Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Develop upstream sync 230814 #2193

Merged
merged 360 commits into from
Aug 22, 2023
Merged
This pull request is big! We’re only showing the most recent 250 commits.

Commits on Aug 8, 2023

  1. Record num_batch_threads_ as TF does

    PiperOrigin-RevId: 554932893
    AspirinSJL authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    a8286c6 View commit details
    Browse the repository at this point in the history
  2. Remove option to use StreamExecutor Cloud TPU client in JAX

    It's been over three months since the new PJRT C API client was
    enabled by default
    (https://jax.readthedocs.io/en/latest/changelog.html#jax-0-4-8-march-29-2023).
    
    PiperOrigin-RevId: 554935166
    skye authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    74be7fd View commit details
    Browse the repository at this point in the history
  3. Improve some of the documentation around sliced prefetching.

    PiperOrigin-RevId: 554937981
    sparc1998 authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    bd82663 View commit details
    Browse the repository at this point in the history
  4. Remove unnecessary glob in tensorflow/core:lib_internal_impl.

    Since all the subfolders in tensorflow/core/lib have BUILD files, no files are picked up by this glob.
    
    PiperOrigin-RevId: 554943490
    BrianWieder authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    4c08c49 View commit details
    Browse the repository at this point in the history
  5. Code cleanup

    bhavani-subramanian committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    cd1ba12 View commit details
    Browse the repository at this point in the history
  6. Reset context for a couple of python unit tests.

    These tests seem to use different number of GPUs across multiple unit test cases. This causes issues with maintaining the PjRtClient and DeviceCompiler lifetimes.
    
    PiperOrigin-RevId: 554953351
    swachhandl authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    dd3bddb View commit details
    Browse the repository at this point in the history
  7. Configuration menu
    Copy the full SHA
    055b714 View commit details
    Browse the repository at this point in the history
  8. Refine dtensor target_patterns.

    PiperOrigin-RevId: 554958089
    DrMarcII authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    9061817 View commit details
    Browse the repository at this point in the history
  9. No public description

    PiperOrigin-RevId: 554958217
    tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    6da90a9 View commit details
    Browse the repository at this point in the history
  10. Remove unused saved_model_mira_impl

    PiperOrigin-RevId: 554959602
    cky9301 authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    7603dbd View commit details
    Browse the repository at this point in the history
  11. Declare _PforInput parameter types.

    PiperOrigin-RevId: 554965356
    fionalang authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    9c62350 View commit details
    Browse the repository at this point in the history
  12. Add return type annotations related tf.constant.

    PiperOrigin-RevId: 554967781
    fionalang authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    f410be4 View commit details
    Browse the repository at this point in the history
  13. Fix dominance issue after extract_tpu_copy_with_dynamic_shape pass.

    Only move TpuCopyWithDynamicShape out of launches if nothing in the
    launch depends on them.
    
    PiperOrigin-RevId: 554971170
    matthiaskramm authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    0f772dc View commit details
    Browse the repository at this point in the history
  14. Remove 'tensorflow/core:portable_gif_internal' dep where not needed

    PiperOrigin-RevId: 554972724
    LionOfJewdah authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    c302e76 View commit details
    Browse the repository at this point in the history
  15. Add type annotations to AbstractGradientTape.

    PiperOrigin-RevId: 554973779
    tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    794ad9d View commit details
    Browse the repository at this point in the history
  16. #tf-data-service Use absl::Status in distributed_snapshot_test.

    PiperOrigin-RevId: 554976689
    yangustc07 authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    12ac2e2 View commit details
    Browse the repository at this point in the history
  17. [TF:PJRT] Set allowed_devices when gpu_options.visible_device_list is…

    … not empty.
    
    PiperOrigin-RevId: 554976987
    Jieying Luo authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    cad505d View commit details
    Browse the repository at this point in the history
  18. Update tensorboard dependency to >=2.14, < 2.15

    PiperOrigin-RevId: 554977992
    arcra authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    063230f View commit details
    Browse the repository at this point in the history
  19. #tf-data-service Reduce data transfer protocol fallback log severity …

    …to `WARNING`.
    
    PiperOrigin-RevId: 554982965
    mpcallanan authored and tensorflower-gardener committed Aug 8, 2023
    Configuration menu
    Copy the full SHA
    44374f9 View commit details
    Browse the repository at this point in the history

Commits on Aug 9, 2023

  1. Mark all ops that are not currently exported to Python as visibility:…

    … HIDDEN.
    
    Add tf_export'ed raw_ops to modules_with_exports.
    
    PiperOrigin-RevId: 554985433
    DrMarcII authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    4c45953 View commit details
    Browse the repository at this point in the history
  2. Configuration menu
    Copy the full SHA
    1ca3599 View commit details
    Browse the repository at this point in the history
  3. Use imported tf_export directly instead of creating an alias.

    PiperOrigin-RevId: 555002989
    DrMarcII authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    f51e72e View commit details
    Browse the repository at this point in the history
  4. lite:stablehlo: add support for converting stablehlo constants betwee…

    …n MLIR & Flatbuffers
    
    PiperOrigin-RevId: 555004085
    zichuan-wei authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    5b90f56 View commit details
    Browse the repository at this point in the history
  5. Merge pull request tensorflow#61207 from Intel-tensorflow:zhoulong/en…

    …able_xla_auto_cluster_for_pjrt_device
    
    PiperOrigin-RevId: 5550338
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    5b3d815 View commit details
    Browse the repository at this point in the history
  6. Remove psimd dependency

    - psimd was used by XNNPack, but is no longer needed
    
    PiperOrigin-RevId: 555038715
    Maratyszcza authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    50288fe View commit details
    Browse the repository at this point in the history
  7. Cleanup unused headers in UniformQuantizedStablehloToTflPass.

    PiperOrigin-RevId: 555043552
    dansuh17 authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    884d342 View commit details
    Browse the repository at this point in the history
  8. Added test case for Changes in tf.experimental.numpy.vander

    Added a test case for checking the behaviour wrt numpy for the changes done in tf.experimental.numpy.vander.
    SuryanarayanaY authored Aug 9, 2023
    Configuration menu
    Copy the full SHA
    352bbf0 View commit details
    Browse the repository at this point in the history
  9. Resolved indentation errors in np_array_ops.py

    Resolved indentation error in vander function
    SuryanarayanaY authored Aug 9, 2023
    Configuration menu
    Copy the full SHA
    2555ac9 View commit details
    Browse the repository at this point in the history
  10. Make GetFusionEmitter callable before buffer assignment.

    For now, we just disable the in-place DUS/memcpy detection logic.
    
    PiperOrigin-RevId: 555067189
    jreiffers authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    0c49fec View commit details
    Browse the repository at this point in the history
  11. Merge pull request tensorflow#61498 from elfringham:run_build_tests

    PiperOrigin-RevId: 555069429
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    67ada06 View commit details
    Browse the repository at this point in the history
  12. Lower tf.XlaLaunch op to gpurt.compile_and_execute.

    PiperOrigin-RevId: 555077671
    changhuilin authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    6767047 View commit details
    Browse the repository at this point in the history
  13. compat: Update forward compatibility horizon to 2023-08-09

    PiperOrigin-RevId: 555091363
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    d6eb666 View commit details
    Browse the repository at this point in the history
  14. Update GraphDef version to 1583.

    PiperOrigin-RevId: 555091467
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    7944854 View commit details
    Browse the repository at this point in the history
  15. Make the comparison direction part of the CseKey for kCompare.

    This is a small efficiency gain, as we can potentially already detect a
    difference before calling Identical().
    
    PiperOrigin-RevId: 555093049
    akuegel authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    229a87e View commit details
    Browse the repository at this point in the history
  16. [XLA:GPU][NFC] Refactor Triton GEMM rewriter.

    Update comments, rename variables.
    
    PiperOrigin-RevId: 555093353
    Ilia Sergachev authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    170a63c View commit details
    Browse the repository at this point in the history
  17. Fix detection of same op in SliceSinker.

    The logic assumed that we just need to check the opcode, but for several ops we
    also need to check additional attributes. IdenticalSlowPath() is the method to
    check this. Expose this through another SameOp method in HloInstruction.
    
    PiperOrigin-RevId: 555094486
    akuegel authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    44f0ae5 View commit details
    Browse the repository at this point in the history
  18. Don't rewrite unsupported F8 types in gemm_rewriter.

    Fixes crash when using unsupported F8 types.
    
    PiperOrigin-RevId: 555116700
    reedwm authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    c8ad41f View commit details
    Browse the repository at this point in the history
  19. [XLA:GPU] Triton GEMM: support multiple splits of dimensions.

    Change the data structure describing tensors in the rewriter and its handling to make it more flexible and support more kinds transformations.
    
    PiperOrigin-RevId: 555125358
    Ilia Sergachev authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    0c103f7 View commit details
    Browse the repository at this point in the history
  20. [XLA:GPU] Pass fusion roots into HasAnyUnnestedReductionRoot.

    Committing again with a fix.
    
    PiperOrigin-RevId: 555134874
    olegshyshkov authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    b3afe4d View commit details
    Browse the repository at this point in the history
  21. Return error on invalid input in tfl.splitv

    PiperOrigin-RevId: 555138718
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    0051764 View commit details
    Browse the repository at this point in the history
  22. [XLA:GPU] Change the handling of split-K in Triton GEMM.

    Now contracting dimensions with split-K are represented as two fragments of one dimension instead of two separate dimensions.
    
    PiperOrigin-RevId: 555139630
    Ilia Sergachev authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    863e310 View commit details
    Browse the repository at this point in the history
  23. [XLA:GPU] Allow Triton Softmax fusions to be done in place.

    PiperOrigin-RevId: 555145026
    bchetioui authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    2b35d22 View commit details
    Browse the repository at this point in the history
  24. Remove TF Sizetracker helper scripts, which are no longer used

    PiperOrigin-RevId: 555154633
    angerson authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    9db429d View commit details
    Browse the repository at this point in the history
  25. [XLA:GPU] Relaxing shape check on collective ops during setting opera…

    …nd layouts. This was triggering for semantically correct ReduceScatter HLO.
    
    PiperOrigin-RevId: 555169308
    Moerafaat authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    675d73c View commit details
    Browse the repository at this point in the history
  26. [XLA:GPU] Remove --xla_gpu_enable_experimental_block_size flag.

    PiperOrigin-RevId: 555172036
    Aliia Khasanova authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    7f9513f View commit details
    Browse the repository at this point in the history
  27. [pjrt] Enable transpose kernels on machines without AVX

    PiperOrigin-RevId: 555177148
    majnemer authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    e09bd32 View commit details
    Browse the repository at this point in the history
  28. Merge implementations of tsl/platform/{default, google}/cord.h

    PiperOrigin-RevId: 555184894
    ddunl authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    7d5a4e7 View commit details
    Browse the repository at this point in the history
  29. Make all targets under compiler/mlir/tensorflow/tests/tf_saved_model/…

    … have strict dependencies.
    
    PiperOrigin-RevId: 555194981
    jam14j authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    dff18f0 View commit details
    Browse the repository at this point in the history
  30. Reenable flaky flag for ARM64 scripts.

    PiperOrigin-RevId: 555202253
    MichaelHudgins authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    b2e1982 View commit details
    Browse the repository at this point in the history
  31. Refactor conversion patterns in quantization Passses

    This change include modifying 3 passes: ConvertTFQuantTypes, ConvertTFQuantOpsToMHLO and ConvertMHLOQuantToInt.
    
    * In ConvertTFQuantTypes pass, we add qint <-> int tf.cast ops to surround TF UQ ops. We added some special op legality check in ConversionTarget to make sure the resulting TF UQ and tf.cast ops are legal.
    * In ConvertTFQuantOpsToMHLO pass, we add uq <-> int mhlo.convert ops to surround TF UQ ops and convert qint <-> int tf.cast ops to int <-> int mhlo.convert ops.
    * In ConvertMHLOQuantToInt pass, we convert uq <-> int mhlo.convert ops to int <-> int mhlo.convert ops
    
    The above passes may introduce int <-> int mhlo.convert ops. They are no-ops and can be removed in a Canonicalizer pass afterwards.
    
    PiperOrigin-RevId: 555206832
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    6639d33 View commit details
    Browse the repository at this point in the history
  32. [xla] Use latency-hiding-scheduler-preparation to prepare for the lat…

    …ency
    
    hiding scheduler for the GPU.
    
    PiperOrigin-RevId: 555212403
    bixia1 authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    ba411b6 View commit details
    Browse the repository at this point in the history
  33. Grant //visibility:public to xla/stream_executor/tpu:c_api_decl

    PiperOrigin-RevId: 555223964
    LionOfJewdah authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    2589294 View commit details
    Browse the repository at this point in the history
  34. Replaces the O(n^2) convertibility check with a simpler O(n) loop tha…

    …t examines elements along the diagonal only.
    
    PiperOrigin-RevId: 555226962
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    632d43c View commit details
    Browse the repository at this point in the history
  35. PR tensorflow#4864: [ROCm] Fixed amdgpu_compiler

    Imported from GitHub PR openxla/xla#4864
    
    amdgpu_compiler build error due to openxla/xla@6a68378
    
    @akuegel @ddunl Thanks in advance!
    Copybara import of the project:
    
    --
    2479f31ce053a2335e532772afb47d250fe3739f by Chao Chen <cchen104@amd.com>:
    
    add thread_pool to amdgpu_compiler
    
    Merging this change closes tensorflow#4864
    
    PiperOrigin-RevId: 555227394
    i-chaochen authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    14cf4a7 View commit details
    Browse the repository at this point in the history
  36. Adding output of instruction schedule, similar to bufferinfo and all…

    …ocinfo.
    
    PiperOrigin-RevId: 555232216
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    beef7ac View commit details
    Browse the repository at this point in the history
  37. Clean up tpu_configuration_ops

    PiperOrigin-RevId: 555232359
    LionOfJewdah authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    50d2f0d View commit details
    Browse the repository at this point in the history
  38. Merge pull request tensorflow#61079 from tensorflow:SuryanarayanaY-pa…

    …tch-1
    
    PiperOrigin-RevId: 555233600
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    79ae980 View commit details
    Browse the repository at this point in the history
  39. Refactor legalize mlir to hlo to enable combined bridge experiments l…

    …ater
    
    PiperOrigin-RevId: 555235182
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    b700f95 View commit details
    Browse the repository at this point in the history
  40. [XLA/debuggability] Add options to SliceModuleAndExtract() that can…

    … remove custom-call to sharding.
    
    PiperOrigin-RevId: 555242148
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    405a1a0 View commit details
    Browse the repository at this point in the history
  41. Modify requirements lock updater script to run relative to the script…

    … directory.
    
    PiperOrigin-RevId: 555245780
    MichaelHudgins authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    e7f9e0b View commit details
    Browse the repository at this point in the history
  42. [XLA/GPU] Restrict collective schedule linearizer to less cases.

    - Assume that in a multi-controller setting, XLA/GPU should see identical HLOs
      when multiple HLOs are compiled and communicate through collectives.
    - Given that, restrict CollectiveScheduleLinearizer pass to only handle cases
      where compiler auto tuning can case divergence in schedules across different
      compilations of the same HLO.
    
    PiperOrigin-RevId: 555248173
    jurahul authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    07c054b View commit details
    Browse the repository at this point in the history
  43. Add proto_splitter info to tensorflow.org guide.

    PiperOrigin-RevId: 555249702
    BlaziusMaximus authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    752a1ab View commit details
    Browse the repository at this point in the history
  44. [TF:PJRT] Returns early if num_gpus_to_use = 0.

    PiperOrigin-RevId: 555255267
    Jieying Luo authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    e3a6a3b View commit details
    Browse the repository at this point in the history
  45. Do not create dependencies among tf.RandomUniform ops

    This makes the MLIR side effect modelling of the op consistent with ACD, and improves run-time performance of some GPU graphs with these ops.
    
    PiperOrigin-RevId: 555261171
    jcai19 authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    4e2b4ca View commit details
    Browse the repository at this point in the history
  46. lite: Fix build rule of label_image

    Added missing xnnpack_plugin.cc. This PR resolves tensorflow#60261 issue.
    
    PiperOrigin-RevId: 555266459
    terryheo authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    281c2c4 View commit details
    Browse the repository at this point in the history
  47. Fix JAX build failure under Python 3.12.0rc1.

