sm64/src/pc/gfx/gfx_vulkan.cpp

#include <time.h>
#include <errno.h>
#include <vulkan/vulkan.h>
#include <vector>
#include <cstring>
#include <cstdio>
#include <array>
#include <limits>
#include <fstream>
#include <glslang/Public/ShaderLang.h>
#include <glslang/SPIRV/GlslangToSpv.h>

#include "gfx_rendering_api.h"
#include "gfx_pc.h"
#include "gfx_window_manager_api.h"
#include "gfx_cc.h"
#include "defaultresources.hpp"

static VkInstance instance = VK_NULL_HANDLE;
static VkPhysicalDevice physical_device = VK_NULL_HANDLE;
static VkDevice device = VK_NULL_HANDLE;
static VkQueue graphics_queue = VK_NULL_HANDLE;
static VkQueue present_queue = VK_NULL_HANDLE;
static VkCommandPool command_pool = VK_NULL_HANDLE;
static VkSwapchainKHR swapchain = VK_NULL_HANDLE;
static VkFormat surface_format;
static std::vector<VkImage> swapchain_images;
static std::vector<VkImageView> swapchain_views;
static VkRenderPass render_pass = VK_NULL_HANDLE;
static VkImage depth_image = VK_NULL_HANDLE;
static VkDeviceMemory depth_memory = VK_NULL_HANDLE;
static VkImageView depth_view = VK_NULL_HANDLE;
static std::vector<VkFramebuffer> swapchain_framebuffers;

// TODO: removed hardcoded 3 frame queue, should be variable
static std::array<VkCommandBuffer, 3> command_buffers;
static std::array<VkSemaphore, 3> image_available;
static std::array<VkSemaphore, 3> render_finished;
static std::array<VkFence, 3> in_flight;
static uint32_t current_frame = 0;
static uint32_t image_index = 0;

static VkCommandBuffer current_cmd = VK_NULL_HANDLE;

static VkBuffer vertex_buffer = VK_NULL_HANDLE;
static VkDeviceMemory vertex_memory = VK_NULL_HANDLE;

struct ShaderProgramVulkan {
    VkPipeline pipeline = VK_NULL_HANDLE;
    VkPipelineLayout pipeline_layout = VK_NULL_HANDLE;
    uint8_t num_inputs;
    bool used_textures[2];
};

std::unordered_map<uint32_t, ShaderProgramVulkan*> cached_shaders;

static ShaderProgramVulkan* last_program = nullptr;

VKAPI_ATTR VkBool32 VKAPI_CALL
gfx_vulkan_debug_callback(VkDebugUtilsMessageSeverityFlagBitsEXT messageSeverity,
              VkDebugUtilsMessageTypeFlagsEXT messageType,
              const VkDebugUtilsMessengerCallbackDataEXT* pCallbackData,
              void* pUserData) {

    printf("%s\n", pCallbackData->pMessage);

    return VK_FALSE;
}

static bool gfx_vulkan_renderer_z_is_from_0_to_1(void) {
    return false;
}

static VkShaderModule gfx_vulkan_create_shader_module(const uint32_t* code, const int length) {
    VkShaderModuleCreateInfo create_info = {};
    create_info.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
    create_info.codeSize = length;
    create_info.pCode = reinterpret_cast<const uint32_t*>(code);

    VkShaderModule shader_module;
    vkCreateShaderModule(device, &create_info, nullptr, &shader_module);

    return shader_module;
}

// TODO: will be removed, just used for testing
static VkShaderModule gfx_vulkan_load_shader(const char* path) {
    std::ifstream file(path, std::ios::ate | std::ios::binary);

    size_t fileSize = (size_t) file.tellg();
    std::vector<char> buffer(fileSize);

    file.seekg(0);
    file.read(buffer.data(), fileSize);

    return gfx_vulkan_create_shader_module(reinterpret_cast<const uint32_t *>(buffer.data()), fileSize);
}

static std::string gfx_vulkan_create_vertex_shader(const CCFeatures features) {
    std::string src = "#version 460 core\n";

    const bool uses_textures = features.used_textures[0] || features.used_textures[1];
    const bool uses_alpha = features.opt_alpha;
    const bool uses_fog = features.opt_fog;
    const bool needs_screenpos = features.opt_alpha && features.opt_noise;

    src += "layout(location = 0) in vec4 position;\n";

    int next_input = 1;
    if(uses_textures) {
        src += "layout(location = " + std::to_string(next_input++) + ") in vec2 uv;\n";
    }

    /*if(needs_screenpos) {
        src += "layout(location = " + std::to_string(next_input++) + ") in vec4 screen_pos;\n";
    }
*/
    if(uses_fog) {
        src += "layout(location = " + std::to_string(next_input++) + ") in vec4 fog;\n";
    }

    // extra inputs?
    for(int i = 0; i < features.num_inputs; i++) {
        int size = uses_alpha ? 4 : 3;
        src += "layout(location = " + std::to_string(next_input++) + ") in vec" + std::to_string(size) + " input" + std::to_string(i + 1) + ";\n";
    }

    src += "void main() {\n";

    src += "gl_Position = position;\n";

    src += "}";

    return src;
}

static std::string gfx_vulkan_create_fragment_shader(const CCFeatures features) {
    std::string src = "#version 460 core\n";

    const bool uses_textures = features.used_textures[0] || features.used_textures[1];
    const bool uses_alpha = features.opt_alpha;
    const bool uses_fog = features.opt_fog;
    const bool needs_screenpos = features.opt_alpha && features.opt_noise;

    src += "layout(location = 0) out vec4 out_color;\n";

