diff --git a/mindspore/lite/micro/cmake/file_list.cmake b/mindspore/lite/micro/cmake/file_list.cmake
index e9ed0bac18..8a40bd8ed1 100644
--- a/mindspore/lite/micro/cmake/file_list.cmake
+++ b/mindspore/lite/micro/cmake/file_list.cmake
@@ -101,6 +101,7 @@ set(CODER_OPCODERS_SRC
         ${MICRO_DIR}/coder/opcoders/nnacl/int8/conv2d_1x1_int8_coder.cc
         ${MICRO_DIR}/coder/opcoders/nnacl/int8/conv2d_3x3_int8_coder.cc
         ${MICRO_DIR}/coder/opcoders/nnacl/int8/conv2d_int8_coder.cc
+        ${MICRO_DIR}/coder/opcoders/nnacl/int8/convolution_depthwise_int8_coder.cc
         ${MICRO_DIR}/coder/opcoders/nnacl/int8/deconvolution_int8_coder.cc
         ${MICRO_DIR}/coder/opcoders/nnacl/int8/pooling_int8_coder.cc
         ${MICRO_DIR}/coder/opcoders/nnacl/int8/resize_int8_coder.cc
diff --git a/mindspore/lite/micro/cmake/package_wrapper.cmake b/mindspore/lite/micro/cmake/package_wrapper.cmake
index fbaa20b567..54bc68e226 100644
--- a/mindspore/lite/micro/cmake/package_wrapper.cmake
+++ b/mindspore/lite/micro/cmake/package_wrapper.cmake
@@ -7,7 +7,9 @@ set(RUNTIME_SRC
         )
 
 set(WRAPPER_SRC
+        ${WRAPPER_DIR}/base/common_wrapper.c
         ${WRAPPER_DIR}/base/detection_post_process_base_wrapper.c
+        ${WRAPPER_DIR}/base/optimize_handler_wrapper.c
         ${WRAPPER_DIR}/fp32/matmul_fp32_wrapper.c
         ${WRAPPER_DIR}/int8/matmul_int8_wrapper.c
         ${WRAPPER_DIR}/int8/add_int8_wrapper.c
diff --git a/mindspore/lite/micro/coder/opcoders/nnacl/int8/conv2d_int8_coder.cc b/mindspore/lite/micro/coder/opcoders/nnacl/int8/conv2d_int8_coder.cc
index 6f37ec18fe..8f7fc8b6f0 100644
--- a/mindspore/lite/micro/coder/opcoders/nnacl/int8/conv2d_int8_coder.cc
+++ b/mindspore/lite/micro/coder/opcoders/nnacl/int8/conv2d_int8_coder.cc
@@ -21,6 +21,7 @@
 #include "securec/include/securec.h"
 #include "coder/opcoders/nnacl/int8/conv2d_1x1_int8_coder.h"
 #include "coder/opcoders/nnacl/int8/conv2d_3x3_int8_coder.h"
+#include "coder/opcoders/nnacl/int8/convolution_depthwise_int8_coder.h"
 #include "coder/opcoders/serializers/nnacl_serializer/nnacl_int8_serializer.h"
 #include "src/runtime/kernel/arm/base/convolution_base.h"
 #include "src/ops/populate/populate_register.h"
@@ -250,7 +251,39 @@ std::unique_ptr<OperatorCoder> CPUConv2DINT8CoderCreator(const std::vector<Tenso
   return coder;
 }
 
