hummingbird
/
graphengine
Not watched
2
0
Fork 0
Code
Releases 13 Wiki Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain
Browse Source

add error message

tags/v1.3.0
wangxiaotian22 4 years ago
parent
96eaa5364d
commit
ecfdd3ee66
33 changed files with 787 additions and 35 deletions
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +37
    -0
      ge/graph/build/graph_builder.cc
  2. +10
    -0
      ge/graph/build/label_allocator.cc
  3. +9
    -0
      ge/graph/build/logical_stream_allocator.cc
  4. +1
    -0
      ge/graph/build/memory/graph_mem_assigner.cc
  5. +1
    -0
      ge/graph/build/memory/hybrid_mem_assigner.cc
  6. +17
    -4
      ge/graph/build/memory/var_mem_assign_util.cc
  7. +68
    -1
      ge/graph/build/model_builder.cc
  8. +22
    -0
      ge/graph/build/run_context.cc
  9. +65
    -3
      ge/graph/build/stream_allocator.cc
  10. +9
    -0
      ge/graph/build/stream_graph_optimizer.cc
  11. +77
    -4
      ge/graph/build/task_generator.cc
  12. +2
    -0
      ge/graph/common/bcast.cc
  13. +7
    -0
      ge/graph/common/bcast.h
  14. +18
    -0
      ge/graph/common/omg_util.cc
  15. +30
    -0
      ge/graph/execute/graph_execute.cc
  16. +16
    -0
      ge/graph/label/case_label_maker.cc
  17. +22
    -0
      ge/graph/label/if_label_maker.cc
  18. +30
    -0
      ge/graph/label/label_maker.cc
  19. +7
    -0
      ge/graph/label/partitioned_call_label_maker.cc
  20. +20
    -0
      ge/graph/label/while_label_maker.cc
  21. +4
    -0
      ge/graph/load/model_manager/task_info/end_graph_task_info.cc
  22. +4
    -0
      ge/graph/load/model_manager/task_info/event_record_task_info.cc
  23. +7
    -0
      ge/graph/load/model_manager/task_info/event_wait_task_info.cc
  24. +3
    -0
      ge/graph/load/model_manager/task_info/fusion_start_task_info.cc
  25. +3
    -0
      ge/graph/load/model_manager/task_info/fusion_stop_task_info.cc
  26. +20
    -0
      ge/graph/load/model_manager/task_info/hccl_task_info.cc
  27. +67
    -7
      ge/graph/load/model_manager/task_info/kernel_ex_task_info.cc
  28. +175
    -2
      ge/graph/load/model_manager/task_info/kernel_task_info.cc
  29. +12
    -3
      ge/graph/load/model_manager/task_info/super_kernel/super_kernel.cc
  30. +19
    -9
      ge/graph/load/model_manager/task_info/super_kernel/super_kernel_factory.cc
  31. +3
    -0
      inc/framework/common/debug/log.h
  32. +1
    -1
      metadef
  33. +1
    -1
      parser

+ 37
- 0
ge/graph/build/graph_builder.cc View File

@@ -77,6 +77,8 @@ Status HandleSubgraphNode(NodePtr &src_node, OutDataAnchorPtr &src_out_anchor) {
Status HandleSubgraphDataNode(NodePtr &src_node, OutDataAnchorPtr &src_out_anchor) {
uint32_t index = 0;
if (!AttrUtils::GetInt(src_node->GetOpDesc(), ATTR_NAME_PARENT_NODE_INDEX, index)) {
REPORT_INNER_ERROR("E19999", "get attr:%s failed from node:%s when HandleSubgraphDataNode",
ATTR_NAME_PARENT_NODE_INDEX.c_str(), src_node->GetName().c_str());
GELOGE(FAILED, "Get attr ATTR_NAME_PARENT_NODE_INDEX failed, node:%s.", src_node->GetName().c_str());
return FAILED;
}
@@ -109,6 +111,8 @@ Status GraphBuilder::CalcOpParam(const ge::ComputeGraphPtr &graph) {
GE_CHECK_NOTNULL(graph);
auto instance_ptr = ge::GELib::GetInstance();
if (instance_ptr == nullptr || !instance_ptr->InitFlag()) {
REPORT_INNER_ERROR("E19999", "check gelib instance null when CalcOpParam for graph:%s",
graph->GetName().c_str());
GELOGE(GE_CLI_GE_NOT_INITIALIZED, "GraphBuilder: GE is not initialized");
return GE_CLI_GE_NOT_INITIALIZED;
}
@@ -121,6 +125,8 @@ Status GraphBuilder::CalcOpParam(const ge::ComputeGraphPtr &graph) {
(void)instance_ptr->DNNEngineManagerObj().GetDNNEngineName(node_ptr);
kernel_lib_name = node_ptr->GetOpDesc()->GetOpKernelLibName();
if (kernel_lib_name.empty()) {
REPORT_INNER_ERROR("E19999", "op kernel lib is empty in node:%s(%s) when CalcOpParam",
node_ptr->GetName().c_str(), node_ptr->GetType().c_str());
GELOGE(INTERNAL_ERROR, "Get node:%s(%s) kernel lib failed.", node_ptr->GetName().c_str(),
node_ptr->GetType().c_str());
return INTERNAL_ERROR;
@@ -129,12 +135,16 @@ Status GraphBuilder::CalcOpParam(const ge::ComputeGraphPtr &graph) {

auto ret = SetInputSize(node_ptr);
if (ret != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Set node:%s(%s) inputDesc size failed when CalcOpParam",
node_ptr->GetName().c_str(), node_ptr->GetType().c_str());
GELOGE(ret, "Set node inputDesc size failed, node name is %s", node_ptr->GetName().c_str());
return ret;
}

ret = OpsKernelBuilderManager::Instance().CalcOpRunningParam(*node_ptr);
if (ret != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call Calculate op:%s(%s) running param failed",
node_ptr->GetName().c_str(), node_ptr->GetType().c_str());
GELOGE(ret, "Calculate op running param failed, node name is %s", node_ptr->GetName().c_str());
return ret;
}
@@ -191,6 +201,7 @@ Status GraphBuilder::UpdateParentNodeOutputSize(const ge::ComputeGraphPtr &graph

Status GraphBuilder::Build(ComputeGraphPtr &comp_graph, GeRootModelPtr &ge_root_model_ptr, uint64_t session_id) {
if (comp_graph == nullptr) {
REPORT_INNER_ERROR("E19999", "check compute_graph nullptr when BuildGraph, session_id:%lu", session_id);
GELOGE(GE_GRAPH_PARAM_NULLPTR, "Graph build comp_graph is null.");
return GE_GRAPH_PARAM_NULLPTR;
}
@@ -302,6 +313,8 @@ Status GraphBuilder::SetConstantInputOffset(ComputeGraphPtr &comp_graph) {

std::vector<GeTensorPtr> weights = OpDescUtils::MutableWeights(peer_node);
if (weights.empty()) {
REPORT_INNER_ERROR("E19999", "check weights size of node %s(%s) is empty when SetConstantInputOffset",
node->GetName().c_str(), node->GetType().c_str());
GELOGE(FAILED, "weights size of node %s is empty", node->GetName().c_str());
return FAILED;
}
@@ -393,6 +406,7 @@ static Status InsertMemcpyNode(const ComputeGraphPtr &graph, const OutDataAnchor
.Build();
(void)AttrUtils::SetBool(op_desc, ATTR_NO_NEED_CONSTANT_FOLDING, false);
if (GraphUtils::InsertNodeAfter(out_anchor, in_anchors, graph->AddNode(op_desc)) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Insert IDENTITY node %s after %s failed", name.c_str(), in_node->GetName().c_str());
GELOGE(FAILED, "Insert IDENTITY node %s after %s failed.", name.c_str(), in_node->GetName().c_str());
return FAILED;
}
@@ -423,6 +437,8 @@ static Status GenerateTaskForConstant(const std::shared_ptr<ComputeGraph> &graph
GELOGD("Insert MemcpyAsync node between %s and %s.", in_node->GetName().c_str(), node->GetName().c_str());
std::string name = node->GetName() + "_input_" + std::to_string(in_data_anchor->GetIdx()) + "_Memcpy";
if (InsertMemcpyNode(graph, peer_out_anchor, {in_data_anchor}, name) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Insert memcpy between %s and %s failed when GenerateTaskForConstant",
in_node->GetName().c_str(), node->GetName().c_str());
GELOGE(FAILED, "Insert memcpy between %s and %s failed.",
in_node->GetName().c_str(), node->GetName().c_str());
return FAILED;
@@ -470,6 +486,8 @@ Status GraphBuilder::MarkFpBpProfilingTaskAttr(ComputeGraphPtr &com_graph) {
GELOGI("The all reduce node of dynamic graph is %s, idx %u", op_desc->GetName().c_str(), node_index);
(void)ge::AttrUtils::SetBool(op_desc, ATTR_NAME_INSERT_BP_PROFILILNG_TASK, true);
GE_IF_BOOL_EXEC(TypeUtils::CheckUint64MulOverflow(i, kProfilingArStep),
REPORT_INNER_ERROR("E19999", "Multiply result is out of range when calc profiling ar log id "
"for node:%s(%s)", op_desc->GetName().c_str(), op_desc->GetType().c_str());
GELOGE(FAILED, "Multiply result is out of range.");
return FAILED);
int64_t log_id = i * kProfilingArStep + kProfilingArStartLogid;
@@ -549,16 +567,19 @@ Status GraphBuilder::GetTaskInfo(const ge::ModelBuilder &builder, const ModelPtr

int64_t memory_size = 0;
if (!AttrUtils::GetInt(model_ptr, ATTR_MODEL_MEMORY_SIZE, memory_size)) {
REPORT_INNER_ERROR("E19999", "Get Attr:%s fail in model", ATTR_MODEL_MEMORY_SIZE.c_str());
GELOGE(INTERNAL_ERROR, "Get memory size fail.");
return INTERNAL_ERROR;
}
int64_t p2p_memory_size = 0;
if (!AttrUtils::GetInt(model_ptr, ATTR_MODEL_P2P_MEMORY_SIZE, p2p_memory_size)) {
REPORT_INNER_ERROR("E19999", "Get Attr:%s fail in model", ATTR_MODEL_P2P_MEMORY_SIZE.c_str());
GELOGE(INTERNAL_ERROR, "Get p2p memory size fail.");
return INTERNAL_ERROR;
}
int64_t weight_size = 0;
if (!AttrUtils::GetInt(model_ptr, ATTR_MODEL_WEIGHT_SIZE, weight_size)) {
REPORT_INNER_ERROR("E19999", "Get Attr:%s fail in model", ATTR_MODEL_WEIGHT_SIZE.c_str());
GELOGE(INTERNAL_ERROR, "Get weight memory size fail.");
return INTERNAL_ERROR;
}
@@ -668,6 +689,7 @@ Status GraphBuilder::SetInputSize(const ge::NodePtr &node_ptr) {
Status GraphBuilder::UpdateDataInputSize(const ge::NodePtr &node_ptr) {
const auto &op_desc = node_ptr->GetOpDesc();
if (op_desc == nullptr) {
REPORT_INNER_ERROR("E19999", "check op_desc is nullptr when UpdateDataInputSize");
GELOGE(FAILED, "Op desc is nullptr.");
return FAILED;
}
@@ -685,6 +707,8 @@ Status GraphBuilder::UpdateDataInputSize(const ge::NodePtr &node_ptr) {
int64_t real_dim_size = 0;
ge::graphStatus graph_status = TensorUtils::GetTensorSizeInBytes(output_desc, real_dim_size);
if (graph_status != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Get tensor size in bytes failed for op:%s(%s) index:0 when UpdateDataInputSize",
op_desc->GetName().c_str(), op_desc->GetType().c_str());
GELOGE(FAILED, "Get tensor size in bytes failed.");
return FAILED;
}
@@ -692,6 +716,8 @@ Status GraphBuilder::UpdateDataInputSize(const ge::NodePtr &node_ptr) {
ge::GeTensorDesc input_desc = op_desc->GetInputDesc(0);
ge::TensorUtils::SetSize(input_desc, real_dim_size);
if (op_desc->UpdateInputDesc(0, input_desc) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Update input desc size failed for op:%s(%s) index:0 when UpdateDataInputSize",
op_desc->GetName().c_str(), op_desc->GetType().c_str());
GELOGE(FAILED, "Update input desc size failed.");
return FAILED;
}
@@ -720,6 +746,9 @@ Status GraphBuilder::CalcDynShapeRootGraphDataSize(const ge::OpDescPtr &op_desc)
int64_t real_dim_size = 0;
ge::graphStatus graph_status = TensorUtils::GetTensorSizeInBytes(output_desc, real_dim_size);
if (graph_status != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Get tensor size in bytes failed for op:%s(%s) index:0 "
"when CalcDynShapeRootGraphDataSize",
op_desc->GetName().c_str(), op_desc->GetType().c_str());
GELOGE(FAILED, "Get tensor size in bytes failed.");
return FAILED;
}
@@ -727,6 +756,9 @@ Status GraphBuilder::CalcDynShapeRootGraphDataSize(const ge::OpDescPtr &op_desc)
ge::TensorUtils::SetSize(output_desc, real_dim_size);
GELOGI("Update dynamic shape graph data output size to [%ld].", real_dim_size);
if (op_desc->UpdateOutputDesc(0, output_desc) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Update output desc size failed for op:%s(%s) index:0 "
"when CalcDynShapeRootGraphDataSize",
op_desc->GetName().c_str(), op_desc->GetType().c_str());
GELOGE(FAILED, "Update dynamic shape graph data output desc size failed.");
return FAILED;
}
@@ -744,6 +776,8 @@ Status GraphBuilder::SecondPartition(ge::ComputeGraphPtr &comp_graph) {
GE_CHK_STATUS_RET(ret, "Graph partition Failed.");
const auto &graph_2_subgraphlist = graph_partitioner_.GetSubGraphMap();
if (graph_2_subgraphlist.find(comp_graph) == graph_2_subgraphlist.end()) {
REPORT_INNER_ERROR("E19999", "find subgraphlis in graph:%s failed when SecondPartition",
comp_graph->GetName().c_str());
GELOGE(FAILED, "Find subgraph failed.");
return FAILED;
}
@@ -772,6 +806,9 @@ Status GraphBuilder::AddOutputMemTypeForNode(const NodePtr &node) {
mem_type);
if (!AttrUtils::SetInt(src_desc->MutableOutputDesc(src_out_anchor->GetIdx()), ATTR_OUTPUT_MEMORY_TYPE,
mem_type)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s for node:%s(%s) out_index:%u failed when AddOutputMemTypeForNode",
ATTR_OUTPUT_MEMORY_TYPE.c_str(), src_desc->GetName().c_str(), src_desc->GetType().c_str(),
src_out_anchor->GetIdx());
GELOGE(INTERNAL_ERROR, "Set out_memory_type attr for [%s:%d] failed.", src_desc->GetName().c_str(),
src_out_anchor->GetIdx());
return INTERNAL_ERROR;


+ 10
- 0
ge/graph/build/label_allocator.cc View File

@@ -28,6 +28,7 @@ LabelAllocator::LabelAllocator(const ComputeGraphPtr &graph) : compute_graph_(gr

Status LabelAllocator::AssignFunctionalLabels() {
if (compute_graph_ == nullptr) {
REPORT_INNER_ERROR("E19999", "check param compute_graph nullptr when AssignFunctionalLabels");
GELOGE(INTERNAL_ERROR, "ComputeGraph not set, Assign labels failed.");
return INTERNAL_ERROR;
}
@@ -46,11 +47,15 @@ Status LabelAllocator::AssignFunctionalLabels() {
for (auto node : functional_nodes) {
LabelMakerPtr maker = LabelMakerFactory::Instance().Create(node->GetType(), compute_graph_, node);
if (maker == nullptr) {
REPORT_CALL_ERROR("E19999", "Check Node:%s(%s) label maker not registed when AssignFunctionalLabels",
node->GetName().c_str(), node->GetType().c_str());
GELOGE(INTERNAL_ERROR, "Node: %s label maker not registed.", node->GetType().c_str());
return INTERNAL_ERROR;
}

if (maker->Run(label_index) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Node:%s(%s) run label maker failed when AssignFunctionalLabels",
node->GetName().c_str(), node->GetType().c_str());
GELOGE(INTERNAL_ERROR, "Node: %s run label maker failed.", node->GetType().c_str());
return INTERNAL_ERROR;
}
@@ -63,6 +68,7 @@ Status LabelAllocator::AssignFunctionalLabels() {

bool LabelAllocator::CollectFunctionalNode(ComputeGraphPtr &graph, std::set<NodePtr> &functional_nodes) {
if (graph == nullptr) {
REPORT_INNER_ERROR("E19999", "check param compute_graph nullptr when CollectFunctionalNode");
GELOGE(INTERNAL_ERROR, "Sub ComputeGraph is null.");
return false;
}
@@ -74,12 +80,16 @@ bool LabelAllocator::CollectFunctionalNode(ComputeGraphPtr &graph, std::set<Node

NodePtr func_node = graph->GetParentNode();
if (func_node == nullptr) {
REPORT_INNER_ERROR("E19999", "Parent node not set in node:%s(%s), graph:%s",
func_node->GetName().c_str(), func_node->GetType().c_str(), graph->GetName().c_str());
GELOGE(INTERNAL_ERROR, "Parent functional node not set: %s.", graph->GetName().c_str());
return false;
}

ComputeGraphPtr owner_graph = func_node->GetOwnerComputeGraph();
if (owner_graph == nullptr) {
REPORT_INNER_ERROR("E19999", "ComputeGraph owner not set in node:%s(%s), graph:%s",
func_node->GetName().c_str(), func_node->GetType().c_str(), graph->GetName().c_str());
GELOGE(INTERNAL_ERROR, "ComputeGraph owner not set: %s.", func_node->GetName().c_str());
return false;
}


+ 9
- 0
ge/graph/build/logical_stream_allocator.cc View File

@@ -320,6 +320,8 @@ Status SingleStreamPass::Run(ComputeGraphPtr graph, const vector<SubgraphPtr> &s
if (!HasAssignedStream(*subgraph)) {
const string &stream_label = subgraph->subgraph_info.GetStreamLabel();
if (!stream_label.empty()) {
REPORT_INNER_ERROR("E19999", "Stream labels are not supported in SingleStream mode "
"(subgraph: %s, stream label: %s)", subgraph->name.c_str(), stream_label.c_str());
GELOGE(INTERNAL_ERROR, "Stream labels are not supported (subgraph: %s, stream label: %s).",
subgraph->name.c_str(), stream_label.c_str());
return INTERNAL_ERROR;
@@ -337,6 +339,8 @@ Status NodeStreamUpdatePass::Run(ComputeGraphPtr graph, const vector<SubgraphPtr
const string &engine_name = subgraph->engine_conf.id;

if (!IsEngineSkip(*subgraph) && !HasAssignedStream(*subgraph)) {
REPORT_INNER_ERROR("E19999", "Subgraph %s has not yet been assigned a stream (engine: %s) "
" when run NodeStreamUpdatePass", subgraph->name.c_str(), engine_name.c_str());
GELOGE(INTERNAL_ERROR, "Subgraph %s has not yet been assigned a stream (engine: %s).", subgraph->name.c_str(),
engine_name.c_str());
return INTERNAL_ERROR;
@@ -636,6 +640,8 @@ Status LogicalStreamAllocator::DoAssign(const ComputeGraphPtr &graph, const Grap

auto iter = subgraph_map.find(graph);
if (iter == subgraph_map.end()) {
REPORT_INNER_ERROR("E19999", "Graph %s not found in subgraph_map when do logical stream assign ",
graph->GetName().c_str());
GELOGE(FAILED, "Graph %s not found.", graph->GetName().c_str());
return FAILED;
}
@@ -675,6 +681,8 @@ Status LogicalStreamAllocator::ConvertSubgraphs(const vector<SubGraphInfoPtr> &s
const string &engine_name = subgraph_info->GetEngineName();
auto engine_conf_iter = engine_confs.find(engine_name);
if ((engine_conf_iter == engine_confs.end()) || (engine_conf_iter->second == nullptr)) {
REPORT_INNER_ERROR("E19999", "Engine conf of subgraph %s not found (engine name: %s) when ConvertSubgraphs",
subgraph_name.c_str(), engine_name.c_str());
GELOGE(INTERNAL_ERROR, "Engine conf of subgraph %s not found (engine name: %s).", subgraph_name.c_str(),
engine_name.c_str());

@@ -722,6 +730,7 @@ Status LogicalStreamAllocator::RunPasses(const ComputeGraphPtr &graph, const vec
} else if (status == NOT_CHANGED) {
GELOGD("[Show][Status]Stream pass %s return NOT_CHANGED.", pass->GetName().c_str());
} else {
REPORT_CALL_ERROR("E19999", "Stream pass %s run failed.", pass->GetName().c_str());
GELOGE(status, "Stream pass %s failed.", pass->GetName().c_str());
return status;
}


+ 1
- 0
ge/graph/build/memory/graph_mem_assigner.cc View File

@@ -1215,6 +1215,7 @@ Status GraphMemoryAssigner::CheckOffset() {
std::map<std::string, std::string> anchor_to_symbol;
std::map<std::string, std::list<NodeIndexIO>> symbol_to_anchors;
if (GraphUtils::GetRefMapping(compute_graph_, symbol_to_anchors, anchor_to_symbol) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Get ref-mapping for graph %s failed", compute_graph_->GetName().c_str());
GELOGE(FAILED, "[Get][RefMapping]fail for graph %s", compute_graph_->GetName().c_str());
return FAILED;
}


+ 1
- 0
ge/graph/build/memory/hybrid_mem_assigner.cc View File

@@ -42,6 +42,7 @@ Status HybridMemAssigner::AssignMemory(std::unique_ptr<BlockMemAssigner> &block_

Status HybridMemAssigner::Assign() {
if (GraphUtils::GetRefMapping(compute_graph_, symbol_to_anchors_, anchor_to_symbol_) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Get ref-mapping for graph %s failed", compute_graph_->GetName().c_str());
GELOGE(FAILED, "Get ref-mapping for graph %s failed.", compute_graph_->GetName().c_str());
return FAILED;
}


