Index: ps/trunk/source/renderer/PatchRData.cpp =================================================================== --- ps/trunk/source/renderer/PatchRData.cpp (revision 24859) +++ ps/trunk/source/renderer/PatchRData.cpp (revision 24860) @@ -1,1473 +1,1457 @@ /* Copyright (C) 2021 Wildfire Games. * This file is part of 0 A.D. * * 0 A.D. is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation, either version 2 of the License, or * (at your option) any later version. * * 0 A.D. is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with 0 A.D. If not, see . */ #include "precompiled.h" #include "renderer/PatchRData.h" #include "graphics/GameView.h" #include "graphics/LightEnv.h" #include "graphics/LOSTexture.h" #include "graphics/Patch.h" #include "graphics/ShaderManager.h" #include "graphics/Terrain.h" #include "graphics/TerrainTextureEntry.h" #include "graphics/TextRenderer.h" #include "lib/allocators/DynamicArena.h" #include "lib/allocators/STLAllocators.h" #include "maths/MathUtil.h" #include "ps/CLogger.h" #include "ps/Game.h" #include "ps/GameSetup/Config.h" #include "ps/Profile.h" #include "ps/Pyrogenesis.h" #include "ps/World.h" #include "renderer/AlphaMapCalculator.h" #include "renderer/Renderer.h" #include "renderer/TerrainRenderer.h" #include "renderer/WaterManager.h" #include "simulation2/components/ICmpWaterManager.h" #include "simulation2/Simulation2.h" #include #include #include const ssize_t BlendOffsets[9][2] = { { 0, -1 }, { -1, -1 }, { -1, 0 }, { -1, 1 }, { 0, 1 }, { 1, 1 }, { 1, 0 }, { 1, -1 }, { 0, 0 } }; /////////////////////////////////////////////////////////////////// // CPatchRData constructor CPatchRData::CPatchRData(CPatch* patch, CSimulation2* simulation) : m_Patch(patch), m_VBSides(0), m_VBBase(0), m_VBBaseIndices(0), m_VBBlends(0), m_VBBlendIndices(0), m_VBWater(0), m_VBWaterIndices(0), m_VBWaterShore(0), m_VBWaterIndicesShore(0), m_Simulation(simulation) { ENSURE(patch); Build(); } /////////////////////////////////////////////////////////////////// // CPatchRData destructor CPatchRData::~CPatchRData() { // release vertex buffer chunks - if (m_VBSides) g_VBMan.Release(m_VBSides); - if (m_VBBase) g_VBMan.Release(m_VBBase); - if (m_VBBaseIndices) g_VBMan.Release(m_VBBaseIndices); - if (m_VBBlends) g_VBMan.Release(m_VBBlends); - if (m_VBBlendIndices) g_VBMan.Release(m_VBBlendIndices); - if (m_VBWater) g_VBMan.Release(m_VBWater); - if (m_VBWaterIndices) g_VBMan.Release(m_VBWaterIndices); - if (m_VBWaterShore) g_VBMan.Release(m_VBWaterShore); - if (m_VBWaterIndicesShore) g_VBMan.Release(m_VBWaterIndicesShore); + m_VBSides.Reset(); + m_VBBase.Reset(); + m_VBBaseIndices.Reset(); + m_VBBlends.Reset(); + m_VBBlendIndices.Reset(); + m_VBWater.Reset(); + m_VBWaterIndices.Reset(); + m_VBWaterShore.Reset(); + m_VBWaterIndicesShore.Reset(); } /** * Represents a blend for a single tile, texture and shape. */ struct STileBlend { CTerrainTextureEntry* m_Texture; int m_Priority; u16 m_TileMask; // bit n set if this blend contains neighbour tile BlendOffsets[n] struct DecreasingPriority { bool operator()(const STileBlend& a, const STileBlend& b) const { if (a.m_Priority > b.m_Priority) return true; if (a.m_Priority < b.m_Priority) return false; if (a.m_Texture && b.m_Texture) return a.m_Texture->GetTag() > b.m_Texture->GetTag(); return false; } }; struct CurrentTile { bool operator()(const STileBlend& a) const { return (a.m_TileMask & (1 << 8)) != 0; } }; }; /** * Represents the ordered collection of blends drawn on a particular tile. */ struct STileBlendStack { u8 i, j; std::vector blends; // back of vector is lowest-priority texture }; /** * Represents a batched collection of blends using the same texture. */ struct SBlendLayer { struct Tile { u8 i, j; u8 shape; }; CTerrainTextureEntry* m_Texture; std::vector m_Tiles; }; void CPatchRData::BuildBlends() { PROFILE3("build blends"); m_BlendSplats.clear(); std::vector blendVertices; std::vector blendIndices; CTerrain* terrain = m_Patch->m_Parent; std::vector blendStacks; blendStacks.reserve(PATCH_SIZE*PATCH_SIZE); std::vector blends; blends.reserve(9); // For each tile in patch .. for (ssize_t j = 0; j < PATCH_SIZE; ++j) { for (ssize_t i = 0; i < PATCH_SIZE; ++i) { ssize_t gx = m_Patch->m_X * PATCH_SIZE + i; ssize_t gz = m_Patch->m_Z * PATCH_SIZE + j; blends.clear(); // Compute a blend for every tile in the 3x3 square around this tile for (size_t n = 0; n < 9; ++n) { ssize_t ox = gx + BlendOffsets[n][1]; ssize_t oz = gz + BlendOffsets[n][0]; CMiniPatch* nmp = terrain->GetTile(ox, oz); if (!nmp) continue; STileBlend blend; blend.m_Texture = nmp->GetTextureEntry(); blend.m_Priority = nmp->GetPriority(); blend.m_TileMask = 1 << n; blends.push_back(blend); } // Sort the blends, highest priority first std::sort(blends.begin(), blends.end(), STileBlend::DecreasingPriority()); STileBlendStack blendStack; blendStack.i = i; blendStack.j = j; // Put the blends into the tile's stack, merging any adjacent blends with the same texture for (size_t k = 0; k < blends.size(); ++k) { if (!blendStack.blends.empty() && blendStack.blends.back().m_Texture == blends[k].m_Texture) blendStack.blends.back().m_TileMask |= blends[k].m_TileMask; else blendStack.blends.push_back(blends[k]); } // Remove blends that are after (i.e. lower priority than) the current tile // (including the current tile), since we don't want to render them on top of // the tile's base texture blendStack.blends.erase( std::find_if(blendStack.blends.begin(), blendStack.blends.end(), STileBlend::CurrentTile()), blendStack.blends.end()); blendStacks.push_back(blendStack); } } // Given the blend stack per tile, we want to batch together as many blends as possible. // Group them into a series of layers (each of which has a single texture): // (This is effectively a topological sort / linearisation of the partial order induced // by the per-tile stacks, preferring to make tiles with equal textures adjacent.) std::vector blendLayers; while (true) { if (!blendLayers.empty()) { // Try to grab as many tiles as possible that match our current layer, // from off the blend stacks of all the tiles CTerrainTextureEntry* tex = blendLayers.back().m_Texture; for (size_t k = 0; k < blendStacks.size(); ++k) { if (!blendStacks[k].blends.empty() && blendStacks[k].blends.back().m_Texture == tex) { SBlendLayer::Tile t = { blendStacks[k].i, blendStacks[k].j, (u8)blendStacks[k].blends.back().m_TileMask }; blendLayers.back().m_Tiles.push_back(t); blendStacks[k].blends.pop_back(); } // (We've already merged adjacent entries of the same texture in each stack, // so we don't need to bother looping to check the next entry in this stack again) } } // We've grabbed as many tiles as possible; now we need to start a new layer. // The new layer's texture could come from the back of any non-empty stack; // choose the longest stack as a heuristic to reduce the number of layers CTerrainTextureEntry* bestTex = NULL; size_t bestStackSize = 0; for (size_t k = 0; k < blendStacks.size(); ++k) { if (blendStacks[k].blends.size() > bestStackSize) { bestStackSize = blendStacks[k].blends.size(); bestTex = blendStacks[k].blends.back().m_Texture; } } // If all our stacks were empty, we're done if (bestStackSize == 0) break; // Otherwise add the new layer, then loop back and start filling it in SBlendLayer layer; layer.m_Texture = bestTex; blendLayers.push_back(layer); } // Now build outgoing splats m_BlendSplats.resize(blendLayers.size()); for (size_t k = 0; k < blendLayers.size(); ++k) { SSplat& splat = m_BlendSplats[k]; splat.m_IndexStart = blendIndices.size(); splat.m_Texture = blendLayers[k].m_Texture; for (size_t t = 0; t < blendLayers[k].m_Tiles.size(); ++t) { SBlendLayer::Tile& tile = blendLayers[k].m_Tiles[t]; AddBlend(blendVertices, blendIndices, tile.i, tile.j, tile.shape, splat.m_Texture); } splat.m_IndexCount = blendIndices.size() - splat.m_IndexStart; } // Release existing vertex buffer chunks - if (m_VBBlends) - { - g_VBMan.Release(m_VBBlends); - m_VBBlends = 0; - } - - if (m_VBBlendIndices) - { - g_VBMan.Release(m_VBBlendIndices); - m_VBBlendIndices = 0; - } + m_VBBlends.Reset(); + m_VBBlendIndices.Reset(); if (blendVertices.size()) { // Construct vertex buffer - m_VBBlends = g_VBMan.Allocate(sizeof(SBlendVertex), blendVertices.size(), GL_STATIC_DRAW, GL_ARRAY_BUFFER); - m_VBBlends->m_Owner->UpdateChunkVertices(m_VBBlends, &blendVertices[0]); + m_VBBlends = g_VBMan.AllocateChunk(sizeof(SBlendVertex), blendVertices.size(), GL_STATIC_DRAW, GL_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::TERRAIN); + m_VBBlends->m_Owner->UpdateChunkVertices(m_VBBlends.Get(), &blendVertices[0]); // Update the indices to include the base offset of the vertex data for (size_t k = 0; k < blendIndices.size(); ++k) blendIndices[k] += static_cast(m_VBBlends->m_Index); - m_VBBlendIndices = g_VBMan.Allocate(sizeof(u16), blendIndices.size(), GL_STATIC_DRAW, GL_ELEMENT_ARRAY_BUFFER); - m_VBBlendIndices->m_Owner->UpdateChunkVertices(m_VBBlendIndices, &blendIndices[0]); + m_VBBlendIndices = g_VBMan.AllocateChunk(sizeof(u16), blendIndices.size(), GL_STATIC_DRAW, GL_ELEMENT_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::TERRAIN); + m_VBBlendIndices->m_Owner->UpdateChunkVertices(m_VBBlendIndices.Get(), &blendIndices[0]); } } void CPatchRData::AddBlend(std::vector& blendVertices, std::vector& blendIndices, u16 i, u16 j, u8 shape, CTerrainTextureEntry* texture) { CTerrain* terrain = m_Patch->m_Parent; ssize_t gx = m_Patch->m_X * PATCH_SIZE + i; ssize_t gz = m_Patch->m_Z * PATCH_SIZE + j; // uses the current neighbour texture BlendShape8 shape8; for (size_t m = 0; m < 8; ++m) shape8[m] = (shape & (1 << m)) ? 0 : 1; // calculate the required alphamap and the required rotation of the alphamap from blendshape unsigned int alphamapflags; int alphamap = CAlphaMapCalculator::Calculate(shape8, alphamapflags); // now actually render the blend tile (if we need one) if (alphamap == -1) return; float u0 = texture->m_TerrainAlpha->second.m_AlphaMapCoords[alphamap].u0; float u1 = texture->m_TerrainAlpha->second.m_AlphaMapCoords[alphamap].u1; float v0 = texture->m_TerrainAlpha->second.m_AlphaMapCoords[alphamap].v0; float v1 = texture->m_TerrainAlpha->second.m_AlphaMapCoords[alphamap].v1; if (alphamapflags & BLENDMAP_FLIPU) std::swap(u0, u1); if (alphamapflags & BLENDMAP_FLIPV) std::swap(v0, v1); int base = 0; if (alphamapflags & BLENDMAP_ROTATE90) base = 1; else if (alphamapflags & BLENDMAP_ROTATE180) base = 2; else if (alphamapflags & BLENDMAP_ROTATE270) base = 3; SBlendVertex vtx[4]; vtx[(base + 0) % 4].m_AlphaUVs[0] = u0; vtx[(base + 0) % 4].m_AlphaUVs[1] = v0; vtx[(base + 1) % 4].m_AlphaUVs[0] = u1; vtx[(base + 1) % 4].m_AlphaUVs[1] = v0; vtx[(base + 2) % 4].m_AlphaUVs[0] = u1; vtx[(base + 2) % 4].m_AlphaUVs[1] = v1; vtx[(base + 3) % 4].m_AlphaUVs[0] = u0; vtx[(base + 3) % 4].m_AlphaUVs[1] = v1; SBlendVertex dst; CVector3D normal; u16 index = static_cast(blendVertices.size()); terrain->CalcPosition(gx, gz, dst.m_Position); terrain->CalcNormal(gx, gz, normal); dst.m_Normal = normal; dst.m_AlphaUVs[0] = vtx[0].m_AlphaUVs[0]; dst.m_AlphaUVs[1] = vtx[0].m_AlphaUVs[1]; blendVertices.push_back(dst); terrain->CalcPosition(gx + 1, gz, dst.m_Position); terrain->CalcNormal(gx + 1, gz, normal); dst.m_Normal = normal; dst.m_AlphaUVs[0] = vtx[1].m_AlphaUVs[0]; dst.m_AlphaUVs[1] = vtx[1].m_AlphaUVs[1]; blendVertices.push_back(dst); terrain->CalcPosition(gx + 1, gz + 1, dst.m_Position); terrain->CalcNormal(gx + 1, gz + 1, normal); dst.m_Normal = normal; dst.m_AlphaUVs[0] = vtx[2].m_AlphaUVs[0]; dst.m_AlphaUVs[1] = vtx[2].m_AlphaUVs[1]; blendVertices.push_back(dst); terrain->CalcPosition(gx, gz + 1, dst.m_Position); terrain->CalcNormal(gx, gz + 1, normal); dst.m_Normal = normal; dst.m_AlphaUVs[0] = vtx[3].m_AlphaUVs[0]; dst.m_AlphaUVs[1] = vtx[3].m_AlphaUVs[1]; blendVertices.push_back(dst); bool dir = terrain->GetTriangulationDir(gx, gz); if (dir) { blendIndices.push_back(index+0); blendIndices.push_back(index+1); blendIndices.push_back(index+3); blendIndices.push_back(index+1); blendIndices.push_back(index+2); blendIndices.push_back(index+3); } else { blendIndices.push_back(index+0); blendIndices.push_back(index+1); blendIndices.push_back(index+2); blendIndices.push_back(index+2); blendIndices.push_back(index+3); blendIndices.push_back(index+0); } } void