Improve sqrt with SoftFloat-style lookup table and integer arithmetic#1331

Merged

mborland merged 35 commits intoboostorg:developfrom

justinzhuguangwen:develop

Feb 6, 2026

Commits on Jan 24, 2026

Optimize sqrt using hardware binary floating-point sqrt
justinzhu@home
committed

Commits on Jan 25, 2026

feat: optimize sqrt with SoftFloat-style lookup table algorithm
justinzhu@home
committed
Optimize sqrt() using lookup table to reduce Newton iterations
justinzhu@home
committed
fix: remove unused file
justinzhu@home
committed
Optimize sqrt() using SoftFloat-style lookup table
justinzhu@home
committed
fix : delete commit message
justinzhu@home
committed

Commits on Jan 26, 2026

perf(cmath): optimize sqrt with lookup table + remainder elimination
justinzhu@home
committed

Commits on Jan 30, 2026

docs(doc): add decimal vs binary float representation and bit layout
zhuguangwen
committed
perf(cmath): sqrt table + remainder elimination; doc: decimal float and sqrt optimization
zhuguangwen
committed
fix(sqrt): use gx×recip_sqrt for initial value per doc; correct doc and remove sqrt table
zhuguangwen
committed
fix(docs): remove unused doc
zhuguangwen
committed
feat(sqrt): use SoftFloat-style two-table (k0/k1), 128 bins, 2^e→10^e remainder mapping
zhuguangwen
committed
feat(doc):rebuild
zhuguangwen
committed
refactor(sqrt): split into SoftFloat-style sqrt32/64/128 impl and shared tables, C++14 tag dispatch
zhuguangwen
committed

Commits on Jan 31, 2026

feat(sqrt): use integer remainder and 90-entry table
zhuguangwen
committed

Commits on Feb 3, 2026

fix(sqrt): align approx_recip_sqrt with SoftFloat for decimal32/64
zhuguangwen
committed
perf(sqrt): use SoftFloat-style integer arithmetic for sqrt
zhuguangwen
committed
refactor(sqrt): rename files, drop dead code, use snake_case
zhuguangwen
committed

Commits on Feb 4, 2026

Merge upstream/develop, keep our sqrt implementation
justinzhu@home
committed
perf(sqrt): optimize sqrt and remove test files for PR
justinzhu@home
committed
perf(sqrt): optimize sqrt and remove test files for PR
justinzhu@home
committed
chore: trigger CI
zhuguangwen
committed
fix(sqrt128): remove unused scale15 variable
zhuguangwen
committed
Fix -Wsign-conversion in approx_recip_sqrt64
zhuguangwen
committed
fix(sqrt): resolve MSVC C4189 unused scale16 in sqrt64_impl
zhuguangwen
committed
refactor(sqrt): use portable int128 and umul256, remove sqrt64 fallback
justinzhu@home
committed
fix(cmath): restore scale16 constant accidentally removed from sqrt64_impl
justinzhu@home
committed

Commits on Feb 5, 2026

fix(cmath): remove unused variable scale15 in sqrt128_impl.hpp
justinzhu@home
committed
perf(sqrt128): use 64×64→128 for scale33; docs: fix accuracy and deduplicate
zhuguangwen
committed
docs(sqrt): add performance benchmark section
zhuguangwen
committed
docs(sqrt): add performance benchmark section
zhuguangwen
committed
test(sqrt): improve coverage for sqrt32/64/128 implementations
zhuguangwen
committed
refactor(sqrt): remove redundant gx normalization in sqrt32/64/128 impl
zhuguangwen
committed
chore: remove accidentally committed sqrt_baseline.hpp
zhuguangwen
committed
fix(test): use double for decimal128 sqrt tests to avoid ARM64/S390X conversion issues
justinzhu@home
committed