Skip to content

hexagon: Flash Attention optimizations (dma, mpyacc, multi-row) and MatMul updates #20118

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
max-krasnyansky merged 19 commits into from
Mar 5, 2026
Merged

hexagon: Flash Attention optimizations (dma, mpyacc, multi-row) and MatMul updates #20118

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
c367b05
ggml-hexagon: enhance hvx_dot_f16_f16_aa_rx4 for improved performance…
chraac Feb 3, 2026
f445ffe
ggml-hexagon: optimize hvx_dot_f16_f16_aa_rx4 and enhance hvx_vec_red…
chraac Feb 3, 2026
6e0bacc
ggml-hexagon: add hvx_dot_f16_f16_aa_rx32 for enhanced vector process…
chraac Feb 8, 2026
0c0bcf1
optimize hvx_dot_f16_f16_aa_rx4 and hvx_dot_f16_f16_aa_rx32 by removi…
chraac Feb 8, 2026
544dbae
ggml-hexagon: refactor hvx_dot_f16_f16_aa_rx4 for improved readabilit…
chraac Feb 8, 2026
787e393
ggml-hexagon: initialize sums variable in hvx_dot_f16_f16_aa_rx32 for…
chraac Feb 9, 2026
67fa580
ggml-hexagon: fix compiling error
chraac Feb 19, 2026
69e65d1
fix hvx_dot_f16_f16_aa_rx4 to handle leftover elements correctly usin…
chraac Feb 20, 2026
6068664
refactor hvx_dot_f16_f16_aa_rx4 to accept vector and leftover element…
chraac Feb 21, 2026
be35a18
wip
chraac Feb 21, 2026
c4165f0
fa: instrumentation and dma reordering
max-krasnyansky Feb 25, 2026
0978aeb
hex-fa: use block-size 64 to improve DMA pipelining
max-krasnyansky Feb 25, 2026
d0f8d28
hex-fa: optimize vec-dot for v79 and above
max-krasnyansky Feb 25, 2026
0396473
hex-fa: use block size 64
max-krasnyansky Feb 25, 2026
bbfa942
hex-fa: avoid scalar fp32->fp16 conversions
max-krasnyansky Feb 27, 2026
ef65e62
hex-fa: simplify dot_f16 functions using optimized vec_mpyacc
max-krasnyansky Feb 28, 2026
3f60932
hex-fa: rewrite mad_f32_f16 using hvx_vec_mpyacc
max-krasnyansky Mar 4, 2026
7c270ea
hex-mm: use mpyacc in matmul dot functions
max-krasnyansky Mar 4, 2026
56f10cb
Merge branch 'ggml-org:master' into hexagon-fa-updates-dma-mpyacc
max-krasnyansky Mar 4, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Loading
Loading