diff options
Diffstat (limited to 'indra/llrender/llvertexbuffer.cpp')
| -rw-r--r-- | indra/llrender/llvertexbuffer.cpp | 321 |
1 files changed, 231 insertions, 90 deletions
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 8b5503229f..80752231d7 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -38,6 +38,10 @@ #include "llglslshader.h" #include "llmemory.h" +#if LL_DARWIN +#define LL_VBO_POOLING 1 +#else +#endif //Next Highest Power Of Two //helper function, returns first number > v that is a power of 2, or v if v is already a power of 2 U32 nhpo2(U32 v) @@ -49,6 +53,37 @@ U32 nhpo2(U32 v) return r; } +//which power of 2 is i? +//assumes i is a power of 2 > 0 +U32 wpo2(U32 i) +{ + llassert(i > 0); + llassert(nhpo2(i) == i); + + U32 r = 0; + + while (i >>= 1) ++r; + + return r; +} + + +const U32 LL_VBO_BLOCK_SIZE = 2048; +const U32 LL_VBO_POOL_MAX_SEED_SIZE = 256*1024; + +U32 vbo_block_size(U32 size) +{ //what block size will fit size? + U32 mod = size % LL_VBO_BLOCK_SIZE; + return mod == 0 ? size : size + (LL_VBO_BLOCK_SIZE-mod); +} + +U32 vbo_block_index(U32 size) +{ + return vbo_block_size(size)/LL_VBO_BLOCK_SIZE; +} + +const U32 LL_VBO_POOL_SEED_COUNT = vbo_block_index(LL_VBO_POOL_MAX_SEED_SIZE); + //============================================================================ @@ -57,7 +92,16 @@ LLVBOPool LLVertexBuffer::sStreamVBOPool(GL_STREAM_DRAW_ARB, GL_ARRAY_BUFFER_ARB LLVBOPool LLVertexBuffer::sDynamicVBOPool(GL_DYNAMIC_DRAW_ARB, GL_ARRAY_BUFFER_ARB); LLVBOPool LLVertexBuffer::sStreamIBOPool(GL_STREAM_DRAW_ARB, GL_ELEMENT_ARRAY_BUFFER_ARB); LLVBOPool LLVertexBuffer::sDynamicIBOPool(GL_DYNAMIC_DRAW_ARB, GL_ELEMENT_ARRAY_BUFFER_ARB); + U32 LLVBOPool::sBytesPooled = 0; +U32 LLVBOPool::sIndexBytesPooled = 0; +U32 LLVBOPool::sCurGLName = 1; + +std::list<U32> LLVertexBuffer::sAvailableVAOName; +U32 LLVertexBuffer::sCurVAOName = 1; + +U32 LLVertexBuffer::sAllocatedIndexBytes = 0; +U32 LLVertexBuffer::sIndexCount = 0; LLPrivateMemoryPool* LLVertexBuffer::sPrivatePoolp = NULL; U32 LLVertexBuffer::sBindCount = 0; @@ -74,99 +118,88 @@ U32 LLVertexBuffer::sLastMask = 0; bool LLVertexBuffer::sVBOActive = false; bool LLVertexBuffer::sIBOActive = false; U32 LLVertexBuffer::sAllocatedBytes = 0; +U32 LLVertexBuffer::sVertexCount = 0; bool LLVertexBuffer::sMapped = false; bool LLVertexBuffer::sUseStreamDraw = true; bool LLVertexBuffer::sUseVAO = false; bool LLVertexBuffer::sPreferStreamDraw = false; -const U32 FENCE_WAIT_TIME_NANOSECONDS = 10000; //1 ms -class LLGLSyncFence : public LLGLFence +U32 LLVBOPool::genBuffer() { -public: -#ifdef GL_ARB_sync - GLsync mSync; -#endif - - LLGLSyncFence() - { -#ifdef GL_ARB_sync - mSync = 0; -#endif - } + U32 ret = 0; - virtual ~LLGLSyncFence() + if (mGLNamePool.empty()) { -#ifdef GL_ARB_sync - if (mSync) - { - glDeleteSync(mSync); - } -#endif + ret = sCurGLName++; } - - void placeFence() + else { -#ifdef GL_ARB_sync - if (mSync) - { - glDeleteSync(mSync); - } - mSync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); -#endif + ret = mGLNamePool.front(); + mGLNamePool.pop_front(); } - void wait() - { -#ifdef GL_ARB_sync - if (mSync) - { - while (glClientWaitSync(mSync, 0, FENCE_WAIT_TIME_NANOSECONDS) == GL_TIMEOUT_EXPIRED) - { //track the number of times we've waited here - static S32 waits = 0; - waits++; - } - } -#endif - } + return ret; +} +void LLVBOPool::deleteBuffer(U32 name) +{ + if (gGLManager.mInited) + { + LLVertexBuffer::unbind(); -}; + glBindBufferARB(mType, name); + glBufferDataARB(mType, 0, NULL, mUsage); + llassert(std::find(mGLNamePool.begin(), mGLNamePool.end(), name) == mGLNamePool.end()); -//which power of 2 is i? -//assumes i is a power of 2 > 0 -U32 wpo2(U32 i) -{ - llassert(i > 0); - llassert(nhpo2(i) == i); + mGLNamePool.push_back(name); - U32 r = 0; + glBindBufferARB(mType, 0); + } +} - while (i >>= 1) ++r; - return r; +LLVBOPool::LLVBOPool(U32 vboUsage, U32 vboType) +: mUsage(vboUsage), mType(vboType) +{ + mMissCount.resize(LL_VBO_POOL_SEED_COUNT); + std::fill(mMissCount.begin(), mMissCount.end(), 0); } -volatile U8* LLVBOPool::allocate(U32& name, U32 size) +volatile U8* LLVBOPool::allocate(U32& name, U32 size, bool for_seed) { - llassert(nhpo2(size) == size); + llassert(vbo_block_size(size) == size); + + volatile U8* ret = NULL; - U32 i = wpo2(size); + U32 i = vbo_block_index(size); if (mFreeList.size() <= i) { mFreeList.resize(i+1); } - volatile U8* ret = NULL; - - if (mFreeList[i].empty()) + if (mFreeList[i].empty() || for_seed) { //make a new buffer - glGenBuffersARB(1, &name); + name = genBuffer(); + glBindBufferARB(mType, name); - LLVertexBuffer::sAllocatedBytes += size; + + if (!for_seed && i < LL_VBO_POOL_SEED_COUNT) + { //record this miss + mMissCount[i]++; + } + + if (mType == GL_ARRAY_BUFFER_ARB) + { + LLVertexBuffer::sAllocatedBytes += size; + } + else + { + LLVertexBuffer::sAllocatedIndexBytes += size; + } if (LLVertexBuffer::sDisableVBOMapping || mUsage != GL_DYNAMIC_DRAW_ARB) { @@ -179,13 +212,39 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size) } glBindBufferARB(mType, 0); + + if (for_seed) + { //put into pool for future use + llassert(mFreeList.size() > i); + + Record rec; + rec.mGLName = name; + rec.mClientData = ret; + + if (mType == GL_ARRAY_BUFFER_ARB) + { + sBytesPooled += size; + } + else + { + sIndexBytesPooled += size; + } + mFreeList[i].push_back(rec); + } } else { name = mFreeList[i].front().mGLName; ret = mFreeList[i].front().mClientData; - sBytesPooled -= size; + if (mType == GL_ARRAY_BUFFER_ARB) + { + sBytesPooled -= size; + } + else + { + sIndexBytesPooled -= size; + } mFreeList[i].pop_front(); } @@ -195,30 +254,49 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size) void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size) { - llassert(nhpo2(size) == size); - - U32 i = wpo2(size); + llassert(vbo_block_size(size) == size); - llassert(mFreeList.size() > i); + deleteBuffer(name); + ll_aligned_free_16((U8*) buffer); - Record