Feat/vulkan

[thewh1teagle]

Add support for vulkan backend in latest whisper.cpp.

Depends on #169
Improve version of #168

---

Also changed whisper.cpp commit to a bit older commit since latest whisper.cpp is broken with vulkan.

---

Tested on Windows 11. It works fast.

[tazz4843]

Going to make one final release with OpenCL and then another major release with Vulkan, as I found a 80% performance loss in my testing when I tried Vulkan.

[thewh1teagle]

as I found a 80% performance loss in my testing when I tried Vulkan

What specific performance do you mean? the speed or the quality? (Because there's some issues I heard about the quality)
If it's about speed, which OS and what GPU you have?

[tazz4843]

Speed, Arch Linux with an Arc A770

[thewh1teagle]

Speed, Arch Linux with an Arc A770

If it's that drastic like 80% slower maybe the logs suggest that whisper fallback to CPU?

[tazz4843]

The logs do show whisper running on Vulkan. It's not a major hit to my use case to stay on an old version of whisper forever so I'm likely just going to do that.

ggml_vulkan: Found 1 Vulkan devices:
Vulkan0: Intel(R) Arc(tm) A770 Graphics (DG2) (Intel open-source Mesa driver) | uma: 0 | fp16: 1 | warp size: 32
whisper_model_load: Intel(R) Arc(tm) A770 Graphics (DG2) total size =  1176.00 MB
whisper_model_load: model size    = 1176.00 MB
whisper_backend_init_gpu: using Vulkan backend