almosst-dev/cobalt64 is a project that natively accelerates AMD Radeon GPUs that are not officially supported on macOS. The first supported family is AMD's RDNA4 generation GPUs (RX 9070 XT / Navi 48), consisting of a kernel driver, a macOS (Darwin) port of RADV, and a Metal-on-Vulkan bundle. The commit d1f7bad (https://github.com/Almosst-DEV/Cobalt64/commit/d1f7bad8408a0a2c016b067c477422975a59de44) at October 15, 2026 22:17 UTC changed only a single line in tools/ioaccel-layout/README.md. The sentence 'This repository contains only the scripts' was changed to 'This directory contains only the scripts', so that the scope of the accompanying note — that this directory contains no Apple binaries or symbol tables, no vtable dumps, and no generated headers — narrowed from the entire repository to that directory. The change statistics are 1 insertion and 1 deletion.
tools/ioaccel-layout is a set of scripts that reconstruct the C++ vtable layout and instance sizes used by Apple's IOAccelerator family. The README explains that these scripts analyze the Apple software installed on the machine to export headers, and that subclasses built with -fapple-kext can reproduce the same layout without Apple's non-public headers. The same document firmly warns not to keep the produced outputs only inside the machine and not to redistribute them. In other words, this commit is not about code or behavior, but a cleanup that accurately describes the location of the scripts and the scope of use of their outputs in the documentation.
The limitations are clear. This commit changed only one line of documentation, and there is no trace of it touching code on the kernel driver or RADV port side. There is no record of verification or benchmarks in this material as to how accurate the vtable layout and instance sizes described by the README are. License and interest figures are also not presented. Reproducing the IOAccelerator layout as-is is the part that governs compatibility with the macOS side, so this one-line fix should not be read as meaning that driver behavior has changed. To sum up, it is strongly reminiscent of the early-stage work of attaching unsupported AMD GPUs to macOS, and this commit comes close to merely correcting the description of that boundary.