Are you using AMD’s memory allocator? That dramatically simplifies allocation.

Bindless also simplifies all the descriptors set shenanigans

My favourite was doing viewport plugins in Maya, where Maya would leave random bits of GL state dirty for you…

How does it compare to the OpenGL version?

What kind of vulkan are you targeting? “Modern vulkan”, as in bindless everything, buffer device pointers and shader objects is a _lot_ nicer ime

Apparently it’s owned by a third party, not The Mill.

Backprop? Turn the graph into an optimizer?

Debugging why cmake fails to find the correct dependencies in a GitHub Action is a special kind of hell.

I had to implement zoom-around-cursor (don’t know the proper name for it) recently for an image viewer and it took me an embarrassingly long time to figure out.

Yeah having to spin up a whole OpenGL or Vulkan just to boot pixels to the screen is painful. But I’d love to have access to the graphics hardware for exactly the wireframe/simple polygon sorts of use cases.

We really need a CUDA graphics API :)

Time to just write a line rasterizer in CUDA?