experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[Build cleanup] Split "core_cpu_impl" into fine-grained targets (2/n).

Browse Source 
This change splits the executor-related targets into separate cc_library targets.

PiperOrigin-RevId: 307915231
Change-Id: I641dafb6723b4ea77b1ab984c28ab56c3f09eec3
This commit is contained in:
Derek Murray 2020-04-22 15:32:59 -07:00 committed by TensorFlower Gardener
parent f21a440cde
commit a9c4cbb424
11 changed files with 203 additions and 60 deletions

139
tensorflow/core/common_runtime/BUILD
View File

@ -218,16 +218,14 @@ filegroup(
"debugger_state_interface.h",
"device_resolver_local.h",
"dma_helper.h",
"entry.h",
"executor.h",
"executor_factory.h",
"function_optimization_registry.h",
"graph_optimizer.h",
"graph_view.h",
"immutable_executor_state.h",
"input_colocation_exemption_registry.h",
"isolate_placer_inspection_required_ops_pass.h",
"local_device.h",
"local_executor_params.h",
"lower_function_call_op.h",
"lower_if_op.h",
"lower_case_op.h",
@ -236,14 +234,11 @@ filegroup(
"memory_types.h",
"mkl_cpu_allocator.h",
"optimization_registry.h",
"pending_counts.h",
"partitioning_utils.h",
"placer.h",
"process_util.h",
"inspecting_placer.h",
"profile_handler.h",
"propagator_debug_utils.h",
"propagator_state.h",
"renamed_device.h",
"rendezvous_mgr.h",
"rendezvous_util.h",
@ -252,7 +247,6 @@ filegroup(
"ring_alg.h",
"ring_gatherer.h",
"session_factory.h",
"simple_propagator_state.h",
"single_threaded_cpu_device.h",
"stats_publisher_interface.h",
"step_stats_collector.h",
@ -487,6 +481,49 @@ cc_library(
],
)
cc_library(
name = "executor",
srcs = ["executor.cc"],
hdrs = ["executor.h"],
copts = tf_copts(),
deps = [
":costmodel_manager",
":device",
":entry",
":executor_factory",
":graph_view",
":immutable_executor_state",
":local_executor_params",
":pending_counts",
":propagator_state",
":renamed_device",
":simple_propagator_state",
":step_stats_collector",
"//tensorflow/core:framework",
"//tensorflow/core:framework_internal",
"//tensorflow/core:graph",
"//tensorflow/core:lib",
"//tensorflow/core:lib_internal",
"//tensorflow/core:protos_all_cc",
"//tensorflow/core/profiler/lib:annotated_traceme",
"//tensorflow/core/profiler/lib:scoped_annotation",
"//tensorflow/core/profiler/lib:traceme",
"@com_google_absl//absl/memory",
],
alwayslink = 1,
)
cc_library(
name = "executor_factory",
srcs = ["executor_factory.cc"],
hdrs = ["executor_factory.h"],
copts = tf_copts(),
deps = [
"//tensorflow/core:graph",
"//tensorflow/core:lib",
],
)
cc_library(
name = "graph_view",
srcs = ["graph_view.cc"],
@ -538,6 +575,23 @@ cc_library(
alwayslink = 1,
)
cc_library(
name = "immutable_executor_state",
srcs = ["immutable_executor_state.cc"],
hdrs = ["immutable_executor_state.h"],
copts = tf_copts(),
deps = [
":graph_view",
":local_executor_params",
":pending_counts",
"//tensorflow/core:framework",
"//tensorflow/core:framework_internal",
"//tensorflow/core:graph",
"//tensorflow/core:lib",
"@com_google_absl//absl/memory",
],
)
cc_library(
name = "input_colocation_exemption_registry",
srcs = ["input_colocation_exemption_registry.cc"],
@ -564,6 +618,16 @@ cc_library(
],
)
cc_library(
name = "local_executor_params",
hdrs = ["local_executor_params.h"],
copts = tf_copts(),
deps = [
"//tensorflow/core:framework",
"//tensorflow/core:lib",
],
)
cc_library(
name = "memory_types",
srcs = ["memory_types.cc"],
@ -688,6 +752,36 @@ cc_library(
],
)
cc_library(
name = "propagator_debug_utils",
srcs = ["propagator_debug_utils.cc"],
hdrs = ["propagator_debug_utils.h"],
copts = tf_copts(),
deps = [
":entry",
":graph_view",
"//tensorflow/core:framework",
"//tensorflow/core:lib",
],
)
cc_library(
name = "propagator_state",
srcs = ["propagator_state.cc"],
hdrs = ["propagator_state.h"],
copts = tf_copts(),
deps = [
":entry",
":graph_view",
":immutable_executor_state",
":pending_counts",
":propagator_debug_utils",
"//tensorflow/core:framework",
"//tensorflow/core:lib",
"//tensorflow/core/profiler/lib:traceme",
],
)
cc_library(
name = "renamed_device",
srcs = ["renamed_device.cc"],
@ -856,6 +950,23 @@ cc_library(
],
)
cc_library(
name = "simple_propagator_state",
srcs = ["simple_propagator_state.cc"],
hdrs = ["simple_propagator_state.h"],
copts = tf_copts(),
deps = [
":entry",
":graph_view",
":immutable_executor_state",
":pending_counts",
":propagator_debug_utils",
"//tensorflow/core:framework",
"//tensorflow/core:lib",
"//tensorflow/core/profiler/lib:traceme",
],
)
cc_library(
name = "single_threaded_cpu_device",
srcs = ["single_threaded_cpu_device.cc"],
@ -952,13 +1063,10 @@ tf_cuda_library(
"colocation_graph.cc",
"composite_device.cc",
"constant_folding.cc",
"executor.cc",
"executor_factory.cc",
"function.cc",
"function_optimization_registry.cc",
"graph_optimizer.cc",
"graph_runner.cc",
"immutable_executor_state.cc",
"inspecting_placer.cc",
"isolate_placer_inspection_required_ops_pass.cc",
"lower_case_op.cc",
@ -970,9 +1078,6 @@ tf_cuda_library(
"placer_inspection_required_ops_utils.cc",
"placer_inspection_required_ops_utils.h",
"process_function_library_runtime.cc",
"propagator_debug_utils.cc",
"propagator_state.cc",
"simple_propagator_state.cc",
"//tensorflow/core/graph:core_cpu_impl_srcs",
],
hdrs = [":core_cpu_lib_headers"],
@ -980,10 +1085,17 @@ tf_cuda_library(
deps = [
":device",
":entry",
":executor",
":executor_factory",
":graph_view",
":local_executor_params",
":immutable_executor_state",
":input_colocation_exemption_registry",
":pending_counts",
":propagator_debug_utils",
":propagator_state",
":session_options",
":simple_propagator_state",
":single_threaded_cpu_device",
"//tensorflow/core:graph",
"//tensorflow/core:framework",
@ -1277,6 +1389,7 @@ tf_cc_tests(
":core_cpu",
":core_cpu_internal",
":direct_session_internal",
":pending_counts",
"//tensorflow/cc:cc_ops",
"//tensorflow/cc:cc_ops_internal",
"//tensorflow/cc:function_ops",

