mindspore-Ecosystem
/
mindspore
21
0
Fork 27
Code Issues Pull Requests 6 Projects Releases Wiki Activity

This commit is to separate the computation cost and memory cost in auto_parallel. Some related memory correction is removed.

This commit is contained in:
Xiaoda Zhang 2020-04-03 14:33:04 +08:00
parent 0d838c7c9b
commit a153fad874
21 changed files with 313 additions and 363 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
mindspore/ccsrc/parallel/auto_parallel/costmodel.cc
View File

@ -23,8 +23,8 @@
namespace mindspore {
namespace parallel {
void Simplify(CostPtrList* clist_ptrs) {
// Sort the cost_list with the memory_cost increasing, and communication_cost decreasing order. This method
// excludes the cost with greater memory_cost and greater communication_cost.
// Sort the cost_list with the computation_cost_ increasing, and communication_cost decreasing order. This method
// excludes the cost with greater computation_cost_ and greater communication_cost.
// E.g. clist_ptrs = {<100, 20>, <200, 10>, <300, 50>}. After this method, clist_ptrs = {<200, 10>, <100, 20>}
if (!COST_MODEL_SIMPLIFY_CALCULATION) {
return;
@ -33,7 +33,7 @@ void Simplify(CostPtrList* clist_ptrs) {
std::vector<size_t> id(clist_ptrs->size());
std::iota(id.begin(), id.end(), size_t(0));
std::sort(id.begin(), id.end(), [&clist_ptrs](size_t x, size_t y) {
return clist_ptrs->at(x)->memory_cost_ < clist_ptrs->at(y)->memory_cost_;
return clist_ptrs->at(x)->computation_cost_ < clist_ptrs->at(y)->computation_cost_;
});
CostPtrList ret;
for (size_t i = 0; i < clist_ptrs->size(); ++i) {
@ -45,8 +45,8 @@ void Simplify(CostPtrList* clist_ptrs) {
}
void SimplifyForDreasingCommunicationWithPartialPara(CostPtrList* clist_ptrs) {
// Sort the cost_list with the memory_cost increasing, and communication_with_partial_para_cost decreasing order.
// This method excludes the cost with greater memory_cost and greater communication_without_para_cost.
// Sort the cost_list with the computation_cost_ increasing, and communication_with_partial_para_cost decreasing
// order. This method excludes the cost with greater computation_cost_ and greater communication_without_para_cost.
if (!COST_MODEL_SIMPLIFY_CALCULATION) {
return;
}
@ -54,7 +54,7 @@ void SimplifyForDreasingCommunicationWithPartialPara(CostPtrList* clist_ptrs) {
std::vector<size_t> id(clist_ptrs->size());
std::iota(id.begin(), id.end(), size_t(0));
std::sort(id.begin(), id.end(), [&clist_ptrs](size_t x, size_t y) {
return clist_ptrs->at(x)->memory_cost_ < clist_ptrs->at(y)->memory_cost_;
return clist_ptrs->at(x)->computation_cost_ < clist_ptrs->at(y)->computation_cost_;
});
CostPtrList ret;
for (size_t i = 0; i < clist_ptrs->size(); ++i) {

10
mindspore/ccsrc/parallel/auto_parallel/costmodel.h
View File

@ -44,14 +44,18 @@ using RedistributionOpListPtr = std::shared_ptr<std::pair<OperatorVector, OutPut
struct Cost {
Cost();
Cost(double memory, double commuication, const std::shared_ptr<Decision>& decision_ = nullptr)
: memory_cost_(memory), communication_cost_(commuication), decision_ptr_(std::move(decision_)) {
Cost(double computation, double commuication, const std::shared_ptr<Decision>& decision_ = nullptr)
: computation_cost_(computation), communication_cost_(commuication), decision_ptr_(std::move(decision_)) {
memory_with_reuse_ = 0.0;
communication_without_parameter_ = 0.0;
communication_with_partial_para_ = 0.0;
communication_redis_forward_ = 0.0;
communication_redis_backward_ = 0.0;
}
double memory_cost_;
// 'memory_with_reuse_' calculates the peak memory usage in a training phase
double memory_with_reuse_;
// 'computation_cost_' models the training time of an iteration in a training phase
double computation_cost_;
// 'communication_cost_' includes communications from operators (forward and backward) and edges
double communication_cost_;
// communication_without_parameter_ = communication_cost_ - (backward communication from operators)

2
mindspore/ccsrc/parallel/auto_parallel/dp_algo_costmodel.h
View File

@ -35,7 +35,7 @@ namespace parallel {
// interpretation of 6 operations in costmodel.h.
// Phase 2: Search the cost_list in the final graph, and determine the optimal one
// Create the cost_list for the final graph, and choose the optimal one: one the minimum quantity
// COST_MODEL_ALPHA * memory_cost + COST_MODEL_BETA * communication_cost
// COST_MODEL_ALPHA * computation_cost + COST_MODEL_BETA * communication_cost
// Phase 3: Recover the original CostGraph, the determine strategy for each operator
// After determining the optimal cost for the final graph, the algorithm recovers the original graph by applying
// the 4 operations in the reverse order in the Phase 1. Because each operation decision contains the strategy,

51
mindspore/ccsrc/parallel/auto_parallel/edge_costmodel.cc
View File

@ -69,7 +69,7 @@ Status Edge::InitEdgeCost() {
MS_LOG(EXCEPTION) << "Failure: redistribution cost calculation failed";
}
MS_EXCEPTION_IF_NULL(cost);
MS_LOG(DEBUG) << "The redistribution cost: memory_cost: " << cost->memory_cost_
MS_LOG(DEBUG) << "The redistribution cost: computation_cost: " << cost->computation_cost_
<< ", communication_cost: " << cost->communication_cost_
<< ", communication_without_parameter_: " << cost->communication_without_parameter_
<< ", communication_with_partial_para_: " << cost->communication_with_partial_para_ << ".";
@ -117,9 +117,9 @@ Status Edge::GetRedistributionCost(const TensorLayout& prev_op_output_layout, co
double comm_cost = tensor_redistribution.comm_cost();
double forward_comm_cost = tensor_redistribution.forward_comm_cost();
double backward_comm_cost = tensor_redistribution.backward_comm_cost();
double mem_cost = tensor_redistribution.mem_cost();
double computation_cost = tensor_redistribution.computation_cost();
*cost = std::make_shared<Cost>(type_length * mem_cost, type_length * comm_cost);
*cost = std::make_shared<Cost>(type_length * computation_cost, type_length * comm_cost);
(*cost)->communication_without_parameter_ = type_length * comm_cost;
(*cost)->communication_with_partial_para_ =
(*cost)->communication_without_parameter_ +
@ -150,26 +150,26 @@ CostPtrList Edge::CreateEdgeEliminationCostList(const StrategyPtr& output_st_ptr
(void)std::transform(edges.begin(), edges.end(), all_cost_list.begin(), LocalGetCostList);
CostPtrList selected_cost_list(all_cost_list.size(), nullptr);
std::function<void(size_t, double, double, double)> recursive = [&](size_t k, double memory, double communication,
double communication_without_para) {
if (k == edges.size()) {
auto decision = std::make_shared<EdgeEliminationDecision>(selected_cost_list);
CostPtr new_cost = std::make_shared<Cost>(memory, communication);
MS_EXCEPTION_IF_NULL(new_cost);
new_cost->communication_without_parameter_ = communication_without_para;
new_cost->communication_with_partial_para_ =
communication_without_para + COST_MODEL_GAMMA * (communication - communication_without_para);
new_cost->decision_ptr_ = decision;
result.push_back(new_cost);
return;
}
for (auto& c : all_cost_list[k]) {
MS_EXCEPTION_IF_NULL(c);
selected_cost_list[k] = c;
recursive(k + 1, memory + c->memory_cost_, communication + c->communication_cost_,
communication_without_para + c->communication_without_parameter_);
}
};
std::function<void(size_t, double, double, double)> recursive =
[&](size_t k, double computation, double communication, double communication_without_para) {
if (k == edges.size()) {
auto decision = std::make_shared<EdgeEliminationDecision>(selected_cost_list);
CostPtr new_cost = std::make_shared<Cost>(computation, communication);
MS_EXCEPTION_IF_NULL(new_cost);
new_cost->communication_without_parameter_ = communication_without_para;
new_cost->communication_with_partial_para_ =
communication_without_para + COST_MODEL_GAMMA * (communication - communication_without_para);
new_cost->decision_ptr_ = decision;
result.push_back(new_cost);
return;
}
for (auto& c : all_cost_list[k]) {
MS_EXCEPTION_IF_NULL(c);
selected_cost_list[k] = c;
recursive(k + 1, computation + c->computation_cost_, communication + c->communication_cost_,
communication_without_para + c->communication_without_parameter_);
}
};
recursive(0, 0, 0, 0);
SimplifyForDreasingCommunicationWithPartialPara(&result);
return result;
@ -203,7 +203,8 @@ void Edge::CreateOpEliminationSubCostList(StrategyPtr op_strategy, const CostPtr
MS_EXCEPTION_IF_NULL(middle_cost);
for (auto& right_cost : right_cost_list) {
MS_EXCEPTION_IF_NULL(right_cost);
double memory = left_cost->memory_cost_ + middle_cost->memory_cost_ + right_cost->memory_cost_;
double computation =
left_cost->computation_cost_ + middle_cost->computation_cost_ + right_cost->computation_cost_;
double communication =
left_cost->communication_cost_ + middle_cost->communication_cost_ + right_cost->communication_cost_;
double communication_without_para = left_cost->communication_without_parameter_ +
@ -211,7 +212,7 @@ void Edge::CreateOpEliminationSubCostList(StrategyPtr op_strategy, const CostPtr
right_cost->communication_without_parameter_;
auto decision = std::make_shared<OpEliminationDecision>(op_strategy, left_cost, middle_cost, right_cost);
auto cost = std::make_shared<Cost>(memory, communication, decision);
auto cost = std::make_shared<Cost>(computation, communication, decision);
MS_EXCEPTION_IF_NULL(cost);
cost->communication_without_parameter_ = communication_without_para;
cost->communication_with_partial_para_ =

