!3701 lite int8 pad op

Merge pull request !3701 from ling/pad

mindspore/lite/schema/ops.fbs

@@ -99,8 +99,9 @@ enum PaddingMode : byte {
 }
 
 table Pad {
-    paddingmode: PaddingMode;
     paddings: [int];
+    paddingMode: PaddingMode;
+    constantValue: float;
 }
 
 table Maximum {

mindspore/lite/src/ops/ops.cc

@@ -45,6 +45,8 @@ Primitive *Primitive::CreatePrimitive(schema::Primitive *primitive) {
       return new lite::FullConnection(const_cast<schema::Primitive *>(primitive));
     case schema::PrimitiveType_Power:
       return new lite::Power(const_cast<schema::Primitive *>(primitive));
+    case schema::PrimitiveType_Pad:
+      return new lite::Pad(const_cast<schema::Primitive *>(primitive));
     case schema::PrimitiveType_Range:
       return new lite::Range(const_cast<schema::Primitive *>(primitive));
     case schema::PrimitiveType_Mul:

mindspore/lite/src/populate_parameter.cc

@@ -43,7 +43,7 @@
 #include "src/runtime/kernel/arm/opclib/fp32/local_response_norm.h"
 #include "src/runtime/kernel/arm/opclib/fp32/expandDims.h"
 #include "src/runtime/kernel/arm/opclib/fp32/arithmetic_self.h"
-#include "src/runtime/kernel/arm/opclib/pad.h"
+#include "src/runtime/kernel/arm/opclib/pad_parameter.h"
 #include "src/runtime/kernel/arm/opclib/fp32/fill.h"
 #include "src/runtime/kernel/arm/opclib/transpose.h"
 #include "src/runtime/kernel/arm/opclib/split.h"
@@ -300,6 +300,42 @@ ConvParameter *PopulateDeconvDwParameter(const lite::Primitive *primitive) {
   return parameter;
 }
 
+ConvParameter *PopulateDeconvParameter(const lite::Primitive *primitive) {
+  ConvParameter *parameter = new ConvParameter();
+  auto conv_primitive = primitive->Value()->value_as_DeConv2D();
+  parameter->kernel_h_ = conv_primitive->kernelH();
+  parameter->kernel_w_ = conv_primitive->kernelW();
+  parameter->stride_h_ = conv_primitive->strideH();
+  parameter->stride_w_ = conv_primitive->strideW();
+
+  auto deconv_lite_primitive = (lite::DeConv2D *)primitive;
+  MS_ASSERT(nullptr != deconvdw_lite_primitive);
+  parameter->pad_u_ = deconv_lite_primitive->PadUp();
+  parameter->pad_d_ = deconv_lite_primitive->PadDown();
+  parameter->pad_l_ = deconv_lite_primitive->PadLeft();
+  parameter->pad_r_ = deconv_lite_primitive->PadRight();
+  parameter->pad_h_ = deconv_lite_primitive->PadUp();
+  parameter->pad_w_ = deconv_lite_primitive->PadLeft();
+  parameter->dilation_h_ = conv_primitive->dilateH();
+  parameter->dilation_w_ = conv_primitive->dilateW();
+  auto act_type = conv_primitive->activationType();
+  switch (act_type) {
+    case schema::ActivationType_RELU:
+      parameter->is_relu_ = true;
+      parameter->is_relu6_ = false;
+      break;
+    case schema::ActivationType_RELU6:
+      parameter->is_relu_ = false;
+      parameter->is_relu6_ = true;
+      break;
+    default:
+      parameter->is_relu_ = false;
+      parameter->is_relu6_ = false;
+      break;
+  }
+  return parameter;
+}
+
 SoftmaxParameter *PopulateSoftmaxParameter(const lite::Primitive *primitive) {
   auto softmax_primitive = primitive->Value()->value_as_SoftMax();
   SoftmaxParameter *parameter = new (std::nothrow) SoftmaxParameter();
@@ -335,19 +371,31 @@ ReduceParameter *PopulateReduceParameter(const lite::Primitive *primitive) {
 }
 
 PadParameter *PopulatePadParameter(const lite::Primitive *primitive) {
-  PadParameter *parameter = new (std::nothrow) PadParameter();
-  if (parameter == nullptr) {
+  PadParameter *pad_param = new (std::nothrow) PadParameter();
+  if (pad_param == nullptr) {
     MS_LOG(ERROR) << "new PadParameter failed.";
     return nullptr;
   }
-  auto param = primitive->Value()->value_as_Pad();
-  auto size = param->paddings()->size();
-  parameter->ori_size_ = size;
-  auto valid_size = size <= 8 ? size : 8;
-  for (size_t i = 0; i < valid_size; i++) {
-    parameter->paddings[i] = (*(param->paddings()))[i];
+  auto pad_node = primitive->Value()->value_as_Pad();
+
+  pad_param->pad_mode_ = pad_node->paddingMode();
+  if (pad_param->pad_mode_ == schema::PaddingMode_CONSTANT) {
+    pad_param->constant_value_ = pad_node->constantValue();
+  } else {
+    MS_LOG(ERROR) << "Invalid padding mode: " << pad_param->pad_mode_;
+    return nullptr;
   }
-  return parameter;
+
+  auto size = pad_node->paddings()->size();
+  if (size > MAX_PAD_SIZE) {
+    MS_LOG(ERROR) << "Invalid padding size: " << size;
+    return nullptr;
+  }
+
+  for (size_t i = 0; i < size; i++) {
+    pad_param->paddings_[MAX_PAD_SIZE - size + i] = (*(pad_node->paddings()))[i];
+  }
+  return pad_param;
 }
 
