Why printf is not work in 2D blocks?

Accelerated Computing CUDA CUDA Programming and Performance
1

I tried to print values of parameters. I use kernel as follow:

-------------------------test.cu-----------------------------------

#include <stdio.h>
__global__ void kernel()
{
        int x = threadIdx.x;
        int y = threadIdx.y;

        printf("x = %d, y = %d.\n", x, y);
}

int main()
{
        dim3 dimBlock(32, 3, 1);
        dim3 dimGrid(1, 0, 0);

        kernel<<<dimGrid, dimBlock>>>();
        cudaDeviceSynchronize();

        return 0;
}

Compiled code like this:
nvcc x.cu -arch=sm_35

The program printed nothing.

I used K20m + CUDA 5.5 + SLES11 SP2.

BTW: I found the printf can work fine in 1D blocks.

Why not in 2D blocks?

Thanks

2

in my experience when the printf is not done it is because it is not executed. Your kernel is very simple and I do not see any reason why it is should not run, so maybe the printf is optimized, because there are not calculations.

3

The simpler kernel with only one “printf(“Hello”);” statement was tested in 1d block.
It worked fine in my system.

4

This means your kernel is not executed.

5

Your grid is invalid:

dim3 dimGrid(1, 0, 0);

should be

dim3 dimGrid(1, 1, 1);