platform/upstream/tensorflow.git
6 years ago[TF:XLA] Create Despecializing Pass Pipeline
Nick Desaulniers [Fri, 6 Apr 2018 18:23:40 +0000 (11:23 -0700)]
[TF:XLA] Create Despecializing Pass Pipeline

When comparing backends, it is useful to take an HLO optimized for one backend and perform transformations in order to match numerics.  This can be thought of as finding a lowest common denominator.

Move this grouping of passes into its own HloPassPipeline that can be reused in a few different places.

PiperOrigin-RevId: 191914799

6 years agoAdded headers only version of tensorflow/core/kernels:cwise_lib, cwise_lib_hdrs....
A. Unique TensorFlower [Fri, 6 Apr 2018 18:21:41 +0000 (11:21 -0700)]
Added headers only version of tensorflow/core/kernels:cwise_lib, cwise_lib_hdrs. This is for clients that want to use the cwise_ops machinery when making their own custom ops, including cwise_lib directly causes multiple definition linker errors.

PiperOrigin-RevId: 191914445

6 years agoMinor doc clarification for reduce_sum return type
A. Unique TensorFlower [Fri, 6 Apr 2018 18:21:24 +0000 (11:21 -0700)]
Minor doc clarification for reduce_sum return type

PiperOrigin-RevId: 191914398

6 years agoProperly handle callable objects.
A. Unique TensorFlower [Fri, 6 Apr 2018 18:17:41 +0000 (11:17 -0700)]
Properly handle callable objects.

PiperOrigin-RevId: 191913834

6 years agoTPU Cost Estimator has been modified to also account for the memory cost in the execu...
A. Unique TensorFlower [Fri, 6 Apr 2018 18:16:17 +0000 (11:16 -0700)]
TPU Cost Estimator has been modified to also account for the memory cost in the execution time. Until more sophisticated methods are added, we resort to the roofline model to calculate such cost.

PiperOrigin-RevId: 191913626

6 years agoChange GetInstructionCallContext to take an opcode instead of an
Mark Heffernan [Fri, 6 Apr 2018 17:28:18 +0000 (10:28 -0700)]
Change GetInstructionCallContext to take an opcode instead of an
HloInstruction.
This enables use of the function without an actual instruction (eg, if you just
have an HloProto).

PiperOrigin-RevId: 191905914

6 years agoAllow TFE_NewContext to fail more reasonably when SWIG is checking status.
Akshay Modi [Fri, 6 Apr 2018 17:28:10 +0000 (10:28 -0700)]
Allow TFE_NewContext to fail more reasonably when SWIG is checking status.

Before:
TFE_Context would check nullptr, and the function would fail straight away.

Now:
TFE_Context is nullptr, so it skips down to checking the status, and an error
is raised.

I'm not able to find in SWIG documentation how to order typemaps in the
generated code - ideally, I'd order it to check the status typemap first. This
code makes it not dependent on this ordering either way.

PiperOrigin-RevId: 191905893

6 years agoObject-based checkpointing support for unidirectional cuDNN LSTM cells
Allen Lavoie [Fri, 6 Apr 2018 17:26:40 +0000 (10:26 -0700)]
Object-based checkpointing support for unidirectional cuDNN LSTM cells

Once checked in, this will be the only way I know of to save canonical weights
when executing eagerly. Eager's name-based saving support will only do the
opaque parameter buffer.

I'm not going to try converting everything in one go, but it's a start at
least. And everything else should raise a NotImplementedError rather than
silently not saving correctly.

Single-layer cuDNN cells can be swapped for un-wrapped cuDNN compatible cells or
single cells wrapped in MultiRNNCells. Multi-layer cells need MultiRNNCell
wrapping.

PiperOrigin-RevId: 191905703

6 years agoMaterialize tensor array sizes whenever possible
Benoit Steiner [Fri, 6 Apr 2018 16:45:01 +0000 (09:45 -0700)]
Materialize tensor array sizes whenever possible

PiperOrigin-RevId: 191900015

6 years ago[tf.data] Enable using `tf.contrib.data.prefetch_to_device()` in eager mode.
Derek Murray [Fri, 6 Apr 2018 16:26:26 +0000 (09:26 -0700)]
[tf.data] Enable using `tf.contrib.data.prefetch_to_device()` in eager mode.

