platform/upstream/tensorflow.git
6 years agoInternal change.
A. Unique TensorFlower [Mon, 16 Apr 2018 22:46:42 +0000 (15:46 -0700)]
Internal change.

PiperOrigin-RevId: 193112205

6 years agoIncrease softmax gpu unittest numeric stability
James Qin [Mon, 16 Apr 2018 21:52:41 +0000 (14:52 -0700)]
Increase softmax gpu unittest numeric stability

PiperOrigin-RevId: 193103363

6 years agoPorting tests for `rpc_op` to OS.
Jiri Simsa [Mon, 16 Apr 2018 21:47:31 +0000 (14:47 -0700)]
Porting tests for `rpc_op` to OS.

PiperOrigin-RevId: 193102564

6 years agoExposes InputPipelineConfig as tf.contrib.tpu.InputPipelineConfig.
A. Unique TensorFlower [Mon, 16 Apr 2018 21:41:21 +0000 (14:41 -0700)]
Exposes InputPipelineConfig as tf.contrib.tpu.InputPipelineConfig.
This type is expected by the `per_host_input_for_training` argument of the TPUConfig constructor, but is not currently visible.

PiperOrigin-RevId: 193101540

6 years agodisabling test that fails tensorflow.asan
Olivia Nordquist [Mon, 16 Apr 2018 21:17:02 +0000 (14:17 -0700)]
disabling test that fails tensorflow.asan

PiperOrigin-RevId: 193097794

6 years agoUse fixed sized tensor arrays and max loop iterations in dynamic_decode if the user...
A. Unique TensorFlower [Mon, 16 Apr 2018 21:13:52 +0000 (14:13 -0700)]
Use fixed sized tensor arrays and max loop iterations in dynamic_decode if the user supplies it and if the inputs were created in an XLA context.

PiperOrigin-RevId: 193097293

6 years agoSimplify the recursion when processing unpackings.
A. Unique TensorFlower [Mon, 16 Apr 2018 20:54:03 +0000 (13:54 -0700)]
Simplify the recursion when processing unpackings.

PiperOrigin-RevId: 193094078

6 years ago[XLA] Document and enforce reduction order of init value
Nick Desaulniers [Mon, 16 Apr 2018 20:36:55 +0000 (13:36 -0700)]
[XLA] Document and enforce reduction order of init value

All existing backends apply the init_value on the lhs, except for the evaluator.  This causes problems for reductions which apply an identity function to a reduce or reduce window.

PiperOrigin-RevId: 193091323

6 years agoAdding several utility functions to TF2XLA to help with the Cholesky refactor. Mainl...
A. Unique TensorFlower [Mon, 16 Apr 2018 20:32:12 +0000 (13:32 -0700)]
Adding several utility functions to TF2XLA to help with the Cholesky refactor.  Mainly responsible for handling batching properly.

PiperOrigin-RevId: 193090634

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Mon, 16 Apr 2018 20:22:58 +0000 (13:22 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 193089301

6 years agoSimplify the implementation of break_statements.py
A. Unique TensorFlower [Mon, 16 Apr 2018 20:04:23 +0000 (13:04 -0700)]
Simplify the implementation of break_statements.py

PiperOrigin-RevId: 193086371

6 years agoboosted_trees: Make some regularizer/hyper-params as inputs instead of attributes.
Younghee Kwon [Mon, 16 Apr 2018 19:56:14 +0000 (12:56 -0700)]
boosted_trees: Make some regularizer/hyper-params as inputs instead of attributes.

PiperOrigin-RevId: 193085059

6 years agoEarly TPU distribution strategy and the associated testing infrastructure.
Igor Saprykin [Mon, 16 Apr 2018 19:21:15 +0000 (12:21 -0700)]
Early TPU distribution strategy and the associated testing infrastructure.

