Merge changes from github.

PiperOrigin-RevId: 167401527

README.md

@@ -36,7 +36,15 @@ and discussion, and please direct specific questions to [Stack Overflow](https:/
 
 People who are a little more adventurous can also try our nightly binaries:
 
+**Nightly pip packages**
+* We are pleased to announce that TensorFlow now offers nightly pip packages
+under the [tf-nightly](https://pypi.python.org/pypi/tf-nightly) project on pypi.
+Simply run `pip install tf-nightly` in a clean environment to install the nightly
+tensorflow  build. We currently only support CPU-only packages on Linux and Mac.
+GPU packages on all platforms and Windows CPU-only packages will arrive soon!
 
+
+**Individual whl files**
 * Linux CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-1.3.0-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=cpu-slave)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-1.3.0-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=cpu-slave/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-1.3.0-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-python35-linux-cpu/))
 * Linux GPU: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-1.3.0-cp27-none-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=gpu-linux/)) / [Python 3.4](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-1.3.0-cp34-cp34m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=gpu-linux/)) / [Python 3.5](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow_gpu-1.3.0-cp35-cp35m-linux_x86_64.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3.5,label=gpu-linux/))
 * Mac CPU-only: [Python 2](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-1.3.0-py2-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON2,label=mac-slave/)) / [Python 3](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/lastSuccessfulBuild/artifact/pip_test/whl/tensorflow-1.3.0-py3-none-any.whl) ([build history](https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-cpu/TF_BUILD_IS_OPT=OPT,TF_BUILD_IS_PIP=PIP,TF_BUILD_PYTHON_VERSION=PYTHON3,label=mac-slave/))

RELEASE.md

@@ -46,7 +46,7 @@ See also [TensorBoard 0.1.4](https://github.com/tensorflow/tensorboard/releases/
   * Display feed values with the `print_feed` or `pf` command and clickable links in the curses UI.
   * Runtime profiler at the op level and the Python source line level with the `run -p` command.
 * Initial release of the statistical distribution library `tf.distributions`.
-* GPU kernels and speed improvements for for unary `tf.where` and `tf.nn.top_k`.
+* GPU kernels and speed improvements for unary `tf.where` and `tf.nn.top_k`.
 * Monotonic Attention wrappers added to `tf.contrib.seq2seq`.
 * Added `tf.contrib.signal`, a library for signal processing primitives.
 * Added `tf.contrib.resampler`, containing CPU and GPU ops for differentiable resampling of images.

WORKSPACE

@@ -2,11 +2,11 @@ workspace(name = "org_tensorflow")
 
 http_archive(
     name = "io_bazel_rules_closure",
-    sha256 = "bc41b80486413aaa551860fc37471dbc0666e1dbb5236fb6177cb83b0c105846",
-    strip_prefix = "rules_closure-dec425a4ff3faf09a56c85d082e4eed05d8ce38f",
+    sha256 = "25f5399f18d8bf9ce435f85c6bbf671ec4820bc4396b3022cc5dc4bc66303609",
+    strip_prefix = "rules_closure-0.4.2",
     urls = [
-        "http://mirror.bazel.build/github.com/bazelbuild/rules_closure/archive/dec425a4ff3faf09a56c85d082e4eed05d8ce38f.tar.gz",  # 2017-06-02
-        "https://github.com/bazelbuild/rules_closure/archive/dec425a4ff3faf09a56c85d082e4eed05d8ce38f.tar.gz",
+        "http://mirror.bazel.build/github.com/bazelbuild/rules_closure/archive/0.4.2.tar.gz",  # 2017-08-29
+        "https://github.com/bazelbuild/rules_closure/archive/0.4.2.tar.gz",
     ],
 )
 

configure.py

@@ -143,7 +143,7 @@ def run_shell(cmd, allow_non_zero=False):
 
 def cygpath(path):
   """Convert path from posix to windows."""
-  return run_shell(['cygpath', '-m', path])
+  return os.path.abspath(path).replace('\\', '/')
 
 
 def get_python_path(environ_cp, python_bin_path):
@@ -196,7 +196,7 @@ def setup_python(environ_cp, bazel_version):
     environ_cp['PYTHON_BIN_PATH'] = ''
 
