GPU/HW: Mask bit handling in hardware renderers

Fixes:
 - Menu effect in Ghost in the Shell
 - Incorrect text colours in menu of Dragon Quest VII
 - Fade effect in TwinBee RPG
 - Fog in Silent Hill
 - Water in Duke Nukem - Land of the Babes
 - Shadows in Ultraman - Fighting Evolution

and probably others.

src/common/d3d11/texture.cpp

@@ -27,14 +27,10 @@ D3D11_TEXTURE2D_DESC Texture::GetDesc() const
   return desc;
 }
 
-bool Texture::Create(ID3D11Device* device, u32 width, u32 height, DXGI_FORMAT format, bool shader_resource,
+bool Texture::Create(ID3D11Device* device, u32 width, u32 height, DXGI_FORMAT format, u32 bind_flags,
-                     bool render_target, const void* initial_data, u32 initial_data_stride)
+                     const void* initial_data, u32 initial_data_stride)
 {
-  CD3D11_TEXTURE2D_DESC desc(format, width, height, 1, 1, 0, D3D11_USAGE_DEFAULT, 0, 1, 0, 0);
+  CD3D11_TEXTURE2D_DESC desc(format, width, height, 1, 1, bind_flags, D3D11_USAGE_DEFAULT, 0, 1, 0, 0);
-  if (shader_resource)
-    desc.BindFlags |= D3D11_BIND_SHADER_RESOURCE;
-  if (render_target)
-    desc.BindFlags |= D3D11_BIND_RENDER_TARGET;
 
   D3D11_SUBRESOURCE_DATA srd;
   srd.pSysMem = initial_data;
@@ -50,7 +46,7 @@ bool Texture::Create(ID3D11Device* device, u32 width, u32 height, DXGI_FORMAT fo
   }
 
   ComPtr<ID3D11ShaderResourceView> srv;
-  if (shader_resource)
+  if (bind_flags & D3D11_BIND_SHADER_RESOURCE)
   {
     const CD3D11_SHADER_RESOURCE_VIEW_DESC srv_desc(D3D11_SRV_DIMENSION_TEXTURE2D, desc.Format, 0, desc.MipLevels, 0,
                                                     desc.ArraySize);
@@ -63,7 +59,7 @@ bool Texture::Create(ID3D11Device* device, u32 width, u32 height, DXGI_FORMAT fo
   }
 
   ComPtr<ID3D11RenderTargetView> rtv;
-  if (render_target)
+  if (bind_flags & D3D11_BIND_RENDER_TARGET)
   {
     const CD3D11_RENDER_TARGET_VIEW_DESC rtv_desc(D3D11_RTV_DIMENSION_TEXTURE2D, desc.Format, 0, 0, desc.ArraySize);
     const HRESULT hr = device->CreateRenderTargetView(texture.Get(), &rtv_desc, rtv.GetAddressOf());

src/common/d3d11/texture.h

@@ -31,7 +31,7 @@ public:
   ALWAYS_INLINE operator ID3D11RenderTargetView*() const { return m_rtv.Get(); }
   ALWAYS_INLINE operator bool() const { return static_cast<bool>(m_texture); }
 
-  bool Create(ID3D11Device* device, u32 width, u32 height, DXGI_FORMAT format, bool shader_resource, bool render_target,
+  bool Create(ID3D11Device* device, u32 width, u32 height, DXGI_FORMAT format, u32 bind_flags,
               const void* initial_data = nullptr, u32 initial_data_stride = 0);
   bool Adopt(ID3D11Device* device, ComPtr<ID3D11Texture2D> texture);
 

src/core/gpu_hw.cpp

@@ -28,6 +28,7 @@ bool GPU_HW::Initialize(HostDisplay* host_display, System* system, DMA* dma, Int
 
   const Settings& settings = m_system->GetSettings();
   m_resolution_scale = settings.gpu_resolution_scale;
+  m_render_api = host_display->GetRenderAPI();
   m_true_color = settings.gpu_true_color;
   m_scaled_dithering = settings.gpu_scaled_dithering;
   m_texture_filtering = settings.gpu_texture_filtering;
@@ -46,10 +47,15 @@ void GPU_HW::Reset()
 {
   GPU::Reset();
 
+  m_batch_current_vertex_ptr = m_batch_start_vertex_ptr;
+
   m_vram_shadow.fill(0);
 
   m_batch = {};
   m_batch_ubo_data = {};
+  m_batch_current_vertex_depth_id = 1;
+  m_batch_next_vertex_depth_id = 2;
+  SetBatchUBOVertexDepthID(m_batch_current_vertex_depth_id);
   m_batch_ubo_dirty = true;
 
   SetFullVRAMDirtyRectangle();
@@ -62,7 +68,11 @@ bool GPU_HW::DoState(StateWrapper& sw)
 
   // invalidate the whole VRAM read texture when loading state
   if (sw.IsReading())
+  {
+    m_batch_current_vertex_ptr = m_batch_start_vertex_ptr;
     SetFullVRAMDirtyRectangle();
+    ResetBatchVertexDepthID();
+  }
 
   return true;
 }
@@ -177,12 +187,11 @@ void GPU_HW::LoadVertices()
   const RenderCommand rc{m_render_command.bits};
   const u32 texpage = ZeroExtend32(m_draw_mode.mode_reg.bits) | (ZeroExtend32(m_draw_mode.palette_reg) << 16);
 
-  // TODO: Move this to the GPU..
   switch (rc.primitive)
   {
     case Primitive::Polygon:
     {
-      EnsureVertexBufferSpace(rc.quad_polygon ? 6 : 3);
+      DebugAssert(GetBatchVertexSpace() >= (rc.quad_polygon ? 6u : 3u));
 
       const u32 first_color = rc.color_for_first_vertex;
       const bool shaded = rc.shading_enable;
@@ -308,9 +317,7 @@ void GPU_HW::LoadVertices()
       }
 
       // we can split the rectangle up into potentially 8 quads
-      const u32 required_vertices = 6 * (((rectangle_width + (TEXTURE_PAGE_WIDTH - 1)) / TEXTURE_PAGE_WIDTH) + 1u) *
+      DebugAssert(GetBatchVertexSpace() >= MAX_VERTICES_FOR_RECTANGLE);
-                                    (((rectangle_height + (TEXTURE_PAGE_HEIGHT - 1)) / TEXTURE_PAGE_HEIGHT) + 1u);
-      EnsureVertexBufferSpace(required_vertices);
 
       // Split the rectangle into multiple quads if it's greater than 256x256, as the texture page should repeat.
       u16 tex_top = orig_tex_top;
@@ -361,7 +368,7 @@ void GPU_HW::LoadVertices()
     {
       if (!rc.polyline)
       {
-        EnsureVertexBufferSpace(2);
+        DebugAssert(GetBatchVertexSpace() >= 2);
 
         u32 color0, color1;
         VertexPosition pos0, pos1;
@@ -410,7 +417,7 @@ void GPU_HW::LoadVertices()
       {
         // Multiply by two because we don't use line strips.
         const u32 num_vertices = GetPolyLineVertexCount();
-        EnsureVertexBufferSpace(num_vertices * 2);
+        DebugAssert(GetBatchVertexSpace() >= (num_vertices * 2));
 
         const u32 first_color = rc.color_for_first_vertex;
         const bool shaded = rc.shading_enable;
@@ -534,6 +541,73 @@ void GPU_HW::EnsureVertexBufferSpace(u32 required_vertices)
   MapBatchVertexPointer(required_vertices);
 }
 
+void GPU_HW::EnsureVertexBufferSpaceForCurrentCommand()
+{
+  u32 required_vertices;
+  switch (m_render_command.primitive)
+  {
+    case Primitive::Polygon:
+      required_vertices = m_render_command.quad_polygon ? 6 : 3;
+      break;
+    case Primitive::Rectangle:
+      required_vertices = MAX_VERTICES_FOR_RECTANGLE;
+      break;
+    case Primitive::Line:
+    default:
+      required_vertices = m_render_command.polyline ? (GetPolyLineVertexCount() * 2u) : 2u;
+      break;
+  }
+
+  // can we fit these vertices in the current depth buffer range?
+  if (BatchVertexDepthIDNeedsUpdate() &&
+      (m_batch_next_vertex_depth_id + GetBatchVertexCount() + required_vertices) > MAX_BATCH_VERTEX_COUNTER_IDS)
+  {
+    // implies FlushRender()
+    ResetBatchVertexDepthID();
+  }
+  else if (m_batch_current_vertex_ptr)
+  {
+    if (GetBatchVertexSpace() >= required_vertices)
+      return;
+
+    FlushRender();
+  }
+
+  MapBatchVertexPointer(required_vertices);
+}
+
+void GPU_HW::ResetBatchVertexDepthID()
+{
+  Log_PerfPrint("Resetting batch vertex depth ID");
+  FlushRender();
+  UpdateDepthBufferFromMaskBit();
+
+  m_batch_current_vertex_depth_id = 1;
+  m_batch_next_vertex_depth_id = 2;
+  SetBatchUBOVertexDepthID(m_batch_current_vertex_depth_id);
+}
+
+void GPU_HW::IncrementBatchVertexID(u32 count)
+{
+  DebugAssert((m_batch_next_vertex_depth_id + count) <= MAX_BATCH_VERTEX_COUNTER_IDS);
+  m_batch_next_vertex_depth_id += count;
+}
+
+void GPU_HW::SetBatchUBOVertexDepthID(u32 value)
+{
+  u32 ubo_value;
+
+  // In OpenGL, gl_VertexID is inclusive of the base vertex, whereas SV_VertexID in D3D isn't.
+  // We rely on unsigned overflow to compute the correct value based on the base vertex.
+  if (m_render_api != HostDisplay::RenderAPI::D3D11)
+    ubo_value = m_batch_base_vertex - value;
+  else
+    ubo_value = value;
+
+  m_batch_ubo_dirty |= (m_batch_ubo_data.u_vertex_depth_id != ubo_value);
+  m_batch_ubo_data.u_vertex_depth_id = ubo_value;
+}
+
 void GPU_HW::FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color)
 {
   IncludeVRAMDityRectangle(
@@ -544,12 +618,26 @@ void GPU_HW::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data)
 {
   DebugAssert((x + width) <= VRAM_WIDTH && (y + height) <= VRAM_HEIGHT);
   IncludeVRAMDityRectangle(Common::Rectangle<u32>::FromExtents(x, y, width, height));
+
+  if (m_GPUSTAT.check_mask_before_draw)
+  {
+    // set new vertex counter since we want this to take into consideration previous masked pixels
+    m_batch_current_vertex_depth_id = m_batch_next_vertex_depth_id++;
+    SetBatchUBOVertexDepthID(m_batch_current_vertex_depth_id);
+  }
 }
 
 void GPU_HW::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height)
 {
   IncludeVRAMDityRectangle(
     Common::Rectangle<u32>::FromExtents(dst_x, dst_y, width, height).Clamped(0, 0, VRAM_WIDTH, VRAM_HEIGHT));
+
+  if (m_GPUSTAT.check_mask_before_draw)
+  {
+    // set new vertex counter since we want this to take into consideration previous masked pixels
+    m_batch_current_vertex_depth_id = m_batch_next_vertex_depth_id++;
+    SetBatchUBOVertexDepthID(m_batch_current_vertex_depth_id);
+  }
 }
 
 void GPU_HW::DispatchRenderCommand()
@@ -600,6 +688,8 @@ void GPU_HW::DispatchRenderCommand()
     FlushRender();
   }
 
+  EnsureVertexBufferSpaceForCurrentCommand();
+
   // transparency mode change
   if (m_batch.transparency_mode != transparency_mode && transparency_mode != TransparencyMode::Disabled)
   {
@@ -614,7 +704,8 @@ void GPU_HW::DispatchRenderCommand()
   {
     m_batch.check_mask_before_draw = m_GPUSTAT.check_mask_before_draw;
     m_batch.set_mask_while_drawing = m_GPUSTAT.set_mask_while_drawing;
-    m_batch_ubo_data.u_set_mask_while_drawing = BoolToUInt32(m_GPUSTAT.set_mask_while_drawing);
+    m_batch_ubo_data.u_check_mask_before_draw = BoolToUInt32(m_batch.check_mask_before_draw);
+    m_batch_ubo_data.u_set_mask_while_drawing = BoolToUInt32(m_batch.set_mask_while_drawing);
     m_batch_ubo_dirty = true;
   }
 
@@ -657,6 +748,10 @@ void GPU_HW::FlushRender()
   if (vertex_count == 0)
     return;
 
+  const bool update_depth_id = BatchVertexDepthIDNeedsUpdate();
+  if (update_depth_id)
+    SetBatchUBOVertexDepthID(m_batch_next_vertex_depth_id);
+
   if (m_drawing_area_changed)
   {
     m_drawing_area_changed = false;
@@ -680,6 +775,9 @@ void GPU_HW::FlushRender()
     m_renderer_stats.num_batches++;
     DrawBatchVertices(m_batch.GetRenderMode(), m_batch_base_vertex, vertex_count);
   }
+
+  if (update_depth_id)
+    IncrementBatchVertexID(vertex_count);
 }
 
 void GPU_HW::DrawRendererStats(bool is_idle_frame)

src/core/gpu_hw.h

@@ -1,6 +1,7 @@
 #pragma once
 #include "common/heap_array.h"
 #include "gpu.h"
+#include "host_display.h"
 #include <sstream>
 #include <string>
 #include <tuple>
@@ -42,7 +43,10 @@ protected:
   {
     VRAM_UPDATE_TEXTURE_BUFFER_SIZE = VRAM_WIDTH * VRAM_HEIGHT * sizeof(u32),
     VERTEX_BUFFER_SIZE = 1 * 1024 * 1024,
-    UNIFORM_BUFFER_SIZE = 512 * 1024
+    UNIFORM_BUFFER_SIZE = 512 * 1024,
+    MAX_BATCH_VERTEX_COUNTER_IDS = 65536 - 2,
+    MAX_VERTICES_FOR_RECTANGLE = 6 * (((MAX_PRIMITIVE_WIDTH + (TEXTURE_PAGE_WIDTH - 1)) / TEXTURE_PAGE_WIDTH) + 1u) *
+                                 (((MAX_PRIMITIVE_HEIGHT + (TEXTURE_PAGE_HEIGHT - 1)) / TEXTURE_PAGE_HEIGHT) + 1u)
   };
 
   struct BatchVertex
@@ -102,8 +106,19 @@ protected:
     u32 u_texture_window_offset[2];
     float u_src_alpha_factor;
     float u_dst_alpha_factor;
-    u32 u_set_mask_while_drawing;
     u32 u_interlaced_displayed_field;
+    u32 u_vertex_depth_id;
+    u32 u_check_mask_before_draw;
+    u32 u_set_mask_while_drawing;
+  };
+
+  struct VRAMWriteUBOData
+  {
+    u32 u_base_coords[2];
+    u32 u_size[2];
+    u32 u_buffer_base_offset;
+    u32 u_mask_or_bits;
+    float u_depth_value;
   };
 
   struct VRAMCopyUBOData
@@ -115,6 +130,7 @@ protected:
     u32 u_width;
     u32 u_height;
     u32 u_set_mask_bit;
+    float u_depth_value;
   };
 
   struct RendererStats
@@ -133,6 +149,7 @@ protected:
   }
 
   virtual void UpdateVRAMReadTexture() = 0;
+  virtual void UpdateDepthBufferFromMaskBit() = 0;
   virtual void SetScissorFromDrawingArea() = 0;
   virtual void MapBatchVertexPointer(u32 required_vertices) = 0;
   virtual void UnmapBatchVertexPointer(u32 used_vertices) = 0;
@@ -147,11 +164,28 @@ protected:
   void ClearVRAMDirtyRectangle() { m_vram_dirty_rect.SetInvalid(); }
   void IncludeVRAMDityRectangle(const Common::Rectangle<u32>& rect);
 
+  bool IsFlushed() const { return m_batch_current_vertex_ptr == m_batch_start_vertex_ptr; }
+
   u32 GetBatchVertexSpace() const { return static_cast<u32>(m_batch_end_vertex_ptr - m_batch_current_vertex_ptr); }
   u32 GetBatchVertexCount() const { return static_cast<u32>(m_batch_current_vertex_ptr - m_batch_start_vertex_ptr); }
   void EnsureVertexBufferSpace(u32 required_vertices);
+  void EnsureVertexBufferSpaceForCurrentCommand();
+  void ResetBatchVertexDepthID();
+  void IncrementBatchVertexID(u32 count);
+  void SetBatchUBOVertexDepthID(u32 value);
 
