m53297601
/
Paddle
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

merge develop

Browse Source 
test=develop
panyx0718-patch-1
sneaxiy 6 years ago
parent cf8d2e67e3 080740b340
commit d231e55065
173 changed files with 3318 additions and 1287 deletions
Show Stats Download Patch File Download Diff File

7
paddle/fluid/API.spec
Unescape View File

@ -118,9 +118,10 @@ paddle.fluid.layers.label_smooth ArgSpec(args=['label', 'prior_dist', 'epsilon',
paddle.fluid.layers.roi_pool ArgSpec(args=['input', 'rois', 'pooled_height', 'pooled_width', 'spatial_scale'], varargs=None, keywords=None, defaults=(1, 1, 1.0)) paddle.fluid.layers.roi_pool ArgSpec(args=['input', 'rois', 'pooled_height', 'pooled_width', 'spatial_scale'], varargs=None, keywords=None, defaults=(1, 1, 1.0))
paddle.fluid.layers.roi_align ArgSpec(args=['input', 'rois', 'pooled_height', 'pooled_width', 'spatial_scale', 'sampling_ratio', 'name'], varargs=None, keywords=None, defaults=(1, 1, 1.0, -1, None)) paddle.fluid.layers.roi_align ArgSpec(args=['input', 'rois', 'pooled_height', 'pooled_width', 'spatial_scale', 'sampling_ratio', 'name'], varargs=None, keywords=None, defaults=(1, 1, 1.0, -1, None))
paddle.fluid.layers.dice_loss ArgSpec(args=['input', 'label', 'epsilon'], varargs=None, keywords=None, defaults=(1e-05,)) paddle.fluid.layers.dice_loss ArgSpec(args=['input', 'label', 'epsilon'], varargs=None, keywords=None, defaults=(1e-05,))
paddle.fluid.layers.image_resize ArgSpec(args=['input', 'out_shape', 'scale', 'name', 'resample'], varargs=None, keywords=None, defaults=(None, None, None, 'BILINEAR')) paddle.fluid.layers.image_resize ArgSpec(args=['input', 'out_shape', 'scale', 'name', 'resample', 'actual_shape'], varargs=None, keywords=None, defaults=(None, None, None, 'BILINEAR', None))
paddle.fluid.layers.image_resize_short ArgSpec(args=['input', 'out_short_len', 'resample'], varargs=None, keywords=None, defaults=('BILINEAR',)) paddle.fluid.layers.image_resize_short ArgSpec(args=['input', 'out_short_len', 'resample'], varargs=None, keywords=None, defaults=('BILINEAR',))
paddle.fluid.layers.resize_bilinear ArgSpec(args=['input', 'out_shape', 'scale', 'name'], varargs=None, keywords=None, defaults=(None, None, None)) paddle.fluid.layers.resize_bilinear ArgSpec(args=['input', 'out_shape', 'scale', 'name', 'actual_shape'], varargs=None, keywords=None, defaults=(None, None, None, None))
paddle.fluid.layers.resize_nearest ArgSpec(args=['input', 'out_shape', 'scale', 'name', 'actual_shape'], varargs=None, keywords=None, defaults=(None, None, None, None))
paddle.fluid.layers.gather ArgSpec(args=['input', 'index'], varargs=None, keywords=None, defaults=None) paddle.fluid.layers.gather ArgSpec(args=['input', 'index'], varargs=None, keywords=None, defaults=None)
paddle.fluid.layers.scatter ArgSpec(args=['input', 'index', 'updates', 'name'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.scatter ArgSpec(args=['input', 'index', 'updates', 'name'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.sequence_scatter ArgSpec(args=['input', 'index', 'updates', 'name'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.sequence_scatter ArgSpec(args=['input', 'index', 'updates', 'name'], varargs=None, keywords=None, defaults=(None,))
@ -178,6 +179,7 @@ paddle.fluid.layers.space_to_depth ArgSpec(args=['x', 'blocksize', 'name'], vara
paddle.fluid.layers.affine_grid ArgSpec(args=['theta', 'out_shape', 'name'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.affine_grid ArgSpec(args=['theta', 'out_shape', 'name'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.sequence_reverse ArgSpec(args=['x', 'name'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.sequence_reverse ArgSpec(args=['x', 'name'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.affine_channel ArgSpec(args=['x', 'scale', 'bias', 'data_layout', 'name'], varargs=None, keywords=None, defaults=(None, None, 'NCHW', None)) paddle.fluid.layers.affine_channel ArgSpec(args=['x', 'scale', 'bias', 'data_layout', 'name'], varargs=None, keywords=None, defaults=(None, None, 'NCHW', None))
paddle.fluid.layers.similarity_focus ArgSpec(args=['input', 'axis', 'indexes', 'name'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.hash ArgSpec(args=['input', 'hash_size', 'num_hash', 'name'], varargs=None, keywords=None, defaults=(1, None)) paddle.fluid.layers.hash ArgSpec(args=['input', 'hash_size', 'num_hash', 'name'], varargs=None, keywords=None, defaults=(1, None))
paddle.fluid.layers.grid_sampler ArgSpec(args=['x', 'grid', 'name'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.grid_sampler ArgSpec(args=['x', 'grid', 'name'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.log_loss ArgSpec(args=['input', 'label', 'epsilon', 'name'], varargs=None, keywords=None, defaults=(0.0001, None)) paddle.fluid.layers.log_loss ArgSpec(args=['input', 'label', 'epsilon', 'name'], varargs=None, keywords=None, defaults=(0.0001, None))
@ -200,6 +202,7 @@ paddle.fluid.layers.create_tensor ArgSpec(args=['dtype', 'name', 'persistable'],
paddle.fluid.layers.create_parameter ArgSpec(args=['shape', 'dtype', 'name', 'attr', 'is_bias', 'default_initializer'], varargs=None, keywords=None, defaults=(None, None, False, None)) paddle.fluid.layers.create_parameter ArgSpec(args=['shape', 'dtype', 'name', 'attr', 'is_bias', 'default_initializer'], varargs=None, keywords=None, defaults=(None, None, False, None))
paddle.fluid.layers.create_global_var ArgSpec(args=['shape', 'value', 'dtype', 'persistable', 'force_cpu', 'name'], varargs=None, keywords=None, defaults=(False, False, None)) paddle.fluid.layers.create_global_var ArgSpec(args=['shape', 'value', 'dtype', 'persistable', 'force_cpu', 'name'], varargs=None, keywords=None, defaults=(False, False, None))
paddle.fluid.layers.cast ArgSpec(args=['x', 'dtype'], varargs=None, keywords=None, defaults=None) paddle.fluid.layers.cast ArgSpec(args=['x', 'dtype'], varargs=None, keywords=None, defaults=None)
paddle.fluid.layers.tensor_array_to_tensor ArgSpec(args=['input', 'axis', 'name'], varargs=None, keywords=None, defaults=(1, None))
paddle.fluid.layers.concat ArgSpec(args=['input', 'axis', 'name'], varargs=None, keywords=None, defaults=(0, None)) paddle.fluid.layers.concat ArgSpec(args=['input', 'axis', 'name'], varargs=None, keywords=None, defaults=(0, None))
paddle.fluid.layers.sums ArgSpec(args=['input', 'out'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.sums ArgSpec(args=['input', 'out'], varargs=None, keywords=None, defaults=(None,))
paddle.fluid.layers.assign ArgSpec(args=['input', 'output'], varargs=None, keywords=None, defaults=(None,)) paddle.fluid.layers.assign ArgSpec(args=['input', 'output'], varargs=None, keywords=None, defaults=(None,))

