[Windows] misc fixes to enable Windows build #9198

SS-JIA · 2025-03-12T19:40:45Z

Summary:

Context

Implement some small miscellaneous fixes to enable installing ExecuTorch on Windows. Currently, the optimized kernels cannot be built successfully but the installation works in custom ops are not built.

Installation

cd executorch

# Need to disable building custom kernels; there are a bunch of build errors
$env:EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT=0
# For some reason need to explicitly enable tensor extension otherwise I see a linker error
$env:CMAKE_ARGS="-DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON"

python install_executorch.py

# Unset if you want
Remove-Item Env:\EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT
Remove-Item Env:\CMAKE_ARGS

Testing

I am able to follow the Getting Started tutorial successfully.

Outstanding Build Issues

Currently, with CMake I can successfully build with the following settings:

del -Recurse -Force cmake-out; `
  cmake . `
  -DCMAKE_INSTALL_PREFIX=cmake-out `
  -DPYTHON_EXECUTABLE=C:\\Users\\ssjia\\AppData\\Local\\miniconda3\\python.exe `
  -DCMAKE_PREFIX_PATH=C:\\Users\\ssjia\\AppData\\Local\\miniconda3\\Lib\\site-packages `
  -DCMAKE_BUILD_TYPE=Release `
  -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON `
  -DEXECUTORCH_BUILD_PYBIND=ON `
  -DEXECUTORCH_BUILD_XNNPACK=ON `
  -DEXECUTORCH_BUILD_KERNELS_QUANTIZED_AOT=ON `
  -DEXECUTORCH_BUILD_KERNELS_CUSTOM_AOT=ON `
  -DEXECUTORCH_BUILD_KERNELS_CUSTOM=OFF `
  -T ClangCL `
  -Bcmake-out; `
  cmake --build cmake-out -j64 --target install

If I switch EXECUTORCH_BUILD_KERNELS_CUSTOM to ON, then the build fails. The primary offenders appear to be the gelu. The implementation includes a header from ATen which is causing the majority of the errors.

pytorch-bot · 2025-03-12T19:40:49Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/9198

📄 Preview Python docs built from this PR

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

✅ You can merge normally! (1 Unrelated Failure)

As of commit 86e4837 with merge base 58d3582 ():

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

pull / android / run-emulator (gh) (trunk failure)
Process completed with exit code 9.

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

SS-JIA · 2025-03-12T19:44:43Z

extension/llm/custom_ops/CMakeLists.txt

@@ -49,7 +49,7 @@ if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm64|armv7)$")
  list(APPEND _custom_ops__srcs
       "extension/llm/custom_ops/spinquant/third-party/FFHT/fht_neon.c"
  )
-elseif(CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64")
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64)")


apparently AMD64 is synonymous with x86_64. On my Windows machine this value is AMD64.

SS-JIA · 2025-03-12T21:02:53Z

backends/xnnpack/CMakeLists.txt

 add_library(xnnpack_schema INTERFACE ${_xnnpack_schema__outputs})
 set_target_properties(xnnpack_schema PROPERTIES LINKER_LANGUAGE CXX)
 target_include_directories(
  xnnpack_schema INTERFACE ${_xnnpack_schema__include_dir}
                           ${EXECUTORCH_ROOT}/third-party/flatbuffers/include
 )

-set(xnnpack_third_party pthreadpool cpuinfo)
+set(xnnpack_third_party pthreadpool extension_threadpool cpuinfo)


required due to this recent refactor

without it, I see a linking error complaining that pthreadpool symbols cannot be found.

SS-JIA · 2025-03-12T21:03:15Z

backends/xnnpack/CMakeLists.txt


 include(cmake/Dependencies.cmake)

 list(TRANSFORM _xnnpack_backend__srcs PREPEND "${EXECUTORCH_ROOT}/")
 add_library(xnnpack_backend STATIC ${_xnnpack_backend__srcs})
 target_link_libraries(
-  xnnpack_backend PRIVATE ${xnnpack_third_party} executorch_core xnnpack_schema
+  xnnpack_backend PUBLIC ${xnnpack_third_party} executorch_core xnnpack_schema


changing to PUBLIC as suggested by @swolchok

SS-JIA · 2025-03-12T21:04:19Z

setup.py

-                relative to the cmake-out directory. May be an fnmatch-style
-                glob that matches exactly one file. If the path ends in `.so`,
-                this class will also look for similarly-named `.dylib` files.
+            src_dir: The directory of the file to install, relative to the cmake-out


these changes are lifed from #6979

tdakhran · 2025-03-13T06:12:35Z

backends/xnnpack/CMakeLists.txt

@@ -73,34 +73,42 @@ foreach(fbs_file ${_xnnpack_schema__srcs})
  )
 endforeach()

+if(WIN32)
+  set(MV_COMMAND powershell -Command "Move-Item -Path ${_xnnpack_flatbuffer__outputs} -Destination ${_xnnpack_schema__outputs}")


When cross-compiling for Windows, WIN32 is set in cmake, but there is no powershell

good point. I may address this in another PR though. There are some other places in our CMake files that don't account for cross compiling for windows. So currently I prefer to address it in a follow up.

do you mind adding this as a comment in this cmake? worthwhile to keep this info for the follow up next time.

yes, for sure.

