platform/upstream/tensorflow.git
6 years agoSilence compilation warning in allocator.h
Derek Murray [Fri, 23 Mar 2018 17:12:05 +0000 (10:12 -0700)]
Silence compilation warning in allocator.h

Some builds report the following warning (even though LOG(FATAL) is terminal):

```
./tensorflow/core/framework/device_base.h(188): warning: missing return statement at end of non-void function "tensorflow::DeviceBase::GetScopedAllocator"
```

PiperOrigin-RevId: 190235863

6 years agoMerged commit includes the following changes:
A. Unique TensorFlower [Fri, 23 Mar 2018 16:15:58 +0000 (09:15 -0700)]
Merged commit includes the following changes:
190228434  by A. Unique TensorFlower:

    internal change only

--
190215102  by A. Unique TensorFlower:

    Remove placement strategy from Fisher estimator and add it in as Mixin.
    A new Kfac Optimizer sub class which runs covariance and inverse ops periodically. This should deprecate lazykfac.

--
190211603  by A. Unique TensorFlower:

    Internal only

--

PiperOrigin-RevId: 190228434

6 years agoFix HloSharding::ToString to be compatible with the HLO text parser
A. Unique TensorFlower [Fri, 23 Mar 2018 12:43:13 +0000 (05:43 -0700)]
Fix HloSharding::ToString to be compatible with the HLO text parser

The incompatibility made it impossible to parse HLO modules with
sharding attributes printed from a previous run of XLA. The previous
format generated by this function was ambigious so changing the HLO
parser only wouldn't be feasible.

PiperOrigin-RevId: 190208069

6 years agoUse int32 GPU kernel for StopGradient
A. Unique TensorFlower [Fri, 23 Mar 2018 11:41:47 +0000 (04:41 -0700)]
Use int32 GPU kernel for StopGradient

PiperOrigin-RevId: 190204238

6 years agoImplement operator<< for HloSharding
A. Unique TensorFlower [Fri, 23 Mar 2018 10:55:03 +0000 (03:55 -0700)]
Implement operator<< for HloSharding

The new operator makes the error messages coming from gtest more
readable as well as making logging easier.

PiperOrigin-RevId: 190200926

6 years agoInternal change
A. Unique TensorFlower [Fri, 23 Mar 2018 10:37:05 +0000 (03:37 -0700)]
Internal change

PiperOrigin-RevId: 190199534

6 years agoEnable fusion of element-wise ops using XLA (Off by default, can be enabled by settin...
A. Unique TensorFlower [Fri, 23 Mar 2018 06:25:37 +0000 (23:25 -0700)]
Enable fusion of element-wise ops using XLA (Off by default, can be enabled by setting the flag tf_xla_fusion_only=true).

PiperOrigin-RevId: 190179459

6 years agoAdd cast op support for bfloat16 in CPU.
A. Unique TensorFlower [Fri, 23 Mar 2018 04:46:31 +0000 (21:46 -0700)]
Add cast op support for bfloat16 in CPU.

PiperOrigin-RevId: 190173501

6 years agoAdded experimental C APIs to build a hard-coded stack of dataset + iterator.
Mingsheng Hong [Fri, 23 Mar 2018 03:51:51 +0000 (20:51 -0700)]
Added experimental C APIs to build a hard-coded stack of dataset + iterator.

PiperOrigin-RevId: 190170332

6 years agoImprovements to quantization matching code:
Suharsh Sivakumar [Fri, 23 Mar 2018 03:40:55 +0000 (20:40 -0700)]
Improvements to quantization matching code:

- Guarantee that we do the largest matches first.
- Don't allow matching layers multiple times.
- Don't allow adding quantization ops to the same node multiple times.
- Return a list of match results rather than yielding. This is much easier to reason about.
- Only require ReadVariableOp when matching resource variables, since the input to ReadVariableOps don't necessarily have to be a VarHandleOp.
- Place post activation bypass ops quantization nodes in the same post activation bypass op's name scope for better viewing.

