Offload computation to GPU using Vulkan

[farooqkz]

Is your feature request related to a problem? Please describe.
Inference of neural networks such as RNNoise on CPU is slow and expensive. Inference on GPU consumes less power, has better execution speed and then CPU is free to do other tasks.

Describe the solution you'd like
Add an option to offload RNNoise inference to GPU using Vulkan.

Additional context
Another option to offload to GPU is OpenCL. But I've heard OpenCL is not good on Nvidia GPUs. Both Vulkan and OpenCL are supported on a vast range of GPUs(excluding Nvidia of course). Maybe it might be a good idea to add both OpenCL and Vulkan and let the user choose, with Vulkan being the default.

[Technetium1]

Is that supported upstream? Maybe related: #440

[TheDukeofErl]

Is that supported upstream?

I was wondering the same. The only things I see mentioned in #440 are SSE4.1 and AVX2, which are CPU-related technologies, not GPU.

It's not like running RNNoise on the CPU is particularly taxing to begin with.

[farooqkz]

Maybe open an issue upstream?

Edit: It seems the original repo is here: https://gitlab.xiph.org/xiph/rnnoise/