double buffered vertex buffer in two threads with two different queues if possible

include/application.hpp

@@ -44,9 +44,12 @@ private:
 	void createSyncObjects();
 	unique_ptr<Semaphore> imageAvailable;
 	unique_ptr<Semaphore> renderFinished;
-	unique_ptr<Semaphore> computeFinished;
-	unique_ptr<Fence> renderInFlight;
-	unique_ptr<Fence> computeInFlight;
+	unique_ptr<Semaphore> computeSemaphore;
+	unique_ptr<Semaphore> transferFinished;
+	unique_ptr<Fence> renderFence;
+	unique_ptr<Fence> computeFence;
+	unique_ptr<Fence> transferFence;
+	std::mutex submitMutex;
 
 	unique_ptr<Swapchain> swapchain;
 	unique_ptr<DescriptorPool> descriptorPool;
@@ -56,7 +59,8 @@ private:
 	unique_ptr<Camera> camera;
 
 	void createMeshBuffers();
-	unique_ptr<Buffer> vertexBuffer;
+	size_t currentDrawVertexBuffer = 0;
+	unique_ptr<Buffer> vertexBuffers[2];
 	unique_ptr<Buffer> faceBuffer;
 	unique_ptr<Buffer> edgeBuffer;
 	unique_ptr<Buffer> triangleBuffer;
@@ -75,6 +79,7 @@ private:
 
 	struct Properties {
 		glm::vec3 gravity;
+		// Delta time in seconds
 		float dt;
 		uint32_t k;
 	};
@@ -86,8 +91,7 @@ private:
 	unique_ptr<ComputePipeline> normalPipeline;
 
 	void updateUniformBuffer();
-	void recordGraphicsCommandBuffer(uint32_t imageIndex);
-	void recordDrawCommands();
+	void recordDrawCommands(VkCommandBuffer cmdBuffer);
 	void drawFrame();
 
 	void recordComputeCommands(VkCommandBuffer cmdBuffer);

include/vulkan/command_pool.hpp

@@ -7,12 +7,10 @@ class Instance;
 
 class CommandPool {
 public:
-	explicit CommandPool(VkSurfaceKHR surface);
+	explicit CommandPool(uint32_t queueFamilyIndex, uint32_t bufferCount);
 	~CommandPool();
-	VkCommandBuffer graphicsBuffer = VK_NULL_HANDLE;
-	VkCommandBuffer computeBuffer = VK_NULL_HANDLE;
 	VkCommandPool handle = VK_NULL_HANDLE;
+	std::vector<VkCommandBuffer> buffers;
 private:
-
-	void createBuffers();
+	void allocateBuffers(uint32_t count);
 };

include/vulkan/instance.hpp

@@ -4,8 +4,11 @@
 #include <GLFW/glfw3.h>
 #include <optional>
 #include <set>
+#include <vector>
 #include "vk_mem_alloc.h"
 
+using std::optional, std::vector;
+
 class CommandPool;
 
 class Instance {
@@ -14,35 +17,37 @@ public:
 	~Instance();
 
 	GLFWwindow *window = nullptr;
-	VkQueue graphicsQueue = VK_NULL_HANDLE;
-	VkQueue presentQueue = VK_NULL_HANDLE;
-	VkQueue computeQueue = VK_NULL_HANDLE;
+	VkQueue graphicsAndPresentQueue = VK_NULL_HANDLE;
+	VkQueue computeAndTransferQueue = VK_NULL_HANDLE;
 
 	bool windowResized = false;
 
-	CommandPool* commandPool = nullptr;
+	CommandPool* renderingCommandPool = nullptr;
+	CommandPool* computeCommandPool = nullptr;
+
 	
