/*
* Vulkan Example - Compute shader N-body simulation using two passes and shared compute shader memory
*
* Copyright (C) 2016 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 <vector>
#include <random>

#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"
#include "VulkanTexture.hpp"

#define VERTEX_BUFFER_BIND_ID 0
#define ENABLE_VALIDATION false
#if defined(__ANDROID__)
// Lower particle count on Android for performance reasons
#define PARTICLES_PER_ATTRACTOR 3 * 1024
#else
#define PARTICLES_PER_ATTRACTOR 4 * 1024
#endif

class VulkanExample : public VulkanExampleBase
{
public:
	uint32_t numParticles;

	struct {
		vks::Texture2D particle;
		vks::Texture2D gradient;
	} textures;

	struct {
		VkPipelineVertexInputStateCreateInfo inputState;
		std::vector<VkVertexInputBindingDescription> bindingDescriptions;
		std::vector<VkVertexInputAttributeDescription> attributeDescriptions;
	} vertices;

	// Resources for the graphics part of the example
	struct {
		vks::Buffer uniformBuffer;					// Contains scene matrices
		VkDescriptorSetLayout descriptorSetLayout;	// Particle system rendering shader binding layout
		VkDescriptorSet descriptorSet;				// Particle system rendering shader bindings
		VkPipelineLayout pipelineLayout;			// Layout of the graphics pipeline
		VkPipeline pipeline;						// Particle rendering pipeline
		struct {
			glm::mat4 projection;
			glm::mat4 view;
			glm::vec2 screenDim;
		} ubo;
	} graphics;

	// Resources for the compute part of the example
	struct {
		vks::Buffer storageBuffer;					// (Shader) storage buffer object containing the particles
		vks::Buffer uniformBuffer;					// Uniform buffer object containing particle system parameters
		VkQueue queue;								// Separate queue for compute commands (queue family may differ from the one used for graphics)
		VkCommandPool commandPool;					// Use a separate command pool (queue family may differ from the one used for graphics)
		VkCommandBuffer commandBuffer;				// Command buffer storing the dispatch commands and barriers
		VkFence fence;								// Synchronization fence to avoid rewriting compute CB if still in use
		VkDescriptorSetLayout descriptorSetLayout;	// Compute shader binding layout
		VkDescriptorSet descriptorSet;				// Compute shader bindings
		VkPipelineLayout pipelineLayout;			// Layout of the compute pipeline
		VkPipeline pipelineCalculate;				// Compute pipeline for N-Body velocity calculation (1st pass)
		VkPipeline pipelineIntegrate;				// Compute pipeline for euler integration (2nd pass)
		VkPipeline blur;
		VkPipelineLayout pipelineLayoutBlur;
		VkDescriptorSetLayout descriptorSetLayoutBlur;
		VkDescriptorSet descriptorSetBlur;
		struct computeUBO {							// Compute shader uniform block object
			float deltaT;							//		Frame delta time
			float destX;							//		x position of the attractor
			float destY;							//		y position of the attractor
			int32_t particleCount;
		} ubo;
	} compute;

	// SSBO particle declaration
	struct Particle {
		glm::vec4 pos;								// xyz = position, w = mass
		glm::vec4 vel;								// xyz = velocity, w = gradient texture position
	};

	VulkanExample() : VulkanExampleBase(ENABLE_VALIDATION)
	{
		title = "Compute shader N-body system";
		settings.overlay = true;
		camera.type = Camera::CameraType::lookat;
		camera.setPerspective(60.0f, (float)width / (float)height, 0.1f, 512.0f);
		camera.setRotation(glm::vec3(-26.0f, 75.0f, 0.0f));
		camera.setTranslation(glm::vec3(0.0f, 0.0f, -14.0f));
		camera.movementSpeed = 2.5f;
	}

	~VulkanExample()
	{
		// Graphics
		graphics.uniformBuffer.destroy();
		vkDestroyPipeline(device, graphics.pipeline, nullptr);
		vkDestroyPipelineLayout(device, graphics.pipelineLayout, nullptr);
		vkDestroyDescriptorSetLayout(device, graphics.descriptorSetLayout, nullptr);