   # Convert python path to Windows style before checking lib and version
-  if is_cygwin():
+  if is_windows() or is_cygwin():
     python_bin_path = cygpath(python_bin_path)
 
   # Get PYTHON_LIB_PATH
@@ -219,7 +219,7 @@ def setup_python(environ_cp, bazel_version):
   python_major_version = get_python_major_version(python_bin_path)
 
   # Convert python path to Windows style before writing into bazel.rc
-  if is_cygwin():
+  if is_windows() or is_cygwin():
     python_lib_path = cygpath(python_lib_path)
 
   # Set-up env variables used by python_configure.bzl
@@ -600,7 +600,7 @@ def set_tf_cuda_version(environ_cp):
 
     # Find out where the CUDA toolkit is installed
     default_cuda_path = _DEFAULT_CUDA_PATH
-    if is_cygwin():
+    if is_windows() or is_cygwin():
       default_cuda_path = cygpath(
           environ_cp.get('CUDA_PATH', _DEFAULT_CUDA_PATH_WIN))
     elif is_linux():
@@ -660,7 +660,7 @@ def set_tf_cunn_version(environ_cp):
     # unusable. Going through one more level of expansion to handle that.
     cudnn_install_path = os.path.realpath(
         os.path.expanduser(cudnn_install_path))
-    if is_cygwin():
+    if is_windows() or is_cygwin():
       cudnn_install_path = cygpath(cudnn_install_path)
 
     if is_windows():

tensorflow/cc/gradients/nn_grad.cc

@@ -95,6 +95,14 @@ Status SeluGradHelper(const Scope& scope, const Operation& op,
 }
 REGISTER_GRADIENT_OP("Selu", SeluGradHelper);
 
+Status L2LossGrad(const Scope& scope, const Operation& op,
+                  const std::vector<Output>& grad_inputs,
+                  std::vector<Output>* grad_outputs) {
+  grad_outputs->push_back(Mul(scope, op.input(0), grad_inputs[0]));
+  return scope.status();
+}
+REGISTER_GRADIENT_OP("L2Loss", L2LossGrad);
+
 Status BiasAddGradHelper(const Scope& scope, const Operation& op,
                          const std::vector<Output>& grad_inputs,
                          std::vector<Output>* grad_outputs) {

tensorflow/cc/gradients/nn_grad_test.cc

@@ -122,6 +122,14 @@ TEST_F(NNGradTest, SeluGrad) {
   RunTest(x, x_init_value, y, shape);
 }
 
+TEST_F(NNGradTest, L2LossGrad) {
+  TensorShape x_shape({5, 2});
+  TensorShape y_shape({1});
+  auto x = Placeholder(scope_, DT_FLOAT, Placeholder::Shape(x_shape));
+  auto y = L2Loss(scope_, x);
+  RunTest(x, x_shape, y, y_shape);
+}
+
 TEST_F(NNGradTest, BiasAddGradHelper) {
   TensorShape shape({4, 5});
   TensorShape bias_shape({5});

tensorflow/compiler/xla/service/gpu/llvm_gpu_backend/gpu_backend_lib.cc

@@ -389,7 +389,7 @@ StatusOr<string> CompileModuleToPtx(llvm::Module* module,
 
   // Loop unrolling exposes more opportunities for SROA. Therefore, we run SROA
   // again after the standard optimization passes [http://b/13329423].
-  // TODO(jingyue): SROA may further expose more optimization opportunities, such
+  // TODO(jingyue): SROA may further expose more optimization opportunities such
   // as more precise alias analysis and more function inlining (SROA may change
   // the inlining cost of a function). For now, running SROA already emits good
   // enough code for the evaluated benchmarks. We may want to run more

tensorflow/contrib/cmake/tf_tests.cmake

@@ -82,6 +82,7 @@ function(AddTest)
   set_tests_properties(${_AT_TARGET}
     PROPERTIES ENVIRONMENT "TEST_TMPDIR=${tempdir};TEST_SRCDIR=${testdir}"
   )
+  set_tests_properties(${_AT_TARGET} PROPERTIES TIMEOUT "600")
 
