vulkan/vulkan_swapchain.cc - mirrors/engine - Git at Google

 // Copyright 2013 The Flutter Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.

 #include "vulkan_swapchain.h"

 #include "flutter/vulkan/procs/vulkan_proc_table.h"

 #include "third_party/skia/include/core/SkColorSpace.h"
 #include "third_party/skia/include/core/SkSurface.h"
 #include "third_party/skia/include/gpu/GrBackendSurface.h"
 #include "third_party/skia/include/gpu/GrDirectContext.h"
 #include "third_party/skia/include/gpu/ganesh/SkSurfaceGanesh.h"
 #include "third_party/skia/include/gpu/ganesh/vk/GrVkBackendSurface.h"
 #include "third_party/skia/include/gpu/vk/GrVkTypes.h"

 #include "vulkan_backbuffer.h"
 #include "vulkan_device.h"
 #include "vulkan_image.h"
 #include "vulkan_surface.h"

 namespace vulkan {

 namespace {
 struct FormatInfo {
   VkFormat format_;
   SkColorType color_type_;
   sk_sp<SkColorSpace> color_space_;
 };
 }  // namespace

 static std::vector<FormatInfo> DesiredFormatInfos() {
   return {{VK_FORMAT_R8G8B8A8_SRGB, kRGBA_8888_SkColorType,
            SkColorSpace::MakeSRGB()},
           {VK_FORMAT_B8G8R8A8_SRGB, kRGBA_8888_SkColorType,
            SkColorSpace::MakeSRGB()},
           {VK_FORMAT_R16G16B16A16_SFLOAT, kRGBA_F16_SkColorType,
            SkColorSpace::MakeSRGBLinear()},
           {VK_FORMAT_R8G8B8A8_UNORM, kRGBA_8888_SkColorType,
            SkColorSpace::MakeSRGB()},
           {VK_FORMAT_B8G8R8A8_UNORM, kRGBA_8888_SkColorType,
            SkColorSpace::MakeSRGB()}};
 }

 VulkanSwapchain::VulkanSwapchain(const VulkanProcTable& p_vk,
                                  const VulkanDevice& device,
                                  const VulkanSurface& surface,
                                  GrDirectContext* skia_context,
                                  std::unique_ptr<VulkanSwapchain> old_swapchain,
                                  uint32_t queue_family_index)
     : vk(p_vk),
       device_(device),
       capabilities_(),
       surface_format_(),
       current_pipeline_stage_(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
       current_backbuffer_index_(0),
       current_image_index_(0),
       valid_(false) {
   if (!device_.IsValid() || !surface.IsValid() || skia_context == nullptr) {
     FML_DLOG(INFO) << "Device or surface is invalid.";
     return;
   }

   if (!device_.GetSurfaceCapabilities(surface, &capabilities_)) {
     FML_DLOG(INFO) << "Could not find surface capabilities.";
     return;
   }

   const auto format_infos = DesiredFormatInfos();
   std::vector<VkFormat> desired_formats(format_infos.size());
   for (size_t i = 0; i < format_infos.size(); ++i) {
     if (skia_context->colorTypeSupportedAsSurface(
             format_infos[i].color_type_)) {
       desired_formats[i] = format_infos[i].format_;
     } else {
       desired_formats[i] = VK_FORMAT_UNDEFINED;
     }
   }

   int format_index =
       device_.ChooseSurfaceFormat(surface, desired_formats, &surface_format_);
   if (format_index < 0) {
     FML_DLOG(INFO) << "Could not choose surface format.";
     return;
   }

   VkPresentModeKHR present_mode = VK_PRESENT_MODE_FIFO_KHR;
   if (!device_.ChoosePresentMode(surface, &present_mode)) {
     FML_DLOG(INFO) << "Could not choose present mode.";
     return;
   }

   // Check if the surface can present.

   VkBool32 supported = VK_FALSE;
   if (VK_CALL_LOG_ERROR(vk.GetPhysicalDeviceSurfaceSupportKHR(
           device_.GetPhysicalDeviceHandle(),  // physical device
           queue_family_index,                 // queue family
           surface.Handle(),                   // surface to test
           &supported)) != VK_SUCCESS) {
     FML_DLOG(INFO) << "Could not get physical device surface support.";
     return;
   }

   if (supported != VK_TRUE) {
     FML_DLOG(INFO) << "Surface was not supported by the physical device.";
     return;
   }

   // Construct the Swapchain

   VkSwapchainKHR old_swapchain_handle = VK_NULL_HANDLE;

   if (old_swapchain != nullptr && old_swapchain->IsValid()) {
     old_swapchain_handle = old_swapchain->swapchain_;
     // The unique pointer to the swapchain will go out of scope here
     // and its handle collected after the appropriate device wait.
   }

   VkSurfaceKHR surface_handle = surface.Handle();

   VkImageUsageFlags usage_flags = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT |
                                   VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
                                   VK_IMAGE_USAGE_TRANSFER_DST_BIT;

   const VkSwapchainCreateInfoKHR create_info = {
       .sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR,
       .pNext = nullptr,
       .flags = 0,
       .surface = surface_handle,
       .minImageCount = capabilities_.minImageCount,
       .imageFormat = surface_format_.format,
       .imageColorSpace = surface_format_.colorSpace,
       .imageExtent = capabilities_.currentExtent,
       .imageArrayLayers = 1,
       .imageUsage = usage_flags,
       .imageSharingMode = VK_SHARING_MODE_EXCLUSIVE,
       .queueFamilyIndexCount = 0,  // Because of the exclusive sharing mode.
       .pQueueFamilyIndices = nullptr,
       .preTransform = VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR,
       .compositeAlpha = VK_COMPOSITE_ALPHA_INHERIT_BIT_KHR,
       .presentMode = present_mode,
       .clipped = VK_FALSE,
       .oldSwapchain = old_swapchain_handle,
   };

