From 4b3f6ef8d7f1d0d94d18c363dd622eb41dced352 Mon Sep 17 00:00:00 2001
From: Han Yin <han.yin@arm.com>
Date: Thu, 26 Jun 2025 13:21:17 -0700
Subject: [PATCH] misc: rename LlamaAndroid related class to InferenceEngine
 prefixes

---
 .../app/src/main/AndroidManifest.xml          |  4 ++--
 .../java/com/example/llama/di/AppModule.kt    | 11 +++++-----
 .../app/src/main/res/values/strings.xml       |  2 +-
 .../app/src/main/res/values/themes.xml        |  2 +-
 .../llama/src/main/cpp/cpu_detector.cpp       |  4 ++--
 .../llama/src/main/cpp/llama-android.cpp      | 22 +++++++++----------
 ...LLamaAndroid.kt => InferenceEngineImpl.kt} | 14 +++++++-----
 ...raryLoader.kt => InferenceEngineLoader.kt} | 20 ++++++++---------
 8 files changed, 41 insertions(+), 38 deletions(-)
 rename examples/llama.android/llama/src/main/java/android/llama/cpp/{LLamaAndroid.kt => InferenceEngineImpl.kt} (95%)
 rename examples/llama.android/llama/src/main/java/android/llama/cpp/{LLamaLibraryLoader.kt => InferenceEngineLoader.kt} (88%)

diff --git a/examples/llama.android/app/src/main/AndroidManifest.xml b/examples/llama.android/app/src/main/AndroidManifest.xml
index e85d4899b2..616cf384e5 100644
--- a/examples/llama.android/app/src/main/AndroidManifest.xml
+++ b/examples/llama.android/app/src/main/AndroidManifest.xml
@@ -13,14 +13,14 @@
         android:label="@string/app_name"
         android:roundIcon="@mipmap/ic_launcher_round"
         android:supportsRtl="true"
-        android:theme="@style/Theme.LlamaAndroid"
+        android:theme="@style/Theme.KleidiLlama"
         >
 
         <activity
             android:name=".MainActivity"
             android:exported="true"
             android:screenOrientation="portrait"
-            android:theme="@style/Theme.LlamaAndroid">
+            android:theme="@style/Theme.KleidiLlama">
             <intent-filter>
                 <action android:name="android.intent.action.MAIN" />
 
diff --git a/examples/llama.android/app/src/main/java/com/example/llama/di/AppModule.kt b/examples/llama.android/app/src/main/java/com/example/llama/di/AppModule.kt
index 5914ec7a00..24f6549afb 100644
--- a/examples/llama.android/app/src/main/java/com/example/llama/di/AppModule.kt
+++ b/examples/llama.android/app/src/main/java/com/example/llama/di/AppModule.kt
@@ -2,7 +2,7 @@ package com.example.llama.di
 
 import android.content.Context
 import android.llama.cpp.InferenceEngine
-import android.llama.cpp.LLamaLibraryLoader
+import android.llama.cpp.InferenceEngineLoader
 import com.example.llama.data.local.AppDatabase
 import com.example.llama.data.remote.HuggingFaceApiService
 import com.example.llama.data.remote.HuggingFaceRemoteDataSource
@@ -58,14 +58,15 @@ internal abstract class AppModule {
     ): HuggingFaceRemoteDataSource
 
     companion object {
-        private const val USE_REAL_ENGINE = true
+        private const val USE_STUB_ENGINE = false
 
         @Provides
         fun provideInferenceEngine(@ApplicationContext context: Context): InferenceEngine {
-            return if (USE_REAL_ENGINE) {
-                LLamaLibraryLoader.createInstance(context) ?: throw InstantiationException("Cannot instantiate LlamaAndroid!")
-            } else {
+            return if (USE_STUB_ENGINE) {
                 StubInferenceEngine()
+            } else {
+                InferenceEngineLoader.createInstance(context)
+                    ?: throw InstantiationException("Cannot instantiate InferenceEngine!")
             }
         }
 