 ActivationParameter *PopulateActivationParameter(const lite::Primitive *primitive) {
@@ -891,7 +939,7 @@ FlattenParameter *PopulateFlattenParameter(const lite::Primitive *primitive) {
     MS_LOG(ERROR) << "new FlattenParameter fail!";
     return nullptr;
   }
-    return parameter;
+  return parameter;
 }
 
 StridedSliceParameter *PopulateStridedSliceParam(const lite::Primitive *primitive) {
@@ -932,6 +980,8 @@ OpParameter *PopulateParameter(const lite::Primitive *primitive) {
       return reinterpret_cast<OpParameter *>(PopulateConvDwParameter(primitive));
     case schema::PrimitiveType_DeDepthwiseConv2D:
       return reinterpret_cast<OpParameter *>(PopulateDeconvDwParameter(primitive));
+    case schema::PrimitiveType_DeConv2D:
+      return reinterpret_cast<OpParameter *>(PopulateDeconvParameter(primitive));
     case schema::PrimitiveType_FusedBatchNorm:
       return reinterpret_cast<OpParameter *>(PopulateFusedBatchNorm(primitive));
     case schema::PrimitiveType_FullConnection:

mindspore/lite/src/runtime/kernel/arm/base/convolution_base.cc

@@ -49,41 +49,51 @@ namespace mindspore::kernel {
 ConvolutionBaseCPUKernel::~ConvolutionBaseCPUKernel() {
   if (bias_data_ != nullptr) {
     free(bias_data_);
+    bias_data_ = nullptr;
   }
   if (nhwc4_input_ != nullptr) {
     free(nhwc4_input_);
+    nhwc4_input_ = nullptr;
   }
 }
 
 void ConvolutionBaseCPUKernel::FreeQuantParam() {
-  if (quant_args_ != nullptr) {
+  ConvQuantArg *conv_quant_arg_ = &conv_param_->conv_quant_arg_;
+  if (conv_quant_arg_ == nullptr) {
+    return;
+  }
+  if (conv_quant_arg_->real_multiplier_ != nullptr) {
+    free(conv_quant_arg_->real_multiplier_);
+    conv_quant_arg_->real_multiplier_ = nullptr;
+  }
+  if (conv_quant_arg_->left_shift_ != nullptr) {
+    free(conv_quant_arg_->left_shift_);
+    conv_quant_arg_->left_shift_ = nullptr;
+  }
+  if (conv_quant_arg_->right_shift_ != nullptr) {
+    free(conv_quant_arg_->right_shift_);
+    conv_quant_arg_->right_shift_ = nullptr;
+  }
+  if (conv_quant_arg_->quant_multiplier_ != nullptr) {
+    free(conv_quant_arg_->quant_multiplier_);
+    conv_quant_arg_->quant_multiplier_ = nullptr;
+  }
+  if (conv_quant_arg_->out_act_min_ != nullptr) {
+    free(conv_quant_arg_->out_act_min_);
+    conv_quant_arg_->out_act_min_ = nullptr;
+  }
+  if (conv_quant_arg_->out_act_max_ != nullptr) {
+    free(conv_quant_arg_->out_act_max_);
+    conv_quant_arg_->out_act_max_ = nullptr;
+  }
+
+  if (conv_quant_arg_->quant_args_ != nullptr) {
     for (int i = 0; i < 3; ++i) {
-      if (*(quant_args_ + i) != nullptr) {
-        free(*(quant_args_ + i));
+      if (*(conv_quant_arg_->quant_args_ + i) != nullptr) {
+        free(*(conv_quant_arg_->quant_args_ + i));
       }
     }
   }
-  if (conv_quant_arg_ != nullptr) {
-    if (conv_quant_arg_->real_multiplier_ != nullptr) {
-      free(conv_quant_arg_->real_multiplier_);
-    }
-    if (conv_quant_arg_->left_shift_ != nullptr) {
-      free(conv_quant_arg_->left_shift_);
-    }
-    if (conv_quant_arg_->right_shift_ != nullptr) {
-      free(conv_quant_arg_->right_shift_);
-    }
-    if (conv_quant_arg_->quant_multiplier_ != nullptr) {
-      free(conv_quant_arg_->quant_multiplier_);
-    }
-    if (conv_quant_arg_->out_act_min_ != nullptr) {
-      free(conv_quant_arg_->out_act_min_);
-    }
-    if (conv_quant_arg_->out_act_max_ != nullptr) {
-      free(conv_quant_arg_->out_act_max_);
-    }
-    free(conv_quant_arg_);
-  }
 }
 
 int ConvolutionBaseCPUKernel::Init() {
@@ -116,11 +126,19 @@ int ConvolutionBaseCPUKernel::CheckLayout(lite::tensor::Tensor *input_tensor) {
 }
 
