diff --git a/mindspore/lite/nnacl/errorcode.h b/mindspore/lite/nnacl/errorcode.h
index f80ce1bb67..09ee45ff68 100644
--- a/mindspore/lite/nnacl/errorcode.h
+++ b/mindspore/lite/nnacl/errorcode.h
@@ -40,6 +40,7 @@ typedef enum ErrorCodeFp32OpEnum {
 
 typedef enum ErrorCodeFp16OpEnum {
   NNACL_ERRCODE_OP_FP16_START = 20000,
+  NNACL_ERRCODE_OP_FP16_WINOGRAD_GENERATOR,
   NNACL_ERRCODE_OP_FP16_END = 29999
 } ErrorCodeFp16OpEnum;
 
diff --git a/mindspore/lite/nnacl/fp16/common_func_fp16.c b/mindspore/lite/nnacl/fp16/common_func_fp16.c
new file mode 100644
index 0000000000..1d1283b6cc
--- /dev/null
+++ b/mindspore/lite/nnacl/fp16/common_func_fp16.c
@@ -0,0 +1,54 @@
+/**
+ * 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/fp16/common_func_fp16.h"
+
+void PostConvFuncCommFp16(float16_t *out_ptr, const float16_t *src_ptr_, const float16_t *bias_ptr,
+                          size_t output_channel, size_t plane_size, size_t oc_stride, size_t hw_stride,
+                          ActType act_type, int size) {
+  if (size == 0) {
+    return;
+  }
+  for (int oc = 0; oc < output_channel; oc++) {
+    int oc_div = oc / size, oc_mod = oc % size;
+    for (int hw = 0; hw < plane_size; hw++) {
+      int src_index = oc_div * size * hw_stride + hw * size + oc_mod;
+      int dst_index = hw * oc_stride + oc;
+      float16_t value = src_ptr_[src_index];
+      if (bias_ptr != NULL) {
+        value = value + bias_ptr[oc];
+      }
+      value = (act_type == ActType_Relu || act_type == ActType_Relu6) ? (MSMAX(0.f, value)) : (value);
+      value = (act_type == ActType_Relu6) ? (MSMIN(6.f, value)) : (value);
+      out_ptr[dst_index] = value;
+    }
+  }
+  return;
+}
+
+void PostConvFuncFp16C8(const float16_t *c8_out, float16_t *nhwc_out, const float16_t *bias, size_t oc, size_t plane,
+                        size_t oc_stride, ActType act_type) {
+  size_t oc8mod = oc % C8NUM;
+  size_t oc8div = oc - oc8mod;
+  size_t stride_size = oc_stride * sizeof(float16_t);
+  PostFuncBiasReluC8Fp16(nhwc_out, c8_out, bias, oc8div, oc8mod, plane, stride_size, act_type);
+  return;
+}
+
+void PostConvFuncFp16C4(const float16_t *c4_out, float16_t *nhwc_out, const float16_t *bias, size_t oc, size_t plane,
+                        size_t plane_stride, ActType act_type) {
+  PostConvFuncCommFp16(nhwc_out, c4_out, bias, oc, plane, oc, plane_stride, act_type, C4NUM);
+}
diff --git a/mindspore/lite/nnacl/fp16/common_func_fp16.h b/mindspore/lite/nnacl/fp16/common_func_fp16.h
new file mode 100644
index 0000000000..b9b811dcdb
--- /dev/null
+++ b/mindspore/lite/nnacl/fp16/common_func_fp16.h
@@ -0,0 +1,39 @@
+/**
+ * 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_FP16_COMMON_FUNC_FP16_H_
+#define MINDSPORE_LITE_NNACL_FP16_COMMON_FUNC_FP16_H_
+
+#include <arm_neon.h>
+#include "nnacl/op_base.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* deconv common */
+void PostConvFuncFp16C8(const float16_t *c8_out_ptr, float16_t *out_ptr, const float16_t *bias_ptr,
+                        size_t output_channel, size_t plane_size, size_t stride, ActType act_type);
+void PostFuncBiasReluC8Fp16(float16_t *dst, const float16_t *src, const float16_t *bias, size_t oc8div, size_t oc8mod,
+                            size_t plane_size, size_t stride, size_t relu_type);
+
+/* deconv winograd */
+void PostConvFuncFp16C4(const float16_t *c4_out, float16_t *nhwc_out, const float16_t *bias, size_t output_channel,
+                        size_t plane_size, size_t plane_stride, ActType act_type);
+
+#ifdef __cplusplus
+}
+#endif
+#endif  // MINDSPORE_LITE_NNACL_FP16_COMMON_FUNC_FP16_H_
diff --git a/mindspore/lite/nnacl/fp16/deconv_fp16.c b/mindspore/lite/nnacl/fp16/deconv_fp16.c
index d235d748ab..d29dc2b2bf 100644
--- a/mindspore/lite/nnacl/fp16/deconv_fp16.c
+++ b/mindspore/lite/nnacl/fp16/deconv_fp16.c
@@ -13,41 +13,8 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#include "nnacl/fp16/deconv_fp16.h"
-
-void PostConvFuncCommFp16(float16_t *out_ptr, const float16_t *src_ptr_, const float16_t *bias_ptr,
-                          size_t output_channel, size_t plane_size, size_t stride, bool is_relu, bool is_relu6,
-                          int size) {
-  if (size == 0) {
-    return;
-  }
-  for (int oc = 0; oc < output_channel; oc++) {
-    int oc_div = oc / size, oc_mod = oc % size;
-    for (int hw = 0; hw < plane_size; hw++) {
-      int src_index = oc_div * size * plane_size + hw * size + oc_mod;
-      int dst_index = hw * stride + oc;
-      float16_t value = src_ptr_[src_index];
-      if (bias_ptr != NULL) {
-        value = value + bias_ptr[oc];
-      }
-      value = (is_relu || is_relu6) ? (MSMAX(0.f, value)) : (value);
-      value = (is_relu6) ? (MSMIN(6.f, value)) : (value);
-      out_ptr[dst_index] = value;
-    }
-  }
-  return;
-}
 
-void PostConvFuncFp16C8(const float16_t *c8_out_ptr, float16_t *out_ptr, const float16_t *bias_ptr,
-                        size_t output_channel, size_t plane_size, size_t stride, bool is_relu, bool is_relu6) {
-  size_t oc8mod = output_channel % C8NUM;
-  size_t oc8div = output_channel - oc8mod;
-  size_t stride_size = stride * sizeof(float16_t);
-  size_t relu_type = is_relu ? 1 : 0;
-  relu_type = is_relu6 ? 3 : relu_type;
-  PostFuncBiasReluC8Fp16(out_ptr, c8_out_ptr, bias_ptr, oc8div, oc8mod, plane_size, stride_size, relu_type);
-  return;
-}
+#include "nnacl/fp16/deconv_fp16.h"
 
 int DeConvPostFp16(const float16_t *src, float16_t *tmp, const float16_t *bias, float16_t *dst, int output_channel,
                    ConvParameter *conv_param) {
@@ -112,7 +79,6 @@ int DeConvPostFp16(const float16_t *src, float16_t *tmp, const float16_t *bias,
     }       /*ih*/
   }         /*oc8*/
 
-  PostConvFuncFp16C8(tmp, dst, bias, output_channel, output_plane, conv_param->output_channel_,
-                     conv_param->act_type_ == ActType_Relu, conv_param->act_type_ == ActType_Relu6);
+  PostConvFuncFp16C8(tmp, dst, bias, output_channel, output_plane, conv_param->output_channel_, conv_param->act_type_);
   return NNACL_OK;
 }
diff --git a/mindspore/lite/nnacl/fp16/deconv_fp16.h b/mindspore/lite/nnacl/fp16/deconv_fp16.h
index 068e51a412..f095930543 100644
--- a/mindspore/lite/nnacl/fp16/deconv_fp16.h
+++ b/mindspore/lite/nnacl/fp16/deconv_fp16.h
@@ -13,27 +13,23 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
+
 #ifndef MINDSPORE_LITE_NNACL_FP16_DECONV_FP16_H_
 #define MINDSPORE_LITE_NNACL_FP16_DECONV_FP16_H_
 
-#include <string.h>
 #include <arm_neon.h>
+#include <string.h>
 #include "nnacl/conv_parameter.h"
-#include "nnacl/matmul_parameter.h"
-#include "nnacl/fp16/matmul_fp16.h"
+#include "nnacl/errorcode.h"
+#include "nnacl/fp16/common_func_fp16.h"
 
 #ifdef __cplusplus
 extern "C" {
 #endif
+
 int DeConvPostFp16(const float16_t *src, float16_t *tmp, const float16_t *bias, float16_t *dst, int output_channel,
                    ConvParameter *conv_param);
 
