Skip to content

M
mindspore

magicwindyyd / mindspore
与 Fork 源项目一致

Fork自 MindSpore / mindspore

通知 1
Star 1
Fork 0
M
mindspore
提交 5e2f440e 编写于 作者: M mindspore-ci-bot 提交者： Gitee
浏览文件
操作

!1271 refactor of memreuse allocator to adapt the control stream 

Merge pull request !1271 from yangjie159/refactor_memreuse_allocator
上级 e7936ded cbf5390b
隐藏空白更改
内联 并排
Showing with 319 addition and 536 deletion
mindspore/ccsrc/pre_activate/mem_reuse/kernel_refcount.cc
浏览文件 @ 5e2f440e
......@@ -47,7 +47,7 @@ std::vector<int> KernelDef::GetOutputRefIndexs() const {
return output_ref_indexs;
}
std::vector<int> KernelDef::GetWkRefIndexs() const {
std::vector<int> KernelDef::GetWorkspaceRefIndexs() const {
std::vector<int> wk_ref_indexs;
if (wk_space_.empty()) {
return wk_ref_indexs;
......
mindspore/ccsrc/pre_activate/mem_reuse/kernel_refcount.h
浏览文件 @ 5e2f440e
......@@ -20,6 +20,7 @@
#include <map>
#include <string>
#include <memory>
#include <set>
namespace mindspore {
namespace memreuse {
......@@ -73,13 +74,15 @@ class KernelDef {
KernelRefCountPtrList output_refs() const { return output_refs_; }
std::vector<int> GetInputRefIndexs() const;
std::vector<int> GetOutputRefIndexs() const;
std::vector<int> GetWkRefIndexs() const;
std::vector<int> GetWorkspaceRefIndexs() const;
void set_stream_id(uint32_t stream_id) { stream_id_ = stream_id; }
uint32_t stream_id() const { return stream_id_; }
void set_kernel_name(const std::string &kernel_name) { kernel_name_ = kernel_name; }
std::string kernel_name() const { return kernel_name_; }
void set_scope_full_name(const std::string &scop_name) { scop_full_name_ = scop_name; }
std::string scope_full_name() const { return scop_full_name_; }
void InsertInputKernel(const std::shared_ptr<KernelDef> &input_kernel) { input_kernels_.insert(input_kernel); }
const std::set<std::shared_ptr<KernelDef>> &input_kernels() { return input_kernels_; }
private:
std::string scop_full_name_;
......@@ -87,6 +90,7 @@ class KernelDef {
uint32_t stream_id_{0};
KernelRefCountPtrList input_refs_;
KernelRefCountPtrList output_refs_;
std::set<std::shared_ptr<KernelDef>> input_kernels_;
};
using KernelDefPtr = std::shared_ptr<KernelDef>;
} // namespace memreuse
......
mindspore/ccsrc/pre_activate/mem_reuse/mem_reuse.cc
浏览文件 @ 5e2f440e
......@@ -245,6 +245,34 @@ void MemReuseUtil::SetKernelDefMap() {
kernel_def_ptr->set_input_refs(kernel_def_ptr->inputs_[key]);
kernel_def_ptr->set_output_refs(kernel_def_ptr->outputs_[key]);
kernel_def_ptr_list_.push_back(kernel_def_ptr);
kernel_map_[key] = kernel_def_ptr;
}
SetKernelDefInputs();
}
void MemReuseUtil::SetKernelDefInputs() {
for (const auto &kernel : graph_->execution_order()) {
auto key = kernel.get();
// find kernel_def according to cnode addr
auto iter = kernel_map_.find(key);
if (iter == kernel_map_.end()) {
MS_LOG(EXCEPTION) << "kernel [" << kernel->fullname_with_scope() << "] is not init.";
}
auto kernel_def = iter->second;
for (size_t i = 0; i < AnfAlgo::GetInputTensorNum(kernel); ++i) {
auto ref_ptr = GetKernelInputRef(kernel, i);
if (ref_ptr != nullptr) {
// set the inputs of this kernel_def
auto input_node = AnfAlgo::GetInputNode(kernel, i);
auto input = AnfAlgo::VisitKernel(input_node, 0);
auto input_key = (input.first).get();
auto input_iter = kernel_map_.find(input_key);
if (input_iter == kernel_map_.end()) {
MS_LOG(EXCEPTION) << "kernel [" << (input.first)->fullname_with_scope() << "] is not init.";
}
kernel_def->InsertInputKernel(input_iter->second);
}
}
}
}
......
mindspore/ccsrc/pre_activate/mem_reuse/mem_reuse.h
浏览文件 @ 5e2f440e
......@@ -61,6 +61,7 @@ class MemReuseUtil {
void SetInputMap(const CNodePtr &kernel, KernelDef *kernel_def_ptr);
void SetOutputMap(const CNodePtr &kernel, KernelDef *kernel_def_ptr);
void SetWkMap(const CNodePtr &kernel, KernelDef *kernel_def_ptr);
void SetKernelDefInputs();
void SetReuseRefCount();
// Set the reference count of graph output specially.
void SetGraphOutputRefCount();
......@@ -94,6 +95,8 @@ class MemReuseUtil {
size_t total_workspace_size_ = 0;
size_t total_reuseworkspace_size_ = 0;
uint8_t *mem_base_{nullptr};
// kernel_map_: key is the AnfNodePtr addr, value is the KernelDef
std::map<KernelKey, KernelDefPtr> kernel_map_;
};
using MemReuseUtilPtr = std::shared_ptr<MemReuseUtil>;
} // namespace memreuse
......
mindspore/ccsrc/pre_activate/mem_reuse/mem_reuse_allocator.cc
浏览文件 @ 5e2f440e
......@@ -15,9 +15,6 @@
*/
#include "pre_activate/mem_reuse/mem_reuse_allocator.h"
#include <memory>
#include <algorithm>
#include <set>
#include "pre_activate/mem_reuse/mem_reuse.h"
#include "pre_activate/mem_reuse/mem_reuse_checker.h"
......@@ -25,9 +22,9 @@ namespace mindspore {
namespace memreuse {
void BestFitMemReuse::InitMemReuseInfo(const MemReuseUtil *mem_reuse_util_ptr) {
MS_EXCEPTION_IF_NULL(mem_reuse_util_ptr);
tensor_ptr_list_ = mem_reuse_util_ptr->total_refs_list();
wk_tensor_list_ = mem_reuse_util_ptr->total_wk_ref_list();
op_ptr_list_ = mem_reuse_util_ptr->kernel_def_ptr_list();
set_tensor_ptr_list(mem_reuse_util_ptr->total_refs_list());
set_workspace_ptr_list(mem_reuse_util_ptr->total_wk_ref_list());
set_op_ptr_list(mem_reuse_util_ptr->kernel_def_ptr_list());
// check info Correctness
for (auto &tensor : tensor_ptr_list_) {
tensor->size_ = AlignMemorySize(tensor->size_);
......@@ -37,63 +34,65 @@ void BestFitMemReuse::InitMemReuseInfo(const MemReuseUtil *mem_reuse_util_ptr) {
wk->size_ = AlignMemorySize(wk->size_);
wk->ref_count_ = 1;
}
auto stream_reuse = std::make_shared<StreamReuse>();
stream_reuse->SetStreamReuseResource();
parallel_streams_map_ = stream_reuse->parallel_streams_map();
}
bool BestFitMemReuse::CheckMembufIndx(const std::vector<MembufPtr> &membuf_ptr_list, size_t check_idx) const {
return check_idx < membuf_ptr_list.size();
