stillwater-sc / universal / 28298610954

85%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

feat(efloat): implement logarithmic suite (log2, log10, log1p) (#1123)

* feat(efloat): implement logarithmic suite (log2, log10, log1p) (#1108)

Implement log2() (logarithm base 2) and log10() (logarithm base 10) inside math/logarithm.hpp
utilizing high-precision ln2 and ln10 constants.

Implement log1p() (natural logarithm of 1 + x) inside math/logarithm.hpp, utilizing standard
Taylor series expansion for small inputs (|x| < 0.375) to prevent catastrophic cancellation as
x -> 0.

Fix power-of-2 sign-clobbering bug in operator*=: inside the second power-of-2 bypass block,
save the original sign of *this to original_sign before the copy-assignment *this = rhs, and
evaluate the sign XOR using original_sign. This prevents the product of a power of 2 by any
negative number from erroneously becoming positive, fully resolving Taylor series convergence for
negative arguments.

Expand math.cpp to test log2(4.0), log10(100.0), and log1p() under extremely small inputs and
exceptional boundaries (-2.0 nan / -1.0 -inf).

Resolves #1108
Relates to Epic #1101, PR #1122

* fix(efloat): resolve CodeRabbit review comments on logarithm suite

1. Implement exact operator<(): replace the operator<() stub in efloat_impl.hpp with a complete,
   constexpr-compatible, standard-conforming implementation. It handles NaN unorderedness,
   zero sign equivalence, exponent scales, and lexical limb comparisons under ties. This
   makes ordering comparisons in logarithm.hpp fully functional.

2. Fix negative-infinity domain checks: replace 'x.isneg()' with 'x.sign() == -1' inside
   log(), log2(), log10(), and log1p() to ensure negative infinity is caught and converted
   to NaN with InvalidOperation instead of bypassing domain checks.

3. Implement precision-safe log1p(): replace double-casting range checks with exact
   efloat-space comparisons 'abs(x) < efloat<nlimbs>(0.375)'.

4. Implement dynamic termination: replace the hardcoded iteration cap of 100 with a
   precision-safe ... (continued)

Coverage Stats

80 of 116 new or added lines in 2 files covered. (68.97%)

5 existing lines in 2 files now uncovered.

48769 of 57645 relevant lines covered (84.6%)

6737472.27 hits per line