Mesa OpenGL Threading Work Sees Much Reduced Memory Footprint For OpenGL Calls

[phoronix]

Phoronix: Mesa OpenGL Threading Work Sees Much Reduced Memory Footprint For OpenGL Calls

Longtime AMD open-source Mesa developer Marek Olšák after more than one decade working officially for AMD and years before that as an independent open-source contributor going back to the R300g days still has not run out of new performance optimizations to pursue. The most recent accomplishment for this leading Mesa contributor are some refinements to the OpenGL threading "glthread" code for lowering the memory footprint...

https://www.phoronix.com/news/Mesa-GL-Thread-Lower-Footprint

[user1]

Is glthread currently disabled by default on RadeonSi? I remember it was enabled in late 2022 but was disabled again shortly after.

[zboszor]

Is glthread currently disabled by default on RadeonSi? I remember it was enabled in late 2022 but was disabled again shortly after.

It was never disabled in master as further fixes made it unnecessary. It was only disabled for the 22.3 release series.

radeonsi has it enabled by default. zink, asahi and maybe some others as well.

[bug77]

Back when Vulkan was in its infancy and it was believed it will cure cancer and everything, there was a presentation floating around saying OpenGL itself can trim fat to become nimble to the point of rivaling Vulkan. Of course, OpenGL won't do everything Vulkan can, but this improvement right here says that presentation was at least partially right.

[user1]

Back when Vulkan was in its infancy and it was believed it will cure cancer and everything, there was a presentation floating around saying OpenGL itself can trim fat to become nimble to the point of rivaling Vulkan. Of course, OpenGL won't do everything Vulkan can, but this improvement right here says that presentation was at least partially right.

this one? Although it was at least 2 years before Vulkan was released, so idk if Vulkan was even known back then,

[cb88]

this one? Although it was at least 2 years before Vulkan was released, so idk if Vulkan was even known back then,

Mantle existed and is from a high level very similar. The PS3/PS4 graphics apis also existed and were probably one of the main motivations behind all this because GL was becoming uncompetitive.

[rmfx]

Back when Vulkan was in its infancy and it was believed it will cure cancer and everything, there was a presentation floating around saying OpenGL itself can trim fat to become nimble to the point of rivaling Vulkan. Of course, OpenGL won't do everything Vulkan can, but this improvement right here says that presentation was at least partially right.

The advantage of Vulkan lies not only in performance but also in simplifying drivers and their maintenance, by leveraging complexity.
Additionally, it established a new standard built upon fresh modern foundations.

[bearoso]

The advantage of Vulkan lies not only in performance but also in simplifying drivers and their maintenance, by leveraging complexity.
Additionally, it established a new standard built upon fresh modern foundations.

From my experience, the advantage of Vulkan is that it's stateless. Even OpenGL DSA has some global state. But removing that was the primary goal for any true OpenGL successor. Without state you can do a lot more threading. Making threading first-class meant Vulkan's synchronization had to become manual. They added the render pass stuff to cater to to mobile, and that's how we ended up with the complexity we got with Vulkan 1.0.

I'm glad they're backpedalling a bit on the mobile focus with dynamic rendering and shader objects. The extreme verbosity has to stay because, with state attached to objects, it has to all be defined, but Vulkan is now and will be easier to use in the future.

[mxan]

Mantle existed and is from a high level very similar. The PS3/PS4 graphics apis also existed and were probably one of the main motivations behind all this because GL was becoming uncompetitive.

Mantle isn't just "similar", it's the direct predecessor to Vulkan. AMD donated it to Khronos.

[cb88]

Mantle isn't just "similar", it's the direct predecessor to Vulkan. AMD donated it to Khronos.

Firstly chill. Vulkan and mantle are different enough that similar is the right word. If you acutally check the Mantle API reference you'll see almost nothing from Vulkan was directly copied into mantle without at least changing the name of it to VK something even basic stuff like GPU enumeration has a completely different name.

So the things that are there, have completely different API calls even if they are functionally identical.