diff --git a/examples/llama.android/app/src/main/res/values/strings.xml b/examples/llama.android/app/src/main/res/values/strings.xml
index 7a9d314e29..25c3cc42d2 100644
--- a/examples/llama.android/app/src/main/res/values/strings.xml
+++ b/examples/llama.android/app/src/main/res/values/strings.xml
@@ -1,3 +1,3 @@
 <resources>
-    <string name="app_name">LlamaAndroid</string>
+    <string name="app_name">Kleidi Llama</string>
 </resources>
diff --git a/examples/llama.android/app/src/main/res/values/themes.xml b/examples/llama.android/app/src/main/res/values/themes.xml
index 8a24fda566..031444ffd7 100644
--- a/examples/llama.android/app/src/main/res/values/themes.xml
+++ b/examples/llama.android/app/src/main/res/values/themes.xml
@@ -1,5 +1,5 @@
 <?xml version="1.0" encoding="utf-8"?>
 <resources>
 
-    <style name="Theme.LlamaAndroid" parent="android:Theme.Material.Light.NoActionBar" />
+    <style name="Theme.KleidiLlama" parent="android:Theme.Material.Light.NoActionBar" />
 </resources>
diff --git a/examples/llama.android/llama/src/main/cpp/cpu_detector.cpp b/examples/llama.android/llama/src/main/cpp/cpu_detector.cpp
index d8be2a4f72..2913036fa7 100644
--- a/examples/llama.android/llama/src/main/cpp/cpu_detector.cpp
+++ b/examples/llama.android/llama/src/main/cpp/cpu_detector.cpp
@@ -12,7 +12,7 @@ static const Aarch64Info info = GetAarch64Info();
 static const Aarch64Features features = info.features;
 
 extern "C" JNIEXPORT jint JNICALL
-Java_android_llama_cpp_LLamaLibraryLoader_getOptimalTier(
+Java_android_llama_cpp_InferenceEngineLoader_getOptimalTier(
         JNIEnv* env,
         jclass clazz) {
     int tier = 0;  // Default to T0 (baseline)
@@ -46,7 +46,7 @@ Java_android_llama_cpp_LLamaLibraryLoader_getOptimalTier(
 
 // Optional: Keep a feature string function for debugging
 extern "C" JNIEXPORT jstring JNICALL
-Java_android_llama_cpp_LLamaLibraryLoader_getCpuFeaturesString(
+Java_android_llama_cpp_InferenceEngineLoader_getCpuFeaturesString(
         JNIEnv* env,
         jclass clazz) {
     std::string text;
diff --git a/examples/llama.android/llama/src/main/cpp/llama-android.cpp b/examples/llama.android/llama/src/main/cpp/llama-android.cpp
index 3e797baa74..e06150572a 100644
--- a/examples/llama.android/llama/src/main/cpp/llama-android.cpp
+++ b/examples/llama.android/llama/src/main/cpp/llama-android.cpp
@@ -72,7 +72,7 @@ static void log_callback(ggml_log_level level, const char *fmt, void *data) {
 
 extern "C"
 JNIEXPORT void JNICALL
-Java_android_llama_cpp_LLamaAndroid_init(JNIEnv *env, jobject /*unused*/) {
+Java_android_llama_cpp_InferenceEngineImpl_init(JNIEnv *env, jobject /*unused*/) {
     // Set llama log handler to Android
     llama_log_set(log_callback, nullptr);
 
@@ -83,7 +83,7 @@ Java_android_llama_cpp_LLamaAndroid_init(JNIEnv *env, jobject /*unused*/) {
 
 extern "C"
 JNIEXPORT jint JNICALL
-Java_android_llama_cpp_LLamaAndroid_load(JNIEnv *env, jobject, jstring jmodel_path) {
+Java_android_llama_cpp_InferenceEngineImpl_load(JNIEnv *env, jobject, jstring jmodel_path) {
     llama_model_params model_params = llama_model_default_params();
 
     const auto *model_path = env->GetStringUTFChars(jmodel_path, 0);
@@ -137,7 +137,7 @@ static common_sampler *new_sampler(float temp) {
 