}
bool BestFitMemReuse::IsMembufListEmpty(const std::vector<MembufPtr> &membuf_ptr_list) const {
return membuf_ptr_list.empty();
}
int BestFitMemReuse::GetFacIdx(size_t real_idx, int flag) const {
if (flag == kDyFac) {
return SizeToInt(real_idx);
} else if (flag == kWkFac) {
auto wk_fac_idx = kWkIndexFactor * SizeToInt(real_idx + 1);
return wk_fac_idx;
} else {
MS_LOG(EXCEPTION) << "flag " << flag << " is invalid";
void BestFitMemReuse::InitKernelDependence() {
for (const auto &kernel : op_ptr_list_) {
std::set<KernelDefPtr> front;
std::queue<KernelDefPtr> to_visit;
to_visit.push(kernel);
// find all kernels before current kernel
while (!to_visit.empty()) {
auto curr = to_visit.front();
to_visit.pop();
if (front.count(curr)) {
continue;
}
front.insert(curr);
auto iter = kernel_front_map_.find(curr);
if (iter != kernel_front_map_.end()) {
auto visited_front = iter->second;
front.insert(visited_front.begin(), visited_front.end());
continue;
}
for (const auto &input : curr->input_kernels()) {
to_visit.push(input);
}
}
kernel_front_map_[kernel] = front;
}
}
int BestFitMemReuse::GetRealIdx(int fac_idx, int flag) const {
// membuf index maybe invalid_index
if (fac_idx == kInvalidIndex) {
MS_LOG(EXCEPTION) << "this membuf index is invalid";
bool BestFitMemReuse::IsUsable(const KernelDefPtr &kernel_curr, const KernelDefPtr &kernel_prev) {
// determine whether the kernel_curr can reuse kernel_prev's output tensor membuf
MS_EXCEPTION_IF_NULL(kernel_curr);
MS_EXCEPTION_IF_NULL(kernel_prev);
auto curr_stream_id = kernel_curr->stream_id();
auto prev_stream_id = kernel_prev->stream_id();
if (curr_stream_id == prev_stream_id) {
return true;
}
if (flag == kDyFac) {
return fac_idx;
} else if (flag == kWkFac) {
if (fac_idx % 10 == 0) {
auto wk_fac_idx = fac_idx / kWkIndexFactor + 1;
return wk_fac_idx;
} else {
MS_LOG(EXCEPTION) << "fac_idx: " << fac_idx << "is invalid";
}
} else {
MS_LOG(EXCEPTION) << "flag: " << flag << " is invalid";
auto iter = kernel_front_map_.find(kernel_curr);
if (iter == kernel_front_map_.end()) {
MS_LOG(EXCEPTION) << kernel_curr->scope_full_name() << " is not init.";
}
auto kernel_curr_front = iter->second;
return kernel_curr_front.count(kernel_prev);
}
void BestFitMemReuse::AssignNodeOutputOffset(const KernelDef *kernel_def_ptr) {
MS_EXCEPTION_IF_NULL(kernel_def_ptr);
for (auto &tensor_idx : kernel_def_ptr->GetOutputRefIndexs()) {
CheckTensorIndex(tensor_idx);
auto tensor_desc = tensor_ptr_list_[IntToSize(tensor_idx)];
void BestFitMemReuse::AssignNodeOutputOffset() {
for (auto &tensor_idx : current_kernel_->GetOutputRefIndexs()) {
size_t index = GetTensorIndex(tensor_idx);
auto tensor_desc = tensor_ptr_list_[index];
MS_EXCEPTION_IF_NULL(tensor_desc);
auto reusable_membuf_map = GetReusableMembufMap(tensor_desc->size_);
if (!reusable_membuf_map.empty()) {
auto membuf_index = reusable_membuf_map.begin()->second;
// find the best suitable membuf in membuf list, and reuse it
ReuseExistMembuf(tensor_desc.get(), membuf_index, kDyFac);
ReuseExistMembuf(tensor_desc.get(), membuf_index, kDynamicMem);
} else {
// no membuf can reuse, add new membuf after the membuf_ptr_list
AddNewMembufPtr(tensor_desc.get(), kDyFac);
AddNewMembufPtr(tensor_desc.get(), kDynamicMem);
#ifdef MEM_REUSE_DEBUG
MemReuseChecker::GetInstance().IsAddNewMembuf_ = true;
#endif
......@@ -101,43 +100,24 @@ void BestFitMemReuse::AssignNodeOutputOffset(const KernelDef *kernel_def_ptr) {
}
}
void BestFitMemReuse::AssignNodeWkOffset(const KernelDef *kernel_def_ptr) {
MS_EXCEPTION_IF_NULL(kernel_def_ptr);
for (auto &wk_idx : kernel_def_ptr->GetWkRefIndexs()) {
if (IntToSize(wk_idx) >= wk_tensor_list_.size()) {
MS_LOG(EXCEPTION) << "wk_idx: " << wk_idx << " is invalid";
}
auto wk_ref = wk_tensor_list_[IntToSize(wk_idx)];
void BestFitMemReuse::AssignNodeWorkspaceOffset() {
for (auto &wk_idx : current_kernel_->GetWorkspaceRefIndexs()) {
size_t index = GetWorkspaceIndex(wk_idx);
auto wk_ref = wk_tensor_list_[index];
MS_EXCEPTION_IF_NULL(wk_ref);
auto re_wk_membuf_map = GetReusableMembufMap(wk_ref->size_);
if (!re_wk_membuf_map.empty()) {
auto membuf_index = re_wk_membuf_map.begin()->second;
ReuseExistMembuf(wk_ref.get(), membuf_index, kWkFac);
ReuseExistMembuf(wk_ref.get(), membuf_index, kWorkspaceMem);
} else {
AddNewMembufPtr(wk_ref.get(), kWkFac);
}
}
}
// releas pre node wk
void BestFitMemReuse::ReleasePreNodeWkSpace(const KernelDef *kernel_def_ptr) {
for (auto &wk_idx : kernel_def_ptr->GetWkRefIndexs()) {
auto wk_index = IntToSize(wk_idx);
if (wk_index >= wk_tensor_list_.size()) {
MS_LOG(EXCEPTION) << "wk_index: " << wk_index << " is larger than wk_tensor_list size" << wk_tensor_list_.size();
}
auto wk_tensor = wk_tensor_list_[wk_index];
wk_tensor->ref_count_--;
if (wk_tensor->ref_count_ == 0) {
ReleaseMembuf(wk_index, kWkFac);
AddNewMembufPtr(wk_ref.get(), kWorkspaceMem);
}
}
}
void BestFitMemReuse::ReuseExistMembuf(KernelRefCount *tensor_desc, size_t membuf_index, int flag) {
MS_EXCEPTION_IF_NULL(tensor_desc);
if (!CheckMembufIndx(membuf_ptr_list_, membuf_index)) {
return;
}
CheckMembufIndx(membuf_index);
auto membuf = membuf_ptr_list_[membuf_index];
MS_EXCEPTION_IF_NULL(membuf);
// first to split && then update membuf_info
......@@ -153,11 +133,9 @@ std::map<size_t, size_t> BestFitMemReuse::GetReusableMembufMap(size_t tensor_siz
std::map<size_t, size_t> size_map;
for (size_t i = 0; i < membuf_ptr_list_.size(); ++i) {
auto membuf = membuf_ptr_list_[i];
auto called_ids = membuf->called_stream_ids_;
auto index = i;
bool IsMembufOk = membuf->status_ == kUnused && membuf->size_ >= tensor_size;
bool has_parallel_id = HasParallelId(called_ids, current_stream_id_);
if (IsMembufOk && !has_parallel_id) {
bool is_membuf_ok = membuf->status_ == kUnused && membuf->size_ >= tensor_size;