 	struct QueueFamilyIndices {
-		std::optional<uint32_t> graphicsFamily;
-		std::optional<uint32_t> computeFamily;
-		std::optional<uint32_t> presentFamily;
-		std::optional<uint32_t> graphicsAndComputeFamily;
-		bool isComplete() const {
-			return graphicsFamily.has_value() &&
-			computeFamily.has_value() &&
-			presentFamily.has_value() &&
-			graphicsAndComputeFamily.has_value();
-		}
+		vector<uint32_t> graphicsAndPresent;
+		vector<uint32_t> computeAndTransfer;
 		std::set<uint32_t> uniqueQueueFamilies(){
-			return {
-				graphicsFamily.value(),
-				presentFamily.value(),
-				computeFamily.value(),
-				graphicsAndComputeFamily.value()
-			};
+			std::set<uint32_t> unique;
+			unique.insert(graphicsAndPresent.begin(), graphicsAndPresent.end());
+			unique.insert(computeAndTransfer.begin(), computeAndTransfer.end());
+			return unique;
+		}
+		uint32_t tryComputeAndTransferDedicated(){
+			for (uint32_t family : computeAndTransfer){
+				if (std::find(graphicsAndPresent.begin(), graphicsAndPresent.end(), family) == graphicsAndPresent.end()){
+					return family;
+				}
+			}
+			return computeAndTransfer[0];
+		}
+		bool isEnough(){
+			return !graphicsAndPresent.empty() && !computeAndTransfer.empty();
 		}
 	};
-	static QueueFamilyIndices findQueueFamilies(VkPhysicalDevice device, VkSurfaceKHR surface);
+	QueueFamilyIndices indices {};
 
 	static Instance* instance;
 	static VkDevice GetDevice();
@@ -67,5 +72,6 @@ private:
 	bool isDeviceSuitable(VkPhysicalDevice potentialPhysicalDevice);
 
 	static bool checkDeviceExtensionSupport(VkPhysicalDevice device);
+	static QueueFamilyIndices findQueueFamilies(VkPhysicalDevice device, VkSurfaceKHR surface);
 
 };

shaders/pbd.comp

@@ -68,8 +68,8 @@ void preSolve(uint vID){
 	if (vertices[vID].w == 0){
 		return;
 	}
-	vertices[vID].velocity += dt * gravity;
-	vertices[vID].position += dt * vertices[vID].velocity;
+	vertices[vID].velocity += dt / k * gravity;
+	vertices[vID].position += dt / k * vertices[vID].velocity;
 
 	float dist = vertices[vID].position.y + 5;
 	if (dist < 0){
@@ -102,7 +102,7 @@ void solveEdge(uint eID){
 	vec3 diff = v1.position - v2.position;
 	float currentLength = length(diff);
 
-	float alpha = edge.compliance / dt / dt;
+	float alpha = edge.compliance / (dt / k) / (dt / k);
 
 	float s = -(currentLength - edge.restLength) / (v1.w + v2.w + alpha);
 
@@ -144,7 +144,7 @@ void solveTetrahedron(uint tetID){
 
 	if (w == 0) return;
 
-	float alpha = tetrahedron.compliance / dt / dt;
+	float alpha = tetrahedron.compliance / (dt / k) / (dt / k);
 
 	float s = -volumeError / (w + alpha);
 
@@ -163,7 +163,7 @@ void postSolve(uint vID){
 	if (vertices[vID].w == 0){
 		return;
 	}
-	vertices[vID].velocity = (vertices[vID].position - vertices[vID].prevPosition) / dt;
+	vertices[vID].velocity = (vertices[vID].position - vertices[vID].prevPosition) / (dt / k);
 }
 
 void main() {

src/application.cpp

@@ -12,6 +12,7 @@
 #include "soft_body.hpp"
 #include "mesh.hpp"
 #include "constraints.hpp"
+#include "timer.hpp"
 
 Application::Application() {
 	createSyncObjects();
@@ -34,7 +35,7 @@ Application::Application() {
 	createMeshBuffers();
 
 	SizeInformation sizeInformation {};
-	sizeInformation.vertexCount = vertexBuffer->size / sizeof(Vertex);
+	sizeInformation.vertexCount = vertexBuffers[0]->size / sizeof(Vertex);
 	sizeInformation.faceCount = faceBuffer->size / sizeof(Face);
 
 	sizeInformationBuffer = make_unique<Buffer>(
@@ -44,14 +45,14 @@ Application::Application() {
 
 	properties.gravity = {0, -9.81, 0};
 	properties.k = 10;
-	properties.dt = 1.f / 60.f / static_cast<float>(properties.k);
+	properties.dt = 1.f / 60.f;
 
 	propertiesBuffer = make_unique<Buffer>(
 			sizeof(Properties), &properties, sizeof(properties),
 			VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT,
 			VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0);
 
