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: Github Executorch

.ci/scripts/build_llama_android.sh

@@ -38,6 +38,8 @@ install_executorch_and_backend_lib() {
 build_llama_runner() {
     echo "Building llama runner for Android..."
     ANDROID_ABI=arm64-v8a
+    SITE_PACKAGES="$(${PYTHON_EXECUTABLE} -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
+    CMAKE_PREFIX_PATH="${SITE_PACKAGES}/torch"
     cmake -DBUCK2="${BUCK2}" \
     -DCMAKE_TOOLCHAIN_FILE="$ANDROID_NDK"/build/cmake/android.toolchain.cmake  \
     -DANDROID_ABI="${ANDROID_ABI}" \
@@ -47,6 +49,7 @@ build_llama_runner() {
     -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
     -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+    -DCMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH}" \
     -Bcmake-android-out/examples/models/llama examples/models/llama
 
     cmake --build cmake-android-out/examples/models/llama -j4 --config Release

.ci/scripts/test_model.sh

@@ -50,10 +50,13 @@ prepare_artifacts_upload() {
 
 build_cmake_executor_runner() {
   echo "Building executor_runner"
+  SITE_PACKAGES="$(${PYTHON_EXECUTABLE} -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
+  CMAKE_PREFIX_PATH="${SITE_PACKAGES}/torch"
   rm -rf ${CMAKE_OUTPUT_DIR}
   cmake -DCMAKE_BUILD_TYPE=Debug \
       -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
       -DPYTHON_EXECUTABLE="$PYTHON_EXECUTABLE" \
+      -DCMAKE_PREFIX_PATH="$CMAKE_PREFIX_PATH" \
       -B${CMAKE_OUTPUT_DIR} .
 
   cmake --build ${CMAKE_OUTPUT_DIR} -j4 --config Debug

.ci/scripts/test_phi_3_mini.sh

@@ -22,8 +22,11 @@ NPROC=8
 if hash nproc &> /dev/null; then NPROC=$(nproc); fi
 
 cmake_install_executorch_libraries() {
+  SITE_PACKAGES="$(${PYTHON_EXECUTABLE} -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
+  CMAKE_PREFIX_PATH="${SITE_PACKAGES}/torch"
   cmake -DPYTHON_EXECUTABLE=python \
       -DCMAKE_INSTALL_PREFIX=${BUILD_DIR} \
+      -DCMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH}" \
       -DEXECUTORCH_ENABLE_LOGGING=1 \
       -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
       -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
@@ -39,8 +42,11 @@ cmake_install_executorch_libraries() {
 }
 
 cmake_build_phi_3_mini() {
+  SITE_PACKAGES="$(${PYTHON_EXECUTABLE} -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
+  CMAKE_PREFIX_PATH="${SITE_PACKAGES}/torch"
   cmake -DPYTHON_EXECUTABLE=$PYTHON_EXECUTABLE \
       -DCMAKE_INSTALL_PREFIX=${BUILD_DIR} \
+      -DCMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH}" \
       -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
       -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
       -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \

test/run_oss_cpp_tests.sh

@@ -27,8 +27,11 @@ build_executorch() {
   if [ -x "$(command -v glslc)" ]; then
     BUILD_VULKAN="ON"
   fi
+  SITE_PACKAGES="$(${PYTHON_EXECUTABLE} -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())')"
+  CMAKE_PREFIX_PATH="${SITE_PACKAGES}/torch"
   cmake . \
     -DCMAKE_INSTALL_PREFIX=cmake-out \
+    -DCMAKE_PREFIX_PATH="${CMAKE_PREFIX_PATH}" \
     -DEXECUTORCH_USE_CPP_CODE_COVERAGE=ON \
     -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
     -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \