Update on "Reuse GELU implementation from PyTorch core"

kernels/optimized doesn't need to support embedded systems, so it can just take a header-only dep on PyTorch.

Note that, because we will pick up Sleef internally and ignore it
externally thanks to ATen vec, this PR gets to enable optimized GELU in OSS.

Testing: CI to make sure this doesn't break mobile build modes; happy to take advice on anything not currently covered that might break.

Differential Revision: [D66335522](https://our.internmc.facebook.com/intern/diff/D66335522/)

[ghstack-poisoned]

Author: swolchok

.ci/scripts/build_llama_android.sh

@@ -15,6 +15,7 @@ if [[ -z "${PYTHON_EXECUTABLE:-}" ]]; then
 fi
 which "${PYTHON_EXECUTABLE}"
 CMAKE_PREFIX_PATH="$(python3 -c 'import torch as _; print(_.__path__[0])')"
+SITE_PACKAGES="$(${PYTHON_EXECUTABLE} -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
 
 install_executorch_and_backend_lib() {
   echo "Installing executorch and xnnpack backend"
@@ -28,6 +29,7 @@ install_executorch_and_backend_lib() {
     -DANDROID_ABI="${ANDROID_ABI}" \
     -DCMAKE_INSTALL_PREFIX=cmake-android-out \
     -DCMAKE_BUILD_TYPE=Release \
+    -DCMAKE_FIND_ROOT_PATH="${SITE_PACKAGES}" \
     -DCMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH}" \
     -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
     -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \

build/build_android_llm_demo.sh

@@ -40,12 +40,10 @@ build_android_native_library() {
   else
     EXECUTORCH_BUILD_NEURON=OFF
   fi
+  SITE_PACKAGES="$(${PYTHON_EXECUTABLE} -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
 
-  echo "BEGIN DEBUG COMMANDS"
-  ls "${CMAKE_PREFIX_PATH}"
-  find "${CMAKE_PREFIX_PATH}" -name "*.cmake" || true
-  echo "END DEBUG COMMANDS"
   cmake . -DCMAKE_INSTALL_PREFIX="${CMAKE_OUT}" \
+    -DCMAKE_FIND_ROOT_PATH="${SITE_PACKAGES}" \
     -DCMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH}" \
     -DCMAKE_TOOLCHAIN_FILE="${ANDROID_NDK}/build/cmake/android.toolchain.cmake" \
     -DANDROID_ABI="${ANDROID_ABI}" \
@@ -80,6 +78,7 @@ build_android_native_library() {
     -DANDROID_ABI="${ANDROID_ABI}" \
     -DANDROID_PLATFORM=android-26 \
     -DCMAKE_INSTALL_PREFIX="${CMAKE_OUT}" \
+    -DCMAKE_FIND_ROOT_PATH="${SITE_PACKAGES}" \
     -DCMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH}" \
     -DEXECUTORCH_ENABLE_LOGGING=ON \
     -DEXECUTORCH_LOG_LEVEL=Info \

kernels/optimized/CMakeLists.txt

@@ -61,7 +61,6 @@ generate_bindings_for_kernels(
 )
 message("Generated files ${gen_command_sources}")
 
-message("DEBUG: cmake_prefix_path is ${CMAKE_PREFIX_PATH}")
 list(TRANSFORM _optimized_kernels__srcs PREPEND "${EXECUTORCH_ROOT}/")
 add_library(optimized_kernels ${_optimized_kernels__srcs})
 find_package(Torch CONFIG REQUIRED)