-	descriptorPool->bindBuffer(*vertexBuffer, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, DescriptorSet::MESH, 0);
+	descriptorPool->bindBuffer(*vertexBuffers[1 - currentDrawVertexBuffer], VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, DescriptorSet::MESH, 0);
 	descriptorPool->bindBuffer(*faceBuffer, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, DescriptorSet::MESH, 1);
 	descriptorPool->bindBuffer(*edgeBuffer, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, DescriptorSet::MESH, 2);
 	descriptorPool->bindBuffer(*triangleBuffer, VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, DescriptorSet::MESH, 3);
@@ -68,12 +69,32 @@ Application::Application() {
 	vmaFreeStatsString(Instance::GetAllocator(), stats);
 }
 
+#include <future>
+#include <chrono>
+
+using namespace std::chrono;
+
 void Application::mainLoop() {
+	std::future compute = std::async(std::launch::async, [this](){
 		while (!glfwWindowShouldClose(Instance::instance->window)){
-		glfwPollEvents();
+			Timer timer;
+
+			auto t1 = system_clock::now();
 			update();
+			auto t2 = system_clock::now();
+
+			microseconds updateDuration = duration_cast<microseconds>(t2 - t1);
+
+			microseconds sleepDuration(static_cast<int64_t>(properties.dt * 1000 * 1000));
+			std::this_thread::sleep_for(sleepDuration - updateDuration);
+		}
+	});
+
+	while (!glfwWindowShouldClose(Instance::instance->window)){
+		glfwPollEvents();
 		drawFrame();
 	}
+	compute.wait();
 	vkDeviceWaitIdle(Instance::GetDevice());
 }
 
@@ -84,9 +105,11 @@ Application::~Application() {
 void Application::createSyncObjects() {
 	imageAvailable = make_unique<Semaphore>();
 	renderFinished = make_unique<Semaphore>();
-	computeFinished = make_unique<Semaphore>();
-	renderInFlight = make_unique<Fence>(true);
-	computeInFlight = make_unique<Fence>(true);
+	computeSemaphore = make_unique<Semaphore>();
+	transferFinished = make_unique<Semaphore>();
+	renderFence = make_unique<Fence>(true);
+	computeFence = make_unique<Fence>(true);
+	transferFence = make_unique<Fence>(true);
 }
 
 void Application::createMeshBuffers() {
@@ -102,7 +125,7 @@ void Application::createMeshBuffers() {
 	}
 
 	body = std::make_unique<SoftBody>(&bunny, 1.f / 3);
-	for (size_t i = 0; i < 5; i++){
+	for (size_t i = 0; i < 10; i++){
 		auto copy = std::make_unique<SoftBody>(*body.get());
 		copy->applyVertexOffset({i * 2, 0, 2});
 		softBodies.push_back(std::move(copy));
@@ -196,7 +219,10 @@ void Application::createMeshBuffers() {
 				: Buffer(size, data, size, VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | additionalUsageFlags, VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE, 0) {}
 	};
 
-	vertexBuffer = make_unique<SimulationBuffer>(vertices.data(), vertices.size() * sizeof(Vertex), VK_BUFFER_USAGE_VERTEX_BUFFER_BIT);
+	vertexBuffers[0] = make_unique<SimulationBuffer>(vertices.data(), vertices.size() * sizeof(Vertex),
+												  VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT);
+	vertexBuffers[1] = make_unique<SimulationBuffer>(vertices.data(), vertices.size() * sizeof(Vertex),
+												  VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT);
 	faceBuffer = make_unique<SimulationBuffer>(faces.data(), faces.size() * sizeof(Face), VK_BUFFER_USAGE_INDEX_BUFFER_BIT);
 	edgeBuffer = make_unique<SimulationBuffer>(constraintData.edges.data(), constraintData.edges.size() * sizeof(Edge));
 	triangleBuffer = make_unique<SimulationBuffer>(constraintData.triangles.data(), constraintData.triangles.size() * sizeof(Triangle));
@@ -246,16 +272,62 @@ void Application::updateUniformBuffer() {
 	ubo.projection[1][1] *= -1;
 
 	memcpy(uniformBuffer->allocationInfo.pMappedData, &ubo, sizeof(UniformBufferObject));
+
+	VkMappedMemoryRange mappedMemoryRange {};
+	mappedMemoryRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE;
+	mappedMemoryRange.offset = uniformBuffer->allocationInfo.offset;
+	mappedMemoryRange.size = uniformBuffer->allocationInfo.size;
+	mappedMemoryRange.memory = uniformBuffer->allocationInfo.deviceMemory;
+
+	vkFlushMappedMemoryRanges(Instance::GetDevice(), 1, &mappedMemoryRange);
 }
 
