From babbdadc36f9dd937c7d4593d134c851f690e623 Mon Sep 17 00:00:00 2001
From: baldurk <baldurk@baldurk.org>
Date: Mon, 17 Aug 2020 12:43:36 +0100
Subject: [PATCH] Cache a single value inline in ImageSubresourceMap

* Having a single subresource (range) is a common case, so avoid allocating
  storage in an array for that, only switch to the array when we have more than
  one range to store.
---
 renderdoc/driver/vulkan/vk_image_states.cpp | 12 +++++++-----
 renderdoc/driver/vulkan/vk_resources.h      | 20 +++++++++++---------
 2 files changed, 18 insertions(+), 14 deletions(-)

diff --git a/renderdoc/driver/vulkan/vk_image_states.cpp b/renderdoc/driver/vulkan/vk_image_states.cpp
index 93397a1ad..2dbfb8a24 100644
--- a/renderdoc/driver/vulkan/vk_image_states.cpp
+++ b/renderdoc/driver/vulkan/vk_image_states.cpp
@@ -169,13 +169,15 @@ void ImageSubresourceMap::Split(bool splitAspects, bool splitLevels, bool splitL
   uint32_t newSplitSliceCount = splitDepth ? GetImageInfo().extent.depth : oldSplitSliceCount;
 
   uint32_t oldSize = (uint32_t)m_values.size();
-  RDCASSERT(oldSize > 0);
 
   uint32_t newSize =
       newSplitAspectCount * newSplitLevelCount * newSplitLayerCount * newSplitSliceCount;
-  RDCASSERT(newSize > oldSize);
+  RDCASSERT(newSize > RDCMAX(oldSize, 1U));
 
   m_values.resize(newSize);
+  // if m_values was empty before, copy the first value from our inline storage
+  if(oldSize == 0)
+    m_values[0] = m_value;
 
   uint32_t newAspectIndex = newSplitAspectCount - 1;
   uint32_t oldAspectIndex = AreAspectsSplit() ? newAspectIndex : 0;
@@ -322,7 +324,7 @@ void ImageSubresourceMap::Unsplit(bool unsplitAspects, bool unsplitLevels, bool
 
 void ImageSubresourceMap::Unsplit()
 {
-  if(m_values.size() == 1)
+  if(m_values.size() <= 1)
     return;
 
   uint32_t aspectCount = AreAspectsSplit() ? m_aspectCount : 1;
@@ -449,7 +451,7 @@ size_t ImageSubresourceMap::SubresourceIndex(uint32_t aspectIndex, uint32_t leve
 
 void ImageSubresourceMap::ToArray(rdcarray<ImageSubresourceStateForRange> &arr)
 {
-  arr.reserve(arr.size() + m_values.size());
+  arr.reserve(arr.size() + size());
   for(auto src = begin(); src != end(); ++src)
   {
     arr.push_back(*src);
@@ -464,7 +466,7 @@ void ImageSubresourceMap::FromArray(const rdcarray<ImageSubresourceStateForRange
     return;
   }
   Split(arr.front().range);
-  if(m_values.size() != arr.size())
+  if(size() != arr.size())
   {
     RDCERR("Incorrect number of values for ImageSubresourceMap");
     return;
diff --git a/renderdoc/driver/vulkan/vk_resources.h b/renderdoc/driver/vulkan/vk_resources.h
index 97eba44ca..b3f7c2b91 100644
--- a/renderdoc/driver/vulkan/vk_resources.h
+++ b/renderdoc/driver/vulkan/vk_resources.h
@@ -1397,6 +1397,10 @@ class ImageSubresourceMap
   // `m_values` and the `*Split` flags in `m_flags`.
   rdcarray<ImageSubresourceState> m_values;
 
+  // commonly there will only be one value, in that case we inline it here. This is only valid if
+  // m_values is empty.
+  ImageSubresourceState m_value;
+
   // The bit count of `m_aspectMask`
   uint16_t m_aspectCount = 0;
 
@@ -1443,8 +1447,8 @@ public:
     for(auto it = ImageAspectFlagIter::begin(GetImageInfo().Aspects());
         it != ImageAspectFlagIter::end(); ++it)
       ++m_aspectCount;
-    m_values.push_back(
-        ImageSubresourceState(VK_QUEUE_FAMILY_IGNORED, UNKNOWN_PREV_IMG_LAYOUT, refType));
+
+    m_value = ImageSubresourceState(VK_QUEUE_FAMILY_IGNORED, UNKNOWN_PREV_IMG_LAYOUT, refType);
   }
 
   void ToArray(rdcarray<ImageSubresourceStateForRange> &arr);
@@ -1456,11 +1460,15 @@ public:
   inline ImageSubresourceState &SubresourceIndexValue(uint32_t aspectIndex, uint32_t level,
                                                       uint32_t layer, uint32_t slice)
   {
+    if(m_values.empty())
+      return m_value;
     return m_values[SubresourceIndex(aspectIndex, level, layer, slice)];
   }
   inline const ImageSubresourceState &SubresourceIndexValue(uint32_t aspectIndex, uint32_t level,
                                                             uint32_t layer, uint32_t slice) const
   {
+    if(m_values.empty())
+      return m_value;
     return m_values[SubresourceIndex(aspectIndex, level, layer, slice)];
   }
   inline ImageSubresourceState &SubresourceAspectValue(VkImageAspectFlagBits aspect, uint32_t level,
@@ -1493,12 +1501,6 @@ public:
           range.baseDepthSlice != 0u || range.sliceCount < GetImageInfo().extent.depth);
   }
   void Unsplit();
-  inline void Clear()
-  {
-    m_values.clear();
-    m_values.resize(1);
-    m_flags = 0;
-  }
   FrameRefType Merge(const ImageSubresourceMap &other, FrameRefCompFunc compose);
 
   template <typename Map, typename Pair>
@@ -1588,7 +1590,7 @@ public:
   inline SubresourceRangeConstIter begin() const { return RangeBegin(GetImageInfo().FullRange()); }
   inline SubresourceRangeIter end() { return SubresourceRangeIter(); }
   inline SubresourceRangeConstIter end() const { return SubresourceRangeConstIter(); }
-  inline size_t size() const { return m_values.size(); }
+  inline size_t size() const { return RDCMAX(m_values.size(), (size_t)1); }
 };
 
 template <typename Barrier>