!15176 [MS][LITE]Change lite_kernel.h interface of InnerContext

From: @gongdaguo
Reviewed-by: @jpc_chenjianping,@zhanghaibo5
Signed-off-by: @jpc_chenjianping

mindspore/lite/schema/ops.fbs

@@ -841,7 +841,7 @@ table Rsqrt {
 }
 
 table QuantDTypeCast {
-    src_t: long; // deprecated
+    src_t: long;
     dst_t: long;
 }
 

mindspore/lite/src/lite_kernel.h

@@ -30,6 +30,7 @@
 #include "src/tensor.h"
 #include "include/errorcode.h"
 #include "schema/model_generated.h"
+#include "include/context.h"
 
 namespace mindspore::kernel {
 enum KERNEL_ARCH {
@@ -64,7 +65,7 @@ class LiteKernel {
  public:
   LiteKernel() = default;
   LiteKernel(OpParameter *parameter, std::vector<lite::Tensor *> in_tensors, std::vector<lite::Tensor *> out_tensors,
-             const lite::InnerContext *ctx)
+             const lite::Context *ctx)
       : op_parameter_(parameter),
         in_tensors_(std::move(in_tensors)),
         out_tensors_(std::move(out_tensors)),
@@ -175,7 +176,7 @@ class LiteKernel {
 
   SubGraphType subgraph_type() const { return this->subgraph_type_; }
 
-  const lite::InnerContext *context() const { return this->context_; }
+  const lite::Context *context() const { return this->context_; }
 
   virtual std::string ToString() const;
 
@@ -202,7 +203,7 @@ class LiteKernel {
   // tensor will free in ~lite_session()
   std::vector<lite::Tensor *> in_tensors_;
   std::vector<lite::Tensor *> out_tensors_;
-  const lite::InnerContext *context_ = nullptr;
+  const lite::Context *context_ = nullptr;
   std::vector<LiteKernel *> in_kernels_;
   std::vector<LiteKernel *> out_kernels_;
   bool train_mode_ = false;
@@ -217,13 +218,13 @@ class LiteKernel {
 
 typedef LiteKernel *(*KernelCreator)(const std::vector<lite::Tensor *> &inputs,
                                      const std::vector<lite::Tensor *> &outputs, OpParameter *parameter,
-                                     const lite::InnerContext *ctx, const KernelKey &desc);
+                                     const lite::Context *ctx, const KernelKey &desc);
 
 template <class T>
 kernel::LiteKernel *LiteKernelCreator(const std::vector<lite::Tensor *> &inputs,
                                       const std::vector<lite::Tensor *> &outputs, OpParameter *parameter,
-                                      const lite::InnerContext *ctx, const kernel::KernelKey &desc) {
-  auto *kernel = new (std::nothrow) T(parameter, inputs, outputs, ctx);
+                                      const lite::Context *ctx, const kernel::KernelKey &desc) {
+  auto *kernel = new (std::nothrow) T(parameter, inputs, outputs, static_cast<const lite::InnerContext *>(ctx));
   if (kernel == nullptr) {
     MS_LOG(ERROR) << "kernel: " << parameter->name_ << "is nullptr.";
     free(parameter);

mindspore/lite/src/runtime/agent/npu/subgraph_npu_kernel.cc

@@ -206,7 +206,8 @@ int SubGraphNpuKernel::Init() {
 
     MS_ASSERT(npu_manager_ != nullptr);
 
-    npu_manager_->AddModel(model_buffer_data, GetOMModelName(), context_->GetNpuInfo().frequency_);
+    npu_manager_->AddModel(model_buffer_data, GetOMModelName(),
+                           static_cast<const lite::InnerContext *>(context_)->GetNpuInfo().frequency_);
 
     executor_ = new (std::nothrow) mindspore::lite::NPUExecutor(GetOMModelName(), npu_manager_);
 

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

@@ -73,7 +73,8 @@ int ConstantOfShapeCPUKernel::Run() {
   int thread_count = MSMIN(op_parameter_->thread_num_, param_->element_size_);
   thread_stride_ = UP_DIV(param_->element_size_, thread_count);
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ConstantOfShapeRun, this, thread_count);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConstantOfShapeRun,
+                            this, thread_count);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ConstantOfShapeRun error error_code[" << ret << "]";
     return ret;

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

@@ -144,17 +144,7 @@ void DetectionPostProcessBaseCPUKernel::FreeAllocatedBuffer() {
   }
 }
 
-int DetectionPostProcessBaseCPUKernel::Run() {
-  MS_ASSERT(context_->allocator != nullptr);
-  int status = GetInputData();
-  if (status != RET_OK) {
-    return status;
-  }
-  auto output_boxes = reinterpret_cast<float *>(out_tensors_.at(0)->data_c());
-  auto output_classes = reinterpret_cast<float *>(out_tensors_.at(1)->data_c());
-  auto output_scores = reinterpret_cast<float *>(out_tensors_.at(2)->data_c());
-  auto output_num = reinterpret_cast<float *>(out_tensors_.at(3)->data_c());
-
+int DetectionPostProcessBaseCPUKernel::ParamInit() {
   num_boxes_ = in_tensors_.at(0)->shape().at(1);
   num_classes_with_bg_ = in_tensors_.at(1)->shape().at(2);
   params_->decoded_boxes_ = context_->allocator->Malloc(num_boxes_ * 4 * sizeof(float));
@@ -221,6 +211,24 @@ int DetectionPostProcessBaseCPUKernel::Run() {
       return RET_ERROR;
     }
   }
+  return RET_OK;
+}
+
+int DetectionPostProcessBaseCPUKernel::Run() {
+  MS_ASSERT(context_->allocator != nullptr);
+  int status = GetInputData();
+  if (status != RET_OK) {
+    return status;
+  }
+  auto output_boxes = reinterpret_cast<float *>(out_tensors_.at(0)->data_c());
+  auto output_classes = reinterpret_cast<float *>(out_tensors_.at(1)->data_c());
+  auto output_scores = reinterpret_cast<float *>(out_tensors_.at(2)->data_c());
+  auto output_num = reinterpret_cast<float *>(out_tensors_.at(3)->data_c());
+
+  if (ParamInit() != RET_OK) {
+    MS_LOG(ERROR) << "ParamInit error";
+    return status;
+  }
 
   status = DecodeBoxes(num_boxes_, input_boxes_, params_->anchors_, params_);
   if (status != RET_OK) {
@@ -238,7 +246,8 @@ int DetectionPostProcessBaseCPUKernel::Run() {
       return status;
     }
   } else {
-    status = ParallelLaunch(this->context_->thread_pool_, NmsMultiClassesFastCoreRun, this, op_parameter_->thread_num_);
+    status = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                            NmsMultiClassesFastCoreRun, this, op_parameter_->thread_num_);
     if (status != RET_OK) {
       MS_LOG(ERROR) << "NmsMultiClassesFastCoreRun error error_code[" << status << "]";
       FreeAllocatedBuffer();

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

@@ -47,6 +47,7 @@ class DetectionPostProcessBaseCPUKernel : public LiteKernel {
 
  protected:
   virtual int GetInputData() = 0;
+  int ParamInit();
 
  private:
   void FreeAllocatedBuffer();

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

@@ -166,7 +166,8 @@ int RunPriorBox(void *cdata, int task_id) {
 }
 
 int PriorBoxCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, RunPriorBox, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, RunPriorBox,
+                                  this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "PriorBox run error, error_code[" << error_code << "]";
     return RET_ERROR;

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

@@ -172,7 +172,8 @@ int QuantDTypeCastCPUKernel::Run() {
     uint8_ptr_ = reinterpret_cast<uint8_t *>(out_tensors_[0]->data_c());
   }
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, QuantDTypeCastRun, this, thread_n_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, QuantDTypeCastRun,
+                            this, thread_n_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Scale error error_code[" << ret << "]";
     if (in_tensors_[0]->data_type() == TypeId::kNumberTypeInt8 &&

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

@@ -66,7 +66,8 @@ int ReshapeRun(void *cdata, int task_id) {
 int ReshapeBaseCPUKernel::Run() {
   input_ptr_ = reinterpret_cast<uint8_t *>(in_tensors_.at(kInputIndex)->data_c());
   output_ptr_ = reinterpret_cast<uint8_t *>(out_tensors_.at(kOutputIndex)->data_c());
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ReshapeRun, this, context_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ReshapeRun, this,
+                            context_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Reshape run error error_code[" << ret << "]";
     return ret;

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

@@ -81,7 +81,8 @@ int SliceCPUKernel::Run() {
                       lite::DataTypeSize(in_tensors_.at(0)->data_type()));
     return RET_OK;
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, SliceLaunch, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SliceLaunch, this,
+                            op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "slice launch fail!ret: " << ret;
     return RET_ERROR;

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

@@ -120,7 +120,8 @@ int SplitBaseCPUKernel::Run() {
     output_ptr_.at(i) = output_tensor->data_c();
   }
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, SplitRun, this, thread_n_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SplitRun, this,
+                            thread_n_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "split error error_code[" << ret << "]";
   }