PiperOrigin-RevId: 193080098

6 years agoTighten label check in BinaryLogisticHeadWithSigmoidCrossEntropyLoss
A. Unique TensorFlower [Mon, 16 Apr 2018 19:12:46 +0000 (12:12 -0700)]
Tighten label check in BinaryLogisticHeadWithSigmoidCrossEntropyLoss

PiperOrigin-RevId: 193078844

6 years agoAdd a common transformer feature that allows keeping temporary state across nodes...
A. Unique TensorFlower [Mon, 16 Apr 2018 19:09:36 +0000 (12:09 -0700)]
Add a common transformer feature that allows keeping temporary state across nodes. To be used in the break, continue and return canonicalizers.

PiperOrigin-RevId: 193078359

6 years agoCopy the if statement handlers over to the operators module. They will enabled in...
A. Unique TensorFlower [Mon, 16 Apr 2018 19:09:33 +0000 (12:09 -0700)]
Copy the if statement handlers over to the operators module. They will enabled in a follow-up CL.

PiperOrigin-RevId: 193078348

6 years agoRemove obsolete tests. Patch the unexpected print output in Python 3.
A. Unique TensorFlower [Mon, 16 Apr 2018 19:09:24 +0000 (12:09 -0700)]
Remove obsolete tests. Patch the unexpected print output in Python 3.

PiperOrigin-RevId: 193078330

6 years agoMark the parent list as modified for index writes. Add special case for constructors...
A. Unique TensorFlower [Mon, 16 Apr 2018 18:56:15 +0000 (11:56 -0700)]
Mark the parent list as modified for index writes. Add special case for constructors where we know setting an attribute actually creates a new symbol. Clean up the tests a bit.

PiperOrigin-RevId: 193075909

6 years agoRefactoring: Rename the __ops module to ag__ (double underscore prefix has special...
A. Unique TensorFlower [Mon, 16 Apr 2018 18:47:11 +0000 (11:47 -0700)]
Refactoring: Rename the __ops module to ag__ (double underscore prefix has special meaning in Python). Consolidate all internal API calls through the ag__ module.

PiperOrigin-RevId: 193074379

6 years agoInternal testing changes
Austin Anderson [Mon, 16 Apr 2018 18:33:14 +0000 (11:33 -0700)]
Internal testing changes

PiperOrigin-RevId: 193071881

6 years agoRemove the hidden replicate_model_fn copy from core.
Igor Saprykin [Mon, 16 Apr 2018 18:27:09 +0000 (11:27 -0700)]
Remove the hidden replicate_model_fn copy from core.

PiperOrigin-RevId: 193070799

6 years agoPorting tests for the `decode_proto` and `encode_proto` to OS.
Jiri Simsa [Mon, 16 Apr 2018 18:24:43 +0000 (11:24 -0700)]
Porting tests for the `decode_proto` and `encode_proto` to OS.

PiperOrigin-RevId: 193070420

6 years agoPrefix clip_by_value with underscore in gen_math_ops so that it doesn't
Anna R [Mon, 16 Apr 2018 18:20:40 +0000 (11:20 -0700)]
Prefix clip_by_value with underscore in gen_math_ops so that it doesn't
interfere with clip_ops.clip_by_value when importing.

PiperOrigin-RevId: 193069700

6 years agoPiperOrigin-RevId: 193059174
Anna R [Mon, 16 Apr 2018 17:22:29 +0000 (10:22 -0700)]
PiperOrigin-RevId: 193059174

6 years agoUpdate for ObjectMemoryBuffer.h rename in upstream LLVM.
Benjamin Kramer [Mon, 16 Apr 2018 16:34:27 +0000 (09:34 -0700)]
Update for ObjectMemoryBuffer.h rename in upstream LLVM.

This will require a version bump in workspace.bzl

PiperOrigin-RevId: 193052084

6 years agoInternal change
A. Unique TensorFlower [Mon, 16 Apr 2018 04:52:17 +0000 (21:52 -0700)]
Internal change

PiperOrigin-RevId: 192981122

6 years agoMinor cleanup.
A. Unique TensorFlower [Mon, 16 Apr 2018 00:35:39 +0000 (17:35 -0700)]
Minor cleanup.

PiperOrigin-RevId: 192971080

6 years agoStyle nit: avoid creating local variables when not required.
A. Unique TensorFlower [Sat, 14 Apr 2018 09:15:58 +0000 (02:15 -0700)]
Style nit: avoid creating local variables when not required.

