llama.cpp
hexagon: improve buffer allocations and mappings (mmap stacking)
#32144
Merged
Merged
hexagon: improve buffer allocations and mappings (mmap stacking) #32144
max-krasnyansky
merged
5 commits into
ggml-org:master
from
qualcomm:hexagon-shared-bufs
hex-bufs: combine queue and fence buffers
5b8c872d
hex-bufs: use n_threads for queue sizing
d8bb1b93
hex-mmap: optimize buffer stacking at the host to avoid iova fragment…
2e6b360b
hex-mmap: add simply retry for spurios mmap failures
9a8f1e84
hex-mmap: use eager mode for extended mmap retries
ed952501
lhez
approved these changes
on 2026-11-12
jhen0409
approved these changes
on 2026-11-13
Login to write a write a comment.
Login via GitHub