|
- /**
- * Copyright 2020 Huawei Technologies Co., Ltd
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
- #include "graph/build/memory/block_mem_assigner.h"
- #include <algorithm>
- #include <sstream>
-
- #include "external/ge/ge_api_types.h"
- #include "framework/common/debug/ge_log.h"
- #include "graph/anchor.h"
- #include "graph/buffer.h"
- #include "graph/ge_attr_value.h"
- #include "graph/ge_context.h"
- #include "graph/node.h"
- #include "graph/utils/graph_utils.h"
- #include "graph/utils/node_utils.h"
- #include "graph/utils/op_desc_utils.h"
- #include "graph/utils/tensor_utils.h"
-
- #include "graph/debug/ge_attr_define.h"
-
- #include "graph/common/local_context.h"
- #include "graph/optimize/common/params.h"
- #include "omg/omg_inner_types.h"
- #include "runtime/mem.h"
-
- using std::map;
- using std::set;
- using std::list;
- using std::pair;
- using std::string;
- using std::stringstream;
- using std::unordered_map;
- using std::unordered_set;
- using std::vector;
-
- namespace {
- const char *const kAttrNameWorkspaceReuseFlag = "workspace_reuse_flag";
- const char *const kL2FusionDynamicConvergeOp = "l2fusion_dynamic_converge_op";
- const char *const kOpNoReuseMem = "no_reuse_mem_flag";
- const char *const OP_NO_REUSE_MEM = "OP_NO_REUSE_MEM";
- const int kReuseMaxOpNum = 10;
- const int kReuseMaxCharNum = 2000;
- } // namespace
-
- namespace ge {
- void AlignMemOffset(size_t &mem_align_size) {
- if (mem_align_size <= 0) {
- return;
- }
- mem_align_size = (mem_align_size + MEM_ALIGN_SIZE - 1) / MEM_ALIGN_SIZE * MEM_ALIGN_SIZE;
- }
-
- static bool CompareLifeTime(const NodeTypeIndex &left, const NodeTypeIndex &right) {
- auto left_node_op_desc = left.node->GetOpDesc();
- auto right_node_op_desc = right.node->GetOpDesc();
- if ((left_node_op_desc != nullptr) && (right_node_op_desc != nullptr)
- && (left_node_op_desc->GetId() < right_node_op_desc->GetId())) {
- return true;
- }
- return false;
- }
-
- void GetLifeList(const MemoryBlock &block, std::vector<NodeTypeIndex> &life_list, bool child) {
- for (auto &node : block.NodeTypeIndexList()) {
- life_list.emplace_back(node);
- }
-
- if (child) {
- for (auto child_block : block.ChildBlockList()) {
- if (child_block == nullptr) {
- continue;
- }
- if (block.stream_id_ != child_block->stream_id_ || !block.same_stream_ || !child_block->same_stream_) {
- life_list.clear();
- return;
- }
- GetLifeList(*child_block, life_list, child);
- }
- }
- }
-
- bool CrossLifeTime(const NodeTypeIndex &left, const NodeTypeIndex &right) {
- if ((left.node == nullptr) || (right.node == nullptr)) {
- return true;
- }
- auto left_node_op_desc = left.node->GetOpDesc();
- auto right_node_op_desc = right.node->GetOpDesc();
- if ((left_node_op_desc != nullptr) && (right_node_op_desc != nullptr)) {
- if (left_node_op_desc->GetId() < right_node_op_desc->GetId()) {
- if (left.life_time_end >= static_cast<size_t>(right_node_op_desc->GetId())) {
- return true;
- }
- } else if (left_node_op_desc->GetId() == right_node_op_desc->GetId()) {
- return true;
- } else {
- if (right.life_time_end >= static_cast<size_t>(left_node_op_desc->GetId())) {
- return true;
- }
- }
- }
- return false;
- }
-
- ///
- /// When child block's life time are not cross with parent block, they can be reused(only same stream).
- /// |-----------------------------parent block---------------------|
- /// |------child block1--------------||------child block2------|
- /// |--child block1-1-|
- ///
- bool CanIntervalLifeReuse(MemoryBlock &parent_block, MemoryBlock &child_block) {
- // judge by interval life time, only same stream can be judged by interval life time
- if (parent_block.stream_id_ != child_block.stream_id_ || !parent_block.same_stream_ || !child_block.same_stream_
- || parent_block.NodeTypeIndexList().empty() || child_block.NodeTypeIndexList().empty()) {
- return false;
- }
-
- // quick judge by front and back node
- if (CrossLifeTime(parent_block.NodeTypeIndexList().front(), child_block.NodeTypeIndexList().front())) {
- return false;
- }
- if (CrossLifeTime(parent_block.NodeTypeIndexList().back(), child_block.NodeTypeIndexList().back())) {
- return false;
- }
-
- std::vector<NodeTypeIndex> life_list;
- GetLifeList(parent_block, life_list, false);
- GetLifeList(child_block, life_list, true);
- if (life_list.empty()) {
- return false;
- }
- std::sort(life_list.begin(), life_list.end(), CompareLifeTime);
- size_t pre_life_end = 0;
- for (auto &node : life_list) {
- auto node_op_desc = node.node->GetOpDesc();
- if (node_op_desc != nullptr && pre_life_end >= static_cast<size_t>(node_op_desc->GetId())) {
- // life time cross
- return false;
- }
- pre_life_end = node.life_time_end;
- }
- GELOGI("Block size[%zu, %zu] life time are not cross.", parent_block.Size(), child_block.Size());
- return true;
- }
-
- void MemoryBlock::SetHeadOffset(size_t offset) {
- head_offset_ = offset;
- size_t child_offset = head_offset_;
- for (auto block : child_blocks_) {
- if (block != nullptr) {
- block->SetHeadOffset(child_offset);
- child_offset += block->Size();
- }
- }
- }
-
- void MemoryBlock::SetTailOffset(size_t offset) {
- tail_offset_ = offset;
- size_t child_offset = head_offset_;
- for (auto block : child_blocks_) {
- if (block != nullptr) {
- child_offset += block->Size();
- block->SetTailOffset(child_offset - 1);
- }
- }
- }
-
- void MemoryBlock::Resize() {
- size_t child_block_size = 0;
- for (auto block : child_blocks_) {
- if (block != nullptr) {
- block->Resize();
- child_block_size += block->Size();
- }
- }
- auto iter = std::max_element(real_size_list_.begin(), real_size_list_.end());
- if (iter == real_size_list_.end()) {
- GELOGW("real_size_list_ is empty");
- return;
- } else {
- size_t block_size = (child_block_size > *iter) ? child_block_size : *iter;
- if ((block_size > 0) && (block_size % MEM_ALIGN_SIZE != 0)) {
- AlignMemOffset(block_size);
- }
- block_size_ = block_size;
- if (last_continuous_block_) {
- block_size_ += MEM_ALIGN_SIZE;
- }
- }
- }
-
- size_t MemoryBlock::AlignSize() const {
- size_t align_block_size = 0;
- auto iter = std::max_element(real_size_list_.begin(), real_size_list_.end());
- if (iter == real_size_list_.end()) {
- GELOGW("real_size_list_ is empty");
- } else {
- align_block_size = *iter;
- if ((align_block_size > 0) && (align_block_size % MEM_ALIGN_SIZE != 0)) {
- AlignMemOffset(align_block_size);
- }
- }
- return align_block_size;
- }
-
- bool MemoryBlock::IsSameBatchLabel() {
- // only same batch label can reuse
- if (batch_label_.empty() || node_type_index_list_.empty()) {
- return false;
- }
-
- bool all_same_label = true;
- for (size_t index = 1; index < node_type_index_list_.size(); ++index) {
- if (node_type_index_list_[index].node == nullptr) {
- continue;
- }
- std::string batch_label;
- auto index_op_desc = node_type_index_list_[index].node->GetOpDesc();
- GE_IF_BOOL_EXEC(index_op_desc == nullptr, continue);
- // not all op has ATTR_NAME_BATCH_LABEL, no need check return value, only check out parameter
- (void)ge::AttrUtils::GetStr(index_op_desc, ATTR_NAME_BATCH_LABEL, batch_label);
- if (batch_label_ != batch_label) {
- all_same_label = false;
- break;
- }
- }
- return all_same_label;
- }
-
- bool CanNotLifeReuse(MemoryBlock *block) {
- if ((block == nullptr) || !block->reuse_mem_ || block->deleted_block_) {
- return true;
- }
- return false;
- }
-
- void MemoryBlock::AddContinuousLifeReuseBlock(MemoryBlock *block, DependStreamLife &total_node_depend_stream_life) {
- // continuous memory case:only real_size is maximum can be reused and only one continuous memory in one block
- auto it_block = std::max_element(std::begin(block->NoAlignSizeList()), std::end(block->NoAlignSizeList()));
- auto it_this = std::max_element(std::begin(NoAlignSizeList()), std::end(NoAlignSizeList()));
- if (it_block != std::end(block->NoAlignSizeList()) && it_this != std::end(NoAlignSizeList())) {
- if ((continuous_block_ && block->continuous_block_) ||
- (continuous_block_ && (*it_this < *it_block)) || (block->continuous_block_ && (*it_this > *it_block))) {
- GELOGD("Conflict current block size:%zu continuous:%d, reuse block max size:%zu continuous:%d",
- *it_this, continuous_block_, *it_block, block->continuous_block_);
- return;
- }
- }
-
- MemoryBlock *parent = nullptr;
- MemoryBlock *child = nullptr;
- // merge small block to large block
- if (block->GetDependLifeBegin(stream_id_, total_node_depend_stream_life) > GetLifeEnd()) {
- if ((block->child_offset_ + AlignSize()) <= *it_block) {
- parent = block;
- child = this;
- }
- }
- if ((parent != nullptr) && (child != nullptr) && child->child_blocks_.empty()) {
- parent->child_blocks_.emplace_back(child);
- parent->child_offset_ += child->AlignSize();
- child->deleted_block_ = true;
- GELOGI("Add continuous block[%p size:%zu, stream id:%ld life time[begin:%zu, end:%zu]] to"
- " block[%p size:%zu, stream id:%ld, life time[begin:%zu, end:%zu]]", child, child->block_size_,
- child->stream_id_, child->GetLifeBegin(), child->GetLifeEnd(), parent, parent->block_size_,
