Render to optimally tiled images and blit to linear before scanout

triangle.cc

@@ -7,8 +7,8 @@
 #include <string>
 #include <vector>
 
-const uint32_t kWidth = 128;
-const uint32_t kHeight = 128;
+const uint32_t kWidth = 256;
+const uint32_t kHeight = 256;
 const VkFormat kVulkanFormat = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
 const std::string kVertexShaderPath = "triangle.vert.spv";
 const std::string kFragmentShaderPath = "triangle.frag.spv";
@@ -135,6 +135,21 @@ VkCommandPool CreateVkCommandPool(VkDevice device, uint32_t device_queue_family_
   return command_pool;
 }
 
+VkCommandBuffer CreateVkCommandBuffer(VkDevice device, VkCommandPool command_pool) {
+  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;
+
+  VkCommandBuffer command_buffer;
+  VkResult result = vkAllocateCommandBuffers( device, &allocate_info, &command_buffer);
+  if (result != VK_SUCCESS) {
+    ERROR("Unable to create VkCommandBuffer: " << result);
+  }
+  return command_buffer;
+}
+
 VkRenderPass CreateVkRenderPass(VkDevice device) {
   VkAttachmentDescription attachment_description = {};
   attachment_description.format = kVulkanFormat;
@@ -144,7 +159,7 @@ VkRenderPass CreateVkRenderPass(VkDevice device) {
   attachment_description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
   attachment_description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
   attachment_description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
-  attachment_description.finalLayout = VK_IMAGE_LAYOUT_GENERAL;
+  attachment_description.finalLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
 
   VkAttachmentReference attachment_reference = {};
   attachment_reference.attachment = 0;
@@ -319,7 +334,7 @@ VkPipeline CreateVkPipeline(VkDevice device, VkRenderPass render_pass) {
   return pipeline;
 }
 
-VkImage CreateVkImage(VkDevice device) {
+VkImage CreateVkImage(VkDevice device, VkImageTiling image_tiling, VkImageUsageFlags usage) {
   VkImageCreateInfo create_info = {};
   create_info.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
   create_info.pNext = nullptr;
@@ -334,9 +349,8 @@ VkImage CreateVkImage(VkDevice device) {
   create_info.arrayLayers = 1;
   create_info.samples = VK_SAMPLE_COUNT_1_BIT;
   create_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
-  // create_info.tiling = VK_IMAGE_TILING_OPTIMAL;
-  create_info.tiling = VK_IMAGE_TILING_LINEAR;
-  create_info.usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
+  create_info.tiling = image_tiling;
+  create_info.usage = usage;
   create_info.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
 
   VkImage image;
@@ -347,13 +361,24 @@ VkImage CreateVkImage(VkDevice device) {
   return image;
 }
 
+inline VkImage CreateRenderVkImage(VkDevice device) {
+  return CreateVkImage(device, VK_IMAGE_TILING_OPTIMAL,
+      VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT);
+}
+
+inline VkImage CreateScanoutVkImage(VkDevice device) {
+  return CreateVkImage(device, VK_IMAGE_TILING_LINEAR,
+      VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT);
+}
+
 uint32_t FindMemoryType(uint32_t valid_image_memory_types,
-    VkPhysicalDeviceMemoryProperties device_memory_properties) {
+    VkPhysicalDeviceMemoryProperties device_memory_properties,
+    VkMemoryPropertyFlags memory_property_flags) {
   for (uint32_t i = 0; i < device_memory_properties.memoryTypeCount; i++) {
     // We don't care about performance, so just choose the first mappable memory type.
     if ((valid_image_memory_types % (1 << i)) &&
-        (device_memory_properties.memoryTypes[i].propertyFlags &
-        VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT)) {
+        ((device_memory_properties.memoryTypes[i].propertyFlags & memory_property_flags) ==
+        memory_property_flags)) {
       return i;
     }
   }
@@ -361,7 +386,7 @@ uint32_t FindMemoryType(uint32_t valid_image_memory_types,
 }
 
 VkDeviceMemory AllocateAndBindMemory(VkPhysicalDevice physical_device, VkDevice device,
-    VkImage image) {
+    VkImage image, VkMemoryPropertyFlags memory_property_flags) {
   VkMemoryRequirements image_memory_requirements;
   vkGetImageMemoryRequirements(device, image, &image_memory_requirements);
 
@@ -372,7 +397,8 @@ VkDeviceMemory AllocateAndBindMemory(VkPhysicalDevice physical_device, VkDevice
   memory_allocate_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
   memory_allocate_info.allocationSize = image_memory_requirements.size;
   memory_allocate_info.memoryTypeIndex = FindMemoryType(
-      image_memory_requirements.memoryTypeBits, device_memory_properties);
+      image_memory_requirements.memoryTypeBits, device_memory_properties,
+      memory_property_flags);
 
