE)
find_package(VisionWorks-NVXIO NO_MODULE)

# Compile-time definitions
add_definitions(-std=c++11 -DNVX)

set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} -DMY_DEF=1")
set( CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DMY_DEF=1" )
set( CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DMY_DEF=1" )

include_directories(
  ${catkin_INCLUDE_DIRS} 
  ${CUDA_INCLUDE_DIRS}
  ${VisionWorks_INCLUDE_DIRS}
  ${VisionWorks-NVXIO_INCLUDE_DIRS}
)

cuda_add_executable(add_exec src/main.cu )

target_link_libraries(add_exec 
    ${catkin_LIBRARIES} 
    ${CUDA_LIBS} 
    ${VisionWorks_LIBRARIES} )

is like as:

0 + 0 = 2566178
1 + 1 = 2566178
2 + 2 = 2566178
3 + 3 = 2566178
4 + 4 = 2566178
5 + 5 = 2566178
6 + 6 = 2566178
7 + 7 = 2566178
8 + 8 = 2566178
9 + 9 = 2566178
10 + 10 = 2566178
11 + 11 = 2566178
12 + 12 = 2566178
13 + 13 = 2566178
14 + 14 = 2566178
15 + 15 = 2566178
16 + 16 = 2566178
17 + 17 = 2566178
18 + 18 = 2566178
19 + 19 = 2566178
20 + 20 = 2566178
21 + 21 = 2566178
22 + 22 = 2566178
23 + 23 = 2566178
24 + 24 = 2566178
25 + 25 = 2566178
26 + 26 = 2566178
....

or

0 + 0 = 0
1 + 1 = 0
2 + 2 = 0
3 + 3 = 0
4 + 4 = 0
5 + 5 = 0
6 + 6 = 0
7 + 7 = 0
8 + 8 = 0
9 + 9 = 0
10 + 10 = 0
11 + 11 = 0
12 + 12 = 0
13 + 13 = 0
14 + 14 = 0
15 + 15 = 0
16 + 16 = 0