Universal v3.95

Universal Numbers Library v3.95

What's New

Changes since v3.94:

• V3.95: Block formats and CI compile caches (#502)

Installation

Universal is a header-only C++ template library. To use:

find_package(UNIVERSAL CONFIG REQUIRED)
target_link_libraries(your_target universal::universal)

Or simply add include/sw to your include path.

Requirements

• C++20 compatible compiler (GCC 10+, Clang 12+, MSVC 2019+)
• CMake 3.16+

What's Changed

• V3.95: Block formats and CI compile caches by @Ravenwater in #502

Full Changelog: v3.94...v3.95

Universal v3.94

Universal Numbers Library v3.94

What's New

Changes since v3.93:

• V3.94: interval arithmetic for mixed-precision algorithm SDK (#501)

Installation

Universal is a header-only C++ template library. To use:

find_package(UNIVERSAL CONFIG REQUIRED)
target_link_libraries(your_target universal::universal)

Or simply add include/sw to your include path.

Requirements

• C++20 compatible compiler (GCC 10+, Clang 12+, MSVC 2019+)
• CMake 3.16+

What's Changed

• V3.94: interval arithmetic for mixed-precision algorithm SDK by @Ravenwater in #501

Full Changelog: v3.93...v3.94

Universal v3.93

Universal Numbers Library v3.93

What's New

Changes since v3.92:

• V3.93: UBSan fixes, image demo, and coverage workflow (#499)

Installation

Universal is a header-only C++ template library. To use:

find_package(UNIVERSAL CONFIG REQUIRED)
target_link_libraries(your_target universal::universal)

Or simply add include/sw to your include path.

Requirements

• C++20 compatible compiler (GCC 10+, Clang 12+, MSVC 2019+)
• CMake 3.16+

What's Changed

• V3.93: UBSan fixes, image demo, and coverage workflow by @Ravenwater in #499

Full Changelog: v3.92...v3.93

Universal v3.92

Universal Numbers Library v3.92

What's New

Changes since v3.91:

• V3.92: process improvements (#497)

Installation

Universal is a header-only C++ template library. To use:

find_package(UNIVERSAL CONFIG REQUIRED)
target_link_libraries(your_target universal::universal)

Or simply add include/sw to your include path.

Requirements

• C++20 compatible compiler (GCC 10+, Clang 12+, MSVC 2019+)
• CMake 3.16+

What's Changed

• V3.92: process improvements by @Ravenwater in #497

Full Changelog: v3.91...v3.92

v3.91: quality upgrade

V3.91 (#495)

* Incrementing SEMVER to v3.91

* RCA'ed the GCC array-bound warnings to be a GCC bug: mitigating with pragma guards

* unicode clean up

* Incrementing SEMVER to v3.91

* RCA'ed the GCC array-bound warnings to be a GCC bug: mitigating with pragma guards

* unicode clean up

* [#490] regression fixes for Apple CLang

* WIP: adding complex<> library to deal with non-native floating-point types

* change and session logs for Jan 11th, 2026

* WIP: posit float conversion debug utilities

* WIP: resurrecting the docker builder and release container automation

* Clang x86 test failure fixes (#491)

* Add temporary Clang x86 workaround toggle for positRegime and bfloat16 cos

* Clang on x86: CI fixes - PR update

* Fixes for several compiler warnings. (#493)

* UBSan / ASan fixes uncovered by intermittent test failures (#492)

* Fixes for several undefined behavior issues caught by UBSan

* UBSan fix: *WIP* Grisu gfp broken (Negative array index fix)

* [Codex]Fixes for several undefined behavior and undefined address issues caught by UBSan and UASan

These fixes were made by OpenAI Codex and could probably use additional review

* fixing incorrect suggestion

if (v == -v) finds the special case 0x100...000
if (v == numeric_limits<Arith>::min()) finds 0 as Arith == bool

Signed-off-by: Theodore Omtzigt <theo@stillwater-sc.com>

---------

Signed-off-by: Theodore Omtzigt <theo@stillwater-sc.com>
Co-authored-by: Theodore Omtzigt <theo@stillwater-sc.com>

* Making all Dockerfiles consistent again with current build targets

* code hygiene and operationalizing the sanitizers

* Listen to the tools when they tell you something: UB fixed

* Adding x86/Mac to the platform matrix

* adding a x86 Linux with Clang

* macos-13 is deprecated, moving to macos-15-intel

* Enhancing the labeling of the CI jobs

---------

Signed-off-by: Theodore Omtzigt <theo@stillwater-sc.com>
Co-authored-by: Loren Osborn <linux_dr@yahoo.com>

v3.80: double-double and quad-double updates

Fleshed out double-double and quad-double types.

New math lib structure that is modeled after Boost.MATH but optimized for mixed-precision algorithm design and optimization.

v3.79: quad-double type

A new quad-double floating-point type and math library.

The quad-double is a segmented floating-point type consisting of four (4) doubles capturing 212 bits of precision.

v3.78 double-double type

v3.78 presents a new double-double type

v3.75: new posit infrastructure

added a fast posit<16,2> and restructured the regression environment to support more of these specialized posit implementations.

v3.74: new posit R&D

Added a new fast standard posit<16,2>