itycodes
/
wlseeds
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

render/vulkan: switch to timeline semaphores

Browse Source 
Up until now we were using a VkFence for GPU-to-CPU
synchronization. This has some limitations that become a blocker
when trying to have multiple command buffers in flight at once
(e.g. for multi-output). It's desirable to implement a command
buffer pool [1], but VkFence cannot be used to track command buffer
completion for individual subpasses.

Let's just switch to timeline semaphores [2], which fix this issue,
make synchronization a lot more ergonomic and are a core Vulkan 1.2
feature.

[1]: https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/3802
[2]: https://www.khronos.org/blog/vulkan-timeline-semaphores
Simon Ser 2 years ago
parent f92d1499cd
commit a8a194d695
3 changed files with 62 additions and 27 deletions
Show Stats Download Patch File Download Diff File

4
include/render/vulkan.h
Unescape View File

@ -44,6 +44,7 @@ struct wlr_vk_device {
struct {
PFN_vkGetMemoryFdPropertiesKHR getMemoryFdPropertiesKHR;
PFN_vkWaitSemaphoresKHR waitSemaphoresKHR;
} api;
uint32_t format_prop_count;
@ -151,7 +152,8 @@ struct wlr_vk_renderer {
VkPipelineLayout pipe_layout;
VkSampler sampler;
VkFence fence;
VkSemaphore timeline_semaphore;
uint64_t timeline_point;
struct wlr_vk_render_buffer *current_render_buffer;

78
render/vulkan/renderer.c
Unescape View File

@ -349,32 +349,41 @@ bool vulkan_submit_stage_wait(struct wlr_vk_renderer *renderer) {
vkEndCommandBuffer(renderer->stage.cb);
renderer->stage.recording = false;
renderer->timeline_point++;
VkTimelineSemaphoreSubmitInfoKHR timeline_submit_info = {
.sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR,
.signalSemaphoreValueCount = 1,
.pSignalSemaphoreValues = &renderer->timeline_point,
};
VkSubmitInfo submit_info = {
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
.commandBufferCount = 1u,
.pNext = &timeline_submit_info,
.commandBufferCount = 1,
.pCommandBuffers = &renderer->stage.cb,
.signalSemaphoreCount = 1,
.pSignalSemaphores = &renderer->timeline_semaphore,
};
VkResult res = vkQueueSubmit(renderer->dev->queue, 1,
&submit_info, renderer->fence);
VkResult res = vkQueueSubmit(renderer->dev->queue, 1, &submit_info, NULL);
if (res != VK_SUCCESS) {
wlr_vk_error("vkQueueSubmit", res);
return false;
}
res = vkWaitForFences(renderer->dev->dev, 1, &renderer->fence, true,
UINT64_MAX);
VkSemaphoreWaitInfoKHR wait_info = {
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_WAIT_INFO_KHR,
.semaphoreCount = 1,
.pSemaphores = &renderer->timeline_semaphore,
.pValues = &renderer->timeline_point,
};
res = renderer->dev->api.waitSemaphoresKHR(renderer->dev->dev, &wait_info, UINT64_MAX);
if (res != VK_SUCCESS) {
wlr_vk_error("vkWaitForFences", res);
wlr_vk_error("vkWaitSemaphoresKHR", res);
return false;
}
// NOTE: don't release stage allocations here since they may still be
// used for reading. Will be done next frame.
res = vkResetFences(renderer->dev->dev, 1, &renderer->fence);
if (res != VK_SUCCESS) {
wlr_vk_error("vkResetFences", res);
return false;
}
return true;
}
@ -726,14 +735,25 @@ static void vulkan_end(struct wlr_renderer *wlr_renderer) {
++submit_count;
}
renderer->timeline_point++;
VkTimelineSemaphoreSubmitInfoKHR timeline_submit_info = {
.sType = VK_STRUCTURE_TYPE_TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR,
.signalSemaphoreValueCount = 1,
.pSignalSemaphoreValues = &renderer->timeline_point,
};
VkSubmitInfo *render_sub = &submit_infos[submit_count];
render_sub->sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
render_sub->pNext = &timeline_submit_info;
render_sub->pCommandBuffers = &render_cb;
render_sub->commandBufferCount = 1u;
render_sub->signalSemaphoreCount = 1;
render_sub->pSignalSemaphores = &renderer->timeline_semaphore,
++submit_count;
VkResult res = vkQueueSubmit(renderer->dev->queue, submit_count,
submit_infos, renderer->fence);
submit_infos, NULL);
if (res != VK_SUCCESS) {
wlr_vk_error("vkQueueSubmit", res);
return;
@ -742,10 +762,15 @@ static void vulkan_end(struct wlr_renderer *wlr_renderer) {
// sadly this is required due to the current api/rendering model of wlr
// ideally we could use gpu and cpu in parallel (_without_ the
// implicit synchronization overhead and mess of opengl drivers)
res = vkWaitForFences(renderer->dev->dev, 1, &renderer->fence, true,
UINT64_MAX);
VkSemaphoreWaitInfoKHR wait_info = {
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_WAIT_INFO_KHR,
.semaphoreCount = 1,
.pSemaphores = &renderer->timeline_semaphore,
.pValues = &renderer->timeline_point,
};
res = renderer->dev->api.waitSemaphoresKHR(renderer->dev->dev, &wait_info, UINT64_MAX);
if (res != VK_SUCCESS) {
wlr_vk_error("vkWaitForFences", res);
wlr_vk_error("vkWaitSemaphoresKHR", res);
return;
}
@ -758,11 +783,6 @@ static void vulkan_end(struct wlr_renderer *wlr_renderer) {
}
wl_list_init(&renderer->destroy_textures); // reset the list
res = vkResetFences(renderer->dev->dev, 1, &renderer->fence);
if (res != VK_SUCCESS) {
wlr_vk_error("vkResetFences", res);
return;
}
}
static bool vulkan_render_subtexture_with_matrix(struct wlr_renderer *wlr_renderer,
@ -979,7 +999,7 @@ static void vulkan_destroy(struct wlr_renderer *wlr_renderer) {
vkDestroyShaderModule(dev->dev, renderer->tex_frag_module, NULL);
vkDestroyShaderModule(dev->dev, renderer->quad_frag_module, NULL);
vkDestroyFence(dev->dev, renderer->fence, NULL);
vkDestroySemaphore(dev->dev, renderer->timeline_semaphore, NULL);
vkDestroyPipelineLayout(dev->dev, renderer->pipe_layout, NULL);
vkDestroyDescriptorSetLayout(dev->dev, renderer->ds_layout, NULL);
vkDestroySampler(dev->dev, renderer->sampler, NULL);
@ -1745,13 +1765,19 @@ struct wlr_renderer *vulkan_renderer_create_for_device(struct wlr_vk_device *dev
goto error;
}
VkFenceCreateInfo fence_info = {
.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO,
VkSemaphoreTypeCreateInfoKHR semaphore_type_info = {
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_TYPE_CREATE_INFO_KHR,
.semaphoreType = VK_SEMAPHORE_TYPE_TIMELINE_KHR,
.initialValue = 0,
};
VkSemaphoreCreateInfo semaphore_info = {
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO,
.pNext = &semaphore_type_info,
};
res = vkCreateFence(dev->dev, &fence_info, NULL,
&renderer->fence);
res = vkCreateSemaphore(dev->dev, &semaphore_info, NULL,
&renderer->timeline_semaphore);
if (res != VK_SUCCESS) {
wlr_vk_error("vkCreateFence", res);
wlr_vk_error("vkCreateSemaphore", res);
goto error;
}

