[Android] Support llama3

Author: kirklandsign

examples/demo-apps/android/LlamaDemo/setup.sh

@@ -34,6 +34,9 @@ cmake examples/models/llama2 \
          -DANDROID_ABI="$ANDROID_ABI" \
          -DCMAKE_INSTALL_PREFIX="${CMAKE_OUT}" \
          -DEXECUTORCH_USE_TIKTOKEN="${EXECUTORCH_USE_TIKTOKEN}" \
+         -DEXECUTORCH_BUILD_CUSTOM=ON \
+         -DEXECUTORCH_BUILD_OPTIMIZED=ON \
+         -DEXECUTORCH_BUILD_XNNPACK=ON \
          -DCMAKE_BUILD_TYPE=Release \
          -B"${CMAKE_OUT}"/examples/models/llama2
 

extension/android/CMakeLists.txt

@@ -68,6 +68,8 @@ if(EXECUTORCH_BUILD_LLAMA_JNI)
   set_property(TARGET custom_ops PROPERTY IMPORTED_LOCATION ${CUSTOM_OPS_PATH})
   target_link_options_shared_lib(custom_ops)
 
+  target_link_options_shared_lib(quantized_ops_lib)
+
   if(TARGET pthreadpool)
     set(LLAMA_JNI_SRCS jni/jni_layer_llama.cpp ../../backends/xnnpack/threadpool/cpuinfo_utils.cpp)
   else()
@@ -83,7 +85,7 @@ if(EXECUTORCH_BUILD_LLAMA_JNI)
   endif()
   target_include_directories(executorch_llama_jni PRIVATE ${_common_include_directories})
   target_link_libraries(executorch_llama_jni ${link_libraries} llama_runner
-                        custom_ops cpublas eigen_blas)
+                        custom_ops cpublas eigen_blas quantized_kernels quantized_ops_lib)
   target_compile_options(executorch_llama_jni PUBLIC ${_common_compile_options})
   if(EXECUTORCH_USE_TIKTOKEN)
     set(ABSL_ENABLE_INSTALL ON)