VideoBackends: Add AbstractStagingTexture class

Can be used for asynchronous readback or upload of textures.

Source/Core/VideoCommon/AbstractStagingTexture.cpp

@@ -0,0 +1,133 @@
+// Copyright 2017 Dolphin Emulator Project
+// Licensed under GPLv2+
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <cstring>
+
+#include "Common/Assert.h"
+#include "Common/MsgHandler.h"
+#include "VideoCommon/AbstractStagingTexture.h"
+#include "VideoCommon/AbstractTexture.h"
+
+AbstractStagingTexture::AbstractStagingTexture(StagingTextureType type, const TextureConfig& c)
+    : m_type(type), m_config(c), m_texel_size(AbstractTexture::GetTexelSizeForFormat(c.format))
+{
+}
+
+AbstractStagingTexture::~AbstractStagingTexture() = default;
+
+void AbstractStagingTexture::CopyFromTexture(const AbstractTexture* src, u32 src_layer,
+                                             u32 src_level)
+{
+  MathUtil::Rectangle<int> src_rect = src->GetConfig().GetMipRect(src_level);
+  MathUtil::Rectangle<int> dst_rect = m_config.GetRect();
+  CopyFromTexture(src, src_rect, src_layer, src_level, dst_rect);
+}
+
+void AbstractStagingTexture::CopyToTexture(AbstractTexture* dst, u32 dst_layer, u32 dst_level)
+{
+  MathUtil::Rectangle<int> src_rect = m_config.GetRect();
+  MathUtil::Rectangle<int> dst_rect = dst->GetConfig().GetMipRect(dst_level);
+  CopyToTexture(src_rect, dst, dst_rect, dst_layer, dst_level);
+}
+
+void AbstractStagingTexture::ReadTexels(const MathUtil::Rectangle<int>& rect, void* out_ptr,
+                                        u32 out_stride)
+{
+  _assert_(m_type != StagingTextureType::Upload);
+  if (!PrepareForAccess())
+    return;
+
+  _assert_(rect.left >= 0 && static_cast<u32>(rect.right) <= m_config.width && rect.top >= 0 &&
+           static_cast<u32>(rect.bottom) <= m_config.height);
+
+  // Offset pointer to point to start of region being copied out.
+  const char* current_ptr = m_map_pointer;
+  current_ptr += rect.top * m_map_stride;
+  current_ptr += rect.left * m_texel_size;
+
+  // Optimal path: same dimensions, same stride.
+  if (rect.left == 0 && static_cast<u32>(rect.right) == m_config.width &&
+      m_map_stride == out_stride)
+  {
+    std::memcpy(out_ptr, current_ptr, m_map_stride * rect.GetHeight());
+    return;
+  }
+
+  size_t copy_size = std::min(static_cast<size_t>(rect.GetWidth() * m_texel_size), m_map_stride);
+  int copy_height = rect.GetHeight();
+  char* dst_ptr = reinterpret_cast<char*>(out_ptr);
+  for (int row = 0; row < copy_height; row++)
+  {
+    std::memcpy(dst_ptr, current_ptr, copy_size);
+    current_ptr += m_map_stride;
+    dst_ptr += out_stride;
+  }
+}
+
+void AbstractStagingTexture::ReadTexel(u32 x, u32 y, void* out_ptr)
+{
+  _assert_(m_type != StagingTextureType::Upload);
+  if (!PrepareForAccess())
+    return;
+
+  _assert_(x < m_config.width && y < m_config.height);
+  const char* src_ptr = m_map_pointer + y * m_map_stride + x * m_texel_size;
+  std::memcpy(out_ptr, src_ptr, m_texel_size);
+}
+
+void AbstractStagingTexture::WriteTexels(const MathUtil::Rectangle<int>& rect, const void* in_ptr,
+                                         u32 in_stride)
+{
+  _assert_(m_type != StagingTextureType::Readback);
+  if (!PrepareForAccess())
+    return;
+
+  _assert_(rect.left >= 0 && static_cast<u32>(rect.right) <= m_config.width && rect.top >= 0 &&
+           static_cast<u32>(rect.bottom) <= m_config.height);
+
+  // Offset pointer to point to start of region being copied to.
+  char* current_ptr = m_map_pointer;
+  current_ptr += rect.top * m_map_stride;
+  current_ptr += rect.left * m_texel_size;
+
+  // Optimal path: same dimensions, same stride.
+  if (rect.left == 0 && static_cast<u32>(rect.right) == m_config.width && m_map_stride == in_stride)
+  {
+    std::memcpy(current_ptr, in_ptr, m_map_stride * rect.GetHeight());
+    return;
+  }
+
+  size_t copy_size = std::min(static_cast<size_t>(rect.GetWidth() * m_texel_size), m_map_stride);
+  int copy_height = rect.GetHeight();
+  const char* src_ptr = reinterpret_cast<const char*>(in_ptr);
+  for (int row = 0; row < copy_height; row++)
+  {
+    std::memcpy(current_ptr, src_ptr, copy_size);
+    current_ptr += m_map_stride;
+    src_ptr += in_stride;
+  }
+}
+
+void AbstractStagingTexture::WriteTexel(u32 x, u32 y, const void* in_ptr)
+{
+  _assert_(m_type != StagingTextureType::Readback);
+  if (!PrepareForAccess())
+    return;
+
+  _assert_(x < m_config.width && y < m_config.height);
+  char* dest_ptr = m_map_pointer + y * m_map_stride + x * m_texel_size;
+  std::memcpy(dest_ptr, in_ptr, m_texel_size);
+}
+
+bool AbstractStagingTexture::PrepareForAccess()
+{
+  if (m_needs_flush)
+  {
+    if (IsMapped())
+      Unmap();
+    Flush();
+  }
+  return IsMapped() || Map();
+}