The added functionality is a substitute for the implicit prefetching in
`tfe.Iterator`, and the two paths will converge in a future change.

Fixes #18260.

PiperOrigin-RevId: 191897666

6 years ago[TPUClusterResolver] Start a TFServer when running in GKE
Brennan Saeta [Fri, 6 Apr 2018 16:26:08 +0000 (09:26 -0700)]
[TPUClusterResolver] Start a TFServer when running in GKE

This change allows advanced input pipelines (e.g. StreamingFilesDataset, or split-pipelines that use py_func's) to run in GKE- and GKE-like enviornments.

PiperOrigin-RevId: 191897639

6 years agoTweaked docstrings in LayerCollection.
A. Unique TensorFlower [Fri, 6 Apr 2018 16:21:00 +0000 (09:21 -0700)]
Tweaked docstrings in LayerCollection.

PiperOrigin-RevId: 191897098

6 years agoSupport override of device filters for gRPC, by overriding the requests with default...
A. Unique TensorFlower [Fri, 6 Apr 2018 16:11:32 +0000 (09:11 -0700)]
Support override of device filters for gRPC, by overriding the requests with default session config.

PiperOrigin-RevId: 191895856

6 years agoFix up the support for the case where a given array name occurs multiple
A. Unique TensorFlower [Fri, 6 Apr 2018 15:58:14 +0000 (08:58 -0700)]
Fix up the support for the case where a given array name occurs multiple
times in the inputs/outputs list of an op. The (non-essential) computation
of the optimal workspace size had not been updated for that case, causing it
to fail on a simple test case. Moreover, the initial implementation had some
redundant usage of std::find that this CL moves to a shared helper function.

PiperOrigin-RevId: 191894081

6 years agoThe training model need not be built when the kfac optimizer is initialized so the
A. Unique TensorFlower [Fri, 6 Apr 2018 15:48:16 +0000 (08:48 -0700)]
The training model need not be built when the kfac optimizer is initialized so the
self._variables will be empty list.  So pass a function which returns list of trainable variables to estimator.

PiperOrigin-RevId: 191893084

6 years agoAdd description to the LPIRC 2018 competition benchmarker.
A. Unique TensorFlower [Fri, 6 Apr 2018 15:11:24 +0000 (08:11 -0700)]
Add description to the LPIRC 2018 competition benchmarker.

PiperOrigin-RevId: 191889484

6 years agoAvoid marking clusters containing only Identity ops for compilation.
A. Unique TensorFlower [Fri, 6 Apr 2018 14:50:09 +0000 (07:50 -0700)]
Avoid marking clusters containing only Identity ops for compilation.

This would produce clusters where XLA cannot optimize anything.

PiperOrigin-RevId: 191887414

6 years agoDefine PRNG seeding style for new code in Distributions and TF Probability, with...
A. Unique TensorFlower [Fri, 6 Apr 2018 14:13:49 +0000 (07:13 -0700)]
Define PRNG seeding style for new code in Distributions and TF Probability, with rationales.

Implement lightweight PRNG for seed generation in that style.

Enables incremental refactoring of existing code into this style.

PiperOrigin-RevId: 191884573

6 years agoFix typos in XlaCompilationCache
Igor Ganichev [Fri, 6 Apr 2018 13:30:45 +0000 (06:30 -0700)]
Fix typos in XlaCompilationCache

PiperOrigin-RevId: 191881135

6 years agointernal change
A. Unique TensorFlower [Fri, 6 Apr 2018 10:23:54 +0000 (03:23 -0700)]
internal change

PiperOrigin-RevId: 191869400

6 years agoAdd a test to check graceful handling of out-of-memory conditions.
A. Unique TensorFlower [Fri, 6 Apr 2018 08:46:37 +0000 (01:46 -0700)]
Add a test to check graceful handling of out-of-memory conditions.

PiperOrigin-RevId: 191860462

6 years agoFix StringPiece use-after-free in MasterSession::ReffedClientGraph.
Derek Murray [Fri, 6 Apr 2018 05:37:49 +0000 (22:37 -0700)]
Fix StringPiece use-after-free in MasterSession::ReffedClientGraph.

Use the owned ClientGraph as the source for the node_to_name_ map, rather than the borrowed GraphExecutionState (which can be deleted while the ReffedClientGraph is in use).

PiperOrigin-RevId: 191847023

6 years agoUpgrade libpng
Justine Tunney [Fri, 6 Apr 2018 03:43:47 +0000 (20:43 -0700)]
Upgrade libpng

PiperOrigin-RevId: 191840652

6 years agoSwap in the new implementation of while and for loops.
A. Unique TensorFlower [Fri, 6 Apr 2018 03:14:42 +0000 (20:14 -0700)]
Swap in the new implementation of while and for loops.

PiperOrigin-RevId: 191838806

6 years agoExpose the adaptive sampling option for SDCA and shuffle the data when adaptive sampl...
A. Unique TensorFlower [Fri, 6 Apr 2018 02:33:58 +0000 (19:33 -0700)]
Expose the adaptive sampling option for SDCA and shuffle the data when adaptive sampling is off.

PiperOrigin-RevId: 191836004

6 years ago[XLA] Don't call MultidimensionalIndexToLinearIndex in HloEvaluator's convolution...
Justin Lebar [Fri, 6 Apr 2018 02:30:10 +0000 (19:30 -0700)]
[XLA] Don't call MultidimensionalIndexToLinearIndex in HloEvaluator's convolution routine.

Before: ConvolutionTest/0.StridedFilter (41812 ms)
After:  ConvolutionTest/0.StridedFilter (28054 ms)

Speedup: 42 / 28 = 1.5x
PiperOrigin-RevId: 191835735

6 years agoAdd RunMetadata logging to tf.train.ProfilerHook for Tensorboard Memeory/CPU usage...
A. Unique TensorFlower [Fri, 6 Apr 2018 01:50:35 +0000 (18:50 -0700)]
Add RunMetadata logging to tf.train.ProfilerHook for Tensorboard Memeory/CPU usage visualization

PiperOrigin-RevId: 191832832

6 years agoAdded `drop_final_batch` argument to make_batched_features_dataset. This allows the...
A. Unique TensorFlower [Fri, 6 Apr 2018 01:37:19 +0000 (18:37 -0700)]
Added `drop_final_batch` argument to make_batched_features_dataset. This allows the batch_and_drop_remainder function to be used instead of the default batch function.

PiperOrigin-RevId: 191831842

6 years ago[XLA] Don't call Literal::Get in HloEvaluator's convolution loop.
Justin Lebar [Fri, 6 Apr 2018 01:23:32 +0000 (18:23 -0700)]
[XLA] Don't call Literal::Get in HloEvaluator's convolution loop.

This speeds up the implementation of conv because Literal::Get calls
Literal::Piece::data, which is relatively slow.

Instead, we call Literal::Data() once and cache the result.

Before: ConvolutionTest/0.StridedFilter (59094 ms)
After:  ConvolutionTest/0.StridedFilter (41812 ms)

Speedup: 59/42 = 1.4x
PiperOrigin-RevId: 191830741

6 years agoLazily evaluate shapes with the C API enabled.
Skye Wanderman-Milne [Fri, 6 Apr 2018 01:21:54 +0000 (18:21 -0700)]
Lazily evaluate shapes with the C API enabled.

This change makes it so shapes are computed only when requested with
_USE_C_API = True. Note that the C API will still raise a shape error
if necessary when the op is created.

In addition, it cleans up the logic for _USE_C_SHAPES = True. In this
case, we lazily fetch and cache shapes directly from the C API. We no
longer need set_shapes_for_outputs at all in this case.

PiperOrigin-RevId: 191830565

6 years ago refactor and add proto field required by POD support.
A. Unique TensorFlower [Fri, 6 Apr 2018 00:43:43 +0000 (17:43 -0700)]
  refactor and add proto field required by POD support.

PiperOrigin-RevId: 191826636

6 years agoAdd a command line parameter to toco to change the way toco rescales input and output...
A. Unique TensorFlower [Fri, 6 Apr 2018 00:35:00 +0000 (17:35 -0700)]
Add a command line parameter to toco to change the way toco rescales input and output tensors.

