Skip to content

P
Paddle

BaiXuePrincess / Paddle
与 Fork 源项目一致

Fork自 PaddlePaddle / Paddle

通知 1
Star 1
Fork 0
P
Paddle
未验证 提交 bd7f28bb 编写于 作者: R Ruibiao Chen 提交者： GitHub
浏览文件
操作

Multiplex Workqueue for InterpreterCore (#43660) 

* Multiplex Workqueue for InterpreterCore

* Delete ResetWorkQueueOptions

* Update code format
上级 66a28e13
隐藏空白更改
内联 并排
Showing with 508 addition and 370 deletion
paddle/fluid/framework/new_executor/CMakeLists.txt
浏览文件 @ bd7f28bb
......@@ -110,7 +110,8 @@ if(WITH_GPU
sgd_op
squared_l2_norm_op
memcpy_h2d_op
memcpy_d2h_op)
memcpy_d2h_op
fetch_v2_op)
# All deps of the operators above, part of GLOB_OPERATOR_DEPS.
set(OP_DEPS generator softmax selected_rows_functor jit_kernel_helper
......
paddle/fluid/framework/new_executor/interpretercore.cc
浏览文件 @ bd7f28bb
......@@ -29,9 +29,11 @@
#include "paddle/fluid/platform/mkldnn_helper.h"
#endif
PADDLE_DEFINE_EXPORTED_bool(new_executor_use_inplace, true,
PADDLE_DEFINE_EXPORTED_bool(new_executor_use_inplace,
true,
"Use inplace in new executor");
PADDLE_DEFINE_EXPORTED_bool(new_executor_use_local_scope, true,
PADDLE_DEFINE_EXPORTED_bool(new_executor_use_local_scope,
true,
"Use local_scope in new executor(especially used "
"in UT), can turn off for better performance");
......@@ -99,9 +101,8 @@ InterpreterCore::~InterpreterCore() {
// cancle gc's thread
gc_.reset(nullptr);
async_work_queue_.reset(nullptr);
VLOG(4) << "~InterpreterCore(): " << this;
VLOG(4) << " on" << place_;
async_work_queue_.reset();
VLOG(4) << "~InterpreterCore(): " << this << " on " << place_;
#ifdef PADDLE_WITH_MKLDNN
// Clear mkl-dnn cache,
......@@ -110,8 +111,29 @@ InterpreterCore::~InterpreterCore() {
#endif
}
void InterpreterCore::SetCopyProgram(std::shared_ptr<ProgramDesc> prog) {
copy_program_ = prog;
interpreter::CostInfo InterpreterCore::DryRun(
const std::vector<std::string>& feed_names,
const std::vector<framework::LoDTensor>& feed_tensors) {
global_scope_->SetLocalScope(local_scope_);
Prepare(feed_names, feed_tensors, true);
interpreter::CostInfo cost_info;
{
interpreter::ProfilerGuard(place_, &cost_info);
// For the program that only run once, it is no need to
// create work_queue, so the async_work_queue_ is created
// until the second step run.
async_work_queue_ = GetWorkQueue();
ExecuteInstructionList(vec_instruction_);
platform::DeviceContextPool::Instance().Get(place_)->Wait();
}
if (create_local_scope_) {
ClearLoDTensorArrayInLocalScope();
}
return cost_info;
}
paddle::framework::FetchList InterpreterCore::Run(
......@@ -131,14 +153,7 @@ paddle::framework::FetchList InterpreterCore::Run(
// For the program that only run once, it is no need to
// create work_queue, so the async_work_queue_ is created
// until the second step run.
if (async_work_queue_ == nullptr) {
async_work_queue_ = std::make_unique<interpreter::AsyncWorkQueue>(
kHostNumThreads, kDeviceNumThreads, &main_thread_blocker_);
// prepare for the first time.
async_work_queue_->PrepareAtomicDeps(dependecy_count_);
async_work_queue_->PrepareAtomicVarRef(global_scope_->VecMetaInfo());
}
async_work_queue_ = GetWorkQueue();
ExecuteInstructionList(vec_instruction_);
}
......@@ -172,12 +187,15 @@ paddle::framework::FetchList InterpreterCore::Run(
// scope?
}
global_scope_->SetLocalScope(local_scope_);
paddle::framework::interpreter::build_variable_scope(block_, global_scope_,
create_local_scope_);
paddle::framework::interpreter::build_variable_scope(
block_, global_scope_, create_local_scope_);
std::vector<paddle::framework::OpFuncNode> op_func_nodes;
paddle::framework::interpreter::build_op_func_list(
place_, block_, skip_gc_vars_, &op_func_nodes, global_scope_,
create_local_scope_);
paddle::framework::interpreter::build_op_func_list(place_,
block_,
skip_gc_vars_,
&op_func_nodes,
global_scope_,
create_local_scope_);
is_build_ = true;
SetFeedVarsInplaceSkip(feed_names);
// convert vec func_list to graph
......@@ -190,13 +208,7 @@ paddle::framework::FetchList InterpreterCore::Run(
// For the program that only run once, it is no need to
// create work_queue, so the async_work_queue_ is created
// until the second step run.
if (async_work_queue_ == nullptr) {
async_work_queue_ = std::make_unique<interpreter::AsyncWorkQueue>(
kHostNumThreads, kDeviceNumThreads, &main_thread_blocker_);
// prepare for the first time.
async_work_queue_->PrepareAtomicDeps(dependecy_count_);
async_work_queue_->PrepareAtomicVarRef(global_scope_->VecMetaInfo());
}
async_work_queue_ = GetWorkQueue();
ExecuteInstructionList(vec_instruction_);
}
......@@ -213,15 +225,115 @@ paddle::framework::FetchList InterpreterCore::Run(
return std::move(*fetch_var->GetMutable<framework::FetchList>());
}
// At the end of each step, the holder of Tensor in LoDTensorArray is null.
// Clear these Tensors and leave LoDTensorArray empty, otherwise an exception
// will occur in the next step
void InterpreterCore::ClearLoDTensorArrayInLocalScope() {
auto vars = local_scope_->LocalVars();
for (auto var : vars) {
if (var->IsType<LoDTensorArray>()) {
auto* lod_tensor_arr = var->GetMutable<LoDTensorArray>();
lod_tensor_arr->clear();
void InterpreterCore::SetCopyProgram(std::shared_ptr<ProgramDesc> prog) {
copy_program_ = prog;
}
void InterpreterCore::ShareWorkQueueFrom(std::shared_ptr<InterpreterCore> src) {
async_work_queue_ = src->GetWorkQueue();
VLOG(8) << "Share AsyncWorkQueue from InterpreterCore(" << &src
<< ") to InterpreterCore(" << this << ")";
}
bool InterpreterCore::BuildInplaceCheckVarIsOnlyInput(size_t var_index) {
if (!global_scope_->VarDesc(var_index)) {
return input_var2op_info_.at(var_index).size() == 1;
} else {
int is_input_cnt = 0;
for (auto inst_id : input_var2op_info_.at(var_index)) {
OpInOutInfo info;
info.Build(vec_instruction_.at(inst_id).OpBase());
if (info.IsInArgBufferNeeded(global_scope_->VarDesc(var_index)->Name())) {
is_input_cnt++;
}
}
return is_input_cnt == 1;
}
}
std::shared_ptr<interpreter::AsyncWorkQueue> InterpreterCore::GetWorkQueue() {
if (async_work_queue_ == nullptr) {
async_work_queue_ = std::make_shared<interpreter::AsyncWorkQueue>(
kHostNumThreads, kDeviceNumThreads, &main_thread_blocker_);
}
return async_work_queue_;
}
void InterpreterCore::BuildAndCacheInstructionCtx(Instruction* instr_node) {
VariableValueMap ins_map;
for (auto& var_name_item : instr_node->Inputs()) {
std::vector<Variable*> input_vars;
input_vars.reserve(var_name_item.second.size());
for (auto& id : var_name_item.second) {
input_vars.emplace_back(global_scope_->Var(id));
}
ins_map.emplace(var_name_item.first, std::move(input_vars));
}
VariableValueMap outs_map;
for (auto& var_name_item : instr_node->Outputs()) {
std::vector<Variable*> out_vars;
out_vars.reserve(var_name_item.second.size());
for (auto& id : var_name_item.second) {
out_vars.emplace_back(global_scope_->Var(id));
}
outs_map.emplace(var_name_item.first, std::move(out_vars));
}
// set runtime_ctx and infershape_ctx_
if (instr_node->OpBase()->Type() == "cinn_launch") { // OP use scope in
// kernel
Scope* local_scope = create_local_scope_
? global_scope_->GetMutableLocalScope()
: global_scope_->GetMutableScope();
instr_node->ResetContextWithScope(ins_map, outs_map, *local_scope);
} else {
instr_node->ResetContext(ins_map, outs_map);
}
}
void InterpreterCore::BuildInplace() {
for (size_t i = 0; i < vec_instruction_.size(); ++i) {
auto& instr = vec_instruction_[i];
auto* op_base = instr.OpBase();
if (!op_base->Info().infer_inplace_) {
continue;
}
auto in_to_outs = op_base->Info().infer_inplace_(
platform::is_gpu_place(instr.DeviceContext().GetPlace()));
auto& inputs = instr.Inputs();
auto& outputs = instr.Outputs();
for (auto& pair : in_to_outs) {
auto iter = inputs.find(pair.first);
if (iter != inputs.end() && !iter->second.empty()) {
auto in_var_desc = global_scope_->VarDesc(iter->second[0]);
if (in_var_desc && in_var_desc->Persistable()) {
continue;
}
if (global_scope_->GetVarSikpInplace(iter->second[0])) {
continue;
}
if (BuildInplaceCheckVarIsOnlyInput(iter->second[0])) {
auto iterout = outputs.find(pair.second);
if (iterout != outputs.end() && !iterout->second.empty()) {
auto invar = global_scope_->Var(iter->second[0]);
auto outvar = global_scope_->Var(iterout->second[0]);
if (invar && outvar && invar->IsType<LoDTensor>() &&
outvar->IsType<LoDTensor>()) {
instr.AddInplace(invar, outvar);
VLOG(3) << "inplace " << vec_instruction_[i].OpBase()->Type()
<< " " << global_scope_->GetNameById(iter->second[0])
<< " -> "
<< global_scope_->GetNameById(iterout->second[0])
<< std::endl;
}
}
}
}
}
}
}
......@@ -244,6 +356,19 @@ void InterpreterCore::BuildOperatorDependences() {
}
}
// At the end of each step, the holder of Tensor in LoDTensorArray is null.
// Clear these Tensors and leave LoDTensorArray empty, otherwise an exception
// will occur in the next step
void InterpreterCore::ClearLoDTensorArrayInLocalScope() {
auto vars = local_scope_->LocalVars();
for (auto var : vars) {
if (var->IsType<LoDTensorArray>()) {
auto* lod_tensor_arr = var->GetMutable<LoDTensorArray>();
lod_tensor_arr->clear();
}
}
}
void InterpreterCore::Convert(
std::vector<paddle::framework::OpFuncNode>* op_func_nodes) {
auto& vec_meta_info = global_scope_->MutableVecMetaInfo();
......@@ -376,101 +501,18 @@ void InterpreterCore::Convert(
if (FLAGS_new_executor_use_inplace && !inplaced) {
BuildInplace();
}
}
bool InterpreterCore::BuildInplaceCheckVarIsOnlyInput(size_t var_index) {
if (!global_scope_->VarDesc(var_index)) {
return input_var2op_info_.at(var_index).size() == 1;
} else {
int is_input_cnt = 0;
for (auto inst_id : input_var2op_info_.at(var_index)) {
OpInOutInfo info;
info.Build(vec_instruction_.at(inst_id).OpBase());
if (info.IsInArgBufferNeeded(global_scope_->VarDesc(var_index)->Name())) {
is_input_cnt++;
}
}
return is_input_cnt == 1;
}
}
void InterpreterCore::BuildInplace() {
for (size_t i = 0; i < vec_instruction_.size(); ++i) {
auto& instr = vec_instruction_[i];
auto* op_base = instr.OpBase();
if (!op_base->Info().infer_inplace_) {
continue;
}
auto in_to_outs = op_base->Info().infer_inplace_(
platform::is_gpu_place(instr.DeviceContext().GetPlace()));
auto& inputs = instr.Inputs();
auto& outputs = instr.Outputs();
for (auto& pair : in_to_outs) {
auto iter = inputs.find(pair.first);
if (iter != inputs.end() && !iter->second.empty()) {
auto in_var_desc = global_scope_->VarDesc(iter->second[0]);
if (in_var_desc && in_var_desc->Persistable()) {
continue;
}
if (global_scope_->GetVarSikpInplace(iter->second[0])) {
continue;
}
if (BuildInplaceCheckVarIsOnlyInput(iter->second[0])) {
auto iterout = outputs.find(pair.second);
if (iterout != outputs.end() && !iterout->second.empty()) {
auto invar = global_scope_->Var(iter->second[0]);
auto outvar = global_scope_->Var(iterout->second[0]);
if (invar && outvar && invar->IsType<LoDTensor>() &&
outvar->IsType<LoDTensor>()) {
instr.AddInplace(invar, outvar);
VLOG(3) << "inplace " << vec_instruction_[i].OpBase()->Type()
<< " " << global_scope_->GetNameById(iter->second[0])
<< " -> "
<< global_scope_->GetNameById(iterout->second[0])
<< std::endl;
}
}
}
}
}
}
}
void InterpreterCore::BuildAndCacheInstructionCtx(Instruction* instr_node) {
VariableValueMap ins_map;
for (auto& var_name_item : instr_node->Inputs()) {
std::vector<Variable*> input_vars;
input_vars.reserve(var_name_item.second.size());
for (auto& id : var_name_item.second) {
input_vars.emplace_back(global_scope_->Var(id));
}
ins_map.emplace(var_name_item.first, std::move(input_vars));
}
VariableValueMap outs_map;
for (auto& var_name_item : instr_node->Outputs()) {
std::vector<Variable*> out_vars;
out_vars.reserve(var_name_item.second.size());
for (auto& id : var_name_item.second) {
out_vars.emplace_back(global_scope_->Var(id));
}
outs_map.emplace(var_name_item.first, std::move(out_vars));
}
// set runtime_ctx and infershape_ctx_
if (instr_node->OpBase()->Type() == "cinn_launch") { // OP use scope in
// kernel
Scope* local_scope = create_local_scope_
? global_scope_->GetMutableLocalScope()
: global_scope_->GetMutableScope();
instr_node->ResetContextWithScope(ins_map, outs_map, *local_scope);
} else {
instr_node->ResetContext(ins_map, outs_map);
}
// prepare for the first time.
std::promise<std::unique_ptr<AtomicVectorSizeT>> deps_promise =
std::promise<std::unique_ptr<AtomicVectorSizeT>>();
atomic_deps_ = deps_promise.get_future();
deps_promise.set_value(interpreter::PrepareAtomicDeps(dependecy_count_));
std::promise<std::unique_ptr<AtomicVectorSizeT>> var_ref_promise =
std::promise<std::unique_ptr<AtomicVectorSizeT>>();
atomic_var_ref_ = var_ref_promise.get_future();
var_ref_promise.set_value(
interpreter::PrepareAtomicVarRef(global_scope_->VecMetaInfo()));
}
void InterpreterCore::BuildSkipShareLoDInfo() {
......@@ -505,7 +547,9 @@ void InterpreterCore::RunInstruction(const Instruction& instr_node) {
// If it is OperatorBase, InferShape do nothing.
if (op_with_kernel != nullptr) {
platform::RecordEvent infershape_event(
"infer_shape", platform::TracerEventType::OperatorInner, 1,
"infer_shape",
platform::TracerEventType::OperatorInner,
1,
platform::EventRole::kInnerOp);
// see OperatorWithKernel::RunImpl in operator.cc for why
......@@ -531,7 +575,9 @@ void InterpreterCore::RunInstruction(const Instruction& instr_node) {
{
platform::RecordEvent compute_event(
"compute", platform::TracerEventType::OperatorInner, 1,
"compute",
platform::TracerEventType::OperatorInner,
1,
platform::EventRole::kInnerOp);
if (op_with_kernel == nullptr) {
instr_node.OpBase()->Run(*local_scope, place_);
......@@ -588,7 +634,8 @@ void InterpreterCore::RunInstruction(const Instruction& instr_node) {
if (op_with_kernel != nullptr && FLAGS_check_nan_inf) {
VLOG(4) << "Check nan/inf";
framework::details::CheckOpHasNanOrInf(
*op, *global_scope_,
*op,
*global_scope_,
place); // TODO(xiongkun03) change it to inner scope.
}
}
......@@ -605,11 +652,12 @@ void InterpreterCore::ExecuteInstructionList(
"PrepareAtomic", platform::TracerEventType::UserDefined, 1);
// NOTE(zhiqiu): get the prepared deps from std::future, and async prepare
// those for the next step
auto atomic_deps = async_work_queue_->AtomicDeps();
auto atomic_var_ref = async_work_queue_->AtomicVarRef();
auto atomic_deps = atomic_deps_.get();
auto atomic_var_ref = atomic_var_ref_.get();
async_work_queue_->PrepareAtomicDeps(dependecy_count_);
async_work_queue_->PrepareAtomicVarRef(global_scope_->VecMetaInfo());
atomic_deps_ = async_work_queue_->PrepareAtomicDeps(dependecy_count_);
atomic_var_ref_ =
async_work_queue_->PrepareAtomicVarRef(global_scope_->VecMetaInfo());
record_prepare.End();
exception_holder_.Clear();
......@@ -617,16 +665,19 @@ void InterpreterCore::ExecuteInstructionList(
for (size_t i = 0; i < dependecy_count_.size(); ++i) {
if (dependecy_count_[i] == 0) {
async_work_queue_->AddTask(vec_instr.at(i).KernelType(),
[this, i, atomic_deps = atomic_deps.get(),
[this,
i,
atomic_deps = atomic_deps.get(),
atomic_var_ref = atomic_var_ref.get()] {
RunInstructionAsync(i, atomic_deps,
atomic_var_ref);
RunInstructionAsync(
i, atomic_deps, atomic_var_ref);
});
}
}
auto event_name = main_thread_blocker_.WaitEvent();
VLOG(1) << "event_name: " << event_name;
VLOG(1) << "main_thread_blocker_(" << &main_thread_blocker_
<< ") got event_name: " << event_name;
if (UNLIKELY(exception_holder_.IsCaught())) {
VLOG(1) << "Exception caught " << exception_holder_.Type();
......@@ -637,7 +688,8 @@ void InterpreterCore::ExecuteInstructionList(
}
VLOG(4) << "Cancel ok";
PADDLE_ENFORCE_EQ(
main_thread_blocker_.Clear(), 0,
main_thread_blocker_.Clear(),
0,
platform::errors::PreconditionNotMet(
"main_thread_blocker_.Clear() return -1, clear failed"));
VLOG(4) << "clear ok";
......@@ -646,11 +698,12 @@ void InterpreterCore::ExecuteInstructionList(
}
void InterpreterCore::RunNextInstructions(
const Instruction& instr, std::queue<size_t>* reserved_next_ops,
const Instruction& instr,
std::queue<size_t>* reserved_next_ops,
std::vector<std::atomic<size_t>>* atomic_deps,
std::vector<std::atomic<size_t>>* atomic_var_ref) {
platform::RecordEvent record("RunNextInstructions",
platform::TracerEventType::UserDefined, 10);
platform::RecordEvent record(
"RunNextInstructions", platform::TracerEventType::UserDefined, 10);
VLOG(4) << "atomic 1:" << atomic_deps;
auto& next_instr = instr.NextInstructions();
......@@ -716,7 +769,8 @@ void InterpreterCore::RunNextInstructions(
}
void InterpreterCore::RunInstructionAsync(
size_t instr_id, std::vector<std::atomic<size_t>>* atomic_deps,
size_t instr_id,
std::vector<std::atomic<size_t>>* atomic_deps,
std::vector<std::atomic<size_t>>* atomic_var_ref) {
std::queue<size_t> ready_ops;
ready_ops.push(instr_id);
......@@ -787,8 +841,8 @@ void InterpreterCore::RecordStreamForGC(const Instruction& instr) {
instr.KernelType() != OpFuncType::kQueueAsync) {
return;
}
platform::RecordEvent record("RecordStreamForGC",
platform::TracerEventType::UserDefined, 10);
platform::RecordEvent record(
"RecordStreamForGC", platform::TracerEventType::UserDefined, 10);
gpuStream_t stream = reinterpret_cast<const platform::CUDADeviceContext&>(
instr.DeviceContext())
......@@ -880,8 +934,8 @@ void InterpreterCore::RecordStreamForGC(const Instruction& instr) {
void InterpreterCore::CheckGC(
const Instruction& instr,
std::vector<std::atomic<size_t>>* atomic_var_ref) {
platform::RecordEvent record("CheckGC",
platform::TracerEventType::UserDefined, 10);
platform::RecordEvent record(
"CheckGC", platform::TracerEventType::UserDefined, 10);
size_t instr_id = instr.Id();
auto& var_scope = *global_scope_;
......@@ -906,12 +960,14 @@ void InterpreterCore::CheckGC(
} else {
static_cast<InterpreterCoreEventGarbageCollector*>(gc_.get())->Add(
var_scope.Var(var_id), &gc_event_.at(instr_id),
var_scope.Var(var_id),
&gc_event_.at(instr_id),
&instr.DeviceContext());
}
#else
static_cast<InterpreterCoreEventGarbageCollector*>(gc_.get())->Add(
var_scope.Var(var_id), &gc_event_.at(instr_id),
var_scope.Var(var_id),
&gc_event_.at(instr_id),
&instr.DeviceContext());
#endif
}
......@@ -920,20 +976,24 @@ void InterpreterCore::CheckGC(
void InterpreterCore::Prepare(
const std::vector<std::string>& feed_names,
const std::vector<framework::LoDTensor>& feed_tensors, bool prepare_feed) {
PADDLE_ENFORCE_EQ(feed_names.size(), feed_tensors.size(),
const std::vector<framework::LoDTensor>& feed_tensors,
bool prepare_feed) {
PADDLE_ENFORCE_EQ(feed_names.size(),
feed_tensors.size(),
platform::errors::PreconditionNotMet(
"Required feed_names.size() == feed_tensors.size(), "
"but received %d != %d",
feed_names.size(), feed_tensors.size()));
feed_names.size(),
feed_tensors.size()));
auto FeedInput = [&] {
VLOG(4) << "Feed inputs";
for (size_t i = 0; i < feed_names.size(); ++i) {
auto* feed_var = global_scope_->FindVar(feed_names[i]);
PADDLE_ENFORCE_NOT_NULL(
feed_var, platform::errors::NotFound(
"Variable %s should not be nullptr.", feed_names[i]));
feed_var,
platform::errors::NotFound("Variable %s should not be nullptr.",
feed_names[i]));
auto feed_tensor = feed_var->GetMutable<framework::LoDTensor>();
feed_tensor->ShareDataWith(feed_tensors[i]);
......@@ -942,13 +1002,16 @@ void InterpreterCore::Prepare(
};
if (!is_build_) {
paddle::framework::interpreter::build_variable_scope(block_, global_scope_,
create_local_scope_);
paddle::framework::interpreter::build_variable_scope(
block_, global_scope_, create_local_scope_);
FeedInput();
std::vector<paddle::framework::OpFuncNode> op_func_nodes;
paddle::framework::interpreter::build_op_func_list(
place_, block_, skip_gc_vars_, &op_func_nodes, global_scope_,
create_local_scope_);
paddle::framework::interpreter::build_op_func_list(place_,
block_,
skip_gc_vars_,
&op_func_nodes,
global_scope_,
create_local_scope_);
is_build_ = true;
SetFeedVarsInplaceSkip(feed_names);
// convert vec func_list to graph
......@@ -962,37 +1025,6 @@ void InterpreterCore::Prepare(
}
}
interpreter::CostInfo InterpreterCore::DryRun(
const std::vector<std::string>& feed_names,
const std::vector<framework::LoDTensor>& feed_tensors) {
global_scope_->SetLocalScope(local_scope_);
Prepare(feed_names, feed_tensors, true);
interpreter::CostInfo cost_info;
{
interpreter::ProfilerGuard(place_, &cost_info);
// For the program that only run once, it is no need to
// create work_queue, so the async_work_queue_ is created
// until the second step run.