-  bool IsFlushed() const { return m_batch_current_vertex_ptr == m_batch_start_vertex_ptr; }
+  /// Returns the value to be written to the depth buffer for the current operation for mask bit emulation.
+  ALWAYS_INLINE float GetCurrentNormalizedBatchVertexDepthID() const
+  {
+    return 1.0f - (static_cast<float>(m_batch_next_vertex_depth_id) / 65535.0f);
+  }
+
+  /// Returns true if the batch vertex depth ID needs to be updated.
+  ALWAYS_INLINE bool BatchVertexDepthIDNeedsUpdate() const
+  {
+    // because GL uses base vertex we're incrementing the depth id every draw whether we like it or not
+    return m_batch.check_mask_before_draw || m_render_api != HostDisplay::RenderAPI::D3D11;
+  }
 
   void FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color) override;
   void UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data) override;
@@ -182,9 +216,12 @@ protected:
   BatchVertex* m_batch_end_vertex_ptr = nullptr;
   BatchVertex* m_batch_current_vertex_ptr = nullptr;
   u32 m_batch_base_vertex = 0;
+  u32 m_batch_current_vertex_depth_id = 0;
+  u32 m_batch_next_vertex_depth_id = 0;
 
   u32 m_resolution_scale = 1;
   u32 m_max_resolution_scale = 1;
+  HostDisplay::RenderAPI m_render_api = HostDisplay::RenderAPI::None;
   bool m_true_color = true;
   bool m_scaled_dithering = false;
   bool m_texture_filtering = false;

src/core/gpu_hw_d3d11.cpp

@@ -111,8 +111,7 @@ void GPU_HW_D3D11::RestoreGraphicsAPIState()
   m_context->IASetVertexBuffers(0, 1, m_vertex_stream_buffer.GetD3DBufferArray(), &stride, &offset);
   m_context->IASetInputLayout(m_batch_input_layout.Get());
   m_context->PSSetShaderResources(0, 1, m_vram_read_texture.GetD3DSRVArray());
-  m_context->OMSetDepthStencilState(m_depth_disabled_state.Get(), 0);
+  m_context->OMSetRenderTargets(1, m_vram_texture.GetD3DRTVArray(), m_vram_depth_view.Get());
-  m_context->OMSetRenderTargets(1, m_vram_texture.GetD3DRTVArray(), nullptr);
   m_context->RSSetState(m_cull_none_rasterizer_state.Get());
   SetViewport(0, 0, m_vram_texture.GetWidth(), m_vram_texture.GetHeight());
   SetScissorFromDrawingArea();
@@ -171,16 +170,29 @@ bool GPU_HW_D3D11::CreateFramebuffer()
   const u32 texture_width = VRAM_WIDTH * m_resolution_scale;
   const u32 texture_height = VRAM_HEIGHT * m_resolution_scale;
   const DXGI_FORMAT texture_format = DXGI_FORMAT_R8G8B8A8_UNORM;
+  const DXGI_FORMAT depth_format = DXGI_FORMAT_D16_UNORM;
 
-  if (!m_vram_texture.Create(m_device.Get(), texture_width, texture_height, texture_format, true, true) ||
+  if (!m_vram_texture.Create(m_device.Get(), texture_width, texture_height, texture_format,
-      !m_vram_read_texture.Create(m_device.Get(), texture_width, texture_height, texture_format, true, false) ||
+                             D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_RENDER_TARGET) ||
-      !m_display_texture.Create(m_device.Get(), texture_width, texture_height, texture_format, true, true) ||
+      !m_vram_depth_texture.Create(m_device.Get(), texture_width, texture_height, depth_format,
-      !m_vram_encoding_texture.Create(m_device.Get(), VRAM_WIDTH, VRAM_HEIGHT, texture_format, true, true) ||
+                                   D3D11_BIND_DEPTH_STENCIL) ||
+      !m_vram_read_texture.Create(m_device.Get(), texture_width, texture_height, texture_format,
+                                  D3D11_BIND_SHADER_RESOURCE) ||
+      !m_display_texture.Create(m_device.Get(), texture_width, texture_height, texture_format,
+                                D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_RENDER_TARGET) ||
+      !m_vram_encoding_texture.Create(m_device.Get(), VRAM_WIDTH, VRAM_HEIGHT, texture_format,
+                                      D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_RENDER_TARGET) ||
       !m_vram_readback_texture.Create(m_device.Get(), VRAM_WIDTH, VRAM_HEIGHT, texture_format, false))
   {
     return false;
   }
 
+  const CD3D11_DEPTH_STENCIL_VIEW_DESC depth_view_desc(D3D11_DSV_DIMENSION_TEXTURE2D, depth_format);
+  HRESULT hr =
+    m_device->CreateDepthStencilView(m_vram_depth_texture, &depth_view_desc, m_vram_depth_view.GetAddressOf());
+  if (FAILED(hr))
+    return false;
+
   // do we need to restore the framebuffer after a size change?
   if (old_vram_texture)
   {
@@ -192,10 +204,12 @@ bool GPU_HW_D3D11::CreateFramebuffer()
     BlitTexture(m_vram_texture.GetD3DRTV(), 0, 0, m_vram_texture.GetWidth(), m_vram_texture.GetHeight(),
                 old_vram_texture.GetD3DSRV(), 0, 0, old_vram_texture.GetWidth(), old_vram_texture.GetHeight(),
                 old_vram_texture.GetWidth(), old_vram_texture.GetHeight(), linear_filter);
+    UpdateDepthBufferFromMaskBit();
   }
 
   m_context->OMSetRenderTargets(1, m_vram_texture.GetD3DRTVArray(), nullptr);
   SetFullVRAMDirtyRectangle();
+  RestoreGraphicsAPIState();
   return true;
 }
 
@@ -203,12 +217,16 @@ void GPU_HW_D3D11::ClearFramebuffer()
 {
   static constexpr std::array<float, 4> color = {};
   m_context->ClearRenderTargetView(m_vram_texture.GetD3DRTV(), color.data());
+  m_context->ClearDepthStencilView(m_vram_depth_view.Get(), D3D11_CLEAR_DEPTH, 0.0f, 0);
+  m_context->ClearRenderTargetView(m_display_texture, color.data());
   SetFullVRAMDirtyRectangle();
 }
 
 void GPU_HW_D3D11::DestroyFramebuffer()
 {
   m_vram_read_texture.Destroy();
+  m_vram_depth_view.Reset();
+  m_vram_depth_texture.Destroy();
   m_vram_texture.Destroy();
   m_vram_encoding_texture.Destroy();
   m_display_texture.Destroy();
@@ -289,11 +307,28 @@ bool GPU_HW_D3D11::CreateStateObjects()
   if (FAILED(hr))
     return false;
 
+  ds_desc.DepthEnable = TRUE;
+  ds_desc.DepthWriteMask = D3D11_DEPTH_WRITE_MASK_ALL;
+  ds_desc.DepthFunc = D3D11_COMPARISON_ALWAYS;
+  hr = m_device->CreateDepthStencilState(&ds_desc, m_depth_test_always_state.ReleaseAndGetAddressOf());
+  if (FAILED(hr))
+    return false;
+
+  ds_desc.DepthFunc = D3D11_COMPARISON_GREATER_EQUAL;
+  hr = m_device->CreateDepthStencilState(&ds_desc, m_depth_test_less_state.ReleaseAndGetAddressOf());
+  if (FAILED(hr))
+    return false;
+
   CD3D11_BLEND_DESC bl_desc = CD3D11_BLEND_DESC(CD3D11_DEFAULT());
   hr = m_device->CreateBlendState(&bl_desc, m_blend_disabled_state.ReleaseAndGetAddressOf());
   if (FAILED(hr))
     return false;
 
+  bl_desc.RenderTarget[0].RenderTargetWriteMask = 0;
+  hr = m_device->CreateBlendState(&bl_desc, m_blend_no_color_writes_state.ReleaseAndGetAddressOf());
+  if (FAILED(hr))
+    return false;
+
   CD3D11_SAMPLER_DESC sampler_desc = CD3D11_SAMPLER_DESC(CD3D11_DEFAULT());
   sampler_desc.Filter = D3D11_FILTER_MIN_MAG_MIP_POINT;
   hr = m_device->CreateSamplerState(&sampler_desc, m_point_sampler_state.ReleaseAndGetAddressOf());
@@ -306,12 +341,9 @@ bool GPU_HW_D3D11::CreateStateObjects()
     return false;
 
   for (u8 transparency_mode = 0; transparency_mode < 5; transparency_mode++)
-  {
-    if (transparency_mode == static_cast<u8>(TransparencyMode::Disabled) && !m_texture_filtering)
   {
     bl_desc = CD3D11_BLEND_DESC(CD3D11_DEFAULT());
-    }
+    if (transparency_mode != static_cast<u8>(TransparencyMode::Disabled) || m_texture_filtering)
-    else
     {
       bl_desc.RenderTarget[0].BlendEnable = TRUE;
       bl_desc.RenderTarget[0].SrcBlend = D3D11_BLEND_ONE;
@@ -409,6 +441,11 @@ bool GPU_HW_D3D11::CompileShaders()
   if (!m_vram_copy_pixel_shader)
     return false;
 
