aboutsummaryrefslogtreecommitdiffhomepage
Commit message (Collapse)AuthorAge
* Fix reaching_definitions to correctly mark the definition of modified ↵Gravatar Dan Moldovan2018-07-16
| | | | | | symbols in the statement that replaces them, e.g. a = a. PiperOrigin-RevId: 204749753
* Bring the workshop notebook to date.Gravatar Dan Moldovan2018-07-16
| | | | PiperOrigin-RevId: 204748706
* Fix bug in masked_autoregressive_default_template where custom name was not ↵Gravatar A. Unique TensorFlower2018-07-16
| | | | | | creating custom variable scopes. PiperOrigin-RevId: 204747987
* Bring the dev summit notebook to date.Gravatar Dan Moldovan2018-07-16
| | | | PiperOrigin-RevId: 204741684
* Missing import and pip install of tf-nightly in workshop notebookGravatar A. Unique TensorFlower2018-07-16
| | | | PiperOrigin-RevId: 204736577
* Enable test that was accidentally misnamed.Gravatar Dan Moldovan2018-07-16
| | | | PiperOrigin-RevId: 204734160
* Add experimental config field to output interpolatable error messagesGravatar James Keeling2018-07-16
| | | | | | This is part of our effort to improve Python error messages by allowing the runtime to output formatted messages for the Python layer to interpolate. This will be gated by this config field to begin with. PiperOrigin-RevId: 204731230
* Make the HloDomainRemover pass more configurableGravatar A. Unique TensorFlower2018-07-16
| | | | | | | | | | | | Previously we had two different function to normalize instructions within a domain where one of them was specified inside the metadata while the other one is passed into the domain remover. This change unifies them to use the externally passed in function for both usecase to make it possible to rewrite both of them from the caller of the domain remover (to add special logic). PiperOrigin-RevId: 204715075
* add int32 support for subGravatar A. Unique TensorFlower2018-07-15
| | | | PiperOrigin-RevId: 204681037
* [XLA] Enable the semantic for cross-modeul AllReduce.Gravatar A. Unique TensorFlower2018-07-15
| | | | PiperOrigin-RevId: 204670087
* Comment cleanup.Gravatar Max Galkin2018-07-15
| | | | PiperOrigin-RevId: 204663736
* Merge pull request #20542 from rongjiecomputer:exceptionGravatar TensorFlower Gardener2018-07-15
|\ | | | | | | PiperOrigin-RevId: 204662666
* | Partial update of tf.keras to the Keras 2.2.0 API.Gravatar Pavithra Vijay2018-07-15
| | | | | | | | | | | | | | | | | | | | Changes included are: - Embedding visualization is added to TensorBoard callback (from older Keras API.) - Fix: learning phase info being left out in multi-input models (from older Keras API.) - Fix: Tensorboard callback only supports logging Embeddings layer weights - Fix: Tensorboard callback with layer with multiple outputs PiperOrigin-RevId: 204659796
* | Update default cuda compute capability while installing from sources to ↵Gravatar Smit Hinsu2018-07-14
| | | | | | | | | | | | cover up to Volta PiperOrigin-RevId: 204606836
* | Link NCCL 2.x lib and header where ./configure expects them.Gravatar Toby Boyd2018-07-14
| | | | | | | | PiperOrigin-RevId: 204596391
* | [XLA:AMDGPU] Enable the AMDGPU backend for open source buildsGravatar Benjamin Kramer2018-07-14
| | | | | | | | | | | | Merge pull request #20749 PiperOrigin-RevId: 204581011
* | Disable broken model_analyzer_testGravatar Gunhan Gulsoy2018-07-14
| | | | | | | | PiperOrigin-RevId: 204579728
* | [Java]: Support ConfigProto and RunOptions when loading SavedModels.Gravatar Asim Shankar2018-07-14
| | | | | | | | | | | | | | | | | | Fixes #18143 Fixes #20769 (Similar to #18716 by @raintung) PiperOrigin-RevId: 204575441
* | Add wrapper to help export model trained with estimator as SavedModel for TPU.Gravatar A. Unique TensorFlower2018-07-13
| | | | | | | | PiperOrigin-RevId: 204568222