2
mindspore/ccsrc/parallel/auto_parallel/edge_costmodel.h
View File

@ -133,7 +133,7 @@ class Edge {
void set_parameter_involve(int para_invol) { is_output_parameter_involve_ = para_invol; }
// When the input of a operator contains WEIGHT or a output from other operators involving WEIGHT, then these input
// should stay in memory until it is used in the backward phase, which is kept in memory at the end of forward phase.
Status CorrectStrategyCostForMemoryReuse() const { return SUCCESS; }
Status CalculateMemoryCost() const { return SUCCESS; }
private:
std::string edge_name_;

105
mindspore/ccsrc/parallel/auto_parallel/graph_costmodel.cc
View File

@ -247,7 +247,7 @@ CostPtrList CostGraph::CreateFinalCostList(const OperatorInfoPtr& u, const std::
MS_EXCEPTION_IF_NULL(cost1);
MS_EXCEPTION_IF_NULL(cost2);
MS_EXCEPTION_IF_NULL(cost3);
double memory = cost1->memory_cost_ + cost2->memory_cost_ + cost3->memory_cost_;
double computation = cost1->computation_cost_ + cost2->computation_cost_ + cost3->computation_cost_;
double commmunication =
cost1->communication_cost_ + cost2->communication_cost_ + cost3->communication_cost_;
double communication_without_para = cost1->communication_without_parameter_ +
@ -255,7 +255,7 @@ CostPtrList CostGraph::CreateFinalCostList(const OperatorInfoPtr& u, const std::
cost3->communication_without_parameter_;
auto decision =
std::make_shared<FinalDecision>(u_strategy->strategy_ptr, v_strategy->strategy_ptr, cost1, cost2, cost3);
auto cost = std::make_shared<Cost>(memory, commmunication, decision);
auto cost = std::make_shared<Cost>(computation, commmunication, decision);
MS_EXCEPTION_IF_NULL(cost);
cost->communication_without_parameter_ = communication_without_para;
cost->communication_with_partial_para_ =
@ -282,7 +282,7 @@ CostPtrList CostGraph::CreateFinalSingleCostList(const OperatorInfoPtr& u) {
for (const auto& cost1 : clist1) {
MS_EXCEPTION_IF_NULL(cost1);
auto decision = std::make_shared<FinalSingleDecision>(u_strategy_ptr, cost1);
auto new_cost = std::make_shared<Cost>(cost1->memory_cost_, cost1->communication_cost_, decision);
auto new_cost = std::make_shared<Cost>(cost1->computation_cost_, cost1->communication_cost_, decision);
MS_EXCEPTION_IF_NULL(new_cost);
new_cost->communication_without_parameter_ = cost1->communication_without_parameter_;
new_cost->communication_with_partial_para_ =
@ -297,12 +297,12 @@ CostPtrList CostGraph::CreateFinalSingleCostList(const OperatorInfoPtr& u) {
}
CostPtr CostGraph::SelectCostWithMemoryConstraint(const CostPtrList& cost_list, double memory) {
if (cost_list.empty() || cost_list[0]->memory_cost_ >= memory) {
if (cost_list.empty() || cost_list[0]->computation_cost_ >= memory) {
return nullptr;
}
std::function<CostPtr(CostPtr, const CostPtr&)> LocalCompare = [&](CostPtr init, const CostPtr& cost_x) {
MS_EXCEPTION_IF_NULL(cost_x);
if (init == nullptr || cost_x->memory_cost_ < memory) {
if (init == nullptr || cost_x->computation_cost_ < memory) {
init = cost_x;
}
return init;
@ -313,36 +313,36 @@ CostPtr CostGraph::SelectCostWithMemoryConstraint(const CostPtrList& cost_list,
CostPtr CostGraph::SelectCostWithMinTrainingTime(const CostPtrList& cost_list, double memory) {
// Select the cost with minimum training time. Currently, the training time is modeled as =
// costmodel_alpha_ * memory_cost + costmodel_beta_ * communication_with_partial_para_
// costmodel_alpha_ * computation_cost + costmodel_beta_ * communication_with_partial_para_
if (cost_list.empty()) {
MS_LOG(ERROR) << "Final cost list is null.";
return nullptr;
}
CostPtr ret = cost_list[0];
MS_EXCEPTION_IF_NULL(ret);
if (ret->memory_cost_ >= memory) {
MS_LOG(ERROR) << "No available cost; the minimum cost is " << ret->memory_cost_
if (ret->computation_cost_ >= memory) {
MS_LOG(ERROR) << "No available cost; the minimum cost is " << ret->computation_cost_
<< ", the memory capacity is: " << memory << ".";
return nullptr;
}
double minimum = costmodel_alpha_ * ret->memory_cost_ + costmodel_beta_ * ret->communication_with_partial_para_;
MS_LOG(INFO) << "minimum: " << minimum << ", memory_cost_: " << ret->memory_cost_
double minimum = costmodel_alpha_ * ret->computation_cost_ + costmodel_beta_ * ret->communication_with_partial_para_;
MS_LOG(INFO) << "minimum: " << minimum << ", computation_cost_: " << ret->computation_cost_
<< ", communication_with_partial_para_: " << ret->communication_with_partial_para_
<< ", communication_cost_: " << ret->communication_cost_
<< ", communication_without_parameter_: " << ret->communication_without_parameter_ << ".";
for (size_t i = 1; i < cost_list.size(); ++i) {
MS_EXCEPTION_IF_NULL(cost_list[i]);
if (cost_list[i]->memory_cost_ >= memory) {
MS_LOG(INFO) << "cost_list " << i << " memory_cost_: " << cost_list[i]->memory_cost_
if (cost_list[i]->computation_cost_ >= memory) {
MS_LOG(INFO) << "cost_list " << i << " computation_cost_: " << cost_list[i]->computation_cost_
<< ", is larger than the memory capacity: " << memory << ".";
break;
}
MS_LOG(INFO) << "cost_list " << i << " memory_cost_: " << cost_list[i]->memory_cost_
MS_LOG(INFO) << "cost_list " << i << " computation_cost_: " << cost_list[i]->computation_cost_
<< ", communication_with_partial_para_: " << cost_list[i]->communication_with_partial_para_
<< ", communication_cost_: " << cost_list[i]->communication_cost_
<< ", communication_without_parameter_: " << cost_list[i]->communication_without_parameter_ << ".";
auto tmp =
costmodel_alpha_ * cost_list[i]->memory_cost_ + costmodel_beta_ * cost_list[i]->communication_with_partial_para_;
auto tmp = costmodel_alpha_ * cost_list[i]->computation_cost_ +
costmodel_beta_ * cost_list[i]->communication_with_partial_para_;
MS_LOG(INFO) << "tmp: " << tmp;
if (minimum > tmp) {
minimum = tmp;
@ -363,8 +363,8 @@ CostPtrList CostGraph::SelectCostListWithMinTrainingTimeMultiple(const std::vect
MS_LOG(ERROR) << "The cost list " << i << " is empty.";
return ret;
} else {
total_memory += all_cost_list[i][0]->memory_cost_;
minimum += costmodel_alpha_ * all_cost_list[i][0]->memory_cost_ +
total_memory += all_cost_list[i][0]->computation_cost_;
minimum += costmodel_alpha_ * all_cost_list[i][0]->computation_cost_ +
costmodel_beta_ * all_cost_list[i][0]->communication_with_partial_para_;
ret[i] = all_cost_list[i][0];
}
@ -381,8 +381,8 @@ CostPtrList CostGraph::SelectCostListWithMinTrainingTimeMultiple(const std::vect
double tmp_memory = 0.0, tmp_minimum = 0.0;
for (size_t i = 0; i < selected_cost_list.size(); ++i) {
MS_EXCEPTION_IF_NULL(selected_cost_list[i]);
tmp_memory += selected_cost_list[i]->memory_cost_;
tmp_minimum += costmodel_alpha_ * selected_cost_list[i]->memory_cost_ +
tmp_memory += selected_cost_list[i]->computation_cost_;
tmp_minimum += costmodel_alpha_ * selected_cost_list[i]->computation_cost_ +
costmodel_beta_ * selected_cost_list[i]->communication_with_partial_para_;
}
MS_LOG(INFO) << "tmp_memory: " << tmp_memory << ", tmp_minimum: " << tmp_minimum << ", minimum: " << minimum
@ -394,6 +394,7 @@ CostPtrList CostGraph::SelectCostListWithMinTrainingTimeMultiple(const std::vect
}
return;
}
MS_LOG(DEBUG) << "The value minimum: " << minimum << ", available_memory: " << available_memory << ".";
for (auto& c : all_cost_list[k]) {
selected_cost_list[k] = c;
@ -814,7 +815,7 @@ void CostGraph::CreateMergeEliminationSubCostList(StrategyPtr op_strategy, const
for (size_t k = 0; k < tar_cost_list.size(); ++k) {
auto& tar_cost = tar_cost_list[k];
MS_EXCEPTION_IF_NULL(tar_cost);
double memory = op_cost->memory_cost_ + edge_cost->memory_cost_ + tar_cost->memory_cost_;