   VkSwapchainKHR swapchain = VK_NULL_HANDLE;

   if (VK_CALL_LOG_ERROR(vk.CreateSwapchainKHR(device_.GetHandle(), &create_info,
                                               nullptr, &swapchain)) !=
       VK_SUCCESS) {
     FML_DLOG(INFO) << "Could not create the swapchain.";
     return;
   }

   swapchain_ = VulkanHandle<VkSwapchainKHR>{
       swapchain, [this](VkSwapchainKHR swapchain) {
         FML_ALLOW_UNUSED_LOCAL(device_.WaitIdle());
         vk.DestroySwapchainKHR(device_.GetHandle(), swapchain, nullptr);
       }};

   if (!CreateSwapchainImages(
           skia_context, format_infos[format_index].color_type_,
           format_infos[format_index].color_space_, usage_flags)) {
     FML_DLOG(INFO) << "Could not create swapchain images.";
     return;
   }

   valid_ = true;
 }

 VulkanSwapchain::~VulkanSwapchain() = default;

 bool VulkanSwapchain::IsValid() const {
   return valid_;
 }

 std::vector<VkImage> VulkanSwapchain::GetImages() const {
   uint32_t count = 0;
   if (VK_CALL_LOG_ERROR(vk.GetSwapchainImagesKHR(
           device_.GetHandle(), swapchain_, &count, nullptr)) != VK_SUCCESS) {
     return {};
   }

   if (count == 0) {
     return {};
   }

   std::vector<VkImage> images;

   images.resize(count);

   if (VK_CALL_LOG_ERROR(vk.GetSwapchainImagesKHR(
           device_.GetHandle(), swapchain_, &count, images.data())) !=
       VK_SUCCESS) {
     return {};
   }

   return images;
 }

 SkISize VulkanSwapchain::GetSize() const {
   VkExtent2D extents = capabilities_.currentExtent;

   if (extents.width < capabilities_.minImageExtent.width) {
     extents.width = capabilities_.minImageExtent.width;
   } else if (extents.width > capabilities_.maxImageExtent.width) {
     extents.width = capabilities_.maxImageExtent.width;
   }

   if (extents.height < capabilities_.minImageExtent.height) {
     extents.height = capabilities_.minImageExtent.height;
   } else if (extents.height > capabilities_.maxImageExtent.height) {
     extents.height = capabilities_.maxImageExtent.height;
   }

   return SkISize::Make(extents.width, extents.height);
 }

 sk_sp<SkSurface> VulkanSwapchain::CreateSkiaSurface(
     GrDirectContext* gr_context,
     VkImage image,
     VkImageUsageFlags usage_flags,
     const SkISize& size,
     SkColorType color_type,
     sk_sp<SkColorSpace> color_space) const {
   if (gr_context == nullptr) {
     return nullptr;
   }

   if (color_type == kUnknown_SkColorType) {
     // Unexpected Vulkan format.
     return nullptr;
   }

   GrVkImageInfo image_info;
   image_info.fImage = image;
   image_info.fImageTiling = VK_IMAGE_TILING_OPTIMAL;
   image_info.fImageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
   image_info.fFormat = surface_format_.format;
   image_info.fImageUsageFlags = usage_flags;
   image_info.fSampleCount = 1;
   image_info.fLevelCount = 1;

   // TODO(chinmaygarde): Setup the stencil buffer and the sampleCnt.
   auto backend_render_target =
       GrBackendRenderTargets::MakeVk(size.fWidth, size.fHeight, image_info);
   SkSurfaceProps props(0, kUnknown_SkPixelGeometry);

   return SkSurfaces::WrapBackendRenderTarget(
       gr_context,                // context
       backend_render_target,     // backend render target
       kTopLeft_GrSurfaceOrigin,  // origin
       color_type,                // color type
       std::move(color_space),    // color space
       &props                     // surface properties
   );
 }

 bool VulkanSwapchain::CreateSwapchainImages(
     GrDirectContext* skia_context,
     SkColorType color_type,
     const sk_sp<SkColorSpace>& color_space,
     VkImageUsageFlags usage_flags) {
   std::vector<VkImage> images = GetImages();

   if (images.empty()) {
     return false;
   }

   const SkISize surface_size = GetSize();

   for (const VkImage& image : images) {
     // Populate the backbuffer.
     auto backbuffer = std::make_unique<VulkanBackbuffer>(
         vk, device_.GetHandle(), device_.GetCommandPool());

     if (!backbuffer->IsValid()) {
       return false;
     }

     backbuffers_.emplace_back(std::move(backbuffer));

     // Populate the image.
     VulkanHandle<VkImage> image_handle = VulkanHandle<VkImage>{
         image, [this](VkImage image) {
           vk.DestroyImage(device_.GetHandle(), image, nullptr);
         }};
     auto vulkan_image = std::make_unique<VulkanImage>(std::move(image_handle));

     if (!vulkan_image->IsValid()) {
       return false;
     }

     images_.emplace_back(std::move(vulkan_image));

