Gunhan Gulsoy [Sat, 19 May 2018 03:30:46 +0000 (20:30 -0700)]
Merge pull request #19409 from zheng-xq/branch_197218170
Branch
197218170
annarev [Sat, 19 May 2018 02:10:04 +0000 (19:10 -0700)]
Merge pull request #19408 from annarev/add_more_missing_deps
Add missing dependencies to test_lite_main
zhengxq [Sat, 19 May 2018 00:05:40 +0000 (17:05 -0700)]
Merge commit for internal changes
Xiaoqiang Zheng [Fri, 18 May 2018 23:37:27 +0000 (16:37 -0700)]
Merge pull request #19338 from girving/clip
Make tf.clip_by_value not crash on empty tensors
Igor Saprykin [Fri, 18 May 2018 23:33:19 +0000 (16:33 -0700)]
Automated g4 rollback of changelist
197070234
PiperOrigin-RevId:
197218170
A. Unique TensorFlower [Fri, 18 May 2018 23:28:59 +0000 (16:28 -0700)]
Improve import error messages.
PiperOrigin-RevId:
197217638
Anna R [Fri, 18 May 2018 23:11:30 +0000 (16:11 -0700)]
Add missing dependencies to test_lite_main
Igor Saprykin [Fri, 18 May 2018 22:49:01 +0000 (15:49 -0700)]
Skip tests that require unavailable hardware when not using DistributionStrategy
Right now combinations.py skips tests that do not have the hardware that's
requried by the DistributionStrategy instance that is used in that test. After
this change, the user can trigger such a behavior even when they are not using
DistributionStrategy.
Two new special arguments are added: "required_tpu" and "required_gpus".
PiperOrigin-RevId:
197212466
Yifei Feng [Fri, 18 May 2018 22:43:30 +0000 (15:43 -0700)]
Merge pull request #19404 from yifeif/distributed
Make distributed_test use tf-nightly by default.
Alexandre Passos [Fri, 18 May 2018 22:33:00 +0000 (15:33 -0700)]
Correct dtype in resource_strided_slice_assign
PiperOrigin-RevId:
197210273
A. Unique TensorFlower [Fri, 18 May 2018 22:15:06 +0000 (15:15 -0700)]
Remove unused BUILD dependencies
PiperOrigin-RevId:
197207799
Jianwei Xie [Fri, 18 May 2018 21:59:01 +0000 (14:59 -0700)]
Fixed an issue when add context into params.
PiperOrigin-RevId:
197205327
Yifei Feng [Fri, 18 May 2018 21:25:59 +0000 (14:25 -0700)]
Make distributed_test use tf-nightly by default.
annarev [Fri, 18 May 2018 21:12:26 +0000 (14:12 -0700)]
Adding back abi and stacktrace dependencies to stacktrace_handler (#19401)
Yu-Cheng Ling [Fri, 18 May 2018 20:12:41 +0000 (13:12 -0700)]
Revert a change to fix TFLite iOS demo app.
It depends on released CocoaPod.
PiperOrigin-RevId:
197189734
Suharsh Sivakumar [Fri, 18 May 2018 20:07:18 +0000 (13:07 -0700)]
The quantizer should match the patterns for partition variables.
PiperOrigin-RevId:
197189118
A. Unique TensorFlower [Fri, 18 May 2018 19:41:20 +0000 (12:41 -0700)]
[dataset]: Remove extra `repeat` in the docstring for `shard`.
PiperOrigin-RevId:
197185877
Dimitris Vardoulakis [Fri, 18 May 2018 19:24:20 +0000 (12:24 -0700)]
[TF:XLA] Delete the reverseDFS scheduler. With recent improvements to the List scheduler, reverseDFS is no longer needed.
PiperOrigin-RevId:
197183727
A. Unique TensorFlower [Fri, 18 May 2018 19:21:44 +0000 (12:21 -0700)]
Expose read-only versions of input tensors in tflite.
PiperOrigin-RevId:
197183345
Joshua V. Dillon [Fri, 18 May 2018 19:17:05 +0000 (12:17 -0700)]
Enable `SeedStream` construction from other `SeedStream` instances.
