GitLab

I'd like to start a discussion about a bug I've found on anv before I send a patch because I am not sure about the use of tripple buffering inside the Vulkan driver.

The problem:

When creating a swapchain we usually query the surface capabilities using vkGetPhysicalDeviceSurfaceCapabilitiesKHR. The minimum and maximum images numbers returned by anv are always out of bounds: minImageCount is 3 and maxImageCount is 0! This obviously causes a validation error and maybe further problems.

Checking the anv code I've seen that the maxImageCount is always hard coded to 0 and there's a comment that there's no real maximum. I think this is not correct because 0 will always cause an out of bounds error.

But there's another problem as well: the minimum is taken by the function x11_get_min_image_count which returns 3 and has this comment:

   /* For IMMEDIATE and FIFO, most games work in a pipelined manner where the
    * can produce frames at a rate of 1/MAX(CPU duration, GPU duration), but
    * the render latency is CPU duration + GPU duration.
    *
    * This means that with scanout from pageflipping we need 3 frames to run
    * full speed:
    * 1) CPU rendering work
    * 2) GPU rendering work
    * 3) scanout
    *
    * Once we have a nonblocking acquire that returns a semaphore we can merge
    * 1 and 3. Hence the ideal implementation needs only 2 images, but games
    * cannot tellwe currently do not have an ideal implementation and that
    * hence they need to allocate 3 images. So let us do it for them.
    *
    * This is a tradeoff as it uses more memory than needed for non-fullscreen
    * and non-performance intensive applications.
    */
   return 3;

I believe that: 1- we shouldn't force triple buffering and use a minImageCount < 3 (to avoid the latency problems) 2- we shouldn't set the maximum image count to 0 (because this way we will always get the validation error and out of bounds numbers)

My questions are:

1. Does Vulkan support single buffering? If so, I think we should set the minimum to 1, else to 2. (Nvidia returns 2 so, maybe double buffering is always required?)
2. What about using a maximum image count that makes sense and is >= 3? Nvidia returns 8 but I don't know how this number is selected. Is there a reason we want to always force triple buffering or is it a bug? I think it would make more sense to leave this decision to the user because the latency with 3 images is high.