- parent->stream_id_, parent->GetLifeBegin(), parent->GetLifeEnd());
- }
- }
-
- void MemoryBlock::AddLifeReuseBlock(MemoryBlock *block, DependStreamLife &total_node_depend_stream_life) {
- if (CanNotLifeReuse(this) || CanNotLifeReuse(block) || (batch_label_ != block->batch_label_)) {
- return;
- }
- if (block->continuous_block_) {
- AddContinuousLifeReuseBlock(block, total_node_depend_stream_life);
- return;
- }
- MemoryBlock *parent = nullptr;
- MemoryBlock *child = nullptr;
- // merge small block to large block
- // noalign size 802816 + 802816 = 1605632 can reuse
- // after 32 align size 802848 + 802848 > 1605664 can't reuse
- // after 512 align size 803328 + 803328 > 1606144 can't reuse
- // so 803328 + 803328 = 1606144 + 512 can reuse
- if ((child_offset_ + block->AlignSize()) <= (AlignSize() + MEM_ALIGN_SIZE)) {
- parent = this;
- child = block;
- } else if ((block->child_offset_ + AlignSize()) <= (block->AlignSize() + MEM_ALIGN_SIZE)) {
- parent = block;
- child = this;
- }
-
- if ((parent != nullptr) && (child != nullptr)) {
- // Different streams must use stream dependency to judge the life cycle
- // In case same stream if it has child block, can judge all the child block's life time in CanIntervalLifeReuse
- bool can_block_life_reuse = (child->child_blocks_.empty()
- && (block->GetDependLifeBegin(stream_id_, total_node_depend_stream_life) > GetLifeEnd()));
- if (!can_block_life_reuse && !CanIntervalLifeReuse(*parent, *child)) {
- return;
- }
-
- parent->child_blocks_.emplace_back(child);
- parent->child_offset_ += child->AlignSize();
- child->deleted_block_ = true;
- GELOGI("Add block[%p size:%zu, stream id:%ld life time[begin:%zu, end:%zu]] to"
- " block[%p size:%zu, stream id:%ld, life time[begin:%zu, end:%zu]]", child, child->block_size_,
- child->stream_id_, child->GetLifeBegin(), child->GetLifeEnd(), parent, parent->block_size_,
- parent->stream_id_, parent->GetLifeBegin(), parent->GetLifeEnd());
- }
- }
-
- size_t MemoryBlock::GetLifeBegin() {
- size_t life_time = 0;
- if (!node_type_index_list_.empty()) {
- if (node_type_index_list_.front().node != nullptr) {
- auto node_op_desc = node_type_index_list_.front().node->GetOpDesc();
- if (node_op_desc != nullptr) {
- life_time = node_op_desc->GetId();
- }
- }
- }
- return life_time;
- }
-
- /// |-stream 1-| |-stream 2-|
- /// |--block1--| |--block---|
- /// |--block2--| |--block---|
- /// |--block3--|\ |--block---|
- /// |--block---| \ |--block---|
- /// |--block---| \|--block---|
- /// |--block---| |--block7--|
- /// |--block---| |--block---|
- /// block7's first node's input node's life begin > block2's life end, block7 can reuse block1~block2
- size_t MemoryBlock::GetDependLifeBegin(int64_t stream_id, DependStreamLife &total_node_depend_stream_life) {
- AddDependLifeBegin(total_node_depend_stream_life);
- auto it = depend_stream_life_.find(stream_id);
- if (it == depend_stream_life_.end()) {
- return 0;
- }
- return it->second;
- }
-
- void AddDependLife(const ge::NodePtr &org_node, const ge::NodePtr &node, int64_t stream_id,
- std::map<int64_t, size_t> &depend_stream_life, DependStreamLife &total_node_depend_stream_life) {
- GE_CHECK_NOTNULL_EXEC(node, return);
- GE_CHECK_NOTNULL_EXEC(org_node, return);
- auto node_desc = node->GetOpDesc();
- GE_CHECK_NOTNULL_EXEC(node_desc, return);
- auto node_id = node_desc->GetId();
- auto stream_life = total_node_depend_stream_life.find(node_id);
- if (stream_life != total_node_depend_stream_life.end()) {
- for (auto &it : stream_life->second) {
- if (depend_stream_life.find(it.first) == depend_stream_life.end()) {
- depend_stream_life[it.first] = it.second;
- }
- }
- return;
- }
-
- for (const auto &in_anchor : node->GetAllInAnchors()) {
- GE_CHECK_NOTNULL_EXEC(in_anchor, continue);
- for (auto peer_out_anchor : in_anchor->GetPeerAnchors()) {
- GE_CHECK_NOTNULL_EXEC(peer_out_anchor, continue);
- auto peer_node = peer_out_anchor->GetOwnerNode();
- GE_CHECK_NOTNULL_EXEC(peer_node, continue);
- auto peer_node_desc = peer_node->GetOpDesc();
- GE_CHECK_NOTNULL_EXEC(peer_node_desc, continue);
- auto peer_node_stream_id = peer_node_desc->GetStreamId();
- if (peer_node_stream_id < 0) {
- continue;
- }
- size_t peer_node_life_time = peer_node_desc->GetId();
- auto it = depend_stream_life.find(peer_node_stream_id);
- if (it == depend_stream_life.end() || peer_node_life_time > it->second) {
- depend_stream_life[peer_node_stream_id] = peer_node_life_time;
- if (peer_node_stream_id != stream_id) {
- GELOGI("Node:%s stream id:%ld depend node:%s stream id:%ld index[%d] life time[%zu].",
- org_node->GetName().c_str(), stream_id, peer_node_desc->GetName().c_str(),
- peer_node_stream_id, peer_out_anchor->GetIdx(), peer_node_life_time);
- }
- AddDependLife(org_node, peer_node, stream_id, depend_stream_life, total_node_depend_stream_life);
- }
- }
- }
-
- // save on node to save next calculation
- for (auto &it : depend_stream_life) {
- if (total_node_depend_stream_life[node_id].find(it.first) == total_node_depend_stream_life[node_id].end()) {
- total_node_depend_stream_life[node_id][it.first] = it.second;
- }
- }
- }
-
- void MemoryBlock::AddDependLifeBegin(DependStreamLife &total_node_depend_stream_life) {
- if (!depend_stream_life_.empty()) {
- return;
- }
- if (!node_type_index_list_.empty()) {
- auto node = node_type_index_list_.front().node;
- if (node != nullptr) {
- AddDependLife(node, node, stream_id_, depend_stream_life_, total_node_depend_stream_life);
- }
- }
- depend_stream_life_[stream_id_] = GetLifeBegin();
- }
-
- size_t MemoryBlock::GetLifeEnd() {
- if (!node_type_index_list_.empty()) {
- return node_type_index_list_.back().life_time_end;
- }
- return kMaxLifeTime;
- }
-
- void MemoryBlock::SetLifeTimeEnd(size_t time) {
- if (!node_type_index_list_.empty()) {
- node_type_index_list_.back().life_time_end = time;
- }
- }
-
- void SetLastUsedInputMemAttr(NodePtr &node, int input_index) {
- if (node == nullptr) {
- return;
- }
- auto node_op_desc = node->GetOpDesc();
- if (node_op_desc != nullptr) {
- auto input_desc = node_op_desc->GetInputDesc(input_index);
- if (!ge::AttrUtils::SetInt(input_desc, ATTR_NAME_IS_END_OF_INPUTMEM_LIFECYCLE, true)) {
- GELOGW("Set %s input[%d] ATTR_NAME_IS_END_OF_INPUTMEM_LIFECYCLE to true failed.", node_op_desc->GetName().c_str(),
- input_index);
- return;
- }
- GELOGD("Set %s input[%d] ATTR_NAME_IS_END_OF_INPUTMEM_LIFECYCLE to true success.", node_op_desc->GetName().c_str(),
- input_index);
- if (node_op_desc->UpdateInputDesc(input_index, input_desc) != GRAPH_SUCCESS) {
- GELOGW("Update %s input[%d] desc failed.", node_op_desc->GetName().c_str(), input_index);
- }
- }
- }
-
- Status GetNoAlignSize(const ge::OpDesc &desc, uint32_t index, size_t &size) {
- // calculate tensor real size
- auto output_op_desc = desc.GetOutputDescPtr(index);
- if (output_op_desc == nullptr) {
- GELOGI("GetNoAlignSize failed. OpName: %s, OpType: %s, index: %d",
- desc.GetName().c_str(), desc.GetType().c_str(), index);
- return FAILED;
- }
- int64_t tensor_size = 0;
- GeShape shape = output_op_desc->GetShape();
- Format format = output_op_desc->GetFormat();
- DataType data_type = output_op_desc->GetDataType();
- graphStatus graph_status = TensorUtils::CalcTensorMemSize(shape, format, data_type, tensor_size);
- if (graph_status != GRAPH_SUCCESS) {
- GELOGE(graph_status, "CalcTensorMemSize failed!");
- return FAILED;
- }
- size = static_cast<size_t>(tensor_size);
- return SUCCESS;
- }
-
- string ToString(ge::NodeTypeIndex &x) {
- stringstream ss;
- ss << "[" << x.node->GetName() << "(" << x.node->GetType() << "), ";
- if (x.mem_type == kOutput) {
- ss << "Output, ";
- } else {
- ss << "Workspace, ";
- }
- ss << x.index << "]";
- return ss.str();
- }
-
- string MemoryBlock::String() {
- stringstream ss;
- ss << "Block size: " << Size() << " from " << HeadOffset() << " to " << TailOffset() << " ";
- ss << "real_size_list: " << ToString(real_size_list_) << " ";
- ss << "ref_count: " << ref_count_ << " ";
- ss << "members: ";
- for (auto x : NodeTypeIndexList()) {
- ss << "__node: " << ToString(x) << " ";
- }
- for (const auto& symbol : SymbolList()) {
- ss << "__symbol: " << symbol << " ";
- }
- ss << "memory_type: " << memory_type_ << " ";
- return ss.str();
- }
-
- BlockMemAssigner::BlockMemAssigner(ComputeGraphPtr compute_graph, const map<string, string> &anchor_to_symbol,
- const map<string, list<NodeIndexIO>> &symbol_to_anchors)
- : mem_offset_(0), p2p_mem_offset_(0), compute_graph_(std::move(compute_graph)),
- symbol_to_anchors_(symbol_to_anchors), anchor_to_symbol_(anchor_to_symbol), life_time_(0) {}
-
- BlockMemAssigner::~BlockMemAssigner() {
- GELOGD("blocks_store_ size : %lu", blocks_store_.size());
- for (MemoryBlock *memory_block : blocks_store_) {
- GE_DELETE_NEW_SINGLE(memory_block);
- }
- }
-
- void GetMaxBatchAllMemorySize(std::map<std::string, vector<int64_t>> &batch_all_memory_size,
- std::map<std::string, int64_t> batch_total_size, vector<int64_t> &all_memory_size,
- std::string &max_batch_label) {