CPatchRData::BuildIndices() { PROFILE3("build indices"); CTerrain* terrain = m_Patch->m_Parent; ssize_t px = m_Patch->m_X * PATCH_SIZE; ssize_t pz = m_Patch->m_Z * PATCH_SIZE; // must have allocated some vertices before trying to build corresponding indices ENSURE(m_VBBase); // number of vertices in each direction in each patch ssize_t vsize=PATCH_SIZE+1; // PATCH_SIZE must be 2^8-2 or less to not overflow u16 indices buffer. Thankfully this is always true. ENSURE(vsize*vsize < 65536); std::vector indices; indices.reserve(PATCH_SIZE * PATCH_SIZE * 4); // release existing splats m_Splats.clear(); // build grid of textures on this patch std::vector textures; CTerrainTextureEntry* texgrid[PATCH_SIZE][PATCH_SIZE]; for (ssize_t j=0;jm_MiniPatches[j][i].GetTextureEntry(); texgrid[j][i]=tex; if (std::find(textures.begin(),textures.end(),tex)==textures.end()) { textures.push_back(tex); } } } // now build base splats from interior textures m_Splats.resize(textures.size()); // build indices for base splats size_t base=m_VBBase->m_Index; for (size_t k = 0; k < m_Splats.size(); ++k) { CTerrainTextureEntry* tex = textures[k]; SSplat& splat=m_Splats[k]; splat.m_Texture=tex; splat.m_IndexStart=indices.size(); for (ssize_t j = 0; j < PATCH_SIZE; j++) { for (ssize_t i = 0; i < PATCH_SIZE; i++) { if (texgrid[j][i] == tex) { bool dir = terrain->GetTriangulationDir(px+i, pz+j); if (dir) { indices.push_back(u16(((j+0)*vsize+(i+0))+base)); indices.push_back(u16(((j+0)*vsize+(i+1))+base)); indices.push_back(u16(((j+1)*vsize+(i+0))+base)); indices.push_back(u16(((j+0)*vsize+(i+1))+base)); indices.push_back(u16(((j+1)*vsize+(i+1))+base)); indices.push_back(u16(((j+1)*vsize+(i+0))+base)); } else { indices.push_back(u16(((j+0)*vsize+(i+0))+base)); indices.push_back(u16(((j+0)*vsize+(i+1))+base)); indices.push_back(u16(((j+1)*vsize+(i+1))+base)); indices.push_back(u16(((j+1)*vsize+(i+1))+base)); indices.push_back(u16(((j+1)*vsize+(i+0))+base)); indices.push_back(u16(((j+0)*vsize+(i+0))+base)); } } } } splat.m_IndexCount=indices.size()-splat.m_IndexStart; } // Release existing vertex buffer chunk - if (m_VBBaseIndices) - { - g_VBMan.Release(m_VBBaseIndices); - m_VBBaseIndices = 0; - } + m_VBBaseIndices.Reset(); ENSURE(indices.size()); // Construct vertex buffer - m_VBBaseIndices = g_VBMan.Allocate(sizeof(u16), indices.size(), GL_STATIC_DRAW, GL_ELEMENT_ARRAY_BUFFER); - m_VBBaseIndices->m_Owner->UpdateChunkVertices(m_VBBaseIndices, &indices[0]); + m_VBBaseIndices = g_VBMan.AllocateChunk(sizeof(u16), indices.size(), GL_STATIC_DRAW, GL_ELEMENT_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::TERRAIN); + m_VBBaseIndices->m_Owner->UpdateChunkVertices(m_VBBaseIndices.Get(), &indices[0]); } void CPatchRData::BuildVertices() { PROFILE3("build vertices"); // create both vertices and lighting colors // number of vertices in each direction in each patch ssize_t vsize = PATCH_SIZE + 1; std::vector vertices; vertices.resize(vsize * vsize); // get index of this patch ssize_t px = m_Patch->m_X; ssize_t pz = m_Patch->m_Z; CTerrain* terrain = m_Patch->m_Parent; // build vertices for (ssize_t j = 0; j < vsize; ++j) { for (ssize_t i = 0; i < vsize; ++i) { ssize_t ix = px * PATCH_SIZE + i; ssize_t iz = pz * PATCH_SIZE + j; ssize_t v = j * vsize + i; // calculate vertex data terrain->CalcPosition(ix, iz, vertices[v].m_Position); CVector3D normal; terrain->CalcNormal(ix, iz, normal); vertices[v].m_Normal = normal; } } // upload to vertex buffer if (!m_VBBase) - m_VBBase = g_VBMan.Allocate(sizeof(SBaseVertex), vsize * vsize, GL_STATIC_DRAW, GL_ARRAY_BUFFER); + m_VBBase = g_VBMan.AllocateChunk(sizeof(SBaseVertex), vsize * vsize, GL_STATIC_DRAW, GL_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::TERRAIN); - m_VBBase->m_Owner->UpdateChunkVertices(m_VBBase, &vertices[0]); + m_VBBase->m_Owner->UpdateChunkVertices(m_VBBase.Get(), &vertices[0]); } void CPatchRData::BuildSide(std::vector& vertices, CPatchSideFlags side) { ssize_t vsize = PATCH_SIZE + 1; CTerrain* terrain = m_Patch->m_Parent; CmpPtr cmpWaterManager(*m_Simulation, SYSTEM_ENTITY); for (ssize_t k = 0; k < vsize; k++) { ssize_t gx = m_Patch->m_X * PATCH_SIZE; ssize_t gz = m_Patch->m_Z * PATCH_SIZE; switch (side) { case CPATCH_SIDE_NEGX: gz += k; break; case CPATCH_SIDE_POSX: gx += PATCH_SIZE; gz += PATCH_SIZE-k; break; case CPATCH_SIDE_NEGZ: gx += PATCH_SIZE-k; break; case CPATCH_SIDE_POSZ: gz += PATCH_SIZE; gx += k; break; } CVector3D pos; terrain->CalcPosition(gx, gz, pos); // Clamp the height to the water level float waterHeight = 0.f; if (cmpWaterManager) waterHeight = cmpWaterManager->GetExactWaterLevel(pos.X, pos.Z); pos.Y = std::max(pos.Y, waterHeight); SSideVertex v0, v1; v0.m_Position = pos; v1.m_Position = pos; v1.m_Position.Y = 0; // If this is the start of this tristrip, but we've already got a partial // tristrip, add a couple of degenerate triangles to join the strips properly if (k == 0 && !vertices.empty()) { vertices.push_back(vertices.back()); vertices.push_back(v1); } // Now add the new triangles vertices.push_back(v1); vertices.push_back(v0); } } void CPatchRData::BuildSides() { PROFILE3("build sides"); std::vector sideVertices; int sideFlags = m_Patch->GetSideFlags(); // If no sides are enabled, we don't need to do anything if (!sideFlags) return; // For each side, generate a tristrip by adding a vertex at ground/water // level and a vertex underneath at height 0. if (sideFlags & CPATCH_SIDE_NEGX) BuildSide(sideVertices, CPATCH_SIDE_NEGX); if (sideFlags & CPATCH_SIDE_POSX) BuildSide(sideVertices, CPATCH_SIDE_POSX); if (sideFlags & CPATCH_SIDE_NEGZ) BuildSide(sideVertices, CPATCH_SIDE_NEGZ); if (sideFlags & CPATCH_SIDE_POSZ) BuildSide(sideVertices, CPATCH_SIDE_POSZ); if (sideVertices.empty()) return; if (!m_VBSides) - m_VBSides = g_VBMan.Allocate(sizeof(SSideVertex), sideVertices.size(), GL_STATIC_DRAW, GL_ARRAY_BUFFER); - m_VBSides->m_Owner->UpdateChunkVertices(m_VBSides, &sideVertices[0]); + m_VBSides = g_VBMan.AllocateChunk(sizeof(SSideVertex), sideVertices.size(), GL_STATIC_DRAW, GL_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::DEFAULT); + m_VBSides->m_Owner->UpdateChunkVertices(m_VBSides.Get(), &sideVertices[0]); } void CPatchRData::Build() { BuildVertices(); BuildSides(); BuildIndices(); BuildBlends(); BuildWater(); } void CPatchRData::Update(CSimulation2* simulation) { m_Simulation = simulation; if (m_UpdateFlags!=0) { // TODO,RC 11/04/04 - need to only rebuild necessary bits of renderdata rather // than everything; it's complicated slightly because the blends are dependent // on both vertex and index data BuildVertices(); BuildSides(); BuildIndices(); BuildBlends(); BuildWater(); m_UpdateFlags=0; } } // Types used for glMultiDrawElements batching: // To minimise the cost of memory allocations, everything used for computing // batches uses a arena allocator. (All allocations are short-lived so we can // just throw away the whole arena at the end of each frame.) using Arena = Allocators::DynamicArena<1 * MiB>; // std::map types with appropriate arena allocators and default comparison operator template using PooledBatchMap = std::map, ProxyAllocator, Arena>>; // Equivalent to "m[k]", when it returns a arena-allocated std::map (since we can't // use the default constructor in that case) template typename M::mapped_type& PooledMapGet(M& m, const typename M::key_type& k, Arena& arena) { return m.insert(std::make_pair(k, typename M::mapped_type(typename M::mapped_type::key_compare(), typename M::mapped_type::allocator_type(arena)) )).first->second; } // Equivalent to "m[k]", when it returns a std::pair of arena-allocated std::vectors template typename M::mapped_type& PooledPairGet(M& m, const typename M::key_type& k, Arena& arena) { return m.insert(std::make_pair(k, std::make_pair( typename M::mapped_type::first_type(typename M::mapped_type::first_type::allocator_type(arena)), typename M::mapped_type::second_type(typename M::mapped_type::second_type::allocator_type(arena)) ))).first->second; } // Each multidraw batch has a list of index counts, and a list of pointers-to-first-indexes using BatchElements = std::pair>, std::vector>>; // Group batches by index buffer using IndexBufferBatches = PooledBatchMap; // Group batches by vertex buffer using VertexBufferBatches = PooledBatchMap; // Group batches by texture using TextureBatches = PooledBatchMap; // Group batches by shaders. using ShaderTechniqueBatches = PooledBatchMap; void CPatchRData::RenderBases( const std::vector& patches, const CShaderDefines& context, ShadowMap* shadow) { + PROFILE3("render terrain bases"); + Arena arena; ShaderTechniqueBatches batches(ShaderTechniqueBatches::key_compare(), (ShaderTechniqueBatches::allocator_type(arena))); PROFILE_START("compute batches"); // Collect all the patches' base splats into their appropriate batches for (size_t i = 0; i < patches.size(); ++i) { CPatchRData* patch = patches[i]; for (size_t j = 0; j < patch->m_Splats.size(); ++j) { SSplat& splat = patch->m_Splats[j]; const CMaterial& material = splat.m_Texture->GetMaterial(); if (material.GetShaderEffect().empty()) { LOGERROR("Terrain renderer failed to load shader effect.\n"); continue; } CShaderTechniquePtr techBase = g_Renderer.GetShaderManager().LoadEffect( material.GetShaderEffect(), context, material.GetShaderDefines(0)); BatchElements& batch = PooledPairGet( PooledMapGet( PooledMapGet( PooledMapGet(batches, techBase, arena), splat.m_Texture, arena ), patch->m_VBBase->m_Owner, arena ), patch->m_VBBaseIndices->m_Owner, arena ); batch.first.push_back(splat.m_IndexCount); u8* indexBase = patch->m_VBBaseIndices->m_Owner->GetBindAddress(); batch.second.push_back(indexBase + sizeof(u16)*(patch->m_VBBaseIndices->m_Index + splat.m_IndexStart)); } } PROFILE_END("compute batches"); // Render each batch for (ShaderTechniqueBatches::iterator itTech = batches.begin(); itTech != batches.end(); ++itTech) { const CShaderTechniquePtr& techBase = itTech->first; const int numPasses = techBase->GetNumPasses(); for (int pass = 0; pass < numPasses; ++pass) { techBase->BeginPass(pass); const CShaderProgramPtr& shader = techBase->GetShader(pass); TerrainRenderer::PrepareShader(shader, shadow); TextureBatches& textureBatches = itTech->second; for (TextureBatches::iterator itt = textureBatches.begin(); itt != textureBatches.end(); ++itt) { if (itt->first->GetMaterial().GetSamplers().size() != 0) { const CMaterial::SamplersVector& samplers = itt->first->GetMaterial().GetSamplers(); for(const CMaterial::TextureSampler& samp : samplers) shader->BindTexture(samp.Name, samp.Sampler); itt->first->GetMaterial().GetStaticUniforms().BindUniforms(shader); float c = itt->first->GetTextureMatrix()[0]; float ms = itt->first->GetTextureMatrix()[8]; shader->Uniform(str_textureTransform, c, ms, -ms, 0.f); } else { shader->BindTexture(str_baseTex, g_Renderer.GetTextureManager().GetErrorTexture()); } for (VertexBufferBatches::iterator itv = itt->second.begin(); itv != itt->second.end(); ++itv) { GLsizei stride = sizeof(SBaseVertex); SBaseVertex *base = (SBaseVertex *)itv->first->Bind(); shader->VertexPointer(3, GL_FLOAT, stride, &base->m_Position[0]); shader->NormalPointer(GL_FLOAT, stride, &base->m_Normal[0]); shader->TexCoordPointer(GL_TEXTURE0, 3, GL_FLOAT, stride, &base->m_Position[0]); shader->AssertPointersBound(); for (IndexBufferBatches::iterator it = itv->second.begin(); it != itv->second.end(); ++it) { it->first->Bind(); BatchElements& batch = it->second; if (!g_Renderer.m_SkipSubmit) { // Don't use glMultiDrawElements here since it doesn't have a significant // performance impact and it suffers from various driver bugs (e.g. it breaks // in Mesa 7.10 swrast with index VBOs) for (size_t i = 0; i < batch.first.size(); ++i) glDrawElements(GL_TRIANGLES, batch.first[i], GL_UNSIGNED_SHORT, batch.second[i]); } g_Renderer.m_Stats.m_DrawCalls++; g_Renderer.m_Stats.m_TerrainTris += std::accumulate(batch.first.begin(), batch.first.end(), 0) / 3; } } } techBase->EndPass(); } } CVertexBuffer::Unbind(); } /** * Helper structure for RenderBlends. */ struct SBlendBatch { SBlendBatch(Arena& arena) : m_Batches(VertexBufferBatches::key_compare(), VertexBufferBatches::allocator_type(arena)) { } CTerrainTextureEntry* m_Texture; CShaderTechniquePtr m_ShaderTech; VertexBufferBatches m_Batches; }; /** * Helper structure for RenderBlends. */ struct SBlendStackItem { SBlendStackItem(CVertexBuffer::VBChunk* v, CVertexBuffer::VBChunk* i, const std::vector& s, Arena& arena) : vertices(v), indices(i), splats(s.begin(), s.end(), SplatStack::allocator_type(arena)) { } using SplatStack = std::vector>; CVertexBuffer::VBChunk* vertices; CVertexBuffer::VBChunk* indices; SplatStack splats; }; void CPatchRData::RenderBlends( const std::vector& patches, const CShaderDefines& context, ShadowMap* shadow) { + PROFILE3("render terrain blends"); + Arena arena; using BatchesStack = std::vector>; BatchesStack batches((BatchesStack::allocator_type(arena))); CShaderDefines contextBlend = context; contextBlend.Add(str_BLEND, str_1); PROFILE_START("compute batches"); // Reserve an arbitrary size that's probably big enough in most cases, // to avoid heavy reallocations batches.reserve(256); using BlendStacks = std::vector>; BlendStacks blendStacks((BlendStacks::allocator_type(arena))); blendStacks.reserve(patches.size()); // Extract all the blend splats from each patch for (size_t i = 0; i < patches.size(); ++i) { CPatchRData* patch = patches[i]; if (!patch->m_BlendSplats.empty()) { - blendStacks.push_back(SBlendStackItem(patch->m_VBBlends, patch->m_VBBlendIndices, patch->m_BlendSplats, arena)); + blendStacks.push_back(SBlendStackItem(patch->m_VBBlends.Get(), patch->m_VBBlendIndices.Get(), patch->m_BlendSplats, arena)); // Reverse the splats so the first to be rendered is at the back of the list std::reverse(blendStacks.back().splats.begin(), blendStacks.back().splats.end()); } } // Rearrange the collection of splats to be grouped by texture, preserving // order of splats within each patch: // (This is exactly the same algorithm used in CPatchRData::BuildBlends, // but applied to patch-sized splats rather than to tile-sized splats; // see that function for comments on the algorithm.) while (true) { if (!batches.empty()) { CTerrainTextureEntry* tex = batches.back().m_Texture; for (size_t k = 0; k < blendStacks.size(); ++k) { SBlendStackItem::SplatStack& splats = blendStacks[k].splats; if (!splats.empty() && splats.back().m_Texture == tex) { CVertexBuffer::VBChunk* vertices = blendStacks[k].vertices; CVertexBuffer::VBChunk* indices = blendStacks[k].indices; BatchElements& batch = PooledPairGet(PooledMapGet(batches.back().m_Batches, vertices->m_Owner, arena), indices->m_Owner, arena); batch.first.push_back(splats.back().m_IndexCount); u8* indexBase = indices->m_Owner->GetBindAddress(); batch.second.push_back(indexBase + sizeof(u16)*(indices->m_Index + splats.back().m_IndexStart)); splats.pop_back(); } } } CTerrainTextureEntry* bestTex = NULL; size_t bestStackSize = 0; for (size_t k = 0; k < blendStacks.size(); ++k) { SBlendStackItem::SplatStack& splats = blendStacks[k].splats; if (splats.size() > bestStackSize) { bestStackSize = splats.size(); bestTex = splats.back().m_Texture; } } if (bestStackSize == 0) break; SBlendBatch layer(arena); layer.m_Texture = bestTex; if (!bestTex->GetMaterial().GetSamplers().empty()) { layer.m_ShaderTech = g_Renderer.GetShaderManager().LoadEffect( bestTex->GetMaterial().GetShaderEffect(), contextBlend, bestTex->GetMaterial().GetShaderDefines(0)); } batches.push_back(layer); } PROFILE_END("compute batches"); glEnable(GL_BLEND); glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA); CVertexBuffer* lastVB = nullptr; CShaderProgramPtr previousShader; for (BatchesStack::iterator itTechBegin = batches.begin(), itTechEnd = batches.begin(); itTechBegin != batches.end(); itTechBegin = itTechEnd) { while (itTechEnd != batches.end() && itTechEnd->m_ShaderTech == itTechBegin->m_ShaderTech) ++itTechEnd; const CShaderTechniquePtr& techBase = itTechBegin->m_ShaderTech; const int numPasses = techBase->GetNumPasses(); for (int pass = 0; pass < numPasses; ++pass) { techBase->BeginPass(pass); const CShaderProgramPtr& shader = techBase->GetShader(pass); TerrainRenderer::PrepareShader(shader, shadow); Handle lastBlendTex = 0; for (BatchesStack::iterator itt = itTechBegin; itt != itTechEnd; ++itt) { if (itt->m_Texture->GetMaterial().GetSamplers().empty()) continue; if (itt->m_Texture) { const CMaterial::SamplersVector& samplers = itt->m_Texture->GetMaterial().GetSamplers(); for (const CMaterial::TextureSampler& samp : samplers) shader->BindTexture(samp.Name, samp.Sampler); Handle currentBlendTex = itt->m_Texture->m_TerrainAlpha->second.m_hCompositeAlphaMap; if (currentBlendTex != lastBlendTex) { shader->BindTexture(str_blendTex, currentBlendTex); lastBlendTex = currentBlendTex; } itt->m_Texture->GetMaterial().GetStaticUniforms().BindUniforms(shader); float c = itt->m_Texture->GetTextureMatrix()[0]; float ms = itt->m_Texture->GetTextureMatrix()[8]; shader->Uniform(str_textureTransform, c, ms, -ms, 0.f); } else { shader->BindTexture(str_baseTex, g_Renderer.GetTextureManager().GetErrorTexture()); } for (VertexBufferBatches::iterator itv = itt->m_Batches.begin(); itv != itt->m_Batches.end(); ++itv) { // Rebind the VB only if it changed since the last batch if (itv->first != lastVB || shader != previousShader) { lastVB = itv->first; previousShader = shader; GLsizei stride = sizeof(SBlendVertex); SBlendVertex *base = (SBlendVertex *)itv->first->Bind(); shader->VertexPointer(3, GL_FLOAT, stride, &base->m_Position[0]); shader->NormalPointer(GL_FLOAT, stride, &base->m_Normal[0]); shader->TexCoordPointer(GL_TEXTURE0, 3, GL_FLOAT, stride, &base->m_Position[0]); shader->TexCoordPointer(GL_TEXTURE1, 2, GL_FLOAT, stride, &base->m_AlphaUVs[0]); } shader->AssertPointersBound(); for (IndexBufferBatches::iterator it = itv->second.begin(); it != itv->second.end(); ++it) { it->first->Bind(); BatchElements& batch = it->second; if (!g_Renderer.m_SkipSubmit) { for (size_t i = 0; i < batch.first.size(); ++i) glDrawElements(GL_TRIANGLES, batch.first[i], GL_UNSIGNED_SHORT, batch.second[i]); } g_Renderer.m_Stats.m_DrawCalls++; g_Renderer.m_Stats.m_BlendSplats++; g_Renderer.m_Stats.m_TerrainTris += std::accumulate(batch.first.begin(), batch.first.end(), 0) / 3; } } } techBase->EndPass(); } } glDisable(GL_BLEND); CVertexBuffer::Unbind(); } void CPatchRData::RenderStreams(const std::vector& patches, const CShaderProgramPtr& shader, int streamflags) { + PROFILE3("render terrain streams"); + // Each batch has a list of index counts, and a list of pointers-to-first-indexes using StreamBatchElements = std::pair, std::vector > ; // Group batches by index buffer using StreamIndexBufferBatches = std::map ; // Group batches by vertex buffer using StreamVertexBufferBatches = std::map ; StreamVertexBufferBatches batches; PROFILE_START("compute batches"); // Collect all the patches into their appropriate batches for (const CPatchRData* patch : patches) { StreamBatchElements& batch = batches[patch->m_VBBase->m_Owner][patch->m_VBBaseIndices->m_Owner]; batch.first.push_back(patch->m_VBBaseIndices->m_Count); u8* indexBase = patch->m_VBBaseIndices->m_Owner->GetBindAddress(); batch.second.push_back(indexBase + sizeof(u16)*(patch->m_VBBaseIndices->m_Index)); } PROFILE_END("compute batches"); ENSURE(!(streamflags & ~(STREAM_POS|STREAM_POSTOUV0|STREAM_POSTOUV1))); // Render each batch for (const std::pair& streamBatch : batches) { GLsizei stride = sizeof(SBaseVertex); SBaseVertex *base = (SBaseVertex *)streamBatch.first->Bind(); shader->VertexPointer(3, GL_FLOAT, stride, &base->m_Position); if (streamflags & STREAM_POSTOUV0) shader->TexCoordPointer(GL_TEXTURE0, 3, GL_FLOAT, stride, &base->m_Position); if (streamflags & STREAM_POSTOUV1) shader->TexCoordPointer(GL_TEXTURE1, 3, GL_FLOAT, stride, &base->m_Position); shader->AssertPointersBound(); for (const std::pair& batchIndexBuffer : streamBatch.second) { batchIndexBuffer.first->Bind(); const StreamBatchElements& batch = batchIndexBuffer.second; if (!g_Renderer.m_SkipSubmit) { for (size_t i = 0; i < batch.first.size(); ++i) glDrawElements(GL_TRIANGLES, batch.first[i], GL_UNSIGNED_SHORT, batch.second[i]); } g_Renderer.m_Stats.m_DrawCalls++; g_Renderer.m_Stats.m_TerrainTris += std::accumulate(batch.first.begin(), batch.first.end(), 0) / 3; } } CVertexBuffer::Unbind(); } void CPatchRData::RenderOutline() { CTerrain* terrain = m_Patch->m_Parent; ssize_t gx = m_Patch->m_X * PATCH_SIZE; ssize_t gz = m_Patch->m_Z * PATCH_SIZE; CVector3D pos; std::vector line; for (ssize_t i = 0, j = 0; i <= PATCH_SIZE; ++i) { terrain->CalcPosition(gx + i, gz + j, pos); line.push_back(pos); } for (ssize_t i = PATCH_SIZE, j = 1; j <= PATCH_SIZE; ++j) { terrain->CalcPosition(gx + i, gz + j, pos); line.push_back(pos); } for (ssize_t i = PATCH_SIZE-1, j = PATCH_SIZE; i >= 0; --i) { terrain->CalcPosition(gx + i, gz + j, pos); line.push_back(pos); } for (ssize_t i = 0, j = PATCH_SIZE-1; j >= 0; --j) { terrain->CalcPosition(gx + i, gz + j, pos); line.push_back(pos); } #if CONFIG2_GLES #warning TODO: implement CPatchRData::RenderOutlines for GLES #else glVertexPointer(3, GL_FLOAT, sizeof(CVector3D), &line[0]); glDrawArrays(GL_LINE_STRIP, 0, line.size()); #endif } -void CPatchRData::RenderSides(CShaderProgramPtr& shader) +void CPatchRData::RenderSides(const std::vector& patches, const CShaderProgramPtr& shader) { - ENSURE(m_UpdateFlags==0); - - if (!m_VBSides) - return; - glDisable(GL_CULL_FACE); - SSideVertex *base = (SSideVertex *)m_VBSides->m_Owner->Bind(); + CVertexBuffer* lastVB = nullptr; + for (CPatchRData* patch : patches) + { + ENSURE(patch->m_UpdateFlags == 0); + if (!patch->m_VBSides) + continue; + if (lastVB != patch->m_VBSides->m_Owner) + { + lastVB = patch->m_VBSides->m_Owner; + SSideVertex *base = (SSideVertex*)patch->m_VBSides->m_Owner->Bind(); - // setup data pointers - GLsizei stride = sizeof(SSideVertex); - shader->VertexPointer(3, GL_FLOAT, stride, &base->m_Position); + // setup data pointers + GLsizei stride = sizeof(SSideVertex); + shader->VertexPointer(3, GL_FLOAT, stride, &base->m_Position); + } - shader->AssertPointersBound(); + shader->AssertPointersBound(); - if (!g_Renderer.m_SkipSubmit) - glDrawArrays(GL_TRIANGLE_STRIP, m_VBSides->m_Index, (GLsizei)m_VBSides->m_Count); + if (!g_Renderer.m_SkipSubmit) + glDrawArrays(GL_TRIANGLE_STRIP, patch->m_VBSides->m_Index, (GLsizei)patch->m_VBSides->m_Count); - // bump stats - g_Renderer.m_Stats.m_DrawCalls++; - g_Renderer.m_Stats.m_TerrainTris += m_VBSides->m_Count - 2; + // bump stats + g_Renderer.m_Stats.m_DrawCalls++; + g_Renderer.m_Stats.m_TerrainTris += patch->m_VBSides->m_Count - 2; + } CVertexBuffer::Unbind(); glEnable(GL_CULL_FACE); } void CPatchRData::RenderPriorities(CTextRenderer& textRenderer) { CTerrain* terrain = m_Patch->m_Parent; const CCamera& camera = *(g_Game->GetView()->GetCamera()); for (ssize_t j = 0; j < PATCH_SIZE; ++j) { for (ssize_t i = 0; i < PATCH_SIZE; ++i) { ssize_t gx = m_Patch->m_X * PATCH_SIZE + i; ssize_t gz = m_Patch->m_Z * PATCH_SIZE + j; CVector3D pos; terrain->CalcPosition(gx, gz, pos); // Move a bit towards the center of the tile pos.X += TERRAIN_TILE_SIZE/4.f; pos.Z += TERRAIN_TILE_SIZE/4.f; float x, y; camera.GetScreenCoordinates(pos, x, y); textRenderer.PrintfAt(x, y, L"%d", m_Patch->m_MiniPatches[j][i].Priority); } } } // // Water build and rendering // // Build vertex buffer for water vertices over our patch void CPatchRData::BuildWater() { PROFILE3("build water"); // Number of vertices in each direction in each patch ENSURE(PATCH_SIZE % water_cell_size == 0); - if (m_VBWater) - { - g_VBMan.Release(m_VBWater); - m_VBWater = nullptr; - } - if (m_VBWaterIndices) - { - g_VBMan.Release(m_VBWaterIndices); - m_VBWaterIndices = nullptr; - } - if (m_VBWaterShore) - { - g_VBMan.Release(m_VBWaterShore); - m_VBWaterShore = nullptr; - } - if (m_VBWaterIndicesShore) - { - g_VBMan.Release(m_VBWaterIndicesShore); - m_VBWaterIndicesShore = nullptr; - } + m_VBWater.Reset(); + m_VBWaterIndices.Reset(); + m_VBWaterShore.Reset(); + m_VBWaterIndicesShore.Reset(); + m_WaterBounds.SetEmpty(); // We need to use this to access the water manager or we may not have the // actual values but some compiled-in defaults CmpPtr cmpWaterManager(*m_Simulation, SYSTEM_ENTITY); if (!cmpWaterManager) return; // Build data for water std::vector water_vertex_data; std::vector water_indices; u16 water_index_map[PATCH_SIZE+1][PATCH_SIZE+1]; memset(water_index_map, 0xFF, sizeof(water_index_map)); // Build data for shore std::vector water_vertex_data_shore; std::vector water_indices_shore; u16 