     // Populate the surface.
     auto surface = CreateSkiaSurface(skia_context, image, usage_flags,
                                      surface_size, color_type, color_space);

     if (surface == nullptr) {
       return false;
     }

     surfaces_.emplace_back(std::move(surface));
   }

   FML_DCHECK(backbuffers_.size() == images_.size());
   FML_DCHECK(images_.size() == surfaces_.size());

   return true;
 }

 VulkanBackbuffer* VulkanSwapchain::GetNextBackbuffer() {
   auto available_backbuffers = backbuffers_.size();

   if (available_backbuffers == 0) {
     return nullptr;
   }

   auto next_backbuffer_index =
       (current_backbuffer_index_ + 1) % backbuffers_.size();

   auto& backbuffer = backbuffers_[next_backbuffer_index];

   if (!backbuffer->IsValid()) {
     return nullptr;
   }

   current_backbuffer_index_ = next_backbuffer_index;
   return backbuffer.get();
 }

 VulkanSwapchain::AcquireResult VulkanSwapchain::AcquireSurface() {
   AcquireResult error = {AcquireStatus::ErrorSurfaceLost, nullptr};

   if (!IsValid()) {
     FML_DLOG(INFO) << "Swapchain was invalid.";
     return error;
   }

   // ---------------------------------------------------------------------------
   // Step 0:
   // Acquire the next available backbuffer.
   // ---------------------------------------------------------------------------
   auto backbuffer = GetNextBackbuffer();

   if (backbuffer == nullptr) {
     FML_DLOG(INFO) << "Could not get the next backbuffer.";
     return error;
   }

   // ---------------------------------------------------------------------------
   // Step 1:
   // Wait for use readiness.
   // ---------------------------------------------------------------------------
   if (!backbuffer->WaitFences()) {
     FML_DLOG(INFO) << "Failed waiting on fences.";
     return error;
   }

   // ---------------------------------------------------------------------------
   // Step 2:
   // Put fences in an unsignaled state.
   // ---------------------------------------------------------------------------
   if (!backbuffer->ResetFences()) {
     FML_DLOG(INFO) << "Could not reset fences.";
     return error;
   }

   // ---------------------------------------------------------------------------
   // Step 3:
   // Acquire the next image index.
   // ---------------------------------------------------------------------------
   uint32_t next_image_index = 0;

   VkResult acquire_result = VK_CALL_LOG_ERROR(
       vk.AcquireNextImageKHR(device_.GetHandle(),                   //
                              swapchain_,                            //
                              std::numeric_limits<uint64_t>::max(),  //
                              backbuffer->GetUsageSemaphore(),       //
                              VK_NULL_HANDLE,                        //
                              &next_image_index));

   switch (acquire_result) {
     case VK_SUCCESS:
       break;
     case VK_ERROR_OUT_OF_DATE_KHR:
       return {AcquireStatus::ErrorSurfaceOutOfDate, nullptr};
     case VK_ERROR_SURFACE_LOST_KHR:
       return {AcquireStatus::ErrorSurfaceLost, nullptr};
     default:
       FML_LOG(INFO) << "Unexpected result from AcquireNextImageKHR: "
                     << acquire_result;
       return {AcquireStatus::ErrorSurfaceLost, nullptr};
   }

   // Simple sanity checking of image index.
   if (next_image_index >= images_.size()) {
     FML_DLOG(INFO) << "Image index returned was out-of-bounds.";
     return error;
   }

   auto& image = images_[next_image_index];
   if (!image->IsValid()) {
     FML_DLOG(INFO) << "Image at index was invalid.";
     return error;
   }

   // ---------------------------------------------------------------------------
   // Step 4:
   // Start recording to the command buffer.
   // ---------------------------------------------------------------------------
   if (!backbuffer->GetUsageCommandBuffer().Begin()) {
     FML_DLOG(INFO) << "Could not begin recording to the command buffer.";
     return error;
   }

   // ---------------------------------------------------------------------------
   // Step 5:
   // Set image layout to color attachment mode.
   // ---------------------------------------------------------------------------
   VkPipelineStageFlagBits destination_pipeline_stage =
       VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
   VkImageLayout destination_image_layout =
       VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

   if (!image->InsertImageMemoryBarrier(
           backbuffer->GetUsageCommandBuffer(),   // command buffer
           current_pipeline_stage_,               // src_pipeline_bits
           destination_pipeline_stage,            // dest_pipeline_bits
           VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,  // dest_access_flags
           destination_image_layout               // dest_layout
           )) {
     FML_DLOG(INFO) << "Could not insert image memory barrier.";
     return error;
   } else {
     current_pipeline_stage_ = destination_pipeline_stage;
   }

   // ---------------------------------------------------------------------------
   // Step 6:
   // End recording to the command buffer.
   // ---------------------------------------------------------------------------
   if (!backbuffer->GetUsageCommandBuffer().End()) {
     FML_DLOG(INFO) << "Could not end recording to the command buffer.";
     return error;
   }