-void PostConvFuncFp16C8(const float16_t *c8_out_ptr, float16_t *out_ptr, const float16_t *bias_ptr,
-                        size_t output_channel, size_t plane_size, size_t stride, bool is_relu, bool is_relu6);
-
-void PostFuncBiasReluC8Fp16(float16_t *dst, const float16_t *src, const float16_t *bias, size_t oc8div, size_t oc8mod,
-                            size_t plane_size, size_t stride, size_t relu_type);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/mindspore/lite/nnacl/fp16/deconv_winograd_fp16.c b/mindspore/lite/nnacl/fp16/deconv_winograd_fp16.c
new file mode 100644
index 0000000000..cb8ef91109
--- /dev/null
+++ b/mindspore/lite/nnacl/fp16/deconv_winograd_fp16.c
@@ -0,0 +1,329 @@
+/**
+ * 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/fp16/deconv_winograd_fp16.h"
+#include "nnacl/minimal_filtering_generator.h"
+
+void DeConvWgInputPackFp16(float16_t *src_ptr, float16_t *dst_ptr, int channel, int stride) {
+  int ic4div = channel / C4NUM;
+  int ic4mod = channel % C4NUM;
+  float16_t *src = src_ptr;
+  float16_t *dst = dst_ptr;
+
+  for (int ic = 0; ic < ic4div; ic++) {
+    memcpy(dst, src, C4NUM * sizeof(float16_t));
+    dst += stride;
+    src += C4NUM;
+  }
+
+  if (ic4mod != 0) {
+    int ic_res = 0;
+    for (; ic_res < ic4mod; ic_res++) {
+      dst[ic_res] = src[ic_res];
+    }
+    for (; ic_res < C4NUM; ic_res++) {
+      dst[ic_res] = 0;
+    }
+  }
+  return;
+}
+
+void C4GemmFp16(float16_t *dst, const float16_t *src, const float16_t *weight, size_t src_depth_quad, size_t dst_step,
+                size_t dst_depth_quad, size_t width, size_t weight_depth_offset) {
+  int dx, sz, dz;
+  int src_depth_step = 4 * width;
+  for (dz = 0; dz < dst_depth_quad; ++dz) {
+    float16_t *dst_z = dst + dz * dst_step;
+    const float16_t *weight_dz = weight + dz * (src_depth_quad * 16 + weight_depth_offset);
+    for (dx = 0; dx < width; ++dx) {
+      float16_t *dst_x = dst_z + dx * 4;
+      dst_x[0] = 0.0f;
+      dst_x[1] = 0.0f;
+      dst_x[2] = 0.0f;
+      dst_x[3] = 0.0f;
+      const float16_t *src_dx = src + 4 * dx;
+      for (sz = 0; sz < src_depth_quad; ++sz) {
+        const float16_t *src_z = src_dx + sz * src_depth_step;
+        const float16_t *weight_z = weight_dz + sz * 16;
+        for (int i = 0; i < 4; ++i) {
+          for (int j = 0; j < 4; ++j) {
+            dst_x[j] += src_z[i] * weight_z[4 * i + j];
+          }
+        }
+      }
+    }
+  }
+}
+
+void DeConvWgMergeFp16(const float16_t *src, float16_t *dst, size_t src_stride, size_t dst_stride, size_t count) {
+  for (int i = 0; i < count; ++i) {
+    const float16_t *s = src + i * src_stride;
+    float16_t *d = dst + i * dst_stride;
+    for (int j = 0; j < 4; ++j) {
+      d[j] += s[j];
+    }
+  }
+}
+
+void _deConvWinogradFp16(float16_t *tile_in, float16_t *tile_out, float16_t *weight_buf, float16_t *tmp_buf,
+                         float16_t *at_buf, float16_t *a_mid_buf, float16_t *trans_a_buf, bool a_trans,
+                         float16_t *bt_buf, float16_t *b_tmp_buf, int unit_size, int w_start, int h_start,
+                         ConvParameter *conv_param, DeConvParam *deconv_param) {
+  int winograd_plane = unit_size * unit_size;
+  if (!a_trans) {
+    WinogradMatrixProductLeftFp16(tile_in, at_buf, a_mid_buf, DECONV_WINOGRAD_DEFAULT_UNIT, unit_size,
+                                  DECONV_WINOGRAD_DEFAULT_UNIT, deconv_param->ic_div4_ * DECONV_WINOGRAD_DEFAULT_TILE);
+    WinogradMatrixProductRightFp16(a_mid_buf, at_buf, trans_a_buf, unit_size, unit_size, DECONV_WINOGRAD_DEFAULT_UNIT,
+                                   deconv_param->ic_div4_ * DECONV_WINOGRAD_DEFAULT_TILE);
+  }
+
+  for (int index = 0; index < winograd_plane; index++) {
+    float16_t *src = trans_a_buf + index * DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->ic_up4_;
+    float16_t *dst = tmp_buf + index * deconv_param->oc_up4_ * DECONV_WINOGRAD_DEFAULT_TILE;
+    float16_t *weight = weight_buf + index * deconv_param->ic_up4_ * deconv_param->oc_up4_;
+    C4GemmFp16(dst, src, weight, deconv_param->ic_div4_, DECONV_WINOGRAD_DEFAULT_TILE * C4NUM, deconv_param->oc_div4_,
+               DECONV_WINOGRAD_DEFAULT_TILE, 0);
+  }
+
+  WinogradMatrixProductLeftFp16(tmp_buf, bt_buf, b_tmp_buf, unit_size, unit_size, unit_size,
+                                deconv_param->oc_div4_ * DECONV_WINOGRAD_DEFAULT_TILE);
+  WinogradMatrixProductRightFp16(b_tmp_buf, bt_buf, tmp_buf, unit_size, unit_size, unit_size,
+                                 deconv_param->oc_div4_ * DECONV_WINOGRAD_DEFAULT_TILE);
+
+  // Add to dest
+  for (int uhi = 0; uhi < unit_size; uhi++) {
+    int h_index = uhi * conv_param->stride_h_ + h_start;
+    for (int uwi = 0; uwi < unit_size; uwi++) {
+      int w_index = uwi * conv_param->stride_w_ + w_start;
+
+      float16_t *dst = tile_out + w_index * DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->oc_up4_ +
+                       h_index * deconv_param->out_tile_w_ * DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->oc_up4_;
+      float16_t *src = tmp_buf + (uwi + uhi * unit_size) * DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->oc_up4_;
+      DeConvWgMergeFp16(src, dst, 4, 4, DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->oc_div4_);
+    }
+  }
+  return;
+}
+
+void _deConvCommonFp16(float16_t *tile_in, float16_t *tile_out, float16_t *weight, float16_t *tmp_buf, int h_start,
+                       int w_start, int h_size, int w_size, ConvParameter *conv_param, DeConvParam *deconv_param) {
+  int count = deconv_param->oc_div4_ * w_size * h_size;
+  int in_stride = DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->ic_up4_;
+  int out_stride = DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->oc_up4_;
+
+  for (int hi = 0; hi < DECONV_WINOGRAD_DEFAULT_UNIT; hi++) {
+    for (int wi = 0; wi < DECONV_WINOGRAD_DEFAULT_UNIT; wi++) {
+      float16_t *src_in = tile_in + (wi + hi * DECONV_WINOGRAD_DEFAULT_UNIT) * in_stride;
+      C4GemmFp16(tmp_buf, src_in, weight, deconv_param->ic_div4_, DECONV_WINOGRAD_DEFAULT_TILE * 4, count,
+                 DECONV_WINOGRAD_DEFAULT_TILE, 0);
+
+      for (int uhi = 0; uhi < h_size; uhi++) {
+        for (int uwi = 0; uwi < w_size; uwi++) {
+          int w_index = (wi + uwi) * conv_param->stride_w_ + w_start;
+          int h_index = (hi + uhi) * conv_param->stride_h_ + h_start;
+          float16_t *dst = tile_out + h_index * out_stride * deconv_param->out_tile_w_ + w_index * out_stride;
+          float16_t *src = tmp_buf + (uwi + uhi * w_size) * out_stride;
+          DeConvWgMergeFp16(src, dst, 4, 4, DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->oc_div4_);
+        }
+      }
+    }
+  }
+  return;
+}
+
+int PackDeConvWgDataFp16(float16_t *nhwc_weight, DeConvComputeUnit *unit, ConvParameter *conv_param,
+                         DeConvParam *deconv_param) {
+  int tmp_kernel_plane = unit->w_size_ * unit->h_size_;
+  int output_channel = conv_param->output_channel_;
+  int size = conv_param->input_channel_ * output_channel * tmp_kernel_plane;
+  float16_t *current_unit_weight = (float16_t *)malloc(size * sizeof(float16_t));
+  if (current_unit_weight == NULL) {
+    return NNACL_NULL_PTR;
+  }
+  for (int ic = 0; ic < conv_param->input_channel_; ic++) {
+    float16_t *src_ic = nhwc_weight + deconv_param->kernel_plane_ * output_channel * ic;
+    float16_t *dst_ic = current_unit_weight + tmp_kernel_plane * output_channel * ic;
+    for (int uhi = 0; uhi < unit->h_size_; uhi++) {
+      for (int uwi = 0; uwi < unit->w_size_; uwi++) {
+        int src_h_offset = unit->h_start_ + uhi * conv_param->stride_h_;
+        int src_w_offset = unit->w_start_ + uwi * conv_param->stride_w_;
+        float16_t *src_hw = src_ic + (src_h_offset * conv_param->kernel_w_ + src_w_offset) * output_channel;
+        float16_t *dst_hw = dst_ic + (uhi * unit->w_size_ + uwi) * output_channel;
+        memcpy(dst_hw, src_hw, output_channel * sizeof(float16_t));
+      }
+    }
+  }
+
+  if (unit->use_winograd_) {
+    /* Generate winograd  */
+    float matrix_g[64];
+    float matrix_gt[64];
+    float matrix_a[64];
+    float matrix_at[64];
+    float matrix_b[64];
+    float matrix_bt[64];
+    int ret = CookToomFilter(matrix_a, matrix_at, matrix_b, matrix_bt, matrix_g, matrix_gt, 0.5f,
+                             DECONV_WINOGRAD_DEFAULT_UNIT, unit->h_size_);
+    if (ret != NNACL_OK) {
+      return NNACL_ERRCODE_WINOGRAD_GENERATOR_ERROR;
+    }
+
+    /* winograd AT */
+    unit->winograd_.AT_ = malloc(unit->winograd_.i_ * unit->winograd_.o_ * sizeof(float16_t));
+    if (unit->winograd_.AT_ == NULL) {
+      return NNACL_NULL_PTR;
+    }
+    Float32ToFloat16(matrix_at, unit->winograd_.AT_, unit->winograd_.i_ * unit->winograd_.o_);
+
+    /* winograd BT */
+    unit->winograd_.BT_ = malloc(unit->winograd_.o_ * unit->winograd_.o_ * sizeof(float16_t));
+    if (unit->winograd_.BT_ == NULL) {
+      return NNACL_NULL_PTR;
+    }
+    Float32ToFloat16(matrix_bt, unit->winograd_.BT_, unit->winograd_.o_ * unit->winograd_.o_);
+
+    /* winograd Weight */
+    size = conv_param->input_channel_ * output_channel * unit->winograd_.kh_ * unit->winograd_.kw_;
+    float16_t *winograd_unit_weight = (float16_t *)malloc(size * sizeof(float16_t));
+    if (winograd_unit_weight == NULL) {
+      return NNACL_NULL_PTR;
+    }
+
+    WinogradWeightTransformFp16(current_unit_weight, winograd_unit_weight, matrix_g, matrix_gt, C4NUM,
+                                unit->winograd_.kh_, unit->h_size_, output_channel, conv_param->input_channel_, false);
+
+    /* reset weight data & info */
+    tmp_kernel_plane = unit->winograd_.kh_ * unit->winograd_.kw_;
+    free(current_unit_weight);
+    current_unit_weight = winograd_unit_weight;
+    winograd_unit_weight = NULL;
