Vulkan implementation MinImageCount >= 2

[tim-rex]

Version/Branch of Dear ImGui:

Version: 1.76 WIP
Branch: master

Back-end/Renderer/Compiler/OS

Back-ends: imgui_impl_vulkan.cpp
Operating System: Windows 10

Question:

I'm new to Vulkan (and swapchains) so I apologise if the answer to this should be obvious (and I'm already starting to think maybe the answer is obvious).

I'm using the provided vulkan implementation in my own project (imgui_impl_vulkan.cpp).
When initialising (via ImGui_ImplVulkan_Init) I note that the ImGui_ImplVulkan_InitInfo struct expects both an ImageCount and MinImageCount.

My question is...

Is the intended usage of MinImageCount supposed to represent the minimum number of images supported by the surface that supports the swapchain? Or should it represent the minImageCount of the swapchain itself?

I ask because I encountered an issue while using VK_PRESENT_MODE_FIFO_KHR
When using this mode, the surface that supports the swapchain has minImageCount==1_.
However, I'm still creating the swapchain itself with minImageCount==2.

ImGui_ImplVulkan_Init() asserts to ensure MinImageCount is >= 2 so I presume I should be using the swapchain minImageCount and that should be that.

As an aside... It's not clear to me when minImageCount would not be equal to ImageCount - but I suspect the answer to that is entirely unrelated to Dear ImGui. Please disregard if that is the case.

[ocornut]

I'm not entirely sure but I think if you look at the docking branch, where the back-end has support for multi-viewports, you'll get a fuller understanding of why it is like that.
Happy you hear your thoughts and what you find. If there's anything we can clarify in the back-ends (with renaming or extra comments) we can work on that too.

[tim-rex]

Thanks Omar.
I've had a closer look, and I understand why we need at least 2 images in the swapChain regardless.

I've confirmed for myself that this field should probably represent the minImageCount of the swapchain itself (and not the surface used to create it) and this makes perfect sense.

I say probably because it seems that for external usage, info->MinImageCount is unreferenced outside of the assert. Only info->ImageCount is used (I've confirmed this also on the viewport branch).

Looking closer at the viewport branch, there are situations when info->MinImageCount != info->ImageCount might need to be handled, but it appears this situation is only relevant to the example (not for external usage).

So I'm thinking that instead of:

    IM_ASSERT(info->MinImageCount >= 2);
    IM_ASSERT(info->ImageCount >= info->MinImageCount);

perhaps it makes more sense as:

    IM_ASSERT(info->ImageCount >= 2);
    IM_ASSERT(info->ImageCount >= info->MinImageCount);

And perhaps indicate via a comment that info->MinImageCount is for internal usage only..
Or mandate that both fields are always >= 2, to allow for future usage of info->MinImageCount

[CarloWood]

ImageCount seems to be only used as "Frame Resource Count". That is, resources (like vertex buffers), bound to a command buffer and that need to be left alone until that command buffer signals its fence that it has finished. This has nothing to do with the number of swapchain images, which refer to color attachments that need to be left alone until the presentation engine signals the semaphore saying it is done with that image.

I'd strongly recommend to rename ImageCount to FrameResourceCount. That would be equal to the number of command buffers the application uses, if it uses a single command buffer per frame. This is not true in most cases, but since imgui uses a single command buffer per window surface, you might even call it CommandBufferCount; that would still be less confusing than ImageCount.

In this light, I'd also argue that

IM_ASSERT(info->ImageCount >= info->MinImageCount);

makes no sense. You can't compare the two. In fact, I don't think that imgui has anything to do with MinImageCount at all, except to initialize vulkan examples (aka, to create a swapchain?) Imgui should not be aware of the swapchain and therefore has no business knowing about MinImageCount imho.