!10787 [MSLITE] arithmetic optimize

From: @ling_qiao_min
Reviewed-by: 
Signed-off-by:

mindspore/lite/nnacl/adder.h

@@ -16,9 +16,7 @@
 #ifndef MINDSPORE_LITE_NNACL_ADDER_H_
 #define MINDSPORE_LITE_NNACL_ADDER_H_
 
-#include <math.h>
 #include "nnacl/op_base.h"
-#include "nnacl/quantization/fixed_point.h"
 
 typedef struct AdderParameter {
   OpParameter op_parameter_;

mindspore/lite/nnacl/flatten.c → mindspore/lite/nnacl/arithmetic.c

@@ -13,9 +13,18 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#include "nnacl/flatten.h"
-#include <string.h>
 
-void Flatten(const void *input, void *output, const FlattenParameter *flatten_param) {
-  memcpy(output, input, flatten_param->size);
+#include "nnacl/arithmetic.h"
+
+void CalcMultiplesAndStrides(ArithmeticParameter *param) {
+  NNACL_ASSERT(param->in_shape0_[i] != 0);
+  NNACL_ASSERT(param->in_shape1_[i] != 0);
+  for (size_t i = 0; i < param->ndim_; i++) {
+    param->multiples0_[i] = param->out_shape_[i] / param->in_shape0_[i];
+    param->multiples1_[i] = param->out_shape_[i] / param->in_shape1_[i];
+  }
+  // cal strides
+  ComputeStrides(param->in_shape0_, param->in_strides0_, param->ndim_);
+  ComputeStrides(param->in_shape1_, param->in_strides1_, param->ndim_);
+  ComputeStrides(param->out_shape_, param->out_strides_, param->ndim_);
 }

mindspore/lite/nnacl/flatten.h → mindspore/lite/nnacl/arithmetic.h

@@ -13,23 +13,41 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#ifndef MINDSPORE_LITE_NNACL_FLATTEN_H_
-#define MINDSPORE_LITE_NNACL_FLATTEN_H_
+
+#ifndef MINDSPORE_LITE_NNACL_ARTITHMETIC_H_
+#define MINDSPORE_LITE_NNACL_ARTITHMETIC_H_
+
 #include "nnacl/op_base.h"
+#include "nnacl/common_func.h"
+#include "nnacl/nnacl_utils.h"
 
-typedef struct FlattenParameter {
-  // Primitive parameter
+typedef struct ArithmeticParameter {
   OpParameter op_parameter_;
-  // other parameter
-  int size;
-} FlattenParameter;
+  bool broadcasting_;
+  size_t ndim_;
+  int activation_type_;
+  int in_shape0_[10];
+  int in_elements_num0_;
+  int in_shape1_[10];
+  int in_elements_num1_;
+
+  int out_shape_[10];
+  int out_elements_num_;
+
+  int in_strides0_[10];
+  int in_strides1_[10];
+  int out_strides_[10];
+
+  int multiples0_[10];
+  int multiples1_[10];
+} ArithmeticParameter;
 
 #ifdef __cplusplus
 extern "C" {
 #endif
-void Flatten(const void *input, void *output, const FlattenParameter *flatten_param);
+void CalcMultiplesAndStrides(ArithmeticParameter *param);
 #ifdef __cplusplus
 }
 #endif
 
-#endif  // MINDSPORE_LITE_NNACL_FLATTEN_H_
+#endif  // MINDSPORE_LITE_NNACL_ARTITHMETIC_H_

mindspore/lite/nnacl/arithmetic_common.c

@@ -1,102 +0,0 @@
-/**
- * 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 "nnacl/arithmetic_common.h"
-#include "nnacl/nnacl_utils.h"
-
-void TileOneDimension(const float *inData, float *outData, int dim, size_t ndim, const int *inShape,
-                      const int *inStrides, const int *outStrides, const int *multiple) {
-  int srcDimSize = inShape[dim];
-  if (dim == ndim - 1) {
-    for (int i = 0; i < multiple[dim]; i++) {
-      memcpy(outData, inData, srcDimSize * sizeof(float));
-      outData += srcDimSize;
-    }
-    return;
-  }
-  for (size_t i = 0; i < srcDimSize; i++) {
-    for (size_t j = 0; j < multiple[dim]; j++) {
-      TileOneDimension(inData + inStrides[dim] * i, outData + outStrides[dim] * (i + j * srcDimSize), dim + 1, ndim,
-                       inShape, inStrides, outStrides, multiple);
-    }
-  }
-}
-
-void TileOneDimensionUint8(const uint8_t *inData, uint8_t *outData, int dim, size_t ndim, const int *inShape,
-                           const int *inStrides, const int *outStrides, const int *multiple) {
-  int srcDimSize = inShape[dim];
-  if (dim == ndim - 1) {
-    for (int i = 0; i < multiple[dim]; i++) {
-      memcpy(outData, inData, srcDimSize * sizeof(uint8_t));
-      outData += srcDimSize;
-    }
-    return;
-  }
-  for (size_t i = 0; i < srcDimSize; i++) {
-    for (size_t j = 0; j < multiple[dim]; j++) {
-      TileOneDimensionUint8(inData + inStrides[dim] * i, outData + outStrides[dim] * (i + j * srcDimSize), dim + 1,
-                            ndim, inShape, inStrides, outStrides, multiple);
-    }
-  }
-}
-
-void ComputeStrides(const int *shape, int *strides, const int ndim) {
-  int stride = 1;
-  for (int i = ndim - 1; i >= 0; i--) {
-    strides[i] = stride;
-    stride *= shape[i];
-  }
-}
-
-void CalcMultiplesAndStrides(ArithmeticParameter *param) {
-  NNACL_ASSERT(param->in_shape0_[i] != 0);
-  NNACL_ASSERT(param->in_shape1_[i] != 0);
-  for (size_t i = 0; i < param->ndim_; i++) {
-    param->multiples0_[i] = param->out_shape_[i] / param->in_shape0_[i];
-    param->multiples1_[i] = param->out_shape_[i] / param->in_shape1_[i];
-  }
-  // cal strides
-  ComputeStrides(param->in_shape0_, param->in_strides0_, param->ndim_);
-  ComputeStrides(param->in_shape1_, param->in_strides1_, param->ndim_);
-  ComputeStrides(param->out_shape_, param->out_strides_, param->ndim_);
-}
-
-void TileDimensions(const float *data0, const float *data1, float *tile_data0, float *tile_data1,
-                    ArithmeticParameter *param) {
-  CalcMultiplesAndStrides(param);
-  TileOneDimension(data0, tile_data0, 0, param->ndim_, param->in_shape0_, param->in_strides0_, param->out_strides_,
-                   param->multiples0_);
-  TileOneDimension(data1, tile_data1, 0, param->ndim_, param->in_shape1_, param->in_strides1_, param->out_strides_,
-                   param->multiples1_);
-}
-
-void TileDimensionsUint8(const uint8_t *data0, const uint8_t *data1, uint8_t *tile_data0, uint8_t *tile_data1,
-                         ArithmeticParameter *param) {
-  CalcMultiplesAndStrides(param);
-  TileOneDimensionUint8(data0, tile_data0, 0, param->ndim_, param->in_shape0_, param->in_strides0_, param->out_strides_,
-                        param->multiples0_);
-  TileOneDimensionUint8(data1, tile_data1, 0, param->ndim_, param->in_shape1_, param->in_strides1_, param->out_strides_,
-                        param->multiples1_);
-}
-
-void TileDimensionsInt8(const int8_t *data0, const int8_t *data1, int8_t *tile_data0, int8_t *tile_data1,
-                        ArithmeticParameter *param) {
-  CalcMultiplesAndStrides(param);
-  TileOneDimensionUint8((uint8_t *)(data0), (uint8_t *)(tile_data0), 0, param->ndim_, param->in_shape0_,
-                        param->in_strides0_, param->out_strides_, param->multiples0_);
-  TileOneDimensionUint8((uint8_t *)(data1), (uint8_t *)(tile_data1), 0, param->ndim_, param->in_shape1_,
-                        param->in_strides1_, param->out_strides_, param->multiples1_);
-}