 int ConvolutionBaseCPUKernel::SetQuantParam() {
-  conv_quant_arg_ = new ConvQuantArg();
-  quant_args_ = reinterpret_cast<QuantArg **>(malloc(3 * sizeof(QuantArg *)));
+  ConvQuantArg *conv_quant_arg_ = &conv_param_->conv_quant_arg_;
+  conv_quant_arg_->quant_args_ = reinterpret_cast<QuantArg **>(malloc(3 * sizeof(QuantArg *)));
+  if (conv_quant_arg_->quant_args_ == nullptr) {
+    MS_LOG(ERROR) << "malloc quant_args_ failed.";
+    return RET_ERROR;
+  }
   // per-tensor init
   for (int j = 0; j < 3; ++j) {
-    quant_args_[j] = reinterpret_cast<QuantArg *>(malloc(sizeof(QuantArg)));
+    conv_quant_arg_->quant_args_[j] = reinterpret_cast<QuantArg *>(malloc(sizeof(QuantArg)));
+    if (conv_quant_arg_->quant_args_[j] == nullptr) {
+      MS_LOG(ERROR) << "malloc quant_args_ failed.";
+      return RET_ERROR;
+    }
   }
   auto input_tensor = inputs_.at(kInputIndex);
   auto weight_tensor = inputs_.at(kWeightIndex);
@@ -129,16 +147,15 @@ int ConvolutionBaseCPUKernel::SetQuantParam() {
   auto weight_quant_arg = weight_tensor->GetQuantParams().front();
   auto output_quant_arg = output_tensor->GetQuantParams().front();
   // input
-  quant_args_[0][0].zp_ = input_quant_arg.zeroPoint;
-  quant_args_[0][0].scale_ = input_quant_arg.scale;
+  conv_quant_arg_->quant_args_[0][0].zp_ = input_quant_arg.zeroPoint;
+  conv_quant_arg_->quant_args_[0][0].scale_ = input_quant_arg.scale;
   // weight
-  quant_args_[1][0].zp_ = weight_quant_arg.zeroPoint;
-  quant_args_[1][0].scale_ = weight_quant_arg.scale;
+  conv_quant_arg_->quant_args_[1][0].zp_ = weight_quant_arg.zeroPoint;
+  conv_quant_arg_->quant_args_[1][0].scale_ = weight_quant_arg.scale;
   // output
-  quant_args_[2][0].zp_ = output_quant_arg.zeroPoint;
-  quant_args_[2][0].scale_ = output_quant_arg.scale;
+  conv_quant_arg_->quant_args_[2][0].zp_ = output_quant_arg.zeroPoint;
+  conv_quant_arg_->quant_args_[2][0].scale_ = output_quant_arg.scale;
 
-  conv_quant_arg_->quant_args_ = quant_args_;
   conv_quant_arg_->real_multiplier_ = reinterpret_cast<double *>(malloc(sizeof(double)));
   conv_quant_arg_->left_shift_ = reinterpret_cast<int32_t *>(malloc(sizeof(int32_t)));
   conv_quant_arg_->right_shift_ = reinterpret_cast<int32_t *>(malloc(sizeof(int32_t)));
@@ -151,7 +168,6 @@ int ConvolutionBaseCPUKernel::SetQuantParam() {
   QuantizeRoundParameter(real_multiplier, &conv_quant_arg_->quant_multiplier_[0], &conv_quant_arg_->left_shift_[0],
                          &conv_quant_arg_->right_shift_[0]);
 
-  conv_param_->conv_quant_arg_ = *conv_quant_arg_;
   ComputeQuantOutRange(conv_param_);
   return RET_OK;
 }

mindspore/lite/src/runtime/kernel/arm/base/convolution_base.h

@@ -57,15 +57,12 @@ class ConvolutionBaseCPUKernel : public LiteKernel {
   int thread_count_;
   int tile_num_;
   void *bias_data_ = nullptr;
-  void *nhwc4_input_;
+  void *nhwc4_input_ = nullptr;
   const Context *ctx_;
   ConvParameter *conv_param_;
-  ConvQuantArg *conv_quant_arg_ = nullptr;
-  QuantArg **quant_args_ = nullptr;
   LayoutConvertor convert_func_;
 };
 void ComputeQuantOutRange(ConvParameter *conv_param);
 }  // namespace mindspore::kernel
 
 #endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_BASE_CONVOLUTION_BASE_H_
-

