Repository navigation
[pull] master from tensorflow:master - #8838
Merged
Merged
Conversation
PiperOrigin-RevId: 983936237
…edIndices
When scattering with an empty index vector (`index_vector_length == 0`) and
sorted/distributed indices:
1. Constructing `thread_id_to_update_id_map` by projecting `indices_map`
resulted in an empty `IndexingMap` (`IsKnownEmpty()`) because `indices_map`
has a range variable of size `index_vector_length == 0`, causing the loop
over updates to have 0 iterations.
2. `EmitInequalityCheck` on empty offset vectors always evaluates to `false`,
leaving `offsets_changed == false` on the first iteration
(`iter_slice_id == 0`). Because `is_inbounds_init` starts as `false`,
`UpdateIsInbounds` never evaluated the bounds check and dropped all updates,
and the accumulator would combine with 0 instead of being overwritten on the
first slice.
Fix this by:
- Extracting `ComputeThreadIdToUpdateIdMap` and reusing it to construct
`indices_map` in `ComputeIndexing`.
- Forcing `offsets_changed` to `true` on the first loop iteration
(`iter_slice_id == 0`).
- Simplifying loop initialization and update logic:
- Initialize `indices_init` with `0` instead of `-1` since the first
iteration check is now explicit.
- Remove the redundant `arith::SelectOp` loop updating `new_trimmed_offsets`.
- Remove `is_not_first_iteration` from `write_to_output_required` since
`iter_is_inbounds` is guaranteed `false` on the first iteration.
PiperOrigin-RevId: 983969974
…ottlenecks in TensorFlow CUDA CI. PiperOrigin-RevId: 983993726
PiperOrigin-RevId: 984030093
…eouts: - Increase msan_build memory to 32g for lazy_sstable_reader_test. - Add msan_build memory (20g) to exec_properties for saved_model_test. - Add msan_build memory (30g) to exec_properties for remote_test. - Increase shard_count to 16 for linear_operator_block_diag_test to prevent MSan timeout. PiperOrigin-RevId: 984065765
PiperOrigin-RevId: 984070526
This matches the behavior of some related tooling with which this needs to be compatible. PiperOrigin-RevId: 984075260
The FC -> BMM was looking at T+FC to change that to BMM hoping that if a const + T is already folded by the time it sees it. However, resource + T is not immediately folded by design. So, it was leading to FC -> BMM -> FC -> BMM -> ... This explicitly ensures a FC -> BMM doesn't kick in if rhs is from an eventually foldable chain. PiperOrigin-RevId: 984076078
PiperOrigin-RevId: 984092201
…ogy fingerprint PiperOrigin-RevId: 984102296
PiperOrigin-RevId: 984105430
…computation layout. PiperOrigin-RevId: 984141229
CopyInsertion takes an optional view color. A view colored value is an address stand in that owns no storage and aliases its operand 0's buffer. Copy removal then counts the transitive readers of a value's view colored users as uses of that value, so no producer is merged into the viewed buffer between a view and its readers. Backends without views are unchanged. PiperOrigin-RevId: 984143462
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
10 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
See Commits and Changes for more details.
Created by
pull[bot] (v2.0.0-alpha.4)
Can you help keep this open source service alive? 💖 Please sponsor : )