[drape] Faster RenderBucket

Signed-off-by: x7z4w <x7z4w@noreply.codeberg.org>

CMakeLists.txt

@@ -88,6 +88,11 @@ else()
   add_compile_options(-ffast-math $<$<CXX_COMPILER_ID:GNU>:-Wno-psabi>)
 endif()
 
+# Needed for std::execution::par
+if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+  add_compile_options(-fexperimental-library)
+endif()
+
 if (PLATFORM_WIN)
   add_definitions(
     -DWIN32_LEAN_AND_MEAN

libs/drape/batcher.cpp

@@ -1,11 +1,9 @@
 #include "drape/batcher.hpp"
 #include "drape/batcher_helpers.hpp"
-#include "drape/cpu_buffer.hpp"
 #include "drape/index_storage.hpp"
 #include "drape/vertex_array_buffer.hpp"
 
 #include "base/assert.hpp"
-#include "base/stl_helpers.hpp"
 
 #include <utility>
 
@@ -190,14 +188,6 @@ void Batcher::ResetSession()
   m_buckets.clear();
 }
 
-void Batcher::SetFeatureMinZoom(int minZoom)
-{
-  m_featureMinZoom = minZoom;
-
-  for (auto const & bucket : m_buckets)
-    bucket.second->SetFeatureMinZoom(m_featureMinZoom);
-}
-
 void Batcher::SetBatcherHash(uint64_t batcherHash)
 {
   m_batcherHash = batcherHash;

libs/drape/batcher.hpp

@@ -10,6 +10,7 @@
 
 #include "base/macros.hpp"
 
+#include <execution>
 #include <functional>
 #include <map>
 #include <vector>
@@ -68,7 +69,13 @@ public:
 
   void SetBatcherHash(uint64_t batcherHash);
 
-  void SetFeatureMinZoom(int minZoom);
+  inline void SetFeatureMinZoom(int minZoom)
+  {
+    m_featureMinZoom = minZoom;
+
+    std::for_each(std::execution::par_unseq, m_buckets.begin(), m_buckets.end(),
+                  [this](auto const & bucket) { bucket.second->SetFeatureMinZoom(m_featureMinZoom); });
+  }
 
 private:
   template <typename TBatcher, typename... TArgs>

libs/drape/font_texture.cpp

@@ -5,8 +5,6 @@
 #include "base/logging.hpp"
 
 #include <algorithm>
-#include <functional>
-#include <iterator>
 
 #include "font_constants.hpp"
 

libs/drape/gl_gpu_program.cpp

@@ -1,7 +1,6 @@
 #include "drape/gl_gpu_program.hpp"
 #include "drape/gl_functions.hpp"
 #include "drape/render_state.hpp"
-#include "drape/support_manager.hpp"
 
 #include "base/logging.hpp"
 

libs/drape/gl_gpu_program.hpp

@@ -5,9 +5,8 @@
 #include "drape/pointers.hpp"
 #include "drape/shader.hpp"
 
-#include <map>
+#include <unordered_map>
 #include <string>
-#include <vector>
 
 namespace dp
 {
@@ -30,7 +29,7 @@ public:
     glConst m_type = gl_const::GLFloatType;
   };
 
-  using UniformsInfo = std::map<std::string, UniformInfo>;
+  using UniformsInfo = std::unordered_map<std::string, UniformInfo>;
   UniformsInfo const & GetUniformsInfo() const;
   uint32_t GetNumericUniformsCount() const { return m_numericUniformsCount; }
 

libs/drape/glyph_manager.cpp

@@ -16,6 +16,8 @@
 #include "base/stl_helpers.hpp"
 #include "base/string_utils.hpp"
 
+#include <execution>
+
 #include <ft2build.h>
 #include <hb-ft.h>
 #include <unicode/unistr.h>
@@ -203,9 +205,9 @@ FreetypeError constexpr g_FT_Errors[] =
 
     static void Close(FT_Stream) {}
 
-    void MarkGlyphReady(uint16_t glyphId) { m_readyGlyphs.emplace(glyphId); }
+    inline void MarkGlyphReady(uint16_t glyphId) { m_readyGlyphs.emplace(glyphId); }
 
