DeepSpeed
fix(transformer): use correct stride in Transpose_Kernel shared memory indexing to eliminate bank conflicts
#8055

Merged

fix(transformer): use correct stride in Transpose_Kernel shared memory indexing to eliminate bank conflicts #8055

tohtana merged 2 commits into deepspeedai:master from flutist:fix_shared_memory

flutist requested a review from

tjruwase 40 days ago

Fix shared memory

91c41f5a

flutist force pushed from 85016684 to 91c41f5a 40 days ago

tohtana approved these changes on 2026-06-10

Merge branch 'master' into fix_shared_memory

59eef67e

tohtana enabled auto-merge (squash) 40 days ago

tohtana merged 65e5f6fc into master 39 days ago

Reviewers

tohtana

tjruwase

Assignees

No one assigned

Labels

None yet

Milestone

No milestone