!9159 [lite] fix converter bug

From: @xu_anyue
Reviewed-by: @hangangqiang,@HilbertDavid
Signed-off-by: @hangangqiang,@HilbertDavid

mindspore/lite/src/ops/reshape.cc

@@ -179,6 +179,13 @@ int Reshape::InferShape(std::vector<Tensor *> inputs_, std::vector<Tensor *> out
   std::vector<int> out_shape;
   if (inputs_.size() == kDoubleNum) {
     auto shape_tensor = inputs_.at(1);
+    if (input->ElementsNum() == 1) {
+      if (shape_tensor->shape().empty()) {
+        MS_LOG(DEBUG) << "reshape to a scalar.";
+        output->set_shape(out_shape);
+        return RET_OK;
+      }
+    }
     if (shape_tensor->data_c() == nullptr) {
       MS_LOG(INFO) << "Do infer shape in runtime.";
       return RET_INFER_INVALID;

mindspore/lite/src/ops/split.cc

@@ -52,8 +52,8 @@ int Split::UnPackAttr(const Primitive &prim, const std::vector<AnfNodePtr> &inpu
       MS_LOG(ERROR) << "new primitiveT value failed";
       return RET_ERROR;
     }
-    attr->splitDim = GetValue<int32_t>(prim.GetAttr("axis"));
+    attr->splitDim = CastToInt(prim.GetAttr("axis")).front();
-    attr->numberSplit = GetValue<int32_t>(prim.GetAttr("output_num"));
+    attr->numberSplit = CastToInt(prim.GetAttr("output_num")).front();
     this->primitive_->value.value = attr;
     if (this->primitive_->value.value == nullptr) {
       MS_LOG(ERROR) << "primitive value is nullptr";

mindspore/lite/src/ops/strided_slice.cc

@@ -177,6 +177,15 @@ constexpr size_t kStridedSliceInputNum = 1;
 constexpr size_t kStridedSliceMultiInputNumMin = 3;
 constexpr size_t kStridedSliceMultiInputNumMax = 5;
 }  // namespace
+bool StridedSlice::CheckInputs(std::vector<lite::Tensor *> inputs_) {
+  for (size_t i = 1; i < inputs_.size(); ++i) {
+    if (inputs_[i]->data_c() == nullptr) {
+      MS_LOG(DEBUG) << "strided_slice has input from other node, which only can be obtained when running.";
+      return false;
+    }
+  }
+  return true;
+}
 
 void StridedSlice::ApplyNewAxisMask() {
   for (size_t i = 0; i < new_axis_mask_.size(); i++) {
@@ -365,6 +374,10 @@ int StridedSlice::InferShape(std::vector<lite::Tensor *> inputs, std::vector<lit
       strides_.emplace_back((GetStride())[i]);
     }
   }
+  if (!CheckInputs(inputs)) {
+    MS_LOG(DEBUG) << "Do infer shape in runtime.";
+    return RET_INFER_INVALID;
+  }
   if (inputs.size() == 4) {
     // input order: input, begins, ends, strides.
     auto begin_tensor = inputs.at(1);

mindspore/lite/src/ops/strided_slice.h

@@ -47,6 +47,7 @@ class StridedSlice : public PrimitiveC {
   int UnPackToFlatBuilder(const schema::Primitive *primitive, flatbuffers::FlatBufferBuilder *fbb) override;
 #endif
   int InferShape(std::vector<lite::Tensor *> inputs_, std::vector<lite::Tensor *> outputs_) override;
+  bool CheckInputs(std::vector<lite::Tensor *> inputs_);
   int GetBeginMask() const;
   int GetEndMask() const;
   int GetEllipsisMask() const;

mindspore/lite/tools/converter/parser/tflite/tflite_model_parser.cc

@@ -81,7 +81,6 @@ FuncGraphPtr TfliteModelParser::Parse(const std::string &model_file, const std::
 