PiperOrigin-RevId: 190169622

6 years agoRollback of GCS change.
Jonathan Hseu [Fri, 23 Mar 2018 03:32:03 +0000 (20:32 -0700)]
Rollback of GCS change.
END_PUBLIC

BEGIN_PUBLIC
Automated g4 rollback of changelist 189900977

PiperOrigin-RevId: 190169076

6 years agoCheckpointable: Add a way for subclasses to manage deferred restorations of condition...
Allen Lavoie [Fri, 23 Mar 2018 03:02:45 +0000 (20:02 -0700)]
Checkpointable: Add a way for subclasses to manage deferred restorations of conditional dependencies.
PiperOrigin-RevId: 190166571

6 years agoRefactor keras.Sequential and enable building Sequential models without specifying...
Francois Chollet [Fri, 23 Mar 2018 01:59:48 +0000 (18:59 -0700)]
Refactor keras.Sequential and enable building Sequential models without specifying an input shape in advance (the model gets built on fit/etc, like subclassed models).

PiperOrigin-RevId: 190161771

6 years agoMerge changes from github.
Benoit Steiner [Fri, 23 Mar 2018 01:54:27 +0000 (18:54 -0700)]
Merge changes from github.

PiperOrigin-RevId: 190161440

6 years ago[tf.data] Add `tf.contrib.data.prefetch_to_device()`, which supports prefetching...
Derek Murray [Fri, 23 Mar 2018 01:20:09 +0000 (18:20 -0700)]
[tf.data] Add `tf.contrib.data.prefetch_to_device()`, which supports prefetching to GPU memory.

PiperOrigin-RevId: 190158272

6 years agoAdd asserts for LinearOperatorBlockDiag.
A. Unique TensorFlower [Fri, 23 Mar 2018 01:05:03 +0000 (18:05 -0700)]
Add asserts for LinearOperatorBlockDiag.

All asserts are only dependent on the constituent operators due to properties of the direct sum.

PiperOrigin-RevId: 190156830

6 years agoConvert saved_model to tflite flatbuffer.
Zhixian Yan [Fri, 23 Mar 2018 00:59:19 +0000 (17:59 -0700)]
Convert saved_model to tflite flatbuffer.

PiperOrigin-RevId: 190156133

6 years agoDisable test tensorflow/contrib/linalg:linear_operator_block_diag_test in asan mode
A. Unique TensorFlower [Fri, 23 Mar 2018 00:45:03 +0000 (17:45 -0700)]
Disable test tensorflow/contrib/linalg:linear_operator_block_diag_test in asan mode

PiperOrigin-RevId: 190154645

6 years agoCheckpointable: Fix a bug where SaveableObjects in the Saver's var_list were stale...
Allen Lavoie [Fri, 23 Mar 2018 00:29:47 +0000 (17:29 -0700)]
Checkpointable: Fix a bug where SaveableObjects in the Saver's var_list were stale when executing eagerly

SaveableObjects were recreated each save(), but the Saver was not recreated and so had an old var_list.

PiperOrigin-RevId: 190153011

6 years agoDelete all the name_scopes in the TPU code so that tf.Variable() doesn't inherit...
Jonathan Hseu [Thu, 22 Mar 2018 23:42:49 +0000 (16:42 -0700)]
Delete all the name_scopes in the TPU code so that tf.Variable() doesn't inherit them.

PiperOrigin-RevId: 190146946

6 years agoRemove cached tensors from LinearOperator.
A. Unique TensorFlower [Thu, 22 Mar 2018 23:05:22 +0000 (16:05 -0700)]
Remove cached tensors from LinearOperator.

CSE should already ensure that tensors are deduped (i.e. multiple Cholesky's for the same matrix are not performed).

This also makes it easier to use LinearOperators in tf.while_loop.

PiperOrigin-RevId: 190141732

6 years agoAutomated g4 rollback of changelist 190127730
Mark Heffernan [Thu, 22 Mar 2018 22:49:57 +0000 (15:49 -0700)]
Automated g4 rollback of changelist 190127730

PiperOrigin-RevId: 190139303

6 years agoGo: Update generated wrapper functions for TensorFlow ops.
A. Unique TensorFlower [Thu, 22 Mar 2018 22:46:54 +0000 (15:46 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId: 190138848

6 years agoInstead of depending on ctest to be in PATH, directly reference the binary.
Gunhan Gulsoy [Thu, 22 Mar 2018 22:36:27 +0000 (15:36 -0700)]
Instead of depending on ctest to be in PATH, directly reference the binary.

