VK_EXT_descriptor_heap proposal is a good read.
vulkan has come a long way, to the point that if you use all the modern improvements, it looks almost nothing like what came out in 2016.
dynamic rendering and unified image layouts were my favorites as obviating the most annoying (and often pointless on modern hw) parts of the api. this one is the best though, i've always hated descriptor sets.
https://docs.vulkan.org/features/latest/features/proposals/VK_EXT_descriptor_heap.html
@pythno @nicebyte the Vulkan Guide is okay-ish and tries to keep uptodate with the core API and deprecations: https://docs.vulkan.org/guide/latest/index.html (better than any OpenGL documentation anyway).
For extensions it's the same mess though. The actual extension text is usually extremely sparse (and all in all useless), but sometimes there's the proposal text linked which is much more useful).
@pythno @nicebyte in sokol-gfx I basically use:
- Vulkan 1.3 core API features
- "synchronization2"
- dynamic rendering (e.g. Metal-style transient render passes instead of baked pass objects)
- VK_EXT_descriptor_buffer (eventually to be replaced with VK_EXT_descriptor_heap)
...the biggest surprise so far was how poorly VK_[EXT|KHR]_swapchain_maintenance1 is supported. It reads like it's a natural cleanup extension that everybody supports, but coverage is below 30% and not worth the hassle.
@pythno @nicebyte That's https://www.howtovulkan.com
It's supposed to show how to do Vulkan rasterization in 2026.
nicebyte
Michael Eggers 🇺🇦🇪🇺
Andre Weissflog
Sascha Willems