   // ---------------------------------------------------------------------------
   // Step 7:
   // Submit the command buffer to the device queue.
   // ---------------------------------------------------------------------------
   std::vector<VkSemaphore> wait_semaphores = {backbuffer->GetUsageSemaphore()};
   std::vector<VkSemaphore> signal_semaphores = {};
   std::vector<VkCommandBuffer> command_buffers = {
       backbuffer->GetUsageCommandBuffer().Handle()};

   if (!device_.QueueSubmit(
           {destination_pipeline_stage},  // wait_dest_pipeline_stages
           wait_semaphores,               // wait_semaphores
           signal_semaphores,             // signal_semaphores
           command_buffers,               // command_buffers
           backbuffer->GetUsageFence()    // fence
           )) {
     FML_DLOG(INFO) << "Could not submit to the device queue.";
     return error;
   }

   // ---------------------------------------------------------------------------
   // Step 8:
   // Tell Skia about the updated image layout.
   // ---------------------------------------------------------------------------
   sk_sp<SkSurface> surface = surfaces_[next_image_index];

   if (surface == nullptr) {
     FML_DLOG(INFO) << "Could not access surface at the image index.";
     return error;
   }

   GrBackendRenderTarget backendRT = SkSurfaces::GetBackendRenderTarget(
       surface.get(), SkSurfaces::BackendHandleAccess::kFlushRead);
   if (!backendRT.isValid()) {
     FML_DLOG(INFO) << "Could not get backend render target.";
     return error;
   }
   GrBackendRenderTargets::SetVkImageLayout(&backendRT,
                                            destination_image_layout);

   current_image_index_ = next_image_index;

   return {AcquireStatus::Success, surface};
 }

 bool VulkanSwapchain::Submit() {
   if (!IsValid()) {
     FML_DLOG(INFO) << "Swapchain was invalid.";
     return false;
   }

   sk_sp<SkSurface> surface = surfaces_[current_image_index_];
   const std::unique_ptr<VulkanImage>& image = images_[current_image_index_];
   auto backbuffer = backbuffers_[current_backbuffer_index_].get();

   // ---------------------------------------------------------------------------
   // Step 0:
   // Make sure Skia has flushed all work for the surface to the gpu.
   // ---------------------------------------------------------------------------
   skgpu::ganesh::FlushAndSubmit(surface);

   // ---------------------------------------------------------------------------
   // Step 1:
   // Start recording to the command buffer.
   // ---------------------------------------------------------------------------
   if (!backbuffer->GetRenderCommandBuffer().Begin()) {
     FML_DLOG(INFO) << "Could not start recording to the command buffer.";
     return false;
   }

   // ---------------------------------------------------------------------------
   // Step 2:
   // Set image layout to present mode.
   // ---------------------------------------------------------------------------
   VkPipelineStageFlagBits destination_pipeline_stage =
       VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT;
   VkImageLayout destination_image_layout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;

   if (!image->InsertImageMemoryBarrier(
           backbuffer->GetRenderCommandBuffer(),  // command buffer
           current_pipeline_stage_,               // src_pipeline_bits
           destination_pipeline_stage,            // dest_pipeline_bits
           VK_ACCESS_MEMORY_READ_BIT,             // dest_access_flags
           destination_image_layout               // dest_layout
           )) {
     FML_DLOG(INFO) << "Could not insert memory barrier.";
     return false;
   } else {
     current_pipeline_stage_ = destination_pipeline_stage;
   }

   // ---------------------------------------------------------------------------
   // Step 3:
   // End recording to the command buffer.
   // ---------------------------------------------------------------------------
   if (!backbuffer->GetRenderCommandBuffer().End()) {
     FML_DLOG(INFO) << "Could not end recording to the command buffer.";
     return false;
   }

   // ---------------------------------------------------------------------------
   // Step 4:
   // Submit the command buffer to the device queue. Tell it to signal the render
   // semaphore.
   // ---------------------------------------------------------------------------
   std::vector<VkSemaphore> wait_semaphores = {};
   std::vector<VkSemaphore> queue_signal_semaphores = {
       backbuffer->GetRenderSemaphore()};
   std::vector<VkCommandBuffer> command_buffers = {
       backbuffer->GetRenderCommandBuffer().Handle()};

   if (!device_.QueueSubmit(
           {/* Empty. No wait semaphores. */},  // wait_dest_pipeline_stages
           wait_semaphores,                     // wait_semaphores
           queue_signal_semaphores,             // signal_semaphores
           command_buffers,                     // command_buffers
           backbuffer->GetRenderFence()         // fence
           )) {
     FML_DLOG(INFO) << "Could not submit to the device queue.";
     return false;
   }

   // ---------------------------------------------------------------------------
   // Step 5:
   // Submit the present operation and wait on the render semaphore.
   // ---------------------------------------------------------------------------
   VkSwapchainKHR swapchain = swapchain_;
   uint32_t present_image_index = static_cast<uint32_t>(current_image_index_);
   const VkPresentInfoKHR present_info = {
       .sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR,
       .pNext = nullptr,
       .waitSemaphoreCount =
           static_cast<uint32_t>(queue_signal_semaphores.size()),
       .pWaitSemaphores = queue_signal_semaphores.data(),
       .swapchainCount = 1,
       .pSwapchains = &swapchain,
       .pImageIndices = &present_image_index,
       .pResults = nullptr,
   };

   if (VK_CALL_LOG_ERROR(vk.QueuePresentKHR(device_.GetQueueHandle(),
                                            &present_info)) != VK_SUCCESS) {
     FML_DLOG(INFO) << "Could not submit the present operation.";
     return false;
   }

   return true;
 }

 }  // namespace vulkan
	// Copyright 2013 The Flutter Authors. All rights reserved.
	// Use of this source code is governed by a BSD-style license that can be
	// found in the LICENSE file.

