Repository navigation
hexagon: new HMX-optimized GATED_DELTA_NET - #29199
Merged
Merged
Conversation
github-actions
Bot
added
documentation
max-krasnyansky
force-pushed
the
hexagon-gdn-hmx
branch
from
September 21, 2026 05:04
d40b3fb to
dc18a87
Compare
This was referenced Sep 21, 2026
max-krasnyansky
force-pushed
the
hexagon-gdn-hmx
branch
from
September 21, 2026 19:03
dc18a87 to
68160e5
Compare
CISC
approved these changes
Sep 21, 2026
lhez
approved these changes
Sep 21, 2026
1 task done
LadislavSopko
pushed a commit
to 0ics-srls/llama.cpp
that referenced
this pull request
Oct 5, 2026
* hex-gdn: start putting together HMX support for GDN * hex-gdn: working hmx but not-pipelined and slow for now * hex-gdn: re-write vtcm layout handling and prep for pipelining * hex-gdn: starting to pipeline hmx and dmas * hex-gdn: add hvx threading for most pipeline stages * hex-gdb: add detailed trace events * hex-gdn: vectorize expfs and use aligned hvx reads/writes * hex-gnd: vectorize the rest of expf * hex-gdn: optimize tail processing (pad partial chunks) * hex-gdb: avoid float up/down casts in hot loops * hex-fa: remove float up/down casts from inner loops * hex-gdn: do exp() in f16 to improve HVX utilization * hex-gdn: optimize tiler * hex-hmx: bump hmx-queue to 128 and dispatch all GDN gemms at once * hex-gdn: further pipeline improvements * hex-gdn: optimize gdn prep stage * hex-gdn: yet more tweaks to optimize GND_SOLVE task and pipeline * hex-gdn: improve accuracy and optmize gdn-prep further * hex-gdn: fix rebase conflict * hex-bufs: revert max_bufsize enforcement, it is enough to just enforce max_vmem * hex-scripts: improved inspect script to avoid false alarms in reg spill detector * hex-fa: improve inline softmax with in-reg VKQ32 accum * hex-fa: minor improvement for dma pipeline in hvx kernel * hex-fa: reduce ddr reads by 20-30% during token gen * hex-gdn: proper alignment for hvx vtcm spads
edwardyoon
pushed a commit
to edwardyoon/focus-llama
that referenced
this pull request
Oct 7, 2026
* hex-gdn: start putting together HMX support for GDN * hex-gdn: working hmx but not-pipelined and slow for now * hex-gdn: re-write vtcm layout handling and prep for pipelining * hex-gdn: starting to pipeline hmx and dmas * hex-gdn: add hvx threading for most pipeline stages * hex-gdb: add detailed trace events * hex-gdn: vectorize expfs and use aligned hvx reads/writes * hex-gnd: vectorize the rest of expf * hex-gdn: optimize tail processing (pad partial chunks) * hex-gdb: avoid float up/down casts in hot loops * hex-fa: remove float up/down casts from inner loops * hex-gdn: do exp() in f16 to improve HVX utilization * hex-gdn: optimize tiler * hex-hmx: bump hmx-queue to 128 and dispatch all GDN gemms at once * hex-gdn: further pipeline improvements * hex-gdn: optimize gdn prep stage * hex-gdn: yet more tweaks to optimize GND_SOLVE task and pipeline * hex-gdn: improve accuracy and optmize gdn-prep further * hex-gdn: fix rebase conflict * hex-bufs: revert max_bufsize enforcement, it is enough to just enforce max_vmem * hex-scripts: improved inspect script to avoid false alarms in reg spill detector * hex-fa: improve inline softmax with in-reg VKQ32 accum * hex-fa: minor improvement for dma pipeline in hvx kernel * hex-fa: reduce ddr reads by 20-30% during token gen * hex-gdn: proper alignment for hvx vtcm spads (cherry picked from commit 5836771)
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
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
This PR introduces brand new HMX implementation for
GATED_DELTA_NET.Prompt processing performance improvements range from 1.5-3X depending on the model and device.
Tested with Qwen3.5 and Qwen3.8 (mostly smaller distilled modes) on Galaxy S26/25/24, VentunoQ, IQ9, X2 Elite.
This PR is layered on top of #29197
Once that PR is merged I'll rebase and update this one.
Additional information
Here are some numbers on Galaxy S26
Requirements