Commits

Branch selector

 master

User selector

andyly

Datepicker

All time

Commit History

Commits on Dec 23, 2020

• Migrate TPU resource read for write pass to use declarative pass registration instead of manually defined pass registration (NFC).

  Show description for df37c3e
  

  andyly

  authored and

  tensorflower-gardener

  committed
  df37c3e

  Copy full SHA for df37c3e

• Migrate cluster outlining pass to use declarative pass registration instead of manually defined pass registration (NFC).

  Show description for 2d52be6
  

  andyly

  authored and

  tensorflower-gardener

  committed
  2d52be6

  Copy full SHA for 2d52be6

Commits on Dec 22, 2020

• Update TPU sharding identification pass to support checking for sharding attributes from tf.TPUPartitionedInput/tf.TPUPartitionedOutput ops.

  Show description for 636ca6b
  

  andyly

  authored and

  tensorflower-gardener

  committed
  636ca6b

  Copy full SHA for 636ca6b

Commits on Dec 21, 2020

• Register TPUPartitionedOutput in TF MLIR ODS.

  Show description for 74cda1c
  

  andyly

  authored and

  tensorflower-gardener

  committed
  74cda1c

  Copy full SHA for 74cda1c

• Register TPUPartitionedInput in TF MLIR ODS.

  Show description for 3fb987e
  

  andyly

  authored and

  tensorflower-gardener

  committed
  3fb987e

  Copy full SHA for 3fb987e

• Add shape inference function for tf.XlaSetDynamicDimensionSize and set to not allow constant folding.

  Show description for 0f99ddc
  

  andyly

  authored and

  tensorflower-gardener

  committed
  0f99ddc

  Copy full SHA for 0f99ddc

Commits on Dec 18, 2020

• Migrate constant sinking pass to use declarative pass registration instead of manually defined pass registration (NFC).

  Show description for a0d3f71
  

  andyly

  authored and

  tensorflower-gardener

  committed
  a0d3f71

  Copy full SHA for a0d3f71

Commits on Dec 16, 2020

• Migrate graph pruning pass to use declarative pass registration instead of manually defined pass registration (NFC).

  Show description for 01c2faf
  

  andyly

  authored and

  tensorflower-gardener

  committed
  01c2faf

  Copy full SHA for 01c2faf

Commits on Dec 15, 2020

• Migrate TPU cluster formation pass to use declarative pass registration instead of manually defined pass registration (NFC).

  Show description for a817961
  

  andyly

  authored and

  tensorflower-gardener

  committed
  a817961

  Copy full SHA for a817961

• Migrate TF executor dialect to functional conversion pass to use declarative pass registration instead of manually defined pass registration (NFC).

  Show description for 5e943d0
  

  andyly

  authored and

  tensorflower-gardener

  committed
  5e943d0

  Copy full SHA for 5e943d0

Commits on Dec 14, 2020

• Migrate TF MLIR shape inference pass to use declarative pass registration instead of manually defined pass registration (NFC).

  Show description for e290ea6
  

  andyly

  authored and

  tensorflower-gardener

  committed
  e290ea6

  Copy full SHA for e290ea6

Commits on Dec 10, 2020

• Register tf.EnqueueTPUEmbeddingIntegerBatch, tf.EnqueueTPUEmbeddingSparseBatch and tf.EnqueueTPUEmbeddingBatch and add TPUEmbeddingSideEffect trait to associated ops.

  Show description for e367444
  

  andyly

  authored and

  tensorflower-gardener

  committed
  e367444

  Copy full SHA for e367444

• Fix //tensorflow/compiler/mlir/tensorflow/tests/tf_saved_model:debug_info.py.test for macOS builds.

  Show description for 24e55fb
  

  andyly

  authored and

  tensorflower-gardener

  committed
  24e55fb

  Copy full SHA for 24e55fb

• Disable shape inference on import in ConvertFunctionToMlir.

  Show description for 5d7f57b
  

  andyly

  authored and

  tensorflower-gardener

  committed
  5d7f57b

  Copy full SHA for 5d7f57b

• Switch from explicit argument for inlining functions post TF -> HLO legalization to checking device type (NFC).

  Show description for b43da1e
  

  andyly

  authored and

  tensorflower-gardener

  committed
  b43da1e

  Copy full SHA for b43da1e

Commits on Dec 9, 2020

• Add inliner pass to `CompileSerializedMlirToXlaHlo`.

  Show description for 5eede88
  

  andyly

  authored and

  tensorflower-gardener

  committed
  5eede88

  Copy full SHA for 5eede88

• Add custom builder for tf_device.replicate that accepts a DictionaryAttr instead of a llvm::SmallDenseMap<StringRef, llvm::SmallVector<StringRef, 4>> for devices, and expose WrapsSingleOp, similar …

  Show description for 7d6a9af
  

  andyly

  authored and

  tensorflower-gardener

  committed
  7d6a9af

  Copy full SHA for 7d6a9af

Commits on Dec 7, 2020

• Move tests for arg and retval devices from arg-retval-attrs.pbtxt to existing device attribute test (NFC).

