GPU: Fix batching regression (draws split on texture page)

src/core/gpu.cpp

@@ -67,7 +67,7 @@ bool GPU::DoState(StateWrapper& sw)
   sw.Do(&m_render_state.texture_page_y);
   sw.Do(&m_render_state.texture_palette_x);
   sw.Do(&m_render_state.texture_palette_y);
-  sw.Do(&m_render_state.texture_color_mode);
+  sw.Do(&m_render_state.texture_mode);
   sw.Do(&m_render_state.transparency_mode);
   sw.Do(&m_render_state.texture_window_mask_x);
   sw.Do(&m_render_state.texture_window_mask_y);
@@ -701,7 +701,7 @@ void GPU::RenderState::SetFromPageAttribute(u16 value)
   texture_page_changed |=
     (old_page_attribute & PAGE_ATTRIBUTE_TEXTURE_PAGE_MASK) != (value & PAGE_ATTRIBUTE_TEXTURE_PAGE_MASK);
 
-  texture_color_mode = (static_cast<TextureMode>((value >> 7) & UINT16_C(0x03)));
+  texture_mode = (static_cast<TextureMode>((value >> 7) & UINT16_C(0x03)));
   transparency_mode = (static_cast<TransparencyMode>((value >> 5) & UINT16_C(0x03)));
 }
 

src/core/gpu.h

@@ -337,7 +337,7 @@ protected:
     u32 texture_page_y;
     u32 texture_palette_x;
     u32 texture_palette_y;
-    TextureMode texture_color_mode;
+    TextureMode texture_mode;
     TransparencyMode transparency_mode;
     u8 texture_window_mask_x;   // in 8 pixel steps
     u8 texture_window_mask_y;   // in 8 pixel steps
@@ -354,6 +354,13 @@ protected:
     bool texture_page_changed = false;
     bool texture_window_changed = false;
 
+    /// Returns true if the texture mode requires a palette.
+    bool IsUsingPalette() const
+    {
+      return (static_cast<u8>(texture_mode) &
+              (static_cast<u8>(TextureMode::Palette4Bit) | static_cast<u8>(TextureMode::Palette8Bit))) != 0;
+    }
+
     bool IsTexturePageChanged() const { return texture_page_changed; }
     void ClearTexturePageChangedFlag() { texture_page_changed = false; }
 

src/core/gpu_hw.cpp

@@ -620,7 +620,7 @@ void GPU_HW::DispatchRenderCommand(RenderCommand rc, u32 num_vertices, const u32
         break;
     }
 
-    texture_mode = m_render_state.texture_color_mode;
+    texture_mode = m_render_state.texture_mode;
     if (rc.raw_texture_enable)
     {
       texture_mode =
@@ -633,6 +633,35 @@ void GPU_HW::DispatchRenderCommand(RenderCommand rc, u32 num_vertices, const u32
     texture_mode = TextureMode::Disabled;
   }
 
+  // texture page changed - check that the new page doesn't intersect the drawing area
+  if (m_render_state.IsTexturePageChanged())
+  {
+    m_render_state.ClearTexturePageChangedFlag();
+
+    const u32 texture_page_left = m_render_state.texture_page_x;
+    const u32 texture_page_right = m_render_state.texture_page_y + TEXTURE_PAGE_WIDTH;
+    const u32 texture_page_top = m_render_state.texture_page_y;
+    const u32 texture_page_bottom = texture_page_top + TEXTURE_PAGE_HEIGHT;
+    const bool texture_page_overlaps =
+      (texture_page_left < m_drawing_area.right && texture_page_right > m_drawing_area.left &&
+       texture_page_top > m_drawing_area.bottom && texture_page_bottom < m_drawing_area.top);
+    const u32 texture_palette_left = m_render_state.texture_palette_x;
+    const u32 texture_palette_right = m_render_state.texture_palette_x + 256;
+    const bool texture_palette_overlaps =
+      m_render_state.IsUsingPalette() && texture_palette_left < m_drawing_area.right &&
+      texture_palette_right > m_drawing_area.left && m_render_state.texture_palette_y < m_drawing_area.bottom &&
+      m_render_state.texture_palette_y >= m_drawing_area.top;
+
+    // we only need to update the copy texture if the render area intersects with the texture page
+    if (texture_page_overlaps || texture_palette_overlaps)
+    {
+      Log_WarningPrintf("Invalidating VRAM read cache due to drawing area overlap");
+      if (!IsFlushed())
+        FlushRender();
+      InvalidateVRAMReadCache();
+    }
+  }
+
   // has any state changed which requires a new batch?
   const TransparencyMode transparency_mode =
     rc.transparency_enable ? m_render_state.transparency_mode : TransparencyMode::Disabled;
@@ -644,8 +673,7 @@ void GPU_HW::DispatchRenderCommand(RenderCommand rc, u32 num_vertices, const u32
     const bool buffer_overflow = GetBatchVertexSpace() < max_added_vertices;
     if (buffer_overflow || rc_primitive == HWPrimitive::LineStrip || m_batch.texture_mode != texture_mode ||
         m_batch.transparency_mode != transparency_mode || m_batch.primitive != rc_primitive ||
-        dithering_enable != m_batch.dithering || m_render_state.IsTexturePageChanged() ||
+        dithering_enable != m_batch.dithering || m_render_state.IsTextureWindowChanged())
-        m_render_state.IsTextureWindowChanged())
     {
       FlushRender();
     }
@@ -661,31 +689,6 @@ void GPU_HW::DispatchRenderCommand(RenderCommand rc, u32 num_vertices, const u32
   m_batch.transparency_mode = transparency_mode;
   m_batch.dithering = dithering_enable;
 
-  if (m_render_state.IsTexturePageChanged())
-  {
-    // we only need to update the copy texture if the render area intersects with the texture page
-    const u32 texture_page_left = m_render_state.texture_page_x;
-    const u32 texture_page_right = m_render_state.texture_page_y + TEXTURE_PAGE_WIDTH;
-    const u32 texture_page_top = m_render_state.texture_page_y;
-    const u32 texture_page_bottom = texture_page_top + TEXTURE_PAGE_HEIGHT;
-    const bool texture_page_overlaps =
-      (texture_page_left < m_drawing_area.right && texture_page_right > m_drawing_area.left &&
-       texture_page_top > m_drawing_area.bottom && texture_page_bottom < m_drawing_area.top);
-
-    // TODO: Check palette too.
-    if (texture_page_overlaps)
-    {
-      Log_DebugPrintf("Invalidating VRAM read cache due to drawing area overlap");
-      InvalidateVRAMReadCache();
-    }
-
-    m_batch.texture_page_x = m_render_state.texture_page_x;
-    m_batch.texture_page_y = m_render_state.texture_page_y;
-    m_batch.texture_palette_x = m_render_state.texture_palette_x;
-    m_batch.texture_palette_y = m_render_state.texture_palette_y;
-    m_render_state.ClearTexturePageChangedFlag();
-  }
-
   if (m_render_state.IsTextureWindowChanged())
   {
     m_batch.texture_window_values[0] = m_render_state.texture_window_mask_x;

src/core/gpu_hw.h

@@ -57,10 +57,6 @@ protected:
 
   struct HWBatchConfig
   {
-    u32 texture_page_x;
-    u32 texture_page_y;
-    u32 texture_palette_x;
-    u32 texture_palette_y;
     HWPrimitive primitive;
     TextureMode texture_mode;
     TransparencyMode transparency_mode;

src/core/gpu_sw.cpp

@@ -416,7 +416,7 @@ void GPU_SW::ShadePixel(RenderCommand rc, u32 x, u32 y, u8 color_r, u8 color_g,
                  ((m_render_state.texture_window_offset_y & m_render_state.texture_window_mask_y) * 8u);
 
     VRAMPixel texture_color;
-    switch (m_render_state.texture_color_mode)
+    switch (m_render_state.texture_mode)
     {
       case GPU::TextureMode::Palette4Bit:
       {