+  m_vram_update_depth_pixel_shader =
+    m_shader_cache.GetPixelShader(m_device.Get(), shadergen.GenerateVRAMUpdateDepthFragmentShader());
+  if (!m_vram_update_depth_pixel_shader)
+    return false;
+
   for (u8 depth_24bit = 0; depth_24bit < 2; depth_24bit++)
   {
     for (u8 interlacing = 0; interlacing < 2; interlacing++)
@@ -467,6 +504,7 @@ void GPU_HW_D3D11::BlitTexture(ID3D11RenderTargetView* dst, u32 dst_x, u32 dst_y
                              static_cast<float>(src_height) / static_cast<float>(src_texture_height)};
 
   m_context->OMSetRenderTargets(1, &dst, nullptr);
+  m_context->OMSetDepthStencilState(m_depth_disabled_state.Get(), 0);
   m_context->PSSetShaderResources(0, 1, &src);
   m_context->PSSetSamplers(
     0, 1, linear_filter ? m_linear_sampler_state.GetAddressOf() : m_point_sampler_state.GetAddressOf());
@@ -516,6 +554,8 @@ void GPU_HW_D3D11::DrawBatchVertices(BatchRenderMode render_mode, u32 base_verte
   const TransparencyMode transparency_mode =
     (render_mode == BatchRenderMode::OnlyOpaque) ? TransparencyMode::Disabled : m_batch.transparency_mode;
   m_context->OMSetBlendState(m_batch_blend_states[static_cast<u8>(transparency_mode)].Get(), nullptr, 0xFFFFFFFFu);
+  m_context->OMSetDepthStencilState(
+    m_batch.check_mask_before_draw ? m_depth_test_less_state.Get() : m_depth_test_always_state.Get(), 0);
 
   m_context->Draw(num_vertices, base_vertex);
 }
@@ -567,6 +607,7 @@ void GPU_HW_D3D11::UpdateDisplay()
     else
     {
       m_context->OMSetRenderTargets(1, m_display_texture.GetD3DRTVArray(), nullptr);
+      m_context->OMSetDepthStencilState(m_depth_disabled_state.Get(), 0);
       m_context->PSSetShaderResources(0, 1, m_vram_texture.GetD3DSRVArray());
 
       const u32 reinterpret_field_offset = GetInterlacedField();
@@ -604,6 +645,7 @@ void GPU_HW_D3D11::ReadVRAM(u32 x, u32 y, u32 width, u32 height)
   // Encode the 24-bit texture as 16-bit.
   const u32 uniforms[4] = {copy_rect.left, copy_rect.top, copy_rect.GetWidth(), copy_rect.GetHeight()};
   m_context->OMSetRenderTargets(1, m_vram_encoding_texture.GetD3DRTVArray(), nullptr);
+  m_context->OMSetDepthStencilState(m_depth_disabled_state.Get(), 0);
   m_context->PSSetShaderResources(0, 1, m_vram_texture.GetD3DSRVArray());
   SetViewportAndScissor(0, 0, encoded_width, encoded_height);
   DrawUtilityShader(m_vram_read_pixel_shader.Get(), uniforms, sizeof(uniforms));
@@ -654,6 +696,8 @@ void GPU_HW_D3D11::FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color)
     RGBA8ToFloat(color);
   uniforms.u_interlaced_displayed_field = GetInterlacedField();
 
+  m_context->OMSetDepthStencilState(m_depth_test_always_state.Get(), 0);
+
   SetViewportAndScissor(x * m_resolution_scale, y * m_resolution_scale, width * m_resolution_scale,
                         height * m_resolution_scale);
   DrawUtilityShader(IsInterlacedRenderingEnabled() ? m_vram_interlaced_fill_pixel_shader.Get() :
@@ -682,13 +726,21 @@ void GPU_HW_D3D11::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* d
   std::memcpy(map_result.pointer, data, num_pixels * sizeof(u16));
   m_texture_stream_buffer.Unmap(m_context.Get(), num_pixels * sizeof(u16));
 
-  const u32 uniforms[5] = {x, y, width, height, map_result.index_aligned};
+  const VRAMWriteUBOData uniforms = {x,
+                                     y,
+                                     width,
+                                     height,
+                                     map_result.index_aligned,
+                                     m_GPUSTAT.set_mask_while_drawing ? 0xFFu : 0x00,
+                                     GetCurrentNormalizedBatchVertexDepthID()};
+  m_context->OMSetDepthStencilState(
+    m_GPUSTAT.check_mask_before_draw ? m_depth_test_less_state.Get() : m_depth_test_always_state.Get(), 0);
   m_context->PSSetShaderResources(0, 1, m_texture_stream_buffer_srv_r16ui.GetAddressOf());
 
   // the viewport should already be set to the full vram, so just adjust the scissor
   SetScissor(x * m_resolution_scale, y * m_resolution_scale, width * m_resolution_scale, height * m_resolution_scale);
 
-  DrawUtilityShader(m_vram_write_pixel_shader.Get(), uniforms, sizeof(uniforms));
+  DrawUtilityShader(m_vram_write_pixel_shader.Get(), &uniforms, sizeof(uniforms));
 
   RestoreGraphicsAPIState();
 }
@@ -703,19 +755,20 @@ void GPU_HW_D3D11::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 widt
       UpdateVRAMReadTexture();
     IncludeVRAMDityRectangle(dst_bounds);
 
-    const VRAMCopyUBOData uniforms = {
+    const VRAMCopyUBOData uniforms = {src_x * m_resolution_scale,
-      src_x * m_resolution_scale,
                                       src_y * m_resolution_scale,
                                       dst_x * m_resolution_scale,
                                       dst_y * m_resolution_scale,
                                       width * m_resolution_scale,
                                       height * m_resolution_scale,
                                       m_GPUSTAT.set_mask_while_drawing ? 1u : 0u,
-    };
+                                      GetCurrentNormalizedBatchVertexDepthID()};
 
     const Common::Rectangle<u32> dst_bounds_scaled(dst_bounds * m_resolution_scale);
     SetViewportAndScissor(dst_bounds_scaled.left, dst_bounds_scaled.top, dst_bounds_scaled.GetWidth(),
                           dst_bounds_scaled.GetHeight());
+    m_context->OMSetDepthStencilState(
+      m_GPUSTAT.check_mask_before_draw ? m_depth_test_less_state.Get() : m_depth_test_always_state.Get(), 0);
     m_context->PSSetShaderResources(0, 1, m_vram_read_texture.GetD3DSRVArray());
     DrawUtilityShader(m_vram_copy_pixel_shader.Get(), &uniforms, sizeof(uniforms));
     RestoreGraphicsAPIState();
@@ -728,6 +781,9 @@ void GPU_HW_D3D11::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 widt
   if (m_vram_dirty_rect.Intersects(Common::Rectangle<u32>::FromExtents(src_x, src_y, width, height)))
     UpdateVRAMReadTexture();
 
+  if (m_GPUSTAT.IsMaskingEnabled())
+    Log_WarningPrintf("Masking enabled on VRAM copy - not implemented");
+
   GPU_HW::CopyVRAM(src_x, src_y, dst_x, dst_y, width, height);
 
   src_x *= m_resolution_scale;
@@ -749,6 +805,21 @@ void GPU_HW_D3D11::UpdateVRAMReadTexture()
                                    &src_box);
 }
 
+void GPU_HW_D3D11::UpdateDepthBufferFromMaskBit()
+{
+  SetViewportAndScissor(0, 0, m_vram_texture.GetWidth(), m_vram_texture.GetHeight());
+
+  m_context->OMSetRenderTargets(0, nullptr, m_vram_depth_view.Get());
+  m_context->OMSetDepthStencilState(m_depth_test_always_state.Get(), 0);
+  m_context->OMSetBlendState(m_blend_no_color_writes_state.Get(), nullptr, 0xFFFFFFFFu);
+
+  m_context->PSSetShaderResources(0, 1, m_vram_texture.GetD3DSRVArray());
+  DrawUtilityShader(m_vram_update_depth_pixel_shader.Get(), nullptr, 0);
+
+  m_context->PSSetShaderResources(0, 1, m_vram_read_texture.GetD3DSRVArray());
+  RestoreGraphicsAPIState();
+}
+
 std::unique_ptr<GPU> GPU::CreateHardwareD3D11Renderer()
 {
   return std::make_unique<GPU_HW_D3D11>();

src/core/gpu_hw_d3d11.h

@@ -34,6 +34,7 @@ protected:
   void UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data) override;
   void CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height) override;
   void UpdateVRAMReadTexture() override;
+  void UpdateDepthBufferFromMaskBit() override;
   void SetScissorFromDrawingArea() override;
   void MapBatchVertexPointer(u32 required_vertices) override;
   void UnmapBatchVertexPointer(u32 used_vertices) override;
@@ -77,6 +78,8 @@ private:
 
   // downsample texture - used for readbacks at >1xIR.
   D3D11::Texture m_vram_texture;
+  D3D11::Texture m_vram_depth_texture;
+  ComPtr<ID3D11DepthStencilView> m_vram_depth_view;
   D3D11::Texture m_vram_read_texture;
   D3D11::Texture m_vram_encoding_texture;
   D3D11::Texture m_display_texture;
@@ -94,8 +97,11 @@ private:
   ComPtr<ID3D11RasterizerState> m_cull_none_rasterizer_state;
 