4
paddle/fluid/framework/data_device_transform.cc
Unescape View File

@ -18,8 +18,8 @@ namespace framework {
void TransDataDevice(const Tensor &in, const platform::Place &dst_place, void TransDataDevice(const Tensor &in, const platform::Place &dst_place,
Tensor *out) { Tensor *out) {
VLOG(3) << "DeviceTransform in, src_place " << in.place() VLOG(30) << "DeviceTransform in, src_place " << in.place()
<< " dst_place: " << dst_place; << " dst_place: " << dst_place;
PADDLE_ENFORCE_NE( PADDLE_ENFORCE_NE(
in.place().which(), dst_place.which(), in.place().which(), dst_place.which(),

6
paddle/fluid/framework/data_device_transform_test.cu
Unescape View File

@ -49,10 +49,10 @@ class TestOpWithKernel : public OperatorWithKernel {
OpKernelType GetExpectedKernelType( OpKernelType GetExpectedKernelType(
const ExecutionContext& ctx) const override { const ExecutionContext& ctx) const override {
if (Attr<bool>("use_gpu")) { if (Attr<bool>("use_gpu")) {
VLOG(3) << "force use gpu kernel"; VLOG(30) << "force use gpu kernel";
return OpKernelType(proto::VarType::FP32, platform::CUDAPlace(0)); return OpKernelType(proto::VarType::FP32, platform::CUDAPlace(0));
} else { } else {
VLOG(3) << "use default kernel"; VLOG(30) << "use default kernel";
return OpKernelType(proto::VarType::FP32, return OpKernelType(proto::VarType::FP32,
ctx.Input<Tensor>("input")->place()); ctx.Input<Tensor>("input")->place());
} }
@ -148,7 +148,7 @@ TEST(Operator, CPUtoGPU) {
// get output // get output
auto* output2 = scope.Var("OUT2"); auto* output2 = scope.Var("OUT2");
gpu_op->Run(scope, cuda_place); gpu_op->Run(scope, cuda_place);
VLOG(3) << "after gpu_op run"; VLOG(30) << "after gpu_op run";
// auto* output2_ptr = output2->Get<LoDTensor>().data<float>(); // auto* output2_ptr = output2->Get<LoDTensor>().data<float>();
paddle::platform::DeviceContextPool& pool = paddle::platform::DeviceContextPool& pool =

2
paddle/fluid/framework/details/broadcast_op_handle.cc
Unescape View File

@ -60,7 +60,7 @@ void BroadcastOpHandle::BroadcastOneVar(
PADDLE_ENFORCE_NOT_NULL(in_var); PADDLE_ENFORCE_NOT_NULL(in_var);
Tensor &in_tensor = VariableVisitor::GetMutableTensor(in_var); Tensor &in_tensor = VariableVisitor::GetMutableTensor(in_var);
if (UNLIKELY(!in_tensor.IsInitialized())) { if (UNLIKELY(!in_tensor.IsInitialized())) {
VLOG(3) << "in var " << in_var_handle.name_ << "not inited, return!"; VLOG(30) << "in var " << in_var_handle.name_ << "not inited, return!";
return; return;
} }

4
paddle/fluid/framework/details/modify_op_lock_and_record_event_pass.cc
Unescape View File

@ -45,8 +45,8 @@ std::unique_ptr<ir::Graph> ModifyOpLockAndRecordEventPass::ApplyImpl(
IsLockAndRecordEventFreeComputationOpHandle(compute_op, graph_view); IsLockAndRecordEventFreeComputationOpHandle(compute_op, graph_view);
compute_op->SetLockAndRecordEventFree(is_lock_and_record_event_free); compute_op->SetLockAndRecordEventFree(is_lock_and_record_event_free);
if (is_lock_and_record_event_free) { if (is_lock_and_record_event_free) {
VLOG(10) << "Set is_lock_and_record_event_free be true in op " VLOG(100) << "Set is_lock_and_record_event_free be true in op "
<< compute_op->DebugString(); << compute_op->DebugString();
} }
} }
return ir_graph; return ir_graph;

12
paddle/fluid/framework/details/multi_devices_graph_pass.cc
Unescape View File

@ -399,7 +399,7 @@ std::unique_ptr<ir::Graph> MultiDevSSAGraphBuilder::ApplyImpl(
for (size_t i = 0; i < backward_vars.size(); i += 2) { for (size_t i = 0; i < backward_vars.size(); i += 2) {
auto &p_name = backward_vars[i]; auto &p_name = backward_vars[i];
auto &g_name = backward_vars[i + 1]; auto &g_name = backward_vars[i + 1];
VLOG(10) << "Bcast " << g_name << " for parameter " << p_name; VLOG(100) << "Bcast " << g_name << " for parameter " << p_name;
switch (strategy_.reduce_) { switch (strategy_.reduce_) {
case BuildStrategy::ReduceStrategy::kReduce: case BuildStrategy::ReduceStrategy::kReduce:
@ -809,8 +809,8 @@ int MultiDevSSAGraphBuilder::CreateRPCOp(
node->Op()->GetAttr(OpProtoAndCheckerMaker::OpRoleVarAttrName())); node->Op()->GetAttr(OpProtoAndCheckerMaker::OpRoleVarAttrName()));
PADDLE_ENFORCE_EQ(send_param_grad.size(), 2U); PADDLE_ENFORCE_EQ(send_param_grad.size(), 2U);
op_dev_id = GetAppropriateDeviceID({send_param_grad[1]}); op_dev_id = GetAppropriateDeviceID({send_param_grad[1]});
VLOG(10) << "send grad " << input_var_names[0] << " origin " VLOG(100) << "send grad " << input_var_names[0] << " origin "
<< send_param_grad[1] << " place: " << op_dev_id; << send_param_grad[1] << " place: " << op_dev_id;
for (auto &varname : input_var_names) { for (auto &varname : input_var_names) {
sharded_var_device->emplace(varname, op_dev_id); sharded_var_device->emplace(varname, op_dev_id);
} }
@ -826,9 +826,9 @@ int MultiDevSSAGraphBuilder::CreateRPCOp(
if (recv_param_grad.size() == 2U) { if (recv_param_grad.size() == 2U) {
op_dev_id = op_dev_id =
GetVarDeviceID(*result, recv_param_grad[1], *sharded_var_device); GetVarDeviceID(*result, recv_param_grad[1], *sharded_var_device);
VLOG(10) << "recv param " << recv_param_grad[0] VLOG(100) << "recv param " << recv_param_grad[0]
<< " get grad place: " << recv_param_grad[1] << " get grad place: " << recv_param_grad[1]
<< " place: " << op_dev_id; << " place: " << op_dev_id;
} else { } else {
op_dev_id = GetAppropriateDeviceID(output_var_names); op_dev_id = GetAppropriateDeviceID(output_var_names);
} }