PiperOrigin-RevId: 191825756

6 years agoAutomated g4 rollback of changelist 191605505
Tony Wang [Fri, 6 Apr 2018 00:24:33 +0000 (17:24 -0700)]
Automated g4 rollback of changelist 191605505

PiperOrigin-RevId: 191824447

6 years agoMake concat handler support mixed range input
A. Unique TensorFlower [Fri, 6 Apr 2018 00:08:50 +0000 (17:08 -0700)]
Make concat handler support mixed range input

PiperOrigin-RevId: 191822664

6 years agoAdded an option to inline all functions in aggressive mode.
Benoit Steiner [Thu, 5 Apr 2018 23:46:12 +0000 (16:46 -0700)]
Added an option to inline all functions in aggressive mode.

PiperOrigin-RevId: 191819577

6 years agoAdd link for index file in performance tab.
Mark Daoust [Thu, 5 Apr 2018 22:48:27 +0000 (15:48 -0700)]
Add link for index file in performance tab.

PiperOrigin-RevId: 191811610

6 years agoAdd functions to extract the basic symbols on which a composite name relies. This...
A. Unique TensorFlower [Thu, 5 Apr 2018 22:37:50 +0000 (15:37 -0700)]
Add functions to extract the basic symbols on which a composite name relies. This in turn allows to statically obtain a block's syntactic closure.

PiperOrigin-RevId: 191809965

6 years ago[TF:XLA] No need to set return value in the while loop's condition.
A. Unique TensorFlower [Thu, 5 Apr 2018 22:32:27 +0000 (15:32 -0700)]
[TF:XLA] No need to set return value in the while loop's condition.

PiperOrigin-RevId: 191809110

6 years agoAdd for and while loops to the list of operators. Do not use them yet.
A. Unique TensorFlower [Thu, 5 Apr 2018 22:25:32 +0000 (15:25 -0700)]
Add for and while loops to the list of operators. Do not use them yet.

PiperOrigin-RevId: 191807973

6 years agoAdd Raspberry Pi section and link to github build instructions.
Billy Lamberta [Thu, 5 Apr 2018 22:25:00 +0000 (15:25 -0700)]
Add Raspberry Pi section and link to github build instructions.

PiperOrigin-RevId: 191807862

6 years agoMake tf.contrib.estimator.add_metrics work with warm-starting.
A. Unique TensorFlower [Thu, 5 Apr 2018 22:11:02 +0000 (15:11 -0700)]
Make tf.contrib.estimator.add_metrics work with warm-starting.

PiperOrigin-RevId: 191805682

6 years agoDisable tests that are currently failing with cuda 9
A. Unique TensorFlower [Thu, 5 Apr 2018 22:09:19 +0000 (15:09 -0700)]
Disable tests that are currently failing with cuda 9

PiperOrigin-RevId: 191805453

6 years ago[XLA] Better support for mul reductions in MakeFakeArguments()
Michael Kuperstein [Thu, 5 Apr 2018 21:54:36 +0000 (14:54 -0700)]
[XLA] Better support for mul reductions in MakeFakeArguments()

Mul reductions want a 1 as their init value, not a 0 or a random value.

PiperOrigin-RevId: 191802819

6 years agoFix TF_ImportGraphDefResults and TF_Function leaks in Python API.
Skye Wanderman-Milne [Thu, 5 Apr 2018 21:22:54 +0000 (14:22 -0700)]
Fix TF_ImportGraphDefResults and TF_Function leaks in Python API.

PiperOrigin-RevId: 191797853

6 years agoquantized LSTM support improvements
A. Unique TensorFlower [Thu, 5 Apr 2018 21:05:03 +0000 (14:05 -0700)]
quantized LSTM support improvements

PiperOrigin-RevId: 191794956

6 years agoChanges loss_reduction default to SUM_OVER_BATCH_SIZE for multi_class_head and binary...
A. Unique TensorFlower [Thu, 5 Apr 2018 20:55:45 +0000 (13:55 -0700)]
Changes loss_reduction default to SUM_OVER_BATCH_SIZE for multi_class_head and binary_classification_head.

