Access to lower level HAL

[hmaarrfk]

I read about https://docs.rs/wgpu-hal/latest/wgpu_hal/

and how it might be useful to take a buffer from WGPU and to send it to an other pipeline (perhaps cupy or other cuda accelerated workflow).

Is there any value to this?

I would like to avoid

GPU -> CPU -> GPU just to "move things in and out of WGPU" but not sure how stable the wgpu-hal is and how it would fit into wgpu-py.

Thx for your thoughts.

[almarklein]

I'm not sure how hard/possible is is to get access to the lower-level objects. These issues are relevant here:

• Proposal for Underlying Api Interoperability gfx-rs/wgpu#4067
• Texture memory import API gfx-rs/wgpu#2320