+ 17
- 4
ge/graph/build/memory/var_mem_assign_util.cc View File

@@ -53,6 +53,8 @@ Status VarMemAssignUtil::AssignStaticMemory2Node(ge::ComputeGraphPtr &compute_gr
GE_IF_BOOL_EXEC(ge::AttrUtils::GetStr(n->GetOpDesc(), REF_VAR_SRC_VAR_NAME, ref_var_src_var_name), continue);
string node_name = n->GetName();
GE_IF_BOOL_EXEC(n->GetOpDesc()->GetAllOutputsDesc().empty(),
REPORT_INNER_ERROR("E19999", "check node:%s has no OutputDesc when AssignStaticMemory2Node",
n->GetName().c_str());
GELOGE(FAILED, "node:%s has no OutputDesc.", n->GetName().c_str());
return FAILED);
ge::ConstGeTensorDescPtr tensor_desc = n->GetOpDesc()->GetOutputDescPtr(0);
@@ -116,6 +118,8 @@ Status VarMemAssignUtil::SetOutVariableAttr(const ge::NodePtr &node, const ge::N
GE_CHECK_NOTNULL(node->GetOpDesc());
output_list = node->GetOpDesc()->GetOutputOffset();
if (output_list.empty()) {
REPORT_INNER_ERROR("E19999", "check node:%s output_offset_list is empty when SetOutVariableAttr",
node->GetName().c_str());
GELOGE(PARAM_INVALID, "Output_list is empty");
return PARAM_INVALID;
}
@@ -126,7 +130,12 @@ Status VarMemAssignUtil::SetOutVariableAttr(const ge::NodePtr &node, const ge::N
VarManager::Instance(session_id)->GetVarAddr(var_node->GetName(), var_tensor_desc, &dev_ptr, memory_type));

int out_list_size = static_cast<int>(output_list.size());
GE_CHK_BOOL_RET_STATUS(index < out_list_size, FAILED, "index %d >= output_list.size() %d", index, out_list_size);
if (index < out_list_size) {
REPORT_INNER_ERROR("E19999", "param index:%d >= output_list.size() %d in node %s, "
"check invalid when SetOutVariableAttr", index, out_list_size, node->GetName().c_str());
GELOGE(FAILED, "index %d >= output_list.size() %d", index, out_list_size);
return FAILED;
}

output_list[index] = static_cast<int64_t>(reinterpret_cast<intptr_t>(dev_ptr));
GELOGI("Assign node outputOffset[index] is: %ld", output_list[index]);
@@ -168,9 +177,13 @@ Status VarMemAssignUtil::DealBroadCastNode(uint32_t graph_id, const ge::NodePtr

auto broad_cast_index = static_cast<size_t>(broad_cast_info.idx);
auto input_tensor_desc_ptr_vistor = op_desc->GetAllInputsDescPtr();
GE_CHK_BOOL_RET_STATUS(input_tensor_desc_ptr_vistor.size() > broad_cast_index, FAILED,
"Get broadcast op %s input tensor desc size [%zu] < idx [%d]", node->GetName().c_str(),
input_tensor_desc_ptr_vistor.size(), broad_cast_info.idx);
if (input_tensor_desc_ptr_vistor.size() > broad_cast_index) {
REPORT_INNER_ERROR("E19999", "Get broadcast op %s input tensor desc size [%zu] < idx [%d]",
node->GetName().c_str(), input_tensor_desc_ptr_vistor.size(), broad_cast_info.idx);
GELOGE(FAILED, "Get broadcast op %s input tensor desc size [%zu] < idx [%d]", node->GetName().c_str(),
input_tensor_desc_ptr_vistor.size(), broad_cast_info.idx);
return FAILED;
}
const ge::GeTensorDescPtr input_tensor_desc =
input_tensor_desc_ptr_vistor.at(static_cast<size_t>(broad_cast_info.idx));
int64_t input_size = 0;


+ 68
- 1
ge/graph/build/model_builder.cc View File

@@ -116,11 +116,15 @@ Status ModelBuilder::CalcOutputSize(const ge::NodePtr &n) {
int64_t size_temp = 0;
graphStatus graph_status = TensorUtils::GetTensorMemorySizeInBytes(desc_temp, size_temp);
if (graph_status != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Get tensor size in bytes failed for op:%s(%s) index:%u when CalcOutputSize",
node_op_desc->GetName().c_str(), node_op_desc->GetType().c_str(), index);
GELOGE(graph_status, "GetTensorMemorySizeInBytes failed!");
return FAILED;
}
TensorUtils::SetSize(desc_temp, size_temp);
if (node_op_desc->UpdateOutputDesc(index, desc_temp) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Update Output desc size failed for op:%s(%s) index:%u when CalcOutputSize",
node_op_desc->GetName().c_str(), node_op_desc->GetType().c_str(), index);
GELOGE(FAILED, "UpdateOutputDesc failed.");
return FAILED;
}
@@ -207,11 +211,15 @@ Status ModelBuilder::AdjustConstWeightSize(const ge::NodePtr &node, size_t &mem_
if (node->GetType() == CONSTANT) {
vector<GeTensorPtr> weights = OpDescUtils::MutableWeights(node);
if (weights.empty()) {
REPORT_INNER_ERROR("E19999", "Check weights size of node %s(%s) is empty when AdjustConstWeightSize",
node->GetName().c_str(), node->GetType().c_str());
GELOGE(FAILED, "weights size of node %s is empty", node->GetName().c_str());
return FAILED;
}
GeTensorPtr weight = weights[0];
if (weight == nullptr) {
REPORT_INNER_ERROR("E19999", "Check weight of node %s(%s) is nullptr when AdjustConstWeightSize",
node->GetName().c_str(), node->GetType().c_str());
GELOGE(FAILED, "weights[0] is null.");
return FAILED;
}
@@ -353,6 +361,9 @@ Status ModelBuilder::AdjustInputTensorFlag() {
auto input_desc = owner_node_op_desc->GetInputDesc(in_anchors->GetIdx());
ge::TensorUtils::SetInputTensor(input_desc, true);
if (owner_node_op_desc->UpdateInputDesc(in_anchors->GetIdx(), input_desc) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Update Input desc size failed for op:%s(%s) index:%u when %s",
owner_node_op_desc->GetName().c_str(), owner_node_op_desc->GetType().c_str(),
in_anchors->GetIdx(), __FUNCTION__);
GELOGE(FAILED, "UpdateOutputDesc failed.");
return FAILED;
}
@@ -381,33 +392,51 @@ Status ModelBuilder::BuildModelDef(ge::Model &model) {

max_mem_offset_ = mem_type_to_mem_offset_[RT_MEMORY_HBM];
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetInt(&model, ATTR_MODEL_MEMORY_SIZE, max_mem_offset_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_MEMORY_SIZE.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_MODEL_MEMORY_SIZE failed.");
return FAILED);
if (mem_type_to_mem_offset_.find(RT_MEMORY_P2P_DDR) != mem_type_to_mem_offset_.end()) {
p2p_mem_offset_ = mem_type_to_mem_offset_[RT_MEMORY_P2P_DDR];
}
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetInt(&model, ATTR_MODEL_P2P_MEMORY_SIZE, p2p_mem_offset_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_P2P_MEMORY_SIZE.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_MODEL_P2P_MEMORY_SIZE failed.");
return FAILED);
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetInt(&model, ATTR_MODEL_WEIGHT_SIZE, weight_offset_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_WEIGHT_SIZE.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_MODEL_WEIGHT_SIZE failed.");
return FAILED);
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetInt(&model, ATTR_MODEL_STREAM_NUM, stream_num_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_STREAM_NUM.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_MODEL_STREAM_NUM failed.");
return FAILED);
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetInt(&model, ATTR_MODEL_EVENT_NUM, event_num_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_EVENT_NUM.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_MODEL_EVENT_NUM failed.");
return FAILED);
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetListInt(&model, ATTR_MODEL_HUGE_STREAM_LIST, huge_streams_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_HUGE_STREAM_LIST.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_MODEL_HUGE_STREAM_LIST failed.");
return FAILED);
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetInt(&model, ATTR_MODEL_LABEL_NUM, label_num_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_LABEL_NUM.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_MODEL_LABEL_NUM failed.");
return FAILED);
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetInt(&model, ATTR_MODEL_ZERO_COPY_MEMORY_SIZE, zero_copy_mem_size_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_ZERO_COPY_MEMORY_SIZE.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_MODEL_ZERO_COPY_MEMORY_SIZE failed.");
return FAILED);
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetListStr(&model, ATTR_MODEL_OUT_NODES_NAME, GetLocalOmgContext().net_out_nodes),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_OUT_NODES_NAME.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetListStr of ATTR_MODEL_OUT_NODES_NAME failed.");
return FAILED);
GELOGI("For model, max_mem_offset_: %zu, p2p_mem_size: %zu, zero_copy_mem_size_: %zu", max_mem_offset_,
@@ -415,6 +444,8 @@ Status ModelBuilder::BuildModelDef(ge::Model &model) {
string fp_ceiling_mode;
if (ge::GetContext().GetOption("ge.fpCeilingMode", fp_ceiling_mode) == SUCCESS) {
if (!ge::AttrUtils::SetStr(&model, ATTR_FP_CEILING_MODE, fp_ceiling_mode)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_FP_CEILING_MODE.c_str(), __FUNCTION__);
GELOGE(FAILED, "Failed to set attr ATTR_FP_CEILING_MODE");
return FAILED;
}
@@ -429,22 +460,30 @@ Status ModelBuilder::BuildModelDef(ge::Model &model) {
int64_t core_type = (ge_core_type == kVectorCore) ? 1 : 0;
GELOGI("core_type: %ld", core_type);
if (!ge::AttrUtils::SetInt(&model, ATTR_MODEL_CORE_TYPE, core_type)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_MODEL_CORE_TYPE.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt of ATTR_CORE_TYPE failed.");
}
InitL1FusionOption();
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetBool(&model, ATTR_NAME_SWITCH_FOR_L1_FUSION, is_l1_fusion_enable_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_NAME_SWITCH_FOR_L1_FUSION.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetBool of ATTR_NAME_SWITCH_FOR_L1_FUSION failed.");
return FAILED);
const DumpProperties &dump_properties = DumpManager::GetInstance().GetDumpProperties(session_id_);
bool is_op_debug = dump_properties.IsOpDebugOpen();
if (is_op_debug) {
if (!ge::AttrUtils::SetBool(&model, ATTR_OP_DEBUG_FLAG, is_op_debug)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_OP_DEBUG_FLAG.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetBool of ATTR_OP_DEBUG_FLAG failed.");
return FAILED;
}
uint32_t op_debug_mode = dump_properties.GetOpDebugMode();
GELOGI("Get op debug mode:%d", op_debug_mode);
if (!ge::AttrUtils::SetInt(&model, ATTR_OP_DEBUG_MODE, op_debug_mode)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s in model failed when %s",
ATTR_OP_DEBUG_MODE.c_str(), __FUNCTION__);
GELOGE(FAILED, "SetBool of ATTR_OP_DEBUG_MODE failed.");
return FAILED;
}
@@ -516,6 +555,8 @@ Status ModelBuilder::MergeWeights() {
// If MutableTensor failed, weight is nullptr.
(void)ge::AttrUtils::MutableTensor(op_desc, ATTR_NAME_WEIGHTS, weight);
if (weight == nullptr) {
REPORT_INNER_ERROR("E19999", "Can't get const weight in op:%s(%s) when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Can't get const op weight, name: %s", node->GetName().c_str());
return FAILED;
}
@@ -538,8 +579,15 @@ Status ModelBuilder::MergeWeights() {
continue;
}
if (weight_data.data() != nullptr) {
GE_IF_BOOL_EXEC(base_addr == nullptr, GELOGE(FAILED, "Base addr is nullptr."); return FAILED);
GE_IF_BOOL_EXEC(base_addr == nullptr,
REPORT_INNER_ERROR("E19999", "Check weight in op:%s(%s) is nullptr when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Base addr is nullptr.");
return FAILED);
if (weight_offset_ - offset < weight_data.size()) {
REPORT_INNER_ERROR("E19999", "left weight size not enough for op:%s(%s) left_size:%zu, weight_size:%zu when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(),
weight_offset_ - offset, weight_data.size(), __FUNCTION__);
GELOGE(FAILED, "left weight size not enough. left_size:%lu, weight_size:%lu",
weight_offset_ - offset, weight_data.size());
return FAILED;
@@ -551,6 +599,9 @@ Status ModelBuilder::MergeWeights() {
auto err = memcpy_s(reinterpret_cast<void *>(dst_ptr), SECUREC_MEM_MAX_LEN, reinterpret_cast<void *>(src_ptr),
SECUREC_MEM_MAX_LEN);
if (err != EOK) {
REPORT_CALL_ERROR("E19999", "mem copy failed. errret:%u, "
"dst_ptr:%lx, dst_size:%lu, src_ptr:%lx, src_size:%lu, when %s",
err, dst_ptr, SECUREC_MEM_MAX_LEN, src_ptr, SECUREC_MEM_MAX_LEN, __FUNCTION__);
GELOGE(FAILED, "mem copy failed. errret:%u, "
"dst_ptr:%lx, dst_size:%lu, src_ptr:%lx, src_size:%lu",
err, dst_ptr, SECUREC_MEM_MAX_LEN, src_ptr, SECUREC_MEM_MAX_LEN);
@@ -562,6 +613,9 @@ Status ModelBuilder::MergeWeights() {
}
auto err = memcpy_s(reinterpret_cast<void *>(dst_ptr), left_size, reinterpret_cast<void *>(src_ptr), left_size);
if (err != EOK) {
REPORT_CALL_ERROR("E19999", "mem copy failed. errret:%u, "
"dst_ptr:%lx, dst_size:%lu, src_ptr:%lx, src_size:%lu, when %s",
err, dst_ptr, SECUREC_MEM_MAX_LEN, src_ptr, SECUREC_MEM_MAX_LEN, __FUNCTION__);
GELOGE(FAILED, "mem copy failed. errret:%u, "
"dst_ptr:%lx, dst_size:%lu, src_ptr:%lx, src_size:%lu",
err, dst_ptr, SECUREC_MEM_MAX_LEN, src_ptr, SECUREC_MEM_MAX_LEN);
@@ -602,6 +656,8 @@ Status ModelBuilder::SaveDataToModel(ge::Model &model, ge::GeModel &ge_model) {
}
GE_IF_BOOL_EXEC(tbe_kernel == nullptr, continue);
if (tbe_name_set.count(tbe_kernel->GetName()) > 0) {
REPORT_INNER_ERROR("E19999", "tbe_kernel name %s can't be the same, judge for op:%s(%s), when %s",
tbe_kernel->GetName().c_str(), n->GetName().c_str(), n->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "tbe_kernel name %s can't be the same", tbe_kernel->GetName().c_str());
return FAILED;
}
@@ -618,6 +674,8 @@ Status ModelBuilder::SaveDataToModel(ge::Model &model, ge::GeModel &ge_model) {
node_op_desc->TryGetExtAttr(ge::OP_EXTATTR_CUSTAICPU_KERNEL, CustAICPUKernelPtr());
GE_IF_BOOL_EXEC(cust_aicpu_kernel == nullptr, continue);
if (aicpu_name_set.count(cust_aicpu_kernel->GetName()) > 0) {
REPORT_INNER_ERROR("E19999", "aicpu_kernel name %s can't be the same, judge for op:%s(%s), when %s",
cust_aicpu_kernel->GetName().c_str(), n->GetName().c_str(), n->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "aicpu_kernel name %s can't be the same", cust_aicpu_kernel->GetName().c_str());
return FAILED;
}
@@ -640,6 +698,7 @@ Status ModelBuilder::SaveDataToModel(ge::Model &model, ge::GeModel &ge_model) {
// Add task
GeAttrValue::BYTES task_def_bytes;
if (!AttrUtils::GetZeroCopyBytes(model, MODEL_ATTR_TASKS, task_def_bytes)) {
REPORT_CALL_ERROR("E19999", "Get attr:%s in model fail when %s", MODEL_ATTR_TASKS.c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Get zero copy bytes fail.");
return INTERNAL_ERROR;
}
@@ -675,6 +734,7 @@ void ModelBuilder::SetModelVersion(ge::Model &model) {

Status ModelBuilder::PreBuildModel() {
if ((compute_graph_ == nullptr) || !(compute_graph_->IsValid())) {
REPORT_INNER_ERROR("E19999", "Check compute_graph no valid when %s", __FUNCTION__);
GELOGE(FAILED, "Graph_ is not valid.");
return FAILED;
}
@@ -754,6 +814,7 @@ Status ModelBuilder::CompileSingleOp() {
// Create ge instance
std::shared_ptr<GELib> instance = ge::GELib::GetInstance();
if ((instance == nullptr) || !instance->InitFlag()) {
REPORT_INNER_ERROR("E19999", "Check GELib instance not init before when %s", __FUNCTION__);
GELOGE(ge::GE_CLI_GE_NOT_INITIALIZED, "CompileSingleOp failed.");
return ge::GE_CLI_GE_NOT_INITIALIZED;
}
@@ -775,6 +836,8 @@ Status ModelBuilder::CompileSingleOp() {
(void)instance->DNNEngineManagerObj().GetDNNEngineName(node);
kernel_lib_name = op_desc->GetOpKernelLibName();
if (kernel_lib_name.empty()) {
REPORT_INNER_ERROR("E19999", "Check kernel lib name empty of op:%s(%s) when %s",
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(ge::INTERNAL_ERROR, "Get node:%s(%s) kernel lib failed.", node->GetName().c_str(),
node->GetType().c_str());
return ge::INTERNAL_ERROR;
@@ -785,6 +848,8 @@ Status ModelBuilder::CompileSingleOp() {
if (kernel_info != nullptr) {
node_vector_map[kernel_lib_name].emplace_back(node);
} else {
REPORT_INNER_ERROR("E19999", "Get ops kernel info store failed for op:%s(%s), op_kernel_name:%s, when %s",
node->GetName().c_str(), node->GetType().c_str(), kernel_lib_name.c_str(), __FUNCTION__);
GELOGE(ge::GE_GRAPH_PARAM_NULLPTR, "Get op %s ops kernel info store failed", node->GetName().c_str());
return ge::GE_GRAPH_PARAM_NULLPTR;
}
@@ -800,6 +865,8 @@ Status ModelBuilder::CompileSingleOp() {
GELOGI("[GEPERFTRACE] The node size of compile op of %s is %zu", kernel_lib_name.c_str(), node_vector.size());
GE_TIMESTAMP_ADD(BatchCompileOp);
if (ret != ge::SUCCESS) {
REPORT_CALL_ERROR("E19999", "Batch compile op failed, kernel lib name, node size:%u, when %s",
node_vector.size(), __FUNCTION__);
GELOGE(ret, "Compile op failed, kernel lib name is %s", kernel_lib_name.c_str());
return ret;
}


+ 22
- 0
ge/graph/build/run_context.cc View File

@@ -27,15 +27,21 @@ Status RunContextUtil::InitMemInfo(uint8_t *data_mem_base, uint64_t data_mem_siz
std::map<int64_t, uint64_t> mem_type_to_data_mem_size, uint8_t *weight_mem_base,
uint64_t weight_mem_size) {
if ((data_mem_size > 0) && (data_mem_base == nullptr)) {
REPORT_INNER_ERROR("E19999", "InitMemInfo param data_mem_base is null but data_mem_size = %lu", data_mem_size);
GELOGE(PARAM_INVALID, "InitMemInfo param data_mem_base is null but data_mem_size = %lu.", data_mem_size);
return PARAM_INVALID;
}
if ((weight_mem_size > 0) && (weight_mem_base == nullptr)) {
REPORT_INNER_ERROR("E19999", "InitMemInfo param weight_mem_base is null but weight_mem_size = %lu",
weight_mem_size);
GELOGE(PARAM_INVALID, "InitMemInfo param weight_mem_base is null but weight_mem_size = %lu.", weight_mem_size);
return PARAM_INVALID;
}
if (mem_type_to_data_mem_base.empty() || mem_type_to_data_mem_size.empty() ||
mem_type_to_data_mem_base.size() != mem_type_to_data_mem_size.size()) {
REPORT_INNER_ERROR("E19999", "InitMemInfo param mem_type_to_data_mem_base size[%zu] "
"is not equal to the size of mem_type_to_data_mem_size[%zu].",
mem_type_to_data_mem_base.size(), mem_type_to_data_mem_size.size());
GELOGE(PARAM_INVALID,
"InitMemInfo param mem_type_to_data_mem_base size[%zu] is not equal to the size of "
"mem_type_to_data_mem_size[%zu].",
@@ -55,6 +61,7 @@ Status RunContextUtil::CreateRtModelResources(uint32_t stream_num, uint32_t even
// Create rt model
rtError_t rt_ret = rtModelCreate(&rt_model_, 0);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "call rtModelCreate fail, ret:%d, when %s", static_cast<int>(rt_ret), __FUNCTION__);
GELOGE(RT_FAILED, "rtModelCreate failed. rt_ret = %d", static_cast<int>(rt_ret));
return RT_FAILED;
}
@@ -64,6 +71,8 @@ Status RunContextUtil::CreateRtModelResources(uint32_t stream_num, uint32_t even
rtStream_t stream = nullptr;
rt_ret = rtStreamCreate(&stream, 0);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "call rtStreamCreate fail, ret:%d, index:%u, when %s",
static_cast<int>(rt_ret), i, __FUNCTION__);
GELOGE(RT_FAILED, "rtStreamCreate failed. rt_ret = %d, index = %u", static_cast<int>(rt_ret), i);
return RT_FAILED;
}
@@ -71,6 +80,8 @@ Status RunContextUtil::CreateRtModelResources(uint32_t stream_num, uint32_t even

rt_ret = rtModelBindStream(rt_model_, stream, 0);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "call rtModelBindStream fail, ret:%d, index:%u, when %s",
static_cast<int>(rt_ret), i, __FUNCTION__);
GELOGE(RT_FAILED, "Bind stream and model failed. rt_ret = %d, index = %u", static_cast<int>(rt_ret), i);
return RT_FAILED;
}
@@ -81,6 +92,8 @@ Status RunContextUtil::CreateRtModelResources(uint32_t stream_num, uint32_t even
rtEvent_t event = nullptr;
rt_ret = rtEventCreate(&event);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "call rtEventCreate fail, ret:%d, index:%u, when %s",
static_cast<int>(rt_ret), i, __FUNCTION__);
GELOGE(RT_FAILED, "rtEventCreate failed. rt_ret = %d, index = %u", static_cast<int>(rt_ret), i);
return RT_FAILED;
}
@@ -92,6 +105,8 @@ Status RunContextUtil::CreateRtModelResources(uint32_t stream_num, uint32_t even
rtLabel_t label = nullptr;
rt_ret = rtLabelCreateV2(&label, rt_model_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "call rtLabelCreateV2 fail, ret:%d, index:%u, when %s",
static_cast<int>(rt_ret), i, __FUNCTION__);
GELOGE(RT_FAILED, "rtLabelCreate failed. rt_ret = %d, index = %u", static_cast<int>(rt_ret), i);
return RT_FAILED;
}
@@ -143,12 +158,15 @@ Status RunContextUtil::CreateRunContext(Model &model, const ComputeGraphPtr &gra
GELOGD("Begin to Create RunContext, session_id = %lu", session_id);
// check params
if (graph == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param graph nullptr, session_id:%lu, when %s", session_id, __FUNCTION__);
GELOGE(PARAM_INVALID, "CreateRunContext param graph is null. session_id=%lu", session_id);
return PARAM_INVALID;
}

uint32_t stream_num = 0;
if (!AttrUtils::GetInt(&model, ATTR_MODEL_STREAM_NUM, stream_num)) {
REPORT_INNER_ERROR("Get Attr:%s fail for model, session_id:%lu, when %s",
ATTR_MODEL_STREAM_NUM.c_str(), session_id, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Get stream_num attr from model_def failed. session_id=%lu", session_id);
return INTERNAL_ERROR;
}
@@ -156,6 +174,8 @@ Status RunContextUtil::CreateRunContext(Model &model, const ComputeGraphPtr &gra

uint32_t event_num = 0;
if (!AttrUtils::GetInt(&model, ATTR_MODEL_EVENT_NUM, event_num)) {
REPORT_INNER_ERROR("Get Attr:%s fail for model, session_id:%lu, when %s",
ATTR_MODEL_EVENT_NUM.c_str(), session_id, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Get event_num attr from model failed. session_id=%lu", session_id);
return INTERNAL_ERROR;
}
@@ -163,6 +183,8 @@ Status RunContextUtil::CreateRunContext(Model &model, const ComputeGraphPtr &gra

uint32_t label_num = 0;
if (!AttrUtils::GetInt(&model, ATTR_MODEL_LABEL_NUM, label_num)) {
REPORT_INNER_ERROR("Get Attr:%s fail for model, session_id:%lu, when %s",
ATTR_MODEL_LABEL_NUM.c_str(), session_id, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Get label_num attr from model failed. session_id=%lu", session_id);
return INTERNAL_ERROR;
}


