TensorFlow: upstream changes from the afternoon.

Changes: - futurize --stage2 changes for Python 3 compatibility by @girving. - Small updates to documentation by @vrv, schuster and others - Account for failure of std::thread::hardware_concurrency by @ebrevdo. - More changes for backwards-compatibility tests by Josh - Updates to python op doc generation by Josh - Added support for using the best-fit allocator via ConfigProto by @vrv. - Rename LocalSession to DirectSession, since local was a bad name for it. - Enable tf.nn.moments() to work with tensors of unknown shape by @mrry. GITHUB_ISSUE: 139 - Changes for Android build by Andrew. Base CL: 107645181
codezero00 · Nov 12, 2015 · f2102f4 · f2102f4
1 parent 3961abe
commit f2102f4
Show file tree

Hide file tree

Showing 322 changed files with 2,609 additions and 710 deletions.
diff --git a/README.md b/README.md
@@ -33,7 +33,7 @@ installing from source, GPU-enabled support, etc., see
 
 The TensorFlow Python API currently requires Python 2.7: we are
 [working](https://github.com/tensorflow/tensorflow/issues/1) on adding support
-for Python 3.0.
+for Python 3.
 
 The simplest way to install TensorFlow is using
 [pip](https://pypi.python.org/pypi/pip) for both Linux and Mac.

diff --git a/tensorflow/__init__.py b/tensorflow/__init__.py
@@ -1,4 +1,8 @@
 # Bring in all of the public TensorFlow interface into this
 # module.
 # pylint: disable=wildcard-import
+from __future__ import absolute_import
+from __future__ import division
+from __future__ import print_function
+
 from tensorflow.python import *
diff --git a/tensorflow/cc/BUILD b/tensorflow/cc/BUILD
@@ -71,7 +71,6 @@ cc_binary(
     deps = [
         ":cc_ops",
         "//tensorflow/core:kernels",
-        "//tensorflow/core:local",
         "//tensorflow/core:tensorflow",
     ],
 )

diff --git a/tensorflow/core/BUILD b/tensorflow/core/BUILD
@@ -70,8 +70,8 @@ tf_cuda_library(
             "common_runtime/gpu/*.cc",
             "common_runtime/copy_tensor.cc",
             "common_runtime/gpu_device_factory.cc",
-            "common_runtime/local_session.cc",
-            "common_runtime/local_session.h",
+            "common_runtime/direct_session.cc",
+            "common_runtime/direct_session.h",
         ],
     ),
     hdrs = glob(["public/**/*.h"]),
@@ -113,10 +113,10 @@ tf_cuda_library(
 )
 
 tf_cuda_library(
-    name = "local",
+    name = "direct_session",
     srcs = [
-        "common_runtime/local_session.cc",
-        "common_runtime/local_session.h",
+        "common_runtime/direct_session.cc",
+        "common_runtime/direct_session.h",
     ],
     copts = tf_copts(),
     cuda_deps = [
@@ -200,10 +200,10 @@ tf_cuda_library(
     visibility = ["//visibility:public"],
     deps = [
         ":core",
+        ":direct_session",
         ":gpu_runtime",
         ":kernels",
         ":lib",
-        ":local",
     ],
 )
 
@@ -300,7 +300,7 @@ tf_proto_library(
     cc_api_version = 2,
     go_api_version = 2,
     java_api_version = 2,
-    py_api_version = 2,
+    py_api_version = 2,  # TODO(irving): Handle 3
     visibility = ["//tensorflow:internal"],
 )
 
@@ -402,9 +402,9 @@ tf_cc_tests(
     ),
     deps = [
         ":core",
+        ":direct_session",
         ":kernels",
         ":lib",
-        ":local",
         ":test_main",
         ":testlib",
         "//tensorflow/cc:cc_ops",
@@ -423,8 +423,8 @@ tf_cc_tests(
         ],
     ),
     deps = [
+        ":direct_session",
         ":kernels",
-        ":local",
         ":test_main",
         ":testlib",
         "//tensorflow/cc:cc_ops",

diff --git a/...flow/core/common_runtime/local_session.cc → ...low/core/common_runtime/direct_session.cc b/...flow/core/common_runtime/local_session.cc → ...low/core/common_runtime/direct_session.cc
@@ -1,4 +1,4 @@
-#include "tensorflow/core/common_runtime/local_session.h"
+#include "tensorflow/core/common_runtime/direct_session.h"
 
