Kernel not launching when "if" present...

Accelerated Computing CUDA CUDA Programming and Performance
1

I’m working on some simple program and I’m stumbling on the following situation. My program executes well when the kernel contains:

 unsigned int td = d[tid];

 for (unsigned int j = 0; j < 16; ++j) {

    atomicMax (&(hit[2]), tid);

    atomicMin (&(hit[0]), tid);

    __syncthreads ();

    td <<= 2;

    td += (td_next >> 30);

    td_next <<= 2;

  }

But if the to atomic functions are put within an if, then I get “unspecified launch failure” when adding a CUT_CHECK_ERROR() just after the kernel launch:

 unsigned int td = d[tid];

 for (unsigned int j = 0; j < 16; ++j) {

    if (td == 0) {

      atomicMax (&(hit[2]), tid);

      atomicMin (&(hit[0]), tid);

    }

    __syncthreads ();

    td <<= 2;

    td += (td_next >> 30);

    td_next <<= 2;

  }

Any idea on what might go wrong? Any insights on where to look?