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

@@ -100,7 +100,8 @@ int StackBaseCPUKernel::Run() {
   }
   // run stack
   num_threads_ = MSMIN(UP_DIV(outer_size_, 64), this->context_->thread_num_);
-  auto ret = ParallelLaunch(this->context_->thread_pool_, StackRun, this, num_threads_);
+  auto ret =
+    ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, StackRun, this, num_threads_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "StackBaseCPUKernel Run error: error_code[" << ret << "]";
     return RET_ERROR;

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

@@ -157,7 +157,8 @@ int StridedSliceCPUKernel::FastRun() {
   }
   input_ptr_ = reinterpret_cast<uint8_t *>(in_tensors_.front()->data_c());
   output_ptr_ = reinterpret_cast<uint8_t *>(out_tensors_.front()->data_c());
-  auto ret = ParallelLaunch(this->context_->thread_pool_, StrideRun, this, context_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, StrideRun, this,
+                            context_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Stride run error error_code[" << ret << "]";
     return ret;

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

@@ -127,7 +127,8 @@ int TileCPUKernel::SimpleTileImpl(int task_id) {
 }
 
 int TileCPUKernel::RunSimpleTile() {
-  auto ret = ParallelLaunch(context_->thread_pool_, SimpleTile, this, context_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SimpleTile, this,
+                            context_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "RunSimpleTile error code[" << ret << "]";
     return ret;

mindspore/lite/src/runtime/kernel/arm/fp16/activation_fp16.cc

@@ -100,7 +100,8 @@ int ActivationFp16CPUKernel::Run() {
   fp16_input_ = reinterpret_cast<float16_t *>(input_tensor->data_c());
   fp16_output_ = reinterpret_cast<float16_t *>(output_tensor->data_c());
 
-  int error_code = ParallelLaunch(this->context_->thread_pool_, ActivationFp16Run, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                  ActivationFp16Run, this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Activation function error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp16/arithmetic_compare_fp16.cc

@@ -160,15 +160,16 @@ int ArithmeticCompareFP16CPUKernel::Run() {
   is_input0_fp32_ = in_tensors_.at(0)->data_type() == kNumberTypeFloat32;
   is_input1_fp32_ = in_tensors_.at(1)->data_type() == kNumberTypeFloat32;
 
-  input0_fp16_ = ConvertInputFp32toFp16(in_tensors_.at(0), context_);
-  input1_fp16_ = ConvertInputFp32toFp16(in_tensors_.at(1), context_);
+  input0_fp16_ = ConvertInputFp32toFp16(in_tensors_.at(0), static_cast<const lite::InnerContext *>(this->context_));
+  input1_fp16_ = ConvertInputFp32toFp16(in_tensors_.at(1), static_cast<const lite::InnerContext *>(this->context_));
   output_fp16_ = reinterpret_cast<uint8_t *>(output_tensor->MutableData());
   if (input0_fp16_ == nullptr || input1_fp16_ == nullptr || output_fp16_ == nullptr) {
     MS_LOG(ERROR) << "Memory allocation failed";
     FreeTmpBuffer();
     return RET_ERROR;
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ArithmeticsRunFp16, this, context_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ArithmeticsRunFp16,
+                            this, context_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ArithmeticsRunFp16 run error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/arithmetic_fp16.cc

@@ -127,13 +127,13 @@ void ArithmeticFP16CPUKernel::InitRunFunction(int primitive_type) {
 int ArithmeticFP16CPUKernel::ConstTensorBroadCast() {
   int ret;
   if (in_tensors_[0]->data_c() != nullptr) {
-    ret = ConvertFp32TensorToFp16(in_tensors_[0], context_);
+    ret = ConvertFp32TensorToFp16(in_tensors_[0], static_cast<const lite::InnerContext *>(this->context_));
     if (ret != RET_OK) {
       return ret;
     }
   }
   if (in_tensors_[1]->data_c() != nullptr) {
-    ret = ConvertFp32TensorToFp16(in_tensors_[1], context_);
+    ret = ConvertFp32TensorToFp16(in_tensors_[1], static_cast<const lite::InnerContext *>(this->context_));
     if (ret != RET_OK) {
       return ret;
     }
@@ -167,18 +167,19 @@ int ArithmeticFP16CPUKernel::Run() {
     return RET_ERROR;
   }
   if (!input0_broadcast_) {
-    input0_ptr_ = ConvertInputFp32toFp16(in_tensors_.at(0), context_);
+    input0_ptr_ = ConvertInputFp32toFp16(in_tensors_.at(0), static_cast<const lite::InnerContext *>(this->context_));
   }
   if (!input1_broadcast_) {
-    input1_ptr_ = ConvertInputFp32toFp16(in_tensors_.at(1), context_);
+    input1_ptr_ = ConvertInputFp32toFp16(in_tensors_.at(1), static_cast<const lite::InnerContext *>(this->context_));
   }
   auto output_tensor = out_tensors_.at(0);
-  output_ptr_ = MallocOutputFp16(output_tensor, context_);
+  output_ptr_ = MallocOutputFp16(output_tensor, static_cast<const lite::InnerContext *>(this->context_));
   if (input0_ptr_ == nullptr || input1_ptr_ == nullptr || output_ptr_ == nullptr) {
     FreeFp16Buffer();
     return RET_ERROR;
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ArithmeticsRun, this, context_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ArithmeticsRun, this,
+                            context_->thread_num_);
   if (out_tensors_.at(0)->data_type() == kNumberTypeFloat32) {
     Float16ToFloat32(static_cast<float16_t *>(output_ptr_), reinterpret_cast<float *>(output_tensor->MutableData()),
                      output_tensor->ElementsNum());

mindspore/lite/src/runtime/kernel/arm/fp16/arithmetic_self_fp16.cc

@@ -77,13 +77,14 @@ int ArithmeticSelfFp16CPUKernel::Run() {
   auto output_tensor = out_tensors_.at(0);
 
   if (input_tensor->data_type() == kNumberTypeFloat32) {
-    input_fp16_ptr_ = ConvertInputFp32toFp16(input_tensor, context_);
+    input_fp16_ptr_ = ConvertInputFp32toFp16(input_tensor, static_cast<const lite::InnerContext *>(this->context_));
   } else {
     input_fp16_ptr_ = reinterpret_cast<float16_t *>(input_tensor->data_c());
   }
   output_fp16_ptr_ = reinterpret_cast<float16_t *>(output_tensor->data_c());
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ArithmeticSelfRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ArithmeticSelfRun,
+                            this, op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ArithmeticSelfRun error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/batchnorm_fp16.cc

@@ -51,15 +51,16 @@ int BatchnormFp16CPUKernel::InitConstTensor() {
 int BatchnormFp16CPUKernel::Run() {
   auto input_tensor = in_tensors_.at(0);
   auto output_tensor = out_tensors_.at(0);
-  input_ = ConvertInputFp32toFp16(input_tensor, context_);
-  output_ = MallocOutputFp16(output_tensor, context_);
+  input_ = ConvertInputFp32toFp16(input_tensor, static_cast<const lite::InnerContext *>(this->context_));
+  output_ = MallocOutputFp16(output_tensor, static_cast<const lite::InnerContext *>(this->context_));
   if (input_ == nullptr || output_ == nullptr) {
     FreeInputAndOutput();
     MS_LOG(ERROR) << "input or output is nullptr";
     return RET_ERROR;
   }
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, BatchNormRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, BatchNormRun, this,
+                            op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "BatchnormRun error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/cast_fp16.cc

@@ -132,7 +132,8 @@ int CastFp16CPUKernel::Run() {
   if (data_num_ == 0) {
     return RET_OK;
   }
-  return ParallelLaunch(this->context_->thread_pool_, CastFp16Run, this, op_parameter_->thread_num_);
+  return ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, CastFp16Run, this,
+                        op_parameter_->thread_num_);
 }
 
 REG_KERNEL(kCPU, kNumberTypeFloat16, PrimitiveType_Cast, LiteKernelCreator<CastFp16CPUKernel>)

mindspore/lite/src/runtime/kernel/arm/fp16/convolution_1x1_fp16.cc

@@ -236,14 +236,16 @@ int Convolution1x1FP16CPUKernel::Run() {
 
     int ret = RET_ERROR;
     if (multi_thread_by_hw_) {
-      ret = ParallelLaunch(this->context_->thread_pool_, Convolution1x1Fp16RunHw, this, thread_count_);
+      ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                           Convolution1x1Fp16RunHw, this, thread_count_);
     } else {
 #ifdef ENABLE_ARM64
       RowMajor2Col16MajorFp16Opt(input_ptr_, pack_input_, matmul_param_->row_, matmul_param_->deep_);
 #else
       RowMajor2Col12MajorFp16Opt(input_ptr_, pack_input_, matmul_param_->row_, matmul_param_->deep_);
 #endif
-      ret = ParallelLaunch(this->context_->thread_pool_, Convolution1x1Fp16RunOc, this, thread_count_);
+      ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                           Convolution1x1Fp16RunOc, this, thread_count_);
     }
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "ParallelLaunch failed.";