water_shore_index_map[PATCH_SIZE+1][PATCH_SIZE+1]; memset(water_shore_index_map, 0xFF, sizeof(water_shore_index_map)); WaterManager* WaterMgr = g_Renderer.GetWaterManager(); CPatch* patch = m_Patch; CTerrain* terrain = patch->m_Parent; ssize_t mapSize = terrain->GetVerticesPerSide(); // Top-left coordinates of our patch. ssize_t px = m_Patch->m_X * PATCH_SIZE; ssize_t pz = m_Patch->m_Z * PATCH_SIZE; // To whoever implements different water heights, this is a TODO: water height) float waterHeight = cmpWaterManager->GetExactWaterLevel(0.0f,0.0f); // The 4 points making a water tile. int moves[4][2] = { {0, 0}, {water_cell_size, 0}, {0, water_cell_size}, {water_cell_size, water_cell_size} }; // Where to look for when checking for water for shore tiles. int check[10][2] = { {0, 0}, {water_cell_size, 0}, {water_cell_size*2, 0}, {0, water_cell_size}, {0, water_cell_size*2}, {water_cell_size, water_cell_size}, {water_cell_size*2, water_cell_size*2}, {-water_cell_size, 0}, {0, -water_cell_size}, {-water_cell_size, -water_cell_size} }; // build vertices, uv, and shader varying for (ssize_t z = 0; z < PATCH_SIZE; z += water_cell_size) { for (ssize_t x = 0; x < PATCH_SIZE; x += water_cell_size) { // Check that this tile is close to water bool nearWater = false; for (size_t test = 0; test < 10; ++test) if (terrain->GetVertexGroundLevel(x + px + check[test][0], z + pz + check[test][1]) < waterHeight) nearWater = true; if (!nearWater) continue; // This is actually lying and I should call CcmpTerrain /*if (!terrain->IsOnMap(x+x1, z+z1) && !terrain->IsOnMap(x+x1, z+z1 + water_cell_size) && !terrain->IsOnMap(x+x1 + water_cell_size, z+z1) && !terrain->IsOnMap(x+x1 + water_cell_size, z+z1 + water_cell_size)) continue;*/ for (int i = 0; i < 4; ++i) { if (water_index_map[z+moves[i][1]][x+moves[i][0]] != 0xFFFF) continue; ssize_t xx = x + px + moves[i][0]; ssize_t zz = z + pz + moves[i][1]; SWaterVertex vertex; terrain->CalcPosition(xx,zz, vertex.m_Position); float depth = waterHeight - vertex.m_Position.Y; vertex.m_Position.Y = waterHeight; m_WaterBounds += vertex.m_Position; vertex.m_WaterData = CVector2D(WaterMgr->m_WindStrength[xx + zz*mapSize], depth); water_index_map[z+moves[i][1]][x+moves[i][0]] = static_cast(water_vertex_data.size()); water_vertex_data.push_back(vertex); } water_indices.push_back(water_index_map[z + moves[2][1]][x + moves[2][0]]); water_indices.push_back(water_index_map[z + moves[0][1]][x + moves[0][0]]); water_indices.push_back(water_index_map[z + moves[1][1]][x + moves[1][0]]); water_indices.push_back(water_index_map[z + moves[1][1]][x + moves[1][0]]); water_indices.push_back(water_index_map[z + moves[3][1]][x + moves[3][0]]); water_indices.push_back(water_index_map[z + moves[2][1]][x + moves[2][0]]); // Check id this tile is partly over land. // If so add a square over the terrain. This is necessary to render waves that go on shore. if (terrain->GetVertexGroundLevel(x+px, z+pz) < waterHeight && terrain->GetVertexGroundLevel(x+px + water_cell_size, z+pz) < waterHeight && terrain->GetVertexGroundLevel(x+px, z+pz+water_cell_size) < waterHeight && terrain->GetVertexGroundLevel(x+px + water_cell_size, z+pz+water_cell_size) < waterHeight) continue; for (int i = 0; i < 4; ++i) { if (water_shore_index_map[z+moves[i][1]][x+moves[i][0]] != 0xFFFF) continue; ssize_t xx = x + px + moves[i][0]; ssize_t zz = z + pz + moves[i][1]; SWaterVertex vertex; terrain->CalcPosition(xx,zz, vertex.m_Position); vertex.m_Position.Y += 0.02f; m_WaterBounds += vertex.m_Position; vertex.m_WaterData = CVector2D(0.0f, -5.0f); water_shore_index_map[z+moves[i][1]][x+moves[i][0]] = static_cast(water_vertex_data_shore.size()); water_vertex_data_shore.push_back(vertex); } if (terrain->GetTriangulationDir(x + px, z + pz)) { water_indices_shore.push_back(water_shore_index_map[z + moves[2][1]][x + moves[2][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[0][1]][x + moves[0][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[1][1]][x + moves[1][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[1][1]][x + moves[1][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[3][1]][x + moves[3][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[2][1]][x + moves[2][0]]); } else { water_indices_shore.push_back(water_shore_index_map[z + moves[3][1]][x + moves[3][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[2][1]][x + moves[2][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[0][1]][x + moves[0][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[3][1]][x + moves[3][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[0][1]][x + moves[0][0]]); water_indices_shore.push_back(water_shore_index_map[z + moves[1][1]][x + moves[1][0]]); } } } // No vertex buffers if no data generated if (!water_indices.empty()) { - m_VBWater = g_VBMan.Allocate(sizeof(SWaterVertex), water_vertex_data.size(), GL_STATIC_DRAW, GL_ARRAY_BUFFER); - m_VBWater->m_Owner->UpdateChunkVertices(m_VBWater, &water_vertex_data[0]); + m_VBWater = g_VBMan.AllocateChunk(sizeof(SWaterVertex), water_vertex_data.size(), GL_STATIC_DRAW, GL_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::WATER); + m_VBWater->m_Owner->UpdateChunkVertices(m_VBWater.Get(), &water_vertex_data[0]); - m_VBWaterIndices = g_VBMan.Allocate(sizeof(GLushort), water_indices.size(), GL_STATIC_DRAW, GL_ELEMENT_ARRAY_BUFFER); - m_VBWaterIndices->m_Owner->UpdateChunkVertices(m_VBWaterIndices, &water_indices[0]); + m_VBWaterIndices = g_VBMan.AllocateChunk(sizeof(GLushort), water_indices.size(), GL_STATIC_DRAW, GL_ELEMENT_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::WATER); + m_VBWaterIndices->m_Owner->UpdateChunkVertices(m_VBWaterIndices.Get(), &water_indices[0]); } if (!water_indices_shore.empty()) { - m_VBWaterShore = g_VBMan.Allocate(sizeof(SWaterVertex), water_vertex_data_shore.size(), GL_STATIC_DRAW, GL_ARRAY_BUFFER); - m_VBWaterShore->m_Owner->UpdateChunkVertices(m_VBWaterShore, &water_vertex_data_shore[0]); + m_VBWaterShore = g_VBMan.AllocateChunk(sizeof(SWaterVertex), water_vertex_data_shore.size(), GL_STATIC_DRAW, GL_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::WATER); + m_VBWaterShore->m_Owner->UpdateChunkVertices(m_VBWaterShore.Get(), &water_vertex_data_shore[0]); // Construct indices buffer - m_VBWaterIndicesShore = g_VBMan.Allocate(sizeof(GLushort), water_indices_shore.size(), GL_STATIC_DRAW, GL_ELEMENT_ARRAY_BUFFER); - m_VBWaterIndicesShore->m_Owner->UpdateChunkVertices(m_VBWaterIndicesShore, &water_indices_shore[0]); + m_VBWaterIndicesShore = g_VBMan.AllocateChunk(sizeof(GLushort), water_indices_shore.size(), GL_STATIC_DRAW, GL_ELEMENT_ARRAY_BUFFER, nullptr, CVertexBufferManager::Group::WATER); + m_VBWaterIndicesShore->m_Owner->UpdateChunkVertices(m_VBWaterIndicesShore.Get(), &water_indices_shore[0]); } } void CPatchRData::RenderWater(CShaderProgramPtr& shader, bool onlyShore, bool fixedPipeline) { ASSERT(m_UpdateFlags==0); if (g_Renderer.m_SkipSubmit || (!m_VBWater && !m_VBWaterShore)) return; #if !CONFIG2_GLES if (g_Renderer.GetWaterRenderMode() == WIREFRAME) glPolygonMode(GL_FRONT_AND_BACK, GL_LINE); #endif - if (m_VBWater != 0x0 && !onlyShore) + if (m_VBWater && !onlyShore) { SWaterVertex *base=(SWaterVertex *)m_VBWater->m_Owner->Bind(); // setup data pointers GLsizei stride = sizeof(SWaterVertex); shader->VertexPointer(3, GL_FLOAT, stride, &base[m_VBWater->m_Index].m_Position); if (!fixedPipeline) shader->VertexAttribPointer(str_a_waterInfo, 2, GL_FLOAT, false, stride, &base[m_VBWater->m_Index].m_WaterData); shader->AssertPointersBound(); u8* indexBase = m_VBWaterIndices->m_Owner->Bind(); glDrawElements(GL_TRIANGLES, (GLsizei) m_VBWaterIndices->m_Count, GL_UNSIGNED_SHORT, indexBase + sizeof(u16)*(m_VBWaterIndices->m_Index)); g_Renderer.m_Stats.m_DrawCalls++; g_Renderer.m_Stats.m_WaterTris += m_VBWaterIndices->m_Count / 3; } - if (m_VBWaterShore != 0x0 && + if (m_VBWaterShore && g_Renderer.GetWaterManager()->m_WaterEffects && g_Renderer.GetWaterManager()->m_WaterFancyEffects) { SWaterVertex *base=(SWaterVertex *)m_VBWaterShore->m_Owner->Bind(); GLsizei stride = sizeof(SWaterVertex); shader->VertexPointer(3, GL_FLOAT, stride, &base[m_VBWaterShore->m_Index].m_Position); if (!fixedPipeline) shader->VertexAttribPointer(str_a_waterInfo, 2, GL_FLOAT, false, stride, &base[m_VBWaterShore->m_Index].m_WaterData); shader->AssertPointersBound(); u8* indexBase = m_VBWaterIndicesShore->m_Owner->Bind(); glDrawElements(GL_TRIANGLES, (GLsizei) m_VBWaterIndicesShore->m_Count, GL_UNSIGNED_SHORT, indexBase + sizeof(u16)*(m_VBWaterIndicesShore->m_Index)); g_Renderer.m_Stats.m_DrawCalls++; g_Renderer.m_Stats.m_WaterTris += m_VBWaterIndicesShore->m_Count / 3; } CVertexBuffer::Unbind(); #if !CONFIG2_GLES if (g_Renderer.GetWaterRenderMode() == WIREFRAME) glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); #endif } Index: ps/trunk/source/renderer/PatchRData.h =================================================================== --- ps/trunk/source/renderer/PatchRData.h (revision 24859) +++ ps/trunk/source/renderer/PatchRData.h (revision 24860) @@ -1,181 +1,181 @@ /* Copyright (C) 2021 Wildfire Games. * This file is part of 0 A.D. * * 0 A.D. is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation, either version 2 of the License, or * (at your option) any later version. * * 0 A.D. is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with 0 A.D. If not, see . */ #ifndef INCLUDED_PATCHRDATA #define INCLUDED_PATCHRDATA #include "graphics/Patch.h" #include "graphics/RenderableObject.h" #include "graphics/SColor.h" #include "graphics/ShaderProgramPtr.h" #include "maths/Vector2D.h" #include "maths/Vector3D.h" #include "renderer/VertexBufferManager.h" #include class CPatch; class CShaderDefines; class CSimulation2; class CTerrainTextureEntry; class CTextRenderer; class ShadowMap; ////////////////////////////////////////////////////////////////////////////////////////////////// // CPatchRData: class encapsulating logic for rendering terrain patches; holds per // patch data, plus some supporting static functions for batching, etc class CPatchRData : public CRenderData { public: CPatchRData(CPatch* patch, CSimulation2* simulation); ~CPatchRData(); void Update(CSimulation2* simulation); void RenderOutline(); - void RenderSides(CShaderProgramPtr& shader); void RenderPriorities(CTextRenderer& textRenderer); void RenderWater(CShaderProgramPtr& shader, bool onlyShore = false, bool fixedPipeline = false); CPatch* GetPatch() { return m_Patch; } const CBoundingBoxAligned& GetWaterBounds() const { return m_WaterBounds; } static void RenderBases( const std::vector& patches, const CShaderDefines& context, ShadowMap* shadow); static void RenderBlends( const std::vector& patches, const CShaderDefines& context, ShadowMap* shadow); static void RenderStreams(const std::vector& patches, const CShaderProgramPtr& shader, int streamflags); + static void RenderSides(const std::vector& patches, const CShaderProgramPtr& shader); static void PrepareShader(const CShaderProgramPtr& shader, ShadowMap* shadow); private: friend struct SBlendStackItem; struct SSplat { SSplat() : m_Texture(0), m_IndexCount(0) {} // texture to apply during splat CTerrainTextureEntry* m_Texture; // offset into the index array for this patch where splat starts size_t m_IndexStart; // number of indices used by splat size_t m_IndexCount; }; struct SBaseVertex { // vertex position CVector3D m_Position; CVector3D m_Normal; // pad to a power of two u8 m_Padding[8]; }; cassert(sizeof(SBaseVertex) == 32); struct SSideVertex { // vertex position CVector3D m_Position; // pad to a power of two u8 m_Padding[4]; }; cassert(sizeof(SSideVertex) == 16); struct SBlendVertex { // vertex position CVector3D m_Position; // vertex uvs for alpha texture float m_AlphaUVs[2]; CVector3D m_Normal; }; cassert(sizeof(SBlendVertex) == 32); // Mixed Fancy/Simple water vertex description data structure struct SWaterVertex { // vertex position CVector3D m_Position; CVector2D m_WaterData; // pad to a power of two u8 m_Padding[12]; }; cassert(sizeof(SWaterVertex) == 32); // build this renderdata object void Build(); void AddBlend(std::vector& blendVertices, std::vector& blendIndices, u16 i, u16 j, u8 shape, CTerrainTextureEntry* texture); void BuildBlends(); void BuildIndices(); void BuildVertices(); void BuildSides(); void