diff --git a/mindspore/ccsrc/pre_activate/ascend/ascend_backend_optimization.cc b/mindspore/ccsrc/pre_activate/ascend/ascend_backend_optimization.cc
index 35b6ebd06c..c8d484b81f 100644
--- a/mindspore/ccsrc/pre_activate/ascend/ascend_backend_optimization.cc
+++ b/mindspore/ccsrc/pre_activate/ascend/ascend_backend_optimization.cc
@@ -66,6 +66,8 @@
 #include "pre_activate/ascend/format_type/check_consistency.h"
 #include "pre_activate/ascend/buffer_fusion/ub_pattern_fusion.h"
 #include "pre_activate/ascend/buffer_fusion/eltwise_fusion_pass.h"
+#include "pre_activate/ascend/buffer_fusion/multi_output_fusion_pass.h"
+#include "pre_activate/ascend/buffer_fusion/stridedread_conv_stridedwrite_fusion_pass.h"
 #include "pre_activate/ascend/buffer_fusion/conv2dbackprop_eltwise_eltwise_fusion_pass.h"
 #include "pre_activate/ascend/buffer_fusion/conv2dbackprop_eltwise_fusion_pass.h"
 #include "pre_activate/ascend/buffer_fusion/conv_single_in_fusion_pass.h"
@@ -365,6 +367,7 @@ void AscendBackendUBFusionOptimization(const std::shared_ptr<session::KernelGrap
   auto ub_fusion_pm = std::make_shared<PassManager>("ub_fusion_pm");
   ub_fusion_pm->AddPass(std::make_shared<Conv2DBackpropEltwiseEltwiseFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<Conv2DBackpropEltwiseFusionPass>(fusion_id_allocator));
+  ub_fusion_pm->AddPass(std::make_shared<StridedReadConvStridedWriteFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<ConvBnReduceFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<ConvSingleInFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<BnupdateEltwiseFusionPass>(fusion_id_allocator));
@@ -373,6 +376,7 @@ void AscendBackendUBFusionOptimization(const std::shared_ptr<session::KernelGrap
   ub_fusion_pm->AddPass(std::make_shared<ConvDoubleInFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<ReduceEltwiseFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<SegmentEltwiseFusionPass>(fusion_id_allocator));
+  ub_fusion_pm->AddPass(std::make_shared<MultiOutputFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<EltwiseFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<DepthwiseConvEltwiseFusionPass>(fusion_id_allocator));
   ub_fusion_pm->AddPass(std::make_shared<UbPatternFusion>());
diff --git a/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/fusion_base_pass.cc b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/fusion_base_pass.cc
index 3f5dd98112..7925462f52 100644
--- a/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/fusion_base_pass.cc
+++ b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/fusion_base_pass.cc
@@ -49,6 +49,19 @@ bool FusionBasePass::CheckDoubleInEltWiseNode(FuncGraphManager *manager, const A
          cnode->inputs().size() == ELTWISE_DOUBLE_IN_INPUT_SIZE;
 }
 
+bool FusionBasePass::CheckMultiOutputEltWiseNode(FuncGraphManager *manager, const AnfNodePtr &node) {
+  MS_EXCEPTION_IF_NULL(manager);
+  if (!node->isa<CNode>() || !AnfAlgo::IsRealCNodeKernel(node) || fusion_id_allocator->HasFusionIdAttr(node)) {
+    return false;
+  }
+  auto cnode = node->cast<CNodePtr>();
+  MS_EXCEPTION_IF_NULL(cnode);
+  auto user_nodes = manager->node_users()[node];
+  return AnfAlgo::GetKernelType(node) == KernelType::TBE_KERNEL &&
+         AnfAlgo::GetFusionType(node) == kernel::FusionType::ELEMWISE && user_nodes.size() == ELTWISE_MULTI_USE &&
+         cnode->inputs().size() == ELTWISE_INPUT_SIZE;
+}
+
 void FusionBasePass::SetRecordFusionId(const std::unordered_set<AnfNodePtr> &record) {
   auto id = fusion_id_allocator->AllocateFusionId();
   for (auto node : record) {
diff --git a/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/fusion_base_pass.h b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/fusion_base_pass.h
index 421efa9716..8d6eca774c 100644
--- a/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/fusion_base_pass.h
+++ b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/fusion_base_pass.h