rec; - rec.mGLName = name; - rec.mClientData = buffer; - - if (buffer == NULL) + if (mType == GL_ARRAY_BUFFER_ARB) { - glDeleteBuffersARB(1, &rec.mGLName); + LLVertexBuffer::sAllocatedBytes -= size; } else { - sBytesPooled += size; - mFreeList[i].push_back(rec); + LLVertexBuffer::sAllocatedIndexBytes -= size; + } +} + +void LLVBOPool::seedPool() +{ + U32 dummy_name = 0; + + if (mFreeList.size() < LL_VBO_POOL_SEED_COUNT) + { + mFreeList.resize(LL_VBO_POOL_SEED_COUNT); + } + + for (U32 i = 0; i < LL_VBO_POOL_SEED_COUNT; i++) + { + if (mMissCount[i] > mFreeList[i].size()) + { + U32 size = i*LL_VBO_BLOCK_SIZE; + + S32 count = mMissCount[i] - mFreeList[i].size(); + for (U32 j = 0; j < count; ++j) + { + allocate(dummy_name, size, true); + } + } } } + void LLVBOPool::cleanup() { - U32 size = 1; + U32 size = LL_VBO_BLOCK_SIZE; for (U32 i = 0; i < mFreeList.size(); ++i) { @@ -228,8 +306,8 @@ void LLVBOPool::cleanup() { Record& r = l.front(); - glDeleteBuffersARB(1, &r.mGLName); - + deleteBuffer(r.mGLName); + if (r.mClientData) { ll_aligned_free_16((void*) r.mClientData); @@ -237,12 +315,23 @@ void LLVBOPool::cleanup() l.pop_front(); - LLVertexBuffer::sAllocatedBytes -= size; - sBytesPooled -= size; + if (mType == GL_ARRAY_BUFFER_ARB) + { + sBytesPooled -= size; + LLVertexBuffer::sAllocatedBytes -= size; + } + else + { + sIndexBytesPooled -= size; + LLVertexBuffer::sAllocatedIndexBytes -= size; + } } - size *= 2; + size += LL_VBO_BLOCK_SIZE; } + + //reset miss counts + std::fill(mMissCount.begin(), mMissCount.end(), 0); } @@ -276,6 +365,41 @@ U32 LLVertexBuffer::sGLMode[LLRender::NUM_MODES] = GL_LINE_LOOP, }; +//static +U32 LLVertexBuffer::getVAOName() +{ + U32 ret = 0; + + if (!sAvailableVAOName.empty()) + { + ret = sAvailableVAOName.front(); + sAvailableVAOName.pop_front(); + } + else + { +#ifdef GL_ARB_vertex_array_object + glGenVertexArrays(1, &ret); +#endif + } + + return ret; +} + +//static +void LLVertexBuffer::releaseVAOName(U32 name) +{ + sAvailableVAOName.push_back(name); +} + + +//static +void LLVertexBuffer::seedPools() +{ + sStreamVBOPool.seedPool(); + sDynamicVBOPool.seedPool(); + sStreamIBOPool.seedPool(); + sDynamicIBOPool.seedPool(); +} //static void LLVertexBuffer::setupClientArrays(U32 data_mask) @@ -885,7 +1009,7 @@ LLVertexBuffer::~LLVertexBuffer() if (mGLArray) { #if GL_ARB_vertex_array_object - glDeleteVertexArrays(1, &mGLArray); + releaseVAOName(mGLArray); #endif } @@ -898,6 +1022,9 @@ LLVertexBuffer::~LLVertexBuffer() mFence = NULL; + sVertexCount -= mNumVerts; + sIndexCount -= mNumIndices; + llassert_always(!mMappedData && !mMappedIndexData); }; @@ -929,7 +1056,7 @@ void LLVertexBuffer::waitFence() const void LLVertexBuffer::genBuffer(U32 size) { - mSize = nhpo2(size); + mSize = vbo_block_size(size); if (mUsage == GL_STREAM_DRAW_ARB) { @@ -945,7 +1072,7 @@ void LLVertexBuffer::genBuffer(U32 size) void LLVertexBuffer::genIndices(U32 size) { - mIndicesSize = nhpo2(size); + mIndicesSize = vbo_block_size(size); if (mUsage == GL_STREAM_DRAW_ARB) { @@ -1108,10 +1235,10 @@ void LLVertexBuffer::updateNumVerts(S32 nverts) llassert(nverts >= 0); - if (nverts >= 65535) + if (nverts > 65536) { llwarns << "Vertex buffer overflow!" << llendl; - nverts = 65535; + nverts = 65536; } U32 needed_size = calcOffsets(mTypeMask, mOffsets, nverts); @@ -1121,7 +1248,9 @@ void LLVertexBuffer::updateNumVerts(S32 nverts) createGLBuffer(needed_size); } + sVertexCount -= mNumVerts; mNumVerts = nverts; + sVertexCount += mNumVerts; } void LLVertexBuffer::updateNumIndices(S32 nindices) @@ -1137,7 +1266,9 @@ void LLVertexBuffer::updateNumIndices(S32 nindices) createGLIndices(needed_size); } + sIndexCount -= mNumIndices; mNumIndices = nindices; + sIndexCount += mNumIndices; } void LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create) @@ -1163,7 +1294,7 @@ void LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create) if (gGLManager.mHasVertexArrayObject && useVBOs() && (LLRender::sGLCoreProfile || sUseVAO)) { #if GL_ARB_vertex_array_object - glGenVertexArrays(1, &mGLArray); + mGLArray = getVAOName(); #endif setupVertexArray(); } @@ -1199,7 +1330,7 @@ void LLVertexBuffer::setupVertexArray() 1, //TYPE_WEIGHT, 4, //TYPE_WEIGHT4, 4, //TYPE_CLOTHWEIGHT, - 4, //TYPE_TEXTURE_INDEX + 1, //TYPE_TEXTURE_INDEX }; U32 attrib_type[] = @@ -1216,7 +1347,7 @@ void LLVertexBuffer::setupVertexArray() GL_FLOAT, //TYPE_WEIGHT, GL_FLOAT, //TYPE_WEIGHT4, GL_FLOAT, //TYPE_CLOTHWEIGHT, - GL_UNSIGNED_BYTE, //TYPE_TEXTURE_INDEX + GL_UNSIGNED_INT, //TYPE_TEXTURE_INDEX }; bool attrib_integer[] = @@ -2033,6 +2164,16 @@ void LLVertexBuffer::flush() } } +// bind for transform feedback (quick 'n dirty) +void LLVertexBuffer::bindForFeedback(U32 channel, U32 type, U32 index, U32 count) +{ +#ifdef GL_TRANSFORM_FEEDBACK_BUFFER + U32 offset = mOffsets[type] + sTypeSize[type]*index; + U32 size= (sTypeSize[type]*count); + glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, channel, mGLBuffer, offset, size); +#endif +} + // Set for rendering void LLVertexBuffer::setBuffer(U32 data_mask) { @@ -2184,10 +2325,10 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask) stop_glerror(); volatile U8* base = useVBOs() ? (U8*) mAlignedOffset : mMappedData; - /*if ((data_mask & mTypeMask) != data_mask) + if (gDebugGL && ((data_mask & mTypeMask) != data_mask)) { llerrs << "LLVertexBuffer::setupVertexBuffer missing required components for supplied data mask." << llendl; - }*/ + } if (LLGLSLShader::sNoFixedFunction) { @@ -2263,7 +2404,7 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask) #if !LL_DARWIN S32 loc = TYPE_TEXTURE_INDEX; void *ptr = (void*) (base + mOffsets[TYPE_VERTEX] + 12); - glVertexAttribIPointer(loc, 4, GL_UNSIGNED_BYTE, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); + glVertexAttribIPointer(loc, 1, GL_UNSIGNED_INT, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); #endif } if (data_mask & MAP_VERTEX) |