Hi,

Just wanted to share a quick update on how reMESA is progressing.
The project has made quite a bit of progress on the SGX535 side. I've been
working on the Mesa/Gallium integration, NIR-to-USSE shader compilation,
resource management, and the rendering and presentation paths
On the Dell Inspiron Mini 12, I've successfully executed GPU rendering
tests, including framebuffer readback validation and sampled-texture
workloads. I've also validated a hybrid 2D path combining CPU operations
with GPU-accelerated copies;
The main remaining challenge is getting the normal OpenGL application path
fully working. The experimental rendering paths have produced valid results
on real hardware, but continuous rendering, normal context creation, and
desktop presentation still need further qualification
I've also published an experimental alpha release to make the current
source and development state available. It's not yet a general-purpose
accelerated driver, and I don't want to present it as one.
I'm continuing to work toward a usable Mesa/Gallium implementation, with
the goal of eventually presenting the kernel and userspace sides together,
as suggested in our earlier discussion, I intend to clean up the formatting
code, etc., for MESA, but I see that I might have serious problems using
AI, the provenance shouldn't have any serious problems; I retrieved
everything from open repositories from TI, and reverse-engineered binaries
like xpsb-glx and psb_dri.so

Regarding the AI, I managed to prevent several things before it reached the
repository. I identified and fixed a UAF vulnerability (it literally
generated the UAF code right in front of me), as well as a repository I
call the "bomb repository" full of leaked files, which, if I hadn't been
there, would be in the repository today, None of this content was
consulted; regarding UAF, I reviewed it directly using expand code in the
CLI I use.

I wanted to know about Mesa and its AI restrictions. The best thing is that
Remesa doesn't have its own DRM, but it can have problems precisely because
previous reverse engineering repositories have AI, and almost all of Remesa
code is generated with generative AI (but still revised by me at the time).

I created a provenance file in my repository called PROVENANCE.md where I
put almost all the files I've used. I'll search for more files that I
potentially used; the documentation has many JSON files and such that say
this, but they are obscured by newer documents.

One thing I would greatly appreciate knowing about the use of AI, and
whether this might prevent integration into the MESA. I've been concerned
about this because of past actions of RE projects, blocking AIs, This
really worries me.

Just wanted to keep this discussion updated on the progress. Thanks again
for the feedback and guidance

ty,
gabriel

Em sex., 13 de nov. de 2026 09:30, Gabriel Gama <
[email protected]> escreveu:

> The old 2D acceleration code can rest easy; I'm not going to touch it.
> I've been working on my own hybrid 2D approach instead, with good results.
> The CPU was indeed faster for fills and console scrolling, while the GPU
> performed better for large framebuffer copies.
>
> Ty,
>
> Gabriel
>

Reply via email to