Skip to content

[VFDLIBM-30] Record the sinh and cosh timings on xeon8573c and epyc7763 - #131

Merged
MaxGekk merged 3 commits into
mainfrom
record-sinh-cosh-timings
Nov 22, 2026
Merged

MaxGekk merged 3 commits into
mainfrom
record-sinh-cosh-timings

Conversation

@MaxGekk

@MaxGekk MaxGekk commented Nov 22, 2026 •

Copy link
Copy Markdown
Contributor

What changes were proposed in this pull request?

Part of #30.

Adds six results files, written by bench/record.py write from two benchmark runs and checked with bench/record.py check <file> --run <id>: sinh-, cosh- and cosh-kinds- for xeon8573c (run 44940776246, main at 5d6eb14) and for epyc7763 (run 44929622224, main at 03ea082), all dated 2026-11-21.

Each file's title, reason and reading are written by hand from that file's own tables, by minimums, with a ratio inside about 1.3x read as no direction and no cause given.

The 512-bit rows of the EPYC files are wider than its vectors, run in Java and allocate megabytes per op; the files say so and do not quote them.

No source, benchmark or other document changes. Run 44947864585 (epyc7763, 21 November) is the exp and log10 pair re-run, already recorded in exp-log10-pair-rerun-epyc7763-2026-11-21.txt (#128), so it needs no file here.

Why are the changes needed?

Issue #30 asks for current files for each function on each host, and no timing file of sinh or cosh existed, so the documents that quote or wait on sinh and cosh timings have nothing to cite yet.

Does this PR introduce any user-facing change?

No

How was this patch tested?

bench/record.py check passes for each of the six files against its run: sinh-xeon8573c-2026-11-21.txt, cosh-xeon8573c-2026-11-21.txt and cosh-kinds-xeon8573c-2026-11-21.txt with --run 44940776246, and the three epyc7763 files with --run 44929622224. Every figure in a reading is from the same file's table or summary.

Was this patch authored or co-authored using generative AI tooling?

Generated-by: Claude Code (Sonnet 5.5)

Max Gekk added 3 commits November 22, 2026 08:31
Six results files from runs 44940776246 (xeon8573c) and 44929622224 (epyc7763), written by bench/record.py and checked against the runs' artifacts.

Generated-by: Claude Code (Sonnet 5.5)
… near-1.3x readings

Generated-by: Claude Code (Sonnet 5.5)
…name Exp and Expm1 as loaded code

Generated-by: Claude Code (Sonnet 5.5)
@MaxGekk
MaxGekk merged commit 8a0685e into main Nov 22, 2026
10 checks passed
MaxGekk added a commit that referenced this pull request Nov 22, 2026
### What changes were proposed in this pull request?

`ci.yml` gets a `concurrency` group keyed on the pull request's ref, with `cancel-in-progress` true for `pull_request` events only, so a push to a pull request cancels that pull request's run still going.
Any other run (a push to `main`, a call from `release.yml`) is keyed on its run id, a group of its own, so it is never cancelled.
`release.yml` keeps its group `release-${{ github.ref }}`; `cancel-in-progress` becomes true on pull requests only, so a tag's or a dispatch's dry run is never cancelled while it runs.
Comments in both files state the cases, including that a re-run of an older run of a pull request cancels the head's run, so only the newest is re-run.

### Why are the changes needed?

A stream that pushes several fix commits to one pull request leaves a CI run going per commit; on #131 two ran at once and the stale one was cancelled by hand. Those runs hold runners the nightly and the benchmark dispatches wait for, and their results say nothing about the head.

### Does this PR introduce _any_ user-facing change?

No. CI only.

### How was this patch tested?

`actionlint` on both files (one shellcheck info at `ci.yml` line 193, which this change does not touch).
This pull request was pushed in two commits; the run of the first is cancelled by the push of the second (checked on this pull request's runs).
A `code-review high` found three nits, all in the comments, which the second commit fixes.

Generated-by: Claude Code (Opus 5.5)
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant