Merge pull request #7915 from stenzek/bbox-scaled-update

VideoBackends: Scale bounding box rectangle in the pixel shader

5 files changed, 59 insertions, 145 deletions

diff --git a/Source/Core/VideoBackends/D3D/Render.cpp b/Source/Core/VideoBackends/D3D/Render.cpp
index a172425e3f..82f3cba743 100644
--- a/Source/Core/VideoBackends/D3D/Render.cpp
+++ b/Source/Core/VideoBackends/D3D/Render.cpp
@@ -324,40 +324,11 @@ void Renderer::UnbindTexture(const AbstractTexture* texture)
 
 u16 Renderer::BBoxRead(int index)
 {
-  // Here we get the min/max value of the truncated position of the upscaled framebuffer.
-  // So we have to correct them to the unscaled EFB sizes.
-  int value = BBox::Get(index);
-
-  if (index < 2)
-  {
-    // left/right
-    value = value * EFB_WIDTH / m_target_width;
-  }
-  else
-  {
-    // up/down
-    value = value * EFB_HEIGHT / m_target_height;
-  }
-  if (index & 1)
-    value++;  // fix max values to describe the outer border
-
-  return value;
+  return static_cast<u16>(BBox::Get(index));
 }
 
-void Renderer::BBoxWrite(int index, u16 _value)
+void Renderer::BBoxWrite(int index, u16 value)
 {
-  int value = _value;  // u16 isn't enough to multiply by the efb width
-  if (index & 1)
-    value--;
-  if (index < 2)
-  {
-    value = value * m_target_width / EFB_WIDTH;
-  }
-  else
-  {
-    value = value * m_target_height / EFB_HEIGHT;
-  }
-
   BBox::Set(index, value);
 }
 
diff --git a/Source/Core/VideoBackends/OGL/Render.cpp b/Source/Core/VideoBackends/OGL/Render.cpp
index d0f3e74c5a..fca1de7e7b 100644
--- a/Source/Core/VideoBackends/OGL/Render.cpp
+++ b/Source/Core/VideoBackends/OGL/Render.cpp
@@ -837,49 +837,30 @@ void Renderer::SetScissorRect(const MathUtil::Rectangle<int>& rc)
 
 u16 Renderer::BBoxRead(int index)
 {
-  int swapped_index = index;
+  // swap 2 and 3 for top/bottom
   if (index >= 2)
-    swapped_index ^= 1;  // swap 2 and 3 for top/bottom
+    index ^= 1;
 
-  // Here we get the min/max value of the truncated position of the upscaled and swapped
-  // framebuffer.
-  // So we have to correct them to the unscaled EFB sizes.
-  int value = BoundingBox::Get(swapped_index);
-
-  if (index < 2)
-  {
-    // left/right
-    value = value * EFB_WIDTH / m_target_width;
-  }
-  else
+  int value = BoundingBox::Get(index);
+  if (index >= 2)
   {
     // up/down -- we have to swap up and down
-    value = value * EFB_HEIGHT / m_target_height;
-    value = EFB_HEIGHT - value - 1;
+    value = EFB_HEIGHT - value;
   }
-  if (index & 1)
-    value++;  // fix max values to describe the outer border
 
-  return value;
+  return static_cast<u16>(value);
 }
 
-void Renderer::BBoxWrite(int index, u16 _value)
+void Renderer::BBoxWrite(int index, u16 value)
 {
-  int value = _value;  // u16 isn't enough to multiply by the efb width
-  if (index & 1)
-    value--;
-  if (index < 2)
-  {
-    value = value * m_target_width / EFB_WIDTH;
-  }
-  else
+  s32 swapped_value = value;
+  if (index >= 2)
   {
     index ^= 1;  // swap 2 and 3 for top/bottom
-    value = EFB_HEIGHT - value - 1;
-    value = value * m_target_height / EFB_HEIGHT;
+    swapped_value = EFB_HEIGHT - swapped_value;
   }
 