-    bool IsGlyphReady(uint16_t glyphId) const { return m_readyGlyphs.find(glyphId) != m_readyGlyphs.end(); }
+    inline bool IsGlyphReady(uint16_t glyphId) const { return m_readyGlyphs.find(glyphId) != m_readyGlyphs.end(); }
 
     std::string GetName() const { return std::string(m_fontFace->family_name) + ':' + m_fontFace->style_name; }
 
@@ -229,7 +231,9 @@ FreetypeError constexpr g_FT_Errors[] =
       hb_glyph_info_t const * glyphInfo = hb_buffer_get_glyph_infos(hbBuffer, &glyphCount);
       hb_glyph_position_t const * glyphPos = hb_buffer_get_glyph_positions(hbBuffer, &glyphCount);
 
-      for (unsigned int i = 0; i < glyphCount; ++i)
+      auto const range = std::ranges::views::iota(0u, glyphCount);
+
+      std::for_each(std::execution::par_unseq, range.begin(), range.end(), [&, this](auto const i)
       {
         // TODO(AB): Check for missing glyph ID?
         auto const glyphId = static_cast<uint16_t>(glyphInfo[i].codepoint);
@@ -247,7 +251,7 @@ FreetypeError constexpr g_FT_Errors[] =
         // yAdvance is always zero for horizontal text layouts.
 
         outMetrics.AddGlyphMetrics(static_cast<int16_t>(fontIndex), glyphId, xOffset, yOffset, xAdvance, fontPixelSize);
-      }
+      });
     }
 
   private:
@@ -255,7 +259,7 @@ FreetypeError constexpr g_FT_Errors[] =
     FT_StreamRec_ m_stream;
     FT_Face m_fontFace;
 
-    std::set<uint16_t> m_readyGlyphs;
+    std::unordered_set<uint16_t> m_readyGlyphs;
 
     hb_font_t * m_harfbuzzFont{nullptr};
   };
@@ -332,7 +336,6 @@ FreetypeError constexpr g_FT_Errors[] =
       using is_transparent = void;
     };
 
-    // TODO(AB): Compare performance with std::map.
     std::unordered_map<std::string, text::TextMetrics, StringHash, std::equal_to<>> m_textMetricsCache;
     hb_buffer_t * m_harfbuzzBuffer;
   };
@@ -598,7 +601,7 @@ FreetypeError constexpr g_FT_Errors[] =
 
     allGlyphs.m_glyphs.reserve(strings::CountChar(utf8));
 
-    for (auto const & substring : segments)
+    std::for_each(std::execution::par_unseq, segments.begin(), segments.end(), [&, this](auto const & substring)
     {
       hb_buffer_clear_contents(m_impl->m_harfbuzzBuffer);
 
@@ -626,7 +629,7 @@ FreetypeError constexpr g_FT_Errors[] =
         }
       }
       while (u32CharacterIter != end);
-    }
+    });
 
     // Uncomment utf8 printing for debugging if necessary. It crashes JNI with non-modified UTF-8 strings on Android 5
     // and 6. See https://github.com/organicmaps/organicmaps/issues/10685

libs/drape/render_bucket.cpp

@@ -6,7 +6,7 @@
 #include "drape/overlay_tree.hpp"
 #include "drape/vertex_array_buffer.hpp"
 
-#include "base/stl_helpers.hpp"
+#include <execution>
 
 namespace dp
 {
@@ -51,23 +51,25 @@ void RenderBucket::AddOverlayHandle(drape_ptr<OverlayHandle> && handle)
 
 void RenderBucket::BeforeUpdate()
 {
-  for (auto & overlayHandle : m_overlay)
+  std::for_each(std::execution::par_unseq, m_overlay.begin(), m_overlay.end(),
-    overlayHandle->BeforeUpdate();
+                [](auto & overlayHandle) { overlayHandle->BeforeUpdate(); });
 }
 
 void RenderBucket::Update(ScreenBase const & modelView)
 {
   BeforeUpdate();
-  for (auto & overlayHandle : m_overlay)
+  std::for_each(std::execution::par_unseq, m_overlay.begin(), m_overlay.end(), [&modelView](auto & overlayHandle)
+  {
     if (overlayHandle->IsVisible())
       overlayHandle->Update(modelView);
+  });
 }
 