mindspore/lite/nnacl/arithmetic_common.h

@@ -1,68 +0,0 @@
-/**
- * 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_NNACL_ARITHMETIC_COMMON_H_
-#define MINDSPORE_LITE_NNACL_ARITHMETIC_COMMON_H_
-
-#ifdef ENABLE_NEON
-#include <arm_neon.h>
-#endif
-#include <string.h>
-#include "nnacl/op_base.h"
-#include "nnacl/arithmetic_common.h"
-
-typedef struct ArithmeticParameter {
-  OpParameter op_parameter_;
-  bool broadcasting_;
-  size_t ndim_;
-  int activation_type_;
-  int in_shape0_[10];
-  int in_elements_num0_;
-  int in_shape1_[10];
-  int in_elements_num1_;
-
-  int out_shape_[10];
-  int out_elements_num_;
-
-  int in_strides0_[10];
-  int in_strides1_[10];
-  int out_strides_[10];
-
-  int multiples0_[10];
-  int multiples1_[10];
-} ArithmeticParameter;
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-void TileOneDimension(const float *inData, float *outData, int dim, size_t ndim, const int *inShape,
-                      const int *inStrides, const int *outStrides, const int *multiple);
-void ComputeStrides(const int *shape, int *strides, const int ndim);
-
-void CalcMultiplesAndStrides(ArithmeticParameter *param);
-
-void TileOneDimensionUint8(const uint8_t *inData, uint8_t *outData, int dim, size_t ndim, const int *inShape,
-                           const int *inStrides, const int *outStrides, const int *multiple);
-void TileDimensions(const float *data0, const float *data1, float *tile_data0, float *tile_data1,
-                    ArithmeticParameter *param);
-void TileDimensionsUint8(const uint8_t *data0, const uint8_t *data1, uint8_t *tile_data0, uint8_t *tile_data1,
-                         ArithmeticParameter *param);
-void TileDimensionsInt8(const int8_t *data0, const int8_t *data1, int8_t *tile_data0, int8_t *tile_data1,
-                        ArithmeticParameter *param);
-#ifdef __cplusplus
-}
-#endif
-
-#endif  // MINDSPORE_LITE_NNACL_ARITHMETIC_COMMON_H_

mindspore/lite/nnacl/arithmetic_parameter.h

@@ -1,22 +0,0 @@
-/**
- * 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_NNACL_ARTITHMETIC_PARAMETER_H_
-#define MINDSPORE_LITE_NNACL_ARTITHMETIC_PARAMETER_H_
-
-#include "nnacl/op_attribute.h"
-
-#endif  // MINDSPORE_LITE_NNACL_ARTITHMETIC_PARAMETER_H_

mindspore/lite/nnacl/batch_to_space.c

@@ -15,7 +15,6 @@
  */
 
 #include "nnacl/batch_to_space.h"
-#include "nnacl/arithmetic_common.h"
 
 void BatchToSpaceNoCropForNHWC(const void *input, void *output, const int *in_shape, int out_n, const int *block,
                                int data_size) {

mindspore/lite/nnacl/batch_to_space.h

@@ -15,6 +15,8 @@
  */
 #ifndef MINDSPORE_LITE_NNACL_BATCH_TO_SPACE_H_
 #define MINDSPORE_LITE_NNACL_BATCH_TO_SPACE_H_
+
+#include <string.h>
 #include "nnacl/op_base.h"
 
 #define BATCH_TO_SPACE_BLOCK_SHAPE_SIZE 2

mindspore/lite/nnacl/common_func.h

@@ -63,6 +63,14 @@ static inline int GetStride(int *strides, const int *shape, int length) {
   return stride;
 }
 
+inline void ComputeStrides(const int *shape, int *strides, const int ndim) {
+  int stride = 1;
+  for (int i = ndim - 1; i >= 0; i--) {
+    strides[i] = stride;
+    stride *= shape[i];
+  }
+}
+
 #ifdef ENABLE_ARM64
 void BiasAdd(const float *bias, float *data, size_t oc4, size_t plan_size);
 void BiasAddRelu6(const float *bias, float *data, size_t oc4, size_t plan_size);

mindspore/lite/nnacl/fp16/arithmetic_fp16.c

@@ -16,7 +16,7 @@
 
 #include "nnacl/fp16/arithmetic_fp16.h"
 #include <math.h>
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 #include "nnacl/nnacl_utils.h"
 
 void TileOneDimensionFp16(float16_t *inData, float16_t *outData, int dim, size_t ndim, int *inShape, int *inStrides,

mindspore/lite/nnacl/fp16/arithmetic_fp16.h

@@ -20,7 +20,7 @@
 #include <arm_neon.h>
 #endif
 #include "nnacl/op_base.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 #include "nnacl/errorcode.h"
 
 #ifdef __cplusplus
@@ -107,7 +107,7 @@ int ElementMinimumFp16(float16_t *input0, float16_t *input1, float16_t *output,
 int ElementNotEqualFp16(float16_t *input0, float16_t *input1, uint8_t *output, int element_size);
 int ElementEqualFp16(float16_t *input0, float16_t *input1, uint8_t *output, int element_size);
 int ElementLessFp16(float16_t *input0, float16_t *input1, uint8_t *output, int element_size);
-int ElementLessEqual(float16_t *input0, float16_t *input1, uint8_t *output, int element_size);
+int ElementLessEqualFp16(float16_t *input0, float16_t *input1, uint8_t *output, int element_size);
 int ElementGreaterFp16(float16_t *input0, float16_t *input1, uint8_t *output, int element_size);
 int ElementGreaterEqualFp16(float16_t *input0, float16_t *input1, uint8_t *output, int element_size);
 

mindspore/lite/nnacl/fp16/stack_fp16.c

@@ -15,7 +15,7 @@
  */
 
 #include "nnacl/fp16/stack_fp16.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 
 size_t Fp16GetStackCopyNum(int axis, int *in_shape, size_t shape_size) {
   size_t one_input_size = 1;

mindspore/lite/nnacl/fp32/arithmetic_fp32.h

@@ -20,144 +20,97 @@
 #include <arm_neon.h>
 #endif
 #include "nnacl/op_base.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 #include "nnacl/errorcode.h"
 