-  BoundingBox::Set(index, value);
+  BoundingBox::Set(index, swapped_value);
 }
 
 void Renderer::SetViewport(float x, float y, float width, float height, float near_depth,
diff --git a/Source/Core/VideoBackends/Vulkan/Renderer.cpp b/Source/Core/VideoBackends/Vulkan/Renderer.cpp
index ae3c7c5807..8b560be15d 100644
--- a/Source/Core/VideoBackends/Vulkan/Renderer.cpp
+++ b/Source/Core/VideoBackends/Vulkan/Renderer.cpp
@@ -131,49 +131,12 @@ void Renderer::SetPipeline(const AbstractPipeline* pipeline)
 
 u16 Renderer::BBoxRead(int index)
 {
-  s32 value = m_bounding_box->Get(static_cast<size_t>(index));
-
-  // Here we get the min/max value of the truncated position of the upscaled framebuffer.
-  // So we have to correct them to the unscaled EFB sizes.
-  if (index < 2)
-  {
-    // left/right
-    value = value * EFB_WIDTH / m_target_width;
-  }
-  else
-  {
-    // up/down
-    value = value * EFB_HEIGHT / m_target_height;
-  }
-
-  // fix max values to describe the outer border
-  if (index & 1)
-    value++;
-
-  return static_cast<u16>(value);
+  return static_cast<u16>(m_bounding_box->Get(index));
 }
 
 void Renderer::BBoxWrite(int index, u16 value)
 {
-  s32 scaled_value = static_cast<s32>(value);
-
-  // fix max values to describe the outer border
-  if (index & 1)
-    scaled_value--;
-
-  // scale to internal resolution
-  if (index < 2)
-  {
-    // left/right
-    scaled_value = scaled_value * m_target_width / EFB_WIDTH;
-  }
-  else
-  {
-    // up/down
-    scaled_value = scaled_value * m_target_height / EFB_HEIGHT;
-  }
-
-  m_bounding_box->Set(static_cast<size_t>(index), scaled_value);
+  m_bounding_box->Set(index, value);
 }
 
 void Renderer::BBoxFlush()
diff --git a/Source/Core/VideoCommon/PixelShaderGen.cpp b/Source/Core/VideoCommon/PixelShaderGen.cpp
index 9936bcb85c..4d37418725 100644
--- a/Source/Core/VideoCommon/PixelShaderGen.cpp
+++ b/Source/Core/VideoCommon/PixelShaderGen.cpp
@@ -444,16 +444,46 @@ void WritePixelShaderCommonHeader(ShaderCode& out, APIType ApiType, u32 num_texg
 
   if (bounding_box)
   {
-    if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
-    {
-      out.Write("SSBO_BINDING(0) buffer BBox {\n"
-                "\tint bbox_left, bbox_right, bbox_top, bbox_bottom;\n"
-                "};\n");
-    }
-    else
-    {
-      out.Write("globallycoherent RWBuffer<int> bbox_data : register(u2);\n");
-    }
+    out.Write(R"(
+#ifdef API_D3D
+globallycoherent RWBuffer<int> bbox_data : register(u2);
+#define atomicMin InterlockedMin
+#define atomicMax InterlockedMax
+#define bbox_left bbox_data[0]
+#define bbox_right bbox_data[1]
+#define bbox_top bbox_data[2]
+#define bbox_bottom bbox_data[3]
+#else
+SSBO_BINDING(0) buffer BBox {
+  int bbox_left, bbox_right, bbox_top, bbox_bottom;
+};
+#endif
+
+void UpdateBoundingBox(float2 rawpos) {
+  // The pixel center in the GameCube GPU is 7/12, not 0.5 (see VertexShaderGen.cpp)
+  // Adjust for this by unapplying the offset we added in the vertex shader.
+  const float PIXEL_CENTER_OFFSET = 7.0 / 12.0 - 0.5;
+  float2 offset = float2(PIXEL_CENTER_OFFSET, -PIXEL_CENTER_OFFSET);
+
+#ifdef API_OPENGL
+  // OpenGL lower-left origin means that Y goes in the opposite direction.
+  offset.y = -offset.y;
+#endif
+
+  // The bounding box register is exclusive of the right coordinate, hence the +1.
+  int2 pos = iround(rawpos * cefbscale + offset);
+  int2 pos_offset = pos + int2(1, 1);
+
+  if (bbox_left > pos.x)
+    atomicMin(bbox_left, pos.x);
+  if (bbox_right < pos_offset.x)
+    atomicMax(bbox_right, pos_offset.x);
+  if (bbox_top > pos.y)
+    atomicMin(bbox_top, pos.y);
+  if (bbox_bottom < pos_offset.y)
+    atomicMax(bbox_bottom, pos_offset.y);
+}
+)");
   }
 }
 