PiperOrigin-RevId:
197182686
Igor Ganichev [Fri, 18 May 2018 19:14:47 +0000 (12:14 -0700)]
Add eager tests for automatic tensor flattening
PiperOrigin-RevId:
197182339
Mustafa Ispir [Fri, 18 May 2018 19:09:55 +0000 (12:09 -0700)]
Add an in-memory evaluator for Estimator. It will run evaluation without a checkpoint. This will let user get evaluation metrics on more steps without saving.
Example:
```python
def train_input_fn():
...
return train_dataset
def eval_input_fn():
...
return eval_dataset
estimator = tf.estimator.DNNClassifier(...)
evaluator = tf.contrib.estimator.InMemoryEvaluatorHook(
estimator, eval_input_fn)
estimator.train(train_input_fn, hooks=[evaluator])
```
PiperOrigin-RevId:
197181726
A. Unique TensorFlower [Fri, 18 May 2018 19:08:32 +0000 (12:08 -0700)]
When converting an unsupported op in TOCO, we can possibly infer the output type from an attribute named Tout (this works, e.g., for RandomUniformInt).
PiperOrigin-RevId:
197181552
Sanjoy Das [Fri, 18 May 2018 18:56:28 +0000 (11:56 -0700)]
[TF:XLA] Bump open source llvm revision to r332697
PiperOrigin-RevId:
197179581
Younghee Kwon [Fri, 18 May 2018 18:46:54 +0000 (11:46 -0700)]
boosted_trees: allow labels to have rank 1 as other estimators.
PiperOrigin-RevId:
197178169
Peter Hawkins [Fri, 18 May 2018 18:43:25 +0000 (11:43 -0700)]
[TF:XLA] Remove underscore prefix from XlaLaunch operator.
Minor fixes to comments.
PiperOrigin-RevId:
197177582
Ruoxin Sang [Fri, 18 May 2018 18:28:18 +0000 (11:28 -0700)]
Validate the file signatures before issusing a content read in GCS. Remove the
checkpoint hack. Enable the stat cache by default, as the block cache has
already been enabled by default. Set the default stat cache max_age to 5s,
which is the same in the Cloud TPU server.
PiperOrigin-RevId:
197175258
A. Unique TensorFlower [Fri, 18 May 2018 17:42:50 +0000 (10:42 -0700)]
Automated g4 rollback of changelist
197118212
PiperOrigin-RevId:
197167501
Yu-Cheng Ling [Fri, 18 May 2018 17:39:45 +0000 (10:39 -0700)]
Op version: Populate version in Toco TFLite exporter
PiperOrigin-RevId:
197166962
Nick Desaulniers [Fri, 18 May 2018 17:23:34 +0000 (10:23 -0700)]
[TF:XLA] Allow the HloEvaluator to be changed for the HLO interpreter.
PiperOrigin-RevId:
197164443
Rachel Lim [Fri, 18 May 2018 17:21:59 +0000 (10:21 -0700)]
[tf.data] Changed internal implementation of `make_csv_dataset`, and removed arguments `default_float_type` and `comment` from `make_csv_dataset`
PiperOrigin-RevId:
197164167
Yu-Cheng Ling [Fri, 18 May 2018 17:12:37 +0000 (10:12 -0700)]
Improve TFLite generated example test speed.
PiperOrigin-RevId:
197162686
Brian Patton [Fri, 18 May 2018 15:38:40 +0000 (08:38 -0700)]
Update the docstring on function.Defun to reflect current graph/session freezing behavior.
PiperOrigin-RevId:
197150790
A. Unique TensorFlower [Fri, 18 May 2018 14:52:32 +0000 (07:52 -0700)]
Restructure BUILD rules to be modularly correct
PiperOrigin-RevId:
197145545
A. Unique TensorFlower [Fri, 18 May 2018 14:49:06 +0000 (07:49 -0700)]
Mark link-time dependency library as alwayslink=1.
PiperOrigin-RevId:
197145205
Tatiana Shpeisman [Fri, 18 May 2018 14:50:54 +0000 (07:50 -0700)]
Merge pull request #18240 from Intel-tensorflow/ashraf/direct_session_bugfix
INTEL MKL: Fixing a unit test failure, for direct session memory allocation
A. Unique TensorFlower [Fri, 18 May 2018 14:47:41 +0000 (07:47 -0700)]
Modify PadInsertion pass so that it matches other passes.
Currently, PadInsertion only iterates over the instructions in the
entry_computation. Other passes iterate over MakeNonfusionComputations.
