stillwater-sc / universal / 25944190078

84%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

feat(lns): CORDIC hyperbolic add/sub policy (Phase E of #777) (#836)

* creating a cordic playground example

* feat(lns): CORDIC hyperbolic add/sub policy (Phase E of #777)

Adds the deferred Phase E hardware-codesign tier to the configurable lns
add/sub framework. CORDICAddSub<Lns, MaxIterations = Lns::rbits> implements
classical hyperbolic CORDIC: rotation mode to compute 2^d, vectoring mode
to compute log2(1 +/- v), with the canonical iteration repetitions at
indices 4, 13, and 40 for convergence. MaxIterations is exposed as a
non-type template parameter so a hardware-codesign team can sweep
truncated iteration budgets independently of rbits.

The trigger for this work fired on 2026-05-15 (an internal Stillwater
hardware-codesign project asked for it; see #783 trigger comment), which
expanded the original acceptance criteria with a precision/accuracy
assessment requirement.

Deliverables in this PR:
- include/sw/universal/number/lns/lns_addsub_algorithms.hpp: CORDICAddSub
  policy class with constexpr atanh table generation, K_h gain
  compensation, the required hyperbolic-CORDIC repeat sequence, and a
  cancellation-regime direct-eval fallback (consistent with
  PolynomialAddSub, LookupAddSub, ArnoldBaileyAddSub).
  Plus lns_addsub_log_error_bound<CORDICAddSub<Lns, K>> specialization.
- static/logarithmic/lns/arithmetic/log_add_algorithms.cpp: CORDIC cross-
  validation at lns<8,4> (exhaustive) and corner-case + Tier 1
  cancellation sweeps at lns<32,24>.
- benchmark/performance/arithmetic/lns/log_add_algorithms.cpp: CORDIC row
  added to the per-algorithm throughput + accuracy comparison.
- benchmark/accuracy/lns/cordic_precision_assessment.cpp (new): per-
  iteration convergence study (k = 1..rbits), ULP error histograms, and
  worst-case witness table across the rbits sweep
  lns<8,4>/<12,6>/<16,8>/<24,12>/<32,16>. Emits markdown to stdout.
- docs/design/cordic-precision-assessment.md (new): generated artifact
  from the assessment tool.
- doc... (continued)

Coverage Stats

85 of 86 new or added lines in 1 file covered. (98.84%)

5 existing lines in 2 files now uncovered.

45987 of 54789 relevant lines covered (83.93%)

6470240.9 hits per line