PiperOrigin-RevId: 191793392

6 years agoEnable branch prediction in TensorFlow
Justine Tunney [Thu, 5 Apr 2018 20:23:08 +0000 (13:23 -0700)]
Enable branch prediction in TensorFlow

PiperOrigin-RevId: 191788253

6 years ago[XLA] Remove a dead function and a stale todo.
A. Unique TensorFlower [Thu, 5 Apr 2018 20:12:26 +0000 (13:12 -0700)]
[XLA] Remove a dead function and a stale todo.

PiperOrigin-RevId: 191786563

6 years agoAutomated g4 rollback of changelist 191753026
A. Unique TensorFlower [Thu, 5 Apr 2018 20:00:06 +0000 (13:00 -0700)]
Automated g4 rollback of changelist 191753026

PiperOrigin-RevId: 191784709

6 years agoSave the original from_proto method before calling it to avoid infinite loop.
Priya Gupta [Thu, 5 Apr 2018 19:57:49 +0000 (12:57 -0700)]
Save the original from_proto method before calling it to avoid infinite loop.

PiperOrigin-RevId: 191784430

6 years agoFix regression caused by cl/191020868: Re-use materialized shapes for other broadcast...
A. Unique TensorFlower [Thu, 5 Apr 2018 19:18:34 +0000 (12:18 -0700)]
Fix regression caused by cl/191020868: Re-use materialized shapes for other broadcast gradient shape nodes.

PiperOrigin-RevId: 191779263

6 years agoFix final eval bottleneck creation to work in cases where it isn't cached already.
Suharsh Sivakumar [Thu, 5 Apr 2018 18:36:43 +0000 (11:36 -0700)]
Fix final eval bottleneck creation to work in cases where it isn't cached already.

Fixes #17423

PiperOrigin-RevId: 191773001

6 years agoAutomated g4 rollback of changelist 191761109
A. Unique TensorFlower [Thu, 5 Apr 2018 18:30:35 +0000 (11:30 -0700)]
Automated g4 rollback of changelist 191761109

PiperOrigin-RevId: 191771969

6 years agoExpose odeint_fixed in tf.contrib.integrate
Stephan Hoyer [Thu, 5 Apr 2018 18:18:49 +0000 (11:18 -0700)]
Expose odeint_fixed in tf.contrib.integrate

PiperOrigin-RevId: 191769890

6 years agoInternal change
A. Unique TensorFlower [Thu, 5 Apr 2018 18:17:58 +0000 (11:17 -0700)]
Internal change

PiperOrigin-RevId: 191769724

6 years agoSync only the convolutional_recurrent file to Keras 2.1.5.
Anjali Sridhar [Thu, 5 Apr 2018 17:41:40 +0000 (10:41 -0700)]
Sync only the convolutional_recurrent file to Keras 2.1.5.

PiperOrigin-RevId: 191763101

6 years agoInline more functions
Benoit Steiner [Thu, 5 Apr 2018 17:29:15 +0000 (10:29 -0700)]
Inline more functions

PiperOrigin-RevId: 191761109

6 years agoExpand activity analysis to the test nodes of if and while statements.
A. Unique TensorFlower [Thu, 5 Apr 2018 16:57:58 +0000 (09:57 -0700)]
Expand activity analysis to the test nodes of if and while statements.

PiperOrigin-RevId: 191756234

6 years agoInclude the operators module in the test framework as well.
A. Unique TensorFlower [Thu, 5 Apr 2018 16:56:50 +0000 (09:56 -0700)]
Include the operators module in the test framework as well.

PiperOrigin-RevId: 191756100

6 years agoRefine BatchReshape error messages.
Joshua V. Dillon [Thu, 5 Apr 2018 16:41:31 +0000 (09:41 -0700)]
Refine BatchReshape error messages.

PiperOrigin-RevId: 191754120

6 years agoDocument expected regular structure of the statistical testing library.
A. Unique TensorFlower [Thu, 5 Apr 2018 16:38:40 +0000 (09:38 -0700)]
Document expected regular structure of the statistical testing library.

PiperOrigin-RevId: 191753693

6 years agoAdded a call in CheckpointSaverHook.after_create_session to always save
Sherry Moore [Thu, 5 Apr 2018 16:33:20 +0000 (09:33 -0700)]
Added a call in CheckpointSaverHook.after_create_session to always save
checkpoint before the first training step.