- // use max batch all memory size for reuse range
- int64_t max_batch_size = 0;
- for (const auto &it : batch_total_size) {
- GELOGI("Batch[%s] total memory size[%ld]", it.first.c_str(), it.second);
- // no batch label
- if (it.first.empty()) {
- continue;
- }
- if (it.second > max_batch_size) {
- max_batch_size = it.second;
- max_batch_label = it.first;
- }
- }
- GELOGI("Max batch[%s] total memory size[%ld]", max_batch_label.c_str(), max_batch_size);
-
- for (const auto &it : batch_all_memory_size) {
- if (it.first.empty() || (it.first == max_batch_label)) {
- all_memory_size.insert(all_memory_size.end(), it.second.begin(), it.second.end());
- }
- }
- // all_memory_size can't be empty
- if (all_memory_size.empty()) {
- all_memory_size.emplace_back(MEM_ALIGN_SIZE);
- }
- sort(all_memory_size.begin(), all_memory_size.end());
- GELOGD("All memory size: %s", ToString(all_memory_size).c_str());
-
- for (auto iter = all_memory_size.begin(); iter != all_memory_size.end();) {
- if (*iter == 0) {
- iter = all_memory_size.erase(iter);
- } else {
- ++iter;
- }
- }
- }
-
- void BlockMemAssigner::GetOutAndWorkSpaceMem(vector<int64_t> &all_memory_size) {
- vector<int64_t> temp;
- std::map<std::string, vector<int64_t>> batch_all_memory_size;
- std::map<std::string, int64_t> batch_total_size;
- for (const NodePtr &n : compute_graph_->GetAllNodes()) {
- auto node_op_desc = n->GetOpDesc();
- GE_IF_BOOL_EXEC(node_op_desc == nullptr, continue);
-
- if (CheckIsZeroMemNodeType(node_op_desc->GetType())) {
- continue;
- }
-
- std::string batch_label;
- (void)ge::AttrUtils::GetStr(node_op_desc, ATTR_NAME_BATCH_LABEL, batch_label);
-
- if (node_op_desc->GetType() == ATOMICADDRCLEAN) {
- atomic_addr_clean_id_ = node_op_desc->GetId();
- }
-
- for (auto &out_anchor : n->GetAllOutDataAnchors()) {
- GeTensorDesc output_desc = node_op_desc->GetOutputDesc(out_anchor->GetIdx());
- bool reuse_input = false;
- GE_IF_BOOL_EXEC(ge::TensorUtils::GetReuseInput(output_desc, reuse_input) != SUCCESS,
- GELOGI("Get reuse_input failed"));
-
- if (!reuse_input) {
- int64_t size = 0;
- GE_IF_BOOL_EXEC(ge::TensorUtils::GetSize(output_desc, size) != SUCCESS, GELOGI("Get size failed"));
- batch_all_memory_size[batch_label].emplace_back(size);
- if (batch_total_size.find(batch_label) == batch_total_size.end()) {
- batch_total_size[batch_label] = size;
- } else {
- batch_total_size[batch_label] += size;
- }
-
- if (!anchor_to_symbol_.empty()) {
- auto iter1 = anchor_to_symbol_.find(NodeIndexIO(n, out_anchor->GetIdx(), kOut).ToString());
- if (iter1 == anchor_to_symbol_.end()) {
- continue;
- }
- const std::string &symbol = iter1->second;
- auto iter2 = symbol_size_.find(symbol);
- if (iter2 == symbol_size_.end()) {
- symbol_size_[symbol] = size;
- } else if (size > static_cast<int64_t>(iter2->second)) {
- iter2->second = size;
- }
- }
- }
- }
- temp.clear();
- GetNodeWorkSpaceSize(n, temp, batch_total_size[batch_label]);
- batch_all_memory_size[batch_label].insert(batch_all_memory_size[batch_label].end(), temp.begin(), temp.end());
- }
- GELOGI("The last atomic_addr_clean node id: %ld", atomic_addr_clean_id_);
- GetMaxBatchAllMemorySize(batch_all_memory_size, batch_total_size, all_memory_size, max_batch_label_);
- InitReuseFlag();
- PrintSymbolMap();
- }
-
- ///
- /// @ingroup domi
- /// @brief decide memory size based on actual input memory size
- /// @param [in] size actual memory size in need
- /// @param [in] ranges memory size provided
- /// @return size_t memory size to apply
- ///
- size_t GetBlockSize(size_t size, const vector<int64_t> &ranges) {
- for (int64_t x : ranges) {
- auto x_temp = static_cast<size_t>(x);
- if (size <= x_temp) {
- return x_temp;
- }
- }
-
- GELOGW("Memory needed size:%zu is beyond the biggest block in memory ranges.", size);
- return size;
- }
-
- bool IsDirectOutputNode(const NodePtr &node, int idx) {
- if ((node != nullptr) && (node->GetOpDesc() != nullptr) && (node->GetOpDesc()->GetType() == NETOUTPUT)) {
- GELOGD("This is netoutput node, the input node mem can not be reused");
- return true;
- }
- return false;
- }
-
- void AddReusableBlockCount(const MemoryBlock &mem_block, map<string, uint64_t> &reusable_block_counts) {
- string key = std::to_string(mem_block.Size());
- key += "_" + std::to_string(mem_block.stream_id_);
- key += "_" + std::to_string(mem_block.memory_type_);
- auto it = reusable_block_counts.find(key);
- if (it != reusable_block_counts.end()) {
- it->second++;
- } else {
- reusable_block_counts[key] = 1;
- }
- }
-
- void ReduceReusableBlockCount(const MemoryBlock &mem_block, map<string, uint64_t> &reusable_block_counts) {
- string key = std::to_string(mem_block.Size());
- key += "_" + std::to_string(mem_block.stream_id_);
- key += "_" + std::to_string(mem_block.memory_type_);
- auto it = reusable_block_counts.find(key);
- if (it != reusable_block_counts.end()) {
- if (it->second > 0) {
- it->second--;
- }
- }
- }
-
- bool CanReuseBySize(const map<string, uint64_t> &reusable_block_counts, const MemoryBlock &reusable_block,
- size_t block_size, size_t real_size, bool continuous) {
- bool can_reuse = false;
- if (reusable_block.Size() == block_size) {
- can_reuse = true;
- }
- return can_reuse;
- }
-
- bool BlockMemAssigner::IsOutNodeSetContinuousInput(const NodePtr &n, uint32_t out_index, std::string &peer_name,
- uint32_t &peer_input_index,
- bool &no_need_assign_memory, bool &reset_zero_copy_flag) {
- if (n == nullptr || n->GetAllOutDataAnchors().size() <= 0) {
- return false;
- }
- if (static_cast<size_t>(out_index) < n->GetAllOutDataAnchors().size()) {
- auto out_anchor = n->GetOutDataAnchor(out_index);
- GE_IF_BOOL_EXEC(out_anchor == nullptr,
- GELOGE(FAILED, "Node[%s] output[%u] anchor is null.", n->GetName().c_str(), out_index);
- return false;);
- for (auto const &peer_in_anchor : out_anchor->GetPeerInDataAnchors()) {
- GE_IF_BOOL_EXEC(peer_in_anchor == nullptr,
- GELOGE(FAILED, "Node[%s] output[%u] peer_in_anchor 0 is null.", n->GetName().c_str(), out_index);
- return false;);
- auto peer_node = peer_in_anchor->GetOwnerNode();
- GE_IF_BOOL_EXEC(peer_node == nullptr,
- GELOGE(FAILED, "Node[%s] output[%u] node is null.", n->GetName().c_str(), out_index);
- return false;);
-
- // Get the continuous input type of the node, default is false
- bool is_input_continuous = false;
- auto peer_in_node_desc = peer_node->GetOpDesc();
- GE_IF_BOOL_EXEC(peer_in_node_desc == nullptr,
- GELOGE(FAILED, "Node[%s] output[%u] nodedesc is null.", n->GetName().c_str(), out_index);
- return false;);
-
- // If GetBool fail, is_input_continuous is false.
- bool is_input_continuous_no_padding = false;
- (void)ge::AttrUtils::GetBool(peer_in_node_desc, ATTR_NAME_NOPADDING_CONTINUOUS_INPUT,
- is_input_continuous_no_padding);
- if (is_input_continuous_no_padding) {
- reset_zero_copy_flag = true;
- return false;
- }
- (void)ge::AttrUtils::GetBool(peer_in_node_desc, ATTR_NAME_CONTINUOUS_INPUT, is_input_continuous);
-
- GE_IF_BOOL_EXEC(is_input_continuous && CheckIsZeroMemNodeType(peer_node->GetType()),
- GELOGI("Node[%s] output[%u] no_need_assign_memory.", n->GetName().c_str(), out_index);
- no_need_assign_memory = true;
- return false;);
-
- if (is_input_continuous) {
- if (n->GetOwnerComputeGraph() != nullptr) {
- string graph_name = n->GetOwnerComputeGraph()->GetName();
- GELOGI("%s name[%s] output[%u] node[%s] set input[%d] continuous, input size[%u].", graph_name.c_str(),
- n->GetName().c_str(), out_index, peer_in_node_desc->GetName().c_str(), peer_in_anchor->GetIdx(),
- peer_node->GetAllInDataAnchorsSize());
- // Only set attr one times.
- if (node_continuous_input_blocks_[peer_in_node_desc->GetName()].size() == 0) {
- (void)ge::AttrUtils::SetBool(peer_in_node_desc, ATTR_NAME_CONTINUOUS_INPUT_ALLOC, true);
- node_continuous_input_counts_[peer_in_node_desc->GetName()] = peer_node->GetAllInDataAnchorsSize();
- }
- peer_input_index = peer_in_anchor->GetIdx();
- peer_name = peer_in_node_desc->GetName();
- return true;
- }
- }
- }
- }
- return false;
- }
-
- ///
- /// @ingroup GE
- /// @brief Check pre_reuse flag & post_reuse glag for each symbol
- /// @return void
- ///
- void BlockMemAssigner::InitReuseFlag() {
- static const std::set<std::string> kPreReuseTypes = { ge::DATA_TYPE, ge::AIPP_DATA_TYPE, ge::ANN_DATA_TYPE,
- ge::NETOUTPUT, ge::PROPOSAL, ge::ZEROSLIKE,
- ge::CONSTANT, ge::CONSTANTOP };
- static const std::set<std::string> kPostReuseTypes = { ge::DATA_TYPE, ge::AIPP_DATA_TYPE, ge::ENTER, ge::REFENTER,
- ge::NEXTITERATION, ge::REFNEXTITERATION };
- for (const auto &pair : symbol_to_anchors_) {
- std::string symbol = pair.first;
- bool pre_reuse_flag = true;
- bool post_reuse_flag = true;
- // default memory type
- int64_t mem_type = RT_MEMORY_HBM;
- GetSymbolMemType(pair.second, mem_type);
- GELOGD("The memory type of symbol[%s] is [%ld]].", symbol.c_str(), mem_type);
- if (mem_type == RT_MEMORY_P2P_DDR) {
- UpdateOpTensorMemType(pair.second, mem_type);
- }
- // Only the memory with special requirements is processed. The HBM uses the default processing mode.