double computation = op_cost->computation_cost_ + edge_cost->computation_cost_ + tar_cost->computation_cost_;
double communication =
op_cost->communication_cost_ + edge_cost->communication_cost_ + tar_cost->communication_cost_;
double communication_without_para = op_cost->communication_without_parameter_ +
@ -823,7 +824,7 @@ void CostGraph::CreateMergeEliminationSubCostList(StrategyPtr op_strategy, const
auto decision =
std::make_shared<MergeEliminationDecision>(op_strategy, op_cost, edge_cost, tar_op_strategy, tar_cost);
auto new_cost = std::make_shared<Cost>(memory, communication, decision);
auto new_cost = std::make_shared<Cost>(computation, communication, decision);
MS_EXCEPTION_IF_NULL(new_cost);
new_cost->communication_without_parameter_ = communication_without_para;
new_cost->communication_with_partial_para_ =
@ -891,7 +892,8 @@ void CostGraph::CreateContractEliminationSubCostList(StrategyPtr contract_op_str
for (size_t k = 0; k < tar_cost_list.size(); ++k) {
auto& tar_cost = tar_cost_list[k];
MS_EXCEPTION_IF_NULL(tar_cost);
double memory = contract_op_cost->memory_cost_ + edge_cost->memory_cost_ + tar_cost->memory_cost_;
double computation =
contract_op_cost->computation_cost_ + edge_cost->computation_cost_ + tar_cost->computation_cost_;
double communication =
contract_op_cost->communication_cost_ + edge_cost->communication_cost_ + tar_cost->communication_cost_;
double communication_without_para = contract_op_cost->communication_without_parameter_ +
@ -900,7 +902,7 @@ void CostGraph::CreateContractEliminationSubCostList(StrategyPtr contract_op_str
auto decision = std::make_shared<ContractEliminationDecision>(contract_op_stra, contract_op_cost, edge_cost,
target_op_stra, tar_cost);
auto new_cost = std::make_shared<Cost>(memory, communication, decision);
auto new_cost = std::make_shared<Cost>(computation, communication, decision);
new_cost->communication_without_parameter_ = communication_without_para;
new_cost->communication_with_partial_para_ =
communication_without_para + COST_MODEL_GAMMA * (communication - communication_without_para);
@ -963,9 +965,9 @@ void CostGraph::CreateTriangleEliminationSubCostList(StrategyPtr elimi_op_stra,
MS_EXCEPTION_IF_NULL(left_edge_cost);
for (auto& left_node_cost : left_node_clist_origin) {
MS_EXCEPTION_IF_NULL(left_node_cost);
double new_memory_cost = elimi_op_cost->memory_cost_ + left_edge_cost->memory_cost_ +
left_node_cost->memory_cost_ + right_edge_cost->memory_cost_ +
right_op_cost->memory_cost_;
double new_computation = elimi_op_cost->computation_cost_ + left_edge_cost->computation_cost_ +
left_node_cost->computation_cost_ + right_edge_cost->computation_cost_ +
right_op_cost->computation_cost_;
double new_commu_cost = elimi_op_cost->communication_cost_ + left_edge_cost->communication_cost_ +
left_node_cost->communication_cost_ + right_edge_cost->communication_cost_ +
right_op_cost->communication_cost_;
@ -977,7 +979,7 @@ void CostGraph::CreateTriangleEliminationSubCostList(StrategyPtr elimi_op_stra,
auto decision =
std::make_shared<TriangleEliminationDecision>(elimi_op_stra, elimi_op_cost, left_edge_cost, right_edge_cost,
left_op_stra, left_node_cost, right_op_stra, right_op_cost);
auto new_cost = std::make_shared<Cost>(new_memory_cost, new_commu_cost, decision);
auto new_cost = std::make_shared<Cost>(new_computation, new_commu_cost, decision);
new_cost->communication_without_parameter_ = new_commu_without;
new_cost->communication_with_partial_para_ =
new_commu_without + COST_MODEL_GAMMA * (new_commu_cost - new_commu_without);
@ -1082,11 +1084,12 @@ void CostGraph::CreateStarEliminationSubCostList(const StrategyPtr& first_succ_n
succ_edges_costs[0] = first_succ_edge_cost;
succ_nodes_costs[0] = first_succ_node_cost;
double memory_cost = merged_node_cost->memory_cost_, commu_cost = merged_node_cost->communication_cost_,
double computation_cost = merged_node_cost->computation_cost_,
commu_cost = merged_node_cost->communication_cost_,
commu_without = merged_node_cost->communication_without_parameter_;
for (size_t i = 0; i < succ_nodes_stras.size(); ++i) {
MS_EXCEPTION_IF_NULL(succ_edges_costs[i]);
memory_cost += succ_edges_costs[i]->memory_cost_ + succ_nodes_costs[i]->memory_cost_;
computation_cost += succ_edges_costs[i]->computation_cost_ + succ_nodes_costs[i]->computation_cost_;
commu_cost += succ_edges_costs[i]->communication_cost_ + succ_nodes_costs[i]->communication_cost_;
commu_without += succ_edges_costs[i]->communication_without_parameter_ +
succ_nodes_costs[i]->communication_without_parameter_;
@ -1094,7 +1097,7 @@ void CostGraph::CreateStarEliminationSubCostList(const StrategyPtr& first_succ_n
auto decision = std::make_shared<StarEliminationDecision>(merged_op_stra, merged_node_cost, succ_edges_costs,
succ_nodes_stras, succ_nodes_costs);
auto new_cost = std::make_shared<Cost>(memory_cost, commu_cost, decision);
auto new_cost = std::make_shared<Cost>(computation_cost, commu_cost, decision);
new_cost->communication_without_parameter_ = commu_without;
new_cost->communication_with_partial_para_ = commu_without + COST_MODEL_GAMMA * (commu_cost - commu_without);
first_succ_node_clist_new->emplace_back(std::move(new_cost));
@ -1210,36 +1213,6 @@ Status CostGraph::InitSelectedStrategy() {
return SUCCESS;
}
Status CostGraph::CorrectOpsStrategyCostForMultiOutputUse() {
for (auto& op : ops_) {
MS_EXCEPTION_IF_NULL(op);
if (op->GetAliveSuccEdges().size() > 1) {
// Filter out the case of a output being used by multiple operators
std::map<size_t, int> output_count;
for (size_t i = 0; i < op->GetAliveSuccEdges().size(); ++i) {
auto output_index = op->GetAliveSuccEdges()[i]->prev_op_output_index();
output_count[output_index]++;
}
for (size_t i = 0; i < op->GetAliveSuccEdges().size(); ++i) {
auto output_index = op->GetAliveSuccEdges()[i]->prev_op_output_index();
if (output_count[output_index] <= 1) {
continue;
}
auto next_op = op->GetAliveSuccEdges()[i]->next_operator();
MS_EXCEPTION_IF_NULL(next_op);
auto input_index = op->GetAliveSuccEdges()[i]->next_op_input_index();
if (next_op->CorrectStrategyCostForMultiOutputUse(input_index) != SUCCESS) {
MS_LOG(ERROR) << "The operator name: " << op->name() << ", the next operator name: " << next_op->name()
<< ", the output_index: " << output_index << ", the input_index: " << input_index << ".";
return FAILED;
}
output_count[output_index]--;
}
}
}
return SUCCESS;
}
Status CostGraph::ComputeOpsAndEdgesParameterInvolved() {
for (auto& op : ops_) {
MS_EXCEPTION_IF_NULL(op);
@ -1252,23 +1225,23 @@ Status CostGraph::ComputeOpsAndEdgesParameterInvolved() {
return SUCCESS;
}
Status CostGraph::CorrectOpsStrategyCostForMemoryReuse() {
Status CostGraph::CalculateOpsMemoryCost() {
for (auto& op : ops_) {
MS_EXCEPTION_IF_NULL(op);
if (op->CorrectStrategyCostForMemoryReuse() != SUCCESS) {
MS_LOG(ERROR) << "Correcting Operator: " << op->name() << " cost for memory reuse failed.";
if (op->CalculateMemoryCost() != SUCCESS) {
MS_LOG(ERROR) << "Calculate Operator: " << op->name() << " cost for memory usage failed.";
return FAILED;
}
}
return SUCCESS;
}
Status CostGraph::CorrectEdgesStrategyCostForMemoryReuse() {
Status CostGraph::CalculateEdgesMemoryCost() {
for (auto& edge_pair : edges_) {
const auto& edges = edge_pair.second;
for (auto& one_edge : edges) {
if (one_edge->CorrectStrategyCostForMemoryReuse() != SUCCESS) {
MS_LOG(ERROR) << "Correcting Edge: " << one_edge->edge_name() << " cost for memory reuse failed.";
if (one_edge->CalculateMemoryCost() != SUCCESS) {
MS_LOG(ERROR) << "Calculate Edge: " << one_edge->edge_name() << " cost for memory usage failed.";
return FAILED;
}
}