 #ifdef __cplusplus
 extern "C" {
 #endif
-int ElementOptAdd(const float *input0, const float *input1, float *output, const int element_size,
-                  const ArithmeticParameter *param);
-int ElementOptAddInt(const int *input0, const int *input1, int *output, const int element_size,
-                     const ArithmeticParameter *param);
-int ElementOptAddRelu(const float *input0, const float *input1, float *output, const int element_size,
-                      const ArithmeticParameter *param);
-int ElementOptAddRelu6(const float *input0, const float *input1, float *output, const int element_size,
-                       const ArithmeticParameter *param);
-int ElementOptSub(const float *input0, const float *input1, float *output, const int element_size,
-                  const ArithmeticParameter *param);
-int ElementOptSubInt(const int *input0, const int *input1, int *output, const int element_size,
-                     const ArithmeticParameter *param);
-int ElementOptSubRelu(const float *input0, const float *input1, float *output, const int element_size,
-                      const ArithmeticParameter *param);
-int ElementOptSubRelu6(const float *input0, const float *input1, float *output, const int element_size,
-                       const ArithmeticParameter *param);
-int ElementOptMul(const float *input0, const float *input1, float *output, const int element_size,
-                  const ArithmeticParameter *param);
-int ElementOptMulRelu(const float *input0, const float *input1, float *output, const int element_size,
-                      const ArithmeticParameter *param);
-int ElementOptMulRelu6(const float *input0, const float *input1, float *output, const int element_size,
-                       const ArithmeticParameter *param);
-int ElementOptMulInt(const int *input0, const int *input1, int *output, const int element_size,
-                     const ArithmeticParameter *param);
-int ElementOptMulReluInt(const int *input0, const int *input1, int *output, const int element_size,
-                         const ArithmeticParameter *param);
-int ElementOptMulRelu6Int(const int *input0, const int *input1, int *output, const int element_size,
-                          const ArithmeticParameter *param);
-int ElementOptDiv(const float *input0, const float *input1, float *output, const int element_size,
-                  const ArithmeticParameter *param);
-int ElementOptDivRelu(const float *input0, const float *input1, float *output, const int element_size,
-                      const ArithmeticParameter *param);
-int ElementOptDivRelu6(const float *input0, const float *input1, float *output, const int element_size,
-                       const ArithmeticParameter *param);
-int ElementOptDivInt(const int *input0, const int *input1, int *output, const int element_size,
-                     const ArithmeticParameter *param);
-int ElementMul(const float *input0, const float *input1, float *output, const int element_size);
-int ElementMulRelu(const float *input0, const float *input1, float *output, const int element_size);
-int ElementMulRelu6(const float *input0, const float *input1, float *output, const int element_size);
-int ElementMulInt(const int *input0, const int *input1, int *output, const int element_size);
-int ElementMulReluInt(const int *input0, const int *input1, int *output, const int element_size);
-int ElementMulRelu6Int(const int *input0, const int *input1, int *output, const int element_size);
-int BroadcastMul(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                 int element_size, ArithmeticParameter *param);
-
-int ElementAdd(const float *input0, const float *input1, float *output, const int element_size);
-int ElementAddRelu(const float *input0, const float *input1, float *output, const int element_size);
-int ElementAddRelu6(const float *input0, const float *input1, float *output, const int element_size);
-int ElementAddInt(const int *input0, const int *input1, int *output, const int element_size);
-int BroadcastAdd(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                 int element_size, ArithmeticParameter *param);
-int BroadcastAddInt8(const int8_t *input0, const int8_t *input1, int8_t *tile_input0, int8_t *tile_input1,
-                     int8_t *output, int element_size, ArithmeticParameter *param);
-
-int ElementSub(const float *input0, const float *input1, float *output, const int element_size);
-int ElementSubInt(const int *input0, const int *input1, int *output, const int element_size);
-int ElementSubRelu(const float *input0, const float *input1, float *output, const int element_size);
-int ElementSubRelu6(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastSub(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                 int element_size, ArithmeticParameter *param);
-
-int ElementDiv(const float *input0, const float *input1, float *output, const int element_size);
-int ElementDivRelu(const float *input0, const float *input1, float *output, const int element_size);
-int ElementDivRelu6(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastDiv(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                 int element_size, ArithmeticParameter *param);
-
-int ElementLogicalAnd(const float *input0, const float *input1, float *output, const int element_size);
-int ElementLogicalAndInt(const int *input0, const int *input1, int *output, const int element_size);
-int ElementLogicalAndBool(const bool *input0, const bool *input1, bool *output, const int element_size);
-int BroadcastLogicalAnd(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                        int element_size, ArithmeticParameter *param);
-
-int ElementLogicalOr(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastLogicalOr(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                       int element_size, ArithmeticParameter *param);
-
-int ElementMaximum(const float *input0, const float *input1, float *output, const int element_size);
-int ElementMaximumInt(const int *input0, const int *input1, int *output, const int element_size);
-int BroadcastMaximum(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                     int element_size, ArithmeticParameter *param);
-
-int ElementMinimum(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastMinimum(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                     int element_size, ArithmeticParameter *param);
-
-int ElementFloorDiv(const float *input0, const float *input1, float *output, const int element_size);
-int ElementFloorDivInt(const int *input0, const int *input1, int *output, const int element_size);
-int BroadcastFloorDiv(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                      int element_size, ArithmeticParameter *param);
-
-int ElementFloorMod(const float *input0, const float *input1, float *output, const int element_size);
-int ElementFloorModInt(const int *input0, const int *input1, int *output, const int element_size);
-int BroadcastFloorMod(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                      int element_size, ArithmeticParameter *param);
-
-int ElementMod(const float *input0, const float *input1, float *output, const int element_size);
-int ElementModInt(const int *input0, const int *input1, int *output, const int element_size);
-int ElementOptMod(const float *input0, const float *input1, float *output, const int element_size,
-                  const ArithmeticParameter *param);
-int ElementOptModInt(const int *input0, const int *input1, int *output, const int element_size,
-                     const ArithmeticParameter *param);
-
-int ElementSquaredDifference(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastSquaredDifference(const float *input0, const float *input1, float *tile_input0, float *tile_input1,
-                               float *output, int element_size, ArithmeticParameter *param);
-
-int ElementNotEqual(const float *input0, const float *input1, float *output, const int element_size);
-
-int BroadcastNotEqual(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                      int element_size, ArithmeticParameter *param);
-
-int ElementEqual(const float *input0, const float *input1, float *output, const int element_size);
-
-int BroadcastEqual(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                   int element_size, ArithmeticParameter *param);
-
-int ElementLess(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastLess(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                  int element_size, ArithmeticParameter *param);
-
-int ElementLessEqual(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastLessEqual(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                       int element_size, ArithmeticParameter *param);
-
-int ElementGreater(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastGreater(const float *input0, const float *input1, float *tile_input0, float *tile_input1, float *output,
-                     int element_size, ArithmeticParameter *param);
-
-int ElementGreaterEqual(const float *input0, const float *input1, float *output, const int element_size);
-int BroadcastGreaterEqual(const float *input0, const float *input1, float *tile_input0, float *tile_input1,
-                          float *output, int element_size, ArithmeticParameter *param);
+void TileOneDimensionFp32(const float *inData, float *outData, int dim, size_t ndim, const int *inShape,
+                          const int *inStrides, const int *outStrides, const int *multiple);
+void TileDimensionsFp32(const float *data0, const float *data1, float *tile_data0, float *tile_data1,
+                        ArithmeticParameter *param);
+
+/* Mul */
+int ElementMul(const float *in0, const float *in1, float *out, int size);
+int ElementMulRelu(const float *in0, const float *in1, float *out, int size);
+int ElementMulRelu6(const float *in0, const float *in1, float *out, int size);
+int ElementMulInt(const int *in0, const int *in1, int *out, int size);
+int ElementMulReluInt(const int *in0, const int *in1, int *out, int size);
+int ElementMulRelu6Int(const int *in0, const int *in1, int *out, int size);
+int ElementOptMul(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptMulRelu(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptMulRelu6(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptMulInt(const int *in0, const int *in1, int *out, int size, const ArithmeticParameter *param);
+int ElementOptMulReluInt(const int *in0, const int *in1, int *out, int size, const ArithmeticParameter *param);
+int ElementOptMulRelu6Int(const int *in0, const int *in1, int *out, int size, const ArithmeticParameter *param);
+int BroadcastMul(const float *in0, const float *in1, float *tile_in0, float *tile_in1, float *out, int size,
+                 ArithmeticParameter *param);
+
+/* Add */
+int ElementAdd(const float *in0, const float *in1, float *out, int size);
+int ElementAddRelu(const float *in0, const float *in1, float *out, int size);
+int ElementAddRelu6(const float *in0, const float *in1, float *out, int size);
+int ElementAddInt(const int *in0, const int *in1, int *out, int size);
+int ElementOptAdd(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptAddInt(const int *in0, const int *in1, int *out, int size, const ArithmeticParameter *param);
+int ElementOptAddRelu(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptAddRelu6(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int BroadcastAdd(const float *in0, const float *in1, float *tile_in0, float *tile_in1, float *out, int size,
+                 ArithmeticParameter *param);
+
+/* Sub */
+int ElementSub(const float *in0, const float *in1, float *out, int size);
+int ElementSubInt(const int *in0, const int *in1, int *out, int size);
+int ElementSubRelu(const float *in0, const float *in1, float *out, int size);
+int ElementSubRelu6(const float *in0, const float *in1, float *out, int size);
+int ElementOptSub(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptSubRelu(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptSubRelu6(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptSubInt(const int *in0, const int *in1, int *out, int size, const ArithmeticParameter *param);
+
+/* Div */
+int ElementDiv(const float *in0, const float *in1, float *out, int size);
+int ElementDivRelu(const float *in0, const float *in1, float *out, int size);
+int ElementDivRelu6(const float *in0, const float *in1, float *out, int size);
+int ElementOptDiv(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptDivRelu(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptDivRelu6(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptDivInt(const int *in0, const int *in1, int *out, int size, const ArithmeticParameter *param);
+int BroadcastDiv(const float *in0, const float *in1, float *tile_in0, float *tile_in1, float *out, int size,
+                 ArithmeticParameter *param);
+
+/* logical and */
+int ElementLogicalAnd(const float *in0, const float *in1, float *out, int size);
+int ElementLogicalAndInt(const int *in0, const int *in1, int *out, int size);
+int ElementLogicalAndBool(const bool *in0, const bool *in1, bool *out, int size);
+
+/* logical or */
+int ElementLogicalOr(const float *in0, const float *in1, float *out, int size);
+
+/* Element Squared Difference */
+int ElementSquaredDifference(const float *in0, const float *in1, float *out, int size);
+
+/* max min */
+int ElementMaximum(const float *in0, const float *in1, float *out, int size);
+int ElementMinimum(const float *in0, const float *in1, float *out, int size);
+int ElementMaximumInt(const int *in0, const int *in1, int *out, int size);
+int BroadcastMaximum(const float *in0, const float *in1, float *tile_input0, float *tile_input1, float *out, int size,
+                     ArithmeticParameter *param);
+
+/* floor div */
+int ElementFloorDiv(const float *in0, const float *in1, float *out, int size);
+int ElementFloorDivInt(const int *in0, const int *in1, int *out, int size);
+
+/* floor mod */
+int ElementFloorMod(const float *in0, const float *in1, float *out, int size);
+int ElementFloorModInt(const int *in0, const int *in1, int *out, int size);
+
+/* mod */
+int ElementMod(const float *in0, const float *in1, float *out, int size);
+int ElementModInt(const int *in0, const int *in1, int *out, int size);
+int ElementOptMod(const float *in0, const float *in1, float *out, int size, const ArithmeticParameter *param);
+int ElementOptModInt(const int *in0, const int *in1, int *out, int size, const ArithmeticParameter *param);
 