+ 65
- 3
ge/graph/build/stream_allocator.cc View File

@@ -76,6 +76,7 @@ Status StreamAllocator::AssignLogicalStreams(const std::map<std::string, int> &m

auto gelib = GELib::GetInstance();
if (gelib == nullptr) {
REPORT_INNER_ERROR("E19999", "Check GELib instance nullptr when %s", __FUNCTION__);
GELOGE(FAILED, "Get GELib instance failed.");
return FAILED;
}
@@ -184,6 +185,8 @@ Status StreamAllocator::AssignSingleStream() {
}

if (stream_num_ > 1) {
REPORT_INNER_ERROR("E19999", "The number of ts streams is %ld, only one is supported when %s",
stream_num_, __FUNCTION__);
GELOGE(FAILED, "The number of ts streams is %ld, only one is supported.", stream_num_);
return FAILED;
}
@@ -257,6 +260,9 @@ Status StreamAllocator::SetActiveStreamsByLabel() {
}
}
GE_CHK_BOOL_EXEC(AttrUtils::SetListInt(node->GetOpDesc(), ATTR_NAME_ACTIVE_STREAM_LIST, activated_stream_list),
REPORT_INNER_ERROR("E19999", "Set Attr:%s for op:%s(%s) failed when %s",
ATTR_NAME_ACTIVE_STREAM_LIST.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetListInt failed.");
return FAILED);
}
@@ -307,6 +313,9 @@ Status StreamAllocator::SetActiveStreamsForSubgraphs() {
}

if (!AttrUtils::SetListInt(first_active_node->GetOpDesc(), ATTR_NAME_ACTIVE_STREAM_LIST, active_streams)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s for op:%s(%s) failed when %s",
ATTR_NAME_ACTIVE_STREAM_LIST.c_str(),
first_active_node->GetName().c_str(), first_active_node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Set active streams for node %s failed.", first_active_node->GetName().c_str());
return FAILED;
}
@@ -376,6 +385,8 @@ Status StreamAllocator::InsertOneEventInTwoNodes(const NodePtr &cur_node, const
}

if (next_stream_id == kInvalidStream) {
REPORT_INNER_ERROR("E19999", "Stream id of next_node %s(%s) should not be %ld when %s",
next_node->GetName().c_str(), next_node->GetType().c_str(), kInvalidStream, __FUNCTION__);
GELOGE(FAILED, "Stream id of next_node %s should not be %ld", next_node->GetName().c_str(), kInvalidStream);
return FAILED;
}
@@ -589,8 +600,14 @@ Status StreamAllocator::OptimizeByStreamActivate() {
// -> stream(streamSwitch) -> stream(streamActivate) -> stream(stream true or false)
// No need to insert an event between node in stream(normal) and node in stream(stream true or false)
bool StreamAllocator::IsRecvNodeActivatedBySendNode(const NodePtr &send_node_ptr, const NodePtr &recv_node_ptr) const {
GE_CHECK_NOTNULL_EXEC(send_node_ptr->GetOpDesc(), GELOGE(FAILED, "op desc is nullptr"); return false);
GE_CHECK_NOTNULL_EXEC(recv_node_ptr->GetOpDesc(), GELOGE(FAILED, "op desc is nullptr"); return false);
GE_CHECK_NOTNULL_EXEC(send_node_ptr->GetOpDesc(),
REPORT_INNER_ERROR("E19999", "Check param send_node_ptr nullptr when %s", __FUNCTION__);
GELOGE(FAILED, "op desc is nullptr");
return false);
GE_CHECK_NOTNULL_EXEC(recv_node_ptr->GetOpDesc(),
REPORT_INNER_ERROR("E19999", "Check param recv_node_ptr nullptr when %s", __FUNCTION__);
GELOGE(FAILED, "op desc is nullptr");
return false);
auto cur_stream_id = send_node_ptr->GetOpDesc()->GetStreamId();
if (AttrUtils::HasAttr(recv_node_ptr->GetOpDesc(), ATTR_NAME_STREAM_LABEL)) {
// find streamActivate node
@@ -714,6 +731,8 @@ Status StreamAllocator::SplitStreams(vector<set<int64_t>> &split_streams) {
continue;
}
if (stream_id > last_stream_id) {
REPORT_INNER_ERROR("E19999", "streamid(%ld) > last_stream_id(%ld), check invalid when %s",
stream_id, last_stream_id, __FUNCTION__);
GELOGE(FAILED, "SplitStreams:streamid(%ld) > last_stream_id(%ld)", stream_id, last_stream_id);
return FAILED;
}
@@ -727,6 +746,8 @@ Status StreamAllocator::SplitStreams(vector<set<int64_t>> &split_streams) {
stream_continuous_2_node_num_map[continuous_stream_label]++;
// return error
if (stream_continuous_2_node_num_map[continuous_stream_label] > max_node_num_one_stream) {
REPORT_INNER_ERROR("E19999", "Check node[%s] stream_id[%ld] continuous stream label[%s] unsatisfied when %s",
op_desc->GetName().c_str(), stream_id, continuous_stream_label.c_str(), __FUNCTION__);
GELOGE(FAILED, "SplitStreams:node[%s] stream_id[%ld] continuous stream label[%s] unsatisfied ",
op_desc->GetName().c_str(), stream_id, continuous_stream_label.c_str());
return FAILED;
@@ -881,6 +902,8 @@ Status StreamAllocator::UpdateActiveStreamsForSwitchNode(NodePtr &switch_node) {
GE_CHECK_NOTNULL(op_desc);

if (!AttrUtils::SetListInt(op_desc, ATTR_NAME_ACTIVE_STREAM_LIST, stream_ids)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_ACTIVE_STREAM_LIST.c_str(),
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetListInt failed.");
return FAILED;
}
@@ -895,6 +918,8 @@ Status StreamAllocator::InsertActiveNodesAfterSwitch(NodePtr &switch_node, vecto
vector<string> ori_active_label_list;
if (!AttrUtils::GetListStr(switch_desc, ATTR_NAME_ACTIVE_LABEL_LIST, ori_active_label_list) ||
ori_active_label_list.empty()) {
REPORT_INNER_ERROR("E19999", "Get Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_ACTIVE_LABEL_LIST.c_str(),
switch_node->GetName().c_str(), switch_node->GetType().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Get active label list of switch %s failed.", switch_node->GetName().c_str());
return INTERNAL_ERROR;
}
@@ -918,6 +943,8 @@ Status StreamAllocator::InsertActiveNodesAfterSwitch(NodePtr &switch_node, vecto
for (auto &active_node : added_active_nodes) {
GE_CHECK_NOTNULL(switch_node->GetOutControlAnchor());
if (switch_node->GetOutControlAnchor()->LinkTo(active_node->GetInControlAnchor()) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Link from %s to %s failed when %s",
switch_node->GetName().c_str(), active_node->GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "Link %s to %s failed.", switch_node->GetName().c_str(), active_node->GetName().c_str());
return FAILED;
}
@@ -933,6 +960,8 @@ Status StreamAllocator::UpdateActiveStreamsForActiveNode(const vector<set<int64_
vector<uint32_t> new_active_streams = active_streams;
for (uint32_t logical_stream : active_streams) {
if (static_cast<size_t>(logical_stream) >= split_streams.size()) {
REPORT_INNER_ERROR("E19999", "Check logical stream:%u is out of range:%zu when %s",
logical_stream, split_streams.size(), __FUNCTION__);
GELOGE(FAILED, "logical stream is out of range.");
return FAILED;
}
@@ -951,6 +980,8 @@ Status StreamAllocator::UpdateActiveStreamsForActiveNode(const vector<set<int64_
}
}
if (!AttrUtils::SetListInt(node->GetOpDesc(), ATTR_NAME_ACTIVE_STREAM_LIST, new_active_streams)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_ACTIVE_STREAM_LIST.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Set active streams for node %s failed.", node->GetName().c_str());
return FAILED;
}
@@ -991,6 +1022,8 @@ Status StreamAllocator::UpdateActiveStreamsForSubgraphs() const {
new_active_streams.emplace(static_cast<uint32_t>(new_split_stream));
active_streams.assign(new_active_streams.begin(), new_active_streams.end());
if (!AttrUtils::SetListInt(active_op, ATTR_NAME_ACTIVE_STREAM_LIST, active_streams)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_ACTIVE_STREAM_LIST.c_str(),
active_op->GetName().c_str(), active_op->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Set active streams for node %s failed.", active_node->GetName().c_str());
return FAILED;
}
@@ -1059,6 +1092,8 @@ Status StreamAllocator::SetActiveStreamsForLoop() {

NodePtr pre_switch_node = FindSwitchNodeBeforeLoopActiveNode(node);
if (pre_switch_node == nullptr) {
REPORT_INNER_ERROR("E19999", "Find switch node before loop active node %s fail when %s",
node->GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "find switch node before loop active node %s failed", node->GetName().c_str());
return FAILED;
}
@@ -1066,6 +1101,9 @@ Status StreamAllocator::SetActiveStreamsForLoop() {
if (!AttrUtils::GetListStr(node->GetOpDesc(), ATTR_NAME_ACTIVE_LABEL_LIST, activated_label_list) ||
activated_label_list.empty()) {
GE_CHK_BOOL_EXEC(AttrUtils::SetListInt(node->GetOpDesc(), ATTR_NAME_ACTIVE_STREAM_LIST, loop_active_streams),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s",
ATTR_NAME_ACTIVE_STREAM_LIST.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetListInt failed.");
return FAILED);
for (const auto &stream_id : loop_active_streams) {
@@ -1112,6 +1150,8 @@ Status StreamAllocator::CheckStreamActived() const {
uint32_t stream_id = static_cast<uint32_t>(node->GetOpDesc()->GetStreamId());
auto iter = find(active_streams.begin(), active_streams.end(), stream_id);
if (iter != active_streams.end()) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s) cannot active its own stream %u, check invalid when %s",
node->GetName().c_str(), node->GetType().c_str(), stream_id, __FUNCTION__);
GELOGE(FAILED, "Node %s cannot active its own stream %u.", node->GetName().c_str(), stream_id);
return FAILED;
}
@@ -1139,6 +1179,7 @@ Status StreamAllocator::RefreshContinuousEvents() {
for (size_t i = 0; i < send_events.size(); i++) {
auto find_it = old_to_new_events.find(send_events[i]);
if (find_it == old_to_new_events.end()) {
REPORT_INNER_ERROR("E19999", "Check invalid send event %u when %s", send_events[i], __FUNCTION__);
GELOGE(FAILED, "RefreshContinuousEvents: invalid send event %u", send_events[i]);
return FAILED;
}
@@ -1152,6 +1193,7 @@ Status StreamAllocator::RefreshContinuousEvents() {
for (size_t i = 0; i < recv_events.size(); i++) {
auto find_it = old_to_new_events.find(recv_events[i]);
if (find_it == old_to_new_events.end()) {
REPORT_INNER_ERROR("E19999", "Check invalid recv event %u when %s", recv_events[i], __FUNCTION__);
GELOGE(FAILED, "RefreshContinuousEvents: invalid recv event %u", recv_events[i]);
return FAILED;
}
@@ -1180,7 +1222,11 @@ Status StreamAllocator::InsertSyncEventNodes() {

int64_t temp_stream_id = node->GetOpDesc()->GetStreamId();
op_desc_ptr->SetStreamId(temp_stream_id);
GE_CHK_BOOL_EXEC(AttrUtils::SetInt(op_desc_ptr, RECV_ATTR_EVENT_ID, event_id), GELOGE(FAILED, "SetInt failed.");
GE_CHK_BOOL_EXEC(AttrUtils::SetInt(op_desc_ptr, RECV_ATTR_EVENT_ID, event_id),
REPORT_INNER_ERROR("E19999", "Set Attr:%s for op:%s(%s) failed, event_id:%u, when %s",
RECV_ATTR_EVENT_ID.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt failed.");
return FAILED);
(void)AttrUtils::SetListStr(op_desc_ptr, ATTR_NAME_DATA_DUMP_ORIGIN_OP_NAMES,
std::move(std::vector<std::string>()));
@@ -1189,6 +1235,8 @@ Status StreamAllocator::InsertSyncEventNodes() {
GE_CHECK_NOTNULL(recv_node->GetOutControlAnchor());
Status status = GraphUtils::AddEdge(recv_node->GetOutControlAnchor(), node->GetInControlAnchor());
if (status != SUCCESS) {
REPORT_INNER_ERROR("E19999", "Add edge from node %s to node %s failed when %s",
recv_node->GetName().c_str(), node->GetName().c_str(), __FUNCTION__);
GELOGE(status, "Add edge for node %s and node %s failed.", recv_node->GetName().c_str(),
node->GetName().c_str());
return status;
@@ -1217,6 +1265,8 @@ Status StreamAllocator::InsertSyncEventNodes() {
GE_CHECK_NOTNULL(send_node->GetInControlAnchor());
Status status = GraphUtils::AddEdge(node->GetOutControlAnchor(), send_node->GetInControlAnchor());
if (status != SUCCESS) {
REPORT_INNER_ERROR("E19999", "Add edge from node %s to node %s failed when %s",
node->GetName().c_str(), send_node->GetName().c_str(), __FUNCTION__);
GELOGE(status, "Add edge for node %s and node %s failed.", node->GetName().c_str(),
send_node->GetName().c_str());
return status;
@@ -1228,6 +1278,8 @@ Status StreamAllocator::InsertSyncEventNodes() {

Status status = whole_graph_->InsertGraphEvents();
if (status != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Insert Graph Events fail, graph:%s, when %s",
whole_graph_->GetName().c_str(), __FUNCTION__);
GELOGE(status, "Graph ReorderEventNodes failed");
return status;
}
@@ -1274,6 +1326,8 @@ Status StreamAllocator::GetMaxStreamAndTask(bool huge_stream, uint32_t &max_stre
}
rtError_t ret = rtGetMaxStreamAndTask(stream_type, &max_stream_count, &max_task_count);
if (ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "call rtGetMaxStreamAndTask fail, ret:%d, stream_type:%u, when %s",
static_cast<int>(ret), stream_type, __FUNCTION__);
GELOGE(FAILED, "Get max stream and task count by rts failed.");
return FAILED;
}
@@ -1416,6 +1470,7 @@ Status StreamAllocator::AddActiveNodes(NodePtr &switch_node, const vector<string
for (size_t i = 0; i < label_num; i++) {
const string &active_label = ori_active_label_list[i];
if (labeled_streams_.find(active_label) == labeled_streams_.end()) {
REPORT_INNER_ERROR("E19999", "can not find stream label:%s when %s", active_label.c_str(), __FUNCTION__);
GELOGE(FAILED, "can not find stream label %s", active_label.c_str());
return FAILED;
}
@@ -1442,11 +1497,15 @@ Status StreamAllocator::AddActiveNodes(NodePtr &switch_node, const vector<string
}
GE_CHECK_NOTNULL(switch_node->GetOutControlAnchor());
if (switch_node->GetOutControlAnchor()->Unlink(node->GetInControlAnchor()) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("Unlink %s to %s failed when %s",
switch_node->GetName().c_str(), node->GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "Unlink %s to %s failed.", switch_node->GetName().c_str(), node->GetName().c_str());
return FAILED;
}
GE_CHECK_NOTNULL(active_node->GetOutControlAnchor());
if (active_node->GetOutControlAnchor()->LinkTo(node->GetInControlAnchor()) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("Link %s to %s failed when %s",
active_node->GetName().c_str(), node->GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "Link %s to %s failed.", active_node->GetName().c_str(), node->GetName().c_str());
return FAILED;
}
@@ -1477,12 +1536,15 @@ Status StreamAllocator::AddActiveNodes(NodePtr &switch_node, const vector<string

Status StreamAllocator::SetActiveStreamList(NodePtr &active_node, const string &active_label) {
if (labeled_streams_.find(active_label) == labeled_streams_.end()) {
REPORT_INNER_ERROR("E19999", "Can not find stream label:%s when %s", active_label.c_str(), __FUNCTION__);
GELOGE(FAILED, "Can not find stream label %s.", active_label.c_str());
return FAILED;
}
set<int64_t> &streams = labeled_streams_[active_label];
vector<int64_t> active_streams(streams.begin(), streams.end());
if (!AttrUtils::SetListInt(active_node->GetOpDesc(), ATTR_NAME_ACTIVE_STREAM_LIST, active_streams)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_ACTIVE_STREAM_LIST.c_str(),
active_node->GetName().c_str(), active_node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetListInt of %s failed.", ATTR_NAME_ACTIVE_STREAM_LIST.c_str());
return FAILED;
}


+ 9
- 0
ge/graph/build/stream_graph_optimizer.cc View File

@@ -14,6 +14,9 @@
* limitations under the License.
*/
#include "stream_graph_optimizer.h"

#include <securec.h>

#include "common/util.h"
#include "framework/common/debug/ge_log.h"
#include "graph/utils/node_utils.h"
@@ -122,6 +125,9 @@ Status StreamGraphOptimizer::OptimizeStreamedSubGraph(const ComputeGraphPtr &com
GE_CHECK_NOTNULL(op_desc);
int64_t stream_id = op_desc->GetStreamId();
if (static_cast<size_t>(stream_id) >= run_context.graphStreamList.size()) {
REPORT_INNER_ERROR("E19999", "Check stream_id:%ld in op:%s(%s) is bigger than run_context.graphStreamList.size():%zu "
"when %s", stream_id, op_desc->GetName().c_str(),
op_desc->GetType().c_str(), run_context.graphStreamList.size(), __FUNCTION__);
GELOGE(FAILED, "stream_id %ld is bigger than run_context.graphStreamList.size() %zu", stream_id,
run_context.graphStreamList.size());
return FAILED;
@@ -135,6 +141,9 @@ Status StreamGraphOptimizer::OptimizeStreamedSubGraph(const ComputeGraphPtr &com
for (auto iter = graph_optimizers.begin(); iter != graph_optimizers.end(); ++iter) {
GE_CHECK_NOTNULL(*iter);
Status ret = (*iter)->OptimizeStreamGraph(*subgraph, run_context);
REPORT_CALL_ERROR("E19999", "Call optimize streamed subgraph failed, subgraph: %s, engine_name: %s, graph "
"Optimizer num: %zu, ret: %u", subgraph->GetName().c_str(), engine_name.c_str(),
graph_optimizers.size(), ret);
if (ret != SUCCESS) {
GELOGE(
ret,


+ 77
- 4
ge/graph/build/task_generator.cc View File

@@ -69,6 +69,7 @@ Status TaskGenerator::GetTaskInfo(Model &model, ComputeGraphPtr &graph, uint64_t
GELOGD("Begin to Get TaskInfo. session_id=%lu", session_id);
// Check params
if (graph == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param graph is null, session_id:%lu, when %s", session_id, __FUNCTION__);
GELOGE(PARAM_INVALID, "GetTaskInfo param graph is null. session_id=%lu", session_id);
return PARAM_INVALID;
}
@@ -93,6 +94,8 @@ Status TaskGenerator::GetTaskInfo(Model &model, ComputeGraphPtr &graph, uint64_t
op_name.push_back(iter.second);
}
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetListStr(model, ATTR_MODEL_TASK_INDEX_OP_NAME, op_name),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for model:%s when %s",
ATTR_MODEL_TASK_INDEX_OP_NAME.c_str(), model.GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetListStr failed.");
return FAILED);