		// Compute
		compute.storageBuffer.destroy();
		compute.uniformBuffer.destroy();
		vkDestroyPipelineLayout(device, compute.pipelineLayout, nullptr);
		vkDestroyDescriptorSetLayout(device, compute.descriptorSetLayout, nullptr);
		vkDestroyPipeline(device, compute.pipelineCalculate, nullptr);
		vkDestroyPipeline(device, compute.pipelineIntegrate, nullptr);
		vkDestroyFence(device, compute.fence, nullptr);
		vkDestroyCommandPool(device, compute.commandPool, nullptr);

		textures.particle.destroy();
		textures.gradient.destroy();
	}

	void loadAssets()
	{
		textures.particle.loadFromFile(getAssetPath() + "textures/particle01_rgba.ktx", VK_FORMAT_R8G8B8A8_UNORM, vulkanDevice, queue);
		textures.gradient.loadFromFile(getAssetPath() + "textures/particle_gradient_rgba.ktx", VK_FORMAT_R8G8B8A8_UNORM, vulkanDevice, queue);
	}

	void buildCommandBuffers()
	{
		// Destroy command buffers if already present
		if (!checkCommandBuffers())
		{
			destroyCommandBuffers();
			createCommandBuffers();
		}

		VkCommandBufferBeginInfo cmdBufInfo = vks::initializers::commandBufferBeginInfo();

		VkClearValue clearValues[2];
		clearValues[0].color = { {0.0f, 0.0f, 0.0f, 1.0f} };
		clearValues[1].depthStencil = { 1.0f, 0 };

		VkRenderPassBeginInfo renderPassBeginInfo = vks::initializers::renderPassBeginInfo();
		renderPassBeginInfo.renderPass = renderPass;
		renderPassBeginInfo.renderArea.offset.x = 0;
		renderPassBeginInfo.renderArea.offset.y = 0;
		renderPassBeginInfo.renderArea.extent.width = width;
		renderPassBeginInfo.renderArea.extent.height = height;
		renderPassBeginInfo.clearValueCount = 2;
		renderPassBeginInfo.pClearValues = clearValues;

		for (int32_t i = 0; i < drawCmdBuffers.size(); ++i)
		{
			// Set target frame buffer
			renderPassBeginInfo.framebuffer = frameBuffers[i];

			VK_CHECK_RESULT(vkBeginCommandBuffer(drawCmdBuffers[i], &cmdBufInfo));

			// Draw the particle system using the update vertex buffer

			vkCmdBeginRenderPass(drawCmdBuffers[i], &renderPassBeginInfo, VK_SUBPASS_CONTENTS_INLINE);

			VkViewport viewport = vks::initializers::viewport((float)width, (float)height, 0.0f, 1.0f);
			vkCmdSetViewport(drawCmdBuffers[i], 0, 1, &viewport);

			VkRect2D scissor = vks::initializers::rect2D(width, height, 0, 0);
			vkCmdSetScissor(drawCmdBuffers[i], 0, 1, &scissor);

			vkCmdBindPipeline(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, graphics.pipeline);
			vkCmdBindDescriptorSets(drawCmdBuffers[i], VK_PIPELINE_BIND_POINT_GRAPHICS, graphics.pipelineLayout, 0, 1, &graphics.descriptorSet, 0, NULL);

			VkDeviceSize offsets[1] = { 0 };
			vkCmdBindVertexBuffers(drawCmdBuffers[i], VERTEX_BUFFER_BIND_ID, 1, &compute.storageBuffer.buffer, offsets);
			vkCmdDraw(drawCmdBuffers[i], numParticles, 1, 0, 0);

			vkCmdEndRenderPass(drawCmdBuffers[i]);

			VK_CHECK_RESULT(vkEndCommandBuffer(drawCmdBuffers[i]));
		}

	}

	void buildComputeCommandBuffer()
	{
		VkCommandBufferBeginInfo cmdBufInfo = vks::initializers::commandBufferBeginInfo();

		VK_CHECK_RESULT(vkBeginCommandBuffer(compute.commandBuffer, &cmdBufInfo));

