Merge pull request !38774 from TuDouNi/codeclean
This commit is contained in:
i-robot 2022-07-26 01:19:29 +00:00 committed by Gitee
commit f333b03d4d
No known key found for this signature in database
GPG Key ID: 173E9B9CA92EEF8F
2 changed files with 26 additions and 23 deletions

View File

@ -302,10 +302,10 @@ void GenKernelIoExecInfoMap(const NotNull<KernelGraphPtr> &kernel_graph,
}
}
void AscendStreamAssign::InsertEventsForInputs(const NotNull<KernelGraphPtr> &kernel_graph, const CNodePtr &kernel,
const NodeIoExecInfoPtr &io_exec_info,
void AscendStreamAssign::InsertEventsForInputs(
const NotNull<KernelGraphPtr> &kernel_graph, const CNodePtr &kernel, const NodeIoExecInfoPtr &io_exec_info,
mindspore::HashMap<AnfNodePtr, std::vector<CNodePtr>> *kernel_send,
mindspore::HashMap<AnfNodePtr, std::vector<CNodePtr>> *kernel_recv) {
mindspore::HashMap<AnfNodePtr, std::vector<CNodePtr>> *kernel_recv) const {
auto process_stream_id = AnfAlgo::GetStreamId(kernel);
auto input_exec_info_list = io_exec_info->inputs;
mindspore::HashMap<uint32_t, NodeExecInfoPtr> stream_max_exec_node_map;
@ -333,10 +333,10 @@ void AscendStreamAssign::InsertEventsForInputs(const NotNull<KernelGraphPtr> &ke
}
}
void AscendStreamAssign::InsertEventsForOutputs(const NotNull<KernelGraphPtr> &kernel_graph, const CNodePtr &kernel,
const NodeIoExecInfoPtr &io_exec_info,
void AscendStreamAssign::InsertEventsForOutputs(
const NotNull<KernelGraphPtr> &kernel_graph, const CNodePtr &kernel, const NodeIoExecInfoPtr &io_exec_info,
mindspore::HashMap<AnfNodePtr, std::vector<CNodePtr>> *kernel_send,
mindspore::HashMap<AnfNodePtr, std::vector<CNodePtr>> *kernel_recv) {
mindspore::HashMap<AnfNodePtr, std::vector<CNodePtr>> *kernel_recv) const {
auto process_stream_id = AnfAlgo::GetStreamId(kernel);
auto output_exec_info_list = io_exec_info->outputs;
mindspore::HashMap<uint32_t, NodeExecInfoPtr> stream_min_exec_node_map;
@ -731,7 +731,8 @@ CNodePtr AscendStreamAssign::GetCNodesNeededMoved(vector<CNodePtr> *moved_backwa
}
CNodePtr AscendStreamAssign::GetTargetOutputNode(const vector<CNodePtr> &moved_backward_cnodes,
const CNodePtr first_node, const NotNull<KernelGraphPtr> &graph_ptr) {
const CNodePtr first_node,
const NotNull<KernelGraphPtr> &graph_ptr) const {
auto cnode_ptr_list = graph_ptr->execution_order();
if (moved_backward_cnodes.empty() || !first_node) {
return nullptr;
@ -770,7 +771,7 @@ CNodePtr AscendStreamAssign::GetTargetOutputNode(const vector<CNodePtr> &moved_b
return first_output_node_ptr;
}
bool AscendStreamAssign::FinetuneSubgraphExecOrder(vector<CNodePtr> *cnodes) {
bool AscendStreamAssign::FinetuneSubgraphExecOrder(vector<CNodePtr> *cnodes) const {
MS_EXCEPTION_IF_NULL(cnodes);
auto hcom_pos = find_if(cnodes->begin(), cnodes->end(), [](const CNodePtr &node_ptr) -> bool {
return common::AnfAlgo::GetCNodeName(node_ptr) == "AllReduce";
@ -1005,7 +1006,7 @@ void AscendStreamAssign::ClassifyNodeByGroupAndGraph(const std::vector<CNodePtr>
}