   VkDeviceMemory image_memory;
   VkResult result = vkAllocateMemory(device, &memory_allocate_info, nullptr, &image_memory);
@@ -387,6 +413,16 @@ VkDeviceMemory AllocateAndBindMemory(VkPhysicalDevice physical_device, VkDevice
   return image_memory;
 }
 
+inline VkDeviceMemory AllocateAndBindRenderMemory(VkPhysicalDevice physical_device,
+    VkDevice device, VkImage image) {
+  return AllocateAndBindMemory(physical_device, device, image, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT);
+}
+
+inline VkDeviceMemory AllocateAndBindScanoutMemory(VkPhysicalDevice physical_device,
+    VkDevice device, VkImage image) {
+  return AllocateAndBindMemory(physical_device, device, image, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT);
+}
+
 VkImageView CreateVkImageView(VkDevice device, VkImage image) {
   VkImageViewCreateInfo create_info = {};
   create_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
@@ -436,23 +472,7 @@ VkFramebuffer CreateVkFramebuffer(VkDevice device, VkRenderPass render_pass,
   return framebuffer;
 }
 
-VkCommandBuffer CreateVkCommandBuffer(VkDevice device, VkCommandPool command_pool) {
-  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;
-
-  VkCommandBuffer command_buffer;
-  VkResult result = vkAllocateCommandBuffers( device, &allocate_info, &command_buffer);
-  if (result != VK_SUCCESS) {
-    ERROR("Unable to create VkCommandBuffer: " << result);
-  }
-  return command_buffer;
-}
-
-void Draw(VkDevice device, VkQueue queue, VkRenderPass render_pass, VkPipeline pipeline,
-    VkFramebuffer framebuffer, VkCommandBuffer command_buffer) {
+void BeginCommandBuffer(VkCommandBuffer command_buffer) {
   VkCommandBufferBeginInfo command_buffer_begin_info = {};
   command_buffer_begin_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
   command_buffer_begin_info.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
@@ -460,7 +480,37 @@ void Draw(VkDevice device, VkQueue queue, VkRenderPass render_pass, VkPipeline p
   if (result != VK_SUCCESS) {
     ERROR("Unable to begin command buffer recording: " << result);
   }
+}
+
+void EndCommandBufferAndSubmit(VkCommandBuffer command_buffer, VkQueue queue) {
+  VkResult result = vkEndCommandBuffer(command_buffer);
+  if (result != VK_SUCCESS) {
+    ERROR("Unable to end command buffer recording: " << result);
+  }
 
+  VkSubmitInfo submit_info = {};
+  submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
+  submit_info.commandBufferCount = 1;
+  submit_info.pCommandBuffers = &command_buffer;
+  result = vkQueueSubmit(queue, 1, &submit_info, VK_NULL_HANDLE);
+  if (result != VK_SUCCESS) {
+    ERROR("Error in submitting command buffer to queue: " << result);
+  }
+}
+
+void WaitForIdle(VkQueue queue) {
+  // In a real application, we should use real synchronization primitives to figure out when the
+  // command buffers have been executed. Waiting on an idle queue is simple, but it can cause
+  // deadlock if we continue to submit to the queue while we wait for idle.
+  VkResult result = vkQueueWaitIdle(queue);
+  if (result != VK_SUCCESS) {
+    ERROR("Error in waiting for graphics queue to reach idle state: " << result);
+  }
+}
+
+void Draw(VkDevice device, VkQueue queue, VkRenderPass render_pass, VkPipeline pipeline,
+    VkFramebuffer framebuffer, VkCommandBuffer command_buffer) {
+  BeginCommandBuffer(command_buffer);
   VkRenderPassBeginInfo render_pass_begin_info = {};
   render_pass_begin_info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
   render_pass_begin_info.renderPass = render_pass;
@@ -481,25 +531,58 @@ void Draw(VkDevice device, VkQueue queue, VkRenderPass render_pass, VkPipeline p
   vkCmdBindPipeline(command_buffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
   vkCmdDraw(command_buffer, 3, 1, 0, 0);
   vkCmdEndRenderPass(command_buffer);
+}
 
