signature=e549cd3eb2d433172fe0dd12502b728d,Pull in tensorflow release v1.5.0

Squashed commit of the following:

commit 37aa430d84ced579342a4044c89c236664be7f68

Merge: a539e808f4 dbfa5f5594

Author: Amit Patankar

Date: Thu Jan 25 14:22:10 2018 -0800

Merge pull request #16421 from nfelt/update-tb-dep

Update tensorboard dep to >= 1.5.0, < 1.6.0

commit a539e808f427b6120051fb4b6c6825caba9b836c

Merge: 09e5ff33b9 45a2fe1e98

Author: Amit Patankar

Date: Thu Jan 25 14:21:50 2018 -0800

Merge pull request #16422 from panyx0718/r1.5

Windows: Add missing dependencies in lib_proto_parsing

commit 45a2fe1e98c679e02ba037b4fb1933e16a1e3256

Author: A. Unique TensorFlower

Date: Thu Jan 25 13:28:55 2018 -0800

Windows: Add missing dependencies in lib_proto_parsing

commit dbfa5f559448a9466187be472d8a54eee1f69b46

Author: Nick Felt

Date: Thu Jan 25 13:11:07 2018 -0800

Update tensorboard dep to >= 1.5.0, < 1.6.0

commit 09e5ff33b920c83bb69f04f60eb1ce0e74a33ce6

Merge: ccc289484a e52f17b1e2

Author: Gunhan Gulsoy

Date: Thu Jan 25 10:20:30 2018 -0800

Merge pull request #16389 from panyx0718/r1.5

cherrypick bfloat16 changes

commit e52f17b1e273eaafbddd2581c65a535a198918e0

Author: A. Unique TensorFlower

Date: Tue Jan 16 15:52:12 2018 -0800

Enable bfloat16 for CPU kernels

PiperOrigin-RevId: 182124532

commit 2fad3428e3b13c963375970cbfa9eea554a16486

Author: A. Unique TensorFlower

Date: Tue Dec 19 15:50:58 2017 -0800

Add comparison ufuncs for numpy bfloat16 type.

Fix 2 bfloat16 tests.

PiperOrigin-RevId: 179614898

commit 764f90eca2ac7bdb858c30888ca7b4bb0c47dee1

Author: A. Unique TensorFlower

Date: Fri Dec 15 10:09:07 2017 -0800

Fix bfloat16 numerics issues in the tests.

PiperOrigin-RevId: 179207115

commit 1323fc76958c9bf4907cf8bc53c9db0a85fb7d9b

Author: A. Unique TensorFlower

Date: Tue Jan 16 18:20:02 2018 -0800

Remove a duplicate bfloat16 registration.

PiperOrigin-RevId: 182143007

commit 2c3205a96c50e2488815b6e8dd8d8c18dca5d431

Author: A. Unique TensorFlower

Date: Thu Dec 14 11:18:05 2017 -0800

Enable bfloat16 tests and add a filter for currently

failed tests.

PiperOrigin-RevId: 179069257

commit 95a568febcee480cd6d4e6a6bd687754b1ca1422

Author: A. Unique TensorFlower

Date: Thu Jan 18 14:20:14 2018 -0800

Make bfloat16 work correctly with matmul

PiperOrigin-RevId: 182437226

commit ccc289484a91475d3f80165bb7a590cf9bf00aaf

Merge: 53412ea2ce f3a5ffdf5c

Author: Amit Patankar

Date: Wed Jan 24 15:14:38 2018 -0800

Merge pull request #16376 from angersson/cherrypicks

Apply non-bfloat-related final 1.5.0 cherry-picks.

commit f3a5ffdf5cd368dfaf93c68d35da2e2ed5bcf0dd

Merge: 0c97ad742c 53412ea2ce

Author: Amit Patankar

Date: Wed Jan 24 14:19:22 2018 -0800

Merge branch 'r1.5' into cherrypicks

commit 53412ea2ce71ef988b75eb9c82bbea42ccd1f2fa

Merge: 04b3f6cb70 c75a6087c1

Author: Amit Patankar

Date: Wed Jan 24 14:19:00 2018 -0800

Merge pull request #16378 from angersson/version-names

Update version names to 1.5.0 from 1.5.0-rc1

commit c75a6087c19ed7203081af0b0f0bdb215aedfc00

Author: Austin Anderson

Date: Wed Jan 24 13:24:18 2018 -0800

Update version names to 1.5.0 from 1.5.0-rc1

commit 0c97ad742c176b3854c03067afc42eac20810fe9

Author: Austin Anderson

Date: Wed Jan 24 12:52:02 2018 -0800

Remove note about Ubuntu 14 incompatibility due to future change plans

commit 474929984808916ee89e26c8007081aa0975fe80

Author: A. Unique TensorFlower

Date: Wed Jan 10 18:32:40 2018 -0800

Added the "Getting Started with TensorFlow for ML Beginners" chapter to Get

Started home page.

PiperOrigin-RevId: 181548668

commit a1e777e379170ca10f422d78c75d9c709eb692c1

Author: Nick Felt

Date: Thu Jan 18 15:50:16 2018 -0800

Update tensorboard dependency to minimum of 0.4.0

This should address https://github.com/tensorflow/tensorboard/issues/877.

PiperOrigin-RevId: 182451796

commit b2637597c5f23bbd3f5a71f9ec91b65898ea896f

Author: Yilei Yang

Date: Thu Dec 14 11:12:52 2017 -0800

Continue to allow flag access before explicit parse.

Made tf.flags.FLAGS a wrapper of absl.flags.FLAGS, when the flag is access, parse flags implicitly with sys.argv if not yet.

PiperOrigin-RevId: 179068530

commit 04b3f6cb70b2255f0502b494bcbede2c34e1980a

Merge: c8d15e2963 cc0922c1af

Author: Amit Patankar

Date: Thu Jan 18 12:31:55 2018 -0800

Merge pull request #16223 from av8ramit/docker_cherrypick

Fixing a typo for the argument to docker push. (#16204)

commit cc0922c1af35a850077966d7fe95dfd5c208c4c4

Author: Amit Patankar

Date: Thu Jan 18 10:27:07 2018 -0800

Fixing a typo for the argument to docker push. (#16204)

commit c8d15e296358095ce804f1a5825b2cdc9f551f7c

Author: Amit Patankar

Date: Tue Jan 16 14:58:48 2018 -0800

Updating the docker login command. The email flag is deprecated. (#16171)

PiperOrigin-RevId: 181769938

commit 4b5bdc85e4ba2d2cc98398fc9843db1631e04e80

Author: Asim Shankar

Date: Sat Jan 13 10:36:03 2018 -0800

Address bad merge in Java install instructions (#16096)

commit b1f157f4d2d871f7a6d8eeb21fddf97b5216608a

Author: Amit Patankar

Date: Fri Jan 12 16:29:30 2018 -0800

Disable flaky model_analyzer_test on windows. (#16089)

PiperOrigin-RevId: 180599588

commit f9472619f69cdd224b4801468b19cd578c3e3ec1

Author: Austin Anderson

Date: Fri Jan 12 12:51:35 2018 -0800

Update important RELNOTES and author list (#16078)

commit c678970e351417ba1145fe53c92519f8c64830dc

Author: Austin Anderson

Date: Fri Jan 12 11:24:46 2018 -0800

Update version strings and revert pom.xml changes (#16022)

commit aebb7cc8f5b065de06f9209a9b0b601b5b83cf70

Author: Austin Anderson

Date: Thu Jan 11 16:38:14 2018 -0800

Apply 1.5-rc1 cherry-picks. (#16056)

* Delete empty api_guides.

PiperOrigin-RevId: 179215745

* Java: Instructions for using GPUs via Maven.

GPU support in Maven is being packaged with 1.5.0-rc0 onwards

(for Linux)

Fixes #12909

PiperOrigin-RevId: 180859336

* Fix build issues with cuda 9.1 through updating eigen. (#15796)

* Revert "Fix the headers error due to recent CUDA9.1 change (#15739)"

This reverts commit 3bc4900e7e60f43dc901523f1574f52440e7e701.

* Bump eigen dependency.

* Minor change to make tpu.rewrite compatible with Python 3.

AttrValue is a byte array, and handling this is different between Python 2 and 3.

PiperOrigin-RevId: 180306415

* TensorFlow for NVIDIA Tegra devices with CUDA support (#14167)

This commit enables CUDA support on compatible devices running Android such as the Nvidia TX1 and TX2 when using Makefile builds.

Note that JetPack for Android is required to build/run Android TF binaries with CUDA support. This should be released by Nvidia in the near future.

* Adding cuda_config.h to the pip package. (#15961)

* Adding cuda_config headers to our GPU build.