mindspore/lite/src/runtime/kernel/arm/fp16/convolution_delegate_fp16.cc

@@ -95,10 +95,11 @@ static void SetInputOutputShapeInfo(ConvParameter *conv_param, lite::Tensor *inp
 int ConvolutionDelegateFP16CPUKernel::ReSize() {
   // Update shape info of input and output
   kernel::SetInputOutputShapeInfo(reinterpret_cast<ConvParameter *>(op_parameter_), in_tensors_.front(),
-                                  out_tensors_.front(), context_);
+                                  out_tensors_.front(), static_cast<const lite::InnerContext *>(this->context_));
   if (fp16_conv_kernel_ == nullptr) {
     fp16_conv_kernel_ =
-      CpuConvFp16KernelSelect(in_tensors_, out_tensors_, op_parameter_, context_, origin_weight_, origin_bias_);
+      CpuConvFp16KernelSelect(in_tensors_, out_tensors_, op_parameter_,
+                              static_cast<const lite::InnerContext *>(context_), origin_weight_, origin_bias_);
     if (fp16_conv_kernel_ == nullptr) {
       MS_LOG(ERROR) << "Selecting execute kernel failed for conv_kernel, got a nullptr.";
       return RET_ERROR;
@@ -184,7 +185,7 @@ kernel::LiteKernel *CpuGroupConvFp16KernelCreator(const std::vector<lite::Tensor
 /* creator func */
 kernel::LiteKernel *CpuConvFp16KernelCreator(const std::vector<lite::Tensor *> &inputs,
                                              const std::vector<lite::Tensor *> &outputs, OpParameter *opParameter,
-                                             const InnerContext *ctx, const kernel::KernelKey &desc) {
+                                             const lite::Context *ctx, const kernel::KernelKey &desc) {
   MS_ASSERT(opParameter != nullptr);
   MS_ASSERT(desc.type == schema::PrimitiveType_Conv2DFusion);
 
@@ -200,11 +201,12 @@ kernel::LiteKernel *CpuConvFp16KernelCreator(const std::vector<lite::Tensor *> &
   auto conv_param = reinterpret_cast<ConvParameter *>(opParameter);
   kernel::LiteKernel *kernel = nullptr;
   if (conv_param->group_ == 1) {
-    kernel = new (std::nothrow) kernel::ConvolutionDelegateFP16CPUKernel(opParameter, inputs, outputs, ctx);
+    kernel = new (std::nothrow) kernel::ConvolutionDelegateFP16CPUKernel(opParameter, inputs, outputs,
+                                                                         static_cast<const lite::InnerContext *>(ctx));
   } else if (conv_param->group_ == conv_param->input_channel_ && conv_param->group_ == conv_param->output_channel_) {
-    kernel = CpuConvDwFp16KernelCreator(inputs, outputs, opParameter, ctx);
+    kernel = CpuConvDwFp16KernelCreator(inputs, outputs, opParameter, static_cast<const lite::InnerContext *>(ctx));
   } else {
-    kernel = CpuGroupConvFp16KernelCreator(inputs, outputs, opParameter, ctx);
+    kernel = CpuGroupConvFp16KernelCreator(inputs, outputs, opParameter, static_cast<const lite::InnerContext *>(ctx));
   }
 
   if (kernel == nullptr) {

mindspore/lite/src/runtime/kernel/arm/fp16/convolution_depthwise_fp16.cc

@@ -104,7 +104,8 @@ static int ConvDwFp16Run(void *cdata, int task_id) {
 }
 
 int ConvolutionDepthwiseFp16CPUKernel::Run() {
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ConvDwFp16Run, this, conv_param_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvDwFp16Run, this,
+                            conv_param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ConvDwFp16Run error: error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/convolution_depthwise_slidewindow_fp16.cc

@@ -155,7 +155,8 @@ int ConvolutionDepthwiseSWFp16CPUKernel::Run() {
     packed_output_ = output_ptr;
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, ConvDwSWFp16Run, this, conv_param_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvDwSWFp16Run, this,
+                       conv_param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ConvDwSWFp16Run error: error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/convolution_fp16.cc

@@ -144,7 +144,8 @@ int ConvolutionFP16CPUKernel::Run() {
     return RET_ERROR;
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, ConvolutionFp16Impl, this, thread_count_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvolutionFp16Impl, this,
+                       thread_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "conv fp16 error ret[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/convolution_winograd_fp16.cc

@@ -213,7 +213,8 @@ int ConvolutionWinogradFP16CPUKernel::Run() {
     return RET_ERROR;
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, ConvolutionWinogradFp16Impl, this, thread_count_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                       ConvolutionWinogradFp16Impl, this, thread_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "conv winograd error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/crop_fp16.cc

@@ -53,7 +53,8 @@ int CropFp16CPUKernel::Run() {
   input_ptr_ = reinterpret_cast<float16_t *>(input_tensor->data_c());
   output_ptr_ = reinterpret_cast<float16_t *>(output_tensor->data_c());
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, CropFp16Run, this, crop_para_->thread_count_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, CropFp16Run, this,
+                            crop_para_->thread_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ParallelLaunch failed: " << ret;
   }

mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_depthwise_fp16.cc

@@ -173,7 +173,8 @@ int DeconvolutionDepthwiseFp16CPUKernel::Run() {
     memset(output_ptr, 0, out_tensors_.at(kOutputIndex)->ElementsNum() * sizeof(float16_t));
     packed_output_ = output_ptr;
   }
-  ret = ParallelLaunch(this->context_->thread_pool_, DeconvDwFp16Run, this, conv_param_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, DeconvDwFp16Run, this,
+                       conv_param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "DeconvDwFp16Run error: error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc

@@ -217,7 +217,8 @@ int DeConvolutionFp16CPUKernel::Run() {
 
     RowMajor2Col16MajorFp16Opt(batch_input_, pack_input_, input_plane_, conv_param_->input_channel_);
 
-    error_code = ParallelLaunch(this->context_->thread_pool_, DeConvFp16Run, this, thread_count_);
+    error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, DeConvFp16Run,
+                                this, thread_count_);
     if (error_code != RET_OK) {
       MS_LOG(ERROR) << "deconv fp16 run error! error_code[" << error_code << "]";
     }
@@ -229,7 +230,7 @@ int DeConvolutionFp16CPUKernel::Run() {
 
 kernel::LiteKernel *CpuDeConvFp16KernelCreator(const std::vector<lite::Tensor *> &inputs,
                                                const std::vector<lite::Tensor *> &outputs, OpParameter *op_parameter,
-                                               const lite::InnerContext *ctx, const kernel::KernelKey &desc) {
+                                               const lite::Context *ctx, const kernel::KernelKey &desc) {
   MS_ASSERT(op_parameter != nullptr);
   MS_ASSERT(desc.type == schema::PrimitiveType_Conv2dTransposeFusion);
 
@@ -238,12 +239,15 @@ kernel::LiteKernel *CpuDeConvFp16KernelCreator(const std::vector<lite::Tensor *>
   if (conv_param->group_ == 1) {
     if ((conv_param->stride_h_ != 1 || conv_param->stride_w_ != 1) &&
         (conv_param->dilation_h_ == 1 && conv_param->dilation_w_ == 1)) {
-      kernel = new (std::nothrow) kernel::DeConvWinogradFp16CPUKernel(op_parameter, inputs, outputs, ctx);
+      kernel = new (std::nothrow) kernel::DeConvWinogradFp16CPUKernel(op_parameter, inputs, outputs,
+                                                                      static_cast<const lite::InnerContext *>(ctx));
     } else {
-      kernel = new (std::nothrow) kernel::DeConvolutionFp16CPUKernel(op_parameter, inputs, outputs, ctx);
+      kernel = new (std::nothrow)
+        kernel::DeConvolutionFp16CPUKernel(op_parameter, inputs, outputs, static_cast<const lite::InnerContext *>(ctx));
     }
   } else if (conv_param->group_ == conv_param->input_channel_ && conv_param->group_ == conv_param->output_channel_) {
-    kernel = new (std::nothrow) DeconvolutionDepthwiseFp16CPUKernel(op_parameter, inputs, outputs, ctx);
+    kernel = new (std::nothrow)
+      DeconvolutionDepthwiseFp16CPUKernel(op_parameter, inputs, outputs, static_cast<const lite::InnerContext *>(ctx));
   }
 
   if (kernel == nullptr) {

mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.cc

@@ -392,10 +392,12 @@ int DeConvWinogradFp16CPUKernel::Run() {
     nhwc_output_ = output_ptr + batch_index * deconv_param_->output_plane_ * conv_param_->output_channel_;
 
     ::memset(nc4hw4_output_, 0, deconv_param_->output_plane_ * deconv_param_->oc_div4_ * C4NUM * sizeof(float16_t));
-    ParallelLaunch(this->context_->thread_pool_, DeConvWgFp16Run, this, deconv_param_->thread_num_);
+    ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, DeConvWgFp16Run, this,
+                   deconv_param_->thread_num_);
 
     /*post bias activate and nhwc */
-    ParallelLaunch(this->context_->thread_pool_, DeConvWgPostFp16Run, this, thread_num_hw_);
+    ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, DeConvWgPostFp16Run, this,
+                   thread_num_hw_);
   }
 
   return RET_OK;

mindspore/lite/src/runtime/kernel/arm/fp16/gather_fp16.cc

@@ -151,7 +151,8 @@ int GatherFp16CPUKernel::Run() {
       Float32ToFloat16(reinterpret_cast<float *>(input_tensor->data_c()), input_data_, input_tensor->ElementsNum());
     }
   }
-  ret = ParallelLaunch(this->context_->thread_pool_, GatherRunFp16, this, op_parameter_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, GatherRunFp16, this,
+                       op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Gather function error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp16/instance_norm_fp16.cc

