Merge branch 'development' of gitee.com:mindspore/graphengine into development

build.sh

@@ -23,7 +23,7 @@ export BUILD_PATH="${BASEPATH}/build/"
 usage()
 {
   echo "Usage:"
-  echo "sh build.sh [-j[n]] [-h] [-v] [-s] [-t] [-u] [-c] [-p]"
+  echo "sh build.sh [-j[n]] [-h] [-v] [-s] [-t] [-u] [-c] [-S on|off]"
   echo ""
   echo "Options:"
   echo "    -h Print usage"
@@ -34,9 +34,21 @@ usage()
   echo "    -c Build ut with coverage tag"
   echo "    -p Build inference or train"
   echo "    -v Display build command"
+  echo "    -S Enable enable download cmake compile dependency from gitee , default off"
   echo "to be continued ..."
 }
 
+# check value of input is 'on' or 'off'
+# usage: check_on_off arg_value arg_name
+check_on_off()
+{
+  if [[ "X$1" != "Xon" && "X$1" != "Xoff" ]]; then
+    echo "Invalid value $1 for option -$2"
+    usage
+    exit 1
+  fi
+}
+
 # parse and set options
 checkopts()
 {
@@ -49,8 +61,9 @@ checkopts()
   GE_ONLY="on"
   PLATFORM="inference"
   PRODUCT="normal"
+  ENABLE_GITEE="off"
   # Process the options
-  while getopts 'ustchj:p:g:v' opt
+  while getopts 'ustchj:p:g:vS:' opt
   do
     OPTARG=$(echo ${OPTARG} | tr '[A-Z]' '[a-z]')
     case "${opt}" in
@@ -86,6 +99,11 @@ checkopts()
       g)
         PRODUCT=$OPTARG
         ;;
+      S)
+        check_on_off $OPTARG S
+        ENABLE_GITEE="$OPTARG"
+        echo "enable download from gitee"
+        ;;
       *)
         echo "Undefined option: ${opt}"
         usage
@@ -129,6 +147,9 @@ build_graphengine()
     CMAKE_ARGS="${CMAKE_ARGS} -DENABLE_GE_ST=ON"
   fi
 
+  if [[ "X$ENABLE_GITEE" = "Xon" ]]; then
+    CMAKE_ARGS="${CMAKE_ARGS} -DENABLE_GITEE=ON"
+  fi
   CMAKE_ARGS="${CMAKE_ARGS} -DENABLE_OPEN_SRC=True -DCMAKE_INSTALL_PREFIX=${OUTPUT_PATH} -DPLATFORM=${PLATFORM} -DPRODUCT=${PRODUCT}"
   echo "${CMAKE_ARGS}"
   cmake ${CMAKE_ARGS} ..

cmake/external_libs/gflags.cmake

@@ -11,8 +11,16 @@ if ((${CMAKE_INSTALL_PREFIX} STREQUAL /usr/local) OR
 endif()
 
 set (gflags_CXXFLAGS "-D_GLIBCXX_USE_CXX11_ABI=0 -Dgoogle=ascend_private")
+if (ENABLE_GITEE)
+    set(REQ_URL "https://gitee.com/mirrors/gflags/repository/archive/v2.2.2.tar.gz")
+    set(MD5 "")
+else()
+    set(REQ_URL "https://github.com/gflags/gflags/archive/v2.2.2.tar.gz")
+    set(MD5 "")
+endif ()
+
 ExternalProject_Add(gflags_build
-                    URL https://github.com/gflags/gflags/archive/v2.2.2.tar.gz
+                    URL ${REQ_URL}
                     #URL /home/txd/workspace/linux_cmake/pkg/protobuf-3.8.0.tar.gz
                     #SOURCE_DIR ${METADEF_DIR}/../../third_party/gflags/src/gflags-2.2.2 
                     CONFIGURE_COMMAND ${CMAKE_COMMAND} -DCMAKE_CXX_FLAGS=${gflags_CXXFLAGS} -DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}/gflags <SOURCE_DIR>

cmake/external_libs/json.cmake

@@ -5,8 +5,17 @@ endif()
 include(ExternalProject)
 
 set(JSON_SRC_DIR ${CMAKE_BINARY_DIR}/opensrc/json/include)
+if (ENABLE_GITEE)
+    set(REQ_URL "https://gitee.com/mirrors/JSON-for-Modern-CPP/repository/archive/v3.6.1.zip")
+    set(MD5 "5bda78ce308e6cfcf614dcf1d5ff27a7")
+    set(JSON_INCLUDE_DIR "${JSON_SRC_DIR}/include")
+else()
+    set(REQ_URL "https://github.com/nlohmann/json/releases/download/v3.6.1/include.zip")
+    set(MD5 "0dc903888211db3a0f170304cd9f3a89")
+    set(JSON_INCLUDE_DIR ${JSON_SRC_DIR})
+endif ()
 ExternalProject_Add(json_build
-                    URL https://github.com/nlohmann/json/releases/download/v3.6.1/include.zip
+                    URL ${REQ_URL}
                     #URL /home/txd/workspace/cloud_code/pkg/include.zip
                     SOURCE_DIR  ${JSON_SRC_DIR}
                     CONFIGURE_COMMAND ""
@@ -17,7 +26,7 @@ ExternalProject_Add(json_build
 
 
 add_library(json INTERFACE)
-target_include_directories(json INTERFACE ${JSON_SRC_DIR})
+target_include_directories(json INTERFACE ${JSON_INCLUDE_DIR})
 add_dependencies(json json_build)
 
 #set(HAVE_JSON TRUE CACHE BOOL "json build add")

cmake/external_libs/onnx.cmake

@@ -6,8 +6,16 @@ set(ONNX_PROTO_DIR ${CMAKE_BINARY_DIR}/onnx)
 set(ONNX_PROTO_FILE ${ONNX_PROTO_DIR}/onnx.proto)
 file(MAKE_DIRECTORY ${ONNX_PROTO_DIR})
 