 #include <string>
 #include <vector>
@@ -42,7 +42,7 @@ static bool InitModule(const SessionOptions& options) {
     // Default to using the number of cores available in the process.
     inter_op_parallelism_threads = port::NumSchedulableCPUs();
   }
-  LOG(INFO) << "Local session inter op parallelism threads: "
+  LOG(INFO) << "Direct session inter op parallelism threads: "
             << inter_op_parallelism_threads;
   kernel_thread_pool_ = new thread::ThreadPool(options.env, "Compute",
                                                inter_op_parallelism_threads);
@@ -92,7 +92,7 @@ void SchedClosure(std::function<void()> c) {
 
 }  // namespace
 
-LocalSession::LocalSession(const SessionOptions& options,
+DirectSession::DirectSession(const SessionOptions& options,
                            const DeviceMgr* device_mgr)
     : options_(options),
       device_mgr_(device_mgr),
@@ -124,7 +124,7 @@ LocalSession::LocalSession(const SessionOptions& options,
   }
 }
 
-LocalSession::~LocalSession() {
+DirectSession::~DirectSession() {
   for (auto d : device_mgr_->ListDevices()) {
     d->op_segment()->RemoveHold(session_handle_);
   }
@@ -134,7 +134,7 @@ LocalSession::~LocalSession() {
   delete cancellation_manager_;
 }
 
-Status LocalSession::Create(const GraphDef& graph) {
+Status DirectSession::Create(const GraphDef& graph) {
   mutex_lock l(graph_def_lock_);
   if (graph_created_) {
     return errors::AlreadyExists(
@@ -143,18 +143,18 @@ Status LocalSession::Create(const GraphDef& graph) {
   return ExtendLocked(graph);
 }
 
-Status LocalSession::Extend(const GraphDef& graph) {
+Status DirectSession::Extend(const GraphDef& graph) {
   mutex_lock l(graph_def_lock_);
   return ExtendLocked(graph);
 }
 
-Status LocalSession::ExtendLocked(const GraphDef& graph) {
+Status DirectSession::ExtendLocked(const GraphDef& graph) {
   graph_created_ = true;  // In case this is first call
   graph_def_.MergeFrom(graph);
   return Status::OK();
 }
 
-Status LocalSession::Run(const std::vector<std::pair<string, Tensor>>& inputs,
+Status DirectSession::Run(const std::vector<std::pair<string, Tensor>>& inputs,
                          const std::vector<string>& output_names,
                          const std::vector<string>& target_nodes,
                          std::vector<Tensor>* outputs) {
@@ -250,7 +250,7 @@ Status LocalSession::Run(const std::vector<std::pair<string, Tensor>>& inputs,
   return s;
 }
 
-Status LocalSession::GetOrCreateExecutors(
+Status DirectSession::GetOrCreateExecutors(
     gtl::ArraySlice<string> inputs, gtl::ArraySlice<string> outputs,
     gtl::ArraySlice<string> target_nodes,
     ExecutorsAndKeys** executors_and_keys) {
@@ -358,7 +358,7 @@ Status LocalSession::GetOrCreateExecutors(
   return Status::OK();
 }
 
-void LocalSession::SaveStatefulNodes(Graph* graph) {
+void DirectSession::SaveStatefulNodes(Graph* graph) {
   for (Node* n : graph->nodes()) {
     if (n->op_def().is_stateful()) {
       VLOG(2) << "Saving " << n->DebugString();
@@ -367,7 +367,7 @@ void LocalSession::SaveStatefulNodes(Graph* graph) {
   }
 }
 
-void LocalSession::RestoreStatefulNodes(Graph* graph) {
+void DirectSession::RestoreStatefulNodes(Graph* graph) {
   for (Node* n : graph->nodes()) {
     if (n->op_def().is_stateful()) {
       auto iter = stateful_placements_.find(n->name());
@@ -379,7 +379,7 @@ void LocalSession::RestoreStatefulNodes(Graph* graph) {
   }
 }
 
-Status LocalSession::CreateGraphs(gtl::ArraySlice<string> feeds,
+Status DirectSession::CreateGraphs(gtl::ArraySlice<string> feeds,
                                   gtl::ArraySlice<string> fetches,
                                   gtl::ArraySlice<string> target_nodes,
                                   std::unordered_map<string, Graph*>* outputs) {
@@ -422,7 +422,7 @@ Status LocalSession::CreateGraphs(gtl::ArraySlice<string> feeds,
     return strings::StrCat(prefix, "/_", name_counter_++);
   };
   popts.get_incarnation = [](const string& name) {
-    // The local session does not have changing incarnation numbers.
+    // The direct session does not have changing incarnation numbers.
     // Just return '1'.
     return 1;
   };
@@ -476,29 +476,29 @@ Status LocalSession::CreateGraphs(gtl::ArraySlice<string> feeds,
   return Status::OK();
 }
 