PiperOrigin-RevId: 191753026

6 years agoFix typos in "Profile Model Float Operations" documentation.
A. Unique TensorFlower [Thu, 5 Apr 2018 16:17:57 +0000 (09:17 -0700)]
Fix typos in "Profile Model Float Operations" documentation.

PiperOrigin-RevId: 191751175

6 years agoUse constants in tf.zeros if the constant won't be too big.
Skye Wanderman-Milne [Thu, 5 Apr 2018 15:47:47 +0000 (08:47 -0700)]
Use constants in tf.zeros if the constant won't be too big.

Using fill saves on GraphDef size, but can slow down models since the
total number of ops is greater (fill + shape + constant op). This
change makes us only use fill for large shapes.

PiperOrigin-RevId: 191747456

6 years agoFix docstring.
A. Unique TensorFlower [Thu, 5 Apr 2018 15:47:27 +0000 (08:47 -0700)]
Fix docstring.

PiperOrigin-RevId: 191747417

6 years agoCreate a separate operators module that is to contain all Python constructs that...
A. Unique TensorFlower [Thu, 5 Apr 2018 12:39:21 +0000 (05:39 -0700)]
Create a separate operators module that is to contain all Python constructs that we override: control flow, builtins, operators, etc.

PiperOrigin-RevId: 191729654

6 years agoAdd support for NCCL2. The configure script asks for what version of NCCL to use...
A. Unique TensorFlower [Thu, 5 Apr 2018 10:09:27 +0000 (03:09 -0700)]
Add support for NCCL2. The configure script asks for what version of NCCL to use. The default is still NCCL 1 from GitHub. If the user chooses NCCL 2, it asks for the install directory.

The nccl_configure.bzl generates two different BUILD files based on the chose NCCL version. For NCCL 1, it aliases to the existing 'nccl_archive' http_repo on GitHub. For NCCL 2, it creates a target containing the NCCL 2 library and headers from the chosen install directory.

PiperOrigin-RevId: 191718007

6 years agoUpgrade Snappy to latest version
Justine Tunney [Thu, 5 Apr 2018 07:53:51 +0000 (00:53 -0700)]
Upgrade Snappy to latest version

They added big endian preprocessor macros in recent releases. Hopefully
this should do the right thing on IBM mainframes.

PiperOrigin-RevId: 191705207

6 years agoUpgrade zlib to latest version
Justine Tunney [Thu, 5 Apr 2018 05:37:42 +0000 (22:37 -0700)]
Upgrade zlib to latest version

PiperOrigin-RevId: 191696213

6 years agoSupport arbitrary permutations for DataFormatDimMap.
A. Unique TensorFlower [Thu, 5 Apr 2018 05:37:35 +0000 (22:37 -0700)]
Support arbitrary permutations for DataFormatDimMap.

PiperOrigin-RevId: 191696203

6 years agoAdd a helper function to re-assign colocation in a graph.
A. Unique TensorFlower [Thu, 5 Apr 2018 01:20:36 +0000 (18:20 -0700)]
Add a helper function to re-assign colocation in a graph.

PiperOrigin-RevId: 191679495

6 years agoSort control inputs alphabetically in ToGraphDefSubRange.
Skye Wanderman-Milne [Thu, 5 Apr 2018 01:00:27 +0000 (18:00 -0700)]
Sort control inputs alphabetically in ToGraphDefSubRange.

PiperOrigin-RevId: 191677358

6 years agoBUGFIX: Detect when broadcasting is required and raise NotImplementedError.
Joshua V. Dillon [Thu, 5 Apr 2018 00:28:41 +0000 (17:28 -0700)]
BUGFIX: Detect when broadcasting is required and raise NotImplementedError.

PiperOrigin-RevId: 191673876

6 years agoExclude some gcc options in Windows build
A. Unique TensorFlower [Thu, 5 Apr 2018 00:18:43 +0000 (17:18 -0700)]
Exclude some gcc options in Windows build

PiperOrigin-RevId: 191672761

6 years ago[XLA]: Enable Memory sanitizer for compilation_passes_test.
Bixia Zheng [Thu, 5 Apr 2018 00:14:11 +0000 (17:14 -0700)]
[XLA]: Enable Memory sanitizer for compilation_passes_test.