if (async_work_queue_ == nullptr) {
async_work_queue_ = std::make_unique<interpreter::AsyncWorkQueue>(
kHostNumThreads, kDeviceNumThreads, &main_thread_blocker_);
// prepare for the first time.
async_work_queue_->PrepareAtomicDeps(dependecy_count_);
async_work_queue_->PrepareAtomicVarRef(global_scope_->VecMetaInfo());
}
ExecuteInstructionList(vec_instruction_);
platform::DeviceContextPool::Instance().Get(place_)->Wait();
}
if (create_local_scope_) {
ClearLoDTensorArrayInLocalScope();
}
return cost_info;
}
void InterpreterCore::SetFeedVarsInplaceSkip(
const std::vector<std::string>& feed_names) {
for (auto& feed_name : feed_names) {
......@@ -1001,8 +1033,10 @@ void InterpreterCore::SetFeedVarsInplaceSkip(
}
std::shared_ptr<InterpreterCore> CreateInterpreterCore(
const platform::Place& place, const ProgramDesc& prog,
VariableScope* global_scope, const std::vector<std::string>& fetch_names,
const platform::Place& place,
const ProgramDesc& prog,
VariableScope* global_scope,
const std::vector<std::string>& fetch_names,
const std::set<std::string>& skip_gc_vars) {
std::shared_ptr<InterpreterCore> core = nullptr;
// NOTE(Aurelius84): `add_fetch` will modify BlockDesc, so we should copy
......@@ -1011,8 +1045,8 @@ std::shared_ptr<InterpreterCore> CreateInterpreterCore(
auto* block = new_prog->MutableBlock(0);
interpreter::add_fetch(fetch_names, block);
core = std::make_shared<InterpreterCore>(place, *block, skip_gc_vars,
global_scope);
core = std::make_shared<InterpreterCore>(
place, *block, skip_gc_vars, global_scope);
core->SetCopyProgram(new_prog);
return core;
}
......
paddle/fluid/framework/new_executor/interpretercore.h
浏览文件 @ bd7f28bb
......@@ -34,36 +34,44 @@
namespace paddle {
namespace framework {
using AtomicVectorSizeT = std::vector<std::unique_ptr<std::atomic<size_t>>>;
class InterpreterCore {
public:
InterpreterCore(const platform::Place& place, const BlockDesc& block,
InterpreterCore(const platform::Place& place,
const BlockDesc& block,
const std::set<std::string>& skip_gc_vars,
VariableScope* global_scope);
~InterpreterCore();
interpreter::CostInfo DryRun(
const std::vector<std::string>& feed_names,
const std::vector<framework::LoDTensor>& feed_tensors);
paddle::framework::FetchList Run(
const std::vector<std::string>& feed_names,
const std::vector<framework::LoDTensor>& feed_tensors);
paddle::framework::FetchList Run(const std::vector<std::string>& feed_names);
interpreter::CostInfo DryRun(
const std::vector<std::string>& feed_names,
const std::vector<framework::LoDTensor>& feed_tensors);
void ShareWorkQueueFrom(std::shared_ptr<InterpreterCore> src);
void SetCopyProgram(std::shared_ptr<ProgramDesc> prog);
private:
void Convert(std::vector<paddle::framework::OpFuncNode>* op_func_nodes);
bool BuildInplaceCheckVarIsOnlyInput(size_t var_index);
std::shared_ptr<interpreter::AsyncWorkQueue> GetWorkQueue();
void BuildAndCacheInstructionCtx(Instruction* instr_node);
void BuildInplace();
bool BuildInplaceCheckVarIsOnlyInput(size_t var_index);
void BuildOperatorDependences();
void ClearLoDTensorArrayInLocalScope();
void Convert(std::vector<paddle::framework::OpFuncNode>* op_func_nodes);
void RunInstruction(const Instruction& instr_node);
......@@ -90,12 +98,8 @@ class InterpreterCore {
void BuildSkipShareLoDInfo();
void BuildOperatorDependences();
void SetFeedVarsInplaceSkip(const std::vector<std::string>& feed_names);
void ClearLoDTensorArrayInLocalScope();
bool is_build_;
const platform::Place& place_;
......@@ -123,7 +127,7 @@ class InterpreterCore {
StreamAnalyzer stream_analyzer_;
EventsWaiter main_thread_blocker_;
std::unique_ptr<interpreter::AsyncWorkQueue> async_work_queue_;
std::shared_ptr<interpreter::AsyncWorkQueue> async_work_queue_;
details::ExceptionHolder exception_holder_;
std::shared_ptr<EventsWaiter::EventNotifier> exception_notifier_{nullptr};
std::shared_ptr<EventsWaiter::EventNotifier> completion_notifier_{nullptr};
......@@ -132,10 +136,14 @@ class InterpreterCore {
std::vector<paddle::platform::DeviceEvent> gc_event_;
bool create_local_scope_{true};
Scope* local_scope_{nullptr}; // not owned
std::future<std::unique_ptr<AtomicVectorSizeT>> atomic_deps_;
std::future<std::unique_ptr<AtomicVectorSizeT>> atomic_var_ref_;
};
std::shared_ptr<InterpreterCore> CreateInterpreterCore(
const platform::Place& place, const ProgramDesc& prog,
const platform::Place& place,
const ProgramDesc& prog,
VariableScope* global_scope,
const std::vector<std::string>& fetch_names = {},
const std::set<std::string>& skip_gc_vars = {});
......
paddle/fluid/framework/new_executor/interpretercore_util.cc
浏览文件 @ bd7f28bb
......@@ -32,12 +32,14 @@
// Program, while "serial_run" ensures that all Ops are scheduled in a singal
// thread. In standalone executor, "sequential_run" is also "serial_run", while
// "serial_run" is not necessarily "sequential_run".