 extern "C"
 JNIEXPORT jint JNICALL
-Java_android_llama_cpp_LLamaAndroid_prepare(JNIEnv * /*env*/, jobject /*unused*/) {
+Java_android_llama_cpp_InferenceEngineImpl_prepare(JNIEnv * /*env*/, jobject /*unused*/) {
     auto *context = init_context(g_model);
     if (!context) { return 1; }
     g_context = context;
@@ -161,14 +161,14 @@ static std::string get_backend() {
 
 extern "C"
 JNIEXPORT jstring JNICALL
-Java_android_llama_cpp_LLamaAndroid_systemInfo(JNIEnv *env, jobject /*unused*/) {
+Java_android_llama_cpp_InferenceEngineImpl_systemInfo(JNIEnv *env, jobject /*unused*/) {
     return env->NewStringUTF(llama_print_system_info());
 }
 
 extern "C"
 JNIEXPORT jstring JNICALL
-Java_android_llama_cpp_LLamaAndroid_benchModel(JNIEnv *env, jobject /*unused*/, jint pp, jint tg,
-                                               jint pl, jint nr) {
+Java_android_llama_cpp_InferenceEngineImpl_benchModel(JNIEnv *env, jobject /*unused*/, jint pp, jint tg,
+                                                      jint pl, jint nr) {
     auto *context = init_context(g_model, pp);
     if (!context) {
         const auto err_msg = "Fail to init_context! Bench aborted.";
@@ -377,7 +377,7 @@ static int decode_tokens_in_batches(
 
 extern "C"
 JNIEXPORT jint JNICALL
-Java_android_llama_cpp_LLamaAndroid_processSystemPrompt(
+Java_android_llama_cpp_InferenceEngineImpl_processSystemPrompt(
         JNIEnv *env,
         jobject /*unused*/,
         jstring jsystem_prompt
@@ -426,7 +426,7 @@ Java_android_llama_cpp_LLamaAndroid_processSystemPrompt(
 
 extern "C"
 JNIEXPORT jint JNICALL
-Java_android_llama_cpp_LLamaAndroid_processUserPrompt(
+Java_android_llama_cpp_InferenceEngineImpl_processUserPrompt(
         JNIEnv *env,
         jobject /*unused*/,
         jstring juser_prompt,
@@ -510,7 +510,7 @@ static bool is_valid_utf8(const char *string) {
 
 extern "C"
 JNIEXPORT jstring JNICALL
-Java_android_llama_cpp_LLamaAndroid_generateNextToken(
+Java_android_llama_cpp_InferenceEngineImpl_generateNextToken(
         JNIEnv *env,
         jobject /*unused*/
 ) {
@@ -570,7 +570,7 @@ Java_android_llama_cpp_LLamaAndroid_generateNextToken(
 
 extern "C"
 JNIEXPORT void JNICALL
-Java_android_llama_cpp_LLamaAndroid_unload(JNIEnv * /*unused*/, jobject /*unused*/) {
+Java_android_llama_cpp_InferenceEngineImpl_unload(JNIEnv * /*unused*/, jobject /*unused*/) {
     // Reset long-term & short-term states
     reset_long_term_states();
     reset_short_term_states();
@@ -585,6 +585,6 @@ Java_android_llama_cpp_LLamaAndroid_unload(JNIEnv * /*unused*/, jobject /*unused
 
 extern "C"
 JNIEXPORT void JNICALL
-Java_android_llama_cpp_LLamaAndroid_shutdown(JNIEnv *env, jobject /*unused*/) {
+Java_android_llama_cpp_InferenceEngineImpl_shutdown(JNIEnv *env, jobject /*unused*/) {
     llama_backend_free();
 }
diff --git a/examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt b/examples/llama.android/llama/src/main/java/android/llama/cpp/InferenceEngineImpl.kt
similarity index 95%
rename from examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt
rename to examples/llama.android/llama/src/main/java/android/llama/cpp/InferenceEngineImpl.kt
index 06c30c987e..65954f9388 100644
--- a/examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaAndroid.kt
+++ b/examples/llama.android/llama/src/main/java/android/llama/cpp/InferenceEngineImpl.kt
@@ -36,25 +36,27 @@ import java.io.File
  *
  * @see llama-android.cpp for the native implementation details
  */
-class LLamaAndroid private constructor(private val tier: LLamaTier) : InferenceEngine {
+internal class InferenceEngineImpl private constructor(
+    private val tier: LLamaTier
+) : InferenceEngine {
 