PiperOrigin-RevId: 190137278

6 years agoDisable the test tensorflow/core/grappler/optimizers:memory_optimizer_test for cuda_o...
A. Unique TensorFlower [Thu, 22 Mar 2018 22:32:04 +0000 (15:32 -0700)]
Disable the test tensorflow/core/grappler/optimizers:memory_optimizer_test for cuda_on_cpu

PiperOrigin-RevId: 190136546

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Thu, 22 Mar 2018 22:19:02 +0000 (15:19 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 190134484

6 years agoExplicitly disallow returning variables from a tfe.py_func.
Akshay Agrawal [Thu, 22 Mar 2018 22:18:25 +0000 (15:18 -0700)]
Explicitly disallow returning variables from a tfe.py_func.

PiperOrigin-RevId: 190134382

6 years agoAdd kwargs support to wrap_py_func. Also includes some fixes to the reference tests...
A. Unique TensorFlower [Thu, 22 Mar 2018 22:14:40 +0000 (15:14 -0700)]
Add kwargs support to wrap_py_func. Also includes some fixes to the reference tests and an actual test for wrapping NumPy function that uses kwargs.

PiperOrigin-RevId: 190133811

6 years agoMove cuDNN RNN ops to core, for use in the internal TF codebase only (not publicly...
Martin Wicke [Thu, 22 Mar 2018 21:53:59 +0000 (14:53 -0700)]
Move cuDNN RNN ops to core, for use in the internal TF codebase only (not publicly exposed).

RELNOTES: Moved cuDNN RNN ops to core.
PiperOrigin-RevId: 190130405

6 years agoAutomated g4 rollback of changelist 190021164
Alexandre Passos [Thu, 22 Mar 2018 21:47:22 +0000 (14:47 -0700)]
Automated g4 rollback of changelist 190021164

PiperOrigin-RevId: 190129094

6 years agoMerge consecutive broadcast HLO instructions.
Dimitris Vardoulakis [Thu, 22 Mar 2018 21:38:41 +0000 (14:38 -0700)]
Merge consecutive broadcast HLO instructions.

As an optimization, replace consecutive broadcast instructions with a single equivalent broadcast in algebraic simplification.

PiperOrigin-RevId: 190127730

6 years agoRemove use of deprecated API from RNN Colorbot example.
Akshay Agrawal [Thu, 22 Mar 2018 21:24:23 +0000 (14:24 -0700)]
Remove use of deprecated API from RNN Colorbot example.

PiperOrigin-RevId: 190125356

6 years agoUpdate tf.keras preprocessing to Keras 2.1.5 API
Pavithra Vijay [Thu, 22 Mar 2018 21:14:33 +0000 (14:14 -0700)]
Update tf.keras preprocessing to Keras 2.1.5 API

PiperOrigin-RevId: 190123773

6 years agoAdds float64 support for Conv2d, Conv2dBackpropInput, and Conv2dBackpropFilter
Brian Patton [Thu, 22 Mar 2018 21:11:08 +0000 (14:11 -0700)]
Adds float64 support for Conv2d, Conv2dBackpropInput, and Conv2dBackpropFilter

PiperOrigin-RevId: 190123191

6 years agoFetch C shapes for ops created by import_graph_def with C API enabled.
Skye Wanderman-Milne [Thu, 22 Mar 2018 21:09:59 +0000 (14:09 -0700)]
Fetch C shapes for ops created by import_graph_def with C API enabled.

If _USE_C_API = True, this change makes us always fetch shapes using
the C API after calling TF_ImportGraphDef, even if _USE_C_SHAPES =
False. This is necessary to preserve the shapes specified by the
"_output_shapes" attr on imported NodeDefs (note that this attr isn't
present on the NodeDefs of the imported nodes, so there's no other way
to recover this information after calling TF_ImportGraphDef).

