/*
Created on: Oct 2, 2019
Copyright 2019 flyinghead
This file is part of Flycast.
Flycast is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 2 of the License, or
(at your option) any later version.
Flycast is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with Flycast. If not, see .
*/
#include "vulkan_context.h"
#include "imgui/imgui.h"
#include "imgui_impl_vulkan.h"
#include "../gui.h"
#ifdef USE_SDL
#include
#include
#endif
#include "compiler.h"
#include "texture.h"
#include "utils.h"
VulkanContext *VulkanContext::contextInstance;
static const char *PipelineCacheFileName = DATA_PATH "vulkan_pipeline.cache";
#ifndef __ANDROID__
VKAPI_ATTR static VkBool32 VKAPI_CALL debugUtilsMessengerCallback(VkDebugUtilsMessageSeverityFlagBitsEXT messageSeverity, VkDebugUtilsMessageTypeFlagsEXT messageTypes,
VkDebugUtilsMessengerCallbackDataEXT const * pCallbackData, void * /*pUserData*/)
{
std::string msg = vk::to_string(static_cast(messageSeverity)) + ": "
+ vk::to_string(static_cast(messageTypes)) + ": ";
if (pCallbackData->pMessageIdName)
msg += std::string("messageIDName=") + pCallbackData->pMessageIdName + " ";
// msg += std::string("messageIdNumber=") + pCallbackData->messageIdNumber + " ";
if (pCallbackData->pMessage)
msg += pCallbackData->pMessage;
/* TODO
if (0 < pCallbackData->queueLabelCount)
{
std::cerr << "\t" << "Queue Labels:\n";
for (uint8_t i = 0; i < pCallbackData->queueLabelCount; i++)
{
std::cerr << "\t\t" << "lableName = <" << pCallbackData->pQueueLabels[i].pLabelName << ">\n";
}
}
if (0 < pCallbackData->cmdBufLabelCount)
{
std::cerr << "\t" << "CommandBuffer Labels:\n";
for (uint8_t i = 0; i < pCallbackData->cmdBufLabelCount; i++)
{
std::cerr << "\t\t" << "labelName = <" << pCallbackData->pCmdBufLabels[i].pLabelName << ">\n";
}
}
if (0 < pCallbackData->objectCount)
{
std::cerr << "\t" << "Objects:\n";
for (uint8_t i = 0; i < pCallbackData->objectCount; i++)
{
std::cerr << "\t\t" << "Object " << i << "\n";
std::cerr << "\t\t\t" << "objectType = " << vk::to_string(static_cast(pCallbackData->pObjects[i].objectType)) << "\n";
std::cerr << "\t\t\t" << "objectHandle = " << pCallbackData->pObjects[i].objectHandle << "\n";
if (pCallbackData->pObjects[i].pObjectName)
{
std::cerr << "\t\t\t" << "objectName = <" << pCallbackData->pObjects[i].pObjectName << ">\n";
}
}
}
*/
switch (static_cast(messageSeverity))
{
case vk::DebugUtilsMessageSeverityFlagBitsEXT::eVerbose:
DEBUG_LOG(RENDERER, "%s", msg.c_str());
break;
case vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo:
INFO_LOG(RENDERER, "%s", msg.c_str());
break;
case vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning:
WARN_LOG(RENDERER, "%s", msg.c_str());
break;
case vk::DebugUtilsMessageSeverityFlagBitsEXT::eError:
default:
ERROR_LOG(RENDERER, "%s", msg.c_str());
break;
}
return VK_TRUE;
}
#else
static VkBool32 debugReportCallback(VkDebugReportFlagsEXT flags, VkDebugReportObjectTypeEXT objectType, uint64_t object, size_t location, int32_t messageCode,
const char* pLayerPrefix, const char* pMessage, void* /*pUserData*/)
{
std::string msg = pMessage;
if (flags & VK_DEBUG_REPORT_ERROR_BIT_EXT)
ERROR_LOG(RENDERER, "%s", msg.c_str());
else if (flags & VK_DEBUG_REPORT_WARNING_BIT_EXT)
WARN_LOG(RENDERER, "%s", msg.c_str());
else if (flags & (VK_DEBUG_REPORT_PERFORMANCE_WARNING_BIT_EXT | VK_DEBUG_REPORT_INFORMATION_BIT_EXT))
NOTICE_LOG(RENDERER, "%s", msg.c_str());
else
NOTICE_LOG(RENDERER, "(d) %s", msg.c_str());
return VK_FALSE;
}
#endif
static void CheckImGuiResult(VkResult err)
{
if (err != VK_SUCCESS)
WARN_LOG(RENDERER, "ImGui Vulkan error %d", err);
}
bool VulkanContext::InitInstance(const char** extensions, uint32_t extensions_count)
{
if (volkInitialize() != VK_SUCCESS)
{
ERROR_LOG(RENDERER, "Cannot load Vulkan libraries");
return false;
}
try
{
bool vulkan11 = false;
if (::vkEnumerateInstanceVersion != nullptr)
{
u32 apiVersion;
if (vk::enumerateInstanceVersion(&apiVersion) == vk::Result::eSuccess)
vulkan11 = VK_VERSION_MINOR(apiVersion) == 1;
}
vk::ApplicationInfo applicationInfo("Flycast", 1, "Flycast", 1, vulkan11 ? VK_API_VERSION_1_1 : VK_API_VERSION_1_0);
std::vector vext;
for (uint32_t i = 0; i < extensions_count; i++)
vext.push_back(extensions[i]);
std::vector layer_names;
//layer_names.push_back("VK_LAYER_ARM_AGA");
#ifdef VK_DEBUG
#ifndef __ANDROID__
vext.push_back("VK_EXT_debug_utils");
layer_names.push_back("VK_LAYER_LUNARG_standard_validation");
layer_names.push_back("VK_LAYER_LUNARG_assistant_layer");
#else
vext.push_back("VK_EXT_debug_report"); // NDK <= 19?
