3.1 Multidimensional Grid Organization

• What are the two levels of hierarchy in CUDA’s execution model?
• What built-in CUDA variable gives the index of a thread within a block?
• What built-in CUDA variable gives the index of a block within a grid?
• How do you specify the grid and block dimensions in a CUDA kernel call?
• How many dimensions can CUDA grids and blocks have?
• What is the maximum number of threads per block in CUDA?
• What is the maximum gridDim.x in CUDA?
• What is the maximum gridDim.y and gridDim.z in CUDA?
• What is the default value for the y and z dimensions when using a 1D dim3 grid?
• What happens if a CUDA grid is too large for the hardware?

3.2 Mapping Threads to Multidimensional Data

• What kind of problems benefit from a 2D thread grid?
• How do you compute the global row index of a thread in a 2D block?
• How do you compute the global column index of a thread in a 2D block?
• Why do CUDA image processing kernels often launch more threads than necessary?
• What does an if-condition typically check in a CUDA image processing kernel?
• What happens if threads access out-of-bounds memory in CUDA?
• How does CUDA handle dynamically allocated 2D arrays?