hexagon: eliminate scalar VTCM loads via HVX splat helpers#22993
Merged
Conversation
github-actions
Bot
added
script
Contributor
Author
|
@max-krasnyansky @lhez. could you please review this PR. |
lhez
approved these changes
May 12, 2026
max-krasnyansky
approved these changes
May 13, 2026
xxmustafacooTR
pushed a commit
to xxPlayground/llama-cpp-turboquant
that referenced
this pull request
May 13, 2026
…22993) * hexagon: add hvx_vec_repl helpers and use those for splat-from-vtcm usecase * hmx-mm: optimize per-group scale handling * hmx-fa: optimize slope load from vtcm * hmx-fa: use aligned access where possible in hmx-utils * hexagon: add hvx_vec_repl_2x_f16 helper and consolidate repl helpers --------- Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
rsenthilkumar6
pushed a commit
to rsenthilkumar6/llama.cpp
that referenced
this pull request
May 19, 2026
…22993) * hexagon: add hvx_vec_repl helpers and use those for splat-from-vtcm usecase * hmx-mm: optimize per-group scale handling * hmx-fa: optimize slope load from vtcm * hmx-fa: use aligned access where possible in hmx-utils * hexagon: add hvx_vec_repl_2x_f16 helper and consolidate repl helpers --------- Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
ArberSephirotheca
pushed a commit
to ArberSephirotheca/llama.cpp
that referenced
this pull request
May 19, 2026
…22993) * hexagon: add hvx_vec_repl helpers and use those for splat-from-vtcm usecase * hmx-mm: optimize per-group scale handling * hmx-fa: optimize slope load from vtcm * hmx-fa: use aligned access where possible in hmx-utils * hexagon: add hvx_vec_repl_2x_f16 helper and consolidate repl helpers --------- Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
njsyw1997
added a commit
to aizip/llama.cpp
that referenced
this pull request
May 20, 2026
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
3 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.
Overview
Scalar loads from VTCM are expensive on Hexagon. This PR removes scalar VTCM loads in matmul and flash attention, replacing them with HVX vector loads + splat (
vdelta) operations.Additional information
Results from Snapdragon 8elite Gen 5
unsloth/Qwen3-4B-GGUF/Qwen3-4B-Q4_0.gguf (44% improvement in prefill TPS or TTFT)
Before:
prompt eval time = 2909.89 ms / 946 tokens ( 3.08 ms per token, 325.10 tokens per second)eval time = 3179.95 ms / 63 runs ( 50.48 ms per token, 19.81 tokens per second)After:
prompt eval time = 2020.47 ms / 946 tokens ( 2.14 ms per token, 468.21 tokens per second)eval time = 3213.10 ms / 63 runs ( 51.00 ms per token, 19.61 tokens per second)Requirements