When we run on HloSnapshots derived from TPU benchmarks, this makes a
difference, because it seems none of the convolutions are inside the entry
computation.
PiperOrigin-RevId:
197145067
A. Unique TensorFlower [Fri, 18 May 2018 13:31:20 +0000 (06:31 -0700)]
Dropping support for CUDA < 8.
PiperOrigin-RevId:
197137612
A. Unique TensorFlower [Fri, 18 May 2018 13:27:13 +0000 (06:27 -0700)]
Extracts the following optimizations into methods:
SimplifyArithmeticOperations
ReduceDivToReciprocalMul
PiperOrigin-RevId:
197137281
Ilya Biryukov [Fri, 18 May 2018 10:14:27 +0000 (03:14 -0700)]
Update downloadable clang to a new revision
PiperOrigin-RevId:
197121447
A. Unique TensorFlower [Fri, 18 May 2018 09:37:52 +0000 (02:37 -0700)]
Dropping support for cuDNN < 6.
Enable CUDNN_FFT_TILING_FORWARD for cuDNN >= 7.
PiperOrigin-RevId:
197118212
Pavithra Vijay [Fri, 18 May 2018 04:36:39 +0000 (21:36 -0700)]
Move Keras code out of _impl folder and remove API files.
PiperOrigin-RevId:
197097430
Michael Kuperstein [Fri, 18 May 2018 01:09:45 +0000 (18:09 -0700)]
[XLA] Clamp indices in DynamicSlice and DynamicUpdateSlice instead of wrapping.
This implements the following index clamping in all backends (CPU, GPU, Interpreter):
for(int i = 0; i < rank; ++i)
start_index[i] = clamp(start_index[i], 0, output_dim_size[i] - update_dim_size[i])
Which ensures the slice (or update region) is always inbounds w.r.t the input.
PiperOrigin-RevId:
197082276
A. Unique TensorFlower [Fri, 18 May 2018 00:40:47 +0000 (17:40 -0700)]
Handle HloInstruction::ToString() when literal is missing.
PiperOrigin-RevId:
197079144
Mark Daoust [Fri, 18 May 2018 00:25:47 +0000 (17:25 -0700)]
markdown fixes
PiperOrigin-RevId:
197077588
A. Unique TensorFlower [Fri, 18 May 2018 00:11:47 +0000 (17:11 -0700)]
Do not force default layout when there is no need to.
Allow the inner computations to negotiate a root and parameter layouts different from default.
END_PUBLIC
RELNOTES: n/a
---------------------
BEGIN_PUBLIC
Automated g4 rollback of changelist
194293187
PiperOrigin-RevId:
197076025
Martin Wicke [Fri, 18 May 2018 00:06:37 +0000 (17:06 -0700)]
Merge pull request #19355 from tensorflow/mrry-patch-2
Include <algorithm> in .pb_text.cc generated files
A. Unique TensorFlower [Thu, 17 May 2018 23:59:50 +0000 (16:59 -0700)]
Add NCHW_VECT_W tensor format.
PiperOrigin-RevId:
197074411
A. Unique TensorFlower [Thu, 17 May 2018 23:52:31 +0000 (16:52 -0700)]
Enable optimizations submitted during pre-NIPS freeze:
* Snapshot to Identity promotion.
* Removal of transposes pairs separated by chain.
PiperOrigin-RevId:
197073602
Younghee Kwon [Thu, 17 May 2018 23:52:17 +0000 (16:52 -0700)]
boosted_trees: fixed the crash when eval/prediction is attempted with the initial checkpoints (at step 0) before training.
PiperOrigin-RevId:
197073582
Akshay Modi [Thu, 17 May 2018 23:52:07 +0000 (16:52 -0700)]
Misc changes found when debugging an unrelated problem.
- (pywrap_tfe.i) Improve error message for better debugging TFE_Py_Execute failures.
- (pywrap_tfe_src.cc) Accept _value of None
- (base.i) Remove unnecessary temporary
PiperOrigin-RevId:
197073571
A. Unique TensorFlower [Thu, 17 May 2018 23:26:24 +0000 (16:26 -0700)]
Make V1 metrics distributed-aware. Also fix a bug where assertAllClose
was sometimes ignoring its `msg` parameter.
