Merge a7e97ff798 into f9720f994e

a propose to fix address offset overflow caused by using int

apps/Android/MnnLlmChat/app/src/main/java/com/alibaba/mnnllm/android/chat/input/ChatInputComponent.kt

@@ -174,7 +174,7 @@ class ChatInputComponent(
         } else if (currentUserMessage != null) {
             visible = false
         } else if (!TextUtils.isEmpty(editUserMessage.text.toString())) {
-            visible = false
+            //visible = false
         }
         buttonSwitchVoice!!.visibility =
             if (visible) View.VISIBLE else View.GONE
@@ -227,13 +227,13 @@ class ChatInputComponent(
         voiceRecordingModule.setOnVoiceRecordingListener(object : VoiceRecordingListener {
             override fun onEnterRecordingMode() {
                 updateAudioOutput()
-                binding.btnToggleThinking.visibility = View.GONE
-                editUserMessage.visibility = View.GONE
+//                binding.btnToggleThinking.visibility = View.GONE
+//                editUserMessage.visibility = View.GONE
                 KeyboardUtils.hideKeyboard(editUserMessage)
                 if (attachmentPickerModule != null) {
                     attachmentPickerModule!!.hideAttachmentLayout()
                 }
-                editUserMessage.visibility = View.GONE
+//                editUserMessage.visibility = View.GONE
             }
 
             override fun onLeaveRecordingMode() {
@@ -250,10 +250,11 @@ class ChatInputComponent(
             override fun onRecordSuccess(duration: Float, recordingFilePath: String?) {
                 val chatDataItem = ChatDataItem.createAudioInputData(
                     chatActivity.dateFormat!!.format(Date()),
-                    "",
+                    editUserMessage.text.toString().trim { it <= ' ' },
                     recordingFilePath!!,
                     duration
                 )
+                editUserMessage.setText("")
                 this@ChatInputComponent.onSendMessage?.let { it(chatDataItem) }
             }
 

apps/Android/MnnLlmChat/app/src/main/res/layout/activity_chat.xml

@@ -118,14 +118,16 @@
                     android:padding="12dp"
                     android:textColor="?attr/colorOnSurface"
                     android:textSize="16sp"/>
-                <RelativeLayout
-                    android:layout_width="wrap_content"
-                    android:layout_height="50dp">
+                <LinearLayout
+                    android:layout_width="match_parent"
+                    android:orientation="vertical"
+                    android:layout_height="100dp">
                     <RelativeLayout
                         android:id="@+id/btn_voice_recording"
                         android:layout_width="match_parent"
-                        android:layout_height="match_parent"
-                        android:visibility="gone"
+                        android:layout_height="50dp"
+                        android:visibility="visible"
+                        tools:visibility="visible"
                         >
                         <TextView
                             android:layout_width="wrap_content"
@@ -208,7 +210,7 @@
                             android:visibility="gone" />
                     </FrameLayout>
 
-                </RelativeLayout>
+                </LinearLayout>
                 <LinearLayout
                     android:id="@+id/layout_more_menu"
                     android:layout_width="match_parent"
@@ -255,6 +257,7 @@
         android:visibility="gone"
         tools:visibility="visible"
         android:textColor="?colorOnSurface"
+        android:background="@color/semi_transparent_gray"
         tools:text="@string/release_to_send" />
 
     <RelativeLayout

apps/Android/MnnLlmChat/app/src/main/res/values/colors.xml

@@ -28,7 +28,7 @@
     <color name="color_on_surface_variant">#666666</color>
 
     <color name="color_disabled">#888888</color>
-
+    <color name="semi_transparent_gray">#80888888</color>
 
 
 </resources>

source/backend/coreml/backend/CoreMLBackend.cpp

@@ -133,11 +133,11 @@ namespace MNN {
     void CoreMLBackend::onResizeBegin() {
         mCoreMLLayerPtrs.clear();
     }
-    int CoreMLBackend::getBytes(const halide_type_t& type) {
+    size_t CoreMLBackend::getBytes(const halide_type_t& type) {
         if (type.code == halide_type_float && mPrecision == BackendConfig::Precision_Low) {
             return 1;
         }
-        return type.bytes();
+        return static_cast<size_t>(type.bytes());
     }
 
     ErrorCode CoreMLBackend::onResizeEnd() {

source/backend/coreml/backend/CoreMLBackend.hpp

@@ -107,7 +107,7 @@ namespace MNN {
         void setLayerOutputs(CoreML__Specification__NeuralNetworkLayer* layer, std::vector<std::string>&& outputs);
         void copyName(char** ptr, std::string&& name);
         int getInOutTensorInfo(std::string modelName);
-        int getBytes(const halide_type_t& type);
+        size_t getBytes(const halide_type_t& type);
 
         class Creator {
         public:

source/backend/cpu/CPUBackend.cpp

@@ -606,7 +606,7 @@ static OpType _getRealOpType(OpType opType) {
     }
 }
 void* CPUBackend::onMapTensor(Tensor::MapType mtype, Tensor::DimensionType dtype, const Tensor* srcTensor) {
-    if (getBytes(this, srcTensor) != srcTensor->getType().bytes()) {
+    if (static_cast<int>(getBytes(this, srcTensor)) != srcTensor->getType().bytes()) {
         return nullptr;
     }
     if (OpCommonUtils:: convertDimType(TensorUtils::getDescribe(srcTensor)->dimensionFormat) != dtype) {
@@ -617,7 +617,7 @@ void* CPUBackend::onMapTensor(Tensor::MapType mtype, Tensor::DimensionType dtype
 }
 
 bool CPUBackend::onUnmapTensor(Tensor::MapType mtype, Tensor::DimensionType dtype, const Tensor* dstTensor, void* mapPtr) {
-    if (getBytes(this, dstTensor) != dstTensor->getType().bytes()) {
+    if (static_cast<int>(getBytes(this, dstTensor)) != dstTensor->getType().bytes()) {
         return false;
     }
     if (OpCommonUtils:: convertDimType(TensorUtils::getDescribe(dstTensor)->dimensionFormat) != dtype) {
@@ -651,8 +651,8 @@ size_t CPUBackend::getTensorSize(const Tensor* tensor, bool multiBytes) const {
     return dataSize;
 }
 