-void Application::recordGraphicsCommandBuffer(uint32_t imageIndex) {
-	VkCommandBuffer cmdBuffer = Instance::instance->commandPool->graphicsBuffer;
+void Application::drawFrame() {
+	vkWaitForFences(Instance::GetDevice(), 1, &renderFence->handle, VK_TRUE, UINT64_MAX);
+
+	uint32_t imageIndex;
+	VkResult result = vkAcquireNextImageKHR(Instance::GetDevice(), swapchain->handle, UINT64_MAX, imageAvailable->handle, VK_NULL_HANDLE, &imageIndex);
+	if (result == VK_ERROR_OUT_OF_DATE_KHR){
+		swapchain->recreateSwapchain();
+		return;
+	}
+
+	vkResetFences(Instance::GetDevice(), 1, &renderFence->handle);
+
+	camera->update(0.017);
+	updateUniformBuffer();
+
+	VkCommandBuffer cmdBuffer = Instance::instance->renderingCommandPool->buffers[0];
+	{
+		vkResetCommandBuffer(cmdBuffer, 0);
 
 		VkCommandBufferBeginInfo beginInfo{};
 		beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
 
 		vkBeginCommandBuffer(cmdBuffer, &beginInfo);
 
+		VkBufferMemoryBarrier vertexBufferBarrier{};
+		vertexBufferBarrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
+		vertexBufferBarrier.size = vertexBuffers[currentDrawVertexBuffer]->size;
+		vertexBufferBarrier.offset = 0;
+		vertexBufferBarrier.buffer = vertexBuffers[currentDrawVertexBuffer]->handle;
+		vertexBufferBarrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_SHADER_READ_BIT;
+		vertexBufferBarrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
+
+		vkCmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, 0,
+							 nullptr, 1, &vertexBufferBarrier, 0, nullptr);
+
+		VkBufferMemoryBarrier uniformBufferBarrier {};
+		uniformBufferBarrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
+		uniformBufferBarrier.size = uniformBuffer->size;
+		uniformBufferBarrier.offset = 0;
+		uniformBufferBarrier.buffer = uniformBuffer->handle;
+		uniformBufferBarrier.srcAccessMask = VK_ACCESS_HOST_WRITE_BIT;
+		uniformBufferBarrier.dstAccessMask = VK_ACCESS_UNIFORM_READ_BIT;
+
+		vkCmdPipelineBarrier(cmdBuffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0,
+							 0, nullptr, 1, &uniformBufferBarrier, 0, nullptr);
+
 		VkRenderPassBeginInfo renderPassInfo{};
 		renderPassInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
 		renderPassInfo.renderPass = swapchain->renderPass;
@@ -286,46 +358,30 @@ void Application::recordGraphicsCommandBuffer(uint32_t imageIndex) {
 		scissor.extent = swapchain->extent;
 		vkCmdSetScissor(cmdBuffer, 0, 1, &scissor);
 
-	recordDrawCommands();
+		recordDrawCommands(cmdBuffer);
 
 		vkCmdEndRenderPass(cmdBuffer);
 		vkEndCommandBuffer(cmdBuffer);
 	}
 
-void Application::drawFrame() {
-	vkWaitForFences(Instance::GetDevice(), 1, &renderInFlight->handle, VK_TRUE, UINT64_MAX);
-
-	uint32_t imageIndex;
-	VkResult result = vkAcquireNextImageKHR(Instance::GetDevice(), swapchain->handle, UINT64_MAX, imageAvailable->handle, VK_NULL_HANDLE, &imageIndex);
-	if (result == VK_ERROR_OUT_OF_DATE_KHR){
-		swapchain->recreateSwapchain();
-		return;
-	}
-
-	vkResetFences(Instance::GetDevice(), 1, &renderInFlight->handle);
-
-	vkResetCommandBuffer(Instance::instance->commandPool->graphicsBuffer, 0);
-	recordGraphicsCommandBuffer(imageIndex);
-
-	camera->update(0.017);
-	updateUniformBuffer();
-
 	VkSubmitInfo submitInfo {};
 	submitInfo.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
 
-	VkSemaphore waitSemaphores[] = {imageAvailable->handle, computeFinished->handle};
-	VkPipelineStageFlags waitStages[] = {VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT};
-	submitInfo.waitSemaphoreCount = 2;
+	VkSemaphore waitSemaphores[] = {imageAvailable->handle};
+	VkPipelineStageFlags waitStages[] = {VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT};
+	submitInfo.waitSemaphoreCount = 1;
 	submitInfo.pWaitSemaphores = waitSemaphores;
 	submitInfo.pWaitDstStageMask = waitStages;
 	submitInfo.commandBufferCount = 1;
-	submitInfo.pCommandBuffers = &Instance::instance->commandPool->graphicsBuffer;
+	submitInfo.pCommandBuffers = &cmdBuffer;
 
 	VkSemaphore signalSemaphores[] = {renderFinished->handle};
 	submitInfo.signalSemaphoreCount = 1;
 	submitInfo.pSignalSemaphores = signalSemaphores;
 
-	vkQueueSubmit(Instance::instance->graphicsQueue, 1, &submitInfo, renderInFlight->handle);
+	submitMutex.lock();
+	vkQueueSubmit(Instance::instance->graphicsAndPresentQueue, 1, &submitInfo, renderFence->handle);
+	submitMutex.unlock();
 
 	VkPresentInfoKHR presentInfo {};
 	presentInfo.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;
@@ -337,7 +393,10 @@ void Application::drawFrame() {
 	presentInfo.pSwapchains = swapchains;
 	presentInfo.pImageIndices = &imageIndex;
 
-	result = vkQueuePresentKHR(Instance::instance->presentQueue, &presentInfo);
+	submitMutex.lock();
+	result = vkQueuePresentKHR(Instance::instance->graphicsAndPresentQueue, &presentInfo);
+	submitMutex.unlock();
+
 	if (result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_SUBOPTIMAL_KHR || Instance::instance->windowResized){
 		Instance::instance->windowResized = false;
 		swapchain->recreateSwapchain();
@@ -345,17 +404,19 @@ void Application::drawFrame() {
 }
 
 void Application::update() {
-	vkWaitForFences(Instance::GetDevice(), 1, &computeInFlight->handle, VK_TRUE, UINT64_MAX);
-	vkResetFences(Instance::GetDevice(), 1, &computeInFlight->handle);
+	vkWaitForFences(Instance::GetDevice(), 1, &transferFence->handle, VK_TRUE, UINT64_MAX);
+	vkResetFences(Instance::GetDevice(), 1, &transferFence->handle);
 
-
-	VkCommandBuffer cmdBuffer = Instance::instance->commandPool->computeBuffer;
-	vkResetCommandBuffer(cmdBuffer, 0);
+	currentDrawVertexBuffer = 1 - currentDrawVertexBuffer;
+	descriptorPool->bindBuffer(*vertexBuffers[1 - currentDrawVertexBuffer], VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, DescriptorSet::MESH, 0);
 
 	VkCommandBufferBeginInfo beginInfo {};
 	beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
 	beginInfo.flags = 0;
 
+	VkCommandBuffer cmdBuffer = Instance::instance->computeCommandPool->buffers[0];
+	{
+		vkResetCommandBuffer(cmdBuffer, 0);
 		vkBeginCommandBuffer(cmdBuffer, &beginInfo);
 		recordComputeCommands(cmdBuffer);
 		vkEndCommandBuffer(cmdBuffer);
@@ -363,19 +424,49 @@ void Application::update() {
 		VkSubmitInfo submit {};
 		submit.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
 		submit.commandBufferCount = 1;
-	submit.pCommandBuffers = &Instance::instance->commandPool->computeBuffer;
+		submit.pCommandBuffers = &cmdBuffer;
 		submit.signalSemaphoreCount = 1;
-	submit.pSignalSemaphores = &computeFinished->handle;
+		submit.pSignalSemaphores = &computeSemaphore->handle;
 
-	vkQueueSubmit(Instance::instance->computeQueue, 1, &submit, computeInFlight->handle);
+		submitMutex.lock();
+		vkQueueSubmit(Instance::instance->computeAndTransferQueue, 1, &submit, nullptr);
+		submitMutex.unlock();
 	}
 