BuildSide(std::vector& vertices, CPatchSideFlags side); // owner patch CPatch* m_Patch; // vertex buffer handle for side vertices - CVertexBuffer::VBChunk* m_VBSides; + CVertexBufferManager::Handle m_VBSides; // vertex buffer handle for base vertices - CVertexBuffer::VBChunk* m_VBBase; + CVertexBufferManager::Handle m_VBBase; // vertex buffer handle for base vertex indices - CVertexBuffer::VBChunk* m_VBBaseIndices; + CVertexBufferManager::Handle m_VBBaseIndices; // vertex buffer handle for blend vertices - CVertexBuffer::VBChunk* m_VBBlends; + CVertexBufferManager::Handle m_VBBlends; // vertex buffer handle for blend vertex indices - CVertexBuffer::VBChunk* m_VBBlendIndices; + CVertexBufferManager::Handle m_VBBlendIndices; // list of base splats to apply to this patch std::vector m_Splats; // splats used in blend pass std::vector m_BlendSplats; // boundary of water in this patch CBoundingBoxAligned m_WaterBounds; // Water vertex buffer - CVertexBuffer::VBChunk* m_VBWater; - CVertexBuffer::VBChunk* m_VBWaterShore; + CVertexBufferManager::Handle m_VBWater; + CVertexBufferManager::Handle m_VBWaterShore; // Water indices buffer - CVertexBuffer::VBChunk* m_VBWaterIndices; - CVertexBuffer::VBChunk* m_VBWaterIndicesShore; + CVertexBufferManager::Handle m_VBWaterIndices; + CVertexBufferManager::Handle m_VBWaterIndicesShore; CSimulation2* m_Simulation; // Build water vertices and indices (vertex buffer and data vector) void BuildWater(); // parameter allowing a varying number of triangles per patch for LOD // MUST be an exact divisor of PATCH_SIZE // compiled const for the moment until/if dynamic water LOD is offered // savings would be mostly beneficial for GPU or simple water static const ssize_t water_cell_size = 1; }; #endif // INCLUDED_PATCHRDATA Index: ps/trunk/source/renderer/TerrainRenderer.cpp =================================================================== --- ps/trunk/source/renderer/TerrainRenderer.cpp (revision 24859) +++ ps/trunk/source/renderer/TerrainRenderer.cpp (revision 24860) @@ -1,696 +1,695 @@ /* Copyright (C) 2021 Wildfire Games. * This file is part of 0 A.D. * * 0 A.D. is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation, either version 2 of the License, or * (at your option) any later version. * * 0 A.D. is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with 0 A.D. If not, see . */ /* * Terrain rendering (everything related to patches and water) is * encapsulated in TerrainRenderer */ #include "precompiled.h" #include "graphics/Camera.h" #include "graphics/Decal.h" #include "graphics/LightEnv.h" #include "graphics/LOSTexture.h" #include "graphics/Patch.h" #include "graphics/GameView.h" #include "graphics/Model.h" #include "graphics/ShaderManager.h" #include "renderer/ShadowMap.h" #include "renderer/SkyManager.h" #include "graphics/TerritoryTexture.h" #include "graphics/TextRenderer.h" #include "maths/MathUtil.h" #include "ps/Filesystem.h" #include "ps/CLogger.h" #include "ps/Game.h" #include "ps/Profile.h" #include "ps/World.h" #include "renderer/DecalRData.h" #include "renderer/PatchRData.h" #include "renderer/Renderer.h" #include "renderer/RenderingOptions.h" #include "renderer/ShadowMap.h" #include "renderer/TerrainRenderer.h" #include "renderer/VertexArray.h" #include "renderer/WaterManager.h" #include "tools/atlas/GameInterface/GameLoop.h" extern GameLoopState* g_AtlasGameLoop; /////////////////////////////////////////////////////////////////////////////////////////////// // TerrainRenderer implementation namespace { CShaderProgramPtr GetDummyShader() { const char* shaderName; if (g_RenderingOptions.GetPreferGLSL()) shaderName = "glsl/dummy"; else shaderName = "arb/dummy"; return g_Renderer.GetShaderManager().LoadProgram(shaderName, CShaderDefines()); } } // anonymous namespace /** * TerrainRenderer keeps track of which phase it is in, to detect * when Submit, PrepareForRendering etc. are called in the wrong order. */ enum Phase { Phase_Submit, Phase_Render }; /** * Struct TerrainRendererInternals: Internal variables used by the TerrainRenderer class. */ struct TerrainRendererInternals { /// Which phase (submitting or rendering patches) are we in right now? Phase phase; /// Patches that were submitted for this frame std::vector visiblePatches[CRenderer::CULL_MAX]; /// Decals that were submitted for this frame std::vector visibleDecals[CRenderer::CULL_MAX]; /// Fancy water shader CShaderProgramPtr fancyWaterShader; CSimulation2* simulation; }; /////////////////////////////////////////////////////////////////// // Construction/Destruction TerrainRenderer::TerrainRenderer() { m = new TerrainRendererInternals(); m->phase = Phase_Submit; } TerrainRenderer::~TerrainRenderer() { delete m; } void TerrainRenderer::SetSimulation(CSimulation2* simulation) { m->simulation = simulation; } /////////////////////////////////////////////////////////////////// // Submit a patch for rendering void TerrainRenderer::Submit(int cullGroup, CPatch* patch) { ENSURE(m->phase == Phase_Submit); CPatchRData* data = (CPatchRData*)patch->GetRenderData(); if (data == 0) { // no renderdata for patch, create it now data = new CPatchRData(patch, m->simulation); patch->SetRenderData(data); } data->Update(m->simulation); m->visiblePatches[cullGroup].push_back(data); } /////////////////////////////////////////////////////////////////// // Submit a decal for rendering void TerrainRenderer::Submit(int cullGroup, CModelDecal* decal) { ENSURE(m->phase == Phase_Submit); CDecalRData* data = (CDecalRData*)decal->GetRenderData(); if (data == 0) { // no renderdata for decal, create it now data = new CDecalRData(decal, m->simulation); decal->SetRenderData(data); } data->Update(m->simulation); m->visibleDecals[cullGroup].push_back(data); } /////////////////////////////////////////////////////////////////// // Prepare for rendering void TerrainRenderer::PrepareForRendering() { ENSURE(m->phase == Phase_Submit); m->phase = Phase_Render; } /////////////////////////////////////////////////////////////////// // Clear submissions lists void TerrainRenderer::EndFrame() { ENSURE(m->phase == Phase_Render || m->phase == Phase_Submit); for (int i = 0; i < CRenderer::CULL_MAX; ++i) { m->visiblePatches[i].clear(); m->visibleDecals[i].clear(); } m->phase = Phase_Submit; } void TerrainRenderer::RenderTerrainOverlayTexture(int cullGroup, CMatrix3D& textureMatrix, GLuint texture) { #if CONFIG2_GLES #warning TODO: implement TerrainRenderer::RenderTerrainOverlayTexture for GLES UNUSED2(cullGroup); UNUSED2(textureMatrix); UNUSED2(texture); #else ENSURE(m->phase == Phase_Render); std::vector& visiblePatches = m->visiblePatches[cullGroup]; glEnable(GL_TEXTURE_2D); glEnable(GL_BLEND); glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA); glDepthMask(0); glDisable(GL_DEPTH_TEST); CShaderTechniquePtr debugOverlayTech = g_Renderer.GetShaderManager().LoadEffect(str_debug_overlay); debugOverlayTech->BeginPass(); CShaderProgramPtr debugOverlayShader = debugOverlayTech->GetShader(); debugOverlayShader->Bind(); debugOverlayShader->BindTexture(str_baseTex, texture); debugOverlayShader->Uniform(str_transform, g_Renderer.GetViewCamera().GetViewProjection()); debugOverlayShader->Uniform(str_textureTransform, textureMatrix); CPatchRData::RenderStreams(visiblePatches, debugOverlayShader, STREAM_POS | STREAM_POSTOUV0); glEnable(GL_DEPTH_TEST); // To make the overlay visible over water, render an additional map-sized // water-height patch. CBoundingBoxAligned waterBounds; for (CPatchRData* data : visiblePatches) waterBounds += data->GetWaterBounds(); if (!waterBounds.IsEmpty()) { // Add a delta to avoid z-fighting. const float height = g_Renderer.GetWaterManager()->m_WaterHeight + 0.05f; const float waterPos[] = { waterBounds[0].X, height, waterBounds[0].Z, waterBounds[1].X, height, waterBounds[0].Z, waterBounds[0].X, height, waterBounds[1].Z, waterBounds[1].X, height, waterBounds[1].Z }; const GLsizei stride = sizeof(float) * 3; debugOverlayShader->VertexPointer(3, GL_FLOAT, stride, waterPos); debugOverlayShader->TexCoordPointer(GL_TEXTURE0, 3, GL_FLOAT, stride, waterPos); debugOverlayShader->AssertPointersBound(); glDrawArrays(GL_TRIANGLE_STRIP, 0, 4); } debugOverlayShader->Unbind(); debugOverlayTech->EndPass(); glDepthMask(1); glDisable(GL_BLEND); #endif } /////////////////////////////////////////////////////////////////// /** * Set up all the uniforms for a shader pass. */ void TerrainRenderer::PrepareShader(const CShaderProgramPtr& shader, ShadowMap* shadow) { shader->Uniform(str_transform, g_Renderer.GetViewCamera().GetViewProjection()); shader->Uniform(str_cameraPos, g_Renderer.GetViewCamera().GetOrientation().GetTranslation()); const CLightEnv& lightEnv = g_Renderer.GetLightEnv(); if (shadow) shadow->BindTo(shader); CLOSTexture& los = g_Renderer.GetScene().GetLOSTexture(); shader->BindTexture(str_losTex, los.GetTextureSmooth()); shader->Uniform(str_losTransform, los.GetTextureMatrix()[0], los.GetTextureMatrix()[12], 0.f, 0.f); shader->Uniform(str_ambient, lightEnv.m_AmbientColor); shader->Uniform(str_sunColor, lightEnv.m_SunColor); shader->Uniform(str_sunDir, lightEnv.GetSunDir()); shader->Uniform(str_fogColor, lightEnv.m_FogColor); shader->Uniform(str_fogParams, lightEnv.m_FogFactor, lightEnv.m_FogMax, 0.f, 0.f); } void TerrainRenderer::RenderTerrainShader(const CShaderDefines& context, int cullGroup, ShadowMap* shadow) { ENSURE(m->phase == Phase_Render); std::vector& visiblePatches = m->visiblePatches[cullGroup]; std::vector& visibleDecals = m->visibleDecals[cullGroup]; if (visiblePatches.empty() && visibleDecals.empty()) return; // render the solid black sides of the map first CShaderTechniquePtr techSolid = g_Renderer.GetShaderManager().LoadEffect(str_gui_solid); techSolid->BeginPass(); CShaderProgramPtr shaderSolid = techSolid->GetShader(); shaderSolid->Uniform(str_transform, g_Renderer.GetViewCamera().GetViewProjection()); shaderSolid->Uniform(str_color, 0.0f, 0.0f, 0.0f, 1.0f); PROFILE_START("render terrain sides"); - for (size_t i = 0; i < visiblePatches.size(); ++i) - visiblePatches[i]->RenderSides(shaderSolid); + CPatchRData::RenderSides(visiblePatches, shaderSolid); PROFILE_END("render terrain sides"); techSolid->EndPass(); PROFILE_START("render terrain base"); CPatchRData::RenderBases(visiblePatches, context, shadow); PROFILE_END("render terrain base"); // no need to write to the depth buffer a second time glDepthMask(0); // render blend passes for each patch PROFILE_START("render terrain blends"); CPatchRData::RenderBlends(visiblePatches, context, shadow); PROFILE_END("render terrain blends"); PROFILE_START("render terrain decals"); CDecalRData::RenderDecals(visibleDecals, context, shadow); PROFILE_END("render terrain decals"); // restore OpenGL state g_Renderer.BindTexture(1, 0); g_Renderer.BindTexture(2, 0); g_Renderer.BindTexture(3, 0); glDepthMask(1); glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA); glDisable(GL_BLEND); } /////////////////////////////////////////////////////////////////// // Render un-textured patches as polygons void TerrainRenderer::RenderPatches(int cullGroup, const CColor& color) { ENSURE(m->phase == Phase_Render); std::vector& visiblePatches = m->visiblePatches[cullGroup]; if (visiblePatches.empty()) return; #if CONFIG2_GLES #warning TODO: implement TerrainRenderer::RenderPatches for GLES #else CShaderProgramPtr dummyShader = GetDummyShader(); dummyShader->Bind(); dummyShader->Uniform(str_transform, g_Renderer.GetViewCamera().GetViewProjection()); dummyShader->Uniform(str_color, color); glEnableClientState(GL_VERTEX_ARRAY); CPatchRData::RenderStreams(visiblePatches, dummyShader, STREAM_POS); glDisableClientState(GL_VERTEX_ARRAY); dummyShader->Unbind(); #endif } /////////////////////////////////////////////////////////////////// // Render outlines of submitted patches as lines void TerrainRenderer::RenderOutlines(int cullGroup) { ENSURE(m->phase == Phase_Render); std::vector& visiblePatches = m->visiblePatches[cullGroup]; if (visiblePatches.empty()) return; #if CONFIG2_GLES #warning TODO: implement TerrainRenderer::RenderOutlines for GLES #else glEnableClientState(GL_VERTEX_ARRAY); for (size_t i = 0; i < visiblePatches.size(); ++i) visiblePatches[i]->RenderOutline(); glDisableClientState(GL_VERTEX_ARRAY); #endif } /////////////////////////////////////////////////////////////////// // Scissor