finxed weight tensor format bug and gather parser bug.

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

@@ -57,7 +57,7 @@ STATUS TfliteDoubleInputOpParser::Parse(const std::unique_ptr<tflite::OperatorT>
     }
     if (!x_data->data.empty()) {
       std::vector<tflite::TensorT *> x_tensors{x_tensor.get()};
-      if (RET_OK != ParseTensor(x_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+      if (RET_OK != ParseTensor(x_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
         MS_LOG(ERROR) << "parse the first tensor failed";
         return RET_ERROR;
       }
@@ -76,7 +76,7 @@ STATUS TfliteDoubleInputOpParser::Parse(const std::unique_ptr<tflite::OperatorT>
     }
     if (!y_data->data.empty()) {
       std::vector<tflite::TensorT *> y_tensors{y_tensor.get()};
-      if (RET_OK != ParseTensor(y_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+      if (RET_OK != ParseTensor(y_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
         MS_LOG(ERROR) << "parse the second tensor failed";
         return RET_ERROR;
       }

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

@@ -59,7 +59,7 @@ STATUS TfliteConvParser::Parse(const std::unique_ptr<tflite::OperatorT> &tfliteO
     return RET_NULL_PTR;
   }
   std::vector<tflite::TensorT *> weight_tensors{weight_tensor.get()};
-  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, true)) {
     MS_LOG(ERROR) << "parse weight failed";
     return RET_ERROR;
   }
@@ -79,7 +79,7 @@ STATUS TfliteConvParser::Parse(const std::unique_ptr<tflite::OperatorT> &tfliteO
       return RET_NULL_PTR;
     }
     std::vector<tflite::TensorT *> bias_tensors{bias_tensor.get()};
-    if (RET_OK != ParseTensor(bias_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+    if (RET_OK != ParseTensor(bias_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
       MS_LOG(ERROR) << "parse bias failed";
       return RET_ERROR;
     }

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

@@ -59,7 +59,7 @@ STATUS TfliteDeConvParser::Parse(const std::unique_ptr<tflite::OperatorT> &tflit
     return RET_NULL_PTR;
   }
   std::vector<tflite::TensorT *> weight_tensors{weight_tensor.get()};
-  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, true)) {
     return RET_ERROR;
   }
   auto weight_shape = weight_tensor->shape;

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

@@ -123,7 +123,7 @@ STATUS TfliteDepthwiseConv2DParser::Parse(const std::unique_ptr<tflite::Operator
 
   std::vector<tflite::TensorT *> weight_tensors{weight_tensor.get()};
 
-  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, true)) {
     MS_LOG(ERROR) << "parse weight failed";
     return RET_ERROR;
   }
@@ -133,7 +133,7 @@ STATUS TfliteDepthwiseConv2DParser::Parse(const std::unique_ptr<tflite::Operator
     auto bias_index = tflite_op->inputs[2];
     const auto &bias_tensor = tflite_tensors[bias_index];
     std::vector<tflite::TensorT *> bias_tensors{bias_tensor.get()};
-    if (RET_OK != ParseTensor(bias_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+    if (RET_OK != ParseTensor(bias_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
       MS_LOG(ERROR) << "parse bias failed";
       return RET_ERROR;
     }

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

@@ -44,7 +44,7 @@ STATUS TfliteFakeQuantParser::Parse(const std::unique_ptr<tflite::OperatorT> &tf
     return RET_NULL_PTR;
   }
   std::vector<tflite::TensorT *> weight_tensors{weight_tensor.get()};
-  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, true)) {
     MS_LOG(ERROR) << "parse weight failed";
     return RET_ERROR;
   }
@@ -58,7 +58,7 @@ STATUS TfliteFakeQuantParser::Parse(const std::unique_ptr<tflite::OperatorT> &tf
       return RET_NULL_PTR;
     }
     std::vector<tflite::TensorT *> bias_tensors{bias_tensor.get()};
-    if (RET_OK != ParseTensor(bias_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+    if (RET_OK != ParseTensor(bias_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
       MS_LOG(ERROR) << "parse bias failed";
       return RET_ERROR;
     }