4
paddle/fluid/framework/details/reference_count_pass.cc
Unescape View File

@ -140,8 +140,8 @@ std::unique_ptr<ir::Graph> ReferenceCountPass::ApplyImpl(
if (next_compute_op != nullptr) { if (next_compute_op != nullptr) {
if (compute_ref_cnt_map.count(next_compute_op)) { if (compute_ref_cnt_map.count(next_compute_op)) {
compute_ref_cnt_map[next_compute_op]->AddVar(var_name); compute_ref_cnt_map[next_compute_op]->AddVar(var_name);
VLOG(5) << "Add reference count of " << var_name << " to Operator " VLOG(50) << "Add reference count of " << var_name << " to Operator "
<< next_compute_op->Name(); << next_compute_op->Name();
} else { } else {
// Create new reference_count_op_handle // Create new reference_count_op_handle
ir::Node *ref_cnt_node = graph->CreateEmptyNode( ir::Node *ref_cnt_node = graph->CreateEmptyNode(

2
paddle/fluid/framework/details/scale_loss_grad_op_handle.cc
Unescape View File

@ -51,7 +51,7 @@ void ScaleLossGradOpHandle::RunImpl() {
->stream(); ->stream();
memory::Copy(boost::get<platform::CUDAPlace>(place_), tmp, memory::Copy(boost::get<platform::CUDAPlace>(place_), tmp,
platform::CPUPlace(), &coeff_, sizeof(float), stream); platform::CPUPlace(), &coeff_, sizeof(float), stream);
VLOG(10) << place_ << "RUN Scale loss grad op"; VLOG(100) << place_ << "RUN Scale loss grad op";
}); });
#endif #endif
} }

4
paddle/fluid/framework/details/sequential_execution_pass.cc
Unescape View File

@ -94,8 +94,8 @@ std::unique_ptr<ir::Graph> SequentialExecutionPass::ApplyImpl(
op_node_list[i - 1]->outputs.push_back(dep_var); op_node_list[i - 1]->outputs.push_back(dep_var);
dep_var->outputs.push_back(op_node_list[i]); dep_var->outputs.push_back(op_node_list[i]);
dep_var->inputs.push_back(op_node_list[i - 1]); dep_var->inputs.push_back(op_node_list[i - 1]);
VLOG(10) << "Add dependencies between " << op_node_list[i - 1]->Name() VLOG(100) << "Add dependencies between " << op_node_list[i - 1]->Name()
<< " and " << op_node_list[i]->Name(); << " and " << op_node_list[i]->Name();
} }
return graph; return graph;
} }

8
paddle/fluid/framework/details/threaded_ssa_graph_executor.cc
Unescape View File

@ -210,16 +210,16 @@ void ThreadedSSAGraphExecutor::RunOp(
details::OpHandleBase *op) { details::OpHandleBase *op) {
auto op_run = [ready_var_q, op, this] { auto op_run = [ready_var_q, op, this] {
try { try {
if (VLOG_IS_ON(10)) { if (VLOG_IS_ON(100)) {
VLOG(10) << op << " " << op->Name() << " : " << op->DebugString(); VLOG(100) << op << " " << op->Name() << " : " << op->DebugString();
} }
if (LIKELY(!strategy_.dry_run_)) { if (LIKELY(!strategy_.dry_run_)) {
op->Run(strategy_.use_cuda_); op->Run(strategy_.use_cuda_);
} }
VLOG(10) << op << " " << op->Name() << " Done "; VLOG(100) << op << " " << op->Name() << " Done ";
running_ops_--; running_ops_--;
ready_var_q->Extend(op->Outputs()); ready_var_q->Extend(op->Outputs());
VLOG(10) << op << " " << op->Name() << "Signal posted"; VLOG(100) << op << " " << op->Name() << "Signal posted";
} catch (...) { } catch (...) {
exception_holder_.Catch(std::current_exception()); exception_holder_.Catch(std::current_exception());
} }