layer_names.push_back("VK_LAYER_GOOGLE_threading");
layer_names.push_back("VK_LAYER_LUNARG_parameter_validation");
layer_names.push_back("VK_LAYER_LUNARG_core_validation");
layer_names.push_back("VK_LAYER_GOOGLE_unique_objects");
#endif
#endif
vk::InstanceCreateInfo instanceCreateInfo({}, &applicationInfo, layer_names.size(), layer_names.data(),
vext.size(), vext.data());
// create a UniqueInstance
instance = vk::createInstanceUnique(instanceCreateInfo);
volkLoadInstance(static_cast(*instance));
#ifdef VK_DEBUG
#ifndef __ANDROID__
vk::DebugUtilsMessageSeverityFlagsEXT severityFlags(vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo
| vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning | vk::DebugUtilsMessageSeverityFlagBitsEXT::eError);
vk::DebugUtilsMessageTypeFlagsEXT messageTypeFlags(vk::DebugUtilsMessageTypeFlagBitsEXT::eGeneral
| vk::DebugUtilsMessageTypeFlagBitsEXT::ePerformance | vk::DebugUtilsMessageTypeFlagBitsEXT::eValidation);
debugUtilsMessenger = instance->createDebugUtilsMessengerEXTUnique(vk::DebugUtilsMessengerCreateInfoEXT({}, severityFlags, messageTypeFlags, debugUtilsMessengerCallback));
#else
vk::DebugReportCallbackCreateInfoEXT createInfo(vk::DebugReportFlagBitsEXT::eDebug | vk::DebugReportFlagBitsEXT::eInformation
| vk::DebugReportFlagBitsEXT::ePerformanceWarning | vk::DebugReportFlagBitsEXT::eWarning
| vk::DebugReportFlagBitsEXT::eError, &::debugReportCallback);
debugReportCallback = instance->createDebugReportCallbackEXTUnique(createInfo);
#endif
#endif
// Choose a discrete gpu if there's one, otherwise just pick the first one
physicalDevice = nullptr;
const auto devices = instance->enumeratePhysicalDevices();
for (const auto& phyDev : devices)
{
vk::PhysicalDeviceProperties props;
phyDev.getProperties(&props);
if (props.deviceType == vk::PhysicalDeviceType::eDiscreteGpu)
{
physicalDevice = phyDev;
break;
}
}
if (!physicalDevice)
physicalDevice = instance->enumeratePhysicalDevices().front();
const vk::PhysicalDeviceProperties *properties;
if (vulkan11 && ::vkGetPhysicalDeviceProperties2 != nullptr)
{
static vk::PhysicalDeviceProperties2 properties2;
vk::PhysicalDeviceMaintenance3Properties properties3;
properties2.pNext = &properties3;
physicalDevice.getProperties2(&properties2);
properties = &properties2.properties;
maxMemoryAllocationSize = properties3.maxMemoryAllocationSize;
if (maxMemoryAllocationSize == 0)
// Happens on Windows 7 with NVidia 376.33, ok on 441.66
maxMemoryAllocationSize = 0xFFFFFFFFu;
}
else
{
static vk::PhysicalDeviceProperties phyProperties;
physicalDevice.getProperties(&phyProperties);
properties = &phyProperties;
}
uniformBufferAlignment = properties->limits.minUniformBufferOffsetAlignment;
storageBufferAlignment = properties->limits.minStorageBufferOffsetAlignment;
maxStorageBufferRange = properties->limits.maxStorageBufferRange;
unifiedMemory = properties->deviceType == vk::PhysicalDeviceType::eIntegratedGpu;
vendorID = properties->vendorID;
NOTICE_LOG(RENDERER, "Vulkan API %s. Device %s", vulkan11 ? "1.1" : "1.0", properties->deviceName);
vk::FormatProperties formatProperties = physicalDevice.getFormatProperties(vk::Format::eR5G5B5A1UnormPack16);
if ((formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eSampledImage)
&& (formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eBlitDst)
&& (formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eBlitSrc))
optimalTilingSupported1555 = true;
else
NOTICE_LOG(RENDERER, "eR5G5B5A1UnormPack16 not supported for optimal tiling");
formatProperties = physicalDevice.getFormatProperties(vk::Format::eR5G6B5UnormPack16);
if ((formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eSampledImage)
&& (formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eBlitDst)
&& (formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eBlitSrc))
optimalTilingSupported565 = true;
else
NOTICE_LOG(RENDERER, "eR5G6B5UnormPack16 not supported for optimal tiling");
formatProperties = physicalDevice.getFormatProperties(vk::Format::eR4G4B4A4UnormPack16);
if ((formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eSampledImage)
&& (formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eBlitDst)
&& (formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eBlitSrc))
optimalTilingSupported4444 = true;
else
NOTICE_LOG(RENDERER, "eR4G4B4A4UnormPack16 not supported for optimal tiling");
vk::PhysicalDeviceFeatures features;
physicalDevice.getFeatures(&features);
fragmentStoresAndAtomics = features.fragmentStoresAndAtomics;
samplerAnisotropy = features.samplerAnisotropy;
ShaderCompiler::Init();
return true;
}
catch (const vk::SystemError& err)
{
ERROR_LOG(RENDERER, "Vulkan error: %s", err.what());
}
catch (...)
