Release Mode

[hariravi]

Hi All,

I am unable to get the code to compile in release mode (as opposed to debug), I get EXC_BAD_ACCESS (in run scheme I changed everything from debug to release). How can I overcome this (I have my own personal app using libtorch and I cannot push it to testflight).

Thank you very much for the help, and great demo.

[hariravi]

Is there a way to perhaps compile models in Debug and Release modes? It seems like the underlying structure of the .pt models are just dictionaries (so I think they trace_script.py should be correct for both), I also noticed LibTorch.debug.xcconfig and LibTorch.release.xcconfig seem identical.

Thank you again, been struggling debugging this, was very surprised to see when I released my product to testflight it would crash when the model attempted to predict!

[harshthaker]

I am having the same issue. It crashes in release mode. Any solution?

[hariravi]

In my build settings I set my release flags to mirror debug, and that fixed it!

[harshthaker]

@hariravi thanks. Can you describe it in detail?

[harshthaker]

I can see Header Search Path as: "$(PODS_ROOT)/LibTorch/install/include/"

and PODS_ROOT is set to path-to-project-on-machine/Pods

is it this thing causing the trouble?

[hariravi]

I am actually not sure it might have been the Optimization Level (change it to None on release)? I just tried changing things one by one and eventually it worked, I know the issue is the scheme, and try changing Release flags to mirror the Debug (start with optimization level)!

[harshthaker]

Thanks @hariravi. I changed Optimization level flag to None for the release mode and it runs okay now.

[MrHjalmarsson]

Hi!
Is there another solution to this problem? Is this the expected behaviour?
For me it's not a solution to turn of the optimization for a release app.

[pac00l]

Disable NNPACK

[sasan-soroush]

But is there a better solution?
Changing the optimization level to none is not an option for us.