19
tensorflow/core/common_runtime/executor.h
View File

@ -17,7 +17,7 @@ limitations under the License.
#define TENSORFLOW_CORE_COMMON_RUNTIME_EXECUTOR_H_
#include "tensorflow/core/common_runtime/device.h"
#include "tensorflow/core/common_runtime/rendezvous_mgr.h"
#include "tensorflow/core/common_runtime/local_executor_params.h"
#include "tensorflow/core/framework/rendezvous.h"
#include "tensorflow/core/framework/session_state.h"
#include "tensorflow/core/framework/tensor.h"
@ -131,23 +131,6 @@ class Executor {
//
// "params" provides a set of context for the executor. We expect that
// different context would provide different implementations.
struct LocalExecutorParams {
Device* device;
const SessionMetadata* session_metadata = nullptr;
// The library runtime support.
FunctionLibraryRuntime* function_library = nullptr;
// create_kernel returns an instance of op kernel based on NodeDef.
// delete_kernel is called for every kernel used by the executor
// when the executor is deleted.
std::function<Status(const std::shared_ptr<const NodeProperties>&,
OpKernel**)>
create_kernel;
std::function<void(OpKernel*)> delete_kernel;
};
::tensorflow::Status NewLocalExecutor(const LocalExecutorParams& params,
const Graph& graph, Executor** executor);

