RADV Prepares To Switch Completely To Dynamic Rendering

[schmidtbag]

So... there must be some downside because if it's easier on developers, produces fewer bugs, and can potentially increase performance, then why wasn't this the default for all Vulkan drivers from the beginning?

[F.Ultra]

So... there must be some downside because if it's easier on developers, produces fewer bugs, and can potentially increase performance, then why wasn't this the default for all Vulkan drivers from the beginning?

Looking at the large patch needed for this I would say that it perhaps makes life easier for the user dev but more complicated for the driver dev. OR the idea came long after Vulkan was specified.

[skeevy420]

Looking at the large patch needed for this I would say that it perhaps makes life easier for the user dev but more complicated for the driver dev. OR the idea came long after Vulkan was specified.

It's the latter. Came with Vulkan 1.3. It's almost a year old.

[smitty3268]

and to people not so involved in Vulkan and mesa what does it mean in practice ? like better performance or less code and less bugs ?

The extension proposal is worth a read for anyone interested. https://github.com/KhronosGroup/Vulk...ering.asciidoc

In short, it sounds like developers really didn't like the render pass API in Vulkan, and this is an attempt at replacing it with something much more streamlined and familiar to them.

There are also some comments and people involved that make me think it might be helpful for performance for DXVK/Zink and other projects that map alternative APIs onto Vulkan.