   ComPtr<ID3D11DepthStencilState> m_depth_disabled_state;
+  ComPtr<ID3D11DepthStencilState> m_depth_test_always_state;
+  ComPtr<ID3D11DepthStencilState> m_depth_test_less_state;
 
   ComPtr<ID3D11BlendState> m_blend_disabled_state;
+  ComPtr<ID3D11BlendState> m_blend_no_color_writes_state;
 
   ComPtr<ID3D11SamplerState> m_point_sampler_state;
   ComPtr<ID3D11SamplerState> m_linear_sampler_state;
@@ -114,5 +120,6 @@ private:
   ComPtr<ID3D11PixelShader> m_vram_read_pixel_shader;
   ComPtr<ID3D11PixelShader> m_vram_write_pixel_shader;
   ComPtr<ID3D11PixelShader> m_vram_copy_pixel_shader;
+  ComPtr<ID3D11PixelShader> m_vram_update_depth_pixel_shader;
   std::array<std::array<ComPtr<ID3D11PixelShader>, 2>, 2> m_display_pixel_shaders; // [depth_24][interlaced]
 };

src/core/gpu_hw_opengl.cpp

@@ -11,6 +11,8 @@ GPU_HW_OpenGL::GPU_HW_OpenGL() : GPU_HW() {}
 GPU_HW_OpenGL::~GPU_HW_OpenGL()
 {
   // Destroy objects which don't have destructors to clean them up
+  if (m_vram_fbo_id != 0)
+    glDeleteFramebuffers(1, &m_vram_fbo_id);
   if (m_vao_id != 0)
     glDeleteVertexArrays(1, &m_vao_id);
   if (m_attributeless_vao_id != 0)
@@ -90,7 +92,6 @@ void GPU_HW_OpenGL::ResetGraphicsAPIState()
   glEnable(GL_CULL_FACE);
   glDisable(GL_SCISSOR_TEST);
   glDisable(GL_BLEND);
-  glDepthMask(GL_TRUE);
   if (m_resolution_scale > 1 && !m_supports_geometry_shaders)
     glLineWidth(1.0f);
   glBindVertexArray(0);
@@ -98,13 +99,14 @@ void GPU_HW_OpenGL::ResetGraphicsAPIState()
 
 void GPU_HW_OpenGL::RestoreGraphicsAPIState()
 {
-  m_vram_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
+  glBindFramebuffer(GL_DRAW_FRAMEBUFFER, m_vram_fbo_id);
   glViewport(0, 0, m_vram_texture.GetWidth(), m_vram_texture.GetHeight());
 
   glDisable(GL_CULL_FACE);
-  glDisable(GL_DEPTH_TEST);
+  glEnable(GL_DEPTH_TEST);
   glEnable(GL_SCISSOR_TEST);
-  glDepthMask(GL_FALSE);
+  glDepthMask(GL_TRUE);
+  glDepthFunc(GL_ALWAYS);
   if (m_resolution_scale > 1 && !m_supports_geometry_shaders)
     glLineWidth(static_cast<float>(m_resolution_scale));
   glBindVertexArray(m_vao_id);
@@ -211,34 +213,16 @@ bool GPU_HW_OpenGL::CreateFramebuffer()
 {
   // save old vram texture/fbo, in case we're changing scale
   GL::Texture old_vram_texture = std::move(m_vram_texture);
+  GLuint old_vram_fbo = m_vram_fbo_id;
 
   // scale vram size to internal resolution
   const u32 texture_width = VRAM_WIDTH * m_resolution_scale;
   const u32 texture_height = VRAM_HEIGHT * m_resolution_scale;
 
   if (!m_vram_texture.Create(texture_width, texture_height, GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, nullptr, false) ||
-      !m_vram_texture.CreateFramebuffer())
+      !m_vram_depth_texture.Create(texture_width, texture_height, GL_DEPTH_COMPONENT16, GL_DEPTH_COMPONENT,
-  {
+                                   GL_UNSIGNED_SHORT, nullptr, false) ||
-    return false;
+      !m_vram_read_texture.Create(texture_width, texture_height, GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, nullptr, false) ||
-  }
-
-  // do we need to restore the framebuffer after a size change?
-  if (old_vram_texture.IsValid())
-  {
-    const bool linear_filter = old_vram_texture.GetWidth() > m_vram_texture.GetWidth();
-    Log_DevPrintf("Scaling %ux%u VRAM texture to %ux%u using %s filter", old_vram_texture.GetWidth(),
-                  old_vram_texture.GetHeight(), m_vram_texture.GetWidth(), m_vram_texture.GetHeight(),
-                  linear_filter ? "linear" : "nearest");
-    glDisable(GL_SCISSOR_TEST);
-    old_vram_texture.BindFramebuffer(GL_READ_FRAMEBUFFER);
-    glBlitFramebuffer(0, 0, old_vram_texture.GetWidth(), old_vram_texture.GetHeight(), 0, 0, m_vram_texture.GetWidth(),
-                      m_vram_texture.GetHeight(), GL_COLOR_BUFFER_BIT, linear_filter ? GL_LINEAR : GL_NEAREST);
-
-    glEnable(GL_SCISSOR_TEST);
-    old_vram_texture.Destroy();
-  }
-
-  if (!m_vram_read_texture.Create(texture_width, texture_height, GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, nullptr, false) ||
       !m_vram_read_texture.CreateFramebuffer() ||
       !m_vram_encoding_texture.Create(VRAM_WIDTH, VRAM_HEIGHT, GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, nullptr, false) ||
       !m_vram_encoding_texture.CreateFramebuffer() ||
@@ -248,7 +232,32 @@ bool GPU_HW_OpenGL::CreateFramebuffer()
     return false;
   }
 
-  m_vram_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
+  glGenFramebuffers(1, &m_vram_fbo_id);
+  glBindFramebuffer(GL_DRAW_FRAMEBUFFER, m_vram_fbo_id);
+  glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, m_vram_texture.GetGLId(), 0);
+  glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, m_vram_depth_texture.GetGLId(), 0);
+  Assert(glCheckFramebufferStatus(GL_DRAW_FRAMEBUFFER) == GL_FRAMEBUFFER_COMPLETE);
+
+  // do we need to restore the framebuffer after a size change?
+  if (old_vram_fbo != 0)
+  {
+    const bool linear_filter = old_vram_texture.GetWidth() > m_vram_texture.GetWidth();
+    Log_DevPrintf("Scaling %ux%u VRAM texture to %ux%u using %s filter", old_vram_texture.GetWidth(),
+                  old_vram_texture.GetHeight(), m_vram_texture.GetWidth(), m_vram_texture.GetHeight(),
+                  linear_filter ? "linear" : "nearest");
+    glDisable(GL_SCISSOR_TEST);
+
+    glBindFramebuffer(GL_READ_FRAMEBUFFER, old_vram_fbo);
+    glBlitFramebuffer(0, 0, old_vram_texture.GetWidth(), old_vram_texture.GetHeight(), 0, 0, m_vram_texture.GetWidth(),
+                      m_vram_texture.GetHeight(), GL_COLOR_BUFFER_BIT, linear_filter ? GL_LINEAR : GL_NEAREST);
+
+    glEnable(GL_SCISSOR_TEST);
+    old_vram_texture.Destroy();
+    glDeleteFramebuffers(1, &old_vram_fbo);
+
+    UpdateDepthBufferFromMaskBit();
+  }
+
   SetFullVRAMDirtyRectangle();
   return true;
 }
@@ -257,7 +266,8 @@ void GPU_HW_OpenGL::ClearFramebuffer()
 {
   glDisable(GL_SCISSOR_TEST);
   glClearColor(0.0f, 0.0f, 0.0f, 0.0f);
-  glClear(GL_COLOR_BUFFER_BIT);
+  glClearDepth(0.0f);
+  glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
   glEnable(GL_SCISSOR_TEST);
   SetFullVRAMDirtyRectangle();
 }
@@ -470,6 +480,15 @@ bool GPU_HW_OpenGL::CompilePrograms()
   }
   m_vram_copy_program = std::move(*prog);
 
+  prog = m_shader_cache.GetProgram(shadergen.GenerateScreenQuadVertexShader(), {},
+                                   shadergen.GenerateVRAMUpdateDepthFragmentShader());
+  if (!prog)
+    return false;
+
+  prog->Bind();
+  prog->Uniform1i("samp0", 0);
+  m_vram_update_depth_program = std::move(*prog);
+
   if (m_supports_texture_buffer)
   {
     prog = m_shader_cache.GetProgram(shadergen.GenerateScreenQuadVertexShader(), {},
@@ -519,6 +538,8 @@ void GPU_HW_OpenGL::DrawBatchVertices(BatchRenderMode render_mode, u32 base_vert
     glBlendFuncSeparate(GL_ONE, m_supports_dual_source_blend ? GL_SRC1_ALPHA : GL_SRC_ALPHA, GL_ONE, GL_ZERO);
   }
 