@@ -109,7 +109,8 @@ int InstanceNormFp16Run(void *cdata, int task_id) {
 int InstanceNormFp16CPUKernel::Run() {
   src_data_ = reinterpret_cast<float16_t *>(in_tensors_[0]->data_c());
   dst_data_ = reinterpret_cast<float16_t *>(out_tensors_[0]->data_c());
-  auto ret = ParallelLaunch(this->context_->thread_pool_, InstanceNormFp16Run, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, InstanceNormFp16Run,
+                            this, op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "InstanceNormFp16Run error error_code[" << ret << "]";
     return ret;

mindspore/lite/src/runtime/kernel/arm/fp16/log_softmax_fp16.cc

@@ -95,7 +95,8 @@ int LogSoftmaxLastAxisFp16Run(void *cdata, int task_id) {
 
 int LogSoftmaxFp16CPUKernel::Run() {
   if (in_plane_size_ == 1) {
-    auto ret = ParallelLaunch(this->context_->thread_pool_, LogSoftmaxLastAxisFp16Run, this, context_->thread_num_);
+    auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                              LogSoftmaxLastAxisFp16Run, this, context_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "LogSoftmaxFp16CPUKernel ParallelLaunch failed, ret: " << ret;
     }

mindspore/lite/src/runtime/kernel/arm/fp16/matmul_base_fp16.cc

@@ -286,7 +286,8 @@ int MatmulBaseFP16CPUKernel::Run() {
       batch_b_ptr_ = b_pack_ptr_ + i * params_->deep_ * params_->col_align_;
       batch_c_ptr_ = c_ptr + i * params_->row_ * params_->col_;
     }
-    auto ret = ParallelLaunch(this->context_->thread_pool_, MatmulBaseFP16Run, this, thread_count_);
+    auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, MatmulBaseFP16Run,
+                              this, thread_count_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "MatmulBaseFloatRun failed";
       return ret;

mindspore/lite/src/runtime/kernel/arm/fp16/pad_fp16.cc

@@ -89,7 +89,8 @@ int PadFp16CPUKernel::Run() {
         output_[i] = pad_param_->constant_value_;
       }
     }
-    ret = ParallelLaunch(this->context_->thread_pool_, PadImpl, this, op_parameter_->thread_num_);
+    ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, PadImpl, this,
+                         op_parameter_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "BatchnormRun error error_code[" << ret << "]";
     }
@@ -101,7 +102,8 @@ int PadFp16CPUKernel::Run() {
       return ret;
     }
 
-    ret = ParallelLaunch(this->context_->thread_pool_, MirrorPadImpl, this, context_->thread_num_);
+    ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, MirrorPadImpl, this,
+                         context_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "Pad Reflect or Symmetric mode run error, error_code[" << ret << "]";
     }

mindspore/lite/src/runtime/kernel/arm/fp16/pooling_fp16.cc

@@ -90,7 +90,8 @@ int PoolingFp16CPUKernel::Run() {
   fp16_input_ = reinterpret_cast<float16_t *>(input_tensor->data_c());
   fp16_output_ = reinterpret_cast<float16_t *>(output_tensor->data_c());
 
-  int error_code = ParallelLaunch(this->context_->thread_pool_, PoolingFp16Impl, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                  PoolingFp16Impl, this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "pooling error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp16/power_fp16.cc

@@ -87,7 +87,8 @@ int PowerFp16CPUKernel::Run() {
       return ret;
     }
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, PowerImplFp16, this, thread_count_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, PowerImplFp16, this,
+                            thread_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "PowerFp16CPUKernel error: " << ret;
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp16/quant_dtype_cast_fp16.cc

@@ -164,7 +164,8 @@ int QuantDTypeCastFp16CPUKernel::Run() {
     return RET_ERROR;
   }
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, QuantDTypeCastFP16Run, this, thread_n_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                            QuantDTypeCastFP16Run, this, thread_n_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Scale error error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp16/reduce_fp16.cc

@@ -93,7 +93,8 @@ int ReduceFp16CPUKernel::Run() {
     outer_size_ = outer_sizes_.at(i);
     inner_size_ = inner_sizes_.at(i);
     axis_size_ = axis_sizes_.at(i);
-    auto error_code = ParallelLaunch(this->context_->thread_pool_, ReduceFp16Impl, this, context_->thread_num_);
+    auto error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                     ReduceFp16Impl, this, context_->thread_num_);
     if (error_code != RET_OK) {
       FreeTmpBuffer();
       MS_LOG(ERROR) << "Reduce run error, error_code[" << error_code << "]";
@@ -108,7 +109,8 @@ int ReduceFp16CPUKernel::Run() {
   outer_size_ = outer_sizes_.back();
   inner_size_ = inner_sizes_.back();
   axis_size_ = axis_sizes_.back();
-  auto error_code = ParallelLaunch(this->context_->thread_pool_, ReduceFp16Impl, this, context_->thread_num_);
+  auto error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                   ReduceFp16Impl, this, context_->thread_num_);
   if (error_code != RET_OK) {
     FreeTmpBuffer();
     MS_LOG(ERROR) << "Reduce run error, error_code[" << error_code << "]";

mindspore/lite/src/runtime/kernel/arm/fp16/scale_fp16.cc

@@ -115,7 +115,8 @@ int ScaleFp16CPUKernel::Run() {
     return ret;
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, ScaleFp16Run, this, op_parameter_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ScaleFp16Run, this,
+                       op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Scale error error_code[" << ret << "]";
     FreeTmpBuffer();
@@ -127,12 +128,12 @@ int ScaleFp16CPUKernel::Run() {
 }
 
 int ScaleFp16CPUKernel::MallocAssignTmpBuffer() {
-  scale_ = ConvertInputFp32toFp16(in_tensors_.at(1), context_);
+  scale_ = ConvertInputFp32toFp16(in_tensors_.at(1), static_cast<const lite::InnerContext *>(this->context_));
   if (scale_ == nullptr) {
     return RET_ERROR;
   }
   if (in_tensors_.size() == 3) {
-    offset_ = ConvertInputFp32toFp16(in_tensors_.at(2), context_);
+    offset_ = ConvertInputFp32toFp16(in_tensors_.at(2), static_cast<const lite::InnerContext *>(this->context_));
     if (offset_ == nullptr) {
       return RET_ERROR;
     }

mindspore/lite/src/runtime/kernel/arm/fp16/slice_fp16.cc

@@ -63,7 +63,8 @@ int SliceFp16CPUKernel::Run() {
     DoSliceNoParallel(input_data, out_tensors_.at(0)->data_c(), param_, lite::DataTypeSize(kNumberTypeFloat16));
     return RET_OK;
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, SliceFp16Launch, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SliceFp16Launch,
+                            this, op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "fp16 slice launch fail!ret: " << ret;
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp16/softmax_fp16.cc

@@ -95,7 +95,8 @@ int SoftmaxLastAxisFp16Run(void *cdata, int task_id) {
 
 int SoftmaxFp16CPUKernel::Run() {
   if (in_plane_size_ == 1) {
-    auto ret = ParallelLaunch(this->context_->thread_pool_, SoftmaxLastAxisFp16Run, this, context_->thread_num_);
+    auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                              SoftmaxLastAxisFp16Run, this, context_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "SoftmaxFp16CPUKernel ParallelLaunch failed, ret: " << ret;
     }

mindspore/lite/src/runtime/kernel/arm/fp16/stack_fp16.cc

@@ -40,14 +40,15 @@ void StackFp16CPUKernel::InitMallocFlags() {
 int StackFp16CPUKernel::MallocAssignBuffer() {
   buffers_.resize(in_tensors_.size(), nullptr);
   for (size_t i = 0; i < in_tensors_.size(); ++i) {
-    buffers_.at(i) = reinterpret_cast<char *>(ConvertInputFp32toFp16(in_tensors_.at(i), context_));
+    buffers_.at(i) = reinterpret_cast<char *>(
+      ConvertInputFp32toFp16(in_tensors_.at(i), static_cast<const lite::InnerContext *>(context_)));
     if (buffers_.at(i) == nullptr) {
       return RET_ERROR;
     }
   }
 
   out_buffer_ = nullptr;
-  out_buffer_ = MallocOutputFp16(out_tensors_.at(0), context_);
+  out_buffer_ = MallocOutputFp16(out_tensors_.at(0), static_cast<const lite::InnerContext *>(this->context_));
   if (out_buffer_ == nullptr) {
     return RET_ERROR;
   }
@@ -100,7 +101,8 @@ int StackFp16CPUKernel::Run() {
   }
   // run stack
   num_threads_ = MSMIN(UP_DIV(outer_size_, 64), this->context_->thread_num_);
-  ret = ParallelLaunch(this->context_->thread_pool_, StackRun, this, num_threads_);
+  ret =
+    ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, StackRun, this, num_threads_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "StackBaseCPUKernel Run error: error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp16_grad/activation_fp16_grad.cc