-void Application::recordDrawCommands() {
-	VkCommandBuffer cmdBuffer = Instance::instance->commandPool->graphicsBuffer;
+	cmdBuffer = Instance::instance->computeCommandPool->buffers[1];
+	vkResetCommandBuffer(cmdBuffer, 0);
+	{
+		vkBeginCommandBuffer(cmdBuffer, &beginInfo);
+
+		VkBufferCopy copyRegion {};
+		copyRegion.size = vertexBuffers[0]->size;
+		copyRegion.srcOffset = 0;
+		copyRegion.dstOffset = 0;
+
+		vkCmdCopyBuffer(cmdBuffer, vertexBuffers[1 - currentDrawVertexBuffer]->handle, vertexBuffers[currentDrawVertexBuffer]->handle, 1, &copyRegion);
+
+		vkEndCommandBuffer(cmdBuffer);
+
+		VkPipelineStageFlags waitStage = VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+
+		VkSubmitInfo submit {};
+		submit.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
+		submit.commandBufferCount = 1;
+		submit.pCommandBuffers = &cmdBuffer;
+		submit.waitSemaphoreCount = 1;
+		submit.pWaitSemaphores = &computeSemaphore->handle;
+		submit.pWaitDstStageMask = &waitStage;
+
+		submitMutex.lock();
+		vkQueueSubmit(Instance::instance->computeAndTransferQueue, 1, &submit, transferFence->handle);
+		submitMutex.unlock();
+	}
+}
 
+void Application::recordDrawCommands(VkCommandBuffer cmdBuffer) {
 	vkCmdBindPipeline(cmdBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, graphicsPipeline->handle);
 
-	VkBuffer buffers[] = {vertexBuffer->handle};
+	VkBuffer buffers[] = {vertexBuffers[currentDrawVertexBuffer]->handle};
 	VkDeviceSize offsets[] = {0};
 	vkCmdBindVertexBuffers(cmdBuffer, 0, 1, buffers, offsets);
 	vkCmdBindIndexBuffer(cmdBuffer, faceBuffer->handle, 0, VK_INDEX_TYPE_UINT32);
@@ -393,7 +484,7 @@ void Application::recordComputeCommands(VkCommandBuffer cmdBuffer) {
 		return (threads - 1) / blockSize + 1;
 	};
 
-	uint32_t vertexGroupCount = getGroupCount(vertexBuffer->size / sizeof(Vertex), BlOCK_SIZE);
+	uint32_t vertexGroupCount = getGroupCount(vertexBuffers[1 - currentDrawVertexBuffer]->size / sizeof(Vertex), BlOCK_SIZE);
 	uint32_t faceGroupCount = getGroupCount(faceBuffer->size / sizeof(Face), BlOCK_SIZE);
 
 	VkMemoryBarrier barrier {};

src/vulkan/buffer.cpp

@@ -43,7 +43,7 @@ void Buffer::copyTo(Buffer *dst) {
 	VkCommandBufferAllocateInfo allocateInfo {};
 	allocateInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
 	allocateInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
-	allocateInfo.commandPool = Instance::instance->commandPool->handle;
+	allocateInfo.commandPool = Instance::instance->renderingCommandPool->handle;
 	allocateInfo.commandBufferCount = 1;
 
 	VkCommandBuffer commandBuffer;
@@ -64,8 +64,8 @@ void Buffer::copyTo(Buffer *dst) {
 	submitInfo.commandBufferCount = 1;
 	submitInfo.pCommandBuffers = &commandBuffer;
 
-	vkQueueSubmit(Instance::instance->graphicsQueue, 1, &submitInfo, VK_NULL_HANDLE);
+	vkQueueSubmit(Instance::instance->graphicsAndPresentQueue, 1, &submitInfo, VK_NULL_HANDLE);
 	vkDeviceWaitIdle(Instance::GetDevice());
 
-	vkFreeCommandBuffers(Instance::GetDevice(), Instance::instance->commandPool->handle, 1, &commandBuffer);
+	vkFreeCommandBuffers(Instance::GetDevice(), Instance::instance->renderingCommandPool->handle, 1, &commandBuffer);
 }

src/vulkan/command_pool.cpp

@@ -2,32 +2,29 @@
 #include "application.hpp"
 #include "vulkan/instance.hpp"
 