  Show description for 66a08a0
  

  andyly

  authored and

  tensorflower-gardener

  committed
  66a08a0

  Copy full SHA for 66a08a0

Commits on Dec 4, 2020

• Import shapes from attributes for tf.IteratorGetNext, tf.IteratorGetNextSync, tf.MultiDeviceIteratorGetNextFromShard, tf.InfeedDequeueTuple, and tf.InfeedDequeue when shape inference on import is n…

  Show description for 2022bf3
  

  andyly

  authored and

  tensorflower-gardener

  committed
  2022bf3

  Copy full SHA for 2022bf3

• Define side effects of tf.TPUExecute and tf.TPUExecuteAndUpdateVariables.

  Show description for cd12581
  

  andyly

  authored and

  tensorflower-gardener

  committed
  cd12581

  Copy full SHA for cd12581

Commits on Dec 2, 2020

• Annotate tf_device.return and tf.Yield with NoSideEffect and ReturnLike traits.

  Show description for 8b0eee3
  

  andyly

  authored and

  tensorflower-gardener

  committed
  8b0eee3

  Copy full SHA for 8b0eee3

• Add some verification for tf.TPUExecuteAndUpdateVariables arguments and attributes.

  Show description for 65bff9f
  

  andyly

  authored and

  tensorflower-gardener

  committed
  65bff9f

  Copy full SHA for 65bff9f

• Update tf.While/tf.WhileRegion shape inference to support different operand and result shapes.

  Show description for ad03e24
  

  andyly

  authored and

  tensorflower-gardener

  committed
  ad03e24

  Copy full SHA for ad03e24

• Update side effect analysis to handle TPU Embedding op effects and replace ParallelizeEmbeddingParamsOpsPass with breakup islands pass.

  Show description for 399d51c
  

  andyly

  authored and

  tensorflower-gardener

  committed
  399d51c

  Copy full SHA for 399d51c

• Register TPU retrieve embeddings ops in TF MLIR ODS.

  Show description for f6442b6
  

  andyly

  authored and

  tensorflower-gardener

  committed
  f6442b6

  Copy full SHA for f6442b6

• Update WhileOutlinePass to handle while ops with different operand and result shapes.

  Show description for ce006b9
  

  andyly

  authored and

  tensorflower-gardener

  committed
  ce006b9

  Copy full SHA for ce006b9

Commits on Dec 1, 2020

• Introduce new resource type to model TPU program ops.

  Show description for 98a5efc
  

  andyly

  authored and

  tensorflower-gardener

  committed
  98a5efc

  Copy full SHA for 98a5efc

• Register TPU load embeddings ops in TF MLIR ODS.

  Show description for adc2537
  

  andyly

  authored and

  tensorflower-gardener

  committed
  adc2537

  Copy full SHA for adc2537

• Switch ResourceHandle to use StringRef when storing container, name, and device strings (NFC).

  Show description for d10afb1
  

  andyly

  authored and

  tensorflower-gardener

  committed
  d10afb1

  Copy full SHA for d10afb1

• Update verifier for tf.While/tf.WhileRegion to not check for operand result type compatibility if op is shape invariant.

  Show description for bef3dd3
  

  andyly

  authored and

  tensorflower-gardener

  committed
  bef3dd3

  Copy full SHA for bef3dd3

Commits on Nov 25, 2020

• Update tf.While/tf.WhileRegion shape inference to support different operand and result shapes.

  Show description for 3c572a3
  

  andyly

  authored and

  tensorflower-gardener

  committed
  3c572a3

  Copy full SHA for 3c572a3

Commits on Nov 24, 2020

• Loosen check for removing Cast(Resource) in resource op lifting.

  Show description for 15440a2
  

  andyly

  authored and

  tensorflower-gardener

  committed
  15440a2

  Copy full SHA for 15440a2

• Directly create scalar constant without conversion from TensorFlow tensor to MLIR TF Const op (NFC).

  Show description for 707eb38
  

  andyly

  authored and

  tensorflower-gardener

  committed
  707eb38

  Copy full SHA for 707eb38

• Export result attributes from MLIR function to exported Graph via _Retval node attributes.

  Show description for 5a086ba
  

  andyly

  authored and

  tensorflower-gardener

  committed
  5a086ba

  Copy full SHA for 5a086ba

Commits on Nov 20, 2020

• Update Graph -> TF MLIR importer to always import tf.While output shapes.

  Show description for 7fa0d80
  

  andyly

  authored and

  tensorflower-gardener

  committed
  7fa0d80

  Copy full SHA for 7fa0d80

Pagination

Previous

Next