PiperOrigin-RevId:
197070234
A. Unique TensorFlower [Thu, 17 May 2018 23:23:33 +0000 (16:23 -0700)]
[XLA] Redesign: delete xla::Computation.
PiperOrigin-RevId:
197069851
Craig Citro [Thu, 17 May 2018 23:06:21 +0000 (16:06 -0700)]
Don't add branch prediction hints when compiling under nvcc.
As seen in #19203, the `__builtin_expect` compiler builtin isn't recognized as
a builtin in nvcc8, leading to compilation failures of the form
./tensorflow/core/kernels/gather_functor_gpu.cu.h(57): error: calling a __host__ function("__builtin_
expect") from a __global__ function("tensorflow::GatherOpKernel< ::Eigen::half, int, (bool)1> ") is n
ot allowed
when attempting to build TensorFlow.
This change fixes things by adding an additional check for `__NVCC__`, and
avoiding any branch prediction hints in that case.
PiperOrigin-RevId:
197067418
Olivia Nordquist [Thu, 17 May 2018 22:49:12 +0000 (15:49 -0700)]
fixing some incorrect documentation in test cases.
PiperOrigin-RevId:
197064874
Sanjoy Das [Thu, 17 May 2018 22:47:30 +0000 (15:47 -0700)]
Introduce an "indexed array" analysis
Context: we want to optimize computations hanging off of a embedding lookup from
a constant array. For instance, consider:
embedding = gather from a constant array using non-constant indices
embedding_reshaped = reshape embedding
embedding_reshaped_transposed = transpose embedding_reshaped
result = dot(embedding_reshaped_transposed, constant)
In the graph above, depending on how the details work out, we may be able to
fold `result` into a gather from a precomputed constant array. However, it is
inconvenient to get there by incremental rewrites -- it is probably not
profitable to rewrite embedding_reshaped or embedding_reshaped_transposed [0] as
embedding lookups but we get to "see" that the dot can be rewritten only after
rewriting the reshape and the transpose.
This analysis aims to make the optimization above more straightforward by
allowing a transformation pass (that uses this analysis) to query the analysis
to see if if `result` _can_ be represented as an embedding lookup. If yes it
can then apply some profitability heuristics to decide if it is worth it to
rewrite it as one. This suggested workflow gives us separation of concerns (the
legality of the rewrite is computed separately from its profitability) and, more
importantly, lets us "look ahead" and analyze the dot without rewriting its
operands.
The implementation is far from complete (most of the interesting bits are TODO)
but I wanted to get an early design review before I spent too much time on this.
[0] Under the assumption that transposing or reshaping are not expensive enough
to pay the price of keeping around a new potentially large constant (in
particular, some of these may have been equivalent to free bitcasts).
PiperOrigin-RevId:
197064648
Youlong Cheng [Thu, 17 May 2018 22:47:24 +0000 (15:47 -0700)]
Add more test cases in function_test
PiperOrigin-RevId:
197064629
A. Unique TensorFlower [Thu, 17 May 2018 22:34:00 +0000 (15:34 -0700)]
Delete unused proto schema "graph_explorer.proto".
PiperOrigin-RevId:
197062616
Igor Saprykin [Thu, 17 May 2018 22:29:03 +0000 (15:29 -0700)]
Support combinations stored in a global variable in combinations.py-based tests.
The user's input is modified to give the test a better name. If the user stores combinations in a variable and applies those combinations to more than one test, then the test case name will be a concatenation of of the names of the previous test cases.
PiperOrigin-RevId:
197061902
David Majnemer [Thu, 17 May 2018 22:28:33 +0000 (15:28 -0700)]
[XLA] Use Expm1 in Elu/Selu
exp(x) - 1 is best executed using the composed Expm1 operation as it is better
behaved when exp(x) is near 1.
PiperOrigin-RevId:
197061826
Akshay Modi [Thu, 17 May 2018 22:28:09 +0000 (15:28 -0700)]
Move runtime.{h,cc,_test.cc} into core/common_runtime/eager as attr_builder
I'm not familiar with how the CMake build is set up but from the description
of the problem the dependency graph is coarser than Bazel's, so I think this
should fix #18925.
