Replaces the second ParallelFor barrier/reduction loop in LocalAttentionForAllHeadsTokensAndBatch (tiled_attention.cc) with a lock-free 64-bit atomic compare-and-exchange state machine (AtomicGroupMergeSlot) partitioned into groups of up to 32 queries (kMergeGroupSize = 32). Subtasks hand off completed 32-query group pointers immediately via CAS or claim and merge waiting groups outside any lock. Uses Highway SIMD vectorization (MergeOnlineSoftmaxGroup32, hn::FastExpMinusOrZero) to compute online softmax exponents and scaling factors across all 32 queries simultaneously and combine attention outputs in a single FMA pass. - #1028
Merged
Merged
Conversation
copybara-service
Bot
force-pushed
the
test_979178229
branch
from
September 14, 2026 14:39
1956586 to
a49ee39
Compare
…entionForAllHeadsTokensAndBatch` (`tiled_attention.cc`) with a lock-free 64-bit atomic compare-and-exchange state machine (`AtomicGroupMergeSlot`) partitioned into groups of up to 32 queries (`kMergeGroupSize = 32`). Subtasks hand off completed 32-query group pointers immediately via CAS or claim and merge waiting groups outside any lock. Uses Highway SIMD vectorization (`MergeOnlineSoftmaxGroup32`, `hn::FastExpMinusOrZero`) to compute online softmax exponents and scaling factors across all 32 queries simultaneously and combine attention outputs in a single FMA pass. PiperOrigin-RevId: 981156349
copybara-service
Bot
force-pushed
the
test_979178229
branch
from
September 14, 2026 15:08
a49ee39 to
10ebbdd
Compare
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Replaces the second
ParallelForbarrier/reduction loop inLocalAttentionForAllHeadsTokensAndBatch(tiled_attention.cc) with a lock-free 64-bit atomic compare-and-exchange state machine (AtomicGroupMergeSlot) partitioned into groups of up to 32 queries (kMergeGroupSize = 32). Subtasks hand off completed 32-query group pointers immediately via CAS or claim and merge waiting groups outside any lock. Uses Highway SIMD vectorization (MergeOnlineSoftmaxGroup32,hn::FastExpMinusOrZero) to compute online softmax exponents and scaling factors across all 32 queries simultaneously and combine attention outputs in a single FMA pass.