* Updating the local cuda path for cuda_headers.

* Removing the cuda_config blacklist.

* Buildifier fix.

* Ignoring .so files and manually adding the cuda_config.h file.

* Fixing the path for the src_dir.

* One last minor fix for path.

* Adding brackets.

* Minor fixes for "Linear" tutorial

PiperOrigin-RevId: 179061248

* Sync Premade and Custom estimator docs with example code.

PiperOrigin-RevId: 179404175

* rename files

PiperOrigin-RevId: 179683700

* Modernize old "get_started/get_started.md", as "programmers_guide/low_level_intro.md".

PiperOrigin-RevId: 179807033

* Add links to low level API intro

PiperOrigin-RevId: 179844300

* Make images larger

PiperOrigin-RevId: 181034398

* minor fixes to new "low_level_intro"

PiperOrigin-RevId: 181172455

* typo

PiperOrigin-RevId: 181185642

* Replace get_started

Also add sub-sections to leftnav files,

and sync leftnav and index files.

PiperOrigin-RevId: 181394206

* Added a "Getting Started with TensorFlow for ML Beginners" chapter to Get

Started section.

PiperOrigin-RevId: 181396430

* Add support for CUBLAS_TENSOR_OP_MATH in fp16 GEMM (#13451)

- Applies to matrix multiplications with fp16 input/output.

Computations will fall back to pseudo-fp16 if tensor op math is

disabled or not supported.

- Enabled by default. Tensor ops (both in cublas gemms and cudnn

convolutions) can be disabled globally by setting the

environment variable TF_DISABLE_TENSOR_OP_MATH=1. To disable

tensor ops specifically for gemms or convolutions use

TF_DISABLE_CUBLAS_TENSOR_OP_MATH=1 or

TF_DISABLE_CUDNN_TENSOR_OP_MATH=1, respectively.

- Added CUBLAS 9.0 algorithms to GetBlasGemmAlgorithms().

* Adding page to tensorflow.org with directions for building the TFLite demo on Android.

PiperOrigin-RevId: 179970218

commit 0a1cbfdc6d22b1d64b13a44abbc4a34f2fa02c84

Merge: 7897a4ec85 5aa2baac54

Author: Shanqing Cai

Date: Thu Jan 11 09:48:41 2018 -0500

Merge pull request #15922 from si-you/r1.5

Add clean_dep to a bazel macro.

commit 7897a4ec853d5eb5358c90f85b114eda113818b4

Author: Amit Patankar

Date: Wed Jan 10 14:51:21 2018 -0800

Disable all failing tests to fix TF opensource tests. (#16017)

* Disable all failing tests to fix TF opensource tests.

PiperOrigin-RevId: 181212111

* Disabling the test for python3.5 as well. (#16016)

commit 0d25e135105f336475ef73c7e5b2e32ff60463f3

Author: Amit Patankar

Date: Wed Jan 10 12:15:12 2018 -0800

Adding an install sources line for 1.5.0-rc0. Earlier we only updated… (#15959)

* Adding an install sources line for 1.5.0-rc0. Earlier we only updated this for official.

* Forgot to include windows.

commit 9e2a7172e742c879b32de37ab05ac01f2fcd1c48

Author: Austin Anderson

Date: Wed Jan 10 11:29:27 2018 -0800

Add internal release notes that were previously missing. (#15988)

* Update release notes including missing internal changes.

* Address @av8ramit's comments and add more note cleanup

commit 5aa2baac544621678eddc944e6d8b774a95755cf

Author: Seungil You

Date: Sat Jan 6 20:34:09 2018 -0800

Add clean_dep to a bazel macro.

commit 793280a0a378e3cbaf558a7d8ef320f227287d11

Author: Derek Murray

Date: Fri Jan 5 10:55:50 2018 -0800

Add release note: sparse tensors in `tf.data` (#15884)

commit 622487f55481fd914bbf8f340c44ff2bb1d059de

Author: Amit Patankar

Date: Tue Jan 2 16:00:43 2018 -0800

Fixing the item_test failure. (#15799)

commit 78ab389f8c446ec4f915e5ca5f0cba928762bb41

Author: Amit Patankar

Date: Tue Jan 2 14:43:09 2018 -0800

Fix kmeans gpu and upgrading TF base images to 16 (#15794)

* Upgrade all TF base images to ubuntu 16. (#15458)

* Upgrade all TF base images to ubuntu 16.

* Update bazel version for CI docker image.

* Fix links to pylint in sanity build script.

* Upgrade bazel version to 0.8 on devel docker images.

* Upgrading scikit-learn to avoid the bz2 dependency issue.

commit 68d43cc4af3dfdb2f5e08795b13e76a8a5115ed0

Author: Amit Patankar

Date: Thu Dec 28 10:54:30 2017 -0800

Cherrypicks (#15700)

* Hack around a bug caused by bazel+tempfile+multiprocessing.

PiperOrigin-RevId: 180207077

* Bump the limit on checkpopint_utils_test as it is failing consistently with the current number by a small margin.

PiperOrigin-RevId: 180207280

* Enabling tests to pass with python3.6. Updating dependencies for dock… (#15676)

* Enabling tests to pass with python3.6. Updating dependencies for docker tests.

* Removing an unnecessary update.

commit 0fe7a0b71376a9d33d83b88d804702002b71154c

Author: Amit Patankar

Date: Wed Dec 27 14:43:45 2017 -0800

Revert "Fix a bug: bfloat16 is unsigned on Windows (#15302)" (#15663)

This reverts commit fdf34a88bec9645473f10ba2d52df4cfcb80d582.

commit f826bccee6793e1bdf3f92c92a6d2a2eb7d3fe39

Author: Brennan Saeta

Date: Wed Dec 20 14:47:04 2017 -0800

Truncate status messages before converting to gRPC.

PiperOrigin-RevId: 179736746

commit a37ef3d3302f06879100b77c1a8dd8dd58eb5284

Author: Brennan Saeta

Date: Wed Dec 20 13:29:07 2017 -0800

Add prefetching into parallel_interleave

This change adds 2 parameters to parallel_interleave:

- prefetch_input_elements: determines the number of iterators to prefetch

allowing buffers to warm up and data to be pre-fetched without blocking

the main thread (i.e. the GetNext() call).

- buffer_output_elements: in order to avoid creating thousands of threads, we

fuse in the .prefetch() operator as an additional parameter. The value of

this parameter is identical to the value passed to `.prefetch()`

PiperOrigin-RevId: 179726088

commit 29c74939eed478690410150292f511c70f359ff1

Author: A. Unique TensorFlower

Date: Tue Dec 19 12:38:19 2017 -0800

Protect all calls to launch cuSolver & cuBlas kernels by a lock. The code appears not to be threadsafe pre Cuda 9, and we have several report of crashes. Since the overhead is modest, better to be safe.

PiperOrigin-RevId: 179589983

commit 6aeee3e9cde652466367445c0c9bb08af19bc1fc

Author: Allen Lavoie

Date: Fri Dec 15 13:18:52 2017 -0800

Fix reference counts when watching variables (eager tape)

Adds unit test assertions that variables are properly dealloacted once the tape is deleted and there are no remaining Python references.

PiperOrigin-RevId: 179231752

commit be85b7fd0dfeee6bd13f82b07bc4e2dbddf263af

Author: Allen Lavoie

Date: Fri Dec 15 11:39:54 2017 -0800

Plug an eager memory leak, add tests for reference counts.

There are still some slightly less serious leaks. Will follow up with a fix once I track those down.

