tensorflow / tensorflow

Commits on Oct 14, 2020

1. Add tf.DestroyResource op to MLIR TF Dialect. …

   PiperOrigin-RevId: 337145111
   Change-Id: I69165d0db6aff8e3a9aaf12a3f73812d1ac7c6fc

   

   tensorflower-gardener committed Oct 14, 2020
   3a09bfe

2. Disable broken/flaky TSAN test …

   PiperOrigin-RevId: 337143782
   Change-Id: I46ab677a83678fc37561769f789e74397adea274

   

   mihaimaruseac authored and tensorflower-gardener committed Oct 14, 2020

   Loading status checks…
   86d5f9d

3. Disable broken/flaky TSAN test …

   PiperOrigin-RevId: 337142196
   Change-Id: I44172ecb184ea2efd0685b16b924c91608e26964

   

   mihaimaruseac authored and tensorflower-gardener committed Oct 14, 2020

   Loading status checks…
   cf7b1c2

4. Disable broken/flaky TSAN test …

   PiperOrigin-RevId: 337139985
   Change-Id: I3066f3357f01a1c378269d4c6fc6cdc1d5ace5b1

   

   mihaimaruseac authored and tensorflower-gardener committed Oct 14, 2020

   Loading status checks…
   229d51c

5. Remove data barrier islands (inputs and outputs) creation from Parall… …

   …elExecuteToIslandsPass.
   
   Instead, control dependencies are used, which can minimize data transfers if barrier islands lower to IdentityN receiving data from different devices. This now matches the lowering in ReplicateToIslandPass.
   
   PiperOrigin-RevId: 337136664
   Change-Id: Iec0899e8f2d7dfedb52a8b4b8766ffac76903bff

   

   andyly authored and tensorflower-gardener committed Oct 14, 2020

   Loading status checks…
   908f2c2

6. Fix a bug where vector::reserve was being called on config.sparse ins… …

   …tead of config.ragged.
   
   PiperOrigin-RevId: 337133816
   Change-Id: I4b712c2e141201f21f19fb92627981c53bb4280d

   

   tensorflower-gardener committed Oct 14, 2020

   Loading status checks…
   c346633

7. Fix an incorrect object equality of the static arguments. …

   PiperOrigin-RevId: 337132763
   Change-Id: Id782f638a92e4d3bf86b9c483f675d595647f69e

   

   jblespiau authored and tensorflower-gardener committed Oct 14, 2020

   Loading status checks…
   9eaac42

8. Add GPU specific LMHLO level ops …

   - Introduce operations in a new lmhlo_gpu dialect that map to GPU library function calls
     in the XLA:GPU backend.
   - Add basic unit tests as well.
   
   PiperOrigin-RevId: 337132166
   Change-Id: I02595362e9cf58333932d5dac424b00bae140339

   

   jurahul authored and tensorflower-gardener committed Oct 14, 2020

   Loading status checks…
   04138b6

9. Update ops-related pbtxt files. …

   PiperOrigin-RevId: 337131392
   Change-Id: I62692587def630f82cd7dad22cd210c905ffbf26

   

   tensorflower-gardener committed Oct 14, 2020

   Loading status checks…
   2f643c3

10. Add num_slow_calls to InputPipelineStats. …

    PiperOrigin-RevId: 337129908
    Change-Id: Ic3d6fb58f709fcb86a085e6a9e3b0f0920fdc040

    

    jray319 authored and tensorflower-gardener committed Oct 14, 2020
    3b6dada

11. Remove ThenExecuteFunction references from eager XLA op dispatch. …

    The cleanup behavior is supplied by default by the GPU and XLA backends.
    
    PiperOrigin-RevId: 337127562
    Change-Id: I539251f3932703af27e17a59bd39c120624a143d

    

    rjpower authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    08be643

12. Fix logic for enabling MLIR bridge depending on has_tensor_list_arg. …

    It appears that the polarity of the use of has_tensor_list_arg was
    inadvertently flipped.
    
    Disable any MLIR bridge enabled tests that were passing because they weren't
    using the MLIR bridge due to this issue.
    
    PiperOrigin-RevId: 337125651
    Change-Id: I93e9e61acda9a2aeffaee5cce13e93635d33f5a4

    

    richard-uhler authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    b881485

13. [XLA] Add pattern matcher for select and scatter. …

    PiperOrigin-RevId: 337122945
    Change-Id: I9aaa3cdcb3cb3379492e4697ada514124995a8ce

    

    blakehechtman authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    7e54bf3

14. Disable broken flaky TSAN MLIR/TFR test. …

    PiperOrigin-RevId: 337118470
    Change-Id: I237b9c8f66867ab1850126fc87aad22161d5c6d8

    

    mihaimaruseac authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    6358ad3

15. Revert the toolchains updates. …

    PiperOrigin-RevId: 337117469
    Change-Id: I42af87a955ecb4bc08cafa9cb991bfd8005f2674

    

    av8ramit authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    f4e4758

16. Update _mirrored_variables_indices attribute on TPUReplicate when hea… …

    …d outside compiling.
    
    Head outside compilation may add additional replicated inputs to the _TPUReplicate Op.  These newly added inputs come before the distributed variables that maybe replicated.  Distributed variables may be mirrored and the _mirrored_variables_indices should reflect their location in the full replication inputs list(TInputs attribute).
    
