A grid-stride float sum that ends in one atomicAdd per warp

*total is zeroed before each run, blockDim.x is 256, and the input, grid size, and GPU are identical every time. Comparing runs, the printed sum differs in the last two or three significant digits from run to run. Which statement is correct?

sum.cu
__global__ void sumKernel(const float* x, float* total, int n) {
    float v = 0.0f;
    for (int i = blockIdx.x * blockDim.x + threadIdx.x; i < n;
         i += gridDim.x * blockDim.x)
        v += x[i];

    for (int off = 16; off > 0; off >>= 1)
        v += __shfl_down_sync(0xffffffff, v, off);

    if ((threadIdx.x & 31) == 0)
        atomicAdd(total, v);
}

45 of 46 in GPU / CUDA