[SMI-Lab-Inha]

Summary

First item of the speed programme before 0.3.0.

The problem. Profiling the release driver (IFX, static) on the benchmark decks put the largest share in the vectorised cos/sin/exp of the wave kinematics: 24 % of lazy_wave_vessel_rao and 37 % of als_volturnus_line_break_time. The element kernels came next. The release build targets baseline x86-64, so this vector maths ran only two doubles wide (SSE2).

What changes:

• CABLEDYN_TARGET_AVX2 (CMake option, default OFF) compiles the Fortran core for x86-64-v3 (Intel Haswell and AMD Excavator or newer):
  • IFX: /arch:CORE-AVX2 /Qfma-. The /arch form is the vendor-neutral one; /Qx gave the same code and speed here.
  • gfortran 11+: -march=x86-64-v3 -ffp-contract=off.
  • The static Windows release turns it on.
• FMA contraction stays off. With it, the compiler fuses the same expression differently where it is inlined. A restored instance then no longer continued the run bit for bit (test_line_attachments failed with gfortran x86-64-v3). Contraction was worth only 2–5 % more speed.
• Start-up processor check (src/cabledyn_cpu_guard.c):
  • It is compiled for baseline x86-64 and runs as a CRT initializer (MSVC) or constructor (GNU), before any Fortran code.
  • On a processor or OS without x86-64-v3 it writes the missing instructions to stderr and exits with status 3. Without it, the program would crash at the first AVX2 instruction with no explanation.
  • It uses only operating-system calls. In the Windows GNU build, C library names such as fprintf resolved to openblas.dll's exports and imported it at load time, which defeated the run-time BLAS loading.
  • Setting CABLEDYN_CPU_GUARD_SIMULATE exercises the stop on any processor.
• Identification: CableDyn_driver --version prints the build line (x86-64-v3 (AVX2) or baseline). The OpenFAST module prints CableDyn build: x86-64-v3 (AVX2). at initialisation. These references also keep the check in the static links.
• Release script: builds the reference LAPACK with /arch:CORE-AVX2. It requires both executables to report the x86-64-v3 build, and to stop with exit 3 and the message under the simulated missing feature.
• Docs: processor requirement in doc/installation.rst and the README asset table; the option in DEVELOPMENT.md; CHANGELOG entry.

Speed (IFX static release configuration, this workstation, mean of two runs)

Deck	dev	this PR	change
lazy_wave_vessel_rao	11.9 s	8.8 s	−26 %
buoy_clamped_cable	2.11 s	1.83 s	−13 %
als_volturnus_line_break_time	35.5 s	23.9 s	−33 %

Result changes (round-off)

Deck	Largest change, relative to the channel's range
lazy_wave_vessel_rao	3.0e-7 (last printed digit)
buoy_clamped_cable	7.4e-7 (last printed digit)
als_volturnus_line_break_time	identical traces until about 900 s; after that, the platform's slow drift following the line break at 300 s amplifies the round-off

On the line-break deck, statistics over the whole run match dev:

Channel	Max, dev	Max, this PR	Std, dev	Std, this PR
AnchTen2	1.9177e6	1.9177e6	2.9901e5	2.9902e5
FairTen3	1.2722e7	1.2723e7	1.1631e6	1.1631e6
Body1Rz	4.5699	4.5699	1.4069	1.4068

Verification

• Default build (Windows, gfortran, Release, slow tier included):
  • ctest 297/297;
  • pytest 2051 passed;
  • pre-commit run --all-files and the Sphinx -W build pass.
• -DCABLEDYN_TARGET_AVX2=ON build (gfortran, x86-64-v3): ctest 298/298, including cpu_guard and the end-to-end cpu_guard_stop. libcabledyn.dll imports no openblas.dll.
• IFX static release configuration:
  • --version reports x86-64-v3 (AVX2);
  • with CABLEDYN_CPU_GUARD_SIMULATE the driver prints the message and exits 3 before the banner.
• OpenFAST (gfortran, CMake integration): builds with the module change, and a 2 s coupled IEA-15 MW CompMooring = 5 run terminates normally.

Checklist

• pre-commit run --all-files passes
• ctest --test-dir build --output-on-failure passes
• python -m pytest tests python/tests passes (if Python or validation records changed)
• VALIDATION.md updated if a validation result changed (round-off only; recorded above)
• CHANGELOG.md updated under [Unreleased] (heading added if missing) for user-visible changes
• Documentation in doc/ updated for changed inputs, options, or outputs
• Conventional Commit title (feat:, fix:, docs:, ...)
• Commits are signed off (git commit -s) under the
  Developer Certificate of Origin

[SMI-Lab-Inha]

@codex review

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-22T06:41:14.364386Z	7b6c2cf	Manual request

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" or "@codex security review".

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]

Codex Review: Didn't find any major issues. You're on a roll.

Reviewed commit: 7b6c2cf390

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".

[read-the-docs-community]

Documentation build overview

📚 CableDyn | 🛠️ Build #35317412 | 📁 Comparing 7b6c2cf against latest (d3ddfeb)

  🔍 Preview build  

10 files changed · ± 10 modified

± Modified

• changelog.html
• cli.html
• development.html
• faq.html
• installation.html
• modeling.html
• openfast.html
• options.html
• tutorial_openfast.html
• validation.html