diff options
author | Dave Parks <davep@lindenlab.com> | 2011-10-13 01:19:45 -0500 |
---|---|---|
committer | Dave Parks <davep@lindenlab.com> | 2011-10-13 01:19:45 -0500 |
commit | 7b6723d1e0158d5dc326266a0332e87f634f9755 (patch) | |
tree | 10efed7ccda1a80889bfb65cd64c86b8d33c75ed | |
parent | 1a499c13c813d4aa9631ffae00b36a053f260747 (diff) |
SH-1650 Mitigate memory fragmentation by holding onto and reusing VBOs
-rw-r--r-- | indra/llrender/llvertexbuffer.cpp | 458 | ||||
-rw-r--r-- | indra/llrender/llvertexbuffer.h | 69 | ||||
-rw-r--r-- | indra/newview/lldrawpoolavatar.cpp | 6 | ||||
-rw-r--r-- | indra/newview/lldrawpooltree.cpp | 4 | ||||
-rw-r--r-- | indra/newview/llface.cpp | 13 | ||||
-rw-r--r-- | indra/newview/llfloatermodelpreview.cpp | 2 | ||||
-rw-r--r-- | indra/newview/llspatialpartition.cpp | 10 | ||||
-rw-r--r-- | indra/newview/llviewerdisplay.cpp | 4 | ||||
-rw-r--r-- | indra/newview/llviewerwindow.cpp | 2 | ||||
-rw-r--r-- | indra/newview/llvoavatar.cpp | 4 | ||||
-rw-r--r-- | indra/newview/llvowlsky.cpp | 4 | ||||
-rw-r--r-- | indra/newview/pipeline.cpp | 26 |
12 files changed, 257 insertions, 345 deletions
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 4484a880cc..ad99bd1807 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -38,6 +38,17 @@ #include "llglslshader.h" #include "llmemory.h" +//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) +{ + U32 r = 1; + while (r < v) { + r *= 2; + } + return r; +} + //============================================================================ @@ -46,6 +57,7 @@ LLVBOPool LLVertexBuffer::sStreamVBOPool; LLVBOPool LLVertexBuffer::sDynamicVBOPool; LLVBOPool LLVertexBuffer::sStreamIBOPool; LLVBOPool LLVertexBuffer::sDynamicIBOPool; +U32 LLVBOPool::sBytesPooled = 0; LLPrivateMemoryPool* LLVertexBuffer::sPrivatePoolp = NULL ; U32 LLVertexBuffer::sBindCount = 0; @@ -66,7 +78,6 @@ BOOL LLVertexBuffer::sMapped = FALSE; BOOL LLVertexBuffer::sUseStreamDraw = TRUE; BOOL LLVertexBuffer::sUseVAO = FALSE; BOOL LLVertexBuffer::sPreferStreamDraw = FALSE; -std::vector<U32> LLVertexBuffer::sDeleteList; const U32 FENCE_WAIT_TIME_NANOSECONDS = 10000; //1 ms @@ -122,6 +133,107 @@ public: }; + +//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; +} + +U8* LLVBOPool::allocate(U32& name, U32 size) +{ + llassert(nhpo2(size) == size); + + U32 i = wpo2(size); + + if (mFreeList.size() <= i) + { + mFreeList.resize(i+1); + } + + U8* ret = NULL; + + if (mFreeList[i].empty()) + { + //make a new buffer + glGenBuffersARB(1, &name); + glBindBufferARB(mType, name); + glBufferDataARB(mType, size, 0, mUsage); + LLVertexBuffer::sAllocatedBytes += size; + + if (LLVertexBuffer::sDisableVBOMapping) + { + ret = (U8*) ll_aligned_malloc_16(size); + } + glBindBufferARB(mType, 0); + } + else + { + name = mFreeList[i].front().mGLName; + ret = mFreeList[i].front().mClientData; + + sBytesPooled -= size; + + mFreeList[i].pop_front(); + } + + return ret; +} + +void LLVBOPool::release(U32 name, U8* buffer, U32 size) +{ + llassert(nhpo2(size) == size); + + U32 i = wpo2(size); + + llassert(mFreeList.size() > i); + + Record rec; + rec.mGLName = name; + rec.mClientData = buffer; + + sBytesPooled += size; + + mFreeList[i].push_back(rec); +} + +void LLVBOPool::cleanup() +{ + U32 size = 1; + + for (U32 i = 0; i < mFreeList.size(); ++i) + { + record_list_t& l = mFreeList[i]; + + while (!l.empty()) + { + Record& r = l.front(); + + glDeleteBuffersARB(1, &r.mGLName); + + if (r.mClientData) + { + ll_aligned_free_16(r.mClientData); + } + + l.pop_front(); + + LLVertexBuffer::sAllocatedBytes -= size; + } + + size *= 2; + } +} + + //NOTE: each component must be AT LEAST 4 bytes in size to avoid a performance penalty on AMD hardware S32 LLVertexBuffer::sTypeSize[LLVertexBuffer::TYPE_MAX] = { @@ -374,16 +486,16 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto void LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_offset) const { - if (start >= (U32) mRequestedNumVerts || - end >= (U32) mRequestedNumVerts) + if (start >= (U32) mNumVerts || + end >= (U32) mNumVerts) { - llerrs << "Bad vertex buffer draw range: [" << start << ", " << end << "] vs " << mRequestedNumVerts << llendl; + llerrs << "Bad vertex buffer draw range: [" << start << ", " << end << "] vs " << mNumVerts << llendl; } - llassert(mRequestedNumIndices >= 0); + llassert(mNumIndices >= 0); - if (indices_offset >= (U32) mRequestedNumIndices || - indices_offset + count > (U32) mRequestedNumIndices) + if (indices_offset >= (U32) mNumIndices || + indices_offset + count > (U32) mNumIndices) { llerrs << "Bad index buffer draw range: [" << indices_offset << ", " << indices_offset+count << "]" << llendl; } @@ -407,7 +519,7 @@ void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indi gGL.syncMatrices(); - llassert(mRequestedNumVerts >= 0); + llassert(mNumVerts >= 0); llassert(!LLGLSLShader::sNoFixedFunction || LLGLSLShader::sCurBoundShaderPtr != NULL); if (mGLArray) @@ -462,9 +574,9 @@ void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const gGL.syncMatrices(); - llassert(mRequestedNumIndices >= 0); - if (indices_offset >= (U32) mRequestedNumIndices || - indices_offset + count > (U32) mRequestedNumIndices) + llassert(mNumIndices >= 0); + if (indices_offset >= (U32) mNumIndices || + indices_offset + count > (U32) mNumIndices) { llerrs << "Bad index buffer draw range: [" << indices_offset << ", " << indices_offset+count << "]" << llendl; } @@ -508,9 +620,9 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const gGL.syncMatrices(); - llassert(mRequestedNumVerts >= 0); - if (first >= (U32) mRequestedNumVerts || - first + count > (U32) mRequestedNumVerts) + llassert(mNumVerts >= 0); + if (first >= (U32) mNumVerts || + first + count > (U32) mNumVerts) { llerrs << "Bad vertex buffer draw range: [" << first << ", " << first+count << "]" << llendl; } @@ -546,23 +658,22 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const void LLVertexBuffer::initClass(bool use_vbo, bool no_vbo_mapping) { sEnableVBOs = use_vbo && gGLManager.mHasVertexBufferObject ; - if(sEnableVBOs) - { - //llassert_always(glBindBufferARB) ; //double check the extention for VBO is loaded. - - llinfos << "VBO is enabled." << llendl ; - } - else - { - llinfos << "VBO is disabled." << llendl ; - } - sDisableVBOMapping = sEnableVBOs && no_vbo_mapping ; if(!sPrivatePoolp) { sPrivatePoolp = LLPrivateMemoryPoolManager::getInstance()->newPool(LLPrivateMemoryPool::STATIC) ; } + + sStreamVBOPool.mType = GL_ARRAY_BUFFER_ARB; + sStreamVBOPool.mUsage= GL_STREAM_DRAW_ARB; + sStreamIBOPool.mType = GL_ELEMENT_ARRAY_BUFFER_ARB; + sStreamIBOPool.mUsage= GL_STREAM_DRAW_ARB; + + sDynamicVBOPool.mType = GL_ARRAY_BUFFER_ARB; + sDynamicVBOPool.mUsage= GL_DYNAMIC_DRAW_ARB; + sDynamicIBOPool.mType = GL_ELEMENT_ARRAY_BUFFER_ARB; + sDynamicIBOPool.mUsage= GL_DYNAMIC_DRAW_ARB; } //static @@ -600,7 +711,11 @@ void LLVertexBuffer::cleanupClass() { LLMemType mt2(LLMemType::MTYPE_VERTEX_CLEANUP_CLASS); unbind(); - clientCopy(); // deletes GL buffers + + sStreamIBOPool.cleanup(); + sDynamicIBOPool.cleanup(); + sStreamVBOPool.cleanup(); + sDynamicVBOPool.cleanup(); if(sPrivatePoolp) { @@ -609,15 +724,6 @@ void LLVertexBuffer::cleanupClass() } } -void LLVertexBuffer::clientCopy(F64 max_time) -{ - if (!sDeleteList.empty()) - { - glDeleteBuffersARB(sDeleteList.size(), (GLuint*) &(sDeleteList[0])); - sDeleteList.clear(); - } -} - //---------------------------------------------------------------------------- LLVertexBuffer::LLVertexBuffer(U32 typemask, S32 usage) : @@ -625,8 +731,6 @@ LLVertexBuffer::LLVertexBuffer(U32 typemask, S32 usage) : mNumVerts(0), mNumIndices(0), - mRequestedNumVerts(-1), - mRequestedNumIndices(-1), mUsage(usage), mGLBuffer(0), mGLArray(0), @@ -636,10 +740,7 @@ LLVertexBuffer::LLVertexBuffer(U32 typemask, S32 usage) : mVertexLocked(FALSE), mIndexLocked(FALSE), mFinal(FALSE), - mFilthy(FALSE), mEmpty(TRUE), - mResized(FALSE), - mDynamicSize(FALSE), mFence(NULL) { LLMemType mt2(LLMemType::MTYPE_VERTEX_CONSTRUCTOR); @@ -664,6 +765,11 @@ LLVertexBuffer::LLVertexBuffer(U32 typemask, S32 usage) : mUsage = GL_STREAM_DRAW_ARB; } + if (mUsage && mUsage != GL_STREAM_DRAW_ARB) + { //only stream_draw and dynamic_draw are supported when using VBOs, dynamic draw is the default + mUsage = GL_DYNAMIC_DRAW_ARB; + } + //zero out offsets for (U32 i = 0; i < TYPE_MAX; i++) { @@ -672,6 +778,7 @@ LLVertexBuffer::LLVertexBuffer(U32 typemask, S32 usage) : mTypeMask = typemask; mSize = 0; + mIndicesSize = 0; mAlignedOffset = 0; mAlignedIndexOffset = 0; @@ -775,39 +882,35 @@ void LLVertexBuffer::waitFence() const //---------------------------------------------------------------------------- -void LLVertexBuffer::genBuffer() +void LLVertexBuffer::genBuffer(U32 size) { + mSize = nhpo2(size); + if (mUsage == GL_STREAM_DRAW_ARB) { - mGLBuffer = sStreamVBOPool.allocate(); - } - else if (mUsage == GL_DYNAMIC_DRAW_ARB) - { - mGLBuffer = sDynamicVBOPool.allocate(); + mMappedData = sStreamVBOPool.allocate(mGLBuffer, mSize); } else { - BOOST_STATIC_ASSERT(sizeof(mGLBuffer) == sizeof(GLuint)); - glGenBuffersARB(1, (GLuint*)&mGLBuffer); + mMappedData = sDynamicVBOPool.allocate(mGLBuffer, mSize); } + sGLCount++; } -void LLVertexBuffer::genIndices() +void LLVertexBuffer::genIndices(U32 size) { + mIndicesSize = nhpo2(size); + if (mUsage == GL_STREAM_DRAW_ARB) { - mGLIndices = sStreamIBOPool.allocate(); - } - else if (mUsage == GL_DYNAMIC_DRAW_ARB) - { - mGLIndices = sDynamicIBOPool.allocate(); + mMappedIndexData = sStreamIBOPool.allocate(mGLIndices, mIndicesSize); } else { - BOOST_STATIC_ASSERT(sizeof(mGLBuffer) == sizeof(GLuint)); - glGenBuffersARB(1, (GLuint*)&mGLIndices); + mMappedIndexData = sDynamicIBOPool.allocate(mGLIndices, mIndicesSize); } + sGLCount++; } @@ -815,16 +918,16 @@ void LLVertexBuffer::releaseBuffer() { if (mUsage == GL_STREAM_DRAW_ARB) { - sStreamVBOPool.release(mGLBuffer); - } - else if (mUsage == GL_DYNAMIC_DRAW_ARB) - { - sDynamicVBOPool.release(mGLBuffer); + sStreamVBOPool.release(mGLBuffer, mMappedData, mSize); } else { - sDeleteList.push_back(mGLBuffer); + sDynamicVBOPool.release(mGLBuffer, mMappedData, mSize); } + + mGLBuffer = 0; + mMappedData = NULL; + sGLCount--; } @@ -832,24 +935,23 @@ void LLVertexBuffer::releaseIndices() { if (mUsage == GL_STREAM_DRAW_ARB) { - sStreamIBOPool.release(mGLIndices); + sStreamIBOPool.release(mGLIndices, mMappedIndexData, mIndicesSize); } else if (mUsage == GL_DYNAMIC_DRAW_ARB) { - sDynamicIBOPool.release(mGLIndices); - } - else - { - sDeleteList.push_back(mGLIndices); + sDynamicIBOPool.release(mGLIndices, mMappedIndexData, mIndicesSize); } + + mGLIndices = 0; + mMappedIndexData = NULL; + sGLCount--; } -void LLVertexBuffer::createGLBuffer() +void LLVertexBuffer::createGLBuffer(U32 size) { LLMemType mt2(LLMemType::MTYPE_VERTEX_CREATE_VERTICES); - U32 size = getSize(); if (mGLBuffer) { destroyGLBuffer(); @@ -864,23 +966,21 @@ void LLVertexBuffer::createGLBuffer() if (useVBOs()) { - mMappedData = NULL; - genBuffer(); - mResized = TRUE; + genBuffer(size); } else { static int gl_buffer_idx = 0; mGLBuffer = ++gl_buffer_idx; mMappedData = (U8*)ALLOCATE_MEM(sPrivatePoolp, size); + mSize = size; } } -void LLVertexBuffer::createGLIndices() +void LLVertexBuffer::createGLIndices(U32 size) { LLMemType mt2(LLMemType::MTYPE_VERTEX_CREATE_INDICES); - U32 size = getIndicesSize(); - + if (mGLIndices) { destroyGLIndices(); @@ -900,15 +1000,14 @@ void LLVertexBuffer::createGLIndices() { //pad by another 16 bytes for VBO pointer adjustment size += 16; - mMappedIndexData = NULL; - genIndices(); - mResized = TRUE; + genIndices(size); } else { mMappedIndexData = (U8*)ALLOCATE_MEM(sPrivatePoolp, size); static int gl_buffer_idx = 0; mGLIndices = ++gl_buffer_idx; + mIndicesSize = size; } } @@ -919,12 +1018,6 @@ void LLVertexBuffer::destroyGLBuffer() { if (useVBOs()) { - freeClientBuffer() ; - - if (mMappedData || mMappedIndexData) - { - llerrs << "Vertex buffer destroyed while mapped!" << llendl; - } releaseBuffer(); } else @@ -933,8 +1026,6 @@ void LLVertexBuffer::destroyGLBuffer() mMappedData = NULL; mEmpty = TRUE; } - - sAllocatedBytes -= getSize(); } mGLBuffer = 0; @@ -948,12 +1039,6 @@ void LLVertexBuffer::destroyGLIndices() { if (useVBOs()) { - freeClientBuffer() ; - - if (mMappedData || mMappedIndexData) - { - llerrs << "Vertex buffer destroyed while mapped." << llendl; - } releaseIndices(); } else @@ -962,8 +1047,6 @@ void LLVertexBuffer::destroyGLIndices() mMappedIndexData = NULL; mEmpty = TRUE; } - - sAllocatedBytes -= getIndicesSize(); } mGLIndices = 0; @@ -982,23 +1065,14 @@ void LLVertexBuffer::updateNumVerts(S32 nverts) nverts = 65535; } - mRequestedNumVerts = nverts; + U32 needed_size = calcOffsets(mTypeMask, mOffsets, nverts); - if (!mDynamicSize) + if (needed_size > mSize || needed_size <= mSize/2) { - mNumVerts = nverts; - } - else if (mUsage == GL_STATIC_DRAW_ARB || - nverts > mNumVerts || - nverts < mNumVerts/2) - { - if (mUsage != GL_STATIC_DRAW_ARB && nverts + nverts/4 <= 65535) - { - nverts += nverts/4; - } - mNumVerts = nverts; + createGLBuffer(needed_size); } - mSize = calcOffsets(mTypeMask, mOffsets, mNumVerts); + + mNumVerts = nverts; } void LLVertexBuffer::updateNumIndices(S32 nindices) @@ -1007,22 +1081,14 @@ void LLVertexBuffer::updateNumIndices(S32 nindices) llassert(nindices >= 0); - mRequestedNumIndices = nindices; - if (!mDynamicSize) + U32 needed_size = sizeof(U16) * nindices; + + if (needed_size > mIndicesSize || needed_size <= mIndicesSize/2) { - mNumIndices = nindices; + createGLIndices(needed_size); } - else if (mUsage == GL_STATIC_DRAW_ARB || - nindices > mNumIndices || - nindices < mNumIndices/2) - { - if (mUsage != GL_STATIC_DRAW_ARB) - { - nindices += nindices/4; - } - mNumIndices = nindices; - } + mNumIndices = nindices; } void LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create) @@ -1040,15 +1106,8 @@ void LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create) updateNumVerts(nverts); updateNumIndices(nindices); - if (mMappedData) - { - llerrs << "LLVertexBuffer::allocateBuffer() called redundantly." << llendl; - } if (create && (nverts || nindices)) { - createGLBuffer(); - createGLIndices(); - //actually allocate space for the vertex buffer if using VBO mapping flush(); @@ -1060,8 +1119,6 @@ void LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create) setupVertexArray(); } } - - sAllocatedBytes += getSize() + getIndicesSize(); } void LLVertexBuffer::setupVertexArray() @@ -1151,77 +1208,13 @@ void LLVertexBuffer::resizeBuffer(S32 newnverts, S32 newnindices) llassert(newnverts >= 0); llassert(newnindices >= 0); - mRequestedNumVerts = newnverts; - mRequestedNumIndices = newnindices; - LLMemType mt2(LLMemType::MTYPE_VERTEX_RESIZE_BUFFER); - mDynamicSize = TRUE; - if (mUsage == GL_STATIC_DRAW_ARB) - { //always delete/allocate static buffers on resize - destroyGLBuffer(); - destroyGLIndices(); - allocateBuffer(newnverts, newnindices, TRUE); - mFinal = FALSE; - } - else if (newnverts > mNumVerts || newnindices > mNumIndices || - newnverts < mNumVerts/2 || newnindices < mNumIndices/2) - { - sAllocatedBytes -= getSize() + getIndicesSize(); - - updateNumVerts(newnverts); - updateNumIndices(newnindices); - - S32 newsize = getSize(); - S32 new_index_size = getIndicesSize(); - - sAllocatedBytes += newsize + new_index_size; - - if (newsize) - { - if (!mGLBuffer) - { //no buffer exists, create a new one - createGLBuffer(); - } - else - { - if (!useVBOs()) - { - FREE_MEM(sPrivatePoolp, mMappedData); - mMappedData = (U8*)ALLOCATE_MEM(sPrivatePoolp, newsize); - } - mResized = TRUE; - } - } - else if (mGLBuffer) - { - destroyGLBuffer(); - } - - if (new_index_size) - { - if (!mGLIndices) - { - createGLIndices(); - } - else - { - if (!useVBOs()) - { - FREE_MEM(sPrivatePoolp, mMappedIndexData) ; - mMappedIndexData = (U8*)ALLOCATE_MEM(sPrivatePoolp, new_index_size); - } - mResized = TRUE; - } - } - else if (mGLIndices) - { - destroyGLIndices(); - } - } - - if (mResized && useVBOs()) + + updateNumVerts(newnverts); + updateNumIndices(newnindices); + + if (useVBOs()) { - freeClientBuffer(); flush(); if (mGLArray) @@ -1244,32 +1237,6 @@ BOOL LLVertexBuffer::useVBOs() const } //---------------------------------------------------------------------------- -void LLVertexBuffer::freeClientBuffer() -{ - if(useVBOs() && sDisableVBOMapping && (mMappedData || mMappedIndexData)) - { - FREE_MEM(sPrivatePoolp, mMappedData) ; - FREE_MEM(sPrivatePoolp, mMappedIndexData) ; - mMappedData = NULL ; - mMappedIndexData = NULL ; - } -} - -void LLVertexBuffer::allocateClientVertexBuffer() -{ - if(!mMappedData) - { - mMappedData = (U8*)ALLOCATE_MEM(sPrivatePoolp, getSize()); - } -} - -void LLVertexBuffer::allocateClientIndexBuffer() -{ - if(!mMappedIndexData) - { - mMappedIndexData = (U8*)ALLOCATE_MEM(sPrivatePoolp, getIndicesSize()); - } -} bool expand_region(LLVertexBuffer::MappedRegion& region, S32 index, S32 count) { @@ -1350,7 +1317,6 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran if(sDisableVBOMapping) { map_range = false; - allocateClientVertexBuffer() ; } else { @@ -1535,7 +1501,6 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range) if(sDisableVBOMapping) { map_range = false; - allocateClientIndexBuffer() ; } else { @@ -1772,21 +1737,7 @@ void LLVertexBuffer::unmapBuffer() if(updated_all) { - if(mUsage == GL_STATIC_DRAW_ARB) - { - //static draw buffers can only be mapped a single time - //throw out client data (we won't be using it again) - mEmpty = TRUE; - mFinal = TRUE; - if(sDisableVBOMapping) - { - freeClientBuffer() ; - } - } - else - { - mEmpty = FALSE; - } + mEmpty = FALSE; } } @@ -1965,27 +1916,6 @@ void LLVertexBuffer::flush() { if (useVBOs()) { - if (mResized) - { - if (mGLBuffer) - { - stop_glerror(); - bindGLBuffer(true); - glBufferDataARB(GL_ARRAY_BUFFER_ARB, getSize(), NULL, mUsage); - stop_glerror(); - } - if (mGLIndices) - { - stop_glerror(); - bindGLIndices(true); - glBufferDataARB(GL_ELEMENT_ARRAY_BUFFER_ARB, getIndicesSize(), NULL, mUsage); - stop_glerror(); - } - - mEmpty = TRUE; - mResized = FALSE; - } - unmapBuffer(); } } diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h index b50c409c49..3e6f6a959a 100644 --- a/indra/llrender/llvertexbuffer.h +++ b/indra/llrender/llvertexbuffer.h @@ -51,24 +51,32 @@ //============================================================================ // gl name pools for dynamic and streaming buffers -class LLVBOPool : public LLGLNamePool +class LLVBOPool { -protected: - virtual GLuint allocateName() - { - GLuint name; - stop_glerror(); - glGenBuffersARB(1, &name); - stop_glerror(); - return name; - } +public: + static U32 sBytesPooled; + + U32 mUsage; + U32 mType; - virtual void releaseName(GLuint name) + //size MUST be a power of 2 + U8* allocate(U32& name, U32 size); + + //size MUST be the size provided to allocate that returned the given name + void release(U32 name, U8* buffer, U32 size); + + //destroy all records in mFreeList + void cleanup(); + + class Record { - stop_glerror(); - glDeleteBuffersARB(1, &name); - stop_glerror(); - } + public: + U32 mGLName; + U8* mClientData; + }; + + typedef std::list<Record> record_list_t; + std::vector<record_list_t> mFreeList; }; class LLGLFence @@ -120,8 +128,7 @@ public: static void drawArrays(U32 mode, const std::vector<LLVector3>& pos, const std::vector<LLVector3>& norm); static void drawElements(U32 mode, const LLVector4a* pos, const LLVector2* tc, S32 num_indices, const U16* indicesp); - static void clientCopy(F64 max_time = 0.005); //copy data from client to GL - static void unbind(); //unbind any bound vertex buffer + static void unbind(); //unbind any bound vertex buffer //get the size of a vertex with the given typemask static S32 calcVertexSize(const U32& typemask); @@ -181,25 +188,22 @@ protected: virtual void setupVertexBuffer(U32 