		// Compute particle movement

		// Add memory barrier to ensure that the (graphics) vertex shader has fetched attributes before compute starts to write to the buffer
		VkBufferMemoryBarrier bufferBarrier = vks::initializers::bufferMemoryBarrier();
		bufferBarrier.buffer = compute.storageBuffer.buffer;
		bufferBarrier.size = compute.storageBuffer.descriptor.range;
		bufferBarrier.srcAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;						// Vertex shader invocations have finished reading from the buffer
		bufferBarrier.dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT;								// Compute shader wants to write to the buffer
		// Transfer ownership if compute and graphics queue familiy indices differ
		bufferBarrier.srcQueueFamilyIndex = vulkanDevice->queueFamilyIndices.graphics;			
		bufferBarrier.dstQueueFamilyIndex = vulkanDevice->queueFamilyIndices.compute;			

		vkCmdPipelineBarrier(
			compute.commandBuffer,
			VK_PIPELINE_STAGE_VERTEX_SHADER_BIT,
			VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
			VK_FLAGS_NONE,
			0, nullptr,
			1, &bufferBarrier,
			0, nullptr);

		vkCmdBindPipeline(compute.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, compute.pipelineCalculate);
		vkCmdBindDescriptorSets(compute.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, compute.pipelineLayout, 0, 1, &compute.descriptorSet, 0, 0);

		// First pass: Calculate particle movement
		// -------------------------------------------------------------------------------------------------------
		vkCmdDispatch(compute.commandBuffer, numParticles / 256, 1, 1);

		// Add memory barrier to ensure that compute shader has finished writing to the buffer 
		bufferBarrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;								// Compute shader has finished writes to the buffer
		bufferBarrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;					
		bufferBarrier.buffer = compute.storageBuffer.buffer;
		bufferBarrier.size = compute.storageBuffer.descriptor.range;
		// No ownership transfer necessary
		bufferBarrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; 
		bufferBarrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;

		vkCmdPipelineBarrier(
			compute.commandBuffer,
			VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
			VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
			VK_FLAGS_NONE,
			0, nullptr,
			1, &bufferBarrier,
			0, nullptr);

		// Second pass: Integrate particles
		// -------------------------------------------------------------------------------------------------------
		vkCmdBindPipeline(compute.commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, compute.pipelineIntegrate);
		vkCmdDispatch(compute.commandBuffer, numParticles / 256, 1, 1);

		// Add memory barrier to ensure that compute shader has finished writing to the buffer
		// Without this the (rendering) vertex shader may display incomplete results (partial data from last frame) 
		bufferBarrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;								// Compute shader has finished writes to the buffer
		bufferBarrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;						// Vertex shader invocations want to read from the buffer
		bufferBarrier.buffer = compute.storageBuffer.buffer;
		bufferBarrier.size = compute.storageBuffer.descriptor.range;
		// Transfer ownership if compute and graphics queue familiy indices differ
		bufferBarrier.srcQueueFamilyIndex = vulkanDevice->queueFamilyIndices.compute;
		bufferBarrier.dstQueueFamilyIndex = vulkanDevice->queueFamilyIndices.graphics;

		vkCmdPipelineBarrier(
			compute.commandBuffer,
			VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
			VK_PIPELINE_STAGE_VERTEX_SHADER_BIT,
			VK_FLAGS_NONE,
			0, nullptr,
			1, &bufferBarrier,
			0, nullptr);

		vkEndCommandBuffer(compute.commandBuffer);
	}

	// Setup and fill the compute shader storage buffers containing the particles
	void prepareStorageBuffers()
	{
#if 0
		std::vector<glm::vec3> attractors = {
			glm::vec3(2.5f, 1.5f, 0.0f),
			glm::vec3(-2.5f, -1.5f, 0.0f),
		};
#else
		std::vector<glm::vec3> attractors = {
			glm::vec3(5.0f, 0.0f, 0.0f),
			glm::vec3(-5.0f, 0.0f, 0.0f),
			glm::vec3(0.0f, 0.0f, 5.0f),
			glm::vec3(0.0f, 0.0f, -5.0f),
			glm::vec3(0.0f, 4.0f, 0.0f),
			glm::vec3(0.0f, -8.0f, 0.0f),
		};
#endif

		numParticles = static_cast<uint32_t>(attractors.size()) * PARTICLES_PER_ATTRACTOR;

		// Initial particle positions
		std::vector<Particle> particleBuffer(numParticles);

		std::mt19937 rndGen(static_cast<uint32_t>(time(0)));
		std::normal_distribution<float> rndDist(0.0f, 1.0f);

		for (uint32_t i = 0; i < static_cast<uint32_t>(attractors.size()); i++)
		{
			for (uint32_t j = 0; j < PARTICLES_PER_ATTRACTOR; j++)
			{
				Particle &particle = particleBuffer[i * PARTICLES_PER_ATTRACTOR + j];