+  }
+
+  /* trans mhwc -> hw1:k1-knc0-c4:k1-knc5-c8:hw2:k1-knc0-c4:k1 */
+  float16_t *dst_weight = (float16_t *)unit->weight_;
+  size = deconv_param->ic_up4_ * deconv_param->oc_up4_ * tmp_kernel_plane;
+  memset(dst_weight, 0, size * sizeof(float16_t));
+  for (int ic = 0; ic < conv_param->input_channel_; ic++) {
+    for (int oc = 0; oc < output_channel; oc++) {
+      int oc4div = oc / C4NUM, oc4mod = oc % C4NUM;
+      for (int upi = 0; upi < tmp_kernel_plane; upi++) {
+        int src_index = ic * output_channel * tmp_kernel_plane + upi * output_channel + oc;
+        int dst_index = upi * deconv_param->oc_up4_ * deconv_param->ic_up4_ + oc4div * C4NUM * deconv_param->ic_up4_ +
+                        ic * C4NUM + oc4mod;
+        dst_weight[dst_index] = current_unit_weight[src_index];
+      }
+    }
+  }
+
+  free(current_unit_weight);
+  return NNACL_OK;
+}
+
+void DeconvWgFp16(float16_t *nhwc_input_, float16_t *tile_in, float16_t *tile_out, int start_index, int calculate_count,
+                  ConvParameter *conv_param, DeConvParam *deconv_param, int task_id) {
+  /* pack tile input */
+  int tile_in_unit_stride = deconv_param->ic_up4_ * DECONV_WINOGRAD_DEFAULT_TILE;
+  float16x4_t zero = vdup_n_f16(0.0f);
+
+  for (int unit_index = 0; unit_index < calculate_count; unit_index++) {
+    int plane_index = start_index + unit_index;
+    int w_unit_index = plane_index % deconv_param->in_tile_w_count_;
+    int h_unit_index = plane_index / deconv_param->in_tile_w_count_;
+    int w_start = w_unit_index * DECONV_WINOGRAD_DEFAULT_UNIT;
+    int h_start = h_unit_index * DECONV_WINOGRAD_DEFAULT_UNIT;
+
+    float16_t *dst_unit = tile_in + unit_index * C4NUM;
+    for (int hi = 0; hi < DECONV_WINOGRAD_DEFAULT_UNIT; hi++) {
+      for (int wi = 0; wi < DECONV_WINOGRAD_DEFAULT_UNIT; wi++) {
+        float16_t *dst = dst_unit + (wi + hi * DECONV_WINOGRAD_DEFAULT_UNIT) * tile_in_unit_stride;
+        int w_index = w_start + wi;
+        int h_index = h_start + hi;
+        if (w_index >= conv_param->input_w_ || h_index >= conv_param->input_h_) {
+          for (int ic4_index = 0; ic4_index < deconv_param->ic_div4_; ic4_index++) {
+            vst1_f16(dst + ic4_index * DECONV_WINOGRAD_DEFAULT_TILE * C4NUM, zero);
+          }
+          continue;
+        }
+
+        float16_t *src = nhwc_input_ + (w_index + h_index * conv_param->input_w_) * conv_param->input_channel_;
+        DeConvWgInputPackFp16(src, dst, conv_param->input_channel_, DECONV_WINOGRAD_DEFAULT_TILE * C4NUM);
+      }
+    }
+  }
+
+  /* compute */
+  for (int i = 0; i < deconv_param->compute_size_; i++) {
+    DeConvComputeUnit *unit = &deconv_param->compute_units_[i];
+    if (unit->use_winograd_) {
+      float16_t *tmp_buf = (float16_t *)unit->tmp_buffer_ + task_id * unit->winograd_.kh_ * unit->winograd_.kw_ *
+                                                              deconv_param->oc_up4_ * DECONV_WINOGRAD_DEFAULT_TILE;
+
+      /* winograd a buffer */
+      DeConvWgABuffer *tmp_a = &deconv_param->a_buffer_[unit->winograd_.kh_];
+      float16_t *mid_a = (float16_t *)tmp_a->middle_buffer_ + task_id * unit->winograd_.kw_ * unit->winograd_.kh_ *
+                                                                DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->ic_up4_;
+      float16_t *dst_a = (float16_t *)tmp_a->dest_buffer_ + task_id * unit->winograd_.kw_ * unit->winograd_.kh_ *
+                                                              DECONV_WINOGRAD_DEFAULT_TILE * deconv_param->ic_up4_;
+      float16_t *tmp_b = (float16_t *)unit->winograd_.b_buffer_ + task_id * unit->winograd_.kh_ * unit->winograd_.kw_ *
+                                                                    DECONV_WINOGRAD_DEFAULT_TILE *
+                                                                    deconv_param->oc_up4_;
+      _deConvWinogradFp16(tile_in, tile_out, (float16_t *)unit->weight_, tmp_buf, unit->winograd_.AT_, mid_a, dst_a,
+                          tmp_a->trans_formed_, unit->winograd_.BT_, tmp_b, unit->winograd_.kh_, unit->w_start_,
+                          unit->h_start_, conv_param, deconv_param);
+      tmp_a->trans_formed_ = true;
+    } else {
+      float16_t *tmp_buf = (float16_t *)unit->tmp_buffer_ + task_id * deconv_param->oc_div4_ * unit->w_size_ *
+                                                              unit->h_size_ * DECONV_WINOGRAD_DEFAULT_TILE * C4NUM;
+      _deConvCommonFp16(tile_in, tile_out, (float16_t *)unit->weight_, tmp_buf, unit->h_start_, unit->w_start_,
+                        unit->h_size_, unit->w_size_, conv_param, deconv_param);
+    }
+  }
+  return;
+}
+
+void DeconvWgPostFp16(float16_t *tile_out, float16_t *nc4hw4_output, ConvParameter *conv_param,
+                      DeConvParam *deconv_param, int calculate_count, int tile_index) {
+  /* merge */
+  int src_unit_stride = deconv_param->oc_up4_ * DECONV_WINOGRAD_DEFAULT_TILE;
+
+  int src_stride = DECONV_WINOGRAD_DEFAULT_TILE * C4NUM;
+  int dst_stride = conv_param->output_w_ * conv_param->output_h_ * C4NUM;
+
+  for (int index = 0; index < calculate_count; ++index) {
+    float16_t *src_start = tile_out + index * C4NUM;
+
+    int plane_index = tile_index * DECONV_WINOGRAD_DEFAULT_TILE + index;
+    int w_unit_index = plane_index % deconv_param->in_tile_w_count_;
+    int h_unit_index = plane_index / deconv_param->in_tile_w_count_;
+    int w_start = w_unit_index * DECONV_WINOGRAD_DEFAULT_UNIT * conv_param->stride_w_ - conv_param->pad_l_;
+    int h_start = h_unit_index * DECONV_WINOGRAD_DEFAULT_UNIT * conv_param->stride_h_ - conv_param->pad_u_;
+    float16_t *dst_start = nc4hw4_output + h_start * conv_param->output_w_ * C4NUM + w_start * C4NUM;
+
+    int merge_w_start = MSMAX(-w_start, 0);
+    int merge_h_start = MSMAX(-h_start, 0);
+    int merge_h_end = MSMIN(deconv_param->out_tile_h_, conv_param->output_h_ - h_start);
+    int merge_w_end = MSMIN(deconv_param->out_tile_w_, conv_param->output_w_ - w_start);
+
+    for (int hi = merge_h_start; hi < merge_h_end; hi++) {
+      for (int wi = merge_w_start; wi < merge_w_end; wi++) {
+        float16_t *src = src_start + (hi * deconv_param->out_tile_w_ + wi) * src_unit_stride;
+        float16_t *dst = dst_start + (hi * conv_param->output_w_ + wi) * C4NUM;
+        DeConvWgMergeFp16(src, dst, src_stride, dst_stride, deconv_param->oc_div4_);
+      }
+    }
+  }
+  return;
+}
diff --git a/mindspore/lite/nnacl/fp16/deconv_winograd_fp16.h b/mindspore/lite/nnacl/fp16/deconv_winograd_fp16.h
new file mode 100644
index 0000000000..8ca3c0d1c6
--- /dev/null
+++ b/mindspore/lite/nnacl/fp16/deconv_winograd_fp16.h
@@ -0,0 +1,39 @@
+/**
+ * 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_FP16_DECONV_WINOGRAD_FP16_H_
+#define MINDSPORE_LITE_NNACL_FP16_DECONV_WINOGRAD_FP16_H_
+
+#include "nnacl/fp16/winograd_transform_fp16.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+int PackDeConvWgDataFp16(float16_t *nhwc_weight, DeConvComputeUnit *unit, ConvParameter *conv_param,
+                         DeConvParam *deconv_param);
+
+void DeconvWgFp16(float16_t *nhwc_input_, float16_t *tile_in, float16_t *tile_out, int start_index, int calculate_count,
+                  ConvParameter *conv_param, DeConvParam *deconv_param, int task_id);
+
+void DeconvWgPostFp16(float16_t *tile_out, float16_t *nc4hw4_output, ConvParameter *conv_param,
+                      DeConvParam *deconv_param, int calculate_count, int tile_index);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  // MINDSPORE_LITE_NNACL_FP16_DECONV_WINOGRAD_FP16_H_
diff --git a/mindspore/lite/nnacl/fp16/matrix_fp16.c b/mindspore/lite/nnacl/fp16/matrix_fp16.c
index ce8acfba75..f4ba9d80ef 100644
--- a/mindspore/lite/nnacl/fp16/matrix_fp16.c
+++ b/mindspore/lite/nnacl/fp16/matrix_fp16.c
@@ -81,3 +81,51 @@ void MatrixMultiplyVecFp16(const float16x8_t *matrix_a, const float16x8_t *matri
     }
   }
 }
+
+void WinogradMatrixProductLeftFp16(const float16_t *S, const float16_t *B, float16_t *M, size_t w, size_t h, size_t k,
+                                   size_t length) {
+  int unitStep = 4 * length;
+  for (int y = 0; y < h; ++y) {
+    float16_t *dstY = M + y * w * unitStep;
+    for (int x = 0; x < w; ++x) {
+      float16_t *dstX = dstY + x * unitStep;
+      const float16_t *srcX = S + x * unitStep;
+      memset(dstX, 0, unitStep * sizeof(float16_t));
+      for (int i = 0; i < k; ++i) {
+        float16_t b = B[i * h + y];
+        const float16_t *srcY = srcX + i * w * unitStep;
+        if (0.0f == b) {
+          continue;
+        }
+        for (int j = 0; j < unitStep; ++j) {
+          dstX[j] += srcY[j] * b;
+        }
+      }
+    }
+  }
+}
+
+// M = S * B , M = w*h * l, S = k*h * l, B = w*k
+void WinogradMatrixProductRightFp16(const float16_t *S, const float16_t *B, float16_t *M, size_t w, size_t h, size_t k,
+                                    size_t length) {
+  int unitStep = 4 * length;
+  for (int y = 0; y < h; ++y) {
+    float16_t *dstY = M + y * w * unitStep;
+    const float16_t *srcY = S + y * k * unitStep;
+
+    for (int x = 0; x < w; ++x) {
+      float16_t *dstX = dstY + x * unitStep;
+      memset(dstX, 0, unitStep * sizeof(float16_t));
+      for (int i = 0; i < k; ++i) {
+        const float16_t *srcX = srcY + i * unitStep;
+        float16_t b = B[i * h + x];
+        if (0.0f == b) {
+          continue;
+        }
+        for (int j = 0; j < unitStep; ++j) {
+          dstX[j] += srcX[j] * b;
+        }
+      }
+    }
+  }
+}
diff --git a/mindspore/lite/nnacl/fp16/matrix_fp16.h b/mindspore/lite/nnacl/fp16/matrix_fp16.h
index dfc5f24523..02e0c2a652 100644
--- a/mindspore/lite/nnacl/fp16/matrix_fp16.h
+++ b/mindspore/lite/nnacl/fp16/matrix_fp16.h
@@ -18,6 +18,7 @@
 #define MINDSPORE_LITE_NNACL_FP16_MATRIX_FP16_H_
 