+if (ENABLE_GITEE)
+    set(REQ_URL "https://gitee.com/mirrors/ONNX/repository/archive/v1.6.0.tar.gz")
+    set(MD5 "1bdbcecdd68ea8392630467646776e02")
+else()
+    set(REQ_URL "https://github.com/onnx/onnx/releases/download/v1.6.0/onnx-1.6.0.tar.gz")
+    set(MD5 "512f2779d6215d4a36f366b6b9acdf1e")
+endif ()
+
 ExternalProject_Add(onnx
-                    URL https://github.com/onnx/onnx/releases/download/v1.6.0/onnx-1.6.0.tar.gz
+                    URL ${REQ_URL}
                     #URL /home/txd/workspace/cloud_code/pkg/onnx-1.6.0.tar.gz
                     #URL_HASH SHA256=3b88c3fe521151651a0403c4d131cb2e0311bd28b753ef692020a432a81ce345
                     #SOURCE_DIR ${ONNX_SRC_DIR}

cmake/external_libs/protobuf_shared.cmake

@@ -11,6 +11,13 @@ if ((${CMAKE_INSTALL_PREFIX} STREQUAL /usr/local) OR
     message(STATUS "No install prefix selected, default to ${CMAKE_INSTALL_PREFIX}.")
 endif()
 
+if (ENABLE_GITEE)
+    set(REQ_URL "https://gitee.com/mirrors/protobuf_source/repository/archive/v3.8.0.tar.gz")
+    set(MD5 "eba86ae9f07ba5cfbaf8af3bc4e84236")
+else()
+    set(REQ_URL "https://github.com/protocolbuffers/protobuf/archive/v3.8.0.tar.gz")
+    set(MD5 "3d9e32700639618a4d2d342c99d4507a")
+endif ()
 set(protobuf_CXXFLAGS "-Wno-maybe-uninitialized -Wno-unused-parameter -fPIC -fstack-protector-all -D_FORTIFY_SOURCE=2 -D_GLIBCXX_USE_CXX11_ABI=0 -O2 -Dgoogle=ascend_private")
 set(protobuf_LDFLAGS "-Wl,-z,relro,-z,now,-z,noexecstack")
 ExternalProject_Add(protobuf_build

cmake/external_libs/protobuf_static.cmake

@@ -8,11 +8,18 @@ if ((${CMAKE_INSTALL_PREFIX} STREQUAL /usr/local) OR
     message(STATUS "No install prefix selected, default to ${CMAKE_INSTALL_PREFIX}.")
 endif()
 
+if (ENABLE_GITEE)
+    set(REQ_URL "https://gitee.com/mirrors/protobuf_source/repository/archive/v3.8.0.tar.gz")
+    set(MD5 "eba86ae9f07ba5cfbaf8af3bc4e84236")
+else()
+    set(REQ_URL "https://github.com/protocolbuffers/protobuf/archive/v3.8.0.tar.gz")
+    set(MD5 "3d9e32700639618a4d2d342c99d4507a")
+endif ()
 set(protobuf_CXXFLAGS "-Wno-maybe-uninitialized -Wno-unused-parameter -fPIC -fstack-protector-all -D_FORTIFY_SOURCE=2 -D_GLIBCXX_USE_CXX11_ABI=0 -O2 -Dgoogle=ascend_private")
 set(protobuf_LDFLAGS "-Wl,-z,relro,-z,now,-z,noexecstack")
 set(PROTOBUF_STATIC_PKG_DIR ${CMAKE_INSTALL_PREFIX}/protobuf_static)
 ExternalProject_Add(protobuf_static_build
-                    URL https://github.com/protocolbuffers/protobuf/archive/v3.8.0.tar.gz
+                    URL ${REQ_URL}
                     #URL /home/txd/workspace/linux_cmake/pkg/protobuf-3.8.0.tar.gz
                     #SOURCE_DIR ${GE_CODE_DIR}/../../third_party/protobuf/src/protobuf-3.8.0
                     CONFIGURE_COMMAND ${CMAKE_COMMAND}

cmake/external_libs/protoc.cmake

@@ -12,10 +12,19 @@ if ((${CMAKE_INSTALL_PREFIX} STREQUAL /usr/local) OR
     message(STATUS "No install prefix selected, default to ${CMAKE_INSTALL_PREFIX}.")
 endif()
 
+if (ENABLE_GITEE)
+    set(REQ_URL "https://gitee.com/mirrors/protobuf_source/repository/archive/v3.8.0.tar.gz")
+    set(MD5 "eba86ae9f07ba5cfbaf8af3bc4e84236")
+else()
+    set(REQ_URL "https://github.com/protocolbuffers/protobuf/archive/v3.8.0.tar.gz")
+    set(MD5 "3d9e32700639618a4d2d342c99d4507a")
+endif ()
+
+
 set(protobuf_CXXFLAGS "-Wno-maybe-uninitialized -Wno-unused-parameter -fPIC -fstack-protector-all -D_FORTIFY_SOURCE=2 -D_GLIBCXX_USE_CXX11_ABI=0 -O2")
 set(protobuf_LDFLAGS "-Wl,-z,relro,-z,now,-z,noexecstack")
 ExternalProject_Add(protoc_build
-                    URL https://github.com/protocolbuffers/protobuf/archive/v3.8.0.tar.gz
+                    URL ${REQ_URL}
                     #URL /home/txd/workspace/linux_cmake/pkg/protobuf-3.8.0.tar.gz
                     #SOURCE_DIR ${GE_CODE_DIR}/../third_party/protobuf/src/protobuf-3.8.0
                     CONFIGURE_COMMAND ${CMAKE_COMMAND} -Dprotobuf_WITH_ZLIB=OFF -Dprotobuf_BUILD_TESTS=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_FLAGS=${protobuf_CXXFLAGS} -DCMAKE_CXX_LDFLAGS=${protobuf_LDFLAGS} -DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}/protoc <SOURCE_DIR>/cmake