 #ifdef ENABLE_NNACL_INFER_SHAPE
 int ArithmeticInferShape(int **in_shape, size_t *dim_size, int *out_shape, int *in_format, int *out_format,

mindspore/lite/nnacl/fp32/space_to_batch_fp32.c

@@ -14,7 +14,6 @@
  * limitations under the License.
  */
 #include "nnacl/fp32/space_to_batch_fp32.h"
-#include "nnacl/arithmetic_common.h"
 
 void DoSpaceToBatch(const float *input, float *output, const int *in_shape, const int *out_shape, const int *in_stride,
                     const int *out_stride, const int *blocks, const int *paddings, int thread, int task_id) {

mindspore/lite/nnacl/fp32/space_to_batch_fp32.h

@@ -15,6 +15,8 @@
  */
 #ifndef MINDSPORE_LITE_SRC_BACKEND_ARM_NNACL_FP32_SPACE_TO_BATCH_H_
 #define MINDSPORE_LITE_SRC_BACKEND_ARM_NNACL_FP32_SPACE_TO_BATCH_H_
+
+#include <string.h>
 #include "nnacl/op_base.h"
 
 typedef struct SpaceToBatchParameter {

mindspore/lite/nnacl/fp32/space_to_depth_fp32.c

@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 #include "nnacl/fp32/space_to_depth_fp32.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 #include "nnacl/errorcode.h"
 #include "nnacl/op_base.h"
 

mindspore/lite/nnacl/fp32/stack_fp32.c

@@ -15,7 +15,7 @@
  */
 
 #include "nnacl/fp32/stack_fp32.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 
 size_t GetStackCopyNum(int axis, const int *in_shape, size_t shape_size) {
   size_t one_input_size = 1;

mindspore/lite/nnacl/int8/add_int8.c

@@ -312,3 +312,16 @@ void AddOptInt8(const int8_t *ptr_in, const int8_t element_in, int8_t *output, i
   }
   return;
 }
+
+int ElementAddInt8(const int8_t *in0, const int8_t *in1, int8_t *out, int size) {
+  for (int i = 0; i < size; i++) {
+    out[i] = in0[i] + in1[i];
+  }
+  return NNACL_OK;
+}
+
+int BroadcastAddInt8(const int8_t *in0, const int8_t *in1, int8_t *tile_in0, int8_t *tile_in1, int8_t *out, int size,
+                     ArithmeticParameter *param) {
+  TileDimensionsInt8(in0, in1, tile_in0, tile_in1, param);
+  return ElementAddInt8(tile_in0, tile_in1, out, size);
+}

mindspore/lite/nnacl/int8/add_int8.h

@@ -18,6 +18,9 @@
 #define MINDSPORE_LITE_NNACL_ADD_INT8_H_
 
 #include "nnacl/op_base.h"
+#include "nnacl/errorcode.h"
+#include "nnacl/arithmetic.h"
+#include "nnacl/int8/arithmetic_int8.h"
 
 typedef struct AddQuantQrgs {
   int32_t zp_;
@@ -48,6 +51,10 @@ void AddInt8(const int8_t *input0, const int8_t *input1, int8_t *output, int siz
 void AddOptInt8(const int8_t *ptr_in, const int8_t element_in, int8_t *output, int size, AddQuantParameter *params,
                 AddQuantQrgs *ptr_args, AddQuantQrgs *ele_args);
 
+int ElementAddInt8(const int8_t *in0, const int8_t *in1, int8_t *out, int size);
+int BroadcastAddInt8(const int8_t *in0, const int8_t *in1, int8_t *tile_in0, int8_t *tile_in1, int8_t *out, int size,
+                     ArithmeticParameter *param);
+
 #ifdef __cplusplus
 }
 #endif

mindspore/lite/nnacl/int8/arithmetic_int8.c

@@ -20,6 +20,33 @@
 #endif
 #include "nnacl/errorcode.h"
 