PiperOrigin-RevId:
197061764
A. Unique TensorFlower [Thu, 17 May 2018 22:10:25 +0000 (15:10 -0700)]
Clean up CFG test suite
PiperOrigin-RevId:
197059149
Jiri Simsa [Thu, 17 May 2018 22:12:24 +0000 (15:12 -0700)]
Merge pull request #19294 from yongtang/
03122018-MapAndBatchDataset
Add shape validation in shape function of MapAndBatchDataset
Blake Hechtman [Thu, 17 May 2018 22:08:33 +0000 (15:08 -0700)]
[XLA] Remove eq_shapes from Identical SlowPath since it is already checked in
Identical.
PiperOrigin-RevId:
197058888
Yu-Cheng Ling [Thu, 17 May 2018 21:58:04 +0000 (14:58 -0700)]
Support Bool in Cast (TFLite)
PiperOrigin-RevId:
197056978
Rachel Lim [Thu, 17 May 2018 21:37:34 +0000 (14:37 -0700)]
Added CSV parsing to the tf.data programmers guide
Relevant to #278.
PiperOrigin-RevId:
197053723
Allen Lavoie [Thu, 17 May 2018 21:32:47 +0000 (14:32 -0700)]
Basic usability fixes for RNNCell wrappers
They weren't calling their parent constructors (for the Keras base Layer), so a bunch of their methods threw odd errors. There may still be issues, but hopefully not so blatent.
Fixes #19208. For real this time.
PiperOrigin-RevId:
197052962
A. Unique TensorFlower [Thu, 17 May 2018 21:31:29 +0000 (14:31 -0700)]
Fix L2Normalization.
PiperOrigin-RevId:
197052728
Nick Desaulniers [Thu, 17 May 2018 21:22:49 +0000 (14:22 -0700)]
[TF:STREAM_EXECUTOR] move declaration of multi_platform_manager to header
PiperOrigin-RevId:
197051272
Allen Lavoie [Thu, 17 May 2018 21:11:14 +0000 (14:11 -0700)]
Automated g4 rollback of changelist
197026249
PiperOrigin-RevId:
197049255
Michael Kuperstein [Thu, 17 May 2018 20:37:57 +0000 (13:37 -0700)]
[TF:XLA] Do not rely on implementation-defined semantics of DynamicSlice.
ReverseSequence relies on DynamicSlice wrapping around, which is implementation-defined behavior, and is not guaranteed. Pad the input instead.
PiperOrigin-RevId:
197043307
A. Unique TensorFlower [Thu, 17 May 2018 20:12:51 +0000 (13:12 -0700)]
Allows users to specify allow_custom_ops when calling tf.contrib.lite.toco_convert().
PiperOrigin-RevId:
197039477
Michael Case [Thu, 17 May 2018 20:02:08 +0000 (13:02 -0700)]
Remove -DGEMMLOWP_ALLOW_SLOW_SCALAR_FALLBACK being added by default to all builds.
PiperOrigin-RevId:
197037867
A. Unique TensorFlower [Thu, 17 May 2018 19:55:02 +0000 (12:55 -0700)]
Support functools.partial as callable object in tf_inspect.getargspec.
PiperOrigin-RevId:
197036874
A. Unique TensorFlower [Thu, 17 May 2018 19:31:17 +0000 (12:31 -0700)]
[XLA] Redesign: delete Client::LoadSnapeshot(SessionModule). This is a precondition to delete xla::Computation.
PiperOrigin-RevId:
197033641
Akshay Modi [Thu, 17 May 2018 19:29:54 +0000 (12:29 -0700)]
Test some distributions stuff in Eager as well as Graph
PiperOrigin-RevId:
197033485
A. Unique TensorFlower [Thu, 17 May 2018 18:52:49 +0000 (11:52 -0700)]
Internal Change
PiperOrigin-RevId:
197028096
A. Unique TensorFlower [Thu, 17 May 2018 18:47:16 +0000 (11:47 -0700)]
Support 1x1x1xN bias sizes in TFLite's convolution and FC layers.
PiperOrigin-RevId:
197027135
A. Unique TensorFlower [Thu, 17 May 2018 18:42:18 +0000 (11:42 -0700)]
Internal change
PiperOrigin-RevId:
197026249
Skye Wanderman-Milne [Thu, 17 May 2018 18:33:36 +0000 (11:33 -0700)]
Remove C API staging from importer.py.