Initialize the buffer for a scalar tensor to avoid uninitialized
accesses.

PiperOrigin-RevId: 191672257

6 years agoReplace trivial backend calls with calls to underlying TensorFlow functions - Part 2
Pavithra Vijay [Wed, 4 Apr 2018 23:53:51 +0000 (16:53 -0700)]
Replace trivial backend calls with calls to underlying TensorFlow functions - Part 2

PiperOrigin-RevId: 191669725

6 years agoAddress bug in distributed strategies `Monitor` to allow running for >1 step.
Priya Gupta [Wed, 4 Apr 2018 23:34:54 +0000 (16:34 -0700)]
Address bug in distributed strategies `Monitor` to allow running for >1 step.

PiperOrigin-RevId: 191667378

6 years agoCheck that n + kBlockTrailerSize does not overflow before reading a block
Frank Chen [Wed, 4 Apr 2018 23:26:25 +0000 (16:26 -0700)]
Check that n + kBlockTrailerSize does not overflow before reading a block

PiperOrigin-RevId: 191666300

6 years agoFixed handling of control dependencies in the arithmethic optimizer
Benoit Steiner [Wed, 4 Apr 2018 23:17:46 +0000 (16:17 -0700)]
Fixed handling of control dependencies in the arithmethic optimizer

PiperOrigin-RevId: 191665098

6 years agoRunning sparse_ops_test only in opt mode since the test is flaky (times out) in fastb...
A. Unique TensorFlower [Wed, 4 Apr 2018 23:17:05 +0000 (16:17 -0700)]
Running sparse_ops_test only in opt mode since the test is flaky (times out) in fastbuild mode.

PiperOrigin-RevId: 191665014

6 years agoEnable constant propagation across Enter nodes, but only if is_constant is true.
A. Unique TensorFlower [Wed, 4 Apr 2018 23:05:08 +0000 (16:05 -0700)]
Enable constant propagation across Enter nodes, but only if is_constant is true.
Don't propagate constants with control dependencies through Merge nodes.

PiperOrigin-RevId: 191663396

6 years agoInternal change.
A. Unique TensorFlower [Wed, 4 Apr 2018 22:46:26 +0000 (15:46 -0700)]
Internal change.

PiperOrigin-RevId: 191660588

6 years agoCompile TensorFlow with /arch:AVX on Windows
A. Unique TensorFlower [Wed, 4 Apr 2018 22:45:20 +0000 (15:45 -0700)]
Compile TensorFlow with /arch:AVX on Windows

/arch:AVX is the correpsonding option in MSVC for gcc's --march=native

PiperOrigin-RevId: 191660389

6 years agoAdds commandline option (toco_compatible, bool) that makes the optimize_for_inference...
A. Unique TensorFlower [Wed, 4 Apr 2018 22:45:16 +0000 (15:45 -0700)]
Adds commandline option (toco_compatible, bool) that makes the optimize_for_inference script only use TOCO friendly ops. In particular, FusedResizeAndPadConv2D is not supported by TOCO.

This change does not alter existing behavior (the boolean is set to false by default).

PiperOrigin-RevId: 191660378

6 years agoAdding Operation._control_outputs
Alexandre Passos [Wed, 4 Apr 2018 22:42:14 +0000 (15:42 -0700)]
Adding Operation._control_outputs

PiperOrigin-RevId: 191659944

6 years agoRemove dependency on absl to fix tf-nightly-gpu windows GPU build
Smit Hinsu [Wed, 4 Apr 2018 22:36:52 +0000 (15:36 -0700)]
Remove dependency on absl to fix tf-nightly-gpu windows GPU build

PiperOrigin-RevId: 191659091

6 years agoFix typo in `tf.reduce_mean` documentation.
Daniel Zheng [Wed, 4 Apr 2018 22:28:01 +0000 (15:28 -0700)]
Fix typo in `tf.reduce_mean` documentation.

The range of valid values for the `axis` argument should be `[-rank(input), rank(input))`, just like other reduction ops.