Summary: ## Context This PR is mostly to enable ExecuTorch build for Windows: pytorch/executorch#9198 In ExecuTorch, the optimized GeLU kernel calls the ATen implementation. However, on Windows `math.h` needs to be included with `#define _USE_MATH_DEFINES` in order for math constants to be defined. Test Plan: Rely on CI to make sure existing tests do not break. Tested separately with ExecuTorch to make sure Windows build is successful.

SS-JIA · 2025-03-14T16:22:36Z

install_executorch.bat

-  )
-)
+rem Under windows, it's always python
+set PYTHON_EXECUTABLE=python


these changes are lifed from #6979

mcr229

lgtm stamping to unblock

Summary: ## Context Implement some small miscellaneous fixes to enable installing ExecuTorch on Windows. Currently, the optimized kernels cannot be built successfully but the installation works in custom ops are not built. ## Installation ```powershell cd executorch # Need to disable building custom kernels; there are a bunch of build errors $env:EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT=0 # For some reason need to explicitly enable tensor extension otherwise I see a linker error $env:CMAKE_ARGS="-DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON" python install_executorch.py # Unset if you want Remove-Item Env:\EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT Remove-Item Env:\CMAKE_ARGS ``` ## Testing I am able to follow the [Getting Started tutorial](https://pytorch.org/executorch/stable/getting-started-setup.html#run-your-program) successfully. ## Outstanding Build Issues Currently, with CMake I can successfully build with the following settings: ``` del -Recurse -Force cmake-out; ` cmake . ` -DCMAKE_INSTALL_PREFIX=cmake-out ` -DPYTHON_EXECUTABLE=C:\\Users\\ssjia\\AppData\\Local\\miniconda3\\python.exe ` -DCMAKE_PREFIX_PATH=C:\\Users\\ssjia\\AppData\\Local\\miniconda3\\Lib\\site-packages ` -DCMAKE_BUILD_TYPE=Release ` -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON ` -DEXECUTORCH_BUILD_PYBIND=ON ` -DEXECUTORCH_BUILD_XNNPACK=ON ` -DEXECUTORCH_BUILD_KERNELS_QUANTIZED_AOT=ON ` -DEXECUTORCH_BUILD_KERNELS_CUSTOM_AOT=ON ` -DEXECUTORCH_BUILD_KERNELS_CUSTOM=OFF ` -T ClangCL ` -Bcmake-out; ` cmake --build cmake-out -j64 --target install ``` If I switch `EXECUTORCH_BUILD_KERNELS_CUSTOM` to `ON`, then the build fails. The primary offenders appear to be the `gelu`. The implementation includes a header from ATen which is causing the majority of the errors.

SS-JIA · 2025-03-14T21:25:29Z

CMakeLists.txt

+  if(WIN32)
+    # flatbuffers does not use CMAKE_BUILD_TYPE. Internally, the build forces Release
+    # config, but from CMake's perspective the build type is always Debug.
+    set(FLATC_EXECUTABLE ${BINARY_DIR}/$<CONFIG>/flatc.exe)


@jathu this is the change that was needed

jathu

flatc changes lgtm

Summary: ## Context This PR is mostly to enable ExecuTorch build for Windows: pytorch/executorch#9198 In ExecuTorch, the optimized GeLU kernel calls the ATen implementation. However, on Windows `math.h` needs to be included with `#define _USE_MATH_DEFINES` in order for math constants to be defined. Test Plan: Rely on CI to make sure existing tests do not break. Tested separately with ExecuTorch to make sure Windows build is successful. Pull Request resolved: #149164 Approved by: https://github.com/swolchok

