This project is mirrored from Pull mirroring updated .
  1. 31 May, 2020 2 commits
  2. 30 May, 2020 19 commits
  3. 29 May, 2020 19 commits
    • Yuefeng Zhou's avatar
      fix some linter errors for slurm_cluster_resolver. · 3fbd5ac4
      Yuefeng Zhou authored
      PiperOrigin-RevId: 313873815
      Change-Id: I15ae65bb27af2ee9d60b3629c91c0234fbc8943f
    • Jared Duke's avatar
      Restore `with_select_tf_ops` build flag. · bacd1884
      Jared Duke authored
      This remains useful for testing and development. Restore the ability
      to inject support for TF ops in TFLite using `--define=with_select_tf_ops=true`.
      See also issue #34277.
      PiperOrigin-RevId: 313873470
      Change-Id: I6b68cd863efc17f5ae0667c0d2c9d68958d6e4ad
    • A. Unique TensorFlower's avatar
      Go: Update generated wrapper functions for TensorFlow ops. · d99affc7
      A. Unique TensorFlower authored
      PiperOrigin-RevId: 313873341
      Change-Id: I4ac3ffcf5fc5ed5b1444fc92b2d87988724c310e
    • Jiri Simsa's avatar
      [] Explicitly colocate prefetch dataset op with its input as this... · 8be4d615
      Jiri Simsa authored
      [] Explicitly colocate prefetch dataset op with its input as this collocation only happens automatically in graph mode.
      PiperOrigin-RevId: 313867950
      Change-Id: I88962b96f208b6d9019e0a117715f74efc8fdc67
    • Mehdi Amini's avatar
      Bump the ruy repository reference. · 02dc6f8d
      Mehdi Amini authored
      PiperOrigin-RevId: 313866050
      Change-Id: I6a3c97d6f4e74c6078eb3bcc1607e51fc1f4d784
    • Jiri Simsa's avatar
      [] Remove misleading documentation. · 80671523
      Jiri Simsa authored
      PiperOrigin-RevId: 313862461
      Change-Id: I19720b5a90c251f45ab5bc4d90028481b8964f20
    • A. Unique TensorFlower's avatar
      Add timeout to collective ops to detect deadlocks. · 66529c35
      A. Unique TensorFlower authored
      The timeout is set as an argument to a collective op. When non zero value, a completion timeout is set to detect staleness. If a timeout goes off, the execution is aborted through a DEADLINE_EXCEEDED error.
      PiperOrigin-RevId: 313861868
      Change-Id: I7fee45736608ad7fbcc9dd980db2fd302c9cb4df
    • Guangda Lai's avatar
      Make tf.If work with ConcreteFunction. · 85396efc
      Guangda Lai authored
      PiperOrigin-RevId: 313860966
      Change-Id: I1fccdaf06802511a7020a4045751cdd6b6821687
    • Andy Ly's avatar
      Uniformly import and export _Arg node/FunctionDef arg attributes. · f90c649e
      Andy Ly authored
      In a Function Graph (Graph generated from a Function/FunctionDef), it is possible to have other attributes on the generated _Arg nodes. These attributes are either modeled as fields in FunctionDef ('_resource_arg_unique_id' attributes are stored as FunctionDef::map<uint32, uint32> resource_arg_unique_id) or explicitly in FunctionDef::map<uint32, ArgAttrs> arg_attr. When converting a FunctionDef to a Graph (in import), these attributes are added to generated _Arg node attributes. Some of these attributes should be preserved for downstream users. Currently only '_resource_arg_unique_id' is being imported with special handling.
      This change unifies and imports any _Arg attribute that is not a shape inference based attribute or _Arg op def attribute. On export, attributes of the 'tf' dialect ('tf.' prefix) are added back. For the main function Graph, the attributes are simply added back to the generated _Arg node. For other functions, as a FunctionDef is created instead, '_resource_arg_unique_id' is handled differently, specifically adding it's content to FunctionDef::map<uint32, uint32> resource_arg_unique_id while all other attribute are added to FunctionDef::map<uint32, ArgAttrs> arg_attr.
      PiperOrigin-RevId: 313859301
      Change-Id: I3bb37bb63cc4d401d628c08989900524d0db0572
    • Andy Ly's avatar
      Update TPUExtractHeadTailOutsideCompilation pass in preparation for tail extraction (NFC). · 5a62ab62
      Andy Ly authored
      This simplifies and updates some test cases, and extract some reused logic used by tail extraction.
      PiperOrigin-RevId: 313859255
      Change-Id: I35bb385c0a76aae54cc7836db8a8f549cd9b86ff
    • Eugene Zhulenev's avatar
      Fix inlined function logging · 677c1960
      Eugene Zhulenev authored
      PiperOrigin-RevId: 313858692
      Change-Id: I8823363003eef3a9bf0f7f66322537f2dc3fc8de
    • Raman Sarokin's avatar
      Added missing resource types to arguments. · a6a3a486
      Raman Sarokin authored
      PiperOrigin-RevId: 313858546
      Change-Id: I5a83491728c7f6709994464186725649ad81e3c7
    • Robert Suderman's avatar
      Add xla_hlo.dynamic_iota for non-static cases of iota · a5bd187c
      Robert Suderman authored
      Existing xla_hlo.iota does not cover all use cases. Added an xla_hlo.iota
      operation that supports a dynamially shaped output. This matches the
      behavior for dynamic_broadcast_in_dim.
      PiperOrigin-RevId: 313854741
      Change-Id: Idf8361984d48e30eac9fb22ef3b54b178d925f0d
    • Jiho Choi's avatar
      Register the semantic stats as internal. · 8b3f0347
      Jiho Choi authored
      PiperOrigin-RevId: 313853594
      Change-Id: I4e5ced627e8705cae77230671f70065e3ed25191
    • A. Unique TensorFlower's avatar
      Internal change · 8c63be39
      A. Unique TensorFlower authored
      PiperOrigin-RevId: 313850352
      Change-Id: I89584b0bcb4409eb74d21e31fb0eb68844186707
    • Akshay Naresh Modi's avatar
      Move linspace tests to their own file. · efa88013
      Akshay Naresh Modi authored
      PiperOrigin-RevId: 313843608
      Change-Id: Ifdae2ac60a721795151124a6a2ee643cb0e527ec
    • Nick Kreeger's avatar
      Add special "recording" SimpleMemoryAllocator class to help with logging tail allocations. · 7d0ab617
      Nick Kreeger authored
      This new helper class will enable TFLM to log and record where the allocations in the shared arena are going. A future change will use this new class in a special "recording" MicroAllocator subclass. All these logging mechanisms will be opt-in by code.
      PiperOrigin-RevId: 313843072
      Change-Id: I3fc9205e475e89b4a3795c3cc79c31d2166da2c8
    • A. Unique TensorFlower's avatar
      Fix typo in tf.where documentation. · 3b2674dc
      A. Unique TensorFlower authored
      PiperOrigin-RevId: 313842387
      Change-Id: I255dfad74a2ddc80373504569b07c39636d90cf1
    • Ken Franko's avatar
      Handle output from OutsideCompiled parallel_execute regions. · 26192135
      Ken Franko authored
      Adds ops to send/recv data from host -> device when there are outputs from the OutsideCompiled cluster.  _TPUCompileMlir placeholder ops are also added to be replaced later because host side comm ops require the program_key as input.
      This handles when a the result from OutsideCompiled cluster was originally returned from the TPU cluster.
      PiperOrigin-RevId: 313840240
      Change-Id: I2af37282309dd0998f0c15c0954a855b7bc0ac63