@@ -33,8 +33,12 @@ const int8_t MAX_ELTWISE_NUM = 3;
 const int8_t MIN_ELTWISE_SIZE = 2;
 const int8_t ELTWISE_INPUT_SIZE = 2;
 const int8_t ELTWISE_DOUBLE_IN_INPUT_SIZE = 3;
+const int8_t CONV_DOUBLE_IN_INPUT_SIZE = 3;
+const int8_t CONV_QUART_IN_INPUT_SIZE = 5;
 const int8_t ELTWISE_USE = 1;
+const int8_t ELTWISE_MULTI_USE = 2;
 const int8_t MAX_ELTWISE_SIZE = 6;
+const int8_t MULTI_ELTWISE_SIZE = 4;
 using FusedNodeRecord = std::vector<std::unordered_set<AnfNodePtr>>;
 
 struct BufferFusionInfo_t {
@@ -58,6 +62,7 @@ class FusionBasePass : public Pass {
   void SetRecordFusionId(const std::unordered_set<AnfNodePtr> &record);
   bool CheckEltWiseNode(FuncGraphManager *manager, const AnfNodePtr &node);
   bool CheckDoubleInEltWiseNode(FuncGraphManager *manager, const AnfNodePtr &node);
+  bool CheckMultiOutputEltWiseNode(FuncGraphManager *manager, const AnfNodePtr &node);
   FusionIdAllocatorPtr fusion_id_allocator;
 };
 }  // namespace opt
diff --git a/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/multi_output_fusion_pass.cc b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/multi_output_fusion_pass.cc
new file mode 100644
index 0000000000..01f36d4047
--- /dev/null
+++ b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/multi_output_fusion_pass.cc
@@ -0,0 +1,81 @@
+/**
+ * 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 "pre_activate/ascend/buffer_fusion/multi_output_fusion_pass.h"
+#include <vector>
+#include <unordered_set>
+#include <memory>
+#include <string>
+#include "kernel/kernel_fusion.h"
+#include "debug/anf_ir_dump.h"
+#include "session/anf_runtime_algorithm.h"
+#include "operator/ops.h"
+#include "utils/context/ms_context.h"
+#include "pre_activate/common/fusion_id_allocator.h"
+
+namespace mindspore {
+namespace opt {
+void MultiOutputFusionPass::MatchMultiOutputEltwise(const CNodePtr &cnode, const session::KernelGraph &kernel_graph,
+                                                    FusedNodeRecord *candidate_fusion) {
+  MS_EXCEPTION_IF_NULL(cnode);
+  MS_EXCEPTION_IF_NULL(candidate_fusion);
+  auto manager = kernel_graph.manager();
+  MS_EXCEPTION_IF_NULL(manager);
+  std::unordered_set<AnfNodePtr> record{cnode};
+  auto eltwise_input = cnode->input(1);
+  if (CheckMultiOutputEltWiseNode(manager.get(), eltwise_input)) {
+    (void)record.insert(eltwise_input);
+    auto input_cnode = eltwise_input->cast<CNodePtr>();
+    MS_EXCEPTION_IF_NULL(input_cnode);
+    eltwise_input = input_cnode->input(1);
+  } else {
+    return;
+  }
+  while (CheckEltWiseNode(manager.get(), eltwise_input)) {
+    (void)record.insert(eltwise_input);
+    if (record.size() == MULTI_ELTWISE_SIZE) {
+      break;
+    }
+    auto input_cnode = eltwise_input->cast<CNodePtr>();
+    MS_EXCEPTION_IF_NULL(input_cnode);
+    eltwise_input = input_cnode->input(1);
+  }
+  if (record.size() != MULTI_ELTWISE_SIZE) {
+    return;
+  }
+  candidate_fusion->push_back(record);
+  SetRecordFusionId(record);
+}
+
+void MultiOutputFusionPass::MatchSingleFusionPattern(const session::KernelGraph &kernel_graph,
+                                                     FusedNodeRecord *candidate_fusion) {
+  MS_EXCEPTION_IF_NULL(candidate_fusion);
+  std::vector<AnfNodePtr> node_list = TopoSort(kernel_graph.get_return());
+  std::reverse(node_list.begin(), node_list.end());
+  for (auto &node : node_list) {
+    if (!AnfAlgo::IsRealCNodeKernel(node) || fusion_id_allocator->HasFusionIdAttr(node) ||
+        AnfAlgo::CheckPrimitiveType(node, prim::kPrimReturn)) {
+      continue;
+    }
+    auto cnode = node->cast<CNodePtr>();
+    MS_EXCEPTION_IF_NULL(cnode);
+    if (AnfAlgo::GetKernelType(cnode) == KernelType::TBE_KERNEL &&
+        AnfAlgo::GetFusionType(cnode) == kernel::FusionType::ELEMWISE && cnode->inputs().size() == ELTWISE_INPUT_SIZE) {
+      MatchMultiOutputEltwise(cnode, kernel_graph, candidate_fusion);
+    }