PiperOrigin-RevId: 191657767

6 years agoIterate over a copy of dictionary keys when closing variable subscopes. Otherwise...
Priya Gupta [Wed, 4 Apr 2018 22:15:32 +0000 (15:15 -0700)]
Iterate over a copy of dictionary keys when closing variable subscopes. Otherwise, we run into a "dictionary changed size during iteration" once in a while, as we are modifying the values in the dictionary during the iteration.

PiperOrigin-RevId: 191655599

6 years agoCreates a LinearModel (inherits from keras.training.Model) that creates a linear
Rohan Jain [Wed, 4 Apr 2018 22:14:19 +0000 (15:14 -0700)]
Creates a LinearModel (inherits from keras.training.Model) that creates a linear
model.

Had to modify the __call__ method in the base layer class so that it could work with feature style inputs in which case we lazily convert the inputs to tensors instead of providing tensors as inputs upfront.

PiperOrigin-RevId: 191655445

6 years agoRename `distribute` to `train_distribute` parameter in `RunConfig` to clarify that...
Priya Gupta [Wed, 4 Apr 2018 22:06:08 +0000 (15:06 -0700)]
Rename `distribute` to `train_distribute` parameter in `RunConfig` to clarify that its purpose is only for training.

PiperOrigin-RevId: 191654161

6 years agoAdd quantized uint8 L2Normalization Kernel.
A. Unique TensorFlower [Wed, 4 Apr 2018 21:55:16 +0000 (14:55 -0700)]
Add quantized uint8 L2Normalization Kernel.

PiperOrigin-RevId: 191652174

6 years agoGraphOptimizerStagePipeline to pass through multiple
A. Unique TensorFlower [Wed, 4 Apr 2018 21:45:05 +0000 (14:45 -0700)]
GraphOptimizerStagePipeline to pass through multiple
optimizer stages, skipping stages that return error.

PiperOrigin-RevId: 191650182

6 years agoCheck that the c_api module is not destroyed
Akshay Modi [Wed, 4 Apr 2018 21:41:59 +0000 (14:41 -0700)]
Check that the c_api module is not destroyed

PiperOrigin-RevId: 191649662

6 years agoAdd link to tensorflow.js
Shanqing Cai [Wed, 4 Apr 2018 21:39:40 +0000 (14:39 -0700)]
Add link to tensorflow.js

PiperOrigin-RevId: 191649295

6 years agoInline nested function calls.
A. Unique TensorFlower [Wed, 4 Apr 2018 21:31:29 +0000 (14:31 -0700)]
Inline nested function calls.

PiperOrigin-RevId: 191647899

6 years agoAdding a _create_state method to FeatureColumn so that we can decouple variable ...
Rohan Jain [Wed, 4 Apr 2018 21:29:03 +0000 (14:29 -0700)]
Adding a _create_state method to FeatureColumn so that we can decouple variable (more generally, state) creation from getting dense tensors etc. This lays the groundwork to create a Layer type interface (which is more eager friendly) on top of feature columns where variable creation is separated from getting the tensors out.

PiperOrigin-RevId: 191647517

6 years agorun evaluate nodes on parts of arithmetic optimizer tests.
A. Unique TensorFlower [Wed, 4 Apr 2018 21:28:22 +0000 (14:28 -0700)]
run evaluate nodes on parts of arithmetic optimizer tests.

PiperOrigin-RevId: 191647386

6 years agoPermit use of ArraysExtraInfo/constant_float_value when target model is quantized.
A. Unique TensorFlower [Wed, 4 Apr 2018 20:38:20 +0000 (13:38 -0700)]
Permit use of ArraysExtraInfo/constant_float_value when target model is quantized.

PiperOrigin-RevId: 191639289

6 years agoFix up output array min/max post-quantization if the range was overridden.
A. Unique TensorFlower [Wed, 4 Apr 2018 20:24:06 +0000 (13:24 -0700)]
Fix up output array min/max post-quantization if the range was overridden.

PiperOrigin-RevId: 191637143

6 years agoExpose scatter_add for resource variables.
Adria Puigdomenech [Wed, 4 Apr 2018 20:02:27 +0000 (13:02 -0700)]
Expose scatter_add for resource variables.

PiperOrigin-RevId: 191634030