Another Fun Day In Mesa Git: RADV, NVC0, RadeonSI

[chrisr]

Apitrace fails to trace Dying Light. ...
It looks like this game isn't going to work unless the game developers update the game (alternatively: unless Mesa implements the compatibility mode).

Have you tried exporting "MESA_GL_VERSION_OVERRIDE=4.5COMPAT" and "MESA_GLSL_VERSION_OVERRIDE=450" instead?

[pal666]

I'd like to hear about a clarification and any exceptions for open-source drivers from Khronos as well

opensource drivers should fix their bugs like everyone else, what else you like me to clarify?

[pal666]

Does radeonsi support "GL_ARB_enhanced_layouts: specified vec4-slot component numbers" yet? The features.txt file still says "No" while also claiming that GLSL 4.40 is "DONE" for radeonsi.

yes, and features.txt says yes

[osmotr]

dying light doesn't detect video memory "[OpenGL] Video memory detected: 0 [MB]!"

[Detructor]

Question: What would (legally) happen if someone would just change the supported OpenGL version to 4.5 for radeonsi without contacting Khronos or going through with these conformance tests?

[kmare]

Mostly, yes. Of course the driver quality isnt as high and random GPU crashes can happen or random bugs, but nouveau was able to run bioshock infinite a week after release without any problems except maybe performance, but it was playable in a decent GPU.

Just OpnGL threaded applications are a big issue currently.

Karolherbst and the rest of the nouveau developers, you're doing an amazing job.. seriously... thank you for the awesome job you're doing.

[chrisr]

yes, and features.txt says yes

Really?

GL 4.4, GLSL 4.40 -- all DONE: i965/gen8+, nvc0, radeonsi

  GL_MAX_VERTEX_ATTRIB_STRIDE                           DONE (all drivers)
  GL_ARB_buffer_storage                                 DONE (i965, nv50, r600)
  GL_ARB_clear_texture                                  DONE (i965, nv50, r600)
  GL_ARB_enhanced_layouts                               DONE (i965, nv50, llvmpipe, softpipe)
  - compile-time constant expressions                   DONE
  - explicit byte offsets for blocks                    DONE
  - forced alignment within blocks                      DONE
  - specified vec4-slot component numbers               DONE (i965, nv50, llvmpipe, softpipe)
...

[FLHerne]

opensource drivers should fix their bugs like everyone else, what else you like me to clarify?

Exposing GL4.5 requires passing the conformance tests. Khronos charge tens of thousands of dollars per GL(ES) version [1] just for access to those tests. It's nothing to do with 'fixing bugs', it's the FOSS community not having $200,000 spare for the privilege of providing an open implementation of Khronos' specs.

[1] https://www.khronos.org/conformance/adopters/

[baka0815]

Really?

GL 4.4, GLSL 4.40 -- all DONE: i965/gen8+, nvc0, radeonsi
...

Yes, notice the "all DONE" part at the top.

mesamatrix.net isn't showing radeonsi as green, maybe a bug on the site?

[pal666]

Exposing GL4.5 requires passing the conformance tests. Khronos charge tens of thousands of dollars per GL(ES) version [1] just for access to those tests. It's nothing to do with 'fixing bugs', it's the FOSS community not having $200,000 spare for the privilege of providing an open implementation of Khronos' specs.

your post has nothing to do with reality, since radeonsi is developed by amd and radhat, both of which has access to cts. the only problem is someone has to fix all test failures ffs