   foreach(datafile ${_AT_DATA})
     file(RELATIVE_PATH datafile_rel ${tensorflow_source_dir} ${datafile})
@@ -117,6 +118,7 @@ function(AddPythonTests)
     if (_AT_DEPENDS)
       add_dependencies(${_AT_TARGET} ${_AT_DEPENDS})
     endif()
+    set_tests_properties(${sourcefile} PROPERTIES TIMEOUT "600")
   endforeach()
 endfunction(AddPythonTests)
 

tensorflow/contrib/gdr/BUILD

@@ -62,6 +62,7 @@ tf_cuda_library(
     }),
     deps = [
         ":gdr_proto_cc",
+        "//tensorflow/core:core_cpu_internal",
         "//tensorflow/core:framework",
         "//tensorflow/core:gpu_runtime",
         "//tensorflow/core:lib",

tensorflow/contrib/gdr/gdr_memory_manager.h

@@ -16,14 +16,9 @@ limitations under the License.
 #ifndef GDR_MEMORY_MANAGER_H_
 #define GDR_MEMORY_MANAGER_H_
 
+#include "google/protobuf/any.pb.h"
 #include "tensorflow/core/lib/core/status.h"
 
-namespace google {
-namespace protobuf {
-class Any;
-}
-}
-
 namespace tensorflow {
 
 class Device;

tensorflow/contrib/layers/__init__.py

@@ -115,6 +115,7 @@
                     'legacy_linear',
                     'legacy_relu',
                     'OPTIMIZER_CLS_NAMES',
+                    'OPTIMIZER_SUMMARIES',
                     'regression_target',
                     'SPARSE_FEATURE_CROSS_DEFAULT_HASH_KEY',
                     'summaries']

tensorflow/contrib/layers/python/layers/optimizers.py

@@ -129,8 +129,9 @@ class should be sub-class of `tf.Optimizer` that implements
                `None` to use all trainable variables.
     name: The name for this operation is used to scope operations and summaries.
     summaries: List of internal quantities to visualize on tensorboard. If not
-               set only the loss and the learning rate will be reported. The
-               complete list is in OPTIMIZER_SUMMARIES.
+               set, the loss, the learning rate, and the global norm of the
+               gradients will be reported. The complete list of possible values
+               is in OPTIMIZER_SUMMARIES.
     colocate_gradients_with_ops: If True, try colocating gradients with the
                                  corresponding op.
     increment_global_step: Whether to increment `global_step`. If your model

tensorflow/contrib/learn/python/learn/learn_io/data_feeder.py

@@ -28,7 +28,6 @@
 from six.moves import xrange  # pylint: disable=redefined-builtin
 
 from tensorflow.python.framework import dtypes
-from tensorflow.python.framework import ops
 from tensorflow.python.ops import array_ops
 from tensorflow.python.platform import tf_logging as logging
 
@@ -44,7 +43,7 @@ def _get_in_out_shape(x_shape, y_shape, n_classes, batch_size=None):
   x_is_dict, y_is_dict = isinstance(
       x_shape, dict), y_shape is not None and isinstance(y_shape, dict)
   if y_is_dict and n_classes is not None:
-    assert (isinstance(n_classes, dict))
+    assert isinstance(n_classes, dict)
 
   if batch_size is None:
     batch_size = list(x_shape.values())[0][0] if x_is_dict else x_shape[0]
@@ -322,10 +321,12 @@ def __init__(self,
 
     self._x = dict([(k, check_array(v, v.dtype)) for k, v in list(x.items())
                    ]) if x_is_dict else check_array(x, x.dtype)
-    self._y = None if y is None else \
-      dict([(k, check_array(v, v.dtype)) for k, v in list(y.items())]) if x_is_dict else check_array(y, y.dtype)
+    self._y = None if y is None else (
+        dict([(k, check_array(v, v.dtype)) for k, v in list(y.items())])
+        if y_is_dict else check_array(y, y.dtype))
 