2
tensorflow/core/common_runtime/immutable_executor_state.cc
View File

@ -16,8 +16,8 @@ limitations under the License.
#include "tensorflow/core/common_runtime/immutable_executor_state.h"
#include "absl/memory/memory.h"
#include "tensorflow/core/common_runtime/metrics.h"
#include "tensorflow/core/framework/function.h"
#include "tensorflow/core/framework/metrics.h"
#include "tensorflow/core/framework/node_def_util.h"
#include "tensorflow/core/graph/edgeset.h"
#include "tensorflow/core/graph/graph.h"

2
tensorflow/core/common_runtime/immutable_executor_state.h
View File

@ -20,8 +20,8 @@ limitations under the License.
#include <memory>
#include <vector>
#include "tensorflow/core/common_runtime/executor.h"
#include "tensorflow/core/common_runtime/graph_view.h"
#include "tensorflow/core/common_runtime/local_executor_params.h"
#include "tensorflow/core/common_runtime/pending_counts.h"
#include "tensorflow/core/framework/tensor.h"
#include "tensorflow/core/lib/core/status.h"

54
tensorflow/core/common_runtime/local_executor_params.h Normal file
View File

@ -0,0 +1,54 @@
/* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#ifndef TENSORFLOW_CORE_COMMON_RUNTIME_LOCAL_EXECUTOR_PARAMS_H_
#define TENSORFLOW_CORE_COMMON_RUNTIME_LOCAL_EXECUTOR_PARAMS_H_
#include <functional>
#include <memory>
namespace tensorflow {
class Device;
class StepStatsCollector;
class SessionMetadata;
class FunctionLibraryRuntime;
class NodeProperties;
class OpKernel;
class Status;
// LocalExecutorParams provides arguments that will be shared by all invocations
// of an executor. We expect that different contexts would provide different
// implementations (e.g. local versus distributed).
struct LocalExecutorParams {
Device* device;
const SessionMetadata* session_metadata = nullptr;
// The library runtime support.
FunctionLibraryRuntime* function_library = nullptr;
// create_kernel returns an instance of op kernel based on NodeDef.
// delete_kernel is called for every kernel used by the executor
// when the executor is deleted.
std::function<Status(const std::shared_ptr<const NodeProperties>&,
OpKernel**)>
create_kernel;
std::function<void(OpKernel*)> delete_kernel;
};
} // end namespace tensorflow
#endif // TENSORFLOW_CORE_COMMON_RUNTIME_LOCAL_EXECUTOR_PARAMS_H_

15
tensorflow/core/common_runtime/propagator_debug_utils.cc
View File

@ -14,11 +14,12 @@ limitations under the License.
==============================================================================*/
#include "tensorflow/core/common_runtime/propagator_debug_utils.h"
#include <vector>
#include "tensorflow/core/common_runtime/entry.h"
#include "tensorflow/core/common_runtime/immutable_executor_state.h"
#include "tensorflow/core/common_runtime/graph_view.h"
#include "tensorflow/core/framework/op_kernel.h"
#include "tensorflow/core/framework/tensor.h"
#include "tensorflow/core/platform/logging.h"
#include "tensorflow/core/platform/strcat.h"
namespace tensorflow {
@ -39,10 +40,8 @@ const Tensor* GetTensorValueForDump(const Entry& input) {
}
}
void DumpPendingNodeState(const ImmutableExecutorState& immutable_state,
const int node_id, const Entry* input_vector,
void DumpPendingNodeState(const NodeItem& node_item, const Entry* input_vector,
const bool show_nodes_with_no_ready_inputs) {
const NodeItem& node_item = immutable_state.graph_view().node_ref(node_id);
const int input_base = node_item.input_start;
if (!show_nodes_with_no_ready_inputs) {
bool has_ready_input = false;
@ -73,9 +72,7 @@ void DumpPendingNodeState(const ImmutableExecutorState& immutable_state,
}
}
void DumpActiveNodeState(const ImmutableExecutorState& immutable_state,
const int node_id, const Entry* input_vector) {
const NodeItem& node_item = immutable_state.graph_view().node_ref(node_id);
void DumpActiveNodeState(const NodeItem& node_item, const Entry* input_vector) {
LOG(WARNING) << " Active Node: " << node_item.DebugString();
const int input_base = node_item.input_start;
for (int i = 0; i < node_item.num_inputs; ++i) {

