Change: 149865949

Fixes #8284
Change: 149862437

Change: 149827847

* Creates standalone BUILD target to register tfdbg's Copy* and Debug* ops: tensorflow/core:debug_ops_op_lib
* Makes the public cc BUILD target of tfdbg (tensorflow/core:debug) depend on it
* Through the new dependency path above, the public py BUILD target of tfdbg also obatins the registration of the Copy* and Debug* ops.
* Adds the new debug_ops_op_lib as a dependency of backwards_compatibility_test, so we can continue to track changes to the Copy* and Debug* ops.
* Removes the debug ops from the C and Java JNI library builds. These ops have always been listed as Experimental and were not documented. So this removal should be fine.
* In core/debug/BUILD, add check_deps to assert that tensorflow/core:tensorflow has no dependency on grpc.
Change: 149826469

Change: 149826365

Change: 149823781

Change: 149814250

Enables all names looking like __XXXX__. It also excludes all the symbols defined in "uninteresting" base classes such as Exception, type, object, etc., as well as everything
defined by protobuf, and methods (but not properties) defined by namedtuple.

Sorts methods to have __init__ and __new__ first.

Corrects links to _pb2.py files to point to .proto files instead.

Removes redundant separate pages for class and static methods.
Change: 149813108

Change: 149812908

Change: 149812873

std::set is slow and the iteration order is unstable. A couple other opportunistic changes include consolidating all called computations of an instruction in a single vector. This faciliates fast access to all called computations. Also, replace AddControlSuccessor/Predecessor with Add/RemoveControlDepedencyTo which is less error prone as you can't create a half connected control edge.
Change: 149810889

Benefits of this CL:
1) Removes ~70 python unit test files from the pip wheel.
2) Makes it easier to track which files are included in the pip wheel through bazel queries.
Change: 149810103

* tensor_dtype_regex_whitelist allows filtering by tensor data type during watch_graph() calls.
* tensor_dtype_regex_blacklist allows filtering (out) by tensor data type during watch_graph_with_blacklists() calls.
* Allow watch_fns to return a structured data type (class) WatchOptions for usability and extensibility. The legacy tuple return type (debug_ops, node_name_regex_whitelist, op_type_regex_whitelist) continues to be supported.
Change: 149808584

Change: 149807807

Change: 149803872

Change: 149803866

Change: 149802453

Change: 149800363

Change: 149800303

Change: 149796694

Modify the non-blocking threadpool to make it more flexible and less wasteful of CPU cycles for high-latency use-cases.

* Adds a hint to ThreadPool allowing us to turn off spin waiting. Currently each reader and record yielder op in a graph creates a threadpool with a thread that spins for 1000 iterations through the work stealing loop before yielding. This is wasteful for such ops that process I/O.

This takes advantage of the following additional changes to the Eigen non-blocking threadpool from https://bitbucket.org/eigen/eigen/pull-requests/303/make-the-non-blocking-threadpool-more/diff :

* Changes the number of iterations through the steal loop to be inversely proportional to the number of threads. Since the time of each iteration is proportional to the number of threads, this yields roughly a constant spin time.

* Implement a separate worker loop for the num_threads == 1 case since there is no point in going through the expensive steal loop. Moreover, since Steal() calls PopBack() on the victim queues it might reverse the order in which ops are executed, compared to the order in which they are scheduled, which is usually counter-productive for the types of I/O workloads the single thread pools tend to be used for.

* Store num_threads in a member variable for simplicity and to avoid a data race between the thread creation loop and worker threads calling threads_.size().
Change: 149787520

Previously, when num_threads==1, we put all the features for all the rows into a single shard and then only accumulated a single lhs/rhs pair, which produces the wrong result when there's more than one row in the input.

In practice, when num_threads==1 we would detect this inconsistency and CHECK fail at line 233, specifically
CHECK_EQ(input_index, get_input_index(i));
Change: 149786217

This is similar to MaxBytesInUse.

Also add "jingyue" to contrib/memory_stats/OWNERS.
Change: 149785681

* improve logging during native library loading to reduce chance of RunStats error message being misleading
* catch NPE for unfound ops during Graph loading to make bad node name issues easier to track down
* catch IOException locally in TensorFlowImageClassifier.create() to give better error message
* reduce camera preview size spam by adding all accepted/rejected entries on single lines
* move onPreviewSizeChosen callback outside of try block so that NPE errors don't get interpreted as camera2 problems
* correctly calculate initialization time
* print nodes being used for input/output on inference failure to assist in debugging
Change: 149783208

This is faster, conceptually simpler and avoids finicky issues with dtype
precision.
Change: 149782530

V2 Checkpoints are common, now the default from all TFX
pipelines.  Users of this tool, when filling the 'file_name' flag are
likely to submit an actual filename, rather than the required
prefix.  This error message should catch those cases where the user is
very likely to have meant the prefix.
Change: 149781964

CL/149672538 has made TensorHandles directly feedable for Session.run() calls. This eliminates the need to call TensorHandles.eval() to get the value of a Tensor and feed it back in during tfdbg NodeStepper's cont() (i.e., continue-to) calls.

This change improves the memory efficiency and performance of NodeStepper.cont() by eliminating unnecessary Tensor-numpy and numpy-Tensor copying.
Change: 149769853

Test using this classification signature using the Predict API.
Change: 149769547

Change: 149769532

initializes global variables.
Change: 149769121

Change: 149768595

Add some additional values args to name_scope calls.
Change: 149765769

Rename the new TryConcat()/TrySplit() to Concat()/Split() and mark the existing Concat()/Split() as deprecated.

This is the first change in a sequence that will remove the crashy Concat()/Split().
Change: 149755537

graph_editor: move match.py into the the test folder as it is not meant to be part of the graph_editor public API.
Change: 149741516

Change: 149732502

Change: 149720034

Change: 149714094

Change: 149706759

Add a sample program to evaluate clock cycle profiler and support clock cycle profiling on android 64bit
Change: 149704577