-    # self.n_classes is not None means we're converting raw target indices to one-hot.
+    # self.n_classes is not None means we're converting raw target indices
+    # to one-hot.
     if n_classes is not None:
       if not y_is_dict:
         y_dtype = (np.int64
@@ -344,12 +345,15 @@ def __init__(self,
         x_shape, y_shape, n_classes, batch_size)
 
     # Input dtype matches dtype of x.
-    self._input_dtype = dict([(k, _check_dtype(v.dtype)) for k, v in list(self._x.items())]) if x_is_dict \
-      else _check_dtype(self._x.dtype)
-
-    # note: self._output_dtype = np.float32 when y is None
-    self._output_dtype = dict([(k, _check_dtype(v.dtype)) for k, v in list(self._y.items())]) if y_is_dict \
-      else _check_dtype(self._y.dtype) if y is not None else np.float32
+    self._input_dtype = (
+        dict([(k, _check_dtype(v.dtype)) for k, v in list(self._x.items())])
+        if x_is_dict else _check_dtype(self._x.dtype))
+
+    # self._output_dtype == np.float32 when y is None
+    self._output_dtype = (
+        dict([(k, _check_dtype(v.dtype)) for k, v in list(self._y.items())])
+        if y_is_dict else (
+            _check_dtype(self._y.dtype) if y is not None else np.float32))
 
     # self.n_classes is None means we're passing in raw target indices
     if n_classes is not None and y_is_dict:

tensorflow/contrib/makefile/Makefile

@@ -316,14 +316,14 @@ ifeq ($(TARGET),IOS)
 	IPHONESIMULATOR_SYSROOT := $(shell xcrun --sdk iphonesimulator \
 	--show-sdk-path)
 	IOS_SDK_VERSION := $(shell xcrun --sdk iphoneos --show-sdk-version)
-	MIN_SDK_VERSION := 8.0
+	MIN_SDK_VERSION := 9.0
 # Override IOS_ARCH with ARMV7, ARMV7S, ARM64, or I386.
 	IOS_ARCH := X86_64
 	ifeq ($(IOS_ARCH),ARMV7)
 		CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \
 		-arch armv7 \
 		-fembed-bitcode \
-		-D__thread= \
+		-D__thread=thread_local \
 		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-mno-thumb \
@@ -347,7 +347,7 @@ ifeq ($(TARGET),IOS)
 		CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \
 		-arch armv7s \
 		-fembed-bitcode \
-		-D__thread= \
+		-D__thread=thread_local \
 		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-mno-thumb \
@@ -371,7 +371,7 @@ ifeq ($(TARGET),IOS)
 		CXXFLAGS += -miphoneos-version-min=$(MIN_SDK_VERSION) \
 		-arch arm64 \
 		-fembed-bitcode \
-		-D__thread= \
+		-D__thread=thread_local \
 		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-DTF_LEAN_BINARY \
@@ -395,7 +395,7 @@ ifeq ($(TARGET),IOS)
 		-arch i386 \
 		-mno-sse \
 		-fembed-bitcode \
-		-D__thread= \
+		-D__thread=thread_local \
 		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-DTF_LEAN_BINARY \
@@ -418,7 +418,7 @@ ifeq ($(TARGET),IOS)
 		CXXFLAGS += -mios-simulator-version-min=$(MIN_SDK_VERSION) \
 		-arch x86_64 \
 		-fembed-bitcode \
-		-D__thread= \
+		-D__thread=thread_local \
 		-DUSE_GEMM_FOR_CONV \
 		-Wno-c++11-narrowing \
 		-DTF_LEAN_BINARY \

tensorflow/contrib/makefile/README.md

@@ -201,7 +201,8 @@ tensorflow/contrib/makefile/compile_ios_protobuf.sh
 
 Then, you will need to compile the nsync library for iOS:
 
-```export HOST_NSYNC_LIB=`tensorflow/contrib/makefile/compile_nsync.sh`
+```bash
+export HOST_NSYNC_LIB=`tensorflow/contrib/makefile/compile_nsync.sh`
 export TARGET_NSYNC_LIB=`tensorflow/contrib/makefile/compile_nsync.sh -t ios`
 ```
 

tensorflow/core/distributed_runtime/base_rendezvous_mgr.cc

@@ -35,14 +35,18 @@ limitations under the License.
 
 namespace tensorflow {
 
+static void StartAbortRendevous(Rendezvous* rendez, const Status& s) {
+  rendez->StartAbort(s);
+  rendez->Unref();
+}
+
 BaseRendezvousMgr::BaseRendezvousMgr(const WorkerEnv* worker_env)
     : worker_env_(worker_env) {}
 
 BaseRendezvousMgr::~BaseRendezvousMgr() {
   for (auto& p : table_) {
-    BaseRemoteRendezvous* rendez = p.second;
-    rendez->StartAbort(errors::Aborted("Shutdown"));
-    rendez->Unref();
+    auto rendez = p.second;
+    StartAbortRendevous(rendez, errors::Aborted("Shutdown"));
   }
 }
 
@@ -52,7 +56,7 @@ RemoteRendezvous* BaseRendezvousMgr::Find(int64 step_id) {
 
 BaseRemoteRendezvous* BaseRendezvousMgr::FindOrCreate(int64 step_id) {
   mutex_lock l(mu_);
-  Table::iterator iter = table_.find(step_id);
+  auto iter = table_.find(step_id);
   if (iter == table_.end()) {
     auto rr = Create(step_id, worker_env_);
     iter = table_.insert({step_id, rr}).first;
@@ -64,7 +68,7 @@ BaseRemoteRendezvous* BaseRendezvousMgr::FindOrCreate(int64 step_id) {
 void BaseRendezvousMgr::RecvLocalAsync(int64 step_id,
                                        const Rendezvous::ParsedKey& parsed,
                                        Rendezvous::DoneCallback done) {
-  BaseRemoteRendezvous* rendez = FindOrCreate(step_id);
+  auto rendez = FindOrCreate(step_id);
   using namespace std::placeholders;
   Rendezvous::DoneCallback done_cb = std::bind(
       [rendez](Rendezvous::DoneCallback done,
@@ -101,15 +105,15 @@ void BaseRendezvousMgr::Cleanup(int64 step_id) {
   Rendezvous* rendez = nullptr;
   {
     mutex_lock l(mu_);
-    Table::iterator iter = table_.find(step_id);
+    auto iter = table_.find(step_id);
     if (iter != table_.end()) {
       rendez = iter->second;
       table_.erase(iter);
     }
   }
-  if (!rendez) return;
-  rendez->StartAbort(errors::Aborted("Cleanup ", step_id));
-  rendez->Unref();
+  if (rendez) {
+    StartAbortRendevous(rendez, errors::Aborted("Cleanup ", step_id));
+  }
 }
 
 void BaseRendezvousMgr::CleanupAll() {
@@ -122,8 +126,7 @@ void BaseRendezvousMgr::CleanupAll() {
     table_.clear();
   }
   for (auto rendez : rendezs) {
-    rendez->StartAbort(errors::Aborted("Shutdown"));
-    rendez->Unref();
+    StartAbortRendevous(rendez, errors::Aborted("Shutdown"));
   }
 }
 
@@ -165,7 +168,7 @@ Status BaseRemoteRendezvous::Initialize(WorkerSession* session) {
     session_ = session;
     std::swap(deferred_calls, deferred_calls_);
   }
-  for (DeferredCall& call : deferred_calls) {
+  for (auto& call : deferred_calls) {
     RecvLocalAsyncInternal(call.parsed, std::move(call.done));
   }
   return Status::OK();

tensorflow/core/framework/op_kernel.h

@@ -310,7 +310,7 @@ class OpKernelConstruction {
   FunctionLibraryRuntime* function_library() const { return flib_; }
 
   // The GraphDef version whose behavior we should follow.
-  const int graph_def_version() const { return graph_def_version_; }
+  int graph_def_version() const { return graph_def_version_; }
 
   // Helper routines for the OP_REQUIRES macros
   void CtxFailure(Status s);