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

@@ -45,7 +45,7 @@ STATUS TfliteFullyConnectedParser::Parse(const std::unique_ptr<tflite::OperatorT
     return RET_NULL_PTR;
   }
   std::vector<tflite::TensorT *> weight_tensors{weight_tensor.get()};
-  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, true)) {
     MS_LOG(ERROR) << "parse weight failed";
     return RET_ERROR;
   }
@@ -59,7 +59,7 @@ STATUS TfliteFullyConnectedParser::Parse(const std::unique_ptr<tflite::OperatorT
       return RET_NULL_PTR;
     }
     std::vector<tflite::TensorT *> bias_tensors{bias_tensor.get()};
-    if (RET_OK != ParseTensor(bias_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+    if (RET_OK != ParseTensor(bias_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
       MS_LOG(ERROR) << "parse bias failed";
       return RET_ERROR;
     }

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

@@ -58,7 +58,7 @@ STATUS TfliteGatherNdParser::Parse(const std::unique_ptr<tflite::OperatorT> &tfl
   }
   if (!y_data->data.empty()) {
     std::vector<tflite::TensorT *> y_tensors{y_tensor.get()};
-    if (RET_OK != ParseTensor(y_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+    if (RET_OK != ParseTensor(y_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
       MS_LOG(ERROR) << "parse the second tensor failed";
       return RET_ERROR;
     }

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

@@ -49,6 +49,25 @@ STATUS TfliteGatherParser::Parse(const std::unique_ptr<tflite::OperatorT> &tflit
 
   attr->batchDims = 0;
 
+  auto y_index = tfliteOp->inputs[1];
+  const auto &y_tensor = tfliteTensors[y_index];
+  if (y_tensor == nullptr) {
+    MS_LOG(ERROR) << "the second input is null";
+    return RET_NULL_PTR;
+  }
+  auto &y_data = tfliteModelBuffer.at(y_tensor->buffer);
+  if (y_data == nullptr) {
+    MS_LOG(ERROR) << "the data of the second input is null";
+    return RET_NULL_PTR;
+  }
+  if (!y_data->data.empty()) {
+    std::vector<tflite::TensorT *> y_tensors{y_tensor.get()};
+    if (RET_OK != ParseTensor(y_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
+      MS_LOG(ERROR) << "parse the second tensor failed";
+      return RET_ERROR;
+    }
+  }
+
   op->primitive->value.type = schema::PrimitiveType_Gather;
   op->primitive->value.value = attr.release();
   return RET_OK;

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

@@ -45,7 +45,8 @@ STATUS TfliteNodeParser::CopyTfliteTensorData(const std::vector<std::unique_ptr<
 STATUS TfliteNodeParser::ParseTensor(const std::vector<tflite::TensorT *> &ts,
                                      const std::vector<std::unique_ptr<tflite::BufferT>> &tfliteModelBuffer,
                                      mindspore::lite::TensorCache *tensor_cache,
-                                     int node_type) {
+                                     int node_type,
+                                     bool isWeight) {
   for (const auto &t : ts) {
     auto idx = tensor_cache->FindTensor(t->name);
     if (idx < 0) {
@@ -53,6 +54,12 @@ STATUS TfliteNodeParser::ParseTensor(const std::vector<tflite::TensorT *> &ts,
       tensor->dataType = GetTfliteDataType(t->type);
       tensor->dims = t->shape;
 
+      if (isWeight) {
+        tensor->format = schema::Format_KHWC;
+      } else {
+        tensor->format = schema::Format_NHWC;
+      }
+
       if (t->buffer > 0) {
         CopyTfliteTensorData(tfliteModelBuffer, t, tensor.get());
       }

mindspore/lite/tools/converter/parser/tflite/tflite_node_parser.h

@@ -47,7 +47,8 @@ class TfliteNodeParser {
   STATUS ParseTensor(const std::vector<tflite::TensorT *> &ts,
                      const std::vector<std::unique_ptr<tflite::BufferT>> &tfliteModelBuffer,
                      mindspore::lite::TensorCache *tensor_cache,
-                     int node_type);
+                     int node_type,
+                     bool isWeight);
 
   STATUS CopyTfliteTensorData(const std::vector<std::unique_ptr<tflite::BufferT>> &tfliteModelBuffer,
                               const tflite::TensorT *tflite_tensor,

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

@@ -50,7 +50,7 @@ STATUS TfliteTransposeParser::Parse(const std::unique_ptr<tflite::OperatorT> &tf
     return RET_ERROR;
   }
   std::vector<tflite::TensorT *> weight_tensors{weight_tensor.get()};
-  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST)) {
+  if (RET_OK != ParseTensor(weight_tensors, tfliteModelBuffer, tensor_cache, TF_CONST, false)) {
     MS_LOG(ERROR) << "parse weight failed";
     return RET_ERROR;
   }