KIonutzR - Network

"Pipeline barriers are easy.", right? Zero validation errors, everything renders correctly. Cool. But ...

"Pipeline barriers are easy.", right? Zero validation errors, everything renders correctly. Cool.
But ...

And it took less than 2 hours to get multi-window rendering with Dear ImGui with the new Vulkan backend.All it took was to take command...

And it took less than 2 hours to get multi-window rendering with Dear ImGui with the new Vulkan backend.
All it took was to take command buffers, swap chain data and so on, stick them in a struct and have a struct per window. ImGui's ImGuiPlatformIO code already called all the proper methods and all that was left to do was keep track what the active window we render to is (and keep track of deletion requests related to the window's command buffers).

Wanted to tackle performance, but the surface/window creation code for Vulkan was copied from a tutorial and seemed messy. What better way of cleaning it up than to add a hardcore requirement like "do that thing many times, independently" :D

Been a bit since the last update, but I barely got a chance to code. Between the previous post and this one, I got about a week's worth...

Been a bit since the last update, but I barely got a chance to code. Between the previous post and this one, I got about a week's worth of full-time coding on the Vulkan backend and ... I got stuff to render :D

The Vulkan backend is getting closer and closer to OpenGL (which also has some issues), but everything seems to be going through the proper steps*.
Performance is abysmal, but I'm doing things really naively. No batching of buffer updates (just the barriers), no caching of descriptor sets, no multi-threaded submissions, etc.

*Render targets: MSAA rendering and resolve, HiZ pyramid building, depth-only rendering ,etc
Compute: Clustered Forward shading, IBL convolutions, HiZ culling
Debug stuff: ImGUI, in-scene gizmos
Scene management: Buffer uploads for SSBOs, UBOs, materials, transforms, indirect buffers, etc.
Pipeline barriers: image layout changes, buffer updates, GPU-to-GPU (compute) buffer writes and reads.

Hmm, who would've though that having a newborn will significantly cut down on my coding time?Most I've managed to achieve so far...

Hmm, who would've though that having a newborn will significantly cut down on my coding time?
Most I've managed to achieve so far since the previous post is to get most of my IMGUI based editor rendering under Vulkan.

Not much, I know, but this means that shaders cross compile for OpenGL and Vulkan properly, uniforms map cleanly to descriptor sets, buffers get updated properly, etc.

Clean startup, stable runtime and clean shutdown. I'll take it.

If I manage to implement bliting for all cases (e.g. specific layer of an multisampled array texture, a specific mip level of a depth map, etc), that black screen should draw something.

The editor grid is just another element in the scene, using the regular rendering pipeline, but overlayed post-blit, so if that works, that's great news.

Got my Vulkan backend to run free of validation errors with proper asset/buffer/shader/texture/mesh management. Missing a couple of features...

Got my Vulkan backend to run free of validation errors with proper asset/buffer/shader/texture/mesh management. Missing a couple of features (e.g. render target blitting) to reach feature parity with my OpenGL AZDO backend (and to actually see stuff on screen) but I can now finally see my resources loaded, mipmapped properly, assigned to shaders and so forth.
Performance is ... not great, but I barely cache anything for now, so that's OK.
Got excited so opened a Mastodon account. The small things. 😂