@@ -79,7 +79,8 @@ int ActivationGradRunFp16(void *cdata, int task_id) {
 }
 
 int ActivationGradCPUKernelFp16::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, ActivationGradRunFp16, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                  ActivationGradRunFp16, this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Activation Grad function error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp16_grad/arithmetic_fp16_self_grad.cc

@@ -73,7 +73,8 @@ int ArithmeticSelfGradFp16Run(void *cdata, int task_id) {
 }
 
 int ArithmeticSelfGradFp16CPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, ArithmeticSelfGradFp16Run, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                  ArithmeticSelfGradFp16Run, this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Activation Grad function error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/activation_fp32.cc

@@ -101,7 +101,8 @@ int ActivationRun(void *cdata, int task_id) {
 }
 
 int ActivationCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, ActivationRun, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ActivationRun,
+                                  this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Activation function error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/adder_fp32.cc

@@ -122,7 +122,8 @@ int AdderCPUKernel::Run() {
     return RET_ERROR;
   }
 
-  int error_code = ParallelLaunch(this->context_->thread_pool_, AdderImpl, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, AdderImpl,
+                                  this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "adder error error_code[" << error_code << "]";
     FreeTmpBuffer();

mindspore/lite/src/runtime/kernel/arm/fp32/addn_fp32.cc

@@ -89,7 +89,8 @@ int AddNCPUKernel::Run() {
   in1_addr_ = input0_data;
   in2_addr_ = input1_data;
   out_addr_ = output_data;
-  auto ret = ParallelLaunch(this->context_->thread_pool_, AddNLaunch, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, AddNLaunch, this,
+                            op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "addn launch fail!ret: " << ret;
     return RET_ERROR;
@@ -97,7 +98,8 @@ int AddNCPUKernel::Run() {
   for (size_t i = 2; i < in_tensors_.size(); ++i) {
     in1_addr_ = reinterpret_cast<float *>(in_tensors_[i]->MutableData());
     in2_addr_ = output_data;
-    ret = ParallelLaunch(this->context_->thread_pool_, AddNLaunch, this, op_parameter_->thread_num_);
+    ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, AddNLaunch, this,
+                         op_parameter_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "addn launch fail!ret: " << ret << ", input index: " << i;
       return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/arithmetic_fp32.cc

@@ -418,7 +418,8 @@ int ArithmeticCPUKernel::Run() {
     input1_ptr_ = in_tensors_[1]->data_c();
   }
   output_ptr_ = out_tensors_[0]->data_c();
-  return ParallelLaunch(this->context_->thread_pool_, ArithmeticsRun, this, context_->thread_num_);
+  return ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ArithmeticsRun, this,
+                        context_->thread_num_);
 }
 
 REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_MulFusion, LiteKernelCreator<ArithmeticCPUKernel>)

mindspore/lite/src/runtime/kernel/arm/fp32/arithmetic_self_fp32.cc

@@ -113,7 +113,8 @@ int ArithmeticSelfRun(void *cdata, int task_id) {
 }
 
 int ArithmeticSelfCPUKernel::Run() {
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ArithmeticSelfRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ArithmeticSelfRun,
+                            this, op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ArithmeticSelfRun error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/batchnorm_fp32.cc

@@ -75,7 +75,8 @@ int BatchnormCPUKernel::InitConstTensor() {
 }
 
 int BatchnormCPUKernel::Run() {
-  auto ret = ParallelLaunch(this->context_->thread_pool_, BatchNormRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, BatchNormRun, this,
+                            op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "BatchnormRun error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/cast_fp32.cc

@@ -141,7 +141,8 @@ int CastCPUKernel::Run() {
   if (data_num_ == 0) {
     return RET_OK;
   }
-  return ParallelLaunch(this->context_->thread_pool_, CastRun, this, op_parameter_->thread_num_);
+  return ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, CastRun, this,
+                        op_parameter_->thread_num_);
 }
 
 REG_KERNEL(kCPU, kNumberTypeFloat32, PrimitiveType_Cast, LiteKernelCreator<CastCPUKernel>)

mindspore/lite/src/runtime/kernel/arm/fp32/concat_fp32.cc

@@ -69,7 +69,8 @@ int ConcatRun(void *cdata, int task_id) {
 }
 
 int ConcatCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, ConcatRun, this, op_parameter_->thread_num_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConcatRun,
+                                  this, op_parameter_->thread_num_);
   return error_code;
 }
 

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_1x1_fp32.cc

@@ -247,10 +247,12 @@ int Convolution1x1CPUKernel::Run() {
     }
 
     if (multi_thread_by_hw_) {
-      ParallelLaunch(this->context_->thread_pool_, Convolution1x1RunHw, this, thread_count_);
+      ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, Convolution1x1RunHw, this,
+                     thread_count_);
     } else {
       PackMatmulInput(input_ptr_, pack_input_, matmul_param_->row_, matmul_param_->deep_);
-      ParallelLaunch(this->context_->thread_pool_, Convolution1x1Run, this, thread_count_);
+      ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, Convolution1x1Run, this,
+                     thread_count_);
     }
   }
 

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_delegate_fp32.cc

@@ -138,16 +138,19 @@ kernel::LiteKernel *ConvolutionDelegateCPUKernel::CpuConvFp32KernelSelect() {
   kernel::LiteKernel *kernel = nullptr;
   auto conv_param = reinterpret_cast<ConvParameter *>(op_parameter_);
   if (conv_param->kernel_h_ == 1 && conv_param->kernel_w_ == 1) {
-    kernel = new (std::nothrow)
-      kernel::Convolution1x1CPUKernel(op_parameter_, in_tensors_, out_tensors_, context_, origin_weight_, origin_bias_);
+    kernel = new (std::nothrow) kernel::Convolution1x1CPUKernel(op_parameter_, in_tensors_, out_tensors_,
+                                                                static_cast<const lite::InnerContext *>(this->context_),
+                                                                origin_weight_, origin_bias_);
   } else {
     int out_unit;
     if (CheckIfUseWinograd(&out_unit, conv_param)) {
       kernel = new (std::nothrow) kernel::ConvolutionWinogradCPUKernel(
-        op_parameter_, in_tensors_, out_tensors_, context_, out_unit, origin_weight_, origin_bias_);
+        op_parameter_, in_tensors_, out_tensors_, static_cast<const lite::InnerContext *>(this->context_), out_unit,
+        origin_weight_, origin_bias_);
     } else {
-      kernel = new (std::nothrow)
-        kernel::ConvolutionCPUKernel(op_parameter_, in_tensors_, out_tensors_, context_, origin_weight_, origin_bias_);
+      kernel = new (std::nothrow) kernel::ConvolutionCPUKernel(op_parameter_, in_tensors_, out_tensors_,
+                                                               static_cast<const lite::InnerContext *>(this->context_),
+                                                               origin_weight_, origin_bias_);
     }
   }
 
@@ -214,7 +217,7 @@ kernel::LiteKernel *CpuGroupConvFp32KernelCreator(const std::vector<lite::Tensor
 /* creator func */
 kernel::LiteKernel *CpuConvFp32KernelCreator(const std::vector<lite::Tensor *> &inputs,
                                              const std::vector<lite::Tensor *> &outputs, OpParameter *op_parameter,
-                                             const InnerContext *ctx, const kernel::KernelKey &desc) {
+                                             const lite::Context *ctx, const kernel::KernelKey &desc) {
   MS_ASSERT(op_parameter != nullptr);
   MS_ASSERT(desc.type == schema::PrimitiveType_Conv2DFusion);
   MS_ASSERT(desc.data_type == kNumberTypeFloat32);
@@ -222,11 +225,12 @@ kernel::LiteKernel *CpuConvFp32KernelCreator(const std::vector<lite::Tensor *> &
   auto conv_param = reinterpret_cast<ConvParameter *>(op_parameter);
   kernel::LiteKernel *kernel = nullptr;
   if (conv_param->group_ == 1) {
-    kernel = new (std::nothrow) kernel::ConvolutionDelegateCPUKernel(op_parameter, inputs, outputs, ctx);
+    kernel = new (std::nothrow)
+      kernel::ConvolutionDelegateCPUKernel(op_parameter, inputs, outputs, static_cast<const lite::InnerContext *>(ctx));
   } else if (conv_param->group_ == conv_param->input_channel_ && conv_param->group_ == conv_param->output_channel_) {
-    kernel = CpuConvDwFp32KernelCreator(inputs, outputs, op_parameter, ctx);
+    kernel = CpuConvDwFp32KernelCreator(inputs, outputs, op_parameter, static_cast<const lite::InnerContext *>(ctx));
   } else {
-    kernel = CpuGroupConvFp32KernelCreator(inputs, outputs, op_parameter, ctx);
+    kernel = CpuGroupConvFp32KernelCreator(inputs, outputs, op_parameter, static_cast<const lite::InnerContext *>(ctx));
   }
 
   if (kernel == nullptr) {

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_depthwise_3x3_fp32.cc

@@ -126,7 +126,8 @@ int ConvolutionDepthwise3x3CPUKernel::Run() {
   auto output_tensor = out_tensors_.at(kOutputIndex);
   output_ptr_ = reinterpret_cast<float *>(output_tensor->data_c());
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ConvDw3x3Run, this, conv_param_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvDw3x3Run, this,
+                            conv_param_->thread_num_);
   ctx_->allocator->Free(buffer_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ConvDw3x3Run error: error_code[" << ret << "]";