ge/CMakeLists.txt

@@ -239,6 +239,7 @@ set(TRAIN_SRC_LIST
     "graph/passes/switch_data_edges_bypass.cc"
     "graph/passes/switch_logic_remove_pass.cc"
     "graph/passes/merge_to_stream_merge_pass.cc"
+    "graph/passes/merge_input_memcpy_pass.cc"
     "graph/passes/switch_to_stream_switch_pass.cc"
     "graph/passes/attach_stream_label_pass.cc"
     "graph/passes/switch_dead_branch_elimination.cc"
@@ -559,6 +560,7 @@ set(INFER_SRC_LIST
     "graph/passes/cast_translate_pass.cc"
     "graph/passes/prune_pass.cc"
     "graph/passes/merge_to_stream_merge_pass.cc"
+    "graph/passes/merge_input_memcpy_pass.cc"
     "graph/passes/switch_to_stream_switch_pass.cc"
     "graph/passes/attach_stream_label_pass.cc"
     "graph/passes/multi_batch_pass.cc"

ge/ge_inference.mk

@@ -177,6 +177,7 @@ OMG_HOST_SRC_FILES := \
     graph/passes/cast_translate_pass.cc \
     graph/passes/prune_pass.cc \
     graph/passes/merge_to_stream_merge_pass.cc \
+    graph/passes/merge_input_memcpy_pass.cc \
     graph/passes/switch_to_stream_switch_pass.cc \
     graph/passes/attach_stream_label_pass.cc \
     graph/passes/multi_batch_pass.cc \

ge/ge_runner.mk

@@ -212,6 +212,7 @@ LIBGE_LOCAL_SRC_FILES := \
     graph/passes/switch_data_edges_bypass.cc \
     graph/passes/switch_logic_remove_pass.cc \
     graph/passes/merge_to_stream_merge_pass.cc \
+    graph/passes/merge_input_memcpy_pass.cc \
     graph/passes/switch_to_stream_switch_pass.cc \
     graph/passes/attach_stream_label_pass.cc \
     graph/passes/switch_dead_branch_elimination.cc \

ge/graph/manager/graph_manager.cc

@@ -69,6 +69,7 @@
 #include "graph/passes/link_gen_mask_nodes_pass.h"
 #include "graph/passes/mark_graph_unknown_status_pass.h"
 #include "graph/passes/merge_pass.h"
+#include "graph/passes/merge_input_memcpy_pass.h"
 #include "graph/passes/merge_to_stream_merge_pass.h"
 #include "graph/passes/multi_batch_pass.h"
 #include "graph/passes/next_iteration_pass.h"
@@ -1961,6 +1962,8 @@ Status GraphManager::OptimizeStage1(ge::ComputeGraphPtr &compute_graph) {
   }
   PassManager after_merge_passes;
   GE_CHK_STATUS_RET(
+      after_merge_passes.AddPass("OptimizeStage1_1::MergeInputMemcpyPass", new (std::nothrow) MergeInputMemcpyPass));
+  GE_CHK_STATUS_RET(
       after_merge_passes.AddPass("OptimizeStage1_1::SwitchDataEdgesBypass", new (std::nothrow) SwitchDataEdgesBypass));
   GE_CHK_STATUS_RET(
       after_merge_passes.AddPass("OptimizeStage1_1::ConstantFuseSamePass", new (std::nothrow) ConstantFuseSamePass));