    /*int next_input = 0;
    if(uses_textures) {
        src += "layout(location = " + std::to_string(next_input++) + ") in vec2 uv;\n";
    }

    if(needs_screenpos) {
        src += "layout(location = " + std::to_string(next_input++) + ") in vec4 screen_pos;\n";
    }

    if(uses_fog) {
        src += "layout(location = " + std::to_string(next_input++) + ") in vec4 fog;\n";
    }

    // extra inputs?
    for(int i = 0; i < features.num_inputs; i++) {
        int size = uses_alpha ? 4 : 3;
        src += "layout(location = " + std::to_string(next_input++) + ") in vec" + std::to_string(size) + " input" + std::to_string(i + 1) + ";\n";
    }*/

    src += "void main() {\n";

    src += "out_color = vec4(1, 1, 0, 1);\n";

    src += "}";

    return src;
}

static VkShaderModule gfx_vulkan_compile_glsl(const std::string& shader_src, const EShLanguage language) {
    glslang::InitializeProcess(); // TODO: lol

    const char* InputCString = shader_src.c_str();

    glslang::TShader shader(language);
    shader.setStrings(&InputCString, 1);

    int ClientInputSemanticsVersion = 100; // maps to, say, #define VULKAN 100

    shader.setEnvInput(glslang::EShSourceGlsl, language, glslang::EShClientVulkan, ClientInputSemanticsVersion);

    // we are targeting vulkan 1.1, so that uses SPIR-V 1.3
    shader.setEnvClient(glslang::EShClientVulkan, glslang::EShTargetVulkan_1_0);
    shader.setEnvTarget(glslang::EShTargetSpv, glslang::EShTargetSpv_1_0);

    if (!shader.parse(&DefaultTBuiltInResource, 100, false, EShMsgDefault)) {
        printf("%s\n", shader.getInfoLog());

        return {};
    }

    glslang::TProgram Program;
    Program.addShader(&shader);

    if (!Program.link(EShMsgDefault)) {
        printf("failed to link %s %s %s\n", shader_src.data(), shader.getInfoLog(), shader.getInfoDebugLog());

        return VK_NULL_HANDLE;
    }

    std::vector<unsigned int> SpirV;
    spv::SpvBuildLogger logger;

    glslang::SpvOptions spvOptions;
    spvOptions.generateDebugInfo = true;

    glslang::GlslangToSpv(*Program.getIntermediate(language), SpirV, &logger, &spvOptions);

    return gfx_vulkan_create_shader_module(SpirV.data(), SpirV.size() * sizeof(uint32_t));
}

static std::pair<VkPipeline, VkPipelineLayout> gfx_vulkan_create_pipeline(const CCFeatures features) {
    const std::string vertex_src = gfx_vulkan_create_vertex_shader(features);
    const std::string fragment_src = gfx_vulkan_create_fragment_shader(features);

    VkPipelineShaderStageCreateInfo vertex_stage = {};
    vertex_stage.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
    vertex_stage.stage = VK_SHADER_STAGE_VERTEX_BIT;
    vertex_stage.module = gfx_vulkan_compile_glsl(vertex_src, EShLanguage::EShLangVertex);
    vertex_stage.pName = "main";

    VkPipelineShaderStageCreateInfo fragment_stage = {};
    fragment_stage.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
    fragment_stage.stage = VK_SHADER_STAGE_FRAGMENT_BIT;
    fragment_stage.module = gfx_vulkan_compile_glsl(fragment_src, EShLanguage::EShLangFragment);
    fragment_stage.pName = "main";

    std::array<VkPipelineShaderStageCreateInfo, 2> shader_stages = { vertex_stage, fragment_stage };

    const bool uses_textures = features.used_textures[0] || features.used_textures[1];
    const bool uses_alpha = features.opt_alpha;
    const bool uses_fog = features.opt_fog;
    const bool needs_screenpos = features.opt_alpha && features.opt_noise;

    int num_floats = 0;
    int current_location = 1;

    std::vector<VkVertexInputAttributeDescription> attributes;

    VkVertexInputAttributeDescription position_attribute = {};
    position_attribute.format = VK_FORMAT_R32G32B32A32_SFLOAT;
    num_floats += 4;

    attributes.push_back(position_attribute);

    if(uses_textures) {
        VkVertexInputAttributeDescription uv_attribute = {};
        uv_attribute.format = VK_FORMAT_R32G32_SFLOAT;
        uv_attribute.location = current_location++;
        uv_attribute.offset = num_floats * sizeof(float);

        num_floats += 2;

        attributes.push_back(uv_attribute);
    }

    if(uses_fog) {
        VkVertexInputAttributeDescription fog_attribute = {};
        fog_attribute.format = VK_FORMAT_R32G32B32A32_SFLOAT;
        fog_attribute.location = current_location++;
        fog_attribute.offset = num_floats * sizeof(float);

        num_floats += 4;

        attributes.push_back(fog_attribute);
    }

    for(int i = 0; i < features.num_inputs; i++) {
        VkVertexInputAttributeDescription extra_attribute = {};
        extra_attribute.location = current_location++;
        extra_attribute.offset = num_floats * sizeof(float);

        if(uses_alpha) {
            extra_attribute.format = VK_FORMAT_R32G32B32A32_SFLOAT;
            num_floats += 4;
        } else {
            extra_attribute.format = VK_FORMAT_R32G32B32_SFLOAT;
            num_floats += 3;
        }

        attributes.push_back(extra_attribute);
    }

    printf("uses textures: %d\n", uses_textures);
    printf("needs screenpos: %d\n", needs_screenpos);
    printf("uses fog: %d\n", uses_fog);
    printf("extra inputs: %d\n", features.num_inputs);