PiperOrigin-RevId: 192876802

6 years agoMake sure that same nodes are not optimized as part of multiple groups.
A. Unique TensorFlower [Sat, 14 Apr 2018 08:22:59 +0000 (01:22 -0700)]
Make sure that same nodes are not optimized as part of multiple groups.

Replace recusrsion with iteration in AbsorbInputByOptimizedNodesGroup.

PiperOrigin-RevId: 192874364

6 years ago[XLA] Use pattern matcher in algebraic simplifier
A. Unique TensorFlower [Sat, 14 Apr 2018 04:15:59 +0000 (21:15 -0700)]
[XLA] Use pattern matcher in algebraic simplifier

PiperOrigin-RevId: 192862841

6 years agoUpgrade gRPC version used in OSS Tensorflow
A. Unique TensorFlower [Sat, 14 Apr 2018 04:09:37 +0000 (21:09 -0700)]
Upgrade gRPC version used in OSS Tensorflow

PiperOrigin-RevId: 192862541

6 years agoCheck there are no duplicate entries in sparse features as this would invalidate...
A. Unique TensorFlower [Sat, 14 Apr 2018 04:00:02 +0000 (21:00 -0700)]
Check there are no duplicate entries in sparse features as this would invalidate the example norm computation in SDCA.

PiperOrigin-RevId: 192861834

6 years agoAdding 1d and 3d orthogonal kernels convolutions.
A. Unique TensorFlower [Sat, 14 Apr 2018 03:59:14 +0000 (20:59 -0700)]
Adding 1d and 3d orthogonal kernels convolutions.

PiperOrigin-RevId: 192861809

6 years agoInternal Change
A. Unique TensorFlower [Sat, 14 Apr 2018 02:22:02 +0000 (19:22 -0700)]
Internal Change

PiperOrigin-RevId: 192856330

6 years agoUpdate ops-related pbtxt files.
A. Unique TensorFlower [Sat, 14 Apr 2018 02:18:28 +0000 (19:18 -0700)]
Update ops-related pbtxt files.

PiperOrigin-RevId: 192856167

6 years agoClarify a caveat about metrics.
Igor Saprykin [Sat, 14 Apr 2018 02:09:07 +0000 (19:09 -0700)]
Clarify a caveat about metrics.

PiperOrigin-RevId: 192855733

6 years agoMerge changes from github.
Scott Zhu [Sat, 14 Apr 2018 00:52:20 +0000 (17:52 -0700)]
Merge changes from github.

PiperOrigin-RevId: 192850372

6 years agoJava: Bump release to 1.8.0-rc0
Asim Shankar [Sat, 14 Apr 2018 00:51:37 +0000 (17:51 -0700)]
Java: Bump release to 1.8.0-rc0

PiperOrigin-RevId: 192850310

6 years agoAdd a caveat about make_initiliazable_iterator to the README.
Igor Saprykin [Sat, 14 Apr 2018 00:47:58 +0000 (17:47 -0700)]
Add a caveat about make_initiliazable_iterator to the README.

PiperOrigin-RevId: 192850014

6 years agoVLOG(1) all OutOfRange CtxFailures, and LOG(WARNING) all other CtxFailures. This
Jeremy Lau [Sat, 14 Apr 2018 00:36:00 +0000 (17:36 -0700)]
VLOG(1) all OutOfRange CtxFailures, and LOG(WARNING) all other CtxFailures. This
unifies the logging behavior of the OP_REQUIRES and OP_REQUIRES_OK macros.

PiperOrigin-RevId: 192848921

6 years agoRevamp Cudnn RNN kernels for incoming autotune changes.
James Qin [Sat, 14 Apr 2018 00:26:46 +0000 (17:26 -0700)]
Revamp Cudnn RNN kernels for incoming autotune changes.

* Create DoForward() and DoBackward() to be used by fwd/bak kernels and later autotune.
* Simplify CudnnRnnForward Comupute() function. Offload the majority of its logic to other member functions.