16
tensorflow/core/common_runtime/propagator_debug_utils.h
View File

@ -18,22 +18,20 @@ limitations under the License.
namespace tensorflow {
struct Entry;
class ImmutableExecutorState;
struct NodeItem;
class Tensor;
// Returns a pointer to the tensor in `input` if one exists, or `nullptr`.
const Tensor* GetTensorValueForDump(const Entry& input);
// Writes a LOG(WARNING) message describing the state of the pending node
// `node_id` in the graph described by `immutable_state`.
void DumpPendingNodeState(const ImmutableExecutorState& immutable_state,
const int node_id, const Entry* input_vector,
// Writes a LOG(WARNING) message describing the state of the given pending node
// in the graph described by `immutable_state`.
void DumpPendingNodeState(const NodeItem& node_item, const Entry* input_vector,
const bool show_nodes_with_no_ready_inputs);
// Writes a LOG(WARNING) message describing the state of the active node
// `node_id` in the graph described by `immutable_state`.
void DumpActiveNodeState(const ImmutableExecutorState& immutable_state,
const int node_id, const Entry* input_vector);
// Writes a LOG(WARNING) message describing the state of the given active node
// in the graph described by `immutable_state`.
void DumpActiveNodeState(const NodeItem& node_item, const Entry* input_vector);
} // namespace tensorflow

7
tensorflow/core/common_runtime/propagator_state.cc
View File

@ -17,6 +17,7 @@ limitations under the License.
#include "tensorflow/core/common_runtime/graph_view.h"
#include "tensorflow/core/common_runtime/propagator_debug_utils.h"
#include "tensorflow/core/framework/op_kernel.h"
#include "tensorflow/core/lib/hash/hash.h"
#include "tensorflow/core/profiler/lib/traceme.h"
@ -177,8 +178,7 @@ void PropagatorState::DumpIterationState(const FrameState* frame,
immutable_state_.pending_ids()[node->node_id];
if (iteration->node_state(pending_id) == PendingCounts::PENDING_NOTREADY ||
iteration->node_state(pending_id) == PendingCounts::PENDING_READY) {
DumpPendingNodeState(immutable_state_, node->node_id,
iteration->input_tensors, false);
DumpPendingNodeState(*node, iteration->input_tensors, false);
}
}
// Then the active nodes.
@ -186,8 +186,7 @@ void PropagatorState::DumpIterationState(const FrameState* frame,
PendingCounts::Handle pending_id =
immutable_state_.pending_ids()[node->node_id];
if (iteration->node_state(pending_id) == PendingCounts::STARTED) {
DumpActiveNodeState(immutable_state_, node->node_id,
iteration->input_tensors);
DumpActiveNodeState(*node, iteration->input_tensors);
}
}
// Show all input tensors in use.

1
tensorflow/core/common_runtime/propagator_state.h
View File

@ -33,7 +33,6 @@ limitations under the License.
namespace tensorflow {
typedef gtl::InlinedVector<TensorValue, 4> TensorValueVec;
typedef gtl::InlinedVector<AllocatorAttributes, 4> AllocatorAttributeVec;
// Represents the ephemeral "edge state" associated with one invocation of

7
tensorflow/core/common_runtime/simple_propagator_state.cc
View File

@ -17,6 +17,7 @@ limitations under the License.
#include <atomic>
#include "tensorflow/core/common_runtime/propagator_debug_utils.h"
#include "tensorflow/core/framework/op_kernel.h"
#include "tensorflow/core/profiler/lib/traceme.h"
namespace tensorflow {
@ -107,15 +108,13 @@ void SimplePropagatorState::DumpState() {
// Dump any waiting nodes that are holding on to tensors.
for (const NodeItem* node : *nodes_) {
if (pending_[node->node_id]) {
DumpPendingNodeState(immutable_state_, node->node_id,
input_tensors_.data(), false);
DumpPendingNodeState(*node, input_tensors_.data(), false);
}
}
// Then the active nodes.
for (const NodeItem* node : *nodes_) {
if ((*active_)[node->node_id]) {
DumpActiveNodeState(immutable_state_, node->node_id,
input_tensors_.data());
DumpActiveNodeState(*node, input_tensors_.data());
}
}
// Show all input tensors in use.

1
tensorflow/core/kernels/data/BUILD
View File

@ -170,6 +170,7 @@ cc_library(
"//tensorflow/core:core_cpu",
"//tensorflow/core:core_cpu_internal",
"//tensorflow/core:lib",
"//tensorflow/core/common_runtime:local_executor_params",
],
alwayslink = 1,
)