Question

I’m using Visual Studio 2010 and a GTX480 with compute capability 2.0.

I have tried setting sm to 2.0, but when I attempt to use printf() in a kernel, I get:

error : calling a host function(“printf”) from a __device__/__global__
function(“test”) is not allowed

This is my code:

#include "util\cuPrintf.cu"
#include <cuda.h>
#include <iostream>
#include <stdio.h>
#include <conio.h>
#include <cuda_runtime.h>

__global__ void test (void)
{
  printf("Hello, world from the device!\n");
}

void main(void)
{
    test<<<1,1>>>();
    getch();
}

I find a example here: “CUDA_C_Programming_Guide” ‘page _106’ “B.16.4 Examples”
at last,it is work for me 😀 thank you.

#include "stdio.h"
#include <conio.h>

// printf() is only supported
// for devices of compute capability 2.0 and higher

  #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ < 200)
      #define printf(f, ...) ((void)(f, __VA_ARGS__),0)
  #endif


__global__ void helloCUDA(float f)
{
    printf("Hello thread %d, f=%f\n", threadIdx.x, f);
}

int main()
{
    helloCUDA<<<1, 5>>>(1.2345f);
    cudaDeviceSynchronize();
    getch();
    return 0;
}

Answer 1

To use printf in kernel code, you have to do three things:

1. make sure that cstdio or stdio.h are included in the kernel compilation unit. CUDA implements kernel printf by overloading, so you must include that file
2. Compile your code for compute capability 2.x or 3.x and run it on a supported GPU (so pass something like -arch=sm_20 to nvcc or the IDE equivalent in Visual Studio or Nsight Eclipse edition)
3. Ensure that the kernel has finished running by including an explicit or implicit synchronization point in your host code (cudaDeviceSynchronize for example).