Repository navigation
Record both sides at the fork's compact-factor region floor - #9
Merged
Merged
Conversation
tensor4all/faer#9 removes a per-block rayon region from the compact Householder factor build. The pair is two tlinalg checkouts differing only in the fork revision: bf88c0a (fork 22f23cf) and 7fdba8a (fork 6e8e178), MKL arm, 8T, the qr / householder families, one complete set each. `qr` faer-pool medians after/before are 0.576 (f64) and 0.742 (c64), the 4096x256 and 1024x64 tall cells are unchanged, and the unchanged MKL binary and the sequential `faer-1lane` rows sit at 1.00 as the floor. It is a scan, not a paired comparison: one complete set, so no pooled cell is claimed.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Records both sides of tensor4all/faer#9 at 8T on
zen5-cpu, MKL arm, one complete set each. The two cells arebf88c0a(fork22f23cf) and7fdba8a(fork6e8e178) — tlinalg checkouts that differ only in the fork revision theirCargo.tomlnames — so the pair isolates the fork change.result/zen5-cpu/tlinalg-kernels/bf88c0a6e568-mkl.md— beforeresult/zen5-cpu/tlinalg-kernels/7fdba8a53ba3-mkl.md— afterWhat it says, medians of
after / beforeforfaer-poolat 8T:qr0.576 (f64) / 0.742 (c64),householder_factor0.984,householder_apply0.992. The floor is in the same pair: everyfaer-1lanerow (sequential, never threads) reads 0.988–1.003 and everylapack-mklrow 0.999–1.002.qr128x128 b128 c64qr256x256 b1 c64qr512x512 b4 c64qr4096x256 b1 f64 / c64Both pages are
partialcoverage (three families, one thread budget) and the runs are a scan, not a paired comparison:--aa 1, so no pooled cell is claimed. One pooled cell moved the wrong way (+92%,householder_apply32x32 b128 f64) — a shape that fans the batch out over lanes with a sequential inner call, so the fork change is not in that path; the worklog entry says so rather than leaving it unexplained.The two revisions
The pair's checkouts are branches in
tlinalg-rs, differing only in the fork revision theirCargo.tomlnames, so both resolve for the validator:region-floor-before=bf88c0a6e568b70d5bc724d925da3e99b1601609(fork22f23cf)bump-faer-region-floor=7fdba8a53ba35fd36a088695741bef270c3548ce(fork6e8e178, re-pinned to the merged04542faeasd576230after perf(householder): no parallel region below the compact factor's own floor faer#9 landed)