    VkVertexInputBindingDescription binding = {};
    binding.stride = sizeof(float) * num_floats;
    binding.inputRate = VK_VERTEX_INPUT_RATE_VERTEX;

    VkPipelineVertexInputStateCreateInfo vertex_input_state = {};
    vertex_input_state.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO;
    vertex_input_state.vertexBindingDescriptionCount = 1;
    vertex_input_state.pVertexBindingDescriptions = &binding;
    vertex_input_state.vertexAttributeDescriptionCount = attributes.size();
    vertex_input_state.pVertexAttributeDescriptions = attributes.data();

    VkPipelineInputAssemblyStateCreateInfo input_assembly = {};
    input_assembly.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO;
    input_assembly.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;

    VkViewport viewport = {};
    viewport.width = 640;
    viewport.height = 480;
    viewport.maxDepth = 1.0f;

    VkRect2D scissor = {};
    scissor.extent.width = 640;
    scissor.extent.height = 480;

    VkPipelineViewportStateCreateInfo viewport_state = {};
    viewport_state.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO;
    viewport_state.viewportCount = 1;
    viewport_state.pViewports = &viewport;
    viewport_state.scissorCount = 1;
    viewport_state.pScissors = &scissor;

    VkPipelineRasterizationStateCreateInfo rasterizer = {};
    rasterizer.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO;
    rasterizer.lineWidth = 1.0f;
    rasterizer.cullMode = VK_CULL_MODE_NONE;
    rasterizer.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE;

    VkPipelineMultisampleStateCreateInfo multisampling = {};
    multisampling.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO;
    multisampling.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;

    VkPipelineColorBlendAttachmentState color_blend_attachment = {};
    color_blend_attachment.colorWriteMask = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT | VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT;

    VkPipelineColorBlendStateCreateInfo color_blending = {};
    color_blending.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO;
    color_blending.attachmentCount = 1;
    color_blending.pAttachments = &color_blend_attachment;

    VkPipelineDynamicStateCreateInfo dynamic_state = {};
    dynamic_state.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO;

    VkPipelineLayoutCreateInfo pipeline_layout_create_info = {};
    pipeline_layout_create_info.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;

    VkPipelineLayout layout = VK_NULL_HANDLE;

    vkCreatePipelineLayout(device, &pipeline_layout_create_info, nullptr, &layout);

    VkPipelineDepthStencilStateCreateInfo depth_stencil = {};
    depth_stencil.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO;
    depth_stencil.depthTestEnable = VK_TRUE;
    depth_stencil.depthWriteEnable = VK_TRUE;
    depth_stencil.depthCompareOp = VK_COMPARE_OP_LESS;
    depth_stencil.maxDepthBounds = 1.0f;

    VkGraphicsPipelineCreateInfo pipeline_create_info = {};
    pipeline_create_info.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO;
    pipeline_create_info.stageCount = shader_stages.size();
    pipeline_create_info.pStages = shader_stages.data();
    pipeline_create_info.pVertexInputState = &vertex_input_state;
    pipeline_create_info.pInputAssemblyState = &input_assembly;
    pipeline_create_info.pViewportState = &viewport_state;
    pipeline_create_info.pRasterizationState = &rasterizer;
    pipeline_create_info.pMultisampleState = &multisampling;
    pipeline_create_info.pColorBlendState = &color_blending;
    pipeline_create_info.pDynamicState = &dynamic_state;
    pipeline_create_info.pDepthStencilState = &depth_stencil;
    pipeline_create_info.layout = layout;
    pipeline_create_info.renderPass = render_pass;

    VkPipeline pipeline = VK_NULL_HANDLE;
    vkCreateGraphicsPipelines(device,
                              VK_NULL_HANDLE,
                              1,
                              &pipeline_create_info,
                              nullptr,
                              &pipeline);

    return {pipeline, layout};
}

// FIXME: reorganize
static void gfx_vulkan_renderer_load_shader(struct ShaderProgram *new_prg);

static struct ShaderProgram *gfx_vulkan_renderer_create_and_load_new_shader(uint32_t shader_id) {
    ShaderProgramVulkan* shader = new ShaderProgramVulkan();