-::tensorflow::Status LocalSession::Close() {
+::tensorflow::Status DirectSession::Close() {
   cancellation_manager_->StartCancel();
   return ::tensorflow::Status::OK();
 }
 
-class LocalSessionFactory : public SessionFactory {
+class DirectSessionFactory : public SessionFactory {
  public:
-  LocalSessionFactory() {}
+  DirectSessionFactory() {}
 
   Session* NewSession(const SessionOptions& options) override {
     std::vector<Device*> devices;
     DeviceFactory::AddDevices(options, "/job:localhost/replica:0/task:0",
                               &devices);
-    return new LocalSession(options, new DeviceMgr(devices));
+    return new DirectSession(options, new DeviceMgr(devices));
   }
 };
 
-class LocalSessionRegistrar {
+class DirectSessionRegistrar {
  public:
-  LocalSessionRegistrar() {
-    SessionFactory::Register("LOCAL_SESSION", new LocalSessionFactory());
+  DirectSessionRegistrar() {
+    SessionFactory::Register("DIRECT_SESSION", new DirectSessionFactory());
   }
 };
-static LocalSessionRegistrar registrar;
+static DirectSessionRegistrar registrar;
 
 }  // namespace tensorflow
diff --git a/...rflow/core/common_runtime/local_session.h → ...flow/core/common_runtime/direct_session.h b/...rflow/core/common_runtime/local_session.h → ...flow/core/common_runtime/direct_session.h
@@ -1,5 +1,5 @@
-#ifndef TENSORFLOW_COMMON_RUNTIME_LOCAL_SESSION_H_
-#define TENSORFLOW_COMMON_RUNTIME_LOCAL_SESSION_H_
+#ifndef TENSORFLOW_COMMON_RUNTIME_DIRECT_SESSION_H_
+#define TENSORFLOW_COMMON_RUNTIME_DIRECT_SESSION_H_
 
 #include <memory>
 #include <string>
@@ -22,11 +22,11 @@ namespace tensorflow {
 
 class Device;
 
-class LocalSession : public Session {
+class DirectSession : public Session {
  public:
   // Takes ownership of 'device_mgr'.
-  LocalSession(const SessionOptions& options, const DeviceMgr* device_mgr);
-  ~LocalSession() override;
+  DirectSession(const SessionOptions& options, const DeviceMgr* device_mgr);
+  ~DirectSession() override;
 
   ::tensorflow::Status Create(const GraphDef& graph) override;
   ::tensorflow::Status Extend(const GraphDef& graph) override;
@@ -101,9 +101,9 @@ class LocalSession : public Session {
   // For generating unique names.
   int64 name_counter_ GUARDED_BY(mu_) = 0;
 
-  TF_DISALLOW_COPY_AND_ASSIGN(LocalSession);
+  TF_DISALLOW_COPY_AND_ASSIGN(DirectSession);
 };
 
 }  // end namespace tensorflow
 
-#endif  // TENSORFLOW_COMMON_RUNTIME_LOCAL_SESSION_H_
+#endif  // TENSORFLOW_COMMON_RUNTIME_DIRECT_SESSION_H_
diff --git a/...core/common_runtime/local_session_test.cc → ...ore/common_runtime/direct_session_test.cc b/...core/common_runtime/local_session_test.cc → ...ore/common_runtime/direct_session_test.cc
@@ -1,4 +1,4 @@
-#include "tensorflow/core/common_runtime/local_session.h"
+#include "tensorflow/core/common_runtime/direct_session.h"
 
 #include <map>
 #include <string>
@@ -29,7 +29,7 @@ Session* CreateSession() {
   return NewSession(options);
 }
 
-class LocalSessionMinusAXTest : public ::testing::Test {
+class DirectSessionMinusAXTest : public ::testing::Test {
  public:
   void Initialize(std::initializer_list<float> a_values) {
     RequireDefaultOps();
@@ -64,7 +64,7 @@ class LocalSessionMinusAXTest : public ::testing::Test {
   GraphDef def_;
 };
 
-TEST_F(LocalSessionMinusAXTest, RunSimpleNetwork) {
+TEST_F(DirectSessionMinusAXTest, RunSimpleNetwork) {
   Initialize({3, 2, -1, 0});
   std::unique_ptr<Session> session(CreateSession());
   ASSERT_TRUE(session != nullptr);
@@ -86,7 +86,7 @@ TEST_F(LocalSessionMinusAXTest, RunSimpleNetwork) {
   EXPECT_FLOAT_EQ(5.0, mat(0, 0));
 }
 