	#include "vulkan_swapchain.h"

	#include "flutter/vulkan/procs/vulkan_proc_table.h"

	#include "third_party/skia/include/core/SkColorSpace.h"
	#include "third_party/skia/include/core/SkSurface.h"
	#include "third_party/skia/include/gpu/GrBackendSurface.h"
	#include "third_party/skia/include/gpu/GrDirectContext.h"
	#include "third_party/skia/include/gpu/ganesh/SkSurfaceGanesh.h"
	#include "third_party/skia/include/gpu/ganesh/vk/GrVkBackendSurface.h"
	#include "third_party/skia/include/gpu/vk/GrVkTypes.h"

	#include "vulkan_backbuffer.h"
	#include "vulkan_device.h"
	#include "vulkan_image.h"
	#include "vulkan_surface.h"

	namespace vulkan {

	namespace {
	struct FormatInfo {
	VkFormat format_;
	SkColorType color_type_;
	sk_sp<SkColorSpace> color_space_;
	};
	} // namespace

	static std::vector<FormatInfo> DesiredFormatInfos() {
	return {{VK_FORMAT_R8G8B8A8_SRGB, kRGBA_8888_SkColorType,
	SkColorSpace::MakeSRGB()},
	{VK_FORMAT_B8G8R8A8_SRGB, kRGBA_8888_SkColorType,
	SkColorSpace::MakeSRGB()},
	{VK_FORMAT_R16G16B16A16_SFLOAT, kRGBA_F16_SkColorType,
	SkColorSpace::MakeSRGBLinear()},
	{VK_FORMAT_R8G8B8A8_UNORM, kRGBA_8888_SkColorType,
	SkColorSpace::MakeSRGB()},
	{VK_FORMAT_B8G8R8A8_UNORM, kRGBA_8888_SkColorType,
	SkColorSpace::MakeSRGB()}};
	}

	VulkanSwapchain::VulkanSwapchain(const VulkanProcTable& p_vk,
	const VulkanDevice& device,
	const VulkanSurface& surface,
	GrDirectContext* skia_context,
	std::unique_ptr<VulkanSwapchain> old_swapchain,
	uint32_t queue_family_index)
	: vk(p_vk),
	device_(device),
	capabilities_(),
	surface_format_(),
	current_pipeline_stage_(VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT),
	current_backbuffer_index_(0),
	current_image_index_(0),
	valid_(false) {
	if (!device_.IsValid() \|\| !surface.IsValid() \|\| skia_context == nullptr) {
	FML_DLOG(INFO) << "Device or surface is invalid.";
	return;
	}

	if (!device_.GetSurfaceCapabilities(surface, &capabilities_)) {
	FML_DLOG(INFO) << "Could not find surface capabilities.";
	return;
	}

	const auto format_infos = DesiredFormatInfos();
	std::vector<VkFormat> desired_formats(format_infos.size());
	for (size_t i = 0; i < format_infos.size(); ++i) {
	if (skia_context->colorTypeSupportedAsSurface(
	format_infos[i].color_type_)) {
	desired_formats[i] = format_infos[i].format_;
	} else {
	desired_formats[i] = VK_FORMAT_UNDEFINED;
	}
	}

	int format_index =
	device_.ChooseSurfaceFormat(surface, desired_formats, &surface_format_);
	if (format_index < 0) {
	FML_DLOG(INFO) << "Could not choose surface format.";
	return;
	}

	VkPresentModeKHR present_mode = VK_PRESENT_MODE_FIFO_KHR;
	if (!device_.ChoosePresentMode(surface, &present_mode)) {
	FML_DLOG(INFO) << "Could not choose present mode.";
	return;
	}

	// Check if the surface can present.

	VkBool32 supported = VK_FALSE;
	if (VK_CALL_LOG_ERROR(vk.GetPhysicalDeviceSurfaceSupportKHR(
	device_.GetPhysicalDeviceHandle(), // physical device
	queue_family_index, // queue family
	surface.Handle(), // surface to test
	&supported)) != VK_SUCCESS) {
	FML_DLOG(INFO) << "Could not get physical device surface support.";
	return;
	}

	if (supported != VK_TRUE) {
	FML_DLOG(INFO) << "Surface was not supported by the physical device.";
	return;
	}

	// Construct the Swapchain

	VkSwapchainKHR old_swapchain_handle = VK_NULL_HANDLE;

	if (old_swapchain != nullptr && old_swapchain->IsValid()) {
	old_swapchain_handle = old_swapchain->swapchain_;
	// The unique pointer to the swapchain will go out of scope here
	// and its handle collected after the appropriate device wait.
	}

	VkSurfaceKHR surface_handle = surface.Handle();