    The distutils module has been removed under Python 3.12, and the code here fails with a ModuleNotFoundError. This occurs because:
    
    ```
    $ ipython
    Python 3.12.0rc1 (main, Aug  7 2023, 13:07:47) [GCC 12.2.0]
    Type 'copyright', 'credits' or 'license' for more information
    IPython 8.14.0 -- An enhanced Interactive Python. Type '?' for help.
    
    In [1]: import importlib.util
    
    In [2]: importlib.util.find_spec('distutils.sysconfig')
    ---------------------------------------------------------------------------
    ModuleNotFoundError                       Traceback (most recent call last)
    Cell In[2], line 1
    ----> 1 importlib.util.find_spec('distutils.sysconfig')
    
    File <frozen importlib.util>:91, in find_spec(name, package)
    
    ModuleNotFoundError: No module named 'distutils'
    ```
    
    However, as far as I can tell, we can just unconditionally use sysconfig since Python 3.2, which is well below the minimum Python version requirement of this code.
    
    This change effectively reverts: tensorflow#54292
    That change will no longer be tenable once Python 3.12 is released.
    
    PiperOrigin-RevId: 555273752
    hawkinsp authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    87ec04d View commit details
    Browse the repository at this point in the history
  48. Add a TPU nightly job

    PiperOrigin-RevId: 555274419
    kanglant authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    1074a7c View commit details
    Browse the repository at this point in the history
  49. Split tensorflow/compiler/tf2xla/kernels:xla_ops into fine grained ta…

    …rgets.
    
    PiperOrigin-RevId: 555277315
    BrianWieder authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    90de990 View commit details
    Browse the repository at this point in the history
  50. Remove tf_exports that are not being exported.

    PiperOrigin-RevId: 555277542
    BrianWieder authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    c09c4d5 View commit details
    Browse the repository at this point in the history
  51. Add a configuration to SliceModuleAndExtract() that removes the tup…

    …le element in the parameter instructions that have no users.
    
    PiperOrigin-RevId: 555286907
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    eb653bf View commit details
    Browse the repository at this point in the history
  52. Add declaration for a dynamic ragged tensor op, which accepts a devic…

    …e ordinal
    
    as an input.
    
    PiperOrigin-RevId: 555289711
    rjpower authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    80fe343 View commit details
    Browse the repository at this point in the history
  53. [xla:gpu] Disable concurrent regions in op-by-op execution

    PiperOrigin-RevId: 555292150
    anlunx authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    3ce4a89 View commit details
    Browse the repository at this point in the history
  54. Creating Converter from python list of tensors to c++ vector of tensors

    PiperOrigin-RevId: 555296787
    Malik Stewart authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    1748e75 View commit details
    Browse the repository at this point in the history
  55. Update ops-related pbtxt files.

    PiperOrigin-RevId: 555298522
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    f922c21 View commit details
    Browse the repository at this point in the history
  56. Enable XLA detailed logging when using PJRT

    PiperOrigin-RevId: 555302357
    swachhandl authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    b6f9258 View commit details
    Browse the repository at this point in the history
  57. lite:stablehlo: add serialization for stablehlo add, multiply, divide…

    … and maximum op
    
    PiperOrigin-RevId: 555305030
    zichuan-wei authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    9b5e827 View commit details
    Browse the repository at this point in the history
  58. Fix in-depth-guide links.

    PiperOrigin-RevId: 555305766
    BlaziusMaximus authored and tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    ee02ea9 View commit details
    Browse the repository at this point in the history
  59. Adding new device flag for compilation cache

    If the device flag is specified, then the cache will only be used for that specific device. By default, this change does impact how the compilation cache works.
    
    This flag is being added to prevent errors from occurring when running on hardware with multiple device types.
    
    PiperOrigin-RevId: 555311811
    tensorflower-gardener committed Aug 9, 2023
    Configuration menu
    Copy the full SHA
    fa481b4 View commit details
    Browse the repository at this point in the history

Commits on Aug 10, 2023

  1. [JAX] Introduce DeviceList backed by C++ xla::ifrt::DeviceList

    This change adds `xla_client.DeviceList` that is implemented in C++
    `jax::PyDeviceList`. `jax::PyDeviceList` implements the features of
    `pxla._DeviceAssignment` as a functional drop-in replacement.
    `jax::PyDeviceList` internally has `xla::ifrt::DeviceList`, which will be used
    when using IFRT APIs without having to construct a new copy of a potentially
    large device list.
    
    `pxla._DeviceAssignment`'s interface is changed slightly to encourage avoiding
    conversion to tuple.
    
    Note that for the backward compatibility (and fast `xla_client.Device`
    conversion), `jax::PyDeviceList` still uses a Python tuple whose element can be
    any Python object matches `xla_client.Device` interface with duck typing. This
    duck typing support will be removed when such use case is deprecated.
    Eventually, we can try to avoid any type conversion to remove a shadow copy of
    device list in JAX.
    
    PiperOrigin-RevId: 555317152
    hyeontaek authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    faae949 View commit details
    Browse the repository at this point in the history
  2. [XLA/debuggability] Extend hlo_extractor to support replacing an ex…

    …cluded instruction with a constant broadcasted from a random number.
    
    PiperOrigin-RevId: 555317504
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    631d582 View commit details
    Browse the repository at this point in the history
  3. Configuration menu
    Copy the full SHA
    cbf180c View commit details
    Browse the repository at this point in the history
  4. Passing un-parsed key through C API

    The parsed key solution won't have a correct buf_ and could break the object lifespan and ParsedKey::FullKey()
    
    PiperOrigin-RevId: 555327807
    hhb authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    349fcb4 View commit details
    Browse the repository at this point in the history
  5. Integrate LLVM at llvm/llvm-project@6556e2902570

    Updates LLVM usage to match
    [6556e2902570](llvm/llvm-project@6556e2902570)
    
    PiperOrigin-RevId: 555348867
    slackito authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    eef530d View commit details
    Browse the repository at this point in the history
  6. Configuration menu
    Copy the full SHA
    2a0efa4 View commit details
    Browse the repository at this point in the history
  7. Add String based category time to GenericStepBreakdown

    PiperOrigin-RevId: 555370639
    cliveverghese authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    e3a6a75 View commit details
    Browse the repository at this point in the history
  8. Merge pull request tensorflow#61503 from matthew-olson-intel:master

    PiperOrigin-RevId: 555379542
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    7903f31 View commit details
    Browse the repository at this point in the history
  9. Don't recompute reduction heroes O(n) times.

    PiperOrigin-RevId: 555392263
    jreiffers authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    347c568 View commit details
    Browse the repository at this point in the history
  10. Add a pattern that converts stablehlo.dot_general to `tfl.fully_con…

    …nected` when the filter is per-axis quantized.
    
    This change adds a new pattern in `UniformQuantizedStablehloToTflPass` that converts quantized `stablehlo.dot_general` to `tfl.fully_connected`. This pattern is applied when the filter is per-axis quantized.
    
    PiperOrigin-RevId: 555405376
    dansuh17 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    11a7b51 View commit details
    Browse the repository at this point in the history
  11. [xla:gpu] NFC: Switch from CUDA runtime API to CUDA driver API in CUD…

    …A graph integration
    
    Similar to other Gpu libraries move the platform specific implementation to GpuDriver.
    
    PiperOrigin-RevId: 555408360
    ezhulenev authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    e479b33 View commit details
    Browse the repository at this point in the history
  12. [XLA:GPU] Triton GEMM: decouple properties of block pointers.

    Derive block pointer parameters separately for each input tensor so that one scope of a dot can access tensors with different shapes and layouts. This is ~NFC for now, it is required to enable more fusions but they will only come with other updates to the rewriter and the analysis.
    
    PiperOrigin-RevId: 555414634
    Ilia Sergachev authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    2929172 View commit details
    Browse the repository at this point in the history
  13. Update GraphDef version to 1584.

    PiperOrigin-RevId: 555416852
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    363d047 View commit details
    Browse the repository at this point in the history
  14. compat: Update forward compatibility horizon to 2023-08-10

    PiperOrigin-RevId: 555417053
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    bd2329d View commit details
    Browse the repository at this point in the history
  15. Allow all elementwise unary ops in Reduce Epilogue fusion.