PiperOrigin-RevId: 190122991

6 years agoSupports optimizer arg in head.create_estimator_spec.
A. Unique TensorFlower [Thu, 22 Mar 2018 21:01:10 +0000 (14:01 -0700)]
Supports optimizer arg in head.create_estimator_spec.

PiperOrigin-RevId: 190121386

6 years agoFix cases where we export incorrect symbol with tf_export. This can happen when
Anna R [Thu, 22 Mar 2018 20:55:48 +0000 (13:55 -0700)]
Fix cases where we export incorrect symbol with tf_export. This can happen when
both generated op and its python wrapper have tf_export decorator.
create_python_api.py now checks that we don't export different symbols with same name. Also, simplified some logic.

PiperOrigin-RevId: 190120505

6 years agoCode cleanup: rather than storing the outside_compilation shape inference graph as...
A. Unique TensorFlower [Thu, 22 Mar 2018 20:41:13 +0000 (13:41 -0700)]
Code cleanup: rather than storing the outside_compilation shape inference graph as a serialized GraphDef in an attr, put it into the function library.

PiperOrigin-RevId: 190118116

6 years agoDisable all the automatic optimizations when testing, to ensure that we can
Benoit Steiner [Thu, 22 Mar 2018 20:24:51 +0000 (13:24 -0700)]
Disable all the automatic optimizations when testing, to ensure that we can
properly compare the results of the original graph against that of the hand
optimized graph.

PiperOrigin-RevId: 190115606

6 years agoAllow specifying in the arrays extra info file:
A. Unique TensorFlower [Thu, 22 Mar 2018 20:21:57 +0000 (13:21 -0700)]
Allow specifying in the arrays extra info file:
 - the shape of the array
 - the hardcoding of the values of the array as a single repeated constant
   scalar value, turning an activations array into a constant array.

PiperOrigin-RevId: 190115218

6 years agoDisable testing flaky tensorflow/contrib/eager/python/examples/spinn:spinn_test under py3
A. Unique TensorFlower [Thu, 22 Mar 2018 20:05:30 +0000 (13:05 -0700)]
Disable testing flaky tensorflow/contrib/eager/python/examples/spinn:spinn_test under py3

PiperOrigin-RevId: 190112748

6 years agoMake api_compatibility_test output verbose by default.
Martin Wicke [Thu, 22 Mar 2018 19:52:25 +0000 (12:52 -0700)]
Make api_compatibility_test output verbose by default.

PiperOrigin-RevId: 190110866

6 years agoAdd tf.contrib.framework.argsort, wrapping tf.nn.top_k (#288).
Dan Ringwalt [Thu, 22 Mar 2018 19:45:24 +0000 (12:45 -0700)]
Add tf.contrib.framework.argsort, wrapping tf.nn.top_k (#288).

Comparable to np.argsort.

PiperOrigin-RevId: 190109968

6 years agoMerges predict export_outputs in multi_head.
A. Unique TensorFlower [Thu, 22 Mar 2018 19:34:02 +0000 (12:34 -0700)]
Merges predict export_outputs in multi_head.

PiperOrigin-RevId: 190108434

6 years agoMake GetLocalTemporaryDirectories() a virtual method of Env, that is overriden
A. Unique TensorFlower [Thu, 22 Mar 2018 18:55:34 +0000 (11:55 -0700)]
Make GetLocalTemporaryDirectories() a virtual method of Env, that is overriden
by the implementations.

PiperOrigin-RevId: 190102851

6 years agoGo: Update generated wrapper functions for TensorFlow ops.
A. Unique TensorFlower [Thu, 22 Mar 2018 18:55:11 +0000 (11:55 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId: 190102805

6 years agoCleanup: replace an errant `in_eager_mode()` with `executing_eagerly()`.
Akshay Agrawal [Thu, 22 Mar 2018 18:29:25 +0000 (11:29 -0700)]
Cleanup: replace an errant `in_eager_mode()` with `executing_eagerly()`.