-TEST_F(LocalSessionMinusAXTest, TestFeed) {
+TEST_F(DirectSessionMinusAXTest, TestFeed) {
   Initialize({1, 2, 3, 4});
   std::unique_ptr<Session> session(CreateSession());
   ASSERT_TRUE(session != nullptr);
@@ -115,7 +115,7 @@ TEST_F(LocalSessionMinusAXTest, TestFeed) {
   EXPECT_FLOAT_EQ(39.0, mat(1, 0));
 }
 
-TEST_F(LocalSessionMinusAXTest, TestConcurrency) {
+TEST_F(DirectSessionMinusAXTest, TestConcurrency) {
   Initialize({1, 2, 3, 4});
   std::unique_ptr<Session> session(CreateSession());
   ASSERT_TRUE(session != nullptr);
@@ -147,7 +147,7 @@ TEST_F(LocalSessionMinusAXTest, TestConcurrency) {
   delete tp;
 }
 
-TEST_F(LocalSessionMinusAXTest, TwoCreateCallsFails) {
+TEST_F(DirectSessionMinusAXTest, TwoCreateCallsFails) {
   Initialize({1, 2, 3, 4});
   std::unique_ptr<Session> session(CreateSession());
   ASSERT_TRUE(session != nullptr);
@@ -157,7 +157,7 @@ TEST_F(LocalSessionMinusAXTest, TwoCreateCallsFails) {
   ASSERT_FALSE(session->Create(def_).ok());
 }
 
-TEST_F(LocalSessionMinusAXTest, ForgetToCreate) {
+TEST_F(DirectSessionMinusAXTest, ForgetToCreate) {
   Initialize({1, 2, 3, 4});
   std::unique_ptr<Session> session(CreateSession());
   ASSERT_TRUE(session != nullptr);
@@ -166,7 +166,7 @@ TEST_F(LocalSessionMinusAXTest, ForgetToCreate) {
   ASSERT_FALSE(session->Run(inputs, {y_ + ":0"}, {y_neg_}, &outputs).ok());
 }
 
-TEST_F(LocalSessionMinusAXTest, InvalidDevice) {
+TEST_F(DirectSessionMinusAXTest, InvalidDevice) {
   GraphDef def;
   Graph graph(OpRegistry::Global());
 
@@ -203,7 +203,7 @@ TEST_F(LocalSessionMinusAXTest, InvalidDevice) {
   ASSERT_OK(session->Run(inputs, output_names, {}, &outputs));
 }
 
-TEST(LocalSessionTest, KeepsStateAcrossRunsOfSession) {
+TEST(DirectSessionTest, KeepsStateAcrossRunsOfSession) {
   GraphDef def;
   Graph g(OpRegistry::Global());
   Node* var = test::graph::Var(&g, DT_FLOAT, TensorShape({10}));
@@ -242,7 +242,7 @@ TEST(LocalSessionTest, KeepsStateAcrossRunsOfSession) {
   EXPECT_EQ(20.0, outputs[0].flat<float>()(0));
 }
 
-TEST(LocalSessionTest, MultipleFeedTest) {
+TEST(DirectSessionTest, MultipleFeedTest) {
   GraphDef def;
   Graph g(OpRegistry::Global());
   Node* var = test::graph::Var(&g, DT_FLOAT, TensorShape({10}));

diff --git a/tensorflow/core/common_runtime/gpu/gpu_device.cc b/tensorflow/core/common_runtime/gpu/gpu_device.cc
@@ -558,11 +558,12 @@ LocalDevice* BaseGPUDeviceFactory::CreateGPUDevice(
           << " numa: " << desc.numa_node() << " pci: " << desc.pci_bus_id();
 
   ProcessState* process_state = ProcessState::singleton();
-  return CreateGPUDevice(
-      options, name, allocated_bytes, bus_adjacency, gpu_id,
-      GetShortDeviceDescription(gpu_id, desc),
-      process_state->GetGPUAllocator(gpu_id, allocated_memory),
-      process_state->GetCPUAllocator(desc.numa_node()));
+  return CreateGPUDevice(options, name, allocated_bytes, bus_adjacency, gpu_id,
+                         GetShortDeviceDescription(gpu_id, desc),
+                         process_state->GetGPUAllocator(
+                             gpu_id, allocated_memory,
+                             options.config.gpu_options().allocator_type()),
+                         process_state->GetCPUAllocator(desc.numa_node()));
 }
 
 static int GetMinGPUMultiprocessorCount() {