Possible typo in CUDA Programming Guide v13.3: cuda.threadIdx and cuda.gridDim descriptions appear to be swapped in the Python table

Accelerated Computing GPU-Accelerated Libraries
1

I think there may be a documentation typo in the CUDA Programming Guide v13.3.

Page:

Section:

2.3.2. Thread Hierarchy

In the Python list of CUDA built-in variables, the descriptions of cuda.threadIdx.[xyz] and cuda.gridDim.[xyz] appear to be swapped.

The current text says:

  • cuda.threadIdx.[xyz]: Size of the grid in the x, y and z dimension respectively.
  • cuda.gridDim.[xyz]: Index of the thread in the x, y and z dimension respectively.

I believe this should be:

  • cuda.threadIdx.[xyz]: Index of the thread within its thread block in the x, y, and z dimensions respectively.
  • cuda.gridDim.[xyz]: Size of the grid in the x, y, and z dimensions respectively.

This interpretation is consistent with the preceding paragraph in the same section, which says that the grid size can be queried with gridDim, the block size with blockDim, the block index with blockIdx, and the thread index with threadIdx.

It is also consistent with the C++ list immediately above the Python list:

  • gridDim.[x|y|z]: Size of the grid
  • threadIdx.[x|y|z]: Index of the thread

Could someone from NVIDIA confirm whether this is a typo in the Python list?

2

Yes, I believe you are correct. I suggest filing a bug.

3

Thank you for confirming.

I have filed a bug report:

https://developer.nvidia.com/bugs/6264401

Thanks for your suggestion and assistance.