Revision history [back]

It is not impossible to pass and manipulate multi-dimensional arrays in CUDA. In fact, CUDA provides support for multi-dimensional arrays through the use of the "dim3" type, which represents a 3D block or grid configuration. However, it is important to note that in many cases, manipulating multi-dimensional arrays in CUDA will require additional memory access and computation overhead, which can impact performance. Therefore, it is often recommended to use flattened 1D arrays instead, for better data locality and cache utilization.