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_depthwise_fp32.cc

@@ -107,7 +107,8 @@ int ConvolutionDepthwiseCPUKernel::Run() {
   auto output_tensor = out_tensors_.at(kOutputIndex);
   output_ptr_ = reinterpret_cast<float *>(output_tensor->MutableData());
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ConvDwRun, this, conv_param_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvDwRun, this,
+                            conv_param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ConvDwRun error: error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_depthwise_indirect_fp32.cc

@@ -194,7 +194,8 @@ int ConvolutionDepthwiseIndirectCPUKernel::Run() {
 
   ConvDwInitIndirection(indirect_buffer_, packed_input_, zero_ptr_, conv_param_, step_h, step_w);
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ConvDwIndirectRun, this, conv_param_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvDwIndirectRun,
+                            this, conv_param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ConvDwIndirectRun error: error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_depthwise_slidewindow_fp32.cc

@@ -163,7 +163,8 @@ int ConvolutionDepthwiseSWCPUKernel::Run() {
     packed_output_ = output_ptr;
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, ConvDwSWRun, this, conv_param_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvDwSWRun, this,
+                       conv_param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ConvDwSWRun error: error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_fp32.cc

@@ -152,7 +152,8 @@ int ConvolutionCPUKernel::Run() {
     PackWeight();
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, ConvolutionImpl, this, thread_count_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvolutionImpl, this,
+                       thread_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "conv error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/convolution_winograd_fp32.cc

@@ -219,7 +219,8 @@ int ConvolutionWinogradCPUKernel::Run() {
     InitWeightBias();
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, ConvolutionWinogradImpl, this, thread_count_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ConvolutionWinogradImpl,
+                       this, thread_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "conv winograd error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/crop_and_resize_fp32.cc

@@ -151,7 +151,8 @@ int CropAndResizeCPUKernel::Run() {
     return ret;
   }
 
-  int error_code = ParallelLaunch(this->context_->thread_pool_, CropAndResizeImpl, this, context_->thread_num_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                  CropAndResizeImpl, this, context_->thread_num_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "CropAndResize run error, error_code[" << error_code << "]";
     FreeTmpBuffer();

mindspore/lite/src/runtime/kernel/arm/fp32/crop_fp32.cc

@@ -62,7 +62,8 @@ int CropCPUKernel::Run() {
     return RET_OK;
   }
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, CropLaunch, this, crop_para_->thread_count_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, CropLaunch, this,
+                            crop_para_->thread_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Crop launch fail!ret: " << ret;
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/deconvolution_depthwise_fp32.cc

@@ -168,7 +168,8 @@ int DeconvolutionDepthwiseCPUKernel::Run() {
     packed_output_ = output_addr;
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, DeconvDwRun, this, conv_param_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, DeconvDwRun, this,
+                       conv_param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "DeconvDwRun error: error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/deconvolution_fp32.cc

@@ -222,7 +222,8 @@ int DeConvolutionCPUKernel::Run() {
     RowMajor2Col12Major(input_ptr_, pack_input_, matmul_param_->row_, matmul_param_->deep_);
 #endif
 
-    error_code = ParallelLaunch(this->context_->thread_pool_, DeConvFp32Run, this, thread_count_);
+    error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, DeConvFp32Run,
+                                this, thread_count_);
     if (error_code != RET_OK) {
       MS_LOG(ERROR) << "deconv fp32 run error! error_code[" << error_code << "]";
       FreeRunBuf();
@@ -236,7 +237,7 @@ int DeConvolutionCPUKernel::Run() {
 
 kernel::LiteKernel *CpuDeConvFp32KernelCreator(const std::vector<lite::Tensor *> &inputs,
                                                const std::vector<lite::Tensor *> &outputs, OpParameter *op_parameter,
-                                               const lite::InnerContext *ctx, const kernel::KernelKey &desc) {
+                                               const lite::Context *ctx, const kernel::KernelKey &desc) {
   MS_ASSERT(op_parameter != nullptr);
   MS_ASSERT(desc.type == schema::PrimitiveType_Conv2dTransposeFusion);
 
@@ -245,12 +246,15 @@ kernel::LiteKernel *CpuDeConvFp32KernelCreator(const std::vector<lite::Tensor *>
   if (conv_param->group_ == 1) {
     if ((conv_param->stride_h_ != 1 || conv_param->stride_w_ != 1) &&
         (conv_param->dilation_w_ == 1 && conv_param->dilation_h_ == 1)) {
-      kernel = new (std::nothrow) kernel::DeConvolutionWinogradCPUKernel(op_parameter, inputs, outputs, ctx);
+      kernel = new (std::nothrow) kernel::DeConvolutionWinogradCPUKernel(op_parameter, inputs, outputs,
+                                                                         static_cast<const lite::InnerContext *>(ctx));
     } else {
-      kernel = new (std::nothrow) kernel::DeConvolutionCPUKernel(op_parameter, inputs, outputs, ctx);
+      kernel = new (std::nothrow)
+        kernel::DeConvolutionCPUKernel(op_parameter, inputs, outputs, static_cast<const lite::InnerContext *>(ctx));
     }
   } else if (conv_param->group_ == conv_param->input_channel_ && conv_param->group_ == conv_param->output_channel_) {
-    kernel = new (std::nothrow) kernel::DeconvolutionDepthwiseCPUKernel(op_parameter, inputs, outputs, ctx);
+    kernel = new (std::nothrow) kernel::DeconvolutionDepthwiseCPUKernel(op_parameter, inputs, outputs,
+                                                                        static_cast<const lite::InnerContext *>(ctx));
   } else {
     MS_LOG(ERROR) << "deconv do not support group deconv!";
     kernel = nullptr;

mindspore/lite/src/runtime/kernel/arm/fp32/deconvolution_winograd_fp32.cc

@@ -411,10 +411,12 @@ int DeConvolutionWinogradCPUKernel::Run() {
     nhwc_output_ = src_out + batch_index * deconv_param_->output_plane_ * conv_param_->output_channel_;
 
     ::memset(nc4hw4_output_, 0, deconv_param_->output_plane_ * deconv_param_->oc_div4_ * C4NUM * sizeof(float));
-    ParallelLaunch(this->context_->thread_pool_, DeConvWgFp32Run, this, deconv_param_->thread_num_);
+    ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, DeConvWgFp32Run, this,
+                   deconv_param_->thread_num_);
 
     /*post bias activate and nhwc */
-    ParallelLaunch(this->context_->thread_pool_, DeConvWgPostFp32Run, this, thread_num_hw_);
+    ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, DeConvWgPostFp32Run, this,
+                   thread_num_hw_);
   }
 
   FreeRunBuf();

mindspore/lite/src/runtime/kernel/arm/fp32/elu_fp32.cc

@@ -55,7 +55,8 @@ int EluRun(void *cdata, int task_id) {
 }
 
 int EluCPUKernel::Run() {
-  auto ret = ParallelLaunch(this->context_->thread_pool_, EluRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, EluRun, this,
+                            op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Elu error: error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/embedding_lookup_fp32.cc

@@ -87,7 +87,8 @@ int EmbeddingLookupCPUKernel::Run() {
     memcpy(input_addr_ + dest_loc, input_t, sizeof(float) * in_tensors_.at(i)->ElementsNum());
     dest_loc += in_tensors_.at(i)->ElementsNum();
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, EmbeddingLookupRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, EmbeddingLookupRun,
+                            this, op_parameter_->thread_num_);
   FreeRunBuff();
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "EmbeddingLookup error: error_code[" << ret << "]";

mindspore/lite/src/runtime/kernel/arm/fp32/exp_fp32.cc

@@ -73,7 +73,8 @@ int ExpCPUKernel::Run() {
   output_addr_ = reinterpret_cast<float *>(out_tensors_.front()->MutableData());
   exp_parameter_->element_num_ = in_tensors_.front()->ElementsNum();
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ExpRun, this, exp_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ExpRun, this,
+                            exp_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Exp error: error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/fill_fp32.cc

@@ -91,7 +91,8 @@ int FillCPUKernel::Run() {
     MS_LOG(ERROR) << "unsupported fill data type " << fill_input->data_type();
     return RET_ERROR;
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, FillRun, this, thread_sz_count_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, FillRun, this,
+                            thread_sz_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "FillRun error error_code[" << ret << "]";
     return ret;

mindspore/lite/src/runtime/kernel/arm/fp32/fused_batchnorm_fp32.cc

@@ -93,7 +93,8 @@ int FusedBatchnormCPUKernel::Run() {
 
     trained_ = true;  // trained at least once
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, BatchNormRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, BatchNormRun, this,
+                            op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "BatchnormRun error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/gatherNd_fp32.cc