    CCFeatures cc_features;
    gfx_cc_get_features(shader_id, &cc_features);

    auto [pipeline, layout] = gfx_vulkan_create_pipeline(cc_features);
    shader->pipeline = pipeline;
    shader->pipeline_layout = layout;

    cached_shaders[shader_id] = shader;

    gfx_vulkan_renderer_load_shader(reinterpret_cast<ShaderProgram *>(shader));

    shader->used_textures[0] = cc_features.used_textures[0];
    shader->used_textures[1] = cc_features.used_textures[1];
    shader->num_inputs = cc_features.num_inputs;

    return reinterpret_cast<ShaderProgram *>(shader);
}

static struct ShaderProgram *gfx_vulkan_renderer_lookup_shader(uint32_t shader_id) {
    return reinterpret_cast<ShaderProgram *>(cached_shaders[shader_id]);
}

static void gfx_vulkan_renderer_shader_get_info(struct ShaderProgram *prg, uint8_t *num_inputs, bool used_textures[2]) {
    ShaderProgramVulkan* vulkan = reinterpret_cast<ShaderProgramVulkan *>(prg);

    *num_inputs = vulkan->num_inputs;
    used_textures[0] = vulkan->used_textures[0];
    used_textures[1] = vulkan->used_textures[1];
}

static uint32_t gfx_vulkan_renderer_new_texture(void) {
    return 0;
}

static uint32_t gfx_vulkan_find_memory_type(uint32_t type_filter, VkMemoryPropertyFlags properties) {
    VkPhysicalDeviceMemoryProperties memory_properties;
    vkGetPhysicalDeviceMemoryProperties(physical_device, &memory_properties);

    for (uint32_t i = 0; i < memory_properties.memoryTypeCount; i++) {
        if ((type_filter & (1 << i)) &&
            (memory_properties.memoryTypes[i].propertyFlags & properties) ==
                properties) {
            return i;
        }
    }

    return -1;
}

static void gfx_vulkan_create_instance() {
    std::vector<const char*> instance_extensions = {"VK_EXT_debug_utils"};

    uint32_t extension_count = 0;
    vkEnumerateInstanceExtensionProperties(nullptr, &extension_count, nullptr);

    std::vector<VkExtensionProperties> extensions(extension_count);
    vkEnumerateInstanceExtensionProperties(nullptr, &extension_count, extensions.data());

    for(auto& extension : extensions) {
        if (strstr(extension.extensionName, "surface") != nullptr) {
            instance_extensions.push_back(extension.extensionName);
        }

        if (strstr(extension.extensionName, "VK_KHR_get_physical_device_properties2") != nullptr) {
            instance_extensions.push_back(extension.extensionName);
        }
    }

    VkDebugUtilsMessengerCreateInfoEXT debug_create_info = {};
    debug_create_info.sType =
        VK_STRUCTURE_TYPE_DEBUG_UTILS_MESSENGER_CREATE_INFO_EXT;
    debug_create_info.messageSeverity =
        VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT |
        VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT;
    debug_create_info.messageType =
        VK_DEBUG_UTILS_MESSAGE_TYPE_VALIDATION_BIT_EXT;
    debug_create_info.pfnUserCallback = gfx_vulkan_debug_callback;

    VkInstanceCreateInfo create_info = {};
    create_info.pNext = &debug_create_info;
    create_info.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO;
    create_info.ppEnabledExtensionNames = instance_extensions.data();
    create_info.enabledExtensionCount = instance_extensions.size();

    vkCreateInstance(&create_info, nullptr, &instance);

    printf("Created vulkan instance!\n");
}

static void gfx_vulkan_create_command_pool(const int graphics_family_index) {
    VkCommandPoolCreateInfo pool_create_info = {};
    pool_create_info.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
    pool_create_info.queueFamilyIndex = graphics_family_index;
    pool_create_info.flags = VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;

    vkCreateCommandPool(device, &pool_create_info, nullptr, &command_pool);
}

static void gfx_vulkan_create_device() {
    // pick physical device
    uint32_t device_count = 0;
    vkEnumeratePhysicalDevices(instance, &device_count, nullptr);

    std::vector<VkPhysicalDevice> devices(device_count);
    vkEnumeratePhysicalDevices(instance, &device_count, devices.data());

    for (auto device : devices) {
        VkPhysicalDeviceProperties device_properties;
        vkGetPhysicalDeviceProperties(device, &device_properties);

        printf("GPU Found: %s\n", device_properties.deviceName);
    }

    physical_device = devices[0];

    uint32_t extension_count = 0;
    vkEnumerateDeviceExtensionProperties(physical_device, nullptr,
                                         &extension_count, nullptr);

    std::vector<VkExtensionProperties> extension_properties(extension_count);
    vkEnumerateDeviceExtensionProperties(
        physical_device, nullptr, &extension_count,
                                         extension_properties.data());

    uint32_t graphics_family_index = 0, present_family_index = 0;