mindspore/lite/src/runtime/kernel/arm/base/pad.cc

@@ -0,0 +1,90 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <vector>
+#include "src/runtime/kernel/arm/fp32/pad.h"
+#include "src/runtime/kernel/arm/int8/pad_int8.h"
+#include "schema/model_generated.h"
+#include "src/kernel_factory.h"
+#include "include/errorcode.h"
+#include "include/context.h"
+
+using mindspore::lite::KernelRegistrar;
+using mindspore::lite::RET_ERROR;
+using mindspore::lite::RET_OK;
+using mindspore::schema::PrimitiveType_Pad;
+
+namespace mindspore::kernel {
+
+kernel::LiteKernel *CpuPadInt8KernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
+                                            const std::vector<lite::tensor::Tensor *> &outputs,
+                                            OpParameter *opParameter, const lite::Context *ctx,
+                                            const kernel::KernelKey &desc) {
+  auto *kernel = new (std::nothrow) PadInt8CPUKernel(opParameter, inputs, outputs, ctx);
+  if (kernel == nullptr) {
+    MS_LOG(ERROR) << "new PadCPUKernel failed.";
+    return nullptr;
+  }
+  return kernel;
+}
+
+kernel::LiteKernel *CpuPadFp32KernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
+                                            const std::vector<lite::tensor::Tensor *> &outputs,
+                                            OpParameter *opParameter, const lite::Context *ctx,
+                                            const kernel::KernelKey &desc) {
+  auto *kernel = new (std::nothrow) PadCPUKernel(opParameter, inputs, outputs, ctx);
+  if (kernel == nullptr) {
+    MS_LOG(ERROR) << "new PadCPUKernel failed.";
+    return nullptr;
+  }
+  return kernel;
+}
+
+kernel::LiteKernel *CpuPadKernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
+                                        const std::vector<lite::tensor::Tensor *> &outputs, OpParameter *opParameter,
+                                        const lite::Context *ctx, const kernel::KernelKey &desc) {
+  MS_ASSERT(opParameter != nullptr);
+  MS_ASSERT(desc.type == schema::PrimitiveType_Concat);
+  auto input_tensor = inputs.at(kInputIndex);
+  auto data_type = input_tensor->data_type();
+  kernel::LiteKernel *kernel = nullptr;
+  switch (data_type) {
+    case kNumberTypeInt8:
+      kernel = CpuPadInt8KernelCreator(inputs, outputs, opParameter, ctx, desc);
+      break;
+    case kNumberTypeFloat32:
+      kernel = CpuPadFp32KernelCreator(inputs, outputs, opParameter, ctx, desc);
+      break;
+    default:
+      break;
+  }
+
+  if (kernel == nullptr) {
+    MS_LOG(ERROR) << "kernel is nullptr.";
+    return nullptr;
+  }
+
+  auto ret = kernel->Init();
+  if (ret != RET_OK) {
+    delete kernel;
+    MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_ << ", type: "
+                  << schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(opParameter->type_));
+    return nullptr;
+  }
+  return kernel;
+}
+
+REG_KERNEL(kCPU, PrimitiveType_Pad, CpuPadKernelCreator)
+}  // namespace mindspore::kernel

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_1x1.cc

@@ -39,10 +39,6 @@ Convolution1x1CPUKernel::~Convolution1x1CPUKernel() {
     free(tmp_ptr_);
     tmp_ptr_ = nullptr;
   }
-  if (bias_ptr_ != nullptr) {
-    free(bias_ptr_);
-    bias_ptr_ = nullptr;
-  }
   if (weight_ptr_ != nullptr) {
     free(weight_ptr_);
     weight_ptr_ = nullptr;
@@ -64,15 +60,15 @@ void Convolution1x1CPUKernel::InitConv1x1MatmulParam() {
 
 int Convolution1x1CPUKernel::InitConv1x1BiasWeight() {
   if (inputs_.size() == 3) {
-    bias_ptr_ = reinterpret_cast<float *>(malloc(matmul_param_->col_ * C4NUM * sizeof(float)));
-    if (bias_ptr_ == nullptr) {
+    bias_data_ = malloc(matmul_param_->col_ * C4NUM * sizeof(float));
+    if (bias_data_ == nullptr) {
       MS_LOG(ERROR) << "Conv1x1 Malloc bias_ptr_ error!";
       return RET_ERROR;
     }
-    memset(bias_ptr_, 0, matmul_param_->col_ * C4NUM * sizeof(float));
-    memcpy(bias_ptr_, inputs_[2]->Data(), conv_param_->output_channel_ * sizeof(float));
+    memset(bias_data_, 0, matmul_param_->col_ * C4NUM * sizeof(float));
+    memcpy(bias_data_, inputs_[2]->Data(), conv_param_->output_channel_ * sizeof(float));
   } else {
-    bias_ptr_ = nullptr;
+    bias_data_ = nullptr;
   }
 
   weight_ptr_ = reinterpret_cast<float *>(
@@ -109,15 +105,15 @@ int Convolution1x1CPUKernel::InitConv1x1Param() {
     MS_LOG(ERROR) << "Conv1x1 Malloc tmp_ptr_ error!";
     return RET_MEMORY_FAILED;
   }
-  c4_output_ = reinterpret_cast<float *>(malloc(outputs_[0]->ElementsC4Num() / conv_param_->output_batch_ *
-          sizeof(float)));
+  c4_output_ =
+    reinterpret_cast<float *>(malloc(outputs_[0]->ElementsC4Num() / conv_param_->output_batch_ * sizeof(float)));
   if (c4_output_ == nullptr) {
     MS_LOG(ERROR) << "Conv1x1 Malloc c4_output_ error!";
     return RET_MEMORY_FAILED;
   }
 
-  c4_input_ = reinterpret_cast<float *>(malloc(inputs_[0]->ElementsC4Num() / conv_param_->input_batch_ *
-          sizeof(float)));
+  c4_input_ =
+    reinterpret_cast<float *>(malloc(inputs_[0]->ElementsC4Num() / conv_param_->input_batch_ * sizeof(float)));
   if (c4_input_ == nullptr) {
     MS_LOG(ERROR) << "Conv1x1 Malloc c4_input_ error!";
     return RET_MEMORY_FAILED;
@@ -189,9 +185,12 @@ int Convolution1x1CPUKernel::DoPostFunc(int task_id) {
     return RET_OK;
   }
 