Source/Core/VideoCommon/AbstractStagingTexture.h

@@ -0,0 +1,86 @@
+// Copyright 2017 Dolphin Emulator Project
+// Licensed under GPLv2+
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <cstddef>
+#include <string>
+
+#include "Common/CommonTypes.h"
+#include "Common/MathUtil.h"
+#include "VideoCommon/TextureConfig.h"
+
+class AbstractTexture;
+
+class AbstractStagingTexture
+{
+public:
+  explicit AbstractStagingTexture(StagingTextureType type, const TextureConfig& c);
+  virtual ~AbstractStagingTexture();
+
+  const TextureConfig& GetConfig() const { return m_config; }
+  StagingTextureType GetType() const { return m_type; }
+  size_t GetTexelSize() const { return m_texel_size; }
+  bool IsMapped() const { return m_map_pointer != nullptr; }
+  char* GetMappedPointer() const { return m_map_pointer; }
+  size_t GetMappedStride() const { return m_map_stride; }
+  // Copies from the GPU texture object to the staging texture, which can be mapped/read by the CPU.
+  // Both src_rect and dst_rect must be with within the bounds of the the specified textures.
+  virtual void CopyFromTexture(const AbstractTexture* src, const MathUtil::Rectangle<int>& src_rect,
+                               u32 src_layer, u32 src_level,
+                               const MathUtil::Rectangle<int>& dst_rect) = 0;
+
+  // Wrapper for copying a whole layer of a texture to a readback texture.
+  // Assumes that the level of src texture and this texture have the same dimensions.
+  void CopyFromTexture(const AbstractTexture* src, u32 src_layer = 0, u32 src_level = 0);
+
+  // Copies from this staging texture to a GPU texture.
+  // Both src_rect and dst_rect must be with within the bounds of the the specified textures.
+  virtual void CopyToTexture(const MathUtil::Rectangle<int>& src_rect, AbstractTexture* dst,
+                             const MathUtil::Rectangle<int>& dst_rect, u32 dst_layer,
+                             u32 dst_level) = 0;
+
+  // Wrapper for copying a whole layer of a texture to a readback texture.
+  // Assumes that the level of src texture and this texture have the same dimensions.
+  void CopyToTexture(AbstractTexture* dst, u32 dst_layer = 0, u32 dst_level = 0);
+
+  // Maps the texture into the CPU address space, enabling it to read the contents.
+  // The Map call may not perform synchronization. If the contents of the staging texture
+  // has been updated by a CopyFromTexture call, you must call Flush() first.
+  // If persistent mapping is supported in the backend, this may be a no-op.
+  virtual bool Map() = 0;
+
+  // Unmaps the CPU-readable copy of the texture. May be a no-op on backends which
+  // support persistent-mapped buffers.
+  virtual void Unmap() = 0;
+
+  // Flushes pending writes from the CPU to the GPU, and reads from the GPU to the CPU.
+  // This may cause a command buffer flush depending on if one has occurred between the last
+  // call to CopyFromTexture()/CopyToTexture() and the Flush() call.
+  virtual void Flush() = 0;
+
+  // Reads the specified rectangle from the staging texture to out_ptr, with the specified stride
+  // (length in bytes of each row). CopyFromTexture must be called first. The contents of any
+  // texels outside of the rectangle used for CopyFromTexture is undefined.
+  void ReadTexels(const MathUtil::Rectangle<int>& rect, void* out_ptr, u32 out_stride);
+  void ReadTexel(u32 x, u32 y, void* out_ptr);
+
+  // Copies the texels from in_ptr to the staging texture, which can be read by the GPU, with the
+  // specified stride (length in bytes of each row). After updating the staging texture with all
+  // changes, call CopyToTexture() to update the GPU copy.
+  void WriteTexels(const MathUtil::Rectangle<int>& rect, const void* in_ptr, u32 in_stride);
+  void WriteTexel(u32 x, u32 y, const void* in_ptr);
+
+protected:
+  bool PrepareForAccess();
+
+  const StagingTextureType m_type;
+  const TextureConfig m_config;
+  const size_t m_texel_size;
+
+  char* m_map_pointer = nullptr;
+  size_t m_map_stride = 0;
+
+  bool m_needs_flush = false;
+};