PiperOrigin-RevId: 190098277

6 years agoAdd new Ops for ScopedAllocator and the associated Concat and Split. The
Ayush Dubey [Thu, 22 Mar 2018 18:25:58 +0000 (11:25 -0700)]
Add new Ops for ScopedAllocator and the associated Concat and Split.  The
ScopedAllocatorOp allocates a large backing tensor whose slices may be
concatenated or splitted with ScopedAllocatorConcatOp and
ScopedAllocatorSplitOp respectively.  These ops should only be added via
Grappler optimizations on the dataflow graph provided by the user.

PiperOrigin-RevId: 190097586

6 years agoSupports PReLU in TFLite & Toco.
Yu-Cheng Ling [Thu, 22 Mar 2018 18:25:49 +0000 (11:25 -0700)]
Supports PReLU in TFLite & Toco.

PiperOrigin-RevId: 190097557

6 years agoRun the grappler optimizer tests both on GPU and CPU
Benoit Steiner [Thu, 22 Mar 2018 18:23:26 +0000 (11:23 -0700)]
Run the grappler optimizer tests both on GPU and CPU

PiperOrigin-RevId: 190097168

6 years agoEager/g3doc: Gradients with respect to constants are None and not 0.
Asim Shankar [Thu, 22 Mar 2018 18:21:52 +0000 (11:21 -0700)]
Eager/g3doc: Gradients with respect to constants are None and not 0.

Same behavior as tf.gradients() for graphs.
Some discussion of this choice in #783

PiperOrigin-RevId: 190096919

6 years agoAdds remaining validations in sequence_numeric_column.
A. Unique TensorFlower [Thu, 22 Mar 2018 18:12:10 +0000 (11:12 -0700)]
Adds remaining validations in sequence_numeric_column.

PiperOrigin-RevId: 190094883

6 years agoAutomated g4 rollback of changelist 189819449
A. Unique TensorFlower [Thu, 22 Mar 2018 17:28:56 +0000 (10:28 -0700)]
Automated g4 rollback of changelist 189819449

PiperOrigin-RevId: 190087074

6 years agoSmall cleanup StrCat related number formatting
A. Unique TensorFlower [Thu, 22 Mar 2018 17:21:59 +0000 (10:21 -0700)]
Small cleanup StrCat related number formatting

- Resolve inconsistency in return values (pointer to start vs end of buffer)
- Instead, return the number of chars written as this turns out to be most useful to callers
- Removes the need for redundant strlen calls.

PiperOrigin-RevId: 190085812

6 years agoMark tensor as const in a function that does not mutate a tensor.
A. Unique TensorFlower [Thu, 22 Mar 2018 17:21:39 +0000 (10:21 -0700)]
Mark tensor as const in a function that does not mutate a tensor.

PiperOrigin-RevId: 190085757

6 years agoOmit tensorflow/python/estimator:replicate_model_fn_test from asan builds.
A. Unique TensorFlower [Thu, 22 Mar 2018 17:17:18 +0000 (10:17 -0700)]
Omit tensorflow/python/estimator:replicate_model_fn_test from asan builds.
It gets flaky timeouts.

PiperOrigin-RevId: 190084932

6 years agoTemplate system improvements: allow replacing keyword args. Allow using function...
A. Unique TensorFlower [Thu, 22 Mar 2018 17:10:18 +0000 (10:10 -0700)]
Template system improvements: allow replacing keyword args. Allow using function calls and dicts in name replacements.

PiperOrigin-RevId: 190083700

6 years agoSimplifying "is" and "is not" dispatch
A. Unique TensorFlower [Thu, 22 Mar 2018 16:43:32 +0000 (09:43 -0700)]
Simplifying "is" and "is not" dispatch

PiperOrigin-RevId: 190078959

6 years ago[XLA] Only overwrite the hlo_profiling flag when it's not enabled by default.
Benjamin Kramer [Thu, 22 Mar 2018 16:34:29 +0000 (09:34 -0700)]
[XLA] Only overwrite the hlo_profiling flag when it's not enabled by default.

This got broken in 504d103a405654f029e8902d97d4dd8f3aa07513

PiperOrigin-RevId: 190077360

6 years agoDisable over-aggressive shape inference
A. Unique TensorFlower [Thu, 22 Mar 2018 16:14:07 +0000 (09:14 -0700)]
Disable over-aggressive shape inference

PiperOrigin-RevId: 190074445

6 years agoMethods to work with symbolic tensor shapes.
A. Unique TensorFlower [Thu, 22 Mar 2018 15:51:43 +0000 (08:51 -0700)]
Methods to work with symbolic tensor shapes.

