llama-mmap: add MADV_HUGEPAGE hint for THP on Linux#22022
Open
Marxist-Leninist wants to merge 1 commit intoggml-org:masterfrom
Open
llama-mmap: add MADV_HUGEPAGE hint for THP on Linux#22022Marxist-Leninist wants to merge 1 commit intoggml-org:masterfrom
Marxist-Leninist wants to merge 1 commit intoggml-org:masterfrom
Conversation
Issue madvise(MADV_HUGEPAGE) on the read-only file mapping used for model weights on Linux. For a 1 GB model this drops the potential page count from ~262K 4KB pages to ~512 2MB pages, reducing TLB pressure and (more importantly) reducing the number of re-faults when pages get evicted under memory pressure. No-op on kernels where THP is disabled. On 'madvise' mode (the common modern default for desktop distros), this is opt-in and requires the caller to ask. Guarded by defined(MADV_HUGEPAGE) so it compiles cleanly on non-Linux. Benchmark on a Skylake-SP VM, Bonsai-8B Q1_0, -fa on -ctk q8_0 -ctv q8_0 -t 12 -ub 128: neutral on this machine (~9.5 t/s tg128 both before and after) because the VM isn't memory-constrained. The change is intended for systems where the mapping does get evicted and re-faulted under pressure.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Calls madvise(MADV_HUGEPAGE) on the read-only mmap region used for model weights on Linux.
When THP is in 'madvise' mode (the default on many desktop distros), this opts the mapping
into transparent huge page promotion. For a 4-5 GB model weight map the page count drops from
~1M 4 KB pages to ~2K 2 MB pages, which reduces TLB pressure and the number of minor faults
on the hot inference path.
No configuration required. The call is advisory: if THP is disabled
(transparent_hugepage/enabled == never) or pages cannot be promoted, madvise returns EINVAL
and a debug-level log message is emitted. NUMA-aware allocations are skipped (existing !numa
guard). Guarded by defined(MADV_HUGEPAGE) for portability.
Tested on Linux x86_64 with THP=madvise. Neutral on an unloaded machine (pages stay resident);
reduces re-fault latency spikes under memory pressure.
Complements #21821 which adds HugeTLB support; that approach requires pre-allocated pages in
the HugeTLB pool, while THP requires no system configuration.