@@ -128,7 +128,8 @@ int GatherNdCPUKernel::Run() {
   in_ptr_ = reinterpret_cast<float *>(in_tensors_.front()->MutableData());
   out_ptr_ = reinterpret_cast<float *>(out_tensors_.front()->MutableData());
   InitOffset();
-  auto ret = ParallelLaunch(this->context_->thread_pool_, GatherNdRun, this, thread_sz_count_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, GatherNdRun, this,
+                            thread_sz_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "gatherNd error error_code[" << ret << "]";
     return ret;

mindspore/lite/src/runtime/kernel/arm/fp32/gather_fp32.cc

@@ -92,7 +92,8 @@ int GatherCPUKernel::Run() {
     return ret;
   }
 
-  ret = ParallelLaunch(this->context_->thread_pool_, GatherRun, this, op_parameter_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, GatherRun, this,
+                       op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Gather function error error_code[" << ret << "]";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/instance_norm_fp32.cc

@@ -66,7 +66,8 @@ int InstanceNormCPUKernel::Run() {
   gamma_data_ = reinterpret_cast<float *>(in_tensors_.at(1)->data_c());
   beta_data_ = reinterpret_cast<float *>(in_tensors_.at(2)->data_c());
   dst_data_ = reinterpret_cast<float *>(out_tensors_.at(0)->data_c());
-  auto ret = ParallelLaunch(this->context_->thread_pool_, InstanceNormRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, InstanceNormRun,
+                            this, op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "InstanceNormRun error error_code[" << ret << "]";
     return ret;

mindspore/lite/src/runtime/kernel/arm/fp32/l2_norm_fp32.cc

@@ -146,7 +146,8 @@ int L2NormCPUKernel::Run() {
   output_ptr_ = reinterpret_cast<float *>(out_tensors_.at(kOutputIndex)->MutableData());
   if (l2_norm_param_->axis_num_ == 0 || l2_norm_param_->axis_num_ == input_shape.size()) {
     // all axis
-    auto ret = ParallelLaunch(this->context_->thread_pool_, SquareSumRun, this, context_->thread_num_);
+    auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SquareSumRun, this,
+                              context_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "L2Norm error: error_code[" << ret << "]";
       return RET_ERROR;
@@ -156,13 +157,15 @@ int L2NormCPUKernel::Run() {
       sum += tmp_sum_[i];
     }
     sqrt_sum_ = sqrt(sum > l2_norm_param_->epsilon_ ? sum : l2_norm_param_->epsilon_);
-    ret = ParallelLaunch(this->context_->thread_pool_, L2NormRun, this, context_->thread_num_);
+    ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, L2NormRun, this,
+                         context_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "L2Norm error: error_code[" << ret << "]";
       return RET_ERROR;
     }
   } else if (l2_norm_param_->axis_num_ == 1 && l2_norm_param_->axis_[0] == static_cast<int>(input_shape.size()) - 1) {
-    auto ret = ParallelLaunch(this->context_->thread_pool_, L2NormTrailingAxisRun, this, context_->thread_num_);
+    auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                              L2NormTrailingAxisRun, this, context_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "L2Norm error: error_code[" << ret << "]";
       return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/layer_norm_fp32.cc

@@ -92,7 +92,8 @@ int LayerNormCPUKernel::Run() {
     mean_data_ = reinterpret_cast<float *>(context_->allocator->Malloc(param_->norm_outer_size_ * sizeof(float)));
     var_data_ = reinterpret_cast<float *>(context_->allocator->Malloc(param_->norm_outer_size_ * sizeof(float)));
   }
-  ret = ParallelLaunch(this->context_->thread_pool_, LayerNormRun, this, op_parameter_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, LayerNormRun, this,
+                       op_parameter_->thread_num_);
   if (out_tensors_.size() != 3) {
     context_->allocator->Free(mean_data_);
     context_->allocator->Free(var_data_);

mindspore/lite/src/runtime/kernel/arm/fp32/local_response_norm_fp32.cc

@@ -74,7 +74,8 @@ int LocalResponseNormRun(void *cdata, int task_id) {
 }
 
 int LocalResponseNormCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, LocalResponseNormRun, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                  LocalResponseNormRun, this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "LocalResponseNorm function error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/log_softmax_fp32.cc

@@ -96,7 +96,8 @@ int LogSoftmaxLastAxisRun(void *cdata, int task_id) {
 int LogSoftmaxCPUKernel::Run() {
   int ret = RET_OK;
   if (in_plane_size_ == 1) {
-    ret = ParallelLaunch(this->context_->thread_pool_, LogSoftmaxLastAxisRun, this, context_->thread_num_);
+    ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, LogSoftmaxLastAxisRun,
+                         this, context_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "LogSoftmaxCPUKernel ParallelLaunch failed, ret: " << ret;
     }

mindspore/lite/src/runtime/kernel/arm/fp32/lsh_projection_fp32.cc

@@ -61,7 +61,8 @@ int LshProjectionCPUKernel::Run() {
   if (ret != RET_OK) {
     return ret;
   }
-  ret = ParallelLaunch(this->context_->thread_pool_, LshProjectionRun, this, op_parameter_->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, LshProjectionRun, this,
+                       op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "LshProjection kernel parallel launch failed";
   }

mindspore/lite/src/runtime/kernel/arm/fp32/matmul_fp32_base.cc

@@ -332,7 +332,8 @@ int MatmulFp32BaseCPUKernel::Run() {
       batch_b_ptr_ = b_pack_ptr_ + i * params_->deep_ * params_->col_align_;
       batch_c_ptr_ = c_ptr + i * params_->row_ * params_->col_;
     }
-    auto ret = ParallelLaunch(this->context_->thread_pool_, MatmulBaseFloatRun, this, thread_count_);
+    auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, MatmulBaseFloatRun,
+                              this, thread_count_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "MatmulBaseFloatRun failed";
       return ret;

mindspore/lite/src/runtime/kernel/arm/fp32/one_hot_fp32.cc

@@ -181,7 +181,8 @@ int OneHotCPUKernel::GetParams() {
 }
 
 int OneHotCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, RunOneHot, this, context_->thread_num_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, RunOneHot,
+                                  this, context_->thread_num_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "OneHot function error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/pad_fp32.cc

@@ -395,7 +395,8 @@ int PadCPUKernel::Run() {
         output_data[i] = pad_param_->constant_value_;
       }
     }
-    error_code = ParallelLaunch(this->context_->thread_pool_, PadImpl, this, context_->thread_num_);
+    error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, PadImpl, this,
+                                context_->thread_num_);
     if (error_code != RET_OK) {
       MS_LOG(ERROR) << "Pad run error, error_code[" << error_code << "]";
       return RET_ERROR;
@@ -408,7 +409,8 @@ int PadCPUKernel::Run() {
       return error_code;
     }
 
-    error_code = ParallelLaunch(this->context_->thread_pool_, MirrorPadImpl, this, context_->thread_num_);
+    error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, MirrorPadImpl,
+                                this, context_->thread_num_);
     if (error_code != RET_OK) {
       MS_LOG(ERROR) << "Pad Reflect or Symmetric mode run error, error_code[" << error_code << "]";
       return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/pooling_fp32.cc

@@ -85,7 +85,8 @@ int PoolingImpl(void *cdata, int task_id) {
 }
 
 int PoolingCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, PoolingImpl, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, PoolingImpl,
+                                  this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "pooling error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/power_fp32.cc

@@ -41,7 +41,8 @@ int PowerImpl(void *cdata, int task_id) {
 }
 
 int PowerCPUKernel::Run() {
-  auto ret = ParallelLaunch(this->context_->thread_pool_, PowerImpl, this, thread_count_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, PowerImpl, this,
+                            thread_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "PowerCPUKernel error: " << ret;
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/prelu_fp32.cc

@@ -93,7 +93,8 @@ int PReluCPUKernel::Run() {
   auto negative_slope_tensor = in_tensors_.at(1);
   prelu_param_->slope_ = reinterpret_cast<float *>(negative_slope_tensor->data_c());
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, PReluRun, this, prelu_param_->op_parameter_.thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, PReluRun, this,
+                            prelu_param_->op_parameter_.thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "PRelu Run error: error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/reduce_fp32.cc