+  float *cur_bias =
+    (bias_data_ == nullptr) ? nullptr : reinterpret_cast<float *>(bias_data_) + task_id * thread_oc_stride_;
+
   PostConvFuncFp32(c4_output_ + matmul_param_->row_ * thread_oc_stride_ * task_id,
-                   output_ptr_ + task_id * thread_oc_stride_, bias_ptr_ + task_id * thread_oc_stride_, cur_oc,
-                   matmul_param_->row_, conv_param_->output_channel_, conv_param_->is_relu_, conv_param_->is_relu6_);
+                   output_ptr_ + task_id * thread_oc_stride_, cur_bias, cur_oc, matmul_param_->row_,
+                   conv_param_->output_channel_, conv_param_->is_relu_, conv_param_->is_relu6_);
   return RET_OK;
 }
 
@@ -228,4 +227,3 @@ int Convolution1x1CPUKernel::Run() {
   return RET_OK;
 }
 }  // namespace mindspore::kernel
-

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_1x1.h

@@ -56,7 +56,6 @@ class Convolution1x1CPUKernel : public ConvolutionBaseCPUKernel {
   int thread_hw_stride_ = 0;
   int thread_oc4_count_ = 0;
   int thread_oc_stride_ = 0;
-  float *bias_ptr_ = nullptr;
   float *weight_ptr_ = nullptr;
   float *tmp_ptr_ = nullptr;
   float *c4_input_ = nullptr;

mindspore/lite/src/runtime/kernel/arm/fp32/deconvolution.cc

@@ -169,10 +169,12 @@ int DeConvolutionCPUKernel::DoPostFunc(int task_id) {
     return RET_OK;
   }
 
+  float *cur_bias =
+    (bias_data_ == nullptr) ? nullptr : reinterpret_cast<float *>(bias_data_) + thread_co_stride_ * task_id;
+
   DeConvPostFp32(tmp_output_ + thread_co_stride_ * task_id * input_plane * kernel_plane,
                  c4_output_ + thread_co_stride_ * task_id * output_plane, output_ptr_ + thread_co_stride_ * task_id,
-                 reinterpret_cast<float *>(bias_data_) + thread_co_stride_ * task_id, cur_oc, input_plane, kernel_plane,
-                 output_plane, conv_param_);
+                 cur_bias, cur_oc, input_plane, kernel_plane, output_plane, conv_param_);
   return RET_OK;
 }
 
@@ -224,4 +226,3 @@ int DeConvolutionCPUKernel::Run() {
   return RET_OK;
 }
 }  // namespace mindspore::kernel
-

mindspore/lite/src/runtime/kernel/arm/fp32/pad.cc

@@ -37,7 +37,7 @@ constexpr int kInputRank = 4;
 constexpr int kPaddingsSize = 8;
 }  // namespace
 
-int PadCPUKernel::CheckInputsOutputsParams() {
+int PadCPUKernel::Init() {
   if (inputs_.size() != kInputNum || outputs_.size() != kOutputNum) {
     MS_LOG(ERROR) << "Pad input size should be " << kInputNum << ", got " << inputs_.size() << ", output size should be"
                   << kOutputNum << ", got " << outputs_.size();
@@ -71,42 +71,6 @@ int PadCPUKernel::CheckInputsOutputsParams() {
   return RET_OK;
 }
 
-int PadCPUKernel::MaybeConvertInputLayout() {
-  auto input = inputs_.at(0);
-  auto input_format = input->GetFormat();
-  if (input_format != exec_format_) {
-    auto input_type = input->data_type();
-    layout_convertor_ = LayoutTransform(input_type, input_format, exec_format_);
-    if (layout_convertor_ == nullptr) {
-      MS_LOG(ERROR) << "Pad lack layout convertor from " << input_format << " to " << exec_format_;
-      return RET_ERROR;
-    }
-    exec_input_data_ = reinterpret_cast<float *>(malloc(input->DataSize() * sizeof(float)));
-    if (exec_input_data_ == nullptr) {
-      MS_LOG(ERROR) << "Pad malloc failed.";
-      return RET_ERROR;
-    }
-  }
-  return RET_OK;
-}
-
-int PadCPUKernel::Init() {
-  auto ret = CheckInputsOutputsParams();
-  if (ret != RET_OK) {
-    return ret;
-  }
-
-  ret = MaybeConvertInputLayout();
-  if (ret != RET_OK) {
-    return ret;
-  }
-
-  auto output = outputs_.at(0);
-  output->SetFormat(exec_format_);
-
-  return RET_OK;
-}
-
 int PadImpl(int task_id, LiteParallelGroupEnv *penv, void *cdata) {
   auto padKernel = reinterpret_cast<PadCPUKernel *>(cdata);
   int error_code = padKernel->RunImpl(task_id);
@@ -125,11 +89,8 @@ int PadCPUKernel::RunImpl(int task_id) {
   auto output_data = reinterpret_cast<float *>(output->Data());
   auto input_shape = input->shape().data();
   auto output_shape = output->shape().data();
-  if (exec_input_data_ != nullptr) {
-    Pad(exec_input_data_, output_data, input_shape, output_shape, paddings_.data(), task_id, context_->threadNum);
-  } else {
-    Pad(input_data, output_data, input_shape, output_shape, paddings_.data(), task_id, context_->threadNum);
-  }
+
+  Pad(input_data, output_data, input_shape, output_shape, paddings_.data(), task_id, context_->threadNum);
 
   return RET_OK;
 }
@@ -142,15 +103,6 @@ int PadCPUKernel::Run() {
   // todo parallel memset to save time
   memset(output_data, 0, output_size * sizeof(float));
 
