Support build on native Windows

[python3kgae]

🚀 The feature, motivation and pitch

Currently, executorch supports Windows via WSL. It would be beneficial to enable native Windows builds as well, as this would assist users who do not have WSL installed.

Alternatives

Executorch already supports Windows via WSL.

Additional context

No response

RFC (Optional)

The plan is to create a PowerShell version of install_requirements.sh.

Here is a list of things that need to be updated for building on native Windows:

• [ ] Resolve build issues in file_data_loader.cpp, which includes , and invoke ::close on -1.
• [ ] Get the Windows version of buck2 for Windows build in build/resolve_buck.py.
• [x] Check the path based on the build configuration for Windows in setup.py. For example, third-party/flatbuffers/flatc should be third-party/flatbuffers/Debug/flatc.exe for Debug builds.
• [x] Avoid generating 'def flat.exe()' which is illegal python in build/pip_data_bin_init.py.in.
• [x] Add the Windows version of kernel_link_options in build/Utils.cmake.
• [x] Add links of pthreadpool and cpuinfo for custom_ops_aot_lib in extension/llm/custom_ops/CMakeLists.txt.
• [x] Define ssize_t for Windows in runtime/core/portable_type/tensor_impl.h and runtime/core/exec_aten/util/tensor_util.h.
• [x] #4659

[iseeyuan]

Hi @python3kgae , thanks for submitting this issue and proposing the list of items. @daveboat and @malfet , does the list make sense to you?

We are limited in resource on Windows. @python3kgae , are you willing to directly contribute to the ET windows build? We are happy to support/unblock you.

[python3kgae]

PR created. Need some help with #4659. Not sure where choose_qparams_tensor_out is defined for WSL.