Summary: ## Context Implement some small miscellaneous fixes to enable installing ExecuTorch on Windows. Currently, the optimized kernels cannot be built successfully but the installation works in custom ops are not built. ## Installation ```powershell cd executorch # Need to disable building custom kernels; there are a bunch of build errors $env:EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT=0 # For some reason need to explicitly enable tensor extension otherwise I see a linker error $env:CMAKE_ARGS="-DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON" python install_executorch.py # Unset if you want Remove-Item Env:\EXECUTORCH_BUILD_KERNELS_CUSTOM_AOT Remove-Item Env:\CMAKE_ARGS ``` ## Testing I am able to follow the [Getting Started tutorial](https://pytorch.org/executorch/stable/getting-started-setup.html#run-your-program) successfully. ## Outstanding Build Issues Currently, with CMake I can successfully build with the following settings: ``` del -Recurse -Force cmake-out; ` cmake . ` -DCMAKE_INSTALL_PREFIX=cmake-out ` -DPYTHON_EXECUTABLE=C:\\Users\\ssjia\\AppData\\Local\\miniconda3\\python.exe ` -DCMAKE_PREFIX_PATH=C:\\Users\\ssjia\\AppData\\Local\\miniconda3\\Lib\\site-packages ` -DCMAKE_BUILD_TYPE=Release ` -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON ` -DEXECUTORCH_BUILD_PYBIND=ON ` -DEXECUTORCH_BUILD_XNNPACK=ON ` -DEXECUTORCH_BUILD_KERNELS_QUANTIZED_AOT=ON ` -DEXECUTORCH_BUILD_KERNELS_CUSTOM_AOT=ON ` -DEXECUTORCH_BUILD_KERNELS_CUSTOM=OFF ` -T ClangCL ` -Bcmake-out; ` cmake --build cmake-out -j64 --target install ``` If I switch `EXECUTORCH_BUILD_KERNELS_CUSTOM` to `ON`, then the build fails. The primary offenders appear to be the `gelu`. The implementation includes a header from ATen which is causing the majority of the errors.

SS-JIA requested review from jackzhxng, iseeyuan, larryliu0820, swolchok, digantdesai and mcr229 as code owners March 12, 2025 19:40

facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Mar 12, 2025

SS-JIA commented Mar 12, 2025

View reviewed changes

SS-JIA added the release notes: build Changes related to build, including dependency upgrades, build flags, optimizations, etc. label Mar 12, 2025

SS-JIA force-pushed the pr9198 branch from 07b9477 to f0f1878 Compare March 12, 2025 20:59

SS-JIA commented Mar 12, 2025

View reviewed changes

SS-JIA force-pushed the pr9198 branch 2 times, most recently from 566a7ed to b802f6e Compare March 12, 2025 22:55

tdakhran reviewed Mar 13, 2025

View reviewed changes

SS-JIA mentioned this pull request Mar 13, 2025

[ATen-CPU] Add math.h for Gelu pytorch/pytorch#149164

Closed

SS-JIA force-pushed the pr9198 branch from b802f6e to 224931d Compare March 14, 2025 16:20

SS-JIA commented Mar 14, 2025

View reviewed changes

mcr229 approved these changes Mar 14, 2025

View reviewed changes

SS-JIA force-pushed the pr9198 branch from 224931d to 86e4837 Compare March 14, 2025 21:24

SS-JIA commented Mar 14, 2025

View reviewed changes

jathu approved these changes Mar 14, 2025

View reviewed changes

SS-JIA merged commit 3137889 into main Mar 14, 2025
81 of 154 checks passed

SS-JIA deleted the pr9198 branch March 14, 2025 23:36

This was referenced Mar 17, 2025

[Build] Support windows for pybindings install in setup.py #5382

Closed

Support build on native Windows x64 #7217

Closed

Windows x64 Build with support for xnnpack and llama example #6979

Closed

cbilgin mentioned this pull request Mar 21, 2025

Openvino backend for Executorch to enable inference on Intel CPUs, GPUs, NPUs #8573

Merged

cavusmustafa mentioned this pull request Mar 25, 2025

Windows build failure: Failed to query buck for sources. #9616

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Windows] misc fixes to enable Windows build #9198

[Windows] misc fixes to enable Windows build #9198

SS-JIA commented Mar 12, 2025 •

edited

Loading

pytorch-bot bot commented Mar 12, 2025 •

edited

Loading

SS-JIA Mar 12, 2025

SS-JIA Mar 12, 2025

SS-JIA Mar 12, 2025

SS-JIA Mar 12, 2025

SS-JIA Mar 12, 2025

tdakhran Mar 13, 2025

SS-JIA Mar 13, 2025

mcr229 Mar 14, 2025

SS-JIA Mar 14, 2025

SS-JIA Mar 14, 2025

mcr229 left a comment

SS-JIA Mar 14, 2025

jathu left a comment

[Windows] misc fixes to enable Windows build #9198

[Windows] misc fixes to enable Windows build #9198

Conversation

SS-JIA commented Mar 12, 2025 • edited Loading

Context

Installation

Testing

Outstanding Build Issues

pytorch-bot bot commented Mar 12, 2025 • edited Loading

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/9198

✅ You can merge normally! (1 Unrelated Failure)

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

mcr229 left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jathu left a comment

Choose a reason for hiding this comment

SS-JIA commented Mar 12, 2025 •

edited

Loading

pytorch-bot bot commented Mar 12, 2025 •

edited

Loading