-  auto input = inputs_.at(0);
-  if (exec_input_data_ != nullptr) {
-    if (layout_convertor_ == nullptr) {
-      return RET_NULL_PTR;
-    }
-    layout_convertor_(inputs_.at(0), exec_input_data_, input->Batch(), input->Height() * input->Width(),
-                      input->Channel());
-  }
-
   int error_code = LiteBackendParallelLaunch(PadImpl, this, context_->threadNum);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Pad run error, error_code[" << error_code << "]";
@@ -158,30 +110,4 @@ int PadCPUKernel::Run() {
   }
   return RET_OK;
 }
-
-kernel::LiteKernel *CpuPadFp32KernelCreator(const std::vector<lite::tensor::Tensor *> &inputs,
-                                            const std::vector<lite::tensor::Tensor *> &outputs,
-                                            OpParameter *opParameter, const lite::Context *ctx,
-                                            const kernel::KernelKey &desc) {
-  if (opParameter == nullptr) {
-    MS_LOG(ERROR) << "Pad opParameter nullptr";
-    return nullptr;
-  }
-  MS_ASSERT(desc.type == PrimitiveType_Pad);
-  auto *kernel = new (std::nothrow) PadCPUKernel(opParameter, inputs, outputs, ctx);
-  if (kernel == nullptr) {
-    MS_LOG(ERROR) << "new PadCPUKernel failed.";
-    return nullptr;
-  }
-  auto ret = kernel->Init();
-  if (ret != RET_OK) {
-    MS_LOG(ERROR) << "Init kernel failed, name: " << opParameter->name_ << ", type: "
-                  << schema::EnumNamePrimitiveType(static_cast<schema::PrimitiveType>(opParameter->type_));
-    return nullptr;
-  }
-  return kernel;
-}
-
-REG_KERNEL(kCPU, PrimitiveType_Pad, CpuPadFp32KernelCreator)
 }  // namespace mindspore::kernel
-

mindspore/lite/src/runtime/kernel/arm/fp32/pad.h

@@ -19,7 +19,7 @@
 #include <vector>
 #include "src/lite_kernel.h"
 
-#include "src/runtime/kernel/arm/opclib/pad.h"
+#include "src/runtime/kernel/arm/opclib/fp32/pad.h"
 #include "src/runtime/kernel/arm/base/layout_transform.h"
 
 namespace mindspore::kernel {
@@ -29,31 +29,18 @@ class PadCPUKernel : public LiteKernel {
                const std::vector<lite::tensor::Tensor *> &outputs, const lite::Context *ctx)
       : LiteKernel(parameter, inputs, outputs), context_(ctx) {}
 
-  ~PadCPUKernel() {
-    if (exec_input_data_ != nullptr) {
-      free(exec_input_data_);
-      exec_input_data_ = nullptr;
-    }
-  }
+  ~PadCPUKernel() {}
 
   int Init() override;
   int ReSize() override { return 0; };
   int Run() override;
   int RunImpl(int task_id);
 
- private:
-  int CheckInputsOutputsParams();
-  int MaybeConvertInputLayout();
-
  private:
   std::vector<int> paddings_;
   size_t paddings_size_;
   const lite::Context *context_;
-  schema::Format exec_format_ = schema::Format_NHWC;
-  LayoutConvertor layout_convertor_ = nullptr;
-  float *exec_input_data_ = nullptr;
 };
 }  // namespace mindspore::kernel
 