+void TileOneDimensionInt8(const int8_t *inData, int8_t *outData, int dim, size_t ndim, const int *inShape,
+                          const int *inStrides, const int *outStrides, const int *multiple) {
+  int srcDimSize = inShape[dim];
+  if (dim == ndim - 1) {
+    for (int i = 0; i < multiple[dim]; i++) {
+      memcpy(outData, inData, srcDimSize * sizeof(int8_t));
+      outData += srcDimSize;
+    }
+    return;
+  }
+  for (size_t i = 0; i < srcDimSize; i++) {
+    for (size_t j = 0; j < multiple[dim]; j++) {
+      TileOneDimensionInt8(inData + inStrides[dim] * i, outData + outStrides[dim] * (i + j * srcDimSize), dim + 1, ndim,
+                           inShape, inStrides, outStrides, multiple);
+    }
+  }
+}
+
+void TileDimensionsInt8(const int8_t *data0, const int8_t *data1, int8_t *tile_data0, int8_t *tile_data1,
+                        ArithmeticParameter *param) {
+  CalcMultiplesAndStrides(param);
+  TileOneDimensionInt8(data0, tile_data0, 0, param->ndim_, param->in_shape0_, param->in_strides0_, param->out_strides_,
+                       param->multiples0_);
+  TileOneDimensionInt8(data1, tile_data1, 0, param->ndim_, param->in_shape1_, param->in_strides1_, param->out_strides_,
+                       param->multiples1_);
+}
+
 #define ACCURACY_DATA 0.00000001
 
 int ElementNotEqualInt8(int8_t *input0, int8_t *input1, uint8_t *output, int element_size,

mindspore/lite/nnacl/int8/arithmetic_int8.h

@@ -17,11 +17,17 @@
 #define MINDSPORE_LITE_NNACL_INT8_ARITHMETIC_INT8_H_
 
 #include "nnacl/op_base.h"
+#include "nnacl/arithmetic.h"
 #include "nnacl/quantization/quantize.h"
 
 #ifdef __cplusplus
 extern "C" {
 #endif
+void TileOneDimensionInt8(const int8_t *inData, int8_t *outData, int dim, size_t ndim, const int *inShape,
+                          const int *inStrides, const int *outStrides, const int *multiple);
+void TileDimensionsInt8(const int8_t *data0, const int8_t *data1, int8_t *tile_data0, int8_t *tile_data1,
+                        ArithmeticParameter *param);
+
 int ElementNotEqualInt8(int8_t *input0, int8_t *input1, uint8_t *output, int element_size,
                         ArithmeticQuantArg *quant_arg);
 

mindspore/lite/nnacl/int8/space_to_batch_int8.c

@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 #include "nnacl/int8/space_to_batch_int8.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 
 void DoSpaceToBatchNHWCInt8(const int8_t *input, int8_t *output, const int *block_sizes, const int *in_shape,
                             const int *out_shape) {

mindspore/lite/nnacl/reverse_sequence.c

@@ -16,7 +16,7 @@
 
 #include "nnacl/reverse_sequence.h"
 #include <string.h>
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 
 void ReverseSequence(float *input0, const void *input1, float *output, ReverseSequenceParameter *para) {
   (void)memcpy(output, input0, para->total_data_size_);

mindspore/lite/src/ops/arithmetic.h

@@ -21,7 +21,7 @@
 #include <set>
 #include <cmath>
 #include "src/ops/primitive_c.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 
 namespace mindspore {
 namespace lite {

mindspore/lite/src/ops/populate/add_populate.cc

@@ -17,7 +17,7 @@
 #include "src/ops/add.h"
 #include "src/ops/primitive_c.h"
 #include "src/ops/populate/populate_register.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 #include "src/ops/populate/arithmetic_populate.h"
 
 namespace mindspore {

mindspore/lite/src/ops/populate/bias_add_populate.cc

@@ -16,7 +16,7 @@
 
 #include "src/ops/primitive_c.h"
 #include "src/ops/populate/populate_register.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 
 namespace mindspore {
 namespace lite {

mindspore/lite/src/ops/populate/bias_grad_populate.cc

@@ -16,7 +16,7 @@
 
 #include "src/ops/primitive_c.h"
 #include "src/ops/populate/populate_register.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 
 namespace mindspore {
 namespace lite {

mindspore/lite/src/ops/populate/flatten_populate.cc

@@ -16,18 +16,17 @@
 
 #include "src/ops/primitive_c.h"
 #include "src/ops/populate/populate_register.h"
-#include "nnacl/flatten.h"
 
 namespace mindspore {
 namespace lite {
 OpParameter *PopulateFlattenParameter(const mindspore::lite::PrimitiveC *primitive) {
-  FlattenParameter *flatten_param = reinterpret_cast<FlattenParameter *>(malloc(sizeof(FlattenParameter)));
+  OpParameter *flatten_param = reinterpret_cast<OpParameter *>(malloc(sizeof(OpParameter)));
   if (flatten_param == nullptr) {
     MS_LOG(ERROR) << "malloc FlattenParameter failed.";
     return nullptr;
   }
-  memset(flatten_param, 0, sizeof(FlattenParameter));
-  flatten_param->op_parameter_.type_ = primitive->Type();
+  memset(flatten_param, 0, sizeof(OpParameter));
+  flatten_param->type_ = primitive->Type();
   return reinterpret_cast<OpParameter *>(flatten_param);
 }
 

mindspore/lite/src/ops/populate/mul_populate.cc

@@ -15,7 +15,7 @@
  */
 
 #include "src/ops/mul.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 #include "src/ops/primitive_c.h"
 #include "src/ops/populate/populate_register.h"
 #include "src/ops/populate/arithmetic_populate.h"

mindspore/lite/src/ops/populate/squeeze_populate.cc

@@ -23,13 +23,13 @@ namespace mindspore {
 namespace lite {
 
 OpParameter *PopulateSqueezeParameter(const mindspore::lite::PrimitiveC *primitive) {
-  SqueezeParameter *squeeze_param = reinterpret_cast<SqueezeParameter *>(malloc(sizeof(SqueezeParameter)));
+  OpParameter *squeeze_param = reinterpret_cast<OpParameter *>(malloc(sizeof(OpParameter)));
   if (squeeze_param == nullptr) {
     MS_LOG(ERROR) << "malloc SqueezeParameter failed.";
     return nullptr;
   }
-  memset(squeeze_param, 0, sizeof(SqueezeParameter));
-  squeeze_param->op_parameter_.type_ = primitive->Type();
+  memset(squeeze_param, 0, sizeof(OpParameter));
+  squeeze_param->type_ = primitive->Type();
   return reinterpret_cast<OpParameter *>(squeeze_param);
 }
 Registry SqueezeParameterRegistry(schema::PrimitiveType_Squeeze, PopulateSqueezeParameter);

mindspore/lite/src/ops/populate/sub_populate.cc

@@ -17,7 +17,7 @@
 #include "src/ops/sub.h"
 #include "src/ops/primitive_c.h"
 #include "src/ops/populate/populate_register.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 #include "src/ops/populate/arithmetic_populate.h"
 
 namespace mindspore {

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

@@ -16,7 +16,7 @@
 #include "src/runtime/kernel/arm/base/depth_to_space_base.h"
 #include "nnacl/depth_to_space.h"
 #include "src/runtime/kernel/arm/fp32/depth_to_space_fp32.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 #include "schema/model_generated.h"
 #include "src/kernel_registry.h"
 #include "include/errorcode.h"

