[Windows] [mmap_data_loader.cpp] mmap equivalent for Windows

[SamGondelman]

Summary

#5164 with the code suggestions applied.

Test plan

I am not familiar with how to test this area of the code. On non-Windows systems, it should be functionally the same.
#5164 suggests a command to run some relevant tests?

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/8916

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure

As of commit 4e6b973 with merge base 24671a9 ():

NEW FAILURE - The following job has failed:

• pull / test-llava-runner-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 24875dbbbd4d37b7f8c13a9084ba3222d6540f83f900bf4aff304dbe359a3800 /exec failed with exit code 139

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[SamGondelman]

I'm surprised this (and open, etc.) works, since it was the problem in #8913. I haven't built this file on Windows myself yet, but maybe a similar fix is necessary?

[SamGondelman]

hm:

/usr/bin/ld: /pytorch/executorch/build-x86/lib/libextension_module.so: undefined reference to `vtable for executorch::extension::MmapDataLoader'
/usr/bin/ld: /pytorch/executorch/build-x86/lib/libextension_module.so: undefined reference to `executorch::extension::MmapDataLoader::~MmapDataLoader()'
/usr/bin/ld: /pytorch/executorch/build-x86/lib/libextension_module.so: undefined reference to `executorch::extension::MmapDataLoader::from(char const*, executorch::extension::MmapDataLoader::MlockConfig)'

[SS-JIA]

@SamGondelman figured out what was causing the linker error you're seeing.

Btw, if you want to repro it yourself, you can do so on a linux machine (e.g. Devserver). You can follow https://pytorch.org/executorch/main/getting-started-setup.html on how to set up the repo. Then

First, build/install ExecuTorch libraries

(rm -rf cmake-out && \
  pp cmake . \
  -DCMAKE_INSTALL_PREFIX=cmake-out \
  -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
  -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
  -DEXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL=ON \
  -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
  -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
  -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
  -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
  -DEXECUTORCH_BUILD_VULKAN=ON \
  -Bcmake-out && \
  cmake --build cmake-out -j64 --target install)

The above command should work.

Next, try to build the Llama runner binary. This is where it will fail because the binary will try to link the data loader

(rm -rf cmake-out/examples/models/llama && \
  pp cmake examples/models/llama \
    -DCMAKE_INSTALL_PREFIX=cmake-out \
    -Bcmake-out/examples/models/llama && \
  cmake --build cmake-out/examples/models/llama -j16)

The pp in my commands are to set up fwdproxy.

[SS-JIA]

LGTM. The one test failure does not appear to be related...