8
mindspore/ccsrc/parallel/auto_parallel/graph_costmodel.h
View File

@ -175,16 +175,12 @@ class CostGraph {
void CreateStarEliminationSubCostList(const StrategyPtr&, const CostPtrList&, const CostPtrList&, const StrategyPtr&,
const CostPtrList&, std::vector<StrategyPtr>, CostPtrList&, CostPtrList&,
CostPtrList*);
// When a output of a operator is being used by multiple operators, the memory cost of this part should be calculated
// only once. This method is for correcting the 'strategy_cost_' for operators
Status CorrectOpsStrategyCostForMultiOutputUse();
// When the input of a operator is neither a WEIGHT, nor a output of a subsequent operator involving WEIGHT, then
// the memory cost can be resused.
Status CorrectOpsStrategyCostForMemoryReuse();
Status CalculateOpsMemoryCost();
// When the input of the edge is neither a WEIGHT, nor a output of a subsequent operator involving WEIGHT, then
// the memory cost can be resused.
Status CorrectEdgesStrategyCostForMemoryReuse();
Status CalculateEdgesMemoryCost();
Status ComputeOpsAndEdgesParameterInvolved();
std::vector<OperatorInfoPtr> GetOperators() const { return ops_; }

114
mindspore/ccsrc/parallel/auto_parallel/operator_costmodel.cc
View File

@ -74,8 +74,8 @@ double MatMulCost::GetBackwardCommCost(const std::vector<TensorInfo>& inputs, co
// Return the per device memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double MatMulCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t&) const {
double MatMulCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>& outputs, const int32_t&) const {
// In forward phase, the memory cost = slice(A) + slice(B) + (0 or 1) allreduce(slice(C))
double result = 0.0;
TensorInfo output0 = outputs[0];
@ -93,8 +93,8 @@ double MatMulCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, c
// Return the per device memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double MatMulCost::GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t& stage_id) const {
double MatMulCost::GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t& stage_id) const {
// In backward phase, the memory cost = (0 or 1) allreduce(slice(B))
double result = 0.0;
if (is_parameter_[1]) {
@ -147,8 +147,8 @@ double ActivationCost::GetBackwardCommCost(const std::vector<TensorInfo>& inputs
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double ActivationCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
double ActivationCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
TensorInfo input0_info = inputs[0];
Shape input0_slice_shape = input0_info.slice_shape();
return ListProduct(input0_slice_shape) * static_cast<double>(inputs_type_lengths_[0]);
@ -156,8 +156,8 @@ double ActivationCost::GetForwardMemoryCost(const std::vector<TensorInfo>& input
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double ActivationCost::GetBackwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const {
double ActivationCost::GetBackwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const {
return 0.0;
}
@ -191,8 +191,8 @@ double SoftmaxCost::GetBackwardCommCost(const std::vector<TensorInfo>& inputs, c
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double SoftmaxCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
double SoftmaxCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
// In the forward phase, the memory cost = slice(A)
TensorInfo input0 = inputs[0];
Shape input0_slice_shape = input0.slice_shape();
@ -201,8 +201,9 @@ double SoftmaxCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs,
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double SoftmaxCost::GetBackwardMemoryCost(const std::vector<mindspore::parallel::TensorInfo>&,
const std::vector<mindspore::parallel::TensorInfo>&, const int32_t&) const {
double SoftmaxCost::GetBackwardComputationCost(const std::vector<mindspore::parallel::TensorInfo>&,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
return 0.0;
}
@ -222,9 +223,9 @@ double TmpIdentityCost::GetBackwardCommCost(const std::vector<mindspore::paralle
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double TmpIdentityCost::GetForwardMemoryCost(const std::vector<mindspore::parallel::TensorInfo>& inputs,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
double TmpIdentityCost::GetForwardComputationCost(const std::vector<mindspore::parallel::TensorInfo>& inputs,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
TensorInfo input0_info = inputs[0];
Shape input0_slice_shape = input0_info.slice_shape();
return ListProduct(input0_slice_shape) * static_cast<double>(inputs_type_lengths_[0]);
@ -232,15 +233,15 @@ double TmpIdentityCost::GetForwardMemoryCost(const std::vector<mindspore::parall
// Return the per memory cost in the backward phase. The cost is calculated according to the bytes
// this operator uses
double TmpIdentityCost::GetBackwardMemoryCost(const std::vector<mindspore::parallel::TensorInfo>&,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
double TmpIdentityCost::GetBackwardComputationCost(const std::vector<mindspore::parallel::TensorInfo>&,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
return 0.0;
}
double BatchParallelCost::GetForwardMemoryCost(const std::vector<mindspore::parallel::TensorInfo>& inputs,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
double BatchParallelCost::GetForwardComputationCost(const std::vector<mindspore::parallel::TensorInfo>& inputs,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
double cost = 0.0;
for (size_t i = 0; i < inputs.size(); ++i) {
cost += ListProduct(inputs[i].slice_shape()) * static_cast<double>(inputs_type_lengths_[i]);
@ -248,9 +249,9 @@ double BatchParallelCost::GetForwardMemoryCost(const std::vector<mindspore::para
return cost;
}
double BatchParallelCost::GetBackwardMemoryCost(const std::vector<mindspore::parallel::TensorInfo>&,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
double BatchParallelCost::GetBackwardComputationCost(const std::vector<mindspore::parallel::TensorInfo>&,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
return 0.0;
}
@ -285,8 +286,8 @@ double PReLUCost::GetBackwardCommCost(const std::vector<TensorInfo>& inputs, con
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double PReLUCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
double PReLUCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
// In forward phase, the memory cost = slice(A) + slice(B)
Shape input0_slice_shape = inputs[0].slice_shape();
Shape input1_slice_shape = inputs[1].slice_shape();
@ -297,9 +298,9 @@ double PReLUCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, co
// Return the per memory cost in the backward phase. The cost is calculated according to the bytes
// this operator uses
double PReLUCost::GetBackwardMemoryCost(const std::vector<mindspore::parallel::TensorInfo>& inputs,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t& stage_id) const {
double PReLUCost::GetBackwardComputationCost(const std::vector<mindspore::parallel::TensorInfo>& inputs,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t& stage_id) const {
// In backward phase, the memory cost = (0 or 1) allreduce(slice(B))
double result = 0.0;
if (is_parameter_[1]) {
@ -338,8 +339,8 @@ double OneHotCost::GetBackwardCommCost(const std::vector<TensorInfo>&, const std
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double OneHotCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
double OneHotCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
// In onehot's forward phase, the memory cost = slice(A)
Shape input0_slice_shape = inputs[0].slice_shape();
return ListProduct(input0_slice_shape) * static_cast<double>(inputs_type_lengths_[0]);
@ -347,8 +348,8 @@ double OneHotCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, c
// Return the per memory cost in the backward phase. The cost is calculated according to the bytes
// this operator uses
double OneHotCost::GetBackwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const {
double OneHotCost::GetBackwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const {
return 0.0;
}
@ -368,8 +369,9 @@ double SoftmaxCrossEntropyWithLogitsCost::GetBackwardCommCost(const std::vector<
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double SoftmaxCrossEntropyWithLogitsCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>&, const int32_t&) const {
double SoftmaxCrossEntropyWithLogitsCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>&,
const int32_t&) const {
// In forward phase, the memory cost = slice(A) + slice(B)
Shape input0_slice_shape = inputs[0].slice_shape();
Shape input1_slice_shape = inputs[1].slice_shape();
@ -380,8 +382,9 @@ double SoftmaxCrossEntropyWithLogitsCost::GetForwardMemoryCost(const std::vector
// Return the per memory cost in the backward phase. The cost is calculated according to the bytes
// this operator uses
double SoftmaxCrossEntropyWithLogitsCost::GetBackwardMemoryCost(const std::vector<TensorInfo>&,
const std::vector<TensorInfo>&, const int32_t&) const {
double SoftmaxCrossEntropyWithLogitsCost::GetBackwardComputationCost(const std::vector<TensorInfo>&,
const std::vector<TensorInfo>&,
const int32_t&) const {
return 0.0;
}
@ -409,8 +412,8 @@ double ReshapeCost::GetBackwardCommCost(const std::vector<TensorInfo>&, const st
// Return the per memory cost in the forward phase. The cost is calculated according to the bytes
// this operator uses
double ReshapeCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const {
double ReshapeCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>& outputs, const int32_t& stage_id) const {
CheckGlobalDeviceManager();
MS_EXCEPTION_IF_NULL(g_device_manager);