22
paddle/fluid/framework/executor.cc
Unescape View File

@ -43,7 +43,7 @@ ExecutorPrepareContext::ExecutorPrepareContext(
} }
ExecutorPrepareContext::~ExecutorPrepareContext() { ExecutorPrepareContext::~ExecutorPrepareContext() {
VLOG(5) << "destroy ExecutorPrepareContext"; VLOG(50) << "destroy ExecutorPrepareContext";
} }
template <typename RefCntMap> template <typename RefCntMap>
@ -60,7 +60,7 @@ static void DeleteUnusedTensors(const Scope& scope, const OperatorBase* op,
if ((it->second)-- == 1) { if ((it->second)-- == 1) {
auto* var = scope.FindVar(name); auto* var = scope.FindVar(name);
if (var != nullptr) { if (var != nullptr) {
VLOG(10) << "Erase tensor \'" << name << "\'"; VLOG(100) << "Erase tensor \'" << name << "\'";
if (var->IsType<LoDTensor>()) { if (var->IsType<LoDTensor>()) {
erase_tensors.insert(var->GetMutable<LoDTensor>()); erase_tensors.insert(var->GetMutable<LoDTensor>());
} else if (var->IsType<SelectedRows>()) { } else if (var->IsType<SelectedRows>()) {
@ -141,21 +141,21 @@ void Executor::CreateVariables(const ProgramDesc& pdesc, Scope* scope,
if (var->Persistable()) { if (var->Persistable()) {
auto* ptr = const_cast<Scope*>(ancestor_scope)->Var(var->Name()); auto* ptr = const_cast<Scope*>(ancestor_scope)->Var(var->Name());
InitializeVariable(ptr, var->GetType()); InitializeVariable(ptr, var->GetType());
VLOG(3) << "Create Variable " << var->Name() VLOG(30) << "Create Variable " << var->Name()
<< " global, which pointer is " << ptr; << " global, which pointer is " << ptr;
} else { } else {
auto* ptr = scope->Var(var->Name()); auto* ptr = scope->Var(var->Name());
InitializeVariable(ptr, var->GetType()); InitializeVariable(ptr, var->GetType());
VLOG(3) << "Create Variable " << var->Name() VLOG(30) << "Create Variable " << var->Name()
<< " locally, which pointer is " << ptr; << " locally, which pointer is " << ptr;
} }
} }
} else { } else {
for (auto& var : global_block.AllVars()) { for (auto& var : global_block.AllVars()) {
auto* ptr = scope->Var(var->Name()); auto* ptr = scope->Var(var->Name());
InitializeVariable(ptr, var->GetType()); InitializeVariable(ptr, var->GetType());
VLOG(3) << "Create variable " << var->Name() << ", which pointer is " VLOG(30) << "Create variable " << var->Name() << ", which pointer is "
<< ptr; << ptr;
} }
} }
} }
@ -286,7 +286,7 @@ void Executor::Run(const ProgramDesc& program, Scope* scope,
int i = 0; int i = 0;
for (auto& feed_target : (*feed_targets)) { for (auto& feed_target : (*feed_targets)) {
std::string var_name = feed_target.first; std::string var_name = feed_target.first;
VLOG(3) << "feed target's name: " << var_name; VLOG(30) << "feed target's name: " << var_name;
// prepend feed op // prepend feed op
auto* op = global_block->PrependOp(); auto* op = global_block->PrependOp();
@ -309,7 +309,7 @@ void Executor::Run(const ProgramDesc& program, Scope* scope,
int i = 0; int i = 0;
for (auto& fetch_target : (*fetch_targets)) { for (auto& fetch_target : (*fetch_targets)) {
std::string var_name = fetch_target.first; std::string var_name = fetch_target.first;
VLOG(3) << "fetch target's name: " << var_name; VLOG(30) << "fetch target's name: " << var_name;
// append fetch op // append fetch op
auto* op = global_block->AppendOp(); auto* op = global_block->AppendOp();
@ -459,7 +459,7 @@ void Executor::RunPreparedContext(
void Executor::EnableMKLDNN(const ProgramDesc& program) { void Executor::EnableMKLDNN(const ProgramDesc& program) {
#ifdef PADDLE_WITH_MKLDNN #ifdef PADDLE_WITH_MKLDNN
VLOG(3) << "use_mkldnn=True"; VLOG(30) << "use_mkldnn=True";
for (size_t bid = 0; bid < program.Size(); ++bid) { for (size_t bid = 0; bid < program.Size(); ++bid) {
auto* block = const_cast<ProgramDesc&>(program).MutableBlock(bid); auto* block = const_cast<ProgramDesc&>(program).MutableBlock(bid);
for (auto* op : block->AllOps()) { for (auto* op : block->AllOps()) {

6
paddle/fluid/framework/feed_fetch_method.cc
Unescape View File

@ -25,7 +25,7 @@ void SetFeedVariable(Scope* scope, const LoDTensor& input,
const std::string& var_name, size_t index) { const std::string& var_name, size_t index) {
// If var_name Variable is not found in GlobalScope, a new variable will // If var_name Variable is not found in GlobalScope, a new variable will
// be created. // be created.
VLOG(3) << "SetFeedVariable name=" << var_name << " index=" << index; VLOG(30) << "SetFeedVariable name=" << var_name << " index=" << index;
Variable* g_feed_value = scope->Var(var_name); Variable* g_feed_value = scope->Var(var_name);
auto& feed_inputs = *(g_feed_value->GetMutable<FeedFetchList>()); auto& feed_inputs = *(g_feed_value->GetMutable<FeedFetchList>());
if (index >= feed_inputs.size()) { if (index >= feed_inputs.size()) {
@ -47,8 +47,8 @@ LoDTensor& GetFetchVariable(const Scope& scope, const std::string& var_name,
typeid(FeedFetchList).name()); typeid(FeedFetchList).name());
auto& fetch_outputs = *g_fetch_value->GetMutable<FeedFetchList>(); auto& fetch_outputs = *g_fetch_value->GetMutable<FeedFetchList>();
auto& tensor = fetch_outputs[index]; auto& tensor = fetch_outputs[index];
VLOG(3) << "Fetch " << var_name << " with index " << index VLOG(30) << "Fetch " << var_name << " with index " << index
<< " shape= " << tensor.dims(); << " shape= " << tensor.dims();
PADDLE_ENFORCE_LT(index, fetch_outputs.size()); PADDLE_ENFORCE_LT(index, fetch_outputs.size());
return tensor; return tensor;
} }

28
paddle/fluid/framework/ir/attention_lstm_fuse_pass.cc
Unescape View File

@ -147,19 +147,19 @@ void PrepareParameters(Graph* graph, const Param& param) {
scope->Var(param.LSTMX)->GetMutable<LoDTensor>(); scope->Var(param.LSTMX)->GetMutable<LoDTensor>();
scope->Var(param.LSTMOUT)->GetMutable<LoDTensor>(); scope->Var(param.LSTMOUT)->GetMutable<LoDTensor>();
#define GATE_W(name__) \ #define GATE_W(name__) \
auto* W_##name__##_w0 = scope->FindVar(#name__ ".w_0"); \ auto* W_##name__##_w0 = scope->FindVar(#name__ ".w_0"); \
auto* W_##name__##_w1 = scope->FindVar(#name__ ".w_1"); \ auto* W_##name__##_w1 = scope->FindVar(#name__ ".w_1"); \
auto* W_##name__##_b0 = scope->FindVar(#name__ ".b_0"); \ auto* W_##name__##_b0 = scope->FindVar(#name__ ".b_0"); \
CHECK_P3(W_##name__##_w0, W_##name__##_w1, W_##name__##_b0); \ CHECK_P3(W_##name__##_w0, W_##name__##_w1, W_##name__##_b0); \
VLOG(4) << #name__ "_w0" \ VLOG(40) << #name__ "_w0" \
<< " shape: " << W_##name__##_w0->Get<LoDTensor>().dims(); \ << " shape: " << W_##name__##_w0->Get<LoDTensor>().dims(); \
VLOG(4) << #name__ "_w1" \ VLOG(40) << #name__ "_w1" \
<< " shape: " << W_##name__##_w1->Get<LoDTensor>().dims(); \ << " shape: " << W_##name__##_w1->Get<LoDTensor>().dims(); \
VLOG(4) << #name__ "_b0" \ VLOG(40) << #name__ "_b0" \
<< " shape: " << W_##name__##_b0->Get<LoDTensor>().dims(); \ << " shape: " << W_##name__##_b0->Get<LoDTensor>().dims(); \
auto& W_##name__##_w0_t = W_##name__##_w0->Get<LoDTensor>(); \ auto& W_##name__##_w0_t = W_##name__##_w0->Get<LoDTensor>(); \
auto& W_##name__##_w1_t = W_##name__##_w1->Get<LoDTensor>(); \ auto& W_##name__##_w1_t = W_##name__##_w1->Get<LoDTensor>(); \
auto& W_##name__##_b0_t = W_##name__##_b0->Get<LoDTensor>(); auto& W_##name__##_b0_t = W_##name__##_b0->Get<LoDTensor>();
GATE_W(forget); GATE_W(forget);
@ -208,7 +208,7 @@ void PrepareLSTMWeight(const LoDTensor& W_forget_w0,
int D = W_forget_w0.dims()[0]; int D = W_forget_w0.dims()[0];
int M = W_forget_w1.dims()[0]; int M = W_forget_w1.dims()[0];
out->Resize(make_ddim({D + M, 4 * D})); out->Resize(make_ddim({D + M, 4 * D}));
VLOG(3) << "LSTMWeight resized to " << out->dims(); VLOG(30) << "LSTMWeight resized to " << out->dims();
float* out_data = out->mutable_data<float>(platform::CPUPlace()); float* out_data = out->mutable_data<float>(platform::CPUPlace());
std::array<const float*, 4> tensors( std::array<const float*, 4> tensors(