				// First particle in group as heavy center of gravity
				if (j == 0)
				{
					particle.pos = glm::vec4(attractors[i] * 1.5f, 90000.0f);
					particle.vel = glm::vec4(glm::vec4(0.0f));
				}
				else
				{
					// Position					
					glm::vec3 position(attractors[i] + glm::vec3(rndDist(rndGen), rndDist(rndGen), rndDist(rndGen)) * 0.75f);
					float len = glm::length(glm::normalize(position - attractors[i]));
					position.y *= 2.0f - (len * len);

					// Velocity
					glm::vec3 angular = glm::vec3(0.5f, 1.5f, 0.5f) * (((i % 2) == 0) ? 1.0f : -1.0f);
					glm::vec3 velocity = glm::cross((position - attractors[i]), angular) + glm::vec3(rndDist(rndGen), rndDist(rndGen), rndDist(rndGen) * 0.025f);

					float mass = (rndDist(rndGen) * 0.5f + 0.5f) * 75.0f;
					particle.pos = glm::vec4(position, mass);
					particle.vel = glm::vec4(velocity, 0.0f);
				}

				// Color gradient offset
				particle.vel.w = (float)i * 1.0f / static_cast<uint32_t>(attractors.size());
			}
		}

		compute.ubo.particleCount = numParticles;

		VkDeviceSize storageBufferSize = particleBuffer.size() * sizeof(Particle);

		// Staging
		// SSBO won't be changed on the host after upload so copy to device local memory 

		vks::Buffer stagingBuffer;

		vulkanDevice->createBuffer(
			VK_BUFFER_USAGE_TRANSFER_SRC_BIT,
			VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
			&stagingBuffer,
			storageBufferSize,
			particleBuffer.data());

		vulkanDevice->createBuffer(
			// The SSBO will be used as a storage buffer for the compute pipeline and as a vertex buffer in the graphics pipeline
			VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
			VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
			&compute.storageBuffer,
			storageBufferSize);

		// Copy to staging buffer
		VkCommandBuffer copyCmd = VulkanExampleBase::createCommandBuffer(VK_COMMAND_BUFFER_LEVEL_PRIMARY, true);
		VkBufferCopy copyRegion = {};
		copyRegion.size = storageBufferSize;
		vkCmdCopyBuffer(copyCmd, stagingBuffer.buffer, compute.storageBuffer.buffer, 1, &copyRegion);
		VulkanExampleBase::flushCommandBuffer(copyCmd, queue, true);

		stagingBuffer.destroy();

		// Binding description
		vertices.bindingDescriptions.resize(1);
		vertices.bindingDescriptions[0] =
			vks::initializers::vertexInputBindingDescription(
				VERTEX_BUFFER_BIND_ID,
				sizeof(Particle),
				VK_VERTEX_INPUT_RATE_VERTEX);

		// Attribute descriptions
		// Describes memory layout and shader positions
		vertices.attributeDescriptions.resize(2);
		// Location 0 : Position
		vertices.attributeDescriptions[0] =
			vks::initializers::vertexInputAttributeDescription(
				VERTEX_BUFFER_BIND_ID,
				0,
				VK_FORMAT_R32G32B32A32_SFLOAT,
				offsetof(Particle, pos));
		// Location 1 : Velocity (used for gradient lookup)
		vertices.attributeDescriptions[1] =
			vks::initializers::vertexInputAttributeDescription(
				VERTEX_BUFFER_BIND_ID,
				1,
				VK_FORMAT_R32G32B32A32_SFLOAT,
				offsetof(Particle, vel));

		// Assign to vertex buffer
		vertices.inputState = vks::initializers::pipelineVertexInputStateCreateInfo();
		vertices.inputState.vertexBindingDescriptionCount = static_cast<uint32_t>(vertices.bindingDescriptions.size());
		vertices.inputState.pVertexBindingDescriptions = vertices.bindingDescriptions.data();
		vertices.inputState.vertexAttributeDescriptionCount = static_cast<uint32_t>(vertices.attributeDescriptions.size());
		vertices.inputState.pVertexAttributeDescriptions = vertices.attributeDescriptions.data();
	}

	void setupDescriptorPool()
	{
		std::vector<VkDescriptorPoolSize> poolSizes =
		{
			vks::initializers::descriptorPoolSize(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, 2),
			vks::initializers::descriptorPoolSize(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 1),
			vks::initializers::descriptorPoolSize(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 2)
		};

