[code-reflection] RFR: [hat][cuda] 2D/3D NDrange supported for CUDA

[Gary Frost]

On Wed, 30 Jul 2025 12:19:58 GMT, Juan Fumero <duke at openjdk.org> wrote:

> This PR extends the 2D and 3D NDRange developed for the OpenCL backend/runtime for CUDA devices. 
> In addition, it improves the thread-block selection. 
> 
> The heuristic for thread-block selection here is quite simple (static decision based on the number of dimensions). However, in the future we can improve this. My goal with this PR is to showcase 2D and 2D ranges for CUDA compatible devices. 
> 
> How to test?
> 
> 
> java @hat/run ffi-cuda matmul 1D
> 
> java @hat/run ffi-cuda matmul 2D
> 
> # Old examples should continue to work
> java @hat/run ffi-cuda squares

/Sponsor

-------------

PR Comment: https://git.openjdk.org/babylon/pull/504#issuecomment-3137354168