 #include <arm_neon.h>
+#include <string.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -28,6 +29,13 @@ void MatrixMultiplyVecFp16(const float16x8_t *matrix_a, const float16x8_t *matri
                            const float16_t *bias, int m, int k, int n);
 void MatrixMultiplyWinogradFp16(const float16_t *matix_a, const float16_t *matrix_b, float16_t *matrix_c, int m, int k,
                                 int n, int in_channel);
+
+void WinogradMatrixProductLeftFp16(const float16_t *S, const float16_t *B, float16_t *M, size_t w, size_t h, size_t k,
+                                   size_t length);
+
+void WinogradMatrixProductRightFp16(const float16_t *S, const float16_t *B, float16_t *M, size_t w, size_t h, size_t k,
+                                    size_t length);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/mindspore/lite/nnacl/fp16/winograd_transform_fp16.c b/mindspore/lite/nnacl/fp16/winograd_transform_fp16.c
index e10a76b45b..26a6d5775c 100644
--- a/mindspore/lite/nnacl/fp16/winograd_transform_fp16.c
+++ b/mindspore/lite/nnacl/fp16/winograd_transform_fp16.c
@@ -712,3 +712,102 @@ void WinogradOutputTransformFp16(const float16_t *gemm_out, float16_t *tmp_out_d
     out_tile_index++;
   }
 }
+
+int WinogradWeightTransformFp16(const float16_t *weight_data, float16_t *winograd_data, float *matrix_g,
+                                float *matrix_gt, int oc_block, int input_unit, int kernel_unit, int filter_channel,
+                                int filter_batch, bool pack) {
+  // original weight format : ohwi
+  int oc_block_num = UP_DIV(filter_batch, oc_block);
+  int block_stride = filter_channel * oc_block;
+  int block_num_stride = block_stride * oc_block_num;
+
+  float16_t *matrix_gt_data_fp16 = (float16_t *)(malloc(input_unit * kernel_unit * sizeof(float16_t)));
+  if (matrix_gt_data_fp16 == NULL) {
+    return NNACL_ERRCODE_OP_FP16_WINOGRAD_GENERATOR;
+  }
+  Float32ToFloat16(matrix_gt, matrix_gt_data_fp16, input_unit * kernel_unit);
+
+  // trans_filter = G*g*GT (g represents weight_data) = [(g * (G)T)T * (G)T]T
+  // separate into two steps ===> tmp = (g * (G)T)T ===> out = [tmp * (G)T]T
+  float16_t *tmp_data = (float16_t *)(malloc(filter_channel * input_unit * kernel_unit * sizeof(float16_t)));
+  if (tmp_data == NULL) {
+    free(matrix_gt_data_fp16);
+    return NNACL_ERRCODE_OP_FP16_WINOGRAD_GENERATOR;
+  }
+  float16_t *trans_out_data = (float16_t *)(malloc(filter_channel * input_unit * input_unit * sizeof(float16_t)));
+  if (trans_out_data == NULL) {
+    free(tmp_data);
+    free(matrix_gt_data_fp16);
+    return NNACL_ERRCODE_OP_FP16_WINOGRAD_GENERATOR;
+  }
+
+#ifndef ENABLE_ARM64
+  float16_t *tmp_data1 = (float16_t *)(malloc(filter_channel * input_unit * kernel_unit * sizeof(float16_t)));
+  if (tmp_data1 == NULL) {
+    free(tmp_data);
+    free(matrix_gt_data_fp16);
+    free(trans_out_data);
+    return NNACL_ERRCODE_OP_FP16_WINOGRAD_GENERATOR;
+  }
+  float16_t *trans_out_data1 = (float16_t *)(malloc(filter_channel * input_unit * input_unit * sizeof(float16_t)));
+  if (trans_out_data1 == NULL) {
+    free(tmp_data);
+    free(tmp_data1);
+    free(matrix_gt_data_fp16);
+    free(trans_out_data);
+    return NNACL_ERRCODE_OP_FP16_WINOGRAD_GENERATOR;
+  }
+#endif
+
+  int input_oz_offset = kernel_unit * kernel_unit * filter_channel;
+  for (int i = 0; i < filter_batch; i++) {
+    int out_c_block = i / oc_block;
+    int out_c_res = i % oc_block;
+    int output_oz_offset = out_c_block * block_stride + out_c_res;
+
+#ifndef ENABLE_ARM64
+    // tmp_data = g * GT
+    MatrixMultiplyWinogradFp16(weight_data + i * input_oz_offset, matrix_gt_data_fp16, tmp_data, kernel_unit,
+                               kernel_unit, input_unit, filter_channel);
+    // tmp_data1 = (tmp_data)T
+    PackHWCToWHCFp16(tmp_data, tmp_data1, kernel_unit, input_unit, filter_channel);
+    // trans_out_data1 = tmp * GT
+    MatrixMultiplyWinogradFp16(tmp_data1, matrix_gt_data_fp16, trans_out_data1, input_unit, kernel_unit, input_unit,
+                               filter_channel);
+    // trans_out_data = (trans_out_data1)T
+    PackHWCToWHCFp16(trans_out_data1, trans_out_data, input_unit, input_unit, filter_channel);
+#else
+    // tmp = (g * GT)T
+    MatrixMultiplyWinogradFp16(weight_data + i * input_oz_offset, matrix_gt_data_fp16, tmp_data, kernel_unit,
+                               kernel_unit, input_unit, filter_channel);
+    // trans = (tmp * GT)T
+    MatrixMultiplyWinogradFp16(tmp_data, matrix_gt_data_fp16, trans_out_data, input_unit, kernel_unit, input_unit,
+                               filter_channel);
+#endif
+
+    if (pack) {
+      int in_offset = 0;
+      for (int j = 0; j < input_unit; ++j) {
+        for (int k = 0; k < input_unit; ++k) {
+          for (int c = 0; c < filter_channel; ++c) {
+            *(winograd_data + output_oz_offset + c * oc_block) = trans_out_data[in_offset + c];
+          }
+          in_offset += filter_channel;
+          output_oz_offset += block_num_stride;
+        }
+      }
+    } else {
+      memcpy(winograd_data + i * filter_channel * input_unit * input_unit, trans_out_data,
+             filter_channel * input_unit * input_unit * sizeof(float16_t));
+    }
+  }
+
+#ifndef ENABLE_ARM64
+  free(tmp_data1);
+  free(trans_out_data1);
+#endif
+  free(tmp_data);
+  free(trans_out_data);
+  free(matrix_gt_data_fp16);
+  return NNACL_OK;
+}
diff --git a/mindspore/lite/nnacl/fp16/winograd_transform_fp16.h b/mindspore/lite/nnacl/fp16/winograd_transform_fp16.h
index eaedbc498d..42d997a3cc 100644
--- a/mindspore/lite/nnacl/fp16/winograd_transform_fp16.h
+++ b/mindspore/lite/nnacl/fp16/winograd_transform_fp16.h
@@ -19,9 +19,10 @@
 
