Repository navigation
defk-w8r.2: record the handwired study's speed lessons in docs/performance.md - #77
Merged
Merged
Conversation
Five entries in docs/performance.md (2.19, 2.20, 6.43, 6.44 and a pointer in 3.1): anchor-then-verify with :binary.matches beats :re and a defk NFA, literal search ties a shift-and kernel, a backtracking regex spelling costs 8x, the NFA kernel's per-byte cost, and exact float32 through an [f32] scratch write. The GPU data point goes to docs/research/nx-boundary.md; the scope line (GPU is Nx's) is unchanged. The research note and its scripts were deleted; the ticket is the record.
DanielLuning
pushed a commit
that referenced
this pull request
Oct 21, 2026
…ired-lessons, worktree ../defk-w8r.2). Five entries in docs/performance.md (2.19, 2.20, 3.1 pointer, 6.43, 6.44) and the GPU data point in docs/research/nx-boundary.md; scope.md unchanged. mix ci green locally. Waits for the owner's merge; then bw close, bw sync, remove the worktree."
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What changed
Records the speed lessons of the handwired study (a deleted prototype; the ticket defk-w8r.2 is the record) in
docs/performance.md::binary.matches/2, then verify in Elixir. Beats:reand a defk bit-parallel NFA 1.5 to 2x; a backtracking regex spelling cost 8x in:re.:binary.matches/3with a scope matches a shift-and kernel. A pointer bullet in 3.1.[f32]scratch write is free (0.91x the f64 time), bit-exact in 15,000,030 results, 60x faster than rounding in Elixir; the 5.5x gap to C float is the runtime checks (defk-w8r.3); the builtin decision is defk-vcx.5.The GPU data point (best CPU times 7 to 15 ms in 12 chunks against handwired's WebGPU 4.5 to 5.2 ms, an order of magnitude only) goes to
docs/research/nx-boundary.md, replacing its "GPU not measured" line.docs/scope.mdis unchanged: GPU stays Nx's, and a user page gets no ticket ids.Results
No new measurement. Every number is copied from the ticket; each entry says one run, quiet machine, and that the source cannot be re-run. The three sketches carry
defk-example skipmarkers with reasons, because they are fragments.How it was checked
mix cion the M-series Mac (2305 passed, 0 failures);mix defk.docs --checkreports every generated region up to date.CHANGELOG.mdentry.test/stability/? No.