* | Broad refactor (part 7): Swap in the new CFG implementation, along with the ↵Gravatar Dan Moldovan2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | new directives support. Simplify the process by which transformers are applied and their testing. Changes to converters: * all tests have been refactored to use the new helpers; functionally they remain the same * several converters have been renamed to be more consistent with the overall naming patterns: single_return.py -> return_statements.py; list_comprehension.py -> list_comprehensions.py; if_exp.py -> conditional_expressions.py * conditional expression converter has been rewritten to use local functions instead of lambdas, which are not yet supported by the static analyzers * the handling of if statements in control_flow.py now uses both the liveness and reaching definitions analyses to more robustly detect output variables * a new directives.py converter is introduced, that transforms user directives (see lang/directives.py) into AST annotations and removes the function calls from the code * list_comprehensions.py is slightly refactored for better readability; it is still not yet enabled * slices.py and lists.py now use the new directives mechanism to obtain list type information * side_effects_guards_test.py has been refactored to avoid flakiness Changes to AG core: * conversion.py now uses the helpers from converter.py and removes the redundant function * converter_testing.py now offers a more unified helper that can be used to test most use cases with less boilerplate * the new directives module now completely replaces the obsolete type_hints.py Changes to static analysis: * activity.py no longer considers index mutation to modify the parent (e.g. a[0] = 1 is not considered to modify a) * activity.py no longer detects the local symbol changes (the IS_MODIFIED_SINCE_ENTRY annotation), which is now handled by reaching_definitions.py * activity.py marks the created field as obsolete, to be removed once the loops in control_flow are migrated to use the dataflow analyses * the obsolete annos.py extension has been removes in favor of the consolidated tags in anno.py * the new CFG implementation now completely replaces the old version * live_values.py and type_info.py also use the symbol definitions from reaching_definitions.py PiperOrigin-RevId: 204563046
* | Merge pull request #20759 from ↵Gravatar TensorFlower Gardener2018-07-13
|\ \ | | | | | | | | | | | | | | | ROCmSoftwarePlatform:upstream-staging-xla-gpu-transfer-manager PiperOrigin-RevId: 204562235
* \ \ Merge pull request #20754 from ↵Gravatar TensorFlower Gardener2018-07-13
|\ \ \ | | | | | | | | | | | | | | | | | | | | ROCmSoftwarePlatform:upstream-staging-xla-globalvariable-addrspacecast PiperOrigin-RevId: 204560820
* | | | Automated rollback of commit fde3f09e3080a28b9a06ee219474957ba149a20dGravatar Revan Sopher2018-07-13
| | | | | | | | | | | | | | | | PiperOrigin-RevId: 204560078
* | | | Internal change.Gravatar A. Unique TensorFlower2018-07-13
| | | | | | | | | | | | | | | | PiperOrigin-RevId: 204560026
* | | | [XLA] Move implementation of ThreeFry stateless PRNG into xla/client/libGravatar Peter Hawkins2018-07-13
| | | | | | | | | | | | | | | | PiperOrigin-RevId: 204557470
* | | | Exclude util/stats_calculator.* from :framework_internal_implGravatar A. Unique TensorFlower2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | Otherwise both :framework_internal_impl and :stats_calculator_portable compile this file and cause multiple definition. PiperOrigin-RevId: 204555332
* | | | Add linear estimator colab entry to tutorials.Gravatar Billy Lamberta2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | FIx other colab links. PiperOrigin-RevId: 204555150
* | | | Qualify all names fully in registration macro. Otherwise the macro cannot be ↵Gravatar A. Unique TensorFlower2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | used from other namespaces without additional contortions. PiperOrigin-RevId: 204554899
* | | | [Cloud TPU / Keras]: Support experimental tf.data integration.Gravatar Brennan Saeta2018-07-13
| | | | | | | | | | | | | | | | PiperOrigin-RevId: 204552928
* | | | [XLA] Use shfl.sync.down instead of shfl.sync.Gravatar Justin Lebar2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | shfl.down is deprecated and ptxas 9.2 emits a loud warning when you use it. Convert XLA to shfl.sync.down. This change makes XLA:GPU require CUDA 9. PiperOrigin-RevId: 204546742