ge/graph/passes/merge_input_memcpy_pass.cc

@@ -0,0 +1,98 @@
+/**
+ * Copyright 2019-2020 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 "graph/passes/merge_input_memcpy_pass.h"
+#include "common/ge/ge_util.h"
+#include "ge/ge_api_types.h"
+#include "graph/common/omg_util.h"
+
+namespace ge {
+Status MergeInputMemcpyPass::Run(ComputeGraphPtr graph) {
+  GELOGD("MergeInputMemcpyPass Enter");
+  for (const auto &node : graph->GetDirectNode()) {
+    if ((node->GetType() != MERGE) && (node->GetType() != REFMERGE)) {
+      continue;
+    }
+    GE_CHECK_NOTNULL(node->GetOpDesc());
+    GE_CHK_STATUS_RET(AddMemcpyAsyncNodes(graph, node, node->GetOpDesc()->HasAttr(ATTR_INSERT_BY_MBATCH)),
+                      "Merge add memcpy node failed.");
+  }
+  GELOGD("MergeInputMemcpyPass Leave");
+  return SUCCESS;
+}
+
+///
+/// @brief Add MemcpyAsync Op as Merge in_node
+/// @param [in] graph
+/// @param [in] node
+/// @param [in] multi_batch_flag
+/// @return Status
+///
+Status MergeInputMemcpyPass::AddMemcpyAsyncNodes(const ComputeGraphPtr &graph, const NodePtr &node,
+                                                 bool multi_batch_flag) {
+  for (const InDataAnchorPtr &in_data_anchor : node->GetAllInDataAnchors()) {
+    OutDataAnchorPtr peer_out_anchor = in_data_anchor->GetPeerOutAnchor();
+    GE_IF_BOOL_EXEC(peer_out_anchor == nullptr, continue);
+    NodePtr in_node = peer_out_anchor->GetOwnerNode();
+    const std::string &type = in_node->GetType();
+    // For WhileLoop no need memcpy for merge.
+    GE_IF_BOOL_EXEC((type == ENTER) || (type == REFENTER) || (type == NEXTITERATION) || (type == REFNEXTITERATION),
+                    continue);
+
+    const std::string &memcpy_name = node->GetName() + "_input_" + std::to_string(in_data_anchor->GetIdx());
+    NodePtr memcpy_node = CreateMemcpyAsyncNode(graph, memcpy_name, peer_out_anchor, multi_batch_flag);
+    GE_CHK_BOOL_EXEC(memcpy_node != nullptr, return FAILED, "Create MemcpyAsync node failed.");
+    GE_CHK_STATUS(GraphUtils::RemoveEdge(peer_out_anchor, in_data_anchor), "MemcpyAsync node remove edge failed.");
+    GE_CHK_STATUS(GraphUtils::AddEdge(peer_out_anchor, memcpy_node->GetInDataAnchor(0)),
+                  "MemcpyAsync node add edge failed.");
+    GE_CHK_STATUS(GraphUtils::AddEdge(memcpy_node->GetOutDataAnchor(0), in_data_anchor),
+                  "MemcpyAsync node add edge failed.");
+  }
+
+  return SUCCESS;
+}
+
+///
+/// @brief Add MemcpyAsync Node
+/// @param [in] graph
+/// @param [in] name
+/// @param [in] out_data_anchor
+/// @param [in] multi_batch_flag
+/// @return ge::NodePtr
+///
+NodePtr MergeInputMemcpyPass::CreateMemcpyAsyncNode(const ComputeGraphPtr &graph, const std::string &name,
+                                                    const OutDataAnchorPtr &out_data_anchor, bool multi_batch_flag) {
+  OpDescPtr pre_op_desc = out_data_anchor->GetOwnerNode()->GetOpDesc();
+  GE_CHK_BOOL_EXEC(pre_op_desc != nullptr, return nullptr, "OpDesc of pre node is invalid.");
+
+  const std::string &memcpy_type = multi_batch_flag ? MEMCPYADDRASYNC : MEMCPYASYNC;
+  const std::string &node_name = name + "_" + memcpy_type;
+  GELOGI("Create MemcpyAsync op:%s.", node_name.c_str());
+  OpDescPtr op_desc = MakeShared<OpDesc>(node_name, memcpy_type);
+  if (op_desc == nullptr) {
+    GELOGE(FAILED, "Create op_desc failed, MemcpyAsync:%s.", node_name.c_str());
+    return nullptr;
+  }
+
+  GE_CHK_BOOL_EXEC(op_desc->AddInputDesc(pre_op_desc->GetOutputDesc(out_data_anchor->GetIdx())) == GRAPH_SUCCESS,
+                   return nullptr, "Create MemcpyAsync op: add input desc failed.");
+  GE_CHK_BOOL_EXEC(op_desc->AddOutputDesc(pre_op_desc->GetOutputDesc(out_data_anchor->GetIdx())) == GRAPH_SUCCESS,
+                   return nullptr, "Create MemcpyAsync op: add output desc failed.");
+
+  return graph->AddNode(op_desc);
+}
+}  // namespace ge
+

ge/graph/passes/merge_input_memcpy_pass.h

@@ -0,0 +1,49 @@
+/**
+ * Copyright 2019-2020 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 GE_GRAPH_PASSES_MERGE_ADD_INPUT_MEMCPY_PASS_H_
+#define GE_GRAPH_PASSES_MERGE_ADD_INPUT_MEMCPY_PASS_H_
+
+#include "inc/graph_pass.h"
+
+namespace ge {
+class MergeInputMemcpyPass : public GraphPass {
+ public:
+  Status Run(ComputeGraphPtr graph);
+
+ private:
+  ///
+  /// @brief Add MemcpyAsync Op as Merge in_node
+  /// @param [in] graph
+  /// @param [in] node
+  /// @param [in] multi_batch_flag
+  /// @return Status
+  ///
+  Status AddMemcpyAsyncNodes(const ComputeGraphPtr &graph, const NodePtr &node, bool multi_batch_flag);
+
+  ///
+  /// @brief Add MemcpyAsync Node
+  /// @param [in] graph
+  /// @param [in] name
+  /// @param [in] out_data_anchor
+  /// @param [in] multi_batch_flag
+  /// @return ge::NodePtr
+  ///
+  NodePtr CreateMemcpyAsyncNode(const ComputeGraphPtr &graph, const std::string &name,
+                                const OutDataAnchorPtr &out_data_anchor, bool multi_batch_flag);
+};
+}  // namespace ge
+#endif  // GE_GRAPH_PASSES_MERGE_ADD_INPUT_MEMCPY_PASS_H_

ge/graph/passes/merge_to_stream_merge_pass.cc

@@ -32,7 +32,7 @@ Status MergeToStreamMergePass::Run(ComputeGraphPtr graph) {
     OpDescPtr merge_op_desc = node->GetOpDesc();
     GE_CHECK_NOTNULL(merge_op_desc);
     if (merge_op_desc->HasAttr(ATTR_INSERT_BY_MBATCH)) {
-      GE_CHK_STATUS_RET(AddMemcpyAsyncNodes(graph, node, true), "Merge add memcpy node failed.");
+      GE_CHK_STATUS_RET(AddActiveNodes(graph, node), "Merge add active node failed.");
       GE_CHK_STATUS_RET(SetStreamLabel(node, node->GetName()), "Set stream label failed");
     } else {
       GE_CHK_STATUS_RET(ReplaceMergeNode(graph, node), "Add StreamMerge node failed.");
@@ -99,38 +99,26 @@ Status MergeToStreamMergePass::ReplaceMergeNode(const ComputeGraphPtr &graph, co
     }
   }
 
-  return AddMemcpyAsyncNodes(graph, stream_merge, false);
+  return AddActiveNodes(graph, stream_merge);
 }
 
