deconv adapter

mindspore/lite/src/model_impl.cc

@@ -76,6 +76,8 @@ lite::Primitive *ModelImpl::CopyPrimitive(const schema::Primitive *srcPrim) {
       return new lite::Activation(const_cast<schema::Primitive *>(srcPrim));
     case schema::PrimitiveType_Conv2D:
       return new lite::Conv2D(const_cast<schema::Primitive *>(srcPrim));
+    case schema::PrimitiveType_DeConv2D:
+      return new lite::DeConv2D(const_cast<schema::Primitive *>(srcPrim));
     case schema::PrimitiveType_Reduce:
       return new lite::Reduce(const_cast<schema::Primitive *>(srcPrim));
     case schema::PrimitiveType_Pooling:

mindspore/lite/tools/converter/optimizer/fusion/conv_biasadd_fusion_pass.cc

@@ -81,7 +81,7 @@ STATUS ConvBiasAddFusionPass::DoFusion(MetaGraphT *graph, const std::string &pat
   }
   auto baNodeBiasTensor = graph->allTensors.at(baNodeInputIndex[BIASADD_OP_CONST_TENSOR_INDEX]).get();
   MS_ASSERT(baNodeBiasTensor != nullptr);
-  if (baNodeBiasTensor->refCount != schema::NodeType_ValueNode) {
+  if (baNodeBiasTensor->nodeType != schema::NodeType_ValueNode) {
     // dont fusion, return
     return RET_OK;
   }
@@ -215,7 +215,9 @@ STATUS ConvBiasAddFusionPass::GenConvBiasTensor(std::shared_ptr<Path> convPath,
                   << ". or bias tensor is a scaler";
     return RET_ERROR;
   }
-  if (!biasDims.empty() && biasDims.at(BIASADD_BIAS_DIM_INDEX) != kernelNum) {
+
+  bool bias_const = !biasDims.empty() && biasDims.size() == 1 && biasDims[0] == 1;
+  if (!biasDims.empty() && !bias_const && biasDims.at(BIASADD_BIAS_DIM_INDEX) != kernelNum) {
     MS_LOG(ERROR) << "Size(%d) of BiasAdd(%s) bias tensor should be equal to kernelNum(%d)"
                   << biasDims.at(BIASADD_BIAS_DIM_INDEX) << baNode->name.c_str() << kernelNum;
     return RET_ERROR;
@@ -234,6 +236,11 @@ STATUS ConvBiasAddFusionPass::GenConvBiasTensor(std::shared_ptr<Path> convPath,
       MS_LOG(ERROR) << "memset_s newBiasData failed";
       return RET_ERROR;
     }
+  } else if (bias_const) {
+    auto *biasData = reinterpret_cast<float *>(biasTensor->data.data());
+    for (size_t i = 0; i < kernelNum; i++) {
+      newBiasData[i] = *biasData;
+    }
   } else {
     if (0 != memcpy_s(newBiasData, kernelNum * sizeof(float), biasTensor->data.data(), kernelNum * sizeof(float))) {
       MS_LOG(ERROR) << "memcpy_s newBiasData failed";

mindspore/lite/tools/converter/optimizer/node/weight_format_pass.cc

@@ -152,6 +152,8 @@ int WeightFormatPass::ShapeFormatTrans(GraphNode *graphNode) {
           weightTensor->format = schema::Format_KHWC;
         } else if (opType == schema::PrimitiveType_DepthwiseConv2D) {
           weightTensor->format = schema::Format_CHWK;
+        } else if (opType == schema::PrimitiveType_DeConv2D) {
+          weightTensor->format = schema::Format_KHWC;
         } else {
           MS_LOG(ERROR) << "unsupport format";
           return -1;
@@ -355,18 +357,18 @@ int WeightFormatPass::NonQuantDataFormatTrans(GraphNode *graphNode) {
       MS_LOG(WARNING) << "TransFilter HWCKToCKHW failed, node : " << node->name.c_str();
       // todo(00445839): consider varible weight condition
     }
-  } else if (opType == schema::PrimitiveType_DeConv2D) {  // weight should be KCHW
-    if (weightTensor->format == schema::Format_KCHW) {    // from caffe or onnx
-      return 0;
-    } else if (weightTensor->format == schema::Format_HWKC) {  // from tf
-      status = TransFilterFormat<float>(weightTensor.get(), kHWKC2KCHW);
+  } else if (opType == schema::PrimitiveType_DeConv2D) {  // weight should be KHWC
+    if (weightTensor->format == schema::Format_KCHW) {  // from caffe or onnx or ms
+      status = TransFilterFormat<float>(weightTensor.get(), kKCHW2KHWC);
+    } else if (weightTensor->format == schema::Format_KHWC) {  // from tf
+      status = RET_OK;
     } else {
       MS_LOG(ERROR) << "Unsupported weightTensor format: " << weightTensor->format;
       return -1;
     }
     if (status == 0) {
-      node->primitive->value.AsDepthwiseConv2D()->format = schema::Format_NCHW;
-      weightTensor->format = schema::Format_KCHW;
+      node->primitive->value.AsDeConv2D()->format = schema::Format_NCHW;
+      weightTensor->format = schema::Format_KHWC;
     } else {
       MS_LOG(WARNING) << "TransFilter HWKCToKCHW failed, node : " << node->name.c_str();
       // todo(00445839): consider varible weight condition

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

@@ -27,8 +27,16 @@ STATUS TfliteAddParser::Parse(const std::unique_ptr<tflite::OperatorT> &tfliteOp
                               schema::CNodeT *op,
                               TensorCache *tensor_cache,
                               bool quantizedModel) {
-  // MS_LOGD("parse TfliteAddParser");
+  MS_LOG(DEBUG) << "parse TfliteAddParser";
   std::unique_ptr<schema::AddT> attr(new schema::AddT());
+  auto weight_index = tfliteOp->inputs[1];
+  const auto &weight_tensor = tfliteTensors[weight_index];
+  std::vector<tflite::TensorT *> weight_tensors{weight_tensor.get()};
+
+  if (RET_OK != ParseWeight(weight_tensors, tfliteModelBuffer, tensor_cache, schema::Format_KHWC)) {
+    return RET_ERROR;
+  }
+
   if (op != nullptr) {
     op->primitive = std::make_unique<schema::PrimitiveT>();
     op->primitive->value.type = schema::PrimitiveType_Add;