    // create logical device
    uint32_t queue_family_count = 0;
    vkGetPhysicalDeviceQueueFamilyProperties(physical_device, &queue_family_count,
                                             nullptr);

    std::vector<VkQueueFamilyProperties> queueFamilies(queue_family_count);
    vkGetPhysicalDeviceQueueFamilyProperties(physical_device, &queue_family_count,
                                             queueFamilies.data());

    int i = 0;
    for (const auto&queue_family : queueFamilies) {
        if (queue_family.queueCount > 0 && queue_family.queueFlags & VK_QUEUE_GRAPHICS_BIT) {
            graphics_family_index = i;
        }

        i++;
    }

    std::vector<VkDeviceQueueCreateInfo> queue_create_infos;

    if (graphics_family_index == present_family_index) {
        VkDeviceQueueCreateInfo queue_create_info = {};
        queue_create_info.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
        queue_create_info.queueFamilyIndex = graphics_family_index;
        queue_create_info.queueCount = 1;

        float queuePriority = 1.0f;
        queue_create_info.pQueuePriorities = &queuePriority;

        queue_create_infos.push_back(queue_create_info);
    } else {
        // graphics
        {
            VkDeviceQueueCreateInfo queue_create_info = {};
            queue_create_info.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
            queue_create_info.queueFamilyIndex = graphics_family_index;
            queue_create_info.queueCount = 1;

            float queuePriority = 1.0f;
            queue_create_info.pQueuePriorities = &queuePriority;

            queue_create_infos.push_back(queue_create_info);
        }

        // present
        {
            VkDeviceQueueCreateInfo queue_create_info = {};
            queue_create_info.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
            queue_create_info.queueFamilyIndex = present_family_index;
            queue_create_info.queueCount = 1;

            float queuePriority = 1.0f;
            queue_create_info.pQueuePriorities = &queuePriority;

            queue_create_infos.push_back(queue_create_info);
        }
    }

    std::array<const char*, 1> device_extensions = {"VK_KHR_swapchain"};

    VkDeviceCreateInfo device_create_info = {};
    device_create_info.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
    device_create_info.pQueueCreateInfos = queue_create_infos.data();
    device_create_info.queueCreateInfoCount =
        static_cast<uint32_t>(queue_create_infos.size());
    device_create_info.ppEnabledExtensionNames = device_extensions.data();
    device_create_info.enabledExtensionCount = device_extensions.size();

    vkCreateDevice(physical_device, &device_create_info, nullptr, &device);

    printf("Created vulkan device!\n");

    vkGetDeviceQueue(device, graphics_family_index, 0, &graphics_queue);
    vkGetDeviceQueue(device, present_family_index, 0, &present_queue);

    gfx_vulkan_create_command_pool(graphics_family_index);
}

static void gfx_vulkan_create_swapchain() {
    vkQueueWaitIdle(present_queue);

    // create surface from wapi
    VkSurfaceKHR surface =
        static_cast<VkSurfaceKHR>(gfx_get_current_windowing_api()->create_surface(instance));

    VkBool32 supported;
    vkGetPhysicalDeviceSurfaceSupportKHR(physical_device, 0,
                                         surface, &supported);

    // query swapchain support
    VkSurfaceCapabilitiesKHR capabilities;
    vkGetPhysicalDeviceSurfaceCapabilitiesKHR(
        physical_device, surface, &capabilities);

    std::vector<VkSurfaceFormatKHR> formats;

    uint32_t format_count;
    vkGetPhysicalDeviceSurfaceFormatsKHR(
        physical_device, surface, &format_count, nullptr);

    formats.resize(format_count);
    vkGetPhysicalDeviceSurfaceFormatsKHR(
        physical_device, surface, &format_count, formats.data());

    std::vector<VkPresentModeKHR> present_modes;
    uint32_t present_mode_count;
    vkGetPhysicalDeviceSurfacePresentModesKHR(
        physical_device, surface, &present_mode_count, nullptr);

    present_modes.resize(present_mode_count);
    vkGetPhysicalDeviceSurfacePresentModesKHR(
        physical_device, surface, &present_mode_count,
                                              present_modes.data());

    // choosing swapchain features
    surface_format = formats[0].format;
    for (const auto& available_format : formats) {
        if (available_format.format == VK_FORMAT_B8G8R8A8_UNORM && available_format.colorSpace == VK_COLOR_SPACE_SRGB_NONLINEAR_KHR) {
            surface_format = available_format.format;
        }
    }

    VkPresentModeKHR present_mode = VK_PRESENT_MODE_FIFO_KHR;
    for (const auto& available_mode : present_modes) {
        if (available_mode == VK_PRESENT_MODE_MAILBOX_KHR) {
            present_mode = available_mode;
        }
    }

    uint32_t image_count = capabilities.minImageCount + 1;
    if (capabilities.maxImageCount > 0 && image_count > capabilities.maxImageCount) {
        image_count = capabilities.maxImageCount;
    }

    // create swapchain
    VkSwapchainCreateInfoKHR swapchain_create_info = {};
    swapchain_create_info.sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR;
    swapchain_create_info.surface = surface;
    swapchain_create_info.minImageCount = image_count;
    swapchain_create_info.imageFormat = surface_format;
    swapchain_create_info.imageColorSpace = VK_COLORSPACE_SRGB_NONLINEAR_KHR; // FIXME: hardcoded
    swapchain_create_info.imageExtent.width = 640; // FIXME: hardcoded
    swapchain_create_info.imageExtent.height = 480;
    swapchain_create_info.imageArrayLayers = 1;
    swapchain_create_info.imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
    swapchain_create_info.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE;
    swapchain_create_info.preTransform = capabilities.currentTransform;
    swapchain_create_info.compositeAlpha = VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR;
    swapchain_create_info.presentMode = present_mode;
    swapchain_create_info.clipped = VK_TRUE;