RankList dev_list = g_device_manager->GetDeviceListByStageId(stage_id);
@ -421,26 +424,27 @@ double ReshapeCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs,
if (tensor_redistribution.ComputeCost() == FAILED) {
MS_LOG(EXCEPTION) << "Failure: tensor_redistribution ComputeCost failed.";
}
return (inputs_type_lengths_[0] * tensor_redistribution.mem_cost());
return (inputs_type_lengths_[0] * tensor_redistribution.computation_cost());
}
// Return the per memory cost in the backward phase. The cost is calculated according to the bytes
// this operator uses
double ReshapeCost::GetBackwardMemoryCost(const std::vector<mindspore::parallel::TensorInfo>&,
const std::vector<mindspore::parallel::TensorInfo>&, const int32_t&) const {
double ReshapeCost::GetBackwardComputationCost(const std::vector<mindspore::parallel::TensorInfo>&,
const std::vector<mindspore::parallel::TensorInfo>&,
const int32_t&) const {
return 0.0;
}
double ArithmeticCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
double ArithmeticCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
double result;
result = ListProduct(inputs[0].slice_shape()) * static_cast<double>(inputs_type_lengths_[0]) +
ListProduct(inputs[1].slice_shape()) * static_cast<double>(inputs_type_lengths_[1]);
return result;
}
double ArithmeticCost::GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t& stage_id) const {
double ArithmeticCost::GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t& stage_id) const {
double result = 0.0;
CheckGlobalDeviceManager();
MS_EXCEPTION_IF_NULL(g_device_manager);
@ -533,15 +537,15 @@ double L2NormalizeCost::GetBackwardCommCost(const std::vector<TensorInfo>& input
return result;
}
double L2NormalizeCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
double L2NormalizeCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t&) const {
TensorInfo input0_info = inputs[0];
Shape input0_slice_shape = input0_info.slice_shape();
return ListProduct(input0_slice_shape) * static_cast<double>(inputs_type_lengths_[0]);
}
double L2NormalizeCost::GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>&,
const int32_t& stage_id) const {
double L2NormalizeCost::GetBackwardComputationCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>&, const int32_t& stage_id) const {
double result = 0.0;
if (is_parameter_[0]) {
@ -618,8 +622,9 @@ double ReduceMethodCost::GetBackwardCommCost(const std::vector<TensorInfo>& inpu
return result;
}
double ReduceMethodCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>& outputs, const int32_t& stage_id) const {
double ReduceMethodCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const {
double result = 0.0;
TensorInfo input0 = inputs[0];
TensorInfo output0 = outputs[0];
@ -640,8 +645,9 @@ double ReduceMethodCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inp
return result;
}
double ReduceMeanCost::GetForwardMemoryCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>& outputs, const int32_t& stage_id) const {
double ReduceMeanCost::GetForwardComputationCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const {
double result = 0.0;
TensorInfo input0 = inputs[0];
TensorInfo output0 = outputs[0];

226
mindspore/ccsrc/parallel/auto_parallel/operator_costmodel.h
View File

@ -65,12 +65,12 @@ class OperatorCost {
virtual double GetBackwardCommCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const = 0;
// per device computation cost
virtual double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const = 0;
virtual double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const = 0;
virtual double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const = 0;
virtual double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const = 0;
virtual double GetForwardComputationCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>& outputs, const int32_t& stage_id) const = 0;
virtual double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs,
const std::vector<TensorInfo>& outputs, const int32_t& stage_id) const = 0;
protected:
// for each input in 'inputs_', there is a bool variable indicating whether that the corresponding input is parameter
@ -96,14 +96,14 @@ class MatMulCost : public OperatorCost {
const int32_t& stage_id) const override;
// per device computation cost
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using MatMulCostPtr = std::shared_ptr<MatMulCost>;
@ -121,14 +121,14 @@ class ActivationCost : public OperatorCost {
const int32_t& stage_id) const override;
double GetBackwardCommCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using ActivationCostPtr = std::shared_ptr<ActivationCost>;
@ -146,14 +146,14 @@ class SoftmaxCost : public OperatorCost {
const int32_t& stage_id) const override;
double GetBackwardCommCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t&) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t&) const override;
};
using SoftmaxCostPtr = std::shared_ptr<SoftmaxCost>;
@ -171,14 +171,14 @@ class TmpIdentityCost : public OperatorCost {
const int32_t& stage_id) const override;
double GetBackwardCommCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using TmpIdentityCostPtr = std::shared_ptr<TmpIdentityCost>;
@ -199,14 +199,14 @@ class BatchParallelCost : public OperatorCost {
const int32_t&) const override {
return 0.0;
}
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using BatchParallelCostPtr = std::shared_ptr<BatchParallelCost>;
@ -227,16 +227,16 @@ class VirtualDatasetCost : public OperatorCost {
const int32_t&) const override {
return 0.0;
}
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
double GetForwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
return 0.0;
}
double GetBackwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
double GetBackwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
return 0.0;
}
};
@ -259,18 +259,18 @@ class GeneratorBaseCost : public OperatorCost {
const int32_t&) const override {
return 0.0;
}
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
// Inputs vector is empty for generator ops.
double GetForwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
double GetForwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
return 0.0;
}
// Generator ops don't have backward steps.
double GetBackwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
double GetBackwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
return 0.0;
}
};
@ -292,14 +292,14 @@ class PReLUCost : public OperatorCost {
const int32_t& stage_id) const override;
// per device computation cost
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using PReLUCostPtr = std::shared_ptr<PReLUCost>;
@ -319,14 +319,14 @@ class OneHotCost : public OperatorCost {
const int32_t& stage_id) const override;
// per device computation cost
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using OneHotCostPtr = std::shared_ptr<OneHotCost>;
@ -346,14 +346,14 @@ class SoftmaxCrossEntropyWithLogitsCost : public OperatorCost {
const int32_t& stage_id) const override;
// per device computation cost
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using SoftmaxCrossEntropyWithLogitsCostPtr = std::shared_ptr<SoftmaxCrossEntropyWithLogitsCost>;
@ -376,16 +376,16 @@ class ReshapeCost : public OperatorCost {
const int32_t& stage_id) const override;
// per device computation cost
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using ReshapeCostPtr = std::shared_ptr<ReshapeCost>;
@ -405,14 +405,14 @@ class ArithmeticCost : public OperatorCost {
double GetBackwardCommCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override;
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using ArithmeticCostPtr = std::shared_ptr<ArithmeticCost>;
@ -431,14 +431,14 @@ class L2NormalizeCost : public OperatorCost {
}
double GetBackwardCommCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using L2NormalizeCostPtr = std::shared_ptr<L2NormalizeCost>;
@ -455,14 +455,14 @@ class ReduceMethodCost : public OperatorCost {
const int32_t& stage_id) const override;
double GetBackwardCommCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardCommCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardCommCost(inputs, outputs, stage_id);
}
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetBackwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
return 0.0;
}
void set_cross_batch(bool cb) { cross_batch_ = cb; }
@ -477,8 +477,8 @@ class ReduceMeanCost : public ReduceMethodCost {
ReduceMeanCost() = default;
~ReduceMeanCost() override = default;
double GetForwardMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
double GetForwardComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override;
};
using ReduceMeanCostPtr = std::shared_ptr<ReduceMeanCost>;
@ -499,18 +499,18 @@ class GetNextCost : public OperatorCost {
const int32_t&) const override {
return 0.0;
}
double GetMemoryCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardMemoryCost(inputs, outputs, stage_id) + GetBackwardMemoryCost(inputs, outputs, stage_id);
double GetComputationCost(const std::vector<TensorInfo>& inputs, const std::vector<TensorInfo>& outputs,
const int32_t& stage_id) const override {
return GetForwardComputationCost(inputs, outputs, stage_id) + GetBackwardComputationCost(inputs, outputs, stage_id);
}
// Inputs vector is empty for generator ops.
double GetForwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
double GetForwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
return 0.0;
}
// Generator ops don't have backward steps.
double GetBackwardMemoryCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
double GetBackwardComputationCost(const std::vector<TensorInfo>&, const std::vector<TensorInfo>&,
const int32_t&) const override {
return 0.0;
}
};

