From 95740eb78c40d4a4fb8a2b5882e696f729fb6f51 Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Sat, 6 Mar 2021 02:33:39 +0100
Subject: [PATCH] [dxvk] Use clears to impelemnt image copies if possible

Optimizes away a large amount of redundant depth-stencil clear+copy
madness in Final Fantasy XIV for a small performance improvement.
---
 src/dxvk/dxvk_context.cpp | 71 +++++++++++++++++++++++++++++++++++++++
 src/dxvk/dxvk_context.h   |  8 +++++
 src/vulkan/vulkan_util.h  |  9 +++++
 3 files changed, 88 insertions(+)

diff --git a/src/dxvk/dxvk_context.cpp b/src/dxvk/dxvk_context.cpp
index 67d3c826c..72da2a7d3 100644
--- a/src/dxvk/dxvk_context.cpp
+++ b/src/dxvk/dxvk_context.cpp
@@ -766,6 +766,10 @@ namespace dxvk {
           VkOffset3D            srcOffset,
           VkExtent3D            extent) {
     this->spillRenderPass(true);
+
+    if (this->copyImageClear(dstImage, dstSubresource, dstOffset, extent, srcImage, srcSubresource))
+      return;
+
     this->prepareImage(m_execBarriers, dstImage, vk::makeSubresourceRange(dstSubresource));
     this->prepareImage(m_execBarriers, srcImage, vk::makeSubresourceRange(srcSubresource));
 
@@ -3183,6 +3187,73 @@ namespace dxvk {
   }
 
 
+  bool DxvkContext::copyImageClear(
+    const Rc<DxvkImage>&        dstImage,
+          VkImageSubresourceLayers dstSubresource,
+          VkOffset3D            dstOffset,
+          VkExtent3D            dstExtent,
+    const Rc<DxvkImage>&        srcImage,
+          VkImageSubresourceLayers srcSubresource) {
+    // If the source image has a pending deferred clear, we can
+    // implement the copy by clearing the destination image to
+    // the same clear value.
+    const VkImageUsageFlags attachmentUsage
+      = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT
+      | VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
+
+    if (!(dstImage->info().usage & attachmentUsage)
+     || !(srcImage->info().usage & attachmentUsage))
+      return false;
+
+    // Ignore 3D images since those are complicated to handle
+    if (dstImage->info().type == VK_IMAGE_TYPE_3D
+     || srcImage->info().type == VK_IMAGE_TYPE_3D)
+      return false;
+
+    // Find a pending clear that overlaps with the source image
+    const DxvkDeferredClear* clear = nullptr;
+
+    for (const auto& entry : m_deferredClears) {
+      // Entries in the deferred clear array cannot overlap, so
+      // if we find an entry covering all source subresources,
+      // it's the only one in the list that does.
+      if ((entry.imageView->image() == srcImage) && ((srcSubresource.aspectMask & entry.clearAspects) == srcSubresource.aspectMask)
+       && (vk::checkSubresourceRangeSuperset(entry.imageView->subresources(), vk::makeSubresourceRange(srcSubresource)))) {
+        clear = &entry;
+        break;
+      }
+    }
+
+    if (!clear)
+      return false;
+
+    // Create a view for the destination image with the general
+    // properties ofthe source image view used for the clear
+    DxvkImageViewCreateInfo viewInfo = clear->imageView->info();
+    viewInfo.type = dstImage->info().type == VK_IMAGE_TYPE_1D
+      ? VK_IMAGE_VIEW_TYPE_1D_ARRAY
+      : VK_IMAGE_VIEW_TYPE_2D_ARRAY;
+    viewInfo.minLevel = dstSubresource.mipLevel;
+    viewInfo.numLevels = 1;
+    viewInfo.minLayer = dstSubresource.baseArrayLayer;
+    viewInfo.numLayers = dstSubresource.layerCount;
+
+    // That is, if the formats are actually compatible
+    // so that we can safely use the same clear value
+    if (!dstImage->isViewCompatible(viewInfo.format))
+      return false;
+
+    // Ignore mismatched size for now, needs more testing since we'd
+    // need to prepare the image first and then call clearImageViewFb
+    if (dstImage->mipLevelExtent(dstSubresource.mipLevel) != dstExtent)
+      return false;
+
+    auto view = m_device->createImageView(dstImage, viewInfo);
+    this->deferClear(view, clear->clearAspects, clear->clearValue);
+    return true;
+  }
+
+
   void DxvkContext::resolveImageHw(
     const Rc<DxvkImage>&            dstImage,
     const Rc<DxvkImage>&            srcImage,
diff --git a/src/dxvk/dxvk_context.h b/src/dxvk/dxvk_context.h
index 36c37112f..4b5231f36 100644
--- a/src/dxvk/dxvk_context.h
+++ b/src/dxvk/dxvk_context.h
@@ -1067,6 +1067,14 @@ namespace dxvk {
             VkOffset3D            srcOffset,
             VkExtent3D            extent);
     
+    bool copyImageClear(
+      const Rc<DxvkImage>&        dstImage,
+            VkImageSubresourceLayers dstSubresource,
+            VkOffset3D            dstOffset,
+            VkExtent3D            dstExtent,
+      const Rc<DxvkImage>&        srcImage,
+            VkImageSubresourceLayers srcSubresource);
+
     void resolveImageHw(
       const Rc<DxvkImage>&            dstImage,
       const Rc<DxvkImage>&            srcImage,
diff --git a/src/vulkan/vulkan_util.h b/src/vulkan/vulkan_util.h
index e8e318047..a2899d3ef 100644
--- a/src/vulkan/vulkan_util.h
+++ b/src/vulkan/vulkan_util.h
@@ -67,6 +67,15 @@ namespace dxvk::vk {
         && a.baseArrayLayer + a.layerCount > b.baseArrayLayer;
   }
 
+  inline bool checkSubresourceRangeSuperset(
+    const VkImageSubresourceRange&  a,
+    const VkImageSubresourceRange&  b) {
+    return a.baseMipLevel                <= b.baseMipLevel
+        && a.baseMipLevel + a.levelCount >= b.baseMipLevel + b.levelCount
+        && a.baseArrayLayer                <= b.baseArrayLayer
+        && a.baseArrayLayer + a.layerCount >= b.baseArrayLayer + b.layerCount;
+  }
+
 }