Struct wgpu_hal::vulkan::RelaySemaphores

struct RelaySemaphores {
    wait: Option<Semaphore>,
    signal: Semaphore,
}

Semaphores for forcing queue submissions to run in order.

The wgpu_hal::Queue trait promises that if two calls to submit are ordered, then the first submission will finish on the GPU before the second submission begins. To get this behavior on Vulkan we need to pass semaphores to vkQueueSubmit for the commands to wait on before beginning execution, and to signal when their execution is done.

Normally this can be done with a single semaphore, waited on and then signalled for each submission. At any given time there’s exactly one submission that would signal the semaphore, and exactly one waiting on it, as Vulkan requires.

However, as of Oct 2021, bug #5508 in the Mesa ANV drivers caused them to hang if we use a single semaphore. The workaround is to alternate between two semaphores. The bug has been fixed in Mesa, but we should probably keep the workaround until, say, Oct 2026.

Fields

wait: Option<Semaphore>

The semaphore the next submission should wait on before beginning execution on the GPU. This is None for the first submission, which should not wait on anything at all.

signal: Semaphore

The semaphore the next submission should signal when it has finished execution on the GPU.

Implementations

impl RelaySemaphores

fn new(device: &DeviceShared) -> Result<Self, DeviceError>

fn advance(&mut self, device: &DeviceShared) -> Result<Self, DeviceError>

Advances the semaphores, returning the semaphores that should be used for a submission.

unsafe fn destroy(&self, device: &Device)

Destroys the semaphores.

Trait Implementations

impl Clone for RelaySemaphores

fn clone(&self) -> RelaySemaphores

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.