 ///
-/// @brief Add MemcpyAsync Op as StreamMerge in_node
+/// @brief Add StreamActive Op before StreamMerge/Merge
 /// @param [in] graph
 /// @param [in] node
-/// @param [in] multi_batch_flag
 /// @return Status
 ///
-Status MergeToStreamMergePass::AddMemcpyAsyncNodes(const ComputeGraphPtr &graph, const NodePtr &node,
-                                                   bool multi_batch_flag) {
+Status MergeToStreamMergePass::AddActiveNodes(const ComputeGraphPtr &graph, const NodePtr &node) {
   GE_CHK_BOOL_EXEC(node != nullptr, return FAILED, "Param of pre node is null.");
   for (const InDataAnchorPtr &in_data_anchor : node->GetAllInDataAnchors()) {
     OutDataAnchorPtr peer_out_anchor = in_data_anchor->GetPeerOutAnchor();
     GE_IF_BOOL_EXEC(peer_out_anchor == nullptr, continue);
     NodePtr in_node = peer_out_anchor->GetOwnerNode();
     const std::string &type = in_node->GetType();
-    // For WhileLoop no need memcpy & active for merge.
+    // For WhileLoop, no need to add active nodes here, since which have been added in NextIterationPass.
     GE_IF_BOOL_EXEC((type == ENTER) || (type == REFENTER) || (type == NEXTITERATION) || (type == REFNEXTITERATION),
                     continue);
-
-    const std::string &memcpy_name = node->GetName() + "_input_" + std::to_string(in_data_anchor->GetIdx());
-    NodePtr memcpy_node = CreateMemcpyAsyncNode(graph, memcpy_name, peer_out_anchor, multi_batch_flag);
-    GE_CHK_BOOL_EXEC(memcpy_node != nullptr, return FAILED, "Create MemcpyAsync node failed.");
-    GE_CHK_STATUS(GraphUtils::RemoveEdge(peer_out_anchor, in_data_anchor), "MemcpyAsync node remove edge failed.");
-    GE_CHK_STATUS(GraphUtils::AddEdge(peer_out_anchor, memcpy_node->GetInDataAnchor(0)),
-                  "MemcpyAsync node add edge failed.");
-    GE_CHK_STATUS(GraphUtils::AddEdge(memcpy_node->GetOutDataAnchor(0), in_data_anchor),
-                  "MemcpyAsync node add edge failed.");
-
-    NodePtr active_node = CreateActiveNode(graph, memcpy_node);
+    NodePtr active_node = CreateActiveNode(graph, in_node);
     GE_CHK_BOOL_EXEC(active_node != nullptr, return FAILED, "Create StreamActive node failed.");
     GE_CHK_STATUS(GraphUtils::AddEdge(active_node->GetOutControlAnchor(), node->GetInControlAnchor()),
                   "StreamActive add ctrl edge failed.");
@@ -144,37 +132,6 @@ Status MergeToStreamMergePass::AddMemcpyAsyncNodes(const ComputeGraphPtr &graph,
 }
 
 ///
-/// @brief Add MemcpyAsync Node
-/// @param [in] graph
-/// @param [in] name
-/// @param [in] out_data_anchor
-/// @param [in] multi_batch_flag
-/// @return ge::NodePtr
-///
-NodePtr MergeToStreamMergePass::CreateMemcpyAsyncNode(const ComputeGraphPtr &graph, const std::string &name,
-                                                      const OutDataAnchorPtr &out_data_anchor, bool multi_batch_flag) {
-  GE_CHK_BOOL_EXEC(out_data_anchor != nullptr, return nullptr, "Param of input node is null.");
-  OpDescPtr pre_op_desc = out_data_anchor->GetOwnerNode()->GetOpDesc();
-  GE_CHK_BOOL_EXEC(pre_op_desc != nullptr, return nullptr, "OpDesc of pre node is invalid.");
-
-  const std::string &memcpy_type = multi_batch_flag ? MEMCPYADDRASYNC : MEMCPYASYNC;
-  const std::string &node_name = name + "_" + memcpy_type;
-  GELOGI("Create MemcpyAsync op:%s.", node_name.c_str());
-  OpDescPtr op_desc = MakeShared<OpDesc>(node_name, memcpy_type);
-  if (op_desc == nullptr) {
-    GELOGE(FAILED, "Create op_desc failed, MemcpyAsync:%s.", node_name.c_str());
-    return nullptr;
-  }
-
-  GE_CHK_BOOL_EXEC(op_desc->AddInputDesc(pre_op_desc->GetOutputDesc(out_data_anchor->GetIdx())) == GRAPH_SUCCESS,
-                   return nullptr, "Create MemcpyAsync op: add input desc failed.");
-  GE_CHK_BOOL_EXEC(op_desc->AddOutputDesc(pre_op_desc->GetOutputDesc(out_data_anchor->GetIdx())) == GRAPH_SUCCESS,
-                   return nullptr, "Create MemcpyAsync op: add output desc failed.");
-
-  return graph->AddNode(op_desc);
-}
-
-///
 /// @brief Create Active Op
 /// @param [in] graph
 /// @param [in] node
@@ -193,7 +150,7 @@ NodePtr MergeToStreamMergePass::CreateActiveNode(const ComputeGraphPtr &graph, c
   GE_CHK_BOOL_EXEC(active_node != nullptr, return nullptr, "Create StreamActive node failed.");
   GE_IF_BOOL_EXEC(GraphUtils::AddEdge(node->GetOutControlAnchor(), active_node->GetInControlAnchor()) != SUCCESS,
                   GELOGE(INTERNAL_ERROR, "add edge failed");
-                   return nullptr);
+                  return nullptr);
   GE_IF_BOOL_EXEC(SetSwitchBranchNodeLabel(active_node, node_name) != SUCCESS,
                   GELOGE(INTERNAL_ERROR, "set switch branch node label failed");
                   return nullptr);

ge/graph/passes/merge_to_stream_merge_pass.h

@@ -34,24 +34,12 @@ class MergeToStreamMergePass : public GraphPass {
   Status ReplaceMergeNode(const ComputeGraphPtr &graph, const NodePtr &merge_node);
 