{
ERROR_LOG(RENDERER, "Unknown error");
}
return false;
}
vk::Format VulkanContext::FindDepthFormat()
{
const vk::Format depthFormats[] = { vk::Format::eD32SfloatS8Uint, vk::Format::eD24UnormS8Uint, vk::Format::eD16UnormS8Uint };
vk::ImageTiling tiling;
depthFormat = vk::Format::eUndefined;
for (size_t i = 0; i < ARRAY_SIZE(depthFormats); i++)
{
vk::FormatProperties formatProperties = physicalDevice.getFormatProperties(depthFormats[i]);
if (formatProperties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eDepthStencilAttachment)
{
tiling = vk::ImageTiling::eOptimal;
depthFormat = depthFormats[i];
break;
}
}
if (depthFormat == vk::Format::eUndefined)
{
// Try to find a linear format
for (size_t i = 0; i < ARRAY_SIZE(depthFormats); i++)
{
vk::FormatProperties formatProperties = physicalDevice.getFormatProperties(depthFormats[i]);
if (formatProperties.linearTilingFeatures & vk::FormatFeatureFlagBits::eDepthStencilAttachment)
{
tiling = vk::ImageTiling::eLinear;
depthFormat = depthFormats[i];
break;
}
}
if (depthFormat == vk::Format::eUndefined)
die("No supported depth/stencil format found");
}
NOTICE_LOG(RENDERER, "Using depth format %s tiling %s", vk::to_string(depthFormat).c_str(), vk::to_string(tiling).c_str());
return depthFormat;
}
void VulkanContext::InitImgui()
{
gui_init();
ImGui_ImplVulkan_InitInfo initInfo = {};
initInfo.Instance = (VkInstance)*instance;
initInfo.PhysicalDevice = (VkPhysicalDevice)physicalDevice;
initInfo.Device = (VkDevice)*device;
initInfo.QueueFamily = graphicsQueueIndex;
initInfo.Queue = (VkQueue)graphicsQueue;
initInfo.PipelineCache = (VkPipelineCache)*pipelineCache;
initInfo.DescriptorPool = (VkDescriptorPool)*descriptorPool;
#ifdef VK_DEBUG
initInfo.CheckVkResultFn = &CheckImGuiResult;
#endif
if (!ImGui_ImplVulkan_Init(&initInfo, (VkRenderPass)*renderPass, 0))
{
die("ImGui initialization failed");
}
if (ImGui::GetIO().Fonts->TexID == 0)
{
// Upload Fonts
device->resetFences(1, &(*drawFences.front()));
device->resetCommandPool(*commandPools.front(), vk::CommandPoolResetFlagBits::eReleaseResources);
vk::CommandBuffer& commandBuffer = *commandBuffers.front();
commandBuffer.begin(vk::CommandBufferBeginInfo(vk::CommandBufferUsageFlagBits::eOneTimeSubmit));
ImGui_ImplVulkan_CreateFontsTexture((VkCommandBuffer)commandBuffer);
commandBuffer.end();
vk::SubmitInfo submitInfo(0, nullptr, nullptr, 1, &commandBuffer);
graphicsQueue.submit(1, &submitInfo, *drawFences.front());
device->waitIdle();
ImGui_ImplVulkan_InvalidateFontUploadObjects();
}
}
bool VulkanContext::InitDevice()
{
if (!instance)
return false;
try
{
std::vector queueFamilyProperties = physicalDevice.getQueueFamilyProperties();
#ifdef VK_DEBUG
std::for_each(queueFamilyProperties.begin(), queueFamilyProperties.end(),
[](vk::QueueFamilyProperties const& qfp) { INFO_LOG(RENDERER, "Queue Family: count %d flags %s minImgGranularity %d x %d x %d",
qfp.queueCount, vk::to_string(qfp.queueFlags).c_str(), qfp.minImageTransferGranularity.width, qfp.minImageTransferGranularity.height,
qfp.minImageTransferGranularity.depth); });
#endif
// get the first index into queueFamiliyProperties which supports graphics
graphicsQueueIndex = (u32)std::distance(queueFamilyProperties.begin(),
std::find_if(queueFamilyProperties.begin(), queueFamilyProperties.end(),
[](vk::QueueFamilyProperties const& qfp) { return qfp.queueFlags & vk::QueueFlagBits::eGraphics; }));
verify(graphicsQueueIndex < queueFamilyProperties.size());
// determine a queueFamilyIndex that supports present
// first check if the graphicsQueueFamilyIndex is good enough
presentQueueIndex = physicalDevice.getSurfaceSupportKHR(graphicsQueueIndex, GetSurface()) ? graphicsQueueIndex : queueFamilyProperties.size();
if (presentQueueIndex == queueFamilyProperties.size())
{
// the graphicsQueueFamilyIndex doesn't support present -> look for an other family index that supports both graphics and present
for (size_t i = 0; i < queueFamilyProperties.size(); i++)
{
if ((queueFamilyProperties[i].queueFlags & vk::QueueFlagBits::eGraphics) && physicalDevice.getSurfaceSupportKHR((u32)i, GetSurface()))
{
graphicsQueueIndex = (u32)i;
presentQueueIndex = (u32)i;
break;
}
}
if (presentQueueIndex == queueFamilyProperties.size())
{
// there's nothing like a single family index that supports both graphics and present -> look for an other family index that supports present
DEBUG_LOG(RENDERER, "Using separate Graphics and Present queue families");
for (size_t i = 0; i < queueFamilyProperties.size(); i++)
{
if (physicalDevice.getSurfaceSupportKHR((u32)i, GetSurface()))
{
presentQueueIndex = (u32)i;
break;
}
}
}
}
if (graphicsQueueIndex == queueFamilyProperties.size() || presentQueueIndex == queueFamilyProperties.size())
die("Could not find a queue for graphics or present -> terminating");
if (graphicsQueueIndex == presentQueueIndex)
DEBUG_LOG(RENDERER, "Using Graphics+Present queue family");