-CommandPool::CommandPool(VkSurfaceKHR surface) {
-	Instance::QueueFamilyIndices indices = Instance::findQueueFamilies(Instance::GetPhysicalDevice(), surface);
-
+CommandPool::CommandPool(uint32_t queueFamilyIndex, uint32_t bufferCount) {
 	VkCommandPoolCreateInfo poolInfo {};
 	poolInfo.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
 	poolInfo.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;
-	poolInfo.queueFamilyIndex = indices.graphicsAndComputeFamily.value();
+	poolInfo.queueFamilyIndex = queueFamilyIndex;
 
 	vkCreateCommandPool(Instance::GetDevice(), &poolInfo, nullptr, &handle);
 
-	createBuffers();
+	allocateBuffers(bufferCount);
 }
 
-void CommandPool::createBuffers() {
+void CommandPool::allocateBuffers(uint32_t count) {
 	VkCommandBufferAllocateInfo allocateInfo {};
 	allocateInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
 	allocateInfo.commandPool = handle;
 	allocateInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
-	allocateInfo.commandBufferCount = 1;
+	allocateInfo.commandBufferCount = count;
 
-	vkAllocateCommandBuffers(Instance::GetDevice(), &allocateInfo, &graphicsBuffer);
-	vkAllocateCommandBuffers(Instance::GetDevice(), &allocateInfo, &computeBuffer);
+	buffers.resize(count);
+	vkAllocateCommandBuffers(Instance::GetDevice(), &allocateInfo, buffers.data());
 }
 
 CommandPool::~CommandPool() {
-	vkFreeCommandBuffers(Instance::GetDevice(), handle, 1, &graphicsBuffer);
-	vkFreeCommandBuffers(Instance::GetDevice(), handle, 1, &computeBuffer);
+	vkFreeCommandBuffers(Instance::GetDevice(), handle, buffers.size(), buffers.data());
 	vkDestroyCommandPool(Instance::GetDevice(), handle, nullptr);
 }

src/vulkan/instance.cpp

@@ -59,7 +59,8 @@ Instance::Instance() {
 	pickPhysicalDevice();
 	createLogicalDevice();
 	createAllocator();
-	commandPool = new CommandPool(surface);
+	renderingCommandPool = new CommandPool(indices.graphicsAndPresent[0], 1);
+	computeCommandPool = new CommandPool(indices.tryComputeAndTransferDedicated(), 2);
 }
 
 void Instance::initWindow() {
@@ -118,9 +119,10 @@ void Instance::pickPhysicalDevice() {
 	std::vector<VkPhysicalDevice> devices(deviceCount);
 	vkEnumeratePhysicalDevices(handle, &deviceCount, devices.data());
 
-	for (const VkPhysicalDevice &device : devices){
-		if (isDeviceSuitable(device)){
-			physicalDevice = device;
+	for (const VkPhysicalDevice &potentialPhysicalDevice : devices){
+		if (isDeviceSuitable(potentialPhysicalDevice)){
+			physicalDevice = potentialPhysicalDevice;
+			indices = findQueueFamilies(physicalDevice, surface);
 			break;
 		}
 	}
@@ -135,8 +137,6 @@ void Instance::pickPhysicalDevice() {
 }
 
 void Instance::createLogicalDevice() {
-	QueueFamilyIndices indices = findQueueFamilies(physicalDevice, surface);
-
 	std::vector<VkDeviceQueueCreateInfo> queueCreateInfos;
 
 	float queuePriority = 1.0f;
@@ -167,9 +167,8 @@ void Instance::createLogicalDevice() {
 
 	vkCreateDevice(physicalDevice, &createInfo, nullptr, &device);
 
-	vkGetDeviceQueue(device, indices.graphicsFamily.value(), 0, &graphicsQueue);
-	vkGetDeviceQueue(device, indices.presentFamily.value(), 0, &presentQueue);
-	vkGetDeviceQueue(device, indices.graphicsAndComputeFamily.value(), 0, &computeQueue);
+	vkGetDeviceQueue(device, indices.graphicsAndPresent[0], 0, &graphicsAndPresentQueue);
+	vkGetDeviceQueue(device, indices.tryComputeAndTransferDedicated(), 0, &computeAndTransferQueue);
 }
 
 void Instance::createAllocator() {
@@ -196,23 +195,18 @@ Instance::QueueFamilyIndices Instance::findQueueFamilies(VkPhysicalDevice device
 
