Esta madrugada salió nueva versión de #freedesktop 25.08. La decimotercera revisión.
Aquí, en mi caso, es relevante por traer #Mesa3D 26.0.8 que viene con correcciones para #ForzaHorizon6. No lo voy a comprar. Pero bueno, por si alguien quiere. Es buen momento.
https://gitlab.com/freedesktop-sdk/freedesktop-sdk/-/releases/freedesktop-sdk-25.08.13
Los pichiquitas de #freedesktop se podrían plantear estar más al día en ciertos componentes como #Mesa3D. Tener que esperar unas tres o seis semanas tiene cojones.
Flatpak "deduplica" y da igual si sacan una actualización que solo modifica Mesa. Además, ya mismo, Mesa va a tener como una rama con correcciones / arreglos para juegos específicos.
@thekhronosgroup is there an easy option for the general public to have access to issues in the Khronos issue tracker for #Vulkan ? Within #Mesa3d I always stumble across links like https://gitlab.khronos.org/Tracker/vk-gl-cts/-/issues/6580 that I can't view.
(Yes, some stuff needs to be internal, but GitLab does allow internal notes/comments, why not use that and push the "normal" stuff out into the open?)
@austriancoder Wow, great to hear! Although I do not understand all the technical details of what you are working on, I'm following your work with high interest. And I'm not the only one :-)
https://forums.puri.sm/t/gles-3-0-for-librem-5/23919
#etnaviv #mesa3d #Librem5 #mntreform #gpu #opensource #mobilelinux #FOSS
I am transforming a ~7month old wip branch into something upstream-able. Lot more GLES3 tests are passing with these commits. But I need to spend some more hours to test 2-3 different GPU models, before I will create a MR.
dEQP-GLES31.functional.image_load_store.* 699/747 pass
With the help of some freshly reverse-engineered ISA opcodes :)
It’s always a pain to emulate missing hardware features.
dEQP-GLES3, old etnaviv compiler vs new one:
+6140 passes, −3004 crashes. 🚀
I built infrastructure to validate ISA-specific behavior on hardware against a `Foldable` CPU reference - the idea is straight out of NAK by @gfxstrand . Write the per-opcode reference once, run a matrix of cells (src mods × swizzles × write masks) on the GPU and compare.
It saved me so much guesswork about how each ISA instruction behaves with input values and rounding modes - hello subnormal IEEE 754 pain.
I should have started this earlier.
RenézuCode
GNU/Trinukso ۞
Kai Wasserbäch
Jan Vlug
Christian Gmeiner