else
DEBUG_LOG(RENDERER, "Using distinct Graphics and Present queue families");
// Enable VK_KHR_dedicated_allocation if available
bool getMemReq2Supported = false;
dedicatedAllocationSupported = false;
std::vector deviceExtensions = { VK_KHR_SWAPCHAIN_EXTENSION_NAME };
for (const auto& property : physicalDevice.enumerateDeviceExtensionProperties())
{
if (!strcmp(property.extensionName, VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME))
{
deviceExtensions.push_back(VK_KHR_GET_MEMORY_REQUIREMENTS_2_EXTENSION_NAME);
getMemReq2Supported = true;
}
else if (!strcmp(property.extensionName, VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME))
{
deviceExtensions.push_back(VK_KHR_DEDICATED_ALLOCATION_EXTENSION_NAME);
dedicatedAllocationSupported = true;
}
}
dedicatedAllocationSupported &= getMemReq2Supported;
// create a UniqueDevice
float queuePriority = 1.0f;
vk::DeviceQueueCreateInfo deviceQueueCreateInfo(vk::DeviceQueueCreateFlags(), graphicsQueueIndex, 1, &queuePriority);
vk::PhysicalDeviceFeatures features;
if (fragmentStoresAndAtomics)
features.fragmentStoresAndAtomics = true;
if (samplerAnisotropy)
features.samplerAnisotropy = true;
const char *layers[] = { "VK_LAYER_ARM_AGA" };
device = physicalDevice.createDeviceUnique(vk::DeviceCreateInfo(vk::DeviceCreateFlags(), 1, &deviceQueueCreateInfo,
0, layers, deviceExtensions.size(), &deviceExtensions[0], &features));
// This links entry points directly from the driver and isn't absolutely necessary
volkLoadDevice(static_cast(*device));
// Queues
graphicsQueue = device->getQueue(graphicsQueueIndex, 0);
presentQueue = device->getQueue(presentQueueIndex, 0);
// Descriptor pool
vk::DescriptorPoolSize pool_sizes[] =
{
{ vk::DescriptorType::eSampler, 2 },
{ vk::DescriptorType::eCombinedImageSampler, 4000 },
{ vk::DescriptorType::eSampledImage, 2 },
{ vk::DescriptorType::eStorageImage, 12 },
{ vk::DescriptorType::eUniformTexelBuffer, 2 },
{ vk::DescriptorType::eStorageTexelBuffer, 2 },
{ vk::DescriptorType::eUniformBuffer, 36 },
{ vk::DescriptorType::eStorageBuffer, 36 },
{ vk::DescriptorType::eUniformBufferDynamic, 2 },
{ vk::DescriptorType::eStorageBufferDynamic, 2 },
{ vk::DescriptorType::eInputAttachment, 36 }
};
descriptorPool = device->createDescriptorPoolUnique(vk::DescriptorPoolCreateInfo(vk::DescriptorPoolCreateFlagBits::eFreeDescriptorSet,
10000, ARRAY_SIZE(pool_sizes), pool_sizes));
std::string cachePath = get_writable_data_path(PipelineCacheFileName);
FILE *f = fopen(cachePath.c_str(), "rb");
if (f == nullptr)
pipelineCache = device->createPipelineCacheUnique(vk::PipelineCacheCreateInfo());
else
{
fseek(f, 0, SEEK_END);
size_t cacheSize = ftell(f);
fseek(f, 0, SEEK_SET);
u8 *cacheData = new u8[cacheSize];
if (fread(cacheData, 1, cacheSize, f) != cacheSize)
cacheSize = 0;
fclose(f);
pipelineCache = device->createPipelineCacheUnique(vk::PipelineCacheCreateInfo(vk::PipelineCacheCreateFlags(), cacheSize, cacheData));
delete [] cacheData;
INFO_LOG(RENDERER, "Vulkan pipeline cache loaded from %s: %zd bytes", cachePath.c_str(), cacheSize);
}
allocator.Init(physicalDevice, *device);
shaderManager = std::unique_ptr(new ShaderManager());
quadPipeline = std::unique_ptr(new QuadPipeline());
quadDrawer = std::unique_ptr(new QuadDrawer());
CreateSwapChain();
return true;
}
catch (const vk::SystemError& err)
{
ERROR_LOG(RENDERER, "Vulkan error: %s", err.what());
}
catch (...)
{
ERROR_LOG(RENDERER, "Unknown error");
}
return false;
}
void VulkanContext::CreateSwapChain()
{
try
{
device->waitIdle();
framebuffers.clear();
drawFences.clear();
imageAcquiredSemaphores.clear();
renderCompleteSemaphores.clear();
commandBuffers.clear();
commandPools.clear();
imageViews.clear();
// get the supported VkFormats
std::vector formats = physicalDevice.getSurfaceFormatsKHR(GetSurface());
assert(!formats.empty());
for (const auto& f : formats)
{
DEBUG_LOG(RENDERER, "Supported surface format: %s", vk::to_string(f.format).c_str());
// Try to find an non-sRGB color format
if (f.format == vk::Format::eB8G8R8A8Unorm || f.format == vk::Format::eR8G8B8A8Unorm)
{
colorFormat = f.format;
break;
}
}
if (colorFormat == vk::Format::eUndefined)
{
colorFormat = (formats[0].format == vk::Format::eUndefined) ? vk::Format::eB8G8R8A8Unorm : formats[0].format;
}
vk::SurfaceCapabilitiesKHR surfaceCapabilities = physicalDevice.getSurfaceCapabilitiesKHR(GetSurface());
DEBUG_LOG(RENDERER, "Surface capabilities: %d x %d, %s, image count: %d - %d", surfaceCapabilities.currentExtent.width, surfaceCapabilities.currentExtent.height,
vk::to_string(surfaceCapabilities.currentTransform).c_str(), surfaceCapabilities.minImageCount, surfaceCapabilities.maxImageCount);
VkExtent2D swapchainExtent;
if (surfaceCapabilities.currentExtent.width == std::numeric_limits::max())
{
// If the surface size is undefined, the size is set to the size of the images requested.
swapchainExtent.width = std::min(std::max(640u, surfaceCapabilities.minImageExtent.width), surfaceCapabilities.maxImageExtent.width);
swapchainExtent.height = std::min(std::max(480u, surfaceCapabilities.minImageExtent.height), surfaceCapabilities.maxImageExtent.height);
}
else
{
// If the surface size is defined, the swap chain size must match
swapchainExtent = surfaceCapabilities.currentExtent;
}
SetWindowSize(swapchainExtent.width, swapchainExtent.height);
// The FIFO present mode is guaranteed by the spec to be supported
vk::PresentModeKHR swapchainPresentMode = vk::PresentModeKHR::eFifo;
// Use FIFO on mobile, prefer Mailbox on desktop
#if HOST_CPU != CPU_ARM && HOST_CPU != CPU_ARM64 && !defined(__ANDROID__)
for (auto& presentMode : physicalDevice.getSurfacePresentModesKHR(GetSurface()))
{
if (presentMode == vk::PresentModeKHR::eMailbox)
{
INFO_LOG(RENDERER, "Using mailbox present mode");
swapchainPresentMode = vk::PresentModeKHR::eMailbox;
break;
}
#ifdef TEST_AUTOMATION
if (presentMode == vk::PresentModeKHR::eImmediate)
{
INFO_LOG(RENDERER, "Using immediate present mode");
swapchainPresentMode = vk::PresentModeKHR::eImmediate;
break;
}
#endif
}
#endif
vk::SurfaceTransformFlagBitsKHR preTransform = (surfaceCapabilities.supportedTransforms & vk::SurfaceTransformFlagBitsKHR::eIdentity) ? vk::SurfaceTransformFlagBitsKHR::eIdentity : surfaceCapabilities.currentTransform;
vk::CompositeAlphaFlagBitsKHR compositeAlpha =
(surfaceCapabilities.supportedCompositeAlpha & vk::CompositeAlphaFlagBitsKHR::ePreMultiplied) ? vk::CompositeAlphaFlagBitsKHR::ePreMultiplied :
(surfaceCapabilities.supportedCompositeAlpha & vk::CompositeAlphaFlagBitsKHR::ePostMultiplied) ? vk::CompositeAlphaFlagBitsKHR::ePostMultiplied :
(surfaceCapabilities.supportedCompositeAlpha & vk::CompositeAlphaFlagBitsKHR::eInherit) ? vk::CompositeAlphaFlagBitsKHR::eInherit : vk::CompositeAlphaFlagBitsKHR::eOpaque;
u32 imageCount = std::max(3u, surfaceCapabilities.minImageCount);
if (surfaceCapabilities.maxImageCount != 0)
imageCount = std::min(imageCount, surfaceCapabilities.maxImageCount);
vk::ImageUsageFlags usage = vk::ImageUsageFlagBits::eColorAttachment;
#ifdef TEST_AUTOMATION
// for final screenshot
usage |= vk::ImageUsageFlagBits::eTransferSrc;
#endif
vk::SwapchainCreateInfoKHR swapChainCreateInfo(vk::SwapchainCreateFlagsKHR(), GetSurface(), imageCount, colorFormat, vk::ColorSpaceKHR::eSrgbNonlinear,
swapchainExtent, 1, usage, vk::SharingMode::eExclusive, 0, nullptr, preTransform, compositeAlpha, swapchainPresentMode, true, nullptr);
u32 queueFamilyIndices[2] = { graphicsQueueIndex, presentQueueIndex };
if (graphicsQueueIndex != presentQueueIndex)
{
// If the graphics and present queues are from different queue families, we either have to explicitly transfer ownership of images between
// the queues, or we have to create the swapchain with imageSharingMode as VK_SHARING_MODE_CONCURRENT
swapChainCreateInfo.imageSharingMode = vk::SharingMode::eConcurrent;
swapChainCreateInfo.queueFamilyIndexCount = 2;
swapChainCreateInfo.pQueueFamilyIndices = queueFamilyIndices;
}
swapChain.reset();
swapChain = device->createSwapchainKHRUnique(swapChainCreateInfo);
std::vector swapChainImages = device->getSwapchainImagesKHR(*swapChain);
imageViews.reserve(swapChainImages.size());
commandPools.reserve(swapChainImages.size());
commandBuffers.reserve(swapChainImages.size());
vk::ComponentMapping componentMapping(vk::ComponentSwizzle::eR, vk::ComponentSwizzle::eG, vk::ComponentSwizzle::eB, vk::ComponentSwizzle::eA);
vk::ImageSubresourceRange subResourceRange(vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1);
for (auto image : swapChainImages)
{
vk::ImageViewCreateInfo imageViewCreateInfo(vk::ImageViewCreateFlags(), image, vk::ImageViewType::e2D, colorFormat, componentMapping, subResourceRange);
imageViews.push_back(device->createImageViewUnique(imageViewCreateInfo));
// create a UniqueCommandPool to allocate a CommandBuffer from
commandPools.push_back(device->createCommandPoolUnique(vk::CommandPoolCreateInfo(vk::CommandPoolCreateFlagBits::eTransient, graphicsQueueIndex)));
// allocate a CommandBuffer from the CommandPool
commandBuffers.push_back(std::move(device->allocateCommandBuffersUnique(vk::CommandBufferAllocateInfo(*commandPools.back(), vk::CommandBufferLevel::ePrimary, 1)).front()));
}
FindDepthFormat();
// Render pass
vk::AttachmentDescription attachmentDescription = vk::AttachmentDescription(vk::AttachmentDescriptionFlags(), colorFormat, vk::SampleCountFlagBits::e1,
vk::AttachmentLoadOp::eClear, vk::AttachmentStoreOp::eStore, vk::AttachmentLoadOp::eDontCare, vk::AttachmentStoreOp::eDontCare,
vk::ImageLayout::eUndefined, vk::ImageLayout::ePresentSrcKHR);
vk::AttachmentReference colorReference(0, vk::ImageLayout::eColorAttachmentOptimal);
vk::SubpassDescription subpass(vk::SubpassDescriptionFlags(), vk::PipelineBindPoint::eGraphics, 0, nullptr, 1, &colorReference,
nullptr, nullptr);
renderPass = device->createRenderPassUnique(vk::RenderPassCreateInfo(vk::RenderPassCreateFlags(),
1, &attachmentDescription, 1, &subpass));
// Framebuffers, fences, semaphores
framebuffers.reserve(imageViews.size());
drawFences.reserve(imageViews.size());
renderCompleteSemaphores.reserve(imageViews.size());
imageAcquiredSemaphores.reserve(imageViews.size());
for (auto const& view : imageViews)
{
framebuffers.push_back(device->createFramebufferUnique(vk::FramebufferCreateInfo(vk::FramebufferCreateFlags(), *renderPass,
1, &view.get(), width, height, 1)));
drawFences.push_back(device->createFenceUnique(vk::FenceCreateInfo(vk::FenceCreateFlagBits::eSignaled)));
renderCompleteSemaphores.push_back(device->createSemaphoreUnique(vk::SemaphoreCreateInfo()));
imageAcquiredSemaphores.push_back(device->createSemaphoreUnique(vk::SemaphoreCreateInfo()));
}
quadPipeline->Init(shaderManager.get(), *renderPass);
quadDrawer->Init(quadPipeline.get());
vmus->Init(quadPipeline.get());
InitImgui();
currentImage = GetSwapChainSize() - 1;
INFO_LOG(RENDERER, "Vulkan swap chain created: %d x %d, swap chain size %d", width, height, (int)imageViews.size());
}
catch (const vk::SystemError& err)
{
ERROR_LOG(RENDERER, "Vulkan error: %s", err.what());
}
catch (...)
{
ERROR_LOG(RENDERER, "Unknown error");
}
}
bool VulkanContext::Init()
{
std::vector extensions;
extensions.push_back(VK_KHR_SURFACE_EXTENSION_NAME);
#if defined(USE_SDL)
sdl_recreate_window(SDL_WINDOW_VULKAN);
uint32_t extensionsCount = 0;
SDL_Vulkan_GetInstanceExtensions((SDL_Window *)window, &extensionsCount, NULL);
extensions.resize(extensionsCount + 1);
SDL_Vulkan_GetInstanceExtensions((SDL_Window *)window, &extensionsCount, &extensions[1]);
#elif defined(_WIN32)
extern void CreateMainWindow();
CreateMainWindow();
extensions.push_back(VK_KHR_WIN32_SURFACE_EXTENSION_NAME);
#elif defined(__MACH__)
extensions.push_back(VK_MVK_MACOS_SURFACE_EXTENSION_NAME);
#elif defined(SUPPORT_X11)
extensions.push_back(VK_KHR_XLIB_SURFACE_EXTENSION_NAME);
#elif defined(__ANDROID__)
extensions.push_back(VK_KHR_ANDROID_SURFACE_EXTENSION_NAME);
#endif
if (!InitInstance(&extensions[0], extensions.size()))
return false;
#if defined(USE_SDL)
VkSurfaceKHR surface;
if (SDL_Vulkan_CreateSurface((SDL_Window *)window, *instance, (VkSurfaceKHR *)&this->surface) == 0)
return false;
#elif defined(_WIN32)
vk::Win32SurfaceCreateInfoKHR createInfo(vk::Win32SurfaceCreateFlagsKHR(), GetModuleHandle(NULL), (HWND)window);
surface = instance->createWin32SurfaceKHRUnique(createInfo);
#elif defined(SUPPORT_X11)
vk::XlibSurfaceCreateInfoKHR createInfo(vk::XlibSurfaceCreateFlagsKHR(), (Display*)display, (Window)window);
surface = instance->createXlibSurfaceKHRUnique(createInfo);
#elif defined(__ANDROID__)
vk::AndroidSurfaceCreateInfoKHR createInfo(vk::AndroidSurfaceCreateFlagsKHR(), (struct ANativeWindow*)window);
surface = instance->createAndroidSurfaceKHRUnique(createInfo);
#endif
vmus = std::unique_ptr(new VulkanVMUs());
return InitDevice();
}
void VulkanContext::NewFrame()
{
if (HasSurfaceDimensionChanged())
{
CreateSwapChain();
rend_resize(width, height);
}
device->acquireNextImageKHR(*swapChain, UINT64_MAX, *imageAcquiredSemaphores[currentSemaphore], nullptr, ¤tImage);
device->waitForFences(1, &(*drawFences[currentImage]), true, UINT64_MAX);
device->resetFences(1, &(*drawFences[currentImage]));
device->resetCommandPool(*commandPools[currentImage], vk::CommandPoolResetFlagBits::eReleaseResources);
vk::CommandBuffer commandBuffer = *commandBuffers[currentImage];
commandBuffer.begin(vk::CommandBufferBeginInfo(vk::CommandBufferUsageFlagBits::eOneTimeSubmit));
verify(!rendering);
rendering = true;
}
void VulkanContext::BeginRenderPass()
{
const vk::ClearValue clear_colors[] = { vk::ClearColorValue(std::array{0.f, 0.f, 0.f, 1.f}), vk::ClearDepthStencilValue{ 0.f, 0 } };
vk::CommandBuffer commandBuffer = *commandBuffers[currentImage];
commandBuffer.beginRenderPass(vk::RenderPassBeginInfo(*renderPass, *framebuffers[currentImage], vk::Rect2D({0, 0}, {width, height}), 2, clear_colors),
vk::SubpassContents::eInline);
}
void VulkanContext::EndFrame(const std::vector *cmdBuffers)
{
vk::CommandBuffer commandBuffer = *commandBuffers[currentImage];
commandBuffer.endRenderPass();