PiperOrigin-RevId:
197024708
Alexandre Passos [Thu, 17 May 2018 18:17:58 +0000 (11:17 -0700)]
Fix the fizzbuzz example
PiperOrigin-RevId:
197021930
A. Unique TensorFlower [Thu, 17 May 2018 18:13:00 +0000 (11:13 -0700)]
Use integral power function rather than floating point version. The integral version is faster.
PiperOrigin-RevId:
197021020
Benjamin Kramer [Thu, 17 May 2018 18:06:05 +0000 (11:06 -0700)]
[XLA:GPU] Unroll multi-output loop fusions
This is easier than I thought because we can assume that all tuple members have
the same number of elements. LLVM doesn't do a great job of vectorizing the
resulting stores, but otherwise this is working fine.
PiperOrigin-RevId:
197019718
A. Unique TensorFlower [Thu, 17 May 2018 17:56:36 +0000 (10:56 -0700)]
Change loop variable type to be deduced.
PiperOrigin-RevId:
197017789
Xiaoqiang Zheng [Thu, 17 May 2018 17:53:40 +0000 (10:53 -0700)]
Merge pull request #19321 from krantideep95/patch-1
core/framework/op_gen_lib.cc: include <algorithm>
A. Unique TensorFlower [Thu, 17 May 2018 17:15:45 +0000 (10:15 -0700)]
Change traverse_test.test_module to traverse a constructed dummy module rather than testcase itself.
PiperOrigin-RevId:
197010681
Mark Ryan [Thu, 17 May 2018 17:17:39 +0000 (18:17 +0100)]
Fix alignment crashes in AVX512 builds (#19121)
* Fix issue #15588 by simplifying the code
The allocator.h code tried to be clever and use 32 byte alignment for SSE/AVX2/etc use,
and 64 byte alignment for AVX512.
Unfortunately, the #ifdef in use (from EIGEN) is not useful; the bazel BUILD files do
not propagate the tf_copts() compiler flags when the allocator.cc/allocator.h files get
compiled, to EIGEN does not see the actual AVX512 using compiler flags...
Rather than changing compiler flag propagation throughout a whole bunch of code,
there's an opportunity to just simplify the code and always use 64 byte alignment.
Yes it wastes a bit of space, but on the other hand now these allocations are
cache line aligned which isn't a bad thing... and an ifdef can be dropped
Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
* Set EIGEN_MAX_ALIGN_BYTES=64
This patch sets a 64 byte upper bound on the alignment of memory allocated by
eigen. This is necessary to prevent crashes during the execution of the unit
tests when they are compiled with AVX512 support.
Signed-off-by: Mark Ryan <mark.d.ryan@intel.com>
* Update the tensorflow/compiler/aot tests for 64 byte alignment
Modifications to the tensorflow/core/framework/allocator.h to always
use 64 byte alignment causes failures in the tensorflow/compiler/aot
unit tests. This patch updates these tests so that they pass with
64 byte aligned allocated memory.
Signed-off-by: Mark Ryan <mark.d.ryan@intel.com>
* Update Tensor.Slice_Basic for 64 byte alignment
The test case
//tensorflow/core:framework_tensor_test:Tensor.Slice_Basic
fails with EIGEN_MAX_ALIGN_BYTES set to 64. The reason is that the
slices it takes of the sample tensor are 32 byte and not 64 byte
aligned. This commit increases one of the dimensions of the original
tensor to ensure that the slices taken by the test cases are indeed 64
byte aligned.
Signed-off-by: Mark Ryan <mark.d.ryan@intel.com>
* Update ScopedAllocatorConcatOpTest.Reshape for 64 byte alignment
The ScopedAllocatorConcatOpTest.Reshape test requires that the elements
of the field_shapes parameter of ExecOp are multiples of
Allocator::kAllocatorAlignment in size. If they are not, the backing
tensor allocated by PrepOp will have too many elements and reshaping
will fail. This commit modifies the test case, making the elements
64 bytes in size, the new value for Allocator::kAllocatorAlignment.
Signed-off-by: Mark Ryan <mark.d.ryan@intel.com>
Geoffrey Irving [Thu, 17 May 2018 17:16:07 +0000 (10:16 -0700)]
Add tf.print an an alias for tf.Print (#19234)
Users with Python 3 or `from __future__ import print_function` can now
use lowercase `tf.print`. `create_python_api.py` needed some adjustment
to ensure that `print_function` doesn't appear as part of the API.
Fixes #18053.