    VkSwapchainKHR old_swapchain = swapchain;
    swapchain_create_info.oldSwapchain = old_swapchain;

    vkCreateSwapchainKHR(device, &swapchain_create_info, nullptr, &swapchain);

    if(old_swapchain != VK_NULL_HANDLE) {
        vkDestroySwapchainKHR(device, old_swapchain, nullptr);
    }

    vkGetSwapchainImagesKHR(device, swapchain, &image_count,
                            nullptr);
    swapchain_images.resize(image_count);
    vkGetSwapchainImagesKHR(device, swapchain, &image_count, swapchain_images.data());

    swapchain_views.resize(swapchain_images.size());

    for (size_t i = 0; i < swapchain_images.size(); i++) {
        VkImageViewCreateInfo view_create_info = {};
        view_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
        view_create_info.image = swapchain_images[i];
        view_create_info.viewType = VK_IMAGE_VIEW_TYPE_2D;
        view_create_info.format = surface_format;
        view_create_info.components.r = VK_COMPONENT_SWIZZLE_IDENTITY;
        view_create_info.components.g = VK_COMPONENT_SWIZZLE_IDENTITY;
        view_create_info.components.b = VK_COMPONENT_SWIZZLE_IDENTITY;
        view_create_info.components.a = VK_COMPONENT_SWIZZLE_IDENTITY;
        view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
        view_create_info.subresourceRange.baseMipLevel = 0;
        view_create_info.subresourceRange.levelCount = 1;
        view_create_info.subresourceRange.baseArrayLayer = 0;
        view_create_info.subresourceRange.layerCount = 1;

        vkCreateImageView(device, &view_create_info, nullptr,&swapchain_views[i]);
    }

    printf("Initialized swap chain!\n");
}

static void gfx_vulkan_create_render_pass() {
    VkAttachmentDescription color_attachment = {};
    color_attachment.format = surface_format;
    color_attachment.samples = VK_SAMPLE_COUNT_1_BIT;
    color_attachment.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
    color_attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
    color_attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
    color_attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
    color_attachment.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
    color_attachment.finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;

    VkAttachmentReference color_attachment_ref = {};
    color_attachment_ref.attachment = 0;
    color_attachment_ref.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

    VkAttachmentDescription depth_attachment = {};
    depth_attachment.format = VK_FORMAT_D32_SFLOAT;
    depth_attachment.samples = VK_SAMPLE_COUNT_1_BIT;
    depth_attachment.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
    depth_attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
    depth_attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
    depth_attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
    depth_attachment.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
    depth_attachment.finalLayout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

    VkAttachmentReference depth_attachment_ref = {};
    depth_attachment_ref.attachment = 1;
    depth_attachment_ref.layout = VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;

    VkSubpassDependency dependency = {};
    dependency.srcSubpass = VK_SUBPASS_EXTERNAL;
    dependency.dstSubpass = 0;
    dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
    dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT;
    dependency.srcAccessMask = 0;
    dependency.dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
    dependency.dependencyFlags = 0;

    VkSubpassDescription subpass = {};
    subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
    subpass.colorAttachmentCount = 1;
    subpass.pColorAttachments = &color_attachment_ref;
    subpass.pDepthStencilAttachment = &depth_attachment_ref;

    std::array<VkAttachmentDescription, 2> attachments = { color_attachment, depth_attachment };

    VkRenderPassCreateInfo render_pass_create_info = {};
    render_pass_create_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
    render_pass_create_info.attachmentCount = attachments.size();
    render_pass_create_info.pAttachments = attachments.data();
    render_pass_create_info.subpassCount = 1;
    render_pass_create_info.pSubpasses = &subpass;
    render_pass_create_info.dependencyCount = 1;
    render_pass_create_info.pDependencies = &dependency;

    vkCreateRenderPass(device, &render_pass_create_info, nullptr, &render_pass);
}

static void gfx_vulkan_create_sync() {
    for(int i = 0; i < 3; i++) {
        VkCommandBufferAllocateInfo allocate_info = {};
        allocate_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO;
        allocate_info.commandPool = command_pool;
        allocate_info.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY;
        allocate_info.commandBufferCount = 1;

        vkAllocateCommandBuffers(device, &allocate_info, &command_buffers[i]);
    }

    VkSemaphoreCreateInfo semaphore_create_info = {};
    semaphore_create_info.sType = VK_STRUCTURE_TYPE_SEMAPHORE_CREATE_INFO;

    VkFenceCreateInfo fence_create_info = {};
    fence_create_info.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
    fence_create_info.flags = VK_FENCE_CREATE_SIGNALED_BIT;

    for (size_t i = 0; i < 3; i++) {
        vkCreateSemaphore(device, &semaphore_create_info, nullptr, &image_available[i]);
        vkCreateSemaphore(device, &semaphore_create_info, nullptr, &render_finished[i]);
        vkCreateFence(device, &fence_create_info, nullptr, &in_flight[i]);
    }
}

static void gfx_vulkan_create_depth() {
    VkImageCreateInfo image_create_info = {};
    image_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
    image_create_info.imageType = VK_IMAGE_TYPE_2D;
    image_create_info.extent.width = 640;
    image_create_info.extent.height = 480;
    image_create_info.extent.depth = 1;
    image_create_info.mipLevels = 1;
    image_create_info.arrayLayers = 1;
    image_create_info.format = VK_FORMAT_D32_SFLOAT;
    image_create_info.tiling = VK_IMAGE_TILING_OPTIMAL;
    image_create_info.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
    image_create_info.usage = VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
    image_create_info.samples = VK_SAMPLE_COUNT_1_BIT;
    image_create_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;

    vkCreateImage(device, &image_create_info, nullptr, &depth_image);