-  result = vkEndCommandBuffer(command_buffer);
-  if (result != VK_SUCCESS) {
-    ERROR("Unable to end command buffer recording: " << result);
-  }
-
-  VkSubmitInfo submit_info = {};
-  submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
-  submit_info.commandBufferCount = 1;
-  submit_info.pCommandBuffers = &command_buffer;
-  result = vkQueueSubmit(queue, 1, &submit_info, VK_NULL_HANDLE);
-  if (result != VK_SUCCESS) {
-    ERROR("Error in submitting command buffer to queue: " << result);
-  }
-
-  result = vkQueueWaitIdle(queue);
-  if (result != VK_SUCCESS) {
-    ERROR("Error in waiting for graphics queue to reach idle state: " << result);
-  }
+void BlitToScanoutImage(VkQueue queue, VkCommandBuffer command_buffer, VkImage source_image,
+    VkImage dest_image) {
+  // We need to make sure the writes of the render pass have executed before actually scanning out
+  // the rendered image. Also, we need to transition the layout of the scanout image to
+  // VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL.
+  VkImageMemoryBarrier image_memory_barrier = {};
+  image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+  image_memory_barrier.pNext = nullptr;
+  image_memory_barrier.srcAccessMask = 0;
+  image_memory_barrier.dstAccessMask = 0;
+  image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
+  image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
+  image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+  image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+  image_memory_barrier.image = dest_image;
+  VkImageSubresourceRange subresource_range = {};
+  subresource_range.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+  subresource_range.baseMipLevel = 0;
+  subresource_range.levelCount = 1;
+  subresource_range.baseArrayLayer = 0;
+  subresource_range.layerCount = 1;
+  image_memory_barrier.subresourceRange = subresource_range;
+  vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
+      VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+
+  VkOffset3D blit_start = {};
+  blit_start.x = 0;
+  blit_start.y = 0;
+  blit_start.z = 0;
+  VkOffset3D blit_end = {};
+  blit_end.x = kWidth;
+  blit_end.y = kHeight;
+  blit_end.z = 1;
+  VkImageSubresourceLayers subresource_layers = {};
+  subresource_layers.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+  subresource_layers.mipLevel = 0;
+  subresource_layers.baseArrayLayer = 0;
+  subresource_layers.layerCount = 1;
+  VkImageBlit image_blit = {};
+  image_blit.srcSubresource = subresource_layers;
+  image_blit.srcOffsets[0] = blit_start;
+  image_blit.srcOffsets[1] = blit_end;
+  image_blit.dstSubresource = subresource_layers;
+  image_blit.dstOffsets[0] = blit_start;
+  image_blit.dstOffsets[1] = blit_end;
+  // TODO(brkho): Support multi-planar formats.
+  vkCmdBlitImage(command_buffer, source_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dest_image,
+      VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &image_blit, VK_FILTER_NEAREST);
+
+  EndCommandBufferAndSubmit(command_buffer, queue);
 }
 
 uint8_t* MapVkDeviceMemory(VkDevice device, VkDeviceMemory image_memory) {
@@ -533,17 +616,27 @@ int main() {
   VkDevice device = CreateVkDevice(physical_device, device_queue_family_index);
   VkQueue queue = GetVkQueue(device, device_queue_family_index);
   VkCommandPool command_pool = CreateVkCommandPool(device, device_queue_family_index);
+  VkCommandBuffer command_buffer = CreateVkCommandBuffer(device, command_pool);
+
   VkRenderPass render_pass = CreateVkRenderPass(device);
   VkPipeline pipeline = CreateVkPipeline(device, render_pass);
-  VkImage image = CreateVkImage(device);
-  VkDeviceMemory image_memory = AllocateAndBindMemory(physical_device, device, image);
-  VkImageView image_view = CreateVkImageView(device, image);
-  VkFramebuffer framebuffer = CreateVkFramebuffer(device, render_pass, image_view);
-  VkCommandBuffer command_buffer = CreateVkCommandBuffer(device, command_pool);
+
+  VkImage render_image = CreateRenderVkImage(device);
+  AllocateAndBindRenderMemory(physical_device, device, render_image);
+  VkImageView render_image_view = CreateVkImageView(device, render_image);
+  VkFramebuffer framebuffer = CreateVkFramebuffer(device, render_pass, render_image_view);
+
+  VkImage scanout_image = CreateScanoutVkImage(device);
+  VkDeviceMemory scanout_image_memory =
+      AllocateAndBindScanoutMemory(physical_device, device, scanout_image);
 
   Draw(device, queue, render_pass, pipeline, framebuffer, command_buffer);
+  // Since the render target is created with VK_IMAGE_TILING_OPTIMAL, we need to copy it to a linear
+  // format before scanning out.
+  BlitToScanoutImage(queue, command_buffer, render_image, scanout_image);
+  WaitForIdle(queue);
 
-  uint8_t* pixels = MapVkDeviceMemory(device, image_memory);
+  uint8_t* pixels = MapVkDeviceMemory(device, scanout_image_memory);
   WriteImage(pixels);
 
   return EXIT_SUCCESS;