4
paddle/fluid/framework/ir/conv_bias_mkldnn_fuse_pass.cc
Unescape View File

@ -57,7 +57,7 @@ std::unique_ptr<ir::Graph> ConvBiasFusePass::ApplyImpl(
int found_conv_bias_count = 0; int found_conv_bias_count = 0;
auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
Graph* g) { Graph* g) {
VLOG(4) << "handle ConvBias fuse"; VLOG(40) << "handle ConvBias fuse";
GET_IR_NODE_FROM_SUBGRAPH(conv_weight, conv_weight, GET_IR_NODE_FROM_SUBGRAPH(conv_weight, conv_weight,
conv_bias_pattern); // Filter conv_bias_pattern); // Filter
GET_IR_NODE_FROM_SUBGRAPH(conv_out, conv_out, conv_bias_pattern); // tmp GET_IR_NODE_FROM_SUBGRAPH(conv_out, conv_out, conv_bias_pattern); // tmp
@ -74,7 +74,7 @@ std::unique_ptr<ir::Graph> ConvBiasFusePass::ApplyImpl(
// check if fuse can be done and if MKL-DNN should be used // check if fuse can be done and if MKL-DNN should be used
FuseOptions fuse_option = FindFuseOption(*conv, *eltwise); FuseOptions fuse_option = FindFuseOption(*conv, *eltwise);
if (fuse_option == DO_NOT_FUSE || fuse_option == FUSE_NATIVE) { if (fuse_option == DO_NOT_FUSE || fuse_option == FUSE_NATIVE) {
VLOG(3) << "do not perform conv+bias fuse"; VLOG(30) << "do not perform conv+bias fuse";
return; return;
} }

6
paddle/fluid/framework/ir/conv_bn_fuse_pass.cc
Unescape View File

@ -121,7 +121,7 @@ std::unique_ptr<ir::Graph> ConvBNFusePass::ApplyImpl(
int found_conv_bn_count = 0; int found_conv_bn_count = 0;
auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
Graph* g) { Graph* g) {
VLOG(4) << "handle ConvBN fuse"; VLOG(40) << "handle ConvBN fuse";
// conv, batch_norm, // conv, batch_norm,
// conv_weight, conv_out, // conv_weight, conv_out,
@ -133,7 +133,7 @@ std::unique_ptr<ir::Graph> ConvBNFusePass::ApplyImpl(
// check if fuse can be done and if MKL-DNN should be used // check if fuse can be done and if MKL-DNN should be used
FuseOptions fuse_option = FindFuseOption(*conv, *batch_norm); FuseOptions fuse_option = FindFuseOption(*conv, *batch_norm);
if (fuse_option == DO_NOT_FUSE) { if (fuse_option == DO_NOT_FUSE) {
VLOG(3) << "do not perform conv+bn fuse"; VLOG(30) << "do not perform conv+bn fuse";
return; return;
} }
@ -241,7 +241,7 @@ std::unique_ptr<ir::Graph> ConvEltwiseAddBNFusePass::ApplyImpl(
int found_conv_bn_count = 0; int found_conv_bn_count = 0;
auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
Graph* g) { Graph* g) {
VLOG(4) << "handle ConvBN fuse"; VLOG(40) << "handle ConvBN fuse";
// conv, batch_norm, // conv, batch_norm,
// conv_weight, conv_out, // conv_weight, conv_out,

4
paddle/fluid/framework/ir/conv_relu_mkldnn_fuse_pass.cc
Unescape View File

@ -38,7 +38,7 @@ std::unique_ptr<ir::Graph> ConvReLUFusePass::ApplyImpl(
int found_conv_relu_count = 0; int found_conv_relu_count = 0;
auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
Graph* g) { Graph* g) {
VLOG(4) << "handle ConvReLU fuse"; VLOG(40) << "handle ConvReLU fuse";
GET_IR_NODE_FROM_SUBGRAPH(conv_weight, conv_weight, GET_IR_NODE_FROM_SUBGRAPH(conv_weight, conv_weight,
conv_relu_pattern); // Filter conv_relu_pattern); // Filter
GET_IR_NODE_FROM_SUBGRAPH(conv_out, conv_out, conv_relu_pattern); // tmp GET_IR_NODE_FROM_SUBGRAPH(conv_out, conv_out, conv_relu_pattern); // tmp
@ -48,7 +48,7 @@ std::unique_ptr<ir::Graph> ConvReLUFusePass::ApplyImpl(
FuseOptions fuse_option = FindFuseOption(*conv, *relu); FuseOptions fuse_option = FindFuseOption(*conv, *relu);
if (fuse_option == DO_NOT_FUSE) { if (fuse_option == DO_NOT_FUSE) {
VLOG(3) << "do not perform conv+relu fuse"; VLOG(30) << "do not perform conv+relu fuse";
return; return;
} }

2
paddle/fluid/framework/ir/depthwise_conv_mkldnn_pass.cc
Unescape View File

@ -39,7 +39,7 @@ std::unique_ptr<ir::Graph> DepthwiseConvMKLDNNPass::ApplyImpl(
int found_depthwise_conv_mkldnn_count = 0; int found_depthwise_conv_mkldnn_count = 0;
auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
Graph* g) { Graph* g) {
VLOG(3) << "handle DepthwiseConvMKLDNN fuse"; VLOG(30) << "handle DepthwiseConvMKLDNN fuse";
GET_NODE(depthwise_conv, (*pattern)); GET_NODE(depthwise_conv, (*pattern));
depthwise_conv->Op()->SetType("conv2d"); depthwise_conv->Op()->SetType("conv2d");
found_depthwise_conv_mkldnn_count++; found_depthwise_conv_mkldnn_count++;