* | | | Automated rollback of commit d98b99d1cd4337ee11e7cbc4c9b6324f0e381502Gravatar Reed Wanderman-Milne2018-07-13
| | | | | | | | | | | | | | | | PiperOrigin-RevId: 204544587
* | | | [XLA] Update bad-ptxas-version warning to warn for anything less than 9.2.88.Gravatar Justin Lebar2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | We have hit another ptxas bug, which appears to be fixed in 9.2.88. We're no longer testing 8.x, so we can't say those versions work either. Just warn if it's less than 9.2.88. Also update the warning to suggest that people can cherry-pick the new ptxas if they can't upgrade to 9.2.88 properly. PiperOrigin-RevId: 204541857
* | | | Add support for metric_ops that are Operations when exporting TRAIN or EVAL ↵Gravatar Karmel Allison2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | SavedModels. PiperOrigin-RevId: 204540825
* | | | Merge pull request #20284 from perfinion:unbundleGravatar TensorFlower Gardener2018-07-13
|\ \ \ \ | | | | | | | | | | | | | | | PiperOrigin-RevId: 204539752
* | | | | Add a few TODOs based on discussion with Andrew.Gravatar Dan Moldovan2018-07-13
| | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204538445
* | | | | Merge pull request #20576 from Intel-tensorflow:mabuzain/avx-performance-fixGravatar TensorFlower Gardener2018-07-13
|\ \ \ \ \ | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204530874
* | | | | | TPUEstimator: Run tpu.initialize_system() in its own graph whenever the mainGravatar Jeremy Lau2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | graph is finalized. PiperOrigin-RevId: 204529164
* | | | | | Disable flaky tests on windows.Gravatar Gunhan Gulsoy2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204527084
* | | | | | Allow non-index slices to be used without conversion, instead of rejecting ↵Gravatar Dan Moldovan2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | them - they are widely used in TF code. PiperOrigin-RevId: 204526510
* | | | | | Merge pull request #20725 from yongtang:20722-TFSA-2018-001Gravatar TensorFlower Gardener2018-07-13
|\ \ \ \ \ \ | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204525677
* | | | | | | Disable flaky random seed testGravatar Gunhan Gulsoy2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204525365
* | | | | | | [XLA] Don't call VisibleDeviceCount from GetSupportedPlatformsGravatar A. Unique TensorFlower2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204523318
* | | | | | | Re-enabling delete on exit true by default.Gravatar A. Unique TensorFlower2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204520442
* | | | | | | tfdbg: remove Experimental tags and obsolete libraryGravatar Shanqing Cai2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | * debug_gateway and the related node_outputs_callback are not used and hence are removed in this CL. PiperOrigin-RevId: 204519574
* | | | | | | Reorder estimator guides in index and nav based on feedback.Gravatar Billy Lamberta2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204518322
* | | | | | | Change size of linear_operator_low_rank_update_test to large as it is ↵Gravatar Smit Hinsu2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | currently timing out in windows PiperOrigin-RevId: 204518052
* | | | | | | Automated rollback of commit 57527f7e47e3e67966b432065f510a601a4d8647Gravatar A. Unique TensorFlower2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204516578
* | | | | | | Update default NCCL version while installing from source to 2.2Gravatar Smit Hinsu2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This is to keep the default configuration consistent with prebuilt TensorFlow. PiperOrigin-RevId: 204513386
* | | | | | | Merge pull request #20667 from mktozk:patch-1Gravatar TensorFlower Gardener2018-07-13
|\ \ \ \ \ \ \ | | | | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204508607
* | | | | | | | [TF:XLA] Bump open source llvm revision to r336991Gravatar Sanjoy Das2018-07-13
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | PiperOrigin-RevId: 204506936