	VkImageUsageFlags usage_flags = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT \|
	VK_IMAGE_USAGE_TRANSFER_SRC_BIT \|
	VK_IMAGE_USAGE_TRANSFER_DST_BIT;

	const VkSwapchainCreateInfoKHR create_info = {
	.sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR,
	.pNext = nullptr,
	.flags = 0,
	.surface = surface_handle,
	.minImageCount = capabilities_.minImageCount,
	.imageFormat = surface_format_.format,
	.imageColorSpace = surface_format_.colorSpace,
	.imageExtent = capabilities_.currentExtent,
	.imageArrayLayers = 1,
	.imageUsage = usage_flags,
	.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE,
	.queueFamilyIndexCount = 0, // Because of the exclusive sharing mode.
	.pQueueFamilyIndices = nullptr,
	.preTransform = VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR,
	.compositeAlpha = VK_COMPOSITE_ALPHA_INHERIT_BIT_KHR,
	.presentMode = present_mode,
	.clipped = VK_FALSE,
	.oldSwapchain = old_swapchain_handle,
	};

	VkSwapchainKHR swapchain = VK_NULL_HANDLE;

	if (VK_CALL_LOG_ERROR(vk.CreateSwapchainKHR(device_.GetHandle(), &create_info,
	nullptr, &swapchain)) !=
	VK_SUCCESS) {
	FML_DLOG(INFO) << "Could not create the swapchain.";
	return;
	}

	swapchain_ = VulkanHandle<VkSwapchainKHR>{
	swapchain, [this](VkSwapchainKHR swapchain) {
	FML_ALLOW_UNUSED_LOCAL(device_.WaitIdle());
	vk.DestroySwapchainKHR(device_.GetHandle(), swapchain, nullptr);
	}};

	if (!CreateSwapchainImages(
	skia_context, format_infos[format_index].color_type_,
	format_infos[format_index].color_space_, usage_flags)) {
	FML_DLOG(INFO) << "Could not create swapchain images.";
	return;
	}

	valid_ = true;
	}

	VulkanSwapchain::~VulkanSwapchain() = default;

	bool VulkanSwapchain::IsValid() const {
	return valid_;
	}

	std::vector<VkImage> VulkanSwapchain::GetImages() const {
	uint32_t count = 0;
	if (VK_CALL_LOG_ERROR(vk.GetSwapchainImagesKHR(
	device_.GetHandle(), swapchain_, &count, nullptr)) != VK_SUCCESS) {
	return {};
	}

	if (count == 0) {
	return {};
	}

	std::vector<VkImage> images;

	images.resize(count);

	if (VK_CALL_LOG_ERROR(vk.GetSwapchainImagesKHR(
	device_.GetHandle(), swapchain_, &count, images.data())) !=
	VK_SUCCESS) {
	return {};
	}

	return images;
	}

	SkISize VulkanSwapchain::GetSize() const {
	VkExtent2D extents = capabilities_.currentExtent;

	if (extents.width < capabilities_.minImageExtent.width) {
	extents.width = capabilities_.minImageExtent.width;
	} else if (extents.width > capabilities_.maxImageExtent.width) {
	extents.width = capabilities_.maxImageExtent.width;
	}

	if (extents.height < capabilities_.minImageExtent.height) {
	extents.height = capabilities_.minImageExtent.height;
	} else if (extents.height > capabilities_.maxImageExtent.height) {
	extents.height = capabilities_.maxImageExtent.height;
	}

	return SkISize::Make(extents.width, extents.height);
	}

	sk_sp<SkSurface> VulkanSwapchain::CreateSkiaSurface(
	GrDirectContext* gr_context,
	VkImage image,
	VkImageUsageFlags usage_flags,
	const SkISize& size,
	SkColorType color_type,
	sk_sp<SkColorSpace> color_space) const {
	if (gr_context == nullptr) {
	return nullptr;
	}

	if (color_type == kUnknown_SkColorType) {
	// Unexpected Vulkan format.
	return nullptr;
	}

	GrVkImageInfo image_info;
	image_info.fImage = image;
	image_info.fImageTiling = VK_IMAGE_TILING_OPTIMAL;
	image_info.fImageLayout = VK_IMAGE_LAYOUT_UNDEFINED;
	image_info.fFormat = surface_format_.format;
	image_info.fImageUsageFlags = usage_flags;
	image_info.fSampleCount = 1;
	image_info.fLevelCount = 1;

	// TODO(chinmaygarde): Setup the stencil buffer and the sampleCnt.
	auto backend_render_target =
	GrBackendRenderTargets::MakeVk(size.fWidth, size.fHeight, image_info);
	SkSurfaceProps props(0, kUnknown_SkPixelGeometry);

	return SkSurfaces::WrapBackendRenderTarget(
	gr_context, // context
	backend_render_target, // backend render target
	kTopLeft_GrSurfaceOrigin, // origin
	color_type, // color type
	std::move(color_space), // color space
	&props // surface properties
	);
	}

	bool VulkanSwapchain::CreateSwapchainImages(
	GrDirectContext* skia_context,
	SkColorType color_type,
	const sk_sp<SkColorSpace>& color_space,
	VkImageUsageFlags usage_flags) {
	std::vector<VkImage> images = GetImages();

	if (images.empty()) {
	return false;
	}

	const SkISize surface_size = GetSize();

	for (const VkImage& image : images) {
	// Populate the backbuffer.
	auto backbuffer = std::make_unique<VulkanBackbuffer>(
	vk, device_.GetHandle(), device_.GetCommandPool());

	if (!backbuffer->IsValid()) {
	return false;
	}

	backbuffers_.emplace_back(std::move(backbuffer));

	// Populate the image.
	VulkanHandle<VkImage> image_handle = VulkanHandle<VkImage>{
	image, [this](VkImage image) {
	vk.DestroyImage(device_.GetHandle(), image, nullptr);
	}};
	auto vulkan_image = std::make_unique<VulkanImage>(std::move(image_handle));

	if (!vulkan_image->IsValid()) {
	return false;
	}

	images_.emplace_back(std::move(vulkan_image));