8
mindspore/ccsrc/parallel/ops_info/matmul_info.cc
View File

@ -592,10 +592,10 @@ Status MatMulBase::SetCostUnderStrategy(const mindspore::parallel::StrategyPtr&
int32_t stage_id = strategy->GetInputStage();
// Here, we use the origin outputs_, because we only use the slice size of the output tensor.
// It does not matter whether the output tensor is transposed or not.
double memory_cost =
matmulcost_ptr->GetForwardMemoryCost(relica_inputs_tensor_vector, outputs_tensor_info_, stage_id);
double computation_cost =
matmulcost_ptr->GetForwardComputationCost(relica_inputs_tensor_vector, outputs_tensor_info_, stage_id);
double communication_cost = matmulcost_ptr->GetCommCost(relica_inputs_tensor_vector, outputs_tensor_info_, stage_id);
std::shared_ptr<Cost> result = std::make_shared<Cost>(memory_cost, communication_cost);
std::shared_ptr<Cost> result = std::make_shared<Cost>(computation_cost, communication_cost);
result->communication_without_parameter_ =
matmulcost_ptr->GetForwardCommCost(relica_inputs_tensor_vector, outputs_tensor_info_, stage_id);
result->communication_with_partial_para_ =
@ -604,7 +604,7 @@ Status MatMulBase::SetCostUnderStrategy(const mindspore::parallel::StrategyPtr&
// Breaking ties for preferring data parallelization
BreakingTiesForPerferringDataParallel(strategy, result);
MS_LOG(DEBUG) << name_ << " : memory_cost: " << result->memory_cost_
MS_LOG(DEBUG) << name_ << " : computation_cost: " << result->computation_cost_
<< ", communication_cost: " << result->communication_cost_
<< ", communication_without_parameter_: " << result->communication_without_parameter_
<< ", communication_with_partial_para_: " << result->communication_with_partial_para_;

25
mindspore/ccsrc/parallel/ops_info/operator_info.cc
View File

@ -1034,9 +1034,10 @@ Status OperatorInfo::SetCostUnderStrategyBase(const StrategyPtr& strategy) {
return FAILED;
}
int32_t stage_id = strategy->GetInputStage();
double memory_cost = GetOperatorCost()->GetForwardMemoryCost(inputs_tensor_info_, outputs_tensor_info_, stage_id);
double computation_cost =
GetOperatorCost()->GetForwardComputationCost(inputs_tensor_info_, outputs_tensor_info_, stage_id);
double communication_cost = GetOperatorCost()->GetCommCost(inputs_tensor_info_, outputs_tensor_info_, stage_id);
std::shared_ptr<Cost> result = std::make_shared<Cost>(memory_cost, communication_cost);
std::shared_ptr<Cost> result = std::make_shared<Cost>(computation_cost, communication_cost);
result->communication_without_parameter_ =
GetOperatorCost()->GetForwardCommCost(inputs_tensor_info_, outputs_tensor_info_, stage_id);
result->communication_with_partial_para_ =
@ -1056,22 +1057,6 @@ Status OperatorInfo::SetCostUnderStrategyBase(const StrategyPtr& strategy) {
return SUCCESS;
}
Status OperatorInfo::CorrectStrategyCostForMultiOutputUse(size_t input_index) {
for (auto& swc : strategy_cost_) {
double parameter_memory_cost = ListProduct(swc->inputs_ptr[input_index].slice_shape()) *
static_cast<double>(GetOperatorCost()->inputs_type_lengths()[input_index]);
// remove the parameter memory cost
swc->cost_list[0]->memory_cost_ -= parameter_memory_cost;
if (swc->cost_list[0]->memory_cost_ < -1) {
MS_LOG(ERROR) << "The memory cost after correction is " << swc->cost_list[0]->memory_cost_
<< ", the parameter_memory_cost is " << parameter_memory_cost;
return FAILED;
}
}
corrected_input_indices_.push_back(input_index);
return SUCCESS;
}
int OperatorInfo::ComputeOpAndPrevEdgeParameterInvolved() {
if (is_output_parameter_involve_ != -1) {
return is_output_parameter_involve_;
@ -1217,7 +1202,7 @@ void OperatorInfo::BreakingTiesForPerferringDataParallel(const StrategyPtr& stra
CheckGlobalDeviceManager();
auto total_device_num = g_device_manager->GetDeviceListByStageId(stra->GetInputStage()).size();
if (IntToSize(stra->GetInputDim()[0][0]) == total_device_num) {
cost->memory_cost_ -= 1.0;
cost->computation_cost_ -= 1.0;
cost->communication_cost_ -= 1.0;
cost->communication_with_partial_para_ -= 1.0;
cost->communication_without_parameter_ -= 1.0;
@ -1226,7 +1211,7 @@ void OperatorInfo::BreakingTiesForPerferringDataParallel(const StrategyPtr& stra
}
double OperatorInfo::GetForwardMemoryCostFromCNode() {
return GetOperatorCost()->GetForwardMemoryCost(inputs_tensor_info_, outputs_tensor_info_, 0);
return GetOperatorCost()->GetForwardComputationCost(inputs_tensor_info_, outputs_tensor_info_, 0);
}
} // namespace parallel

6
mindspore/ccsrc/parallel/ops_info/operator_info.h
View File

@ -87,13 +87,9 @@ class OperatorInfo {
// is checked
Status SetCostUnderStrategyBase(const StrategyPtr& strategy);
std::vector<std::shared_ptr<StrategyWithCost>> GetStrategyCost() { return strategy_cost_; }
// In the case of a Parameter (or a output) being used by multiple operators, the memory cost induced by
// the parameter (or a output) should be calculated only once. This method is used to
// remove this part from the 'strategy_cost_'.
Status CorrectStrategyCostForMultiOutputUse(size_t input_index);
// When the input of a operator contains WEIGHT or a output from other operators involving WEIGHT, then these input
// should stay in memory until it is used in the backward phase, which is kept in memory at the end of forward phase.
Status CorrectStrategyCostForMemoryReuse() const { return SUCCESS; }
Status CalculateMemoryCost() const { return SUCCESS; }
int ComputeOpAndPrevEdgeParameterInvolved();
ForwardOp forward_op() const { return forward_op_; }