rectangle of water patches CBoundingBoxAligned TerrainRenderer::ScissorWater(int cullGroup, const CMatrix3D &viewproj) { std::vector& visiblePatches = m->visiblePatches[cullGroup]; CBoundingBoxAligned scissor; for (size_t i = 0; i < visiblePatches.size(); ++i) { CPatchRData* data = visiblePatches[i]; const CBoundingBoxAligned& waterBounds = data->GetWaterBounds(); if (waterBounds.IsEmpty()) continue; CVector4D v1 = viewproj.Transform(CVector4D(waterBounds[0].X, waterBounds[1].Y, waterBounds[0].Z, 1.0f)); CVector4D v2 = viewproj.Transform(CVector4D(waterBounds[1].X, waterBounds[1].Y, waterBounds[0].Z, 1.0f)); CVector4D v3 = viewproj.Transform(CVector4D(waterBounds[0].X, waterBounds[1].Y, waterBounds[1].Z, 1.0f)); CVector4D v4 = viewproj.Transform(CVector4D(waterBounds[1].X, waterBounds[1].Y, waterBounds[1].Z, 1.0f)); CBoundingBoxAligned screenBounds; #define ADDBOUND(v1, v2, v3, v4) \ if (v1.Z >= -v1.W) \ screenBounds += CVector3D(v1.X, v1.Y, v1.Z) * (1.0f / v1.W); \ else \ { \ float t = v1.Z + v1.W; \ if (v2.Z > -v2.W) \ { \ CVector4D c2 = v1 + (v2 - v1) * (t / (t - (v2.Z + v2.W))); \ screenBounds += CVector3D(c2.X, c2.Y, c2.Z) * (1.0f / c2.W); \ } \ if (v3.Z > -v3.W) \ { \ CVector4D c3 = v1 + (v3 - v1) * (t / (t - (v3.Z + v3.W))); \ screenBounds += CVector3D(c3.X, c3.Y, c3.Z) * (1.0f / c3.W); \ } \ if (v4.Z > -v4.W) \ { \ CVector4D c4 = v1 + (v4 - v1) * (t / (t - (v4.Z + v4.W))); \ screenBounds += CVector3D(c4.X, c4.Y, c4.Z) * (1.0f / c4.W); \ } \ } ADDBOUND(v1, v2, v3, v4); ADDBOUND(v2, v1, v3, v4); ADDBOUND(v3, v1, v2, v4); ADDBOUND(v4, v1, v2, v3); #undef ADDBOUND if (screenBounds[0].X >= 1.0f || screenBounds[1].X <= -1.0f || screenBounds[0].Y >= 1.0f || screenBounds[1].Y <= -1.0f) continue; scissor += screenBounds; } return CBoundingBoxAligned(CVector3D(Clamp(scissor[0].X, -1.0f, 1.0f), Clamp(scissor[0].Y, -1.0f, 1.0f), -1.0f), CVector3D(Clamp(scissor[1].X, -1.0f, 1.0f), Clamp(scissor[1].Y, -1.0f, 1.0f), 1.0f)); } // Render fancy water bool TerrainRenderer::RenderFancyWater(const CShaderDefines& context, int cullGroup, ShadowMap* shadow) { PROFILE3_GPU("fancy water"); WaterManager* WaterMgr = g_Renderer.GetWaterManager(); CShaderDefines defines = context; // If we're using fancy water, make sure its shader is loaded if (!m->fancyWaterShader || WaterMgr->m_NeedsReloading) { if (WaterMgr->m_WaterRealDepth) defines.Add(str_USE_REAL_DEPTH, str_1); if (WaterMgr->m_WaterFancyEffects) defines.Add(str_USE_FANCY_EFFECTS, str_1); if (WaterMgr->m_WaterRefraction) defines.Add(str_USE_REFRACTION, str_1); if (WaterMgr->m_WaterReflection) defines.Add(str_USE_REFLECTION, str_1); // haven't updated the ARB shader yet so I'll always load the GLSL /*if (!g_RenderingOptions.GetPreferGLSL() && !superFancy) m->fancyWaterShader = g_Renderer.GetShaderManager().LoadProgram("arb/water_high", defines); else*/ m->fancyWaterShader = g_Renderer.GetShaderManager().LoadProgram("glsl/water_high", defines); if (!m->fancyWaterShader) { LOGERROR("Failed to load water shader. Falling back to fixed pipeline water.\n"); WaterMgr->m_RenderWater = false; return false; } WaterMgr->m_NeedsReloading = false; } CLOSTexture& losTexture = g_Renderer.GetScene().GetLOSTexture(); // Calculating the advanced informations about Foam and all if the quality calls for it. /*if (WaterMgr->m_NeedInfoUpdate && (WaterMgr->m_WaterFoam || WaterMgr->m_WaterCoastalWaves)) { WaterMgr->m_NeedInfoUpdate = false; WaterMgr->CreateSuperfancyInfo(); }*/ double time = WaterMgr->m_WaterTexTimer; double period = 8; int curTex = (int)(time*60/period) % 60; int nexTex = (curTex + 1) % 60; float repeatPeriod = WaterMgr->m_RepeatPeriod; // Render normals and foam to a framebuffer if we're in fancy effects if (WaterMgr->m_WaterFancyEffects) { // Save the post-processing framebuffer. GLint fbo; glGetIntegerv(GL_FRAMEBUFFER_BINDING_EXT, &fbo); pglBindFramebufferEXT(GL_FRAMEBUFFER_EXT, WaterMgr->m_FancyEffectsFBO); glDisable(GL_BLEND); glEnable(GL_DEPTH_TEST); glDepthFunc(GL_LEQUAL); glDisable(GL_CULL_FACE); // Overwrite waves that would be behind the ground. CShaderProgramPtr dummyShader = g_Renderer.GetShaderManager().LoadProgram("glsl/gui_solid", CShaderDefines()); dummyShader->Bind(); dummyShader->Uniform(str_transform, g_Renderer.GetViewCamera().GetViewProjection()); dummyShader->Uniform(str_color, 0.0f, 0.0f, 0.0f, 0.0f); std::vector& visiblePatches = m->visiblePatches[cullGroup]; for (size_t i = 0; i < visiblePatches.size(); ++i) { CPatchRData* data = visiblePatches[i]; data->RenderWater(dummyShader, true, true); } dummyShader->Unbind(); glEnable(GL_CULL_FACE); pglBindFramebufferEXT(GL_FRAMEBUFFER_EXT, fbo); } glEnable(GL_BLEND); glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA); glEnable(GL_DEPTH_TEST); glDepthFunc(GL_LEQUAL); m->fancyWaterShader->Bind(); const CCamera& camera = g_Renderer.GetViewCamera(); m->fancyWaterShader->BindTexture(str_normalMap, WaterMgr->m_NormalMap[curTex]); m->fancyWaterShader->BindTexture(str_normalMap2, WaterMgr->m_NormalMap[nexTex]); if (WaterMgr->m_WaterFancyEffects) { m->fancyWaterShader->BindTexture(str_waterEffectsTex, WaterMgr->m_FancyTexture); } if (WaterMgr->m_WaterRefraction && WaterMgr->m_WaterRealDepth) { m->fancyWaterShader->BindTexture(str_depthTex, WaterMgr->m_RefrFboDepthTexture); m->fancyWaterShader->Uniform(str_projInvTransform, WaterMgr->m_RefractionProjInvMatrix); m->fancyWaterShader->Uniform(str_viewInvTransform, WaterMgr->m_RefractionViewInvMatrix); } if (WaterMgr->m_WaterRefraction) m->fancyWaterShader->BindTexture(str_refractionMap, WaterMgr->m_RefractionTexture); if (WaterMgr->m_WaterReflection) m->fancyWaterShader->BindTexture(str_reflectionMap, WaterMgr->m_ReflectionTexture); m->fancyWaterShader->BindTexture(str_losTex, losTexture.GetTextureSmooth()); const CLightEnv& lightEnv = g_Renderer.GetLightEnv(); m->fancyWaterShader->Uniform(str_transform, g_Renderer.GetViewCamera().GetViewProjection()); //TODO: bind only what's needed if (WaterMgr->m_WaterRefraction || WaterMgr->m_WaterReflection) { m->fancyWaterShader->BindTexture(str_skyCube, g_Renderer.GetSkyManager()->GetSkyCube()); // TODO: check that this rotates in the right direction. CMatrix3D skyBoxRotation; skyBoxRotation.SetIdentity(); skyBoxRotation.RotateY(M_PI + lightEnv.GetRotation()); m->fancyWaterShader->Uniform(str_skyBoxRot, skyBoxRotation); if (WaterMgr->m_WaterRefraction) m->fancyWaterShader->Uniform(str_refractionMatrix, WaterMgr->m_RefractionMatrix); if (WaterMgr->m_WaterReflection) m->fancyWaterShader->Uniform(str_reflectionMatrix, WaterMgr->m_ReflectionMatrix); } m->fancyWaterShader->Uniform(str_ambient, lightEnv.m_AmbientColor); m->fancyWaterShader->Uniform(str_sunDir, lightEnv.GetSunDir()); m->fancyWaterShader->Uniform(str_sunColor, lightEnv.m_SunColor); m->fancyWaterShader->Uniform(str_color, WaterMgr->m_WaterColor); m->fancyWaterShader->Uniform(str_tint, WaterMgr->m_WaterTint); m->fancyWaterShader->Uniform(str_waviness, WaterMgr->m_Waviness); m->fancyWaterShader->Uniform(str_murkiness, WaterMgr->m_Murkiness); m->fancyWaterShader->Uniform(str_windAngle, WaterMgr->m_WindAngle); m->fancyWaterShader->Uniform(str_repeatScale, 1.0f / repeatPeriod); m->fancyWaterShader->Uniform(str_losTransform, losTexture.GetTextureMatrix()[0], losTexture.GetTextureMatrix()[12], 0.f, 0.f); m->fancyWaterShader->Uniform(str_cameraPos, camera.GetOrientation().GetTranslation()); m->fancyWaterShader->Uniform(str_fogColor, lightEnv.m_FogColor); m->fancyWaterShader->Uniform(str_fogParams, lightEnv.m_FogFactor, lightEnv.m_FogMax, 0.f, 0.f); m->fancyWaterShader->Uniform(str_time, (float)time); m->fancyWaterShader->Uniform(str_screenSize, (float)g_Renderer.GetWidth(), (float)g_Renderer.GetHeight(), 0.0f, 0.0f); if (WaterMgr->m_WaterType == L"clap") { m->fancyWaterShader->Uniform(str_waveParams1, 30.0f,1.5f,20.0f,0.03f); m->fancyWaterShader->Uniform(str_waveParams2, 0.5f,0.0f,0.0f,0.0f); } else if (WaterMgr->m_WaterType == L"lake") { m->fancyWaterShader->Uniform(str_waveParams1, 8.5f,1.5f,15.0f,0.03f); m->fancyWaterShader->Uniform(str_waveParams2, 0.2f,0.0f,0.0f,0.07f); } else { m->fancyWaterShader->Uniform(str_waveParams1, 15.0f,0.8f,10.0f,0.1f); m->fancyWaterShader->Uniform(str_waveParams2, 0.3f,0.0f,0.1f,0.3f); } if (shadow) shadow->BindTo(m->fancyWaterShader); std::vector& visiblePatches = m->visiblePatches[cullGroup]; for (size_t i = 0; i < visiblePatches.size(); ++i) { CPatchRData* data = visiblePatches[i]; data->RenderWater(m->fancyWaterShader); } m->fancyWaterShader->Unbind(); glDepthFunc(GL_LEQUAL); glDisable(GL_BLEND); return true; } void TerrainRenderer::RenderSimpleWater(int cullGroup) { #if CONFIG2_GLES UNUSED2(cullGroup); #else PROFILE3_GPU("simple water"); WaterManager* WaterMgr = g_Renderer.GetWaterManager(); CLOSTexture& losTexture = g_Game->GetView()->GetLOSTexture(); glEnable(GL_DEPTH_TEST); glDepthFunc(GL_LEQUAL); double time = WaterMgr->m_WaterTexTimer; double period = 1.6f; int curTex = (int)(time*60/period) % 60; CShaderTechniquePtr waterSimpleTech = g_Renderer.GetShaderManager().LoadEffect(str_water_simple); waterSimpleTech->BeginPass(); CShaderProgramPtr waterSimpleShader = waterSimpleTech->GetShader(); waterSimpleShader->Bind(); waterSimpleShader->BindTexture(str_baseTex, WaterMgr->m_WaterTexture[curTex]); waterSimpleShader->BindTexture(str_losTex, losTexture.GetTextureSmooth()); waterSimpleShader->Uniform(str_transform, g_Renderer.GetViewCamera().GetViewProjection()); waterSimpleShader->Uniform(str_losTransform, losTexture.GetTextureMatrix()[0], losTexture.GetTextureMatrix()[12], 0.f, 0.f); waterSimpleShader->Uniform(str_time, static_cast(time)); waterSimpleShader->Uniform(str_color, WaterMgr->m_WaterColor); glEnableClientState(GL_VERTEX_ARRAY); std::vector& visiblePatches = m->visiblePatches[cullGroup]; for (size_t i = 0; i < visiblePatches.size(); ++i) { CPatchRData* data = visiblePatches[i]; data->RenderWater(waterSimpleShader, false, true); } glDisableClientState(GL_VERTEX_ARRAY); waterSimpleShader->Unbind(); g_Renderer.BindTexture(1, 0); pglActiveTextureARB(GL_TEXTURE0_ARB); glDisable(GL_TEXTURE_2D); waterSimpleTech->EndPass(); #endif } /////////////////////////////////////////////////////////////////// // Render water that is part of the terrain void TerrainRenderer::RenderWater(const CShaderDefines& context, int cullGroup, ShadowMap* shadow) { WaterManager* WaterMgr = g_Renderer.GetWaterManager(); WaterMgr->UpdateQuality(); if (!WaterMgr->WillRenderFancyWater()) RenderSimpleWater(cullGroup); else RenderFancyWater(context, cullGroup, shadow); } void TerrainRenderer::RenderPriorities(int cullGroup) { PROFILE("priorities"); ENSURE(m->phase == Phase_Render); CShaderTechniquePtr tech = g_Renderer.GetShaderManager().LoadEffect(str_gui_text); tech->BeginPass(); CTextRenderer textRenderer(tech->GetShader()); textRenderer.Font(CStrIntern("mono-stroke-10")); textRenderer.Color(1.0f, 1.0f, 0.0f); std::vector& visiblePatches = m->visiblePatches[cullGroup]; for (size_t i = 0; i < visiblePatches.size(); ++i) visiblePatches[i]->RenderPriorities(textRenderer); textRenderer.Render(); tech->EndPass(); } Index: ps/trunk/source/renderer/VertexBuffer.cpp =================================================================== --- ps/trunk/source/renderer/VertexBuffer.cpp (revision 24859) +++ ps/trunk/source/renderer/VertexBuffer.cpp (revision 24860) @@ -1,373 +1,378 @@ /* Copyright (C) 2021 Wildfire Games. * This file is part of 0 A.D. * * 0 A.D. is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation, either version 2 of the License, or * (at your option) any later version. * * 0 A.D. is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with 0 A.D. If not, see . */ #include "precompiled.h" #include "VertexBuffer.h" #include "lib/ogl.h" #include "lib/sysdep/cpu.h" #include "Renderer.h" #include "ps/CLogger.h" #include "ps/Errors.h" #include #include // Absolute maximum (bytewise) size of each GL vertex buffer object. // Make it large enough for the maximum feasible mesh size (64K vertexes, // 64 bytes per vertex in InstancingModelRenderer). // TODO: measure what influence this has on performance -#define MAX_VB_SIZE_BYTES (4*1024*1024) +constexpr std::size_t MAX_VB_SIZE_BYTES = 4 * 1024 * 1024; CVertexBuffer::CVertexBuffer(size_t vertexSize, GLenum usage, GLenum target) + : CVertexBuffer(vertexSize, usage, target, MAX_VB_SIZE_BYTES) +{ +} + +CVertexBuffer::CVertexBuffer(size_t vertexSize, GLenum usage, GLenum target, size_t maximumBufferSize) : m_VertexSize(vertexSize), m_Handle(0), m_SysMem(0), m_Usage(usage), m_Target(target), m_HasNeededChunks(false) { - size_t size = MAX_VB_SIZE_BYTES; + size_t size = maximumBufferSize; if (target == GL_ARRAY_BUFFER) // vertex data buffer { // We want to store 16-bit indices to any vertex in a buffer, so the // buffer must never be bigger than vertexSize*64K bytes since we can // address at most 64K of them with 16-bit indices size = std::min(size, vertexSize*65536); } // store max/free vertex counts m_MaxVertices = m_FreeVertices = size / vertexSize; // allocate raw buffer if (g_Renderer.m_Caps.m_VBO) { pglGenBuffersARB(1, &m_Handle); pglBindBufferARB(m_Target, m_Handle); pglBufferDataARB(m_Target, m_MaxVertices * m_VertexSize, 0, m_Usage); pglBindBufferARB(m_Target, 0); } else { m_SysMem = new u8[m_MaxVertices * m_VertexSize]; } // create sole free chunk VBChunk* chunk = new VBChunk; chunk->m_Owner = this; chunk->m_Count = m_FreeVertices; chunk->m_Index = 0; m_FreeList.emplace_back(chunk); } CVertexBuffer::~CVertexBuffer() { // Must have released all chunks before destroying the buffer ENSURE(m_AllocList.empty()); if (m_Handle) pglDeleteBuffersARB(1, &m_Handle); SAFE_ARRAY_DELETE(m_SysMem); for (VBChunk* const& chunk : m_FreeList) delete chunk; } bool CVertexBuffer::CompatibleVertexType(size_t vertexSize, GLenum usage, GLenum target) const { return usage == m_Usage && target == m_Target && vertexSize == m_VertexSize; } /////////////////////////////////////////////////////////////////////////////// // Allocate: try to allocate a buffer of given number of vertices (each of // given size), with the given type, and using the given texture - return null // if no free chunks available CVertexBuffer::VBChunk* CVertexBuffer::Allocate(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore) { // check this is the right kind of buffer if (!CompatibleVertexType(vertexSize, usage, target)) return nullptr; if (UseStreaming(usage)) ENSURE(backingStore != nullptr); // quick check there's enough vertices spare to allocate if (numVertices > m_FreeVertices) return nullptr; // trawl free list looking for first free chunk with enough space std::vector::iterator best_iter = m_FreeList.end(); for (std::vector::iterator iter = m_FreeList.begin(); iter != m_FreeList.end(); ++iter) { if (numVertices == (*iter)->m_Count) { best_iter = iter; break; } else if (numVertices < (*iter)->m_Count && (best_iter == m_FreeList.end() || (*best_iter)->m_Count < (*iter)->m_Count)) best_iter = iter; } // We could not find a large enough chunk. if (best_iter == m_FreeList.end()) return nullptr; VBChunk* chunk = *best_iter; m_FreeList.erase(best_iter); m_FreeVertices -= chunk->m_Count; chunk->m_BackingStore = backingStore; chunk->m_Dirty = false; chunk->m_Needed = false; // split chunk into two; - allocate a new chunk using all unused vertices in the // found chunk, and add it to the free list if (chunk->m_Count > numVertices) { VBChunk* newchunk = new VBChunk; newchunk->m_Owner = this; newchunk->m_Count = chunk->m_Count - numVertices; newchunk->m_Index = chunk->m_Index + numVertices; m_FreeList.emplace_back(newchunk); m_FreeVertices += newchunk->m_Count; // resize given chunk chunk->m_Count = numVertices; } // return found chunk m_AllocList.push_back(chunk); return chunk; } /////////////////////////////////////////////////////////////////////////////// // Release: return given chunk to this buffer void CVertexBuffer::Release(VBChunk* chunk) { // Update total free count before potentially modifying this chunk's count m_FreeVertices += chunk->m_Count; m_AllocList.erase(std::find(m_AllocList.begin(), m_AllocList.end(), chunk)); // Sorting O(nlogn) shouldn't be too far from O(n) by performance, because // the container is partly sorted already. std::sort( m_FreeList.begin(), m_FreeList.end(), [](const VBChunk* chunk1, const VBChunk* chunk2) -> bool { return chunk1->m_Index < chunk2->m_Index; }); // Coalesce with any free-list items that are adjacent to this chunk; // merge the found chunk with the new one, and remove the old one // from the list. for (std::vector::iterator iter = m_FreeList.begin(); iter != m_FreeList.end();) { if ((*iter)->m_Index == chunk->m_Index + chunk->m_Count || (*iter)->m_Index + (*iter)->m_Count == chunk->m_Index) { chunk->m_Index = std::min(chunk->m_Index, (*iter)->m_Index); chunk->m_Count += (*iter)->m_Count; delete *iter; iter = m_FreeList.erase(iter); if (!m_FreeList.empty() && iter != m_FreeList.begin()) iter = std::prev(iter); } else { ++iter; } } m_FreeList.emplace_back(chunk); } /////////////////////////////////////////////////////////////////////////////// // UpdateChunkVertices: update vertex data for given chunk void CVertexBuffer::UpdateChunkVertices(VBChunk* chunk, void* data) { if (g_Renderer.m_Caps.m_VBO) { ENSURE(m_Handle); if (UseStreaming(m_Usage)) { // The VBO is now out of sync with the backing store chunk->m_Dirty = true; // Sanity check: Make sure the caller hasn't tried to reallocate // their backing store ENSURE(data == chunk->m_BackingStore); } else { pglBindBufferARB(m_Target, m_Handle); pglBufferSubDataARB(m_Target, chunk->m_Index * m_VertexSize, chunk->m_Count * m_VertexSize, data); pglBindBufferARB(m_Target, 0); } } else { ENSURE(m_SysMem); memcpy(m_SysMem + chunk->m_Index * m_VertexSize, data, chunk->m_Count * m_VertexSize); } } /////////////////////////////////////////////////////////////////////////////// // Bind: bind to this buffer; return pointer to address required as parameter // to glVertexPointer ( + etc) calls u8* CVertexBuffer::Bind() { if (!g_Renderer.m_Caps.m_VBO) return m_SysMem; pglBindBufferARB(m_Target, m_Handle); if (UseStreaming(m_Usage)) { if (!m_HasNeededChunks) return nullptr; // If any chunks are out of sync with the current VBO, and are // needed for rendering this frame, we'll need to re-upload the VBO bool needUpload = false; for (VBChunk* const& chunk : m_AllocList) { if (chunk->m_Dirty && chunk->m_Needed) { needUpload = true; break; } } if (needUpload) { // Tell the driver that it can reallocate the whole VBO pglBufferDataARB(m_Target, m_MaxVertices * m_VertexSize, NULL, m_Usage); // (In theory, glMapBufferRange with GL_MAP_INVALIDATE_BUFFER_BIT could be used // here instead of glBufferData(..., NULL, ...) plus glMapBuffer(), but with // current Intel Windows GPU drivers (as of 2015-01) it's much faster if you do // the explicit glBufferData.) while (true) { void* p = pglMapBufferARB(m_Target, GL_WRITE_ONLY); if (p == NULL) { // This shouldn't happen unless we run out of virtual address space LOGERROR("glMapBuffer failed"); break; } #ifndef NDEBUG // To help detect bugs where PrepareForRendering() was not called, // force all not-needed data to 0, so things won't get rendered // with undefined (but possibly still correct-looking) data. memset(p, 0, m_MaxVertices * m_VertexSize); #endif // Copy only the chunks we need. (This condition is helpful when // the VBO contains data for every unit in the world, but only a // handful are visible on screen and we don't need to bother copying // the rest.) for (VBChunk* const& chunk : m_AllocList) if (chunk->m_Needed) memcpy((u8 *)p + chunk->m_Index * m_VertexSize, chunk->m_BackingStore, chunk->m_Count * m_VertexSize); if (pglUnmapBufferARB(m_Target) == GL_TRUE) break; // Unmap might fail on e.g. resolution switches, so just try again // and hope it will eventually succeed debug_printf("glUnmapBuffer failed, trying again...\n"); } // Anything we just uploaded is clean; anything else is dirty // since the rest of the VBO content is now undefined for (VBChunk* const& chunk : m_AllocList) { if (chunk->m_Needed) { chunk->m_Dirty = false; chunk->m_Needed = false; } else chunk->m_Dirty = true; } } else { // Reset the flags for the next phase. for (VBChunk* const& chunk : m_AllocList) chunk->m_Needed = false; } m_HasNeededChunks = false; } return nullptr; } u8* CVertexBuffer::GetBindAddress() { if (g_Renderer.m_Caps.m_VBO) return nullptr; else return m_SysMem; } void CVertexBuffer::Unbind() { if (g_Renderer.m_Caps.m_VBO) { pglBindBufferARB(GL_ARRAY_BUFFER, 0); pglBindBufferARB(GL_ELEMENT_ARRAY_BUFFER, 0); } } size_t CVertexBuffer::GetBytesReserved() const { return MAX_VB_SIZE_BYTES; } size_t CVertexBuffer::GetBytesAllocated() const { return (m_MaxVertices - m_FreeVertices) * m_VertexSize; } -void CVertexBuffer::DumpStatus() +void CVertexBuffer::DumpStatus() const { debug_printf("freeverts = %d\n", static_cast(m_FreeVertices)); size_t maxSize = 0; for (VBChunk* const& chunk : m_FreeList) { debug_printf("free chunk %p: size=%d\n", static_cast(chunk), static_cast(chunk->m_Count)); maxSize = std::max(chunk->m_Count, maxSize); } debug_printf("max size = %d\n", static_cast(maxSize)); } bool CVertexBuffer::UseStreaming(GLenum usage) { return usage == GL_DYNAMIC_DRAW || usage == GL_STREAM_DRAW; } void CVertexBuffer::PrepareForRendering(VBChunk* chunk) { chunk->m_Needed = true; m_HasNeededChunks = true; } Index: ps/trunk/source/renderer/VertexBuffer.h =================================================================== --- ps/trunk/source/renderer/VertexBuffer.h (revision 24859) +++ ps/trunk/source/renderer/VertexBuffer.h (revision 24860) @@ -1,163 +1,164 @@ /* Copyright (C) 2021 Wildfire Games. * This file is part of 0 A.D. * * 0 A.D. is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation, either version 2 of the License, or * (at your option) any later version. * * 0 A.D. is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with 0 A.D. If not, see . */ /* * encapsulation of VBOs with batching and sharing */ #ifndef INCLUDED_VERTEXBUFFER #define INCLUDED_VERTEXBUFFER #include "lib/res/graphics/ogl_tex.h" #include /** * CVertexBuffer: encapsulation of ARB_vertex_buffer_object, also supplying * some additional functionality for sharing buffers between multiple objects. * * The class can be used in two modes, depending on the usage parameter: * * GL_STATIC_DRAW: Call Allocate() with backingStore = NULL. Then call * UpdateChunkVertices() with any pointer - the data will be immediately copied * to the VBO. This should be used for vertex data that rarely changes. * * GL_DYNAMIC_DRAW, GL_STREAM_DRAW: Call Allocate() with backingStore pointing * at some memory that will remain valid for the lifetime of the CVertexBuffer. * This should be used for vertex data that may change every frame. * Rendering is expected to occur in two phases: * - "Prepare" phase: * If this chunk is going to be used for rendering during the next Bind phase, * you must call PrepareForRendering(). * If the vertex data in backingStore has been modified since the last Bind phase, * you must call UpdateChunkVertices(). * - "Bind" phase: * Bind() can be called (multiple times). The vertex data will be uploaded * to the GPU if necessary. * It is okay to have multiple prepare/bind cycles per frame (though slightly less * efficient), but they must occur sequentially. */ class CVertexBuffer { NONCOPYABLE(CVertexBuffer); public: /// VBChunk: describes a portion of this vertex buffer struct VBChunk { /// Owning (parent) vertex buffer CVertexBuffer* m_Owner; /// Start index of this chunk in owner size_t m_Index; /// Number of vertices used by chunk size_t m_Count; /// If UseStreaming() is true, points at the data for this chunk void* m_BackingStore; /// If true, the VBO is not consistent with the chunk's backing store /// (and will need to be re-uploaded before rendering with this chunk) bool m_Dirty; /// If true, we have been told this chunk is going to be used for /// rendering in the next bind phase and will need to be uploaded bool m_Needed; private: // Only CVertexBuffer can construct/delete these // (Other people should use g_VBMan.Allocate, g_VBMan.Release) friend class CVertexBuffer; VBChunk() {} ~VBChunk() {} }; public: // constructor, destructor CVertexBuffer(size_t vertexSize, GLenum usage, GLenum target); + CVertexBuffer(size_t vertexSize, GLenum usage, GLenum target, size_t maximumBufferSize); ~CVertexBuffer(); /// Bind to this buffer; return pointer to address required as parameter /// to glVertexPointer ( + etc) calls u8* Bind(); /// Get the address that Bind() will return, without actually binding u8* GetBindAddress(); /// Unbind any