-REG_OPERATOR_CODER(kX86, kNumberTypeInt8, PrimitiveType_Conv2DFusion, CPUConv2DINT8CoderCreator)
-REG_OPERATOR_CODER(kARM32A, kNumberTypeInt8, PrimitiveType_Conv2DFusion, CPUConv2DINT8CoderCreator)
-REG_OPERATOR_CODER(kARM64, kNumberTypeInt8, PrimitiveType_Conv2DFusion, CPUConv2DINT8CoderCreator)
+std::unique_ptr<OperatorCoder> CPUConv2DFusionINT8CoderCreator(const std::vector<Tensor *> &in_tensors,
+                                                               const std::vector<Tensor *> &out_tensors,
+                                                               const Model::Node *node, size_t node_index,
+                                                               Target target) {
+  const void *primitive = node->primitive_;
+  if (primitive == nullptr) {
+    return nullptr;
+  }
+  int schema_version = VersionManager::GetInstance()->GetSchemaVersion();
+  ParameterGen paramGen =
+    PopulateRegistry::GetInstance()->GetParameterCreator(GetPrimitiveType(node->primitive_), schema_version);
+  if (paramGen == nullptr) {
+    MS_LOG(ERROR) << "parameter generator is null";
+    return nullptr;
+  }
+  auto conv_param = reinterpret_cast<ConvParameter *>(paramGen(node->primitive_));
+  std::unique_ptr<OperatorCoder> coder;
+  if (conv_param->group_ == 1) {
+    coder = CPUConv2DINT8CoderCreator(in_tensors, out_tensors, node, node_index, target);
+  } else if (conv_param->group_ == conv_param->input_channel_ && conv_param->group_ == conv_param->output_channel_) {
+    coder = CPUOpCoderCreator<ConvolutionDepthwiseINT8Coder>(in_tensors, out_tensors, node, node_index, target);
+  } else {
+    // group conv
+  }
+  free(conv_param);
+  if (coder == nullptr) {
+    MS_LOG(ERROR) << "create conv2d int8 coder failed";
+    return nullptr;
+  }
+  return coder;
+}
+
+REG_OPERATOR_CODER(kX86, kNumberTypeInt8, PrimitiveType_Conv2DFusion, CPUConv2DFusionINT8CoderCreator)
+REG_OPERATOR_CODER(kARM32A, kNumberTypeInt8, PrimitiveType_Conv2DFusion, CPUConv2DFusionINT8CoderCreator)
+REG_OPERATOR_CODER(kARM64, kNumberTypeInt8, PrimitiveType_Conv2DFusion, CPUConv2DFusionINT8CoderCreator)
 }  // namespace mindspore::lite::micro::nnacl
diff --git a/mindspore/lite/micro/coder/opcoders/nnacl/int8/convolution_depthwise_int8_coder.cc b/mindspore/lite/micro/coder/opcoders/nnacl/int8/convolution_depthwise_int8_coder.cc
index b40c68782f..1ba7f604b1 100644
--- a/mindspore/lite/micro/coder/opcoders/nnacl/int8/convolution_depthwise_int8_coder.cc
+++ b/mindspore/lite/micro/coder/opcoders/nnacl/int8/convolution_depthwise_int8_coder.cc
@@ -22,8 +22,6 @@
 #include "coder/opcoders/serializers/nnacl_serializer/nnacl_int8_serializer.h"
 #include "nnacl/int8/conv_depthwise_int8.h"
 
-using mindspore::schema::PrimitiveType_DepthwiseConv2D;
-
 namespace mindspore::lite::micro {
 
 int ConvolutionDepthwiseINT8Coder::Prepare(CoderContext *const context) {
@@ -105,6 +103,4 @@ int ConvolutionDepthwiseINT8Coder::DoCode(CoderContext *const context) {
   return RET_OK;
 }
 
-REG_OPERATOR_CODER(kAllTargets, kNumberTypeInt8, PrimitiveType_DepthwiseConv2D,
-                   CPUOpCoderCreator<ConvolutionDepthwiseINT8Coder>)
 }  // namespace mindspore::lite::micro
diff --git a/mindspore/lite/micro/coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.cc b/mindspore/lite/micro/coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.cc
index a73fd82044..414e07ee34 100644
--- a/mindspore/lite/micro/coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.cc
+++ b/mindspore/lite/micro/coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.cc
@@ -17,6 +17,7 @@
 #include "coder/opcoders/serializers/nnacl_serializer/nnacl_fp32_serializer.h"
 #include "src/common/log_adapter.h"
 #include "coder/log.h"
+#include "coder/opcoders/parallel.h"
 #include "nnacl/pooling_parameter.h"
 