@@ -117,7 +117,8 @@ int ReduceCPUKernel::Run() {
     outer_size_ = outer_sizes_.at(i);
     inner_size_ = inner_sizes_.at(i);
     axis_size_ = axis_sizes_.at(i);
-    auto error_code = ParallelLaunch(this->context_->thread_pool_, ReduceImpl, this, context_->thread_num_);
+    auto error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ReduceImpl,
+                                     this, context_->thread_num_);
     if (error_code != RET_OK) {
       MS_LOG(ERROR) << "Reduce run error, error_code[" << error_code << "]";
       FreeTmpBuffer();

mindspore/lite/src/runtime/kernel/arm/fp32/resize_fp32.cc

@@ -205,7 +205,8 @@ int ResizeCPUKernel::RunImpl(int task_id) {
 }
 
 int ResizeCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, ResizeImpl, this, context_->thread_num_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ResizeImpl,
+                                  this, context_->thread_num_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Resize run error, error_code[" << error_code << "]";
     FreeTmpBuffer();

mindspore/lite/src/runtime/kernel/arm/fp32/reverse_fp32.cc

@@ -129,7 +129,8 @@ int ReverseCPUKernel::DoReverse(int task_id) {
 int ReverseCPUKernel::Run() {
   in_ptr_ = reinterpret_cast<float *>(in_tensors_[0]->MutableData());
   out_ptr_ = reinterpret_cast<float *>(out_tensors_[0]->MutableData());
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ReverseRun, this, thread_sz_count_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ReverseRun, this,
+                            thread_sz_count_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Reverse run error error_code[" << ret << "]";
     return ret;

mindspore/lite/src/runtime/kernel/arm/fp32/roi_pooling_fp32.cc

@@ -101,7 +101,8 @@ int ROIPoolingCPUKernel::Run() {
   in_ptr_ = reinterpret_cast<float *>(in_tensors_.front()->MutableData());
   out_ptr_ = reinterpret_cast<float *>(out_tensors_.front()->MutableData());
   roi_ptr_ = reinterpret_cast<float *>(in_tensors_.at(1)->MutableData());
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ROIPoolingRun, this, param_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ROIPoolingRun, this,
+                            param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ROIPooling error: error_code[" << ret << "]";
     return ret;

mindspore/lite/src/runtime/kernel/arm/fp32/scale_fp32.cc

@@ -188,7 +188,8 @@ int ScaleCPUKernel::Run() {
   auto out_tensor = out_tensors_.front();
   output_ptr_ = reinterpret_cast<float *>(out_tensor->MutableData());
 
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ScaleRun, this, op_parameter_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ScaleRun, this,
+                            op_parameter_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "Scale error error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/scatter_nd_fp32.cc

@@ -149,7 +149,8 @@ int ScatterNDRun(void *cdata, int task_id) {
 }
 
 int ScatterNDCPUKernel::Run() {
-  auto ret = ParallelLaunch(this->context_->thread_pool_, ScatterNDRun, this, thread_n_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, ScatterNDRun, this,
+                            thread_n_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "ScatterND error error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/softmax_fp32.cc

@@ -96,7 +96,8 @@ int SoftmaxLastAxisRun(void *cdata, int task_id) {
 int SoftmaxCPUKernel::Run() {
   int ret = RET_OK;
   if (in_plane_size_ == 1) {
-    ret = ParallelLaunch(this->context_->thread_pool_, SoftmaxLastAxisRun, this, context_->thread_num_);
+    ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SoftmaxLastAxisRun,
+                         this, context_->thread_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "SoftmaxCPUKernel ParallelLaunch failed, ret: " << ret;
     }

mindspore/lite/src/runtime/kernel/arm/fp32/space_to_batch_fp32.cc

@@ -102,7 +102,8 @@ int SpaceToBatchCPUKernel::Run() {
     }
   }
 
-  ParallelLaunch(this->context_->thread_pool_, SpaceToBatchFp32Run, this, op_parameter_->thread_num_);
+  ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SpaceToBatchFp32Run, this,
+                 op_parameter_->thread_num_);
 
   return RET_OK;
 }

mindspore/lite/src/runtime/kernel/arm/fp32/space_to_depth_fp32.cc

@@ -93,7 +93,8 @@ int SpaceToDepthCPUKernel::Run() {
   input_ptr_ = reinterpret_cast<float *>(in_tensors_.at(0)->data_c());
   output_ptr_ = reinterpret_cast<float *>(out_tensors_.at(0)->data_c());
   if (in_tensors_.at(0)->format() == schema::Format::Format_NHWC) {
-    auto ret = ParallelLaunch(this->context_->thread_pool_, SpaceToDepthRun, this, thread_h_num_);
+    auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SpaceToDepthRun,
+                              this, thread_h_num_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "SpaceToDepth error error_code[" << ret << "]";
       return ret;

mindspore/lite/src/runtime/kernel/arm/fp32/sparse_to_dense_fp32.cc

@@ -175,7 +175,8 @@ int SparseToDenseCPUKernel::Run() {
   }
   output_data = reinterpret_cast<float *>(out_tensors_.at(0)->MutableData());
   count_unit_ = thread_count_ > 1 ? UP_DIV(index_num, thread_count_) : index_num;
-  ret = ParallelLaunch(this->context_->thread_pool_, SparseToDenseRun, this, s2d_param->thread_num_);
+  ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, SparseToDenseRun, this,
+                       s2d_param->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "SparseToDenseRun error: error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32/transpose_fp32.cc

@@ -159,7 +159,8 @@ int TransposeCPUKernel::Run() {
   thread_count_ = op_parameter_->thread_num_;
   GetNHNCTransposeFunc(in_tensor, out_tensor, param_);
   if (NHNCTransposeFunc_ != nullptr) {
-    auto ret = ParallelLaunch(this->context_->thread_pool_, TransposeImpl, this, thread_count_);
+    auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, TransposeImpl,
+                              this, thread_count_);
     if (ret != RET_OK) {
       MS_LOG(ERROR) << "NHNCTransposeFunc_ is error!";
     }
@@ -187,7 +188,8 @@ int TransposeCPUKernel::Run() {
   }
   int ret;
   if (dims_ > MAX_TRANSPOSE_DIM_SIZE) {
-    ret = ParallelLaunch(this->context_->thread_pool_, TransposeImpl, this, thread_count_);
+    ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, TransposeImpl, this,
+                         thread_count_);
   } else {
     ret = DoTransposeFp32(in_data_, out_data_, out_shape_, param_);
   }

mindspore/lite/src/runtime/kernel/arm/fp32/where_fp32.cc

@@ -133,7 +133,8 @@ int WhereCPUKernel::RunWithTripleInputs() {
     MS_LOG(ERROR) << "Error, inputs' length are zero !!!";
     return RET_ERROR;
   }
-  auto ret = ParallelLaunch(this->context_->thread_pool_, WhereRun, this, where_param_->thread_num_);
+  auto ret = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, WhereRun, this,
+                            where_param_->thread_num_);
   if (ret != RET_OK) {
     MS_LOG(ERROR) << "WhereDwRun error: error_code[" << ret << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32_grad/activation_grad.cc

@@ -98,7 +98,8 @@ int ActivationGradRun(void *cdata, int task_id) {
 }
 
 int ActivationGradCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, ActivationGradRun, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                  ActivationGradRun, this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Activation Grad function error error_code[" << error_code << "]";
     return RET_ERROR;

mindspore/lite/src/runtime/kernel/arm/fp32_grad/adam.cc

@@ -102,7 +102,8 @@ int AdamRun(void *cdata, int task_id) {
 }
 
 int AdamCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, AdamRun, this, thread_count_);
+  int error_code =
+    ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_, AdamRun, this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Adam function error error_code[" << error_code << "]";
     return RET_ERROR;
@@ -145,9 +146,10 @@ int AdamCPUKernel::OptimizerStep() {
 
 kernel::LiteKernel *CpuAdamFp32KernelCreator(const std::vector<lite::Tensor *> &inputs,
                                              const std::vector<lite::Tensor *> &outputs, OpParameter *opParameter,
-                                             const lite::InnerContext *ctx, const kernel::KernelKey &desc) {
+                                             const lite::Context *ctx, const kernel::KernelKey &desc) {
   MS_ASSERT(desc.type == schema::PrimitiveType_Adam);
-  auto *kernel = new (std::nothrow) AdamCPUKernel(opParameter, inputs, outputs, ctx);
+  auto *kernel =
+    new (std::nothrow) AdamCPUKernel(opParameter, inputs, outputs, static_cast<const lite::InnerContext *>(ctx));
   if (kernel == nullptr) {
     MS_LOG(ERROR) << "new AdamCPUKernel fail!";
     free(opParameter);

mindspore/lite/src/runtime/kernel/arm/fp32_grad/apply_momentum.cc

@@ -82,7 +82,8 @@ int ApplyMomentumRun(void *cdata, int task_id) {
 }
 
 int ApplyMomentumCPUKernel::Run() {
-  int error_code = ParallelLaunch(this->context_->thread_pool_, ApplyMomentumRun, this, thread_count_);
+  int error_code = ParallelLaunch(static_cast<const lite::InnerContext *>(this->context_)->thread_pool_,
+                                  ApplyMomentumRun, this, thread_count_);
   if (error_code != RET_OK) {
     MS_LOG(ERROR) << "Apply Momentum function error error_code[" << error_code << "]";
     return RET_ERROR;
@@ -119,10 +120,11 @@ int ApplyMomentumCPUKernel::OptimizerStep() {
 
 kernel::LiteKernel *CpuApplyMomentumFp32KernelCreator(const std::vector<lite::Tensor *> &inputs,
                                                       const std::vector<lite::Tensor *> &outputs,
-                                                      OpParameter *opParameter, const lite::InnerContext *ctx,
+                                                      OpParameter *opParameter, const lite::Context *ctx,
                                                       const kernel::KernelKey &desc) {
   MS_ASSERT(desc.type == schema::PrimitiveType_ApplyMomentum);
-  auto *kernel = new (std::nothrow) ApplyMomentumCPUKernel(opParameter, inputs, outputs, ctx);
+  auto *kernel = new (std::nothrow)
+    ApplyMomentumCPUKernel(opParameter, inputs, outputs, static_cast<const lite::InnerContext *>(ctx));
   if (kernel == nullptr) {
     MS_LOG(ERROR) << "new ApplyMomentumCPUKernel fail!";
     free(opParameter);