+  }
+}
+}  // namespace opt
+}  // namespace mindspore
diff --git a/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/multi_output_fusion_pass.h b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/multi_output_fusion_pass.h
new file mode 100644
index 0000000000..0e2510128a
--- /dev/null
+++ b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/multi_output_fusion_pass.h
@@ -0,0 +1,48 @@
+/**
+ * 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_CCSRC_PRE_ACTIVATE_ASCEND_BUFFER_FUSION_PASS_MULTI_OUTPUT_FUSION_PASS_H_
+#define MINDSPORE_CCSRC_PRE_ACTIVATE_ASCEND_BUFFER_FUSION_PASS_MULTI_OUTPUT_FUSION_PASS_H_
+
+#include <unordered_set>
+#include <vector>
+
+#include "pre_activate/ascend/buffer_fusion/fusion_base_pass.h"
+#include "ir/anf.h"
+#include "pre_activate/common/pass.h"
+#include "pre_activate/common/fusion_id_allocator.h"
+#include "device/kernel_info.h"
+#include "kernel/kernel.h"
+#include "session/kernel_graph.h"
+
+namespace mindspore {
+namespace opt {
+using FusedNodeRecord = std::vector<std::unordered_set<AnfNodePtr>>;
+
+class MultiOutputFusionPass : public FusionBasePass {
+ public:
+  explicit MultiOutputFusionPass(FusionIdAllocatorPtr idAllocator)
+      : FusionBasePass("MultiOutputFusionPass", idAllocator) {}
+  ~MultiOutputFusionPass() override = default;
+  void MatchSingleFusionPattern(const session::KernelGraph &kernel_graph, FusedNodeRecord *candidate_fusion) override;
+
+ private:
+  void MatchMultiOutputEltwise(const CNodePtr &cnode, const session::KernelGraph &kernel_graph,
+                               FusedNodeRecord *candidate_fusion);
+};
+}  // namespace opt
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CCSRC_PRE_ACTIVATE_ASCEND_BUFFER_FUSION_PASS_MULTI_OUTPUT_FUSION_PASS_H_
diff --git a/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/stridedread_conv_stridedwrite_fusion_pass.cc b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/stridedread_conv_stridedwrite_fusion_pass.cc
new file mode 100644
index 0000000000..eb23b3d9cf
--- /dev/null
+++ b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/stridedread_conv_stridedwrite_fusion_pass.cc
@@ -0,0 +1,90 @@
+/**
+ * Copyright 2020 Huawei Technologies Co., Ltd
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include "pre_activate/ascend/buffer_fusion/stridedread_conv_stridedwrite_fusion_pass.h"
+
+#include <vector>
+#include <unordered_set>
+#include <memory>
+#include <string>
+#include "kernel/kernel_fusion.h"
+#include "debug/anf_ir_dump.h"
+#include "session/anf_runtime_algorithm.h"
+#include "operator/ops.h"
+#include "utils/context/ms_context.h"
+#include "pre_activate/common/fusion_id_allocator.h"
+
+namespace mindspore {
+namespace opt {
+void StridedReadConvStridedWriteFusionPass::MatchStridedReadConvStridedWrite(const CNodePtr &cnode,
+                                                                             const session::KernelGraph &kernel_graph,
+                                                                             FusedNodeRecord *candidate_fusion) {
+  MS_EXCEPTION_IF_NULL(cnode);
+  MS_EXCEPTION_IF_NULL(candidate_fusion);
+  auto manager = kernel_graph.manager();
+  MS_EXCEPTION_IF_NULL(manager);
+  std::unordered_set<AnfNodePtr> record{cnode};
+  auto write_input = cnode->input(1);
+
+  if (CheckEltWiseNode(manager.get(), write_input)) {
+    (void)record.insert(write_input);
+    auto input_cnode = write_input->cast<CNodePtr>();
+    MS_EXCEPTION_IF_NULL(input_cnode);
+    write_input = input_cnode->input(1);
+  }
+
+  if (!write_input->isa<CNode>() || !AnfAlgo::IsRealCNodeKernel(write_input) ||
+      fusion_id_allocator->HasFusionIdAttr(write_input)) {
+    return;
+  }
+  auto conv_cnode = write_input->cast<CNodePtr>();
+  MS_EXCEPTION_IF_NULL(conv_cnode);
+  if (AnfAlgo::GetKernelType(conv_cnode) == KernelType::TBE_KERNEL &&
+      AnfAlgo::GetFusionType(conv_cnode) == kernel::FusionType::CONVLUTION &&
