Intel Wires Up Dual-SIMD8 Dispatch For Mesa Drivers

[coder]

ANV was developed by third party.

Easy to say, now let's see some proof.

they still have to develop their amdvlk, just to have fallback implementation, because they can't rely on mesa implementation quality.

What are some things it does that Mesa does slower, or not at all?

[Khrundel]

Easy to say, now let's see some proof.

LOL. https://www.phoronix.com/news/LunarG-Vulkan-AMA

What are some things it does that Mesa does slower, or not at all?

You are literally writing this comment under the post about example of such "slower" case. You may remember how much time it took to implement RDNA RT. But you still asking for examples because you can't face reality. How they say... "feelings don't care about your facts"?

And in reality you don't need any example. You can easily browse phoronix for news like "feature X support was merged to mesa" and calculate lag between hardware availability and release date of this mesa version, but it would be useless info. Neither you nor I can for example understand how important is wave32 support for one of RDNA GPUs or is NGG supports works in later GPUs only because it is broken at hardware level or radv developers just unable to find correct cases where it can be used? So what is the point in this game? You can just look at most competent party here: GPU producer. AMD do not want to drop their AMDVLK. This means they know how badly mesa supports their GPUs.

[Svyatko]

You can just look at most competent party here: GPU producer. AMD do not want to drop their AMDVLK. This means they know how badly mesa supports their GPUs.

Really?
Intel fails: https://www.phoronix.com/news/Torval...-Linux-6.8-DRM
AMD fails: https://bugzilla.opensuse.org/show_bug.cgi?id=1168611
I've created these instructions to overcome AMD's ignorance of users needs for GPGPU on Linux: https://en.opensuse.org/SDB:AMD_GPGPU