[nnc][mobile] temporarily disable quantization external functions (#68029)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68029
Temporarily disable quantization external functions with a new macro DISABLE_NNC_QUANTIZATION.
The ATen CPU library consists of two parts:
A. Common operator functions, e.g. "at::empty()", the list of sources can be found at "aten_cpu_source_list" in "tools/build_variables.bzl";
B. Implementations of these operators, e.g. "at::native::empty()", the list of sources is defined at "aten_native_source_list" in "tools/build_variables.bzl";
Note that A does not directly depend on B. A calls B via dispatch table. The dependency is injected into the dispatch table by B during its static initialization.
For internal mobile builds, B is built on a per-app basis. A is the public library for other libraries to depend on. Because these external functions call quantization functions that are not part of A, the NNC kernel library cannot resolve the missing symbols.
Use this PR to unblock the internal experiment until we figure out a better solution (e.g. move quantization API to A).
ghstack-source-id: 142868370
Test Plan: Make sure it can build with the stacked diff.
Reviewed By: IvanKobzarev
Differential Revision: D32239783
fbshipit-source-id: 3797b14104b0f54fb527bc3fc5be7f09cc93d9e4