PiperOrigin-RevId: 192848100

6 years agoTFLite: Copy output data from BufferHandle to CPU memory by default.
Yu-Cheng Ling [Sat, 14 Apr 2018 00:13:45 +0000 (17:13 -0700)]
TFLite: Copy output data from BufferHandle to CPU memory by default.
PiperOrigin-RevId: 192846824

6 years agoAutomated g4 rollback of changelist 192784701
Yifei Feng [Fri, 13 Apr 2018 23:38:29 +0000 (16:38 -0700)]
Automated g4 rollback of changelist 192784701

PiperOrigin-RevId: 192842670

6 years agoImprove layout optimizer tests -- Evaluate nodes before and after optimization, to...
A. Unique TensorFlower [Fri, 13 Apr 2018 23:38:12 +0000 (16:38 -0700)]
Improve layout optimizer tests -- Evaluate nodes before and after optimization, to confirm the graph's behavior is maintained after optimization.

PiperOrigin-RevId: 192842623

6 years agoUpgrade SQLite
Justine Tunney [Fri, 13 Apr 2018 22:57:45 +0000 (15:57 -0700)]
Upgrade SQLite

PiperOrigin-RevId: 192837358

6 years agoRestore definitions of static members in MklCpuAllocator.
A. Unique TensorFlower [Fri, 13 Apr 2018 22:53:05 +0000 (15:53 -0700)]
Restore definitions of static members in MklCpuAllocator.

These were removed in #17396 which made the static member variables of
MklCpuAllocator into inline variables, which are a C++17 feature, and not
properly restored in #18006 which reverted the inline declarations, leading to
an ODR violation that is apparently ignored with some compilers.
END_PUBLIC

RELNOTES: n/a

BEGIN_PUBLIC
Automated g4 rollback of changelist 191305220

PiperOrigin-RevId: 192836808

6 years agoEnable remote functions for TPU_SYSTEM.
Akshay Agrawal [Fri, 13 Apr 2018 22:47:37 +0000 (15:47 -0700)]
Enable remote functions for TPU_SYSTEM.

PiperOrigin-RevId: 192836098

6 years agoAvoid mixing `Dimension` type and `int` when defining kernel shapes in conv layers.
Francois Chollet [Fri, 13 Apr 2018 22:33:07 +0000 (15:33 -0700)]
Avoid mixing `Dimension` type and `int` when defining kernel shapes in conv layers.

PiperOrigin-RevId: 192834255

6 years ago[XLA] Pattern matcher for HLO, Shapes, Layouts
A. Unique TensorFlower [Fri, 13 Apr 2018 22:32:11 +0000 (15:32 -0700)]
[XLA] Pattern matcher for HLO, Shapes, Layouts

PiperOrigin-RevId: 192834129

6 years agoAdd broadcasting to all LinearOperators.
A. Unique TensorFlower [Fri, 13 Apr 2018 22:23:08 +0000 (15:23 -0700)]
Add broadcasting to all LinearOperators.

This will broadcast in cases where batch shapes are not equal (but tries to determine statically if this is the case). The broadcasting is not as efficient as doing the broadcast in C++, but makes for the API to at least be completely broadcastable.

PiperOrigin-RevId: 192832919

6 years ago[TF] Enable half precision XLA compiler tests for the gpu backend.
Bixia Zheng [Fri, 13 Apr 2018 22:15:44 +0000 (15:15 -0700)]
[TF] Enable half precision XLA compiler tests for the gpu backend.

Modify some tests to allow larger error for half precision.

Enable half precision SpaceToBatchNDTest for the cpu backend.

PiperOrigin-RevId: 192831909

6 years agoAdded PmfToQuantizedCdf op to contrib/coder in TensorFlow.
Sung Jin Hwang [Fri, 13 Apr 2018 21:51:16 +0000 (14:51 -0700)]
Added PmfToQuantizedCdf op to contrib/coder in TensorFlow.

The added op transforms probability mass functions (PMF) to quantized
cumulative distribution function (CDF), which can be used by range coder ops in
contrib/coder.