PADDLE_DEFINE_EXPORTED_bool(new_executor_sequential_run, false,
PADDLE_DEFINE_EXPORTED_bool(new_executor_sequential_run,
false,
"Enable sequential execution for standalone "
"executor, only applied to GPU OPs.");
PADDLE_DEFINE_EXPORTED_bool(
new_executor_serial_run, false,
new_executor_serial_run,
false,
"Enable serial execution for standalone executor, used for debug.");
DECLARE_bool(use_mkldnn);
......@@ -46,8 +48,47 @@ namespace paddle {
namespace framework {
namespace interpreter {
using VariableIdMap = std::map<std::string, std::vector<int>>;
constexpr size_t kPrepareWorkQueueIdx = 2;
const std::vector<WorkQueueOptions> ConstructWorkQueueOptions(
size_t host_num_threads, size_t device_num_threads, EventsWaiter* waiter) {
std::vector<WorkQueueOptions> group_options;
// for execute host Kernel
group_options.emplace_back(/*name*/ "HostTasks",
/*num_threads*/ host_num_threads,
/*allow_spinning*/ true,
/*always_spinning*/ false,
/*track_task*/ false,
/*detached*/ true,
/*events_waiter*/ waiter);
// for launch device Kernel
group_options.emplace_back(/*name*/ "DeviceKernelLaunch",
/*num_threads*/ device_num_threads,
/*allow_spinning*/ true,
/*always_spinning*/ true,
/*track_task*/ false,
/*detached*/ true,
/*events_waiter*/ waiter);
// for prepare deps and others
group_options.emplace_back(/*name*/ "Prepare",
/*num_threads*/ 1,
/*allow_spinning*/ true,
/*always_spinning*/ false,
/*track_task*/ false,
/*detached*/ true,
/*events_waiter*/ waiter);
return group_options;
}
AsyncWorkQueue::AsyncWorkQueue(size_t host_num_threads,
size_t device_num_threads,
EventsWaiter* waiter)
: host_num_thread_(host_num_threads) {
queue_group_ = CreateWorkQueueGroup(
ConstructWorkQueueOptions(host_num_threads, device_num_threads, waiter));
}
void AsyncWorkQueue::AddTask(const OpFuncType& op_func_type,
std::function<void()> fn) {
VLOG(4) << "Add task: " << static_cast<size_t>(op_func_type) << " ";
......@@ -60,36 +101,42 @@ void AsyncWorkQueue::AddTask(const OpFuncType& op_func_type,
}
}
using VariableIdMap = std::map<std::string, std::vector<int>>;
std::future<std::unique_ptr<AtomicVectorSizeT>>
AsyncWorkQueue::PrepareAtomicDeps(const std::vector<size_t>& dependecy_count) {
VLOG(4) << "PrepareAtomicDeps";
return queue_group_->AddAwaitableTask(
kPrepareWorkQueueIdx, interpreter::PrepareAtomicDeps, dependecy_count);
}
std::future<std::unique_ptr<AtomicVectorSizeT>>
AsyncWorkQueue::PrepareAtomicVarRef(
const std::vector<VariableMetaInfo>& vec_meta_info) {
VLOG(4) << "PrepareAtomicVarRef";
return queue_group_->AddAwaitableTask(
kPrepareWorkQueueIdx, interpreter::PrepareAtomicVarRef, vec_meta_info);
}
void AsyncWorkQueue::PrepareAtomicDeps(
std::unique_ptr<AtomicVectorSizeT> PrepareAtomicDeps(
const std::vector<size_t>& dependecy_count) {
VLOG(4) << "PrepareAtomicDeps";
atomic_deps_ =
queue_group_->AddAwaitableTask(kPrepareWorkQueueIdx, [&dependecy_count] {
auto op_deps = std::make_unique<std::vector<std::atomic<size_t>>>(
dependecy_count.size());
for (size_t i = 0; i < dependecy_count.size(); ++i) {
(*op_deps)[i] = dependecy_count[i];
}
VLOG(4) << "AtomicDeps:" << op_deps.get() << " " << op_deps->size();
return op_deps;
});
auto op_deps = std::make_unique<AtomicVectorSizeT>(dependecy_count.size());
for (size_t i = 0; i < dependecy_count.size(); ++i) {
(*op_deps)[i] = dependecy_count[i];
}
VLOG(4) << "AtomicDeps:" << op_deps.get() << " " << op_deps->size();
return op_deps;
}
void AsyncWorkQueue::PrepareAtomicVarRef(
std::unique_ptr<AtomicVectorSizeT> PrepareAtomicVarRef(
const std::vector<VariableMetaInfo>& vec_meta_info) {
VLOG(4) << "PrepareAtomicVarRef";
atomic_var_ref_ =
queue_group_->AddAwaitableTask(kPrepareWorkQueueIdx, [&vec_meta_info] {
auto var_ref = std::make_unique<std::vector<std::atomic<size_t>>>(
vec_meta_info.size());
for (size_t i = 0; i < vec_meta_info.size(); ++i) {
(*var_ref)[i] = vec_meta_info[i].var_ref_count_;
}
VLOG(4) << "AtomicVarRef:" << var_ref.get() << " " << var_ref->size();
return var_ref;
});
auto var_ref = std::make_unique<AtomicVectorSizeT>(vec_meta_info.size());
for (size_t i = 0; i < vec_meta_info.size(); ++i) {
(*var_ref)[i] = vec_meta_info[i].var_ref_count_;
}
VLOG(4) << "AtomicVarRef:" << var_ref.get() << " " << var_ref->size();
return var_ref;
}
bool var_can_be_deleted(const std::string& name, const BlockDesc& block) {
......@@ -160,7 +207,8 @@ get_unused_vars(const BlockDesc& block,
}
void build_variable_scope(const framework::BlockDesc& block,
VariableScope* var_scope, bool use_local_scope) {
VariableScope* var_scope,
bool use_local_scope) {
VLOG(3) << "Creating Variables";
auto inner_scope = var_scope->GetMutableScope();
......@@ -229,7 +277,8 @@ void create_all_ops(const framework::BlockDesc& block,
}
std::tuple<VariableValueMap, VariableIdMap> build_variable_map(
const VariableNameMap& var_name_map, VariableScope* var_scope,
const VariableNameMap& var_name_map,
VariableScope* var_scope,
bool enforce_exist = true) {
VariableValueMap name2var;
VariableIdMap name2id;
......@@ -293,7 +342,8 @@ void apply_device_guard(const OperatorBase* op_base,
void deal_operator_base(const platform::Place& place,
const VariableScope* var_scope,
std::shared_ptr<OperatorBase> op_base,
OpFuncNode* op_func_node, Scope* local_scope) {
OpFuncNode* op_func_node,
Scope* local_scope) {
platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
auto* dev_ctx = pool.Get(place);
// input, output is prepared. set the other attributes.
