# \[ANN\] geomancy-0.4 -- GPU-friendly types and fast operations

**URL:** <https://discourse.haskell.org/t/ann-geomancy-0-4-gpu-friendly-types-and-fast-operations/15076>\
**Category:** Announcements\
**Created:** [November 12, 2026, 7:32pm UTC](https://discourse.haskell.org/t/ann-geomancy-0-4-gpu-friendly-types-and-fast-operations/15076 "2026-11-12T19:32:14Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![wiz](https://sea2.discourse-cdn.com/flex002/user_avatar/discourse.haskell.org/wiz/32/2408_2.png) [@wiz](https://discourse.haskell.org/u/wiz)\
**Post date:** [November 12, 2026, 7:32pm UTC](https://discourse.haskell.org/t/ann-geomancy-0-4-gpu-friendly-types-and-fast-operations/15076/1 "2026-11-12T19:32:14Z")

</div>

> **[Changelog for geomancy-0.4.0.0 | Hackage](https://hackage.haskell.org/package/geomancy-0.4.0.0/changelog)**

A few highlights:

- Double-based vectors and dmat4 (for your very large worlds)
- ~~OverloadedRecordForth~~ Element accessors and expanding/shuffling/narrowing swizzles (`magenta = blue.xyz1.wyz`) with rewrite rules to do the right thing.
- NaN-detecting pointwise Eq and Ord (don’t use floats as map keys please)
- Fixes for silly errors in ops for rare types.
- Letting specialized CPU hardware doing its job instead of pre-aligning matrix buffers in software (a bit faster now, and Storable buffers are still aligned).