if (is_membuf_ok && IsUsable(current_kernel_, membuf->used_kernel_)) {
(void)size_map.insert(std::make_pair(membuf->size_, index));
break;
}
......@@ -168,13 +146,10 @@ std::map<size_t, size_t> BestFitMemReuse::GetReusableMembufMap(size_t tensor_siz
void BestFitMemReuse::UpdateMembufInfo(KernelRefCount *tensor_desc, Membuf *membuf, int flag) {
MS_EXCEPTION_IF_NULL(tensor_desc);
MS_EXCEPTION_IF_NULL(membuf);
auto fac_idx = GetFacIdx(IntToSize(tensor_desc->index_), flag);
auto real_index = GetRealIndex(IntToSize(tensor_desc->index_), flag);
membuf->status_ = kReused;
membuf->stream_id_ = current_stream_id_;
// clear before called_ids
membuf->called_stream_ids_.clear();
(void)membuf->called_stream_ids_.insert(current_stream_id_);
membuf->index_ = fac_idx;
membuf->index_ = real_index;
membuf->used_kernel_ = current_kernel_;
tensor_desc->offset_ = membuf->offset_;
}
......@@ -182,52 +157,39 @@ bool BestFitMemReuse::IsSplit(size_t tensor_size, size_t membuf_size) const { re
void BestFitMemReuse::SplitMembuf(const KernelRefCount *tensor_desc, size_t membuf_index) {
MS_EXCEPTION_IF_NULL(tensor_desc);
if (!CheckMembufIndx(membuf_ptr_list_, membuf_index)) {
return;
}
CheckMembufIndx(membuf_index);
auto membuf = membuf_ptr_list_[membuf_index];
MS_EXCEPTION_IF_NULL(membuf);
auto bias = membuf->size_ - tensor_desc->size_;
membuf->size_ = tensor_desc->size_;
// to check if spilt membuf can be merge
auto new_membuf =
std::make_shared<Membuf>(current_stream_id_, kUnused, bias, membuf->offset_ + membuf->size_, kInvalidIndex);
std::make_shared<Membuf>(kUnused, bias, membuf->offset_ + membuf->size_, kInvalidIndex, current_kernel_);
(void)membuf_ptr_list_.insert(membuf_ptr_list_.begin() + SizeToInt(membuf_index + 1), new_membuf);
MergeCalledIds(membuf.get(), new_membuf.get());
}
void BestFitMemReuse::AddNewMembufPtr(KernelRefCount *tensor_desc, int flag) {
MS_EXCEPTION_IF_NULL(tensor_desc);
size_t membuf_offset = std::accumulate(membuf_ptr_list_.begin(), membuf_ptr_list_.end(), IntToSize(0),
[](size_t sum, MembufPtr &membuf) { return sum + membuf->size_; });
size_t membuf_size = tensor_desc->size_;
auto fac_idx = GetFacIdx(IntToSize(tensor_desc->index_), flag);
auto membuf = std::make_shared<Membuf>(current_stream_id_, kReused, membuf_size, membuf_offset, fac_idx);
size_t membuf_offset = 0;
if (!membuf_ptr_list_.empty()) {
membuf_offset = membuf_ptr_list_.back()->offset_ + membuf_ptr_list_.back()->size_;
}
auto membuf_size = tensor_desc->size_;
auto real_index = GetRealIndex(IntToSize(tensor_desc->index_), flag);
auto membuf = std::make_shared<Membuf>(kReused, membuf_size, membuf_offset, real_index, current_kernel_);
membuf_ptr_list_.push_back(membuf);
tensor_desc->offset_ = membuf_offset;
(void)membuf->called_stream_ids_.insert(current_stream_id_);
}
void BestFitMemReuse::UpdateNodeInputAndMembuf(const KernelDef *kernel_def_ptr) {
void BestFitMemReuse::UpdateNodeInputAndMembuf() {
// process node input tensor
for (const auto &tensor_idx : kernel_def_ptr->GetInputRefIndexs()) {
auto tensor_index = IntToSize(tensor_idx);
CheckTensorIndex(tensor_idx);
for (const auto &tensor_idx : current_kernel_->GetInputRefIndexs()) {
size_t tensor_index = GetTensorIndex(tensor_idx);
auto tensor_desc = tensor_ptr_list_[tensor_index];
auto fac_idx = GetFacIdx(tensor_index, kDyFac);
MS_EXCEPTION_IF_NULL(tensor_desc);
tensor_desc->ref_count_--;
// find tensor_index -> membuf update it's called_ids
for (size_t i = 0; i < membuf_ptr_list_.size(); ++i) {
auto membuf = membuf_ptr_list_[i];
// find it
if (membuf->index_ == fac_idx) {
(void)membuf->called_stream_ids_.insert(current_stream_id_);
break;
}
}
if (tensor_desc->ref_count_ == 0) {
ReleaseMembuf(tensor_index, kDyFac);
ReleaseMembuf(tensor_index, kDynamicMem);
} else if (tensor_desc->ref_count_ < 0) {
MS_LOG(EXCEPTION) << "tensor: " << tensor_desc->index_ << " refcount: " << tensor_desc->ref_count_
<< " check error";
......@@ -235,14 +197,13 @@ void BestFitMemReuse::UpdateNodeInputAndMembuf(const KernelDef *kernel_def_ptr)
}
}
void BestFitMemReuse::ReleaseNodeUnusedOutput(const KernelDef *kernel_def_ptr) {
for (auto &tensor_idx : kernel_def_ptr->GetOutputRefIndexs()) {
auto tensor_index = IntToSize(tensor_idx);
CheckTensorIndex(tensor_idx);
void BestFitMemReuse::ReleaseNodeUnusedOutput() {
for (auto &tensor_idx : current_kernel_->GetOutputRefIndexs()) {
size_t tensor_index = GetTensorIndex(tensor_idx);
auto tensor_desc = tensor_ptr_list_[tensor_index];
MS_EXCEPTION_IF_NULL(tensor_desc);
if (tensor_desc->ref_count_ == 0) {
ReleaseMembuf(tensor_index, kDyFac);
ReleaseMembuf(tensor_index, kDynamicMem);
} else if (tensor_desc->ref_count_ < 0) {
MS_LOG(EXCEPTION) << "tensor: " << tensor_desc->index_ << " refcount: " << tensor_desc->ref_count_
<< " check error";
......@@ -250,124 +211,57 @@ void BestFitMemReuse::ReleaseNodeUnusedOutput(const KernelDef *kernel_def_ptr) {
}
}
size_t BestFitMemReuse::FindIndx(const std::vector<MembufPtr> &membuf_ptr_list, int fac_idx) const {
size_t membuf_index = membuf_ptr_list.size();
for (size_t n = 0; n < membuf_ptr_list.size(); ++n) {
auto membuf = membuf_ptr_list[n];
MS_EXCEPTION_IF_NULL(membuf);
if (membuf->index_ == fac_idx) {
membuf_index = n;
break;
void BestFitMemReuse::ReleasePreNodeWorkspace(const KernelDef *kernel_def_ptr) {
for (auto &workspace_index : kernel_def_ptr->GetWorkspaceRefIndexs()) {
size_t index = GetWorkspaceIndex(workspace_index);
auto wk_tensor = wk_tensor_list_[index];
wk_tensor->ref_count_--;
if (wk_tensor->ref_count_ == 0) {
ReleaseMembuf(index, kWorkspaceMem);
} else if (wk_tensor->ref_count_ < 0) {
MS_LOG(EXCEPTION) << "tensor: " << wk_tensor->index_ << " refcount: " << wk_tensor->ref_count_ << " check error";
}
}
return membuf_index;
}
