bench: headline fallback for pre-#147 submissions; add M1 datapoint (#148) - #150
Merged
sbryngelson merged 1 commit intoAug 1, 2026
Conversation
…datapoint (sbryngelson#148) _perf_headline now falls back to the raw saturation + bandwidth sweeps for GEMM / bandwidth / ridge when roofline_analysis is absent or errored -- so a submission from before sbryngelson#147 (where roofline_analysis KeyError'd on a device a sweep could not measure) still shows a full headline row instead of three blanks. Also widen the bandwidth cell format so sub-1 GB/s ANE values read correctly (0.86, not 1). Incorporates the Apple M1 (MacBook Pro 17,1) submission from sbryngelson#148 (@diegobauavi); its roofline_analysis errored the same way M2 Pro's did, now recovered by the fallback. Table now spans M1 / M2 Pro / M5 Pro, and the slice-x16 clamp shows present on M1+M2, exact on M5 -- the pre-A16 -> A16 transition bracketed by data. Co-authored-by: diegobauavi <diegobauavi@users.noreply.github.com>
2 tasks
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Two things, both consequences of #146/#147 landing while data was already in flight.
1. Headline fallback in
aggregate_rooflines.py._perf_headlinenow derives GEMM / bandwidth / ridge from the raw saturation + bandwidth sweeps whenroofline_analysisis absent or errored. Submissions taken before #147 (whereroofline_analysisKeyError'd on a device a sweep could not measure) carry a rc=1 blob for that step, so their headline GEMM/bandwidth/ridge were blank even though the numbers sit in the report. The fallback recovers them, and future-proofs the table against that whole failure class. Also widened the bandwidth cell format so sub-1 GB/s ANE values render correctly (0.86, not a rounded 1).2. Incorporates the M1 datapoint from #148 (@diegobauavi, MacBook Pro 17,1). Its
roofline_analysiserrored exactly like M2 Pro's; the fallback recovers it. I pulled the submission JSON in here and regenerated the combined table because #148 conflicted onROOFLINES.mdafter #145 merged (both regenerate it) -- closing #148 in favor of this, with credit preserved on the row and in the co-author trailer.The table now spans M1 / M2 Pro / M5 Pro, and the science shows: matmul cliff ~32759 on all three (silicon-invariant), and the slice-x16 crop-DMA clamp present on M1 + M2, exact on M5 -- the pre-A16 -> A16/M5 transition bracketed by real data.
ruff clean;
aggregate_rooflines.py --checkgreen; ASCII-clean.