The op takes greedy approach to ensure that the post-quantization probability
masses do not sum over the maximum quantized value. The op does not make any
adjustment when the post-quantization probability masses already sum less than
the maximum value.

PiperOrigin-RevId: 192827779

6 years agoAllow tf.train.Saver to load object-based checkpoints (using names)
Allen Lavoie [Fri, 13 Apr 2018 21:32:45 +0000 (14:32 -0700)]
Allow tf.train.Saver to load object-based checkpoints (using names)

This is the second part of the compatibility story. Object-based checkpointing APIs can already read name-based checkpoints, and now the name-based APIs can read object-based checkpoints by looking up the modified keys in the object graph proto.

PiperOrigin-RevId: 192824907

6 years ago[XLA] Redesign: add ~XlaOp() and ~XlaComputation().
A. Unique TensorFlower [Fri, 13 Apr 2018 21:17:31 +0000 (14:17 -0700)]
[XLA] Redesign: add ~XlaOp() and ~XlaComputation().

PiperOrigin-RevId: 192822559

6 years agoSmall tag change
Austin Anderson [Fri, 13 Apr 2018 21:13:12 +0000 (14:13 -0700)]
Small tag change

PiperOrigin-RevId: 192821895

6 years agoExtend Keras symbol-feeding to dynamic-length tensors and tensors of different dtypes...
Francois Chollet [Fri, 13 Apr 2018 21:12:16 +0000 (14:12 -0700)]
Extend Keras symbol-feeding to dynamic-length tensors and tensors of different dtypes from the target placeholders.

PiperOrigin-RevId: 192821770

6 years agoInternal change.
Anna R [Fri, 13 Apr 2018 21:09:58 +0000 (14:09 -0700)]
Internal change.

PiperOrigin-RevId: 192821482

6 years agoSupport scalar mean in resolve_batch_normalization
A. Unique TensorFlower [Fri, 13 Apr 2018 20:40:28 +0000 (13:40 -0700)]
Support scalar mean in resolve_batch_normalization

PiperOrigin-RevId: 192816848

6 years agoRemove closure_js_proto_library rule for tf.example protos.
James Wexler [Fri, 13 Apr 2018 19:44:41 +0000 (12:44 -0700)]
Remove closure_js_proto_library rule for tf.example protos.

PiperOrigin-RevId: 192809073

6 years ago[XLA] Redesign: add a constructor: XlaComputation(HloModuleProto).
A. Unique TensorFlower [Fri, 13 Apr 2018 19:37:04 +0000 (12:37 -0700)]
[XLA] Redesign: add a constructor: XlaComputation(HloModuleProto).

PiperOrigin-RevId: 192808038

6 years ago[XLA] Redesign: add a method that creates fake data for XlaComputation.
A. Unique TensorFlower [Fri, 13 Apr 2018 19:35:32 +0000 (12:35 -0700)]
[XLA] Redesign: add a method that creates fake data for XlaComputation.

PiperOrigin-RevId: 192807851

6 years agoExpose tf.decode_compressed to the public API.
A. Unique TensorFlower [Fri, 13 Apr 2018 19:18:53 +0000 (12:18 -0700)]
Expose tf.decode_compressed to the public API.

PiperOrigin-RevId: 192805605

6 years agoFixing output alternatives
A. Unique TensorFlower [Fri, 13 Apr 2018 18:16:36 +0000 (11:16 -0700)]
Fixing output alternatives

PiperOrigin-RevId: 192795596

6 years ago[TF:XLA] Start a TensorFlow library that contains direct wrappers for XLA operators.
Peter Hawkins [Fri, 13 Apr 2018 18:14:09 +0000 (11:14 -0700)]
[TF:XLA] Start a TensorFlow library that contains direct wrappers for XLA operators.

Add new XlaReduceWindow and XlaDynamicUpdateSlice operators.
Add new tests for the existing XlaWhile operator.
Add wrappers for XlaSend and XlaRecv.