 #endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_PAD_H_
-

mindspore/lite/src/runtime/kernel/arm/int8/pad_int8.cc

@@ -0,0 +1,119 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "src/runtime/kernel/arm/int8/pad_int8.h"
+
+using mindspore::lite::RET_ERROR;
+using mindspore::lite::RET_MEMORY_FAILED;
+using mindspore::lite::RET_OK;
+
+namespace mindspore::kernel {
+void PadInt8CPUKernel::FreeQuantParam() {
+  if (pad_param_->pad_quant_arg_.in_quant_args_ != nullptr) {
+    free(pad_param_->pad_quant_arg_.in_quant_args_);
+    pad_param_->pad_quant_arg_.in_quant_args_ = nullptr;
+  }
+  if (pad_param_->pad_quant_arg_.out_quanr_args_ != nullptr) {
+    free(pad_param_->pad_quant_arg_.out_quanr_args_);
+    pad_param_->pad_quant_arg_.out_quanr_args_ = nullptr;
+  }
+}
+
+int PadInt8CPUKernel::SetQuantParam() {
+  PadQuantArg *pad_quant_args = &pad_param_->pad_quant_arg_;
+  pad_quant_args->in_quant_args_ = reinterpret_cast<QuantArg *>(malloc(sizeof(QuantArg)));
+  if (pad_quant_args->in_quant_args_ == nullptr) {
+    return RET_MEMORY_FAILED;
+  }
+  pad_quant_args->out_quanr_args_ = reinterpret_cast<QuantArg *>(malloc(sizeof(QuantArg)));
+  if (pad_quant_args->out_quanr_args_ == nullptr) {
+    return RET_MEMORY_FAILED;
+  }
+  pad_quant_args->constant_value_ = reinterpret_cast<int8_t *>(malloc(sizeof(int8_t)));
+  if (pad_quant_args->constant_value_ == nullptr) {
+    return RET_MEMORY_FAILED;
+  }
+
+  auto *input_tensor = inputs_.at(kInputIndex);
+  auto *out_tensor = outputs_.at(kOutputIndex);
+  auto in_quant_arg = input_tensor->GetQuantParams();
+  auto out_quant_arg = out_tensor->GetQuantParams();
+
+  pad_quant_args->in_quant_args_->zp_ = in_quant_arg.front().zeroPoint;
+  pad_quant_args->in_quant_args_->scale_ = in_quant_arg.front().scale;
+  pad_quant_args->out_quanr_args_->zp_ = out_quant_arg.front().zeroPoint;
+  pad_quant_args->out_quanr_args_->scale_ = out_quant_arg.front().scale;
+
+  if (pad_quant_args->in_quant_args_->scale_ != pad_quant_args->out_quanr_args_->scale_ ||
+      pad_quant_args->in_quant_args_->zp_ != pad_quant_args->out_quanr_args_->zp_) {
+    MS_LOG(ERROR) << "Pad int8 op : scale & zp of output and input must be equal.";
+    return RET_ERROR;
+  }
+
+  pad_quant_args->constant_value_[0] = QuantizeToInt8(
+    pad_param_->constant_value_, pad_quant_args->in_quant_args_->scale_, pad_quant_args->in_quant_args_->zp_);
+  return RET_OK;
+}
+
+int PadInt8CPUKernel::InitPadParam() {
+  auto in_dims = inputs_[0]->shape();
+  auto out_dims = outputs_[0]->shape();
+  int ndims = in_dims.size();
+
+  int in[] = {1, 1, 1, 1};
+  int out[] = {1, 1, 1, 1};
+
+  for (int i = 0; i < ndims; i++) {
+    in[DEFAULT_PAD_NDIMS - ndims + i] = in_dims[i];
+    out[DEFAULT_PAD_NDIMS - ndims + i] = out_dims[i];
+  }
+
+  memcpy(in_dims_, in, DEFAULT_PAD_NDIMS * sizeof(int));
+  memcpy(out_dims_, out, DEFAULT_PAD_NDIMS * sizeof(int));
+
+  return RET_OK;
+}
+
+int PadInt8CPUKernel::ReSize() {
+  InitPadParam();
+  return RET_OK;
+}
+
+int PadInt8CPUKernel::Init() {
+  int error_code = InitPadParam();
+  if (error_code != RET_OK) {
+    MS_LOG(ERROR) << "InitPadParam failed. errorcode: " << error_code;
+    return error_code;
+  }
+
+  error_code = SetQuantParam();
+  if (error_code != RET_OK) {
+    MS_LOG(ERROR) << "SetQuantParam failed. errorcode: " << error_code;
+    return error_code;
+  }
+  return RET_OK;
+}
+
+int PadInt8CPUKernel::Run() {
+  int8_t *in_data = reinterpret_cast<int8_t *>(inputs_[0]->Data());
+  int8_t *out_data = reinterpret_cast<int8_t *>(outputs_[0]->Data());
+
+  memset(out_data, pad_param_->pad_quant_arg_.constant_value_[0], outputs_[0]->ElementsNum() * sizeof(int8_t));
+  PadConstant4D(in_data, out_data, in_dims_, out_dims_, pad_param_->paddings_);
+  return RET_OK;
+}
+
+}  // namespace mindspore::kernel

mindspore/lite/src/runtime/kernel/arm/int8/pad_int8.h

@@ -0,0 +1,52 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_PAD_INT8_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_PAD_INT8_H_
+
+#include <vector>
+#include "include/errorcode.h"
+#include "src/lite_kernel.h"
+#include "src/runtime/runtime_api.h"
+#include "src/runtime/kernel/arm/opclib/pad_parameter.h"
+#include "src/runtime/kernel/arm/opclib/int8/pad.h"
+
+namespace mindspore::kernel {
+class PadInt8CPUKernel : public LiteKernel {
+ public:
+  explicit PadInt8CPUKernel(OpParameter *parameter, const std::vector<lite::tensor::Tensor *> &inputs,
+                            const std::vector<lite::tensor::Tensor *> &outputs, const lite::Context *ctx)
+      : LiteKernel(parameter, inputs, outputs) {
+    opParameter->thread_num_ = ctx->threadNum;
+    pad_param_ = reinterpret_cast<PadParameter *>(opParameter);
+  }
+  ~PadInt8CPUKernel() override { FreeQuantParam(); };
+
+  int Init() override;
+  int ReSize() override;
+  int Run() override;
+
+ private:
+  int SetQuantParam();
+  int InitPadParam();
+  void FreeQuantParam();
+
+ private:
+  PadParameter *pad_param_;
+  int in_dims_[DEFAULT_PAD_NDIMS];
+  int out_dims_[DEFAULT_PAD_NDIMS];
+};
+}  // namespace mindspore::kernel
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_PAD_INT8_H_

mindspore/lite/src/runtime/kernel/arm/opclib/pad.cc → mindspore/lite/src/runtime/kernel/arm/opclib/fp32/pad.cc