 void RenderBucket::CollectOverlayHandles(ref_ptr<OverlayTree> tree)
 {
   BeforeUpdate();
-  for (auto const & overlayHandle : m_overlay)
+  std::for_each(std::execution::par_unseq, m_overlay.begin(), m_overlay.end(),
-    tree->Add(make_ref(overlayHandle));
+                [&tree](auto & overlayHandle) { tree->Add(make_ref(overlayHandle)); });
 }
 
 bool RenderBucket::HasOverlayHandles() const
@@ -102,7 +104,8 @@ void RenderBucket::Render(ref_ptr<GraphicsContext> context, bool drawAsLine)
     ref_ptr<AttributeBufferMutator> rfpAttrib = make_ref(&attributeMutator);
 
     bool hasIndexMutation = false;
-    for (drape_ptr<OverlayHandle> const & handle : m_overlay)
+    std::for_each(std::execution::par_unseq, m_overlay.begin(), m_overlay.end(),
+                  [&, this](drape_ptr<OverlayHandle> const & handle)
     {
       if (handle->IndexesRequired())
       {
@@ -113,7 +116,7 @@ void RenderBucket::Render(ref_ptr<GraphicsContext> context, bool drawAsLine)
 
       if (handle->HasDynamicAttributes())
         handle->GetAttributeMutation(rfpAttrib);
-    }
+    });
 
     m_buffer->ApplyMutation(context, hasIndexMutation ? rfpIndex : nullptr, rfpAttrib);
   }
@@ -129,7 +132,7 @@ void RenderBucket::SetFeatureMinZoom(int minZoom)
 void RenderBucket::RenderDebug(ref_ptr<GraphicsContext> context, ScreenBase const & screen,
                                ref_ptr<DebugRenderer> debugRectRenderer) const
 {
-  if (!debugRectRenderer || !debugRectRenderer->IsEnabled() || m_overlay.empty())
+  if (m_overlay.empty())
     return;
 
   for (auto const & handle : m_overlay)

libs/drape/symbols_texture.cpp

@@ -264,39 +264,4 @@ void SymbolsTexture::Fail(ref_ptr<dp::GraphicsContext> context)
 
   Create(context, p, make_ref(&alphaTexture));
 }
-
-bool SymbolsTexture::IsSymbolContained(std::string const & symbolName) const
-{
-  return m_definition.find(symbolName) != m_definition.end();
-}
-
-bool SymbolsTexture::DecodeToMemory(std::string const & skinPathName, std::string const & textureName,
-                                    std::vector<uint8_t> & symbolsSkin, std::map<std::string, m2::RectU> & symbolsIndex,
-                                    uint32_t & skinWidth, uint32_t & skinHeight)
-{
-  auto definitionInserter = [&symbolsIndex](std::string const & name, m2::RectF const & rect)
-  { symbolsIndex.insert(make_pair(name, m2::RectU(rect))); };
-
-  bool result = true;
-  auto completionHandler =
-      [&result, &symbolsSkin, &skinWidth, &skinHeight](unsigned char * data, uint32_t width, uint32_t height)
-  {
-    size_t size = 4 * width * height;
-    symbolsSkin.resize(size);
-    memcpy(symbolsSkin.data(), data, size);
-    skinWidth = width;
-    skinHeight = height;
-    result = true;
-  };
-
-  auto failureHandler = [&result](std::string const & reason)
-  {
-    LOG(LERROR, (reason));
-    result = false;
-  };
-
-  LoadSymbols(skinPathName, textureName, false /* convertToUV */, definitionInserter, completionHandler,
-              failureHandler);
-  return result;
-}
 }  // namespace dp

libs/drape/symbols_texture.hpp

@@ -2,7 +2,7 @@
 
 #include "drape/texture.hpp"
 
-#include <map>
+#include <unordered_map>
 #include <string>
 #include <vector>
 
@@ -39,19 +39,17 @@ public:
   void Invalidate(ref_ptr<dp::GraphicsContext> context, std::string const & skinPathName,
                   ref_ptr<HWTextureAllocator> allocator, std::vector<drape_ptr<HWTexture>> & internalTextures);
 