 namespace mindspore::lite::micro::nnacl {
@@ -34,7 +35,7 @@ void NNaclFp32Serializer::CodeStruct(const std::string &name, const PoolingParam
                  pooling_parameter.output_batch_, pooling_parameter.output_channel_, pooling_parameter.pad_u_,
                  pooling_parameter.pad_d_, pooling_parameter.pad_l_, pooling_parameter.pad_r_,
                  // other parameter
-                 pooling_parameter.thread_num_, nullptr, pooling_parameter.quantize_);
+                 gThreadNum, nullptr, pooling_parameter.quantize_);
 }
 
 void NNaclFp32Serializer::CodeStruct(const std::string &name, const BatchNormParameter &batch_norm_parameter) {
@@ -60,14 +61,16 @@ void NNaclFp32Serializer::CodeStruct(const std::string &name, const SoftmaxParam
 }
 
 void NNaclFp32Serializer::CodeStruct(const std::string &name, const ConvParameter &conv_parameter) {
+  code << "int thread_num = MSMIN(" << gThreadNum << ", " << conv_parameter.output_h_ << ");\n";
   CodeBaseStruct("ConvParameter", name, conv_parameter.op_parameter_, "{}", conv_parameter.kernel_h_,
                  conv_parameter.kernel_w_, conv_parameter.stride_h_, conv_parameter.stride_w_,
                  conv_parameter.dilation_h_, conv_parameter.dilation_w_, conv_parameter.pad_u_, conv_parameter.pad_d_,
                  conv_parameter.pad_l_, conv_parameter.pad_r_, conv_parameter.group_, conv_parameter.tile_num_,
                  conv_parameter.input_batch_, conv_parameter.input_h_, conv_parameter.input_w_,
                  conv_parameter.input_channel_, conv_parameter.output_batch_, conv_parameter.output_h_,
-                 conv_parameter.output_w_, conv_parameter.output_channel_, conv_parameter.op_parameter_.thread_num_,
-                 conv_parameter.input_unit_, conv_parameter.output_unit_, conv_parameter.act_type_);
+                 conv_parameter.output_w_, conv_parameter.output_channel_, "thread_num", conv_parameter.input_unit_,
+                 conv_parameter.output_unit_, conv_parameter.pad_mode_, conv_parameter.act_type_,
+                 conv_parameter.channel_multiplie_, conv_parameter.output_padding_w_, conv_parameter.output_padding_h_);
 }
 
 void NNaclFp32Serializer::CodeStruct(const std::string &name, const MatMulParameter &mat_mul_parameter) {
diff --git a/mindspore/lite/micro/coder/opcoders/serializers/nnacl_serializer/nnacl_int8_serializer.cc b/mindspore/lite/micro/coder/opcoders/serializers/nnacl_serializer/nnacl_int8_serializer.cc
index 82eca83215..2235d1e8e4 100644
--- a/mindspore/lite/micro/coder/opcoders/serializers/nnacl_serializer/nnacl_int8_serializer.cc
+++ b/mindspore/lite/micro/coder/opcoders/serializers/nnacl_serializer/nnacl_int8_serializer.cc
@@ -57,7 +57,8 @@ void NNaclInt8Serializer::CodeStruct(const std::string &name, const ConvParamete
                  conv_parameter.input_batch_, conv_parameter.input_h_, conv_parameter.input_w_,
                  conv_parameter.input_channel_, conv_parameter.output_batch_, conv_parameter.output_h_,
                  conv_parameter.output_w_, conv_parameter.output_channel_, "thread_num", conv_parameter.input_unit_,
-                 conv_parameter.output_unit_, conv_parameter.pad_mode_, conv_parameter.act_type_);
+                 conv_parameter.output_unit_, conv_parameter.pad_mode_, conv_parameter.act_type_,
+                 conv_parameter.channel_multiplie_, conv_parameter.output_padding_w_, conv_parameter.output_padding_h_);
 }
 
 void NNaclInt8Serializer::CodeStruct(const std::string &name, const MatMulParameter &matmul_parameter) {