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

@@ -39,7 +39,7 @@ ARITHMETIC_COMPARE_FUNC_INFO_FP16 arithmetic_cp_fun_table_fp16[] = {
   {PrimitiveType_NotEqual, schema::ActivationType_NO_ACTIVATION, ElementNotEqualFp16, ElementOptNotEqualFp16},
   {PrimitiveType_Equal, schema::ActivationType_NO_ACTIVATION, ElementEqualFp16, ElementOptEqualFp16},
   {PrimitiveType_Less, schema::ActivationType_NO_ACTIVATION, ElementLessFp16, ElementOptLessFp16},
-  {PrimitiveType_LessEqual, schema::ActivationType_NO_ACTIVATION, ElementLessEqual, ElementOptLessEqualFp16},
+  {PrimitiveType_LessEqual, schema::ActivationType_NO_ACTIVATION, ElementLessEqualFp16, ElementOptLessEqualFp16},
   {PrimitiveType_Greater, schema::ActivationType_NO_ACTIVATION, ElementGreaterFp16, ElementOptGreaterFp16},
   {PrimitiveType_GreaterEqual, schema::ActivationType_NO_ACTIVATION, ElementGreaterEqualFp16,
    ElementOptGreaterEqualFp16}};

mindspore/lite/src/runtime/kernel/arm/fp32/argminmax_fp32.h

@@ -19,7 +19,7 @@
 #include <vector>
 #include "include/errorcode.h"
 #include "nnacl/fp32/arg_min_max_fp32.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 #include "src/lite_kernel.h"
 
 namespace mindspore::kernel {

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

@@ -29,7 +29,6 @@ using mindspore::lite::RET_OK;
 using mindspore::schema::PrimitiveType_Eltwise;
 
 namespace mindspore::kernel {
-
 ArithmeticCPUKernel::~ArithmeticCPUKernel() {
   FreeTmpPtr();
   return;
@@ -72,9 +71,10 @@ int ArithmeticCPUKernel::InitBroadCastCase() {
     if (input0_ptr_ == nullptr) {
       return RET_ERROR;
     }
-    TileOneDimension(reinterpret_cast<float *>(in_tensors_[0]->data_c()), reinterpret_cast<float *>(input0_ptr_), 0,
-                     arithmeticParameter_->ndim_, arithmeticParameter_->in_shape0_, arithmeticParameter_->in_strides0_,
-                     arithmeticParameter_->out_strides_, arithmeticParameter_->multiples0_);
+    TileOneDimensionFp32(reinterpret_cast<float *>(in_tensors_[0]->data_c()), reinterpret_cast<float *>(input0_ptr_), 0,
+                         arithmeticParameter_->ndim_, arithmeticParameter_->in_shape0_,
+                         arithmeticParameter_->in_strides0_, arithmeticParameter_->out_strides_,
+                         arithmeticParameter_->multiples0_);
     arithmeticParameter_->broadcasting_ = false;
     input0_broadcast_ = true;
   }
@@ -85,9 +85,10 @@ int ArithmeticCPUKernel::InitBroadCastCase() {
       FreeTmpPtr();
       return RET_ERROR;
     }
-    TileOneDimension(reinterpret_cast<float *>(in_tensors_[1]->data_c()), reinterpret_cast<float *>(input1_ptr_), 0,
-                     arithmeticParameter_->ndim_, arithmeticParameter_->in_shape1_, arithmeticParameter_->in_strides1_,
-                     arithmeticParameter_->out_strides_, arithmeticParameter_->multiples1_);
+    TileOneDimensionFp32(reinterpret_cast<float *>(in_tensors_[1]->data_c()), reinterpret_cast<float *>(input1_ptr_), 0,
+                         arithmeticParameter_->ndim_, arithmeticParameter_->in_shape1_,
+                         arithmeticParameter_->in_strides1_, arithmeticParameter_->out_strides_,
+                         arithmeticParameter_->multiples1_);
     arithmeticParameter_->broadcasting_ = false;
     input1_broadcast_ = true;
   }

mindspore/lite/src/runtime/kernel/arm/fp32/depth_to_space_fp32.h

@@ -18,7 +18,6 @@
 
 #include <vector>
 #include "include/errorcode.h"
-#include "nnacl/arithmetic_common.h"
 #include "nnacl/depth_to_space.h"
 #include "src/runtime/kernel/arm/base/depth_to_space_base.h"
 

mindspore/lite/src/runtime/kernel/arm/fp32/flatten_fp32.cc

@@ -17,7 +17,6 @@
 #include "src/runtime/kernel/arm/fp32/flatten_fp32.h"
 #include "schema/model_generated.h"
 #include "src/kernel_registry.h"
-#include "nnacl/flatten.h"
 #include "include/errorcode.h"
 
 using mindspore::kernel::KERNEL_ARCH::kCPU;
@@ -34,19 +33,12 @@ int FlattenCPUKernel::Init() {
   return ReSize();
 }
 
-int FlattenCPUKernel::ReSize() {
-  auto output_shape = out_tensors_.at(0)->shape();
-  flatten_param_->size = sizeof(float);
-  for (size_t i = 0; i < output_shape.size(); i++) {
-    flatten_param_->size *= output_shape.at(i);
-  }
-  return RET_OK;
-}
+int FlattenCPUKernel::ReSize() { return RET_OK; }
 
 int FlattenCPUKernel::Run() {
-  auto input = reinterpret_cast<float *>(in_tensors_.at(0)->MutableData());
-  auto output = reinterpret_cast<float *>(out_tensors_.at(0)->MutableData());
-  Flatten(input, output, flatten_param_);
+  auto input = in_tensors_.at(0);
+  auto output = out_tensors_.at(0);
+  memcpy(output->data_c(), input->data_c(), output->Size());
   return RET_OK;
 }
 

mindspore/lite/src/runtime/kernel/arm/fp32/flatten_fp32.h

@@ -18,9 +18,7 @@
 
 #include <vector>
 #include "src/lite_kernel.h"
-
 #include "include/context.h"
-#include "nnacl/flatten.h"
 
 using mindspore::lite::InnerContext;
 
@@ -30,17 +28,12 @@ class FlattenCPUKernel : public LiteKernel {
   FlattenCPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,
                    const std::vector<lite::Tensor *> &outputs, const lite::InnerContext *ctx,
                    const mindspore::lite::PrimitiveC *primitive)
-      : LiteKernel(parameter, inputs, outputs, ctx, primitive) {
-    flatten_param_ = reinterpret_cast<FlattenParameter *>(parameter);
-  }
+      : LiteKernel(parameter, inputs, outputs, ctx, primitive) {}
   ~FlattenCPUKernel() override = default;
 
   int Init() override;
   int ReSize() override;
   int Run() override;
-
- private:
-  FlattenParameter *flatten_param_;
 };
 }  // namespace mindspore::kernel
 

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

@@ -19,7 +19,7 @@
 #include <vector>
 #include "src/lite_kernel.h"
 #include "nnacl/fp32/space_to_batch_fp32.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 
 namespace mindspore::kernel {
 class SpaceToBatchCPUKernel : public LiteKernel {

mindspore/lite/src/runtime/kernel/arm/int8/add_int8.cc

@@ -15,9 +15,6 @@
  */
 
 #include "src/runtime/kernel/arm/int8/add_int8.h"
-#include <limits>
-#include <algorithm>
-#include "nnacl/arithmetic_common.h"
 #include "nnacl/quantization/quantize.h"
 #include "src/runtime/runtime_api.h"
 #include "src/kernel_registry.h"

mindspore/lite/src/runtime/kernel/arm/int8/add_int8.h

@@ -17,9 +17,11 @@
 #define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_ADD_INT8_H_
 
 #include <vector>
+#include <limits>
+#include <algorithm>
 #include "src/lite_kernel.h"
 #include "nnacl/int8/add_int8.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 #include "src/runtime/runtime_api.h"
 
 namespace mindspore::kernel {

mindspore/lite/src/runtime/kernel/arm/int8/argminmax_int8.h

@@ -19,7 +19,7 @@
 #include <vector>
 #include "nnacl/quantization/quantize.h"
 #include "nnacl/int8/arg_min_max_int8.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/common_func.h"
 #include "include/errorcode.h"
 #include "src/lite_kernel.h"
 

mindspore/lite/src/runtime/kernel/arm/int8/arithmetic_int8.cc

@@ -17,7 +17,7 @@
 #include "src/runtime/kernel/arm/int8/arithmetic_int8.h"
 #include "src/runtime/kernel/arm/int8/add_int8.h"
 #include "src/runtime/kernel/arm/int8/mul_int8.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 #include "schema/model_generated.h"
 #include "src/kernel_registry.h"
 #include "src/runtime/runtime_api.h"