    Also fix IsInputFusibleReduction(), it was using HasRealReductionHero()
    instead of IsReductionFromOrToContiguousDimensions()
    
    PiperOrigin-RevId: 555458297
    akuegel authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    ab146e6 View commit details
    Browse the repository at this point in the history
  16. Update to use keras-nightly~=2.15.0.dev

    PiperOrigin-RevId: 555458586
    sampathweb authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    193d5f2 View commit details
    Browse the repository at this point in the history
  17. Return error on invalid input in tfl.atan2

    PiperOrigin-RevId: 555461404
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    cf5c14e View commit details
    Browse the repository at this point in the history
  18. Fork a tensor_matcher to tensorflow. It supports easier comparison of…

    … POD and string type of
    
    tensor match on list of tensors.
    
    PiperOrigin-RevId: 555508096
    deqiangc authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    f3a0e6b View commit details
    Browse the repository at this point in the history
  19. [XLA] Matching partial replicated sharding.

    Adds detection of partial replicated input compatible with fully sharded output
    for contracting dimension matching case.
    
    PiperOrigin-RevId: 555512892
    Marcello Maggioni authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    caa25d5 View commit details
    Browse the repository at this point in the history
  20. Integrate LLVM at llvm/llvm-project@6448d5ba581a

    Updates LLVM usage to match
    [6448d5ba581a](llvm/llvm-project@6448d5ba581a)
    
    PiperOrigin-RevId: 555515803
    gribozavr authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    6e0495a View commit details
    Browse the repository at this point in the history
  21. [XLA:GPU] Add verification pass to CompileAndPrintLlvmIr.

    PiperOrigin-RevId: 555527275
    bchetioui authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    15d1d0e View commit details
    Browse the repository at this point in the history
  22. Clean up package tensorflow/core/tpu/kernels/xla

    PiperOrigin-RevId: 555546999
    LionOfJewdah authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    1d1f64d View commit details
    Browse the repository at this point in the history
  23. Add some return type annotations for dataset_ops.py.

    PiperOrigin-RevId: 555552149
    fionalang authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    52f9856 View commit details
    Browse the repository at this point in the history
  24. Configuration menu
    Copy the full SHA
    966a270 View commit details
    Browse the repository at this point in the history
  25. Replace a test case that causes an overflow on TAP ASAN.

    PiperOrigin-RevId: 555563851
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    36b7de6 View commit details
    Browse the repository at this point in the history
  26. #tf-data Prune an irrelevant dependency for standalone.

    PiperOrigin-RevId: 555567528
    yangustc07 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    045ed57 View commit details
    Browse the repository at this point in the history
  27. Move out OpMetricsDb builder into utility class.

    PiperOrigin-RevId: 555570649
    cliveverghese authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    34c9c26 View commit details
    Browse the repository at this point in the history
  28. Legalize mhlo.dot_general with dynamic shaped inputs. tf.unsorted_seg…

    …ment_prod, tf.gather and tf.concat are used to calculate the operand flattened shapes.
    
    Here's how the dot_general legalization work:
    
    1. lhs_transposed = transpose (lhs, (batch_dims, out_dims, contracting_dims))
    2. rhs_transposed = transpose (rhs, (batch_dims, contracting_dims, out_dims))
    3. lhs_reshaped = reshape(lhs_transposed, (batch_dims, flattened_out_dim, flattened_contracting_dim))
    4. rhs_reshaped = reshape(rhs_transposed, (batch_dims, flattened_contracting_dim, flattened_out_dim))
    5. result = BMM(lhs_reshaped, rhs_reshaped)
    6. result_reshaped = reshape(result, (batch_dims, lhs_out_dims, rhs_out_dims))
    
    PiperOrigin-RevId: 555571369
    yishuangP authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    cff924e View commit details
    Browse the repository at this point in the history
  29. Improve documentation for bincount mathops

    PiperOrigin-RevId: 555571653
    SeeForTwo authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    65bd674 View commit details
    Browse the repository at this point in the history
  30. Enable MLIR_BRIDGE_LOG_PASS_FILTER for DataDumperLoggerConfig.

    This functionality is in the base class and should also be in the derived class.  It's useful when only one of the intermediate MLIR dumps is needed.
    
    PiperOrigin-RevId: 555571800
    kenfranko authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    e78cfa0 View commit details
    Browse the repository at this point in the history
  31. Remove Conv2D and Conv3D GPU kernels and replace them with general ke…

    …rnel.
    
    PiperOrigin-RevId: 555582618
    nicolagp authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    026a36e View commit details
    Browse the repository at this point in the history
  32. Add _copy_trackable_to_cpu() to Trackable, refactor AsyncCheckpoint…

    … accordingly
    
    PiperOrigin-RevId: 555586385
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    aef0c37 View commit details
    Browse the repository at this point in the history
  33. #tf-data Prune an irrelevant dependency for server_lib.

    PiperOrigin-RevId: 555586536
    yangustc07 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    6e08a1b View commit details
    Browse the repository at this point in the history
  34. Add a warning to ensure no TF APIs have been called before enabling t…

    …he new type promotion.
    
    The APIs supported by the new type promotion are monkey-patched. However, if the API has been called before monkey-patching happens, it creates a reference in local namespace and monkey-patching has no effect: http://shortn/_uho9H0wY2U
    
    PiperOrigin-RevId: 555587891
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    56a9dbc View commit details
    Browse the repository at this point in the history
  35. PR tensorflow#4889: [ROCm] rocm updated graph api and fixed hlo_op_pr…

    …ofiler_test
    
    Imported from GitHub PR openxla/xla#4889
    
    1. rocm updates graph api due to openxla/xla@214a67c
    2. rocm adds hlo_op_profiler_test due to openxla/xla@b79e02f#diff-a20d455ecfbd936acdc085c4e33efef62a01739020d74fd716d7eea27680c09c
    
    @akuegel @ddunl @ezhulenev
    
    Thanks in advance!
    Copybara import of the project:
    
    --
    9ea7cbda5746cab11348246ebe5b343a80a0f373 by Chao Chen <cchen104@amd.com>:
    
    rocm updated graph api and fixed hlo_op_profiler_test
    
    Merging this change closes tensorflow#4889
    
    PiperOrigin-RevId: 555595849
    i-chaochen authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    8e1f218 View commit details
    Browse the repository at this point in the history
  36. #tf-data Prune dependencies for rewrite_utils.

    PiperOrigin-RevId: 555606211
    yangustc07 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    adfacb2 View commit details
    Browse the repository at this point in the history
  37. [Memories] Check if the pjrt_buffer memory_space's devices match the …

    …devices of the sharding along with memory_kind
    
    PiperOrigin-RevId: 555608103
    yashk2810 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    96f5899 View commit details
    Browse the repository at this point in the history
  38. Update find_packages() to find_namespace_packages()

    This seems to resolve the issue where assembling the TF pip wheel dumps thousands of lines of warnings about deprecated behavior. It does not appear to change the contents of the pip wheel, from what I can tell, so I think it's fine to do.
    
    PiperOrigin-RevId: 555612180
    angerson authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    9b0d268 View commit details
    Browse the repository at this point in the history
  39. [PJRT C API] Simplify TPU initialization logic in the framework side.

    - Move Logics in `InitializeTpuLibrary` to `PJRT_Plugin_Initialize` of TPU plugin.
    - The framework still needs to call `InitializeTpuStructFns` to `dlsym` some C APIs that are not part of PJRT C APIs.
    
    PiperOrigin-RevId: 555612657
    Jieying Luo authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    208a6f3 View commit details
    Browse the repository at this point in the history
  40. [XLA/layout assignment] Explicitly handle output-operand aliasing whe…

    …n propagating buffer layouts. Do not alias buffers connected by def-use relations.
    
    PiperOrigin-RevId: 555612801
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    7da9e88 View commit details
    Browse the repository at this point in the history
  41. [PJRT C API] Implement AcquireExternalReference functionalities for P…

    …JRT C API.
    
    PiperOrigin-RevId: 555619438
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    df938c1 View commit details
    Browse the repository at this point in the history
  42. #tf-data Prune dependencies for compression_utils.

