llama.cpp
hexagon: eliminate scalar VTCM loads via HVX splat helpers
#22993

Merged

hexagon: eliminate scalar VTCM loads via HVX splat helpers #22993

max-krasnyansky merged 5 commits into ggml-org:master from qualcomm:tr/hvx-splat-vtcm

hexagon: add hvx_vec_repl helpers and use those for splat-from-vtcm u…

219d281b

hmx-mm: optimize per-group scale handling

24485261

hmx-fa: optimize slope load from vtcm

ce799b9e

hmx-fa: use aligned access where possible in hmx-utils

e6171dd4

hexagon: add hvx_vec_repl_2x_f16 helper and consolidate repl helpers

72ebe045

trivikram-reddy1 requested a review 27 days ago

github-actions added script

github-actions added ggml

github-actions added Hexagon

lhez approved these changes on 2026-05-12

max-krasnyansky approved these changes on 2026-05-13

max-krasnyansky merged 856c3ada into master 27 days ago

trivikram-reddy1 deleted the tr/hvx-splat-vtcm branch 26 days ago

Reviewers

max-krasnyansky

lhez

Assignees

No one assigned

Labels

script ggml Hexagon

Milestone

No milestone