 	uint32_t i = 0;
 	for (const VkQueueFamilyProperties& queueFamilyProperties : queueFamilies){
-		if (queueFamilyProperties.queueFlags & VK_QUEUE_GRAPHICS_BIT){
-			indices.graphicsFamily = i;
-		}
-		if (queueFamilyProperties.queueFlags & VK_QUEUE_COMPUTE_BIT){
-			indices.computeFamily = i;
-		}
-		if (indices.graphicsFamily == i && indices.computeFamily == i){
-			indices.graphicsAndComputeFamily = i;
-		}
-		VkBool32 presentSupport = false;
-		vkGetPhysicalDeviceSurfaceSupportKHR(device, i, surface, &presentSupport);
-		if (presentSupport){
-			indices.presentFamily = i;
-		}
-		if (indices.isComplete()){
-			break;
-		}
+		VkBool32 present = false;
+		vkGetPhysicalDeviceSurfaceSupportKHR(device, i, surface, &present);
+
+		bool graphics = queueFamilyProperties.queueFlags & VK_QUEUE_GRAPHICS_BIT;
+		bool compute = queueFamilyProperties.queueFlags & VK_QUEUE_COMPUTE_BIT;
+		bool transfer = queueFamilyProperties.queueFlags & VK_QUEUE_TRANSFER_BIT;
+
+		if (graphics && present)
+			indices.graphicsAndPresent.push_back(i);
+		if (compute && transfer)
+			indices.computeAndTransfer.push_back(i);
+
 		i++;
 	}
 	return indices;
@@ -225,7 +219,7 @@ bool Instance::isDeviceSuitable(VkPhysicalDevice potentialPhysicalDevice) {
 	VkPhysicalDeviceFeatures deviceFeatures;
 	vkGetPhysicalDeviceFeatures(potentialPhysicalDevice, &deviceFeatures);
 
-	QueueFamilyIndices indices = findQueueFamilies(potentialPhysicalDevice, surface);
+	QueueFamilyIndices potentialIndices = findQueueFamilies(potentialPhysicalDevice, surface);
 
 	bool extensionsSupported = checkDeviceExtensionSupport(potentialPhysicalDevice);
 
@@ -235,7 +229,7 @@ bool Instance::isDeviceSuitable(VkPhysicalDevice potentialPhysicalDevice) {
 		swapChainAdequate = !details.formats.empty() && !details.presentModes.empty();
 	}
 
-	return indices.isComplete() && extensionsSupported && swapChainAdequate;
+	return potentialIndices.isEnough() && extensionsSupported && swapChainAdequate;
 }
 
 bool Instance::checkDeviceExtensionSupport(VkPhysicalDevice device) {
@@ -271,7 +265,8 @@ VkSurfaceKHR Instance::GetSurface() {
 }
 
 Instance::~Instance() {
-	delete commandPool;
+	delete renderingCommandPool;
+	delete computeCommandPool;
 	vmaDestroyAllocator(allocator);
 	vkDestroyDevice(device, nullptr);
 	vkDestroySurfaceKHR(handle, surface, nullptr);

src/vulkan/swapchain.cpp

@@ -80,16 +80,7 @@ void Swapchain::createSwapchain() {
 	createInfo.imageArrayLayers = 1;
 	createInfo.imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
 
-	Instance::QueueFamilyIndices indices = Instance::findQueueFamilies(Instance::GetPhysicalDevice(), Instance::GetSurface());
-	uint32_t queueFamilyIndices[] = {indices.graphicsFamily.value(), indices.presentFamily.value()};
-
-	if (indices.graphicsFamily != indices.presentFamily){
-		createInfo.imageSharingMode = VK_SHARING_MODE_CONCURRENT;
-		createInfo.queueFamilyIndexCount = 2;
-		createInfo.pQueueFamilyIndices = queueFamilyIndices;
-	} else {
 	createInfo.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE;
-	}
 
 	createInfo.preTransform = swapchainSupport.capabilities.currentTransform;
 	createInfo.compositeAlpha = VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR;
@@ -118,7 +109,7 @@ void Swapchain::cleanupSwapchain() {
 }
 
 void Swapchain::recreateSwapchain() {
-	vkDeviceWaitIdle(Instance::GetDevice());
+	vkQueueWaitIdle(Instance::instance->graphicsAndPresentQueue);
 
 	cleanupSwapchain();