- if (mem_type == RT_MEMORY_P2P_DDR) {
- symbol_to_mem_type_[symbol] = mem_type;
- }
-
- for (const auto &node_index_io : pair.second) {
- if (node_index_io.io_type_ == kIn) {
- continue;
- }
-
- OutDataAnchorPtr out_anchor = node_index_io.node_->GetOutDataAnchor(node_index_io.index_);
- if (out_anchor == nullptr) {
- continue;
- }
-
- bool out_flg = false;
- if (node_index_io.node_->GetOutDataNodes().empty()) {
- out_flg = true;
- }
- for (const auto &in_anchor : out_anchor->GetPeerInDataAnchors()) {
- if (IsDirectOutputNode(in_anchor->GetOwnerNode(), in_anchor->GetIdx())) {
- out_flg = true;
- break;
- }
- }
- const std::string &type = out_anchor->GetOwnerNode()->GetType();
- pre_reuse_flag = pre_reuse_flag && !out_flg && (kPreReuseTypes.count(type) == 0);
- post_reuse_flag = post_reuse_flag && (kPostReuseTypes.count(type) == 0);
- if (!pre_reuse_flag && !post_reuse_flag) {
- break;
- }
- }
- pre_reuse_flag_[symbol] = pre_reuse_flag;
- post_reuse_flag_[symbol] = post_reuse_flag;
- }
- }
-
- ///
- /// @ingroup GE
- /// @brief get pre_reuse flag
- /// @param [in] node
- /// @param [in] out_index
- /// @return bool
- ///
- bool BlockMemAssigner::IsPreReuse(const NodePtr &node, uint32_t out_index) const {
- OutDataAnchorPtr out_data_anchor = nullptr;
- if (static_cast<size_t>(out_index) < node->GetAllOutDataAnchors().size()) {
- out_data_anchor = node->GetOutDataAnchor(out_index);
- }
- if (out_data_anchor == nullptr) {
- return false;
- }
- NodeIndexIO cur_node_index_io(out_data_anchor->GetOwnerNode(), out_data_anchor->GetIdx(), kOut);
- auto iter1 = anchor_to_symbol_.find(cur_node_index_io.ToString());
- if (iter1 == anchor_to_symbol_.end()) {
- return false;
- }
-
- const std::string &symbol = iter1->second;
- auto iter2 = pre_reuse_flag_.find(symbol);
- if (iter2 == pre_reuse_flag_.end()) {
- return false;
- }
- return iter2->second;
- }
-
- ///
- /// @ingroup GE
- /// @brief get post_reuse flag
- /// @param [in] mem_block
- /// @return bool
- ///
- bool BlockMemAssigner::IsPostReuse(const MemoryBlock *mem_block) const {
- if (mem_block == nullptr) {
- return false;
- }
- for (const auto &symbol : mem_block->SymbolList()) {
- auto iter = post_reuse_flag_.find(symbol);
- if (iter == post_reuse_flag_.end()) {
- continue;
- }
- if (!iter->second) {
- return false;
- }
- }
- return true;
- }
-
- ///
- /// @ingroup GE
- /// @brief check if symbol of cur node_index_io has block
- /// @param [in] node_index_io
- /// @param [out] symbol
- /// @return bool
- ///
- bool BlockMemAssigner::IsSymbolExist(const NodeIndexIO &node_index_io, string &symbol) {
- auto iter = anchor_to_symbol_.find(node_index_io.ToString());
- if (iter == anchor_to_symbol_.end()) {
- return false;
- }
-
- symbol = iter->second;
- return symbol_blocks_.find(iter->second) != symbol_blocks_.end();
- }
-
- ///
- /// @ingroup GE
- /// @brief Print symbol
- /// @return void
- ///
- void BlockMemAssigner::PrintSymbolMap() {
- for (const auto &pair : symbol_to_anchors_) {
- GELOGD("symbol=%s, max_size=%zu, pre_reuse=%s, post_reuse=%s", pair.first.c_str(), symbol_size_[pair.first],
- pre_reuse_flag_[pair.first] ? "true" : "false", post_reuse_flag_[pair.first] ? "true" : "false");
- for (const auto &node_index_io : pair.second) {
- GELOGD("anchor:%s", node_index_io.ToString().c_str());
- }
- }
- }
-
- void BlockMemAssigner::GetSymbolMemType(std::list<NodeIndexIO> node_index_io_list, int64_t &memory_type) {
- memory_type = RT_MEMORY_HBM;
- vector<int64_t> memory_types;
- for (auto &node_index_io : node_index_io_list) {
- auto op_desc = node_index_io.node_->GetOpDesc();
- if (op_desc == nullptr) {
- GELOGW("Node[%s] op desc is null.", node_index_io.node_->GetName().c_str());
- return;
- }
-
- if (node_index_io.io_type_ == kIn) {
- vector<int64_t> input_memory_types;
- (void) ge::AttrUtils::GetListInt(op_desc, ATTR_NAME_INPUT_MEM_TYPE_LIST, input_memory_types);
- if (!input_memory_types.empty() && node_index_io.index_ < input_memory_types.size()) {
- int64_t input_memory_type = input_memory_types[node_index_io.index_];
- GELOGD("Node[%s]: the memory type of input index [%u] is [%ld]].", op_desc->GetName().c_str(),
- node_index_io.index_, input_memory_type);
- memory_types.emplace_back(input_memory_type);
- }
- }
- if (node_index_io.io_type_ == kOut) {
- vector<int64_t> output_memory_types;
- (void) ge::AttrUtils::GetListInt(op_desc, ATTR_NAME_OUTPUT_MEM_TYPE_LIST, output_memory_types);
- if (!output_memory_types.empty() && node_index_io.index_ < output_memory_types.size()) {
- int64_t output_memory_type = output_memory_types[node_index_io.index_];
- GELOGD("Node[%s]: the memory type of output index [%u] is [%ld]].", op_desc->GetName().c_str(),
- node_index_io.index_, output_memory_type);
- memory_types.emplace_back(output_memory_type);
- }
- }
- }
-
- // memory priority
- for (auto node_memory_type : memory_types) {
- if (node_memory_type > memory_type) {
- memory_type = node_memory_type;
- }
- }
- }
-
- void BlockMemAssigner::UpdateOpTensorMemType(std::list<NodeIndexIO> node_index_io_list, int64_t memory_type) {
- for (auto &node_index_io : node_index_io_list) {
- auto op_desc = node_index_io.node_->GetOpDesc();
- if (op_desc == nullptr) {
- GELOGW("Node[%s] op desc is null.", node_index_io.node_->GetName().c_str());
- return;
- }
-
- if (node_index_io.io_type_ == kIn) {
- auto input_desc = op_desc->MutableInputDesc(node_index_io.index_);
- (void) AttrUtils::SetInt(input_desc, ATTR_NAME_TENSOR_MEM_TYPE, memory_type);
- }
-
- if (node_index_io.io_type_ == kOut) {
- auto output_desc = op_desc->MutableOutputDesc(node_index_io.index_);
- (void) AttrUtils::SetInt(output_desc, ATTR_NAME_TENSOR_MEM_TYPE, memory_type);
- }
- }
- }
-
- bool BlockMemAssigner::IsContinuousOutput(const NodePtr &n) {
- if (n == nullptr) {
- GELOGE(FAILED, "Node is null.");
- return false;
- }
-
- // Get the continuous output type of the node, default is false
- bool is_output_continuous = false;
- auto node_desc = n->GetOpDesc();
- if (node_desc == nullptr) {
- GELOGE(FAILED, "Node[%s] nodedesc is null.", n->GetName().c_str());
- return false;
- }
-
- // If GetBool fail, is_output_continuous is false.
- (void)ge::AttrUtils::GetBool(node_desc, ATTR_NAME_CONTINUOUS_OUTPUT, is_output_continuous);
- if (is_output_continuous) {
- if (n->GetOwnerComputeGraph() != nullptr) {
- string graph_name = n->GetOwnerComputeGraph()->GetName();
- GELOGI("%s name[%s] set continuous, output size[%u].", graph_name.c_str(),
- n->GetName().c_str(), n->GetAllOutDataAnchorsSize());
- return true;
- }
- }
-
- return false;
- }
-
- bool BlockMemAssigner::IsZeroCopyBlock(const NodePtr &node, bool continuous) {
- if (NodeUtils::IsDynamicShape(node)) {
- return ((node->GetType() == DATA_TYPE) && !continuous) || (node->GetType() == NETOUTPUT);
- }
-
- if ((node->GetType() == DATA_TYPE) && !continuous) {
- return !node->GetOpDesc()->HasAttr(ATTR_NAME_PARENT_NODE_INDEX);
- }
-
- if (node->GetType() == NETOUTPUT) {
- const auto &owner = node->GetOwnerComputeGraph();
- return owner->GetParentGraph() == nullptr;
- }
-
- return false;
- }
-
- MemoryBlock *BlockMemAssigner::ApplyMemory(size_t block_size, size_t real_size, size_t no_align_size,
- OpMemoryType mem_type, const NodePtr &n, uint32_t out_index,
- const vector<bool> &workspace_reuse_flag, const bool is_op_reuse_mem,
- const bool continuous, int64_t memory_type) {
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(n == nullptr, return nullptr, "Input parameter n is null.");
- auto node_op_desc = n->GetOpDesc();
- GE_IF_BOOL_EXEC(node_op_desc == nullptr, return nullptr);
- std::string batch_label;
- (void)ge::AttrUtils::GetStr(node_op_desc, ATTR_NAME_BATCH_LABEL, batch_label);
- if (batch_label.empty() || (batch_label == max_batch_label_)) {
- size_t align_size = real_size;
- AlignMemOffset(align_size);
- theory_memory_size_ += align_size;
- if (theory_memory_size_ > theory_min_memory_size_) {
- theory_min_memory_size_ = theory_memory_size_;
- }
- }
-
- bool is_reuse_memory = false;
- if (ge_disable_reuse_mem_env_ != "1") {
- bool reuse_mem_flag = (mem_type == kOutput) ? IsPreReuse(n, out_index) :
- !((workspace_reuse_flag.size() > out_index) && !workspace_reuse_flag[out_index]);
- is_reuse_memory = !node_op_desc->HasAttr(kL2FusionDynamicConvergeOp) &&
- !node_op_desc->HasAttr(kOpNoReuseMem) && reuse_mem_flag && is_op_reuse_mem;
- bool do_reuse = is_reuse_memory && !continuous && !reusable_blocks_[memory_type].empty();
- if (do_reuse) {
- auto stream_id = node_op_desc->GetStreamId();
- for (auto it = reusable_blocks_[memory_type][stream_id].rbegin();
- it != reusable_blocks_[memory_type][stream_id].rend(); ++it) {
- MemoryBlock *reusable_block = *it;
- if (!IsPostReuse(reusable_block)) {
- reusable_block->reuse_mem_ = false;
- GELOGI("Unreusable block.");
- continue;
- }
- GE_IF_BOOL_EXEC(reusable_block->batch_label_ != batch_label, continue);