 #include <arm_neon.h>
 #include <string.h>
-#include "nnacl/fp16/pack_fp16.h"
+#include "nnacl/errorcode.h"
+#include "nnacl/fp16/cast_fp16.h"
 #include "nnacl/fp16/conv_fp16.h"
-#include "nnacl/fp16/winograd_utils_fp16.h"
+#include "nnacl/fp16/matrix_fp16.h"
 
 #ifdef __cplusplus
 extern "C" {
@@ -49,6 +50,12 @@ void WinogradInputTransformFp16(const float16_t *input_data, float16_t *trans_in
 void WinogradOutputTransformFp16(const float16_t *gemm_out, float16_t *tmp_out_data, const float16_t *bias_data,
                                  int cal_num, int out_tile_index, int output_unit_num, ConvParameter *conv_param,
                                  OutputTransFp16Func func);
+
+// fp16 winograd weight trans
+int WinogradWeightTransformFp16(const float16_t *weight_data, float16_t *winograd_data, float *matrix_g,
+                                float *matrix_gt, int oc_block, int input_unit, int kernel_unit, int filter_channel,
+                                int filter_batch, bool pack);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp16/convolution_winograd_fp16.cc b/mindspore/lite/src/runtime/kernel/arm/fp16/convolution_winograd_fp16.cc
index 2aa733b4cf..37364467eb 100644
--- a/mindspore/lite/src/runtime/kernel/arm/fp16/convolution_winograd_fp16.cc
+++ b/mindspore/lite/src/runtime/kernel/arm/fp16/convolution_winograd_fp16.cc
@@ -15,23 +15,10 @@
  */
 
 #include "src/runtime/kernel/arm/fp16/convolution_winograd_fp16.h"
-#include "nnacl/fp16/matrix_fp16.h"
-#include "nnacl/fp16/conv_fp16.h"
-#include "nnacl/fp16/cast_fp16.h"
-#include "nnacl/fp16/pack_fp16.h"
-#include "nnacl/fp16/winograd_transform_fp16.h"
-#include "nnacl/fp16/winograd_utils_fp16.h"
-#include "src/runtime/kernel/arm/fp16/layout_transform_fp16.h"
-#include "schema/model_generated.h"
-#include "src/kernel_registry.h"
-#include "include/errorcode.h"
 #include "src/runtime/runtime_api.h"
 
-using mindspore::kernel::KERNEL_ARCH::kCPU;
-using mindspore::lite::KernelRegistrar;
 using mindspore::lite::RET_ERROR;
 using mindspore::lite::RET_OK;
-using mindspore::schema::PrimitiveType_Conv2D;
 
 namespace mindspore::kernel {
 int ConvolutionWinogradFP16CPUKernel::WinogradFilterTransformFp16(const float16_t *weight_data, float *matrix_g,
@@ -40,104 +27,9 @@ int ConvolutionWinogradFP16CPUKernel::WinogradFilterTransformFp16(const float16_
     MS_LOG(ERROR) << "Divide by zero";
     return RET_ERROR;
   }
-  // original weight format : ohwi
-  auto channel_in = conv_param_->input_channel_;
-  auto channel_out = conv_param_->output_channel_;
-  int oc_block_num = UP_DIV(channel_out, oc_block);
-  int block_stride = channel_in * oc_block;
-  int block_num_stride = block_stride * oc_block_num;
 
-  auto matrix_gt_data_fp16 = reinterpret_cast<float16_t *>(malloc(input_unit_ * kernel_unit_ * sizeof(float16_t)));
-  if (matrix_gt_data_fp16 == nullptr) {
-    MS_LOG(ERROR) << "malloc matrix_gt_data_fp16 failed.";
-    return RET_ERROR;
-  }
-  Float32ToFloat16(matrix_gt, matrix_gt_data_fp16, input_unit_ * kernel_unit_);
-
-  // trans_filter = G*g*GT (g represents weight_data) = [(g * (G)T)T * (G)T]T
-  // separate into two steps ===> tmp = (g * (G)T)T ===> out = [tmp * (G)T]T
-  auto tmp_data = reinterpret_cast<float16_t *>(malloc(channel_in * input_unit_ * kernel_unit_ * sizeof(float16_t)));
-  if (tmp_data == nullptr) {
-    free(matrix_gt_data_fp16);
-    MS_LOG(ERROR) << "malloc tmp_data failed.";
-    return RET_ERROR;
-  }
-  auto trans_out_data =
-    reinterpret_cast<float16_t *>(malloc(channel_in * input_unit_ * input_unit_ * sizeof(float16_t)));
-  if (trans_out_data == nullptr) {
-    free(tmp_data);
-    free(matrix_gt_data_fp16);
-    MS_LOG(ERROR) << "malloc trans_out_data failed.";
-    return RET_ERROR;
-  }
-
-#ifndef ENABLE_ARM64
-  auto tmp_data1 = reinterpret_cast<float16_t *>(malloc(channel_in * input_unit_ * kernel_unit_ * sizeof(float16_t)));
-  if (tmp_data1 == nullptr) {
-    free(tmp_data);
-    free(matrix_gt_data_fp16);
-    free(trans_out_data);
-    MS_LOG(ERROR) << "malloc tmp_data1 failed.";
-    return RET_ERROR;
-  }
-  auto trans_out_data1 =
-    reinterpret_cast<float16_t *>(malloc(channel_in * input_unit_ * input_unit_ * sizeof(float16_t)));
-  if (trans_out_data1 == nullptr) {
-    free(tmp_data);
-    free(tmp_data1);
-    free(matrix_gt_data_fp16);
-    free(trans_out_data);
-    MS_LOG(ERROR) << "malloc trans_out_data1 failed.";
-    return RET_ERROR;
-  }
-#endif
-
-  int input_oz_offset = kernel_unit_ * kernel_unit_ * channel_in;
-  for (int i = 0; i < channel_out; i++) {
-    int out_c_block = i / oc_block;
-    int out_c_res = i % oc_block;
-    int output_oz_offset = out_c_block * block_stride + out_c_res;
-
-#ifndef ENABLE_ARM64
-    // tmp_data = g * GT
-    MatrixMultiplyWinogradFp16(weight_data + i * input_oz_offset, matrix_gt_data_fp16, tmp_data, kernel_unit_,
-                               kernel_unit_, input_unit_, channel_in);
-    // tmp_data1 = (tmp_data)T
-    PackHWCToWHCFp16(tmp_data, tmp_data1, kernel_unit_, input_unit_, channel_in);
-    // trans_out_data1 = tmp * GT
-    MatrixMultiplyWinogradFp16(tmp_data1, matrix_gt_data_fp16, trans_out_data1, input_unit_, kernel_unit_, input_unit_,
-                               channel_in);
-    // trans_out_data = (trans_out_data1)T
-    PackHWCToWHCFp16(trans_out_data1, trans_out_data, input_unit_, input_unit_, channel_in);
-#else
-    // tmp = (g * GT)T
-    MatrixMultiplyWinogradFp16(weight_data + i * input_oz_offset, matrix_gt_data_fp16, tmp_data, kernel_unit_,
-                               kernel_unit_, input_unit_, channel_in);
-    // trans = (tmp * GT)T
-    MatrixMultiplyWinogradFp16(tmp_data, matrix_gt_data_fp16, trans_out_data, input_unit_, kernel_unit_, input_unit_,
-                               channel_in);
-#endif
-
-    int in_offset = 0;
-    for (int j = 0; j < input_unit_; ++j) {
-      for (int k = 0; k < input_unit_; ++k) {
-        for (int c = 0; c < channel_in; ++c) {
-          *(trans_weight_ + output_oz_offset + c * oc_block) = trans_out_data[in_offset + c];
-        }
-        in_offset += channel_in;
-        output_oz_offset += block_num_stride;
-      }
-    }
-  }
-
-#ifndef ENABLE_ARM64
-  free(tmp_data1);
-  free(trans_out_data1);
-#endif
-  free(tmp_data);
-  free(trans_out_data);
-  free(matrix_gt_data_fp16);
-  return RET_OK;
+  return WinogradWeightTransformFp16(weight_data, trans_weight_, matrix_g, matrix_gt, oc_block, input_unit_,
+                                     kernel_unit_, conv_param_->input_channel_, conv_param_->output_channel_, true);
 }
 
