frames in flight == swapchain images

[alexwennstrom]

This is a great example, I don't quite understand why author choose to have equal frames in flight as swapchain images, whereever I look there is usually recomended to have not more than 2 frames in flight and 2 + x swapchain images?

vk_minimal_latest/src/minimal_latest.cpp

Line 2448 in 72a37ca

createFrameSubmission(m_swapchain.getMaxFramesInFlight());

[mklefrancois]

Thanks for the heads-up! It's a good observation, though I'd nuance it a bit.

The sample wasn't incorrect; it was implicit. The frames-in-flight were tied to the swapchain image count via a single variable. getMaxFramesInFlight() returned the image count. With three swapchain images, three frames were in flight. While this can be a reasonable setup for heavy workloads, it's not the right default for interactive/latency-sensitive apps.

The real issue your comment points out is the coupling, not the specific number. Having one variable perform two functions obscures the fact that there are two independent tuning knobs: swapchain image count (presentation parallelism) and frames in flight (CPU parallelism). Once the two are separated, the per-image versus per-slot resource ownership also becomes visible (renderFinished lives with the image, and imageAvailable lives with the slot).

I'm working on a broader branch that touches on several parts of the sample, including shader objects, BDA opt-in, documentation updates, and a few bug fixes. As part of that, the swapchain is refactored as follows:

• m_imageCount and m_framesInFlight are independent members with their own getters.
• The SwapchainImage {image, view, renderFinished} is sized by m_imageCount(3).
• InFlightSlot {imageAvailable} is sized by m_framesInFlight(2).
• A block comment on the class spells out this distinction so that readers don't have to rediscover it.

The fix you want is incorporated into the larger cleanup. I appreciate the note.

Thanks for flagging it; I'm happy to have the extra set of eyes.

[alexwennstrom]

Thank you for your reply. The frame synchronization aspect of Vulkan is far more confusing than most recent tutorials convey. I really appreciate the planned improvements.