texture_cache: Flush 3D textures in the order they are drawn

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -235,8 +235,9 @@ void SwizzleFunc(MortonSwizzleMode mode, u8* memory, const SurfaceParams& params
 
 } // Anonymous namespace
 
-CachedSurface::CachedSurface(const SurfaceParams& params)
+CachedSurface::CachedSurface(TextureCacheOpenGL& texture_cache, const SurfaceParams& params)
-    : VideoCommon::SurfaceBaseContextless<CachedSurfaceView>{params} {
+    : VideoCommon::SurfaceBaseContextless<TextureCacheOpenGL, CachedSurfaceView>{texture_cache,
+                                                                                 params} {
     const auto& tuple{GetFormatTuple(params.GetPixelFormat(), params.GetComponentType())};
     internal_format = tuple.internal_format;
     format = tuple.format;

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -26,15 +26,17 @@ using VideoCore::Surface::SurfaceType;
 
 class CachedSurfaceView;
 class CachedSurface;
+class TextureCacheOpenGL;
 
 using Surface = std::shared_ptr<CachedSurface>;
 using TextureCacheBase = VideoCommon::TextureCacheContextless<CachedSurface, CachedSurfaceView>;
 
-class CachedSurface final : public VideoCommon::SurfaceBaseContextless<CachedSurfaceView> {
+class CachedSurface final
+    : public VideoCommon::SurfaceBaseContextless<TextureCacheOpenGL, CachedSurfaceView> {
     friend CachedSurfaceView;
 
 public:
-    explicit CachedSurface(const SurfaceParams& params);
+    explicit CachedSurface(TextureCacheOpenGL& texture_cache, const SurfaceParams& params);
     ~CachedSurface();
 
     void LoadBuffer();

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -460,7 +460,6 @@ static void APIENTRY DebugHandler(GLenum source, GLenum type, GLuint id, GLenum
     switch (severity) {
     case GL_DEBUG_SEVERITY_HIGH:
         LOG_CRITICAL(Render_OpenGL, format, str_source, str_type, id, message);
-        __debugbreak();
         break;
     case GL_DEBUG_SEVERITY_MEDIUM:
         LOG_WARNING(Render_OpenGL, format, str_source, str_type, id, message);

src/video_core/texture_cache.cpp

@@ -154,8 +154,8 @@ bool SurfaceParams::IsLayered() const {
     switch (target) {
     case SurfaceTarget::Texture1DArray:
     case SurfaceTarget::Texture2DArray:
-    case SurfaceTarget::TextureCubeArray:
     case SurfaceTarget::TextureCubemap:
+    case SurfaceTarget::TextureCubeArray:
         return true;
     default:
         return false;
@@ -192,9 +192,11 @@ u32 SurfaceParams::GetMipBlockDepth(u32 level) const {
     while (block_depth > 1 && depth * 2 <= block_depth) {
         block_depth >>= 1;
     }
+
     if (block_depth == 32 && GetMipBlockHeight(level) >= 4) {
         return 16;
     }
+
     return block_depth;
 }
 
@@ -227,7 +229,7 @@ std::size_t SurfaceParams::GetLayerSize(bool as_host_size, bool uncompressed) co
     for (u32 level = 0; level < num_levels; ++level) {
         size += GetInnerMipmapMemorySize(level, as_host_size, uncompressed);
     }
-    if (is_tiled && IsLayered()) {
+    if (is_tiled && (IsLayered() || target == SurfaceTarget::Texture3D)) {
         return Common::AlignUp(size, Tegra::Texture::GetGOBSize() * block_height * block_depth);
     }
     return size;
@@ -312,9 +314,10 @@ void SurfaceParams::CalculateCachedValues() {
 
     guest_size_in_bytes = GetInnerMemorySize(false, false, false);
 
-    // ASTC is uncompressed in software, in emulated as RGBA8
     if (IsPixelFormatASTC(pixel_format)) {
-        host_size_in_bytes = static_cast<std::size_t>(width * height * depth * 4U);
+        // ASTC is uncompressed in software, in emulated as RGBA8
+        host_size_in_bytes = static_cast<std::size_t>(width) * static_cast<std::size_t>(height) *
+                             static_cast<std::size_t>(depth) * 4ULL;
     } else {
         host_size_in_bytes = GetInnerMemorySize(true, false, false);
     }

src/video_core/texture_cache.h

@@ -273,7 +273,7 @@ struct hash<VideoCommon::ViewKey> {
 
 namespace VideoCommon {
 
-template <typename TView, typename TExecutionContext>
+template <typename TTextureCache, typename TView, typename TExecutionContext>
 class SurfaceBase {
     static_assert(std::is_trivially_copyable_v<TExecutionContext>);
 
