fix: Build TensorFlow Lite C Framework for x86_64 (simulator) as well

[mrousavy]

Builds the TensorFlow Lite C Framework for x86 (64-bit) multi-CPU as well (used in the iOS Simulator)

[mrousavy]

nvm the command I added does not work. I am not even sure if this really is a problem, I had this running on an iOS simulator on arm64. I think this might be an issue with the user's RN app, not with Fast TFLite.

[Chasty]

I could not run in simulator for arm64 arch.
@mrousavy

[computerjazz]

There are several open issues suggesting that tflite does not currently work with arm64 sims:

tensorflow/tensorflow#47400
am15h/tflite_flutter_plugin#153 (comment)

In the meantime, i can get it up and running via a Rosetta simulator. And I've seen some mention that the upstream issue has been fixed, but not released yet -- although this comment is over a year old now: tensorflow/tensorflow#47400 (comment)

[computerjazz]

I just spent a few hours going town a tflite rabbit hole and verified that if I build tflite ONLY for ios_sim_arm64 it builds and runs in the simulator. However, building for ios_multi_cpus and including ios_sim_arm64 as one of the options fails to build.

working build command for sim only:

bazel build --config=ios_sim_arm64  -c opt --cxxopt=--std=c++17 \                                
  //tensorflow/lite/ios:TensorFlowLiteC_framework

non-working build command, hoping to cover all architectures:

bazel build --ios_multi_cpus=armv7,arm64,i386,x86_64,ios_sim_arm64  -c opt --cxxopt=--std=c++17 \                                               
  //tensorflow/lite/ios:TensorFlowLiteC_framework

So, I'm not seeing a path forward for being able to run on ios sim outside of Rosetta, unless there's a way to include both sim-only and other-cpu binaries in the project and choose which one to include at build time, based on destination architecture.

PR that added arm64 sim support: tensorflow/tensorflow#55777