PiperOrigin-RevId: 179220052

commit 6e08980837212c56c73d2b6dca024f2a91831c8f

Author: Gunhan Gulsoy

Date: Wed Dec 20 13:37:38 2017 -0800

Revert more changes to tested source configurations table.

commit 1237158f4127570686a1549359b6e7620726dcff

Author: Gunhan Gulsoy

Date: Wed Dec 20 11:18:04 2017 -0800

Revert changes to tested source configurations table.

commit cf1e7d6249b0f14528654efed4735c37a7b5c43c

Author: Gunhan Gulsoy

Date: Mon Dec 18 22:29:17 2017 -0800

Linkify eager and tflite in release notes.

commit 9da14e50c374e252190af5ea7256d12af9054ff7

Author: Gunhan Gulsoy

Date: Mon Dec 18 22:25:58 2017 -0800

Update version string to 1.5.

commit 97608913fa5f9db1a6101b7b66b3de5d58b64a95

Author: Gunhan Gulsoy

Date: Mon Dec 18 20:47:24 2017 -0800

Update release notes for 1.5 release.

commit a57ba6e9337e217ed159378aabe8c851d7770aca

Author: Gunhan Gulsoy

Date: Mon Dec 18 16:56:32 2017 -0800

Disable failing testcases on windows. (#15462)

commit 0d3713e808ac27b547619bd850713eb28ff26eb4

Author: Yan Facai (颜发才)

Date: Tue Dec 19 08:52:09 2017 +0800

`variable_scope` use `auxiliary_name_scope` to control whether to create new name scope (#14390)

* TST: add test case

* ENH: variable_scope supports auxiliary_name_scope

* DOC: add document

* ENH: support to reuse name scope given by user

* TST: add test for name scope

* DOC: revise docment for name scope

* CLN: revise TypeError for None

* TST: add test for reentering variable_scope

* BUG: root scope reuse name scope

* TST: add test case for root scope

* TST: remove name scope tests

* ENH: remove name scope support

* DOC: remove name scope support

* TST: split and add more test cases

* BUG: root name scope is empty string

* CLN: remove comment

commit 662ac41e2513756ee63467c313eaad8e53e05dcd

Merge: acaabdfe58 c25bf74a3e

Author: Yifei Feng

Date: Mon Dec 18 16:27:41 2017 -0800

Merge pull request #15344 from javierluraschi/lite-readme-toco-link

Fix broken link in tensorflow lite readme

commit acaabdfe587de35ee66a612b3bbcbafef2dcca89

Merge: 96cfe81f7b b9e021d07e

Author: Yifei Feng

Date: Mon Dec 18 14:27:33 2017 -0800

Merge pull request #15444 from snnn/fix1

Fix lib_strings_str_util_test on Windows

commit 96cfe81f7b013e326582de8abcc8c474eb3e065a

Merge: 212cf55130 1aa3b54977

Author: Yifei Feng

Date: Mon Dec 18 14:10:20 2017 -0800

Merge pull request #15439 from snnn/m5

Add an is_external arg to tf_copts

commit 212cf55130f54f41bc57821fd247188fba7868a8

Merge: 4cdc52964a 768729ac47

Author: Martin Wicke

Date: Mon Dec 18 14:09:51 2017 -0800

Merge pull request #15459 from gunan/api

Enable api compatibility test to also run on macos.

commit 4cdc52964ae24f2c2b863939b4dc9cbe9cd9d104

Merge: 4d75d36f9d 88bec2d66a

Author: Yifei Feng

Date: Mon Dec 18 13:52:44 2017 -0800

Merge pull request #15394 from qmick/patch-2

Fix typo of tf.abs docstring

commit 4d75d36f9d17072500c3e8db259617c0e75a18c4

Merge: af0a705184 7d2a601fb5

Author: Yifei Feng

Date: Mon Dec 18 13:50:54 2017 -0800

Merge pull request #15386 from betterenvi/fix-gan-api-example

Fix api usage in examples of GAN

commit af0a705184fdce8cbcfc41fb79d21f2a93645e57

Merge: cfd0d3f2aa bcb326c26e

Author: Yifei Feng

Date: Mon Dec 18 13:47:28 2017 -0800

Merge pull request #15454 from AndreiCostinescu/patch-3

Update core.py

commit 768729ac47fd8bb84a195a1c32954de84434977b

Author: Gunhan Gulsoy

Date: Mon Dec 18 13:09:23 2017 -0800

Enable api compatibility test to also run on macos.

commit cfd0d3f2aa24b3078d2e79ad0a212c7c53916de9

Merge: aed215d721 5287eb13f3

Author: Martin Wicke

Date: Mon Dec 18 10:37:25 2017 -0800

Merge pull request #12369 from yongtang/12344-RecordInput-compression

Add compression support to RecordInput

commit aed215d72110269fa4ba513092717a62bbbb0af5

Author: Neal Wu

Date: Mon Dec 18 09:28:58 2017 -0800

Fix the CODEOWNERS file syntax (#15411)

commit bcb326c26e7986e7bb411e639ee3fb2fd9f53c04

Author: Andrei Costinescu

Date: Mon Dec 18 18:14:04 2017 +0100

Update core.py

Corrected the documentation of the Dense layer, regarding the computation performed by the layer

commit b9e021d07eb92c3f9fa6302224da7ecca66d7545

Author: Changming Sun

Date: Mon Dec 18 19:40:16 2017 +0800

Fix lib_strings_str_util_test on Windows

commit 1aa3b549770d05a665e8075322e84e8b8c787088

Author: Changming Sun

Date: Mon Dec 18 16:53:20 2017 +0800

Add an is_external arg to tf_copts

commit fdf34a88bec9645473f10ba2d52df4cfcb80d582

Author: Changming Sun

Date: Mon Dec 18 14:05:28 2017 +0800

Fix a bug: bfloat16 is unsigned on Windows (#15302)

commit 0f2f5b978524f3306e415d18701ea64bd2c688b3

Author: Taehoon Lee

Date: Mon Dec 18 15:03:46 2017 +0900

Fix PEP8 (#15378)

commit 943201bf1a959acf6a08b88a488b3db55404835c

Author: Gunhan Gulsoy

Date: Sun Dec 17 21:44:42 2017 -0800

Revert "Add customized kernel implementation for clip_by_value (#13998)"

This reverts commit 71d3c8e636efcf0bc3f87ed9888cca71fde8cc8e.

commit 2853a0d4a5b0d1d11aa4c68548a250b1b8084bb4

Author: Gunhan Gulsoy

Date: Sun Dec 17 21:29:01 2017 -0800

Revert "Initial SRU Implementation (#13978)"

This reverts commit e3e2ac9181c42eb82548726d8a250944b56180fd.

commit 17e725c0558581cba19bd6c409698b2c3f88efe5

Merge: 3a7c03afdf 444e319256

Author: Shanqing Cai

Date: Sun Dec 17 00:12:19 2017 -0500

Merge pull request #15313 from rajendraarora16/patch-5

Fixing typo

commit 3a7c03afdfd8d2c51bb3997a4daec01874862e71

Merge: 92233820e6 d2185de970

Author: Shanqing Cai

Date: Sat Dec 16 20:37:35 2017 -0500

Merge pull request #15333 from taehoonlee/fix_typos

Fix typos

commit 92233820e6256ffb428650e67dc8c6dc4bbc7074

Author: Yong Tang

Date: Sat Dec 16 19:04:59 2017 -0600

Fix issues in doc `tf.Placeholder` should be `tf.placeholder` (#15330)

This fix fixes issues in the doc (data_feeder.py) where

`tf.Placeholder` should be `tf.placeholder`

Signed-off-by: Yong Tang

commit e3e2ac9181c42eb82548726d8a250944b56180fd

Author: Tian Jin

Date: Sat Dec 16 20:00:27 2017 -0500

Initial SRU Implementation (#13978)

* initial tf sru implementation

* fix equation

* Refactor to BasicLSTM like

* Refactor to BasicLSTM like

* address comments and explain constraint

* address pylint concern

* fix wrong merge

commit 71d3c8e636efcf0bc3f87ed9888cca71fde8cc8e

Author: Yong Tang

Date: Sat Dec 16 17:02:58 2017 -0600

Add customized kernel implementation for clip_by_value (#13998)

This fix tries to address the issue raised in 7225 where

`tf.clip_by_value` does not have a custom kernel and reused

`tf.maximum` and `tf.mimimum`. In case scalar values are passed

to `tf.clip_by_value`, unnecessary memory might incur.

commit 5287eb13f352680449d2f759c379f459c434fdad

Author: Yong Tang

Date: Sat Dec 16 00:23:49 2017 +0000

Update API with `tensorflow/core/api_def/update_api_def.sh`

Signed-off-by: Yong Tang

commit 0b80606c4d635250810d5b2c950986f91fe1cfae

Author: Yong Tang

Date: Fri Dec 15 18:06:13 2017 -0600

Load boundaries array into shared memory before hand for `bucketize` (#14774)

This fix is a follow up to 13922. This fix loads boundaries

array into shared memory before each thread, in order to improve

performance for `bucketize` op.

The fix is based on feedback 13922.

Signed-off-by: Yong Tang

commit fe54ab55eee43814b033d0efa6334e81cced59e6

Author: Yong Tang

Date: Thu Aug 17 19:31:11 2017 +0000

Add test cases of compression_type for RecordInput.

Signed-off-by: Yong Tang

commit b831830334bc7a57cd3052f5e1ce39cb9e16f363

Author: Yong Tang

Date: Thu Aug 17 19:30:49 2017 +0000

Add compression support to RecordInput

This fix tries to fix the request raised in 12344 so that

it is possible to process RecordInput with compressions.

An attr of `compression_type` has been added.