currently-bound buffer, so glVertexPointer etc calls will not attempt to use it static void Unbind(); /// Make the vertex data available for the next call to Bind() void PrepareForRendering(VBChunk* chunk); /// Update vertex data for given chunk. Transfers the provided data to the actual OpenGL vertex buffer. void UpdateChunkVertices(VBChunk* chunk, void* data); size_t GetVertexSize() const { return m_VertexSize; } size_t GetBytesReserved() const; size_t GetBytesAllocated() const; /// Returns true if this vertex buffer is compatible with the specified vertex type and intended usage. bool CompatibleVertexType(size_t vertexSize, GLenum usage, GLenum target) const; - void DumpStatus(); + void DumpStatus() const; /** * Given the usage flags of a buffer that has been (or will be) allocated: * * If true, we assume the buffer is going to be modified on every frame, * so we will re-upload the entire buffer every frame using glMapBuffer. * This requires the buffer's owner to hold onto its backing store. * * If false, we assume it will change rarely, and use glSubBufferData to * update it incrementally. The backing store can be freed to save memory. */ static bool UseStreaming(GLenum usage); protected: friend class CVertexBufferManager; // allow allocate only via CVertexBufferManager /// Try to allocate a buffer of given number of vertices (each of given size), /// and with the given type - return null if no free chunks available VBChunk* Allocate(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore); /// Return given chunk to this buffer void Release(VBChunk* chunk); private: /// Vertex size of this vertex buffer size_t m_VertexSize; /// Number of vertices of above size in this buffer size_t m_MaxVertices; /// List of free chunks in this buffer std::vector m_FreeList; /// List of allocated chunks std::vector m_AllocList; /// Available free vertices - total of all free vertices in the free list size_t m_FreeVertices; /// Handle to the actual GL vertex buffer object GLuint m_Handle; /// Raw system memory for systems not supporting VBOs u8* m_SysMem; /// Usage type of the buffer (GL_STATIC_DRAW etc) GLenum m_Usage; /// Buffer target (GL_ARRAY_BUFFER, GL_ELEMENT_ARRAY_BUFFER) GLenum m_Target; bool m_HasNeededChunks; }; #endif Index: ps/trunk/source/renderer/VertexBufferManager.cpp =================================================================== --- ps/trunk/source/renderer/VertexBufferManager.cpp (revision 24859) +++ ps/trunk/source/renderer/VertexBufferManager.cpp (revision 24860) @@ -1,131 +1,169 @@ -/* Copyright (C) 2017 Wildfire Games. +/* Copyright (C) 2021 Wildfire Games. * This file is part of 0 A.D. * * 0 A.D. is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation, either version 2 of the License, or * (at your option) any later version. * * 0 A.D. is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with 0 A.D. If not, see . */ /* * Allocate and destroy CVertexBuffers */ #include "precompiled.h" #include "VertexBufferManager.h" #include "lib/ogl.h" #include "ps/CLogger.h" #define DUMP_VB_STATS 0 // for debugging CVertexBufferManager g_VBMan; -/////////////////////////////////////////////////////////////////////////////// +CVertexBufferManager::Handle::Handle(Handle&& other) + : m_Chunk(other.m_Chunk) +{ + other.m_Chunk = nullptr; +} + +CVertexBufferManager::Handle& CVertexBufferManager::Handle::operator=(Handle&& other) +{ + if (&other == this) + return *this; + + if (IsValid()) + Reset(); + + Handle tmp(std::move(other)); + swap(*this, tmp); + + return *this; +} + +CVertexBufferManager::Handle::Handle(CVertexBuffer::VBChunk* chunk) + : m_Chunk(chunk) +{ +} + +void CVertexBufferManager::Handle::Reset() +{ + if (!IsValid()) + return; + + g_VBMan.Release(m_Chunk); + m_Chunk = nullptr; +} + // Explicit shutdown of the vertex buffer subsystem. // This avoids the ordering issues that arise when using destructors of // global instances. void CVertexBufferManager::Shutdown() { - typedef std::list::iterator Iter; - for (Iter iter = m_Buffers.begin(); iter != m_Buffers.end(); ++iter) - delete *iter; - m_Buffers.clear(); + for (int group = static_cast(Group::DEFAULT); group < static_cast(Group::COUNT); ++group) + m_Buffers[group].clear(); +} + +CVertexBuffer::VBChunk* CVertexBufferManager::Allocate(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore) +{ + return AllocateImpl(vertexSize, numVertices, usage, target, backingStore, Group::DEFAULT); } +CVertexBufferManager::Handle CVertexBufferManager::AllocateChunk(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore, Group group) +{ + CVertexBuffer::VBChunk* chunk = AllocateImpl(vertexSize, numVertices, usage, target, backingStore, group); + if (!chunk) + return Handle(); + return Handle(chunk); +} -/////////////////////////////////////////////////////////////////////////////// // Allocate: try to allocate a buffer of given number of vertices (each of // given size), with the given type, and using the given texture - return null // if no free chunks available -CVertexBuffer::VBChunk* CVertexBufferManager::Allocate(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore) +CVertexBuffer::VBChunk* CVertexBufferManager::AllocateImpl(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore, Group group) { - CVertexBuffer::VBChunk* result=0; + CVertexBuffer::VBChunk* result = nullptr; ENSURE(usage == GL_STREAM_DRAW || usage == GL_STATIC_DRAW || usage == GL_DYNAMIC_DRAW); ENSURE(target == GL_ARRAY_BUFFER || target == GL_ELEMENT_ARRAY_BUFFER); if (CVertexBuffer::UseStreaming(usage)) ENSURE(backingStore != NULL); // TODO, RC - run some sanity checks on allocation request - typedef std::list::iterator Iter; + std::list>& buffers = m_Buffers[static_cast(group)]; #if DUMP_VB_STATS debug_printf("\n============================\n# allocate vsize=%zu nverts=%zu\n\n", vertexSize, numVertices); - for (Iter iter = m_Buffers.begin(); iter != m_Buffers.end(); ++iter) { - CVertexBuffer* buffer = *iter; + for (const std::unique_ptr& buffer : buffers) + { if (buffer->CompatibleVertexType(vertexSize, usage, target)) { - debug_printf("%p\n", buffer); + debug_printf("%p\n", buffer.get()); buffer->DumpStatus(); } } #endif // iterate through all existing buffers testing for one that'll // satisfy the allocation - for (Iter iter = m_Buffers.begin(); iter != m_Buffers.end(); ++iter) { - CVertexBuffer* buffer = *iter; + for (const std::unique_ptr& buffer : buffers) + { result = buffer->Allocate(vertexSize, numVertices, usage, target, backingStore); if (result) return result; } // got this far; need to allocate a new buffer - CVertexBuffer* buffer = new CVertexBuffer(vertexSize, usage, target); - m_Buffers.push_front(buffer); - result = buffer->Allocate(vertexSize, numVertices, usage, target, backingStore); + buffers.emplace_back( + group == Group::DEFAULT + ? std::make_unique(vertexSize, usage, target) + // Reduces the buffer size for not so frequent buffers. + : std::make_unique(vertexSize, usage, target, 1024 * 1024)); + result = buffers.back()->Allocate(vertexSize, numVertices, usage, target, backingStore); if (!result) { LOGERROR("Failed to create VBOs (%zu*%zu)", vertexSize, numVertices); } return result; } -/////////////////////////////////////////////////////////////////////////////// -// Release: return given chunk to its owner void CVertexBufferManager::Release(CVertexBuffer::VBChunk* chunk) { ENSURE(chunk); #if DUMP_VB_STATS debug_printf("\n============================\n# release %p nverts=%zu\n\n", chunk, chunk->m_Count); #endif chunk->m_Owner->Release(chunk); } - -size_t CVertexBufferManager::GetBytesReserved() +size_t CVertexBufferManager::GetBytesReserved() const { size_t total = 0; - - typedef std::list::iterator Iter; - for (Iter iter = m_Buffers.begin(); iter != m_Buffers.end(); ++iter) - total += (*iter)->GetBytesReserved(); - + for (int group = static_cast(Group::DEFAULT); group < static_cast(Group::COUNT); ++group) + for (const std::unique_ptr& buffer : m_Buffers[static_cast(group)]) + total += buffer->GetBytesReserved(); return total; } -size_t CVertexBufferManager::GetBytesAllocated() +size_t CVertexBufferManager::GetBytesAllocated() const { size_t total = 0; - - typedef std::list::iterator Iter; - for (Iter iter = m_Buffers.begin(); iter != m_Buffers.end(); ++iter) - total += (*iter)->GetBytesAllocated(); - + for (int group = static_cast(Group::DEFAULT); group < static_cast(Group::COUNT); ++group) + for (const std::unique_ptr& buffer : m_Buffers[static_cast(group)]) + total += buffer->GetBytesAllocated(); return total; } Index: ps/trunk/source/renderer/VertexBufferManager.h =================================================================== --- ps/trunk/source/renderer/VertexBufferManager.h (revision 24859) +++ ps/trunk/source/renderer/VertexBufferManager.h (revision 24860) @@ -1,67 +1,114 @@ -/* Copyright (C) 2012 Wildfire Games. +/* Copyright (C) 2021 Wildfire Games. * This file is part of 0 A.D. * * 0 A.D. is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation, either version 2 of the License, or * (at your option) any later version. * * 0 A.D. is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with 0 A.D. If not, see . */ /* * Allocate and destroy CVertexBuffers */ #ifndef INCLUDED_VERTEXBUFFERMANAGER #define INCLUDED_VERTEXBUFFERMANAGER -#include "VertexBuffer.h" +#include "lib/types.h" +#include "renderer/VertexBuffer.h" + +#include +#include -/////////////////////////////////////////////////////////////////////////////// // CVertexBufferManager: owner object for CVertexBuffer objects; acts as // 'front end' for their allocation and destruction class CVertexBufferManager { public: + enum class Group : u32 + { + DEFAULT, + TERRAIN, + WATER, + COUNT + }; + + // Helper for automatic VBChunk lifetime management. + class Handle + { + public: + Handle() = default; + Handle(const Handle&) = delete; + Handle& operator=(const Handle&) = delete; + + explicit Handle(CVertexBuffer::VBChunk* chunk); + Handle(Handle&& other); + Handle& operator=(Handle&& other); + + ~Handle() { Reset(); } + + bool IsValid() const { return m_Chunk != nullptr; } + explicit operator bool() const { return IsValid(); } + bool operator!() const { return !static_cast(*this); } + void Reset(); + + friend void swap(Handle& lhs, Handle& rhs) + { + using std::swap; + + swap(lhs.m_Chunk, rhs.m_Chunk); + } + + CVertexBuffer::VBChunk& operator*() const { return *m_Chunk; } + CVertexBuffer::VBChunk* operator->() const { return m_Chunk; } + CVertexBuffer::VBChunk* Get() const { return m_Chunk; } + + private: + CVertexBuffer::VBChunk* m_Chunk = nullptr; + }; /** * Try to allocate a vertex buffer of the given size and type. * * @param vertexSize size of each vertex in the buffer * @param numVertices number of vertices in the buffer * @param usage GL_STATIC_DRAW, GL_DYNAMIC_DRAW, GL_STREAM_DRAW * @param target typically GL_ARRAY_BUFFER or GL_ELEMENT_ARRAY_BUFFER * @param backingStore if usage is STATIC, this is NULL; else for DYNAMIC/STREAM, * this must be a copy of the vertex data that remains valid for the * lifetime of the VBChunk * @return chunk, or NULL if no free chunks available */ - CVertexBuffer::VBChunk* Allocate(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore = NULL); + CVertexBuffer::VBChunk* Allocate(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore = nullptr); /// Returns the given @p chunk to its owning buffer void Release(CVertexBuffer::VBChunk* chunk); - /// Returns a list of all buffers - const std::list& GetBufferList() const { return m_Buffers; } + // Same as the Allocate function but returns Handle. Should be used instead + // of the Allocate. + Handle AllocateChunk(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore = nullptr, Group group = Group::DEFAULT); - size_t GetBytesReserved(); - size_t GetBytesAllocated(); + size_t GetBytesReserved() const; + size_t GetBytesAllocated() const; /// Explicit shutdown of the vertex buffer subsystem; releases all currently-allocated buffers. void Shutdown(); private: + CVertexBuffer::VBChunk* AllocateImpl(size_t vertexSize, size_t numVertices, GLenum usage, GLenum target, void* backingStore = nullptr, Group group = Group::DEFAULT); + /// List of all known vertex buffers - std::list m_Buffers; + std::list> m_Buffers[static_cast(Group::COUNT)]; }; extern CVertexBufferManager g_VBMan; #endif