Merge pull request #12825 from iwubcode/texture_cache_sampler_split

VideoCommon: pull texture sampler out of texture cache for later reuse

Source/Core/VideoCommon/TextureCacheBase.cpp

@@ -1012,8 +1012,8 @@ static bool IsAnisostropicEnhancementSafe(const TexMode0& tm0)
   return !(tm0.min_filter == FilterMode::Near && tm0.mag_filter == FilterMode::Near);
 }
 
-static void SetSamplerState(u32 index, float custom_tex_scale, bool custom_tex,
+SamplerState TextureCacheBase::GetSamplerState(u32 index, float custom_tex_scale, bool custom_tex,
-                            bool has_arbitrary_mips)
+                                               bool has_arbitrary_mips)
 {
   const TexMode0& tm0 = bpmem.tex.GetUnit(index).texMode0;
 
@@ -1073,13 +1073,11 @@ static void SetSamplerState(u32 index, float custom_tex_scale, bool custom_tex,
     state.tm0.anisotropic_filtering = false;
   }
 
-  g_gfx->SetSamplerState(index, state);
+  return state;
-  auto& system = Core::System::GetInstance();
-  auto& pixel_shader_manager = system.GetPixelShaderManager();
-  pixel_shader_manager.SetSamplerState(index, state.tm0.hex, state.tm1.hex);
 }
 
-void TextureCacheBase::BindTextures(BitSet32 used_textures)
+void TextureCacheBase::BindTextures(BitSet32 used_textures,
+                                    const std::array<SamplerState, 8>& samplers)
 {
   auto& system = Core::System::GetInstance();
   auto& pixel_shader_manager = system.GetPixelShaderManager();
@@ -1091,8 +1089,9 @@ void TextureCacheBase::BindTextures(BitSet32 used_textures)
       g_gfx->SetTexture(i, tentry->texture.get());
       pixel_shader_manager.SetTexDims(i, tentry->native_width, tentry->native_height);
 
-      const float custom_tex_scale = tentry->GetWidth() / float(tentry->native_width);
+      auto& state = samplers[i];
-      SetSamplerState(i, custom_tex_scale, tentry->is_custom_tex, tentry->has_arbitrary_mips);
+      g_gfx->SetSamplerState(i, state);
+      pixel_shader_manager.SetSamplerState(i, state.tm0.hex, state.tm1.hex);
     }
   }
 

Source/Core/VideoCommon/TextureCacheBase.h

@@ -33,6 +33,7 @@
 class AbstractFramebuffer;
 class AbstractStagingTexture;
 class PointerWrap;
+struct SamplerState;
 struct VideoConfig;
 
 namespace VideoCommon
@@ -282,7 +283,7 @@ public:
   RcTcacheEntry GetXFBTexture(u32 address, u32 width, u32 height, u32 stride,
                               MathUtil::Rectangle<int>* display_rect);
 
-  virtual void BindTextures(BitSet32 used_textures);
+  virtual void BindTextures(BitSet32 used_textures, const std::array<SamplerState, 8>& samplers);
   void CopyRenderTargetToTexture(u32 dstAddr, EFBCopyFormat dstFormat, u32 width, u32 height,
                                  u32 dstStride, bool is_depth_copy,
                                  const MathUtil::Rectangle<int>& srcRect, bool isIntensity,
@@ -308,6 +309,10 @@ public:
   static bool AllCopyFilterCoefsNeeded(const std::array<u32, 3>& coefficients);
   static bool CopyFilterCanOverflow(const std::array<u32, 3>& coefficients);
 
+  // Get a new sampler state
+  static SamplerState GetSamplerState(u32 index, float custom_tex_scale, bool custom_tex,
+                                      bool has_arbitrary_mips);
+
 protected:
   // Decodes the specified data to the GPU texture specified by entry.
   // Returns false if the configuration is not supported.

Source/Core/VideoCommon/VertexManagerBase.cpp

@@ -563,13 +563,19 @@ void VertexManagerBase::Flush()
   const auto used_textures = UsedTextures();
   std::vector<std::string> texture_names;
   Common::SmallVector<u32, 8> texture_units;
+  std::array<SamplerState, 8> samplers;
   if (!m_cull_all)
   {
     if (!g_ActiveConfig.bGraphicMods)
     {
       for (const u32 i : used_textures)
       {
-        g_texture_cache->Load(TextureInfo::FromStage(i));
+        const auto cache_entry = g_texture_cache->Load(TextureInfo::FromStage(i));
+        if (!cache_entry)
+          continue;
+        const float custom_tex_scale = cache_entry->GetWidth() / float(cache_entry->native_width);
+        samplers[i] = TextureCacheBase::GetSamplerState(
+            i, custom_tex_scale, cache_entry->is_custom_tex, cache_entry->has_arbitrary_mips);
       }
     }
     else
@@ -585,6 +591,10 @@ void VertexManagerBase::Flush()
             texture_names.push_back(cache_entry->texture_info_name);
             texture_units.push_back(i);
           }
+
+          const float custom_tex_scale = cache_entry->GetWidth() / float(cache_entry->native_width);
+          samplers[i] = TextureCacheBase::GetSamplerState(
+              i, custom_tex_scale, cache_entry->is_custom_tex, cache_entry->has_arbitrary_mips);
         }
       }
     }
@@ -633,7 +643,7 @@ void VertexManagerBase::Flush()
     // Texture loading can cause palettes to be applied (-> uniforms -> draws).
     // Palette application does not use vertices, only a full-screen quad, so this is okay.
     // Same with GPU texture decoding, which uses compute shaders.
-    g_texture_cache->BindTextures(used_textures);
+    g_texture_cache->BindTextures(used_textures, samplers);
 
     if (PerfQueryBase::ShouldEmulate())
       g_perf_query->EnableQuery(bpmem.zcontrol.early_ztest ? PQG_ZCOMP_ZCOMPLOC : PQG_ZCOMP);