 int ConvolutionWinogradFP16CPUKernel::InitWeightBias() {
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc
index 963f929f91..8ca9749ae1 100644
--- a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc
+++ b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.cc
@@ -15,6 +15,7 @@
  */
 
 #include "src/runtime/kernel/arm/fp16/deconvolution_fp16.h"
+#include "src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.h"
 #include "src/runtime/runtime_api.h"
 
 using mindspore::kernel::KERNEL_ARCH::kCPU;
@@ -63,7 +64,7 @@ int DeConvolutionFp16CPUKernel::InitWeightBias() {
   memset(bias_data_, 0, UP_ROUND(output_channel, C4NUM) * sizeof(float16_t));
   if (in_tensors_.size() == 3) {
     Float32ToFloat16(reinterpret_cast<float *>(in_tensors_[2]->MutableData()),
-                     reinterpret_cast<float16_t *>(bias_data_), conv_param_->output_channel_);
+                     reinterpret_cast<float16_t *>(bias_data_), output_channel);
   }
 
   size_t weight_pack_size = input_channel * kernel_w * kernel_h * UP_ROUND(output_channel, C8NUM) * sizeof(float16_t);
@@ -157,9 +158,10 @@ int DeConvolutionFp16CPUKernel::DoDeconv(int task_id) {
   MatMulFp16(pack_input_, execute_weight_ + task_id * thread_stride_ * C8NUM * kernel_plane_ * matmul_param_->deep_,
              tmp_buf, nullptr, ActType_No, matmul_param_->deep_, matmul_param_->row_, oc * C8NUM * kernel_plane_, 0,
              OutType_C8);
+
   DeConvPostFp16(tmp_buf, pack_output_ + task_id * thread_stride_ * C8NUM * output_plane_,
                  reinterpret_cast<float16_t *>(bias_data_) + task_id * thread_stride_ * C8NUM,
-                 execute_output_ + task_id * thread_stride_ * C8NUM, oc_res, conv_param_);
+                 batch_output_ + task_id * thread_stride_ * C8NUM, oc_res, conv_param_);
   return RET_OK;
 }
 
@@ -190,7 +192,10 @@ int DeConvolutionFp16CPUKernel::Run() {
   }
 
   for (int batch_index = 0; batch_index < conv_param_->input_batch_; batch_index++) {
-    RowMajor2Col16MajorFp16Opt(execute_input_, pack_input_, input_plane_, conv_param_->input_channel_);
+    batch_input_ = execute_input_ + batch_index * conv_param_->input_channel_ * input_plane_;
+    batch_output_ = execute_output_ + batch_index * conv_param_->output_channel_ * output_plane_;
+
+    RowMajor2Col16MajorFp16Opt(batch_input_, pack_input_, input_plane_, conv_param_->input_channel_);
 
     error_code = ParallelLaunch(this->context_->thread_pool_, DeConvFp16Run, this, thread_count_);
     if (error_code != RET_OK) {
@@ -227,7 +232,16 @@ kernel::LiteKernel *CpuDeConvFp16KernelCreator(const std::vector<lite::Tensor *>
     weight_tensor->SetData(dequant_weight);
   }
 
-  auto kernel = new (std::nothrow) DeConvolutionFp16CPUKernel(opParameter, inputs, outputs, ctx, primitive);
+  kernel::LiteKernel *kernel;
+  auto conv_param = reinterpret_cast<ConvParameter *>(opParameter);
+  if ((conv_param->stride_h_ != 1 || conv_param->stride_w_ != 1) &&
+      (conv_param->dilation_w_ == 1 && conv_param->dilation_h_ == 1)) {
+    /* DeConvWinogradFp16CPUKernel */
+    kernel = new (std::nothrow) kernel::DeConvolutionFp16CPUKernel(opParameter, inputs, outputs, ctx, primitive);
+  } else {
+    kernel = new (std::nothrow) kernel::DeConvolutionFp16CPUKernel(opParameter, inputs, outputs, ctx, primitive);
+  }
+
   if (kernel == nullptr) {
     MS_LOG(ERROR) << "kernel is nullptr.";
     if (dequant_flag) {
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.h b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.h
index 0dbab90335..f9f0d0fc2b 100644
--- a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.h
+++ b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_fp16.h
@@ -17,17 +17,11 @@
 #ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_DECONVOLUTION_H_
 #define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_DECONVOLUTION_H_
 
-#include <float.h>
 #include <vector>
-#include "src/lite_kernel.h"
-#include "src/kernel_registry.h"
-#include "include/errorcode.h"
-#include "schema/model_generated.h"
-#include "src/runtime/kernel/arm/fp16/convolution_base_fp16.h"
 #include "nnacl/fp16/deconv_fp16.h"
 #include "nnacl/fp16/matmul_fp16.h"
-#include "nnacl/fp16/pack_fp16.h"
-#include "nnacl/fp16/cast_fp16.h"
+#include "src/kernel_registry.h"
+#include "src/runtime/kernel/arm/fp16/convolution_base_fp16.h"
 