-
- // A node can reuse blocks of the same stream and preorder streams
- if (CanReuseBySize(reusable_block_counts_, *reusable_block, block_size, real_size, continuous)) {
- reusable_block->AddNodeTypeIndex({n, mem_type, out_index, false}, real_size, no_align_size);
- if (mem_type == kOutput) {
- auto iter = anchor_to_symbol_.find(NodeIndexIO(n, out_index, kOut).ToString());
- if (iter != anchor_to_symbol_.end()) {
- reusable_block->AddSymbol(iter->second);
- }
- }
- reusable_block->continuous_block_ = continuous;
- reusable_block->ref_count_++;
- ReduceReusableBlockCount(*reusable_block, reusable_block_counts_);
- reusable_blocks_[memory_type][stream_id].erase((++it).base());
- return reusable_block;
- }
- }
- }
- }
-
- auto block = new (std::nothrow) MemoryBlock(block_size, node_op_desc->GetStreamId(), is_reuse_memory, memory_type);
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(block == nullptr, return nullptr, "new an object failed.");
-
- // Data and netoutput need zero copy block
- block->is_zero_copy_ = IsZeroCopyBlock(n, continuous);
-
- block->Init(real_size, mem_type, n, out_index, no_align_size, node_op_desc->GetStreamId());
- block->stream_id_ = node_op_desc->GetStreamId();
- block->ref_count_++;
- block->continuous_block_ = continuous;
- block->batch_label_ = batch_label;
- if (mem_type == kOutput) {
- auto iter = anchor_to_symbol_.find(NodeIndexIO(n, out_index, kOut).ToString());
- if (iter != anchor_to_symbol_.end()) {
- block->AddSymbol(iter->second);
- }
- }
- memory_blocks_.emplace_back(block);
- // cause memory_blocks_ may reduce when swap after,
- // create blocks_store_ to assure blocks deleted finally
- blocks_store_.emplace_back(block);
- return block;
- }
-
- MemoryBlock *BlockMemAssigner::ApplyContinuousMemory(const NodePtr &n, const vector<int64_t> &ranges,
- const bool is_op_reuse_mem) {
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(n == nullptr, return nullptr, "input node is null.");
- auto node_op_desc = n->GetOpDesc();
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(node_op_desc == nullptr, return nullptr, "node_op_desc is null.");
- MemoryBlock *block = nullptr;
- int64_t total_size = 0;
- int64_t memory_type = RT_MEMORY_HBM;
- for (uint32_t index = 0; index < static_cast<uint32_t>(node_op_desc->GetOutputsSize()); index++) {
- auto output_op_desc = node_op_desc->GetOutputDescPtr(index);
- if (output_op_desc == nullptr) {
- return nullptr;
- }
-
- if (CheckIsZeroMemNodeType(n->GetType())) {
- zero_memory_list_.emplace_back(n, kOutput, index);
- continue;
- }
-
- int64_t size = 0;
- if (ge::TensorUtils::GetSize(*output_op_desc, size) != SUCCESS) {
- GELOGI("Get size failed");
- return nullptr;
- }
- size_t align_size = static_cast<size_t>(size);
- AlignMemOffset(align_size);
- total_size += align_size;
-
- // only apply total size in first block
- if (index != 0) {
- zero_memory_list_.emplace_back(n, kOutput, index);
- } else {
- NodeIndexIO node_index_io(n, index, kOut);
- auto iter = anchor_to_symbol_.find(node_index_io.ToString());
- if (iter != anchor_to_symbol_.end()) {
- string symbol = iter->second;
- if (symbol_to_mem_type_.find(symbol) != symbol_to_mem_type_.end()) {
- memory_type = symbol_to_mem_type_[symbol];
- GELOGD("Continuous out memory symbol is [%s], memory type is [%ld]", symbol.c_str(), memory_type);
- }
- }
- }
- }
-
- if (total_size == 0) {
- return nullptr;
- }
-
- auto block_size = GetBlockSize(total_size, ranges);
- GELOGI("Node[%s] continuous out memory size[%ld] block size[%zu]", node_op_desc->GetName().c_str(),
- total_size, block_size);
-
- vector<bool> workspace_reuse_flag;
- block = ApplyMemory(block_size, total_size, total_size, kOutput, n, 0, workspace_reuse_flag, is_op_reuse_mem, true,
- memory_type);
- if (block != nullptr) {
- // hccl task need align header and tail
- block->first_continuous_block_ = true;
- block->last_continuous_block_ = true;
- }
- return block;
- }
-
- MemoryBlock *BlockMemAssigner::ApplyOutMemory(const NodePtr &n, uint32_t index, const vector<int64_t> &ranges,
- const bool is_op_reuse_mem, const bool continuous) {
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(n == nullptr, return nullptr, "input node is null.");
- auto node_op_desc = n->GetOpDesc();
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(node_op_desc == nullptr, return nullptr, "node_op_desc is null.");
- MemoryBlock *block = nullptr;
- NodeIndexIO node_index_io(n, index, kOut);
- int64_t size = 0;
- auto output_op_desc = node_op_desc->GetOutputDescPtr(index);
- if (output_op_desc != nullptr) {
- GE_IF_BOOL_EXEC(ge::TensorUtils::GetSize(*output_op_desc, size) != SUCCESS, GELOGI("Get size failed"));
- }
- size_t no_align_size = 0;
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(GetNoAlignSize(*node_op_desc, index, no_align_size) != SUCCESS,
- return nullptr, "Get no align size failed");
-
- std::string symbol;
- if (IsSymbolExist(node_index_io, symbol)) {
- block = symbol_blocks_[symbol];
- block->AddNodeTypeIndex({n, kOutput, index, true}, size, no_align_size);
- block->ref_count_++;
- } else {
- int64_t max_size = size;
- int64_t memory_type = RT_MEMORY_HBM;
- auto iter1 = anchor_to_symbol_.find(node_index_io.ToString());
- if (iter1 != anchor_to_symbol_.end()) {
- auto iter2 = symbol_size_.find(iter1->second);
- if (iter2 != symbol_size_.end()) {
- max_size = iter2->second;
- }
- auto iter3 = symbol_to_mem_type_.find(iter1->second);
- if (iter3 != symbol_to_mem_type_.end()) {
- memory_type = iter3->second;
- }
- }
-
- auto block_size = GetBlockSize(max_size, ranges);
- vector<bool> workspace_reuse_flag;
- block = ApplyMemory(block_size, size, no_align_size, kOutput, n, index,
- workspace_reuse_flag, is_op_reuse_mem, continuous, memory_type);
- }
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(block == nullptr, return nullptr, "Block is nullptr.");
- int out_count_reuse_input = block->ref_count_;
- int out_count = 0;
- GE_IF_BOOL_EXEC(index >= n->GetAllOutDataAnchors().size(), GELOGE(FAILED, "index is out of range."); return nullptr);
- auto out_data_anchor = n->GetOutDataAnchor(index);
- GE_IF_BOOL_EXEC(out_data_anchor == nullptr, GELOGE(FAILED, "Out data anchor is nullptr."); return nullptr);
- for (const auto &in_anchor : out_data_anchor->GetPeerInDataAnchors()) {
- auto owner_node = in_anchor->GetOwnerNode();
- auto op_desc = owner_node->GetOpDesc();
- GE_IF_BOOL_EXEC(op_desc == nullptr, continue);
- Params *instance = Params::Instance();
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(instance == nullptr, return nullptr, "Params instance is nullptr.");
- if (!((instance->GetTarget() == TARGET_TYPE_TINY) && (op_desc->GetType() == NETOUTPUT))) {
- out_count++;
- }
- }
- bool reuse_input = false;
- for (const auto &in_anchor : out_data_anchor->GetPeerInDataAnchors()) {
- auto owner_node = in_anchor->GetOwnerNode();
- GE_IF_BOOL_EXEC(owner_node == nullptr, continue);
- auto op_desc = owner_node->GetOpDesc();
- GE_IF_BOOL_EXEC(op_desc == nullptr, continue);
- for (uint32_t i = 0; i < static_cast<uint32_t>(op_desc->GetOutputsSize()); i++) {
- bool dst_reuse_input = false;
- uint32_t dst_reuse_input_index = 0;
- auto owner_node_op_desc = op_desc->GetOutputDescPtr(i);
- GE_IF_BOOL_EXEC(owner_node_op_desc == nullptr, continue);
- GE_IF_BOOL_EXEC(ge::TensorUtils::GetReuseInput(*owner_node_op_desc, dst_reuse_input) != SUCCESS,
- GELOGI("Get dst_reuse_input failed"));
- GE_IF_BOOL_EXEC(ge::TensorUtils::GetReuseInputIndex(*owner_node_op_desc, dst_reuse_input_index) != SUCCESS,
- GELOGI("Get dst_reuse_input_index failed"));
- if (dst_reuse_input && (dst_reuse_input_index == static_cast<uint32_t>(in_anchor->GetIdx()))) {
- block->AddNodeTypeIndex({owner_node, kOutput, i, true}, block->Size(), block->Size());
- out_count_reuse_input += 1;
- reuse_input = true;
- }
- }
- }
- block->ref_count_ = reuse_input ? out_count_reuse_input + out_count - 1 : out_count;
- return block;
- }
-
- bool IsOutputBlock(const ge::InDataAnchorPtr &in_data_anchor) {
- auto peer_out_anchor = in_data_anchor->GetPeerOutAnchor();
- GE_IF_BOOL_EXEC(peer_out_anchor == nullptr, GELOGE(FAILED, "Peer out anchor is nullptr."); return false);
- auto src = peer_out_anchor->GetOwnerNode();
- int32_t index = peer_out_anchor->GetIdx();
- auto iter = GetLocalOmgContext().out_nodes_map.find(src->GetName());
- if (iter != GetLocalOmgContext().out_nodes_map.end()) {
- for (auto id : iter->second) {
- if (index == id) {
- return true;
- }
- }
- }
- return false;
- }
-
- // atomic out memory will be reassigned
- bool IsAtomicOutputMemory(const ge::NodePtr &node, uint32_t output_index, bool is_atomic,
- bool out_node_set_continuous_input) {
- auto op_desc = node->GetOpDesc();
- if (op_desc == nullptr) {
- return false;
- }
- vector<int64_t> atomic_output_index;
- // If GetListInt fail, atomic_output_index is empty.