Alexandre Passos [Thu, 17 May 2018 17:02:06 +0000 (10:02 -0700)]
Allows the fizzbuzz example to work when called as fizzbuzz(tf.constant(10)).
Fixes #18960
PiperOrigin-RevId:
197008373
Tom Hennigan [Thu, 17 May 2018 16:57:16 +0000 (09:57 -0700)]
Rename private push/pop API and use from `stop_recording` method.
PiperOrigin-RevId:
197007561
Nick Desaulniers [Thu, 17 May 2018 16:54:17 +0000 (09:54 -0700)]
[TF:XLA] remove re-initializations of Literals
It's an antipattern to have:
auto x = Literal::CreateFromShape(my_shape);
x->Populate();
as that results in initialization followed by reinitialization. Can be replaced
with:
auto x = MakeUnique<Literal>(my_shape);
x->Populate();
Suggested-by: Kay Zhu <kayzhu@google.com>
PiperOrigin-RevId:
197007127
Derek Murray [Thu, 17 May 2018 16:32:02 +0000 (09:32 -0700)]
Improve the error message printed when a WorkerService::GetStatus() call fails on session creation.
PiperOrigin-RevId:
197003951
Skye Wanderman-Milne [Thu, 17 May 2018 16:28:35 +0000 (09:28 -0700)]
Improvements to function._FuncGraph.
* Adds 'inputs', 'outputs', and 'name' field to _FuncGraph. This
allows _FuncGraph to encapsulate all the information needed to
convert it to a FunctionDef.
* Refactor logic for converting a Python callable to a _FuncGraph into
a new method, func_graph_from_py_func().
These changes are in preparation for converting tf.cond to emit an If
op. By exposing _FuncGraph functionality outside of _DefinedFunction,
_FuncGraphs can be used to represent functions that are manipulated
(e.g. to output intermediate tensors) before being converted to
FunctionDef protos.
PiperOrigin-RevId:
197003496
A. Unique TensorFlower [Thu, 17 May 2018 16:26:18 +0000 (09:26 -0700)]
Remove misleading declaration-as-default that results in a deleted constructor, and a misguided comment.
PiperOrigin-RevId:
197003162
A. Unique TensorFlower [Thu, 17 May 2018 16:25:14 +0000 (09:25 -0700)]
[XLA] Adds HloLivenessAnalysis and HloModuleDCE.
HloLivenessAnalysis marks all live instruction outputs (i.e. tuple elements) for all instructions in an HloModule, propagating live values across computation boundaries.
HloModuleDCE sweeps through each instructions dead tuple elements, eliminating dead code (currently removes dead tuple elements from while loops, but could be extended to do the same for call instructions).
PiperOrigin-RevId:
197003043
Skye Wanderman-Milne [Thu, 17 May 2018 16:22:24 +0000 (09:22 -0700)]
Update SessionTest.testFeedShapeCompatibility to work with C API enabled.
This test got lost in the transition. Prior to enabling the C API,
some constant node whose values were used for shape inference would be
marked as unfeedable in tensor_util.constant_value
(https://github.com/tensorflow/tensorflow/blob/r1.8/tensorflow/python/framework/tensor_util.py#L810).
This shape inference path is no longer used with the C API enabled, so
the constant node is successfully fed, triggering a runtime shape error.
This is arguably a regression, but given that the Python code wouldn't
mark all nodes evaluated during shape inference as unfeedable, it
seems ok to relax the check a little more.
PiperOrigin-RevId:
197002741
A. Unique TensorFlower [Thu, 17 May 2018 16:11:16 +0000 (09:11 -0700)]
Avoid accessing platform/default directly.
PiperOrigin-RevId:
197001347
A. Unique TensorFlower [Thu, 17 May 2018 15:52:36 +0000 (08:52 -0700)]
Fixed bug in tf.pad shape logic that made it more restrictive than necessary by interpreting 0 as False.
PiperOrigin-RevId:
196998883
Alexandre Passos [Thu, 17 May 2018 15:23:10 +0000 (08:23 -0700)]
Methods to stop and reset tf.GradientTape()
PiperOrigin-RevId:
196995160
Benjamin Kramer [Thu, 17 May 2018 15:00:54 +0000 (08:00 -0700)]
[TF:XLA] Bump open source llvm revision to r332584
PiperOrigin-RevId:
196992500