CUDA threading allocation

I have gone through the CUDA programming guide and I cannot understand the t开发者_Go百科hread allocation method shown below:

dim3 dimGrid( 2, 2, 1 );
dim3 dimBlock( 4, 2, 2 );
KernelFunction<<< dimGrid, dimBlock >>>(. . .);

Can some explain how threads are allocated for the above condition?

An intuitive way to think about grid and block is to visualize them:

• Grid: A grid is a lattice of horizontal and vertical lines. Thus it has only 2 dimensions.
• Block: Think of a block of wood. It has all 3 dimensions: length, width and height.
• A block is made up of threads.
• A grid is made up of blocks.

Your dimBlock( 4, 2, 2 ) means that each block has 4 x 2 x 2 = 16 threads. Your dimGrid( 2, 2, 1 ) means that the grid has 2 x 2 x 1 = 4 blocks. Thus, your kernel is launched on a grid of 4 blocks, where each block has 16 threads. To conclude, your kernel will be launched with 16 x 4 = 64 threads.