PiperOrigin-RevId: 190071400

6 years agoFix QN for Calls.
A. Unique TensorFlower [Thu, 22 Mar 2018 15:19:21 +0000 (08:19 -0700)]
Fix QN for Calls.

PiperOrigin-RevId: 190067548

6 years agoAutomatically insert api-links.
Mark Daoust [Thu, 22 Mar 2018 12:50:56 +0000 (05:50 -0700)]
Automatically insert api-links.

If the contents of a pair of back-ticks match a public api symbol name insert a link.

PiperOrigin-RevId: 190051941

6 years agoUpdate file due to changes in Bazel (PACKAGE_NAME is deprecated)
A. Unique TensorFlower [Thu, 22 Mar 2018 12:46:07 +0000 (05:46 -0700)]
Update file due to changes in Bazel (PACKAGE_NAME is deprecated)

PiperOrigin-RevId: 190051589

6 years agoAllow to download clang and use clang for CPU builds.
Ilya Biryukov [Thu, 22 Mar 2018 12:33:42 +0000 (05:33 -0700)]
Allow to download clang and use clang for CPU builds.

Previously we only allowed to download clang when doing GPU builds.

The added skylark files use bazel's autoconf scripts, which were only added in
0.10.0. To provide nice error message for older versions of bazel
(i.e. 'version is less than 0.10' vs 'can't load @bazel_tools/cpp/...'), we
move the bazel version check into WORKSPACE file from workspace.bzl.

PiperOrigin-RevId: 190050798

6 years agoAdd a utility that converts call keyword arguments into dicts, in AST space.
A. Unique TensorFlower [Thu, 22 Mar 2018 11:50:42 +0000 (04:50 -0700)]
Add a utility that converts call keyword arguments into dicts, in AST space.

PiperOrigin-RevId: 190047495

6 years agoJava: Release 1.7.0-rc1
Asim Shankar [Thu, 22 Mar 2018 07:48:30 +0000 (00:48 -0700)]
Java: Release 1.7.0-rc1

PiperOrigin-RevId: 190028714

6 years agoTFLite: Ensure only 1 scale/zero_point is in QuantizationParam.
Yu-Cheng Ling [Thu, 22 Mar 2018 07:26:33 +0000 (00:26 -0700)]
TFLite: Ensure only 1 scale/zero_point is in QuantizationParam.
PiperOrigin-RevId: 190027163

6 years agoMinor style improvement in TFLite interpreter_test.py
Yu-Cheng Ling [Thu, 22 Mar 2018 07:26:31 +0000 (00:26 -0700)]
Minor style improvement in TFLite interpreter_test.py

PiperOrigin-RevId: 190027161

6 years agoAutomated g4 rollback of changelist 190001737
A. Unique TensorFlower [Thu, 22 Mar 2018 06:11:40 +0000 (23:11 -0700)]
Automated g4 rollback of changelist 190001737

PiperOrigin-RevId: 190021164

6 years agoMerge changes from github.
Sourabh Bajaj [Thu, 22 Mar 2018 06:04:59 +0000 (23:04 -0700)]
Merge changes from github.

PiperOrigin-RevId: 190020572

6 years agoSimplified the experimental APIs related to TPU execution, by moving the graph
Mingsheng Hong [Thu, 22 Mar 2018 05:11:10 +0000 (22:11 -0700)]
Simplified the experimental APIs related to TPU execution, by moving the graph
rewrite functionality out of it.

PiperOrigin-RevId: 190016936

6 years agoAdd tensor quantization info to python wrapper
A. Unique TensorFlower [Thu, 22 Mar 2018 02:12:18 +0000 (19:12 -0700)]
Add tensor quantization info to python wrapper

PiperOrigin-RevId: 190005998

6 years agoMoves TFE_Executor to common_runtime
Alexandre Passos [Thu, 22 Mar 2018 01:22:36 +0000 (18:22 -0700)]
Moves TFE_Executor to common_runtime

PiperOrigin-RevId: 190001737

6 years ago[tf.data] Do not crash when combining .cache().take().repeat()
Brennan Saeta [Thu, 22 Mar 2018 01:02:01 +0000 (18:02 -0700)]
[tf.data] Do not crash when combining .cache().take().repeat()

Currently, if the .cache() iterator is not fully consumed before
being repeated, it will cause an exception to be raised to Python.
Instead, cache should act as an identity transformation and log
an error, as this will not affect the correctness of the user's
program (at the cost of an unexpected performance cost: i.e. not
actually caching).

PiperOrigin-RevId: 189999552

6 years agoAdded an experimental C API to dump TF_Graph in a human-readable format, for
Mingsheng Hong [Thu, 22 Mar 2018 00:38:04 +0000 (17:38 -0700)]
Added an experimental C API to dump TF_Graph in a human-readable format, for
debugging purposes.

PiperOrigin-RevId: 189997099

6 years agoSmall convenience changes.
A. Unique TensorFlower [Thu, 22 Mar 2018 00:34:52 +0000 (17:34 -0700)]
Small convenience changes.

PiperOrigin-RevId: 189996801

6 years agoShorter definitions for elementwise_ops in op_level_cost_estimator.
Max Galkin [Wed, 21 Mar 2018 23:00:44 +0000 (16:00 -0700)]
Shorter definitions for elementwise_ops in op_level_cost_estimator.

PiperOrigin-RevId: 189983460

6 years agoChange Softmax on CUDA to use fp32 for denominator when input/output are fp16.
James Qin [Wed, 21 Mar 2018 22:55:30 +0000 (15:55 -0700)]
Change Softmax on CUDA to use fp32 for denominator when input/output are fp16.

This avoids potential overflow in the denominator, also makes sure accumulation is done
in high precision.

PiperOrigin-RevId: 189982655

6 years ago[TF:XLA] do not emit bfloat16 sum reductions from tf2xla
Nick Desaulniers [Wed, 21 Mar 2018 22:23:07 +0000 (15:23 -0700)]
[TF:XLA] do not emit bfloat16 sum reductions from tf2xla

bfloat16 is a storage format, not a computation format. Doing reductions in
this reduced precision is prone to quickly overflow.  Instead, emit a float32
computation, and wrap the reduce params and result in conversions to and from float32.

PiperOrigin-RevId: 189977590

6 years agoAdd an alternative "no_contrib" BUILD target to tensorflow/python to avoid including...
A. Unique TensorFlower [Wed, 21 Mar 2018 21:59:06 +0000 (14:59 -0700)]
Add an alternative "no_contrib" BUILD target to tensorflow/python to avoid including contrib packages

PiperOrigin-RevId: 189973359

6 years agoAvoid taking a reference on a temporary value
Benoit Steiner [Wed, 21 Mar 2018 21:26:14 +0000 (14:26 -0700)]
Avoid taking a reference on a temporary value

PiperOrigin-RevId: 189967517

6 years ago[XLA][BF16] Add bf16 rounding function.
Yunxing Dai [Wed, 21 Mar 2018 21:12:07 +0000 (14:12 -0700)]
[XLA][BF16] Add bf16 rounding function.

We now use truncation to convert a F32 to BF16 by default. This CL adds a rounding method (basically a roll forward of the rounding part in cl/175252067).

PiperOrigin-RevId: 189965138

6 years ago[docs] Update explicit latest_checkpoint with `path` returned while saving for object...
Shivani Agrawal [Wed, 21 Mar 2018 20:57:29 +0000 (13:57 -0700)]
[docs] Update explicit latest_checkpoint with `path` returned while saving for object-based checkpointing.