7
render/vulkan/vulkan.c
Unescape View File

@ -424,6 +424,7 @@ struct wlr_vk_device *vulkan_device_create(struct wlr_vk_instance *ini,
VK_EXT_EXTERNAL_MEMORY_DMA_BUF_EXTENSION_NAME,
VK_EXT_QUEUE_FAMILY_FOREIGN_EXTENSION_NAME,
VK_EXT_IMAGE_DRM_FORMAT_MODIFIER_EXTENSION_NAME,
VK_KHR_TIMELINE_SEMAPHORE_EXTENSION_NAME, // or vulkan 1.2
};
size_t extensions_len = sizeof(extensions) / sizeof(extensions[0]);
@ -464,8 +465,13 @@ struct wlr_vk_device *vulkan_device_create(struct wlr_vk_instance *ini,
.pQueuePriorities = &prio,
};
VkPhysicalDeviceTimelineSemaphoreFeaturesKHR timeline_features = {
.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR,
.timelineSemaphore = VK_TRUE,
};
VkDeviceCreateInfo dev_info = {
.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO,
.pNext = &timeline_features,
.queueCreateInfoCount = 1u,
.pQueueCreateInfos = &qinfo,
.enabledExtensionCount = extensions_len,
@ -482,6 +488,7 @@ struct wlr_vk_device *vulkan_device_create(struct wlr_vk_instance *ini,
load_device_proc(dev, "vkGetMemoryFdPropertiesKHR",
&dev->api.getMemoryFdPropertiesKHR);
load_device_proc(dev, "vkWaitSemaphoresKHR", &dev->api.waitSemaphoresKHR);
// - check device format support -
size_t max_fmts;

Write Preview
Loading…
Cancel
Save