-  bool IsSymbolContained(std::string const & symbolName) const;
+  inline bool IsSymbolContained(std::string const & symbolName) const
-
+  {
-  static bool DecodeToMemory(std::string const & skinPathName, std::string const & textureName,
+    return m_definition.find(symbolName) != m_definition.end();
-                             std::vector<uint8_t> & symbolsSkin, std::map<std::string, m2::RectU> & symbolsIndex,
+  }
-                             uint32_t & skinWidth, uint32_t & skinHeight);
 
 private:
   void Fail(ref_ptr<dp::GraphicsContext> context);
   void Load(ref_ptr<dp::GraphicsContext> context, std::string const & skinPathName,
             ref_ptr<HWTextureAllocator> allocator);
 
-  using TSymDefinition = std::map<std::string, SymbolInfo>;
   std::string m_name;
-  mutable TSymDefinition m_definition;
+  mutable std::unordered_map<std::string, SymbolInfo> m_definition;
 };
 }  // namespace dp

libs/drape/texture_manager.cpp

@@ -539,12 +539,6 @@ GlyphFontAndId TextureManager::GetSpaceGlyph() const
   return m_spaceGlyph;
 }
 
-bool TextureManager::AreGlyphsReady(TGlyphs const & glyphs) const
-{
-  CHECK(m_isInitialized, ());
-  return m_glyphManager->AreGlyphsReady(glyphs);
-}
-
 ref_ptr<Texture> TextureManager::GetSymbolsTexture() const
 {
   CHECK(m_isInitialized, ());

libs/drape/texture_manager.hpp

@@ -95,7 +95,11 @@ public:
                                       TMultilineGlyphsBuffer & multilineGlyphRegions);
 
   // This method must be called only on Frontend renderer's thread.
-  bool AreGlyphsReady(TGlyphs const & glyphs) const;
+  inline bool AreGlyphsReady(TGlyphs const & glyphs) const
+  {
+    CHECK(m_isInitialized, ());
+    return m_glyphManager->AreGlyphsReady(glyphs);
+  }
 
   GlyphFontAndId GetSpaceGlyph() const;
 

libs/drape/vulkan/vulkan_vertex_array_buffer_impl.cpp

@@ -3,16 +3,15 @@
 #include "drape/vulkan/vulkan_base_context.hpp"
 #include "drape/vulkan/vulkan_gpu_buffer_impl.hpp"
 #include "drape/vulkan/vulkan_param_descriptor.hpp"
-#include "drape/vulkan/vulkan_utils.hpp"
 
 #include "base/assert.hpp"
 #include "base/macros.hpp"
+#include "drape/vulkan/vulkan_staging_buffer.hpp"
 
 #include <array>
 #include <cstdint>
 #include <cstring>
 #include <utility>
-#include <vector>
 
 namespace dp
 {
@@ -44,11 +43,11 @@ public:
 
   void RenderRange(ref_ptr<GraphicsContext> context, bool drawAsLine, IndicesRange const & range) override
   {
-    CHECK(m_vertexArrayBuffer->HasBuffers(), ());
+    ASSERT(m_vertexArrayBuffer->HasBuffers(), ());
 
     ref_ptr<dp::vulkan::VulkanBaseContext> vulkanContext = context;
     VkCommandBuffer commandBuffer = vulkanContext->GetCurrentRenderingCommandBuffer();
-    CHECK(commandBuffer != nullptr, ());
+    ASSERT(commandBuffer != nullptr, ());
 
     vulkanContext->SetPrimitiveTopology(drawAsLine ? VK_PRIMITIVE_TOPOLOGY_LINE_LIST
                                                    : VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST);
@@ -71,23 +70,22 @@ public:
     for (auto & buffer : m_vertexArrayBuffer->m_staticBuffers)
     {
       ref_ptr<VulkanGpuBufferImpl> b = buffer.second->GetBuffer();
-      CHECK_LESS(bufferIndex, kMaxBuffersCount, ());
+      ASSERT_LESS(bufferIndex, kMaxBuffersCount, ());
       buffers[bufferIndex++] = b->GetVulkanBuffer();
     }
     for (auto & buffer : m_vertexArrayBuffer->m_dynamicBuffers)
     {
       ref_ptr<VulkanGpuBufferImpl> b = buffer.second->GetBuffer();
-      CHECK_LESS(bufferIndex, kMaxBuffersCount, ());
+      ASSERT_LESS(bufferIndex, kMaxBuffersCount, ());
       buffers[bufferIndex++] = b->GetVulkanBuffer();
     }
     vkCmdBindVertexBuffers(commandBuffer, 0, bufferIndex, buffers.data(), offsets.data());
 