30
mindspore/ccsrc/parallel/step_auto_parallel.cc
View File

@ -387,7 +387,7 @@ OperatorInfoPtr CreateTheOperatorInfo(const PrimitivePtr &prim, const CNodePtr &
operator_info->set_outputs_dtype(cnode->Type());
operator_info->set_cnode(cnode);
// If no strategy has been configured for this operator, then candidate strategies are generated for
// auto-strategy searchingm if this primitive is Cast, we ignore the user-specified strategy
// auto-strategy searching; if this primitive is CAST, we ignore the user-specified strategy
if (!StrategyFound(attrs) || prim->name() == CAST) {
// Compute split_flag_list_, indicating which input has batch dimension. This is ONLY used for preparation for
// BatchParallelInfo operator
@ -600,13 +600,7 @@ void ConstructCostGraphEdges(const std::vector<AnfNodePtr> &all_nodes) {
}
MS_LOG(INFO) << "Successfully created " << edge_count << " edges for: " << cnode->operator_info()->name();
}
// For the case of a output being used by multiple subsequent operators, the output induced memory cost should be
// calculated only once. This method is for correct the operators' memory cost calculation.
if (entire_costgraph->CorrectOpsStrategyCostForMultiOutputUse() != SUCCESS) {
MS_LOG(EXCEPTION) << "Correcting strategy_cost_ for operators failed.";
} else {
MS_LOG(INFO) << "Correcting strategy_cost_ for operators succeeded.";
}
MS_LOG(INFO) << "Constructing edges for cost graph ends.";
}
@ -803,14 +797,6 @@ void AugmentCostGraph(const std::vector<AnfNodePtr> &all_nodes) {
std::shared_ptr<Edge> edge_ptr = std::make_shared<Edge>(
edge_name, tmp_identity_ptr, target_cnode->operator_info(), 0, input_index - 1, false, true);
// Correct the memory calculation for a parameter being used by multiple operators. The parameter is calculated
// only once
if (target_cnode->operator_info()->CorrectStrategyCostForMultiOutputUse(IntToSize(input_index - 1)) != SUCCESS) {
MS_LOG(EXCEPTION) << "Correcting strategy_cost_ failed : " << prim->name();
} else {
MS_LOG(INFO) << "Correcting strategy_cost_ succeeded. " << prim->name();
}
if (edge_ptr->InitEdgeCost() != SUCCESS) {
MS_LOG(EXCEPTION) << "Edge cost initialization failed";
}
@ -840,7 +826,7 @@ Status ParallelStrategySearch(const std::vector<AnfNodePtr> &all_nodes, const Fu
// taking care for the case of a single Parameter being used by multiple operators. Create a TmpIdentity
// operator for this Parameter, and add an edge for the use of this Parameter by each
// subsequent operator;
// Step 3.1: Correct the memory calculation for memory reuse
// Step 3.1: Calculate memory usage
// Step 4: Run the Dynamic Programming algorithm:
// in this process, cost is calculated based on not only the operators, but also the edges. Here, the edge
// cost is caused by the redistribution of a operator's output tensor layout to the next operator's input
@ -867,14 +853,14 @@ Status ParallelStrategySearch(const std::vector<AnfNodePtr> &all_nodes, const Fu
MS_LOG(INFO) << "After the augmenting procedure, there are " << entire_costgraph->GetOperators().size()
<< " operators, and " << entire_costgraph->GetNumPairs() << " edges.";
// Step 3.1: Correcting calculation for memory reuse
// Step 3.1: Calculate the memory usage
if (entire_costgraph->ComputeOpsAndEdgesParameterInvolved() == SUCCESS) {
// Correcting operators' memory usage
if (entire_costgraph->CorrectOpsStrategyCostForMemoryReuse() != SUCCESS) {
// Calculate operators' memory usage
if (entire_costgraph->CalculateOpsMemoryCost() != SUCCESS) {
MS_LOG(EXCEPTION) << "Correcting operators' cost for memory reuse failed.";
}
// Correcting edges' memory usage
if (entire_costgraph->CorrectEdgesStrategyCostForMemoryReuse() != SUCCESS) {
// Calculate edges' memory usage
if (entire_costgraph->CalculateEdgesMemoryCost() != SUCCESS) {
MS_LOG(EXCEPTION) << "Correcting edges' cost for memory reuse failed.";
}
} else {

24
mindspore/ccsrc/parallel/tensor_layout/tensor_redistribution.cc
View File

@ -144,7 +144,7 @@ Status TensorRedistribution::ComputeCost() {
MS_LOG(ERROR) << "Failure: InferTensorRedistribution failed";
return Status::FAILED;
}
// Compute redistribution communication cost and memory cost
// Compute redistribution communication cost and computation cost
for (auto& op_cost : operator_list_) {
OperatorR op = op_cost.first;
Shape slice_shape = op_cost.second;
@ -154,14 +154,14 @@ Status TensorRedistribution::ComputeCost() {
if (str == PERMUTE_BY_AXIS) {
// The shape does not change after PermuteByAxis operation.
// communication cost = all_to_all + all_to_all = 2 * slice_shape
// memory cost = slice_shape
// computation cost = slice_shape
forward_comm_cost_ += prod;
backward_comm_cost_ += prod;
comm_cost_ += 2.0 * prod;
mem_cost_ += prod;
computation_cost_ += prod;
} else if (str == CONCAT_BY_AXIS) {
// communication cost = all_gather + reduce_scatter = before_slice_shape + after_slice_shape
// memory cost = before_slice_shape
// computation cost = before_slice_shape
if (op.second.size() < 3) {
MS_LOG(ERROR) << "op.second size should not be less than 3!";
return Status::FAILED;
@ -173,22 +173,22 @@ Status TensorRedistribution::ComputeCost() {
comm_cost_ += prod * (dev_num + 1.0);
int32_t concat_dim = op.second[0];
if (concat_dim == 0) {
// memory cost = all_gather
mem_cost_ += prod;
// computation cost = all_gather
computation_cost_ += prod;
} else {
// memory cost = all_gather + split + concat
mem_cost_ += (prod + prod * dev_num + prod * dev_num);
// computation cost = all_gather + split + concat
computation_cost_ += (prod + prod * dev_num + prod * dev_num);
}
} else {
// There is only memory cost in SplitByAxis.
// memory cost = before_slice_shape
mem_cost_ += prod;
// There is only computation cost in SplitByAxis.
// computation cost = before_slice_shape
computation_cost_ += prod;
}
}
if (reshape_flag()) {
Shape prev_slice_shape = from_.slice_shape().array();
double prev_prod = std::accumulate(prev_slice_shape.begin(), prev_slice_shape.end(), 1, std::multiplies<int>());
mem_cost_ += 2.0 * prev_prod;
computation_cost_ += 2.0 * prev_prod;
}
return Status::SUCCESS;
}

9
mindspore/ccsrc/parallel/tensor_layout/tensor_redistribution.h
View File

@ -41,7 +41,7 @@ class TensorRedistribution {
comm_cost_(0.0),
forward_comm_cost_(0.0),
backward_comm_cost_(0.0),
mem_cost_(0.0),
computation_cost_(0.0),
construct_op_flag_(construct_op_flag),
keep_reshape_(keep_reshape) {}
Status Init(const TensorLayout& from, const TensorLayout& to, const RankList& dev_list);
@ -51,7 +51,7 @@ class TensorRedistribution {
bool reshape_flag() const { return reshape_flag_; }
Status ComputeCost();
double comm_cost() const { return comm_cost_; }
double mem_cost() const { return mem_cost_; }
double computation_cost() const { return computation_cost_; }
double forward_comm_cost() const { return forward_comm_cost_; }
double backward_comm_cost() const { return backward_comm_cost_; }
@ -66,10 +66,13 @@ class TensorRedistribution {
RankList dev_list_;
OperatorList operator_list_;
bool reshape_flag_;
// communication cost
double comm_cost_;
// forward communication cost
double forward_comm_cost_;
// backward communication cost
double backward_comm_cost_;
double mem_cost_;
double computation_cost_;
bool construct_op_flag_;
bool keep_reshape_;
};