Source/Core/VideoCommon/AbstractTexture.cpp

@@ -100,7 +100,7 @@ AbstractTexture::MapRegionImpl(u32 level, u32 x, u32 y, u32 width, u32 height)
   return {};
 }
 
-bool AbstractTexture::IsCompressedHostTextureFormat(AbstractTextureFormat format)
+bool AbstractTexture::IsCompressedFormat(AbstractTextureFormat format)
 {
   switch (format)
   {
@@ -115,7 +115,7 @@ bool AbstractTexture::IsCompressedHostTextureFormat(AbstractTextureFormat format
   }
 }
 
-size_t AbstractTexture::CalculateHostTextureLevelPitch(AbstractTextureFormat format, u32 row_length)
+size_t AbstractTexture::CalculateStrideForFormat(AbstractTextureFormat format, u32 row_length)
 {
   switch (format)
   {
@@ -134,6 +134,25 @@ size_t AbstractTexture::CalculateHostTextureLevelPitch(AbstractTextureFormat for
   }
 }
 
+size_t AbstractTexture::GetTexelSizeForFormat(AbstractTextureFormat format)
+{
+  switch (format)
+  {
+  case AbstractTextureFormat::DXT1:
+    return 8;
+  case AbstractTextureFormat::DXT3:
+  case AbstractTextureFormat::DXT5:
+  case AbstractTextureFormat::BPTC:
+    return 16;
+  case AbstractTextureFormat::RGBA8:
+  case AbstractTextureFormat::BGRA8:
+    return 4;
+  default:
+    PanicAlert("Unhandled texture format.");
+    return 0;
+  }
+}
+
 const TextureConfig& AbstractTexture::GetConfig() const
 {
   return m_config;

Source/Core/VideoCommon/AbstractTexture.h

@@ -39,8 +39,9 @@ public:
   virtual void Load(u32 level, u32 width, u32 height, u32 row_length, const u8* buffer,
                     size_t buffer_size) = 0;
 
-  static bool IsCompressedHostTextureFormat(AbstractTextureFormat format);
-  static size_t CalculateHostTextureLevelPitch(AbstractTextureFormat format, u32 row_length);
+  static bool IsCompressedFormat(AbstractTextureFormat format);
+  static size_t CalculateStrideForFormat(AbstractTextureFormat format, u32 row_length);
+  static size_t GetTexelSizeForFormat(AbstractTextureFormat format);
 
   const TextureConfig& GetConfig() const;
 

Source/Core/VideoCommon/CMakeLists.txt

@@ -1,4 +1,5 @@
 set(SRCS
+  AbstractStagingTexture.cpp
   AbstractTexture.cpp
   AsyncRequests.cpp
   AsyncShaderCompiler.cpp

