stillwater-sc / universal / 26489863160

84%

master: 84%

Build Type

push

github

Committed by web-flow

Commit Message

perf(bfloat16): marshal exponent functions through float; add IEEE special-case tests (#1017)

Addresses PR #1015 review feedback on the new bfloat16 exponent-manipulation
functions (ldexp/frexp/scalbn/logb/ilogb, issue #941).

- Maintainer (Ravenwater): marshal through float instead of double. A bfloat16
  is the high 16 bits of an IEEE float32 and shares its 8-bit exponent field, so
  bfloat16 -> float is exact and covers bfloat16's entire exponent range. float
  is therefore the natural, cheaper marshalling type for exponent manipulation
  (no widening to double); results are identical, only the conversion is cheaper.
- CodeRabbit: extend the test corpus with IEEE special cases. Added a section
  exercising +0/-0, +/-inf, and NaN: frexp/ldexp/scalbn preserve the value class,
  and ilogb returns the std sentinels (FP_ILOGB0, INT_MAX, FP_ILOGBNAN).

Verified gcc+clang.

Relates to #941

Coverage Stats

5 of 5 new or added lines in 1 file covered. (100.0%)

20 existing lines in 3 files now uncovered.

46866 of 55805 relevant lines covered (83.98%)

5776355.7 hits per line