		VkDescriptorPoolCreateInfo descriptorPoolInfo =
			vks::initializers::descriptorPoolCreateInfo(
				static_cast<uint32_t>(poolSizes.size()),
				poolSizes.data(),
				2);

		VK_CHECK_RESULT(vkCreateDescriptorPool(device, &descriptorPoolInfo, nullptr, &descriptorPool));
	}

	void setupDescriptorSetLayout()
	{
		std::vector<VkDescriptorSetLayoutBinding> setLayoutBindings;
		setLayoutBindings = {
			vks::initializers::descriptorSetLayoutBinding(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, VK_SHADER_STAGE_FRAGMENT_BIT, 0),
			vks::initializers::descriptorSetLayoutBinding(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, VK_SHADER_STAGE_FRAGMENT_BIT, 1),
			vks::initializers::descriptorSetLayoutBinding(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, VK_SHADER_STAGE_VERTEX_BIT, 2),
		};

		VkDescriptorSetLayoutCreateInfo descriptorLayout =
			vks::initializers::descriptorSetLayoutCreateInfo(
				setLayoutBindings.data(),
				static_cast<uint32_t>(setLayoutBindings.size()));

		VK_CHECK_RESULT(vkCreateDescriptorSetLayout(device, &descriptorLayout, nullptr, &graphics.descriptorSetLayout));

		VkPipelineLayoutCreateInfo pipelineLayoutCreateInfo =
			vks::initializers::pipelineLayoutCreateInfo(
				&graphics.descriptorSetLayout,
				1);

		VK_CHECK_RESULT(vkCreatePipelineLayout(device, &pipelineLayoutCreateInfo, nullptr, &graphics.pipelineLayout));
	}

	void setupDescriptorSet()
	{
		VkDescriptorSetAllocateInfo allocInfo =
			vks::initializers::descriptorSetAllocateInfo(
				descriptorPool,
				&graphics.descriptorSetLayout,
				1);

		VK_CHECK_RESULT(vkAllocateDescriptorSets(device, &allocInfo, &graphics.descriptorSet));

		std::vector<VkWriteDescriptorSet> writeDescriptorSets;
		writeDescriptorSets = {
			vks::initializers::writeDescriptorSet(graphics.descriptorSet, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 0, &textures.particle.descriptor),
			vks::initializers::writeDescriptorSet(graphics.descriptorSet, VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, 1, &textures.gradient.descriptor),
			vks::initializers::writeDescriptorSet(graphics.descriptorSet, VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, 2, &graphics.uniformBuffer.descriptor),
		};
		vkUpdateDescriptorSets(device, static_cast<uint32_t>(writeDescriptorSets.size()), writeDescriptorSets.data(), 0, NULL);
	}

	void preparePipelines()
	{
		VkPipelineInputAssemblyStateCreateInfo inputAssemblyState =
			vks::initializers::pipelineInputAssemblyStateCreateInfo(
				VK_PRIMITIVE_TOPOLOGY_POINT_LIST,
				0,
				VK_FALSE);

		VkPipelineRasterizationStateCreateInfo rasterizationState =
			vks::initializers::pipelineRasterizationStateCreateInfo(
				VK_POLYGON_MODE_FILL,
				VK_CULL_MODE_NONE,
				VK_FRONT_FACE_COUNTER_CLOCKWISE,
				0);

		VkPipelineColorBlendAttachmentState blendAttachmentState =
			vks::initializers::pipelineColorBlendAttachmentState(
				0xf,
				VK_FALSE);

		VkPipelineColorBlendStateCreateInfo colorBlendState =
			vks::initializers::pipelineColorBlendStateCreateInfo(
				1,
				&blendAttachmentState);

		VkPipelineDepthStencilStateCreateInfo depthStencilState =
			vks::initializers::pipelineDepthStencilStateCreateInfo(
				VK_FALSE,
				VK_FALSE,
				VK_COMPARE_OP_ALWAYS);

