Merge pull request #6 from ZephyrEngine/unified-geometry-buffers

Unified Geometry Buffers

app/next/src/main_window.cpp

@@ -153,6 +153,9 @@ namespace zephyr {
     gltf_scene_1->GetTransform().GetRotation().SetFromEuler(1.5f, 0.0f, 0.0f);
     m_scene_root->Add(std::move(gltf_scene_1));
 
+    m_scene_root->Add(gltf_loader.Parse("models/triangleWithoutIndices/TriangleWithoutIndices.gltf"));
+    //m_scene_root->Add(gltf_loader.Parse("models/triangle/Triangle.gltf"));
+
     m_behemoth_scene = gltf_loader.Parse("models/Behemoth/scene.gltf");
     m_behemoth_scene->GetTransform().GetPosition() = Vector3{-1.0f, 0.0f, -5.0f};
     m_behemoth_scene->GetTransform().GetRotation().SetFromEuler(-M_PI * 0.5, M_PI, 0.0f);

zephyr/renderer/CMakeLists.txt

@@ -3,6 +3,7 @@
 # TODO: remove dependency on SDL2 if possible
 
 set(SOURCES
+  src/backend/opengl/dynamic_gpu_array.cpp
   src/backend/opengl/render_backend.cpp
   src/backend/opengl/render_geometry.cpp
   src/backend/vulkan/render_backend.cpp
@@ -12,8 +13,9 @@ set(SOURCES
 )
 
 set(HEADERS
-  src/backend/opengl/render_geometry.hpp
+  src/backend/opengl/dynamic_gpu_array.hpp
   src/backend/opengl/render_backend.hpp
+  src/backend/opengl/render_geometry.hpp
   src/backend/vulkan/shader/triangle.frag.h
   src/backend/vulkan/shader/triangle.vert.h
 )

zephyr/renderer/include/zephyr/renderer/backend/render_backend.hpp

@@ -57,7 +57,7 @@ namespace zephyr {
       virtual RenderGeometry* CreateRenderGeometry(RenderGeometryLayout layout, size_t number_of_vertices, size_t number_of_indices) = 0;
       virtual void UpdateRenderGeometryIndices(RenderGeometry* render_geometry, std::span<const u8> data) = 0;
       virtual void UpdateRenderGeometryVertices(RenderGeometry* render_geometry, std::span<const u8> data) = 0;
-      virtual void DestroyRenderGeometry(RenderGeometry* geometry) = 0;
+      virtual void DestroyRenderGeometry(RenderGeometry* render_geometry) = 0;
 
       /// Just a quick thing for testing the rendering.
       virtual void Render(const Matrix4& view_projection, std::span<const RenderObject> render_objects) = 0;