@@ -859,23 +889,7 @@ ShaderCode GeneratePixelShaderCode(APIType ApiType, const ShaderHostConfig& host
     WriteBlend(out, uid_data);
 
   if (uid_data->bounding_box)
-  {
-    if (ApiType == APIType::D3D)
-    {
-      out.Write(
-          "\tif(bbox_data[0] > int(rawpos.x)) InterlockedMin(bbox_data[0], int(rawpos.x));\n"
-          "\tif(bbox_data[1] < int(rawpos.x)) InterlockedMax(bbox_data[1], int(rawpos.x));\n"
-          "\tif(bbox_data[2] > int(rawpos.y)) InterlockedMin(bbox_data[2], int(rawpos.y));\n"
-          "\tif(bbox_data[3] < int(rawpos.y)) InterlockedMax(bbox_data[3], int(rawpos.y));\n");
-    }
-    else
-    {
-      out.Write("\tif(bbox_left > int(rawpos.x)) atomicMin(bbox_left, int(rawpos.x));\n"
-                "\tif(bbox_right < int(rawpos.x)) atomicMax(bbox_right, int(rawpos.x));\n"
-                "\tif(bbox_top > int(rawpos.y)) atomicMin(bbox_top, int(rawpos.y));\n"
-                "\tif(bbox_bottom < int(rawpos.y)) atomicMax(bbox_bottom, int(rawpos.y));\n");
-    }
-  }
+    out.Write("\tUpdateBoundingBox(rawpos.xy);\n");
 
   out.Write("}\n");
 
diff --git a/Source/Core/VideoCommon/UberShaderPixel.cpp b/Source/Core/VideoCommon/UberShaderPixel.cpp
index 1b9695f5ce..097b6456bd 100644
--- a/Source/Core/VideoCommon/UberShaderPixel.cpp
+++ b/Source/Core/VideoCommon/UberShaderPixel.cpp
@@ -1250,24 +1250,9 @@ ShaderCode GenPixelShader(APIType ApiType, const ShaderHostConfig& host_config,
 
   if (bounding_box)
   {
-    out.Write("  if (bpmem_bounding_box) {\n");
-    if (ApiType == APIType::D3D)
-    {
-      out.Write(
-          "    if(bbox_data[0] > int(rawpos.x)) InterlockedMin(bbox_data[0], int(rawpos.x));\n"
-          "    if(bbox_data[1] < int(rawpos.x)) InterlockedMax(bbox_data[1], int(rawpos.x));\n"
-          "    if(bbox_data[2] > int(rawpos.y)) InterlockedMin(bbox_data[2], int(rawpos.y));\n"
-          "    if(bbox_data[3] < int(rawpos.y)) InterlockedMax(bbox_data[3], int(rawpos.y));\n");
-    }
-    else
-    {
-      out.Write("\tif(bbox_left > int(rawpos.x)) atomicMin(bbox_left, int(rawpos.x));\n"
-                "\tif(bbox_right < int(rawpos.x)) atomicMax(bbox_right, int(rawpos.x));\n"
-                "\tif(bbox_top > int(rawpos.y)) atomicMin(bbox_top, int(rawpos.y));\n"
-                "\tif(bbox_bottom < int(rawpos.y)) atomicMax(bbox_bottom, int(rawpos.y));\n");
-    }
-
-    out.Write("  }\n");
+    out.Write("  if (bpmem_bounding_box) {\n"
+              "    UpdateBoundingBox(rawpos.xy);\n"
+              "  }\n");
   }
 
   if (use_shader_blend)