@@ -106,6 +109,8 @@ Status TaskGenerator::GetTaskInfo(Model &model, ComputeGraphPtr &graph, uint64_t
for (const TaskDef &task_def_temp : task_def_list) {
TaskDef *task_def = model_task_def.add_task();
if (task_def == nullptr) {
REPORT_INNER_ERROR("E19999", "Add task_def in ModelTaskDef fail, session_id:%lu, graph:%s, model:%s, when %s",
session_id, graph->GetName().c_str(), model.GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "task_def is nullptr.");
return FAILED;
}
@@ -126,30 +131,44 @@ Status TaskGenerator::AddModelTaskToModel(const ModelTaskDef &model_task_def, ui
RunContext &run_context) {
GE_CHK_BOOL_EXEC(
AttrUtils::SetInt(model, MODEL_ATTR_TASK_GEN_BASE_ADDR, reinterpret_cast<uintptr_t>(run_context.dataMemBase)),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for model:%s when %s",
MODEL_ATTR_TASK_GEN_BASE_ADDR.c_str(), model.GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt MODEL_ATTR_TASK_GEN_BASE_ADDR failed.");
return FAILED);
GE_CHK_BOOL_EXEC(
AttrUtils::SetInt(model, MODEL_ATTR_TASK_GEN_WEIGHT_ADDR, reinterpret_cast<uintptr_t>(run_context.weightMemBase)),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for model:%s when %s",
MODEL_ATTR_TASK_GEN_WEIGHT_ADDR.c_str(), model.GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt MODEL_ATTR_TASK_GEN_WEIGHT_ADDR failed.");
return FAILED);
GE_CHK_BOOL_EXEC(AttrUtils::SetInt(model, ATTR_MODEL_TASK_GEN_VAR_ADDR, reinterpret_cast<uintptr_t>(var_mem_base_)),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for model:%s when %s",
ATTR_MODEL_TASK_GEN_VAR_ADDR.c_str(), model.GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt ATTR_MODEL_TASK_GEN_VAR_ADDR failed.");
return FAILED);
GE_CHK_BOOL_EXEC(AttrUtils::SetInt(model, ATTR_MODEL_VAR_SIZE, var_mem_size_),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for model:%s when %s",
ATTR_MODEL_VAR_SIZE.c_str(), model.GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt ATTR_MODEL_VAR_SIZE failed.");
return FAILED);
GE_CHK_BOOL_EXEC(AttrUtils::SetInt(model, MODEL_ATTR_SESSION_ID, session_id),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for mode:%s when %s",
MODEL_ATTR_SESSION_ID.c_str(), model.GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetInt MODEL_ATTR_SESSION_ID failed.");
return FAILED);

size_t task_size = model_task_def.ByteSizeLong();
ge::Buffer serial_buff(task_size);
if (!model_task_def.SerializePartialToArray(serial_buff.GetData(), static_cast<int>(task_size))) {
REPORT_INNER_ERROR("E19999", "model_task_def's serialize failed, model name = %s, task_size=%zu when %s",
model.GetName().c_str(), task_size, __FUNCTION__);
GELOGE(FAILED, "model_task_def's serialize failed, model name = %s, task_size=%zu.", model.GetName().c_str(),
task_size);
return FAILED;
}
if (!AttrUtils::SetZeroCopyBytes(model, MODEL_ATTR_TASKS, std::move(serial_buff))) {
REPORT_INNER_ERROR("E19999", "Set model task to model failed, model name = %s, task_size=%zu.",
model.GetName().c_str(), task_size, __FUNCTION__);
GELOGE(FAILED, "Set model task to model failed, model name = %s, task_size=%zu.", model.GetName().c_str(),
task_size);
return FAILED;
@@ -167,7 +186,10 @@ Status TaskGenerator::UpdateOpIsVarAttr(const OpDescPtr &op_desc, uint64_t sessi
for (int64_t input : input_offsets) {
input_var.push_back(VarManager::Instance(session_id)->IsVarAddr(input));
}
GE_CHK_BOOL_EXEC(AttrUtils::SetListBool(op_desc, kIsInputVar, input_var), GELOGE(FAILED, "SetListBool failed.");
GE_CHK_BOOL_EXEC(AttrUtils::SetListBool(op_desc, kIsInputVar, input_var),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", kIsInputVar,
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetListBool failed.");
return FAILED);
}

@@ -177,7 +199,10 @@ Status TaskGenerator::UpdateOpIsVarAttr(const OpDescPtr &op_desc, uint64_t sessi
for (int64_t output : output_offsets) {
output_var.push_back(VarManager::Instance(session_id)->IsVarAddr(output));
}
GE_CHK_BOOL_EXEC(AttrUtils::SetListBool(op_desc, kIsOutputVar, output_var), GELOGE(FAILED, "SetListBool failed.");
GE_CHK_BOOL_EXEC(AttrUtils::SetListBool(op_desc, kIsOutputVar, output_var),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", kIsOutputVar,
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "SetListBool failed.");
return FAILED);
}
return SUCCESS;
@@ -252,6 +277,7 @@ Status TaskGenerator::GenerateTask(RunContext &run_context, ComputeGraphPtr &gra
GELOGD("Beign to generate task, graph name is %s.", graph->GetName().c_str());
std::shared_ptr<GELib> ge_lib = GELib::GetInstance();
if ((ge_lib == nullptr) || !ge_lib->InitFlag()) {
REPORT_INNER_ERROR("E19999", "Check GELib instance not init before when %s", __FUNCTION__);
GELOGE(GE_CLI_GE_NOT_INITIALIZED, "GenerateTask failed.");
return GE_CLI_GE_NOT_INITIALIZED;
}
@@ -319,6 +345,8 @@ Status TaskGenerator::GenerateTask(RunContext &run_context, ComputeGraphPtr &gra
}
auto kernel_info_store = ops_kernel_manager.GetOpsKernelInfoStore(op_kernel_lib_name);
if (kernel_info_store == nullptr) {
REPORT_INNER_ERROR("E19999", "Get ops kernel info store failed for op:%s(%s), op_kernel_name:%s when %s",
node->GetName().c_str(), node->GetType().c_str(), op_kernel_lib_name.c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR,
"No ops kernel store or ops kernel builder found. node:%s(%s), op_kernel_lib_name=%s.",
name.c_str(),
@@ -344,6 +372,8 @@ Status TaskGenerator::GenerateTask(RunContext &run_context, ComputeGraphPtr &gra
auto ret = OpsKernelBuilderManager::Instance().GenerateTask(*node, run_context, task_def_list);
GE_TIMESTAMP_ADD(GenerateTask);
if (ret != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call OpsKernelBuilderManager GenerateTask fail for op:%s(%s) when %s",
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(ret, "Call %s to generate node[name:%s(%s), id:%ld, stream_id:%ld] task failed.",
op_kernel_lib_name.c_str(), name.c_str(), type.c_str(), op_id, stream_id);
return ret;
@@ -353,6 +383,9 @@ Status TaskGenerator::GenerateTask(RunContext &run_context, ComputeGraphPtr &gra
size_t task_list_size_after = task_def_list.size();
// If tasks is reduced
if (task_list_size_after < task_list_size_before) {
REPORT_INNER_ERROR("E19999", "Call %s to generate node[name:%s(%s), id:%ld, stream_id:%ld] task "
"but task num from %zu to %zu, check invalid", op_kernel_lib_name.c_str(), name.c_str(),
type.c_str(), op_id, stream_id, task_list_size_before, task_list_size_after);
GELOGE(FAILED, "Call %s to generate node[name:%s(%s), id:%ld, stream_id:%ld] task. but task num from %zu to %zu.",
op_kernel_lib_name.c_str(), name.c_str(), type.c_str(), op_id, stream_id, task_list_size_before,
task_list_size_after);
@@ -417,6 +450,9 @@ Status TaskGenerator::GenerateTaskForFusionNode(FusionTaskInfo &fusion_task_info
size_t task_list_size_before = task_def_list.size();
OpsKernelInfoStorePtr kernel_info_store = ops_kernel_manager.GetOpsKernelInfoStore(op_kernel_lib_name);
if (kernel_info_store == nullptr) {
REPORT_INNER_ERROR("E19999", "Get ops kernel info store failed for op:%s(%s), op_kernel_name:%s when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(),
op_kernel_lib_name.c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR,
"Fusion: No ops kernel store or ops kernel builder found. fusion_node:%s(%s), op_kernel_lib_name=%s.",
fusion_node_name.c_str(), fusion_node_type.c_str(), op_kernel_lib_name.c_str());
@@ -433,6 +469,9 @@ Status TaskGenerator::GenerateTaskForFusionNode(FusionTaskInfo &fusion_task_info
int64_t op_id = op_desc->GetId();
int64_t stream_id = op_desc->GetStreamId();
if (stream_id < 0 || stream_id >= (int64_t)run_context.graphStreamList.size()) {
REPORT_INNER_ERROR("E19999", "Fusion: fusion_node[name:%s(%s), id:%ld] stream id is invalid, "
"stream list size=%zu, when %s", fusion_node_name.c_str(), fusion_node_type.c_str(),
op_id, run_context.graphStreamList.size(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Fusion: fusion_node[name:%s(%s), id:%ld] stream id is invalid, stream list size=%zu",
fusion_node_name.c_str(), fusion_node_type.c_str(), op_id, run_context.graphStreamList.size());
return INTERNAL_ERROR;
@@ -444,6 +483,9 @@ Status TaskGenerator::GenerateTaskForFusionNode(FusionTaskInfo &fusion_task_info
op_kernel_lib_name.c_str(), fusion_node_name.c_str(), fusion_node_type.c_str(), op_id, stream_id);
ret = OpsKernelBuilderManager::Instance().GenerateTask(*fusion_node, run_context, task_def_list);
if (ret != SUCCESS) {
REPORT_CALL_ERROR("E19999", " Call %s to generate fusion_node:[fusion_node_name:%s(%s), "
"id:%ld, stream_id:%ld] task failed when %s", op_kernel_lib_name.c_str(),
fusion_node_name.c_str(), fusion_node_type.c_str(), op_id, stream_id, __FUNCTION__);
GELOGE(ret,
"Fusion: Call %s to generate fusion_node:[fusion_node_name:%s(%s), "
"id:%ld, stream_id:%ld] task failed.",
@@ -455,6 +497,10 @@ Status TaskGenerator::GenerateTaskForFusionNode(FusionTaskInfo &fusion_task_info
size_t task_list_size_after = task_def_list.size();
// if tasks is reduced
if (task_list_size_after < task_list_size_before) {
REPORT_INNER_ERROR("E19999", "InsertProfilingTask for fusion_node:[fusion_node_name:%s(%s), "
"id:%ld, stream_id:%ld] task, but task num from %zu to %zu, check invalid when %s",
op_kernel_lib_name.c_str(), fusion_node_name.c_str(), fusion_node_type.c_str(),
op_id, stream_id, task_list_size_before, task_list_size_after, __FUNCTION__);
GELOGE(FAILED,
"Fusion: Call %s to generate fusion_node:[fusion_node_name:%s(%s), "
"id:%ld, stream_id:%ld] task. but task num from %zu to %zu.",
@@ -489,6 +535,8 @@ Status TaskGenerator::GenerateTaskForFusionNode(FusionTaskInfo &fusion_task_info

Status TaskGenerator::UpdateAnchorStatus(const NodePtr &node) {
if (NodeUtils::SetAllAnchorStatus(node) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "SetAllAnchorStatus fail for op:%s(%s) when %s",
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "NodeUtils::SetAllAnchorStatus failed.");
return INTERNAL_ERROR;
}
@@ -496,6 +544,8 @@ Status TaskGenerator::UpdateAnchorStatus(const NodePtr &node) {
auto peer_anchor = anchor->GetPeerOutAnchor();
if (peer_anchor == nullptr) {
if (AnchorUtils::SetStatus(anchor, ANCHOR_SUSPEND) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Set in peer anchor status fail for op:%s(%s), anchor_index:%d, when %s",
node->GetName().c_str(), node->GetType().c_str(), anchor->GetIdx(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "AnchorUtils::SetStatus failed.");
return INTERNAL_ERROR;
}
@@ -506,11 +556,15 @@ Status TaskGenerator::UpdateAnchorStatus(const NodePtr &node) {
bool is_const = NodeUtils::GetConstOpType(peer_anchor->GetOwnerNode(), const_type);
if (is_const && (const_type == CONSTANT)) {
if (AnchorUtils::SetStatus(anchor, ANCHOR_CONST) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Set in anchor CONST status fail for op:%s(%s), anchor_index:%d, when %s",
node->GetName().c_str(), node->GetType().c_str(), anchor->GetIdx(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "AnchorUtils::SetStatus failed.");
return INTERNAL_ERROR;
}
} else {
if (AnchorUtils::SetStatus(anchor, ANCHOR_DATA) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Set in anchor DATA status fail for op:%s(%s), anchor_index:%d, when %s",
node->GetName().c_str(), node->GetType().c_str(), anchor->GetIdx(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "AnchorUtils::SetStatus failed.");
return INTERNAL_ERROR;
}
@@ -523,12 +577,15 @@ Status TaskGenerator::UpdateAnchorStatus(const NodePtr &node) {
Status TaskGenerator::MarkNodeAndSetIndex(ComputeGraphPtr &graph) {
auto ge_lib = GELib::GetInstance();
if ((ge_lib == nullptr) || !ge_lib->InitFlag()) {
REPORT_INNER_ERROR("E19999", "Check GELib instance not init before when %s", __FUNCTION__);
GELOGE(GE_CLI_GE_NOT_INITIALIZED, "GE is not initialized or is finalized.");
return GE_CLI_GE_NOT_INITIALIZED;
}

const auto all_nodes = graph->GetNodes(graph->GetGraphUnknownFlag());
if (all_nodes.empty()) {
REPORT_INNER_ERROR("E19999", "Check param all_nodes empty in graph:%s when %s",
graph->GetName().c_str(), __FUNCTION__);
GELOGE(GE_GRAPH_GRAPH_NODE_NULL, "Graph's node is empty");
return GE_GRAPH_GRAPH_NODE_NULL;
}
@@ -584,6 +641,9 @@ Status TaskGenerator::MarkFirstAndLastOps(const vector<OpDescPtr> &ops, bool is_
for (auto &op_desc : continuous_ops) {
string op_kernel_lib_name = op_desc->GetOpKernelLibName();
if (op_kernel_lib_name.empty()) {
REPORT_INNER_ERROR("E19999", "Get ops kernel info store failed for op:%s(%s), op_kernel_name:%s when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(),
op_kernel_lib_name.c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "node:%s(%s) get op kernel lib failed.", op_desc->GetName().c_str(),
op_desc->GetType().c_str());
return INTERNAL_ERROR;
@@ -599,9 +659,17 @@ Status TaskGenerator::MarkFirstAndLastOps(const vector<OpDescPtr> &ops, bool is_

for (auto &it : first_and_last_ops) {
auto &op_pair = it.second;
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetBool(op_pair.first, kIsFirstNode, true), GELOGE(FAILED, "SetBool failed.");
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetBool(op_pair.first, kIsFirstNode, true),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", kIsFirstNode,
op_pair.first->GetName().c_str(), op_pair.first->GetType().c_str(),
__FUNCTION__);
GELOGE(FAILED, "SetBool failed.");
return FAILED);
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetBool(op_pair.second, kIsLastNode, true), GELOGE(FAILED, "SetBool failed.");
GE_CHK_BOOL_EXEC(ge::AttrUtils::SetBool(op_pair.second, kIsLastNode, true),
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", kIsLastNode,
op_pair.second->GetName().c_str(), op_pair.second->GetType().c_str(),
__FUNCTION__);
GELOGE(FAILED, "SetBool failed.");
return FAILED);
}
}
@@ -906,6 +974,8 @@ Status TaskGenerator::InsertProfilingArTaskBefore(const OpDescPtr &op_desc, std:
for (size_t i = 0; i < all_reduce_nodes.size(); i++) {
if (all_reduce_nodes[i] == node_index) {
GE_IF_BOOL_EXEC(TypeUtils::CheckUint64MulOverflow(i, kProfilingArStep),
REPORT_INNER_ERROR("E19999", "Multiply result is out of range when calc profiling ar log id "
"for node:%s(%s)", op_desc->GetName().c_str(), op_desc->GetType().c_str());
GELOGE(FAILED, "Multiply result is out of range.");
return FAILED);
ar_log_id = i * kProfilingArStep + kProfilingArStartLogid;
@@ -998,6 +1068,8 @@ Status TaskGenerator::InsertProfilingArTaskAfter(const OpDescPtr &op_desc, std::
for (size_t i = 0; i < all_reduce_nodes.size(); i++) {
if (all_reduce_nodes[i] == node_index) {
GE_IF_BOOL_EXEC(TypeUtils::CheckUint64MulOverflow(i, kProfilingArStep),
REPORT_INNER_ERROR("E19999", "Multiply result is out of range when calc profiling ar log id "
"for node:%s(%s)", op_desc->GetName().c_str(), op_desc->GetType().c_str());
GELOGE(FAILED, "Multiply result is out of range.");
return FAILED);
ar_log_id = i * kProfilingArStep + kProfilingArEndLogid;
@@ -1107,6 +1179,7 @@ Status TaskGenerator::SetUnknownShapeStream(RunContext &run_context, rtStream_t
run_context.stream = stream;
rtError_t rt_ret = rtModelBindStream(run_context.model, stream, 0);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtModelBindStream fail, ret:0x%X when %s", rt_ret, __FUNCTION__);
GELOGE(FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
GE_CHK_RT_RET(rtStreamDestroy(stream));
return FAILED;


+ 2
- 0
ge/graph/common/bcast.cc View File

@@ -73,6 +73,8 @@ Status BCast::SetShapeDifferentInfo(const kVecInt &x, const kVecInt &y) {
y_bcast_i = x_i;
grad_y_reduce_idx_.push_back(n - 1 - i);
} else {
REPORT_INNER_ERROR("E19999", "SetShapeDifferentInfo failed. Two tensor shapes are not compatible "
"according to the broadcasting rule.");
GELOGE(domi::PARAM_INVALID,
"SetShapeDifferentInfo failed. Two tensor shapes are not compatible "
"according to the broadcasting rule.");


+ 7
- 0
ge/graph/common/bcast.h View File

@@ -111,11 +111,14 @@ class BCast {
const std::function<OutT(InT const &, InT const &)> &func) {
Status ret;
if (func == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param func nullptr when %s", __FUNCTION__);
GELOGE(domi::PARAM_INVALID, "Param func is null");
return domi::PARAM_INVALID;
}
// Min input num is 2
if (input.size() < kMinDimNum) {
REPORT_INNER_ERROR("E19999", "Param input.size():%zu < %zu, check invalid when %s",
input.size(), kMinDimNum, __FUNCTION__);
GELOGE(domi::PARAM_INVALID, "Input size is smaller than two.");
return domi::PARAM_INVALID;
}
@@ -149,11 +152,14 @@ class BCast {
Status BCastComputeCheck(const std::vector<ConstGeTensorPtr> &input, std::vector<OutT> &v_output,
const std::function<OutT(InT const &, InT const &, DataType &type, Status &)> &func) {
if (func == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param func nullptr when %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "Param func is null");
return PARAM_INVALID;
}
// Min input num is 2
if (input.size() < kMinDimNum) {
REPORT_INNER_ERROR("E19999", "Param input.size():%zu < %zu, check invalid when %s",
input.size(), kMinDimNum, __FUNCTION__);
GELOGE(PARAM_INVALID, "Input size is smaller than two.");
return PARAM_INVALID;
}
@@ -179,6 +185,7 @@ class BCast {
auto value = func((*(reinterpret_cast<const InT *>(x1_data) + x_index)),
(*(reinterpret_cast<const InT *>(x2_data) + y_index)), data_type, ret);
if (ret != SUCCESS) {
REPORT_INNER_ERROR("E19999", "BCastComputeCheck func execute failed, datatype is %d.", data_type);
GELOGE(ret, "BCastComputeCheck func execute failed, datatype is %d.", data_type);
return ret;
}


+ 18
- 0
ge/graph/common/omg_util.cc View File

@@ -36,6 +36,8 @@ Status GetOriginalType(const ge::NodePtr &node, string &type) {
GE_CHECK_NOTNULL(node->GetOpDesc());
bool ret = ge::AttrUtils::GetStr(node->GetOpDesc(), ATTR_NAME_FRAMEWORK_ORIGINAL_TYPE, type);
if (!ret) {
REPORT_INNER_ERROR("E19999", "Get Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_FRAMEWORK_ORIGINAL_TYPE.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Get FrameWorkOp original type [%s]", type.c_str());
return INTERNAL_ERROR;
}
@@ -55,6 +57,8 @@ Status SetStreamLabel(const ge::NodePtr &node, const std::string &label) {
GE_CHECK_NOTNULL(tmp_desc);

if (!AttrUtils::SetStr(tmp_desc, ge::ATTR_NAME_STREAM_LABEL, label)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_STREAM_LABEL.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Op: %s set ATTR_NAME_STREAM_LABEL failed", node->GetName().c_str());
return FAILED;
}
@@ -72,6 +76,8 @@ Status SetCycleEvent(const ge::NodePtr &node) {
OpDescPtr tmp_desc = node->GetOpDesc();
GE_CHECK_NOTNULL(tmp_desc);
if (!AttrUtils::SetBool(tmp_desc, ge::ATTR_NAME_STREAM_CYCLE_EVENT_FLAG, true)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_STREAM_CYCLE_EVENT_FLAG.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Op: %s set ATTR_NAME_STREAM_CYCLE_EVENT_FLAG failed", node->GetName().c_str());
return FAILED;
}
@@ -90,6 +96,8 @@ Status SetActiveLabelList(const ge::NodePtr &node, const std::vector<std::string
OpDescPtr tmp_desc = node->GetOpDesc();
GE_CHECK_NOTNULL(tmp_desc);
if (!AttrUtils::SetListStr(tmp_desc, ge::ATTR_NAME_ACTIVE_LABEL_LIST, active_label_list)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_ACTIVE_LABEL_LIST.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Op: %s set ATTR_NAME_ACTIVE_LABEL_LIST failed", node->GetName().c_str());
return FAILED;
}
@@ -108,6 +116,8 @@ Status SetSwitchBranchNodeLabel(const ge::NodePtr &node, const std::string &bran
OpDescPtr tmp_desc = node->GetOpDesc();
GE_CHECK_NOTNULL(tmp_desc);
if (!AttrUtils::SetStr(tmp_desc, ge::ATTR_NAME_SWITCH_BRANCH_NODE_LABEL, branch_label)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_SWITCH_BRANCH_NODE_LABEL.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Op: %s set ATTR_NAME_SWITCH_BRANCH_NODE_LABEL failed", node->GetName().c_str());
return FAILED;
}
@@ -126,6 +136,8 @@ Status SetSwitchTrueBranchFlag(const ge::NodePtr &node, bool value) {
OpDescPtr tmp_desc = node->GetOpDesc();
GE_CHECK_NOTNULL(tmp_desc);
if (!AttrUtils::SetBool(tmp_desc, ge::ATTR_NAME_SWITCH_TRUE_BRANCH_FLAG, value)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_SWITCH_TRUE_BRANCH_FLAG.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Op: %s set ATTR_NAME_SWITCH_TRUE_BRANCH_FLAG failed", node->GetName().c_str());
return FAILED;
}
@@ -144,6 +156,8 @@ Status SetOriginalNodeName(const ge::NodePtr &node, const std::string &orig_name
OpDescPtr tmp_desc = node->GetOpDesc();
GE_CHECK_NOTNULL(tmp_desc);
if (!AttrUtils::SetStr(tmp_desc, ge::ATTR_NAME_ORIG_NODE_NAME, orig_name)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_ORIG_NODE_NAME.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Op: %s set ATTR_NAME_ORIG_NODE_NAME failed", node->GetName().c_str());
return FAILED;
}
@@ -161,6 +175,8 @@ Status SetCyclicDependenceFlag(const ge::NodePtr &node) {
OpDescPtr tmp_desc = node->GetOpDesc();
GE_CHECK_NOTNULL(tmp_desc);
if (!AttrUtils::SetBool(tmp_desc, ge::ATTR_NAME_CYCLIC_DEPENDENCE_FLAG, true)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_CYCLIC_DEPENDENCE_FLAG.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Op: %s set ATTR_NAME_CYCLIC_DEPENDENCE_FLAG failed", node->GetName().c_str());
return FAILED;
}
@@ -180,6 +196,8 @@ Status SetNextIteration(const ge::NodePtr &node, const std::string &next) {
GE_CHECK_NOTNULL(tmp_desc);

if (!AttrUtils::SetStr(tmp_desc, ge::ATTR_NAME_NEXT_ITERATION, next)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_NEXT_ITERATION.c_str(),
node->GetName().c_str(), node->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Op: %s set ATTR_NAME_NEXT_ITERATION failed", node->GetName().c_str());
return FAILED;
}


