Repository navigation
linux: drop the never-shipped LUT kernel and stop Nix builds compiling the benchmarks - #108
Merged
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
The speech-kernel crate
lf-cpu, which ships in the daemon, still carried benchmark-only code:gemm_lut: the exact-FP32 lookup-table GEMM from October.peak: register-only throughput probes that onlylf-gemm-bench --peakcalls.Separately, the Nix package builds only
-p lf-daemon, but its check phase rancargo test --workspace. Everynix buildtherefore also compiled and tested the benchmark tools.The changes:
lf-cpu/src/gemm_lut.rs, and the test-onlyavx512_poolhelper inlib.rsthat only its tests used. Its measurements stay in git history.lutvariant and the--lut-*options. f32 and i8x1–3,--export(paired timing across exports),--synthetic,--accuracy,--peakand the memory table are unchanged.peak.rsmoves into lf-bench as a private module of lf-gemm-bench, without its LUT probes.deny(clippy::undocumented_unsafe_blocks, clippy::missing_safety_doc).linux/flake.nix:cargoTestFlags = [ "--workspace" "--exclude" "lf-bench" ], with the comment updated. CI still tests the whole workspace.default-memberstolinux/Cargo.toml.linux/is a virtual workspace, so it would silently shrink what CI'scargo clippy --all-targetsandcargo testcover.Tests removed:
gemm_lut's own unit tests, together with the kernel they pinned. No other test changed.Not changed:
linux/CHANGELOG.md, because nothing user-visible changes.linux/PLAN.mdand the READMEs are left alone; the docs cleanup PR handles them. The nixpkgs derivation (nixpkgs PR 579077, localflow 0.1.0) is separate and could take the same--exclude lf-benchat its next version bump.Verification
Proof: AGENTS.md's three cargo commands with the private integration suites included, a sandboxed
nix build, and smoke runs of every remaining lf-gemm-bench mode.nix develop ./linux, noLF_SKIP_*set): fmt is clean and clippy has no warnings.real_model).nix build ./linux -L, before and after:--workspace, compiled lf-bench, and ran 50 suites (592 passed).--workspace --exclude lf-benchand never compiled lf-bench: 46 suites, 567 passed, 0 failed, 2 ignored.gemm_lut's 9.--peakworks at both tiers (LOCALFLOW_CPU_TIER=avx2for the AVX2 probes).--synthetic ternary|binary --accuracyworks, as does a paired run over two kea-b exports.--variants lutand--lut-panelare now rejected as unknown.nixfmt --check linux/flake.nixandgit diff --checkpass.make checkwas not run locally (it needs macOS); CI runs it. This PR touches no Swift.Risk and privacy
localflowdandlocalflowctlare built exactly as before.peak.rs's inline-asm probes moved crates unchanged, with their// SAFETY:comments, and the lint that enforces those comments came with them.Reconciled with #105–#112. Every pair, and the full set, merge onto main without conflicts, so they can merge in any order. All eight merged together pass
make checkandgit diff --checkon the Mac. The seven other than #112 also passmake -C iOS checktogether; #112 changes no file the iOS build compiles.Created with Claude Opus 5.5 in T3 Code.
Sol Loop: 1 rounds, 0 fixes (0 high, 0 medium, 0 low, 0 rejected)