Additional tests have been created to cover the changes.

This fix fixes 12344.

Signed-off-by: Yong Tang

commit 908343b5c0f957224ee44512d7ccfc81a8349e2f

Author: Gunhan Gulsoy

Date: Fri Dec 15 15:02:02 2017 -0800

Bump the eigen dependency version. (#15405)

Fixes #12052

commit d06ebf2ae4e37ae74317bc364d5605994169fc50

Author: Asim Shankar

Date: Fri Dec 15 13:51:03 2017 -0800

[Go]: Make op wrapper generation more robust. (#15353)

- Since Go 1.8, GOPATH has a default value, so handle

that (https://golang.org/doc/go1.8#gopath)

- generate.sh expected bash (for the string substitution syntax)

while 'sh' may point to another shell. So explicitly require bash.

commit faf7f05f5ed3d92405656a318fb2d571a7d31532

Author: Yan Facai (颜发才)

Date: Sat Dec 16 03:54:37 2017 +0800

GPU: Add Complex kernel for tf.exp() (#15182)

* TST: complex case

commit 4952f981be07b8bf508f8226f83c10cdafa3f0c4

Author: Yong Tang

Date: Fri Dec 15 13:02:54 2017 -0600

Replace loop iteration with `chip` (#15289)

In unique_op.cc, the ouput tensor was generated

through loop iteration. It seems that this could be

improved through Eigen's `chip`.

The fix addresses this improvement.

Signed-off-by: Yong Tang

commit 5ea0404f5e01ca12739df73dc86b9ac58db58f83

Merge: 51c1e36cdf 979436cd29

Author: Martin Wicke

Date: Fri Dec 15 10:37:35 2017 -0800

Merge pull request #15377 from tensorflow/yifeif-patch-1

Add description for new PR workflow.

commit 979436cd29cbe908614f1475cca344e89fcea1bc

Author: Yifei Feng

Date: Fri Dec 15 10:27:54 2017 -0800

Update CONTRIBUTING.md

commit 51c1e36cdf888abe3b51965a5ce650308c655ab8

Merge: 8d3bd70a54 9e966e9e54

Author: Martin Wicke

Date: Fri Dec 15 10:20:52 2017 -0800

Merge pull request #14510 from facaiy/BUG/index_to_string_table_from_file

index_to_string_table_from_file can use tf.string as vocabulary file

commit 8d3bd70a547fa843ba3e3da6f1379c4fdd836f2b

Author: Alex Sergeev

Date: Fri Dec 15 08:59:53 2017 -0800

Support --config=monolithic in tf.sysconfig.get_link_flags() (#15139)

* Don't add -ltensorflow_framework to sysconfig LFLAGS if the build is monolithic

commit 88bec2d66a5a2a9ebca8db9cfa4c7e95e48d2d4f

Author: Jiongyan Zhang

Date: Fri Dec 15 20:12:19 2017 +0800

Fix typo of tf.abs docstring

Change "'", which causes incorrect highlight, to "`"

commit 7d2a601fb5c5cb06173ec4fa3737a363fce58f5b

Author: CQY

Date: Fri Dec 15 15:43:55 2017 +0800

Fix api usage in examples of gan

commit 798fa36d11119e6fdc13b90a14abfe1805e7de90

Author: Gunhan Gulsoy

Date: Thu Dec 14 22:02:41 2017 -0800

Instead of "option" use "set" to define non-bool cmake build args. (#15372)

commit 00f8b97fc601381546aea89315dee549bdbbbdfc

Merge: 9713156818 5de6f68848

Author: Martin Wicke

Date: Thu Dec 14 17:57:44 2017 -0800

Merge pull request #14389 from cthom055/master

Forward declare condition_variable, fix for #14388

commit 97131568180379d416e3405e55d2e288d0f17d80

Merge: 0237257f6a c534baccd7

Author: Martin Wicke

Date: Thu Dec 14 17:57:13 2017 -0800

Merge pull request #14399 from PW486/fixed-typos

Fixed typos in comments

commit c534baccd744740147ace5ec765d02cb8a6c8366

Merge: e058a030f8 0237257f6a

Author: Martin Wicke

Date: Thu Dec 14 17:34:24 2017 -0800

Merge branch 'master' into fixed-typos

commit 0237257f6ac8021c7e4062f8f0694804a1dcbbef

Merge: 888c1dc189 c25cd200dd

Author: Martin Wicke

Date: Thu Dec 14 17:33:16 2017 -0800

Merge pull request #14401 from myungjoo/fix/options

CMake: configure default string values of options properly

commit 1d976cbd2ab76c778de994e8b23534f684f65521

Author: Yifei Feng

Date: Thu Dec 14 17:16:46 2017 -0800

Add description for new PR workflow.

commit 888c1dc18948802e20f42a710f1853a784883f72

Author: Chris Donahue

Date: Thu Dec 14 15:10:02 2017 -0800

Changed ffmpeg verbosity semantics (#14582)

commit c41b43b920fd9ee188f862e1e35d5ac5852f1747

Merge: a44c2f3281 40fc0cb025

Author: Martin Wicke

Date: Thu Dec 14 12:53:15 2017 -0800

Merge pull request #14321 from yongtang/11052017-tf.Constant

Fix issue in the `Defun` docs

commit a44c2f3281453c5210cd6d8336744b7264aa3e96

Author: Felix Abecassis

Date: Thu Dec 14 12:38:56 2017 -0800

Dockerfile.devel-gpu: optimize the size of the generated image (#15355)

- Use `nvidia/cuda:9.0-base-ubuntu16.04` as the base image to select

just the CUDA libraries we need.

- Remove the installed static libraries.

- Remove the dependency on openjdk-8 since Bazel ships with a local copy.

- Perform a shallow clone of the repository.

The image is 2.94GB, down from 4.87GB.

Signed-off-by: Felix Abecassis

commit 73658420db2498ad7f07363bfa72cba6e2d9fdd2

Author: Jonathan Hseu

Date: Thu Dec 14 11:33:34 2017 -0800

Benchmarks for flipping and random flipping (#15348)

* Random flip benchmark

* More benchmarks

commit 2fb40006976490d7323eb6be8e2bae2275948551

Author: Jonathan Hseu

Date: Thu Dec 14 11:33:21 2017 -0800

Revert "Add batch support for various image_ops (#14854)" (#15349)

This reverts commit 20aa9e0a9f129ed929cea1fb45ec12b7be3ac68e.

commit f9a173a9fa9a96eb65781b25ce29785c5f46788a

Merge: f5f2f789ea 9e993d2dc6

Author: Shanqing Cai

Date: Thu Dec 14 14:16:31 2017 -0500

Merge pull request #15173 from JohanJu/master

Fix for #12537

commit f5f2f789ea395e585ddcbc43e088fa63d6b41d0e

Merge: 2bb302ee26 d9769ce641

Author: Gunhan Gulsoy

Date: Thu Dec 14 11:10:32 2017 -0800

Merge pull request #15356 from dandelionmane/branch_178996911

Branch 178996911

commit 2bb302ee26c9389df138b82f0bebac5ee01f3c30

Author: David Norman

Date: Thu Dec 14 18:08:05 2017 +0000

[XLA] Add fast path cases for common scatter and gather operations (#15185)

* Add fast path cases for common scatter and gather operations

* Updates following code review

commit d9769ce6410c2d95f371941855c46a0b2c461c0e

Author: Dandelion Mané

Date: Wed Dec 13 22:21:25 2017 -0800

Fix syntax error I introduced when re-applying my merge fix.

commit 7cb8f7f4fb2a82d90e55ea3e71b8a45bcdf609aa

Merge: 18c864cdae eea28cb2d4

Author: Dandelion Mané

Date: Wed Dec 13 22:03:27 2017 -0800

Resolve conflicts

commit eea28cb2d458d41dfa8129afb34e4420fefc6997

Author: Eugene Brevdo

Date: Wed Dec 13 20:50:37 2017 -0800

[TF Variants] The ResourceAssignOp now supports Variants on GPU.

This allows Variants to sit on resource variables; before, though the

ReadValue op was enabled for Variants on GPU, because assignment happened

on CPU, Variant-based Resource Variables always had to reside on CPU

due to the associated colocation constraints.

PiperOrigin-RevId: 178996911

commit ed0e2508183ce1bdc1b06cbb752f6a8ed9af99e4

Author: Eugene Brevdo

Date: Wed Dec 13 20:27:14 2017 -0800

Bugfix in variable naming of GRUBlockCell.

PiperOrigin-RevId: 178995589

commit b8831d34a24b10a84c711362ae810e5faf3f27ae

Author: A. Unique TensorFlower

Date: Wed Dec 13 18:46:18 2017 -0800

Minor fix on degenerative case.

