Skip to content

Conversation

@dan-zheng
Copy link
Contributor

  • Create SILSerializer::DifferentiabilityWitnessesToEmit to track
    differentiability witnesses referenced by
    differentiability_witness_function instructions.
    These witnesses need to be serialized.
  • Move differentiability witness serialization before SIL function serialization
    but after visiting SIL functions (differentiability_witness_function
    instructions).
  • Use -emit-sorted-sil in test/AutoDiff/sil_differentiability_witness.sil
    for deterministic ordering for printing and deserialization.

Mirror of #28461 applied to tensorflow branch to keep orthogonal PRs small.

test/AutoDiff/sil_differentiability_witness_reference_serialization.sil was created by @marcrasi.

- Create `SILSerializer::DifferentiabilityWitnessesToEmit` to track
  differentiability witnesses referenced by
  `differentiability_witness_function` instructions.
  These witnesses need to be serialized.
- Move differentiability witness serialization before SIL function serialization
  but after visiting SIL functions (`differentiability_witness_function`
  instructions).
- Use `-emit-sorted-sil` in test/AutoDiff/sil_differentiability_witness.sil
  for deterministic ordering for printing and deserialization.
@dan-zheng dan-zheng added the tensorflow This is for "tensorflow" branch PRs. label Nov 24, 2019
@dan-zheng
Copy link
Contributor Author

@swift-ci Please clean test tensorflow

@dan-zheng dan-zheng requested review from marcrasi and rxwei and removed request for rxwei November 24, 2019 18:57
@dan-zheng
Copy link
Contributor Author

Merging since mirror patch #28461 was approved.

@dan-zheng dan-zheng merged commit 21a4bc5 into swiftlang:tensorflow Nov 24, 2019
@dan-zheng dan-zheng deleted the captain-crunch branch November 24, 2019 19:23
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

tensorflow This is for "tensorflow" branch PRs.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant