Repository navigation
hexagon: overhaul of buffer and DMA handling to support 64bit mappings and general improvements - #29197
Merged
Conversation
hex-dma64: expand binary ops to support more DMA scenarios hex-dma64: add binary-ops.h hex-dma64: add --hex-dma64 to run.py and fix minor issues hex-dma64: update SSM_CONV to use dma with proper support for 64bit hex-ops: remove obsolete gate for % 128 in binary ops hex-l2: dont check weight tensors against dirty ranges hex-dma64: most binary ops now support dma hex-dma: use dma_addr_t instead of plain uint64_t to avoid overhead on older targets hex-dma: update all dma users to use dma_data (instead of pointers) hex-dma64: simplify lazy buffer mapping and clonning hex-fusion: factor out try_fuse_common that checks for dma64 buffers hex-bufs: minor cleanup for mmaping logic hex-bufs: simplify buffer clonning hex-ssm-conv: tighten gating checks and check vtcm size in kparams hex-binary: fix incorred mod/wrap in scalar ops hex-binary: make sure to call precompute kparams in support checks hex-dma64: update addr handling in mm,concat,binary hex-dma64: fixing up leftover of dma_addr_t conversion hex-binary: redo the kernel selection again and fix regressions in MOEs hex-binary: specialize per-type/per-op hex-binary: vtcm-layout and per-src dma-queue hex-dma64: update dma_push to transparently handle 64bit/extended
…to 64bit addr math
… limits much better
…ls and float promos
github-actions
Bot
added
documentation
max-krasnyansky
changed the title
Member
Author
|
@lhez please review/ack when you get the chance |
max-krasnyansky
added
the
merge ready
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
…s + improvements (ggml-org#29197) * hex-dma64: enable support extended buffer mappings and 64bit dma hex-dma64: expand binary ops to support more DMA scenarios hex-dma64: add binary-ops.h hex-dma64: add --hex-dma64 to run.py and fix minor issues hex-dma64: update SSM_CONV to use dma with proper support for 64bit hex-ops: remove obsolete gate for % 128 in binary ops hex-l2: dont check weight tensors against dirty ranges hex-dma64: most binary ops now support dma hex-dma: use dma_addr_t instead of plain uint64_t to avoid overhead on older targets hex-dma: update all dma users to use dma_data (instead of pointers) hex-dma64: simplify lazy buffer mapping and clonning hex-fusion: factor out try_fuse_common that checks for dma64 buffers hex-bufs: minor cleanup for mmaping logic hex-bufs: simplify buffer clonning hex-ssm-conv: tighten gating checks and check vtcm size in kparams hex-binary: fix incorred mod/wrap in scalar ops hex-binary: make sure to call precompute kparams in support checks hex-dma64: update addr handling in mm,concat,binary hex-dma64: fixing up leftover of dma_addr_t conversion hex-binary: redo the kernel selection again and fix regressions in MOEs hex-binary: specialize per-type/per-op hex-binary: vtcm-layout and per-src dma-queue hex-dma64: update dma_push to transparently handle 64bit/extended * hex-cpy: fix improper rebase with the fixes for cont. tensors * hex-dma-cpy: update CPY to use safe dma rows/size limits * hex-mmap: bump number of mmaps to 64 to allow avoid eviction in larger models * hex-dma: add support for the secondary ring as a fallback for too-large transactions * hex-rope: fix freq_factors access with 64bit dma * hex-dma: audit all ops for proper use/gards for 64bit addresses * hex-dma64: uninline glu-compute funcs to avoid register pressure due to 64bit addr math * hex-dma64: refactor binary ops to separate dma loops * hex-devel: add inspect script to help with dbg and analysis * hex-dma: refactor dma-pipelines in unary-ops * hex-dma: rewrite softmax to use dma * hex-dma: rewrite GDN dma loops and improve HVX register usage * hex-gdn: fuse GDN+CPY * hex-mm: factor out HVX solver * hex-mm: remove hvx-flat kernels, the chunked version now handles vtcm limits much better * hex-buffs: reject huge buffer allocations that we cannot memory map * hex-inspect: add logic to look for float promo calls * hex-mm: reduce HVX register spills in HVX prompt kernels * hex-bufs: do not double count buffers from tensors in the same op * hex-roll: fix merge conflict * hex-dma: reroute all matmul ddr kernels to new chunked dma/vtcm kernels * hex-dev: update developer docs to include inspection for register spils and float promos * hex-ops: forgot to add new headers * hex-softmax: fix gpt-oss dims * hex-dma64: cleanup dma_addr_t casts * hex-dma64: add support for dma/vtcm for flash-atten with sinks * hex-mm-add: fix MUL_MAT+ADD fusion with bias.weights in extended bufs * hex-add-id: add support for dma for src1 (exp. table) * hex-dma: imrpove v73 fallback paths * hex-bufs: do not drop extended mappings during va defrag * hex-scripts: fix flake8 warnings * hex-docs: fix editor-config warnings * hex-inspect: fix warnings from ty
frostyautumnleaf
pushed a commit
to frostyautumnleaf/llama.cpp
that referenced
this pull request
Oct 5, 2026
…s + improvements (ggml-org#29197) * hex-dma64: enable support extended buffer mappings and 64bit dma hex-dma64: expand binary ops to support more DMA scenarios hex-dma64: add binary-ops.h hex-dma64: add --hex-dma64 to run.py and fix minor issues hex-dma64: update SSM_CONV to use dma with proper support for 64bit hex-ops: remove obsolete gate for % 128 in binary ops hex-l2: dont check weight tensors against dirty ranges hex-dma64: most binary ops now support dma hex-dma: use dma_addr_t instead of plain uint64_t to avoid overhead on older targets hex-dma: update all dma users to use dma_data (instead of pointers) hex-dma64: simplify lazy buffer mapping and clonning hex-fusion: factor out try_fuse_common that checks for dma64 buffers hex-bufs: minor cleanup for mmaping logic hex-bufs: simplify buffer clonning hex-ssm-conv: tighten gating checks and check vtcm size in kparams hex-binary: fix incorred mod/wrap in scalar ops hex-binary: make sure to call precompute kparams in support checks hex-dma64: update addr handling in mm,concat,binary hex-dma64: fixing up leftover of dma_addr_t conversion hex-binary: redo the kernel selection again and fix regressions in MOEs hex-binary: specialize per-type/per-op hex-binary: vtcm-layout and per-src dma-queue hex-dma64: update dma_push to transparently handle 64bit/extended * hex-cpy: fix improper rebase with the fixes for cont. tensors * hex-dma-cpy: update CPY to use safe dma rows/size limits * hex-mmap: bump number of mmaps to 64 to allow avoid eviction in larger models * hex-dma: add support for the secondary ring as a fallback for too-large transactions * hex-rope: fix freq_factors access with 64bit dma * hex-dma: audit all ops for proper use/gards for 64bit addresses * hex-dma64: uninline glu-compute funcs to avoid register pressure due to 64bit addr math * hex-dma64: refactor binary ops to separate dma loops * hex-devel: add inspect script to help with dbg and analysis * hex-dma: refactor dma-pipelines in unary-ops * hex-dma: rewrite softmax to use dma * hex-dma: rewrite GDN dma loops and improve HVX register usage * hex-gdn: fuse GDN+CPY * hex-mm: factor out HVX solver * hex-mm: remove hvx-flat kernels, the chunked version now handles vtcm limits much better * hex-buffs: reject huge buffer allocations that we cannot memory map * hex-inspect: add logic to look for float promo calls * hex-mm: reduce HVX register spills in HVX prompt kernels * hex-bufs: do not double count buffers from tensors in the same op * hex-roll: fix merge conflict * hex-dma: reroute all matmul ddr kernels to new chunked dma/vtcm kernels * hex-dev: update developer docs to include inspection for register spils and float promos * hex-ops: forgot to add new headers * hex-softmax: fix gpt-oss dims * hex-dma64: cleanup dma_addr_t casts * hex-dma64: add support for dma/vtcm for flash-atten with sinks * hex-mm-add: fix MUL_MAT+ADD fusion with bias.weights in extended bufs * hex-add-id: add support for dma for src1 (exp. table) * hex-dma: imrpove v73 fallback paths * hex-bufs: do not drop extended mappings during va defrag * hex-scripts: fix flake8 warnings * hex-docs: fix editor-config warnings * hex-inspect: fix warnings from ty
edwardyoon
pushed a commit
to edwardyoon/focus-llama
that referenced
this pull request
Oct 7, 2026
…s + improvements (ggml-org#29197) * hex-dma64: enable support extended buffer mappings and 64bit dma hex-dma64: expand binary ops to support more DMA scenarios hex-dma64: add binary-ops.h hex-dma64: add --hex-dma64 to run.py and fix minor issues hex-dma64: update SSM_CONV to use dma with proper support for 64bit hex-ops: remove obsolete gate for % 128 in binary ops hex-l2: dont check weight tensors against dirty ranges hex-dma64: most binary ops now support dma hex-dma: use dma_addr_t instead of plain uint64_t to avoid overhead on older targets hex-dma: update all dma users to use dma_data (instead of pointers) hex-dma64: simplify lazy buffer mapping and clonning hex-fusion: factor out try_fuse_common that checks for dma64 buffers hex-bufs: minor cleanup for mmaping logic hex-bufs: simplify buffer clonning hex-ssm-conv: tighten gating checks and check vtcm size in kparams hex-binary: fix incorred mod/wrap in scalar ops hex-binary: make sure to call precompute kparams in support checks hex-dma64: update addr handling in mm,concat,binary hex-dma64: fixing up leftover of dma_addr_t conversion hex-binary: redo the kernel selection again and fix regressions in MOEs hex-binary: specialize per-type/per-op hex-binary: vtcm-layout and per-src dma-queue hex-dma64: update dma_push to transparently handle 64bit/extended * hex-cpy: fix improper rebase with the fixes for cont. tensors * hex-dma-cpy: update CPY to use safe dma rows/size limits * hex-mmap: bump number of mmaps to 64 to allow avoid eviction in larger models * hex-dma: add support for the secondary ring as a fallback for too-large transactions * hex-rope: fix freq_factors access with 64bit dma * hex-dma: audit all ops for proper use/gards for 64bit addresses * hex-dma64: uninline glu-compute funcs to avoid register pressure due to 64bit addr math * hex-dma64: refactor binary ops to separate dma loops * hex-devel: add inspect script to help with dbg and analysis * hex-dma: refactor dma-pipelines in unary-ops * hex-dma: rewrite softmax to use dma * hex-dma: rewrite GDN dma loops and improve HVX register usage * hex-gdn: fuse GDN+CPY * hex-mm: factor out HVX solver * hex-mm: remove hvx-flat kernels, the chunked version now handles vtcm limits much better * hex-buffs: reject huge buffer allocations that we cannot memory map * hex-inspect: add logic to look for float promo calls * hex-mm: reduce HVX register spills in HVX prompt kernels * hex-bufs: do not double count buffers from tensors in the same op * hex-roll: fix merge conflict * hex-dma: reroute all matmul ddr kernels to new chunked dma/vtcm kernels * hex-dev: update developer docs to include inspection for register spils and float promos * hex-ops: forgot to add new headers * hex-softmax: fix gpt-oss dims * hex-dma64: cleanup dma_addr_t casts * hex-dma64: add support for dma/vtcm for flash-atten with sinks * hex-mm-add: fix MUL_MAT+ADD fusion with bias.weights in extended bufs * hex-add-id: add support for dma for src1 (exp. table) * hex-dma: imrpove v73 fallback paths * hex-bufs: do not drop extended mappings during va defrag * hex-scripts: fix flake8 warnings * hex-docs: fix editor-config warnings * hex-inspect: fix warnings from ty (cherry picked from commit 0c3626e)
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
This a long overdue update of our buffer and DMA handling to support 64-bit (aka extended) mappings.
64-bit DMA is supported on Hexagon v81 and up (Gen5, X2-Elite, IQ10) and allows us to memory map buffers above the 4GB virtual address space on the NPU. Those buffers can be used only with the DMA engine (
DDR -> DMA[in] -> VTCM -> DMA[out] -> DDR).The performance improvements are noticeable with larger models like gemma-4-26B, gpt-oss-20b, ...
64-bit DMA support removes all overhead of having to dynamically map/unmap the tensor buffers during inference.
This update required quite a few changes in many ops. Some ops like
MUL_MAT,FA,GDN,SSM_CONV, ... had to be completely overhauled to avoid reading tensors directly viaDDR->L2->HVX. This is a good thing in general (not specific to 64bit) and allows for better memory BW utilization.MUL_MAThas seen the most updates and I was able to get rid of all the legacy paths.For the ops that already supported the DMA in all paths the changes are focused on properly maintaining
dma_addr_tfor the base addresses. This is a no-op on the older Hexagon arches.This change uncovered some issues with the register spills (both scalar and HVX) and requires a slightly different approach to optimizing the kernels (ie not inlining the inner micro-kernels into the dma pipeline/loop). I added a new little tool
ggml-hexagon-inspect.pythat can do disasm, detect register spills and unnecessary float conversions (snapdragon/developer.mdhas been updated to include this info).As part of the DMA overhaul I also completely re-wrote large-transaction fallback path for v73 (ie when row size or n_rows exceeds DMA engine limits). It's now implemented as a separate ring buffer that is used to flush out large transaction without affecting the rest of the pipeline sequence (helps with newer/larger models on X-Elite and IQ9).
This feature has been extensively tested on all supported devices and a good set of the latest models. So it's enabled by default. It's possible that I missed some scenarios though. So if folks see issues please report and you can disable it with
GGML_HEXAGON_DMA64=0orscripts/snapdragon/run.py --hex-dma64 0 ....Requirements
I have read and agree with the contributing guidelines
AI usage disclosure: Yes, lots of help from Antigravity (3.8-Flash) with refactoring, testing, review, etc.