		VkPipelineViewportStateCreateInfo viewportState =
			vks::initializers::pipelineViewportStateCreateInfo(1, 1, 0);

		VkPipelineMultisampleStateCreateInfo multisampleState =
			vks::initializers::pipelineMultisampleStateCreateInfo(
				VK_SAMPLE_COUNT_1_BIT,
				0);

		std::vector<VkDynamicState> dynamicStateEnables = {
			VK_DYNAMIC_STATE_VIEWPORT,
			VK_DYNAMIC_STATE_SCISSOR
		};
		VkPipelineDynamicStateCreateInfo dynamicState =
			vks::initializers::pipelineDynamicStateCreateInfo(
				dynamicStateEnables.data(),
				static_cast<uint32_t>(dynamicStateEnables.size()),
				0);

		// Rendering pipeline
		// Load shaders
		std::array<VkPipelineShaderStageCreateInfo,2> shaderStages;

		shaderStages[0] = loadShader(getAssetPath() + "shaders/computenbody/particle.vert.spv", VK_SHADER_STAGE_VERTEX_BIT);
		shaderStages[1] = loadShader(getAssetPath() + "shaders/computenbody/particle.frag.spv", VK_SHADER_STAGE_FRAGMENT_BIT);

		VkGraphicsPipelineCreateInfo pipelineCreateInfo =
			vks::initializers::pipelineCreateInfo(
				graphics.pipelineLayout,
				renderPass,
				0);

		pipelineCreateInfo.pVertexInputState = &vertices.inputState;
		pipelineCreateInfo.pInputAssemblyState = &inputAssemblyState;
		pipelineCreateInfo.pRasterizationState = &rasterizationState;
		pipelineCreateInfo.pColorBlendState = &colorBlendState;
		pipelineCreateInfo.pMultisampleState = &multisampleState;
		pipelineCreateInfo.pViewportState = &viewportState;
		pipelineCreateInfo.pDepthStencilState = &depthStencilState;
		pipelineCreateInfo.pDynamicState = &dynamicState;
		pipelineCreateInfo.stageCount = static_cast<uint32_t>(shaderStages.size());
		pipelineCreateInfo.pStages = shaderStages.data();
		pipelineCreateInfo.renderPass = renderPass;

		// Additive blending
		blendAttachmentState.colorWriteMask = 0xF;
		blendAttachmentState.blendEnable = VK_TRUE;
		blendAttachmentState.colorBlendOp = VK_BLEND_OP_ADD;
		blendAttachmentState.srcColorBlendFactor = VK_BLEND_FACTOR_ONE;
		blendAttachmentState.dstColorBlendFactor = VK_BLEND_FACTOR_ONE;
		blendAttachmentState.alphaBlendOp = VK_BLEND_OP_ADD;
		blendAttachmentState.srcAlphaBlendFactor = VK_BLEND_FACTOR_SRC_ALPHA;
		blendAttachmentState.dstAlphaBlendFactor = VK_BLEND_FACTOR_DST_ALPHA;

		VK_CHECK_RESULT(vkCreateGraphicsPipelines(device, pipelineCache, 1, &pipelineCreateInfo, nullptr, &graphics.pipeline));
	}

	void prepareCompute()
	{
		// Create a compute capable device queue
		// The VulkanDevice::createLogicalDevice functions finds a compute capable queue and prefers queue families that only support compute
		// Depending on the implementation this may result in different queue family indices for graphics and computes,
		// requiring proper synchronization (see the memory barriers in buildComputeCommandBuffer)
		VkDeviceQueueCreateInfo queueCreateInfo = {};
		queueCreateInfo.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
		queueCreateInfo.pNext = NULL;
		queueCreateInfo.queueFamilyIndex = vulkanDevice->queueFamilyIndices.compute;
		queueCreateInfo.queueCount = 1;
		vkGetDeviceQueue(device, vulkanDevice->queueFamilyIndices.compute, 0, &compute.queue);

		// Create compute pipeline
		// Compute pipelines are created separate from graphics pipelines even if they use the same queue (family index)

		std::vector<VkDescriptorSetLayoutBinding> setLayoutBindings = {
			// Binding 0 : Particle position storage buffer
			vks::initializers::descriptorSetLayoutBinding(
				VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
				VK_SHADER_STAGE_COMPUTE_BIT,
				0),
			// Binding 1 : Uniform buffer
			vks::initializers::descriptorSetLayoutBinding(
				VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
				VK_SHADER_STAGE_COMPUTE_BIT,
				1),
		};

