Orthogonal BlockScan's for Multidimensional Thread Layout

[Hprairie]

I found working with multi-dimensional data with cub to be confusing, but think I got most of my questions answered by #1653 .

However, I am curious if there is a way to essentially do parallel scan operations along a specific thread axis. For example, lets say that I have the following setup.

• block size = (32, 32, 1)
• grid size = (1, 1, 1)
• Items / thread = 1
  -thread_data = [1]

Here for exclusive scan we would end up with something like this when looking at it from a thread view.

{ { 0, 1, 2, 3, .... , 31},
{32, 33, 34, ....., 63},
....}

However, I was wondering how I would go about doing an orthogonal scan across a specific thread axis. I.e. if I want to get something like this instead.

{ { 0, 1, 2, 3, .... , 31},
{ 0, 1, 2, 3, .... , 31},
....
{ 0, 1, 2, 3, .... , 31}}

I would do a scan across the x thread axis.

I am newish to CUDA, so any help with this would be very much appreciated!

[Hprairie]

Ahh from my understanding I should be able to use LOGICAL_WARP_THREADS for cub::WarpScan.