From 8bcb176a19297bca4f55f03a5244a507f8bf9174 Mon Sep 17 00:00:00 2001 From: madmaxoft Date: Sat, 12 Apr 2014 00:04:50 +0200 Subject: [PATCH] Lighting reads blocktypes only for blocks under heightmap. This should theoretically speed it up, since less data is copied back and forth. Also implemented a possibly more cache-friendly blocklight starter algorithm (PrepareBlockLight2()), is disabled by default, needs perf testing. --- src/LightingThread.cpp | 92 +++++++++++++++++++++++++++++++++++------- src/LightingThread.h | 11 ++++- 2 files changed, 87 insertions(+), 16 deletions(-) diff --git a/src/LightingThread.cpp b/src/LightingThread.cpp index 302473d71..f23f0c5f4 100644 --- a/src/LightingThread.cpp +++ b/src/LightingThread.cpp @@ -27,7 +27,8 @@ class cReader : ROW * OutputRows = (ROW *)m_BlockTypes; int InputIdx = 0; int OutputIdx = m_ReadingChunkX + m_ReadingChunkZ * cChunkDef::Width * 3; - for (int y = 0; y < cChunkDef::Height; y++) + int MaxHeight = std::min(cChunkDef::Height, m_MaxHeight + 16); // Need 16 blocks above the highest + for (int y = 0; y < MaxHeight; y++) { for (int z = 0; z < cChunkDef::Width; z++) { @@ -43,6 +44,7 @@ class cReader : virtual void HeightMap(const cChunkDef::HeightMap * a_Heightmap) override { + // Copy the entire heightmap, distribute it into the 3x3 chunk blob: typedef struct {HEIGHTTYPE m_Row[16]; } ROW; ROW * InputRows = (ROW *)a_Heightmap; ROW * OutputRows = (ROW *)m_HeightMap; @@ -53,13 +55,32 @@ class cReader : OutputRows[OutputIdx] = InputRows[InputIdx++]; OutputIdx += 3; } // for z + + // Find the highest block in the entire chunk, use it as a base for m_MaxHeight: + HEIGHTTYPE MaxHeight = m_MaxHeight; + for (size_t i = 0; i < ARRAYCOUNT(*a_Heightmap); i++) + { + if ((*a_Heightmap)[i] > MaxHeight) + { + MaxHeight = (*a_Heightmap)[i]; + } + } + m_MaxHeight = MaxHeight; } public: int m_ReadingChunkX; // 0, 1 or 2; x-offset of the chunk we're reading from the BlockTypes start int m_ReadingChunkZ; // 0, 1 or 2; z-offset of the chunk we're reading from the BlockTypes start + HEIGHTTYPE m_MaxHeight; // Maximum value in this chunk's heightmap BLOCKTYPE * m_BlockTypes; // 3x3 chunks of block types, organized as a single XZY blob of data (instead of 3x3 XZY blobs) HEIGHTTYPE * m_HeightMap; // 3x3 chunks of height map, organized as a single XZY blob of data (instead of 3x3 XZY blobs) + + cReader(BLOCKTYPE * a_BlockTypes, HEIGHTTYPE * a_HeightMap) : + m_BlockTypes(a_BlockTypes), + m_HeightMap(a_HeightMap), + m_MaxHeight(0) + { + } } ; @@ -225,7 +246,7 @@ void cLightingThread::LightChunk(cLightingChunkStay & a_Item) // DEBUG: Save chunk data with highlighted seeds for visual inspection: cFile f4; if ( - f4.Open(Printf("Chunk_%d_%d_seeds.grab", a_Item.x, a_Item.z), cFile::fmWrite) + f4.Open(Printf("Chunk_%d_%d_seeds.grab", a_Item.m_ChunkX, a_Item.m_ChunkZ), cFile::fmWrite) ) { for (int z = 0; z < cChunkDef::Width * 3; z++) @@ -244,6 +265,7 @@ void cLightingThread::LightChunk(cLightingChunkStay & a_Item) f4.Write(Seeds, cChunkDef::Width * 3); } } + f4.Close(); } //*/ @@ -253,9 +275,9 @@ void cLightingThread::LightChunk(cLightingChunkStay & a_Item) // DEBUG: Save XY slices of the chunk data and lighting for visual inspection: cFile f1, f2, f3; if ( - f1.Open(Printf("Chunk_%d_%d_data.grab", a_Item.x, a_Item.z), cFile::fmWrite) && - f2.Open(Printf("Chunk_%d_%d_sky.grab", a_Item.x, a_Item.z), cFile::fmWrite) && - f3.Open(Printf("Chunk_%d_%d_glow.grab", a_Item.x, a_Item.z), cFile::fmWrite) + f1.Open(Printf("Chunk_%d_%d_data.grab", a_Item.m_ChunkX, a_Item.m_ChunkZ), cFile::fmWrite) && + f2.Open(Printf("Chunk_%d_%d_sky.grab", a_Item.m_ChunkX, a_Item.m_ChunkZ), cFile::fmWrite) && + f3.Open(Printf("Chunk_%d_%d_glow.grab", a_Item.m_ChunkX, a_Item.m_ChunkZ), cFile::fmWrite) ) { for (int z = 0; z < cChunkDef::Width * 3; z++) @@ -274,6 +296,9 @@ void