[AOTI] Add back include_pytorch for specifying link paths (#126802)
Summary: Running dashboard with the cpp wrapper mode sometimes hit erros like "undefined symbol: aoti_torch_empty_stride", although it can not be reproduced locally and seems only happen on the dashboard CI.
Differential Revision: [D57911442](https://our.internmc.facebook.com/intern/diff/D57911442)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/126802
Approved by: https://github.com/chenyang78
ghstack dependencies: #126916, #127037
diff --git a/torch/_inductor/codecache.py b/torch/_inductor/codecache.py
index c47f017..b5aa1d1 100644
--- a/torch/_inductor/codecache.py
+++ b/torch/_inductor/codecache.py
@@ -2538,7 +2538,7 @@
cache: Dict[str, Callable[[], Union[CDLL, ModuleType]]] = {}
cache_clear = staticmethod(cache.clear)
cpp_compile_command_flags = {
- "include_pytorch": not config.abi_compatible,
+ "include_pytorch": True,
"shared": True,
}
entry_function = "inductor_entry_cpp"