2
paddle/fluid/framework/ir/fc_fuse_pass.cc
Unescape View File

@ -39,7 +39,7 @@ std::unique_ptr<ir::Graph> FCFusePass::ApplyImpl(
int found_fc_count = 0; int found_fc_count = 0;
auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t& subgraph,
Graph* g) { Graph* g) {
VLOG(4) << "handle FC fuse"; VLOG(40) << "handle FC fuse";
GET_IR_NODE_FROM_SUBGRAPH(w, w, fc_pattern); GET_IR_NODE_FROM_SUBGRAPH(w, w, fc_pattern);
GET_IR_NODE_FROM_SUBGRAPH(fc_bias, bias, fc_pattern); GET_IR_NODE_FROM_SUBGRAPH(fc_bias, bias, fc_pattern);
GET_IR_NODE_FROM_SUBGRAPH(fc_out, Out, fc_pattern); GET_IR_NODE_FROM_SUBGRAPH(fc_out, Out, fc_pattern);

28
paddle/fluid/framework/ir/fuse_elewise_add_act_pass.cc
Unescape View File

@ -61,7 +61,7 @@ std::unique_ptr<ir::Graph> FuseElewiseAddActPass::FuseElewiseAddAct(
auto handler = [&](const GraphPatternDetector::subgraph_t &subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t &subgraph,
Graph *g) { Graph *g) {
VLOG(4) << "handle FuseElewiseAddAct fuse"; VLOG(40) << "handle FuseElewiseAddAct fuse";
GET_IR_NODE_FROM_SUBGRAPH(ele_y, ele_y, elewise_add_act_pattern); GET_IR_NODE_FROM_SUBGRAPH(ele_y, ele_y, elewise_add_act_pattern);
GET_IR_NODE_FROM_SUBGRAPH(ele_out, elewise_add_out, GET_IR_NODE_FROM_SUBGRAPH(ele_out, elewise_add_out,
elewise_add_act_pattern); elewise_add_act_pattern);
@ -77,10 +77,10 @@ std::unique_ptr<ir::Graph> FuseElewiseAddActPass::FuseElewiseAddAct(
Node *elewise_add_act_node = CreateFuseElewiseAddActNode( Node *elewise_add_act_node = CreateFuseElewiseAddActNode(
g, act, ele_add, ele_x_n, ele_y_n, ele_out_n, act_out_n); g, act, ele_add, ele_x_n, ele_y_n, ele_out_n, act_out_n);
VLOG(4) << "\n\t " << ele_x_n << " and " << ele_y_n << " -> " VLOG(40) << "\n\t " << ele_x_n << " and " << ele_y_n << " -> "
<< ele_add->Name() << " -> " << ele_out_n << "\n" << ele_add->Name() << " -> " << ele_out_n << "\n"
<< "\t " << ele_out_n << " -> " << act->Name() << " -> " << "\t " << ele_out_n << " -> " << act->Name() << " -> "
<< act_out_n; << act_out_n;
ReLinkNodes(g, ele_out, ele_add, act, elewise_add_act_node); ReLinkNodes(g, ele_out, ele_add, act, elewise_add_act_node);
found_elewise_add_act_count++; found_elewise_add_act_count++;
@ -113,7 +113,7 @@ std::unique_ptr<ir::Graph> FuseElewiseAddActPass::FuseActElewiseAdd(
auto handler = [&](const GraphPatternDetector::subgraph_t &subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t &subgraph,
Graph *g) { Graph *g) {
VLOG(4) << "handle FuseElewiseAddAct fuse"; VLOG(40) << "handle FuseElewiseAddAct fuse";
GET_IR_NODE_FROM_SUBGRAPH(act_out, act_out, act_elewise_add_pattern); GET_IR_NODE_FROM_SUBGRAPH(act_out, act_out, act_elewise_add_pattern);
GET_IR_NODE_FROM_SUBGRAPH(ele_x, ele_x, act_elewise_add_pattern); GET_IR_NODE_FROM_SUBGRAPH(ele_x, ele_x, act_elewise_add_pattern);
GET_IR_NODE_FROM_SUBGRAPH(ele_out, elewise_add_out, GET_IR_NODE_FROM_SUBGRAPH(ele_out, elewise_add_out,
@ -129,9 +129,9 @@ std::unique_ptr<ir::Graph> FuseElewiseAddActPass::FuseActElewiseAdd(
Node *elewise_add_act_node = CreateFuseElewiseAddActNode( Node *elewise_add_act_node = CreateFuseElewiseAddActNode(
g, ele_add, act, elewise_add_x_n, act_i_n, act_o_n, elewise_add_out_n); g, ele_add, act, elewise_add_x_n, act_i_n, act_o_n, elewise_add_out_n);
VLOG(4) << "\n\t " << act_i_n << " -> " << act->Name() << " -> " << act_o_n VLOG(40) << "\n\t " << act_i_n << " -> " << act->Name() << " -> " << act_o_n
<< "\n\t " << act_o_n << " and " << elewise_add_x_n << " -> " << "\n\t " << act_o_n << " and " << elewise_add_x_n << " -> "
<< ele_add->Name() << " -> " << elewise_add_out_n; << ele_add->Name() << " -> " << elewise_add_out_n;
ReLinkNodes(g, act_out, act, ele_add, elewise_add_act_node); ReLinkNodes(g, act_out, act, ele_add, elewise_add_act_node);
found_elewise_add_act_count++; found_elewise_add_act_count++;
@ -165,7 +165,7 @@ std::unique_ptr<ir::Graph> FuseElewiseAddActPass::FuseElewiseAddActInplaceGrad(
auto handler = [&](const GraphPatternDetector::subgraph_t &subgraph, auto handler = [&](const GraphPatternDetector::subgraph_t &subgraph,
Graph *g) { Graph *g) {
VLOG(4) << "handle FuseElewiseAddActGrad1 fuse"; VLOG(40) << "handle FuseElewiseAddActGrad1 fuse";
GET_IR_NODE_FROM_SUBGRAPH(act_out, act_out, elewise_add_act_grad_pattern); GET_IR_NODE_FROM_SUBGRAPH(act_out, act_out, elewise_add_act_grad_pattern);
GET_IR_NODE_FROM_SUBGRAPH(act_grad, act_grad, elewise_add_act_grad_pattern); GET_IR_NODE_FROM_SUBGRAPH(act_grad, act_grad, elewise_add_act_grad_pattern);
GET_IR_NODE_FROM_SUBGRAPH(d_itermediate_out, d_itermediate_out, GET_IR_NODE_FROM_SUBGRAPH(d_itermediate_out, d_itermediate_out,
@ -208,10 +208,10 @@ std::unique_ptr<ir::Graph> FuseElewiseAddActPass::FuseElewiseAddActInplaceGrad(
auto fused_node = g->CreateOpNode(&desc); auto fused_node = g->CreateOpNode(&desc);
VLOG(4) << "\n\t " << d_act_out_n << " and " << act_out_n << " -> " VLOG(40) << "\n\t " << d_act_out_n << " and " << act_out_n << " -> "
<< act_grad->Name() << " -> " << d_itermediate_out_n << "\n\t " << act_grad->Name() << " -> " << d_itermediate_out_n << "\n\t "
<< d_itermediate_out_n << " and " << act_out_n << " -> " << d_itermediate_out_n << " and " << act_out_n << " -> "
<< ele_add_grad->Name() << " -> " << d_itermediate_out_n; << ele_add_grad->Name() << " -> " << d_itermediate_out_n;
ReLinkNodes(g, d_itermediate_out, act_grad, ele_add_grad, fused_node); ReLinkNodes(g, d_itermediate_out, act_grad, ele_add_grad, fused_node);
found_elewise_add_act_count++; found_elewise_add_act_count++;