   ///
-  /// @brief Add MemcpyAsync Op as StreamMerge in_node
+  /// @brief Add StreamActive Op as StreamMerge in_node
   /// @param [in] graph
   /// @param [in] node
-  /// @param [in] multi_batch_flag
   /// @return Status
   ///
-  Status AddMemcpyAsyncNodes(const ComputeGraphPtr &graph, const NodePtr &node, bool multi_batch_flag);
-
-  ///
-  /// @brief Add MemcpyAsync Node
-  /// @param [in] graph
-  /// @param [in] name
-  /// @param [in] out_data_anchor
-  /// @param [in] multi_batch_flag
-  /// @return ge::NodePtr
-  ///
-  NodePtr CreateMemcpyAsyncNode(const ComputeGraphPtr &graph, const std::string &name,
-                                const OutDataAnchorPtr &out_data_anchor, bool multi_batch_flag);
+  Status AddActiveNodes(const ComputeGraphPtr &graph, const NodePtr &node);
 
   ///
   /// @brief Create Active Op

ge/offline/main.cc

@@ -67,19 +67,19 @@ static bool is_dynamic_input = false;
 const char *const kGraphMemoryManagerMallocMaxSize = "8*1024*1024*1024";
 const char *const kModeSupport = "only support 0(model to framework model), "
                                  "1(framework model to json), 3(only pre-check), 5(pbtxt to json)";
-const char *const kModelToJsonSupport = "only support 0(Caffe) 3(TensorFlow)";
+const char *const kModelToJsonSupport = "only support 0(Caffe) 3(TensorFlow) 5(Onnx)";
 
 // limit available mem size 2G
 const long kMinAvailableMem = 2 * 1024 * 1024;
 
 DEFINE_string(model, "", "The model file.");
 DEFINE_string(output, "", "The output file path&name.");
-DEFINE_int32(framework, -1, "Framework type(0:Caffe; 1:MindSpore; 3:Tensorflow).");
+DEFINE_int32(framework, -1, "Framework type(0:Caffe; 1:MindSpore; 3:Tensorflow; 5:Onnx).");
 DEFINE_string(weight, "", "Optional; weight file. Required when framework is Caffe.");
 
