Conversation
ankus-qti
requested review from
qti-ashwshan,
qti-chuteng,
qti-jkilpatrick,
qti-kromero,
qti-yuduo,
tirupath-qti and
yath1
as code owners
August 19, 2026 09:44
ankus-qti
force-pushed
the
dev/ankus/QKV_fusion_window_partition_fusion
branch
from
August 19, 2026 12:04
66ffc3a to
4d7446e
Compare
Adrressing claude review comments. Signed-off-by: ankus <ankus@qti.qualcomm.com>
ankus-qti
force-pushed
the
dev/ankus/QKV_fusion_window_partition_fusion
branch
from
August 21, 2026 03:32
4d7446e to
3e633ec
Compare
ankus-qti
requested review from
minfhong-qti,
qti-shubham and
yuhuchua-qti
as code owners
September 9, 2026 05:51
This branch has not been deployed
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Description
Rewrite high-rank tensor patterns in Swin-Transformer-style attention into rank-≤4 QNN ops on the HTP backend.
Motivation and Context
Rank-5 tensors are supported on HTP but perform poorly; these fusions eliminate them from the hot path and achieve ~2x perf gain.
Before: 1005 ms
After: 502 ms