+  glDepthFunc(m_GPUSTAT.check_mask_before_draw ? GL_GEQUAL : GL_ALWAYS);
+
   static constexpr std::array<GLenum, 4> gl_primitives = {{GL_LINES, GL_LINE_STRIP, GL_TRIANGLES, GL_TRIANGLE_STRIP}};
   glDrawArrays(gl_primitives[static_cast<u8>(m_batch.primitive)], m_batch_base_vertex, num_vertices);
 }
@@ -590,6 +611,7 @@ void GPU_HW_OpenGL::UpdateDisplay()
     {
       glDisable(GL_BLEND);
       glDisable(GL_SCISSOR_TEST);
+      glDisable(GL_DEPTH_TEST);
 
       m_display_programs[BoolToUInt8(m_GPUSTAT.display_area_color_depth_24)][BoolToUInt8(interlaced)].Bind();
       m_display_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
@@ -607,6 +629,7 @@ void GPU_HW_OpenGL::UpdateDisplay()
       m_batch_ubo_dirty = true;
 
       glViewport(0, reinterpret_field_offset, reinterpret_width, scaled_display_height);
+      glBindVertexArray(m_attributeless_vao_id);
       glDrawArrays(GL_TRIANGLES, 0, 3);
 
       m_host_display->SetDisplayTexture(reinterpret_cast<void*>(static_cast<uintptr_t>(m_display_texture.GetGLId())),
@@ -615,8 +638,10 @@ void GPU_HW_OpenGL::UpdateDisplay()
                                         scaled_display_width, -static_cast<s32>(scaled_display_height));
 
       // restore state
-      m_vram_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
+      glBindFramebuffer(GL_DRAW_FRAMEBUFFER, m_vram_fbo_id);
+      glBindVertexArray(m_vao_id);
       glViewport(0, 0, m_vram_texture.GetWidth(), m_vram_texture.GetHeight());
+      glEnable(GL_DEPTH_TEST);
       glEnable(GL_SCISSOR_TEST);
     }
 
@@ -644,6 +669,7 @@ void GPU_HW_OpenGL::ReadVRAM(u32 x, u32 y, u32 width, u32 height)
   glDisable(GL_BLEND);
   glDisable(GL_SCISSOR_TEST);
   glViewport(0, 0, encoded_width, encoded_height);
+  glBindVertexArray(m_attributeless_vao_id);
   glDrawArrays(GL_TRIANGLES, 0, 3);
 
   // Readback encoded texture.
@@ -688,7 +714,8 @@ void GPU_HW_OpenGL::FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color)
   {
     const auto [r, g, b, a] = RGBA8ToFloat(color);
     glClearColor(r, g, b, a);
-    glClear(GL_COLOR_BUFFER_BIT);
+    glClearDepth(a);
+    glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
     SetScissorFromDrawingArea();
   }
   else
@@ -705,6 +732,9 @@ void GPU_HW_OpenGL::FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color)
 
     m_vram_interlaced_fill_program.Bind();
     UploadUniformBuffer(&uniforms, sizeof(uniforms));
+    glDisable(GL_BLEND);
+    glDepthFunc(GL_ALWAYS);
+    glBindVertexArray(m_attributeless_vao_id);
     glDrawArrays(GL_TRIANGLES, 0, 3);
 
     RestoreGraphicsAPIState();
@@ -743,13 +773,21 @@ void GPU_HW_OpenGL::UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void*
     glViewport(scaled_x, scaled_flipped_y, scaled_width, scaled_height);
     glDisable(GL_BLEND);
     glDisable(GL_SCISSOR_TEST);
+    glDepthFunc(m_GPUSTAT.check_mask_before_draw ? GL_GEQUAL : GL_ALWAYS);
 
     m_vram_write_program.Bind();
     glBindTexture(GL_TEXTURE_BUFFER, m_texture_buffer_r16ui_texture);
 
-    const u32 uniforms[5] = {x, flipped_y, width, height, map_result.index_aligned};
+    const VRAMWriteUBOData uniforms = {x,
-    UploadUniformBuffer(uniforms, sizeof(uniforms));
+                                       flipped_y,
+                                       width,
+                                       height,
+                                       map_result.index_aligned,
+                                       m_GPUSTAT.set_mask_while_drawing ? 0xFFu : 0x00,
+                                       GetCurrentNormalizedBatchVertexDepthID()};
+    UploadUniformBuffer(&uniforms, sizeof(uniforms));
 
+    glBindVertexArray(m_attributeless_vao_id);
     glDrawArrays(GL_TRIANGLES, 0, 3);
 
     RestoreGraphicsAPIState();
@@ -822,21 +860,21 @@ void GPU_HW_OpenGL::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 wid
       UpdateVRAMReadTexture();
     IncludeVRAMDityRectangle(dst_bounds);
 
-    VRAMCopyUBOData uniforms = {
+    VRAMCopyUBOData uniforms = {src_x * m_resolution_scale,
-      src_x * m_resolution_scale,
                                 src_y * m_resolution_scale,
                                 dst_x * m_resolution_scale,
                                 dst_y * m_resolution_scale,
                                 width * m_resolution_scale,
                                 height * m_resolution_scale,
                                 m_GPUSTAT.set_mask_while_drawing ? 1u : 0u,
-    };
+                                GetCurrentNormalizedBatchVertexDepthID()};
     uniforms.u_src_y = m_vram_texture.GetHeight() - uniforms.u_src_y - uniforms.u_height;
     uniforms.u_dst_y = m_vram_texture.GetHeight() - uniforms.u_dst_y - uniforms.u_height;
     UploadUniformBuffer(&uniforms, sizeof(uniforms));
 
     glDisable(GL_SCISSOR_TEST);
     glDisable(GL_BLEND);
+    glDepthFunc(m_GPUSTAT.check_mask_before_draw ? GL_GEQUAL : GL_ALWAYS);
 
     const Common::Rectangle<u32> dst_bounds_scaled(dst_bounds * m_resolution_scale);
     glViewport(dst_bounds_scaled.left,
@@ -876,7 +914,7 @@ void GPU_HW_OpenGL::CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 wid
   else
   {
     glDisable(GL_SCISSOR_TEST);
-    m_vram_texture.BindFramebuffer(GL_READ_FRAMEBUFFER);
+    glBindFramebuffer(GL_READ_FRAMEBUFFER, m_vram_fbo_id);
     glBlitFramebuffer(src_x, src_y, src_x + width, src_y + height, dst_x, dst_y, dst_x + width, dst_y + height,
                       GL_COLOR_BUFFER_BIT, GL_NEAREST);
     glEnable(GL_SCISSOR_TEST);
@@ -904,14 +942,31 @@ void GPU_HW_OpenGL::UpdateVRAMReadTexture()
   else
   {
     m_vram_read_texture.BindFramebuffer(GL_DRAW_FRAMEBUFFER);
-    m_vram_texture.BindFramebuffer(GL_READ_FRAMEBUFFER);
+    glBindFramebuffer(GL_READ_FRAMEBUFFER, m_vram_fbo_id);
     glDisable(GL_SCISSOR_TEST);
     glBlitFramebuffer(x, y, x + width, y + height, x, y, x + width, y + height, GL_COLOR_BUFFER_BIT, GL_NEAREST);
     glEnable(GL_SCISSOR_TEST);
-    m_vram_texture.BindFramebuffer(GL_FRAMEBUFFER);
+    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, m_vram_fbo_id);
   }
 }
 
+void GPU_HW_OpenGL::UpdateDepthBufferFromMaskBit()
+{
+  glDisable(GL_SCISSOR_TEST);
+  glDisable(GL_BLEND);
+  glColorMask(GL_FALSE, GL_FALSE, GL_FALSE, GL_FALSE);
+  glDepthFunc(GL_ALWAYS);
+
+  m_vram_texture.Bind();
+  m_vram_update_depth_program.Bind();
+  glBindVertexArray(m_attributeless_vao_id);
+  glDrawArrays(GL_TRIANGLES, 0, 3);
+
+  glBindVertexArray(m_vao_id);
+  glColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE);
+  glEnable(GL_SCISSOR_TEST);
+}
+
 std::unique_ptr<GPU> GPU::CreateHardwareOpenGLRenderer()
 {
   return std::make_unique<GPU_HW_OpenGL>();

src/core/gpu_hw_opengl.h

@@ -30,6 +30,7 @@ protected:
   void UpdateVRAM(u32 x, u32 y, u32 width, u32 height, const void* data) override;
   void CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height) override;
   void UpdateVRAMReadTexture() override;
+  void UpdateDepthBufferFromMaskBit() override;
   void SetScissorFromDrawingArea() override;
   void MapBatchVertexPointer(u32 required_vertices) override;
   void UnmapBatchVertexPointer(u32 used_vertices) override;
@@ -63,11 +64,13 @@ private:
 