......@@ -325,7 +375,8 @@ void build_op_func_list(const platform::Place& place,
const framework::BlockDesc& block,
const std::set<std::string>& skip_gc_vars,
std::vector<OpFuncNode>* vec_func_list,
VariableScope* var_scope, bool use_local_scope) {
VariableScope* var_scope,
bool use_local_scope) {
Scope* local_scope = use_local_scope ? var_scope->GetMutableLocalScope()
: var_scope->GetMutableScope();
auto& all_op_kernels = OperatorWithKernel::AllOpKernels();
......@@ -428,9 +479,14 @@ void build_op_func_list(const platform::Place& place,
// NOTE(zhiqiu): op_func_node->operator_base_ maybe changed in
// ApplyDataTransform
ApplyDataTransform(expected_kernel_key, place, &ins_map_temp,
&outs_map_temp, var_scope, &op_func_node,
vec_func_list, use_local_scope);
ApplyDataTransform(expected_kernel_key,
place,
&ins_map_temp,
&outs_map_temp,
var_scope,
&op_func_node,
vec_func_list,
use_local_scope);
op_with_kernel = const_cast<framework::OperatorWithKernel*>(
static_cast<const framework::OperatorWithKernel*>(
op_func_node.operator_base_.get()));
......@@ -463,8 +519,8 @@ void build_op_func_list(const platform::Place& place,
op_with_kernel->Info().infer_shape_(&infer_shape_ctx);
}
auto exec_ctx = ExecutionContext(*op_with_kernel, *runtime_scope,
*dev_ctx, runtime_context);
auto exec_ctx = ExecutionContext(
*op_with_kernel, *runtime_scope, *dev_ctx, runtime_context);
auto run_phi_kernel = false;
if (phi::KernelFactory::Instance().HasCompatiblePhiKernel(
......@@ -498,25 +554,27 @@ void build_op_func_list(const platform::Place& place,
<< " : expected_kernel_key : " << expected_kernel_key;
if (run_phi_kernel) {
phi::KernelContext pt_kernel_context;
op_with_kernel->BuildPhiKernelContext(runtime_context, dev_ctx,
&pt_kernel_context);
op_with_kernel->BuildPhiKernelContext(
runtime_context, dev_ctx, &pt_kernel_context);
op_func_node.pt_kernel_ = op_with_kernel->PhiKernel();
(*op_func_node.pt_kernel_)(&pt_kernel_context);
} else {
auto kernels_iter = all_op_kernels.find(op->Type());
PADDLE_ENFORCE_NE(
kernels_iter, all_op_kernels.end(),
kernels_iter,
all_op_kernels.end(),
platform::errors::Unavailable(
"There are no kernels which are registered in the %s operator.",
op->Type()));
OpKernelMap& kernels = kernels_iter->second;
auto kernel_iter = kernels.find(expected_kernel_key);
PADDLE_ENFORCE_NE(
kernel_iter, kernels.end(),
platform::errors::NotFound(
"Operator (%s) does not have kernel for %s.", op->Type(),
KernelTypeToString(expected_kernel_key)));
PADDLE_ENFORCE_NE(kernel_iter,
kernels.end(),
platform::errors::NotFound(
"Operator (%s) does not have kernel for %s.",
op->Type(),
KernelTypeToString(expected_kernel_key)));
// TODO(zhiqiu): add fallback logic
op_func_node.kernel_func_ = OpKernelComputeFunc(kernel_iter->second);
op_func_node.kernel_func_(exec_ctx);
......@@ -525,9 +583,13 @@ void build_op_func_list(const platform::Place& place,
// post-process grad_op.outputs if need cast complex grad into real grad.
// NOTE(Aurelius84): insert a transfer_dtype_op inplacely to cast it.
if (framework::IsComplexType(expected_kernel_key.data_type_)) {
interpreter::HandleComplexGradToRealGrad(
op_func_node, place, outputs_names, &runtime_context.outputs,
var_scope, vec_func_list, local_scope);
interpreter::HandleComplexGradToRealGrad(op_func_node,
place,
outputs_names,
&runtime_context.outputs,
var_scope,
vec_func_list,
local_scope);
}
if (!op_func_node.inplace_back_map.empty()) {
auto& m = op_func_node.inplace_back_map;
......@@ -583,7 +645,8 @@ void build_op_func_list(const platform::Place& place,
} else {
PADDLE_THROW(platform::errors::Unimplemented(
"Type %s of variable %s is not supported eager deletion.",
framework::ToTypeName(var->Type()), var_name));
framework::ToTypeName(var->Type()),
var_name));
}
}
delete garbages; // free mem
......@@ -612,8 +675,8 @@ void add_fetch(const std::vector<std::string>& fetch_names,
std::vector<size_t> merge_vector(const std::vector<size_t>& first,
const std::vector<size_t>& second) {
std::vector<size_t> out(first.size() + second.size());
std::merge(first.begin(), first.end(), second.begin(), second.end(),
out.begin());
std::merge(
first.begin(), first.end(), second.begin(), second.end(), out.begin());
std::vector<size_t>::iterator it;
it = std::unique(out.begin(), out.end());
......@@ -625,7 +688,8 @@ std::vector<size_t> merge_vector(const std::vector<size_t>& first,
void update_var_min_rw_op(const std::map<int, std::set<int>>& op2dependences,
std::map<int, std::list<int>>* var2min_rw_op,
int cur_op, int rw_var) {
int cur_op,
int rw_var) {
// rw_var is inputs or outputs of cur_op
// this function update the var2min_rw_op set .
if (var2min_rw_op->find(rw_var) == var2min_rw_op->end()) {
......@@ -637,7 +701,8 @@ void update_var_min_rw_op(const std::map<int, std::set<int>>& op2dependences,
var2min_rw_op->at(rw_var).push_back(cur_op);
}
void AddDownstreamOp(int prior_op_idx, int posterior_op_idx,
void AddDownstreamOp(int prior_op_idx,
int posterior_op_idx,
std::map<int, std::list<int>>* op_downstream_map) {
if (op_downstream_map->find(prior_op_idx) == op_downstream_map->end()) {
op_downstream_map->emplace(std::make_pair(prior_op_idx, std::list<int>()));
......@@ -645,7 +710,8 @@ void AddDownstreamOp(int prior_op_idx, int posterior_op_idx,
op_downstream_map->at(prior_op_idx).push_back(posterior_op_idx);
}
void AddDownstreamOp(int prior_op_idx, int posterior_op_idx,
void AddDownstreamOp(int prior_op_idx,
int posterior_op_idx,
std::map<int, std::list<int>>* op_downstream_map,
const std::vector<std::vector<bool>>& op_happens_before) {
if (op_downstream_map->find(prior_op_idx) != op_downstream_map->end()) {
......@@ -675,7 +741,8 @@ const std::string StringizeDownstreamMap(
std::ostringstream oss;
for (auto pair : downstream_map) {
oss << pair.first << " -> ";
std::copy(pair.second.begin(), pair.second.end(),
std::copy(pair.second.begin(),
pair.second.end(),
std::ostream_iterator<int>(oss, " "));
oss << std::endl;
}
......@@ -717,8 +784,8 @@ void ShrinkDownstreamMap(std::map<int, std::list<int>>* downstream_map,
op_happens_before->resize(op_num);
for (size_t i = 0; i < op_num; ++i) {
(*op_happens_before)[i].resize(op_num);
std::fill((*op_happens_before)[i].begin(), (*op_happens_before)[i].end(),
false);
std::fill(
(*op_happens_before)[i].begin(), (*op_happens_before)[i].end(), false);
}
// bfs to get all next ops
......@@ -735,11 +802,15 @@ void ShrinkDownstreamMap(std::map<int, std::list<int>>* downstream_map,
}
for (auto next : downstream_map->at(op)) {
if (!visited[next]) {
PADDLE_ENFORCE_EQ((*op_happens_before)[next][op_idx], false,
PADDLE_ENFORCE_EQ((*op_happens_before)[next][op_idx],
false,
paddle::platform::errors::AlreadyExists(
"There exists circle in graph, expected "
"%d->%d, but already got %d->%d",
op_idx, next, next, op_idx));
op_idx,
next,
next,
op_idx));
(*op_happens_before)[op_idx][next] = true;
VLOG(8) << "happens before: " << op_idx << " " << next;
q.push(next);
......@@ -892,8 +963,8 @@ std::map<int, std::list<int>> build_op_downstream_map(
for (size_t op_idx = 0; op_idx < op_num; ++op_idx) {
if (random_op_set.count(vec_instruction[op_idx].OpBase()->Type())) {
if (dependence_op_idx != -1) {
AddDownstreamOp(dependence_op_idx, op_idx, &op_downstream_map,
*op_happens_before);
AddDownstreamOp(
dependence_op_idx, op_idx, &op_downstream_map, *op_happens_before);
}
dependence_op_idx = op_idx;
}
......@@ -919,8 +990,8 @@ std::map<int, std::list<int>> build_op_downstream_map(
for (size_t op_idx = 0; op_idx < op_num; ++op_idx) {
if (is_comm_op(vec_instruction[op_idx].OpBase()->Type())) {
if (dependence_op_idx != -1) {
AddDownstreamOp(dependence_op_idx, op_idx, &op_downstream_map,
*op_happens_before);
AddDownstreamOp(
dependence_op_idx, op_idx, &op_downstream_map, *op_happens_before);
VLOG(4) << "Add depend from "
<< vec_instruction[dependence_op_idx].OpBase()->Type() << " to "
<< vec_instruction[op_idx].OpBase()->Type();
......@@ -948,8 +1019,8 @@ std::map<int, std::list<int>> build_op_downstream_map(
VLOG(4) << "Add depend from "
<< vec_instruction[dependence_op_idx].OpBase()->Type() << " to "
<< vec_instruction[op_idx].OpBase()->Type();
AddDownstreamOp(dependence_op_idx, op_idx, &op_downstream_map,
*op_happens_before);
AddDownstreamOp(
dependence_op_idx, op_idx, &op_downstream_map, *op_happens_before);
}
}
}
......@@ -1002,10 +1073,13 @@ std::map<int, std::list<int>> build_op_downstream_map(
// first_read_fused_out_op)
// add depend: them->first_read_fused_out_op
for (auto j = op_idx + 1;
j < static_cast<size_t>(first_read_fused_out_op); ++j) {
j < static_cast<size_t>(first_read_fused_out_op);
++j) {
for (auto var_id : outputs) {
if (is_write(vec_instruction[j], var_id)) {
AddDownstreamOp(j, first_read_fused_out_op, &op_downstream_map,
AddDownstreamOp(j,
first_read_fused_out_op,
&op_downstream_map,
*op_happens_before);
VLOG(4) << j << " -> " << first_read_fused_out_op;
VLOG(4)
......@@ -1055,7 +1129,9 @@ std::map<int, std::list<int>> build_op_downstream_map(
for (size_t op_idx = 0; op_idx < op_num; ++op_idx) {
if (!IsCpuOp(vec_instruction[op_idx])) {
if (dependence_op_idx != -1) {
AddDownstreamOp(dependence_op_idx, op_idx, &op_downstream_map,
AddDownstreamOp(dependence_op_idx,
op_idx,
&op_downstream_map,
*op_happens_before);
VLOG(4) << "Add depend from "
<< vec_instruction[dependence_op_idx].OpBase()->Type() << "("
......