    PiperOrigin-RevId: 337117443
    Change-Id: I085affe3b316ef80a864a9847bc014168b9719e6

    

    kenfranko authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    7ca0b1c

17. Disable distribute_strategy_test_gpu on Guitar. …

    PiperOrigin-RevId: 337115998
    Change-Id: Id7ec174920a98d7bf0b13d9bed63b853136707fb

    

    crccw authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    5f98faf

18. [lite] Add unit test for stderr_reporter. …

    PiperOrigin-RevId: 337112639
    Change-Id: I0cc00cd9ead8a4ca81f32a0153398aad3506eb30

    

    fergushenderson authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    b7d9054

19. Add support for RaggedTensor properties, instance methods, and class … …

    …methods in the Keras functional API.
    
    PiperOrigin-RevId: 337109937
    Change-Id: Ieb5bdc0d74ff6cedf37945a065766eaf7ec30154

    

    tomerk authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    e9c5b8b

20. Go: Update generated wrapper functions for TensorFlow ops. …

    PiperOrigin-RevId: 337109867
    Change-Id: I0ba02cbb60ad868f9a71b202d700a5facb204b85

    

    tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    e14093e

21. Support batching all-reduce with concat and split …

    Collective v2 doesn't support scoped allocator. While it's possible to make scoped allocator work with it. Concat/split is much simpler.
    
    PiperOrigin-RevId: 337109439
    Change-Id: I3535f5e0b090696f3bb620617f2b57f1f4b78b22

    

    crccw authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    56124bd

22. Added gradients for RaggedTensorToVariant and RaggedTensorFromVariant… …

    …. (This allows gradients to pass through map_fn when it is applied to ragged tensors.)
    
    PiperOrigin-RevId: 337108621
    Change-Id: I73d5f3296181877f0cc4c7a6273b693bcf8310ab

    

    edloper authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    be6b1fd

23. Fix collective_all_reduce_strategy_test breakages …

    We need to respect collective_key_base in local mode as well.
    
    PiperOrigin-RevId: 337108529
    Change-Id: Ib9e4bc4b6a38f0b7c87d74bdc6001709ded32523

    

    crccw authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    be99659

24. Added lambda variant to ActivityStart in TraceMe. …

    For example metadata can be added to the trace like
    ```
    uint64 id = ActivityStart([]() {
        return TraceMeEncode("my_trace", {{"key1", value1}, {"key2", 42}});
    });
    ```
    
    PiperOrigin-RevId: 337108432
    Change-Id: I7bba63595ed80f289c65b6086a15af104648845e

    

    tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    e56256e

25. Support XlaGather and XlaScatter ops for bool. …

    The undelying XLA ops do work for bool.
    
    PiperOrigin-RevId: 337106941
    Change-Id: Ifd920287f1113b9d6140489e8518613547289977

    

    tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    377ef73

26. Use double to represent milliseconds in time_utils.h …

    PiperOrigin-RevId: 337106111
    Change-Id: Ia4aebd0e9e92a62e5c88db9825fd6096317a6c59

    

    jbaiocchi authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    668c686

27. Remove test_suite rules. nothing seems to depend on these. …

    PiperOrigin-RevId: 337105125
    Change-Id: Ic26701d7f7e6e884e64b7ba4cf766d641becf871

    

    gunan authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    eae81ad

28. Add BUILD rules to build a benchmark model APK with flex delegate. …

    PiperOrigin-RevId: 337102779
    Change-Id: I6584c608ccbd81f47aa99827342bfb40cd430d78

    

    mihaimaruseac authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    090b691

29. Disable distribute_strategy_test_gpu on Guitar. …

    PiperOrigin-RevId: 337101527
    Change-Id: Ia02ffa0244a78cbf1c920d47a773516e80d57aed

    

    pavithrasv authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    98e3858

30. Merge pull request #43852 from ROCmSoftwarePlatform:google_upstream_r… …

    …ocm_fix_201007
    
    PiperOrigin-RevId: 337099240
    Change-Id: I62843e9f095a9ec8d8189f5a66ac85610273a839

    

    tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    18cbca2

31. Update TFLite schema generated header file …

    PiperOrigin-RevId: 337097065
    Change-Id: I16ff571a1adea6b9d2866a360874104554c3a89e

    

    abattery authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    dd1764b

32. [XLA:GPU] Do not create multi-output fusions that would cause exponen… …

    …tial compile behavior (time and size).
    
    PiperOrigin-RevId: 337095307
    Change-Id: I5fd2d2c169ce80b26f99c62749eb19843d42be07

    

    thomasjoerg authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    379bcfa

33. Disable tensorflow/python/keras/layers/preprocessing/reduction_test o… …

    …n tsan as it's failing.
    
    PiperOrigin-RevId: 337094272
    Change-Id: I8a6fc3f571e67551a7e410fd10124ee71f55e431

    

    rchao authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    43e3def

34. [XLA:CPU] Replace local copy of linalg codegen strategy with the upst… …

    …ream mlir one
    
    PiperOrigin-RevId: 337089868
    Change-Id: I476a6756fdb558beb7d454d02837fdea57a2cfef

    

    d0k authored and tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    07159df

35. Merge pull request #43935 from vnvo2409:gradients …

    PiperOrigin-RevId: 337081150
    Change-Id: I79311086235fca9be19d10926167653bf2d91a94

    

    tensorflower-gardener committed Oct 14, 2020

    Loading status checks…
    0602ad7