[xnnpack] New script to build llama_runner for Android

This will eventually be merged with the build_llama_android.sh.

Since Kleidi is not enabled by default, we want two paths for testing.

Author: digantdesai

.ci/scripts/build_llama_android_xnnpack.sh

@@ -0,0 +1,62 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+set -exu
+
+BUILD_TYPE=${1:-Release}
+ENABLE_KLEIDI=${2:-ON}
+NDK=${ANDROID_NDK:-/opt/ndk}
+BUILD_DIR=cmake-android-out-${BUILD_TYPE}-Kleidi-${ENABLE_KLEIDI}
+
+install_executorch_and_backend_lib() {
+  echo "Installing executorch and xnnpack backend"
+  ANDROID_ABI=arm64-v8a
+  cmake \
+    -DCMAKE_TOOLCHAIN_FILE="${NDK}/build/cmake/android.toolchain.cmake" \
+    -DEXECUTORCH_ENABLE_LOGGING=ON \
+    -DANDROID_ABI="${ANDROID_ABI}" \
+    -DANDROID_PLATFORM=android-23 \
+    -DCMAKE_INSTALL_PREFIX=${BUILD_DIR} \
+    -DCMAKE_BUILD_TYPE=${BUILD_TYPE} \
+    -DEXECUTORCH_BUILD_EXTENSION_DATA_LOADER=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
+    -DEXECUTORCH_BUILD_EXTENSION_TENSOR=ON \
+    -DEXECUTORCH_BUILD_XNNPACK=ON \
+    -DEXECUTORCH_XNNPACK_ENABLE_KLEIDI=${ENABLE_KLEIDI} \
+    -DEXECUTORCH_XNNPACK_WORKSPACE_SHARING=ON \
+    -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
+    -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
+    -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+    -DXNNPACK_ENABLE_ARM_BF16=OFF \
+    -B${BUILD_DIR} .
+
+  cmake --build ${BUILD_DIR} -j4 --target install --config ${BUILD_TYPE}
+}
+
+build_llama_runner() {
+    echo "Building llama runner for Android..."
+    ANDROID_ABI=arm64-v8a
+cmake \
+    -DCMAKE_TOOLCHAIN_FILE="$NDK"/build/cmake/android.toolchain.cmake  \
+    -DEXECUTORCH_ENABLE_LOGGING=ON \
+    -DANDROID_ABI="${ANDROID_ABI}" \
+    -DANDROID_PLATFORM=android-23 \
+    -DCMAKE_INSTALL_PREFIX=${BUILD_DIR} \
+    -DCMAKE_BUILD_TYPE=${BUILD_TYPE} -DPYTHON_EXECUTABLE=python \
+    -DEXECUTORCH_BUILD_XNNPACK=ON \
+    -DEXECUTORCH_XNNPACK_ENABLE_KLEIDI=${ENABLE_KLEIDI} \
+    -DEXECUTORCH_XNNPACK_WORKSPACE_SHARING=ON \
+    -DEXECUTORCH_BUILD_KERNELS_OPTIMIZED=ON \
+    -DEXECUTORCH_BUILD_KERNELS_QUANTIZED=ON \
+    -DEXECUTORCH_BUILD_KERNELS_CUSTOM=ON \
+    -B${BUILD_DIR}/examples/models/llama2 examples/models/llama2
+
+    cmake --build ${BUILD_DIR}/examples/models/llama2 -j4 --config ${BUILD_TYPE}
+}
+
+install_executorch_and_backend_lib
+build_llama_runner