PiperOrigin-RevId: 192795174

6 years agoAdd more logging for failure cases in CUDATimer
James Qin [Fri, 13 Apr 2018 18:06:49 +0000 (11:06 -0700)]
Add more logging for failure cases in CUDATimer

PiperOrigin-RevId: 192793983

6 years agoFix comment in xla_data.proto related to padding value for Windows.
Mark Heffernan [Fri, 13 Apr 2018 18:02:08 +0000 (11:02 -0700)]
Fix comment in xla_data.proto related to padding value for Windows.

PiperOrigin-RevId: 192792971

6 years agoInternal change.
Anna R [Fri, 13 Apr 2018 17:52:56 +0000 (10:52 -0700)]
Internal change.

PiperOrigin-RevId: 192791493

6 years agoDisable x * x -> square(x) Grapler rewrite for complex types unless the op is on...
A. Unique TensorFlower [Fri, 13 Apr 2018 17:30:32 +0000 (10:30 -0700)]
Disable x * x -> square(x) Grapler rewrite for complex types unless the op is on CPU. Square is not registered for complex types on GPU, and doing so produces a crash in with CUDA_ILLEGAL_INSTRUCTION when running it on open source ubuntu.

PiperOrigin-RevId: 192788160

6 years agoDepthwiseConv Optimization Fixes
A. Unique TensorFlower [Fri, 13 Apr 2018 17:27:11 +0000 (10:27 -0700)]
DepthwiseConv Optimization Fixes

PiperOrigin-RevId: 192787669

6 years agoEnable GCS remote cache in Windows Bazel Build
A. Unique TensorFlower [Fri, 13 Apr 2018 17:07:10 +0000 (10:07 -0700)]
Enable GCS remote cache in Windows Bazel Build

PiperOrigin-RevId: 192784701

6 years agoAdd debugging checks for setting cuda stream, so it will check fail if the
Guangda Lai [Fri, 13 Apr 2018 17:02:25 +0000 (10:02 -0700)]
Add debugging checks for setting cuda stream, so it will check fail if the
stream is not set or set to a wrong one when running cudnn methods that
conceptually require a stream.

Also add missing cudnnSetStream()s for DoRnnForwardImpl() and
DoRnnBackwardImpl().

Implementation details:
1. a current_cudnn_stream_ member is added which will be set in cudnnSetStream()
2. a different macro is used to wrap cudnn methods that require a stream in
   order to verify whether the provided stream is same as current_cudnn_stream_,
   and the program will check fail if not

PiperOrigin-RevId: 192783913

6 years agoExperiment with pre-shuffled fully-connected weights
A. Unique TensorFlower [Fri, 13 Apr 2018 15:22:06 +0000 (08:22 -0700)]
Experiment with pre-shuffled fully-connected weights

PiperOrigin-RevId: 192771889

6 years agoRefactor to remove the duplicate calls to obtain a function's namespace. This removes...
A. Unique TensorFlower [Fri, 13 Apr 2018 15:17:49 +0000 (08:17 -0700)]
Refactor to remove the duplicate calls to obtain a function's namespace. This removes the need to explicitly import internal components (barring the tf module which cannot be imported directly).

PiperOrigin-RevId: 192771440

6 years agoAdd support to TFLite for dilated convolution.
A. Unique TensorFlower [Fri, 13 Apr 2018 15:12:42 +0000 (08:12 -0700)]
Add support to TFLite for dilated convolution.

PiperOrigin-RevId: 192770919

6 years agoAutomated g4 rollback of changelist 192768744
A. Unique TensorFlower [Fri, 13 Apr 2018 15:10:57 +0000 (08:10 -0700)]
Automated g4 rollback of changelist 192768744

PiperOrigin-RevId: 192770717

6 years agoSplit byte_order.h off cpu_info.h
A. Unique TensorFlower [Fri, 13 Apr 2018 14:55:46 +0000 (07:55 -0700)]
Split byte_order.h off cpu_info.h

PiperOrigin-RevId: 192768744

6 years agoDemo: RNN colorbot with Estimators.
A. Unique TensorFlower [Fri, 13 Apr 2018 14:24:15 +0000 (07:24 -0700)]
Demo: RNN colorbot with Estimators.