	// Populate the surface.
	auto surface = CreateSkiaSurface(skia_context, image, usage_flags,
	surface_size, color_type, color_space);

	if (surface == nullptr) {
	return false;
	}

	surfaces_.emplace_back(std::move(surface));
	}

	FML_DCHECK(backbuffers_.size() == images_.size());
	FML_DCHECK(images_.size() == surfaces_.size());

	return true;
	}

	VulkanBackbuffer* VulkanSwapchain::GetNextBackbuffer() {
	auto available_backbuffers = backbuffers_.size();

	if (available_backbuffers == 0) {
	return nullptr;
	}

	auto next_backbuffer_index =
	(current_backbuffer_index_ + 1) % backbuffers_.size();

	auto& backbuffer = backbuffers_[next_backbuffer_index];

	if (!backbuffer->IsValid()) {
	return nullptr;
	}

	current_backbuffer_index_ = next_backbuffer_index;
	return backbuffer.get();
	}

	VulkanSwapchain::AcquireResult VulkanSwapchain::AcquireSurface() {
	AcquireResult error = {AcquireStatus::ErrorSurfaceLost, nullptr};

	if (!IsValid()) {
	FML_DLOG(INFO) << "Swapchain was invalid.";
	return error;
	}

	// ---------------------------------------------------------------------------
	// Step 0:
	// Acquire the next available backbuffer.
	// ---------------------------------------------------------------------------
	auto backbuffer = GetNextBackbuffer();

	if (backbuffer == nullptr) {
	FML_DLOG(INFO) << "Could not get the next backbuffer.";
	return error;
	}

	// ---------------------------------------------------------------------------
	// Step 1:
	// Wait for use readiness.
	// ---------------------------------------------------------------------------
	if (!backbuffer->WaitFences()) {
	FML_DLOG(INFO) << "Failed waiting on fences.";
	return error;
	}

	// ---------------------------------------------------------------------------
	// Step 2:
	// Put fences in an unsignaled state.
	// ---------------------------------------------------------------------------
	if (!backbuffer->ResetFences()) {
	FML_DLOG(INFO) << "Could not reset fences.";
	return error;
	}

	// ---------------------------------------------------------------------------
	// Step 3:
	// Acquire the next image index.
	// ---------------------------------------------------------------------------
	uint32_t next_image_index = 0;

	VkResult acquire_result = VK_CALL_LOG_ERROR(
	vk.AcquireNextImageKHR(device_.GetHandle(), //
	swapchain_, //
	std::numeric_limits<uint64_t>::max(), //
	backbuffer->GetUsageSemaphore(), //
	VK_NULL_HANDLE, //
	&next_image_index));

	switch (acquire_result) {
	case VK_SUCCESS:
	break;
	case VK_ERROR_OUT_OF_DATE_KHR:
	return {AcquireStatus::ErrorSurfaceOutOfDate, nullptr};
	case VK_ERROR_SURFACE_LOST_KHR:
	return {AcquireStatus::ErrorSurfaceLost, nullptr};
	default:
	FML_LOG(INFO) << "Unexpected result from AcquireNextImageKHR: "
	<< acquire_result;
	return {AcquireStatus::ErrorSurfaceLost, nullptr};
	}

	// Simple sanity checking of image index.
	if (next_image_index >= images_.size()) {
	FML_DLOG(INFO) << "Image index returned was out-of-bounds.";
	return error;
	}

	auto& image = images_[next_image_index];
	if (!image->IsValid()) {
	FML_DLOG(INFO) << "Image at index was invalid.";
	return error;
	}

	// ---------------------------------------------------------------------------
	// Step 4:
	// Start recording to the command buffer.
	// ---------------------------------------------------------------------------
	if (!backbuffer->GetUsageCommandBuffer().Begin()) {
	FML_DLOG(INFO) << "Could not begin recording to the command buffer.";
	return error;
	}

	// ---------------------------------------------------------------------------
	// Step 5:
	// Set image layout to color attachment mode.
	// ---------------------------------------------------------------------------
	VkPipelineStageFlagBits destination_pipeline_stage =
	VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
	VkImageLayout destination_image_layout =
	VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;

	if (!image->InsertImageMemoryBarrier(
	backbuffer->GetUsageCommandBuffer(), // command buffer
	current_pipeline_stage_, // src_pipeline_bits
	destination_pipeline_stage, // dest_pipeline_bits
	VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, // dest_access_flags
	destination_image_layout // dest_layout
	)) {
	FML_DLOG(INFO) << "Could not insert image memory barrier.";
	return error;
	} else {
	current_pipeline_stage_ = destination_pipeline_stage;
	}

	// ---------------------------------------------------------------------------
	// Step 6:
	// End recording to the command buffer.
	// ---------------------------------------------------------------------------
	if (!backbuffer->GetUsageCommandBuffer().End()) {
	FML_DLOG(INFO) << "Could not end recording to the command buffer.";
	return error;
	}