 namespace mindspore::kernel {
 class DeConvolutionFp16CPUKernel : public ConvolutionBaseFP16CPUKernel {
@@ -65,6 +59,8 @@ class DeConvolutionFp16CPUKernel : public ConvolutionBaseFP16CPUKernel {
   float16_t *pack_input_;
   float16_t *pack_output_;
   float16_t *tmp_buffer_;
+  float16_t *batch_input_;
+  float16_t *batch_output_;
 };
 }  // namespace mindspore::kernel
 #endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_DECONVOLUTION_H_
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.cc b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.cc
new file mode 100644
index 0000000000..35f8f5f188
--- /dev/null
+++ b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.cc
@@ -0,0 +1,311 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.h"
+#include "src/runtime/runtime_api.h"
+
+using mindspore::lite::RET_ERROR;
+using mindspore::lite::RET_NULL_PTR;
+using mindspore::lite::RET_OK;
+using mindspore::schema::PrimitiveType_DeConv2D;
+using mindspore::schema::Format::Format_NHWC;
+
+namespace mindspore::kernel {
+
+DeConvWinogradFp16CPUKernel::~DeConvWinogradFp16CPUKernel() {
+  FreeResizeBuf();
+  FreeDeconvParam();
+  return;
+}
+
+void DeConvWinogradFp16CPUKernel::FreeResizeBuf() { return; }
+
+void DeConvWinogradFp16CPUKernel::FreeDeconvParam() {
+  if (deconv_param_ != nullptr) {
+    delete deconv_param_;
+    deconv_param_ = nullptr;
+  }
+  return;
+}
+
+int DeConvWinogradFp16CPUKernel::InitParameter() {
+  deconv_param_->input_plane_ = conv_param_->input_h_ * conv_param_->input_w_;
+  deconv_param_->output_plane_ = conv_param_->output_h_ * conv_param_->output_w_;
+
+  nc4hw4_output_ =
+    reinterpret_cast<float16_t *>(malloc(deconv_param_->oc_up4_ * deconv_param_->output_plane_ * sizeof(float16_t)));
+
+  deconv_param_->in_tile_w_count_ = UP_DIV(conv_param_->input_w_, DECONV_WINOGRAD_DEFAULT_UNIT);
+  deconv_param_->in_tile_h_count_ = UP_DIV(conv_param_->input_h_, DECONV_WINOGRAD_DEFAULT_UNIT);
+
+  deconv_param_->in_tile_count_ =
+    UP_DIV(deconv_param_->in_tile_w_count_ * deconv_param_->in_tile_h_count_, DECONV_WINOGRAD_DEFAULT_TILE);
+  deconv_param_->thread_num_ = MSMAX(1, op_parameter_->thread_num_);
+  deconv_param_->thread_num_ = MSMIN(deconv_param_->thread_num_, deconv_param_->in_tile_count_);
+
+  thread_num_hw_ = MSMIN(op_parameter_->thread_num_, deconv_param_->output_plane_);
+  thread_stride_hw_ = UP_DIV(deconv_param_->output_plane_, thread_num_hw_);
+
+  int size = deconv_param_->thread_num_ * DECONV_WINOGRAD_DEFAULT_UNIT * DECONV_WINOGRAD_DEFAULT_UNIT *
+             DECONV_WINOGRAD_DEFAULT_TILE * deconv_param_->ic_up4_;
+  tile_input_ = reinterpret_cast<float16_t *>(malloc(size * sizeof(float16_t)));
+  memset(tile_input_, 0, size * sizeof(float16_t));
+
+  deconv_param_->out_tile_w_ = (DECONV_WINOGRAD_DEFAULT_UNIT - 1) * conv_param_->stride_w_ + conv_param_->kernel_w_;
+  deconv_param_->out_tile_h_ = (DECONV_WINOGRAD_DEFAULT_UNIT - 1) * conv_param_->stride_h_ + conv_param_->kernel_h_;
+  size = deconv_param_->thread_num_ * deconv_param_->out_tile_w_ * deconv_param_->out_tile_h_ *
+         DECONV_WINOGRAD_DEFAULT_TILE * deconv_param_->oc_up4_;
+  tile_output_ = reinterpret_cast<float16_t *>(malloc(size * sizeof(float16_t)));
+
+  for (int i = 0; i < deconv_param_->compute_size_; i++) {
+    DeConvComputeUnit &unit = deconv_param_->compute_units_[i];
+    if (unit.use_winograd_) {
+      if (deconv_param_->a_buffer_[unit.winograd_.kh_].buf_init_ == false) {
+        deconv_param_->a_buffer_[unit.winograd_.kh_].buf_init_ = true;
+        deconv_param_->a_buffer_[unit.winograd_.kh_].trans_formed_ = false;
+
+        size = unit.winograd_.kh_ * unit.winograd_.kw_ * DECONV_WINOGRAD_DEFAULT_TILE * deconv_param_->ic_up4_;
+        deconv_param_->a_buffer_[unit.winograd_.kh_].middle_buffer_ =
+          malloc(deconv_param_->thread_num_ * size * sizeof(float16_t));
+        deconv_param_->a_buffer_[unit.winograd_.kh_].dest_buffer_ =
+          malloc(deconv_param_->thread_num_ * size * sizeof(float16_t));
+      }
+
+      unit.winograd_.b_buffer_ = malloc(deconv_param_->thread_num_ * unit.winograd_.kh_ * unit.winograd_.kw_ *
+                                        deconv_param_->oc_up4_ * DECONV_WINOGRAD_DEFAULT_TILE * sizeof(float16_t));
+      unit.tmp_buffer_ = malloc(deconv_param_->thread_num_ * unit.winograd_.kh_ * unit.winograd_.kw_ *
+                                deconv_param_->oc_div4_ * DECONV_WINOGRAD_DEFAULT_TILE * C4NUM * sizeof(float16_t));
+
+    } else {
+      unit.tmp_buffer_ = malloc(deconv_param_->thread_num_ * deconv_param_->oc_div4_ * unit.w_size_ * unit.h_size_ *
+                                DECONV_WINOGRAD_DEFAULT_TILE * C4NUM * sizeof(float16_t));
+    }
+  }
+
+  return RET_OK;
+}
+
+int DeConvWinogradFp16CPUKernel::DoDeconv(int task_id) {
+  for (int tile_index = task_id; tile_index < deconv_param_->in_tile_count_; tile_index += deconv_param_->thread_num_) {
+    float16_t *tile_in = tile_input_ + task_id * DECONV_WINOGRAD_DEFAULT_UNIT * DECONV_WINOGRAD_DEFAULT_UNIT *
+                                         DECONV_WINOGRAD_DEFAULT_TILE * deconv_param_->ic_up4_;
+    int size = deconv_param_->out_tile_w_ * deconv_param_->out_tile_h_ * DECONV_WINOGRAD_DEFAULT_TILE *
+               deconv_param_->oc_div4_ * C4NUM;
+    float16_t *tile_out = tile_output_ + task_id * size;
+    memset(tile_out, 0, size * sizeof(float16_t));
+
+    int start_index = tile_index * DECONV_WINOGRAD_DEFAULT_TILE;
+    int calculate_count = MSMIN(DECONV_WINOGRAD_DEFAULT_TILE,
+                                deconv_param_->in_tile_w_count_ * deconv_param_->in_tile_h_count_ - start_index);
+
+    for (int i = 0; i < DECONV_WINOGRAD_BUFFER_COUNT; i++) {
+      deconv_param_->a_buffer_[i].trans_formed_ = false;
+    }
+    DeconvWgFp16(nhwc_input_, tile_in, tile_out, start_index, calculate_count, conv_param_, deconv_param_, task_id);
+
+    std::unique_lock<std::mutex> merge_lock(lock_);
+    DeconvWgPostFp16(tile_out, nc4hw4_output_, conv_param_, deconv_param_, calculate_count, tile_index);
+  }
+  return RET_OK;
+}
+
+int DeConvWinogradFp16CPUKernel::DeDeconvPost(int task_id) {
+  int rest_plane = deconv_param_->output_plane_ - task_id * thread_stride_hw_;
+  int current_plane = MSMIN(rest_plane, thread_stride_hw_);
+  if (current_plane <= 0) {
+    return RET_OK;
+  }
+
+  PostConvFuncFp16C4(nc4hw4_output_ + task_id * thread_stride_hw_ * C4NUM,
+                     nhwc_output_ + task_id * thread_stride_hw_ * conv_param_->output_channel_,
+                     reinterpret_cast<float16_t *>(bias_data_), conv_param_->output_channel_, current_plane,
+                     deconv_param_->output_plane_, conv_param_->act_type_);
+  return RET_OK;
+}
+
+int DeConvWgFp16Run(void *cdata, int task_id) {
+  auto deconvWg = reinterpret_cast<DeConvWinogradFp16CPUKernel *>(cdata);
+  deconvWg->DoDeconv(task_id);
+  return RET_OK;
+}
+
+int DeConvWgPostFp16Run(void *cdata, int task_id) {
+  auto deconvWg = reinterpret_cast<DeConvWinogradFp16CPUKernel *>(cdata);
+  deconvWg->DeDeconvPost(task_id);
+  return RET_OK;
+}
+
+int DeConvWinogradFp16CPUKernel::InitComputeParam() {
+  auto weight_tensor = in_tensors_[1];
+
+  conv_param_->input_channel_ = weight_tensor->Batch();
+  conv_param_->output_channel_ = weight_tensor->Channel();
+  conv_param_->kernel_w_ = weight_tensor->Width();
+  conv_param_->kernel_h_ = weight_tensor->Height();
+
+  deconv_param_->kernel_plane_ = conv_param_->kernel_w_ * conv_param_->kernel_h_;
+  deconv_param_->ic_div4_ = UP_DIV(conv_param_->input_channel_, C4NUM);
+  deconv_param_->oc_div4_ = UP_DIV(conv_param_->output_channel_, C4NUM);
+  deconv_param_->ic_up4_ = deconv_param_->ic_div4_ * C4NUM;
+  deconv_param_->oc_up4_ = deconv_param_->oc_div4_ * C4NUM;
+
+  deconv_param_->compute_size_ = 0;
+  for (int si_h = 0; si_h < conv_param_->stride_h_; si_h++) {
+    for (int si_w = 0; si_w < conv_param_->stride_w_; si_w++) {