paddle/fluid/framework/new_executor/interpretercore_util.h
浏览文件 @ bd7f28bb
......@@ -44,49 +44,23 @@
#include "paddle/fluid/platform/device_context.h"
#include "paddle/fluid/platform/init.h"
using AtomicVectorSizeT = std::vector<std::atomic<size_t>>;
namespace paddle {
namespace framework {
namespace interpreter {
using AtomicVectorSizeT =
std::future<std::unique_ptr<std::vector<std::atomic<size_t>>>>;
class AsyncWorkQueue {
public:
AsyncWorkQueue(size_t host_num_threads, size_t deivce_num_threads,
EventsWaiter* waiter)
: host_num_thread_(host_num_threads) {
std::vector<WorkQueueOptions> group_options;
// for execute host Kernel
group_options.emplace_back(/*name*/ "HostTasks",
/*num_threads*/ host_num_threads,
/*allow_spinning*/ true,
/*always_spinning*/ false,
/*track_task*/ false,
/*detached*/ true,
/*events_waiter*/ waiter);
// for launch device Kernel
group_options.emplace_back(/*name*/ "DeviceKernelLaunch",
/*num_threads*/ deivce_num_threads,
/*allow_spinning*/ true,
/*always_spinning*/ true,
/*track_task*/ false,
/*detached*/ true,
/*events_waiter*/ waiter);
// for prepare deps and others
group_options.emplace_back(/*name*/ "Prepare",
/*num_threads*/ 1,
/*allow_spinning*/ true,
/*always_spinning*/ false,
/*track_task*/ false,
/*detached*/ true,
/*events_waiter*/ waiter);
queue_group_ = CreateWorkQueueGroup(group_options);
}
void PrepareAtomicDeps(const std::vector<size_t>& dependecy_count);
void PrepareAtomicVarRef(const std::vector<VariableMetaInfo>& vec_meta_info);
AsyncWorkQueue(size_t host_num_threads,
size_t deivce_num_threads,
EventsWaiter* waiter);
std::future<std::unique_ptr<AtomicVectorSizeT>> PrepareAtomicDeps(
const std::vector<size_t>& dependecy_count);
std::future<std::unique_ptr<AtomicVectorSizeT>> PrepareAtomicVarRef(
const std::vector<VariableMetaInfo>& vec_meta_info);
// void WaitEmpty() { queue_group_->WaitQueueGroupEmpty(); }
......@@ -94,20 +68,16 @@ class AsyncWorkQueue {
void Cancel() { queue_group_->Cancel(); }
std::unique_ptr<std::vector<std::atomic<size_t>>> AtomicDeps() {
return atomic_deps_.get();
}
std::unique_ptr<std::vector<std::atomic<size_t>>> AtomicVarRef() {
return atomic_var_ref_.get();
}
private:
size_t host_num_thread_;
std::unique_ptr<WorkQueueGroup> queue_group_;
AtomicVectorSizeT atomic_deps_;
AtomicVectorSizeT atomic_var_ref_;
};
std::unique_ptr<AtomicVectorSizeT> PrepareAtomicDeps(
const std::vector<size_t>& dependecy_count);
std::unique_ptr<AtomicVectorSizeT> PrepareAtomicVarRef(
const std::vector<VariableMetaInfo>& vec_meta_info);
void build_variable_scope(const framework::BlockDesc& block,
VariableScope* var_scope,
bool use_local_scope = true);
......@@ -116,7 +86,8 @@ void build_op_func_list(const platform::Place& place,
const framework::BlockDesc& block,
const std::set<std::string>& skip_gc_vars,
std::vector<OpFuncNode>* vec_func_list,
VariableScope* var_scope, bool use_local_scope = true);
VariableScope* var_scope,
bool use_local_scope = true);
std::map<int, std::list<int>> build_op_downstream_map(
const std::vector<Instruction>& vec_instruction,
......
