diff options
Diffstat (limited to 'indra/llrender/llvertexbuffer.cpp')
| -rw-r--r-- | indra/llrender/llvertexbuffer.cpp | 273 |
1 files changed, 152 insertions, 121 deletions
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 823c6b9dc5..eadef93c89 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -38,12 +38,6 @@ #include "llglslshader.h" #include "llmemory.h" -#if LL_DARWIN -#define LL_VBO_POOLING 1 -#else -#define LL_VBO_POOLING 0 -#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) @@ -71,6 +65,7 @@ U32 wpo2(U32 i) 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? @@ -83,6 +78,7 @@ 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); //============================================================================ @@ -95,6 +91,11 @@ LLVBOPool LLVertexBuffer::sDynamicIBOPool(GL_DYNAMIC_DRAW_ARB, GL_ELEMENT_ARRAY_ 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; @@ -119,69 +120,55 @@ 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() + return ret; +} + +void LLVBOPool::deleteBuffer(U32 name) +{ + if (gGLManager.mInited) { -#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 - } + LLVertexBuffer::unbind(); + glBindBufferARB(mType, name); + glBufferDataARB(mType, 0, NULL, mUsage); -}; + llassert(std::find(mGLNamePool.begin(), mGLNamePool.end(), name) == mGLNamePool.end()); + mGLNamePool.push_back(name); + + glBindBufferARB(mType, 0); + } +} -volatile U8* LLVBOPool::allocate(U32& name, U32 size) + +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, bool for_seed) { llassert(vbo_block_size(size) == size); volatile U8* ret = NULL; -#if LL_VBO_POOLING - U32 i = vbo_block_index(size); if (mFreeList.size() <= i) @@ -189,12 +176,18 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size) mFreeList.resize(i+1); } - if (mFreeList[i].empty()) + if (mFreeList[i].empty() || for_seed) { //make a new buffer - glGenBuffersARB(1, &name); + name = genBuffer(); + glBindBufferARB(mType, name); + if (!for_seed && i < LL_VBO_POOL_SEED_COUNT) + { //record this miss + mMissCount[i]++; + } + if (mType == GL_ARRAY_BUFFER_ARB) { LLVertexBuffer::sAllocatedBytes += size; @@ -215,6 +208,25 @@ 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 { @@ -232,33 +244,6 @@ volatile U8* LLVBOPool::allocate(U32& name, U32 size) mFreeList[i].pop_front(); } -#else //no pooling - - glGenBuffersARB(1, &name); - glBindBufferARB(mType, name); - - if (mType == GL_ARRAY_BUFFER_ARB) - { - LLVertexBuffer::sAllocatedBytes += size; - } - else - { - LLVertexBuffer::sAllocatedIndexBytes += size; - } - - if (LLVertexBuffer::sDisableVBOMapping || mUsage != GL_DYNAMIC_DRAW_ARB) - { - glBufferDataARB(mType, size, 0, mUsage); - ret = (U8*) ll_aligned_malloc_16(size); - } - else - { //always use a true hint of static draw when allocating non-client-backed buffers - glBufferDataARB(mType, size, 0, GL_STATIC_DRAW_ARB); - } - - glBindBufferARB(mType, 0); - -#endif return ret; } @@ -267,50 +252,48 @@ void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size) { llassert(vbo_block_size(size) == size); -#if LL_VBO_POOLING - - U32 i = vbo_block_index(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 { - if (mType == GL_ARRAY_BUFFER_ARB) - { - sBytesPooled += size; - } - else - { - sIndexBytesPooled += size; - } - mFreeList[i].push_back(rec); + LLVertexBuffer::sAllocatedIndexBytes -= size; } -#else //no pooling - glDeleteBuffersARB(1, &name); - ll_aligned_free_16((U8*) buffer); +} - if (mType == GL_ARRAY_BUFFER_ARB) +void LLVBOPool::seedPool() +{ + U32 dummy_name = 0; + + if (mFreeList.size() < LL_VBO_POOL_SEED_COUNT) { - LLVertexBuffer::sAllocatedBytes -= size; + mFreeList.resize(LL_VBO_POOL_SEED_COUNT); } - else + + for (U32 i = 0; i < LL_VBO_POOL_SEED_COUNT; i++) { - LLVertexBuffer::sAllocatedIndexBytes -= size; + 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); + } + } } -#endif } + + void LLVBOPool::cleanup() { - U32 size = 1; + U32 size = LL_VBO_BLOCK_SIZE; for (U32 i = 0; i < mFreeList.size(); ++i) { @@ -320,8 +303,8 @@ void LLVBOPool::cleanup() { Record& r = l.front(); - glDeleteBuffersARB(1, &r.mGLName); - + deleteBuffer(r.mGLName); + if (r.mClientData) { ll_aligned_free_16((void*) r.mClientData); @@ -341,8 +324,11 @@ void LLVBOPool::cleanup() } } - size *= 2; + size += LL_VBO_BLOCK_SIZE; } + + //reset miss counts + std::fill(mMissCount.begin(), mMissCount.end(), 0); } @@ -376,6 +362,41 @@ 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) @@ -985,7 +1006,7 @@ LLVertexBuffer::~LLVertexBuffer() if (mGLArray) { #if GL_ARB_vertex_array_object - glDeleteVertexArrays(1, &mGLArray); + releaseVAOName(mGLArray); #endif } @@ -1211,10 +1232,10 @@ 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); @@ -1270,7 +1291,7 @@ 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(); } @@ -1306,7 +1327,7 @@ void LLVertexBuffer::setupVertexArray() 1, //TYPE_WEIGHT, 4, //TYPE_WEIGHT4, 4, //TYPE_CLOTHWEIGHT, - 4, //TYPE_TEXTURE_INDEX + 1, //TYPE_TEXTURE_INDEX }; U32 attrib_type[] = @@ -1323,7 +1344,7 @@ 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[] = @@ -2140,6 +2161,16 @@ 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) { @@ -2291,10 +2322,10 @@ 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) { @@ -2370,7 +2401,7 @@ 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) |