void BestFitMemReuse::ReleaseMembuf(size_t tensor_index, int flag) {
auto fac_idex = GetFacIdx(tensor_index, flag);
auto membuf_index = FindIndx(membuf_ptr_list_, fac_idex);
if (!CheckMembufIndx(membuf_ptr_list_, membuf_index)) {
if (membuf_ptr_list_.empty()) {
return;
}
auto membuf = membuf_ptr_list_[membuf_index];
auto real_index = GetRealIndex(tensor_index, flag);
auto membuf_iter = std::find_if(membuf_ptr_list_.begin(), membuf_ptr_list_.end(),
[real_index](const MembufPtr &membuf) { return membuf->index_ == real_index; });
if (membuf_iter == membuf_ptr_list_.end()) {
return;
}
auto membuf = (*membuf_iter);
MS_EXCEPTION_IF_NULL(membuf);
membuf->status_ = kUnused;
if (membuf_index != (membuf_ptr_list_.size() - 1)) {
auto membuf_next = membuf_ptr_list_[membuf_index + 1];
if (membuf_iter != membuf_ptr_list_.end() - 1) {
auto next_iter = membuf_iter + 1;
auto membuf_next = (*next_iter);
MS_EXCEPTION_IF_NULL(membuf_next);
bool has_parallel_id = false;
for (auto &cal_id : membuf->called_stream_ids_) {
has_parallel_id = HasParallelId(membuf_next->called_stream_ids_, cal_id);
if (has_parallel_id) {
break;
}
}
if (membuf_next->status_ == kUnused && !has_parallel_id) {
membuf->size_ += membuf_next->size_;
MergeCalledIds(membuf_next.get(), membuf.get());
auto it = membuf_ptr_list_.begin() + SizeToInt(membuf_index + 1);
(void)membuf_ptr_list_.erase(it);
}
}
if (membuf_index != 0) {
if (!CheckMembufIndx(membuf_ptr_list_, membuf_index - 1)) {
return;
}
auto membuf_prev = membuf_ptr_list_[membuf_index - 1];
MS_EXCEPTION_IF_NULL(membuf_prev);
bool has_parallel_id = false;
for (auto &cal_id : membuf->called_stream_ids_) {
has_parallel_id = HasParallelId(membuf_prev->called_stream_ids_, cal_id);
if (has_parallel_id) {
break;
}
}
if (membuf_prev->status_ == kUnused && !has_parallel_id) {
membuf->size_ += membuf_prev->size_;
membuf->offset_ = membuf_prev->offset_;
MergeCalledIds(membuf_prev.get(), membuf.get());
auto it = membuf_ptr_list_.begin() + SizeToInt(membuf_index - 1);
(void)membuf_ptr_list_.erase(it);
}
}
}
bool BestFitMemReuse::HasParallelId(const std::set<uint32_t> &called_ids, uint32_t curr_id) {
if (called_ids.empty()) {
MS_LOG(EXCEPTION) << "There is a invalid WkMembuf,called_ids is empty";
}
for (auto item : called_ids) {
if (!IsReusableStream(curr_id, item)) {
return true;
}
}
return false;
}
void BestFitMemReuse::MergeCalledIds(const Membuf *membuf_target, Membuf *membuf) {
MS_EXCEPTION_IF_NULL(membuf_target);
MS_EXCEPTION_IF_NULL(membuf);
for (auto target : membuf_target->called_stream_ids_) {
(void)membuf->called_stream_ids_.insert(target);
}
}
void BestFitMemReuse::ReleaseParallStream() {
std::vector<size_t> target_relea_idxs;
for (size_t i = 0; i < membuf_ptr_list_.size(); ++i) {
auto membuf = membuf_ptr_list_[i];
if (membuf->status_ == kReused) {
continue;
}
// for begin to end, so no need merge pre_membuf
if (i != (membuf_ptr_list_.size() - 1)) {
auto membuf_next = membuf_ptr_list_[i + 1];
if (membuf_next->status_ == kReused) {
continue;
}
MS_EXCEPTION_IF_NULL(membuf_next);
// judge current id no parallel fro membuf && membuf_next
bool has_parallel_id_crr = HasParallelId(membuf->called_stream_ids_, current_stream_id_);
bool has_parallel_id_next = HasParallelId(membuf_next->called_stream_ids_, current_stream_id_);
if (membuf->status_ == kUnused && membuf_next->status_ == kUnused && !has_parallel_id_crr &&
!has_parallel_id_next) {
if (membuf_next->status_ == kUnused) {
bool is_merge = IsUsable(current_kernel_, membuf_next->used_kernel_);
if (is_merge) {
membuf->size_ += membuf_next->size_;
MergeCalledIds(membuf_next.get(), membuf.get());
target_relea_idxs.push_back(i + 1);
(void)membuf_ptr_list_.erase(next_iter);
}
}
}
// erase all target membuf
std::vector<MembufPtr> membuf_ptr_list_tmp;
for (size_t j = 0; j < membuf_ptr_list_.size(); ++j) {
for (auto idx : target_relea_idxs) {
if (j != idx) {
membuf_ptr_list_tmp.push_back(membuf_ptr_list_[j]);
if (membuf_iter != membuf_ptr_list_.begin()) {
auto prev_iter = membuf_iter - 1;
auto membuf_prev = (*prev_iter);
MS_EXCEPTION_IF_NULL(membuf_prev);
if (membuf_prev->status_ == kUnused) {
bool is_merge = IsUsable(current_kernel_, membuf_prev->used_kernel_);
if (is_merge) {
membuf->size_ += membuf_prev->size_;
membuf->offset_ = membuf_prev->offset_;
(void)membuf_ptr_list_.erase(prev_iter);
}
}
}
membuf_ptr_list_.clear();
(void)std::copy(membuf_ptr_list_tmp.begin(), membuf_ptr_list_tmp.end(), back_inserter(membuf_ptr_list_));
}
size_t BestFitMemReuse::AlignMemorySize(size_t size) const {
......@@ -380,74 +274,83 @@ size_t BestFitMemReuse::GetAllocatedSize() {
if (membuf_ptr_list_.empty()) {
return AllocatedSize;
}
AllocatedSize = (*membuf_ptr_list_.rbegin())->offset_ + (*membuf_ptr_list_.rbegin())->size_;
AllocatedSize = membuf_ptr_list_.back()->offset_ + membuf_ptr_list_.back()->size_;
MS_LOG(INFO) << "MemReuse Allocated Dynamic Size: " << AllocatedSize;
return AllocatedSize;
}
/**
* parallel_streams_map: key, current_stream_id; value, streams parallel to current stream
* @param curr_stream_id
* @param target_stream_id
* @return bool, if the target stream can be reused by current stream
*/
bool BestFitMemReuse::IsReusableStream(uint32_t curr_stream_id, uint32_t target_stream_id) {
auto iter_parall = parallel_streams_map_.find(curr_stream_id);
if (parallel_streams_map_.empty() || (iter_parall == parallel_streams_map_.end())) {
// no parallel stream exists
return true;
}
auto curr_parallel_set = iter_parall->second;
return curr_parallel_set.find(target_stream_id) == curr_parallel_set.end();
}
bool BestFitMemReuse::IsRelease(const std::string &kernel_name) {
bool BestFitMemReuse::IsRelease() {
// unable_used_node include the node type that output tensor cannot be released,
// even if its refcount is equal to zero.