 DEFINE_string(input_shape, "",
               "Optional; shape of input data. Required when framework is caffe "
-              "or TensorFLow or MindSpore."
+              "or TensorFLow or MindSpore or Onnx. "
               "Format: \"input_name1:n1,c1,h1,w1;input_name2:n2,c2,h2,w2\"");
 DEFINE_bool(h, false, "show this help message");
 DEFINE_string(cal_conf, "", "Optional; the calibration config file.");
@@ -225,31 +225,29 @@ class GFlagUtils {
         "  --model             Model file\n"
         "  --weight            Weight file. Required when framework is Caffe\n"
         "  --om                The model file to be converted to json\n"
-        "  --framework         Framework type. 0:Caffe; 1:MindSpore; 3:Tensorflow\n"
+        "  --framework         Framework type. 0:Caffe; 1:MindSpore; 3:Tensorflow; 5:Onnx\n"
         "  --input_format      Format of input data. E.g.: \"NCHW\"\n"
-        "  --input_shape       Shape of input data. Separate multiple nodes with semicolons (;)."
+        "  --input_shape       Shape of input data. Separate multiple nodes with semicolons (;). "
         "Use double quotation marks (\") to enclose each argument.\n"
         "                      E.g.: \"input_name1:n1,c1,h1,w1;input_name2:n2,c2,h2,w2\"\n"
-        "  --dynamic_batch_size Set dynamic batch size. E.g: \"batchsize1,batchsize2,batchsize3\"\n"
-        "  --dynamic_image_size Set dynamic image size. Separate multiple nodes with semicolons (;)."
+        "  --dynamic_batch_size Set dynamic batch size. E.g.: \"batchsize1,batchsize2,batchsize3\"\n"
+        "  --dynamic_image_size Set dynamic image size. Separate multiple nodes with semicolons (;). "
         "Use double quotation marks (\") to enclose each argument.\n"
-        "                      E.g: \"imagesize1_height,imagesize1_width;imagesize2_height,imagesize2_width\"\n"
-        "  --dynamic_dims      Set dynamic dims. Separate multiple nodes with semicolons (;)."
-        "Use double quotation marks (\") to enclose each argument. E.g: \"dims1_n1,dims1_n2;dims2_n1,dims2_n2\"\n"
+        "                       E.g.: \"imagesize1_height,imagesize1_width;imagesize2_height,imagesize2_width\"\n"
+        "  --dynamic_dims      Set dynamic dims. Separate multiple nodes with semicolons (;). "
+        "Use double quotation marks (\") to enclose each argument.\n"
+        "                      E.g.: \"dims1_n1,dims1_n2;dims2_n1,dims2_n2\"\n"
         "  --singleop          Single op definition file. atc will generate offline "
         "model(s) for single op if --singleop is set.\n"
         "\n[Output]\n"
-        "  --output            Output file path&name(needn't suffix, will add "
-        ".om automatically). \n"
+        "  --output            Output file path&name(needn't suffix, will add .om automatically). \n"
         "                      If --singleop is set, this arg specifies the directory to "
         "which the single op offline model will be generated\n"
-        "  --output_type       Set net output type. Support FP32, FP16, UINT8."
+        "  --output_type       Set net output type. Support FP32, FP16, UINT8. "
         "E.g.: FP16, indicates that all out nodes are set to FP16.\n"
         "                      \"node1:0:FP16;node2:1:FP32\", indicates setting the datatype of multiple out nodes.\n"
-        "  --check_report      The pre-checking report file. Default value is: "
-        "\"check_result.json\"\n"
-        "  --json              The output json file path&name which is "
-        "converted from a model\n"
+        "  --check_report      The pre-checking report file. Default value is: \"check_result.json\"\n"
+        "  --json              The output json file path&name which is converted from a model\n"
         "\n[Target]\n"
         "  --soc_version       The soc version.\n"
         "  --core_type         Set core type AiCore or VectorCore. VectorCore: use vector core. "
@@ -260,23 +258,22 @@ class GFlagUtils {
         "  --out_nodes         Output nodes designated by users. Separate multiple nodes with semicolons (;)."
         "Use double quotation marks (\") to enclose each argument.\n"
         "                      E.g.: \"node_name1:0;node_name1:1;node_name2:0\"\n"
-        "  --input_fp16_nodes  Input node datatype is fp16. Separate multiple nodes with semicolons "
-        "(;)."
-        "Use double quotation marks (\") to enclose each argument."
+        "  --input_fp16_nodes  Input node datatype is fp16. Separate multiple nodes with semicolons (;)."
+        "Use double quotation marks (\") to enclose each argument. "
         "E.g.: \"node_name1;node_name2\"\n"
         "  --insert_op_conf    Config file to insert new op\n"
         "  --op_name_map       Custom op name mapping file\n"
         "                      Note: A semicolon(;) cannot be included in each "
         "path, otherwise the resolved path will not match the expected one.\n"
         "  --is_input_adjust_hw_layout    Intput node datatype is fp16 and format is "
-        "NC1HWC0, used with input_fp16_nodes  E.g.: \"true,true,false,true\"\n"
+        "NC1HWC0, used with input_fp16_nodes. E.g.: \"true,true,false,true\"\n"
         "  --is_output_adjust_hw_layout   Net output node datatype is fp16 and format is "
         "NC1HWC0, used with out_nodes. E.g.: \"true,true,false,true\"\n"
         "\n[Model Tuning]\n"
-        "  --disable_reuse_memory    The switch of reuse memory. Default value is : 0."
+        "  --disable_reuse_memory    The switch of reuse memory. Default value is : 0. "
         "0 means reuse memory, 1 means do not reuse memory.\n"
         "  --fusion_switch_file      Set fusion switch file path\n"
-        "  --enable_scope_fusion_passes    validate the non-general scope fusion passes,"
+        "  --enable_scope_fusion_passes    validate the non-general scope fusion passes, "
         "multiple names can be set and separated by ','. E.g.: ScopePass1,ScopePass2,...\n"
         "  --enable_single_stream    Enable single stream. true: enable; false(default): disable\n"
         "  --enable_small_channel    Set enable small channel. 0(default): disable; 1: enable\n"
@@ -287,20 +284,21 @@ class GFlagUtils {
         "  --precision_mode        precision mode, support force_fp16(default), allow_mix_precision, "
         "allow_fp32_to_fp16, must_keep_origin_dtype.\n"
         "  --auto_tune_mode        Set tune mode. E.g.: \"GA,RL\", support configure multiple, spit by ,\n"
-        "  --op_select_implmode    Set op select implmode. Support high_precision, high_performance."
+        "  --op_select_implmode    Set op select implmode. Support high_precision, high_performance. "
         "default: high_performance\n"
         "  --optypelist_for_implmode    Appoint which op to select implmode, cooperated with op_select_implmode.\n"
         "                               Separate multiple nodes with commas (,). Use double quotation marks (\") "
-        "                               to enclose each argument. E.g.: \"node_name1,node_name2\"\n"
+        "to enclose each argument. E.g.: \"node_name1,node_name2\"\n"
         "  --op_debug_level        Debug enable for TBE operator building.\n"
         "                          0 (default): Disable debug; 1: Enable TBE pipe_all, "
         "and generate the operator CCE file and Python-CCE mapping file (.json);\n"
         "                          2: Enable TBE pipe_all, generate the operator CCE file and Python-CCE mapping file "
         "(.json), and enable the CCE compiler -O0-g.\n"
+        "                          3: Disable debug, and keep generating kernel file (.o and .json)\n"
         "\n[Debug]\n"
         "  --save_original_model   Control whether to output original model. E.g.: true: output original model\n"
         "  --log                   Generate log with level. Support debug, info, warning, error, null\n"
-        "  --dump_mode             The switch of dump json with shape, to be used with mode 1."
+        "  --dump_mode             The switch of dump json with shape, to be used with mode 1. "
         "0(default): disable; 1: enable.");
 
     gflags::ParseCommandLineNonHelpFlags(&argc, &argv, true);
@@ -511,9 +509,9 @@ class GFlagUtils {
       // No framework information was passed in or the entered framework is illegal
       ErrorManager::GetInstance().ATCReportErrMessage(
           "E10007", {"parameter", "support"},
-          {"framework", "0(Caffe) or 1(MindSpore) or 3(TensorFlow)"});
+          {"framework", "0(Caffe) or 1(MindSpore) or 3(TensorFlow) or 5(Onnx)"});
       DOMI_LOGE("Input parameter[--framework] is mandatory and it's value must be: "
-                "0(Caffe) or 1(MindSpore) or 3(TensorFlow).");
+                "0(Caffe) or 1(MindSpore) or 3(TensorFlow) or 5(Onnx).");
       return domi::PARAM_INVALID;
     }
 

ge/session/omg.cc

@@ -994,8 +994,9 @@ FMK_FUNC_HOST_VISIBILITY Status ConvertFwkModelToJson(const domi::FrameworkType
 
