Merge pull request #9648 from stenzek/dx11-gl-bbox-caching

Cache bounding box values between register reads in DX11/GL

1 files changed, 80 insertions, 14 deletions

diff --git a/Source/Core/VideoBackends/OGL/OGLBoundingBox.cpp b/Source/Core/VideoBackends/OGL/OGLBoundingBox.cpp
index 9e88767afc..cee200d4ad 100644
--- a/Source/Core/VideoBackends/OGL/OGLBoundingBox.cpp
+++ b/Source/Core/VideoBackends/OGL/OGLBoundingBox.cpp
@@ -2,6 +2,8 @@
 // Licensed under GPLv2+
 // Refer to the license.txt file included.
 
+#include <algorithm>
+#include <array>
 #include <cstring>
 
 #include "Common/GL/GLUtil.h"
@@ -12,7 +14,15 @@
 #include "VideoCommon/DriverDetails.h"
 #include "VideoCommon/VideoConfig.h"
 
+enum : u32
+{
+  NUM_BBOX_VALUES = 4,
+};
+
 static GLuint s_bbox_buffer_id;
+static std::array<s32, NUM_BBOX_VALUES> s_bbox_values;
+static std::array<bool, NUM_BBOX_VALUES> s_bbox_dirty;
+static bool s_bbox_valid = false;
 
 namespace OGL
 {
@@ -21,10 +31,14 @@ void BoundingBox::Init()
   if (!g_ActiveConfig.backend_info.bSupportsBBox)
     return;
 
-  int initial_values[4] = {0, 0, 0, 0};
+  const s32 initial_values[NUM_BBOX_VALUES] = {0, 0, 0, 0};
+  std::memcpy(s_bbox_values.data(), initial_values, sizeof(s_bbox_values));
+  s_bbox_dirty = {};
+  s_bbox_valid = true;
+
   glGenBuffers(1, &s_bbox_buffer_id);
   glBindBuffer(GL_SHADER_STORAGE_BUFFER, s_bbox_buffer_id);
-  glBufferData(GL_SHADER_STORAGE_BUFFER, 4 * sizeof(s32), initial_values, GL_DYNAMIC_DRAW);
+  glBufferData(GL_SHADER_STORAGE_BUFFER, sizeof(initial_values), initial_values, GL_DYNAMIC_DRAW);
   glBindBufferBase(GL_SHADER_STORAGE_BUFFER, 0, s_bbox_buffer_id);
 }
 
@@ -36,22 +50,42 @@ void BoundingBox::Shutdown()
   glDeleteBuffers(1, &s_bbox_buffer_id);
 }
 
-void BoundingBox::Set(int index, int value)
+void BoundingBox::Flush()
 {
-  if (!g_ActiveConfig.backend_info.bSupportsBBox)
+  s_bbox_valid = false;
+
+  if (std::none_of(s_bbox_dirty.begin(), s_bbox_dirty.end(), [](bool dirty) { return dirty; }))
     return;
 
   glBindBuffer(GL_SHADER_STORAGE_BUFFER, s_bbox_buffer_id);
-  glBufferSubData(GL_SHADER_STORAGE_BUFFER, index * sizeof(int), sizeof(int), &value);
+
+  for (u32 start = 0; start < NUM_BBOX_VALUES;)
+  {
+    if (!s_bbox_dirty[start])
+    {
+      start++;
+      continue;
+    }
+
+    u32 end = start + 1;
+    s_bbox_dirty[start] = false;
+    for (; end < NUM_BBOX_VALUES; end++)
+    {
+      if (!s_bbox_dirty[end])
+        break;
+
+      s_bbox_dirty[end] = false;
+    }
+
+    glBufferSubData(GL_SHADER_STORAGE_BUFFER, start * sizeof(s32), (end - start) * sizeof(s32),
+                    &s_bbox_values[start]);
+  }
+
   glBindBuffer(GL_SHADER_STORAGE_BUFFER, 0);
 }
 
-int BoundingBox::Get(int index)
+void BoundingBox::Readback()
 {
-  if (!g_ActiveConfig.backend_info.bSupportsBBox)
-    return 0;
-
-  int data = 0;
   glBindBuffer(GL_SHADER_STORAGE_BUFFER, s_bbox_buffer_id);
   if (!DriverDetails::HasBug(DriverDetails::BUG_SLOW_GETBUFFERSUBDATA) &&
       !static_cast<Renderer*>(g_renderer.get())->IsGLES())
@@ -59,20 +93,52 @@ int BoundingBox::Get(int index)
     // Using glMapBufferRange to read back the contents of the SSBO is extremely slow
     // on nVidia drivers. This is more noticeable at higher internal resolutions.
     // Using glGetBufferSubData instead does not seem to exhibit this slowdown.
-    glGetBufferSubData(GL_SHADER_STORAGE_BUFFER, index * sizeof(int), sizeof(int), &data);
+    std::array<s32, NUM_BBOX_VALUES> gpu_values;
+    glGetBufferSubData(GL_SHADER_STORAGE_BUFFER, 0, sizeof(s32) * NUM_BBOX_VALUES,
+                       gpu_values.data());
+    for (u32 i = 0; i < NUM_BBOX_VALUES; i++)
+    {
+      if (!s_bbox_dirty[i])
+        s_bbox_values[i] = gpu_values[i];
+    }
   }
   else
   {
     // Using glMapBufferRange is faster on AMD cards by a measurable margin.
-    void* ptr = glMapBufferRange(GL_SHADER_STORAGE_BUFFER, index * sizeof(int), sizeof(int),
+    void* ptr = glMapBufferRange(GL_SHADER_STORAGE_BUFFER, 0, sizeof(s32) * NUM_BBOX_VALUES,
                                  GL_MAP_READ_BIT);
     if (ptr)
     {
-      memcpy(&data, ptr, sizeof(int));
+      for (u32 i = 0; i < NUM_BBOX_VALUES; i++)
+      {
+        if (!s_bbox_dirty[i])
+        {
+          std::memcpy(&s_bbox_values[i], reinterpret_cast<const u8*>(ptr) + sizeof(s32) * i,
+                      sizeof(s32));
+        }
+      }
+
       glUnmapBuffer(GL_SHADER_STORAGE_BUFFER);
     }
   }
   glBindBuffer(GL_SHADER_STORAGE_BUFFER, 0);
-  return data;
+  s_bbox_valid = true;
+}
+
+void BoundingBox::Set(int index, int value)
+{
+  if (s_bbox_valid && s_bbox_values[index] == value)
+    return;
+
+  s_bbox_values[index] = value;
+  s_bbox_dirty[index] = true;
+}
+
+int BoundingBox::Get(int index)
+{
+  if (!s_bbox_valid)
+    Readback();
+
+  return s_bbox_values[index];
 }
 };  // namespace OGL