     companion object {
-        private val TAG = LLamaAndroid::class.java.simpleName
+        private val TAG = InferenceEngineImpl::class.java.simpleName
 
         private var initialized = false
 
         /**
-         * Create LLamaAndroid instance with specific tier
+         * Create [InferenceEngineImpl] instance with specific tier
          */
-        internal fun createWithTier(tier: LLamaTier): LLamaAndroid? {
+        internal fun createWithTier(tier: LLamaTier): InferenceEngineImpl? {
             if (initialized) {
                 Log.w(TAG, "LLamaAndroid already initialized")
                 return null
             }
 
             try {
-                Log.i(TAG, "Instantiating LLamaAndroid w/ ${tier.libraryName}")
-                val instance = LLamaAndroid(tier)
+                Log.i(TAG, "Instantiating InferenceEngineImpl w/ ${tier.libraryName}")
+                val instance = InferenceEngineImpl(tier)
                 initialized = true
                 return instance
 
diff --git a/examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaLibraryLoader.kt b/examples/llama.android/llama/src/main/java/android/llama/cpp/InferenceEngineLoader.kt
similarity index 88%
rename from examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaLibraryLoader.kt
rename to examples/llama.android/llama/src/main/java/android/llama/cpp/InferenceEngineLoader.kt
index 23dbdbac30..854726c3f7 100644
--- a/examples/llama.android/llama/src/main/java/android/llama/cpp/LLamaLibraryLoader.kt
+++ b/examples/llama.android/llama/src/main/java/android/llama/cpp/InferenceEngineLoader.kt
@@ -21,10 +21,10 @@ enum class LLamaTier(val rawValue: Int, val libraryName: String, val description
     }
 }
 
-class LLamaLibraryLoader private constructor() {
+class InferenceEngineLoader private constructor() {
 
     companion object {
-        private val TAG = LLamaLibraryLoader::class.simpleName
+        private val TAG = InferenceEngineLoader::class.simpleName
 
         private const val DETECTION_VERSION = 1
         private const val PREFS_NAME = "llama_cpu_detection"
@@ -37,16 +37,16 @@ class LLamaLibraryLoader private constructor() {
         @JvmStatic
         private external fun getCpuFeaturesString(): String
 
-        private var _cachedInstance: LLamaAndroid? = null
+        private var _cachedInstance: InferenceEngineImpl? = null
         private var _detectedTier: LLamaTier? = null
         val detectedTier: LLamaTier? get() = _detectedTier
 
         /**
-         * Factory method to get a configured LLamaAndroid instance.
+         * Factory method to get a configured [InferenceEngineImpl] instance.
          * Handles tier detection, caching, and library loading automatically.
          */
         @Synchronized
-        fun createInstance(context: Context): LLamaAndroid? {
+        fun createInstance(context: Context): InferenceEngine? {
             // Return cached instance if available
             _cachedInstance?.let { return it }
 
@@ -59,18 +59,18 @@ class LLamaLibraryLoader private constructor() {
                 _detectedTier = tier
                 Log.i(TAG, "Using tier: ${tier.name} (${tier.description})")
 
-                // Create and cache LLamaAndroid instance
-                val instance = LLamaAndroid.createWithTier(tier) ?: run {
-                    Log.e(TAG, "Failed to instantiate LLamaAndroid")
+                // Create and cache the inference engine instance
+                val instance = InferenceEngineImpl.createWithTier(tier) ?: run {
+                    Log.e(TAG, "Failed to instantiate InferenceEngineImpl")
                     return null
                 }
                 _cachedInstance = instance
-                Log.i(TAG, "Successfully created LLamaAndroid instance with ${tier.name}")
+                Log.i(TAG, "Successfully created InferenceEngineImpl instance with ${tier.name}")
 
                 return instance
 
             } catch (e: Exception) {
-                Log.e(TAG, "Error creating LLamaAndroid instance", e)
+                Log.e(TAG, "Error creating InferenceEngineImpl instance", e)
                 return null
             }
         }