mindspore/lite/src/runtime/kernel/arm/int8/bias_add_int8.h

@@ -18,8 +18,9 @@
 
 #include <vector>
 #include "src/lite_kernel.h"
-#include "nnacl/fp32/unique_fp32.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
+#include "nnacl/int8/add_int8.h"
+#include "nnacl/int8/arithmetic_int8.h"
 
 namespace mindspore::kernel {
 class BiasAddInt8CPUKernel : public LiteKernel {

mindspore/lite/src/runtime/kernel/arm/int8/deconvolution_int8.h

@@ -27,7 +27,6 @@
 #include "nnacl/int8/matmul_int8.h"
 #include "src/runtime/kernel/arm/base/layout_transform.h"
 #include "src/runtime/kernel/arm/base/convolution_base.h"
-#include "nnacl/arithmetic_common.h"
 
 namespace mindspore::kernel {
 class DeConvInt8CPUKernel : public ConvolutionBaseCPUKernel {

mindspore/lite/src/runtime/kernel/arm/int8/div_int8.cc

@@ -17,7 +17,7 @@
 #include "src/runtime/kernel/arm/int8/div_int8.h"
 #include <limits>
 #include <algorithm>
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/int8/arithmetic_int8.h"
 #include "src/runtime/runtime_api.h"
 #include "src/kernel_registry.h"
 #include "include/errorcode.h"
@@ -114,9 +114,9 @@ int DivInt8CPUKernel::Run() {
       tile1_data_ = nullptr;
       return RET_ERROR;
     }
-    TileDimensionsUint8(static_cast<uint8_t *>(in_tensors_.at(0)->MutableData()),
-                        static_cast<uint8_t *>(in_tensors_.at(1)->MutableData()),
-                        reinterpret_cast<uint8_t *>(tile0_data_), reinterpret_cast<uint8_t *>(tile1_data_), &tile_para);
+    TileDimensionsInt8(static_cast<int8_t *>(in_tensors_.at(0)->MutableData()),
+                       static_cast<int8_t *>(in_tensors_.at(1)->MutableData()), reinterpret_cast<int8_t *>(tile0_data_),
+                       reinterpret_cast<int8_t *>(tile1_data_), &tile_para);
   }
   auto ret = ParallelLaunch(this->context_->thread_pool_, DivInt8Run, this, op_parameter_->thread_num_);
   if (broadcast_) {

mindspore/lite/src/runtime/kernel/arm/int8/mul_int8.cc

@@ -15,10 +15,6 @@
  */
 
 #include "src/runtime/kernel/arm/int8/mul_int8.h"
-#include <limits>
-#include <algorithm>
-#include "nnacl/arithmetic_common.h"
-#include "nnacl/int8/mul_int8.h"
 #include "src/runtime/runtime_api.h"
 #include "src/kernel_registry.h"
 #include "include/errorcode.h"

mindspore/lite/src/runtime/kernel/arm/int8/mul_int8.h

@@ -17,9 +17,12 @@
 #define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_MUL_INT8_H_
 
 #include <vector>
+#include <limits>
+#include <algorithm>
 #include "src/lite_kernel.h"
 #include "nnacl/mul_parameter.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/int8/mul_int8.h"
+#include "nnacl/int8/arithmetic_int8.h"
 #include "src/runtime/runtime_api.h"
 
 namespace mindspore::kernel {

mindspore/lite/src/runtime/kernel/arm/int8/scale_int8.cc

@@ -15,11 +15,6 @@
  */
 
 #include "src/runtime/kernel/arm/int8/scale_int8.h"
-
-#include <string.h>
-#include <vector>
-#include "nnacl/int8/scale_int8.h"
-#include "nnacl/arithmetic_common.h"
 #include "schema/model_generated.h"
 #include "src/kernel_registry.h"
 #include "include/errorcode.h"
@@ -66,9 +61,9 @@ int ScaleInt8CPUKernel::InitScaleOffset() {
         return RET_ERROR;
       }
       malloced_scale_ = true;
-      TileOneDimensionUint8(reinterpret_cast<uint8_t *>(in_tensors_.at(1)->data_c()),
-                            reinterpret_cast<uint8_t *>(input1_data_), 0, tile_para->ndim_, tile_para->in_shape1_,
-                            tile_para->in_strides1_, tile_para->out_strides_, tile_para->multiples1_);
+      TileOneDimensionInt8(reinterpret_cast<int8_t *>(in_tensors_.at(1)->data_c()),
+                           reinterpret_cast<int8_t *>(input1_data_), 0, tile_para->ndim_, tile_para->in_shape1_,
+                           tile_para->in_strides1_, tile_para->out_strides_, tile_para->multiples1_);
     }
   }
 
@@ -93,9 +88,9 @@ int ScaleInt8CPUKernel::InitScaleOffset() {
           return RET_ERROR;
         }
         malloced_offset_ = true;
-        TileOneDimensionUint8(reinterpret_cast<uint8_t *>(in_tensors_.at(2)->data_c()),
-                              reinterpret_cast<uint8_t *>(input2_data_), 0, tile_para->ndim_, tile_para->in_shape1_,
-                              tile_para->in_strides1_, tile_para->out_strides_, tile_para->multiples1_);
+        TileOneDimensionInt8(reinterpret_cast<int8_t *>(in_tensors_.at(2)->data_c()),
+                             reinterpret_cast<int8_t *>(input2_data_), 0, tile_para->ndim_, tile_para->in_shape1_,
+                             tile_para->in_strides1_, tile_para->out_strides_, tile_para->multiples1_);
       }
     }
   }
@@ -305,9 +300,9 @@ int ScaleInt8CPUKernel::Run() {
         MS_LOG(ERROR) << "malloc input1_data_  failed.";
         return RET_ERROR;
       }
-      TileOneDimensionUint8(reinterpret_cast<uint8_t *>(in_tensors_.at(1)->data_c()),
-                            reinterpret_cast<uint8_t *>(input1_data_), 0, tile_para->ndim_, tile_para->in_shape1_,
-                            tile_para->in_strides1_, tile_para->out_strides_, tile_para->multiples1_);
+      TileOneDimensionInt8(reinterpret_cast<int8_t *>(in_tensors_.at(1)->data_c()),
+                           reinterpret_cast<int8_t *>(input1_data_), 0, tile_para->ndim_, tile_para->in_shape1_,
+                           tile_para->in_strides1_, tile_para->out_strides_, tile_para->multiples1_);
     }
 
     // If has bias, bias is passed by previous node case, need do broadcasting online
@@ -319,9 +314,9 @@ int ScaleInt8CPUKernel::Run() {
         input1_data_ = nullptr;
         return RET_ERROR;
       }
-      TileOneDimensionUint8(reinterpret_cast<uint8_t *>(in_tensors_.at(2)->data_c()),
-                            reinterpret_cast<uint8_t *>(input2_data_), 0, tile_para->ndim_, tile_para->in_shape1_,
-                            tile_para->in_strides1_, tile_para->out_strides_, tile_para->multiples1_);
+      TileOneDimensionInt8(reinterpret_cast<int8_t *>(in_tensors_.at(2)->data_c()),
+                           reinterpret_cast<int8_t *>(input2_data_), 0, tile_para->ndim_, tile_para->in_shape1_,
+                           tile_para->in_strides1_, tile_para->out_strides_, tile_para->multiples1_);
     }
 
     auto ret = ParallelLaunch(this->context_->thread_pool_, ScaleRunInt8, this, op_parameter_->thread_num_);

mindspore/lite/src/runtime/kernel/arm/int8/scale_int8.h

@@ -17,14 +17,15 @@
 #ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_SCALE_INT8_H_
 #define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_SCALE_INT8_H_
 
