Matt 2025Q2

Author: mborland

_posts/2025-07-14-Matts2025Q2Update.md

@@ -0,0 +1,48 @@
+---
+layout: post
+nav-class: dark
+categories: matt
+title: Bigger, Faster, Stronger Types
+author-id: matt
+author-name: Matt Borland
+---
+
+We continue to make exciting progress developing new libraries for inclusion in Boost, and expanding those already available.
+
+# New Libraries
+
+## int128
+
+Int128 ([https://github.com/cppalliance/int128](https://github.com/cppalliance/int128)) is a small library that has more or less fallen out of work on Decimal.
+It provides two type: an unsigned 128-bit integer and a signed 128-bit integer.
+Since my last post the library should now be ready for beta, and subsequently production use.
+Much effort was put into optimizing every operation on a multitude of architectures.
+The documentation includes bar charts showing the performance of our types vs Boost.Multiprecision, and built-in types (if available).
+While orgianlly envisioned as just an improvement to the Decimal backend arithmetic, I think this has much more additional usefulness.
+
+## Decimal
+
+Decimal ([https://github.com/cppalliance/decimal](https://github.com/cppalliance/decimal)) is a ground-up implementation of IEEE 754 Decimal Floating Point types in C++14, co-authored with Chris Kormanyos.
+In January we had our formal review for inclusion in Boost.
+As int128 above became more production ready we have integrated it into Decimal as a new backend integer type.
+Not only do we now use the int128 as a backend, we were able to find a few bugs in integration due to the special functions test suite in Decimal.
+The relationship during co-development has worked out really well.
+We also recently merged a new 256-bit integer backend based on developments and lessons learned from `int128`.
+These combined have given us >100% speedups for the 128-bit types, and also benefit the 64-bit types to a lesser degree.
+Discussions are ongoing in the Cpplang Slack channel `#boost-decimal`.
+
+# Existing Libraries
+
+## Math
+
+As posted a few versions ago Boost.Math began offering support to be run on GPU with NVCC, NVRTC, and SYCL.
+We have recieved a few bug reports now that this functionallity is being used more.
+For Boost 1.89 we have put good effort into fixes and internal restructuring to address the issues that people have been having.
+
+## Multiprecision
+
+For the first time in a while Boost.Multiprecision has a new backend type: `cpp_double_fp_backend`.
+This project was started during GSoC 2021 and has finally come to full fruition.
+The `cpp_double_fp_backend` back-end is the sum of two IEEE floating-point numbers combined to create a type having roughly twice the composite width of one of its parts.
+The `cpp_double_fp_backend` back-end is used in conjunction with `number` and acts as an entirely C++ header only floating-point number type.
+If you need more precision than a `double` with less computational expense than arbitrary precision types look out for this in Boost 1.89.