    VkMemoryRequirements memory_requirements;
    vkGetImageMemoryRequirements(device, depth_image, &memory_requirements);

    VkMemoryAllocateInfo allocate_info = {};
    allocate_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
    allocate_info.allocationSize = memory_requirements.size;
    allocate_info.memoryTypeIndex = gfx_vulkan_find_memory_type(memory_requirements.memoryTypeBits, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);

    vkAllocateMemory(device, &allocate_info, nullptr, &depth_memory);

    vkBindImageMemory(device, depth_image, depth_memory, 0);

    VkImageViewCreateInfo view_create_info = {};
    view_create_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
    view_create_info.image = depth_image;
    view_create_info.viewType = VK_IMAGE_VIEW_TYPE_2D;
    view_create_info.format = VK_FORMAT_D32_SFLOAT;
    view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
    view_create_info.subresourceRange.levelCount = 1;
    view_create_info.subresourceRange.layerCount = 1;

    vkCreateImageView(device, &view_create_info, nullptr, &depth_view);
}

static void gfx_vulkan_create_framebuffers() {
    swapchain_framebuffers.resize(swapchain_views.size());

    for (size_t i = 0; i < swapchain_framebuffers.size(); i++) {
        std::array<VkImageView, 2> attachments = {swapchain_views[i], depth_view};

        VkFramebufferCreateInfo create_info = {};
        create_info.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
        create_info.renderPass = render_pass;
        create_info.attachmentCount = attachments.size();
        create_info.pAttachments = attachments.data();
        create_info.width = 640;
        create_info.height = 480;
        create_info.layers = 1;

        vkCreateFramebuffer(device, &create_info, nullptr, &swapchain_framebuffers[i]);
    }
}

static void* mapped_data = nullptr;

static void gfx_vulkan_create_buffers() {
    vkDeviceWaitIdle(device);

    // create buffer
    VkBufferCreateInfo create_info = {};
    create_info.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
    create_info.size = 1048576;
    create_info.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
    create_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;

    vkCreateBuffer(device, &create_info, nullptr, &vertex_buffer);

    // allocate memory
    VkMemoryRequirements memory_requirements;
    vkGetBufferMemoryRequirements(device, vertex_buffer, &memory_requirements);

    VkMemoryAllocateInfo allocate_info = {};
    allocate_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
    allocate_info.allocationSize = memory_requirements.size;
    allocate_info.memoryTypeIndex =
        gfx_vulkan_find_memory_type(memory_requirements.memoryTypeBits,
                       VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);

    vkAllocateMemory(device, &allocate_info, nullptr, &vertex_memory);

    vkBindBufferMemory(device, vertex_buffer, vertex_memory, 0);

    vkMapMemory(device, vertex_memory, 0, VK_WHOLE_SIZE, 0, &mapped_data);
}

static void gfx_vulkan_renderer_init(void) {
    gfx_vulkan_create_instance();
    gfx_vulkan_create_device();
    gfx_vulkan_create_swapchain();
    gfx_vulkan_create_render_pass();
    gfx_vulkan_create_sync();
    gfx_vulkan_create_depth();
    gfx_vulkan_create_framebuffers();
    gfx_vulkan_create_buffers();
}

static void gfx_vulkan_renderer_on_resize(void) {
    // doesn't seem to be called?
}

static int buf_vbo_offset = 0;

static void gfx_vulkan_renderer_start_frame(void) {
    vkWaitForFences(
        device, 1,
        &in_flight[current_frame],
        VK_TRUE, std::numeric_limits<uint64_t>::max());

    image_index = 0;
    VkResult result = vkAcquireNextImageKHR(
        device, swapchain,
        std::numeric_limits<uint64_t>::max(),
        image_available[current_frame],
        VK_NULL_HANDLE, &image_index);

    if (result == VK_ERROR_OUT_OF_DATE_KHR) {
        printf("Swapchain is out of date!\n");
        return;
    }

    current_cmd = command_buffers[current_frame];

    VkCommandBufferBeginInfo buffer_begin = {};
    buffer_begin.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
    buffer_begin.flags = VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT;

    vkBeginCommandBuffer(current_cmd, &buffer_begin);

    VkRenderPassBeginInfo render_pass_begin = {};
    render_pass_begin.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
    render_pass_begin.renderPass = render_pass;
    render_pass_begin.framebuffer = swapchain_framebuffers[image_index];

    std::array<VkClearValue, 2> clear_values = {};
    clear_values[0].color.float32[0] = 0.8;
    clear_values[0].color.float32[1] = 0.8;
    clear_values[0].color.float32[2] = 0.8;
    clear_values[0].color.float32[3] = 1.0;
    clear_values[1].depthStencil = {1.0f, 0};

    render_pass_begin.clearValueCount = clear_values.size();
    render_pass_begin.pClearValues = clear_values.data();
    render_pass_begin.renderArea.extent.width = 640;
    render_pass_begin.renderArea.extent.height = 480;