+#include <string.h>
 #include <vector>
 #include "src/lite_kernel.h"
 #include "nnacl/scale.h"
 #include "nnacl/quantization/quantize.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/int8/arithmetic_int8.h"
+#include "nnacl/int8/scale_int8.h"
 
 namespace mindspore::kernel {
-
 class ScaleInt8CPUKernel : public LiteKernel {
  public:
   ScaleInt8CPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,

mindspore/lite/src/runtime/kernel/arm/int8/sub_int8.cc

@@ -15,10 +15,6 @@
  */
 
 #include "src/runtime/kernel/arm/int8/sub_int8.h"
-#include <limits>
-#include <algorithm>
-#include "nnacl/arithmetic_common.h"
-#include "nnacl/quantization/quantize.h"
 #include "src/runtime/runtime_api.h"
 #include "src/kernel_registry.h"
 #include "include/errorcode.h"
@@ -29,7 +25,6 @@ using mindspore::lite::RET_OK;
 using mindspore::schema::PrimitiveType_Sub;
 
 namespace mindspore::kernel {
-
 int SubInt8CPUKernel::Init() {
   lite::Tensor *input0 = in_tensors_.at(0);
   lite::Tensor *input1 = in_tensors_.at(1);
@@ -142,9 +137,9 @@ int SubInt8CPUKernel::Run() {
       context_->allocator->Free(tile0_data_);
       return RET_ERROR;
     }
-    TileDimensionsUint8(static_cast<uint8_t *>(in_tensors_.at(0)->MutableData()),
-                        static_cast<uint8_t *>(in_tensors_.at(1)->MutableData()),
-                        reinterpret_cast<uint8_t *>(tile0_data_), reinterpret_cast<uint8_t *>(tile1_data_), &tile_para);
+    TileDimensionsInt8(static_cast<int8_t *>(in_tensors_.at(0)->data_c()),
+                       static_cast<int8_t *>(in_tensors_.at(1)->data_c()), reinterpret_cast<int8_t *>(tile0_data_),
+                       reinterpret_cast<int8_t *>(tile1_data_), &tile_para);
   }
   auto ret = ParallelLaunch(this->context_->thread_pool_, SubInt8Run, this, op_parameter_->thread_num_);
   if (broadcast_) {

mindspore/lite/src/runtime/kernel/arm/int8/sub_int8.h

@@ -17,6 +17,10 @@
 #define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_INT8_SUB_INT8_H_
 
 #include <vector>
+#include <limits>
+#include <algorithm>
+#include "nnacl/int8/arithmetic_int8.h"
+#include "nnacl/quantization/quantize.h"
 #include "src/lite_kernel.h"
 #include "nnacl/int8/sub_int8.h"
 #include "src/runtime/runtime_api.h"

mindspore/lite/src/runtime/kernel/npu/eltwise_npu.h

@@ -18,7 +18,6 @@
 #define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_NPU_ELTWISE_NPU_H_
 #include <vector>
 #include "src/ops/eltwise.h"
-#include "nnacl/arithmetic_common.h"
 #include "src/runtime/kernel/npu/npu_kernel.h"
 #include "include/graph/op/all_ops.h"
 namespace mindspore::kernel {

mindspore/lite/src/runtime/kernel/npu/resize_npu.h

@@ -19,7 +19,6 @@
 #include <vector>
 #include "nnacl/resize_parameter.h"
 #include "src/ops/resize.h"
-#include "nnacl/arithmetic_common.h"
 #include "src/runtime/kernel/npu/npu_kernel.h"
 #include "include/graph/op/all_ops.h"
 namespace mindspore::kernel {

mindspore/lite/src/train/train_populate_parameter.cc

@@ -30,7 +30,7 @@
 #include "src/ops/power_grad.h"
 #include "nnacl/power_parameter.h"
 #include "src/ops/bias_grad.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 #include "nnacl/fp32_grad/optimizer.h"
 #include "src/ops/apply_momentum.h"
 #include "src/ops/sgd.h"

mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/batch_to_space_fp32_test.cc

@@ -16,7 +16,7 @@
 #include "src/common/log_adapter.h"
 #include "common/common_test.h"
 #include "mindspore/lite/nnacl/batch_to_space.h"
-#include "mindspore/lite/nnacl/arithmetic_common.h"
+#include "mindspore/lite/nnacl/common_func.h"
 
 namespace mindspore {
 

mindspore/lite/test/ut/src/runtime/kernel/arm/fp32/depth_to_space_fp32_test.cc

@@ -16,7 +16,7 @@
 #include "src/common/log_adapter.h"
 #include "common/common_test.h"
 #include "mindspore/lite/nnacl/depth_to_space.h"
-#include "mindspore/lite/nnacl/arithmetic_common.h"
+#include "mindspore/lite/nnacl/common_func.h"
 
 namespace mindspore {
 

mindspore/lite/test/ut/src/runtime/kernel/arm/int8/bias_add_int8_tests.cc

@@ -1,77 +0,0 @@
-/**
- * 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 <iostream>
-#include <memory>
-#include "schema/inner/model_generated.h"
-#include "common/common_test.h"
-#include "mindspore/lite/src/runtime/kernel/arm/int8/bias_add_int8.h"
-#include "mindspore/lite/src/kernel_registry.h"
-
-using mindspore::lite::DeviceType;
-
-namespace mindspore {
-class TestBiasAddInt8 : public mindspore::CommonTest {
- public:
-  TestBiasAddInt8() {}
-};
-
-TEST_F(TestBiasAddInt8, BiasAdd) {
-  lite::Tensor in_tensor0(kNumberTypeInt8, {1, 2, 3, 2});
-  lite::Tensor in_tensor1(kNumberTypeInt8, {2});
-  lite::Tensor out_tensor(kNumberTypeInt8, {1, 2, 3, 2});
-  int8_t input_data0[] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12};
-  int8_t input_data1[] = {1, 1};
-  int8_t output_data[12] = {0};
-  in_tensor0.set_data(input_data0);
-  in_tensor1.set_data(input_data1);
-  out_tensor.set_data(output_data);
-  std::vector<lite::Tensor *> inputs = {&in_tensor0, &in_tensor1};
-  std::vector<lite::Tensor *> outputs = {&out_tensor};
-
-  ArithmeticParameter parameter = {};
-  int dims[] = {1, 2, 3, 4};
-  parameter.ndim_ = 4;
-  for (int i = 0; i < 4; i++) {
-    parameter.in_shape0_[i] = dims[i];
-    parameter.in_shape1_[i] = 1;
-    parameter.out_shape_[i] = dims[i];
-  }
-  parameter.in_shape1_[3] = dims[3];
-
-  kernel::KernelKey desc = {kernel::KERNEL_ARCH::kCPU, kNumberTypeInt8, schema::PrimitiveType_BiasAdd};
-
-  auto creator = lite::KernelRegistry::GetInstance()->GetCreator(desc);
-  EXPECT_NE(creator, nullptr);
-
-  auto ctx = std::make_shared<lite::InnerContext>();
-  ASSERT_EQ(lite::RET_OK, ctx->Init());
-  auto kernel = creator(inputs, outputs, reinterpret_cast<OpParameter *>(&parameter), ctx.get(), desc, nullptr);
-  EXPECT_NE(kernel, nullptr);
-
-  auto ret = kernel->Run();
-  EXPECT_EQ(0, ret);
-
-  float expect[] = {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13};
-  for (int i = 0; i < 12; ++i) {
-    EXPECT_EQ(output_data[i], expect[i]);
-  }
-
-  in_tensor0.set_data(nullptr);
-  in_tensor1.set_data(nullptr);
-  out_tensor.set_data(nullptr);
-}
-}  // namespace mindspore

mindspore/lite/test/ut/src/runtime/kernel/opencl/arithmetic_tests.cc

@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 #include "ut/src/runtime/kernel/opencl/common.h"
-#include "nnacl/arithmetic_common.h"
+#include "nnacl/arithmetic.h"
 
 namespace mindspore::lite::opencl::test {