PiperOrigin-RevId: 178989673

commit 466926a9900804cd4d5d7a719ee6f8c1b3ff9a75

Author: Igor Saprykin

Date: Wed Dec 13 18:30:33 2017 -0800

Automated g4 rollback of changelist 178963334

PiperOrigin-RevId: 178988579

commit baef8c35900a01d7b9444e001482a8fbd99f1fbe

Author: A. Unique TensorFlower

Date: Wed Dec 13 18:08:58 2017 -0800

Add checkpoint_path to evaluation predicate_fn

PiperOrigin-RevId: 178986670

commit eb1bbef6c291829f2c1856a35b2556fa0f477bdb

Author: Justin Lebar

Date: Wed Dec 13 17:45:55 2017 -0800

[XLA] Add support for CustomCall in HLO parser.

PiperOrigin-RevId: 178984357

commit 69e5969d159fa8560eb61d82ec55b04d19bb0560

Author: Jacques Pienaar

Date: Wed Dec 13 17:09:29 2017 -0800

[TFXLA] Simplify identification of cond branches.

* Remove the clustered graph part as it was difficult to keep it updated with the rest of the graph and instead operate on the graph directly;

PiperOrigin-RevId: 178980836

commit f22809c16ef54be6c515c9889c88ade746146e2e

Author: Derek Murray

Date: Wed Dec 13 17:09:12 2017 -0800

Enable TF_GPU_THREAD_MODE and TF_GPU_THREAD_COUNT support in distributed TensorFlow.

PiperOrigin-RevId: 178980799

commit f675c122de5f336b3eb49bb0a74e0a0e7eb890fc

Author: Anjali Sridhar

Date: Wed Dec 13 16:41:58 2017 -0800

Update tf.keras to the 2.1.2 API.

PiperOrigin-RevId: 178977412

commit 937604abad720291e62635b6623b26ab5fb51a7e

Author: Yao Zhang

Date: Wed Dec 13 16:19:56 2017 -0800

Avoid unnecessary layout transpose to input of ShapeN.

PiperOrigin-RevId: 178974641

commit b3e97d56bd10bdf1976c61aab1f50a8902068c5c

Author: Sanjoy Das

Date: Wed Dec 13 15:54:11 2017 -0800

[XLA:CPU] Implement Ax+b dot output fusion for Matrix-vector products

I had to roll in the change to generalize CPU layout assignment as without it we

lose the make-rhs-column-major optimization and that causes a performance

regression.

PiperOrigin-RevId: 178970986

commit 8a9aaa3bf4e81fef532ea3dad860797ecf85ef93

Author: Shivani Agrawal

Date: Wed Dec 13 15:23:20 2017 -0800

Adds build rule for scan_dataset_op_test.py

PiperOrigin-RevId: 178966883

commit e2e15df4175a3c13aa550b17bad4bb5d92185e7e

Author: A. Unique TensorFlower

Date: Wed Dec 13 15:22:39 2017 -0800

[XLA] Update parser to handle conditional. Also fix the stringification of

conditional HloInstruction.

PiperOrigin-RevId: 178966782

commit d80d6de2890112f0013ffede31767907ec3291ca

Author: Peter Hawkins

Date: Wed Dec 13 15:21:11 2017 -0800

Fix bfloat16 serialization of Tensors.

Previously, Python serialization and deserialization used the half_val field of TensorProto, whereas C++ serialization used the int_val field. However, C++ bfloat16 deserialization was always broken, so it was never possible to correctly deserialize a bfloat16 Tensor.

The only reason serialization worked at all was because of the generic tensor_contents bytes serialization.

PiperOrigin-RevId: 178966536

commit b09be8eff9505486b0f838e2cb281c3ebe8ecfc6

Author: A. Unique TensorFlower

Date: Wed Dec 13 15:13:16 2017 -0800

Enable Div -> Mul by reciprocal strength reduction.

PiperOrigin-RevId: 178965261

commit 828dfee507623c9b8496904e7b94beb9c6ca8306

Author: Igor Saprykin

Date: Wed Dec 13 15:01:14 2017 -0800

Average the loss across `replicate_model_fn`'s towers.

This avoids the need for users to add `loss = loss / num_of_towers` code and is in more in line with the current best practices.

I verified this by running cnn_mnist.

PiperOrigin-RevId: 178963334

commit 57600a8d7739f6fbea445c6efa1f29f12f769748

Author: Nupur Garg

Date: Wed Dec 13 14:54:10 2017 -0800

Add support for "Pad".

PiperOrigin-RevId: 178962340

commit 7fbcd3724379161e379f4693ba66666b6ac4b0aa

Author: Sergio Guadarrama

Date: Wed Dec 13 14:50:24 2017 -0800

Avoid modifying items in dataset_data_provider.

PiperOrigin-RevId: 178961790

commit 82c1af4a693395489cb43c1bbb7c7be41af654b8

Author: A. Unique TensorFlower

Date: Wed Dec 13 14:43:00 2017 -0800

Simplify and improve AdaptiveSharedBatchScheduler implementation. The new implementation will exist alongside the old one (selectable through the scheduler options) until its superiority is confirmed, at which point the old rate-based implementation will be removed.

The new implementation requires fewer options and no user feedback to achieve a low latency batching. Instead of processing batches at an adjustable rate, we limit the number of batches which can be concurrently processed. Below the limit, batches are immediately processed upon creation. At the limit, the oldest batch is processed once an in-processing batch finishes. The scheduler continuously adjusts the limit in order to maintain the smallest overall latency.

PiperOrigin-RevId: 178960621

commit 38b20f83dbaada96902bdd5b419feb5a8e47395c

Author: A. Unique TensorFlower

Date: Wed Dec 13 14:23:40 2017 -0800

- adds support constant cast operators

- adds support for legacy "BatchMatMul" operators

- adds constant scalar values to graphviz output

PiperOrigin-RevId: 178957498

commit 9da6e87662c54d03310fc72fe7efdd2f22b87f85

Author: Akshay Agrawal

Date: Wed Dec 13 14:22:17 2017 -0800

Call convert_to_tensor on all inputs to the op for _backprop_call in graph mode

Fixes a bug in which EagerTensors were provided as input to an op.

PiperOrigin-RevId: 178957283

commit 70062d11bf11d6579bfdbc87c3350a0074a12ae8

Author: A. Unique TensorFlower

Date: Wed Dec 13 13:45:05 2017 -0800

Rename Stream::BlockHostUntilDoneWithStatus to BlockHostUntilDone.

PiperOrigin-RevId: 178951330

commit 553e8f14c8c025a8c09e0a6cb824c786bc258f56

Author: A. Unique TensorFlower

Date: Wed Dec 13 13:39:02 2017 -0800

Update Stream::BlockHostUntilDone examples and documentation.

The new Status return value must be explicitly handled or ignored.

PiperOrigin-RevId: 178950527

commit 18c864cdae3e5a2d444deac2a2b23e08096ee4e2

Merge: df9189cc46 4bc510c17b

Author: Andrew Harp

Date: Wed Dec 13 16:34:33 2017 -0500

Merge pull request #15163 from lihanchen/AndroidTVUSBCamera

Fix problem with camera on Android TV

commit 3d74c715a2ed66ced2bd87c90befb9e70ba8a3d1

Author: Yao Zhang

Date: Wed Dec 13 13:21:49 2017 -0800

Generate a unique name for const used by transpose node, so that we don't use a name that is previously used already. Add dependency for const in binary op if in a loop/frame.

PiperOrigin-RevId: 178948174

commit 2971a85370358789fa07a075be5e265e20dd8c34

Author: Allen Lavoie

Date: Wed Dec 13 13:10:04 2017 -0800

Prototype for serializing the Checkpointable graph.

Includes protocol buffers which will likely be added to BundleHeaderProto. We'll pass this protocol buffer plus the standard map of variable names to variable objects to the Saver, both of which will then be stored along with the checkpoint. This extra information means that the path we stored to a variable in its checkpoint name can be removed as long as there is another in the checkpointed graph.

PiperOrigin-RevId: 178946669

commit df9189cc4671facfecd3e8249c9e8b01b11c0df5

Author: Austin Anderson

Date: Wed Dec 13 12:41:38 2017 -0800

Update location for x86_64 android build (#15346)

See https://github.com/tensorflow/tensorflow/issues/15345

commit ea78050f20f7508537ccd1dee6dd60f0f40829cc

Author: Yao Zhang

Date: Wed Dec 13 12:11:59 2017 -0800

Support Shape and ShapeN ops.