paddle/fluid/framework/new_executor/standalone_executor_test.cc
浏览文件 @ bd7f28bb
......@@ -12,15 +12,14 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#include "paddle/fluid/framework/new_executor/standalone_executor.h"
#include <gtest/gtest.h>
#include <chrono>
#include <iostream>
#include <string>
// #include "gperftools/profiler.h"
#include "paddle/fluid/framework/new_executor/standalone_executor.h"
#include "paddle/phi/core/kernel_registry.h"
USE_OP_ITSELF(fill_constant);
......@@ -61,6 +60,7 @@ USE_OP_ITSELF(sgd);
USE_OP(squared_l2_norm);
USE_OP_ITSELF(memcpy_h2d);
USE_OP_ITSELF(memcpy_d2h);
USE_OP_ITSELF(fetch_v2);
PD_DECLARE_KERNEL(full, GPU, ALL_LAYOUT);
PD_DECLARE_KERNEL(uniform_random_raw, GPU, ALL_LAYOUT);
......@@ -72,6 +72,7 @@ PD_DECLARE_KERNEL(concat_grad, GPU, ALL_LAYOUT);
PD_DECLARE_KERNEL(matmul, GPU, ALL_LAYOUT);
PD_DECLARE_KERNEL(add_raw, KPS, ALL_LAYOUT);
PD_DECLARE_KERNEL(add, KPS, ALL_LAYOUT);
PD_DECLARE_KERNEL(add, CPU, ALL_LAYOUT);
PD_DECLARE_KERNEL(multiply, KPS, ALL_LAYOUT);
PD_DECLARE_KERNEL(multiply_grad, GPU, ALL_LAYOUT);
PD_DECLARE_KERNEL(divide, KPS, ALL_LAYOUT);
......@@ -99,8 +100,6 @@ PD_DECLARE_KERNEL(cross_entropy_with_softmax, GPU, ALL_LAYOUT);
PD_DECLARE_KERNEL(cross_entropy_with_softmax_grad, GPU, ALL_LAYOUT);
PD_DECLARE_KERNEL(sqrt, GPU, ALL_LAYOUT);
DECLARE_double(eager_delete_tensor_gb);
namespace paddle {
namespace framework {
......@@ -115,15 +114,11 @@ ProgramDesc load_from_file(const std::string& file_name) {
return program_desc;
}
TEST(StandaloneExecutor, run) {
FLAGS_eager_delete_tensor_gb = 0.1;
int64_t batch_size = 20;
auto place = platform::CUDAPlace(0);
auto test_prog = load_from_file("lm_startup_program");
auto main_prog = load_from_file("lm_main_program");
ProgramDesc GetLmMainProgram() {
ProgramDesc main_prog = load_from_file("lm_main_program");
auto& global_block = main_prog.Block(0);
int64_t batch_size = 20;
auto& op1 = global_block.AllOps()[1];
auto shape1 = BOOST_GET_CONST(std::vector<int64_t>, op1->GetAttr("shape"));
......@@ -139,6 +134,13 @@ TEST(StandaloneExecutor, run) {
auto shape3 = BOOST_GET_CONST(std::vector<int64_t>, op3->GetAttr("shape"));
shape3[0] = batch_size;
op3->SetAttr("shape", shape3);
return main_prog;
}
TEST(StandaloneExecutor, run) {
auto place = platform::CUDAPlace(0);
ProgramDesc test_prog = load_from_file("lm_startup_program");
ProgramDesc main_prog = GetLmMainProgram();
Scope scope;
StandaloneExecutor exec(place, test_prog, main_prog, &scope);
......@@ -159,31 +161,10 @@ TEST(StandaloneExecutor, run) {
std::cout << "time cost " << diff.count() << std::endl;
}
TEST(StandaloneExecutor, skip_gc_vars) {
FLAGS_eager_delete_tensor_gb = 0;
int64_t batch_size = 20;
TEST(InterpreterCore, skip_gc_vars) {
auto place = platform::CUDAPlace(0);
auto startup_prog = load_from_file("lm_startup_program");
auto main_prog = load_from_file("lm_main_program");
auto& global_block = main_prog.Block(0);
auto& op1 = global_block.AllOps()[1];
auto shape1 = BOOST_GET_CONST(std::vector<int64_t>, op1->GetAttr("shape"));
shape1[0] = batch_size * 20;
op1->SetAttr("shape", shape1);
auto& op2 = global_block.AllOps()[2];
auto shape2 = BOOST_GET_CONST(std::vector<int64_t>, op2->GetAttr("shape"));
shape2[0] = batch_size;
op2->SetAttr("shape", shape2);
auto& op3 = global_block.AllOps()[3];
auto shape3 = BOOST_GET_CONST(std::vector<int64_t>, op3->GetAttr("shape"));
shape3[0] = batch_size;
op3->SetAttr("shape", shape3);
ProgramDesc startup_prog = load_from_file("lm_startup_program");
ProgramDesc main_prog = GetLmMainProgram();
Scope scope;
......@@ -192,20 +173,24 @@ TEST(StandaloneExecutor, skip_gc_vars) {
CreateInterpreterCore(place, startup_prog, &startup_scope);
startup_core->Run({}, {});
std::set<std::string> skip_gc_vars = {"uniform_0.tmp_0", "transpose_0.tmp_0",
"embedding_0.tmp_0", "slice_0.tmp_0",
std::set<std::string> skip_gc_vars = {"uniform_0.tmp_0",
"transpose_0.tmp_0",
"embedding_0.tmp_0",
"slice_0.tmp_0",
"split_1.tmp_2"};
std::set<std::string> gc_vars = {"uniform_1.tmp_0", "matmul_0.tmp_0",
"split_0.tmp_0", "elementwise_add_0.tmp_0",
std::set<std::string> gc_vars = {"uniform_1.tmp_0",
"matmul_0.tmp_0",
"split_0.tmp_0",
"elementwise_add_0.tmp_0",
"tmp_0"};
auto check_gc_result = [](VariableScope& scope, std::set<std::string>& vars,
bool is_skip_gc) {
for (const std::string& var_name : vars) {
ASSERT_EQ(
scope.FindVar(var_name)->GetMutable<LoDTensor>()->IsInitialized(),
is_skip_gc);
}
};
auto check_gc_result =
[](VariableScope& scope, std::set<std::string>& vars, bool is_skip_gc) {
for (const std::string& var_name : vars) {
ASSERT_EQ(
scope.FindVar(var_name)->GetMutable<LoDTensor>()->IsInitialized(),
is_skip_gc);
}
};
VariableScope main_scope(&scope);
std::shared_ptr<InterpreterCore> main_core =
......@@ -220,5 +205,68 @@ TEST(StandaloneExecutor, skip_gc_vars) {
check_gc_result(main_scope, gc_vars, false);
}
void TestShareWorkQueue(const ProgramDesc& prog,
const std::vector<std::string>& feed_names,
const std::vector<LoDTensor>& feed_tensors,
const std::vector<std::string>& fetch_names,
const std::vector<float>& fetch_results) {
const platform::CPUPlace place = platform::CPUPlace();
Scope scope;
VariableScope variable_scope(&scope);
std::shared_ptr<InterpreterCore> core1 =
CreateInterpreterCore(place, prog, &variable_scope, fetch_names);
std::shared_ptr<InterpreterCore> core2 =
CreateInterpreterCore(place, prog, &variable_scope, fetch_names);
core2->ShareWorkQueueFrom(core1);
auto run_and_check = [&feed_names, &feed_tensors, &fetch_results](
std::shared_ptr<InterpreterCore> core) {
FetchList fetch_list = core->Run(feed_names, feed_tensors);
for (size_t i = 0; i < fetch_list.size(); ++i) {
const float* fetch_data =
BOOST_GET_CONST(LoDTensor, fetch_list[i]).data<float>();
ASSERT_FLOAT_EQ(*fetch_data, fetch_results.at(i));
}
};
run_and_check(core1);
run_and_check(core2);
run_and_check(core1);
run_and_check(core2);
}
TEST(InterpreterCore, workqueue_multiplexing) {
ProgramDesc program;
BlockDesc* main_block = program.MutableBlock(0);
VarDesc* var_a = main_block->Var("a");
VarDesc* var_b = main_block->Var("b");
VarDesc* var_c = main_block->Var("c");
var_a->SetType(proto::VarType::LOD_TENSOR);
var_b->SetType(proto::VarType::LOD_TENSOR);
var_c->SetType(proto::VarType::LOD_TENSOR);
OpDesc* add = main_block->AppendOp();
add->SetType("elementwise_add");
add->SetInput("X", {"a"});
add->SetInput("Y", {"b"});
add->SetOutput("Out", {"c"});
float data_a[] = {0, 1, 2, 3};
float data_b[] = {0.0, 0.1, 0.2, 0.3};
phi::DDim dims = phi::make_ddim({2, 2});
const platform::CPUPlace place = platform::CPUPlace();
LoDTensor tensor_a = LoDTensor();
LoDTensor tensor_b = LoDTensor();
std::copy_n(data_a, 4, tensor_a.mutable_data<float>(dims, place));
std::copy_n(data_b, 4, tensor_b.mutable_data<float>(dims, place));
TestShareWorkQueue(
program, {"a", "b"}, {tensor_a, tensor_b}, {"c"}, {0.0, 1.1, 2.2, 3.3});
}
} // namespace framework
} // namespace paddle
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