zephyr/renderer/src/backend/opengl/dynamic_gpu_array.cpp

@@ -0,0 +1,170 @@
+
+#include <zephyr/float.hpp>
+#include <zephyr/panic.hpp>
+#include <algorithm>
+
+#include <fmt/format.h>
+
+#include "dynamic_gpu_array.hpp"
+
+namespace zephyr {
+
+  OpenGLDynamicGPUArray::OpenGLDynamicGPUArray(size_t byte_stride) : m_byte_stride{byte_stride} {
+    ResizeBuffer(k_capacity_increment);
+
+//    const auto PrintAlloc = [](const BufferRange& range) {
+//      fmt::print("allocated: base_element={} \tnumber_of_elements={}\n", range.base_element, range.number_of_elements);
+//    };
+//
+//    const auto range_a = AllocateRange(512);
+//    PrintAlloc(range_a);
+//    PrintFreeRanges();
+//
+//    const auto range_b = AllocateRange(16385);
+//    PrintAlloc(range_b);
+//    PrintFreeRanges();
+//
+//    const auto range_c = AllocateRange(100000);
+//    PrintAlloc(range_c);
+//    PrintFreeRanges();
+//
+//    ReleaseRange(range_b);
+//    PrintFreeRanges();
+//
+//    ReleaseRange(range_c);
+//    PrintFreeRanges();
+//
+//    ReleaseRange(range_a);
+//    PrintFreeRanges();
+  }
+
+  OpenGLDynamicGPUArray::~OpenGLDynamicGPUArray() {
+    glDeleteBuffers(1u, &m_gpu_buffer);
+  }
+
+  OpenGLDynamicGPUArray::BufferRange OpenGLDynamicGPUArray::AllocateRange(size_t number_of_elements) {
+    const auto end = m_free_buffer_ranges.end();
+
+    for(auto it = m_free_buffer_ranges.begin(); it != end; ++it) {
+      BufferRange& free_buffer_range = *it;
+
+      if(free_buffer_range.number_of_elements >= number_of_elements) {
+        const BufferRange allocated_range{free_buffer_range.base_element, number_of_elements};
+
+        if(free_buffer_range.number_of_elements == number_of_elements) {
+          m_free_buffer_ranges.erase(it);
+        } else {
+          free_buffer_range.base_element += number_of_elements;
+          free_buffer_range.number_of_elements -= number_of_elements;
+        }
+
+        return allocated_range;
+      }
+    }
+
+    const size_t required_capacity = m_current_capacity + number_of_elements - m_free_buffer_ranges.back().number_of_elements;
+    const size_t  rounded_capacity = (required_capacity + k_capacity_increment - 1u) / k_capacity_increment * k_capacity_increment;
+    ResizeBuffer(rounded_capacity);
+
+    BufferRange& free_buffer_range = m_free_buffer_ranges.back();
+    const BufferRange allocated_range{free_buffer_range.base_element, number_of_elements};
+
+    if(free_buffer_range.number_of_elements == number_of_elements) {
+      m_free_buffer_ranges.pop_back();
+    } else {
+      free_buffer_range.base_element += number_of_elements;
+      free_buffer_range.number_of_elements -= number_of_elements;
+    }
+
+    return allocated_range;
+  }
+
+  void OpenGLDynamicGPUArray::ReleaseRange(BufferRange free_buffer_range) {
+    fmt::print("release: base={} \tnumber_of_elements={}\n", free_buffer_range.base_element, free_buffer_range.number_of_elements);
+
+    auto neighbour_r_it = std::ranges::find_if(m_free_buffer_ranges, [&](const BufferRange& buffer_range) {
+      return buffer_range.base_element > free_buffer_range.base_element; });
+
+    std::vector<BufferRange>::iterator free_range_it;
+
+    if(neighbour_r_it != m_free_buffer_ranges.end()) {
+      if(neighbour_r_it->base_element == free_buffer_range.base_element + free_buffer_range.number_of_elements) {
+        neighbour_r_it->base_element = free_buffer_range.base_element;
+        neighbour_r_it->number_of_elements += free_buffer_range.number_of_elements;
+        free_range_it = neighbour_r_it;
+      } else {
+        free_range_it = m_free_buffer_ranges.insert(neighbour_r_it, free_buffer_range);
+      }
+    } else {
+      m_free_buffer_ranges.push_back(free_buffer_range);
+      free_range_it = m_free_buffer_ranges.end();
+    }
+
+    if(free_range_it != m_free_buffer_ranges.begin()) {
+      const auto neighbour_l_it = std::prev(free_range_it);
+
+      if(free_range_it->base_element == neighbour_l_it->base_element + neighbour_l_it->number_of_elements) {
+        neighbour_l_it->number_of_elements += free_range_it->number_of_elements;
+        m_free_buffer_ranges.erase(free_range_it);
+      }
+    }
+
+    // TODO(fleroviux): shrink array if possible
+  }
+
+  void OpenGLDynamicGPUArray::Write(std::span<const u8> data, size_t base_element, size_t byte_offset) {
+    const size_t buffer_write_start = base_element * m_byte_stride + byte_offset;
+    const size_t buffer_write_end = buffer_write_start + data.size();
+    const size_t buffer_byte_size = m_current_capacity * m_byte_stride;
+
+    if(buffer_write_end < buffer_write_start || buffer_write_end > buffer_byte_size) {
+      ZEPHYR_PANIC("Out-of-range dynamic GPU array write");
+    }
+
+    glNamedBufferSubData(m_gpu_buffer, (GLintptr)buffer_write_start, (GLsizeiptr)data.size(), data.data());
+  }
+
+  void OpenGLDynamicGPUArray::PrintFreeRanges() {
+    fmt::print("----------------------------------------\n");
+
+    for(const BufferRange& free_buffer_range : m_free_buffer_ranges) {
+      fmt::print("free: \tbase={} \tnumber_of_elements={}\n", free_buffer_range.base_element, free_buffer_range.number_of_elements);
+    }
+
+    fmt::print("----------------------------------------\n");
+  }
+
+  void OpenGLDynamicGPUArray::ResizeBuffer(size_t new_capacity) {
+    if(new_capacity == m_current_capacity) {
+      return;
+    }
+
+    // Create a new VBO that fits the new capacity
+    GLuint new_gpu_buffer;
+    glCreateBuffers(1u, &new_gpu_buffer);
+    glNamedBufferData(new_gpu_buffer, (GLsizeiptr)(new_capacity * m_byte_stride), nullptr, GL_DYNAMIC_DRAW);
+
+    if(m_current_capacity != 0u) {
+      // Copy the contents of the old buffer into the new buffer and delete the old buffer.
+      const size_t copy_size = std::min(new_capacity, m_current_capacity) * m_byte_stride;
+      glCopyNamedBufferSubData(m_gpu_buffer, new_gpu_buffer, 0u, 0u, (GLsizeiptr)copy_size);
+      glDeleteBuffers(1u, &m_gpu_buffer);
+    }
+
+    if(new_capacity > m_current_capacity) {
+      const size_t capacity_increment = new_capacity - m_current_capacity;
+
+      if(m_free_buffer_ranges.empty()) {
+        m_free_buffer_ranges.emplace_back(m_current_capacity, capacity_increment);
+      } else {
+        m_free_buffer_ranges.back().number_of_elements += capacity_increment;
+      }
+    }
+
+    m_gpu_buffer = new_gpu_buffer;
+    m_current_capacity = new_capacity;
+
+    // TODO(fleroviux): emit event for handling rebinding of the new buffer to i.e. a VAO?
+  }
+
+} // namespace zephyr

