Shader Objects

Yep, it’s all merged. That means if your driver supports VK_EXT_shader_object, you can finally enjoy Tomb Raider (2013) without any issues.

NVIDIA has just released a new beta driver that I need to test, but I’m hopeful it will no longer crash when trying to use this extension.

I Didn’t Want To Join Your Club Anyway

Remember that time I mentioned how zink wasn’t allowed to use VK_EXT_vertex_input_dynamic_state on AMDVLK?

We all got a chuckle, and it was sad/funny, and nobody was surprised, but the funnier part is some code that’s been in zink for much longer. Almost a year, in fact:

if (screen->info.have_EXT_extended_dynamic_state) {
   if (screen->info.have_EXT_extended_dynamic_state2) {
      if (screen->info.have_EXT_extended_dynamic_state3) {
         if (screen->info.have_EXT_vertex_input_dynamic_state)
            dynamic = ZINK_DYNAMIC_VERTEX_INPUT;
         else
            dynamic = ZINK_DYNAMIC_STATE3;
      } else {
         if (screen->info.have_EXT_vertex_input_dynamic_state)
            dynamic = ZINK_DYNAMIC_VERTEX_INPUT2;
         else
            dynamic = ZINK_DYNAMIC_STATE2;
      }
   } else {
      dynamic = ZINK_DYNAMIC_STATE;
   }
} else {
   dynamic = ZINK_NO_DYNAMIC_STATE;
}

This is the conditional for enabling dynamic state usage in zink using an enum. As we can see, the only time VK_EXT_vertex_input_dynamic_state is enabled is if either VK_EXT_extended_dynamic_state2 or VK_EXT_extended_dynamic_state3 are also enabled. This cuts down on the number of codepaths that can be used by drivers, which improves performance and debuggability.

AMD’s drivers don’t yet support VK_EXT_extended_dynamic_state3, as anyone can see from gpuinfo. They do, however, support VK_EXT_extended_dynamic_state2, so the driver-side disablement of VK_EXT_vertex_input_dynamic_state does have some effect.

Not.

Way, way back, just over a year ago, I was doing some testing on AMD’s drivers. One thing I noticed was that trying to run any of the GLES CTS caselists on these drivers caused GPU hangs, so I stopped running those, leaving me with just GL4.6 CTS.

And what happened when I enabled VK_EXT_extended_dynamic_state2 there, you ask? Test failures. Lots of test failures.

Thus, AMD got the cone of shame: a driver workaround to explicitly disable this extension.

In conclusion, we all had a good chuckle about AMD blocking zink from using VK_EXT_vertex_input_dynamic_state, but…

Well, there’s nothing in this story that we didn’t already expect.

Tracking

On another topic, I’ve been doing some per-app tracking in zink. Specifically tracking games that don’t work great. If you know of other games with issues, post them there.

But there hasn’t been any sort of canonical list for games that do work great on zink, which leads to a lot of confusion about how useful it is for gaming.

Thus, I’ve created the GAMES THAT WORK tracking ticket. If you’ve played a game on zink and it works, post about it. If you want to know whether a game works, check that ticket and maybe it’ll be updated enough to be useful.

Remember: I don’t play video games, so I can’t fill out this table on my own. The only way I know if a game works is if I spend a comprehensive amount of time benchmarking or debugging it, which is the only reason I have 300+ hours logged in the benchmark mode of Tomb Raider (2013).

Upcoming

Things have been quiet on the surface lately, with nothing big in development.

But there is big work underway. It’s an extremely secret project (not Half Life 3) that (or Portal 3) I’m hoping (or any Valve-related title) can be (it’s not even a game) brought into the light (it’s driver-related) within the next week (involving features) or two (and perf).

I can’t say more on this topic.

Don’t even bother asking.

It’s too secret.

What I can say is that it’s been in development for almost a month. And we all know how much time a month is when it comes to SGC speed.