}
std::set<uint32_t> AscendStreamAssign::AssignNodeStreamInOrder(const std::vector<CNodePtr> node_list) {
std::set<uint32_t> AscendStreamAssign::AssignNodeStreamInOrder(const std::vector<CNodePtr> node_list) const {
AscendStreamMng &resource_manager = AscendStreamMng::GetInstance();
auto cur_stream_id = resource_manager.ApplyNewStream();
std::map<uint32_t, uint32_t> stream_task_map;
@ -1584,7 +1585,7 @@ void AscendStreamAssign::InsertRecvForNotLoopSink(const NotNull<KernelGraphPtr>
}
}
void AscendStreamAssign::GraphLoopSync(const NotNull<KernelGraphPtr> &root_graph, uint32_t graph_id) {
void AscendStreamAssign::GraphLoopSync(const NotNull<KernelGraphPtr> &root_graph, uint32_t graph_id) const {
if (ExistStreamSendAfterLastHcomNode(root_graph, graph_id)) {
return;
}
@ -1733,7 +1734,7 @@ void AscendStreamAssign::InsertEventHcomDependCommonBak(const NotNull<KernelGrap
}
vector<CNodePtr> AscendStreamAssign::GetLastInputCnode(const NotNull<KernelGraphPtr> &graph_ptr,
const CNodePtr &cur_cnode_ptr) {
const CNodePtr &cur_cnode_ptr) const {
auto group_name = GetHcomGroup(cur_cnode_ptr);
auto input_cnodes = GetInputKernels(cur_cnode_ptr);
if (input_cnodes.empty()) {
@ -1894,7 +1895,8 @@ std::vector<std::pair<uint32_t, vector<size_t>>> AscendStreamAssign::GetStreamID
}
void AscendStreamAssign::InsertEventHcomDependHcomAtSameGroup(
const NotNull<KernelGraphPtr> &graph_ptr, std::pair<std::string, std::map<uint32_t, std::set<uint32_t>>> group_item) {
const NotNull<KernelGraphPtr> &graph_ptr,
std::pair<std::string, std::map<uint32_t, std::set<uint32_t>>> group_item) const {
for (const auto &graph_item : std::as_const(group_item.second)) {
auto stream_indices = GetStreamIDHcomMap(graph_ptr->execution_order(), group_item.first, graph_item.first);
constexpr size_t kStreamMax = 2;

View File

@ -87,7 +87,7 @@ class AscendStreamAssign {
~AscendStreamAssign() = default;
void AssignAllNodesStream(const NotNull<KernelGraphPtr> &graph_ptr);
std::set<uint32_t> AssignNodeStreamInOrder(const std::vector<CNodePtr> node_list);
std::set<uint32_t> AssignNodeStreamInOrder(const std::vector<CNodePtr> node_list) const;
void ClassifyNodeByKernel(const NotNull<KernelGraphPtr> &graph_ptr, std::vector<CNodePtr> *common_list,
std::vector<CNodePtr> *hcom_list, std::vector<CNodePtr> *independent_list,
std::vector<CNodePtr> *comm_sub_graph_list) const;
@ -124,8 +124,9 @@ class AscendStreamAssign {
void InsertEventBetweenHcom(const NotNull<KernelGraphPtr> &graph_ptr,
const std::vector<std::pair<uint32_t, vector<size_t>>> &hcom_index) const;
void InsertEventForCallCommSubGraph(const NotNull<KernelGraphPtr> &graph_ptr) const;
void InsertEventHcomDependHcomAtSameGroup(const NotNull<KernelGraphPtr> &graph_ptr,
std::pair<std::string, std::map<uint32_t, std::set<uint32_t>>> group_item);
void InsertEventHcomDependHcomAtSameGroup(
const NotNull<KernelGraphPtr> &graph_ptr,
std::pair<std::string, std::map<uint32_t, std::set<uint32_t>>> group_item) const;