   // downsample texture - used for readbacks at >1xIR.
   GL::Texture m_vram_texture;
+  GL::Texture m_vram_depth_texture;
   GL::Texture m_vram_read_texture;
   GL::Texture m_vram_encoding_texture;
   GL::Texture m_display_texture;
 
   std::unique_ptr<GL::StreamBuffer> m_vertex_stream_buffer;
+  GLuint m_vram_fbo_id = 0;
   GLuint m_vao_id = 0;
   GLuint m_attributeless_vao_id = 0;
 
@@ -85,6 +88,7 @@ private:
   GL::Program m_vram_read_program;
   GL::Program m_vram_write_program;
   GL::Program m_vram_copy_program;
+  GL::Program m_vram_update_depth_program;
 
   u32 m_uniform_buffer_alignment = 1;
   u32 m_max_texture_buffer_size = 0;

src/core/gpu_hw_shadergen.cpp

@@ -319,6 +319,9 @@ void GPU_HW_ShaderGen::DeclareVertexEntryPoint(
   {
     ss << "void main(\n";
 
+    if (declare_vertex_id)
+      ss << "  in uint v_id : SV_VertexID,\n";
+
     u32 attribute_counter = 0;
     for (const char* attribute : attributes)
     {
@@ -326,9 +329,6 @@ void GPU_HW_ShaderGen::DeclareVertexEntryPoint(
       attribute_counter++;
     }
 
-    if (declare_vertex_id)
-      ss << "  in uint v_id : SV_VertexID,\n";
-
     for (u32 i = 0; i < num_color_outputs; i++)
       ss << "  out float4 v_col" << i << " : COLOR" << i << ",\n";
 
@@ -349,7 +349,7 @@ void GPU_HW_ShaderGen::DeclareVertexEntryPoint(
 void GPU_HW_ShaderGen::DeclareFragmentEntryPoint(
   std::stringstream& ss, u32 num_color_inputs, u32 num_texcoord_inputs,
   const std::initializer_list<std::pair<const char*, const char*>>& additional_inputs,
-  bool declare_fragcoord /* = false */, bool dual_color_output /* = false */)
+  bool declare_fragcoord /* = false */, u32 num_color_outputs /* = 1 */, bool depth_output /* = false */)
 {
   if (m_glsl)
   {
@@ -381,23 +381,18 @@ void GPU_HW_ShaderGen::DeclareFragmentEntryPoint(
     if (declare_fragcoord)
       ss << "#define v_pos gl_FragCoord\n";
 
+    if (depth_output)
+      ss << "#define o_depth gl_FragDepth\n";
+
     if (m_use_glsl_binding_layout)
     {
-      if (dual_color_output)
+      for (u32 i = 0; i < num_color_outputs; i++)
-      {
+        ss << "layout(location = 0, index = " << i << ") out float4 o_col" << i << ";\n";
-        ss << "layout(location = 0, index = 0) out float4 o_col0;\n";
-        ss << "layout(location = 0, index = 1) out float4 o_col1;\n";
     }
     else
     {
-        ss << "layout(location = 0) out float4 o_col0;\n";
+      for (u32 i = 0; i < num_color_outputs; i++)
-      }
+        ss << "out float4 o_col" << i << ";\n";
-    }
-    else
-    {
-      ss << "out float4 o_col0;\n";
-      if (dual_color_output)
-        ss << "out float4 o_col1;\n";
     }
 
     ss << "\n";
@@ -425,14 +420,23 @@ void GPU_HW_ShaderGen::DeclareFragmentEntryPoint(
       if (declare_fragcoord)
         ss << "  in float4 v_pos : SV_Position,\n";
 
-      if (dual_color_output)
+      if (depth_output)
       {
-        ss << "  out float4 o_col0 : SV_Target0,\n";
+        ss << "  out float o_depth : SV_Depth";
-        ss << "  out float4 o_col1 : SV_Target1)\n";
+        if (num_color_outputs > 0)
-      }
+          ss << ",\n";
         else
+          ss << ")\n";
+      }
+
+      for (u32 i = 0; i < num_color_outputs; i++)
       {
-        ss << "  out float4 o_col0 : SV_Target)";
+        ss << "  out float4 o_col" << i << " : SV_Target" << i;
+
+        if (i == (num_color_outputs - 1))
+          ss << ")\n";
+        else
+          ss << ",\n";
       }
     }
   }
@@ -440,9 +444,10 @@ void GPU_HW_ShaderGen::DeclareFragmentEntryPoint(
 
 void GPU_HW_ShaderGen::WriteBatchUniformBuffer(std::stringstream& ss)
 {
-  DeclareUniformBuffer(ss, {"uint2 u_texture_window_mask", "uint2 u_texture_window_offset", "float u_src_alpha_factor",
+  DeclareUniformBuffer(ss,
-                            "float u_dst_alpha_factor", "bool u_set_mask_while_drawing",
+                       {"uint2 u_texture_window_mask", "uint2 u_texture_window_offset", "float u_src_alpha_factor",
-                            "uint u_interlaced_displayed_field"});
+                        "float u_dst_alpha_factor", "uint u_interlaced_displayed_field", "uint u_base_vertex_depth_id",
+                        "bool u_check_mask_before_draw", "bool u_set_mask_while_drawing"});
 }
 
 std::string GPU_HW_ShaderGen::GenerateBatchVertexShader(bool textured)
@@ -459,11 +464,11 @@ std::string GPU_HW_ShaderGen::GenerateBatchVertexShader(bool textured)
   if (textured)
   {
     DeclareVertexEntryPoint(ss, {"int2 a_pos", "float4 a_col0", "uint a_texcoord", "uint a_texpage"}, 1, 1,
-                            {{"nointerpolation", "uint4 v_texpage"}});
+                            {{"nointerpolation", "uint4 v_texpage"}, {"nointerpolation", "float v_depth"}}, true);
   }
   else
   {
-    DeclareVertexEntryPoint(ss, {"int2 a_pos", "float4 a_col0"}, 1, 0, {});
+    DeclareVertexEntryPoint(ss, {"int2 a_pos", "float4 a_col0"}, 1, 0, {{"nointerpolation", "float v_depth"}}, true);
   }
 
   ss << R"(
@@ -484,6 +489,12 @@ std::string GPU_HW_ShaderGen::GenerateBatchVertexShader(bool textured)
 #endif
   v_pos = float4(pos_x, pos_y, 0.0, 1.0);
 
+#if API_D3D11
+  v_depth = 1.0 - (float(u_base_vertex_depth_id + (u_check_mask_before_draw ? 0u : v_id)) / 65535.0);
+#else
+  v_depth = 1.0 - (float(v_id - u_base_vertex_depth_id) / 65535.0);
+#endif
+
   v_col0 = a_col0;
   #if TEXTURED
     // Fudge the texture coordinates by half a pixel in screen-space.
@@ -616,11 +627,12 @@ float4 SampleFromVRAM(uint4 texpage, uint2 icoord)
 
   if (textured)
   {
-    DeclareFragmentEntryPoint(ss, 1, 1, {{"nointerpolation", "uint4 v_texpage"}}, true, use_dual_source);
+    DeclareFragmentEntryPoint(ss, 1, 1, {{"nointerpolation", "uint4 v_texpage"}, {"nointerpolation", "float v_depth"}},
+                              true, use_dual_source ? 2 : 1, true);
   }
   else
   {
-    DeclareFragmentEntryPoint(ss, 1, 0, {}, true, use_dual_source);
+    DeclareFragmentEntryPoint(ss, 1, 0, {{"nointerpolation", "float v_depth"}}, true, use_dual_source ? 2 : 1, true);
   }
 
   ss << R"(
@@ -736,6 +748,8 @@ float4 SampleFromVRAM(uint4 texpage, uint2 icoord)
       #else
         o_col0 = float4(color, u_dst_alpha_factor / ialpha);
       #endif
+
+      o_depth = oalpha * v_depth;
     }
     else
     {
@@ -752,6 +766,8 @@ float4 SampleFromVRAM(uint4 texpage, uint2 icoord)
       #else
         o_col0 = float4(color, 1.0 - ialpha);
       #endif
+
+      o_depth = oalpha * v_depth;
     }
   #else
     // Non-transparency won't enable blending so we can write the mask here regardless.
@@ -760,6 +776,8 @@ float4 SampleFromVRAM(uint4 texpage, uint2 icoord)
     #if USE_DUAL_SOURCE
       o_col1 = float4(0.0, 0.0, 0.0, 1.0 - ialpha);
     #endif
+
+    o_depth = oalpha * v_depth;
   #endif
 }
 )";
@@ -783,10 +801,12 @@ CONSTANT float2 WIDTH = (1.0 / float2(VRAM_SIZE)) * float2(RESOLUTION_SCALE, RES
     ss << R"(
 in VertexData {
   float4 v_col0;
+  nointerpolation float v_depth;
 } in_data[];
 
 out VertexData {
   float4 v_col0;
+  nointerpolation float v_depth;
 } out_data;
 
 layout(lines) in;
@@ -799,21 +819,25 @@ void main() {
 
   // top-left
   out_data.v_col0 = in_data[0].v_col0;
+  out_data.v_depth = in_data[0].v_depth;
   gl_Position = gl_in[0].gl_Position - offset;
   EmitVertex();
 