cLightingThread::LightChunk(cLightingChunkStay & a_Item) f3.Write(BlockLight, cChunkDef::Width * 3); } } + f1.Close(); + f2.Close(); + f3.Close(); } //*/ @@ -293,11 +318,9 @@ void cLightingThread::LightChunk(cLightingChunkStay & a_Item) -bool cLightingThread::ReadChunks(int a_ChunkX, int a_ChunkZ) +void cLightingThread::ReadChunks(int a_ChunkX, int a_ChunkZ) { - cReader Reader; - Reader.m_BlockTypes = m_BlockTypes; - Reader.m_HeightMap = m_HeightMap; + cReader Reader(m_BlockTypes, m_HeightMap); for (int z = 0; z < 3; z++) { @@ -305,16 +328,13 @@ bool cLightingThread::ReadChunks(int a_ChunkX, int a_ChunkZ) for (int x = 0; x < 3; x++) { Reader.m_ReadingChunkX = x; - if (!m_World->GetChunkData(a_ChunkX + x - 1, a_ChunkZ + z - 1, Reader)) - { - return false; - } + VERIFY(m_World->GetChunkData(a_ChunkX + x - 1, a_ChunkZ + z - 1, Reader)); } // for z } // for x memset(m_BlockLight, 0, sizeof(m_BlockLight)); memset(m_SkyLight, 0, sizeof(m_SkyLight)); - return true; + m_MaxHeight = Reader.m_MaxHeight; } @@ -405,6 +425,50 @@ void cLightingThread::PrepareBlockLight(void) +void cLightingThread::PrepareBlockLight2(void) +{ + // Clear seeds: + memset(m_IsSeed1, 0, sizeof(m_IsSeed1)); + memset(m_IsSeed2, 0, sizeof(m_IsSeed2)); + m_NumSeeds = 0; + + // Add each emissive block into the seeds: + for (int y = 0; y < m_MaxHeight; y++) + { + int BaseY = y * BlocksPerYLayer; // Partial offset into m_BlockTypes for the Y coord + for (int z = 1; z < cChunkDef::Width * 3 - 1; z++) + { + int HBaseZ = z * cChunkDef::Width * 3; // Partial offset into m_Heightmap for the Z coord + int BaseZ = BaseY + HBaseZ; // Partial offset into m_BlockTypes for the Y and Z coords + for (int x = 1; x < cChunkDef::Width * 3 - 1; x++) + { + int idx = BaseZ + x; + if (y > m_HeightMap[HBaseZ + x]) + { + // We're above the heightmap, ignore the block + continue; + } + if (cBlockInfo::GetLightValue(m_BlockTypes[idx]) == 0) + { + // Not a light-emissive block + continue; + } + + // Add current block as a seed: + m_IsSeed1[idx] = true; + m_SeedIdx1[m_NumSeeds++] = idx; + + // Light it up: + m_BlockLight[idx] = cBlockInfo::GetLightValue(m_BlockTypes[idx]); + } + } + } +} + + + + + void cLightingThread::CalcLight(NIBBLETYPE * a_Light) { int NumSeeds2 = 0; diff --git a/src/LightingThread.h b/src/LightingThread.h index 770ae809f..a484fcbed 100644 --- a/src/LightingThread.h +++ b/src/LightingThread.h @@ -108,6 +108,9 @@ protected: cEvent m_evtItemAdded; // Set when queue is appended, or to stop the thread cEvent m_evtQueueEmpty; // Set when the queue gets empty + /** The highest block in the current 3x3 chunk data */ + HEIGHTTYPE m_MaxHeight; + // Buffers for the 3x3 chunk data // These buffers alone are 1.7 MiB in size, therefore they cannot be located on the stack safely - some architectures may have only 1 MiB for stack, or even less @@ -136,8 +139,8 @@ protected: /** Lights the entire chunk. If neighbor chunks don't exist, touches them and re-queues the chunk */ void LightChunk(cLightingChunkStay & a_Item); - /** Prepares m_BlockTypes and m_HeightMap data; returns false if any of the chunks fail. Zeroes out the light arrays */ - bool ReadChunks(int a_ChunkX, int a_ChunkZ); + /** Prepares m_BlockTypes and m_HeightMap data; zeroes out the light arrays */ + void ReadChunks(int a_ChunkX, int a_ChunkZ); /** Uses m_HeightMap to initialize the m_SkyLight[] data; fills in seeds for the skylight */ void PrepareSkyLight(void); @@ -145,6 +148,10 @@ protected: /** Uses m_BlockTypes to initialize the m_BlockLight[] data; fills in seeds for the blocklight */ void PrepareBlockLight(void); + /** Same as PrepareBlockLight(), but uses a different traversal scheme; possibly better perf cache-wise. + To be compared in perf benchmarks. */ + void PrepareBlockLight2(void); + /** Calculates light in the light array specified, using stored seeds */ void CalcLight(NIBBLETYPE * a_Light);