void InsertRecvForLoopSink(const NotNull<KernelGraphPtr> &root_graph, std::vector<CNodePtr> *cnodes,
uint32_t cur_event_id, uint32_t graph_id) const;
void InsertRecvForNotLoopSink(const NotNull<KernelGraphPtr> &root_graph, std::vector<CNodePtr> *cnodes,
@ -134,7 +135,7 @@ class AscendStreamAssign {
const std::string group, size_t graph_id) const;
void AdjustAtomicAddrCleanOrder(const NotNull<KernelGraphPtr> &graph_ptr) const;
vector<CNodePtr> GetLastInputCnode(const NotNull<KernelGraphPtr> &graph_ptr, const CNodePtr &cur_cnode_ptr);
vector<CNodePtr> GetLastInputCnode(const NotNull<KernelGraphPtr> &graph_ptr, const CNodePtr &cur_cnode_ptr) const;
vector<CNodePtr> GetIndependentNodesNeedsInsertActive(const std::vector<CNodePtr> exe_orders,
const uint32_t graph_id) const;
bool IsSatisfiedHcom(const std::vector<std::pair<uint32_t, vector<size_t>>> &hcom_index, const CNodePtr &node_ptr,
@ -148,8 +149,8 @@ class AscendStreamAssign {
CNodePtr GetCNodesNeededMoved(vector<CNodePtr> *moved_backward_cnodes, vector<CNodePtr> *moved_forward_cnodes,
const vector<CNodePtr> &last_grad_and_status, const NotNull<KernelGraphPtr> &graph_ptr);
CNodePtr GetTargetOutputNode(const vector<CNodePtr> &moved_backward_cnodes, const CNodePtr first_node,
const NotNull<KernelGraphPtr> &graph_ptr);
bool FinetuneSubgraphExecOrder(vector<CNodePtr> *cnodes);
const NotNull<KernelGraphPtr> &graph_ptr) const;
bool FinetuneSubgraphExecOrder(vector<CNodePtr> *cnodes) const;
void TrailingTimeOptimizationByReorder(const NotNull<KernelGraphPtr> &graph_ptr);
uint32_t GetMaxIndexTarget(const NotNull<KernelGraphPtr> &graph_ptr);
@ -185,7 +186,7 @@ class AscendStreamAssign {
bool ExistStreamSendAfterLastHcomNode(const NotNull<KernelGraphPtr> &graph_ptr, uint32_t graph_id) const;
void GetAllGraphID(const NotNull<KernelGraphPtr> &graph_ptr, std::vector<uint32_t> *graphs_id);
void GraphLoopSync(const NotNull<KernelGraphPtr> &root_graph, uint32_t graph_id);
void GraphLoopSync(const NotNull<KernelGraphPtr> &root_graph, uint32_t graph_id) const;
void InsertEventForMicroBatchIndependent(const NotNull<KernelGraphPtr> &graph_ptr) const;
@ -231,11 +232,11 @@ class AscendStreamAssign {
void InsertEventsForInputs(const NotNull<KernelGraphPtr> &kernel_graph, const CNodePtr &kernel,
const NodeIoExecInfoPtr &io_exec_info, HashMap<AnfNodePtr, vector<CNodePtr>> *kernel_send,
HashMap<AnfNodePtr, vector<CNodePtr>> *kernel_recv);
HashMap<AnfNodePtr, vector<CNodePtr>> *kernel_recv) const;
void InsertEventsForOutputs(const NotNull<KernelGraphPtr> &kernel_graph, const CNodePtr &kernel,
const NodeIoExecInfoPtr &io_exec_info, HashMap<AnfNodePtr, vector<CNodePtr>> *kernel_send,
HashMap<AnfNodePtr, vector<CNodePtr>> *kernel_recv);
HashMap<AnfNodePtr, vector<CNodePtr>> *kernel_recv) const;
void InsertEvents(const NotNull<KernelGraphPtr> &kernel_graph, const CNodePtr &parallel_cnode,
const AnfNodePtr &node_before_send, HashMap<mindspore::AnfNodePtr, vector<CNodePtr>> *kernel_send,