llvm-project

274ab686 - [libc][math] Optimize pow 128-bit and 256-bit accurate passes and share exp2_df128. (#237026)

[libc][math] Optimize pow 128-bit and 256-bit accurate passes and share exp2_df128. (#237026) - Refactor 128-bit and 256-bit accurate passes to use multi-tier fixed-point arithmetic (Frac64, Frac128, Frac256) and share double_eval::exp2_df128 with exp/exp2/exp10. - Simplify is_exact_rounding_boundary to operate on normalized (x_e, x_mant), unify power loops, and replace 64-bit hardware divisions with a leading-zero bit-width bound. - Avoid general multi-limb normalization and conversion overhead in pow_accurate and pow_accurate_256. Assisted by: Gemini for analysis and refactoring