4
paddle/fluid/framework/ir/graph.cc
Unescape View File

@ -92,7 +92,7 @@ Graph::Graph(const ProgramDesc &program) : program_(program) {
std::map<std::string, std::vector<ir::Node *>> Graph::InitFromProgram( std::map<std::string, std::vector<ir::Node *>> Graph::InitFromProgram(
const ProgramDesc &program) { const ProgramDesc &program) {
VLOG(3) << "block in program:" << program_.Size(); VLOG(30) << "block in program:" << program_.Size();
std::unordered_map<std::string, VarDesc *> all_vars; std::unordered_map<std::string, VarDesc *> all_vars;
// var nodes for each var name, will have multiple versions in SSA // var nodes for each var name, will have multiple versions in SSA
std::map<std::string, std::vector<ir::Node *>> var_nodes; std::map<std::string, std::vector<ir::Node *>> var_nodes;
@ -160,7 +160,7 @@ void Graph::ResolveHazard(
auto it_old = versions.rbegin(); auto it_old = versions.rbegin();
++it_old; ++it_old;
for (; it_old != versions.rend(); it_new = it_old, ++it_old) { for (; it_old != versions.rend(); it_new = it_old, ++it_old) {
VLOG(3) << "deal with var: " << (*it_new)->Name(); VLOG(30) << "deal with var: " << (*it_new)->Name();
ir::Node *write_op = ir::Node *write_op =
(*it_new)->inputs.empty() ? nullptr : (*it_new)->inputs[0]; (*it_new)->inputs.empty() ? nullptr : (*it_new)->inputs[0];
const auto &read_ops = (*it_old)->outputs; const auto &read_ops = (*it_old)->outputs;

2
paddle/fluid/framework/ir/graph.h
Unescape View File

@ -89,7 +89,7 @@ class Graph {
attr_name); attr_name);
attrs_[attr_name] = attr; attrs_[attr_name] = attr;
attr_dels_[attr_name] = [attr, attr_name]() { attr_dels_[attr_name] = [attr, attr_name]() {
VLOG(3) << "deleting " << attr_name; VLOG(30) << "deleting " << attr_name;
delete attr; delete attr;
}; };
} }

19
paddle/fluid/framework/ir/graph_helper.cc
Unescape View File

@ -33,8 +33,9 @@ void SortHelper(
} }
} }
VLOG(3) << "topology sort insert: " << node->Name() VLOG(30) << "topology sort insert: " << node->Name()
<< reinterpret_cast<void *>(node) << " input " << node->inputs.size(); << reinterpret_cast<void *>(node) << " input "
<< node->inputs.size();
ret->push_back(node); ret->push_back(node);
} }
@ -103,9 +104,9 @@ std::map<ir::Node *, std::unordered_set<ir::Node *>> BuildOperationAdjList(
for (auto &var : n->inputs) { for (auto &var : n->inputs) {
for (auto &adj_n : var->inputs) { for (auto &adj_n : var->inputs) {
PADDLE_ENFORCE(adj_n->NodeType() == ir::Node::Type::kOperation); PADDLE_ENFORCE(adj_n->NodeType() == ir::Node::Type::kOperation);
VLOG(4) << "adj " << adj_n->Name() << reinterpret_cast<void *>(adj_n) VLOG(40) << "adj " << adj_n->Name() << reinterpret_cast<void *>(adj_n)
<< " -> " << n->Name() << reinterpret_cast<void *>(n) << " -> " << n->Name() << reinterpret_cast<void *>(n)
<< " via " << var->Name() << reinterpret_cast<void *>(var); << " via " << var->Name() << reinterpret_cast<void *>(var);
adj_list[n].insert(adj_n); adj_list[n].insert(adj_n);
} }
} }
@ -163,10 +164,10 @@ size_t GraphNum(const Graph &graph) {
graph_nodes.emplace_back(g_nodes); graph_nodes.emplace_back(g_nodes);
} }
if (VLOG_IS_ON(10)) { if (VLOG_IS_ON(100)) {
VLOG(10) << "graph_num: " << graph_nodes.size(); VLOG(100) << "graph_num: " << graph_nodes.size();
for (auto &g_n : graph_nodes) { for (auto &g_n : graph_nodes) {
VLOG(10) << "graph_nodes: " << g_n.size(); VLOG(100) << "graph_nodes: " << g_n.size();
if (g_n.size() < 10) { if (g_n.size() < 10) {
std::stringstream out; std::stringstream out;
for (auto &node : g_n) { for (auto &node : g_n) {
@ -180,7 +181,7 @@ size_t GraphNum(const Graph &graph) {
} }
out << "]"; out << "]";
} }
VLOG(10) << out.str(); VLOG(100) << out.str();
} }
} }
} }

22
paddle/fluid/framework/ir/graph_pattern_detector.cc
Unescape View File

@ -12,6 +12,7 @@
// See the License for the specific language governing permissions and // See the License for the specific language governing permissions and
// limitations under the License. // limitations under the License.
#include <algorithm>
#include <array> #include <array>
#include <string> #include <string>
#include <vector> #include <vector>
@ -91,19 +92,19 @@ void GraphPatternDetector::operator()(Graph *graph,
PrettyLogEndl(Style::detail(), "--- detect %d subgraphs", subgraphs.size()); PrettyLogEndl(Style::detail(), "--- detect %d subgraphs", subgraphs.size());
int id = 0; int id = 0;
for (auto &g : subgraphs) { for (auto &g : subgraphs) {
VLOG(3) << "optimizing #" << id++ << " subgraph"; VLOG(30) << "optimizing #" << id++ << " subgraph";
handler(g, graph); handler(g, graph);
} }
} }
bool GraphPatternDetector::MarkPDNodesInGraph(const ir::Graph &graph) { bool GraphPatternDetector::MarkPDNodesInGraph(const ir::Graph &graph) {
VLOG(3) << "mark pdnodes in graph"; VLOG(30) << "mark pdnodes in graph";
if (graph.Nodes().empty()) return false; if (graph.Nodes().empty()) return false;
for (auto &node : GraphTraits::DFS(graph)) { for (auto &node : GraphTraits::DFS(graph)) {
for (const auto &pdnode : pattern_.nodes()) { for (const auto &pdnode : pattern_.nodes()) {
if (pdnode->Tell(&node)) { if (pdnode->Tell(&node)) {
VLOG(4) << "pdnode " << pdnode->name() << " marked"; VLOG(40) << "pdnode " << pdnode->name() << " marked";
pdnodes2nodes_[pdnode.get()].insert(&node); pdnodes2nodes_[pdnode.get()].insert(&node);
} }
} }
@ -111,7 +112,7 @@ bool GraphPatternDetector::MarkPDNodesInGraph(const ir::Graph &graph) {
// Check to early stop if some PDNode can't find matched Node. // Check to early stop if some PDNode can't find matched Node.
for (auto &pdnode : pattern_.nodes()) { for (auto &pdnode : pattern_.nodes()) {
if (!pdnodes2nodes_.count(pdnode.get())) { if (!pdnodes2nodes_.count(pdnode.get())) {
VLOG(4) << pdnode->name() << " can't find matched Node, early stop"; VLOG(40) << pdnode->name() << " can't find matched Node, early stop";
// return false; // return false;
} }
} }
@ -120,7 +121,7 @@ bool GraphPatternDetector::MarkPDNodesInGraph(const ir::Graph &graph) {
GetMarkedNodes(const_cast<Graph *>(&graph)).insert(n); GetMarkedNodes(const_cast<Graph *>(&graph)).insert(n);
} }
} }
VLOG(3) << pdnodes2nodes_.size() << " nodes marked"; VLOG(30) << pdnodes2nodes_.size() << " nodes marked";
return !pdnodes2nodes_.empty(); return !pdnodes2nodes_.empty();
} }
@ -213,7 +214,7 @@ GraphPatternDetector::DetectPatterns() {
// Extend a PDNode to subgraphs by deducing the connection relations defined // Extend a PDNode to subgraphs by deducing the connection relations defined
// in edges of PDNodes. // in edges of PDNodes.
for (const auto &edge : pattern_.edges()) { for (const auto &edge : pattern_.edges()) {
VLOG(4) << "check " << edge.first->name() << " -> " << edge.second->name(); VLOG(40) << "check " << edge.first->name() << " -> " << edge.second->name();
// TODO(Superjomn) Fix bug here, the groups might be duplicate here. // TODO(Superjomn) Fix bug here, the groups might be duplicate here.
// Each role has two PDNodes, which indicates two roles. // Each role has two PDNodes, which indicates two roles.
// Detect two Nodes that can match these two roles and they are connected. // Detect two Nodes that can match these two roles and they are connected.
@ -224,7 +225,7 @@ GraphPatternDetector::DetectPatterns() {
// source -> target // source -> target
for (Node *source : pdnodes2nodes_[edge.first]) { for (Node *source : pdnodes2nodes_[edge.first]) {
for (Node *target : pdnodes2nodes_[edge.second]) { for (Node *target : pdnodes2nodes_[edge.second]) {
VLOG(8) << "check " << source->id() << " -- " << target->id(); VLOG(80) << "check " << source->id() << " -- " << target->id();
// TODO(Superjomn) add some prune strategies. // TODO(Superjomn) add some prune strategies.
for (const auto &group : pre_groups) { for (const auto &group : pre_groups) {
HitGroup new_group = group; HitGroup new_group = group;
@ -240,12 +241,13 @@ GraphPatternDetector::DetectPatterns() {
} }
} }
} }
VLOG(3) << "step " << step << " get records: " << cur_groups.size(); VLOG(30) << "step " << step << " get records: " << cur_groups.size();
for (auto &group : cur_groups) { for (auto &group : cur_groups) {
for (auto &item : group.roles) { for (auto &item : group.roles) {
VLOG(4) << "node " << item.second->id() << " as " << item.first->name(); VLOG(40) << "node " << item.second->id() << " as "
<< item.first->name();
} }
VLOG(4) << "========================================================="; VLOG(40) << "=========================================================";
} }
} }

2
paddle/fluid/framework/ir/graph_viz_pass.cc
Unescape View File

@ -41,7 +41,7 @@ std::string FormatName(const Node* node) {
std::unique_ptr<ir::Graph> GraphVizPass::ApplyImpl( std::unique_ptr<ir::Graph> GraphVizPass::ApplyImpl(
std::unique_ptr<ir::Graph> graph) const { std::unique_ptr<ir::Graph> graph) const {
const std::string graph_viz_path = Get<std::string>(kGraphVizPath); const std::string graph_viz_path = Get<std::string>(kGraphVizPath);
VLOG(3) << "draw IR graph viz to " << graph_viz_path; VLOG(30) << "draw IR graph viz to " << graph_viz_path;
std::unique_ptr<std::ostream> fout(new std::ofstream(graph_viz_path)); std::unique_ptr<std::ostream> fout(new std::ofstream(graph_viz_path));
PADDLE_ENFORCE(fout->good()); PADDLE_ENFORCE(fout->good());
std::ostream& sout = *fout; std::ostream& sout = *fout;

2
paddle/fluid/framework/ir/mkldnn_placement_pass.cc
Unescape View File

@ -20,7 +20,7 @@ namespace ir {
std::unique_ptr<ir::Graph> MKLDNNPlacementPass::ApplyImpl( std::unique_ptr<ir::Graph> MKLDNNPlacementPass::ApplyImpl(
std::unique_ptr<ir::Graph> graph) const { std::unique_ptr<ir::Graph> graph) const {
VLOG(3) << "Aplies MKL-DNN placement strategy."; VLOG(30) << "Aplies MKL-DNN placement strategy.";
for (const Node* n : graph->Nodes()) { for (const Node* n : graph->Nodes()) {
if (n->IsOp() && n->Op()->HasAttr("use_mkldnn")) { if (n->IsOp() && n->Op()->HasAttr("use_mkldnn")) {
n->Op()->SetAttr("use_mkldnn", true); n->Op()->SetAttr("use_mkldnn", true);

Some files were not shown because too many files have changed in this diff Show More

Write Preview
Loading…
Cancel
Save