PiperOrigin-RevId: 178939498

commit ac4d418e3cd1d3236037508b815db4cff82bcfda

Author: A. Unique TensorFlower

Date: Wed Dec 13 12:10:18 2017 -0800

Test consistently that the strings passed in input_arrays and output_arrays

consist of printable ASCII characters (this is motivated by a user having

unwittingly passed unicode zero-width characters, probably by copy-pasting),

and are names of arrays actually existing in the model.

Centralize these tests in CheckInvariants.

This can be overridden with new model flags: --allow_nonascii_arrays,

--allow_nonexistent_arrays. These are model flags because this is about

self-consistency of the model and its existing modelflags.

This CL partly undoes a recent relaxation of checks on input arrays that

was done to support getting graphviz out of incorrectly specified graphs.

Such users will now have to pass --allow_nonexistent_arrays.

PiperOrigin-RevId: 178939235

commit e256c813f1d1cdb857014a8617628c7c812d98c6

Author: Allen Lavoie

Date: Wed Dec 13 12:03:20 2017 -0800

A start on a prototype for object-based variable saving.

Doesn't do anything useful yet, but has an API for Checkpointable:

- add_variable (compatible with Layer)

- track_checkpointable (kinda like tfe.Network's track_layer)

- a property indicating checkpoint dependencies and their names

Also has some logic for collecting variables (using a private attribute).

Main TODOs:

- Prototype checkpointing a full object DAG (to eventually be added to BundleHeaderProto?). This is needed in case one path to an object with multiple paths is broken, and it happens to be the path we used in the checkpoint keys for that object's variables.

- Logic to attribute human-readable checkpoint names to objects in this full checkpointed DAG, and from the checkpointed DAG to the Python DAG being loaded into. Need some escaping when writing checkpoint names for this to work.

- Deferred loading (copy from tfe.Network?), including some management of the creation of slot variables

- APIs for saving/loading, some end-to-end examples

PiperOrigin-RevId: 178938256

commit f9a88f8a2b9fc371bc5dbd7a9404494c29857f6e

Author: Austin Anderson

Date: Wed Dec 13 11:49:40 2017 -0800

Cap prompt attempts before aborting, like other prompts

PiperOrigin-RevId: 178936195

commit 3425ae40aa54f16776bfcfe3be9c42bdb426cc20

Author: Austin Anderson

Date: Wed Dec 13 11:27:29 2017 -0800

Update docs as per #15325

PiperOrigin-RevId: 178932328

commit 52a44f28174f3a08fa92c3d43a9531c7c1101666

Author: Eugene Brevdo

Date: Wed Dec 13 11:17:07 2017 -0800

Convert LSTMFusedBlockCell to a plain Layer; it is not really an RNNCell.

This allows us to revert a change to the public API for most RNNCells.

That breaking change was introduced yesterday (wherein scope argument had to be

passed by keyword arg).

PiperOrigin-RevId: 178930316

commit dcbf6c972d7b4203735bca04f4d33d575ef7b22b

Author: A. Unique TensorFlower

Date: Wed Dec 13 11:06:55 2017 -0800

Run HardcodeMinMax graph transformation again after applying

--default_ranges_{min,max}. The specific issue that this fixes was

that with a Concatenation op involving one constant and one

non-constant input, the non-constant input got its minmax from

--default_ranges_{min,max}, but then it was too late for HardcodeMinMax

to propagate that to the other constant input of this Concatenation.

PiperOrigin-RevId: 178928533

commit c25bf74a3ebb13e4fd8a46022bc4e231edd43869

Author: Javier Luraschi

Date: Wed Dec 13 10:19:14 2017 -0800

Fix link in tensorflow lite readme

commit 5ecc3bf148f3e260352638c40f0787f57e4be382

Author: Yao Zhang

Date: Wed Dec 13 10:12:42 2017 -0800

Simplify the naming of added transpose node. Refactor dimension check and the node creation for data format transform for better code reuse (data format transform to be used for both inputs and outputs).

PiperOrigin-RevId: 178918771

commit 10d45f9ca118ed37b190140f9310e58f95d4d52c

Author: A. Unique TensorFlower

Date: Wed Dec 13 10:01:47 2017 -0800

Make bfloat16 works with complex

PiperOrigin-RevId: 178917043

commit e31f38913d4018c2cee094e05a04833ac96f8b68

Author: A. Unique TensorFlower

Date: Wed Dec 13 09:54:52 2017 -0800

Fix 'tags' parameter in predictor_factories.load_from_model.

tags was incorrectly being mapped to inputs.

Added basic unit tests.

PiperOrigin-RevId: 178916192

commit 185c593cb71cb6d8116ba05c97e9385642648f1b

Author: Derek Murray

Date: Wed Dec 13 08:56:20 2017 -0800

Automated g4 rollback of changelist 178759398

PiperOrigin-RevId: 178909147

commit 2b1b7dffcd2c76876efdbcfc431424e259da3bf4

Author: Benoit Steiner

Date: Wed Dec 13 08:52:40 2017 -0800

Check that all the inputs to a Concat op are of the same rank.

PiperOrigin-RevId: 178908773

commit 246d5b95723a275b80f4b803024182b4221b4e43

Author: A. Unique TensorFlower

Date: Wed Dec 13 08:43:09 2017 -0800

Stream::BlockHostUntilDone now returns Status rather than bool.

The now-deprecated Stream::BlockHostUntilDoneWithStatus remains, to facilitate a

multi-CL renaming transition. Once all callers have been renamed to

BlockHostUntilDone, *WithStatus will be removed.

The StreamExecutor (private) method has also been renamed to BlockHostUntilDone.

It's only used by Stream.

The StreamExecutorInterface method will be renamed in a separate atomic CL.

It's harder to perform that transition gradually, and we've already performed an

atomic change previously, so we might as well fix it up in one shot.

PiperOrigin-RevId: 178907807

commit 9e993d2dc641c4fef0837b1109d99ceddc2a1c21

Author: Johan Ju

Date: Wed Dec 13 17:33:03 2017 +0100

Always use std::cerr

commit 2eae1ac21ce28f3b2cafe9e12a25b3bddc475847

Author: A. Unique TensorFlower

Date: Wed Dec 13 08:03:03 2017 -0800

Standardize attribute naming for operators specifying a dimension to "axis". This mirrors TensorFlow's attribute naming.

PiperOrigin-RevId: 178903728

commit d2185de9703d9a1b98f0e41bcc78ee493639bdaa

Author: Taehoon Lee

Date: Wed Dec 13 16:58:06 2017 +0900

Fix typos

commit d7dbf21bda3fae90ced99db1bb5c592264c02526

Author: Austin Anderson

Date: Tue Dec 12 21:24:11 2017 -0800

Explicitly specify CUDA and CUDNN versions (#15328)

A workaround to the base problem causing https://github.com/tensorflow/tensorflow/issues/15291

commit 8f19188a14b62f2612783f3ebba0cd1c9d08aba8

Author: Igor Ganichev

Date: Tue Dec 12 21:12:14 2017 -0800

Create global_step when recording summaries if needed

User might have not created global_step prior to using

some summary method.

PiperOrigin-RevId: 178857144

commit d109bd2f3fc6a018629469a6497dfe953bf5e305

Author: A. Unique TensorFlower

Date: Tue Dec 12 20:19:12 2017 -0800

Disable a test case in params_test for CPU.

This test has thousands of parameters, and the resulting graph takes take too long

to compile on the CPU backend.

PiperOrigin-RevId: 178853687

commit 216878ea3dafdc5fbe6a15d389edb003ad2fd4b4

Author: Alexander Gorban

Date: Tue Dec 12 20:09:45 2017 -0800

Simplify tf.case implementation.

PiperOrigin-RevId: 178853258

commit b7308e3bd69349e9023497948a6bf55d3b0895d9

Author: Igor Ganichev

Date: Tue Dec 12 19:28:44 2017 -0800

Correctly pass name in layers.util.smart_cond

Before this change arguments were passed positionally and "name"

argument was wrongly mapped to "strict" argument of tf.cond instead

of the itended "name". Such a fix could potentially change operation

names and cause an error when restoring a graph, but it seems like

this particular change is safe for the following reasons.

- smart_cond is not a public API. So users should not be calling it

directly.

- smart_cond is used in 38 places internally. All of them, except for

in tf.contrib.summary don't use the "name" parameter. Such usage leads

to the same names before and after this change. The names will change

for users of tf.contrib.summary. Luckily, this is a very recent

addition and has utility only in context of eager execution, which is

in pre-alpha stage yet.

Because this change reroutes the wrong "name" -> "strict" mapping to

"name" -> "name", the value of "strict" is changing from "None" to

"False". Luckily, this has no effect on the function's behavior.