PiperOrigin-RevId: 192765203

6 years agoKeep function doc string at the top of the function.
A. Unique TensorFlower [Fri, 13 Apr 2018 13:53:54 +0000 (06:53 -0700)]
Keep function doc string at the top of the function.

PiperOrigin-RevId: 192761604

6 years ago- Fixed small bug in example script
A. Unique TensorFlower [Fri, 13 Apr 2018 12:47:21 +0000 (05:47 -0700)]
- Fixed small bug in example script

PiperOrigin-RevId: 192756152

6 years agoUpdate for upstream LLVM *.def -> *.inc rename
A. Unique TensorFlower [Fri, 13 Apr 2018 12:05:12 +0000 (05:05 -0700)]
Update for upstream LLVM *.def -> *.inc rename

PiperOrigin-RevId: 192752798

6 years agoFix bug in converted_call, and add tests for it.
A. Unique TensorFlower [Fri, 13 Apr 2018 11:44:10 +0000 (04:44 -0700)]
Fix bug in converted_call, and add tests for it.

PiperOrigin-RevId: 192751211

6 years agoImplementation of Less
A. Unique TensorFlower [Fri, 13 Apr 2018 07:12:41 +0000 (00:12 -0700)]
Implementation of Less

PiperOrigin-RevId: 192728635

6 years ago-- Add a new histogram/cdf computation method compatible with the TPU.
A. Unique TensorFlower [Fri, 13 Apr 2018 07:03:48 +0000 (00:03 -0700)]
-- Add a new histogram/cdf computation method compatible with the TPU.
-- Refactor utility functions into pruning_utils.py and add tests

PiperOrigin-RevId: 192727737

6 years ago[XLA] Rename Interpreter{Executor,Platform} -> XlaInterpreter{Executor,Platform}.
Justin Lebar [Fri, 13 Apr 2018 06:05:35 +0000 (23:05 -0700)]
[XLA] Rename Interpreter{Executor,Platform} -> XlaInterpreter{Executor,Platform}.

These types live inside StreamExecutor's namespace, but they are
specific to XLA.  Therefore they either shouldn't live in SE's namespace
or should have "XLA" in the name.

Moving them out of SE's namespace is ugly, because almost every type
used inside of these headers then needs to be qualified.  So name-change
it is.

This patch was generated by a mechanical find/replace.

PiperOrigin-RevId: 192724238

6 years agoAutomated g4 rollback of changelist 192698931
Jiri Simsa [Fri, 13 Apr 2018 04:39:26 +0000 (21:39 -0700)]
Automated g4 rollback of changelist 192698931

PiperOrigin-RevId: 192718697

6 years agoAutomated g4 rollback of changelist 192504411
Jiri Simsa [Fri, 13 Apr 2018 02:56:38 +0000 (19:56 -0700)]
Automated g4 rollback of changelist 192504411

PiperOrigin-RevId: 192711501

6 years agoAdd boolean type to tflite in favor of comparison implementations.
A. Unique TensorFlower [Fri, 13 Apr 2018 02:52:18 +0000 (19:52 -0700)]
Add boolean type to tflite in favor of comparison implementations.

PiperOrigin-RevId: 192711203

6 years agoAutomated g4 rollback of changelist 192691078
Jiri Simsa [Fri, 13 Apr 2018 02:13:18 +0000 (19:13 -0700)]
Automated g4 rollback of changelist 192691078

PiperOrigin-RevId: 192708480

6 years agoEnable efficient feeding of symbolic tensors to placeholders in the Keras backend.
Francois Chollet [Fri, 13 Apr 2018 02:01:10 +0000 (19:01 -0700)]
Enable efficient feeding of symbolic tensors to placeholders in the Keras backend.

PiperOrigin-RevId: 192707345

6 years agoAvoid calling K.learning_phase() when not necessary in Dropout layer since it instant...
Francois Chollet [Fri, 13 Apr 2018 01:36:13 +0000 (18:36 -0700)]
Avoid calling K.learning_phase() when not necessary in Dropout layer since it instantiates a placeholder_with_default, which is not supported by TPU compilation.

