Commands and synchronization

VulkanCommandBuffer owns a command pool and one primary command buffer for the queue mode selected by its VulkanContext.

Reusable command buffers

Call Reset(), begin the native command buffer, record work, and submit through one of the Submit() overloads. SubmitSyncInfo groups wait semaphores, pipeline stage masks, and signal semaphores for rendering submissions.

core::vulkan::VulkanCommandBuffer command_buffer(&context);
core::vulkan::VulkanFence fence(&context);

command_buffer.Reset();
VkCommandBufferBeginInfo begin_info{};
begin_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
VK_CHECK(vkBeginCommandBuffer(command_buffer.buffer(), &begin_info));

// Record vkCmd* operations here.

fence.Reset();
command_buffer.Submit(fence.fence);
VK_CHECK(vkWaitForFences(
    context.logical_device, 1, &fence.fence, VK_TRUE, UINT64_MAX));

Note

CORE’s Submit() ends the command buffer internally. Do not call vkEndCommandBuffer() before it. Reset() only resets the command buffer; wait for earlier work that uses it before recording again.

One-time commands

Resource transfers and layout transitions can use:

auto command =
    core::vulkan::VulkanCommandBuffer::BeginOneTimeCommands(&context);
// Record transfer or transition commands.
command.EndOneTimeCommands();

EndOneTimeCommands() submits to the context’s selected queue and waits on a temporary fence. It is convenient for setup operations, but each call blocks the CPU until that queue completes the command.

Fences and semaphores

• VulkanFence lets the CPU wait for a submitted command buffer and can be reset before reuse.

• VulkanSemaphore orders GPU operations, such as image acquisition, rendering, and presentation.

The graphics examples use a fence for frames in flight, an image-available semaphore, and one render-finished semaphore per swapchain image.

command_buffer.Submit(
    in_flight_fence.fence,
    core::vulkan::SubmitSyncInfo{
        .wait_semaphores = {image_available.semaphore},
        .wait_stage_masks = {
            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT},
        .signal_semaphores = {
            render_finished[image_index]->semaphore},
    });

Warning

wait_semaphores and wait_stage_masks must have the same number of entries. Keep a render-finished semaphore per swapchain image as the current triangle example does; reusing a semaphore before presentation has consumed its signal is invalid.