PiperOrigin-RevId: 178850766

commit 192bcb86d9e988d4f3b12bd2677e3638a9a0cae3

Author: A. Unique TensorFlower

Date: Tue Dec 12 18:44:25 2017 -0800

Fully-qualify function call in TF_CHECK_OK macro implementation, so that it can

be safely used outside of the tensorflow namespace.

Note that the StreamExecutor SE_CHECK_OK simply uses TF_CHECK_OK, so this helps

those cases.

PiperOrigin-RevId: 178847904

commit 24fda91f22e3df55187199c99b3e4a6bf5c0d1a1

Author: A. Unique TensorFlower

Date: Tue Dec 12 18:19:29 2017 -0800

Allow Tensor::bit_casted_shaped() to take type parameter T with different size

from the buffer data type size.

PiperOrigin-RevId: 178845870

commit dd6f9d5f43870dc39dbed91c6897dc4bb22ca495

Author: Eugene Brevdo

Date: Tue Dec 12 17:45:50 2017 -0800

Move more contrib RNN objects to be Layers.

PiperOrigin-RevId: 178842373

commit e115b064f57f5c373f1acdb56b210c541ccf63fb

Author: Eugene Brevdo

Date: Tue Dec 12 17:01:02 2017 -0800

[TF] Mark DT_STRING and DT_RESOURCE types as always sitting on host memory.

This is important when these arguments may appear in op input lists or output lists,

where the signature may not be able to declare them as sitting on host.

For DT_RESOURCE types, just the handles are marked as sitting on host memory;

the actual data may reside on GPU.

PiperOrigin-RevId: 178837213

commit e2174cc943f60d0e0bb47abd26fc2fe9772a27ef

Merge: 10c97ff4da 7ba5810c10

Author: Martin Wicke

Date: Tue Dec 12 16:40:31 2017 -0800

Merge pull request #13453 from TTrapper/sampled_sparse_softmax_loss

sampled version of sparse_softmax_cross_entropy_with_logits

commit 618d5c5fad4f70456856625322db104b851a399d

Author: Ian Langmore

Date: Tue Dec 12 16:30:09 2017 -0800

BUGFIX: MVN Full Covariance: Use dtype dependent tolerance to verify symmetric.

PiperOrigin-RevId: 178833453

commit c373a16f61bff835181163dc07417e3cba6f47bc

Author: Yangzihao Wang

Date: Tue Dec 12 16:21:26 2017 -0800

Return unimplemented error when trying to use dilated rate > 1 combined with NHWC format on the CPU.

Add test for unimplemented errors in Conv2D op.

PiperOrigin-RevId: 178832407

commit 47b674c938a38c6d88f27244a12ce3944c2f0464

Author: A. Unique TensorFlower

Date: Tue Dec 12 16:09:47 2017 -0800

[XLA] Remove a source of nondeterminism in HLO clustering.

Record the HLO clusters with std::set instead of std::unordered_set to ensure

that the algorithm to assign each cluster a sequence number during a set

traversal is deterministic.

PiperOrigin-RevId: 178830794

commit 7d8ab0e440cf895dceb31f260ce5ae177b84c05d

Author: A. Unique TensorFlower

Date: Tue Dec 12 16:06:31 2017 -0800

For many requests, the GCS filesystem client did not provide DNS lookup hints. This change allows all GCS HTTP requests to use the GCS DNS cache. It also simplifies the code, and eliminates a lot of redundant code.

The GCS DNS cache has been simplified and made more general. It is now easy to add more DNS names, simply by adding an entry to the GcsDnsCache::names_ list.

PiperOrigin-RevId: 178830317

commit 82afab2e9c9fd0f5da7bf6c27c597a9320a14d84

Author: A. Unique TensorFlower

Date: Tue Dec 12 15:58:15 2017 -0800

Only require validation that a fetch is requested for tf2xla::Config.

It is legitimate to convert a graph with only fetches, eg in that case where

the inputs to the graph are supplied by the infeed rather than by a feed node.

PiperOrigin-RevId: 178828952

commit 87cfa5696122c2173902accd47418ee4f25995d7

Author: Guangda Lai

Date: Tue Dec 12 15:39:52 2017 -0800

Refactor helper functions a bit for virtual gpu changes later.

PiperOrigin-RevId: 178826426

commit a94506b831aac707979aa615b0f424f2d4b2f316

Author: Yao Zhang

Date: Tue Dec 12 15:29:16 2017 -0800

Support permutation from NCHW to NHWC.

PiperOrigin-RevId: 178824999

commit 1c2bcf947f2e192512857887fb1301d13fe332ec

Author: Derek Murray

Date: Tue Dec 12 14:57:55 2017 -0800

Fix bug in kernel creation with functions marked "stateful".

The CallOp kernel caches a handle for invoking the function. This

handle is only valid in a single subgraph (it is scoped to the

FunctionLibraryRuntime). Marking a function as stateful causes its

CallOp kernel to be shared between multiple subgraphs. Therefore, this

change overrides the kernel creation logic to ensure that each

subgraph gets its own CallOp.

PiperOrigin-RevId: 178820064

commit d0a4a79c02b8c2a64763e1ef02c878f34e9defff

Author: Igor Ganichev

Date: Tue Dec 12 14:33:55 2017 -0800

Add test case for record_summaries_every_n_global_steps

This test case illustrates how to use

record_summaries_every_n_global_steps and tf.all_summaries()

in graph mode. There are no tests using

record_summaries_every_n_global_steps. All existing graph

based tests don't use tf.all_summaries() creating the impression

that summary ops will somehow always run, which is not the case.

PiperOrigin-RevId: 178816316

commit 389ffa842b0c0e7344ac721bbee8639de96cc81e

Author: Benoit Steiner

Date: Tue Dec 12 14:15:09 2017 -0800

Added a debug mode to the model analyzer to make it easier to figure out why shapes are missing.

PiperOrigin-RevId: 178813305

commit 9810da1b87aae689cac42bae754e7e4cb5a99d57

Author: A. Unique TensorFlower

Date: Tue Dec 12 13:03:52 2017 -0800

Adds XLA support for tf.nn.dynamic_rnn

Changes tf.nn.dynamic_rnn to specify `maximum_iterations` argument for the while_loop.

When `maximum_iterations` argument is supplied to tf.while_loop, use this to provide an upper bound on the size of Stacks used for gradient computation.

By specifying the stack limit we can generate gradient code for while loops that uses fixed shape TensorArrays and hence can be compiled with XLA.

PiperOrigin-RevId: 178802710

commit 50784e4ef8d920ed5fe7b3a9d0bafbac073ebb26

Author: Brennan Saeta

Date: Tue Dec 12 12:51:41 2017 -0800

Automated g4 rollback of changelist 177619402

PiperOrigin-RevId: 178800980

commit 10c97ff4dafffcb5c407c205d55c56f3d7dd7635

Author: Gunhan Gulsoy

Date: Tue Dec 12 12:26:15 2017 -0800

Parameterize tensorflow CUDA and cudnn versions in cmake build. (#15298)

Also upgrade the defaults to cuda9 and cudnn 7.

commit bb70caa0fb79c9f6cc8756816849d09a690fe198

Author: David Majnemer

Date: Tue Dec 12 11:35:20 2017 -0800

[XLA] Always fold transposes into convs or dots regardless of use count

PiperOrigin-RevId: 178790193

commit a6af4dc3a3c116638f97df01b9e1972fcddea488

Author: A. Unique TensorFlower

Date: Tue Dec 12 11:26:41 2017 -0800

Sliced Wasserstein Distance metric for GANs evaluation.

PiperOrigin-RevId: 178788810

commit 4e1a7a74b61aa02bc9c3104706afb2153faefddf

Author: A. Unique TensorFlower

Date: Tue Dec 12 11:18:28 2017 -0800

Add CompositeNodeManager for Grappler VirtualScheduler.

CompositeNodeManager has per-device LIFO manager, FirstReadyManagers for _Send

and _Recv ops, and chooses FirstReady among the ops from per-device LIFOManager

and _Send and _Recv FirstReadyManagers.

This one can maximizes producer-consumer locality within a device (with LIFO),

but does not introduce previously reported scheduling inefficiency w.r.t.

multi-device execution with separately managing _Send and _Recv ops and global

FirstReady policy across devices.

It's implemented, but not enabled; VirtualScheduler still uses

FirstReadyManager.

