procedural-3d-engine/examples/trianglevulkan13/trianglevulkan13.cpp

/*
* Vulkan Example - Basic indexed triangle rendering using Vulkan 1.3
*
* Note:
* This is a variation of the the triangle sample that makes use of Vulkan 1.3 features
* This simplifies the api a bit, esp. with dynamic rendering replacing render passes (and with that framebuffers)
*
* Copyright (C) 2024 by Sascha Willems - www.saschawillems.de
*
* This code is licensed under the MIT license (MIT) (http://opensource.org/licenses/MIT)
*/

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <fstream>
#include <vector>
#include <exception>

#define GLM_FORCE_RADIANS
#define GLM_FORCE_DEPTH_ZERO_TO_ONE
#include <glm/glm.hpp>
#include <glm/gtc/matrix_transform.hpp>

#include <vulkan/vulkan.h>
#include "vulkanexamplebase.h"

// We want to keep GPU and CPU busy. To do that we may start building a new command buffer while the previous one is still being executed
// This number defines how many frames may be worked on simultaneously at once
// Increasing this number may improve performance but will also introduce additional latency
#define MAX_CONCURRENT_FRAMES 2

class VulkanExample : public VulkanExampleBase
{
public:
	// Vertex layout used in this example
	struct Vertex {
		float position[3];
		float color[3];
	};

	struct VulkanBuffer {
		VkDeviceMemory memory{ VK_NULL_HANDLE };
		VkBuffer handle{ VK_NULL_HANDLE };
	};

	VulkanBuffer vertexBuffer;
	VulkanBuffer indexBuffer;
	uint32_t indexCount{ 0 };

	// Uniform buffer block object
	struct UniformBuffer : VulkanBuffer {
		// The descriptor set stores the resources bound to the binding points in a shader
		// It connects the binding points of the different shaders with the buffers and images used for those bindings
		VkDescriptorSet descriptorSet;
		// We keep a pointer to the mapped buffer, so we can easily update it's contents via a memcpy
		uint8_t* mapped{ nullptr };
	};
	// We use one UBO per frame, so we can have a frame overlap and make sure that uniforms aren't updated while still in use
	std::array<UniformBuffer, MAX_CONCURRENT_FRAMES> uniformBuffers;

	// For simplicity we use the same uniform block layout as in the shader
	// This way we can just memcpy the data to the ubo
	// Note: You should use data types that align with the GPU in order to avoid manual padding (vec4, mat4)
	struct ShaderData {
		glm::mat4 projectionMatrix;
		glm::mat4 modelMatrix;
		glm::mat4 viewMatrix;
	};

	// The pipeline layout is used by a pipeline to access the descriptor sets
	// It defines interface (without binding any actual data) between the shader stages used by the pipeline and the shader resources
	// A pipeline layout can be shared among multiple pipelines as long as their interfaces match
	VkPipelineLayout pipelineLayout{ VK_NULL_HANDLE };

	// Pipelines (often called "pipeline state objects") are used to bake all states that affect a pipeline
	// While in OpenGL every state can be changed at (almost) any time, Vulkan requires to layout the graphics (and compute) pipeline states upfront
	// So for each combination of non-dynamic pipeline states you need a new pipeline (there are a few exceptions to this not discussed here)
	// Even though this adds a new dimension of planning ahead, it's a great opportunity for performance optimizations by the driver
	VkPipeline pipeline{ VK_NULL_HANDLE };

	// The descriptor set layout describes the shader binding layout (without actually referencing descriptor)
	// Like the pipeline layout it's pretty much a blueprint and can be used with different descriptor sets as long as their layout matches
	VkDescriptorSetLayout descriptorSetLayout{ VK_NULL_HANDLE };

	// Synchronization primitives
	// Synchronization is an important concept of Vulkan that OpenGL mostly hid away. Getting this right is crucial to using Vulkan.

	// Semaphores are used to coordinate operations within the graphics queue and ensure correct command ordering
	std::array<VkSemaphore, MAX_CONCURRENT_FRAMES> presentCompleteSemaphores{};
	std::array<VkSemaphore, MAX_CONCURRENT_FRAMES> renderCompleteSemaphores{};
	// Fences are used to make sure command buffers aren't rerecorded until they've finished executing
	std::array<VkFence, MAX_CONCURRENT_FRAMES> waitFences{};

	VkCommandPool commandPool{ VK_NULL_HANDLE };
	std::array<VkCommandBuffer, MAX_CONCURRENT_FRAMES> commandBuffers{};

	// To select the correct sync objects, we need to keep track of the current frame
	uint32_t currentFrame{ 0 };

	VkPhysicalDeviceVulkan13Features enabledFeatures{ VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_3_FEATURES };

	VulkanExample() : VulkanExampleBase()
	{
		title = "Vulkan Example - Basic indexed triangle using Vulkan 1.3";
		// To keep things simple, we don't use the UI overlay from the framework
		settings.overlay = false;
		// Setup a default look-at camera
		camera.type = Camera::CameraType::lookat;
		camera.setPosition(glm::vec3(0.0f, 0.0f, -2.5f));
		camera.setRotation(glm::vec3(0.0f));
		camera.setPerspective(60.0f, (float)width / (float)height, 1.0f, 256.0f);
		// We want to use Vulkan 1.3 with the dynamic rendering and sync 2 features
		apiVersion = VK_API_VERSION_1_3;
		enabledFeatures.dynamicRendering = VK_TRUE;
		enabledFeatures.synchronization2 = VK_TRUE;
		deviceCreatepNextChain = &enabledFeatures;
	}

	~VulkanExample()
	{
		// Clean up used Vulkan resources
		// Note: Inherited destructor cleans up resources stored in base class
		if (device) {
			vkDestroyPipeline(device, pipeline, nullptr);

			vkDestroyPipelineLayout(device, pipelineLayout, nullptr);
			vkDestroyDescriptorSetLayout(device, descriptorSetLayout, nullptr);

			vkDestroyBuffer(device, vertexBuffer.handle, nullptr);
			vkFreeMemory(device, vertexBuffer.memory, nullptr);

			vkDestroyBuffer(device, indexBuffer.handle, nullptr);
			vkFreeMemory(device, indexBuffer.memory, nullptr);

			vkDestroyCommandPool(device, commandPool, nullptr);

			for (uint32_t i = 0; i < MAX_CONCURRENT_FRAMES; i++) {
				vkDestroyFence(device, waitFences[i], nullptr);
				vkDestroySemaphore(device, presentCompleteSemaphores[i], nullptr);
				vkDestroySemaphore(device, renderCompleteSemaphores[i], nullptr);
				vkDestroyBuffer(device, uniformBuffers[i].handle, nullptr);
				vkFreeMemory(device, uniformBuffers[i].memory, nullptr);
			}
		}
	}

	// This function is used to request a device memory type that supports all the property flags we request (e.g. device local, host visible)
	// Upon success it will return the index of the memory type that fits our requested memory properties
	// This is necessary as implementations can offer an arbitrary number of memory types with different
	// memory properties.
	// You can check https://vulkan.gpuinfo.org/ for details on different memory configurations
	uint32_t getMemoryTypeIndex(uint32_t typeBits, VkMemoryPropertyFlags properties)
	{
		// Iterate over all memory types available for the device used in this example
		for (uint32_t i = 0; i < deviceMemoryProperties.memoryTypeCount; i++) {
			if ((typeBits & 1) == 1) {
				if ((deviceMemoryProperties.memoryTypes[i].propertyFlags & properties) == properties) {
					return i;
				}
			}
			typeBits >>= 1;
		}
		throw "Could not find a suitable memory type!";
	}

	// Create the per-frame (in flight) sVulkan synchronization primitives used in this example
	void createSynchronizationPrimitives()
	{
		// Semaphores are used for correct command ordering within a queue
		VkSemaphoreCreateInfo semaphoreCI{ VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO };

		// Fences are used to check draw command buffer completion on the host
		VkFenceCreateInfo fenceCI{ VK_STRUCTURE_TYPE_FENCE_CREATE_INFO };
		// Create the fences in signaled state (so we don't wait on first render of each command buffer)
		fenceCI.flags = VK_FENCE_CREATE_SIGNALED_BIT;

		for (uint32_t i = 0; i < MAX_CONCURRENT_FRAMES; i++) {		
			// Semaphore used to ensure that image presentation is complete before starting to submit again
			VK_CHECK_RESULT(vkCreateSemaphore(device, &semaphoreCI, nullptr, &presentCompleteSemaphores[i]));
			// Semaphore used to ensure that all commands submitted have been finished before submitting the image to the queue
			VK_CHECK_RESULT(vkCreateSemaphore(device, &semaphoreCI, nullptr, &renderCompleteSemaphores[i]));
			// Fence used to ensure that command buffer has completed exection before using it again
			VK_CHECK_RESULT(vkCreateFence(device, &fenceCI, nullptr, &waitFences[i]));
		}
	}

	void createCommandBuffers()
	{
		// All command buffers are allocated from the same command pool
		VkCommandPoolCreateInfo commandPoolCI{ VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO };
		commandPoolCI.queueFamilyIndex = swapChain.queueNodeIndex;
		commandPoolCI.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;
		VK_CHECK_RESULT(vkCreateCommandPool(device, &commandPoolCI, nullptr, &commandPool));
		// Allocate one command buffer per max. concurrent frame from above pool
		VkCommandBufferAllocateInfo cmdBufAllocateInfo = vks::initializers::commandBufferAllocateInfo(commandPool, VK_COMMAND_BUFFER_LEVEL_PRIMARY, MAX_CONCURRENT_FRAMES);
		VK_CHECK_RESULT(vkAllocateCommandBuffers(device, &cmdBufAllocateInfo, commandBuffers.data()));
	}

	// Prepare vertex and index buffers for an indexed triangle
	// Also uploads them to device local memory using staging and initializes vertex input and attribute binding to match the vertex shader
	void createVertexBuffer()
	{
		// A note on memory management in Vulkan in general:
		//	This is a very complex topic and while it's fine for an example application to small individual memory allocations that is not
		//	what should be done a real-world application, where you should allocate large chunks of memory at once instead.

		// Setup vertices
		const std::vector<Vertex> vertices{
			{ {  1.0f,  1.0f, 0.0f }, { 1.0f, 0.0f, 0.0f } },
			{ { -1.0f,  1.0f, 0.0f }, { 0.0f, 1.0f, 0.0f } },
			{ {  0.0f, -1.0f, 0.0f }, { 0.0f, 0.0f, 1.0f } }
		};
		uint32_t vertexBufferSize = static_cast<uint32_t>(vertices.size()) * sizeof(Vertex);

		// Setup indices
		// We do this for demonstration purposes, a triangle doesn't require indices to be rendered (because of the 1:1 mapping), but more complex shapes usually make use of indices
		std::vector<uint32_t> indices{ 0, 1, 2 };
		indexCount = static_cast<uint32_t>(indices.size());
		uint32_t indexBufferSize = indexCount * sizeof(uint32_t);

		VkMemoryAllocateInfo memAlloc{ VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO };
		VkMemoryRequirements memReqs;

		// Static data like vertex and index buffer should be stored on the device memory for optimal (and fastest) access by the GPU
		//
		// To achieve this we use so-called "staging buffers" :
		// - Create a buffer that's visible to the host (and can be mapped)
		// - Copy the data to this buffer
		// - Create another buffer that's local on the device (VRAM) with the same size
		// - Copy the data from the host to the device using a command buffer
		// - Delete the host visible (staging) buffer
		// - Use the device local buffers for rendering
		//
		// Note: On unified memory architectures where host (CPU) and GPU share the same memory, staging is not necessary
		// To keep this sample easy to follow, there is no check for that in place

		struct {
			VulkanBuffer vertexBuffer;
			VulkanBuffer indexBuffer;
		} stagingBuffers;

		void* data;

		// Vertex buffer
		VkBufferCreateInfo vertexBufferInfoCI{ VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO };
		vertexBufferInfoCI.size = vertexBufferSize;
		// Buffer is used as the copy source
		vertexBufferInfoCI.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
		// Create a host-visible buffer to copy the vertex data to (staging buffer)
		VK_CHECK_RESULT(vkCreateBuffer(device, &vertexBufferInfoCI, nullptr, &stagingBuffers.vertexBuffer.handle));
		vkGetBufferMemoryRequirements(device, stagingBuffers.vertexBuffer.handle, &memReqs);
		memAlloc.allocationSize = memReqs.size;
		// Request a host visible memory type that can be used to copy our data do
		// Also request it to be coherent, so that writes are visible to the GPU right after unmapping the buffer
		memAlloc.memoryTypeIndex = getMemoryTypeIndex(memReqs.memoryTypeBits, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
		VK_CHECK_RESULT(vkAllocateMemory(device, &memAlloc, nullptr, &stagingBuffers.vertexBuffer.memory));
		// Map and copy
		VK_CHECK_RESULT(vkMapMemory(device, stagingBuffers.vertexBuffer.memory, 0, memAlloc.allocationSize, 0, &data));
		memcpy(data, vertices.data(), vertexBufferSize);
		vkUnmapMemory(device, stagingBuffers.vertexBuffer.memory);
		VK_CHECK_RESULT(vkBindBufferMemory(device, stagingBuffers.vertexBuffer.handle, stagingBuffers.vertexBuffer.memory, 0));

		// Create a device local buffer to which the (host local) vertex data will be copied and which will be used for rendering
		vertexBufferInfoCI.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
		VK_CHECK_RESULT(vkCreateBuffer(device, &vertexBufferInfoCI, nullptr, &vertexBuffer.handle));
		vkGetBufferMemoryRequirements(device, vertexBuffer.handle, &memReqs);
		memAlloc.allocationSize = memReqs.size;
		memAlloc.memoryTypeIndex = getMemoryTypeIndex(memReqs.memoryTypeBits, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
		VK_CHECK_RESULT(vkAllocateMemory(device, &memAlloc, nullptr, &vertexBuffer.memory));
		VK_CHECK_RESULT(vkBindBufferMemory(device, vertexBuffer.handle, vertexBuffer.memory, 0));

		// Index buffer
		VkBufferCreateInfo indexbufferCI{ VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO };
		indexbufferCI.size = indexBufferSize;
		indexbufferCI.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
		// Copy index data to a buffer visible to the host (staging buffer)
		VK_CHECK_RESULT(vkCreateBuffer(device, &indexbufferCI, nullptr, &stagingBuffers.indexBuffer.handle));
		vkGetBufferMemoryRequirements(device, stagingBuffers.indexBuffer.handle, &memReqs);
		memAlloc.allocationSize = memReqs.size;
		memAlloc.memoryTypeIndex = getMemoryTypeIndex(memReqs.memoryTypeBits, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
		VK_CHECK_RESULT(vkAllocateMemory(device, &memAlloc, nullptr, &stagingBuffers.indexBuffer.memory));
		VK_CHECK_RESULT(vkMapMemory(device, stagingBuffers.indexBuffer.memory, 0, indexBufferSize, 0, &data));
		memcpy(data, indices.data(), indexBufferSize);
		vkUnmapMemory(device, stagingBuffers.indexBuffer.memory);
		VK_CHECK_RESULT(vkBindBufferMemory(device, stagingBuffers.indexBuffer.handle, stagingBuffers.indexBuffer.memory, 0));

		// Create destination buffer with device only visibility
		indexbufferCI.usage = VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
		VK_CHECK_RESULT(vkCreateBuffer(device, &indexbufferCI, nullptr, &indexBuffer.handle));
		vkGetBufferMemoryRequirements(device, indexBuffer.handle, &memReqs);
		memAlloc.allocationSize = memReqs.size;
		memAlloc.memoryTypeIndex = getMemoryTypeIndex(memReqs.memoryTypeBits, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
		VK_CHECK_RESULT(vkAllocateMemory(device, &memAlloc, nullptr, &indexBuffer.memory));
		VK_CHECK_RESULT(vkBindBufferMemory(device, indexBuffer.handle, indexBuffer.memory, 0));

		// Buffer copies have to be submitted to a queue, so we need a command buffer for them
		VkCommandBuffer copyCmd;

		VkCommandBufferAllocateInfo cmdBufAllocateInfo{ VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO };
		cmdBufAllocateInfo.commandPool = commandPool;
		cmdBufAllocateInfo.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
		cmdBufAllocateInfo.commandBufferCount = 1;
		VK_CHECK_RESULT(vkAllocateCommandBuffers(device, &cmdBufAllocateInfo, &copyCmd));

		VkCommandBufferBeginInfo cmdBufInfo = vks::initializers::commandBufferBeginInfo();
		VK_CHECK_RESULT(vkBeginCommandBuffer(copyCmd, &cmdBufInfo));
		// Copy vertex and index buffers to the device
		VkBufferCopy copyRegion{};
		copyRegion.size = vertexBufferSize;
		vkCmdCopyBuffer(copyCmd, stagingBuffers.vertexBuffer.handle, vertexBuffer.handle, 1, &copyRegion);
		copyRegion.size = indexBufferSize;
		vkCmdCopyBuffer(copyCmd, stagingBuffers.indexBuffer.handle, indexBuffer.handle,	1, &copyRegion);
		VK_CHECK_RESULT(vkEndCommandBuffer(copyCmd));

		// Submit the command buffer to the queue to finish the copy
		VkSubmitInfo submitInfo{ VK_STRUCTURE_TYPE_SUBMIT_INFO };
		submitInfo.commandBufferCount = 1;
		submitInfo.pCommandBuffers = &copyCmd;

		// Create fence to ensure that the command buffer has finished executing
		VkFenceCreateInfo fenceCI{ VK_STRUCTURE_TYPE_FENCE_CREATE_INFO };
		VkFence fence;
		VK_CHECK_RESULT(vkCreateFence(device, &fenceCI, nullptr, &fence));

		// Submit copies to the queue
		VK_CHECK_RESULT(vkQueueSubmit(queue, 1, &submitInfo, fence));
		// Wait for the fence to signal that command buffer has finished executing
		VK_CHECK_RESULT(vkWaitForFences(device, 1, &fence, VK_TRUE, DEFAULT_FENCE_TIMEOUT));

		vkDestroyFence(device, fence, nullptr);
		vkFreeCommandBuffers(device, commandPool, 1, &copyCmd);

		// The fence made sure copies are finished, so we can safely delete the staging buffer
		vkDestroyBuffer(device, stagingBuffers.vertexBuffer.handle, nullptr);
		vkFreeMemory(device, stagingBuffers.vertexBuffer.memory, nullptr);
		vkDestroyBuffer(device, stagingBuffers.indexBuffer.handle, nullptr);
		vkFreeMemory(device, stagingBuffers.indexBuffer.memory, nullptr);
	}

	// Decriptors are used to pass data to shaders, for our sample we use a descriptor to pass parameters like matrices to the shader
	void createDescriptors()
	{
		// Descriptors are allocated from a pool, that tells the implementation how many and what types of descriptors we are going to use (at maximum)
		VkDescriptorPoolSize descriptorTypeCounts[1]{};
		// This example only one descriptor type (uniform buffer)
		descriptorTypeCounts[0].type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
		// We have one buffer (and as such descriptor) per frame
		descriptorTypeCounts[0].descriptorCount = MAX_CONCURRENT_FRAMES;
		// For additional types you need to add new entries in the type count list
		// E.g. for two combined image samplers :
		// typeCounts[1].type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
		// typeCounts[1].descriptorCount = 2;

		// Create the global descriptor pool
		// All descriptors used in this example are allocated from this pool
		VkDescriptorPoolCreateInfo descriptorPoolCI{ VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO };
		descriptorPoolCI.poolSizeCount = 1;
		descriptorPoolCI.pPoolSizes = descriptorTypeCounts;
		// Set the max. number of descriptor sets that can be requested from this pool (requesting beyond this limit will result in an error)
		// Our sample will create one set per uniform buffer per frame
		descriptorPoolCI.maxSets = MAX_CONCURRENT_FRAMES;
		VK_CHECK_RESULT(vkCreateDescriptorPool(device, &descriptorPoolCI, nullptr, &descriptorPool));

		// Descriptor set layouts define the interface between our application and the shader
		// Basically connects the different shader stages to descriptors for binding uniform buffers, image samplers, etc.
		// So every shader binding should map to one descriptor set layout binding
		// Binding 0: Uniform buffer (Vertex shader)
		VkDescriptorSetLayoutBinding layoutBinding{};
		layoutBinding.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
		layoutBinding.descriptorCount = 1;
		layoutBinding.stageFlags = VK_SHADER_STAGE_VERTEX_BIT;

		VkDescriptorSetLayoutCreateInfo descriptorLayoutCI{ VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO };
		descriptorLayoutCI.bindingCount = 1;
		descriptorLayoutCI.pBindings = &layoutBinding;
		VK_CHECK_RESULT(vkCreateDescriptorSetLayout(device, &descriptorLayoutCI, nullptr, &descriptorSetLayout));

		// Where the descriptor set layout is the interface, the descriptor set points to actual data
		// Descriptors that are changed per frame need to be multiplied, so we can update descriptor n+1 while n is still used by the GPU, so we create one per max frame in flight
		for (uint32_t i = 0; i < MAX_CONCURRENT_FRAMES; i++) {
			VkDescriptorSetAllocateInfo allocInfo{ VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO };
			allocInfo.descriptorPool = descriptorPool;
			allocInfo.descriptorSetCount = 1;
			allocInfo.pSetLayouts = &descriptorSetLayout;
			VK_CHECK_RESULT(vkAllocateDescriptorSets(device, &allocInfo, &uniformBuffers[i].descriptorSet));

			// Update the descriptor set determining the shader binding points
			// For every binding point used in a shader there needs to be one
			// descriptor set matching that binding point
			VkWriteDescriptorSet writeDescriptorSet{ VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET };

			// The buffer's information is passed using a descriptor info structure
			VkDescriptorBufferInfo bufferInfo{};
			bufferInfo.buffer = uniformBuffers[i].handle;
			bufferInfo.range = sizeof(ShaderData);

			// Binding 0 : Uniform buffer
			writeDescriptorSet.dstSet = uniformBuffers[i].descriptorSet;
			writeDescriptorSet.descriptorCount = 1;
			writeDescriptorSet.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
			writeDescriptorSet.pBufferInfo = &bufferInfo;
			writeDescriptorSet.dstBinding = 0;
			vkUpdateDescriptorSets(device, 1, &writeDescriptorSet, 0, nullptr);
		}
	}

	// Create the depth (and stencil) buffer attachments
	// Note: Override of virtual function in the base class and called from within VulkanExampleBase::prepare
	void setupDepthStencil()
	{
		// Create an optimal image used as the depth stencil attachment
		VkImageCreateInfo imageCI{ VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO };
		imageCI.imageType = VK_IMAGE_TYPE_2D;
		imageCI.format = depthFormat;
		// Use example's height and width
		imageCI.extent = { width, height, 1 };
		imageCI.mipLevels = 1;
		imageCI.arrayLayers = 1;
		imageCI.samples = VK_SAMPLE_COUNT_1_BIT;
		imageCI.tiling = VK_IMAGE_TILING_OPTIMAL;
		imageCI.usage = VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
		imageCI.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
		VK_CHECK_RESULT(vkCreateImage(device, &imageCI, nullptr, &depthStencil.image));

		// Allocate memory for the image (device local) and bind it to our image
		VkMemoryAllocateInfo memAlloc{ VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO };
		VkMemoryRequirements memReqs;
		vkGetImageMemoryRequirements(device, depthStencil.image, &memReqs);
		memAlloc.allocationSize = memReqs.size;
		memAlloc.memoryTypeIndex = getMemoryTypeIndex(memReqs.memoryTypeBits, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
		VK_CHECK_RESULT(vkAllocateMemory(device, &memAlloc, nullptr, &depthStencil.memory));
		VK_CHECK_RESULT(vkBindImageMemory(device, depthStencil.image, depthStencil.memory, 0));

		// Create a view for the depth stencil image
		// Images aren't directly accessed in Vulkan, but rather through views described by a subresource range
		// This allows for multiple views of one image with differing ranges (e.g. for different layers)
		VkImageViewCreateInfo depthStencilViewCI{ VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO };
		depthStencilViewCI.viewType = VK_IMAGE_VIEW_TYPE_2D;
		depthStencilViewCI.format = depthFormat;
		depthStencilViewCI.subresourceRange = {};
		depthStencilViewCI.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
		// Stencil aspect should only be set on depth + stencil formats (VK_FORMAT_D16_UNORM_S8_UINT..VK_FORMAT_D32_SFLOAT_S8_UINT)
		if (depthFormat >= VK_FORMAT_D16_UNORM_S8_UINT) {
			depthStencilViewCI.subresourceRange.aspectMask |= VK_IMAGE_ASPECT_STENCIL_BIT;
		}
		depthStencilViewCI.subresourceRange.baseMipLevel = 0;
		depthStencilViewCI.subresourceRange.levelCount = 1;
		depthStencilViewCI.subresourceRange.baseArrayLayer = 0;
		depthStencilViewCI.subresourceRange.layerCount = 1;
		depthStencilViewCI.image = depthStencil.image;
		VK_CHECK_RESULT(vkCreateImageView(device, &depthStencilViewCI, nullptr, &depthStencil.view));
	}

	// Vulkan loads its shaders from an immediate binary representation called SPIR-V
	// Shaders are compiled offline from e.g. GLSL using the reference glslang compiler
	// This function loads such a shader from a binary file and returns a shader module structure
	VkShaderModule loadSPIRVShader(std::string filename)
	{
		size_t shaderSize;
		char* shaderCode{ nullptr };

#if defined(__ANDROID__)
		// Load shader from compressed asset
		AAsset* asset = AAssetManager_open(androidApp->activity->assetManager, filename.c_str(), AASSET_MODE_STREAMING);
		assert(asset);
		shaderSize = AAsset_getLength(asset);
		assert(shaderSize > 0);

		shaderCode = new char[shaderSize];
		AAsset_read(asset, shaderCode, shaderSize);
		AAsset_close(asset);
#else
		std::ifstream is(filename, std::ios::binary | std::ios::in | std::ios::ate);

		if (is.is_open()) {
			shaderSize = is.tellg();
			is.seekg(0, std::ios::beg);
			// Copy file contents into a buffer
			shaderCode = new char[shaderSize];
			is.read(shaderCode, shaderSize);
			is.close();
			assert(shaderSize > 0);
		}
#endif
		if (shaderCode) {
			// Create a new shader module that will be used for pipeline creation
			VkShaderModuleCreateInfo shaderModuleCI{ VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO };
			shaderModuleCI.codeSize = shaderSize;
			shaderModuleCI.pCode = (uint32_t*)shaderCode;

			VkShaderModule shaderModule;
			VK_CHECK_RESULT(vkCreateShaderModule(device, &shaderModuleCI, nullptr, &shaderModule));

			delete[] shaderCode;

			return shaderModule;
		} else {
			std::cerr << "Error: Could not open shader file \"" << filename << "\"" << std::endl;
			return VK_NULL_HANDLE;
		}
	}

	void createPipeline()
	{
		// The pipeline layout is the interface telling the pipeline what type of descriptors will later be bound
		VkPipelineLayoutCreateInfo pipelineLayoutCI{ VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO };
		pipelineLayoutCI.setLayoutCount = 1;
		pipelineLayoutCI.pSetLayouts = &descriptorSetLayout;
		VK_CHECK_RESULT(vkCreatePipelineLayout(device, &pipelineLayoutCI, nullptr, &pipelineLayout));

		// Create the graphics pipeline used in this example
		// Vulkan uses the concept of rendering pipelines to encapsulate fixed states, replacing OpenGL's complex state machine
		// A pipeline is then stored and hashed on the GPU making pipeline changes very fast

		VkGraphicsPipelineCreateInfo pipelineCI{ VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO };
		// The layout used for this pipeline (can be shared among multiple pipelines using the same layout)
		pipelineCI.layout = pipelineLayout;

		// Construct the different states making up the pipeline

		// Input assembly state describes how primitives are assembled
		// This pipeline will assemble vertex data as a triangle lists (though we only use one triangle)
		VkPipelineInputAssemblyStateCreateInfo inputAssemblyStateCI{ VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO };
		inputAssemblyStateCI.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;

		// Rasterization state
		VkPipelineRasterizationStateCreateInfo rasterizationStateCI{ VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO };
		rasterizationStateCI.polygonMode = VK_POLYGON_MODE_FILL;
		rasterizationStateCI.cullMode = VK_CULL_MODE_NONE;
		rasterizationStateCI.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE;
		rasterizationStateCI.depthClampEnable = VK_FALSE;
		rasterizationStateCI.rasterizerDiscardEnable = VK_FALSE;
		rasterizationStateCI.depthBiasEnable = VK_FALSE;
		rasterizationStateCI.lineWidth = 1.0f;

		// Color blend state describes how blend factors are calculated (if used)
		// We need one blend attachment state per color attachment (even if blending is not used)
		VkPipelineColorBlendAttachmentState blendAttachmentState{};
		blendAttachmentState.colorWriteMask = 0xf;
		blendAttachmentState.blendEnable = VK_FALSE;
		VkPipelineColorBlendStateCreateInfo colorBlendStateCI{ VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO };
		colorBlendStateCI.attachmentCount = 1;
		colorBlendStateCI.pAttachments = &blendAttachmentState;

		// Viewport state sets the number of viewports and scissor used in this pipeline
		// Note: This is actually overridden by the dynamic states (see below)
		VkPipelineViewportStateCreateInfo viewportStateCI{ VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO };
		viewportStateCI.viewportCount = 1;
		viewportStateCI.scissorCount = 1;

		// Enable dynamic states
		// Most states are baked into the pipeline, but there is somee state that can be dynamically changed within the command buffer to mak e things easuer
		// To be able to change these we need do specify which dynamic states will be changed using this pipeline. Their actual states are set later on in the command buffer
		std::vector<VkDynamicState> dynamicStateEnables = { VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR };
		VkPipelineDynamicStateCreateInfo dynamicStateCI{ VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO };
		dynamicStateCI.pDynamicStates = dynamicStateEnables.data();
		dynamicStateCI.dynamicStateCount = static_cast<uint32_t>(dynamicStateEnables.size());

		// Depth and stencil state containing depth and stencil compare and test operations
		// We only use depth tests and want depth tests and writes to be enabled and compare with less or equal
		VkPipelineDepthStencilStateCreateInfo depthStencilStateCI{ VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO };
		depthStencilStateCI.depthTestEnable = VK_TRUE;
		depthStencilStateCI.depthWriteEnable = VK_TRUE;
		depthStencilStateCI.depthCompareOp = VK_COMPARE_OP_LESS_OR_EQUAL;
		depthStencilStateCI.depthBoundsTestEnable = VK_FALSE;
		depthStencilStateCI.back.failOp = VK_STENCIL_OP_KEEP;
		depthStencilStateCI.back.passOp = VK_STENCIL_OP_KEEP;
		depthStencilStateCI.back.compareOp = VK_COMPARE_OP_ALWAYS;
		depthStencilStateCI.stencilTestEnable = VK_FALSE;
		depthStencilStateCI.front = depthStencilStateCI.back;

		// This example does not make use of multi sampling (for anti-aliasing), the state must still be set and passed to the pipeline
		VkPipelineMultisampleStateCreateInfo multisampleStateCI{ VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO };
		multisampleStateCI.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;

		// Vertex input descriptions
		// Specifies the vertex input parameters for a pipeline

		// Vertex input binding
		// This example uses a single vertex input binding at binding point 0 (see vkCmdBindVertexBuffers)
		VkVertexInputBindingDescription vertexInputBinding{};
		vertexInputBinding.binding = 0;
		vertexInputBinding.stride = sizeof(Vertex);
		vertexInputBinding.inputRate = VK_VERTEX_INPUT_RATE_VERTEX;

		// Input attribute bindings describe shader attribute locations and memory layouts
		std::array<VkVertexInputAttributeDescription, 2> vertexInputAttributs;
		// These match the following shader layout (see triangle.vert):
		//	layout (location = 0) in vec3 inPos;
		//	layout (location = 1) in vec3 inColor;
		// Attribute location 0: Position
		vertexInputAttributs[0].binding = 0;
		vertexInputAttributs[0].location = 0;
		// Position attribute is three 32 bit signed (SFLOAT) floats (R32 G32 B32)
		vertexInputAttributs[0].format = VK_FORMAT_R32G32B32_SFLOAT;
		vertexInputAttributs[0].offset = offsetof(Vertex, position);
		// Attribute location 1: Color
		vertexInputAttributs[1].binding = 0;
		vertexInputAttributs[1].location = 1;
		// Color attribute is three 32 bit signed (SFLOAT) floats (R32 G32 B32)
		vertexInputAttributs[1].format = VK_FORMAT_R32G32B32_SFLOAT;
		vertexInputAttributs[1].offset = offsetof(Vertex, color);

		// Vertex input state used for pipeline creation
		VkPipelineVertexInputStateCreateInfo vertexInputStateCI{ VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO };
		vertexInputStateCI.vertexBindingDescriptionCount = 1;
		vertexInputStateCI.pVertexBindingDescriptions = &vertexInputBinding;
		vertexInputStateCI.vertexAttributeDescriptionCount = 2;
		vertexInputStateCI.pVertexAttributeDescriptions = vertexInputAttributs.data();

		// Shaders
		std::array<VkPipelineShaderStageCreateInfo, 2> shaderStages{};

		// Vertex shader
		shaderStages[0].sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
		shaderStages[0].stage = VK_SHADER_STAGE_VERTEX_BIT;
		shaderStages[0].module = loadSPIRVShader(getShadersPath() + "triangle/triangle.vert.spv");
		shaderStages[0].pName = "main";
		assert(shaderStages[0].module != VK_NULL_HANDLE);

		// Fragment shader
		shaderStages[1].sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
		shaderStages[1].stage = VK_SHADER_STAGE_FRAGMENT_BIT;
		shaderStages[1].module = loadSPIRVShader(getShadersPath() + "triangle/triangle.frag.spv");
		shaderStages[1].pName = "main";
		assert(shaderStages[1].module != VK_NULL_HANDLE);

		// Set pipeline shader stage info
		pipelineCI.stageCount = static_cast<uint32_t>(shaderStages.size());
		pipelineCI.pStages = shaderStages.data();

		// Attachment information for dynamic rendering
		VkPipelineRenderingCreateInfoKHR pipelineRenderingCI{ VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO_KHR };
		pipelineRenderingCI.colorAttachmentCount = 1;
		pipelineRenderingCI.pColorAttachmentFormats = &swapChain.colorFormat;
		pipelineRenderingCI.depthAttachmentFormat = depthFormat;
		pipelineRenderingCI.stencilAttachmentFormat = depthFormat;

		// Assign the pipeline states to the pipeline creation info structure
		pipelineCI.pVertexInputState = &vertexInputStateCI;
		pipelineCI.pInputAssemblyState = &inputAssemblyStateCI;
		pipelineCI.pRasterizationState = &rasterizationStateCI;
		pipelineCI.pColorBlendState = &colorBlendStateCI;
		pipelineCI.pMultisampleState = &multisampleStateCI;
		pipelineCI.pViewportState = &viewportStateCI;
		pipelineCI.pDepthStencilState = &depthStencilStateCI;
		pipelineCI.pDynamicState = &dynamicStateCI;
		pipelineCI.pNext = &pipelineRenderingCI;

		// Create rendering pipeline using the specified states
		VK_CHECK_RESULT(vkCreateGraphicsPipelines(device, pipelineCache, 1, &pipelineCI, nullptr, &pipeline));

		// Shader modules can safely be destroyed when the pipeline has been created
		vkDestroyShaderModule(device, shaderStages[0].module, nullptr);
		vkDestroyShaderModule(device, shaderStages[1].module, nullptr);
	}

	void createUniformBuffers()
	{
		// Prepare and initialize the per-frame uniform buffer blocks containing shader uniforms
		// Single uniforms like in OpenGL are no longer present in Vulkan. All Shader uniforms are passed via uniform buffer blocks

		// Vertex shader uniform buffer block
		VkBufferCreateInfo bufferInfo{ VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO };
		VkMemoryAllocateInfo allocInfo{ VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO };
		allocInfo.allocationSize = 0;
		allocInfo.memoryTypeIndex = 0;

		bufferInfo.size = sizeof(ShaderData);
		// This buffer will be used as a uniform buffer
		bufferInfo.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;

		// Create the buffers
		for (uint32_t i = 0; i < MAX_CONCURRENT_FRAMES; i++) {
			VK_CHECK_RESULT(vkCreateBuffer(device, &bufferInfo, nullptr, &uniformBuffers[i].handle));
			// Get memory requirements including size, alignment and memory type
			VkMemoryRequirements memReqs;
			vkGetBufferMemoryRequirements(device, uniformBuffers[i].handle, &memReqs);
			allocInfo.allocationSize = memReqs.size;
			// Get the memory type index that supports host visible memory access
			// Most implementations offer multiple memory types and selecting the correct one to allocate memory from is crucial
			// We also want the buffer to be host coherent so we don't have to flush (or sync after every update).
			allocInfo.memoryTypeIndex = getMemoryTypeIndex(memReqs.memoryTypeBits, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
			// Allocate memory for the uniform buffer
			VK_CHECK_RESULT(vkAllocateMemory(device, &allocInfo, nullptr, &(uniformBuffers[i].memory)));
			// Bind memory to buffer
			VK_CHECK_RESULT(vkBindBufferMemory(device, uniformBuffers[i].handle, uniformBuffers[i].memory, 0));
			// We map the buffer once, so we can update it without having to map it again
			VK_CHECK_RESULT(vkMapMemory(device, uniformBuffers[i].memory, 0, sizeof(ShaderData), 0, (void**)&uniformBuffers[i].mapped));
		}

	}

	void prepare()
	{
		VulkanExampleBase::prepare();
		createSynchronizationPrimitives();
		createCommandBuffers();
		createVertexBuffer();
		createUniformBuffers();
		createDescriptors();
		createPipeline();
		prepared = true;
	}

	virtual void render() override
	{
		// Use a fence to wait until the command buffer has finished execution before using it again
		vkWaitForFences(device, 1, &waitFences[currentFrame], VK_TRUE, UINT64_MAX);
		VK_CHECK_RESULT(vkResetFences(device, 1, &waitFences[currentFrame]));

		// Get the next swap chain image from the implementation
		// Note that the implementation is free to return the images in any order, so we must use the acquire function and can't just cycle through the images/imageIndex on our own
		uint32_t imageIndex;
		VkResult result = vkAcquireNextImageKHR(device, swapChain.swapChain, UINT64_MAX, presentCompleteSemaphores[currentFrame], VK_NULL_HANDLE, &imageIndex);
		if (result == VK_ERROR_OUT_OF_DATE_KHR) {
			windowResize();
			return;
		} else if ((result != VK_SUCCESS) && (result != VK_SUBOPTIMAL_KHR)) {
			throw "Could not acquire the next swap chain image!";
		}

		// Update the uniform buffer for the next frame
		ShaderData shaderData{};
		shaderData.projectionMatrix = camera.matrices.perspective;
		shaderData.viewMatrix = camera.matrices.view;
		shaderData.modelMatrix = glm::mat4(1.0f);
		// Copy the current matrices to the current frame's uniform buffer. As we requested a host coherent memory type for the uniform buffer, the write is instantly visible to the GPU.
		memcpy(uniformBuffers[currentFrame].mapped, &shaderData, sizeof(ShaderData));

		// Build the command buffer for the next frame to render
		vkResetCommandBuffer(commandBuffers[currentFrame], 0);
		VkCommandBufferBeginInfo cmdBufInfo{ VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO };
		const VkCommandBuffer commandBuffer = commandBuffers[currentFrame];
		VK_CHECK_RESULT(vkBeginCommandBuffer(commandBuffer, &cmdBufInfo));

		// With dynamic rendering we need to explicitly add layout transitions by using barriers, this set of barriers prepares the color and depth images for output
		vks::tools::insertImageMemoryBarrier(commandBuffer, swapChain.buffers[imageIndex].image, 0, VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_ATTACHMENT_OPTIMAL, VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, VkImageSubresourceRange{ VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1 });
		vks::tools::insertImageMemoryBarrier(commandBuffer, depthStencil.image, 0, VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT, VK_IMAGE_LAYOUT_UNDEFINED, VK_IMAGE_LAYOUT_ATTACHMENT_OPTIMAL, VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, VkImageSubresourceRange{ VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT, 0, 1, 0, 1 });

		// New structures are used to define the attachments used in dynamic rendering
		// Color attachment
		VkRenderingAttachmentInfo colorAttachment{ VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
		colorAttachment.imageView = swapChain.buffers[imageIndex].view;
		colorAttachment.imageLayout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
		colorAttachment.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
		colorAttachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
		colorAttachment.clearValue.color = { 0.0f, 0.0f, 0.2f, 0.0f };
		// Depth/stencil attachment
		VkRenderingAttachmentInfo depthStencilAttachment{ VK_STRUCTURE_TYPE_RENDERING_ATTACHMENT_INFO };
		depthStencilAttachment.imageView = depthStencil.view;
		depthStencilAttachment.imageLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
		depthStencilAttachment.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
		depthStencilAttachment.storeOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
		depthStencilAttachment.clearValue.depthStencil = { 1.0f,  0 };

		VkRenderingInfo renderingInfo{ VK_STRUCTURE_TYPE_RENDERING_INFO_KHR };
		renderingInfo.renderArea = { 0, 0, width, height };
		renderingInfo.layerCount = 1;
		renderingInfo.colorAttachmentCount = 1;
		renderingInfo.pColorAttachments = &colorAttachment;
		renderingInfo.pDepthAttachment = &depthStencilAttachment;
		renderingInfo.pStencilAttachment = &depthStencilAttachment;

		// Start a dynamic rendering section
		vkCmdBeginRendering(commandBuffer, &renderingInfo);
		// Update dynamic viewport state
		VkViewport viewport{ 0.0f, 0.0f, (float)width, (float)height, 0.0f, 1.0f };
		vkCmdSetViewport(commandBuffer, 0, 1, &viewport);
		// Update dynamic scissor state
		VkRect2D scissor{ 0, 0, width, height };
		vkCmdSetScissor(commandBuffer, 0, 1, &scissor);
		// Bind descriptor set for the currrent frame's uniform buffer, so the shader uses the data from that buffer for this draw
		vkCmdBindDescriptorSets(commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipelineLayout, 0, 1, &uniformBuffers[currentFrame].descriptorSet, 0, nullptr);
		// The pipeline (state object) contains all states of the rendering pipeline, binding it will set all the states specified at pipeline creation time
		vkCmdBindPipeline(commandBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
		// Bind triangle vertex buffer (contains position and colors)
		VkDeviceSize offsets[1]{ 0 };
		vkCmdBindVertexBuffers(commandBuffer, 0, 1, &vertexBuffer.handle, offsets);
		// Bind triangle index buffer
		vkCmdBindIndexBuffer(commandBuffer, indexBuffer.handle, 0, VK_INDEX_TYPE_UINT32);
		// Draw indexed triangle
		vkCmdDrawIndexed(commandBuffer, indexCount, 1, 0, 0, 1);
		// Finish the current dynamic rendering section
		vkCmdEndRendering(commandBuffer);

		// This barrier prepares the color image for presentation, we don't need to care for the depth image
		vks::tools::insertImageMemoryBarrier(commandBuffer, swapChain.buffers[imageIndex].image, VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, 0, VK_IMAGE_LAYOUT_ATTACHMENT_OPTIMAL, VK_IMAGE_LAYOUT_PRESENT_SRC_KHR, VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT, VK_PIPELINE_STAGE_2_NONE, VkImageSubresourceRange{ VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1 });
		VK_CHECK_RESULT(vkEndCommandBuffer(commandBuffer));

		// Submit the command buffer to the graphics queue

		// Pipeline stage at which the queue submission will wait (via pWaitSemaphores)
		VkPipelineStageFlags waitStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
		// The submit info structure specifies a command buffer queue submission batch
		VkSubmitInfo submitInfo{ VK_STRUCTURE_TYPE_SUBMIT_INFO };
		submitInfo.pWaitDstStageMask = &waitStageMask;      // Pointer to the list of pipeline stages that the semaphore waits will occur at
		submitInfo.pCommandBuffers = &commandBuffer;		// Command buffers(s) to execute in this batch (submission)
		submitInfo.commandBufferCount = 1;                  // We submit a single command buffer

		// Semaphore to wait upon before the submitted command buffer starts executing
		submitInfo.pWaitSemaphores = &presentCompleteSemaphores[currentFrame]; 
		submitInfo.waitSemaphoreCount = 1;
		// Semaphore to be signaled when command buffers have completed
		submitInfo.pSignalSemaphores = &renderCompleteSemaphores[currentFrame];
		submitInfo.signalSemaphoreCount = 1;

		// Submit to the graphics queue passing a wait fence
		VK_CHECK_RESULT(vkQueueSubmit(queue, 1, &submitInfo, waitFences[currentFrame]));

		// Present the current frame buffer to the swap chain
		// Pass the semaphore signaled by the command buffer submission from the submit info as the wait semaphore for swap chain presentation
		// This ensures that the image is not presented to the windowing system until all commands have been submitted
		VkPresentInfoKHR presentInfo{ VK_STRUCTURE_TYPE_PRESENT_INFO_KHR };
		presentInfo.waitSemaphoreCount = 1;
		presentInfo.pWaitSemaphores = &renderCompleteSemaphores[currentFrame];
		presentInfo.swapchainCount = 1;
		presentInfo.pSwapchains = &swapChain.swapChain;
		presentInfo.pImageIndices = &imageIndex;
		result = vkQueuePresentKHR(queue, &presentInfo);
		if ((result == VK_ERROR_OUT_OF_DATE_KHR) || (result == VK_SUBOPTIMAL_KHR)) {
			windowResize();
		} else if (result != VK_SUCCESS) {
			throw "Could not present the image to the swap chain!";
		}

		// Select the next frame to render to, based on the max. no. of concurrent frames
		currentFrame = (currentFrame + 1) % MAX_CONCURRENT_FRAMES;
	}

	// Override these as otherwise the base class would generate frame buffers and render passes
	void setupFrameBuffer() override {}
	void setupRenderPass() override {}
};

// OS specific main entry points
// Most of the code base is shared for the different supported operating systems, but stuff like message handling differs

#if defined(_WIN32)
// Windows entry point
VulkanExample *vulkanExample;
LRESULT CALLBACK WndProc(HWND hWnd, UINT uMsg, WPARAM wParam, LPARAM lParam)
{
	if (vulkanExample != NULL)
	{
		vulkanExample->handleMessages(hWnd, uMsg, wParam, lParam);
	}
	return (DefWindowProc(hWnd, uMsg, wParam, lParam));
}
int APIENTRY WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR pCmdLine, int nCmdShow)
{
	for (size_t i = 0; i < __argc; i++) { VulkanExample::args.push_back(__argv[i]); };
	vulkanExample = new VulkanExample();
	vulkanExample->initVulkan();
	vulkanExample->setupWindow(hInstance, WndProc);
	vulkanExample->prepare();
	vulkanExample->renderLoop();
	delete(vulkanExample);
	return 0;
}

#elif defined(__ANDROID__)
// Android entry point
VulkanExample *vulkanExample;
void android_main(android_app* state)
{
	vulkanExample = new VulkanExample();
	state->userData = vulkanExample;
	state->onAppCmd = VulkanExample::handleAppCommand;
	state->onInputEvent = VulkanExample::handleAppInput;
	androidApp = state;
	vulkanExample->renderLoop();
	delete(vulkanExample);
}
#elif defined(_DIRECT2DISPLAY)

// Linux entry point with direct to display wsi
// Direct to Displays (D2D) is used on embedded platforms
VulkanExample *vulkanExample;
static void handleEvent()
{
}
int main(const int argc, const char *argv[])
{
	for (size_t i = 0; i < argc; i++) { VulkanExample::args.push_back(argv[i]); };
	vulkanExample = new VulkanExample();
	vulkanExample->initVulkan();
	vulkanExample->prepare();
	vulkanExample->renderLoop();
	delete(vulkanExample);
	return 0;
}
#elif defined(VK_USE_PLATFORM_DIRECTFB_EXT)
VulkanExample *vulkanExample;
static void handleEvent(const DFBWindowEvent *event)
{
	if (vulkanExample != NULL)
	{
		vulkanExample->handleEvent(event);
	}
}
int main(const int argc, const char *argv[])
{
	for (size_t i = 0; i < argc; i++) { VulkanExample::args.push_back(argv[i]); };
	vulkanExample = new VulkanExample();
	vulkanExample->initVulkan();
	vulkanExample->setupWindow();
	vulkanExample->prepare();
	vulkanExample->renderLoop();
	delete(vulkanExample);
	return 0;
}
#elif defined(VK_USE_PLATFORM_WAYLAND_KHR)
VulkanExample *vulkanExample;
int main(const int argc, const char *argv[])
{
	for (size_t i = 0; i < argc; i++) { VulkanExample::args.push_back(argv[i]); };
	vulkanExample = new VulkanExample();
	vulkanExample->initVulkan();
	vulkanExample->setupWindow();
	vulkanExample->prepare();
	vulkanExample->renderLoop();
	delete(vulkanExample);
	return 0;
}
#elif defined(__linux__) || defined(__FreeBSD__)

// Linux entry point
VulkanExample *vulkanExample;
#if defined(VK_USE_PLATFORM_XCB_KHR)
static void handleEvent(const xcb_generic_event_t *event)
{
	if (vulkanExample != NULL)
	{
		vulkanExample->handleEvent(event);
	}
}
#else
static void handleEvent()
{
}
#endif
int main(const int argc, const char *argv[])
{
	for (size_t i = 0; i < argc; i++) { VulkanExample::args.push_back(argv[i]); };
	vulkanExample = new VulkanExample();
	vulkanExample->initVulkan();
	vulkanExample->setupWindow();
	vulkanExample->prepare();
	vulkanExample->renderLoop();
	delete(vulkanExample);
	return 0;
}
#elif (defined(VK_USE_PLATFORM_MACOS_MVK) || defined(VK_USE_PLATFORM_METAL_EXT)) && defined(VK_EXAMPLE_XCODE_GENERATED)
VulkanExample *vulkanExample;
int main(const int argc, const char *argv[])
{
	@autoreleasepool
	{
		for (size_t i = 0; i < argc; i++) { VulkanExample::args.push_back(argv[i]); };
		vulkanExample = new VulkanExample();
		vulkanExample->initVulkan();
		vulkanExample->setupWindow(nullptr);
		vulkanExample->prepare();
		vulkanExample->renderLoop();
		delete(vulkanExample);
	}
	return 0;
}
#elif defined(VK_USE_PLATFORM_SCREEN_QNX)
VULKAN_EXAMPLE_MAIN()
#endif