- (void)ge::AttrUtils::GetListInt(op_desc, ATOMIC_ATTR_OUTPUT_INDEX, atomic_output_index);
- if (!out_node_set_continuous_input && is_atomic) {
- for (auto &index : atomic_output_index) {
- if (static_cast<uint32_t>(index) == output_index) {
- if (node->GetOwnerComputeGraph() != nullptr) {
- string graph_name = node->GetOwnerComputeGraph()->GetName();
- GELOGD("[IMAS]Atomic no assign %s name[%s] output[%ld] streamid[%ld].", graph_name.c_str(),
- op_desc->GetName().c_str(), index, op_desc->GetStreamId());
- }
- return true;
- }
- }
- }
- return false;
- }
-
- bool IsKnownSubgraphData(const NodePtr &node) {
- if (NodeUtils::IsDynamicShape(node)) {
- return false;
- }
-
- return node->GetOpDesc()->HasAttr(ATTR_NAME_PARENT_NODE_INDEX);
- }
-
- void BlockMemAssigner::ReleaseMemory(MemoryBlock *to_release, vector<MemoryBlock *> &reusable_memory,
- bool same_stream) {
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(to_release == nullptr, return, "Input parameter to_release is null.");
- GE_CHK_TRUE_EXEC_INFO(to_release->ref_count_ <= 0, return, "Release memory");
- GE_CHK_TRUE_EXEC_INFO(!to_release->reuse_mem_, return, "doesn't reuse memory");
- --to_release->ref_count_;
- if (!same_stream) {
- to_release->same_stream_ = false;
- }
- if (to_release->ref_count_ == 0) {
- if (to_release->reuse_mem_ && !to_release->RealSizeList().empty()) {
- if (to_release->batch_label_.empty() || (to_release->batch_label_ == max_batch_label_)) {
- size_t align_size = to_release->RealSizeList().back();
- AlignMemOffset(align_size);
- theory_memory_size_ -= align_size;
- }
- }
- if (to_release->same_stream_) {
- to_release->SetLifeTimeEnd(life_time_);
- reusable_memory.emplace_back(to_release);
- AddReusableBlockCount(*to_release, reusable_block_counts_);
- }
- }
- }
-
- void BlockMemAssigner::ReleaseMemorys(const vector<MemoryBlock *> &to_releases,
- vector<MemoryBlock *> &reusable_memory) {
- for (auto mem_block : to_releases) {
- ReleaseMemory(mem_block, reusable_memory);
- }
- }
-
- void BlockMemAssigner::ReleaseInputNodeOutMemory(const unordered_map<string, vector<MemoryBlock *>> &node_out_blocks,
- vector<MemoryBlock *> &reusable_memory, NodePtr &node) {
- for (const auto &in_anchor : node->GetAllInDataAnchors()) {
- if ((in_anchor->GetPeerOutAnchor() == nullptr) ||
- (in_anchor->GetPeerOutAnchor()->GetOwnerNode()->GetOpDesc() == nullptr) || (node->GetOpDesc() == nullptr)) {
- return;
- }
- GE_IF_BOOL_EXEC(IsOutputBlock(in_anchor), continue);
-
- auto node_name = in_anchor->GetPeerOutAnchor()->GetOwnerNode()->GetName();
-
- GE_IF_BOOL_EXEC((in_anchor->GetPeerOutAnchor()->GetOwnerNode()->GetType() == CONSTANT) ||
- (in_anchor->GetPeerOutAnchor()->GetOwnerNode()->GetType() == FASTRCNNPREDICTIONS) ||
- (in_anchor->GetPeerOutAnchor()->GetOwnerNode()->GetType() == CONSTANTOP),
- continue);
-
- auto it = node_out_blocks.find(node_name);
- if (it == node_out_blocks.end()) {
- continue;
- }
- for (auto block : it->second) {
- const vector<NodeTypeIndex> &node_type_indexs = block->NodeTypeIndexList();
- if (node_type_indexs.empty()) {
- continue;
- }
- GELOGD("node_type_indexs: %d, %s", node_type_indexs.back().index,
- node_type_indexs.back().node->GetName().c_str());
-
- if ((node_type_indexs.back().node == in_anchor->GetPeerOutAnchor()->GetOwnerNode()) &&
- (node_type_indexs.back().index == static_cast<uint32_t>(in_anchor->GetPeerOutAnchor()->GetIdx()))) {
- ReleaseMemory(block, reusable_memory, (node->GetOpDesc()->GetStreamId() == block->stream_id_));
- if (block->ref_count_ == 0 && block->same_stream_) {
- SetLastUsedInputMemAttr(node, in_anchor->GetIdx());
- }
- }
- }
- }
- }
-
- void SplitStringByComma(const string &str, vector<string> &sub_str_vec) {
- std::string tmp_string = str + ",";
- std::string::size_type start_pos = 0;
- std::string::size_type cur_pos = tmp_string.find(',', 0);
- while (cur_pos != std::string::npos) {
- std::string sub_str = tmp_string.substr(start_pos, cur_pos - start_pos);
- if (!sub_str.empty()) {
- vector<string>::iterator ret = std::find(sub_str_vec.begin(), sub_str_vec.end(), sub_str);
- if (ret == sub_str_vec.end()) {
- sub_str_vec.push_back(sub_str);
- }
- }
- start_pos = cur_pos + 1;
- cur_pos = tmp_string.find(',', start_pos);
- }
- }
-
- void CheckAndGetOpReuseEnv(const string &env, vector<string> &env_vec, bool &op_reuse_env_valid) {
- string env_str;
- env_str = string(env);
- if (env_str.size() > kReuseMaxCharNum) {
- GELOGE(FAILED, "The OP_NO_REUSE_MEM has more than %d characters.", kReuseMaxCharNum);
- return;
- }
-
- SplitStringByComma(env_str, env_vec);
- if (env_vec.size() > kReuseMaxOpNum) {
- GELOGE(FAILED, "The OP_NO_REUSE_MEM has more than %d nodes.", kReuseMaxOpNum);
- return;
- }
-
- op_reuse_env_valid = true;
- return;
- }
-
- Status BlockMemAssigner::AssignOutputMemoryWithReuse(const NodePtr &node, vector<int64_t> &ranges) {
- auto op_desc = node->GetOpDesc();
- int64_t stream_id = op_desc->GetStreamId();
- vector<int64_t> memorys_type;
- bool has_mem_type_attr = ge::AttrUtils::GetListInt(op_desc, ATTR_NAME_OUTPUT_MEM_TYPE_LIST, memorys_type);
- GELOGD("Assign memory node[%s], output size[%zu], output memory type size[%zu]", op_desc->GetName().c_str(),
- op_desc->GetOutputsSize(), memorys_type.size());
- if (has_mem_type_attr && (memorys_type.size() != op_desc->GetOutputsSize())) {
- GELOGE(INTERNAL_ERROR, "fusion: node[%s], output memory size err[outputsize:%zu, memorysize:%zu]",
- op_desc->GetName().c_str(), op_desc->GetOutputsSize(), memorys_type.size());
- return INTERNAL_ERROR;
- }
-
- is_op_reuse_mem_ = true;
- if (op_reuse_env_valid_ == true) {
- vector<string>::iterator it_name =
- std::find(op_no_reuse_mem_vec_.begin(), op_no_reuse_mem_vec_.end(), op_desc->GetName());
- vector<string>::iterator it_type =
- std::find(op_no_reuse_mem_vec_.begin(), op_no_reuse_mem_vec_.end(), op_desc->GetType());
- GE_IF_BOOL_EXEC(it_name != op_no_reuse_mem_vec_.end() || it_type != op_no_reuse_mem_vec_.end(),
- is_op_reuse_mem_ = false;);
- }
-
- bool is_atomic = false;
- // If GetBool fail, is_atomic is false.
- (void)ge::AttrUtils::GetBool(op_desc, ATOMIC_ATTR_IS_ATOMIC_NODE, is_atomic);
- // Allocate memory for the current node and release node memory of the same size in the workspace
- GE_IF_BOOL_EXEC(ge_disable_reuse_mem_env_ != "1",
- for (auto iter = stream_workspace_blocks_.begin(); iter != stream_workspace_blocks_.end();
- ++iter) { ReleaseMemorys(iter->second[stream_id], reusable_blocks_[iter->first][stream_id]); });
- if (IsContinuousOutput(node)) {
- (void)ApplyContinuousMemory(node, ranges, is_op_reuse_mem_);
- return SUCCESS;
- }
- for (uint32_t i = 0; i < static_cast<uint32_t>(op_desc->GetOutputsSize()); i++) {
- int64_t size = 0;
- auto output_op_desc = op_desc->GetOutputDescPtr(i);
- if (output_op_desc != nullptr) {
- GE_IF_BOOL_EXEC(ge::TensorUtils::GetSize(*output_op_desc, size) != SUCCESS, GELOGI("Get size failed"));
- }
- // fusion: other type's size not means malloc HBM memory
- bool l1_flag = has_mem_type_attr && memorys_type[i] == RT_MEMORY_L1;
- if (l1_flag) {
- GELOGI("fusion: node[%s], output[%s], output memory type [%ld]",
- op_desc->GetName().c_str(), op_desc->GetOutputNameByIndex(i).c_str(), memorys_type[i]);
- size = 0;
- }
- std::string peer_name;
- uint32_t peer_input_index = 0;
- bool out_node_set_continuous_input = false;
- bool reset_zero_copy_flag = false;
- bool no_need_assign_memory = ((size == 0) || CheckIsZeroMemNodeType(node->GetType()));
- if (!no_need_assign_memory) {
- out_node_set_continuous_input =
- IsOutNodeSetContinuousInput(node, i, peer_name, peer_input_index,
- no_need_assign_memory, reset_zero_copy_flag);
- GE_IF_BOOL_EXEC(!no_need_assign_memory,
- no_need_assign_memory = IsAtomicOutputMemory(node, i, is_atomic, out_node_set_continuous_input););
- }
- no_need_assign_memory = (no_need_assign_memory || IsKnownSubgraphData(node));
- if (no_need_assign_memory) {
- zero_memory_list_.emplace_back(node, kOutput, i, false);
- continue;
- }
- // atomic can't be reused
- bool need_change = is_op_reuse_mem_ && out_node_set_continuous_input && is_atomic;
- if (need_change) {
- is_op_reuse_mem_ = false;
- }
-
- MemoryBlock *mem_block = ApplyOutMemory(node, i, ranges, is_op_reuse_mem_, out_node_set_continuous_input);
- if (mem_block != nullptr) {
- GE_IF_BOOL_EXEC(reset_zero_copy_flag,
- mem_block->is_zero_copy_ = false;
- GELOGI("Node[%s] output[%u] need assign memory before reassign.", op_desc->GetName().c_str(), i););
- node_out_blocks_[node->GetName()].emplace_back(mem_block);
- if (out_node_set_continuous_input) {
- node_continuous_input_blocks_[peer_name][peer_input_index] = mem_block;
- }
- NodeIndexIO node_index_io(node, i, kOut);