   // top-right
   out_data.v_col0 = in_data[0].v_col0;
+  out_data.v_depth = in_data[0].v_depth;
   gl_Position = gl_in[0].gl_Position + offset;
   EmitVertex();
 
   // bottom-left
   out_data.v_col0 = in_data[1].v_col0;
+  out_data.v_depth = in_data[1].v_depth;
   gl_Position = gl_in[1].gl_Position - offset;
   EmitVertex();
 
   // bottom-right
   out_data.v_col0 = in_data[1].v_col0;
+  out_data.v_depth = in_data[1].v_depth;
   gl_Position = gl_in[1].gl_Position + offset;
   EmitVertex();
 
@@ -827,6 +851,7 @@ void main() {
 struct Vertex
 {
   float4 col0 : COLOR0;
+  float depth : TEXCOORD0;
   float4 pos : SV_Position;
 };
 
@@ -841,21 +866,25 @@ void main(line Vertex input[2], inout TriangleStream<Vertex> output)
 
   // top-left
   v.col0 = input[0].col0;
+  v.depth = input[0].depth;
   v.pos = input[0].pos - offset;
   output.Append(v);
 
   // top-right
   v.col0 = input[0].col0;
+  v.depth = input[0].depth;
   v.pos = input[0].pos + offset;
   output.Append(v);
 
   // bottom-left
   v.col0 = input[1].col0;
+  v.depth = input[1].depth;
   v.pos = input[1].pos - offset;
   output.Append(v);
 
   // bottom-right
   v.col0 = input[1].col0;
+  v.depth = input[1].depth;
   v.pos = input[1].pos + offset;
   output.Append(v);
 
@@ -890,11 +919,12 @@ std::string GPU_HW_ShaderGen::GenerateFillFragmentShader()
   std::stringstream ss;
   WriteHeader(ss);
   DeclareUniformBuffer(ss, {"float4 u_fill_color"});
-  DeclareFragmentEntryPoint(ss, 0, 1, {}, false, false);
+  DeclareFragmentEntryPoint(ss, 0, 1, {}, false, 1, true);
 
   ss << R"(
 {
   o_col0 = u_fill_color;
+  o_depth = u_fill_color.a;
 }
 )";
 
@@ -907,7 +937,7 @@ std::string GPU_HW_ShaderGen::GenerateInterlacedFillFragmentShader()
   WriteHeader(ss);
   WriteCommonFunctions(ss);
   DeclareUniformBuffer(ss, {"float4 u_fill_color", "uint u_interlaced_displayed_field"});
-  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, false);
+  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, 1, true);
 
   ss << R"(
 {
@@ -915,6 +945,7 @@ std::string GPU_HW_ShaderGen::GenerateInterlacedFillFragmentShader()
     discard;
 
   o_col0 = u_fill_color;
+  o_depth = u_fill_color.a;
 }
 )";
 
@@ -927,7 +958,7 @@ std::string GPU_HW_ShaderGen::GenerateCopyFragmentShader()
   WriteHeader(ss);
   DeclareUniformBuffer(ss, {"float4 u_src_rect"});
   DeclareTexture(ss, "samp0", 0);
-  DeclareFragmentEntryPoint(ss, 0, 1, {}, false, false);
+  DeclareFragmentEntryPoint(ss, 0, 1, {}, false, 1);
 
   ss << R"(
 {
@@ -950,7 +981,7 @@ std::string GPU_HW_ShaderGen::GenerateDisplayFragmentShader(bool depth_24bit, bo
   DeclareUniformBuffer(ss, {"uint2 u_vram_offset", "uint u_field_offset"});
   DeclareTexture(ss, "samp0", 0);
 
-  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, false);
+  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, 1);
   ss << R"(
 {
   uint2 icoords = uint2(v_pos.xy) + u_vram_offset;
@@ -1013,7 +1044,7 @@ uint SampleVRAM(uint2 coords)
 }
 )";
 
-  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, false);
+  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, 1);
   ss << R"(
 {
   uint2 sample_coords = uint2(uint(v_pos.x) * 2u, uint(v_pos.y));
@@ -1043,10 +1074,11 @@ std::string GPU_HW_ShaderGen::GenerateVRAMWriteFragmentShader()
   std::stringstream ss;
   WriteHeader(ss);
   WriteCommonFunctions(ss);
-  DeclareUniformBuffer(ss, {"uint2 u_base_coords", "uint2 u_size", "uint u_buffer_base_offset"});
+  DeclareUniformBuffer(ss, {"uint2 u_base_coords", "uint2 u_size", "uint u_buffer_base_offset", "uint u_mask_or_bits",
+                            "float u_depth_value"});
 
   DeclareTextureBuffer(ss, "samp0", 0, true, true);
-  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, false);
+  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, 1, true);
   ss << R"(
 {
   uint2 coords = uint2(v_pos.xy) / uint2(RESOLUTION_SCALE, RESOLUTION_SCALE);
@@ -1058,9 +1090,10 @@ std::string GPU_HW_ShaderGen::GenerateVRAMWriteFragmentShader()
   #endif
 
   uint buffer_offset = u_buffer_base_offset + (offset.y * u_size.x) + offset.x;
-  uint value = LOAD_TEXTURE_BUFFER(samp0, int(buffer_offset)).r;
+  uint value = LOAD_TEXTURE_BUFFER(samp0, int(buffer_offset)).r | u_mask_or_bits;
   
   o_col0 = RGBA5551ToRGBA8(value);
+  o_depth = (o_col0.a == 1.0) ? u_depth_value : 0.0;
 })";
 
   return ss.str();
@@ -1071,10 +1104,11 @@ std::string GPU_HW_ShaderGen::GenerateVRAMCopyFragmentShader()
   std::stringstream ss;
   WriteHeader(ss);
   WriteCommonFunctions(ss);
-  DeclareUniformBuffer(ss, {"uint2 u_src_coords", "uint2 u_dst_coords", "uint2 u_size", "bool u_set_mask_bit"});
+  DeclareUniformBuffer(
+    ss, {"uint2 u_src_coords", "uint2 u_dst_coords", "uint2 u_size", "bool u_set_mask_bit", "float u_depth_value"});
 
   DeclareTexture(ss, "samp0", 0);
-  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, false);
+  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, 1, true);
   ss << R"(
 {
   uint2 dst_coords = uint2(v_pos.xy);
@@ -1090,7 +1124,24 @@ std::string GPU_HW_ShaderGen::GenerateVRAMCopyFragmentShader()
   // sample and apply mask bit
   float4 color = LOAD_TEXTURE(samp0, int2(src_coords), 0);
   o_col0 = float4(color.xyz, u_set_mask_bit ? 1.0 : color.a);
+  o_depth = (u_set_mask_bit ? 1.0f : ((o_col0.a == 1.0) ? u_depth_value : 0.0));
 })";
 
   return ss.str();
 }
+
+std::string GPU_HW_ShaderGen::GenerateVRAMUpdateDepthFragmentShader()
+{
+  std::stringstream ss;
+  WriteHeader(ss);
+  DeclareTexture(ss, "samp0", 0);
+  DeclareFragmentEntryPoint(ss, 0, 1, {}, true, 0, true);
+
+  ss << R"(
+{
+  o_depth = LOAD_TEXTURE(samp0, int2(v_pos.xy), 0).a;
+}
+)";
+
+  return ss.str();
+}

src/core/gpu_hw_shadergen.h

@@ -25,6 +25,7 @@ public:
   std::string GenerateVRAMReadFragmentShader();
   std::string GenerateVRAMWriteFragmentShader();
   std::string GenerateVRAMCopyFragmentShader();
+  std::string GenerateVRAMUpdateDepthFragmentShader();
 
 private:
   void SetGLSLVersionString();
@@ -38,7 +39,7 @@ private:
                                bool declare_vertex_id = false);
   void DeclareFragmentEntryPoint(std::stringstream& ss, u32 num_color_inputs, u32 num_texcoord_inputs,
                                  const std::initializer_list<std::pair<const char*, const char*>>& additional_inputs,
-                                 bool declare_fragcoord = false, bool dual_color_output = false);
+                                 bool declare_fragcoord = false, u32 num_color_outputs = 1, bool depth_output = false);
 
   void WriteCommonFunctions(std::stringstream& ss);
   void WriteBatchUniformBuffer(std::stringstream& ss);