    PiperOrigin-RevId: 555620750
    yangustc07 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    6d55033 View commit details
    Browse the repository at this point in the history
  43. [stream_executor] Add support for graph building with cuGraphAddKerne…

    …lNode
    
    PiperOrigin-RevId: 555622861
    ezhulenev authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    204b924 View commit details
    Browse the repository at this point in the history
  44. Add cleanup, resultstore options & fix bugs

    This bundles a few changes:
    
    - Added --config=resultstore option to all bazel invocations, which will
      create easy-to-read resultstore logs for individual invocations in official
      jobs
    - Fix an issue in rename_and_verify_wheels.sh where the wheel would be deleted
    - Copy all script output for all scripts to build/script.log for easier review
      and post-processing
    - Renamed DOCKER_GPU_ARGS to DOCKER_ARGS, since it's a convenient place to put
      other args as well, such as volume mounts
    - Added cleanup scripts that dump useful information after a job is complete,
      including a list of any remote resultstore invocations for easy reading
    - Re-use the same docker container if it still exists, which makes re-running
      easier
    
    PiperOrigin-RevId: 555625021
    angerson authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    0fae071 View commit details
    Browse the repository at this point in the history
  45. #tf-data-service Add return type annotations for tf.data service publ…

    …ic APIs.
    
    PiperOrigin-RevId: 555626619
    yangustc07 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    0abdf37 View commit details
    Browse the repository at this point in the history
  46. Implement RegionBranchOpInterface for IfRegionOp.

    PiperOrigin-RevId: 555633866
    matthiaskramm authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    8326102 View commit details
    Browse the repository at this point in the history
  47. Export original tf function if tf._original_func_name is set

    PiperOrigin-RevId: 555639415
    cky9301 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    4d5807f View commit details
    Browse the repository at this point in the history
  48. Add a haswell CPU config to Tensorflow and link PJRT deps for Haswe…

    …ll in gpu_device.
    
    This is for internal use only.
    
    PiperOrigin-RevId: 555641900
    swachhandl authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    f079f12 View commit details
    Browse the repository at this point in the history
  49. Adding read-only feature to XLA Compilation Cache

    When the `tf_xla_persistent_cache_directory_read_only` flag is set to true, the XLA compilation cache will only read from the cache. This is useful in cases where the xla cache directory is read-only.
    
    By default, there is no change to how the XLA compilation cache behaves.
    
    PiperOrigin-RevId: 555643678
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    ab0b3e6 View commit details
    Browse the repository at this point in the history
  50. Fix internal tests.

    PiperOrigin-RevId: 555650825
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    31f9fa2 View commit details
    Browse the repository at this point in the history
  51. Support dynamic shapes in infeed

    PiperOrigin-RevId: 555653779
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    e79c98c View commit details
    Browse the repository at this point in the history
  52. Internal TFRT change.

    PiperOrigin-RevId: 555660748
    cky9301 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    d93239d View commit details
    Browse the repository at this point in the history
  53. Turn off pipelining assertions for num_steps>=2

    PiperOrigin-RevId: 555664892
    patnotz authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    3d92ebb View commit details
    Browse the repository at this point in the history
  54. Allow destruction of RendezvousInterface

    PiperOrigin-RevId: 555674689
    hhb authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    c3be981 View commit details
    Browse the repository at this point in the history
  55. [Memories] Allow device_put outside jax.jit to work with different me…

    …mory kinds.
    
    Currently only jax.Arrays work. Other types will be added in subsequent CLs.
    
    PiperOrigin-RevId: 555677540
    yashk2810 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    78abbb4 View commit details
    Browse the repository at this point in the history
  56. Fix internal tests.

    PiperOrigin-RevId: 555679734
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    edbbabf View commit details
    Browse the repository at this point in the history
  57. Add more subgraph variations in tflite tests

    PiperOrigin-RevId: 555680178
    majiddadashi authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    9b36fd6 View commit details
    Browse the repository at this point in the history
  58. Use TF quantization option directly in Inference Converter V1

    PiperOrigin-RevId: 555686551
    thaink authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    8923f0e View commit details
    Browse the repository at this point in the history
  59. Add support for NumPy Dtype class in flexible_dtypes.

    PiperOrigin-RevId: 555687566
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    a8aadc9 View commit details
    Browse the repository at this point in the history
  60. New option '--tf_mlir_enable_strict_clusters'.

    PiperOrigin-RevId: 555690457
    matthiaskramm authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    4988fed View commit details
    Browse the repository at this point in the history
  61. Clean up package tensorflow/core/tpu/ops

    PiperOrigin-RevId: 555694915
    LionOfJewdah authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    29dd40f View commit details
    Browse the repository at this point in the history
  62. Configuration menu
    Copy the full SHA
    5ca436f View commit details
    Browse the repository at this point in the history
  63. Clean up package tensorflow/core/tpu/graph_rewrite

    PiperOrigin-RevId: 555700124
    LionOfJewdah authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    e4e3439 View commit details
    Browse the repository at this point in the history
  64. [Memories] Allow device_put outside jax.jit to work with numpy arrays…

    …, numpy scalars, python scalars and python ints with memory kinds
    
    PiperOrigin-RevId: 555703325
    yashk2810 authored and tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    1540038 View commit details
    Browse the repository at this point in the history
  65. Move the mlir bridge second phase metric definition into the framewor…

    …k metrics file.
    
    PiperOrigin-RevId: 555707138
    tensorflower-gardener committed Aug 10, 2023
    Configuration menu
    Copy the full SHA
    d0165a2 View commit details
    Browse the repository at this point in the history

Commits on Aug 11, 2023

  1. #tf-data-service Ramp up "no_compression" experiment to 50%.

    PiperOrigin-RevId: 555717012
    mpcallanan authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    e03250c View commit details
    Browse the repository at this point in the history
  2. Update tensorboard tb-nightly dependency, as a follow-up to the 2.14.…

    …0 release.
    
    The tb-nightly 2.15.0.a* pypi package is now available: https://pypi.org/project/tb-nightly/2.15.0a20230810/
    
    PiperOrigin-RevId: 555718964
    arcra authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    e42b64a View commit details
    Browse the repository at this point in the history
  3. Fix a typo

    PiperOrigin-RevId: 555726754
    angerson authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    11b885c View commit details
    Browse the repository at this point in the history
  4. Restore if_cuda_or_rocm and if_static checks to tf2xla/kernels tf…

    …_kernel_library BUILD targets.
    
    PiperOrigin-RevId: 555726777
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    9e4b9da View commit details
    Browse the repository at this point in the history
  5. Configuration menu
    Copy the full SHA
    0236c9b View commit details
    Browse the repository at this point in the history
  6. [Memories][PJRT:C] Add PJRT_Client_AddressableMemories and PJRT_Memor…

    …y_AddressableByDevices.
    
    Also refactors the code to make the relationship between client, device, and memory space more clear, i.e., device and memory space are owned and attached to each other by client.
    
    PiperOrigin-RevId: 555734053
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    a39e7f9 View commit details
    Browse the repository at this point in the history
  7. Added unit tests, aot check for loadsavedmodel, deserialization proce…

    …ss for aot_packages if found, a new aot model and according unit testing.
    
    PiperOrigin-RevId: 555736121
    IMMZEK authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    85d5a35 View commit details
    Browse the repository at this point in the history
  8. address comments

    jzhoulon committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    49e285d View commit details
    Browse the repository at this point in the history
  9. Integrate LLVM at llvm/llvm-project@f2d32ddcec82

    Updates LLVM usage to match
    [f2d32ddcec82](llvm/llvm-project@f2d32ddcec82)
    
    PiperOrigin-RevId: 555771026
    slackito authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    2b3936e View commit details
    Browse the repository at this point in the history
  10. Configuration menu
    Copy the full SHA
    4a5a9a8 View commit details
    Browse the repository at this point in the history
  11. Remove unneeded parameter.

    PiperOrigin-RevId: 555802731
    changhuilin authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    a67a92c View commit details
    Browse the repository at this point in the history
  12. [Refactoring] Separate Xla op conversion from PrepareLiftingPass

    This CL introduces a new pass called ConvertXlaOpToTfOpPass that
    handles Xla op to TF op conversion separately from PrepareLiftingPass.
    No behavior change is expected from this change.
    
    PiperOrigin-RevId: 555811966
    sngyhan authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    94d53dc View commit details
    Browse the repository at this point in the history
  13. [XLA:Python] Fix strict-weak-ordering problem in dictionary key sort.

    If a key comparison fails, it's simplest just to throw an exception right there and then, rather than return an arbitrary comparison value which might violate the strict-weak-ordering contract of C++ sorts.
    
    PiperOrigin-RevId: 555819401
    hawkinsp authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    429aa41 View commit details
    Browse the repository at this point in the history
  14. Allow composing quantized stablehlo.dot_general when the i8->f32 ca…

    …st isn't present.
    