-int CPUBackend::getBytes(const Backend* backend, const Tensor* output) {
-    auto bytes = output->getType().bytes();
+size_t CPUBackend::getBytes(const Backend* backend, const Tensor* output) {
+    size_t bytes = output->getType().bytes();
     auto core = static_cast<const CPUBackend*>(backend)->functions();
     auto quant = TensorUtils::getDescribe(output)->quantAttr.get();
     if (output->getType().code == halide_type_float) {

source/backend/cpu/CPUBackend.hpp

@@ -175,7 +175,7 @@ public:
     inline int taskIndex() const {return mRuntime->mTaskIndex;}
 #endif
     static void initCreatorMap();
-    static int getBytes(const Backend* backend, const Tensor* output);
+    static size_t getBytes(const Backend* backend, const Tensor* output);
     static DataType getDataType(const Tensor* tensor);
     friend class CPURuntime;
     void enqueueTask(std::function<int()>&& task);

source/backend/cpu/compute/ConvInt8TiledExecutor.cpp

@@ -1048,7 +1048,7 @@ ErrorCode DenseConvInt8TiledExecutor::onExecute(const std::vector<Tensor*>& inpu
     const auto kernelCountUnit       = mIm2ColParamter.kernelCountUnit;
     const auto unitColBufferSize  = kernelCountUnit * DST_XUNIT * SRC_UNIT * sizeof(int8_t);
     const auto colBufferSize       = unitColBufferSize * mIm2ColCount;
-    const int dstBytes               = static_cast<CPUBackend*>(backend())->getBytes(backend(), output);
+    const auto dstBytes               = static_cast<CPUBackend*>(backend())->getBytes(backend(), output);
     const int blockL                 = kernelCountUnit / mBlockNum; // source depthQuad for each block.
     const int kxky                   = mIm2ColParamter.kernelX * mIm2ColParamter.kernelY;
     const int blocklu                = blockL / kxky;                     // UP_DIV(ic,src_unit) per block

source/backend/cuda/core/CUDABackend.cpp

@@ -169,8 +169,8 @@ private:
     BufferAllocator* mAllocator;
     MemChunk mPoint;
 };
-int CUDABackend::getBytes(const Tensor* tensor) const {
-    auto bytes = tensor->getType().bytes();
+size_t CUDABackend::getBytes(const Tensor* tensor) const {
+    size_t bytes = tensor->getType().bytes();
     if (mPrecision == 2 || mPrecision == 3) {// Fp16 or Bf16
         if (halide_type_float == tensor->getType().code) {
             bytes = 2;

source/backend/cuda/core/CUDABackend.hpp

@@ -89,7 +89,7 @@ public:
         return mStaticBufferPool.get();
     }
     static size_t realSize(const Tensor *tensor);
-    int getBytes(const Tensor* tensor) const;
+    size_t getBytes(const Tensor* tensor) const;
     CPUResizeCache* getCache();
     bool useFp16() const;
     int getPrecision() const;

source/backend/cuda/execution/ConvWinogradExecution.cu

@@ -427,7 +427,6 @@ ErrorCode ConvWinogradExecution::onExecute(const std::vector<Tensor*> &inputs, c
     int co_pack = UP_DIV(mResource->mKernelInfo.kernelN, PACK_NUMBER) * PACK_NUMBER;
     int ci_pack = UP_DIV(mResource->mKernelInfo.kernelC, PACK_NUMBER) * PACK_NUMBER;
 
-    auto bytes = static_cast<CUDABackend*>(backend())->getBytes(input);
     const void *input_addr = (const void*)input->deviceId();
     const void *mGgGt_Buffer = mResource->mFilter;
     const void *bias_addr = mResource->mBias;

source/backend/cuda/execution/MatMulExecution.cu

@@ -224,7 +224,6 @@ MatMulExecution::~ MatMulExecution() {
 
 void MatMulExecution::setArguments(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
     auto runtime = static_cast<CUDABackend*>(backend())->getCUDARuntime();
-    auto bytes = static_cast<CUDABackend*>(backend())->getBytes(inputs[0]);
     auto pool = static_cast<CUDABackend*>(backend())->getBufferPool();
 
     const Tensor* A = inputs[0];
@@ -971,7 +970,6 @@ void MatMulExecution::setArguments(const std::vector<Tensor *> &inputs, const st
 
 ErrorCode MatMulExecution::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
     auto runtime = static_cast<CUDABackend*>(backend())->getCUDARuntime();
-    auto bytes = static_cast<CUDABackend*>(backend())->getBytes(inputs[0]);
 
     const Tensor* A = inputs[0];
     const Tensor* B = inputs[1];
@@ -1060,7 +1058,6 @@ ErrorCode MatMulExecution::onResize(const std::vector<Tensor *> &inputs, const s
 }
 
 ErrorCode MatMulExecution::onExecute(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
-    auto bytes = static_cast<CUDABackend*>(backend())->getBytes(inputs[0]);
     auto runtime = static_cast<CUDABackend*>(backend())->getCUDARuntime();
     bool hAlignment = (mGemmInfo.elhPad[2] == mGemmInfo.elh[2]);