+      conv_cnode->inputs().size() >= CONV_DOUBLE_IN_INPUT_SIZE &&
+      conv_cnode->inputs().size() <= CONV_QUART_IN_INPUT_SIZE) {
+    (void)record.insert(write_input);
+    auto conv_input = conv_cnode->input(1);
+    if (!conv_input->isa<CNode>() || !AnfAlgo::IsRealCNodeKernel(conv_input) ||
+        fusion_id_allocator->HasFusionIdAttr(conv_input)) {
+      return;
+    }
+
+    if (AnfAlgo::GetCNodeName(conv_input) == kStridedReadOpName) {
+      (void)record.insert(conv_input);
+      candidate_fusion->push_back(record);
+      SetRecordFusionId(record);
+    }
+  }
+}
+
+void StridedReadConvStridedWriteFusionPass::MatchSingleFusionPattern(const session::KernelGraph &kernel_graph,
+                                                                     FusedNodeRecord *candidate_fusion) {
+  MS_EXCEPTION_IF_NULL(candidate_fusion);
+  std::vector<AnfNodePtr> node_list = TopoSort(kernel_graph.get_return());
+  for (auto &node : node_list) {
+    if (!AnfAlgo::IsRealCNodeKernel(node) || fusion_id_allocator->HasFusionIdAttr(node) ||
+        AnfAlgo::CheckPrimitiveType(node, prim::kPrimReturn)) {
+      continue;
+    }
+    auto cnode = node->cast<CNodePtr>();
+    MS_EXCEPTION_IF_NULL(cnode);
+    if (AnfAlgo::GetCNodeName(cnode) == kStridedWriteOpName) {
+      MatchStridedReadConvStridedWrite(cnode, kernel_graph, candidate_fusion);
+    }
+  }
+}
+}  // namespace opt
+}  // namespace mindspore
diff --git a/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/stridedread_conv_stridedwrite_fusion_pass.h b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/stridedread_conv_stridedwrite_fusion_pass.h
new file mode 100644
index 0000000000..c6c5fe88dc
--- /dev/null
+++ b/mindspore/ccsrc/pre_activate/ascend/buffer_fusion/stridedread_conv_stridedwrite_fusion_pass.h
@@ -0,0 +1,48 @@
+/**
+ * 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_CCSRC_PRE_ACTIVATE_ASCEND_BUFFER_FUSION_STRIDEDREAD_CONV_STRIDEDWRITE_FUSION_PASS_H_
+#define MINDSPORE_CCSRC_PRE_ACTIVATE_ASCEND_BUFFER_FUSION_STRIDEDREAD_CONV_STRIDEDWRITE_FUSION_PASS_H_
+
+#include <unordered_set>
+#include <vector>
+
+#include "pre_activate/ascend/buffer_fusion/fusion_base_pass.h"
+#include "ir/anf.h"
+#include "pre_activate/common/pass.h"
+#include "pre_activate/common/fusion_id_allocator.h"
+#include "device/kernel_info.h"
+#include "kernel/kernel.h"
+#include "session/kernel_graph.h"
+
+namespace mindspore {
+namespace opt {
+using FusedNodeRecord = std::vector<std::unordered_set<AnfNodePtr>>;
+
+class StridedReadConvStridedWriteFusionPass : public FusionBasePass {
+ public:
+  explicit StridedReadConvStridedWriteFusionPass(FusionIdAllocatorPtr idAllocator)
+      : FusionBasePass("StridedReadConvStridedWriteFusionPass", idAllocator) {}
+  ~StridedReadConvStridedWriteFusionPass() override = default;
+  void MatchSingleFusionPattern(const session::KernelGraph &kernel_graph, FusedNodeRecord *candidate_fusion) override;
+
+ private:
+  void MatchStridedReadConvStridedWrite(const CNodePtr &cnode, const session::KernelGraph &kernel_graph,
+                                        FusedNodeRecord *candidate_fusion);
+};
+}  // namespace opt
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CCSRC_PRE_ACTIVATE_ASCEND_BUFFER_FUSION_STRIDEDREAD_CONV_STRIDEDWRITE_FUSION_PASS_H_
diff --git a/mindspore/ccsrc/utils/utils.h b/mindspore/ccsrc/utils/utils.h
index 2ea3e7ba94..ff2ba05c84 100644
--- a/mindspore/ccsrc/utils/utils.h
+++ b/mindspore/ccsrc/utils/utils.h
@@ -154,6 +154,8 @@ constexpr auto kLarsV2UpdateOpName = "LarsV2Update";
 constexpr auto kSquareSumAllOpName = "SquareSumAll";
 constexpr auto kNMSWithMaskOpName = "NMSWithMask";
 constexpr auto kSoftmaxGradExtOpName = "SoftmaxGradExt";
+constexpr auto kStridedReadOpName = "StridedRead";
+constexpr auto kStridedWriteOpName = "StridedWrite";
 
 // attr key name
 constexpr auto kAttrInputNames = "input_names";