- auto iter = anchor_to_symbol_.find(node_index_io.ToString());
- if (iter == anchor_to_symbol_.end()) {
- continue;
- }
- symbol_blocks_[iter->second] = mem_block;
- }
- }
- return SUCCESS;
- }
-
- ///
- /// @ingroup domi
- /// @brief traverse all nodes outputs and workspace in need, apply memory block considering memory reuse
- /// @param [in/out] ranges memory size provided
- /// @return Status result
- ///
- void BlockMemAssigner::AssignMemoryWithReuse(vector<int64_t> &ranges) {
- (void)ge::GetContext().GetOption(OPTION_EXEC_DISABLE_REUSED_MEMORY, ge_disable_reuse_mem_env_);
- GELOGD("Reuse memory %s", ge_disable_reuse_mem_env_ == "1" ? "close" : "open");
- string op_no_reuse_mem_str;
- const char *op_no_reuse_mem = std::getenv(OP_NO_REUSE_MEM);
- GE_IF_BOOL_EXEC(op_no_reuse_mem != nullptr, op_no_reuse_mem_str = string(op_no_reuse_mem);
- CheckAndGetOpReuseEnv(op_no_reuse_mem_str, op_no_reuse_mem_vec_, op_reuse_env_valid_););
-
- for (NodePtr &n : compute_graph_->GetAllNodes()) {
- auto node_op_desc = n->GetOpDesc();
- GE_IF_BOOL_EXEC(node_op_desc == nullptr, continue);
- life_time_ = node_op_desc->GetId();
- int64_t stream_id = node_op_desc->GetStreamId();
- if (AssignOutputMemoryWithReuse(n, ranges) != SUCCESS) {
- return;
- }
- for (auto iter = stream_workspace_blocks_.begin(); iter != stream_workspace_blocks_.end(); ++iter) {
- iter->second[stream_id].clear();
- }
- vector<int64_t> temp;
- int64_t tatal_size = 0;
- GetNodeWorkSpaceSize(n, temp, tatal_size);
- vector<int64_t> workspace_bytes;
- vector<int64_t> tvm_workspace_memory_type;
- bool has_tvm_workspace_mem_type_attr =
- ge::AttrUtils::GetListInt(node_op_desc, TVM_ATTR_NAME_WORKSPACE_TYPE, tvm_workspace_memory_type);
- vector<bool> workspace_reuse_flag;
- GE_IF_BOOL_EXEC(!ge::AttrUtils::GetListBool(node_op_desc, kAttrNameWorkspaceReuseFlag, workspace_reuse_flag),
- GELOGD("OP %s get workspace_reuse_flag attr failed", node_op_desc->GetName().c_str()));
- GELOGD("Assign memory node[%s], size [temp:%zu, memory type size:%zu]", node_op_desc->GetName().c_str(),
- temp.size(), tvm_workspace_memory_type.size());
-
- if (has_tvm_workspace_mem_type_attr && (temp.size() != tvm_workspace_memory_type.size())) {
- GELOGE(INTERNAL_ERROR, "fusion: node[%s], tvm workspace memory size error![v_temp:%zu, workspace:%zu]",
- n->GetName().c_str(), temp.size(), tvm_workspace_memory_type.size());
- return;
- }
- for (size_t i = 0; i < temp.size(); i++) {
- // fusion: other type's size not means malloc HBM memory
- bool workspace_skip_flag = false;
- if (has_tvm_workspace_mem_type_attr && tvm_workspace_memory_type[i] == RT_MEMORY_L1) {
- GELOGI(
- "fusion:node[%s]workspace index[%zu] is not hbm type, add to zero_memory_list, workspace memory type [%ld]",
- node_op_desc->GetName().c_str(), i, tvm_workspace_memory_type[i]);
- workspace_skip_flag = true;
- }
- if (temp[i] == 0 || workspace_skip_flag) {
- zero_memory_list_.emplace_back(n, kWorkspace, static_cast<uint32_t>(i), false);
- continue;
- }
- int64_t memory_type = RT_MEMORY_HBM;
- if (!GetWorkSpaceMemoryType(n, i, memory_type)) {
- GELOGW("Get workspace memory type failed.");
- return;
- }
- MemoryBlock *mem_block = ApplyMemory(GetBlockSize(static_cast<size_t>(temp[i]), ranges),
- static_cast<size_t>(temp[i]), static_cast<size_t>(temp[i]),
- kWorkspace, n, static_cast<uint32_t>(i), workspace_reuse_flag,
- is_op_reuse_mem_, false, memory_type);
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(mem_block == nullptr, continue, "failed to apply memory block.");
- CheckWorkspaceReuse(workspace_reuse_flag, i, stream_id, mem_block, memory_type);
- }
- for (auto it = reusable_blocks_.begin(); it != reusable_blocks_.end(); ++it) {
- ReleaseInputNodeOutMemory(node_out_blocks_, it->second[stream_id], n);
- }
- }
-
- GELOGD("Assigned memory blocks:");
- for (auto mem_block : memory_blocks_) {
- GELOGD("%s", mem_block->String().c_str());
- (void)mem_block; // Fix warning
- }
-
- GE_IF_BOOL_EXEC(!(ge_disable_reuse_mem_env_ == "1"), ReuseBlocksByLifeTime(ranges.size()));
- AssignContinuousBlocks();
- ResizeMemoryBlocks();
-
- GELOGD("Memory blocks after resize:");
- for (auto mem_block : memory_blocks_) {
- GELOGD("%s", mem_block->String().c_str());
- (void)mem_block; // Fix warning
- }
- }
-
- void BlockMemAssigner::CheckWorkspaceReuse(const vector<bool> &workspace_reuse_flag, uint32_t index, int64_t stream_id,
- MemoryBlock *mem_block, int64_t memory_type) {
- bool reuse_mem_flag =
- ((workspace_reuse_flag.size() > index) && (workspace_reuse_flag[index] == false)) ? false : true;
- if (reuse_mem_flag) {
- stream_workspace_blocks_[memory_type][stream_id].emplace_back(mem_block);
- }
- }
-
- void BlockMemAssigner::GetNodeWorkSpaceSize(const NodePtr &node, vector<int64_t> &workspace_memory,
- int64_t &total_size) {
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(node->GetOpDesc() == nullptr, return, "Op desc is null.");
- vector<int64_t> workspace_byte_nums = node->GetOpDesc()->GetWorkspaceBytes();
-
- GELOGD("node[%s] size:%zu", node->GetOpDesc()->GetName().c_str(), workspace_byte_nums.size());
- for (int64_t byte_size : workspace_byte_nums) {
- workspace_memory.emplace_back(byte_size);
- total_size += byte_size;
- GELOGD("push back size:%ld", byte_size);
- }
- }
-
- // asending order
- static bool CompareBlockIndex(MemoryBlock *left, MemoryBlock *right) {
- if (left == nullptr || right == nullptr) {
- return false;
- }
- if (left->input_index_ < right->input_index_) {
- return true;
- }
- return false;
- }
- ///
- /// @ingroup domi
- /// @brief order blocks by continuous input index
- /// @param [in] blocks need be processed
- /// @param [in] input blocks need continuous
- /// @param [out] blocks after continuous order
- /// @param [in/out] blocks ordered
- /// @param [in] input or output
- ///
- void ReAssignContinuousBlocks(const std::vector<MemoryBlock *> &org_blocks,
- const std::map<MemoryBlock *, uint32_t> block_map,
- std::vector<MemoryBlock *> &dest_blocks, std::vector<MemoryBlock *> &continuous_blocks,
- const std::string &type) {
- for (auto &memory_block : org_blocks) {
- if (memory_block == nullptr || memory_block->deleted_block_) {
- continue;
- }
- if (block_map.find(memory_block) != block_map.end()) {
- continue;
- }
- dest_blocks.emplace_back(memory_block);
- }
-
- // add continuous block
- std::sort(continuous_blocks.begin(), continuous_blocks.end(), CompareBlockIndex);
- size_t count = 0;
- for (auto &memory_block : continuous_blocks) {
- GE_IF_BOOL_EXEC(memory_block == nullptr, continue);
-
- GELOGI("Block continuous %s index:%d", type.c_str(), memory_block->input_index_);
- count++;
- if (count == 1) {
- memory_block->first_continuous_block_ = true;
- }
- if (count == continuous_blocks.size()) {
- memory_block->last_continuous_block_ = true;
- }
- dest_blocks.emplace_back(memory_block);
- }
- }
-
- void BlockMemAssigner::AssignContinuousBlocks() {
- for (auto &block_map : node_continuous_input_blocks_) {
- std::vector<MemoryBlock *> dest_memory_blocks;
- std::map<MemoryBlock *, uint32_t> continuous_block_map;
- std::vector<MemoryBlock *> continuous_blocks;
- auto it = node_continuous_input_counts_.find(block_map.first);
- GE_IF_BOOL_EXEC(it == node_continuous_input_counts_.end(), continue);
- GELOGI("Node:%s continuous input block count:%zu input count:%u", block_map.first.c_str(), block_map.second.size(),
- it->second);
- GE_IF_BOOL_EXEC(it->second != block_map.second.size(), continue);
-
- for (auto &it : block_map.second) {
- if (it.second != nullptr) {
- continuous_block_map[it.second] = it.first;
- it.second->input_index_ = it.first;
- continuous_blocks.emplace_back(it.second);
- }
- }
- if (continuous_block_map.size() != continuous_blocks.size()) {
- GELOGW("Node:%s continuous input map size:%zu vector size:%zu", block_map.first.c_str(),
- continuous_block_map.size(), continuous_blocks.size());
- continue;
- }
- ReAssignContinuousBlocks(memory_blocks_, continuous_block_map, dest_memory_blocks, continuous_blocks, "input");
- memory_blocks_.swap(dest_memory_blocks);
- }
- }
-
- void BlockMemAssigner::ReuseBlocksByLifeTime(size_t range_size) {
- // 1 means block size is same so no need to do this
- if (range_size <= 1) {
- return;
- }
- for (size_t i = 0; i < memory_blocks_.size(); ++i) {
- auto parent = memory_blocks_[i];
- if (parent == nullptr || parent->deleted_block_ || parent->continuous_block_) {
- continue;
- }
- if (parent->reuse_mem_ && !IsPostReuse(parent)) {
- parent->reuse_mem_ = false;
- }
- for (size_t j = i + 1; j < memory_blocks_.size(); ++j) {
- auto child = memory_blocks_[j];
- if (child == nullptr) {
- continue;
- }
- // If node is before atomic_addr_clean node, the continus memory can't be reused.