zephyr/renderer/src/backend/opengl/dynamic_gpu_array.hpp

@@ -0,0 +1,54 @@
+
+#pragma once
+
+#include <zephyr/renderer/backend/render_backend.hpp>
+#include <GL/glew.h>
+#include <GL/gl.h>
+#include <span>
+#include <vector>
+
+namespace zephyr {
+
+  /**
+   * TODO:
+   * - improve grow strategy (i.e. configurable grow factor)
+   * - implement shrinking
+   */
+
+  class OpenGLDynamicGPUArray {
+    public:
+      struct BufferRange {
+        BufferRange() = default;
+        BufferRange(size_t base_element, size_t number_of_elements) : base_element{base_element}, number_of_elements{number_of_elements} {}
+        size_t base_element;
+        size_t number_of_elements;
+      };
+
+      explicit OpenGLDynamicGPUArray(size_t byte_stride);
+     ~OpenGLDynamicGPUArray();
+
+      [[nodiscard]] GLuint GetBufferHandle() {
+        return m_gpu_buffer;
+      }
+
+      [[nodiscard]] size_t GetByteStride() const {
+        return m_byte_stride;
+      }
+
+      BufferRange AllocateRange(size_t number_of_elements);
+      void ReleaseRange(BufferRange buffer_range);
+      void Write(std::span<const u8> data, size_t base_element, size_t byte_offset = 0u);
+
+    private:
+      static constexpr size_t k_capacity_increment = 16384u;
+
+      void ResizeBuffer(size_t new_capacity);
+      void PrintFreeRanges();
+
+      size_t m_byte_stride{};
+      GLuint m_gpu_buffer{};
+      size_t m_current_capacity{0u};
+      std::vector<BufferRange> m_free_buffer_ranges{};
+  };
+
+} // namespace zephyr

zephyr/renderer/src/backend/opengl/render_backend.cpp

@@ -23,9 +23,13 @@ namespace zephyr {
     glNamedBufferData(m_gl_ubo, sizeof(Matrix4) * 2, nullptr, GL_DYNAMIC_DRAW);
 
     glEnable(GL_DEPTH_TEST);
+
+    m_render_geometry_manager = std::make_unique<OpenGLRenderGeometryManager>();
   }
 
   void OpenGLRenderBackend::DestroyContext() {
+    m_render_geometry_manager.reset();
+
     glDeleteBuffers(1u, &m_gl_ubo);
     glDeleteVertexArrays(1u, &m_gl_vao);
     glDeleteProgram(m_gl_shader_program);
@@ -34,19 +38,19 @@ namespace zephyr {
   }
 