4
tests/ut/cpp/parallel/auto_parallel/graph_costmodel_test.cc
View File

@ -322,8 +322,8 @@ TEST_F(TestCostGraph, test_SelectCostListWithMinTrainingTimeMultiple) {
auto ret_list = entire_cost_graph.SelectCostListWithMinTrainingTimeMultiple(all_list, memory);
ASSERT_EQ(ret_list.size(), 2);
ASSERT_DOUBLE_EQ(ret_list[0]->memory_cost_, 10);
ASSERT_DOUBLE_EQ(ret_list[1]->memory_cost_, 1010);
ASSERT_DOUBLE_EQ(ret_list[0]->computation_cost_, 10);
ASSERT_DOUBLE_EQ(ret_list[1]->computation_cost_, 1010);
}
TEST_F(TestCostGraph, test_CheckOpElimination) {

12
tests/ut/cpp/parallel/auto_parallel/operator_costmodel_test.cc
View File

@ -76,8 +76,8 @@ TEST_F(TestMatMulCost, test_CostGeneration) {
mmcost_.SetInputAndOutputTypeLength(inputs_length, outputs_length);
mmcost_.GetForwardCommCost(inputs, outputs, 0);
mmcost_.GetBackwardCommCost(inputs, outputs, 0);
mmcost_.GetForwardMemoryCost(inputs, outputs, 0);
mmcost_.GetBackwardMemoryCost(inputs, outputs, 0);
mmcost_.GetForwardComputationCost(inputs, outputs, 0);
mmcost_.GetForwardComputationCost(inputs, outputs, 0);
}
class TestActivationCost : public UT::Common {
@ -128,8 +128,8 @@ TEST_F(TestActivationCost, test_CostGeneration) {
std::vector<size_t> inputs_length = {4, 4};
std::vector<size_t> outputs_length = {4};
ac_cost_.SetInputAndOutputTypeLength(inputs_length, outputs_length);
ac_cost_.GetForwardMemoryCost(inputs, outputs, 0);
ac_cost_.GetBackwardMemoryCost(inputs, outputs, 0);
ac_cost_.GetForwardComputationCost(inputs, outputs, 0);
ac_cost_.GetBackwardComputationCost(inputs, outputs, 0);
}
class TestPReLUCost : public UT::Common {
@ -184,8 +184,8 @@ TEST_F(TestPReLUCost, test_CostGeneration) {
prelu_cost_.SetInputAndOutputTypeLength(inputs_length, outputs_length);
double BCC, FMC, GMC;
BCC = prelu_cost_.GetBackwardCommCost(inputs, outputs, 0);
FMC = prelu_cost_.GetForwardMemoryCost(inputs, outputs, 0);
GMC = prelu_cost_.GetBackwardMemoryCost(inputs, outputs, 0);
FMC = prelu_cost_.GetForwardComputationCost(inputs, outputs, 0);
GMC = prelu_cost_.GetBackwardComputationCost(inputs, outputs, 0);
ASSERT_EQ(BCC, 32 * 4);
ASSERT_EQ(FMC, 8 * 32 * 8 * 8 * 4 + 32 * 4);
ASSERT_EQ(GMC, 128);

8
tests/ut/cpp/parallel/ops_info/activation_test.cc
View File

@ -84,8 +84,8 @@ TEST_F(TestActivation, test_activation_strategies) {
act_ptr_->InitForCostModel(sp);
std::vector<TensorInfo> inputs_info = act_ptr_->inputs_tensor_info();
std::vector<TensorInfo> outputs_info = act_ptr_->outputs_tensor_info();
ASSERT_DOUBLE_EQ(act_ptr_->GetOperatorCost()->GetMemoryCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.memory_cost_);
ASSERT_DOUBLE_EQ(act_ptr_->GetOperatorCost()->GetComputationCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.computation_cost_);
ASSERT_DOUBLE_EQ(act_ptr_->GetOperatorCost()->GetCommCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.communication_cost_);
}
@ -109,8 +109,8 @@ TEST_F(TestActivation, test_softmax_strategies) {
soft_ptr_->InitForCostModel(sp);
std::vector<TensorInfo> inputs_info = soft_ptr_->inputs_tensor_info();
std::vector<TensorInfo> outputs_info = soft_ptr_->outputs_tensor_info();
ASSERT_DOUBLE_EQ(soft_ptr_->GetOperatorCost()->GetMemoryCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.memory_cost_);
ASSERT_DOUBLE_EQ(soft_ptr_->GetOperatorCost()->GetComputationCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.computation_cost_);
ASSERT_DOUBLE_EQ(soft_ptr_->GetOperatorCost()->GetCommCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.communication_cost_);
}

8
tests/ut/cpp/parallel/ops_info/matmul_info_test.cc
View File

@ -569,8 +569,8 @@ TEST_F(TestMatmulInfo, test_GenerateStrategies1) {
matmul1->InitForCostModel(sp);
std::vector<TensorInfo> inputs_info = matmul1->inputs_tensor_info();
std::vector<TensorInfo> outputs_info = matmul1->outputs_tensor_info();
ASSERT_DOUBLE_EQ(matmul1->GetOperatorCost()->GetMemoryCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.memory_cost_);
ASSERT_DOUBLE_EQ(matmul1->GetOperatorCost()->GetComputationCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.computation_cost_);
break;
}
}
@ -599,8 +599,8 @@ TEST_F(TestMatmulInfo, test_GenerateStrategies2) {
TensorInfo replica_input1_info(tly, input1_shape, input1_slice_shape);
replica_inputs_info.push_back(replica_input1_info);
ASSERT_DOUBLE_EQ(matmul3->GetOperatorCost()->GetMemoryCost(replica_inputs_info, outputs_info, sp->GetInputStage()),
cost.memory_cost_);
ASSERT_DOUBLE_EQ(matmul3->GetOperatorCost()->GetComputationCost(replica_inputs_info, outputs_info, sp->GetInputStage()),
cost.computation_cost_);
break;
}
}

8
tests/ut/cpp/parallel/ops_info/tensor_add_info_test.cc
View File

@ -188,8 +188,8 @@ TEST_F(TestTensorAddInfo, GenerateStrategies) {
tensor_add->InitForCostModel(sp);
std::vector<TensorInfo> inputs_info = tensor_add->inputs_tensor_info();
std::vector<TensorInfo> outputs_info = tensor_add->outputs_tensor_info();
double memory_cost0 = tensor_add->GetOperatorCost()->GetMemoryCost(inputs_info, outputs_info, sp->GetInputStage());
double memory_cost1 = cost.memory_cost_;
double memory_cost0 = tensor_add->GetOperatorCost()->GetComputationCost(inputs_info, outputs_info, sp->GetInputStage());
double memory_cost1 = cost.computation_cost_;
bool memory = memory_cost0 - memory_cost1 <= 1.0;
double comm_cost0 = tensor_add->GetOperatorCost()->GetCommCost(inputs_info, outputs_info, sp->GetInputStage());
@ -210,8 +210,8 @@ TEST_F(TestTensorAddInfo, GenerateStrategies1) {
tensor_add1->InitForCostModel(sp);
std::vector<TensorInfo> inputs_info = tensor_add1->inputs_tensor_info();
std::vector<TensorInfo> outputs_info = tensor_add1->outputs_tensor_info();
double memory_cost0 = tensor_add1->GetOperatorCost()->GetMemoryCost(inputs_info, outputs_info, sp->GetInputStage());
double memory_cost1 = cost.memory_cost_;
double memory_cost0 = tensor_add1->GetOperatorCost()->GetComputationCost(inputs_info, outputs_info, sp->GetInputStage());
double memory_cost1 = cost.computation_cost_;
bool memory = memory_cost0 - memory_cost1 <= 1.0;
double comm_cost0 = tensor_add1->GetOperatorCost()->GetCommCost(inputs_info, outputs_info, sp->GetInputStage());

4
tests/ut/cpp/parallel/ops_info/tmpidentity_test.cc
View File

@ -145,8 +145,8 @@ TEST_F(TestTmpIdentityInfo, test_generate_strategies) {
identity_ptr->Init(sp);
std::vector<TensorInfo> inputs_info = identity_ptr->inputs_tensor_info();
std::vector<TensorInfo> outputs_info = identity_ptr->outputs_tensor_info();
ASSERT_DOUBLE_EQ(identity_ptr->GetOperatorCost()->GetMemoryCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.memory_cost_);
ASSERT_DOUBLE_EQ(identity_ptr->GetOperatorCost()->GetComputationCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.computation_cost_);
ASSERT_DOUBLE_EQ(identity_ptr->GetOperatorCost()->GetCommCost(inputs_info, outputs_info, sp->GetInputStage()),
cost.communication_cost_);
}