Repository navigation
lab(drf-p2): cold-cache arm; correct the DRAM claim (stranded from #1763) - #1764
Conversation
A focus touches one cache line per distinct ordinal (248-256 KiB here), so the repeated 65 536-row reps were warm, not DRAM. The probe now adds a cold arm that streams a 512 MiB buffer (twice the 260 MiB L3) outside the timed region before each rep, and computes the cold crossover as the smallest N with build_cold + N*fold < direct_cold + (N-1)*direct_warm. Break-even stays 1-3 folds warm and cold. The board entry is rewritten with both arms and the correction. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01G22yT6htkcdyXsihxxXdrg
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars.
 |
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8836b669e1
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01G22yT6htkcdyXsihxxXdrg
|
Add Carrot credits or activate Agent usage billing to use Autofix |
|
@codex review The finding on Generated by Claude Code |
|
Codex Review: Didn't find any major issues. What shall we delve into next? Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
Why
#1763 was merged one commit before the fix for its only review finding was pushed. As a result, main says the 65 536-row timings were taken with rows in DRAM, and they were not. A focus reads one cache line per distinct ordinal, about 248–256 KiB here, so those repeated reps were warm. This is the same commit (
645ec9bf9), rebased onto main.What
examples/p2_breakeven.rs: before each rep, outside the timed region, the probe streams a 512 MiB buffer, twice the 260 MiB L3. It reports cold build and cold direct times.build_cold + N·fold < direct_cold + (N−1)·direct_warm.2026-10-24-drf-p2-breakeven.mdrewritten: it now shows both arms, records the correction, and removes the DRAM conclusion.Result
Break-even stays at 1–3 folds, warm or cold. A cold first touch costs about 30 µs at 1024 rows and 80–150 µs at 65 536 rows, roughly the same on both paths. The full table is in the entry.
Gates: 11/11 tests pass,
clippy --all-targets -D warningsis clean, andentries_index.py --checkreports the index current.Generated by Claude Code