+ 30
- 0
ge/graph/execute/graph_execute.cc View File

@@ -40,6 +40,7 @@ GraphExecutor::~GraphExecutor() {
rtError_t rt_ret;
rt_ret = rtFreeHost(buffer_addr);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtFreeHost fail, ret:0x%X when %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "[GraphManager] subgraph free buffer failed, ret: 0x%X", rt_ret);
}
}
@@ -51,14 +52,17 @@ GraphExecutor::~GraphExecutor() {
Status GraphExecutor::SetCondition(std::mutex *mutex, std::condition_variable *cond,
std::shared_ptr<GraphModelListener> listener) {
if (mutex == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param mutex nullptr when %s", __FUNCTION__);
GELOGE(GE_GRAPH_PARAM_NULLPTR, "[SetCondition] input param mutex is nullptr.");
return GE_GRAPH_PARAM_NULLPTR;
}
if (cond == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param cond nullptr when %s", __FUNCTION__);
GELOGE(GE_GRAPH_PARAM_NULLPTR, "[SetCondition] input param cond is nullptr.");
return GE_GRAPH_PARAM_NULLPTR;
}
if (listener == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param listener nullptr when %s", __FUNCTION__);
GELOGE(GE_GRAPH_PARAM_NULLPTR, "[SetCondition] input param listener is nullptr.");
return GE_GRAPH_PARAM_NULLPTR;
}
@@ -75,6 +79,7 @@ Status GraphExecutor::SetCondition(std::mutex *mutex, std::condition_variable *c

Status GraphExecutor::SetGraphContext(GraphContextPtr graph_context_ptr) {
if (graph_context_ptr == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param graph_context_ptr nullptr when %s", __FUNCTION__);
GELOGE(GE_GRAPH_PARAM_NULLPTR, "[SetGraphContext] input param graph_context_ptr is nullptr");
return GE_GRAPH_PARAM_NULLPTR;
}
@@ -101,6 +106,7 @@ Status GraphExecutor::FreeInOutBuffer() {
rtError_t rt_ret;
rt_ret = rtFreeHost(*iter);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtFreeHost fail, ret:0x%X when %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "[GraphManager] subgraph free buffer failed, ret: 0x%X", rt_ret);
(void)buffer_addr_.erase(buffer_addr_.begin(), iter);
return GE_GRAPH_FREE_FAILED;
@@ -146,6 +152,8 @@ Status GraphExecutor::MallocInOutBuffer(const std::vector<uint64_t> &buffer_size
void *tmp_buf = nullptr;
rt_ret = rtMallocHost(&tmp_buf, buffer_size[i]);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMallocHost fail, size:%lu, ret:0x%X when %s",
buffer_size[i], rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "[GraphManager] subgraph malloc buffer failed, ret: 0x%X", rt_ret);
return GE_GRAPH_MALLOC_FAILED;
}
@@ -191,6 +199,8 @@ Status GraphExecutor::PrepareInputData(const std::vector<GeTensor> &input_tensor
rtError_t rt_ret = rtMemcpy(addrVec[i], bufferSizeVec[i], in_tensor->GetData().data(),
in_tensor->GetData().size(), RT_MEMCPY_HOST_TO_HOST);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail, dst_size:%lu, src_size:%zu, ret:0x%X when %s",
bufferSizeVec[i], in_tensor->GetData().size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_FAILED;
}
@@ -250,6 +260,8 @@ Status GraphExecutor::SyncExecuteModel(uint32_t model_id, const std::vector<GeTe
}

if (graph_run_listener_->ResetResult() != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call graph_run_listener_.ResetResult fail, model_id:%u, when %s",
model_id, __FUNCTION__);
GELOGE(GE_GRAPH_EXECUTE_FAILED, "Reset result failed");
return GE_GRAPH_EXECUTE_FAILED;
}
@@ -273,6 +285,8 @@ Status GraphExecutor::SyncExecuteModel(uint32_t model_id, const std::vector<GeTe
// Run graph return
uint32_t result_code = graph_run_listener_->GetResultCode();
if (result_code != SUCCESS && result_code != END_OF_SEQUENCE) {
REPORT_CALL_ERROR("E19999", "Graph_run_listener_ run fail, result:%u, model_id:%u, when %s",
result_code, model_id, __FUNCTION__);
GELOGE(GE_GRAPH_EXECUTE_FAILED, "[GraphExecutor] execute model failed, ret=%u, modelId=%u.", result_code,
model_id);
return GE_GRAPH_EXECUTE_FAILED;
@@ -281,10 +295,14 @@ Status GraphExecutor::SyncExecuteModel(uint32_t model_id, const std::vector<GeTe
for (size_t i = 0; i < output_data.blobs.size(); ++i) {
DataBuffer outputDataTmp = output_data.blobs[i];
CHECK_FALSE_EXEC(outputDataTmp.length != 0,
REPORT_INNER_ERROR("E19999", "Param output_data.length is 0 in model:%u, check invalid, when %s",
model_id, __FUNCTION__);
GELOGE(GE_GRAPH_EXECUTE_FAILED, "Failed to allocate memory, length is 0.");
return GE_GRAPH_EXECUTE_FAILED);
std::unique_ptr<uint8_t> outBufTmp(new (std::nothrow) uint8_t[outputDataTmp.length]);
if (outBufTmp == nullptr) {
REPORT_INNER_ERROR("E19999", "New output buffer fail, length:%lu, model:%u, when %s",
outputDataTmp.length, model_id, __FUNCTION__);
GELOGE(FAILED, "Failed to allocate memory.");
return FAILED;
}
@@ -292,6 +310,8 @@ Status GraphExecutor::SyncExecuteModel(uint32_t model_id, const std::vector<GeTe
rtError_t ret_value = rtMemcpy(outBufTmp.get(), outputDataTmp.length, outputDataTmp.data, outputDataTmp.length,
RT_MEMCPY_HOST_TO_HOST);
CHECK_FALSE_EXEC(ret_value == RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail, dst_size:%lu, src_size:%zu, ret:0x%X when %s",
outputDataTmp.length, outputDataTmp.length, ret_value, __FUNCTION__);
GELOGE(GE_GRAPH_EXECUTE_FAILED, "Call rt api rtMemcpy failed, ret: 0x%X", ret);
return GE_GRAPH_EXECUTE_FAILED);
GeTensor outTensor;
@@ -344,6 +364,8 @@ Status GraphExecutor::ExecuteGraph(GraphId graph_id, const GeRootModelPtr &ge_ro
last_graph_id_ = graph_id;

if (!init_flag_) {
REPORT_INNER_ERROR("E19999", "No SetCondition called before, graph:%u, check invalid when %s",
graph_id, __FUNCTION__);
GELOGE(GE_GRAPH_EXECUTE_NOT_INIT, "[GraphExecutor] AI Core Engine without calling SetCondition!");
return GE_GRAPH_EXECUTE_NOT_INIT;
}
@@ -392,10 +414,12 @@ Status GraphExecutor::AsyncExecuteModel(uint32_t model_id, const std::vector<Inp

GELOGI("RunAsync success.");
} catch (std::bad_alloc &) {
REPORT_INNER_ERROR("E19999", "Bad memory allocation exception occur when %s failed", __FUNCTION__);
GELOGE(MEMALLOC_FAILED, "RunAsync failed, bad memory allocation occur !");
CsaInteract::GetInstance().WriteErrorCode(FAILED, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC);
return MEMALLOC_FAILED;
} catch (...) {
REPORT_INNER_ERROR("E19999", "Some exceptions occur when %s failed", __FUNCTION__);
GELOGE(FAILED, "RunAsync failed, some exceptions occur !");
CsaInteract::GetInstance().WriteErrorCode(FAILED, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC);
return FAILED;
@@ -415,10 +439,12 @@ Status GraphExecutor::DataInput(const InputData &input_data, OutputData &output_
return ret;
}
} catch (std::bad_alloc &) {
REPORT_INNER_ERROR("E19999", "Bad memory allocation exception occur when %s failed", __FUNCTION__);
GELOGE(MEMALLOC_FAILED, "DataInput failed, bad memory allocation occur !");
CsaInteract::GetInstance().WriteErrorCode(FAILED, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC);
return MEMALLOC_FAILED;
} catch (...) {
REPORT_INNER_ERROR("E19999", "Some exceptions occur when %s failed", __FUNCTION__);
GELOGE(FAILED, "DataInput failed, some exceptions occur !");
CsaInteract::GetInstance().WriteErrorCode(FAILED, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC);
return FAILED;
@@ -439,10 +465,12 @@ Status GraphExecutor::GetInputOutputDescInfo(const uint32_t model_id, vector<Inp
return ret;
}
} catch (std::bad_alloc &) {
REPORT_INNER_ERROR("E19999", "Bad memory allocation exception occur when %s failed", __FUNCTION__);
GELOGE(MEMALLOC_FAILED, "GetInputOutputDescInfo failed, bad memory allocation occur !");
CsaInteract::GetInstance().WriteErrorCode(FAILED, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC);
return MEMALLOC_FAILED;
} catch (...) {
REPORT_INNER_ERROR("E19999", "Some exceptions occur when %s failed", __FUNCTION__);
GELOGE(FAILED, "GetInputOutputDescInfo failed, some exceptions occur !");
CsaInteract::GetInstance().WriteErrorCode(FAILED, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC);
return FAILED;
@@ -466,10 +494,12 @@ Status GraphExecutor::GetInputOutputDescInfo(const uint32_t model_id, vector<Inp
return ret;
}
} catch (std::bad_alloc &) {
REPORT_INNER_ERROR("E19999", "Bad memory allocation exception occur when %s failed", __FUNCTION__);
GELOGE(MEMALLOC_FAILED, "GetInputOutputDescInfo failed, bad memory allocation occur !");
CsaInteract::GetInstance().WriteErrorCode(FAILED, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC);
return MEMALLOC_FAILED;
} catch (...) {
REPORT_INNER_ERROR("E19999", "Some exceptions occur when %s failed", __FUNCTION__);
GELOGE(FAILED, "GetInputOutputDescInfo failed, some exceptions occur !");
CsaInteract::GetInstance().WriteErrorCode(FAILED, ERROR_MODULE_FMK, JOBSUBSTATE_GRAPH_EXEC);
return FAILED;


+ 16
- 0
ge/graph/label/case_label_maker.cc View File

@@ -42,6 +42,8 @@ Status CaseOpLabelMaker::Run(uint32_t &label_index) {

const auto graph_names = case_desc->GetSubgraphInstanceNames();
if (graph_names.empty() || graph_names.size() > kMaxCaseBranch) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s) subgraph size: %zu, check invalid when %s", case_desc->GetName().c_str(),
case_desc->GetType().c_str(), graph_names.size(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Node: %s has invalid subgraph, graph size: %zu.", case_desc->GetName().c_str(),
graph_names.size());
return FAILED;
@@ -67,6 +69,8 @@ Status CaseOpLabelMaker::Run(uint32_t &label_index) {
parent_node_->GetName() + "/StreamActive_" + std::to_string(index); // rtStreamActive
NodePtr stream_active = AddStreamActive(graph, stream_active_name);
if (stream_active == nullptr) {
REPORT_CALL_ERROR("E19999", "Add StreamActive node in graph:%s fail when %s",
graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add stream active failed.", graph->GetName().c_str());
return FAILED;
}
@@ -75,6 +79,8 @@ Status CaseOpLabelMaker::Run(uint32_t &label_index) {
std::string label_set_name = parent_node_->GetName() + "/LabelSet_" + std::to_string(index); // rtLabelSet
NodePtr label = AddLabelSetEnter(graph, label_set_name, curr_label_index, stream_active);
if (label == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSetEnter node in graph:%s fail when %s",
graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label set failed.", graph->GetName().c_str());
return FAILED;
}
@@ -88,6 +94,8 @@ Status CaseOpLabelMaker::Run(uint32_t &label_index) {
// middle node, add goto node to tail.
std::string label_goto_name = parent_node_->GetName() + "/LabelGoto_" + std::to_string(index); // rtLabelGoto
if (AddLabelGotoLeave(graph, label_goto_name, last_label_index) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelGotoLeave node in graph:%s fail when %s",
graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label goto failed.", graph->GetName().c_str());
return FAILED;
}
@@ -95,6 +103,8 @@ Status CaseOpLabelMaker::Run(uint32_t &label_index) {
// last node, add label node to tail.
std::string last_label_name = parent_node_->GetName() + "/LabelSet_Last"; // rtLabelSet
if (AddLabelSetLeave(graph, last_label_name, last_label_index) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSetLeave node in graph:%s fail when %s",
graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label set failed.", graph->GetName().c_str());
return FAILED;
}
@@ -110,12 +120,16 @@ Status CaseOpLabelMaker::Run(uint32_t &label_index) {
const GeTensorDesc &pred_desc = case_desc->GetInputDesc(kCasePredIndex);
NodePtr switch_node = AddLabelSwitchEnter(first_graph, label_switch_name, pred_desc, switch_labels);
if (switch_node == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSwitchEnter node in graph:%s fail when %s",
first_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label switch failed.", first_graph->GetName().c_str());
return FAILED;
}

// Link control edge to then branch head.
if (GraphUtils::AddEdge(switch_node->GetOutControlAnchor(), first_label->GetInControlAnchor()) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add ctrl edge from %s to %s in graph:%s fail when %s", switch_node->GetName().c_str(),
first_label->GetName().c_str(), first_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add ctrl edge to %s failed.", first_label->GetName().c_str());
return FAILED;
}
@@ -123,6 +137,8 @@ Status CaseOpLabelMaker::Run(uint32_t &label_index) {
uint32_t parent_index = 0; // Case cond input is first.
const std::string data_name = parent_node_->GetName() + "/SwitchIndexData";
if (AddLabelSwitchIndex(first_graph, data_name, pred_desc, switch_node, parent_index) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSwitchIndex node in graph:%s fail when %s",
first_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add switch input failed.", first_graph->GetName().c_str());
return FAILED;
}


+ 22
- 0
ge/graph/label/if_label_maker.cc View File

@@ -43,6 +43,10 @@ Status IfOpLabelMaker::Run(uint32_t &label_index) {
const std::string then_branch_name = if_desc->GetSubgraphInstanceName(kThenBranchIndex);
const std::string else_branch_name = if_desc->GetSubgraphInstanceName(kElseBranchIndex);
if (then_branch_name.empty() || else_branch_name.empty()) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s), check subgraph invalid, "
"then branch graph: %s, else branch graph: %s, when %s",
if_desc->GetName().c_str(), if_desc->GetType().c_str(),
then_branch_name.c_str(), else_branch_name.c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Node: %s has invalid subgraph, then branch: %s, else branch: %s.",
if_desc->GetName().c_str(), then_branch_name.c_str(), else_branch_name.c_str());
return FAILED;
@@ -66,32 +70,44 @@ Status IfOpLabelMaker::Run(uint32_t &label_index) {

NodePtr then_stream_active = AddStreamActive(then_sub_graph, then_active_name);
if (then_stream_active == nullptr) {
REPORT_CALL_ERROR("E19999", "Add StreamActive node in graph:%s fail when %s",
then_sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add stream active failed.", then_sub_graph->GetName().c_str());
return FAILED;
}

NodePtr then_enter_label = AddLabelSetEnter(then_sub_graph, then_label_name, then_enter_index, then_stream_active);
if (then_enter_label == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSetEnter node in graph:%s fail when %s",
then_sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label set failed.", then_sub_graph->GetName().c_str());
return FAILED;
}

if (AddLabelGotoLeave(then_sub_graph, then_leave_name, else_leave_index) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelGotoLeave node in graph:%s fail when %s",
then_sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label goto failed.", then_sub_graph->GetName().c_str());
return FAILED;
}

NodePtr else_stream_active = AddStreamActive(else_sub_graph, else_active_name);
if (else_stream_active == nullptr) {
REPORT_CALL_ERROR("E19999", "Add StreamActive node in graph:%s fail when %s",
else_stream_active->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add stream active failed.", else_sub_graph->GetName().c_str());
return FAILED;
}

if (AddLabelSetEnter(else_sub_graph, else_enter_name, else_enter_index, else_stream_active) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSetEnter node in graph:%s fail when %s",
else_sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label set failed.", else_sub_graph->GetName().c_str());
return FAILED;
}
if (AddLabelSetLeave(else_sub_graph, else_leave_name, else_leave_index) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSetLeave node in graph:%s fail when %s",
else_sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label set failed.", else_sub_graph->GetName().c_str());
return FAILED;
}
@@ -103,12 +119,16 @@ Status IfOpLabelMaker::Run(uint32_t &label_index) {
const GeTensorDesc &pred_desc = if_desc->GetInputDesc(kIfPredIndex);
NodePtr switch_node = AddLabelSwitchEnter(then_sub_graph, then_enter_name, pred_desc, switch_labels);
if (switch_node == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSwitchEnter node in graph:%s fail when %s",
then_sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label switch failed.", then_sub_graph->GetName().c_str());
return FAILED;
}

// Link control edge to then branch head.
if (GraphUtils::AddEdge(switch_node->GetOutControlAnchor(), then_enter_label->GetInControlAnchor()) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add ctrl edge from %s to %s in graph:%s fail when %s", switch_node->GetName().c_str(),
then_enter_label->GetName().c_str(), then_sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add ctrl edge to %s failed.", then_enter_label->GetName().c_str());
return FAILED;
}
@@ -116,6 +136,8 @@ Status IfOpLabelMaker::Run(uint32_t &label_index) {
uint32_t parent_index = 0; // If cond input is first.
const std::string data_name = parent_node_->GetName() + "/SwitchIndexData";
if (AddLabelSwitchIndex(then_sub_graph, data_name, pred_desc, switch_node, parent_index) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSwitchIndex node in graph:%s fail when %s",
then_sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add switch input failed.", then_sub_graph->GetName().c_str());
return FAILED;
}


+ 30
- 0
ge/graph/label/label_maker.cc View File

@@ -56,6 +56,8 @@ void LabelMaker::LinkToGraphHead(const ComputeGraphPtr &graph, const NodePtr &no
}

if (GraphUtils::AddEdge(node->GetOutControlAnchor(), n->GetInControlAnchor()) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add ctrl edge from %s to %s in graph:%s fail when %s", node->GetName().c_str(),
n->GetName().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Add ctrl edge from %s to %s failed.", node->GetName().c_str(), n->GetName().c_str());
}
}
@@ -78,6 +80,8 @@ void LabelMaker::LinkToGraphTail(const ComputeGraphPtr &graph, const NodePtr &no
}

if (GraphUtils::AddEdge(tail->GetOutControlAnchor(), node->GetInControlAnchor()) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add ctrl edge from %s to %s in graph:%s fail when %s", tail->GetName().c_str(),
node->GetName().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Add ctrl edge from %s to %s failed.", tail->GetName().c_str(), node->GetName().c_str());
}
return;
@@ -96,6 +100,7 @@ NodePtr LabelMaker::AddStreamActive(const ComputeGraphPtr &graph, const std::str

const auto &node_list = graph->GetDirectNode();
if (node_list.empty()) {
REPORT_INNER_ERROR("E19999", "Check param graph has no node when %s", graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSet: Graph %s node is empty.", graph->GetName().c_str());
return nullptr;
}
@@ -131,6 +136,7 @@ NodePtr LabelMaker::AddLabelSetEnter(const ComputeGraphPtr &graph, const std::st

const auto &node_list = graph->GetDirectNode();
if (node_list.empty()) {
REPORT_INNER_ERROR("E19999", "Check param graph has no node when %s", graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSet: Graph %s node is empty.", graph->GetName().c_str());
return nullptr;
}
@@ -145,6 +151,8 @@ NodePtr LabelMaker::AddLabelSetEnter(const ComputeGraphPtr &graph, const std::st
GE_CHECK_NOTNULL_EXEC(label_set, return nullptr);

if (GraphUtils::AddEdge(label_set->GetOutControlAnchor(), stream_active->GetInControlAnchor()) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add ctrl edge from %s to %s in graph:%s fail when %s", label_set->GetName().c_str(),
stream_active->GetName().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Add ctrl edge from %s to %s failed.", label_set->GetName().c_str(),
stream_active->GetName().c_str());
return nullptr;
@@ -193,6 +201,7 @@ NodePtr LabelMaker::AddLabelGotoEnter(const ComputeGraphPtr &graph, const std::s
const auto &node_list = graph->GetDirectNode();
auto it = node_list.begin();
if (it == node_list.end()) {
REPORT_INNER_ERROR("E19999", "Check param graph has no node when %s", graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelGoto: Graph %s node is empty.", graph->GetName().c_str());
return nullptr;
}
@@ -205,6 +214,8 @@ NodePtr LabelMaker::AddLabelGotoEnter(const ComputeGraphPtr &graph, const std::s
(void)AttrUtils::SetInt(op_desc, ATTR_NAME_LABEL_SWITCH_INDEX, index);
NodePtr label_goto = graph->AddNodeFront(op_desc);
if (label_goto == nullptr) {
REPORT_CALL_ERROR("E19999", "Add node:%s(%s) to graph:%s fail when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelGoto: Add to graph %s failed.", graph->GetName().c_str());
return nullptr;
}
@@ -253,6 +264,7 @@ NodePtr LabelMaker::AddLabelSwitchEnter(const ComputeGraphPtr &graph, const std:
const auto &node_list = graph->GetDirectNode();
auto it = node_list.begin();
if (it == node_list.end()) {
REPORT_INNER_ERROR("E19999", "Check param graph has no node when %s", graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Graph %s node is empty.", graph->GetName().c_str());
return nullptr;
}
@@ -263,17 +275,23 @@ NodePtr LabelMaker::AddLabelSwitchEnter(const ComputeGraphPtr &graph, const std:

GELOGI("LabelSwitchByIndex: Create node %s.", op_desc->GetName().c_str());
if (op_desc->AddInputDesc(desc) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add input desc into node:%s(%s) in graph:%s fail when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add input desc failed.");
return nullptr;
}

if (!AttrUtils::SetListInt(op_desc, ATTR_NAME_LABEL_SWITCH_LIST, labels)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_LABEL_SWITCH_LIST.c_str(),
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add %s failed.", ATTR_NAME_LABEL_SWITCH_INDEX.c_str());
return nullptr;
}

NodePtr label_switch = graph->AddNodeFront(op_desc);
if (label_switch == nullptr) {
REPORT_CALL_ERROR("E19999", "Add node:%s(%s) to graph:%s ahead fail when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add to graph %s failed.", graph->GetName().c_str());
return nullptr;
}
@@ -300,11 +318,15 @@ NodePtr LabelMaker::AddLabelSwitchLeave(const ComputeGraphPtr &graph, const std:

GELOGI("LabelSwitchByIndex: Create node %s.", op_desc->GetName().c_str());
if (op_desc->AddInputDesc(desc) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add input desc into node:%s(%s) in graph:%s fail when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add input desc failed.");
return nullptr;
}

if (!AttrUtils::SetListInt(op_desc, ATTR_NAME_LABEL_SWITCH_LIST, labels)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_LABEL_SWITCH_LIST.c_str(),
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add %s failed.", ATTR_NAME_LABEL_SWITCH_INDEX.c_str());
return nullptr;
}
@@ -336,15 +358,21 @@ NodePtr LabelMaker::AddLabelSwitchIndex(const ComputeGraphPtr &graph, const std:

GELOGI("Data: Create node %s.", op_desc->GetName().c_str());
if (op_desc->AddInputDesc(desc) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add input desc into node:%s(%s) in graph:%s fail when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add data input desc failed.");
return nullptr;
}
if (op_desc->AddOutputDesc(desc) != GRAPH_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add output desc into node:%s(%s) in graph:%s fail when %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add data output desc failed.");
return nullptr;
}

if (!AttrUtils::SetInt(op_desc, ATTR_NAME_PARENT_NODE_INDEX, parent_index)) {
REPORT_INNER_ERROR("E19999", "Set Attr:%s fail for op:%s(%s) when %s", ATTR_NAME_PARENT_NODE_INDEX.c_str(),
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add %s failed.", ATTR_NAME_PARENT_NODE_INDEX.c_str());
return nullptr;
}
@@ -354,6 +382,8 @@ NodePtr LabelMaker::AddLabelSwitchIndex(const ComputeGraphPtr &graph, const std:

// Link control edge to graph head.
if (GraphUtils::AddEdge(op_data->GetOutDataAnchor(0), sw_node->GetInDataAnchor(0)) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add ctrl edge from %s to %s in graph:%s fail when %s", op_data->GetName().c_str(),
sw_node->GetName().c_str(), graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "LabelSwitchByIndex: Add input edge to %s failed.", op_data->GetName().c_str());
return nullptr;
}


+ 7
- 0
ge/graph/label/partitioned_call_label_maker.cc View File

@@ -39,12 +39,17 @@ Status PartitionedCallLabelMaker::Run(uint32_t &label_index) {

std::string sub_graph_name = call_desc->GetSubgraphInstanceName(kSubGraphIndex);
if (sub_graph_name.empty()) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s) subgraph_index:%d name is empty, check invalid when %s",
call_desc->GetName().c_str(), call_desc->GetType().c_str(), kSubGraphIndex, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Node: %s has no subgraph name.", sub_graph_name.c_str());
return FAILED;
}

ComputeGraphPtr sub_graph = parent_graph_->GetSubgraph(sub_graph_name);
if (sub_graph == nullptr) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s) subgraph_name:%s is not exist in parent_graph, check invalid when %s",
call_desc->GetName().c_str(), call_desc->GetType().c_str(),
sub_graph_name.c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Node: %s has no subgraph.", sub_graph_name.c_str());
return FAILED;
}
@@ -52,6 +57,8 @@ Status PartitionedCallLabelMaker::Run(uint32_t &label_index) {
const std::string stream_active_name = parent_node_->GetName() + "/StreamActive"; // rtStreamActive
NodePtr stream_active = AddStreamActive(sub_graph, stream_active_name);
if (stream_active == nullptr) {
REPORT_CALL_ERROR("E19999", "Add StreamActive node in graph:%s fail when %s",
sub_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add stream active node failed.", sub_graph->GetName().c_str());
return FAILED;
}


+ 20
- 0
ge/graph/label/while_label_maker.cc View File

@@ -44,6 +44,9 @@ Status WhileOpLabelMaker::Run(uint32_t &label_index) {
std::string cond_name = while_desc->GetSubgraphInstanceName(kCondBranchIndex);
std::string body_name = while_desc->GetSubgraphInstanceName(kBodyBranchIndex);
if (cond_name.empty() || body_name.empty()) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s) cond subgraph index:%d or body subgraph index:%d name is empty, "
"check invalid when %s", while_desc->GetName().c_str(), while_desc->GetType().c_str(),
kCondBranchIndex, kBodyBranchIndex, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Node: %s has invalid subgraph, cond branch: %s, body branch: %s.",
while_desc->GetName().c_str(), cond_name.c_str(), body_name.c_str());
return FAILED;
@@ -67,32 +70,44 @@ Status WhileOpLabelMaker::Run(uint32_t &label_index) {

NodePtr cond_stream_active = AddStreamActive(cond_graph, cond_active_name);
if (cond_stream_active == nullptr) {
REPORT_CALL_ERROR("E19999", "Add StreamActive node in graph:%s fail when %s",
cond_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add stream active failed.", cond_graph->GetName().c_str());
return FAILED;
}

if (AddLabelSetEnter(cond_graph, cond_enter_name, cond_enter_index, cond_stream_active) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSetEnter node in graph:%s fail when %s",
cond_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label set failed.", cond_graph->GetName().c_str());
return FAILED;
}

NodePtr body_stream_active = AddStreamActive(body_graph, body_active_name);
if (body_stream_active == nullptr) {
REPORT_CALL_ERROR("E19999", "Add StreamActive node in graph:%s fail when %s",
body_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add stream active failed.", body_graph->GetName().c_str());
return FAILED;
}

if (AddLabelSetEnter(body_graph, body_enter_name, body_enter_index, body_stream_active) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSetEnter node in graph:%s fail when %s",
body_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label set failed.", body_graph->GetName().c_str());
return FAILED;
}

if (AddLabelGotoLeave(body_graph, goto_leave_name, cond_enter_index) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelGotoLeave node in graph:%s fail when %s",
body_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label goto failed.", body_graph->GetName().c_str());
return FAILED;
}

if (AddLabelSetLeave(body_graph, body_leave_name, body_leave_index) == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSetLeave node in graph:%s fail when %s",
body_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label set failed.", body_graph->GetName().c_str());
return FAILED;
}
@@ -109,6 +124,8 @@ Status WhileOpLabelMaker::Run(uint32_t &label_index) {
const std::vector<uint32_t> switch_labels = {body_leave_index, body_enter_index};
NodePtr switch_node = AddLabelSwitchLeave(cond_graph, cond_leave_name, pred_desc, switch_labels);
if (switch_node == nullptr) {
REPORT_CALL_ERROR("E19999", "Add LabelSwitchLeave node in graph:%s fail when %s",
cond_graph->GetName().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Subgraph: %s add label switch failed.", cond_graph->GetName().c_str());
return FAILED;
}
@@ -124,6 +141,9 @@ Status WhileOpLabelMaker::Run(uint32_t &label_index) {
InDataAnchorPtr in_anchor = all_in_data.at(kCondOutputIndex);
GE_CHECK_NOTNULL(in_anchor);
if (GraphUtils::AddEdge(in_anchor->GetPeerOutAnchor(), switch_node->GetInDataAnchor(kCondOutputIndex)) != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Add ctrl edge from %s to %s in graph:%s fail when %s",
in_anchor->GetPeerOutAnchor()->GetOwnerNode()->GetName().c_str(),
switch_node->GetName().c_str(), cond_graph ->GetName().c_str(), __FUNCTION__);
GELOGE(FAILED, "Node: %s Add pred data input failed.", switch_node->GetName().c_str());
return FAILED;
}


+ 4
- 0
ge/graph/load/model_manager/task_info/end_graph_task_info.cc View File

@@ -27,6 +27,7 @@ namespace ge {
Status EndGraphTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci_model) {
GELOGI("InitEndGraphTaskInfo Init Start.");
if (davinci_model == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when EndGraphTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return PARAM_INVALID;
}
@@ -52,6 +53,7 @@ Status EndGraphTaskInfo::Distribute() {
GELOGI("Start to call rtEndGraphEx");
rtError_t rt_ret = rtEndGraphEx(model_, stream_, kDumpFlag);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtEndGraphEx fail ret:0x%X, when EndGraphTaskInfo %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rtEndGraphEx failed, ret: 0x%x", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -59,6 +61,7 @@ Status EndGraphTaskInfo::Distribute() {
GELOGI("Start to call rtEndGraph");
rtError_t rt_ret = rtEndGraph(model_, stream_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtEndGraph fail ret:0x%X, when EndGraphTaskInfo %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rtEndGraph failed, ret: 0x%x", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -68,6 +71,7 @@ Status EndGraphTaskInfo::Distribute() {
uint32_t stream_id = 0;
rtError_t rt_ret = rtModelGetTaskId(davinci_model_->GetRtModelHandle(), &task_id, &stream_id);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtModelGetTaskId fail ret:0x%X, when EndGraphTaskInfo %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}


+ 4
- 0
ge/graph/load/model_manager/task_info/event_record_task_info.cc View File

@@ -23,6 +23,7 @@ namespace ge {
Status EventRecordTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci_model) {
GELOGI("EventRecordTaskInfo Init Start.");
if (davinci_model == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when EventRecordTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return PARAM_INVALID;
}
@@ -34,6 +35,8 @@ Status EventRecordTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *da

const auto &eventList = davinci_model->GetEventList();
if (task_def.event_id() >= eventList.size()) {
REPORT_INNER_ERROR("E19999", "Task event_id:%u > model event size:%zu, check invalid when EventRecordTaskInfo %s",
task_def.event_id(), eventList.size(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "event list size:%zu, cur:%u!", eventList.size(), task_def.event_id());
return INTERNAL_ERROR;
}
@@ -47,6 +50,7 @@ Status EventRecordTaskInfo::Distribute() {
GELOGI("EventRecordTaskInfo Distribute Start.");
rtError_t rt_ret = rtEventRecord(event_, stream_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtEventRecord fail ret:0x%X, when EventRecordTaskInfo %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}


+ 7
- 0
ge/graph/load/model_manager/task_info/event_wait_task_info.cc View File

@@ -23,6 +23,7 @@ namespace ge {
Status EventWaitTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci_model) {
GELOGI("EventWaitTaskInfo Init Start.");
if (davinci_model == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when EventWaitTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return PARAM_INVALID;
}
@@ -34,6 +35,8 @@ Status EventWaitTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davi

const auto &eventList = davinci_model->GetEventList();
if (task_def.event_id() >= eventList.size()) {
REPORT_INNER_ERROR("E19999", "Task event_id:%u > model event size:%zu, check invalid when EventWaitTaskInfo %s",
task_def.event_id(), eventList.size(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "event list size:%zu, cur:%u!", eventList.size(), task_def.event_id());
return INTERNAL_ERROR;
}
@@ -48,12 +51,16 @@ Status EventWaitTaskInfo::Distribute() {
GELOGI("EventWaitTaskInfo Distribute Start.");
rtError_t rt_ret = rtStreamWaitEvent(stream_, event_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtStreamWaitEvent fail ret:0x%X, when EventWaitTaskInfo %s",
rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}

rt_ret = rtEventReset(event_, stream_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtEventReset fail ret:0x%X, when EventWaitTaskInfo %s",
rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}


+ 3
- 0
ge/graph/load/model_manager/task_info/fusion_start_task_info.cc View File

@@ -23,6 +23,7 @@ namespace ge {
Status FusionStartTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci_model) {
GELOGI("FusionStartTaskInfo Init Start.");
if (davinci_model == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when FusionStartTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return PARAM_INVALID;
}
@@ -39,6 +40,8 @@ Status FusionStartTaskInfo::Distribute() {
GELOGI("FusionStartTaskInfo Distribute Start.");
rtError_t rt_ret = rtKernelFusionStart(stream_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtKernelFusionStart fail ret:0x%X, when FusionStartTaskInfo %s",
rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}


+ 3
- 0
ge/graph/load/model_manager/task_info/fusion_stop_task_info.cc View File

@@ -23,6 +23,7 @@ namespace ge {
Status FusionStopTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci_model) {
GELOGI("FusionStopTaskInfo Init Start.");
if (davinci_model == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when FusionStopTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return PARAM_INVALID;
}
@@ -39,6 +40,8 @@ Status FusionStopTaskInfo::Distribute() {
GELOGI("FusionStopTaskInfo Distribute Start.");
rtError_t rt_ret = rtKernelFusionEnd(stream_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtKernelFusionEnd fail ret:0x%X, when FusionStopTaskInfo %s",
rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}


+ 20
- 0
ge/graph/load/model_manager/task_info/hccl_task_info.cc View File

@@ -30,6 +30,7 @@ HcclTaskInfo::~HcclTaskInfo() {
if (private_def_ != nullptr) {
rtError_t ret = rtFreeHost(private_def_);
if (ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtFreeHost fail ret:0x%X, when HcclTaskInfo %s", ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rtFree Fail, ret = 0x%X.", ret);
}
private_def_ = nullptr;
@@ -41,6 +42,7 @@ HcclTaskInfo::~HcclTaskInfo() {
Status HcclTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci_model) {
GELOGI("HcclTaskInfo Init Start.");
if (davinci_model == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when HcclTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return PARAM_INVALID;
}
@@ -67,22 +69,30 @@ Status HcclTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci_m
// Only in Horovod scenario should get the inputName and GeShape
ret = HcomOmeUtil::GetHorovodInputs(op_desc, kernel_hccl_infos_);
if (ret != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call GetHorovodInputs fail for op:%s(%s), when HcclTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(ret, "davinci_model: GetHorovodInputs fail! domi error: %u", ret);
return ret;
}
Status dmrt = HcomOmeUtil::GetHcclDataType(op_desc, kernel_hccl_infos_);
if (dmrt != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call GetHcclDataType fail for op:%s(%s), when HcclTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(dmrt, "davinci_model: GetHcomDataType fail! domi error: %u", dmrt);
return dmrt;
}
dmrt = HcomOmeUtil::GetHcclCount(op_desc, kernel_hccl_infos_);
if (dmrt != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call GetHcclCount fail for op:%s(%s), when HcclTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(dmrt, "davinci_model: GetHcomCount fail! domi error: %u", dmrt);
return dmrt;
}
// Only HCOMBROADCAST and HVDCALLBACKBROADCAST need to get the rootId
dmrt = HcomOmeUtil::GetAllRootId(op_desc, kernel_hccl_infos_);
if (dmrt != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call GetAllRootId fail for op:%s(%s), when HcclTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(dmrt, "davinci_model: Get rootId fail! domi error: %u", dmrt);
return dmrt;
}
@@ -169,12 +179,16 @@ Status HcclTaskInfo::CreateStream(int64_t stream_num, DavinciModel *davinci_mode
rtError_t rt_ret =
rtStreamCreateWithFlags(&stream, davinci_model->Priority(), RT_STREAM_PERSISTENT | RT_STREAM_FORCE_COPY);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtStreamCreateWithFlags fail ret:0x%X, stream_idx:%ld, stream_num:%ld, "
"when HcclTaskInfo %s", rt_ret, i, stream_num, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
// Create slave stream, inactive by default, activated by hccl
rt_ret = rtModelBindStream(davinci_model->GetRtModelHandle(), stream, RT_MODEL_WAIT_ACTIVE_STREAM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtModelBindStream fail ret:0x%X, stream_idx:%ld, stream_num:%ld, "
"when HcclTaskInfo %s", rt_ret, i, stream_num, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
(void)rtStreamDestroy(stream);
return RT_ERROR_TO_GE_STATUS(rt_ret);
@@ -192,6 +206,7 @@ Status HcclTaskInfo::CreateStream(int64_t stream_num, DavinciModel *davinci_mode
Status HcclTaskInfo::Distribute() {
GELOGI("HcclTaskInfo Distribute Start. begin to call function LoadTask in hccl.");
if (ops_kernel_store_ == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param ops_kernel_store_ nullptr when HcclTaskInfo %s", __FUNCTION__);
GELOGE(INTERNAL_ERROR, "ops kernel store is null.");
return INTERNAL_ERROR;
}
@@ -201,6 +216,7 @@ Status HcclTaskInfo::Distribute() {
TransToGETaskInfo(ge_task);
auto result = ops_kernel_info_store->LoadTask(ge_task);
if (result != HCCL_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call ops_kernel_info_store LoadTask fail when HcclTaskInfo %s", __FUNCTION__);
GELOGE(INTERNAL_ERROR, "davinci_model : load task fail, return ret: %u", result);
return INTERNAL_ERROR;
}
@@ -316,6 +332,8 @@ void HcclTaskInfo::GetPrivateDefByTaskDef(const domi::TaskDef &task) {
private_def_len_ = private_def_temp.size();
rtError_t ret = rtMallocHost(&private_def_, private_def_len_);
if (ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMallocHost fail ret:0x%X, size:%u, when HcclTaskInfo %s",
ret, private_def_len_, __FUNCTION__);
GELOGE(RT_FAILED, "Call rtMallocHost Fail, ret = 0x%X.", ret);
return;
}
@@ -323,6 +341,8 @@ void HcclTaskInfo::GetPrivateDefByTaskDef(const domi::TaskDef &task) {
ret = rtMemcpy(private_def_, private_def_len_, task.private_def().c_str(), private_def_len_,
RT_MEMCPY_HOST_TO_HOST);
if (ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail ret:0x%X, size:%u, when HcclTaskInfo %s",
ret, private_def_len_, __FUNCTION__);
GELOGE(RT_FAILED, "Call rtMemcpy Fail, ret = 0x%X.", ret);
return;
}


+ 67
- 7
ge/graph/load/model_manager/task_info/kernel_ex_task_info.cc View File

@@ -75,11 +75,15 @@ Status KernelExTaskInfo::InitTaskExtInfo(const std::string &ext_info, const OpDe
}
auto rt_ret = rtMalloc(&ext_info_addr_, ext_handle->GetExtInfoLen(), RT_MEMORY_HBM);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail, size:%zu, ret:0x%X, when KernelExTaskInfo %s",
ext_info.size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMalloc ext_info error: 0x%X, size=%zu", rt_ret, ext_info.size());
return RT_ERROR_TO_GE_STATUS(rt_ret);)
rt_ret = rtMemcpy(ext_info_addr_, ext_handle->GetExtInfoLen(), ext_handle->GetExtInfo(),
ext_handle->GetExtInfoLen(), RT_MEMCPY_HOST_TO_DEVICE);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail, size:%zu, ret:0x%X, when KernelExTaskInfo %s",
ext_handle->GetExtInfoLen(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMemcpy ext_info error: 0x%X, size=%zu", rt_ret, ext_info.size());
return RT_ERROR_TO_GE_STATUS(rt_ret);)
return SUCCESS;
@@ -101,6 +105,8 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin
uint32_t op_index = kernel_ex_def.op_index();
OpDescPtr op_desc = davinci_model_->GetOpByIndex(op_index);
if (op_desc == nullptr) {
REPORT_INNER_ERROR("E19999", "Can't get op_desc from davinci_model by index:%u when KernelExTaskInfo %s",
op_index, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Init aicpu task info error, index is out of range!");
return INTERNAL_ERROR;
}
@@ -108,6 +114,9 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin
// 2. Reconstruct kernelExDef.args to STR_FWK_OP_KERNEL
STR_FWK_OP_KERNEL fwk_op_kernel = {0};
if (sizeof(STR_FWK_OP_KERNEL) < kernel_ex_def.args_size()) {
REPORT_INNER_ERROR("E19999", "Param kernel_ex_def.args_size():%u > sizeof(STR_FWK_OP_KERNEL):%zu, "
"check invalid when KernelExTaskInfo %s", kernel_ex_def.args_size(), sizeof(STR_FWK_OP_KERNEL),
__FUNCTION__);
GELOGE(FAILED, "sizeof STR_FWK_OP_KERNEL is: %zu, but args_size is: %u", sizeof(STR_FWK_OP_KERNEL),
kernel_ex_def.args_size());
return FAILED;
@@ -115,6 +124,8 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin
errno_t sec_ret =
memcpy_s(&fwk_op_kernel, sizeof(STR_FWK_OP_KERNEL), kernel_ex_def.args().data(), kernel_ex_def.args_size());
if (sec_ret != EOK) {
REPORT_CALL_ERROR("E19999", "Call memcpy_s fail, size:%zu, ret:0x%X, when KernelExTaskInfo %s",
sizeof(STR_FWK_OP_KERNEL), sec_ret, __FUNCTION__);
GELOGE(FAILED, "memcpy failed, ret: %d", sec_ret);
return FAILED;
}
@@ -136,12 +147,17 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin
uint64_t kernel_id = fwk_op_kernel.fwkKernelBase.fwk_kernel.kernelID;
GE_IF_BOOL_EXEC(ModelManager::GetInstance()->CreateAicpuKernel(session_id, davinci_model->Id(),
davinci_model->SubModelId(), kernel_id) != SUCCESS,
REPORT_CALL_ERROR("E19999", "CreateAicpuKernel fail, session_id:%lu, model_id:%u, kernel_id:%lu "
"when KernelExTaskInfo %s",
session_id, davinci_model->Id(), kernel_id, __FUNCTION__);
GELOGE(FAILED, "CreateAicpuKernel error.");
return FAILED;)
// 2.3 Create session
GE_CHECK_NOTNULL(ModelManager::GetInstance());
ret = ModelManager::GetInstance()->CreateAicpuSession(session_id);
GE_IF_BOOL_EXEC(ret != SUCCESS,
REPORT_CALL_ERROR("E19999", "CreateAicpuSession fail, session_id:%lu when KernelExTaskInfo %s",
session_id, __FUNCTION__);
GELOGE(ret, "CreateAicpuSession error. session id: %lu", session_id);
return ret;)