commandBuffer.end();
vk::PipelineStageFlags wait_stage = vk::PipelineStageFlagBits::eColorAttachmentOutput;
std::vector allCmdBuffers;
if (cmdBuffers != nullptr)
allCmdBuffers = vk::uniqueToRaw(*cmdBuffers);
allCmdBuffers.push_back(commandBuffer);
vk::SubmitInfo submitInfo(1, &(*imageAcquiredSemaphores[currentSemaphore]), &wait_stage, allCmdBuffers.size(),allCmdBuffers.data(),
1, &(*renderCompleteSemaphores[currentSemaphore]));
graphicsQueue.submit(1, &submitInfo, *drawFences[currentImage]);
verify(rendering);
rendering = false;
renderDone = true;
}
void VulkanContext::Present()
{
if (renderDone)
{
try {
DoSwapAutomation();
presentQueue.presentKHR(vk::PresentInfoKHR(1, &(*renderCompleteSemaphores[currentSemaphore]), 1, &(*swapChain), ¤tImage));
currentSemaphore = (currentSemaphore + 1) % imageViews.size();
} catch (const vk::OutOfDateKHRError& e) {
// Sometimes happens when resizing the window
INFO_LOG(RENDERER, "vk::OutOfDateKHRError");
}
renderDone = false;
}
}
void VulkanContext::DrawFrame(vk::ImageView imageView, vk::Offset2D extent)
{
float marginWidth = ((float)extent.y / extent.x * width / height - 1.f) / 2.f;
QuadVertex vtx[] = {
{ { -1, -1, 0 }, { 0 - marginWidth, 0 } },
{ { 1, -1, 0 }, { 1 + marginWidth, 0 } },
{ { -1, 1, 0 }, { 0 - marginWidth, 1 } },
{ { 1, 1, 0 }, { 1 + marginWidth, 1 } },
};
vk::CommandBuffer commandBuffer = GetCurrentCommandBuffer();
quadPipeline->BindPipeline(commandBuffer);
float blendConstants[4] = { 1.0, 1.0, 1.0, 1.0 };
commandBuffer.setBlendConstants(blendConstants);
vk::Viewport viewport(0, 0, width, height);
commandBuffer.setViewport(0, 1, &viewport);
commandBuffer.setScissor(0, vk::Rect2D(vk::Offset2D(0, 0), vk::Extent2D(width, height)));
quadDrawer->Draw(commandBuffer, imageView, vtx);
}
const std::vector *VulkanContext::PrepareVMUs()
{
return vmus->PrepareVMUs(*commandPools[GetCurrentImageIndex()]);
}
void VulkanContext::DrawVMUs(float scaling)
{
vmus->DrawVMUs(vk::Extent2D(width, height), scaling);
}
extern Renderer *renderer;
void VulkanContext::PresentFrame(vk::ImageView imageView, vk::Offset2D extent)
{
NewFrame();
const std::vector *vmuCmdBuffers = nullptr;
if (settings.rend.FloatVMUs)
vmuCmdBuffers = PrepareVMUs();
BeginRenderPass();
DrawFrame(imageView, extent);
if (settings.rend.FloatVMUs)
DrawVMUs(gui_get_scaling());
renderer->DrawOSD(false);
EndFrame(vmuCmdBuffers);
lastFrameView = imageView;
lastFrameExtent = extent;
}
void VulkanContext::PresentLastFrame()
{
if (lastFrameView)
DrawFrame(lastFrameView, lastFrameExtent);
}
void VulkanContext::Term()
{
lastFrameView = nullptr;
ImGui_ImplVulkan_Shutdown();
gui_term();
if (device && pipelineCache)
{
std::vector cacheData = device->getPipelineCacheData(*pipelineCache);
if (!cacheData.empty())
{
std::string cachePath = get_writable_data_path(PipelineCacheFileName);
FILE *f = fopen(cachePath.c_str(), "wb");
if (f != nullptr)
{
(void)fwrite(&cacheData[0], 1, cacheData.size(), f);
fclose(f);
}
}
}
vmus.reset();
ShaderCompiler::Term();
swapChain.reset();
imageViews.clear();
framebuffers.clear();
renderPass.reset();
quadDrawer.reset();
quadPipeline.reset();
shaderManager.reset();
descriptorPool.reset();
commandBuffers.clear();
commandPools.clear();
imageAcquiredSemaphores.clear();
renderCompleteSemaphores.clear();
drawFences.clear();
allocator.Term();
#ifndef USE_SDL
surface.reset();
#endif
pipelineCache.reset();
device.reset();
#ifdef VK_DEBUG
#ifndef __ANDROID__
debugUtilsMessenger.reset();
#else
debugReportCallback.reset();
#endif
#endif
instance.reset();
#ifdef _WIN32
extern void DestroyMainWindow();
DestroyMainWindow();
#endif
}
void VulkanContext::DoSwapAutomation()
{
#ifdef TEST_AUTOMATION
extern void dc_exit();
extern bool do_screenshot;
if (do_screenshot)
{
bool supportsBlit = true;
vk::FormatProperties properties;
physicalDevice.getFormatProperties(colorFormat, &properties);
if (!(properties.optimalTilingFeatures & vk::FormatFeatureFlagBits::eBlitSrc))
supportsBlit = false;
physicalDevice.getFormatProperties(vk::Format::eR8G8B8A8Unorm, &properties);
if (!(properties.linearTilingFeatures & vk::FormatFeatureFlagBits::eBlitDst))
supportsBlit = false;
{
vk::Image srcImage = device->getSwapchainImagesKHR(*swapChain)[currentImage];
vk::ImageCreateInfo imageCreateInfo(vk::ImageCreateFlags(), vk::ImageType::e2D, vk::Format::eR8G8B8A8Unorm,
vk::Extent3D(width, height, 1), 1, 1,
vk::SampleCountFlagBits::e1, vk::ImageTiling::eLinear, vk::ImageUsageFlagBits::eTransferDst,
vk::SharingMode::eExclusive, 0, nullptr, vk::ImageLayout::eUndefined);
vk::UniqueImage dstImage = device->createImageUnique(imageCreateInfo);