+      if (si_h < conv_param_->kernel_h_ && si_w < conv_param_->kernel_w_) {
+        deconv_param_->compute_size_++;
+      }
+    }
+  }
+
+  int size = deconv_param_->compute_size_ * sizeof(DeConvComputeUnit);
+  deconv_param_->compute_units_ = reinterpret_cast<DeConvComputeUnit *>(malloc(size));
+  if (deconv_param_->compute_units_ == nullptr) {
+    return RET_NULL_PTR;
+  }
+  int cur_count = 0;
+  for (int si_h = 0; si_h < conv_param_->stride_h_; si_h++) {
+    if (si_h >= conv_param_->kernel_h_) {
+      continue;
+    }
+    for (int si_w = 0; si_w < conv_param_->stride_w_; si_w++) {
+      if (si_w >= conv_param_->kernel_w_) {
+        continue;
+      }
+
+      int h_size = 1 + (conv_param_->kernel_h_ - si_h - 1) / conv_param_->stride_h_;
+      int w_size = 1 + (conv_param_->kernel_w_ - si_w - 1) / conv_param_->stride_w_;
+
+      DeConvComputeUnit unit;
+
+      unit.h_start_ = si_h;
+      unit.w_start_ = si_w;
+      unit.h_size_ = h_size;
+      unit.w_size_ = w_size;
+
+      if (h_size == w_size) {
+        unit.use_winograd_ = true;
+
+        unit.winograd_.k_ = unit.h_size_;
+        unit.winograd_.i_ = DECONV_WINOGRAD_DEFAULT_UNIT;
+        unit.winograd_.o_ = DECONV_WINOGRAD_DEFAULT_UNIT + unit.h_size_ - 1;
+        unit.winograd_.kh_ = unit.h_size_ + DECONV_WINOGRAD_DEFAULT_UNIT - 1;
+        unit.winograd_.kw_ = unit.w_size_ + DECONV_WINOGRAD_DEFAULT_UNIT - 1;
+
+        unit.winograd_.b_buffer_ = nullptr;
+        unit.weight_ = malloc(unit.winograd_.kh_ * unit.winograd_.kw_ * deconv_param_->oc_up4_ *
+                              deconv_param_->ic_up4_ * sizeof(float16_t));
+      } else {
+        unit.use_winograd_ = false;
+        unit.weight_ = malloc(h_size * w_size * deconv_param_->ic_up4_ * deconv_param_->oc_up4_ * sizeof(float16_t));
+      }
+      unit.tmp_buffer_ = nullptr;
+      deconv_param_->compute_units_[cur_count] = unit;
+      cur_count++;
+    }
+  }
+
+  return RET_OK;
+}
+
+int DeConvWinogradFp16CPUKernel::InitDataParam() {
+  /* unit data : weight & winograd data*/
+  auto ret = ConvolutionBaseFP16CPUKernel::GetExecuteFilter();
+  if (ret != RET_OK) {
+    MS_LOG(ERROR) << "Get Execute filter failed.";
+    return ret;
+  }
+  for (int i = 0; i < deconv_param_->compute_size_; i++) {
+    DeConvComputeUnit *unit = &deconv_param_->compute_units_[i];
+    ret = PackDeConvWgDataFp16(fp16_weight_, unit, conv_param_, deconv_param_);
+    if (ret != RET_OK) {
+      return ret;
+    }
+  }
+
+  /* bias */
+  bias_data_ = malloc(deconv_param_->oc_up4_ * sizeof(float16_t));
+  if (bias_data_ == nullptr) {
+    MS_LOG(ERROR) << "malloc bias_data_ failed.";
+    return RET_ERROR;
+  }
+  memset(bias_data_, 0, deconv_param_->oc_up4_ * sizeof(float16_t));
+  auto fp16_bias_data = reinterpret_cast<float16_t *>(bias_data_);
+  if (in_tensors_.size() == kInputSize2) {
+    auto src_bias = reinterpret_cast<float *>(in_tensors_.at(kBiasIndex)->MutableData());
+    for (int i = 0; i < conv_param_->output_channel_; ++i) {
+      fp16_bias_data[i] = (float16_t)src_bias[i];
+    }
+  } else {
+    MS_ASSERT(inputs_.size() == kInputSize1);
+  }
+
+  return RET_OK;
+}
+
+int DeConvWinogradFp16CPUKernel::ReSize() {
+  FreeResizeBuf();
+  ConvolutionBaseCPUKernel::Init();
+  InitParameter();
+  return RET_OK;
+}
+
+int DeConvWinogradFp16CPUKernel::Init() {
+  int error_code = InitComputeParam();
+  if (error_code != RET_OK) {
+    MS_LOG(ERROR) << "InitComputeParam error! ret: " << error_code;
+    return error_code;
+  }
+
+  error_code = InitDataParam();
+  if (error_code != RET_OK) {
+    MS_LOG(ERROR) << "InitWeightBias error! ret: " << error_code;
+    return error_code;
+  }
+
+  if (!InferShapeDone()) {
+    return RET_OK;
+  }
+
+  return ReSize();
+}
+
+int DeConvWinogradFp16CPUKernel::Run() {
+  auto prepare_ret = Prepare();
+  if (prepare_ret != RET_OK) {
+    MS_LOG(ERROR) << "Prepare fail!ret: " << prepare_ret;
+    return prepare_ret;
+  }
+
+  ConvolutionBaseFP16CPUKernel::GetExecuteTensor();
+
+  for (int batch_index = 0; batch_index < conv_param_->input_batch_; batch_index++) {
+    nhwc_input_ = execute_input_ + batch_index * deconv_param_->input_plane_ * conv_param_->input_channel_;
+    nhwc_output_ = execute_output_ + 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_);
+
+    /*post bias activate and nhwc */
+    ParallelLaunch(this->context_->thread_pool_, DeConvWgPostFp16Run, this, thread_num_hw_);
+  }
+
+  ConvolutionBaseFP16CPUKernel::IfCastOutput();
+  ConvolutionBaseFP16CPUKernel::FreeTmpBuffer();
+
+  return RET_OK;
+}
+}  // namespace mindspore::kernel
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.h b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.h
new file mode 100644
index 0000000000..bcbca65043
--- /dev/null
+++ b/mindspore/lite/src/runtime/kernel/arm/fp16/deconvolution_winograd_fp16.h
@@ -0,0 +1,67 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_DECONVOLUTION_WINOGRAD_H_
+#define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_DECONVOLUTION_WINOGRAD_H_
+
+#include <vector>
+#include "include/errorcode.h"
+#include "nnacl/fp16/common_func_fp16.h"
+#include "nnacl/fp16/deconv_winograd_fp16.h"
+#include "nnacl/fp16/pack_fp16.h"
+#include "src/runtime/kernel/arm/fp16/convolution_base_fp16.h"
+
+namespace mindspore::kernel {
+class DeConvWinogradFp16CPUKernel : public ConvolutionBaseFP16CPUKernel {
+ public:
+  DeConvWinogradFp16CPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,
+                              const std::vector<lite::Tensor *> &outputs, const lite::InnerContext *ctx,
+                              const mindspore::lite::PrimitiveC *primitive)
+      : ConvolutionBaseFP16CPUKernel(parameter, inputs, outputs, ctx, primitive) {
+    deconv_param_ = new DeConvParam();
+    for (auto &wg : deconv_param_->a_buffer_) {
+      wg.buf_init_ = false;
+    }
+  }
+  ~DeConvWinogradFp16CPUKernel() override;
+  int Init() override;
+  int Run() override;
+  int ReSize() override;
+
+ public:
+  int DoDeconv(int task_id);
+  int DeDeconvPost(int task_id);
+
+ private:
+  int InitComputeParam();
+  int InitDataParam();
+  int InitParameter();
+  void FreeDeconvParam();
+  void FreeResizeBuf();
+
+ private:
+  DeConvParam *deconv_param_;
+  std::mutex lock_;
+  float16_t *nhwc_input_ = nullptr;
+  float16_t *nhwc_output_ = nullptr;
+  float16_t *nc4hw4_output_ = nullptr;
+  float16_t *tile_input_ = nullptr;
+  float16_t *tile_output_ = nullptr;
+  int thread_num_hw_;
+  int thread_stride_hw_;
+};
+}  // namespace mindspore::kernel
+#endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_DECONVOLUTION_WINOGRAD_H_
diff --git a/mindspore/lite/src/runtime/kernel/arm/fp32/deconvolution_winograd.cc b/mindspore/lite/src/runtime/kernel/arm/fp32/deconvolution_winograd.cc
index 48502a385f..73c589cbe6 100644
--- a/mindspore/lite/src/runtime/kernel/arm/fp32/deconvolution_winograd.cc
+++ b/mindspore/lite/src/runtime/kernel/arm/fp32/deconvolution_winograd.cc
@@ -17,16 +17,11 @@
 #include "src/runtime/kernel/arm/fp32/deconvolution_winograd.h"
 #include "src/runtime/runtime_api.h"
 
-using mindspore::kernel::KERNEL_ARCH::kCPU;
-using mindspore::lite::KernelRegistrar;
 using mindspore::lite::RET_ERROR;
 using mindspore::lite::RET_NULL_PTR;
 using mindspore::lite::RET_OK;
-using mindspore::schema::PrimitiveType_DeConv2D;
-using mindspore::schema::Format::Format_NHWC;
 
 namespace mindspore::kernel {
-
 DeConvolutionWinogradCPUKernel::~DeConvolutionWinogradCPUKernel() {
   FreeResizeBuf();
   FreeDeconvParam();
@@ -352,10 +347,7 @@ 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));
-    for (int i = 0; i < deconv_param_->thread_num_; i++) {
-      DoDeconv(i);
-    }
-    //    ParallelLaunch(this->context_->thread_pool_, DeConvWgFp32Run, this, deconv_param_->thread_num_);
+    ParallelLaunch(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_);