	// ---------------------------------------------------------------------------
	// Step 7:
	// Submit the command buffer to the device queue.
	// ---------------------------------------------------------------------------
	std::vector<VkSemaphore> wait_semaphores = {backbuffer->GetUsageSemaphore()};
	std::vector<VkSemaphore> signal_semaphores = {};
	std::vector<VkCommandBuffer> command_buffers = {
	backbuffer->GetUsageCommandBuffer().Handle()};

	if (!device_.QueueSubmit(
	{destination_pipeline_stage}, // wait_dest_pipeline_stages
	wait_semaphores, // wait_semaphores
	signal_semaphores, // signal_semaphores
	command_buffers, // command_buffers
	backbuffer->GetUsageFence() // fence
	)) {
	FML_DLOG(INFO) << "Could not submit to the device queue.";
	return error;
	}

	// ---------------------------------------------------------------------------
	// Step 8:
	// Tell Skia about the updated image layout.
	// ---------------------------------------------------------------------------
	sk_sp<SkSurface> surface = surfaces_[next_image_index];

	if (surface == nullptr) {
	FML_DLOG(INFO) << "Could not access surface at the image index.";
	return error;
	}

	GrBackendRenderTarget backendRT = SkSurfaces::GetBackendRenderTarget(
	surface.get(), SkSurfaces::BackendHandleAccess::kFlushRead);
	if (!backendRT.isValid()) {
	FML_DLOG(INFO) << "Could not get backend render target.";
	return error;
	}
	GrBackendRenderTargets::SetVkImageLayout(&backendRT,
	destination_image_layout);

	current_image_index_ = next_image_index;

	return {AcquireStatus::Success, surface};
	}

	bool VulkanSwapchain::Submit() {
	if (!IsValid()) {
	FML_DLOG(INFO) << "Swapchain was invalid.";
	return false;
	}

	sk_sp<SkSurface> surface = surfaces_[current_image_index_];
	const std::unique_ptr<VulkanImage>& image = images_[current_image_index_];
	auto backbuffer = backbuffers_[current_backbuffer_index_].get();

	// ---------------------------------------------------------------------------
	// Step 0:
	// Make sure Skia has flushed all work for the surface to the gpu.
	// ---------------------------------------------------------------------------
	skgpu::ganesh::FlushAndSubmit(surface);

	// ---------------------------------------------------------------------------
	// Step 1:
	// Start recording to the command buffer.
	// ---------------------------------------------------------------------------
	if (!backbuffer->GetRenderCommandBuffer().Begin()) {
	FML_DLOG(INFO) << "Could not start recording to the command buffer.";
	return false;
	}

	// ---------------------------------------------------------------------------
	// Step 2:
	// Set image layout to present mode.
	// ---------------------------------------------------------------------------
	VkPipelineStageFlagBits destination_pipeline_stage =
	VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT;
	VkImageLayout destination_image_layout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;

	if (!image->InsertImageMemoryBarrier(
	backbuffer->GetRenderCommandBuffer(), // command buffer
	current_pipeline_stage_, // src_pipeline_bits
	destination_pipeline_stage, // dest_pipeline_bits
	VK_ACCESS_MEMORY_READ_BIT, // dest_access_flags
	destination_image_layout // dest_layout
	)) {
	FML_DLOG(INFO) << "Could not insert memory barrier.";
	return false;
	} else {
	current_pipeline_stage_ = destination_pipeline_stage;
	}

	// ---------------------------------------------------------------------------
	// Step 3:
	// End recording to the command buffer.
	// ---------------------------------------------------------------------------
	if (!backbuffer->GetRenderCommandBuffer().End()) {
	FML_DLOG(INFO) << "Could not end recording to the command buffer.";
	return false;
	}

	// ---------------------------------------------------------------------------
	// Step 4:
	// Submit the command buffer to the device queue. Tell it to signal the render
	// semaphore.
	// ---------------------------------------------------------------------------
	std::vector<VkSemaphore> wait_semaphores = {};
	std::vector<VkSemaphore> queue_signal_semaphores = {
	backbuffer->GetRenderSemaphore()};
	std::vector<VkCommandBuffer> command_buffers = {
	backbuffer->GetRenderCommandBuffer().Handle()};

	if (!device_.QueueSubmit(
	{/* Empty. No wait semaphores. */}, // wait_dest_pipeline_stages
	wait_semaphores, // wait_semaphores
	queue_signal_semaphores, // signal_semaphores
	command_buffers, // command_buffers
	backbuffer->GetRenderFence() // fence
	)) {
	FML_DLOG(INFO) << "Could not submit to the device queue.";
	return false;
	}

	// ---------------------------------------------------------------------------
	// Step 5:
	// Submit the present operation and wait on the render semaphore.
	// ---------------------------------------------------------------------------
	VkSwapchainKHR swapchain = swapchain_;
	uint32_t present_image_index = static_cast<uint32_t>(current_image_index_);
	const VkPresentInfoKHR present_info = {
	.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR,
	.pNext = nullptr,
	.waitSemaphoreCount =
	static_cast<uint32_t>(queue_signal_semaphores.size()),
	.pWaitSemaphores = queue_signal_semaphores.data(),
	.swapchainCount = 1,
	.pSwapchains = &swapchain,
	.pImageIndices = &present_image_index,
	.pResults = nullptr,
	};

	if (VK_CALL_LOG_ERROR(vk.QueuePresentKHR(device_.GetQueueHandle(),
	&present_info)) != VK_SUCCESS) {
	FML_DLOG(INFO) << "Could not submit the present operation.";
	return false;
	}

	return true;
	}

	} // namespace vulkan