•_CUDAFor development plan

[ashermancinelli]

The following system modifier is now available with limited support:

$ ./BQN -e '•Show (↕10) + •_CUDAFor ↕10'
⟨ 0 2 4 6 8 10 12 14 16 18 ⟩

What should the development plan be for CUDA/GPU support after the runtime is implemented on the device?

[ashermancinelli]

From Marshall on matrix:

You could handle blocks even, by replacing primitives that are loaded as constants. Well, you have to assume that it doesn't pull in primitives from outside variables. Probably best to scan the variable uses and reject anything that refers to outside variables anyway though.

[ashermancinelli]

This and all other accelerator options should be brought under the namespace •cxbqn. Perhaps:

   cu←•cxbqn.accelerate.cuda
   (↕10) + cu._for ↕10
⟨ 0 2 4 6 8 10 12 14 16 18 ⟩