PiperOrigin-RevId: 178787352

commit a7c11aa3cafea286947a8131523fe1c6dc0b7174

Author: Olivia Nordquist

Date: Tue Dec 12 11:17:00 2017 -0800

disabling flaky test

PiperOrigin-RevId: 178787158

commit 2bf344a6ab8f5465b6d5ed8a98bd7af0083dbe8b

Author: A. Unique TensorFlower

Date: Tue Dec 12 11:04:58 2017 -0800

Associative operator optimization:

Push constants down add/mul to canonicalize chains and possibly create constant nodes at the bottom. Example:

+ + +

/ \ / \ / \

c1 + --> x + --> x c1+c2

/ \ / \

c2 x c2 c1

Small cleanup: Consolidate code for manipulating names of nodes added or modified during constant folding.

PiperOrigin-RevId: 178785218

commit 968da4bf2722b1303cc223e8342357d62c27dfc1

Author: Skye Wanderman-Milne

Date: Tue Dec 12 10:58:31 2017 -0800

Raise exception on missing unused input_map keys with C API enabled.

Without this change, the C++ ImportGraphDef API returns unused

input_map keys (which are plumbed through to the C API as

well). However, the Python import_graph_def API requires slightly

different semantics: it throws an error for unused input_map keys that

are missing from the GraphDef.

This change modifies the C and C++ APIs to limit the returned keys to

those missing from the GraphDef, and plumbs this through to the C

API-enabled import_graph_def implementation.

Note that this is a change to the existing C API. Luckily the modified

method hasn't been released yet, so it's ok to change it.

PiperOrigin-RevId: 178783957

commit 2d988e0172636b9592e23c84e2e51df814b6fabc

Merge: 4806cb0646 8903e318bf

Author: Martin Wicke

Date: Tue Dec 12 10:04:46 2017 -0800

Merge pull request #15316 from Androbin/patch-2

Exclude tests from contrib_py

commit c8a5ffdeb2a17df2d2822c7a6df8a44f3ab85614

Author: A. Unique TensorFlower

Date: Tue Dec 12 09:19:42 2017 -0800

Integrate tensor pool feature to `gan_loss` function.

PiperOrigin-RevId: 178769850

commit 0e1c5577da2c9f4f8bdf3346a12a739e3752db03

Author: A. Unique TensorFlower

Date: Tue Dec 12 09:01:37 2017 -0800

Disable neutral element and reciprocal optimizations again.

PiperOrigin-RevId: 178767676

commit 8693bf519399495cedd91293ec82b492ea401f6f

Author: Jiri Simsa

Date: Tue Dec 12 08:31:48 2017 -0800

This CL makes two improvements to the `map_and_batch` transformation:

1) It fixes a bug that manifested as `OutOfRange` being returned prematurely.

2) It changes the behavior on sequences of elements whose size is not a multiple of `batch_size`. Previously, the implementation would drop the last small batch (similar to `batch_and_drop_remainder). Newly, the implementation returns the last small batch (similar to `batch`).

PiperOrigin-RevId: 178764508

commit 94686be226fc999437618959a791d7bda18e1806

Author: Derek Murray

Date: Tue Dec 12 07:43:24 2017 -0800

Automated g4 rollback of changelist 178675527

PiperOrigin-RevId: 178759398

commit 8903e318bf1840421162dab916c55e8fe1d1bcbc

Author: Robin Richtsfeld

Date: Tue Dec 12 15:05:35 2017 +0100

Exclude tests from contrib_py

commit 5f470cf9b0b8adf9cf74da33e30024c62647165c

Author: Mark Daoust

Date: Tue Dec 12 06:02:03 2017 -0800

Add get started `Datasets` doc

PiperOrigin-RevId: 178751067

commit 444e319256de3229aef688c39b85689d3e2eab4f

Author: Rajendra arora

Date: Tue Dec 12 19:10:23 2017 +0530

Fixing typo

commit 3f5445b3432fdf775bdef289c10338cd20d10edc

Author: Alexandre Passos

Date: Tue Dec 12 03:24:43 2017 -0800

Simplifying tfe function.py

PiperOrigin-RevId: 178740804

commit 65d23686c6404025d0554960d4cc4eda1bbd3c9d

Author: A. Unique TensorFlower

Date: Tue Dec 12 02:33:55 2017 -0800

Remove real-data shape check in GANEstimator. Fixes github issue #14257.

PiperOrigin-RevId: 178737278

commit 3d4aa4eacd4beca250ed52cc284e9caa7fb45e22

Author: Yifei Feng

Date: Tue Dec 12 02:02:28 2017 -0800

Disable flaky //tensorflow/contrib/learn:dnn_linear_combined_test

PiperOrigin-RevId: 178734940

commit 1ffba992df5159cf88c438a62ede4ede3dc5dc2d

Author: A. Unique TensorFlower

Date: Mon Dec 11 23:47:10 2017 -0800

[XLA] Properly set layout requirements in Hlo parser.

PiperOrigin-RevId: 178724659

commit 383a3226a9ad08ac507a3fbd6c220c5c1e15a540

Author: A. Unique TensorFlower

Date: Mon Dec 11 23:35:29 2017 -0800

Use BlockHostUntilDoneWithStatus in various places.

PiperOrigin-RevId: 178723711

commit fcca16c5e3d8168ea95616956fdcf777798b164b

Author: Gunhan Gulsoy

Date: Mon Dec 11 23:26:35 2017 -0800

Disable flaky random ops test.

PiperOrigin-RevId: 178723108

commit f379d3369505457928ccda5adf72583f95709026

Author: Bjarke Hammersholt Roune

Date: Mon Dec 11 22:31:27 2017 -0800

* HloTestBase now prints out the HLO parser error message when there is one.

* TestUtils now supports generating random literals with more than one constraint.

There is still an error if the constraints conflict.

PiperOrigin-RevId: 178720092

commit 913175c2bd38f6e97de399b29cfe1195bffbaa25

Author: Justin Lebar

Date: Mon Dec 11 22:26:00 2017 -0800

[XLA] Add stringification to BatchNormTestParam.

This way when a test fails, it prints out useful information about the

failure, instead of

"<48-byte object with these bytes: de ad be ef ...>"

PiperOrigin-RevId: 178719733

commit 228b3ebc91ff351befcef641d01d1a82d532a6d1

Author: Justin Lebar

Date: Mon Dec 11 21:38:05 2017 -0800

[XLA] Don't call timer->Nanoseconds() on a not-ok stream.

If the stream is not OK, the timer might not have been initialized and

finalized, in which case calling timer->Nanoseconds() is illegal and

will crash.

PiperOrigin-RevId: 178717089

commit 5ff2b9d1c76fa64969238d798b5df94980341988

Author: A. Unique TensorFlower

Date: Mon Dec 11 21:04:20 2017 -0800

Update ops-related pbtxt files.

PiperOrigin-RevId: 178715353

commit 2adbc217b3eeed329d077050e0f1f7d88edd86d7

Author: Sanjoy Das

Date: Mon Dec 11 20:34:08 2017 -0800

[XLA:CPU] Teach the CPU layout assignment about dot dimension numbers

There is no great need for this yet, but I noticed that the test cases were

broken (they were constructing dots with unset dimension numbers), and one thing

led to another.

PiperOrigin-RevId: 178713597

commit f18d23b90a48e51013008d51320f5aedd555cc02

Author: Ian Langmore

Date: Mon Dec 11 19:44:08 2017 -0800

prefer_static_* functions added to CORE/distributions/util.py

PiperOrigin-RevId: 178710439

commit 2a48746874d2043cddb632fe825d88f3da0c185d

Author: Yao Zhang

Date: Mon Dec 11 19:39:35 2017 -0800

Fix the handling of unknown rank. Previous code would wrongly treat a tensor of unknown

rank as a scalar.

PiperOrigin-RevId: 178710185

commit c210009d0a8d40d458c21b1faf7b1adf8e4deaee

Author: A. Unique TensorFlower

Date: Mon Dec 11 19:19:26 2017 -0800

Includes in the TF Lite kernels/op_macros.h to fix a compile errors

when building externally using either the Makefile or Bazel. The macros use

stderr and fprintf which may not be defined depending on the order of

headers included by the .cc files.

PiperOrigin-RevId: 178708839

commit 4806cb0646bd21f713722bd97c0d0262c575f7e0

Merge: fc24c588df e5246fb0b0

Author: Martin Wicke

Date: Mon Dec 11 19:11:57 2017 -0800

Merge pull request #15286 from dandelionmane/branch_178689056

Branch 178689056

commit fc24c588dfb8cd5ff55537e1db79587538a1acff

Author: 张志豪

Date: Tue Dec 12 10:54:10 2017 +0800

Replace `variables.get_global_step()` use `training_util.get_global_step()` (#15247)

* Replace `variables.get_global_step()` use `training_util.get_global_step()`

commit ecdecc5a325482edac2cdef2d40e091ed8b5016d…