std::unordered_set<std::string> unable_used_node = {prim::kPrimBatchNorm->name(), prim::kPrimBatchNormGrad->name(),
prim::kPrimFusedBatchNorm->name(),
prim::kPrimFusedBatchNormGrad->name()};
return unable_used_node.find(kernel_name) == unable_used_node.end();
return unable_used_node.find(current_kernel_->kernel_name()) == unable_used_node.end();
}
void BestFitMemReuse::CheckTensorIndex(int tensor_index) const {
if (tensor_index < 0) {
MS_LOG(EXCEPTION) << "warning, please check tensor info.";
size_t BestFitMemReuse::GetTensorIndex(int index) const {
if (index < 0 || IntToSize(index) >= tensor_ptr_list_.size()) {
MS_LOG(WARNING) << "current cnode: " << current_kernel_->scope_full_name();
MS_LOG(EXCEPTION) << "invalid tensor index";
}
if (IntToSize(tensor_index) >= tensor_ptr_list_.size()) {
return IntToSize(index);
}
size_t BestFitMemReuse::GetWorkspaceIndex(int index) const {
if (index < 0 || IntToSize(index) >= wk_tensor_list_.size()) {
MS_LOG(WARNING) << "current cnode: " << current_kernel_->scope_full_name();
MS_LOG(EXCEPTION) << "invalid tensor index";
}
return IntToSize(index);
}
int BestFitMemReuse::GetRealIndex(size_t index, int flag) const {
if (flag == kDynamicMem) {
return SizeToInt(index);
} else if (flag == kWorkspaceMem) {
return kWorkspaceIndexFactor * SizeToInt(index + 1);
} else {
MS_LOG(EXCEPTION) << "flag " << flag << " is invalid";
}
}
void BestFitMemReuse::CheckMembufIndx(size_t membuf_index) const {
if (membuf_index >= membuf_ptr_list_.size()) {
MS_LOG(WARNING) << "current cnode: " << current_kernel_->scope_full_name();
MS_LOG(EXCEPTION) << "invalid membuf index: " << membuf_index << ", real size: " << membuf_ptr_list_.size();
}
}
void BestFitMemReuse::Reuse(const MemReuseUtil *mem_reuse_util_ptr) {
MS_EXCEPTION_IF_NULL(mem_reuse_util_ptr);
InitMemReuseInfo(mem_reuse_util_ptr);
InitKernelDependence();
KernelDefPtr pre_op = nullptr;
#ifdef MEM_REUSE_DEBUG
size_t op_num = 0;
#endif
for (const auto &op_def_ptr : op_ptr_list_) {
current_stream_id_ = op_def_ptr->stream_id();
current_kernel_ = op_def_ptr;
// releas pre_op_def
if (pre_op != nullptr) {
ReleasePreNodeWkSpace(pre_op.get());
ReleasePreNodeWorkspace(pre_op.get());
}
MemReuseChecker::GetInstance().IsAddNewMembuf_ = false;
// process node output tensor
AssignNodeOutputOffset(op_def_ptr.get());
AssignNodeOutputOffset();
#ifdef MEM_REUSE_DEBUG
if (MemReuseChecker::GetInstance().IsAddNewMembuf_) {
MemReuseChecker::GetInstance().SetAddNewMembuInfos(op_def_ptr.get(), membuf_ptr_list_, op_num);
}
#endif
// deal with current op'workspace
AssignNodeWkOffset(op_def_ptr.get());
AssignNodeWorkspaceOffset();
pre_op = op_def_ptr;
// update node input tensor refcount, and membuf list status
UpdateNodeInputAndMembuf(op_def_ptr.get());
UpdateNodeInputAndMembuf();
// check node output tensor which refcount is equal to zero
if (IsRelease(op_def_ptr->kernel_name())) {
ReleaseNodeUnusedOutput(op_def_ptr.get());
if (IsRelease()) {
ReleaseNodeUnusedOutput();
}
#ifdef MEM_REUSE_DEBUG
MemReuseChecker::GetInstance().SetMembuInfos(op_def_ptr.get(), membuf_ptr_list_);
......@@ -457,6 +360,8 @@ void BestFitMemReuse::Reuse(const MemReuseUtil *mem_reuse_util_ptr) {
#ifdef MEM_REUSE_DEBUG
MemReuseChecker::GetInstance().ExportMembufInfoIR();
MemReuseChecker::GetInstance().ExportAddNewMmebufIR();
MemReuseChecker::GetInstance().set_kernel_front_map(kernel_front_map_);
MemReuseChecker::GetInstance().ExportKernelDependence();
#endif
}
} // namespace memreuse
......
mindspore/ccsrc/pre_activate/mem_reuse/mem_reuse_allocator.h
浏览文件 @ 5e2f440e
......@@ -29,31 +29,30 @@
#include <unordered_map>
#include <unordered_set>
#include <set>
#include <queue>
#include "pre_activate/mem_reuse/kernel_refcount.h"
#include "pre_activate/mem_reuse/mem_reuse.h"
#include "pre_activate/mem_reuse/stream_reuse.h"
namespace mindspore {
namespace memreuse {
static constexpr int kWkIndexFactor = -1000;
static constexpr int kDyFac = -1;
static constexpr int kWkFac = 1;
static constexpr int kWorkspaceIndexFactor = -1000;
static constexpr int kDynamicMem = -1;
static constexpr int kWorkspaceMem = 1;
static constexpr size_t kTotalSize = 0;
enum Status { kUnused, kReused };
class Membuf {
public:
Membuf() = default;
Membuf(uint32_t stream_id, Status status, size_t size, size_t offset, int index)
: stream_id_(stream_id), status_(status), size_(size), offset_(offset), index_(index) {}
Membuf(Status status, size_t size, size_t offset, int index, const KernelDefPtr &used_kernel)
: status_(status), size_(size), offset_(offset), index_(index), used_kernel_(used_kernel) {}
~Membuf() = default;
// Memory block status flags
std::set<uint32_t> called_stream_ids_;
uint32_t stream_id_{0};
Status status_ = kUnused;
size_t size_{0};
size_t offset_{0};
// Store the tensor index stored in this memory block at a certain moment
int index_{0};
KernelDefPtr used_kernel_;
};
using MembufPtr = std::shared_ptr<Membuf>;
......@@ -61,24 +60,45 @@ class BestFitMemReuse {
public:
BestFitMemReuse() = default;
~BestFitMemReuse() { membuf_ptr_list_.clear(); }
// Init all information need by memory reuse
/**
* Init all information need by memory reuse
* @param mem_reuse_util_ptr, initialize in the memreuse.cc
*/
void InitMemReuseInfo(const MemReuseUtil *mem_reuse_util_ptr);
bool CheckMembufIndx(const std::vector<MembufPtr> &membuf_ptr_list, size_t check_idx) const;
bool IsMembufListEmpty(const std::vector<MembufPtr> &membuf_ptr_list) const;
void AssignNodeWkOffset(const KernelDef *kernel_def_ptr);
void ReleasePreNodeWkSpace(const KernelDef *kernel_def_ptr);
// void assign node output tensor memory offset
void AssignNodeOutputOffset(const KernelDef *kernel_def_ptr);
void ReleaseParallStream();
// update node input tensor refcount, and membuf list status
void UpdateNodeInputAndMembuf(const KernelDef *kernel_def_ptr);
// check node output tensor which refcount is equal to zero
void ReleaseNodeUnusedOutput(const KernelDef *kernel_def_ptr);
// If there are memory blocks that can be reused
void CheckMembufIndx(size_t check_idx) const;
void AssignNodeWorkspaceOffset();
void ReleasePreNodeWorkspace(const KernelDef *kernel_def_ptr);
/**
* Assign output tensor memory offset of current kernel
*/
void AssignNodeOutputOffset();
/**
* Update input tensor's status of current kernel, and the status of membuf used by current kernel
*/
void UpdateNodeInputAndMembuf();
/**
* Check whether to release the kernel output tensor which refcount is equal to zero
*/
void ReleaseNodeUnusedOutput();
/**
* Reuse the exist membuf if possible
* @param tensor_desc, the output tensor of current kernel
* @param membuf_index, the index of membuf to be reused
* @param flag
*/
void ReuseExistMembuf(KernelRefCount *tensor_desc, size_t membuf_index, int flag);
// Save memory blocks that can be reused to the map
/**
* Get the membuf that can be reused
* @param tensor_size, the size of the tensor ready to assign memory offset
* @return membuf map, key: the membuf size, value: the membuf index
*/
std::map<size_t, size_t> GetReusableMembufMap(size_t tensor_size);
// Update the status of the reused memory block
/**
* Update the status of the reused memory block
* @param tensor_desc, the tensor ready to assign memory
* @param membuf, the membuf to be reused
* @param flag, distinguish dynamic memory and workspace
*/
void UpdateMembufInfo(KernelRefCount *tensor_desc, Membuf *membuf, int flag);
// If the size of the memory block is greater than the size of the tensor, split the extra memory
void SplitMembuf(const KernelRefCount *tensor_desc, size_t membuf_index);
......@@ -88,30 +108,39 @@ class BestFitMemReuse {
void AddNewMembufPtr(KernelRefCount *tensor_desc, int flag);
// Merge unused membuf
void ReleaseMembuf(size_t tensor_index, int flag);
bool HasParallelId(const std::set<uint32_t> &called_ids, uint32_t curr_id);
void MergeCalledIds(const Membuf *membuf_target, Membuf *membuf);
// Memory address alignment 512
size_t AlignMemorySize(size_t size) const;