Source/Core/VideoCommon/RenderBase.h

@@ -34,9 +34,11 @@
 
 class AbstractRawTexture;
 class AbstractTexture;
+class AbstractStagingTexture;
 class PostProcessingShaderImplementation;
 struct TextureConfig;
 enum class EFBAccessType;
+enum class StagingTextureType;
 
 struct EfbPokeData
 {
@@ -81,6 +83,8 @@ public:
   virtual void ResetAPIState() {}
   virtual void RestoreAPIState() {}
   virtual std::unique_ptr<AbstractTexture> CreateTexture(const TextureConfig& config) = 0;
+  virtual std::unique_ptr<AbstractStagingTexture>
+  CreateStagingTexture(StagingTextureType type, const TextureConfig& config) = 0;
 
   // Ideal internal resolution - multiple of the native EFB resolution
   int GetTargetWidth() const { return m_target_width; }

Source/Core/VideoCommon/TextureConfig.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include "VideoCommon/TextureConfig.h"
+#include "VideoCommon/AbstractTexture.h"
 
 #include <tuple>
 
@@ -12,7 +13,28 @@ bool TextureConfig::operator==(const TextureConfig& o) const
          std::tie(o.width, o.height, o.levels, o.layers, o.format, o.rendertarget);
 }
 
+bool TextureConfig::operator!=(const TextureConfig& o) const
+{
+  return !operator==(o);
+}
+
 MathUtil::Rectangle<int> TextureConfig::GetRect() const
 {
   return {0, 0, static_cast<int>(width), static_cast<int>(height)};
 }
+
+MathUtil::Rectangle<int> TextureConfig::GetMipRect(u32 level) const
+{
+  return {0, 0, static_cast<int>(std::max(width >> level, 1u)),
+          static_cast<int>(std::max(height >> level, 1u))};
+}
+
+size_t TextureConfig::GetStride() const
+{
+  return AbstractTexture::CalculateStrideForFormat(format, width);
+}
+
+size_t TextureConfig::GetMipStride(u32 level) const
+{
+  return AbstractTexture::CalculateStrideForFormat(format, std::max(width >> level, 1u));
+}

Source/Core/VideoCommon/TextureConfig.h

@@ -21,6 +21,13 @@ enum class AbstractTextureFormat : u32
   Undefined
 };
 
+enum class StagingTextureType
+{
+  Readback,  // Optimize for CPU reads, GPU writes, no CPU writes
+  Upload,    // Optimize for CPU writes, GPU reads, no CPU reads
+  Mutable    // Optimize for CPU reads, GPU writes, allow slow CPU reads
+};
+
 struct TextureConfig
 {
   constexpr TextureConfig() = default;
@@ -32,7 +39,11 @@ struct TextureConfig
   }
 
   bool operator==(const TextureConfig& o) const;
+  bool operator!=(const TextureConfig& o) const;
   MathUtil::Rectangle<int> GetRect() const;
+  MathUtil::Rectangle<int> GetMipRect(u32 level) const;
+  size_t GetStride() const;
+  size_t GetMipStride(u32 level) const;
 
   u32 width = 0;
   u32 height = 0;

Source/Core/VideoCommon/VideoCommon.vcxproj

@@ -36,6 +36,7 @@
   </ImportGroup>
   <PropertyGroup Label="UserMacros" />
   <ItemGroup>
+    <ClCompile Include="AbstractStagingTexture.cpp" />
     <ClCompile Include="AbstractTexture.cpp" />
     <ClCompile Include="AsyncRequests.cpp" />
     <ClCompile Include="AsyncShaderCompiler.cpp" />
@@ -96,6 +97,7 @@
     <ClCompile Include="XFStructs.cpp" />
   </ItemGroup>
   <ItemGroup>
+    <ClInclude Include="AbstractStagingTexture.h" />
     <ClInclude Include="AbstractTexture.h" />
     <ClInclude Include="AsyncRequests.h" />
     <ClInclude Include="AsyncShaderCompiler.h" />

Source/Core/VideoCommon/VideoCommon.vcxproj.filters

@@ -191,6 +191,9 @@
     <ClCompile Include="UberShaderVertex.cpp">
       <Filter>Shader Generators</Filter>
     </ClCompile>
+    <ClCompile Include="AbstractStagingTexture.cpp">
+      <Filter>Base</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="CommandProcessor.h" />
@@ -362,6 +365,9 @@
     <ClInclude Include="UberShaderVertex.h">
       <Filter>Shader Generators</Filter>
     </ClInclude>
+    <ClInclude Include="AbstractStagingTexture.h">
+      <Filter>Base</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <Text Include="CMakeLists.txt" />