@@ -13,9 +13,8 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#include "src/runtime/kernel/arm/opclib/pad.h"
-#include <float.h>
-#include "src/runtime/kernel/arm/opclib/offset_utils.h"
+
+#include "src/runtime/kernel/arm/opclib/fp32/pad.h"
 
 void Pad(const float *input_data, float *output_data, const int *input_shape, const int *output_shape,
          const int *paddings, const int tid, const int thread_num) {
@@ -34,4 +33,3 @@ void Pad(const float *input_data, float *output_data, const int *input_shape, co
     }
   }
 }
-

mindspore/lite/src/runtime/kernel/arm/opclib/pad.h → mindspore/lite/src/runtime/kernel/arm/opclib/fp32/pad.h

@@ -13,23 +13,19 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_PAD_H_
-#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_PAD_H_
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_FP32_PAD_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_FP32_PAD_H_
 
 #ifdef ENABLE_NEON
 #include <arm_neon.h>
 #endif
 #include <memory.h>
+#include <float.h>
+#include "src/runtime/kernel/arm/opclib/offset_utils.h"
 #include "src/runtime/kernel/arm/opclib/op_base.h"
-
-struct PadParameter {
-  OpParameter op_parameter_;
-  int paddings[8];
-  size_t ori_size_;
-};
+#include "src/runtime/kernel/arm/opclib/pad_parameter.h"
 
 void Pad(const float *input_data, float *output_data, const int *input_shape, const int *output_shape,
          const int *paddings, const int tid, const int thread_num);
 
-#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_PAD_H_
-
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_FP32_PAD_H_

mindspore/lite/src/runtime/kernel/arm/opclib/int8/pad.cc

@@ -0,0 +1,32 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "src/runtime/kernel/arm/opclib/int8/pad.h"
+
+void PadConstant4D(const int8_t *in_data, int8_t *out_data, const int32_t *in_dims, const int32_t *out_dims,
+                   const int32_t *paddings) {
+  int32_t copy_size = in_dims[3];
+  for (int n = 0; n < in_dims[0]; n++) {
+    for (int h = 0; h < in_dims[1]; h++) {
+      for (int w = 0; w < in_dims[2]; w++) {
+        const int8_t *in = in_data + offset(in_dims, n, h, w, 0);
+        int8_t *out = out_data + offset(out_dims, n + paddings[0], h + paddings[2], w + paddings[4], paddings[6]);
+        memcpy(out, in, copy_size * sizeof(int8_t));
+      }
+    }
+  }
+  return;
+}

mindspore/lite/src/runtime/kernel/arm/opclib/int8/pad.h

@@ -0,0 +1,28 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_INT8_PAD_INT8_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_INT8_PAD_INT8_H_
+
+#include <string.h>
+#include "src/runtime/kernel/arm/opclib/op_base.h"
+#include "src/runtime/kernel/arm/opclib/offset_utils.h"
+#include "src/runtime/kernel/arm/opclib/pad_parameter.h"
+
+void PadConstant4D(const int8_t *in_data, int8_t *out_data, const int32_t *in_dims, const int32_t *out_dims,
+                   const int32_t *paddings);
+
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_INT8_PAD_INT8_H_

mindspore/lite/src/runtime/kernel/arm/opclib/offset_utils.h

@@ -25,6 +25,10 @@ inline int offset(const int *shape, const int dim0, const int dim1, const int di
   return ((dim0 * shape[1] + dim1) * shape[2] + dim2) * shape[3] + dim3;
 }
 
+inline int offsetComm(const int *shape, const int dim0, const int dim1, const int dim2) {
+  return ((dim0 * shape[1] + dim1) * shape[2] + dim2) * shape[3];
+}
+
 inline int offset4d(const int *shape, const int *dims) { return offset(shape, dims[0], dims[1], dims[2], dims[3]); }
 #endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_OFFSET_UTILS_H_
 

mindspore/lite/src/runtime/kernel/arm/opclib/pad_parameter.h

@@ -0,0 +1,32 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_PAD_PARAMETER_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_PAD_PARAMETER_H_
+
+#include "src/runtime/kernel/arm/opclib/op_base.h"
+
+#define MAX_PAD_SIZE 8
+#define DEFAULT_PAD_NDIMS 4
+
+struct PadParameter {
+  OpParameter op_parameter_;
+  PadQuantArg pad_quant_arg_;
+  int paddings_[MAX_PAD_SIZE] = {0};
+  int pad_mode_;
+  float constant_value_;
+};
+
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_OPCLIB_PAD_PARAMETER_H_

mindspore/lite/src/runtime/kernel/arm/opclib/quantization/quantize.h

@@ -58,6 +58,12 @@ struct FcQuantArg {
   int32_t quant_multiplier;
 };
 
+struct PadQuantArg {
+  QuantArg *in_quant_args_ = nullptr;
+  QuantArg *out_quanr_args_ = nullptr;
+  int8_t *constant_value_ = nullptr;
+};
+
 struct MulQuantArg {
   QuantArg in_quant_args_[2];
   QuantArg out_quant_arg_;