int GetFacIdx(size_t real_idx, int flag = kDyFac) const;
int GetRealIdx(int fac_idx, int flag = kDyFac) const;
size_t FindIndx(const std::vector<MembufPtr> &membuf_ptr_list, int fac_idx) const;
void CheckTensorIndex(int tensor_index) const;
int GetRealIndex(size_t index, int flag = kDynamicMem) const;
size_t GetTensorIndex(int index) const;
size_t GetWorkspaceIndex(int index) const;
// Memory reuse main program entry
void Reuse(const MemReuseUtil *mem_reuse_util_ptr);
// Get the total memory that needs to be applied eventually
size_t GetAllocatedSize();
// If the target stream can be reused by current stream
bool IsReusableStream(uint32_t curr_stream_id, uint32_t target_stream_id);
// return false, when the node output cannot be released
bool IsRelease(const std::string &kernel_name);
bool IsRelease();
/**
* determine if the kernel_curr can reuse the output tensor add of kernel_prev
* @param kernel_curr, current kernel
* @param kernel_prev, the membuf used by this kernel
* @return bool
*/
bool IsUsable(const KernelDefPtr &kernel_curr, const KernelDefPtr &kernel_prev);
/**
* init the dependence of all kernels in the graph
*/
void InitKernelDependence();
// set tensor_def and op_def
void set_tensor_ptr_list(const std::vector<KernelRefCountPtr> &tensor_ptr_list) {
tensor_ptr_list_ = tensor_ptr_list;
}
void set_workspace_ptr_list(const std::vector<KernelRefCountPtr> &workspace_ptr_list) {
wk_tensor_list_ = workspace_ptr_list;
}
void set_op_ptr_list(const std::vector<KernelDefPtr> &op_ptr_list) { op_ptr_list_ = op_ptr_list; }
private:
uint32_t current_stream_id_{0};
KernelDefPtr current_kernel_;
// Save all tensor information
std::vector<KernelRefCountPtr> tensor_ptr_list_;
std::vector<KernelRefCountPtr> wk_tensor_list_;
......@@ -119,7 +148,8 @@ class BestFitMemReuse {
std::vector<KernelDefPtr> op_ptr_list_;
// Memory block information sequence, temporary variables
std::vector<MembufPtr> membuf_ptr_list_;
std::unordered_map<uint32_t, std::unordered_set<uint32_t>> parallel_streams_map_;
// kernel_front_map_, key: the kernel_def, value: kernels before this kernel_def
std::map<KernelDefPtr, std::set<KernelDefPtr>> kernel_front_map_;
};
} // namespace memreuse
} // namespace mindspore
......
mindspore/ccsrc/pre_activate/mem_reuse/mem_reuse_checker.cc
浏览文件 @ 5e2f440e
......@@ -19,8 +19,6 @@
#include <vector>
#include <utility>
#include <string>
#include <unordered_map>
#include <unordered_set>
namespace mindspore {
namespace memreuse {
......@@ -188,6 +186,27 @@ void MemReuseChecker::CheckMemReuseIR(const KernelRefCountPtrList &total_refs_li
ofs.close();
}
void MemReuseChecker::ExportKernelDependence() {
std::string filename = "./memreuse_dependence.ir";
std::ofstream ofs(filename);
if (!ofs.is_open()) {
MS_LOG(ERROR) << "Open file [" << filename << "] failed!";
return;
}
size_t i = 0;
for (const auto &kernel_front : kernel_front_map_) {
auto kernel = kernel_front.first;
auto front = kernel_front.second;
ofs << "[" << i++ << "] " << kernel->scope_full_name() << "\n";
for (const auto &node : front) {
ofs << node->scope_full_name() << "\n";
}
ofs << "\n\n";
}
ofs.close();
}
bool MemReuseChecker::CheckGraphOutputAssigned(const session::KernelGraph *graph) {
// set real graph output node to be special who's refcount equal kMaxRefCount
for (const auto &output : graph->outputs()) {
......@@ -393,7 +412,7 @@ void MemReuseChecker::CheckNormalIR(const session::KernelGraph *graph) {
void MemReuseChecker::SetMembuInfos(const KernelDef *op_def, const std::vector<MembufPtr> &membuf_ptr_list) {
std::vector<MembufPtr> curr_mem_infos;
for (const auto &mem : membuf_ptr_list) {
auto mem_checker = std::make_shared<Membuf>(mem->stream_id_, mem->status_, mem->size_, mem->offset_, mem->index_);
auto mem_checker = std::make_shared<Membuf>(mem->status_, mem->size_, mem->offset_, mem->index_, mem->used_kernel_);
curr_mem_infos.push_back(mem_checker);
}
membuf_all_infos_.push_back(curr_mem_infos);
......@@ -407,7 +426,7 @@ void MemReuseChecker::SetAddNewMembuInfos(const KernelDef *op_def, const std::ve
std::vector<MembufPtr> add_new_curr_mem;
for (const auto &mem : membuf_ptr_list) {
auto mem_checker = std::make_shared<Membuf>(mem->stream_id_, mem->status_, mem->size_, mem->offset_, mem->index_);
auto mem_checker = std::make_shared<Membuf>(mem->status_, mem->size_, mem->offset_, mem->index_, mem->used_kernel_);
add_new_curr_mem.push_back(mem_checker);
}
add_new_mem_infos_.push_back(add_new_curr_mem);
......@@ -424,11 +443,11 @@ void MemReuseChecker::ExportMembufInfoIR() {
if (!ofs.is_open()) {
MS_LOG(ERROR) << "Open file [" << ir_file_name << "] failed!";
}
ofs << "total_ori_static_size:" << total_ori_static_size_ << "\n";
ofs << "total_ori_weight_size:" << total_ori_input_size_ << "\n";
ofs << "total_ori_constant_size:" << total_ori_value_size_ << "\n";
ofs << "total_ori_dy_size:" << total_ori_dy_size_ << "\n";
ofs << "total_ori_wkspace_size:" << total_ori_wkspace_size_ << "\n";
ofs << "Total static size:\t" << total_ori_static_size_ << "\n";
ofs << "Graph inputs size:\t" << total_ori_input_size_ << "\n";
ofs << "Value nodes size:\t" << total_ori_value_size_ << "\n";
ofs << "Total dynamic size:\t" << total_ori_dy_size_ << "\n";
ofs << "Total workspace size:\t" << total_ori_wkspace_size_ << "\n";
// get last membuf_list
if (membuf_all_infos_.empty()) {
return;
......@@ -438,8 +457,10 @@ void MemReuseChecker::ExportMembufInfoIR() {
auto checker_size = SizeToLong(membuf->size_);
total_reuse_size += checker_size;
}
ofs << "total_reuse_size:" << total_reuse_size << "\n";
ofs << "After reuse size:\t" << total_reuse_size << "\n\n";
size_t i = 0;
std::vector<size_t> each_node_used_size;
std::vector<size_t> each_node_allocated_size;
for (const auto &curr_membuf_list : membuf_all_infos_) {
ofs << all_split_names_.at(i) << "\n";
++i;
......@@ -449,17 +470,42 @@ void MemReuseChecker::ExportMembufInfoIR() {
<< "tensor_idex\t"
<< "mem_size\t"
<< "mem_head\t"
<< "mem_tail\n";
<< "mem_tail\t"
<< "used_kernel\n";
size_t curr_used = 0;
size_t curr_allocated = 0;
for (size_t j = 0; j < curr_membuf_list.size(); ++j) {
auto membuf = curr_membuf_list.at(j);
auto used_kernel = membuf->used_kernel_->scope_full_name();
ofs << "&" << j << "\t"
<< "streamID[@" << membuf->stream_id_ << "]"
<< "streamID[@" << membuf->used_kernel_->stream_id() << "]"
<< "\t"
<< "#" << static_cast<int>(membuf->status_) << "\t%" << membuf->index_ << "T"
<< "\t" << membuf->size_ << "\t" << membuf->offset_ << "\t" << membuf->offset_ + membuf->size_ << "\n";
<< "\t" << membuf->size_ << "\t" << membuf->offset_ << "\t" << membuf->offset_ + membuf->size_ << "\t"
<< GetSplitName(used_kernel) << "\n";
if (membuf->status_ == kReused) {
curr_used += membuf->size_;
}
}
if (!curr_membuf_list.empty()) {
curr_allocated = curr_membuf_list.back()->offset_ + curr_membuf_list.back()->size_;
}
each_node_used_size.push_back(curr_used);
each_node_allocated_size.push_back(curr_allocated);
ofs << "curr real used size: \t" << curr_used << "\n";
ofs << "curr allocated size: \t" << curr_allocated << "\n";
ofs << "\n\n";
}
ofs << "each node used size: \n";
for (auto size : each_node_used_size) {
ofs << size << "\t";
}
ofs << "\n\n";
ofs << "each node allocated size: \n";
for (auto size : each_node_allocated_size) {
ofs << size << "\t";
}
ofs << "\n\n";
ofs.close();
}
......@@ -479,7 +525,6 @@ void MemReuseChecker::ExportAddNewMmebufIR() {
<< "\n";
i++;
ofs << "mem_num\t"
<< "stream_id\t"
<< "status\t"
<< "tensor_idex\t"
<< "mem_size\t"
......@@ -490,7 +535,6 @@ void MemReuseChecker::ExportAddNewMmebufIR() {
for (size_t j = 0; j < curr_membuf_list.size(); ++j) {
auto membuf = curr_membuf_list.at(j);
ofs << "&" << j << "\t"
<< "streamID[@" << membuf->stream_id_ << "]"
<< "\t"
<< "#" << static_cast<int>(membuf->status_) << "\t%" << membuf->index_ << "T"
<< "\t" << membuf->size_ << "\t" << membuf->offset_ << "\t" << membuf->offset_ + membuf->size_ << "\t";
......