@@ -331,6 +331,9 @@ public:
 
     void MarkAsModified(bool is_modified_) {
         is_modified = is_modified_;
+        if (is_modified_) {
+            modification_tick = texture_cache.Tick();
+        }
     }
 
     const SurfaceParams& GetSurfaceParams() const {
@@ -358,13 +361,18 @@ public:
         is_registered = false;
     }
 
+    u64 GetModificationTick() const {
+        return modification_tick;
+    }
+
     bool IsRegistered() const {
         return is_registered;
     }
 
 protected:
-    explicit SurfaceBase(const SurfaceParams& params)
+    explicit SurfaceBase(TTextureCache& texture_cache, const SurfaceParams& params)
-        : params{params}, view_offset_map{params.CreateViewOffsetMap()} {}
+        : params{params}, texture_cache{texture_cache}, view_offset_map{
+                                                            params.CreateViewOffsetMap()} {}
 
     ~SurfaceBase() = default;
 
@@ -389,12 +397,14 @@ private:
         return view.get();
     }
 
+    TTextureCache& texture_cache;
     const std::map<u64, std::pair<u32, u32>> view_offset_map;
 
     GPUVAddr gpu_addr{};
     VAddr cpu_addr{};
     u8* host_ptr{};
     CacheAddr cache_addr{};
+    u64 modification_tick{};
     bool is_modified{};
     bool is_registered{};
     std::unordered_map<ViewKey, std::unique_ptr<TView>> views;
@@ -475,6 +485,10 @@ public:
         return it != registered_surfaces.end() ? *it->second.begin() : nullptr;
     }
 
+    u64 Tick() {
+        return ++ticks;
+    }
+
 protected:
     TextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
         : system{system}, rasterizer{rasterizer} {}
@@ -521,7 +535,7 @@ private:
 
         const auto host_ptr{memory_manager.GetPointer(gpu_addr)};
         const auto cache_addr{ToCacheAddr(host_ptr)};
-        const auto overlaps{GetSurfacesInRegion(cache_addr, params.GetGuestSizeInBytes())};
+        auto overlaps{GetSurfacesInRegion(cache_addr, params.GetGuestSizeInBytes())};
         if (overlaps.empty()) {
             return LoadSurfaceView(exctx, gpu_addr, *cpu_addr, host_ptr, params, preserve_contents);
         }
@@ -544,8 +558,8 @@ private:
 
         for (const auto& surface : overlaps) {
             if (!fast_view) {
-                // Flush even when we don't care about the contents, to preserve memory not written
+                // Flush even when we don't care about the contents, to preserve memory not
-                // by the new surface.
+                // written by the new surface.
                 exctx = surface->FlushBuffer(exctx);
             }
             Unregister(surface);
@@ -614,6 +628,8 @@ private:
 
     VideoCore::RasterizerInterface& rasterizer;
 
+    u64 ticks{};
+
     IntervalMap registered_surfaces;
 
     /// The surface reserve is a "backup" cache, this is where we put unique surfaces that have
@@ -653,6 +669,10 @@ public:
         return Base::TryFindFramebufferSurface(host_ptr);
     }
 
+    u64 Tick() {
+        return Base::Tick();
+    }
+
 protected:
     explicit TextureCacheContextless(Core::System& system,
                                      VideoCore::RasterizerInterface& rasterizer)
@@ -678,8 +698,8 @@ private:
     }
 };
 
-template <typename TView>
+template <typename TTextureCache, typename TView>
-class SurfaceBaseContextless : public SurfaceBase<TView, DummyExecutionContext> {
+class SurfaceBaseContextless : public SurfaceBase<TTextureCache, TView, DummyExecutionContext> {
 public:
     DummyExecutionContext FlushBuffer(DummyExecutionContext) {
         FlushBufferImpl();
@@ -692,8 +712,8 @@ public:
     }
 
 protected:
-    explicit SurfaceBaseContextless(const SurfaceParams& params)
+    explicit SurfaceBaseContextless(TTextureCache& texture_cache, const SurfaceParams& params)
-        : SurfaceBase<TView, DummyExecutionContext>{params} {}
+        : SurfaceBase<TTextureCache, TView, DummyExecutionContext>{texture_cache, params} {}
 
     virtual void FlushBufferImpl() = 0;