Eglantine nx2 build 177

Eglantine nx2 build 177 is released. Eglantine is a free and opensource (FOSS) Switch 2 emulator, derived from Eden, Yuzu and Sudachi - started by developer Camille LaVey. It's written in C++ with portability in mind, with builds for Windows, Linux and macOS.

Eglantine Changelog:
This is build 177 of nx2 (46967a1): three changes on top of build 175 — an Arch Linux package, a contributing guide that points at this repository, and less per-draw work on the GPU thread in crowded scenes.
Nothing needs configuring: the performance change has no switch, and the Arch package is opt-in.
Build 176 was a CI run cancelled when a newer commit arrived and was never published; this build carries all three changes.
Arch Linux: an eglantine-git package
dist/packaging/arch/eglantine-git/ now holds a PKGBUILD that builds Eglantine from the nx2 branch and installs it as eglantine-git, with both the GUI and eden-cli. It links the system's Qt 6, SDL3 and FFmpeg instead of bundled copies, uses the project's own install rules, disables the built-in updater because the package manager owns updates, and lists qt6-wayland as an optional dependency for native Wayland.
It is in this repository, not in the AUR. It has not been built or run end to end on Arch yet — bash -n PKGBUILD and git diff --check pass, namcap has not been run — so treat it as something to test and report back on rather than a finished package.
Crowded scenes: less per-draw work on the GPU thread
In a Fire Emblem: Fortune's Weave city scene — about 2,100 draws a frame — two pieces of per-draw bookkeeping were a measurable share of the GPUBack thread:
The texture cache's "invalid sampler index" and "invalid image view index" messages fire for unused binding slots on every draw. They were LOG_DEBUG, which is filtered from the console but still costs a call and argument packing per message: 3.1% of the thread's samples. They are LOG_TRACE now, which release builds compile out.
DescriptorTable::Read reads one 32-byte TIC/TSC entry per texture and sampler of every draw through the generic page walk: 5.0% of the thread's samples. An entry that sits within one page is contiguous in host memory, so it is now one page lookup and a memcpy. Entries that cross a page, and pages with no host memory, still take the block read and still read as zero.
The same change in another Eden-based build cut those reads from 6.4% to 0.4% of its GPU thread. This is CPU-side work, so expect it in frame time in draw-heavy scenes rather than in GPU utilisation.
The contributing guide points here now
CONTRIBUTING.md and the pull request template still sent people to Eden's Forgejo and its account-registration instructions. They now describe the flow that applies here: fork Djihads80/Eglantine, branch from nx2, follow the AI and coding policies in docs/policies/, and open a pull request on this repository. The donation links and the Forgejo/Codeberg/email submission routes are gone, and the Discord invite and issue tracker are this project's.
Coming from build 175?
Nothing is missing.
Thanks to @FranzQDH (the Arch package, #323), @Mcm190 (the contributing guide, #382) and @Duck912 (the per-draw work, #374).

Download: Eglantine nx2 build 177
Source: Here