    This pattern may emerge when the filter constant is constant-folded (by e.g. `jax.jit` for small tensors).
    This change also adds an e2e integration test for the quantized dot_general.
    
    PiperOrigin-RevId: 555824049
    dansuh17 authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    15507d4 View commit details
    Browse the repository at this point in the history
  15. [NFC] Use FindNonTrivialHero for reductions.

    FindNonTrivialHero works for reduction fusions as well. It either returns the reduction at the end of the chain of intermediate instructions starting at the root, or a transpose.
    
    Also remove some helper functions. It's too hard to keep track of which one does what.
    
    PiperOrigin-RevId: 555833379
    jreiffers authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    07a0519 View commit details
    Browse the repository at this point in the history
  16. compat: Update forward compatibility horizon to 2023-08-11

    PiperOrigin-RevId: 555854685
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    9a9318b View commit details
    Browse the repository at this point in the history
  17. Update GraphDef version to 1585.

    PiperOrigin-RevId: 555854776
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    4c019a0 View commit details
    Browse the repository at this point in the history
  18. Expand c_api_opaque.h functions, specifically:

    - `TfLiteOpaqueNodeGetInputTensorIndex` to obtain the (global) tensor index of a 'node's input tensor.
    - `TfLiteOpaqueNodeGetOutputTensorIndex` to obtain the (global) tensor index of a 'node's output tensor.
    - `TfLiteOpaqueContextGetSizeOfType` to obtain the size (in bytes) of a `TfLiteType`.
    
    PiperOrigin-RevId: 555860090
    Matt Kreileder authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    3dfc7d1 View commit details
    Browse the repository at this point in the history
  19. [XLA:GPU] Fix integer overflow when number of elements exceeds INT32_…

    …MAX in Softmax
    
    Triton emitter.
    
    Fixes jax-ml/jax#16973.
    
    PiperOrigin-RevId: 555920623
    bchetioui authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    05c1d3c View commit details
    Browse the repository at this point in the history
  20. [XLA:GPU] Generalize propagation of dimensions in Triton GEMM rewriter.

    Change the function interfaces to support HLOs with different dimension orders on operands. At this point this is ~NFC because the actual analysis for such HLOs will be added in next CLs.
    
    PiperOrigin-RevId: 555928758
    Ilia Sergachev authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    555b61b View commit details
    Browse the repository at this point in the history
  21. [XLA:Python] Use consistent return type for PyDeviceList::hash() to f…

    …ix Mac OS build failure.
    
    PiperOrigin-RevId: 555935235
    hawkinsp authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    5c95cd4 View commit details
    Browse the repository at this point in the history
  22. Convert IFRT dtypes directly to NumPy dtypes whenever possible

    This is a no-op change that avoids `xla::PrimitiveType` in the middle when converting IFRT dtypes to NumPy dtypes. IFRT and XLA may not have the same set of dtypes (e.g., `xla::ifrt::DType::kString`), so avoiding the XLA dtype during conversion helps address potential incompatibility issues.
    
    PiperOrigin-RevId: 555937322
    junwhanahn authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    f1b7327 View commit details
    Browse the repository at this point in the history
  23. Return error on invalid input in tfl.sign

    PiperOrigin-RevId: 555938296
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    871c665 View commit details
    Browse the repository at this point in the history
  24. Add a dtype conversion rule for IFRT kString dtype

    String dtypes are converted into "object" dtypes in NumPy in order to be able to express variable-length strings. This is consistent with TensorFlow's `TF_DataType_to_PyArray_TYPE`.
    
    PiperOrigin-RevId: 555949700
    junwhanahn authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    c7c4112 View commit details
    Browse the repository at this point in the history
  25. TfCThunkRendezvous keep a copy of TF_RendezvousThunk

    PiperOrigin-RevId: 555955687
    hhb authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    00013a6 View commit details
    Browse the repository at this point in the history
  26. Pass heroes into helper functions that need them.

    We need to generalize all these helpers to work with partially fused code. There are two ways we could do this:
    
    - Overload all these functions to take boundary functions
    - Hoist the hero computations out of them
    
    The second option is preferable, since we also recompute the heroes unnecessarily in many cases.
    
    PiperOrigin-RevId: 555967200
    jreiffers authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    0a493a0 View commit details
    Browse the repository at this point in the history
  27. Extract some common functions in quantization/stablaehlo/passes/bridg…

    …e into a helper library and add some tests. Also remove GetSameShapeTensorType function and use a simpler alternative.
    
    PiperOrigin-RevId: 555972648
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    ef809fa View commit details
    Browse the repository at this point in the history
  28. [NFC] Clean up dead code.

    PiperOrigin-RevId: 555974072
    jreiffers authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    104f63c View commit details
    Browse the repository at this point in the history
  29. Support tf.LegacyCall when exporting the original func name.

    PiperOrigin-RevId: 555983776
    cky9301 authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    639a8f2 View commit details
    Browse the repository at this point in the history
  30. Fix handling of non-default layouts in DotDimensionMerger.

    PiperOrigin-RevId: 555998149
    akuegel authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    1265de0 View commit details
    Browse the repository at this point in the history
  31. Make all targets under tensorflow/compiler/xla/python/ have strict de…

    …pendencies.
    
    PiperOrigin-RevId: 556000216
    jam14j authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    2ed50f5 View commit details
    Browse the repository at this point in the history
  32. Clean up tensorflow/core/tpu/kernels apart from compilation cache tar…

    …gets
    
    PiperOrigin-RevId: 556000733
    LionOfJewdah authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    e47be9e View commit details
    Browse the repository at this point in the history
  33. Prepare to enable FP8 in XLA runtime.

    This fixes the CublasLtMatmulF8Op support in XLA runtime and adds support for the ConvForwardGraphOp op, which is used for FP8 convolutions. I'll enable FP8 in XLA runtime in a subsequent CL, since it will be easier to rollback the small change if there ends up being issues.
    
    PiperOrigin-RevId: 556000737
    reedwm authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    fd9d01e View commit details
    Browse the repository at this point in the history
  34. Merge pull request tensorflow#61502 from Intel-tensorflow:bhavanis/co…

    …de-cleanup
    
    PiperOrigin-RevId: 556001911
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    181e4da View commit details
    Browse the repository at this point in the history
  35. Fixes a bug that rejects a root instruction being fused into another …

    …called computation.
    
    PiperOrigin-RevId: 556004118
    Yunlong Liu authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    35317d5 View commit details
    Browse the repository at this point in the history
  36. [xla:gpu2] Add gpu graph support to gpu2 backend

    Create Gpu graphs in 3 steps:
    (1) Identify regions of operations that can be captured into a graph and wrap them into xla_gpu.graph.region operation
    (2) When lowering LMHLO operations to API call, select explicit graph building APIs when operation is inside a graph region
    (3) Lower gpu graph regions to API calls that create graph instances
    
    A lot of performance left on the table by not reusing executable graphs, this will come in the follow up changes.
    
    PiperOrigin-RevId: 556007396
    ezhulenev authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    902b786 View commit details
    Browse the repository at this point in the history
  37. Add pattern to convert qint tf.ConstOp to int in ConvertTFQuantTypes …

    …pass.
    
    Also handle the case where RHS of UQ ops are not constant in ConvertTFQuantToMHLO pass.
    
    PiperOrigin-RevId: 556013218
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    2c859ef View commit details
    Browse the repository at this point in the history
  38. Allow WhileRegionOp to have a "yield" in the "cond" block that forwar…

    …ds its
    
    operands.
    
    This makes WhileRegionOp compatible with RegionBranchOpInterface.
    
    PiperOrigin-RevId: 556014608
    matthiaskramm authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    73599e6 View commit details
    Browse the repository at this point in the history
  39. Add return type annotations for tf.ragged.constant and tf.ragged.cons…

    …tant_value.
    
    PiperOrigin-RevId: 556020656
    fionalang authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    87c527c View commit details
    Browse the repository at this point in the history
  40. [XLA/GPU] Add convenience XLA flags for async and pipelined collectives

    - Add flag `xla_gpu_enable_async_collectives` to turn on async variants of
      all collectives.
    - Add flag `xla_gpu_enable_pipelined_collectives` to turn on pipelining of
      all supported collectives.
    - with these flags, we can still have individual control by turning off the
      convenience flag and then using the existing individual flags
    
    PiperOrigin-RevId: 556026882
    jurahul authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    86b994e View commit details
    Browse the repository at this point in the history
  41. Add support for updating an HloSchedule after inserting computations …

    …to the module.
    