PiperOrigin-RevId: 192705478

6 years agoConvert GrapplerFunctionItem to (Specialized)FunctionDef.
A. Unique TensorFlower [Fri, 13 Apr 2018 01:29:05 +0000 (18:29 -0700)]
Convert GrapplerFunctionItem to (Specialized)FunctionDef.

PiperOrigin-RevId: 192704808

6 years agoAdding autograph built-in function checker.
A. Unique TensorFlower [Fri, 13 Apr 2018 01:19:05 +0000 (18:19 -0700)]
Adding autograph built-in function checker.

PiperOrigin-RevId: 192703924

6 years agoReintroducing support for constants as outputs of tf.data.map(). This fixes a regress...
Jiri Simsa [Fri, 13 Apr 2018 01:02:58 +0000 (18:02 -0700)]
Reintroducing support for constants as outputs of tf.data.map(). This fixes a regression introduced by cl/176147440.

PiperOrigin-RevId: 192702279

6 years agoPrint error msg in CUDATimer.Init() when CreateEvent() is not ok().
James Qin [Fri, 13 Apr 2018 00:35:56 +0000 (17:35 -0700)]
Print error msg in CUDATimer.Init() when CreateEvent() is not ok().

PiperOrigin-RevId: 192699277

6 years agoPorting tests for `rpc_op` to OS.
Jiri Simsa [Fri, 13 Apr 2018 00:32:36 +0000 (17:32 -0700)]
Porting tests for `rpc_op` to OS.

PiperOrigin-RevId: 192698931

6 years agoMove dummy AssertOp and CheckNumericsOp to //third_party/tensorflow/compiler/tf2xla...
A. Unique TensorFlower [Fri, 13 Apr 2018 00:07:35 +0000 (17:07 -0700)]
Move dummy AssertOp and CheckNumericsOp to //third_party/tensorflow/compiler/tf2xla/kernels.
Enable type DT_STRING for AssertOp and ConstOp, in order to make dummy Assert compile with a const string (assert message) as its input.

PiperOrigin-RevId: 192695938

6 years agoFix a typo in cross_tower_ops.
Yuefeng Zhou [Fri, 13 Apr 2018 00:01:55 +0000 (17:01 -0700)]
Fix a typo in cross_tower_ops.

PiperOrigin-RevId: 192694794

6 years agoMisc. small optimizations in Grappler and shape inference code.
A. Unique TensorFlower [Thu, 12 Apr 2018 23:59:57 +0000 (16:59 -0700)]
Misc. small optimizations in Grappler and shape inference code.

Impact on time per optimizer on inception graph:

model_pruner:          590 ms -> 550 ms   (-7%)
function_optimizer:    130 ms -> 130 ms   (-0%)
constant_folding:     7600 ms -> 7550 ms  (-0.7%)
arithmetic_optimizer: 1860 ms -> 1550 ms  (-20%)
loop_optimizer:        320 ms -> 320 ms   (-0%)
dependency_optimizer: 1300 ms -> 720 ms   (-45%)
layout:               1400 ms -> 1400 ms  (-0%)
memory_optimizer:     4200 ms -> 3540 ms  (-16%)
PiperOrigin-RevId: 192694528

6 years agoDocument support for boolean values in tf.contrib.training.HParams.
Stephan Hoyer [Thu, 12 Apr 2018 23:57:40 +0000 (16:57 -0700)]
Document support for boolean values in tf.contrib.training.HParams.

PiperOrigin-RevId: 192694244

6 years ago[XLA:GPU] Pass all four args to custom-call convs when they're created.
Justin Lebar [Thu, 12 Apr 2018 23:56:45 +0000 (16:56 -0700)]
[XLA:GPU] Pass all four args to custom-call convs when they're created.

A custom-call-conv should have four arguments: lhs, rhs, algorithm, and
use-tensor-cores.  CudnnAlgorithmPicker did the right thing, and that
path is exercised 99% of the time.  But CudnnAlgorithmPicker can fail,
and if it does, we're stuck with whatever we had before.  What we had
before only had three of the four args, which is bad.

In addition to fixing it, added an e2e test that catches the bug.

PiperOrigin-RevId: 192694119