Use GPU acceleration for Apple Silicon

[rehans]

Copy needed files into bundle

[rehans]

Some observations about performance:

• Whispercpp is faster when it does not need to find out the language spoken in the audio file. So giving the language to whispercpp instead of -l "auto" use -l "en" speeds the process up quite a bit
• The ggml-metal.metal file needs to be put into the Resources folder inside the bundle. (but which ggml-metal.metal file, there are several. Somehow this file can be compiled into .metallibfiles?!)

[rehans]

Learnings

• Processing executed in parallel on the GPU can use a Shading language called Metal’s original Metal Shading Language (MSL).
• A MSL filename extension is .metal e.g. ggml-metal.metal.
• Included into a project XCode can compile .metal files into a library called default.metallib. This library is automatically added to the app bundle's Contents/Resources subfolder. (MSL files can also be built manually: https://github.com/ggerganov/llama.cpp/issues/5977#issuecomment-1987295222)
• MSL files can also be added to the app bundle's Contents/Resources subfolder (as source files WITHOUT being compiled beforehand).

Source: https://gist.github.com/MilesLitteral/1a5d9efade94a11cd0898dbee7727775

Whisper takes all options into account. It searches for

• default.metallib,
• ggml-metal.metal next to the executable (if it is not a bundle) and
• ggml-metal.metal inside the Contents/Resources subfolder of the bundle.

Source: https://github.com/ggerganov/whisper.cpp/blob/master/ggml-metal.m#L313

[rehans]

@MaxDev0101 Sorry, I forgot to make a PR ;) But I just added both files to CMake:

https://github.com/max-and-me/wordify-plugin/commit/625b06edd66fa29671544fee1e2082b268336301#diff-1e7de1ae2d059d21e1dd75d5812d5a34b0222cef273b7c3a2af62eb747f9d20aR236