   ErrorManager::GetInstance().ATCReportErrMessage(
       "E10001", {"parameter", "value", "reason"},
-      {"--framework", std::to_string(framework), "only support 0(Caffe) 3(TensorFlow)"});
-  GELOGE(PARAM_INVALID, "Input parameter[--framework] is mandatory and it's value must be: 0(Caffe) 3(TensorFlow).");
+      {"--framework", std::to_string(framework), "only support 0(Caffe) 3(TensorFlow) 5(Onnx)"});
+  GELOGE(PARAM_INVALID, "Input parameter[--framework] is mandatory and it's value must be: 0(Caffe) 3(TensorFlow) " 
+         "or 5(Onnx).");
   return PARAM_INVALID;
 }
 

ge/single_op/task/op_task.cc

@@ -263,7 +263,8 @@ Status AiCpuBaseTask::SetExtInfoAndType(const std::string &kernel_ext_info) {
 }
 
 Status AiCpuBaseTask::UpdateExtInfo(const std::vector<GeTensorDesc> &input_desc, 
-                                    std::vector<GeTensorDesc> &output_desc) {
+                                    std::vector<GeTensorDesc> &output_desc,
+                                    rtStream_t stream) {
   GELOGI("Update ext info begin, unknown_type=%d.", unknown_type_);
   if (num_inputs_ == 0 && num_outputs_ == 0) {
     GELOGI("No input and output, no need update ext info.");
@@ -283,11 +284,12 @@ Status AiCpuBaseTask::UpdateExtInfo(const std::vector<GeTensorDesc> &input_desc,
     }
   }
 
-  GE_CHK_RT_RET(rtMemcpy(ext_info_addr_dev_,
-                         aicpu_ext_handle_->GetExtInfoLen(), // check size
-                         aicpu_ext_handle_->GetExtInfo(),
-                         aicpu_ext_handle_->GetExtInfoLen(),
-                         RT_MEMCPY_HOST_TO_DEVICE));
+  GE_CHK_RT_RET(rtMemcpyAsync(ext_info_addr_dev_,
+                              aicpu_ext_handle_->GetExtInfoLen(), // check size
+                              aicpu_ext_handle_->GetExtInfo(),
+                              aicpu_ext_handle_->GetExtInfoLen(),
+                              RT_MEMCPY_HOST_TO_DEVICE_EX,
+                              stream));
 
   GELOGI("Update ext info end.");
   return SUCCESS;
@@ -618,7 +620,7 @@ Status AiCpuTask::LaunchKernel(const std::vector<GeTensorDesc> &input_desc,
                                std::vector<GeTensorDesc> &output_desc,
                                std::vector<DataBuffer> &output_buffers,
                                rtStream_t stream) {
-  GE_CHK_STATUS_RET_NOLOG(UpdateExtInfo(input_desc, output_desc));
+  GE_CHK_STATUS_RET_NOLOG(UpdateExtInfo(input_desc, output_desc, stream));
   std::vector<void *> inputs;
   std::vector<void *> outputs;
   for (auto &buffer : input_buffers) {
@@ -629,11 +631,12 @@ Status AiCpuTask::LaunchKernel(const std::vector<GeTensorDesc> &input_desc,
   }
   GE_CHK_STATUS_RET_NOLOG(SetIO(inputs, outputs));
   GE_CHK_STATUS_RET_NOLOG(LaunchKernel(stream));
-  GE_CHK_RT_RET(rtStreamSynchronize(stream));
 
   if (unknown_type_ == DEPEND_SHAPE_RANGE) {
+    GE_CHK_RT_RET(rtStreamSynchronize(stream));
     GE_CHK_STATUS_RET_NOLOG(UpdateOutputShape(output_desc));
   } else if (unknown_type_ == DEPEND_COMPUTE) {
+    GE_CHK_RT_RET(rtStreamSynchronize(stream));
     GE_CHK_STATUS_RET_NOLOG(UpdateShapeAndDataByResultSummary(output_desc, output_buffers, stream));
   }
 
@@ -689,7 +692,7 @@ Status AiCpuCCTask::LaunchKernel(const std::vector<GeTensorDesc> &input_desc,
                          "AiCpuCCTask unknown type[%d] is depend compute, it's not supported now.",
                          unknown_type_);
 
-  GE_CHK_STATUS_RET_NOLOG(UpdateExtInfo(input_desc, output_desc));
+  GE_CHK_STATUS_RET_NOLOG(UpdateExtInfo(input_desc, output_desc, stream));
 
   size_t arg_index = 0;
   auto *task_io_addr = reinterpret_cast<uintptr_t *>(io_addr_);
@@ -702,9 +705,9 @@ Status AiCpuCCTask::LaunchKernel(const std::vector<GeTensorDesc> &input_desc,
   }
 
   GE_CHK_STATUS_RET_NOLOG(LaunchKernel(stream));
-  GE_CHK_RT_RET(rtStreamSynchronize(stream));
 
   if (unknown_type_ == DEPEND_SHAPE_RANGE) {
+    GE_CHK_RT_RET(rtStreamSynchronize(stream));
     GE_CHK_STATUS_RET_NOLOG(UpdateOutputShape(output_desc));
   }
 

ge/single_op/task/op_task.h

@@ -131,7 +131,8 @@ class AiCpuBaseTask : public OpTask {
   Status SetExtInfoAndType(const std::string &kernel_ext_info);
 
   Status UpdateExtInfo(const std::vector<GeTensorDesc> &input_desc,
-                       std::vector<GeTensorDesc> &output_desc);
+                       std::vector<GeTensorDesc> &output_desc,
+                       rtStream_t stream);
   Status UpdateOutputShape(vector<GeTensorDesc> &output_desc);
   Status UpdateShapeToOutputDesc(const GeShape &shape_new, GeTensorDesc &output_desc);
 

metadef

@@ -1 +1 @@
-Subproject commit 384dab951f40b90788cc755e13754ed386ccb9c6
+Subproject commit 5ced68f6e3eff84a28a4d708e9f38a503b20f24d

parser

@@ -1 +1 @@
-Subproject commit f9e30cdc643926d91f7b650b6903a7b3bf6f597b
+Subproject commit 2e47f0ccc3981a3002ac12e3ff6a36fb689968ae