@@ -152,7 +168,10 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin
static_cast<uint64_t>(reinterpret_cast<uintptr_t>(input_output_addr));
void *workspace_base_addr = nullptr;
rtError_t rt_ret = rtMalloc(&workspace_base_addr, kernel_ex_def.task_info_size(), RT_MEMORY_HBM);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMalloc error, ret: Ox%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail, size:%u, ret:0x%X, when KernelExTaskInfo %s",
kernel_ex_def.task_info_size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMalloc error, ret: Ox%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret););
rt_ret = rtMemcpy(workspace_base_addr, kernel_ex_def.task_info_size(), kernel_ex_def.task_info().data(),
kernel_ex_def.task_info_size(), RT_MEMCPY_HOST_TO_DEVICE);
@@ -163,12 +182,18 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin
fwk_op_kernel.fwkKernelBase.fwk_kernel.extInfoAddr = reinterpret_cast<uintptr_t>(ext_info_addr_);

rt_ret = rtMalloc(&kernel_buf_, kernel_buf_size_, RT_MEMORY_HBM);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMalloc error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail ret:0x%X, size:%u, when KernelExTaskInfo %s",
rt_ret, kernel_buf_size_, __FUNCTION__);
GELOGE(RT_FAILED, "rtMalloc error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)

rt_ret = rtMemcpy(kernel_buf_, kernel_buf_size_, static_cast<void *>(&fwk_op_kernel), kernel_buf_size_,
RT_MEMCPY_HOST_TO_DEVICE);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMemcpy error, ret: Ox%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail ret:0x%X, size:%u, when KernelExTaskInfo %s",
rt_ret, kernel_buf_size_, __FUNCTION__);
GELOGE(RT_FAILED, "rtMemcpy error, ret: Ox%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)

SetIoAddrs(op_desc);
@@ -186,6 +211,8 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin

const vector<void *> workspace_data_addrs = ModelUtils::GetWorkspaceDataAddrs(rts_param, op_desc);
if (workspace_data_addrs.empty()) {
REPORT_CALL_ERROR("E19999", "workspace_data_addrs is empty in op:%s(%s), check invalid when KernelExTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "workspace_data_addrs is empty.");
return FAILED;
}
@@ -200,11 +227,17 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin
auto addrs_size = sizeof(uint64_t) * (io_addrs.size());
if (addrs_size > 0) {
rtError_t rt_ret = rtMalloc(&input_output_addr_, addrs_size, RT_MEMORY_HBM);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMalloc error, ret: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail ret:0x%X, size:%u, when KernelExTaskInfo %s",
rt_ret, addrs_size, __FUNCTION__);
GELOGE(RT_FAILED, "rtMalloc error, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)

rt_ret = rtMemcpy(input_output_addr_, addrs_size, io_addrs.data(), addrs_size, RT_MEMCPY_HOST_TO_DEVICE);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMemcpy to input_output_addr_ error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail ret:0x%X, size:%u, when KernelExTaskInfo %s",
rt_ret, addrs_size, __FUNCTION__);
GELOGE(RT_FAILED, "rtMemcpy to input_output_addr_ error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)

InitDumpTask(input_output_addr_, op_desc);
@@ -223,12 +256,18 @@ Status KernelExTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davin

// 4. Return result
rtError_t rt_ret = rtMalloc(&kernel_buf_, sizeof(STR_FWK_OP_KERNEL), RT_MEMORY_HBM);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMalloc error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail ret:0x%X, size:%zu, when KernelExTaskInfo %s",
rt_ret, sizeof(STR_FWK_OP_KERNEL), __FUNCTION__);
GELOGE(RT_FAILED, "rtMalloc error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)

rt_ret = rtMemcpy(kernel_buf_, sizeof(STR_FWK_OP_KERNEL), static_cast<void *>(&fwk_op_kernel),
sizeof(STR_FWK_OP_KERNEL), RT_MEMCPY_HOST_TO_DEVICE);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMemcpy error, ret: Ox%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail ret:0x%X, size:%zu, when KernelExTaskInfo %s",
rt_ret, sizeof(STR_FWK_OP_KERNEL), __FUNCTION__);
GELOGE(RT_FAILED, "rtMemcpy error, ret: Ox%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)

davinci_model_->SetZeroCopyAddr(op_desc, io_addrs, io_addrs.data(), input_output_addr_, addrs_size, 0);
@@ -250,6 +289,8 @@ Status KernelExTaskInfo::CalculateArgs(const domi::TaskDef &task_def, DavinciMod
uint32_t op_index = kernel_ex_def.op_index();
OpDescPtr op_desc = davinci_model->GetOpByIndex(op_index);
if (op_desc == nullptr) {
REPORT_INNER_ERROR("E19999", "Can't get op_desc from davinci_model by index:%u when KernelExTaskInfo %s",
op_index, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "Init aicpu task info error, index is out of range!");
return INTERNAL_ERROR;
}
@@ -267,6 +308,9 @@ Status KernelExTaskInfo::CalculateArgs(const domi::TaskDef &task_def, DavinciMod
if (AttrUtils::GetStr(op_desc, ATTR_DYNAMIC_SHAPE_FIXED_ADDR, peer_input_name) && !peer_input_name.empty()) {
uint32_t output_index = davinci_model->GetFixedAddrOutputIndex(peer_input_name);
if (output_index > outputs_size) {
REPORT_INNER_ERROR("E19999", "The output size[%zu] and output index[%u] in op:%s(%s) are inconsistent, "
"check invalid when KernelExTaskInfo %s", outputs_size, output_index,
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "The output size[%zu] and output index[%u] are inconsistent.", outputs_size, output_index);
return FAILED;
}
@@ -293,6 +337,9 @@ void KernelExTaskInfo::SetIoAddrs(const OpDescPtr &op_desc) {
if (AttrUtils::GetStr(op_desc, ATTR_DYNAMIC_SHAPE_FIXED_ADDR, peer_input_name)) {
uint32_t output_index = davinci_model_->GetFixedAddrOutputIndex(peer_input_name);
if (output_index > output_data_addrs.size()) {
REPORT_INNER_ERROR("E19999", "The output data addr size[%zu] and output index[%u] in op:%s(%s) are inconsistent"
", check invalid when KernelExTaskInfo %s", output_data_addrs.size(), output_index,
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "The output data addr size[%zu] and output index[%u] are inconsistent.",
output_data_addrs.size(), output_index);
return;
@@ -323,17 +370,25 @@ Status KernelExTaskInfo::CopyTaskInfo(const domi::KernelExDef &kernel_def, const
const vector<int64_t> workspace_data_sizes = ModelUtils::GetWorkspaceSize(op_desc);
const vector<void *> workspace_data_addrs = ModelUtils::GetWorkspaceDataAddrs(rts_param, op_desc);
if (workspace_data_addrs.empty() || workspace_data_sizes.empty()) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s) workspace addr:%zu or size:%zu empty, check invalid "
"when KernelExTaskInfo %s", op_desc->GetName().c_str(), op_desc->GetType().c_str(),
workspace_data_addrs.size(), workspace_data_sizes.size(), __FUNCTION__);
GELOGE(FAILED, "Node:%s invalid workspace, addrs is %zu, size is %zu.", op_desc->GetName().c_str(),
workspace_data_addrs.size(), workspace_data_sizes.size());
return FAILED;
}

if (workspace_data_addrs[0] == nullptr) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s) workspace addr is nullptr, check invalid when KernelExTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Node:%s workspace addrs is null.", op_desc->GetName().c_str());
return FAILED;
}

if (workspace_data_sizes[0] < static_cast<int64_t>(kernel_def.task_info_size())) {
REPORT_INNER_ERROR("E19999", "Node:%s(%s) workspace size:%ld < task info size:%d, check invalid "
"when KernelExTaskInfo %s", op_desc->GetName().c_str(), op_desc->GetType().c_str(),
workspace_data_sizes[0], kernel_def.task_info_size(), __FUNCTION__);
GELOGE(FAILED, "Node:%s workspace size is %ld, task info size is %d.", op_desc->GetName().c_str(),
workspace_data_sizes[0], kernel_def.task_info_size());
return FAILED;
@@ -342,6 +397,8 @@ Status KernelExTaskInfo::CopyTaskInfo(const domi::KernelExDef &kernel_def, const
rtError_t rt_ret = rtMemcpy(workspace_data_addrs[0], kernel_def.task_info_size(), kernel_def.task_info().data(),
kernel_def.task_info_size(), RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail ret:0x%X, size:%d, when KernelExTaskInfo %s",
rt_ret, kernel_def.task_info_size(), __FUNCTION__);
GELOGE(RT_FAILED, "rtMemcpy error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -353,11 +410,13 @@ Status KernelExTaskInfo::Distribute() {
GELOGI("KernelExTaskInfo Distribute Start.");
rtError_t rt_ret = rtKernelLaunchEx(kernel_buf_, kernel_buf_size_, dump_flag_, stream_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtKernelLaunchEx fail ret:0x%X when KernelExTaskInfo %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}

if (davinci_model_ == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when KernelExTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model_ is null.");
return PARAM_INVALID;
}
@@ -366,6 +425,7 @@ Status KernelExTaskInfo::Distribute() {
uint32_t stream_id = 0; // for profiling
rt_ret = rtModelGetTaskId(davinci_model_->GetRtModelHandle(), &task_id, &stream_id);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtModelGetTaskId fail ret:0x%X when KernelExTaskInfo %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}


+ 175
- 2
ge/graph/load/model_manager/task_info/kernel_task_info.cc View File

@@ -93,8 +93,13 @@ Status KernelTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci
// new aicpu kernel(rtCpuKernelLaunch) no need to check function
if (kernel_type_ == ccKernelType::CCE_AI_CORE) {
rtError_t rt_ret = rtGetFunctionByName(const_cast<char *>(kernel_def.stub_func().c_str()), &stub_func_);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "execute rtGetFunctionByName failed. stub_func: %s",
kernel_def.stub_func().c_str());
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtGetFunctionByName fail for op:%s(%s), "
"bin_file_key:%s, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
kernel_def.stub_func().c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "execute rtGetFunctionByName failed. stub_func: %s",
kernel_def.stub_func().c_str());
return RT_ERROR_TO_GE_STATUS(rt_ret););
} else if (kernel_type_ == ccKernelType::TE) {
// get bin_file_key
@@ -103,11 +108,18 @@ Status KernelTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci
const char *bin_file_key = davinci_model_->GetRegisterStub(op_desc_->GetName(), session_graph_model_id);
rtError_t rt_ret = rtGetFunctionByName(bin_file_key, &stub_func_);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtGetFunctionByName fail for op:%s(%s), "
"bin_file_key:%s, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
bin_file_key, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "execute rtGetFunctionByName failed. bin_file_key: %s", bin_file_key);
return RT_ERROR_TO_GE_STATUS(rt_ret););
}

if (context.origin_op_index_size() > CC_FUSION_OP_MAX) {
REPORT_INNER_ERROR("E19999", "context.origin_op_index_size():%d is more than CC_FUSION_OP_MAX(%d), op:%s(%s) ,"
"check invalid when KernelTaskInfo %s", context.origin_op_index_size(), CC_FUSION_OP_MAX,
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(PARAM_INVALID, "context.origin_op_index_size() is more than CC_FUSION_OP_MAX(%d)", CC_FUSION_OP_MAX);
return PARAM_INVALID;
}
@@ -120,6 +132,9 @@ Status KernelTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci
ctx_.opIndex = context.op_index();
uint16_t *args_offset_tmp = reinterpret_cast<uint16_t *>(const_cast<char *>(context.args_offset().data()));
if (context.args_offset().size() / sizeof(uint16_t) < 1) {
REPORT_INNER_ERROR("E19999", "context.args_offset().size():%zu / sizeof(uint16_t) less than 1, op:%s(%s) ,"
"check invalid when KernelTaskInfo %s", context.args_offset().size(),
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "context.args_offset().size() / sizeof(uint16_t) less than 1");
return FAILED;
}
@@ -132,6 +147,8 @@ Status KernelTaskInfo::Init(const domi::TaskDef &task_def, DavinciModel *davinci
ret = InitAicpuTask(context.op_index(), kernel_def);
} else {
if (kernel_def.args().empty() || args_size_ == 0) {
REPORT_INNER_ERROR("E19999", "kernel_def.args() is empty, op:%s(%s), check invalid when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "args is null.");
return FAILED;
}
@@ -164,6 +181,8 @@ void KernelTaskInfo::UpdateSKTTaskId() {
if (davinci_model_ != nullptr) {
rtError_t rt_ret = rtModelGetTaskId(davinci_model_->GetRtModelHandle(), &task_id, &stream_id);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtModelGetTaskId fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return;
}
@@ -182,6 +201,8 @@ void KernelTaskInfo::UpdateTaskId() {
if (davinci_model_ != nullptr) {
rtError_t rt_ret = rtModelGetTaskId(davinci_model_->GetRtModelHandle(), &task_id, &stream_id);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtModelGetTaskId fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return;
}
@@ -237,6 +258,8 @@ Status KernelTaskInfo::SuperKernelLaunch() {
static_cast<rtSmDesc_t *>(skt_info.last_sm_desc), skt_info.last_stream,
skt_info.last_dump_flag);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtKernelLaunchWithFlag fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "SuperKernelLaunch: Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -249,6 +272,8 @@ Status KernelTaskInfo::SuperKernelLaunch() {
// Init super kernel factory
Status ge_ret = factory->Init();
if (ge_ret != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call SuperKernelFactory init fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), ge_ret, __FUNCTION__);
GELOGE(ge_ret, "SuperKernelLaunch: SuperKernelFactory init failed");
return ge_ret;
}
@@ -256,6 +281,9 @@ Status KernelTaskInfo::SuperKernelLaunch() {
std::unique_ptr<skt::SuperKernel> superKernel = nullptr;
ge_ret = factory->FuseKernels(skt_kernel_list, skt_arg_list, skt_info.last_block_dim, superKernel);
if (ge_ret != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call SuperKernelFactory FuseKernels fail for op:%s(%s), ret:0x%X, "
"when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), ge_ret, __FUNCTION__);
GELOGE(ge_ret, "SuperKernelLaunch: fuse call failed");
return ge_ret;
}
@@ -263,6 +291,8 @@ Status KernelTaskInfo::SuperKernelLaunch() {
skt_dump_flag_ = GetDumpFlag();
ge_ret = superKernel->Launch(skt_info.last_stream, skt_dump_flag_);
if (ge_ret != SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call SuperKernelFactory Launch fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), ge_ret, __FUNCTION__);
GELOGE(ge_ret, "SuperKernelLaunch: launch failed");
return ge_ret;
}
@@ -301,11 +331,14 @@ Status KernelTaskInfo::SaveSuperKernelInfo() {

bool KernelTaskInfo::IsMarkedLastNode() {
if (davinci_model_ == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when KernelTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return false;
}
OpDescPtr op_desc = davinci_model_->GetOpByIndex(ctx_.opIndex);
if (op_desc == nullptr) {
REPORT_INNER_ERROR("E19999", "Can't get op_desc from davinci_model by index:%u when KernelTaskInfo %s",
ctx_.opIndex, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "InitTVMTaskInfo error, index is out of range!");
return false;
}
@@ -316,11 +349,14 @@ bool KernelTaskInfo::IsMarkedLastNode() {

bool KernelTaskInfo::IsMarkedFirstNode() {
if (davinci_model_ == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when KernelTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return false;
}
OpDescPtr op_desc = davinci_model_->GetOpByIndex(ctx_.opIndex);
if (op_desc == nullptr) {
REPORT_INNER_ERROR("E19999", "Can't get op_desc from davinci_model by index:%u when KernelTaskInfo %s",
ctx_.opIndex, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "InitTVMTaskInfo error, index is out of range!");
return false;
}
@@ -361,6 +397,8 @@ Status KernelTaskInfo::SuperKernelDistribute() {
rtError_t rt_ret = rtKernelLaunchWithFlag(stub_func_, block_dim_, args_, args_size_,
static_cast<rtSmDesc_t *>(sm_desc_), stream_, dump_flag_);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtKernelLaunchWithFlag fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return rt_ret;
}
@@ -425,6 +463,9 @@ Status KernelTaskInfo::Distribute() {
}
}
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtKernelLaunchWithFlag or rtCpuKernelLaunchWithFlag fail for op:%s(%s), "
"ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -459,12 +500,16 @@ Status KernelTaskInfo::CopyNoncontinuousArgs(uint16_t offset) {
// copy io addr
errno_t sec_ret = memcpy_s(args_addr.get() + offset, addr_size, io_addrs_.data(), addr_size);
if (sec_ret != EOK) {
REPORT_CALL_ERROR("E19999", "Call memcpy_s fail, size:%zu, ret:0x%X, when KernelTaskInfo %s",
addr_size, sec_ret, __FUNCTION__);
GELOGE(FAILED, "memcpy failed, ret: %d", sec_ret);
return FAILED;
}

// copy args to device
rtError_t rt_ret = rtMemcpy(args_, args_size_, args_addr.get(), args_size_, RT_MEMCPY_HOST_TO_DEVICE);
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), args_size_, rt_ret, __FUNCTION__);
if (rt_ret != RT_ERROR_NONE) {
GELOGE(RT_FAILED, "Call rt api(rtMemcpy) failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
@@ -514,6 +559,8 @@ Status KernelTaskInfo::Release() {

ret = (sm_desc_ != nullptr) ? rtMemFreeManaged(sm_desc_) : RT_ERROR_NONE;
if (ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemFreeManaged fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", static_cast<int>(ret));
return RT_ERROR_TO_GE_STATUS(ret);
}
@@ -544,12 +591,16 @@ Status KernelTaskInfo::UpdateL2Data(const domi::KernelDef &kernel_def) {

rtError_t rt_ret = rtMemAllocManaged(&sm_desc_, sm_desc.size(), RT_MEMORY_SPM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemAllocManaged fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}

rt_ret = rtMemcpy(sm_desc_, sm_desc.size(), sm_desc.data(), sm_desc.size(), RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), sm_desc.size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -596,6 +647,8 @@ Status KernelTaskInfo::InitTVMTask(uint16_t offset, const domi::KernelDef &kerne
args_addr = std::unique_ptr<uint8_t[]>(new (std::nothrow) uint8_t[args_size_]);
errno_t sec_ret = memcpy_s(args_addr.get(), args_size_, kernel_def.args().data(), args_size_);
if (sec_ret != EOK) {
REPORT_CALL_ERROR("E19999", "Call memcpy_s fail, size:%u, ret:0x%X, when KernelTaskInfo %s",
args_size_, sec_ret, __FUNCTION__);
GELOGE(FAILED, "memcpy failed, ret: %d", sec_ret);
return FAILED;
}
@@ -638,6 +691,8 @@ Status KernelTaskInfo::InitTVMTask(uint16_t offset, const domi::KernelDef &kerne
// malloc args memory
rt_ret = rtMalloc(&args_, args_size_, RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), args_size_, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -645,11 +700,17 @@ Status KernelTaskInfo::InitTVMTask(uint16_t offset, const domi::KernelDef &kerne
// copy orign args
rt_ret = rtMemcpy(args_, args_size_, kernel_def.args().data(), args_size_, RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), args_size_, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}

if ((args_size_ <= offset) || (args_size_ - offset < kAddrLen * tensor_device_addrs.size())) {
REPORT_INNER_ERROR("E19999", "offset:%u >= kernelInfo.argsSize:%u or copy content:%zu beyond applied memory:%u, "
"check invalid in op:%s(%s), when KernelTaskInfo %s",
offset, args_size_, kAddrLen * tensor_device_addrs.size(), args_size_ - offset,
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "offset >= kernelInfo.argsSize or copy content beyond applied memory.");
return FAILED;
}
@@ -658,12 +719,17 @@ Status KernelTaskInfo::InitTVMTask(uint16_t offset, const domi::KernelDef &kerne
rt_ret = rtMemcpy(static_cast<char *>(args_) + offset, args_size_ - offset, tensor_device_addrs.data(),
kAddrLen * tensor_device_addrs.size(), RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
args_size_ - offset, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
sec_ret = memcpy_s(args_addr.get() + offset, args_size_ - offset, tensor_device_addrs.data(),
kAddrLen * tensor_device_addrs.size());
if (sec_ret != EOK) {
REPORT_CALL_ERROR("E19999", "Call memcpy_s fail, size:%u, ret:0x%X, when KernelTaskInfo %s",
args_size_ - offset, sec_ret, __FUNCTION__);
GELOGE(FAILED, "memcpy failed, ret: %d", sec_ret);
return FAILED;
}
@@ -708,6 +774,8 @@ Status KernelTaskInfo::InitAICPUCustomTask(uint32_t op_index, const domi::Kernel
GELOGI("Do InitAICPUCustomTask");
OpDescPtr op_desc = davinci_model_->GetOpByIndex(op_index);
if (op_desc == nullptr) {
REPORT_INNER_ERROR("E19999", "Can't get op_desc from davinci_model by index:%u when KernelTaskInfo %s",
op_index, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "index is out of range, index: %u", op_index);
return INTERNAL_ERROR;
}
@@ -718,11 +786,17 @@ Status KernelTaskInfo::InitAICPUCustomTask(uint32_t op_index, const domi::Kernel
const uint32_t kCustomAicpuArgsLen = 5;
ctx_.argsOffset = new (std::nothrow) uint16_t[kCustomAicpuArgsLen]();
if (ctx_.argsOffset == nullptr) {
REPORT_INNER_ERROR("E19999", "New ctx_.argsOffset fail, size:%u, op:%s(%s), when KernelTaskInfo %s",
kCustomAicpuArgsLen, op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(PARAM_INVALID, "ctx_.argsOffset is null!");
return PARAM_INVALID;
}