- if (!parent->NodeTypeIndexList().empty() && child->continuous_block_) {
- auto node = parent->NodeTypeIndexList()[0].node;
- if (node == nullptr || node->GetOpDesc() == nullptr || (node->GetOpDesc()->GetId() < GetAtomicAddrCleanId())) {
- continue;
- }
- }
- parent->AddLifeReuseBlock(child, total_node_depend_stream_life_);
- }
- }
- }
-
- void AddBlockMemOffset(size_t &mem_offset, size_t &p2p_mem_offset, MemoryBlock &block) {
- if (block.memory_type_ == RT_MEMORY_HBM) {
- if (block.first_continuous_block_) {
- mem_offset += MEM_ALIGN_SIZE;
- }
- block.Resize();
- block.SetHeadOffset(mem_offset);
- mem_offset += block.Size();
- block.SetTailOffset(mem_offset - 1);
- } else if (block.memory_type_ == RT_MEMORY_P2P_DDR) {
- if (block.first_continuous_block_) {
- p2p_mem_offset += MEM_ALIGN_SIZE;
- }
- block.Resize();
- block.SetHeadOffset(p2p_mem_offset);
- p2p_mem_offset += block.Size();
- block.SetTailOffset(p2p_mem_offset - 1);
- }
- }
-
- bool DynamicBatchBlockReuse(MemoryBlock &block) {
- return (block.IsSameBatchLabel() && block.reuse_mem_);
- }
-
- ///
- /// @ingroup domi_omg
- /// @brief get max batch memory size, others reuse this block memory
- /// @param [in&out] memory_blocks_ memory block, after calculating offset
- /// |-dynamic batch block batch1|
- /// |-dynamic batch block batch2----|
- /// |-dynamic batch block batch3--|
- ///
- void BlockMemAssigner::ResizeDynamicBatchBlocks() {
- std::map<std::string, std::vector<MemoryBlock *>> dynamic_batch_blocks;
- for (auto block : memory_blocks_) {
- if (block == nullptr) {
- continue;
- }
- // when memory is not reuseable, it can't be reused by different branch
- if (DynamicBatchBlockReuse(*block)) {
- dynamic_batch_blocks[block->batch_label_].emplace_back(block);
- }
- }
-
- size_t max_mem_offset = mem_offset_;
- size_t max_p2p_mem_offset = p2p_mem_offset_;
- for (auto &batch_blocks : dynamic_batch_blocks) {
- size_t mem_offset = mem_offset_;
- size_t p2p_mem_offset = p2p_mem_offset_;
- for (auto block : batch_blocks.second) {
- if (block == nullptr || block->deleted_block_ || block->is_zero_copy_) {
- continue;
- }
- AddBlockMemOffset(mem_offset, p2p_mem_offset, *block);
- }
- if (mem_offset > max_mem_offset) {
- max_mem_offset = mem_offset;
- }
- if (p2p_mem_offset > max_p2p_mem_offset) {
- max_p2p_mem_offset = p2p_mem_offset;
- }
- GELOGI("Batch[%s] offset[%zu] p2p_offset[%zu]", batch_blocks.first.c_str(), mem_offset, p2p_mem_offset);
- }
- mem_offset_ = max_mem_offset;
- p2p_mem_offset_ = max_p2p_mem_offset;
- }
-
- ///
- /// @ingroup domi_omg
- /// @brief traverse memory size, resize, calculate offset
- /// @param [in&out] memory_blocks_ memory block, after calculating offset
- /// |-not dynamic batch block-||-dynamic batch block batch1| |-zero copy block-|
- /// |-not dynamic batch block-||-dynamic batch block batch2----||-zero copy block-|
- /// |-not dynamic batch block-||-dynamic batch block batch3--| |-zero copy block-|
- ///
- void BlockMemAssigner::ResizeMemoryBlocks() {
- for (auto &memory_block : memory_blocks_) {
- if (memory_block == nullptr || memory_block->deleted_block_ || memory_block->is_zero_copy_
- || DynamicBatchBlockReuse(*memory_block)) {
- continue;
- }
-
- AddBlockMemOffset(mem_offset_, p2p_mem_offset_, *memory_block);
- }
- ResizeDynamicBatchBlocks();
- GELOGI("mem_offset_ exclude zero_copy_memory is %zu, p2p_mem_offset_ exclude zero_copy_memory is %zu,"
- "theory_min_memory_size %zu", mem_offset_, p2p_mem_offset_, theory_min_memory_size_);
- }
-
- ///
- /// @ingroup domi
- /// @brief given NodeTypeIndex, set offset in Op's OpDef
- /// @param [in&out] node_type_index <node, memory type, id>
- /// @param [in] offset offset to be set
- /// @param [in] size memory size
- /// @param [in] real_size memory size in need
- /// @return Status result
- ///
- void SetOffsetSize(const NodeTypeIndex &node_type, const MemoryBlock *block,
- size_t real_size, size_t no_align_size, int32_t child_block_level) {
- ge::OpDescPtr op_desc = node_type.node->GetOpDesc();
- GE_CHK_BOOL_TRUE_EXEC_WITH_LOG(op_desc == nullptr, return, "op_desc is null.");
- string graph_name = node_type.node->GetOwnerComputeGraph()->GetName();
- vector<int64_t> memorys_type;
- int64_t offset = block->HeadOffset();
- size_t end = node_type.life_time_end;
- bool has_mem_type_attr = ge::AttrUtils::GetListInt(op_desc, ATTR_NAME_OUTPUT_MEM_TYPE_LIST, memorys_type);
- if (node_type.mem_type == kOutput) {
- vector<int64_t> output_list = op_desc->GetOutputOffset();
- for (auto i = static_cast<uint32_t>(output_list.size()); i < node_type.index + 1; i++) {
- output_list.emplace_back(kInvalidOffset);
- }
- if (output_list.empty()) {
- GELOGW("Empty output");
- return;
- }
-
- static const set<string> kSetOffsetTypes = { DATA_TYPE, AIPP_DATA_TYPE, MULTISHAPE, NETOUTPUT };
- if ((kSetOffsetTypes.count(op_desc->GetType()) > 0) && !IsKnownSubgraphData(node_type.node)) {
- if ((output_list[node_type.index] == kInvalidOffset) || (output_list[node_type.index] < offset)) {
- output_list.at(node_type.index) = offset;
- }
- } else {
- // fusion: keep the original other type offset value from op_desc
- bool set_out_offset = (!has_mem_type_attr) ||
- (memorys_type.size() > node_type.index && memorys_type[node_type.index] != RT_MEMORY_L1);
- if (set_out_offset) {
- output_list.at(node_type.index) = offset;
- }
- }
- op_desc->SetOutputOffset(output_list);
- } else if (node_type.mem_type == kWorkspace) {
- vector<int64_t> workspace_list;
- workspace_list = op_desc->GetWorkspace();
- for (auto i = static_cast<uint32_t>(workspace_list.size()); i < node_type.index + 1; i++) {
- workspace_list.emplace_back(kInvalidOffset);
- }
- vector<int64_t> workspace_mem_type;
- bool has_workspace_mem_type = ge::AttrUtils::GetListInt(op_desc, TVM_ATTR_NAME_WORKSPACE_TYPE, workspace_mem_type);
- // fusion: keep the original other type offset value from op_desc
- bool set_workspace_offset = (!has_workspace_mem_type) ||
- (workspace_mem_type.size() > node_type.index && workspace_mem_type[node_type.index] != RT_MEMORY_L1);
- if (set_workspace_offset) {
- workspace_list.at(node_type.index) = offset;
- }
- op_desc->SetWorkspace(workspace_list);
- }
- GELOGI("[IMAS]Set %s name[%s] %s[%u] offset to [%ld] streamid[%ld] size[%zu] realsize[%zu] noalignsize[%zu] "
- "life time begin[%zu] life time end[%zu] child[%d:%d:%d:%d:%d] isref[%d] batch[%s]", graph_name.c_str(),
- op_desc->GetName().c_str(), node_type.GetMemType().c_str(), node_type.index, offset, op_desc->GetStreamId(),
- block->Size(), real_size, no_align_size, op_desc->GetId(), end, child_block_level, block->reuse_mem_,
- block->continuous_block_, block->is_zero_copy_, block->same_stream_, node_type.ref_input,
- block->batch_label_.c_str());
- }
-
- void SetBlockOpMemOffset(MemoryBlock *block, int32_t child_block_level) {
- if (block == nullptr) {
- return;
- }
- size_t index = 0;
- size_t real_size = 0;
- size_t no_align_size = 0;
- auto real_size_list_size = block->RealSizeList().size();
- for (const NodeTypeIndex &node_type_index : block->NodeTypeIndexList()) {
- if (index < real_size_list_size) {
- real_size = block->RealSizeList()[index];
- no_align_size = block->NoAlignSizeList()[index];
- }
- SetOffsetSize(node_type_index, block, real_size, no_align_size, child_block_level);
- index++;
- }
-
- child_block_level++;
- for (MemoryBlock *child_block : block->ChildBlockList()) {
- SetBlockOpMemOffset(child_block, child_block_level);
- }
- }
-
- void BlockMemAssigner::SetOpMemOffset(bool is_zero_copy) {
- for (MemoryBlock *memory_block : memory_blocks_) {
- if (memory_block == nullptr || memory_block->deleted_block_) {
- continue;
- }
-
- if ((is_zero_copy && !memory_block->is_zero_copy_) || (!is_zero_copy && memory_block->is_zero_copy_)) {
- continue;
- }
-
- SetBlockOpMemOffset(memory_block, 0);
- }
-
- if (!is_zero_copy) {
- for (const NodeTypeIndex &node_type_index : zero_memory_list_) {
- MemoryBlock block(0, 0);
- SetOffsetSize(node_type_index, &block, 0, 0, 0);
- }
- }
- }
-
- Status BlockMemAssigner::Assign() {
- vector<int64_t> ranges;
- if (GetMemoryRanges(ranges) != SUCCESS) {
- GELOGE(FAILED, "GetMemoryRanges Fail!");
- return FAILED;
- }
- GE_IF_BOOL_EXEC(ranges.empty(), return SUCCESS);
- AssignMemoryWithReuse(ranges);
-
- SetOpMemOffset(false);
-
- return SUCCESS;
- }
-
- bool BlockMemAssigner::CheckIsZeroMemNodeType(const string &node_type) const {
- return (node_type == VARIABLE) || (node_type == CONSTANT) || (node_type == MULTISHAPE) ||
- (node_type == HCOMBROADCAST) || (node_type == CONSTANTOP) ||
- (node_type == ASSIGNADD) || (node_type == ASSIGNSUB) || (node_type == ASSIGN) || (node_type == HVDWAIT) ||
- (node_type == HVDCALLBACKBROADCAST);
- }
-
- bool BlockMemAssigner::GetWorkSpaceMemoryType(const NodePtr &node, size_t index, int64_t &memory_type) {
- memory_type = RT_MEMORY_HBM;
- vector<int64_t> workspace_memory_type;
- auto op_desc = node->GetOpDesc();
- bool has_workspace_mem_type_attr =
- ge::AttrUtils::GetListInt(op_desc, TVM_ATTR_NAME_WORKSPACE_TYPE, workspace_memory_type);
- if (has_workspace_mem_type_attr && (workspace_memory_type.size() <= index)) {
- GELOGE(INTERNAL_ERROR, "node[%s], workspace_memory size error![index:%zu, workspace:%zu]",
- node->GetName().c_str(), index, workspace_memory_type.size());
- return false;
- }
- memory_type = has_workspace_mem_type_attr ? workspace_memory_type[index] : RT_MEMORY_HBM;
- return true;
- }
- } // namespace ge
|