vk::MemoryRequirements memReq = device->getImageMemoryRequirements(*dstImage);
u32 memoryType = findMemoryType(physicalDevice.getMemoryProperties(), memReq.memoryTypeBits,
vk::MemoryPropertyFlagBits::eHostCoherent | vk::MemoryPropertyFlagBits::eHostVisible);
vk::UniqueDeviceMemory deviceMemory = device->allocateMemoryUnique(vk::MemoryAllocateInfo(memReq.size, memoryType));
device->bindImageMemory(dstImage.get(), *deviceMemory, 0);
vk::UniqueCommandBuffer cmdBuffer = std::move(device->allocateCommandBuffersUnique(
vk::CommandBufferAllocateInfo(*commandPools.back(), vk::CommandBufferLevel::ePrimary, 1)).front());
cmdBuffer->begin(vk::CommandBufferBeginInfo(vk::CommandBufferUsageFlagBits::eOneTimeSubmit));
// Transition destination image to transfer destination layout
vk::ImageMemoryBarrier barrier(vk::AccessFlags(), vk::AccessFlagBits::eTransferWrite,
vk::ImageLayout::eUndefined, vk::ImageLayout::eTransferDstOptimal,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED,
*dstImage, vk::ImageSubresourceRange(vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1));
cmdBuffer->pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eTransfer,
vk::DependencyFlags(), nullptr, nullptr, barrier);
// Transition swapchain image from present to transfer source layout
barrier = vk::ImageMemoryBarrier(vk::AccessFlagBits::eMemoryRead, vk::AccessFlagBits::eTransferRead,
vk::ImageLayout::ePresentSrcKHR, vk::ImageLayout::eTransferSrcOptimal,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED,
srcImage, vk::ImageSubresourceRange(vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1));
cmdBuffer->pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eTransfer,
vk::DependencyFlags(), nullptr, nullptr, barrier);
if (supportsBlit)
{
vk::Offset3D blitSize(width, height, 1);
vk::ImageBlit imageBlit(
vk::ImageSubresourceLayers(vk::ImageAspectFlagBits::eColor, 0, 0, 1), { vk::Offset3D(), blitSize },
vk::ImageSubresourceLayers(vk::ImageAspectFlagBits::eColor, 0, 0, 1), { vk::Offset3D(), blitSize });
cmdBuffer->blitImage(srcImage, vk::ImageLayout::eTransferSrcOptimal, *dstImage, vk::ImageLayout::eTransferDstOptimal,
imageBlit, vk::Filter::eNearest);
}
else
{
vk::ImageCopy imageCopy(vk::ImageSubresourceLayers(vk::ImageAspectFlagBits::eColor, 0, 0, 1), vk::Offset3D(),
vk::ImageSubresourceLayers(vk::ImageAspectFlagBits::eColor, 0, 0, 1), vk::Offset3D(), { width, height, 1 });
cmdBuffer->copyImage(srcImage, vk::ImageLayout::eTransferSrcOptimal, *dstImage, vk::ImageLayout::eTransferDstOptimal,
imageCopy);
}
// Transition destination image to general layout, which is the required layout for mapping the image memory later on
barrier = vk::ImageMemoryBarrier(vk::AccessFlagBits::eTransferWrite, vk::AccessFlagBits::eMemoryRead,
vk::ImageLayout::eTransferDstOptimal, vk::ImageLayout::eGeneral,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED,
*dstImage, vk::ImageSubresourceRange(vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1));
cmdBuffer->pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eTransfer,
vk::DependencyFlags(), nullptr, nullptr, barrier);
// Transition back the swap chain image after the blit is done
barrier = vk::ImageMemoryBarrier(vk::AccessFlagBits::eTransferRead, vk::AccessFlagBits::eMemoryRead,
vk::ImageLayout::eTransferSrcOptimal, vk::ImageLayout::ePresentSrcKHR,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED,
srcImage, vk::ImageSubresourceRange(vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1));
cmdBuffer->pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eTransfer,
vk::DependencyFlags(), nullptr, nullptr, barrier);
cmdBuffer->end();
vk::PipelineStageFlags wait_stage = vk::PipelineStageFlagBits::eColorAttachmentOutput;
vk::SubmitInfo submitInfo(0, nullptr, nullptr, 1, &cmdBuffer.get(), 0, nullptr);
graphicsQueue.submit(1, &submitInfo, nullptr);
graphicsQueue.waitIdle();
vk::ImageSubresource subresource(vk::ImageAspectFlagBits::eColor, 0, 0);
vk::SubresourceLayout subresourceLayout;
device->getImageSubresourceLayout(*dstImage, &subresource, &subresourceLayout);
u8* img = (u8*)device->mapMemory(*deviceMemory, 0, VK_WHOLE_SIZE);
img += subresourceLayout.offset;
u8 *end = img + screen_width * screen_height * 4;
if (!supportsBlit && colorFormat == vk::Format::eB8G8R8A8Unorm)
{
for (u8 *p = img; p < end; p += 4)
{
u8 b = p[0];
p[0] = p[2];
p[2] = b;
p[3] = 0xff;
}
}
else
{
for (u8 *p = img; p < end; p += 4)
p[3] = 0xff;
}
dump_screenshot(img, screen_width, screen_height, true, subresourceLayout.rowPitch, false);
device->unmapMemory(*deviceMemory);
}
dc_exit();
Term();
exit(0);
}
#endif
}