if (context.args_offset().size() / sizeof(uint16_t) < kCustomAicpuArgsLen) {
REPORT_INNER_ERROR("E19999", "context.args_offset().size():%zu / sizeof(uint16_t) is less than "
"kCustomAicpuArgsLen:%u, op:%s(%s), check invalid when KernelTaskInfo %s",
context.args_offset().size(), kCustomAicpuArgsLen,
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(PARAM_INVALID, "context.args_offset().size() / sizeof(uint16_t) is less than kCustomAicpuArgsLen");
return PARAM_INVALID;
}
@@ -743,24 +817,32 @@ Status KernelTaskInfo::InitAICPUCustomTask(uint32_t op_index, const domi::Kernel
// attrHandle
Buffer buffer;
if (!AttrUtils::GetBytes(op_desc, ATTR_NAME_OPATTR, buffer)) {
REPORT_INNER_ERROR("E19999", "Get Attr:%s in op:%s(%s) fail when KernelTaskInfo %s", ATTR_NAME_OPATTR.c_str(),
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "can't find opattr bytes!.");
return FAILED;
}

uint32_t op_attr_size = buffer.GetSize();
if (op_attr_size == 0) {
REPORT_INNER_ERROR("E19999", "Attr:%s in op:%s(%s) size is 0, check invalid when KernelTaskInfo %s",
ATTR_NAME_OPATTR.c_str(), op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(PARAM_INVALID, "param op_attr_size is out of range");
return PARAM_INVALID;
}

rtError_t rt_ret = rtMalloc(&custom_info_.attr_handle, op_attr_size, RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), op_attr_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}

rt_ret = rtMemcpy(custom_info_.attr_handle, op_attr_size, buffer.GetData(), op_attr_size, RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), op_attr_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -770,6 +852,10 @@ Status KernelTaskInfo::InitAICPUCustomTask(uint32_t op_index, const domi::Kernel

for (uint32_t i = 0; i < kCustomAicpuArgsLen; ++i) {
if (kernel_def.args().size() < ((size_t)ctx_.argsOffset[i] + sizeof(uint64_t))) {
REPORT_INNER_ERROR("E19999", "ctx.argsOffset[%u]: %u + sizeof(uint64_t): %zu >= kernelDef.args().size():%zu, "
"op:%s(%s) check invalid when KernelTaskInfo %s", i, (uint32_t)ctx_.argsOffset[i],
sizeof(uint64_t), kernel_def.args().size(),
op_desc->GetName().c_str(), op_desc->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "ctx.argsOffset[%u]: %u + sizeof(uint64_t): %zu >= kernelDef.args().size():%zu", i,
(uint32_t)ctx_.argsOffset[i], sizeof(uint64_t), kernel_def.args().size());
return FAILED;
@@ -788,6 +874,8 @@ Status KernelTaskInfo::InitAICPUCustomTask(uint32_t op_index, const domi::Kernel

rt_ret = rtMalloc(&args_, args_size_, RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(), args_size_, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -795,6 +883,9 @@ Status KernelTaskInfo::InitAICPUCustomTask(uint32_t op_index, const domi::Kernel
rt_ret = rtMemcpy(args_, kernel_def.args_size(), kernel_def.args().data(), kernel_def.args_size(),
RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc->GetName().c_str(), op_desc->GetType().c_str(),
kernel_def.args_size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -809,6 +900,7 @@ Status KernelTaskInfo::InitAICPUCustomTask(uint32_t op_index, const domi::Kernel
Status KernelTaskInfo::InitCceTask(const domi::KernelDef &kernel_def) {
GELOGI("Do InitCCETask");
if (davinci_model_ == nullptr) {
REPORT_INNER_ERROR("E19999", "Check param davinci_model nullptr when KernelTaskInfo %s", __FUNCTION__);
GELOGE(PARAM_INVALID, "davinci_model is null!");
return PARAM_INVALID;
}
@@ -823,6 +915,8 @@ Status KernelTaskInfo::InitCceTask(const domi::KernelDef &kernel_def) {

if (context.is_flowtable()) {
if (flowtable.empty()) {
REPORT_INNER_ERROR("E19999", "kernel_def.flowtable is empty, op:%s(%s), check invalid when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "flowtable is null.");
return FAILED;
}
@@ -857,6 +951,9 @@ Status KernelTaskInfo::InitCceTask(const domi::KernelDef &kernel_def) {
// args
rtError_t rt_ret = rtMalloc(&args_, kernel_def.args_size(), RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
kernel_def.args_size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -865,6 +962,9 @@ Status KernelTaskInfo::InitCceTask(const domi::KernelDef &kernel_def) {
rt_ret = rtMemcpy(args_, kernel_def.args_size(), kernel_def.args().data(), kernel_def.args_size(),
RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
kernel_def.args_size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -873,12 +973,16 @@ Status KernelTaskInfo::InitCceTask(const domi::KernelDef &kernel_def) {
if (!sm_desc.empty()) {
rt_ret = rtMemAllocManaged(&sm_desc_, sm_desc.size(), RT_MEMORY_SPM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemAllocManaged fail for op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}

rt_ret = rtMemcpy(sm_desc_, sm_desc.size(), sm_desc.data(), sm_desc.size(), RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), sm_desc.size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -893,6 +997,8 @@ Status KernelTaskInfo::InitAicpuTask(uint32_t op_index, const domi::KernelDef &k

OpDescPtr op_desc = davinci_model_->GetOpByIndex(op_index);
if (op_desc == nullptr) {
REPORT_INNER_ERROR("E19999", "Can't get op_desc from davinci_model by index:%u when KernelTaskInfo %s",
op_index, __FUNCTION__);
GELOGE(INTERNAL_ERROR, "index is out of range, index: %u", op_index);
return INTERNAL_ERROR;
}
@@ -910,6 +1016,8 @@ Status KernelTaskInfo::InitAicpuTask(uint32_t op_index, const domi::KernelDef &k
GE_PRINT_DYNAMIC_MEMORY(new, "cce task physical memory.", sizeof(uint8_t) * args_size_)
errno_t sec_ret = memcpy_s(args_addr.get(), args_size_, kernel_def.args().data(), args_size_);
if (sec_ret != EOK) {
REPORT_CALL_ERROR("E19999", "Call memcpy_s fail, size:%u, ret:0x%X, when KernelTaskInfo %s",
args_size_, sec_ret, __FUNCTION__);
GELOGE(FAILED, "memcpy failed, ret: %d", sec_ret);
return FAILED;
}
@@ -944,6 +1052,8 @@ Status KernelTaskInfo::InitAicpuTask(uint32_t op_index, const domi::KernelDef &k
auto addrs_size = sizeof(uint64_t) * io_addrs.size();
sec_ret = memcpy_s(reinterpret_cast<void *>(io_addr), addrs_size, io_addrs.data(), addrs_size);
if (sec_ret != EOK) {
REPORT_CALL_ERROR("E19999", "Call memcpy_s fail, size:%u, ret:0x%X, when KernelTaskInfo %s",
addrs_size, sec_ret, __FUNCTION__);
GELOGE(FAILED, "memcpy failed, ret: %d", sec_ret);
return FAILED;
}
@@ -952,6 +1062,8 @@ Status KernelTaskInfo::InitAicpuTask(uint32_t op_index, const domi::KernelDef &k
// malloc device memory for args
rtError_t rt_ret = rtMalloc(static_cast<void **>(&args_), args_size_, RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), args_size_, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api(rtMalloc) failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -960,6 +1072,8 @@ Status KernelTaskInfo::InitAicpuTask(uint32_t op_index, const domi::KernelDef &k
// copy args to device
rt_ret = rtMemcpy(args_, args_size_, args_addr.get(), args_size_, RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%u, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), args_size_, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api(rtMemcpy) failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1032,12 +1146,18 @@ Status KernelTaskInfo::InitAicpuTaskExtInfo(const std::string &ext_info) {
}
auto rt_ret = rtMalloc(&aicpu_ext_info_addr_, ext_handle->GetExtInfoLen(), RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
ext_handle->GetExtInfoLen(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMalloc ext_info error: 0x%X, size=%zu", rt_ret, ext_info.size());
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
rt_ret = rtMemcpy(aicpu_ext_info_addr_, ext_handle->GetExtInfoLen(), ext_handle->GetExtInfo(),
ext_handle->GetExtInfoLen(), RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
ext_handle->GetExtInfoLen(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMemcpy ext_info error: 0x%X, size=%zu", rt_ret, ext_info.size());
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1055,6 +1175,9 @@ Status KernelTaskInfo::StoreInputOutputTensor(const std::vector<void *> &input_d
// inputDescs
rtError_t rt_ret = rtMalloc(&custom_info_.input_descs, sizeof(opTensor_t) * input_size, RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
sizeof(opTensor_t) * input_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1063,6 +1186,9 @@ Status KernelTaskInfo::StoreInputOutputTensor(const std::vector<void *> &input_d
rt_ret = rtMemcpy(static_cast<opTensor_t *>(custom_info_.input_descs) + i, sizeof(opTensor_t),
const_cast<tagOpTensor *>(&input_descs[i]), sizeof(opTensor_t), RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
sizeof(opTensor_t), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1071,6 +1197,9 @@ Status KernelTaskInfo::StoreInputOutputTensor(const std::vector<void *> &input_d
// inputAddrs
rt_ret = rtMalloc(&custom_info_.input_addrs, sizeof(opTensor_t) * input_size, RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
sizeof(opTensor_t) * input_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1079,6 +1208,9 @@ Status KernelTaskInfo::StoreInputOutputTensor(const std::vector<void *> &input_d
rt_ret = rtMemcpy(custom_info_.input_addrs, kAddrLen * input_size, &input_data_addrs[0], kAddrLen * input_size,
RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
kAddrLen * input_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1087,6 +1219,9 @@ Status KernelTaskInfo::StoreInputOutputTensor(const std::vector<void *> &input_d
// outputDescs
rt_ret = rtMalloc(&custom_info_.output_descs, sizeof(opTensor_t) * output_size, RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
sizeof(opTensor_t) * output_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1094,6 +1229,9 @@ Status KernelTaskInfo::StoreInputOutputTensor(const std::vector<void *> &input_d
rt_ret = rtMemcpy(static_cast<opTensor_t *>(custom_info_.output_descs) + i, sizeof(opTensor_t),
const_cast<tagOpTensor *>(&input_descs[i]), sizeof(opTensor_t), RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
sizeof(opTensor_t), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1102,6 +1240,9 @@ Status KernelTaskInfo::StoreInputOutputTensor(const std::vector<void *> &input_d
// outputAddrs
rt_ret = rtMalloc(&custom_info_.output_addrs, sizeof(opTensor_t) * output_size, RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
sizeof(opTensor_t) * output_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1110,6 +1251,9 @@ Status KernelTaskInfo::StoreInputOutputTensor(const std::vector<void *> &input_d
rt_ret = rtMemcpy(custom_info_.output_addrs, kAddrLen * output_size, &output_data_addrs[0], kAddrLen * output_size,
RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail for op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
kAddrLen * output_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1126,11 +1270,17 @@ Status KernelTaskInfo::SetContext(const domi::KernelDef &kernel_def) {
ctx_.isFlowtable = context.is_flowtable();
ctx_.argsCount = context.args_count();
if (ctx_.argsCount == 0) {
REPORT_INNER_ERROR("E19999", "kernel_def.context.args_count is 0, op:%s(%s), check invalid when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(INTERNAL_ERROR, "check argsCount fail:%u.", ctx_.argsCount);
return INTERNAL_ERROR;
}

if (context.args_offset().size() / sizeof(uint16_t) < ctx_.argsCount) {
REPORT_INNER_ERROR("E19999", "param [context.args_offset().size():%zu / sizeof(uint16_t)] "
"is less than [ctx_.argsCount:%u], op:%s(%s), check invalid when KernelTaskInfo %s",
context.args_offset().size(), ctx_.argsCount,
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(PARAM_INVALID, "param [context.args_offset().size() / sizeof(uint16_t)] is less than [ctx_.argsCount]");
return PARAM_INVALID;
}
@@ -1138,6 +1288,8 @@ Status KernelTaskInfo::SetContext(const domi::KernelDef &kernel_def) {
// ctx_.argsOffset stores the offset of the internal information of agrs_, equal to the ctx_.argsCount
ctx_.argsOffset = new (std::nothrow) uint16_t[ctx_.argsCount]();
if (ctx_.argsOffset == nullptr) {
REPORT_INNER_ERROR("E19999", "New ctx_.argsOffset fail, size:%u, op:%s(%s), when KernelTaskInfo %s",
ctx_.argsCount, op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(PARAM_INVALID, "(param [ctx_.argsOffset] must not be null.");
return PARAM_INVALID;
}
@@ -1155,6 +1307,7 @@ void KernelTaskInfo::FreeRtMem(void **ptr) {
}
rtError_t ret = rtFree(*ptr);
if (ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtFree fail, ret:0x%X, when KernelTaskInfo %s", ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", ret);
}

@@ -1202,6 +1355,9 @@ Status KernelTaskInfo::CceUpdateKernelArgs(const domi::KernelContext &context, u
if (handle == nullptr) {
error = mmDlerror();
GE_IF_BOOL_EXEC(error == nullptr, error = "");
REPORT_INNER_ERROR("E19999", "Failed in dlopen:%s, dlerror:%s, op:%s(%s), when KernelTaskInfo %s",
canonicalPath.c_str(), error,
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(GE_PLGMGR_SO_NOT_EXIST, "Failed in dlopen %s! ", error);
return FAILED;
}
@@ -1210,6 +1366,9 @@ Status KernelTaskInfo::CceUpdateKernelArgs(const domi::KernelContext &context, u
auto cceUpdateKernelArgs = (ccStatus_t(*)(ccOpContext &, uint64_t, uint64_t,
uint64_t, void *, uint64_t, void *))mmDlsym(handle, const_cast<char *>(update_kernel_args.c_str()));
if (cceUpdateKernelArgs == nullptr) {
REPORT_INNER_ERROR("E19999", "No symbol:%s in %s, op:%s(%s), check invalid when KernelTaskInfo %s",
update_kernel_args.c_str(), canonicalPath.c_str(),
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "Failed to invoke function ccUpdateKernelArgs");
if (mmDlclose(handle) != 0) {
error = mmDlerror();
@@ -1234,6 +1393,8 @@ Status KernelTaskInfo::CceUpdateKernelArgs(const domi::KernelContext &context, u
return FAILED;
}
if (cc_ret != CC_STATUS_SUCCESS) {
REPORT_CALL_ERROR("E19999", "Call cceUpdateKernelArgs fail, op:%s(%s), ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), cc_ret, __FUNCTION__);
GELOGE(CCE_FAILED, "Call cce api failed, ret: 0x%X", cc_ret);
return CCE_FAILED;
}
@@ -1247,6 +1408,9 @@ Status KernelTaskInfo::SetFlowtable(std::string &flowtable, const domi::KernelDe
if (context.is_flowtable()) {
rtError_t rt_ret = rtMalloc(&flowtable_, flowtable.size(), RT_MEMORY_HBM);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail, op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
flowtable.size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1254,6 +1418,9 @@ Status KernelTaskInfo::SetFlowtable(std::string &flowtable, const domi::KernelDe

rt_ret = rtMemcpy(flowtable_, flowtable.size(), flowtable.data(), flowtable.size(), RT_MEMCPY_HOST_TO_DEVICE);
if (rt_ret != RT_ERROR_NONE) {
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail, op:%s(%s), size:%zu, ret:0x%X, when KernelTaskInfo %s",
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(),
flowtable.size(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "Call rt api failed, ret: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);
}
@@ -1263,6 +1430,12 @@ Status KernelTaskInfo::SetFlowtable(std::string &flowtable, const domi::KernelDe

if (kernel_def.args().size() <
((reinterpret_cast<uint16_t *>(const_cast<char *>(context.args_offset().data())))[0] + sizeof(uint64_t))) {
REPORT_INNER_ERROR(
"E19999", "(context.args_offset().data()))[0]:%u + sizeof(uint64_t):%zu > "
"kernelDef.args().size():%zu, op:%s(%s), check invalid when %s",
(uint32_t)((reinterpret_cast<uint16_t *>(const_cast<char *>(context.args_offset().data())))[0]),
sizeof(uint64_t), kernel_def.args().size(),
op_desc_->GetName().c_str(), op_desc_->GetType().c_str(), __FUNCTION__);
GELOGE(FAILED, "(context.args_offset().data()))[0]:%u + sizeof(uint64_t):%zu > kernelDef.args().size():%zu",
(uint32_t)((reinterpret_cast<uint16_t *>(const_cast<char *>(context.args_offset().data())))[0]),
sizeof(uint64_t), kernel_def.args().size());


+ 12
- 3
ge/graph/load/model_manager/task_info/super_kernel/super_kernel.cc View File

@@ -26,15 +26,24 @@ Status SuperKernel::Launch(rtStream_t stream, uint32_t dump_flag) {
reinterpret_cast<const void *>(static_cast<uintptr_t>(this->GetNavTableSize()))};

rtError_t rt_ret = rtMalloc(reinterpret_cast<void **>(&device_args_addr_), sizeof(args), RT_MEMORY_HBM);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMalloc failied. error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail, size:%u, ret:0x%X when %s",
sizeof(args), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMalloc failied. error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)
rt_ret = rtMemcpy(reinterpret_cast<void *>(device_args_addr_), sizeof(args), reinterpret_cast<void *>(args),
sizeof(args), RT_MEMCPY_HOST_TO_DEVICE);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMemcpy failied. error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail, size:%u, ret:0x%X when %s",
sizeof(args), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMemcpy failied. error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)
rt_ret = rtKernelLaunchWithFlag((void *const)func_stub_, block_dim_, device_args_addr_, sizeof(args), NULL, stream,
dump_flag);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtKernelLaunchWithFlag failied. error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtKernelLaunchWithFlag fail, dump_flag:%u, ret:0x%X when %s",
dump_flag, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtKernelLaunchWithFlag failied. error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)
return SUCCESS;
}


+ 19
- 9
ge/graph/load/model_manager/task_info/super_kernel/super_kernel_factory.cc View File

@@ -35,14 +35,16 @@ Status SuperKernelFactory::Init() {
}
rtError_t rt_ret;
rt_ret = rtGetFunctionByName(this->sk_stub_name_.c_str(), &this->func_stub_);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED,
"rtGetFunctionByName "
"failed. stub_func: %s, please export LD_LIBRARY_PATH for "
"libcce_aicore.so",
this->sk_stub_name_.c_str());
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtGetFunctionByName fail, stub_func:%s, ret:0x%X, when %s",
this->sk_stub_name_.c_str(), rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtGetFunctionByName failed. stub_func: %s, please export LD_LIBRARY_PATH for "
"libcce_aicore.so", this->sk_stub_name_.c_str());
return RT_ERROR_TO_GE_STATUS(rt_ret);)
rt_ret = rtGetAddrByFun(this->func_stub_, &this->func_ptr_);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtGetAddrByFun failed. error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtGetAddrByFun fail, ret:0x%X, when %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtGetAddrByFun failed. error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)
GELOGD(
"SKT: fuseKernels super_kernel_template subFunc %p, device func "
@@ -98,7 +100,9 @@ Status SuperKernelFactory::FuseKernels(const std::vector<void *> &stub_func_list
for (unsigned i = 0; i < stub_func_list.size(); i++) {
void *sub_device_func = nullptr;
rt_ret = rtGetAddrByFun(stub_func_list[i], &sub_device_func);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtGetAddrByFun failed. error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtGetAddrByFun fail, ret:0x%X, when %s", rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtGetAddrByFun failed. error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)
GELOGD("SKT: fuseKernels subFunc %p, device func address %p", stub_func_list[i], sub_device_func);
// store two uint64_t address
@@ -109,11 +113,17 @@ Status SuperKernelFactory::FuseKernels(const std::vector<void *> &stub_func_list
GELOGD("SKT: fuseKernels args base address %lu", nav_table[i * kFusedKernelSizeUnit + 1]);
}
rt_ret = rtMalloc(reinterpret_cast<void **>(&hbm_nav_table_addr), nav_table_size, RT_MEMORY_HBM);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMalloc failed. error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMalloc fail, size:%lu, ret:0x%X, when %s",
nav_table_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMalloc failed. error: 0x%X", rt_ret);
return RT_ERROR_TO_GE_STATUS(rt_ret);)
rt_ret = rtMemcpy(reinterpret_cast<void *>(hbm_nav_table_addr), nav_table_size,
reinterpret_cast<void *>(nav_table.get()), nav_table_size, RT_MEMCPY_HOST_TO_DEVICE);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE, GELOGE(RT_FAILED, "rtMemcpy failed. error: 0x%X", rt_ret);
GE_IF_BOOL_EXEC(rt_ret != RT_ERROR_NONE,
REPORT_CALL_ERROR("E19999", "Call rtMemcpy fail, size:%lu, ret:0x%X when %s",
nav_table_size, rt_ret, __FUNCTION__);
GELOGE(RT_FAILED, "rtMemcpy failed. error: 0x%X", rt_ret);
GE_CHK_RT(rtFree(hbm_nav_table_addr)); return RT_ERROR_TO_GE_STATUS(rt_ret);)
// Create the necessary metadata for the super kernel
h =


+ 3
- 0
inc/framework/common/debug/log.h View File

@@ -19,6 +19,7 @@

#include <string>
#include <sstream>
#include <securec.h>

#include "runtime/rt.h"
#include "common/string_util.h"
@@ -105,6 +106,7 @@
do { \
bool b = (expr); \
if (!b) { \
REPORT_INNER_ERROR("E19999", __VA_ARGS__); \
GELOGE(_status, __VA_ARGS__); \
return _status; \
} \
@@ -193,6 +195,7 @@
{ \
bool b = (expr); \
if (b) { \
REPORT_INNER_ERROR("E19999", __VA_ARGS__); \
DOMI_LOGE(__VA_ARGS__); \
exec_expr; \
return _status; \


+ 1
- 1
metadef

@@ -1 +1 @@
Subproject commit ac0de0213755e49360a9467eb5b13e13a752a35b
Subproject commit 99934058bfdceaae740acf2c8cadd316b51c00c7

+ 1
- 1
parser

@@ -1 +1 @@
Subproject commit eff7e2ecc54ef7887581acd6ea66356de6872e3a
Subproject commit 0d4703aa893e90f23ba8a2dbd8903e028680213f

Write Preview
Loading…
Cancel
Save