    PiperOrigin-RevId: 556028521
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    e2d724d View commit details
    Browse the repository at this point in the history
  42. #tf-nightly Increase timeout of flaky tests in the nightly build.

    PiperOrigin-RevId: 556029804
    yangustc07 authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    1467c45 View commit details
    Browse the repository at this point in the history
  43. Update legacy references to tensor.Tensor.

    PiperOrigin-RevId: 556031440
    fionalang authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    86564d0 View commit details
    Browse the repository at this point in the history
  44. Pass a run_mode param in IteratorContext

    This way, we can specify if the tf.data pipeline is running in a `standalone::Iterator` or other future conditions.
    
    PiperOrigin-RevId: 556037598
    armandouv authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    00500bf View commit details
    Browse the repository at this point in the history
  45. Configuration menu
    Copy the full SHA
    2ada77d View commit details
    Browse the repository at this point in the history
  46. [XLA/GPU] Allow ignoring control dependencies during CSE

    - Add a flag to allow ignoring control deps when replacing one instruction
      with another during CSE.
    - Enable this flag when CSE is exercised after collective schedule linearize
      to allow CSE of collectives.
    
    PiperOrigin-RevId: 556041273
    jurahul authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    e3303d9 View commit details
    Browse the repository at this point in the history
  47. Fix internal tests.

    PiperOrigin-RevId: 556041286
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    9d18ab2 View commit details
    Browse the repository at this point in the history
  48. Extract out split_into_island_per_op_pass into its own CC target in p…

    …reparation to make a transforms/BUILD target
    
    PiperOrigin-RevId: 556046506
    changm authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    4e49158 View commit details
    Browse the repository at this point in the history
  49. Delete constructors for CommonOpaqueConversionUtil and OpResolverInte…

    …rnal.
    
    PiperOrigin-RevId: 556046765
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    4fbc35a View commit details
    Browse the repository at this point in the history
  50. Delete legacy reference to tensor.Tensor in framework/ops.py.

    PiperOrigin-RevId: 556053947
    fionalang authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    34a3171 View commit details
    Browse the repository at this point in the history
  51. Extend OptimizeInputOutputBufferAlias, allowing matching the regist…

    …ered buffer donors.
    
    This `HloModulePass` previously considered all the input-output pairs. We add a switch to allow it consider the registered buffer donors only. We also refactor the matching algorithm.
    
    PiperOrigin-RevId: 556055615
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    86cb18a View commit details
    Browse the repository at this point in the history
  52. Examples for bincount with sparse and ragged inputs.

    PiperOrigin-RevId: 556075441
    SeeForTwo authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    8067c7a View commit details
    Browse the repository at this point in the history
  53. Move some files from translate to transform directory in preparation …

    …to make transform a separate BUILD package
    
    PiperOrigin-RevId: 556076698
    changm authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    fa181db View commit details
    Browse the repository at this point in the history
  54. Add additional TF-NumPy methods to WeakTensor.

    In TF-NumPy, ndarray is introduced as an alias of tf.Tensor. A few additional methods that don't exist in tf.Tensor are added to ndarray class in np_array_ops: http://shortn/_FNWKMUHNan. Look for 'setattr' in np_array_ops to see all the methods that are added to ndarray class.
    
    PiperOrigin-RevId: 556078735
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    cf8e6b8 View commit details
    Browse the repository at this point in the history
  55. [NFC] Cleanup unused headers and dependencies in HloCSE

    PiperOrigin-RevId: 556080870
    jurahul authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    7a44dcb View commit details
    Browse the repository at this point in the history
  56. [XLA:Python] Remove some unused includes.

    NFC intended.
    
    PiperOrigin-RevId: 556091429
    hawkinsp authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    73e29d2 View commit details
    Browse the repository at this point in the history
  57. Merge pull request tensorflow#61237 from Intel-tensorflow:amin/pr-xla…

    …-onednn-matmul
    
    PiperOrigin-RevId: 556093307
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    8218b16 View commit details
    Browse the repository at this point in the history
  58. #tf-data-service Add return type annotations for tf.data service.

    PiperOrigin-RevId: 556094425
    yangustc07 authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    d0c505b View commit details
    Browse the repository at this point in the history
  59. Fix compilation error in tpu_compilation_metrics

    PiperOrigin-RevId: 556102155
    LionOfJewdah authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    51ee2f2 View commit details
    Browse the repository at this point in the history
  60. Fix incorrect import in lite.py.

    PiperOrigin-RevId: 556105293
    fionalang authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    9b410ba View commit details
    Browse the repository at this point in the history
  61. Add pattern to lower UQ mhlo Constant op in ConvertMHLOQuantToInt

    PiperOrigin-RevId: 556107715
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    8146900 View commit details
    Browse the repository at this point in the history
  62. PR tensorflow#4912: [ROCm] support for GraphAddKernelNode()

    Imported from GitHub PR openxla/xla#4912
    
    patch for this PR openxla/xla#4894 (comment)
    
    @akuegel @ezhulenev Thanks in advance!
    Copybara import of the project:
    
    --
    27fe0e2e859b5188704f21613283e68d594f4d92 by Chao Chen <cchen104@amd.com>:
    
    rocm graph adds GraphAddKernelNode()
    
    Merging this change closes tensorflow#4912
    
    PiperOrigin-RevId: 556107848
    i-chaochen authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    89fac52 View commit details
    Browse the repository at this point in the history
  63. Refactor the MLIR bridge to add a flag which can be used to disable l…

    …egalization passes.
    
    Added combined bridge legalization that is only included in tests for now
    
    PiperOrigin-RevId: 556114029
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    e525c35 View commit details
    Browse the repository at this point in the history
  64. Merge pull request tensorflow#61358 from Intel-tensorflow:zhoulong/ad…

    …d_TempVariable_C_Api
    
    PiperOrigin-RevId: 556115839
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    64e3be3 View commit details
    Browse the repository at this point in the history
  65. [NFC] Move gpurt kernels to tensorflow/core/tfrt/gpu.

    PiperOrigin-RevId: 556120812
    changhuilin authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    11abca2 View commit details
    Browse the repository at this point in the history
  66. Clean up dependencies of tf2xla/kernels targets.

    PiperOrigin-RevId: 556122962
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    7bbf603 View commit details
    Browse the repository at this point in the history
  67. Add return types for Tensor.shape, Tensor.get_shape(), and EagerTenso…

    …r.get_shape().
    
    PiperOrigin-RevId: 556124653
    fionalang authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    e3bc61f View commit details
    Browse the repository at this point in the history
  68. Automated visibility attribute cleanup.

    PiperOrigin-RevId: 556132467
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    641bd45 View commit details
    Browse the repository at this point in the history
  69. Keep OCParams alive throughout the execution

    PiperOrigin-RevId: 556134141
    hhb authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    e48b3d1 View commit details
    Browse the repository at this point in the history
  70. Add further type hints related to ops.Operation.

    PiperOrigin-RevId: 556136108
    tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    0fd118e View commit details
    Browse the repository at this point in the history
  71. [PJRT C API] Use #if !defined(PLATFORM_GOOGLE) as the condition to …

    …run LoadTpuLibraryAndInitializeTpuStructFns to match the definition of LoadTpuLibraryAndInitializeTpuStructFns in tpu_initializer_framework_helper.cc.
    
    PiperOrigin-RevId: 556136719
    Jieying Luo authored and tensorflower-gardener committed Aug 11, 2023
    Configuration menu
    Copy the full SHA
    7288dce View commit details
    Browse the repository at this point in the history

Commits on Aug 12, 2023

  1. Support AsyncCheckpoint for distributed Trackables by implementing `_…

    …copy_trackable_to_cpu()`
    
    PiperOrigin-RevId: 556141352
    tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    b9a4fc1 View commit details
    Browse the repository at this point in the history
  2. Add WeakTensor support for a tf.math.maximum, tf.math.minimum, and tf…

    ….math.equal.
    