		VkDescriptorSetLayoutCreateInfo descriptorLayout =
			vks::initializers::descriptorSetLayoutCreateInfo(
				setLayoutBindings.data(),
				static_cast<uint32_t>(setLayoutBindings.size()));

		VK_CHECK_RESULT(vkCreateDescriptorSetLayout(device,	&descriptorLayout, nullptr,	&compute.descriptorSetLayout));

		VkPipelineLayoutCreateInfo pPipelineLayoutCreateInfo =
			vks::initializers::pipelineLayoutCreateInfo(
				&compute.descriptorSetLayout,
				1);

		VK_CHECK_RESULT(vkCreatePipelineLayout(device, &pPipelineLayoutCreateInfo, nullptr,	&compute.pipelineLayout));

		VkDescriptorSetAllocateInfo allocInfo =
			vks::initializers::descriptorSetAllocateInfo(
				descriptorPool,
				&compute.descriptorSetLayout,
				1);

		VK_CHECK_RESULT(vkAllocateDescriptorSets(device, &allocInfo, &compute.descriptorSet));

		std::vector<VkWriteDescriptorSet> computeWriteDescriptorSets =
		{
			// Binding 0 : Particle position storage buffer
			vks::initializers::writeDescriptorSet(
				compute.descriptorSet,
				VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
				0,
				&compute.storageBuffer.descriptor),
			// Binding 1 : Uniform buffer
			vks::initializers::writeDescriptorSet(
				compute.descriptorSet,
				VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
				1,
				&compute.uniformBuffer.descriptor)
		};

		vkUpdateDescriptorSets(device, static_cast<uint32_t>(computeWriteDescriptorSets.size()), computeWriteDescriptorSets.data(), 0, NULL);

		// Create pipelines
		VkComputePipelineCreateInfo computePipelineCreateInfo = vks::initializers::computePipelineCreateInfo(compute.pipelineLayout, 0);

		// 1st pass
		computePipelineCreateInfo.stage = loadShader(getAssetPath() + "shaders/computenbody/particle_calculate.comp.spv", VK_SHADER_STAGE_COMPUTE_BIT);

		// Set shader parameters via specialization constants
		struct SpecializationData {
			uint32_t sharedDataSize;
			float gravity;
			float power;
			float soften;
		} specializationData;

		std::vector<VkSpecializationMapEntry> specializationMapEntries;
		specializationMapEntries.push_back(vks::initializers::specializationMapEntry(0, offsetof(SpecializationData, sharedDataSize), sizeof(uint32_t)));
		specializationMapEntries.push_back(vks::initializers::specializationMapEntry(1, offsetof(SpecializationData, gravity), sizeof(float)));
		specializationMapEntries.push_back(vks::initializers::specializationMapEntry(2, offsetof(SpecializationData, power), sizeof(float)));
		specializationMapEntries.push_back(vks::initializers::specializationMapEntry(3, offsetof(SpecializationData, soften), sizeof(float)));

		specializationData.sharedDataSize = std::min((uint32_t)1024, (uint32_t)(vulkanDevice->properties.limits.maxComputeSharedMemorySize / sizeof(glm::vec4)));

		specializationData.gravity = 0.002f;
		specializationData.power = 0.75f;
		specializationData.soften = 0.05f;

		VkSpecializationInfo specializationInfo = 
			vks::initializers::specializationInfo(static_cast<uint32_t>(specializationMapEntries.size()), specializationMapEntries.data(), sizeof(specializationData), &specializationData);
		computePipelineCreateInfo.stage.pSpecializationInfo = &specializationInfo;

		VK_CHECK_RESULT(vkCreateComputePipelines(device, pipelineCache, 1, &computePipelineCreateInfo, nullptr, &compute.pipelineCalculate));

		// 2nd pass
		computePipelineCreateInfo.stage = loadShader(getAssetPath() + "shaders/computenbody/particle_integrate.comp.spv", VK_SHADER_STAGE_COMPUTE_BIT);
		VK_CHECK_RESULT(vkCreateComputePipelines(device, pipelineCache, 1, &computePipelineCreateInfo, nullptr, &compute.pipelineIntegrate));