   RenderGeometry* OpenGLRenderBackend::CreateRenderGeometry(RenderGeometryLayout layout, size_t number_of_vertices, size_t number_of_indices) {
-    return OpenGLRenderGeometry::Build(layout, number_of_vertices, number_of_indices);
+    return m_render_geometry_manager->CreateRenderGeometry(layout, number_of_vertices, number_of_indices);
   }
 
   void OpenGLRenderBackend::UpdateRenderGeometryIndices(RenderGeometry* render_geometry, std::span<const u8> data) {
-    dynamic_cast<OpenGLRenderGeometry*>(render_geometry)->UpdateIndices(data);
+    m_render_geometry_manager->UpdateRenderGeometryIndices(render_geometry, data);
   }
 
   void OpenGLRenderBackend::UpdateRenderGeometryVertices(RenderGeometry* render_geometry, std::span<const u8> data) {
-    dynamic_cast<OpenGLRenderGeometry*>(render_geometry)->UpdateVertices(data);
+    m_render_geometry_manager->UpdateRenderGeometryVertices(render_geometry, data);
   }
 
-  void OpenGLRenderBackend::DestroyRenderGeometry(RenderGeometry* geometry) {
-    delete geometry;
+  void OpenGLRenderBackend::DestroyRenderGeometry(RenderGeometry* render_geometry) {
+    m_render_geometry_manager->DestroyRenderGeometry(render_geometry);
   }
 
   void OpenGLRenderBackend::Render(const Matrix4& view_projection, std::span<const RenderObject> render_objects) {
@@ -58,10 +62,23 @@ namespace zephyr {
     glBindBufferBase(GL_UNIFORM_BUFFER, 0, m_gl_ubo);
     glNamedBufferSubData(m_gl_ubo, 0, sizeof(Matrix4), &view_projection);
 
+    glBindBuffer(GL_DRAW_INDIRECT_BUFFER, m_render_geometry_manager->GetDrawCommandBuffer());
+
     for(const RenderObject& render_object : render_objects) {
       glNamedBufferSubData(m_gl_ubo, sizeof(Matrix4), sizeof(Matrix4), &render_object.local_to_world);
-      dynamic_cast<OpenGLRenderGeometry*>(render_object.render_geometry)->Draw();
+
+      // TODO(fleroviux): avoid constant rebinding of VAO
+      auto render_geometry = dynamic_cast<OpenGLRenderGeometry*>(render_object.render_geometry);
+      auto indirect_buffer_offset = (const void*)(render_geometry->GetDrawCommandID() * sizeof(OpenGLDrawElementsIndirectCommand));
+      glBindVertexArray(m_render_geometry_manager->GetVAOFromLayout(render_geometry->GetLayout()));
+      if(render_geometry->GetNumberOfIndices() > 0) {
+        glDrawElementsIndirect(GL_TRIANGLES, GL_UNSIGNED_INT, indirect_buffer_offset);
+      } else {
+        glDrawArraysIndirect(GL_TRIANGLES, indirect_buffer_offset);
+      }
     }
+
+    glBindBuffer(GL_DRAW_INDIRECT_BUFFER, 0);
   }
 
   void OpenGLRenderBackend::SwapBuffers() {

zephyr/renderer/src/backend/opengl/render_backend.hpp

@@ -22,7 +22,7 @@ namespace zephyr {
       RenderGeometry* CreateRenderGeometry(RenderGeometryLayout layout, size_t number_of_vertices, size_t number_of_indices) override;
       void UpdateRenderGeometryIndices(RenderGeometry* render_geometry, std::span<const u8> data) override;
       void UpdateRenderGeometryVertices(RenderGeometry* render_geometry, std::span<const u8> data) override;
-      void DestroyRenderGeometry(RenderGeometry* geometry) override;
+      void DestroyRenderGeometry(RenderGeometry* render_geometry) override;
 
       void Render(const Matrix4& view_projection, std::span<const RenderObject> render_objects) override;
 
@@ -39,6 +39,7 @@ namespace zephyr {
       GLuint m_gl_shader_program{};
       GLuint m_gl_vao{};
       GLuint m_gl_ubo{};
+      std::unique_ptr<OpenGLRenderGeometryManager> m_render_geometry_manager{};
   };
 
   std::unique_ptr<RenderBackend> CreateOpenGLRenderBackendForSDL2(SDL_Window* sdl2_window) {