Fix remaining build issues on Windows, re-enable Windows pybind preset build CI job (pytorch#13702)

**Note: This is a cherry-pick of
pytorch#13564 and
pytorch#13259. I inadvertently broke
the ghstack when attempting to re-order the stack, leading to the old PR
being merged by ghstack into the ghstack-created branch (not main).**

Add mman_windows.cpp to the CMake sources. Fix a few build-blocking
issues with XNNPACK on Windows. Enable XNNPACK on the Windows preset.
This also gives CI coverage.

I've disabled 4 microkernel families that don't build currently. These
likely need to be gated or fixed upstream, but I'm disabling them
locally for now to unblock usage. They are relatively niche and should
have little to no perf impact on the average user.

With this change, the pybind preset works and we can re-enable it in CI.

Author: GregoryComer

.github/workflows/build-presets.yml

@@ -109,7 +109,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        preset: [windows]
+        preset: [pybind, windows]
     with:
       job-name: build
       ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
@@ -122,14 +122,17 @@ jobs:
           Set-PSDebug -Trace 1
           \$ErrorActionPreference = 'Stop'
           \$PSNativeCommandUseErrorActionPreference = \$true
+
           conda create --yes --quiet -n et python=3.12
           conda activate et
           python install_requirements.py
+
           cmake --preset ${{ matrix.preset }} -T ClangCL
           if (\$LASTEXITCODE -ne 0) {
             Write-Host "CMake configuration was unsuccessful. Exit code: \$LASTEXITCODE."
             exit \$LASTEXITCODE
           }
+
           \$numCores = [System.Environment]::GetEnvironmentVariable('NUMBER_OF_PROCESSORS') - 1
           cmake --build cmake-out -j \$numCores
           if (\$LASTEXITCODE -ne 0) {

backends/xnnpack/cmake/Dependencies.cmake

@@ -55,6 +55,14 @@ else()
   )
 endif()
 
+if(WIN32)
+  # These XNNPACK options don't currently build on Windows.
+  set_overridable_option(XNNPACK_ENABLE_AVX256SKX OFF)
+  set_overridable_option(XNNPACK_ENABLE_AVX256VNNI OFF)
+  set_overridable_option(XNNPACK_ENABLE_AVX256VNNIGFNI OFF)
+  set_overridable_option(XNNPACK_ENABLE_AVX512BF16 OFF)
+endif()
+
 set(XNNPACK_BUILD_ALL_MICROKERNELS
     OFF
     CACHE BOOL ""

extension/data_loader/CMakeLists.txt

@@ -24,6 +24,11 @@ if(NOT ET_HAVE_SYS_MMAN_H AND NOT WIN32)
        "extension/data_loader/mmap_data_loader.cpp"
   )
 endif()
+if(WIN32)
+  list(APPEND _extension_data_loader__srcs
+       "extension/data_loader/mman_windows.cpp"
+  )
+endif()
 list(TRANSFORM _extension_data_loader__srcs PREPEND "${EXECUTORCH_ROOT}/")
 add_library(extension_data_loader ${_extension_data_loader__srcs})
 target_link_libraries(extension_data_loader executorch_core)

tools/cmake/preset/pybind.cmake

@@ -21,12 +21,13 @@ set_overridable_option(EXECUTORCH_BUILD_EXTENSION_FLAT_TENSOR ON)
 set_overridable_option(EXECUTORCH_BUILD_EXTENSION_DATA_LOADER ON)
 set_overridable_option(EXECUTORCH_BUILD_KERNELS_OPTIMIZED ON)
 set_overridable_option(EXECUTORCH_BUILD_EXTENSION_MODULE ON)
-set_overridable_option(EXECUTORCH_BUILD_EXTENSION_TRAINING ON)
 
 if(CMAKE_SYSTEM_NAME STREQUAL "Darwin")
   set_overridable_option(EXECUTORCH_BUILD_COREML ON)
+  set_overridable_option(EXECUTORCH_BUILD_EXTENSION_TRAINING ON)
 elseif(CMAKE_SYSTEM_NAME STREQUAL "Linux")
   set_overridable_option(EXECUTORCH_BUILD_COREML ON)
+  set_overridable_option(EXECUTORCH_BUILD_EXTENSION_TRAINING ON)
 elseif(CMAKE_SYSTEM_NAME STREQUAL "Windows" OR CMAKE_SYSTEM_NAME STREQUAL
                                                "WIN32"
 )

tools/cmake/preset/windows.cmake

@@ -14,6 +14,4 @@ set_overridable_option(EXECUTORCH_BUILD_EXTENSION_RUNNER_UTIL ON)
 set_overridable_option(EXECUTORCH_BUILD_EXTENSION_TENSOR ON)
 set_overridable_option(EXECUTORCH_BUILD_KERNELS_OPTIMIZED ON)
 set_overridable_option(EXECUTORCH_BUILD_KERNELS_QUANTIZED ON)
-
-# Below options are not yet buildable on Windows, but should be.
-# set_overridable_option(EXECUTORCH_BUILD_XNNPACK ON)
+set_overridable_option(EXECUTORCH_BUILD_XNNPACK ON)