#include "cuda.h"
#include <stdio.h>
#include <math.h>

__global__ void fp_test(int a, int b, float c)
{
  int val_gpu = ceilf( (a+b) * c / (1.0f+c) );
  printf("gpu: %dn", val_gpu);
}

int main()
{
  int a = 5;
  int b = 4;
  float c = 0.8;

  fp_test<<<1,1>>>(a, b, c);
  cudaDeviceSynchronize();

  int val_cpu = ceilf( (a+b) * c / (1.0+c) );
  printf("cpu: %dn", val_cpu);
}

$ nvcc -arch=sm_20 gpu_fp_test.cu -o gpu_fp_test
$ gpu_fp_test
gpu: 5
cpu: 4

int val_gpu = ceilf( (a+b) * c / (1.0+c) );
                                     ^
                                     remove f from constant

int val_cpu = ceilf( (a+b) * c / (1.0f+c) );
                                     ^
                                     add f to constant