mindspore/ccsrc/pre_activate/mem_reuse/mem_reuse_checker.h
浏览文件 @ 5e2f440e
......@@ -17,6 +17,7 @@
#ifndef MINDSPORE_CCSRC_PRE_ACTIVATE_MEM_REUSE_MEM_REUSE_CHECKER_H_
#define MINDSPORE_CCSRC_PRE_ACTIVATE_MEM_REUSE_MEM_REUSE_CHECKER_H_
#include <map>
#include <set>
#include <vector>
#include <string>
#include <memory>
......@@ -59,10 +60,14 @@ class MemReuseChecker {
void ExportMembufInfoIR();
void SetAddNewMembuInfos(const KernelDef *op_def, const std::vector<MembufPtr> &membuf_ptr_list, size_t op_idx);
void ExportAddNewMmebufIR();
void set_kernel_front_map(const std::map<KernelDefPtr, std::set<KernelDefPtr>> &kernel_front_map) {
kernel_front_map_ = kernel_front_map;
}
void ExportKernelDependence();
private:
MemReuseChecker() = default;
~MemReuseChecker() { MS_LOG(INFO) << "Total reused workspace size: " << total_re_wkspe_size_checker_; }
~MemReuseChecker() {}
size_t total_re_wkspe_size_checker_{0};
std::vector<std::vector<MembufPtr>> membuf_all_infos_;
std::vector<const void *> nor_output_tensors_;
......@@ -79,6 +84,7 @@ class MemReuseChecker {
std::vector<std::string> all_split_names_;
std::map<int, std::vector<string>> tensor_from_;
std::map<int, std::vector<string>> tensor_to_;
std::map<KernelDefPtr, std::set<KernelDefPtr>> kernel_front_map_;
int64_t total_ori_static_size_ = 0;
int64_t total_ori_input_size_ = 0;
int64_t total_ori_value_size_ = 0;
......
mindspore/ccsrc/pre_activate/mem_reuse/stream_reuse.cc 已删除 100644 → 0
浏览文件 @ e7936ded
/**
* 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 "pre_activate/mem_reuse/stream_reuse.h"
namespace mindspore {
namespace memreuse {
void StreamReuse::SetStreamReuseResource() {
#ifdef ENABLE_D
auto logic_physic_map = device::ascend::AscendStreamAssign::GetInstance().logic_to_physic_map();
auto logic_independent_map = device::ascend::AscendStreamAssign::GetInstance().logic_to_independent_map();
MS_LOG(INFO) << "stream mem reuse for Davici";
if (!logic_independent_map.empty() && !logic_physic_map.empty()) {
set_logic_physic_map(logic_physic_map);
set_logic_independent_map(logic_independent_map);
InitReusableStreamMap();
} else {
MS_LOG(INFO) << "Non task sink or No Parallel stream exists";
}
#endif
MS_LOG(INFO) << "no need to set stream mem reuse resource";
}
std::vector<std::pair<uint32_t, uint32_t>> StreamReuse::SortLogicPhysicMapToList() {
std::vector<std::pair<uint32_t, uint32_t>> logic_physic_list;
(void)std::transform(logic_physic_map_.begin(), logic_physic_map_.end(), std::back_inserter(logic_physic_list),
[](std::pair<uint32_t, uint32_t> log_phy) { return log_phy; });
std::sort(
logic_physic_list.begin(), logic_physic_list.end(),
[](const std::pair<uint32_t, uint32_t> &logic_phyic_pair1, const std::pair<uint32_t, uint32_t> &logic_phyic_pair2) {
return logic_phyic_pair1.second < logic_phyic_pair2.second;
});
return logic_physic_list;
}
std::unordered_map<int, std::set<uint32_t>> StreamReuse::GetLogicPhysicsStreamMap() {
auto logic_physic_list = SortLogicPhysicMapToList();
std::unordered_map<int, std::set<uint32_t>> logic_phyics_map;
for (size_t i = 0; i < logic_physic_list.size() - IntToSize(1); ++i) {
auto curr_logic_physic = logic_physic_list.at(i);
auto next_logic_physic = logic_physic_list.at(i + 1);
for (auto j = curr_logic_physic.second; j < next_logic_physic.second; ++j) {
(void)logic_phyics_map[curr_logic_physic.first].insert(j);
}
}
// sort the logic independ map by value
std::map<uint32_t, uint32_t> temp_map;
for (const auto &logic_independ : logic_independent_map_) {
(void)temp_map.insert(std::make_pair(logic_independ.second, logic_independ.first));
}
auto first_independent_stream_id = (*temp_map.begin()).first;
auto last_physic_logic_stream_id = (*logic_physic_list.rbegin()).second;
for (auto i = last_physic_logic_stream_id; i < first_independent_stream_id; ++i) {
(void)logic_phyics_map[(*logic_physic_list.rbegin()).first].insert(i);
}
return logic_phyics_map;
}
void StreamReuse::InitReusableStreamMap() {
// logic_phyics_map, key, logic_stream_id; value, physic_strema_ids included in that logic stream
auto logic_phyics_map = GetLogicPhysicsStreamMap();
// parallel_streams_map: key, current_stream_id; value, streams parallel to current stream
for (const auto &logic_to_phyics : logic_phyics_map) {
auto logic_stream_id = logic_to_phyics.first;
auto iter_inde = logic_independent_map_.find(logic_stream_id);
if (iter_inde != logic_independent_map_.end()) {
// exist independent steam parallel to these logic streams
auto independent_stream_id = iter_inde->second;
auto physics_stream_id = logic_to_phyics.second;
for (const auto &physic : physics_stream_id) {
(void)parallel_streams_map_[physic].insert(independent_stream_id);
}
}
}
for (const auto &logic_to_independent : logic_independent_map_) {
auto logic_stream_id = logic_to_independent.first;
auto independent_stream_id = logic_to_independent.second;
auto iter_physics = logic_phyics_map.find(logic_stream_id);
if (iter_physics != logic_phyics_map.end()) {
// exist logic steam parallel to these independent streams, default
auto physics_set = iter_physics->second;
for (const auto &physic : physics_set) {
(void)parallel_streams_map_[independent_stream_id].insert(physic);
}
}
}
}
} // namespace memreuse
} // namespace mindspore
mindspore/ccsrc/pre_activate/mem_reuse/stream_reuse.h 已删除 100644 → 0
浏览文件 @ e7936ded
/**
* 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.