    vkCmdBeginRenderPass(current_cmd, &render_pass_begin, VK_SUBPASS_CONTENTS_INLINE);

    buf_vbo_offset = 0;
}

/* draw commands */

static void gfx_vulkan_renderer_select_texture(int tile, uint32_t texture_id) {
}

static void gfx_vulkan_renderer_upload_texture(const uint8_t *rgba32_buf, int width, int height) {
}

static void gfx_vulkan_renderer_set_sampler_parameters(int tile, bool linear_filter, uint32_t cms, uint32_t cmt) {
}

static void gfx_vulkan_renderer_set_depth_test(bool depth_test) {
}

static void gfx_vulkan_renderer_set_depth_mask(bool z_upd) {
}

static void gfx_vulkan_renderer_set_zmode_decal(bool zmode_decal) {
}

static void gfx_vulkan_renderer_set_viewport(int x, int y, int width, int height) {
}

static void gfx_vulkan_renderer_set_scissor(int x, int y, int width, int height) {
}

static void gfx_vulkan_renderer_set_use_alpha(bool use_alpha) {
}

static void gfx_vulkan_renderer_unload_shader(struct ShaderProgram *old_prg) {

}

static void gfx_vulkan_renderer_load_shader(struct ShaderProgram *new_prg) {
    ShaderProgramVulkan* vulkan = (ShaderProgramVulkan*)new_prg;

    last_program = vulkan;
}

static void gfx_vulkan_renderer_draw_triangles(float buf_vbo[], size_t buf_vbo_len, size_t buf_vbo_num_tris) {
    vkCmdBindPipeline(current_cmd, VK_PIPELINE_BIND_POINT_GRAPHICS, last_program->pipeline);

    memcpy((uint8_t *)mapped_data + buf_vbo_offset, buf_vbo, buf_vbo_len * sizeof(float));

    VkDeviceSize offsets[] = {static_cast<VkDeviceSize>(buf_vbo_offset)};
    vkCmdBindVertexBuffers(current_cmd, 0, 1, &vertex_buffer, offsets);

    vkCmdDraw(current_cmd, buf_vbo_num_tris * 3, 1, 0, 0);

    buf_vbo_offset += buf_vbo_len * sizeof(float);
}

/* end draw commands */

static void gfx_vulkan_renderer_end_frame(void) {
    vkCmdEndRenderPass(current_cmd);
    vkEndCommandBuffer(current_cmd);
}

static void gfx_vulkan_renderer_finish_render(void) {
    VkSubmitInfo submit_info = {};
    submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;

    VkSemaphore wait_semaphores[] = {image_available[current_frame]};
    VkPipelineStageFlags wait_stages[] = {
        VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT};
    submit_info.waitSemaphoreCount = 1;
    submit_info.pWaitSemaphores = wait_semaphores;
    submit_info.pWaitDstStageMask = wait_stages;
    submit_info.commandBufferCount = 1;
    submit_info.pCommandBuffers = &current_cmd;

    VkSemaphore signal_semaphores[] = {render_finished[current_frame]};
    submit_info.signalSemaphoreCount = 1;
    submit_info.pSignalSemaphores = signal_semaphores;

    vkResetFences(device, 1, &in_flight[current_frame]);

    if (vkQueueSubmit(graphics_queue, 1, &submit_info, in_flight[current_frame]) != VK_SUCCESS) {
        return;
    }

    // present
    VkPresentInfoKHR present_info = {};
    present_info.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;

    present_info.waitSemaphoreCount = 1;
    present_info.pWaitSemaphores = signal_semaphores;
    VkSwapchainKHR swapChains[] = {swapchain};
    present_info.swapchainCount = 1;
    present_info.pSwapchains = swapChains;
    present_info.pImageIndices = &image_index;

    vkQueuePresentKHR(present_queue, &present_info);

    current_frame = (current_frame + 1) % 3;
    current_cmd = VK_NULL_HANDLE;
}

static const char* gfx_vulkan_get_name() {
    return "Vulkan";
}

struct GfxRenderingAPI gfx_vulkan_api = {
    gfx_vulkan_renderer_z_is_from_0_to_1,
    gfx_vulkan_renderer_unload_shader,
    gfx_vulkan_renderer_load_shader,
    gfx_vulkan_renderer_create_and_load_new_shader,
    gfx_vulkan_renderer_lookup_shader,
    gfx_vulkan_renderer_shader_get_info,
    gfx_vulkan_renderer_new_texture,
    gfx_vulkan_renderer_select_texture,
    gfx_vulkan_renderer_upload_texture,
    gfx_vulkan_renderer_set_sampler_parameters,
    gfx_vulkan_renderer_set_depth_test,
    gfx_vulkan_renderer_set_depth_mask,
    gfx_vulkan_renderer_set_zmode_decal,
    gfx_vulkan_renderer_set_viewport,
    gfx_vulkan_renderer_set_scissor,
    gfx_vulkan_renderer_set_use_alpha,
    gfx_vulkan_renderer_draw_triangles,
    gfx_vulkan_renderer_init,
    gfx_vulkan_renderer_on_resize,
    gfx_vulkan_renderer_start_frame,
    gfx_vulkan_renderer_end_frame,
    gfx_vulkan_renderer_finish_render,
    gfx_vulkan_get_name,
};