macOS arm64 support

[danomatika]

Add support for optimized libtensorflow builds for:

• macOS arm64 (M1)
• iOS (TensorFlow Light) possibly leave for later

References:

• Accelerating TensorFlow Performance on Mac
• Apple macOS Tensorflow Github repo
• TensorFlow iOS quickstart

[danomatika]

I tried out both 2.4.1 and 2.5.0 x86_64 libtensorflow builds in OF 0.11.2 on an M1 MBP 13 (arm64) and the examples build, but crash when running in Rosetta2. Building will only succeed if arm64 is added to the Excluded Architectures in Target's Build Settings.

[danomatika]

The update/macos-arm64 branch adds support for download non-official arm64 macOS builds from https://github.com/vodianyk/libtensorflow-cpu-darwin-arm64. Just run the script on an arm64 system to get arm64 libs, although not all TF versions are there, currently 2.5.0, 2.7.0, and 2.8.0.

I have a new macOS M1 Pro system to test this on soon.

[danomatika]

From doing some research, it looks like there are no plans for libtensorflow to support GPU usage or the Neural Engine on the Apple Silicon CPUs. I will settle for native CPU only builds on arm64.

As for iOS and TF Lite, I think we can leave that for later as a separate issue since there is some difference between TF2 and TF Lite models, as far as I can tell.

[danomatika]

Good news: the arm64 builds work fine on the macOS 12.2 M1 Pro system. I only needed to download the new libs, then change Build Settings->Excluded Architectures to x86_64 in each project. Clean & rebuild, done.

[danomatika]

The update/macos-arm64 branch is now merged with the develop.