-    ref_ptr<VulkanGpuBufferImpl> ib = m_vertexArrayBuffer->m_indexBuffer->GetBuffer();
+    vkCmdBindIndexBuffer(
-    VkBuffer vulkanIndexBuffer = ib->GetVulkanBuffer();
+        commandBuffer, ref_ptr<VulkanGpuBufferImpl>(m_vertexArrayBuffer->m_indexBuffer->GetBuffer())->GetVulkanBuffer(),
-    auto const indexType = dp::IndexStorage::IsSupported32bit() ? VK_INDEX_TYPE_UINT32 : VK_INDEX_TYPE_UINT16;
+        0, m_indexType);
-    vkCmdBindIndexBuffer(commandBuffer, vulkanIndexBuffer, 0, indexType);
 
-    CHECK_LESS_OR_EQUAL(range.m_idxStart + range.m_idxCount,
+    ASSERT_LESS_OR_EQUAL(range.m_idxStart + range.m_idxCount,
                          m_objectManager->GetMemoryManager().GetDeviceLimits().maxDrawIndexedIndexValue, ());
 
     vkCmdDrawIndexed(commandBuffer, range.m_idxCount, 1, range.m_idxStart, 0, 0);
@@ -99,6 +97,7 @@ private:
   BindingInfoArray m_bindingInfo;
   uint8_t m_bindingInfoCount = 0;
   ParamDescriptorUpdater m_descriptorUpdater;
+  VkIndexType const m_indexType = dp::IndexStorage::IsSupported32bit() ? VK_INDEX_TYPE_UINT32 : VK_INDEX_TYPE_UINT16;
 };
 }  // namespace vulkan
 

libs/drape_frontend/frontend_renderer.cpp

@@ -37,14 +37,8 @@
 #include "std/target_os.hpp"
 
 #include <algorithm>
-#include <array>
-#include <chrono>
 #include <cmath>
-#include <functional>
+#include <execution>
-#include <limits>
-#include <memory>
-#include <thread>
-#include <utility>
 
 namespace df
 {
@@ -1813,14 +1807,17 @@ void FrontendRenderer::BuildOverlayTree(ScreenBase const & modelView)
     return;
 
   BeginUpdateOverlayTree(modelView);
-  for (auto const layerId :
+  constexpr std::array<uint8_t, 3> layers = {static_cast<uint8_t>(DepthLayer::OverlayLayer),
-       {DepthLayer::OverlayLayer, DepthLayer::RoutingBottomMarkLayer, DepthLayer::RoutingMarkLayer})
+                                             static_cast<uint8_t>(DepthLayer::RoutingBottomMarkLayer),
+                                             static_cast<uint8_t>(DepthLayer::RoutingMarkLayer)};
+
+  std::for_each(std::execution::par_unseq, layers.begin(), layers.end(), [this, &modelView](uint8_t const layerId)
   {
-    RenderLayer & overlay = m_layers[static_cast<size_t>(layerId)];
+    RenderLayer & overlay = m_layers[layerId];
     overlay.Sort(make_ref(m_overlayTree));
-    for (auto & group : overlay.m_renderGroups)
+    std::for_each(std::execution::par_unseq, overlay.m_renderGroups.begin(), overlay.m_renderGroups.end(),
-      UpdateOverlayTree(modelView, group);
+                  [this, &modelView](auto & group) { UpdateOverlayTree(modelView, group); });
-  }
+  });
   if (m_transitSchemeRenderer->IsSchemeVisible(GetCurrentZoom()) && !HasTransitRouteData())
     m_transitSchemeRenderer->CollectOverlays(make_ref(m_overlayTree), modelView);
   EndUpdateOverlayTree();

libs/drape_frontend/path_text_handle.cpp

@@ -122,11 +122,6 @@ ref_ptr<PathTextLayout> const PathTextContext::GetLayout() const
   return make_ref(m_layout);
 }
 