    PiperOrigin-RevId: 556144527
    tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    ae7bfb5 View commit details
    Browse the repository at this point in the history
  3. Reverting a change that broke some code

    PiperOrigin-RevId: 556162041
    SeeForTwo authored and tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    b178419 View commit details
    Browse the repository at this point in the history
  4. Merge pull request tensorflow#61460 from philipphack:u_fp8_amax_xla

    PiperOrigin-RevId: 556163674
    tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    e61e7e0 View commit details
    Browse the repository at this point in the history
  5. Add buffer_donor in hlo_parser.

    PiperOrigin-RevId: 556166525
    tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    2b19457 View commit details
    Browse the repository at this point in the history
  6. Add AsyncCheckpoint support for iterators by implementing `_copy_trac…

    …kable_to_cpu()`.
    
    PiperOrigin-RevId: 556167773
    tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    1989dff View commit details
    Browse the repository at this point in the history
  7. [xla:gpu2] Add gpu graphs support to gpu2 runtime

    PiperOrigin-RevId: 556195795
    ezhulenev authored and tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    b64e3d4 View commit details
    Browse the repository at this point in the history
  8. Re-enable TF32 tests.

    The disabled tests were fixed by tensorflow@9e4b9da
    
    Also remove unnecessary TF32 dependencies.
    
    PiperOrigin-RevId: 556200766
    reedwm authored and tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    6f14476 View commit details
    Browse the repository at this point in the history
  9. [xla:gpu2] Add support for mempcy nodes in cuda graphs

    PiperOrigin-RevId: 556205122
    ezhulenev authored and tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    715f80c View commit details
    Browse the repository at this point in the history
  10. Create a new BUILD package in tensorflow/transforms/

    PiperOrigin-RevId: 556207509
    changm authored and tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    4f23d95 View commit details
    Browse the repository at this point in the history
  11. Enable 2gpu TAP tests on for Tensorflow

    This adds 2gpu tests to TAP presubmits for Tensorflow, which is required to properly test changes that can only be tested with multiple physical GPUs, e.g. testing control dependencies of collective ops. This also temporarily disables tests that are currently failing.
    
    PiperOrigin-RevId: 556239827
    jcai19 authored and tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    989bd01 View commit details
    Browse the repository at this point in the history
  12. Update GraphDef version to 1586.

    PiperOrigin-RevId: 556264806
    tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    7fed901 View commit details
    Browse the repository at this point in the history
  13. compat: Update forward compatibility horizon to 2023-08-12

    PiperOrigin-RevId: 556264813
    tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    f267523 View commit details
    Browse the repository at this point in the history
  14. Add a serving device selector interface.

    Note that the interface is experimental and subject to change.
    
    PiperOrigin-RevId: 556273411
    changhuilin authored and tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    2539ebc View commit details
    Browse the repository at this point in the history
  15. #tf-data-service Add runtime compression metric helper for recording …

    …an arbitrary string.
    
    PiperOrigin-RevId: 556380189
    mpcallanan authored and tensorflower-gardener committed Aug 12, 2023
    Configuration menu
    Copy the full SHA
    6903aa4 View commit details
    Browse the repository at this point in the history

Commits on Aug 13, 2023

  1. Internal Code Change

    PiperOrigin-RevId: 556442351
    tensorflower-gardener committed Aug 13, 2023
    Configuration menu
    Copy the full SHA
    120b8ba View commit details
    Browse the repository at this point in the history
  2. Update GraphDef version to 1587.

    PiperOrigin-RevId: 556496434
    tensorflower-gardener committed Aug 13, 2023
    Configuration menu
    Copy the full SHA
    a4a8f8b View commit details
    Browse the repository at this point in the history
  3. compat: Update forward compatibility horizon to 2023-08-13

    PiperOrigin-RevId: 556496454
    tensorflower-gardener committed Aug 13, 2023
    Configuration menu
    Copy the full SHA
    8e8633c View commit details
    Browse the repository at this point in the history
  4. Migrate away from removed typed pointer convenience methods

    These methods were removed by
    llvm/llvm-project@899b840
    
    PiperOrigin-RevId: 556497494
    MaskRay authored and tensorflower-gardener committed Aug 13, 2023
    Configuration menu
    Copy the full SHA
    b8a419b View commit details
    Browse the repository at this point in the history
  5. Fix a few inefficiencies within the op_level_cost_estimator.

    PiperOrigin-RevId: 556565038
    tensorflower-gardener committed Aug 13, 2023
    Configuration menu
    Copy the full SHA
    4c562cc View commit details
    Browse the repository at this point in the history

Commits on Aug 14, 2023

  1. Fix a bug in lowering mhlo.convolution in ConvertMHLOQuantToInt pass

    The current impl use the same tensor shape for lhs, rhs and result, even if they might be different in case of Convolution.
    
    PiperOrigin-RevId: 556625694
    tensorflower-gardener committed Aug 14, 2023
    Configuration menu
    Copy the full SHA
    f205faa View commit details
    Browse the repository at this point in the history
  2. Compose stablehlo.dot_general ops when both operands are quantized …

    …activations.
    
    This pattern corresponds to dot_general with both activations quantized, which can also express an einsum expression: `bij,bjd->bid`.
    
    PiperOrigin-RevId: 556701446
    dansuh17 authored and tensorflower-gardener committed Aug 14, 2023
    Configuration menu
    Copy the full SHA
    d51268e View commit details
    Browse the repository at this point in the history
  3. [Refactoring] Replace QuantizationPrecision to QuantizationComponentSpec

    This CL does not change any behavior but just replaces
    QuantizationPrecision to QuantizationComponentSpec where supported
    custom configurations are limited to default ones.
    
    PiperOrigin-RevId: 556702880
    sngyhan authored and tensorflower-gardener committed Aug 14, 2023
    Configuration menu
    Copy the full SHA
    5d6571a View commit details
    Browse the repository at this point in the history
  4. Simplify the code slightly (NFC)

    We don't allow epilogue fusion for reductions that are not race free. Therefore
    we don't need to search for a non-trivial hero in this case.
    
    PiperOrigin-RevId: 556704137
    akuegel authored and tensorflower-gardener committed Aug 14, 2023
    Configuration menu
    Copy the full SHA
    3390dcd View commit details
    Browse the repository at this point in the history
  5. Update GraphDef version to 1588.

    PiperOrigin-RevId: 556710464
    tensorflower-gardener committed Aug 14, 2023
    Configuration menu
    Copy the full SHA
    a42bf96 View commit details
    Browse the repository at this point in the history
  6. compat: Update forward compatibility horizon to 2023-08-14

    PiperOrigin-RevId: 556710468
    tensorflower-gardener committed Aug 14, 2023
    Configuration menu
    Copy the full SHA
    e1db71d View commit details
    Browse the repository at this point in the history
  7. [NFC] Extract fusion traversal logic into a library.

    This will be reused from hlo_fusion_analysis (RowVectorizationEnabled).
    
    PiperOrigin-RevId: 556734204
    jreiffers authored and tensorflower-gardener committed Aug 14, 2023
    Configuration menu
    Copy the full SHA
    1e2ee27 View commit details
    Browse the repository at this point in the history
  8. Bump PThreadPool version

    PiperOrigin-RevId: 556741307
    ngzhian authored and tensorflower-gardener committed Aug 14, 2023
    Configuration menu
    Copy the full SHA
    9194695 View commit details
    Browse the repository at this point in the history
  9. Configuration menu
    Copy the full SHA
    1f8d138 View commit details
    Browse the repository at this point in the history
  10. Configuration menu
    Copy the full SHA
    14bc7f5 View commit details
    Browse the repository at this point in the history
  11. Configuration menu
    Copy the full SHA
    f7aad94 View commit details
    Browse the repository at this point in the history

Commits on Aug 15, 2023

  1. Configuration menu
    Copy the full SHA
    d5cc417 View commit details
    Browse the repository at this point in the history
  2. Configuration menu
    Copy the full SHA
    e4436c7 View commit details
    Browse the repository at this point in the history

Commits on Aug 17, 2023

  1. Configuration menu
    Copy the full SHA
    972bf34 View commit details
    Browse the repository at this point in the history

Commits on Aug 18, 2023

  1. Configuration menu
    Copy the full SHA
    297fc7f View commit details
    Browse the repository at this point in the history

Commits on Aug 22, 2023

  1. turn of conv_ops_test

    weihanmines committed Aug 22, 2023
    Configuration menu
    Copy the full SHA
    9dbc1cc View commit details
    Browse the repository at this point in the history