 STATUS TfliteModelParser::ConvertOps() {
   const auto &tflite_subgraph = tflite_model_->subgraphs.front();
-  const auto &tflite_model_buffers = tflite_model_->buffers;
   NoSupportOp::GetInstance()->SetFmkType("TFLITE");
   STATUS status = RET_OK;
   int op_idx = 0;
@@ -117,6 +116,9 @@ STATUS TfliteModelParser::ConvertOps() {
     std::vector<AnfNodePtr> op_inputs = {NewValueNode(std::shared_ptr<lite::PrimitiveC>(primitiveC))};
     // parse inputs
     for (auto input_idx : op->inputs) {
+      if (tflite_op_type == tflite::BuiltinOperator_FULLY_CONNECTED && input_idx == -1) {
+        continue;
+      }
       if (input_idx < 0) {
         input_idx += tflite_subgraph->tensors.size();
       }
@@ -126,18 +128,14 @@ STATUS TfliteModelParser::ConvertOps() {
         continue;
       }
       // const tensor
-      if (!tflite_model_buffers.at(input_tensor->buffer)->data.empty()) {
+      auto parameter = func_graph_->add_parameter();
-        auto parameter = func_graph_->add_parameter();
+      status = ConvertConstTensor(input_tensor.get(), parameter.get());
-        status = ConvertConstTensor(input_tensor.get(), parameter.get());
+      if (status != RET_OK) {
-        if (status != RET_OK) {
+        MS_LOG(ERROR) << "convert " << op_name << " node: " << input_idx << " const node failed.";
-          MS_LOG(ERROR) << "convert " << op_name << " node: " << input_idx << " const node failed.";
+        return status;
-          return status;
-        }
-        op_inputs.emplace_back(parameter);
-        nodes_.insert(std::pair(input_idx, parameter));
-        continue;
       }
-      MS_LOG(WARNING) << "tensor " << input_idx << " is neither a node output nor a weight tensor.";
+      op_inputs.emplace_back(parameter);
+      nodes_.insert(std::pair(input_idx, parameter));
     }
     auto new_cnode = func_graph_->NewCNode(op_inputs);
     new_cnode->set_fullname_with_scope(op_name);
@@ -268,6 +266,7 @@ STATUS TfliteModelParser::ConvertGraphOutputs() {
     auto make_tuple_prim = NewValueNode(make_tuple_prim_ptr);
     make_tuple_inputs.emplace_back(make_tuple_prim);
     for (auto outputNode : tflite_subgraph->outputs) {
+      outputNode = outputNode < 0 ? outputNode + tflite_subgraph->tensors.size() : outputNode;
       auto cnode = nodes_.at(outputNode);
       if (nullptr == cnode) {
         MS_LOG(ERROR) << "Can't find input node.";
@@ -296,9 +295,12 @@ STATUS TfliteModelParser::ConvertGraphOutputs() {
       MS_LOG(ERROR) << "GetReturnPrim return nullptr";
       return RET_NULL_PTR;
     }
+    int outputNode = tflite_subgraph->outputs.front() < 0
+                       ? static_cast<int>(tflite_subgraph->outputs.front() + tflite_subgraph->tensors.size())
+                       : static_cast<int>(tflite_subgraph->outputs.front());
     auto valueNode = NewValueNode(returnPrim);
     std::vector<AnfNodePtr> op_inputs{valueNode};
-    auto cnode = nodes_.at(tflite_subgraph->outputs.front());
+    auto cnode = nodes_.at(outputNode);
     if (nullptr == cnode) {
       MS_LOG(ERROR) << "Can't find input node.";
       return RET_NOT_FIND_OP;
@@ -345,8 +347,8 @@ STATUS TfliteModelParser::ConvertConstTensor(const tflite::TensorT *tensor, Para
     }
     std::memcpy(tensor_data, data.data(), size);
     param_value->SetTensorData(tensor_data, size);
-    parameter->set_default_param(param_value);
   }
+  parameter->set_default_param(param_value);
   return RET_OK;
 }
 

mindspore/lite/tools/optimizer/graph/group_depthwise_op_convert_pass.cc

@@ -50,8 +50,15 @@ bool GroupDepthwiseOpConvertPass::Run(const FuncGraphPtr &graph) {
     }
 
     auto data_node = depthwise_cnode->input(kConvInputIndex)->abstract();
+    if (data_node == nullptr) {
+      MS_LOG(ERROR) << "the node input is invalid.";
+      return false;
+    }
     auto data_shape = utils::cast<abstract::ShapePtr>(data_node->GetShapeTrack())->shape();
-
+    if (data_shape.empty()) {
+      MS_LOG(DEBUG) << "the tensor's shape is dynamic.";
+      return true;
+    }
     auto conv_attr = std::make_unique<schema::Conv2DT>();
     if (conv_attr == nullptr) {
       MS_LOG(ERROR) << "conv_attr is null";

mindspore/lite/tools/optimizer/graph/infershape_pass.cc

@@ -89,7 +89,7 @@ STATUS InferShapePass::SetParameterAbstract(const ParameterPtr &parameter) {
       return RET_ERROR;
     }
     auto ret = memcpy_s(tensor_data, new_value->tensor_size(), param_value->tensor_addr(), param_value->tensor_size());
-    if (ret != EOK) {
+    if (new_value->tensor_size() != 0 && ret != EOK) {
       MS_LOG(ERROR) << "memcpy error: " << ret;
       delete[] tensor_data;
       return RET_ERROR;
@@ -163,7 +163,7 @@ STATUS InferShapePass::GetCNodeInputTensors(const CNodePtr &cnode, std::vector<l
           return RET_ERROR;
         }
         ret = memcpy_s(tensor->MutableData(), tensor->Size(), param_value->tensor_addr(), param_value->tensor_size());
-        if (ret != EOK) {
+        if (tensor->Size() != 0 && ret != EOK) {
           MS_LOG(ERROR) << "memcpy error: " << ret;
           return RET_ERROR;
         }