PiperOrigin-RevId: 189962437

6 years agoGo: Update generated wrapper functions for TensorFlow ops.
A. Unique TensorFlower [Wed, 21 Mar 2018 20:46:27 +0000 (13:46 -0700)]
Go: Update generated wrapper functions for TensorFlow ops.
PiperOrigin-RevId: 189960595

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Wed, 21 Mar 2018 20:30:26 +0000 (13:30 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 189957912

6 years ago[TF CriticalSection] Bugfix when Execute() inside a while_loop has a dep on a Variabl...
Eugene Brevdo [Wed, 21 Mar 2018 20:28:11 +0000 (13:28 -0700)]
[TF CriticalSection] Bugfix when Execute() inside a while_loop has a dep on a Variable outside of it.

PiperOrigin-RevId: 189957569

6 years agoMake strcat.{h,cc} independent of Eigen.
A. Unique TensorFlower [Wed, 21 Mar 2018 20:08:02 +0000 (13:08 -0700)]
Make strcat.{h,cc} independent of Eigen.

PiperOrigin-RevId: 189954596

6 years agoTemporarily disable constant folding past Enter, since a few breakages have been...
A. Unique TensorFlower [Wed, 21 Mar 2018 19:53:53 +0000 (12:53 -0700)]
Temporarily disable constant folding past Enter, since a few breakages have been observed.

PiperOrigin-RevId: 189952252

6 years agoFurther improve accuracy of op_level_cost_estimator (Gather, GatherV2, Slice).
Max Galkin [Wed, 21 Mar 2018 19:53:04 +0000 (12:53 -0700)]
Further improve accuracy of op_level_cost_estimator (Gather, GatherV2, Slice).

PiperOrigin-RevId: 189952132

6 years ago[XLA] Adding more tests for ReduceWindow.
A. Unique TensorFlower [Wed, 21 Mar 2018 19:40:08 +0000 (12:40 -0700)]
[XLA] Adding more tests for ReduceWindow.

PiperOrigin-RevId: 189950361

6 years agoMade sure to compare all the attributes when comparing nodes
Benoit Steiner [Wed, 21 Mar 2018 19:14:30 +0000 (12:14 -0700)]
Made sure to compare all the attributes when comparing nodes

PiperOrigin-RevId: 189946858

6 years agoMerge changes from github.
Jacques Pienaar [Wed, 21 Mar 2018 19:07:51 +0000 (12:07 -0700)]
Merge changes from github.

PiperOrigin-RevId: 189945839

6 years agoFix compilation error with clang.
Ilya Biryukov [Wed, 21 Mar 2018 19:00:04 +0000 (12:00 -0700)]
Fix compilation error with clang.

Link to breaking CI build:
https://ci.tensorflow.org/view/Nightly/job/nightly-matrix-linux-gpu-clang/232

PiperOrigin-RevId: 189944547

6 years agoDon't run data_utils_test without optimizations.
Justin Lebar [Wed, 21 Mar 2018 18:41:12 +0000 (11:41 -0700)]
Don't run data_utils_test without optimizations.

PiperOrigin-RevId: 189941645

6 years agoMake the graph_editor C-API friendly: always construct ops with their inputs.
Frank Perbet [Wed, 21 Mar 2018 18:40:25 +0000 (11:40 -0700)]
Make the graph_editor C-API friendly: always construct ops with their inputs.

PiperOrigin-RevId: 189941495

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Wed, 21 Mar 2018 18:18:02 +0000 (11:18 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 189937564

6 years agoAllowing dnn tree combined estimator to work with core versions of feature columns...
A. Unique TensorFlower [Wed, 21 Mar 2018 18:16:13 +0000 (11:16 -0700)]
Allowing dnn tree combined estimator to work with core versions of feature columns and losses

PiperOrigin-RevId: 189937063

6 years agoMoves KernelAndDevice to common_runtime
Alexandre Passos [Wed, 21 Mar 2018 18:12:28 +0000 (11:12 -0700)]
Moves KernelAndDevice to common_runtime

PiperOrigin-RevId: 189936396

6 years agoAdding `drop_remainder` option for the `map_and_batch` transformation, which allows...
Jiri Simsa [Wed, 21 Mar 2018 18:10:13 +0000 (11:10 -0700)]
Adding `drop_remainder` option for the `map_and_batch` transformation, which allows the user to express whether they wish to drop the last batch if its size is smaller than desired; the default is not to drop the smaller batch.

PiperOrigin-RevId: 189936029