Fixed child offset computation in nanovdb::tools::cuda::indexToGrid

[heandreas]

I ran into a problem when using nanovdb::tools::cuda::indexToGrid<nanovdb::Vec3f> to build a grid of Vec3fs from an index grid. After debugging for a while I managed to find a fix that works for me. I believe the computation for dstTile.child in processRootTilesKernel is not correct for all types. It seems to work for float and double grids, but not for Vec3f and probably many (or all?) other types that do not align with the size of the index type. My fix is inspired by the code in processNodesKernel that also computes similar byte offsets.

I extended the unit test CudaIndexGridToGrid_ValueOnIndex to show the problem. It only passes with my fix and not the original code. Feedback is highly appreciated!

[linux-foundation-easycla]

• ❌ The email address for the commit (9f73006) is not linked to the GitHub account, preventing the EasyCLA check. Consult this Help Article and GitHub Help to resolve. (To view the commit's email address, add .patch at the end of this PR page's URL.) For further assistance with EasyCLA, please submit a support request ticket.

[kmuseth]

thanks @heandreas ! Looking at your changes now :)

[kmuseth]

yes this does indeed look like a valid fix - great catch!

since you didn't sign the EasyCLA I'm taking the liberty to re-write your PR slightly with more documentation explaining how this fix works

[heandreas]

Sounds great, thanks for looking into it!

[kmuseth]

@heandreas - see PR #2021 - I added a credit to you - let me know if you prefer to be anonymous.

[heandreas]

@kmuseth Thanks, I'd be delighted to show up in the nanovdb changelog :-)

[kmuseth]

@heandreas - take a look here