RadeonSI Gallium3D Receives Some OpenGL 4.5 Conformance Fixes

[Herem]

Yes, very nice. However, even if the "Starter Edition" existed at the time of my anecdote, it doesn't change the fact that the WoW client is a terrible test case because its code is constantly changing in ways that the Mesa developers have no control over. This makes it useless for testing Mesa for regressions.

For example, I was having rendering problems with my Radeon R200 and rv350. But the latest WoW client requires a HD5670 at the very least.

The starter edition was introduced over 5 years ago and prior to that Blizzard used to run 10 day free trials, that started not long after the game left beta.

WoW certainly receives constant content patches but the rendering engine rarely changes outside of major expansion's which only happen on average once every couple of years. AMD and Nvidia have both previously issued plenty of patches to their closed drivers for WoW, so it doesn't seem as if it is too difficult a task.

[chrisr]

The starter edition was introduced over 5 years ago and prior to that Blizzard used to run 10 day free trials, that started not long after the game left beta.

And at the time, I believe the Mesa developers took full advantage of the 10 day trial period and were then stuck at the login screen.

But that is irrelevant. That particular login screen no longer exists and the game can no longer even run on the hardware concerned! So any regression tests for that hardware based on the WoW client would be completely worthless now.

Geez.

[boltronics]

If you can compile Mesa, many can not. Compilation simply fails in Debian testing for example.

I successfully run Mesa from git master, and I'm on Debian stretch. There's no issue, aside from having to compile a newer version of LLVM first.

[ua=42]

Public service announcement: The ignore list appears to be working again. So if you add trolls to it, you won't see their posts ever again.

[M@yeulC]

Speaking as a developer, I would consider a game to be the worst kind of test. For example, when I once tried to demonstrate rendering errors in World of Warcraft to Mesa developers, but none of them had a WoW subscription. The only problems they were able to reproduce and fix easily were those ones visible on the "login" screen. However, that particular login screen for WoW no longer even exists because it changes after each game expansion pack.

I would consider the best tests to be under the developers' control, and testing well defined areas of Mesa.

Well, that's exactly where tools such as apitrace can be useful. You record a few frames, send it to developers, and they can see the output, and replay the frames call by call.
Moreover, I saw at least one commit (I remember of the "bioshock" performance patches: https://cgit.freedesktop.org/mesa/me...2f412904f944a3) where apitrace was mentioned for benchmarking performance speedups. So, you can even time the frame, and/or use it as a test case when you fix a bug

Traces can be quite big, though, IIRC.

[Marc Driftmeyer]

Finally! OpenCL 1.2 support in Mesa for CLOVER:

clover: add missing clGetDeviceInfo CL1.2 queries

https://lists.freedesktop.org/archives/mesa-commit/2016-November/068305.html

clover: Add CL_PROGRAM_BINARY_TYPE support (CL1.2). v3 [Francisco Jerez]: Loosely based on Serge's v1 of this patch in order to avoid CL-specific enums in the clover module binary format. In addition to other changes made in v2: Represent the CL program binary type as the section type instead of adding a CL API-specific enum, check that the binary types of the input objects are valid during clLinkProgram(), pass section type as argument to build_module_library() instead of using separate function.

https://lists.freedesktop.org/archives/mesa-commit/2016-November/068306.html

[mibo]

Maybe this link is outdated?

https://dri.freedesktop.org/wiki/GalliumCompute/

AFAIK without image read functions clover can not accelerate darktable. I'm surprised that version 1.2 doesn't require such essential function.