-void PathTextContext::BeforeUpdate()
-{
-  m_updated = false;
-}
-
 std::vector<double> const & PathTextContext::GetOffsets() const
 {
   return m_globalOffsets;

libs/drape_frontend/path_text_handle.hpp

@@ -21,7 +21,8 @@ public:
 
   bool GetPivot(size_t textIndex, m2::PointD & pivot, m2::Spline::iterator & centerPointIter) const;
 
-  void BeforeUpdate();
+  // TODO: these are unused?
+  inline void BeforeUpdate() { m_updated = false; }
   void Update(ScreenBase const & screen);
 
   std::vector<double> const & GetOffsets() const;

libs/drape_frontend/render_group.cpp

@@ -8,9 +8,8 @@
 
 #include "geometry/screenbase.hpp"
 
-#include "base/stl_helpers.hpp"
-
 #include <algorithm>
+#include <execution>
 #include <sstream>
 #include <utility>
 
@@ -43,8 +42,8 @@ void RenderGroup::CollectOverlay(ref_ptr<dp::OverlayTree> tree)
   if (CanBeDeleted())
     return;
 
-  for (auto & renderBucket : m_renderBuckets)
+  std::for_each(std::execution::par_unseq, m_renderBuckets.begin(), m_renderBuckets.end(),
-    renderBucket->CollectOverlayHandles(tree);
+                [&](auto & renderBucket) { renderBucket->CollectOverlayHandles(tree); });
 }
 
 bool RenderGroup::HasOverlayHandles() const
@@ -78,8 +77,9 @@ void RenderGroup::Render(ref_ptr<dp::GraphicsContext> context, ref_ptr<gpu::Prog
   programPtr->Bind();
   dp::ApplyState(context, programPtr, m_state);
 
-  for (auto & renderBucket : m_renderBuckets)
+  std::for_each(std::execution::par_unseq, m_renderBuckets.begin(), m_renderBuckets.end(),
-    renderBucket->GetBuffer()->Build(context, programPtr);
+                [&context, &programPtr](auto const & renderBucket)
+  { renderBucket->GetBuffer()->Build(context, programPtr); });
 
   auto const program = m_state.GetProgram<gpu::Program>();
   auto const program3d = m_state.GetProgram3d<gpu::Program>();
@@ -100,8 +100,9 @@ void RenderGroup::Render(ref_ptr<dp::GraphicsContext> context, ref_ptr<gpu::Prog
     m_params.m_isOutlinePass = 1.0f;
 
     mng->GetParamsSetter()->Apply(context, programPtr, m_params);
-    for (auto & renderBucket : m_renderBuckets)
+    std::for_each(std::execution::par_unseq, m_renderBuckets.begin(), m_renderBuckets.end(),
-      renderBucket->Render(context, m_state.GetDrawAsLine());
+                  [&context, this](auto const & renderBucket)
+    { renderBucket->Render(context, m_state.GetDrawAsLine()); });
 
     m_params.m_contrastGamma = glsl::vec2(glyphParams.m_contrast, glyphParams.m_gamma);
     m_params.m_isOutlinePass = 0.0f;
@@ -112,11 +113,14 @@ void RenderGroup::Render(ref_ptr<dp::GraphicsContext> context, ref_ptr<gpu::Prog
   }
 
   mng->GetParamsSetter()->Apply(context, programPtr, m_params);
-  for (auto & renderBucket : m_renderBuckets)
+  std::for_each(std::execution::par_unseq, m_renderBuckets.begin(), m_renderBuckets.end(),
-    renderBucket->Render(context, m_state.GetDrawAsLine());
+                [&context, this](auto const & renderBucket)
+  { renderBucket->Render(context, m_state.GetDrawAsLine()); });
 
-  for (auto const & renderBucket : m_renderBuckets)
+  if (debugRectRenderer && debugRectRenderer->IsEnabled())
-    renderBucket->RenderDebug(context, screen, debugRectRenderer);
+    std::for_each(std::execution::par_unseq, m_renderBuckets.begin(), m_renderBuckets.end(),
+                  [&context, &screen, &debugRectRenderer](auto const & renderBucket)
+    { renderBucket->RenderDebug(context, screen, debugRectRenderer); });
 }
 
 void RenderGroup::AddBucket(drape_ptr<dp::RenderBucket> && bucket)