*/
#ifndef MINDSPORE_CCSRC_PRE_ACTIVATE_MEM_REUSE_STREAM_REUSE_H_
#define MINDSPORE_CCSRC_PRE_ACTIVATE_MEM_REUSE_STREAM_REUSE_H_
#include <cmath>
#include <map>
#include <set>
#include <list>
#include <memory>
#include <vector>
#include <numeric>
#include <algorithm>
#include <utility>
#include <fstream>
#include <unordered_set>
#include <unordered_map>
#include "session/anf_runtime_algorithm.h"
#include "pre_activate/mem_reuse/kernel_refcount.h"
#ifdef ENABLE_D
#include "device/ascend/ascend_stream_assign.h"
#endif
namespace mindspore {
namespace memreuse {
class StreamReuse {
public:
StreamReuse() = default;
~StreamReuse() = default;
void SetStreamReuseResource();
void InitReusableStreamMap();
std::vector<std::pair<uint32_t, uint32_t>> SortLogicPhysicMapToList();
std::unordered_map<int, std::set<uint32_t>> GetLogicPhysicsStreamMap();
void set_logic_physic_map(const std::unordered_map<uint32_t, uint32_t> &logic_physic_map) {
logic_physic_map_ = logic_physic_map;
}
void set_logic_independent_map(const std::unordered_map<uint32_t, uint32_t> &logic_independent_map) {
logic_independent_map_ = logic_independent_map;
}
std::unordered_map<uint32_t, std::unordered_set<uint32_t>> parallel_streams_map() { return parallel_streams_map_; }
private:
std::unordered_map<uint32_t, std::unordered_set<uint32_t>> parallel_streams_map_;
std::unordered_map<uint32_t, uint32_t> logic_physic_map_;
std::unordered_map<uint32_t, uint32_t> logic_independent_map_;
};
} // namespace memreuse
} // namespace mindspore
#endif // MINDSPORE_CCSRC_PRE_ACTIVATE_MEM_REUSE_STREAM_REUSE_H_
tests/ut/cpp/pre_activate/mem_reuse/mem_reuse_allocator_test.cc
浏览文件 @ 5e2f440e
......@@ -117,16 +117,13 @@ TEST_F(TestMemReuseAllocator, mem_reuse_allocator) {
MS_LOG(INFO) << "run mem reuse success";
size_t total_allocated_size = best_fit_mem_reuse->GetAllocatedSize();
ASSERT_NE(total_allocated_size, 0);
auto is_reusable_stream = best_fit_mem_reuse->IsReusableStream(1, 3);
ASSERT_EQ(is_reusable_stream, true);
}
TEST_F(TestMemReuseAllocator, mem_reuse_allocator_add_membuf) {
auto best_fit_mem_reuse = std::make_shared<BestFitMemReuse>();
auto tensor_desc = std::make_shared<KernelRefCount>();
tensor_desc->SetKernelRefCountInfo(0, 1024, kDynamicRefCount);
best_fit_mem_reuse->AddNewMembufPtr(tensor_desc.get(), kDyFac);
best_fit_mem_reuse->AddNewMembufPtr(tensor_desc.get(), kDynamicMem);
auto allocated_size = best_fit_mem_reuse->GetAllocatedSize();
ASSERT_EQ(allocated_size, 1024);
}
......@@ -135,7 +132,7 @@ TEST_F(TestMemReuseAllocator, mem_reuse_allocator_split_membuf) {
auto best_fit_mem_reuse = std::make_shared<BestFitMemReuse>();
auto tensor_0 = std::make_shared<KernelRefCount>();
tensor_0->SetKernelRefCountInfo(0, 2048, kDynamicRefCount);
best_fit_mem_reuse->AddNewMembufPtr(tensor_0.get(), kDyFac);
best_fit_mem_reuse->AddNewMembufPtr(tensor_0.get(), kDynamicMem);
auto tensor_1 = std::make_shared<KernelRefCount>();
tensor_1->SetKernelRefCountInfo(1, 800, kDynamicRefCount);
......
tests/ut/cpp/pre_activate/mem_reuse/mem_reuse_test.cc
浏览文件 @ 5e2f440e
......@@ -228,12 +228,6 @@ TEST_F(TestMemReuseWithPy, KernelRef) {
ASSERT_EQ(kernel_def_ptr->dirty, false);
MembufPtr membuf_ptr = std::make_shared<Membuf>();
ASSERT_NE(membuf_ptr, nullptr);
MembufPtr membuf_ptr_x = std::make_shared<Membuf>(0, memreuse::kUnused, 512, 128, 2);
ASSERT_EQ(membuf_ptr_x->status_, memreuse::kUnused);
ASSERT_EQ(membuf_ptr_x->size_, 512);
ASSERT_EQ(membuf_ptr_x->offset_, 128);
ASSERT_EQ(membuf_ptr_x->index_, 2);
ASSERT_EQ(membuf_ptr_x->stream_id_, 0);
}
TEST_F(TestMemReuseWithPy, ReuseAssignDynamicMemory) {
......
tests/ut/cpp/pre_activate/mem_reuse/stream_reuse_test.cc 已删除 100644 → 0
浏览文件 @ e7936ded
/**
* 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 <memory>
#include <vector>
#include <string>
#include "operator/ops.h"
#include "pre_activate/mem_reuse/stream_reuse.h"
#include "common/common_test.h"
#include "common/py_func_graph_fetcher.h"
using mindspore::memreuse::StreamReuse;
namespace mindspore {
class TestStreamMemReuse : public UT::Common {
public:
TestStreamMemReuse() : getPyFun_("gtest_input.mem_reuse.TestMemReuseAllocator", true) {}
void SetUp() {}
public:
UT::PyFuncGraphFetcher getPyFun_;
};
TEST_F(TestStreamMemReuse, init_reusable_stream_map_test) {
std::unordered_map<uint32_t, uint32_t> logic_physic_map{{1, 0}, {2, 8}, {3, 3}};
std::unordered_map<uint32_t, uint32_t> logic_independent_map{{3, 10}, {2, 11}};
auto stream_reuse = std::make_shared<StreamReuse>();
stream_reuse->set_logic_physic_map(logic_physic_map);
stream_reuse->set_logic_independent_map(logic_independent_map);
auto logic_phyics_map = stream_reuse->GetLogicPhysicsStreamMap();
for (const auto &logic_physics : logic_phyics_map) {
MS_LOG(INFO) << "[logic_id: " << logic_physics.first << "]";
for (const auto &physic : logic_physics.second) {
MS_LOG(INFO) << "physic: " << physic;
}
}
MS_LOG(INFO) << "===========UT logic_physic_map size: " << logic_physic_map.size() << "========";
ASSERT_EQ(logic_physic_map.size(), 3);
stream_reuse->InitReusableStreamMap();
auto parallel_streams_map = stream_reuse->parallel_streams_map();
for (const auto &parallel_streams : parallel_streams_map) {
MS_LOG(INFO) << "[stream id: " << parallel_streams.first << "]";
for (const auto &stream : parallel_streams.second) {
MS_LOG(INFO) << "parallel stream id: " << stream;
}
}
ASSERT_EQ(parallel_streams_map[7].size(), 1);
}
} // namespace mindspore
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或