stillwater-sc / universal / 28339727021

85%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

feat(efloat): implement mathematical power and exponential library (#1126)

* feat(efloat): implement mathematical power and exponential library (#1113)

Implement pow(x, int_y) (fast integer power), pow(x, const_y) (transcendental power),
exp2(), exp10(), and expm1() (with Taylor series small-input protection and
dynamic precision-based loop termination) inside math/pow.hpp.

Register math/pow.hpp inside our master efloat mathlib.hpp header.

Create pow.cpp inside elastic/efloat/math/ to verify all power and exponential
functions against extreme boundary conditions, negative base integers, and small Taylor range
cancellation protections.

Resolves #1113
Relates to parent Issue #1092, Epic #1101

* fix(efloat): resolve CodeRabbit reviews and ASCII failure on power library

1. Resolve ASCII Guard check failure in pow.cpp:
   replace the non-ASCII approximation symbol '≈' with standard ASCII '~=' inside
   the comments of pow.cpp, satisfying the repository's 7-bit ASCII-only rule.

2. Correct exp2(0) and exp10(0) return values:
   re-branch exp2 and exp10 in pow.hpp so they branch on x.iszero() separately and return
   exactly 1.0 instead of returning 0.0. Added zero-input tests inside pow.cpp.

3. Reorder NaN fast paths in pow(x, y) around IEEE boundaries:
   put y.iszero() and x == 1.0 checks BEFORE the nan-checks so that pow(NaN, 0.0) and
   pow(1.0, NaN) correctly return 1.0 according to IEEE-754 standards. Added boundary tests.

4. Guard integer_power from INT_MIN negation overflow:
   cast the negated exponent to uint64_t safely ('0ULL - static_cast<uint64_t>(exponent)')
   inside integer_power to prevent signed integer overflow. Added pow(-1, INT_MIN) test.

5. Implement efloat-native integer parity checks in pow(x, y):
   replace narrowing 'llrint(y)' checks with exact, non-narrowing bit-level checks
   checking if the bit at weight 2^0 is 1.

Resolves #1113, PR #1126

Coverage Stats

86 of 97 new or added lines in 1 file covered. (88.66%)

2 existing lines in 1 file now uncovered.

49094 of 57996 relevant lines covered (84.65%)

6697641.53 hits per line