		// Separate command pool as queue family for compute may be different than graphics
		VkCommandPoolCreateInfo cmdPoolInfo = {};
		cmdPoolInfo.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
		cmdPoolInfo.queueFamilyIndex = vulkanDevice->queueFamilyIndices.compute;
		cmdPoolInfo.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;
		VK_CHECK_RESULT(vkCreateCommandPool(device, &cmdPoolInfo, nullptr, &compute.commandPool));

		// Create a command buffer for compute operations
		VkCommandBufferAllocateInfo cmdBufAllocateInfo =
			vks::initializers::commandBufferAllocateInfo(
				compute.commandPool,
				VK_COMMAND_BUFFER_LEVEL_PRIMARY,
				1);	

		VK_CHECK_RESULT(vkAllocateCommandBuffers(device, &cmdBufAllocateInfo, &compute.commandBuffer));

		// Fence for compute CB sync
		VkFenceCreateInfo fenceCreateInfo = vks::initializers::fenceCreateInfo(VK_FENCE_CREATE_SIGNALED_BIT);
		VK_CHECK_RESULT(vkCreateFence(device, &fenceCreateInfo, nullptr, &compute.fence));

		// Build a single command buffer containing the compute dispatch commands
		buildComputeCommandBuffer();
	}

	// Prepare and initialize uniform buffer containing shader uniforms
	void prepareUniformBuffers()
	{
		// Compute shader uniform buffer block
		vulkanDevice->createBuffer(
			VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT,
			VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
			&compute.uniformBuffer,
			sizeof(compute.ubo));

		// Map for host access
		VK_CHECK_RESULT(compute.uniformBuffer.map());

		// Vertex shader uniform buffer block
		vulkanDevice->createBuffer(
			VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT,
			VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
			&graphics.uniformBuffer,
			sizeof(graphics.ubo));

		// Map for host access
		VK_CHECK_RESULT(graphics.uniformBuffer.map());

		updateGraphicsUniformBuffers();
	}

	void updateUniformBuffers()
	{
		compute.ubo.deltaT = paused ? 0.0f : frameTimer * 0.05f;
		compute.ubo.destX = sin(glm::radians(timer * 360.0f)) * 0.75f;
		compute.ubo.destY = 0.0f;
		memcpy(compute.uniformBuffer.mapped, &compute.ubo, sizeof(compute.ubo));
	}

	void updateGraphicsUniformBuffers()
	{
		graphics.ubo.projection = camera.matrices.perspective;
		graphics.ubo.view = camera.matrices.view;
		graphics.ubo.screenDim = glm::vec2((float)width, (float)height);
		memcpy(graphics.uniformBuffer.mapped, &graphics.ubo, sizeof(graphics.ubo));
	}

	void draw()
	{
		// Submit graphics commands
		VulkanExampleBase::prepareFrame();

		submitInfo.commandBufferCount = 1;
		submitInfo.pCommandBuffers = &drawCmdBuffers[currentBuffer];
		VK_CHECK_RESULT(vkQueueSubmit(queue, 1, &submitInfo, VK_NULL_HANDLE));

		VulkanExampleBase::submitFrame();

		// Submit compute commands
		vkWaitForFences(device, 1, &compute.fence, VK_TRUE, UINT64_MAX);
		vkResetFences(device, 1, &compute.fence);

		VkSubmitInfo computeSubmitInfo = vks::initializers::submitInfo();
		computeSubmitInfo.commandBufferCount = 1;
		computeSubmitInfo.pCommandBuffers = &compute.commandBuffer;

		VK_CHECK_RESULT(vkQueueSubmit(compute.queue, 1, &computeSubmitInfo, compute.fence));
	}

	void prepare()
	{
		VulkanExampleBase::prepare();
		loadAssets();
		prepareStorageBuffers();
		prepareUniformBuffers();
		setupDescriptorSetLayout();
		preparePipelines();
		setupDescriptorPool();
		setupDescriptorSet();
		prepareCompute();
		buildCommandBuffers();
		prepared = true;
	}

	virtual void render()
	{
		if (!prepared)
			return;
		draw();
		updateUniformBuffers();
	}

	virtual void viewChanged()
	{
		updateGraphicsUniformBuffers();
	}
};

VULKAN_EXAMPLE_MAIN()