data_mask); // pure virtual, called from mapBuffer() void setupVertexArray(); - void genBuffer(); - void genIndices(); + void genBuffer(U32 size); + void genIndices(U32 size); bool bindGLBuffer(bool force_bind = false); bool bindGLIndices(bool force_bind = false); bool bindGLArray(); void releaseBuffer(); void releaseIndices(); - void createGLBuffer(); - void createGLIndices(); + void createGLBuffer(U32 size); + void createGLIndices(U32 size); void destroyGLBuffer(); void destroyGLIndices(); void updateNumVerts(S32 nverts); void updateNumIndices(S32 nindices); virtual BOOL useVBOs() const; void unmapBuffer(); - void freeClientBuffer() ; - void allocateClientVertexBuffer() ; - void allocateClientIndexBuffer() ; - + public: LLVertexBuffer(U32 typemask, S32 usage); @@ -239,15 +243,13 @@ public: BOOL isLocked() const { return mVertexLocked || mIndexLocked; } S32 getNumVerts() const { return mNumVerts; } S32 getNumIndices() const { return mNumIndices; } - S32 getRequestedVerts() const { return mRequestedNumVerts; } - S32 getRequestedIndices() const { return mRequestedNumIndices; } - + U8* getIndicesPointer() const { return useVBOs() ? (U8*) mAlignedIndexOffset : mMappedIndexData; } U8* getVerticesPointer() const { return useVBOs() ? (U8*) mAlignedOffset : mMappedData; } U32 getTypeMask() const { return mTypeMask; } bool hasDataType(S32 type) const { return ((1 << type) & getTypeMask()); } S32 getSize() const; - S32 getIndicesSize() const { return mNumIndices * sizeof(U16); } + S32 getIndicesSize() const { return mIndicesSize; } U8* getMappedData() const { return mMappedData; } U8* getMappedIndices() const { return mMappedIndexData; } S32 getOffset(S32 type) const { return mOffsets[type]; } @@ -265,12 +267,11 @@ public: protected: S32 mNumVerts; // Number of vertices allocated S32 mNumIndices; // Number of indices allocated - S32 mRequestedNumVerts; // Number of vertices requested - S32 mRequestedNumIndices; // Number of indices requested - + ptrdiff_t mAlignedOffset; ptrdiff_t mAlignedIndexOffset; S32 mSize; + S32 mIndicesSize; U32 mTypeMask; S32 mUsage; // GL usage U32 mGLBuffer; // GL VBO handle @@ -282,10 +283,7 @@ protected: BOOL mVertexLocked; // if TRUE, vertex buffer is being or has been written to in client memory BOOL mIndexLocked; // if TRUE, index buffer is being or has been written to in client memory BOOL mFinal; // if TRUE, buffer can not be mapped again - BOOL mFilthy; // if TRUE, entire buffer must be copied (used to prevent redundant dirty flags) BOOL mEmpty; // if TRUE, client buffer is empty (or NULL). Old values have been discarded. - BOOL mResized; // if TRUE, client buffer has been resized and GL buffer has not - BOOL mDynamicSize; // if TRUE, buffer has been resized at least once (and should be padded) S32 mOffsets[TYPE_MAX]; std::vector<MappedRegion> mMappedVertexRegions; @@ -305,7 +303,6 @@ public: static S32 sGLCount; static S32 sMappedCount; static BOOL sMapped; - static std::vector<U32> sDeleteList; typedef std::list<LLVertexBuffer*> buffer_list_t; static BOOL sDisableVBOMapping; //disable glMapBufferARB diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index a710bdcdbd..7290a48a1a 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -1277,8 +1277,8 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, LLFace* if (buffer.isNull() || buffer->getTypeMask() != data_mask || - buffer->getRequestedVerts() != vol_face.mNumVertices || - buffer->getRequestedIndices() != vol_face.mNumIndices || + buffer->getNumVerts() != vol_face.mNumVertices || + buffer->getNumIndices() != vol_face.mNumIndices || (drawable && drawable->isState(LLDrawable::REBUILD_ALL))) { face->setGeomIndex(0); @@ -1366,7 +1366,7 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, LLFace* LLMatrix4a bind_shape_matrix; bind_shape_matrix.loadu(skin->mBindShapeMatrix); - for (U32 j = 0; j < buffer->getRequestedVerts(); ++j) + for (U32 j = 0; j < buffer->getNumVerts(); ++j) { LLMatrix4a final_mat; final_mat.clear(); diff --git a/indra/newview/lldrawpooltree.cpp b/indra/newview/lldrawpooltree.cpp index 3fe5b4d929..cdf6e1ab52 100644 --- a/indra/newview/lldrawpooltree.cpp +++ b/indra/newview/lldrawpooltree.cpp @@ -113,8 +113,8 @@ void LLDrawPoolTree::render(S32 pass) if(buff) { buff->setBuffer(LLDrawPoolTree::VERTEX_DATA_MASK); - buff->drawRange(LLRender::TRIANGLES, 0, buff->getRequestedVerts()-1, buff->getRequestedIndices(), 0); - gPipeline.addTrianglesDrawn(buff->getRequestedIndices()); + buff->drawRange(LLRender::TRIANGLES, 0, buff->getNumVerts()-1, buff->getNumIndices(), 0); + gPipeline.addTrianglesDrawn(buff->getNumIndices()); } } } diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 36b88ebbd4..eab3dcfadd 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -362,8 +362,8 @@ void LLFace::setSize(S32 num_vertices, S32 num_indices, bool align) { if (align) { - //allocate vertices in blocks of 16 for alignment - num_vertices = (num_vertices + 0xF) & ~0xF; + //allocate vertices in blocks of 4 for alignment + num_vertices = (num_vertices + 0x3) & ~0x3; } if (mGeomCount != num_vertices || @@ -1073,6 +1073,11 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, S32 num_vertices = (S32)vf.mNumVertices; S32 num_indices = (S32) vf.mNumIndices; + if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_OCTREE)) + { + updateRebuildFlags(); + } + bool map_range = gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange; if (mVertexBuffer.notNull()) @@ -2055,7 +2060,7 @@ BOOL LLFace::verify(const U32* indices_array) const } // First, check whether the face data fits within the pool's range. - if ((mGeomIndex + mGeomCount) > mVertexBuffer->getRequestedVerts()) + if ((mGeomIndex + mGeomCount) > mVertexBuffer->getNumVerts()) { ok = FALSE; llinfos << "Face references invalid vertices!" << llendl; @@ -2074,7 +2079,7 @@ BOOL LLFace::verify(const U32* indices_array) const llinfos << "Face has bogus indices count" << llendl; } - if (mIndicesIndex + mIndicesCount > mVertexBuffer->getRequestedIndices()) + if (mIndicesIndex + mIndicesCount > mVertexBuffer->getNumIndices()) { ok = FALSE; llinfos << "Face references invalid indices!" << llendl; diff --git a/indra/newview/llfloatermodelpreview.cpp b/indra/newview/llfloatermodelpreview.cpp index d9ce72a2c2..8f6013f2d9 100644 --- a/indra/newview/llfloatermodelpreview.cpp +++ b/indra/newview/llfloatermodelpreview.cpp @@ -5457,7 +5457,7 @@ BOOL LLModelPreview::render() } } - for (U32 j = 0; j < buffer->getRequestedVerts(); ++j) + for (U32 j = 0; j < buffer->getNumVerts(); ++j) { LLMatrix4 final_mat; final_mat.mMatrix[0][0] = final_mat.mMatrix[1][1] = final_mat.mMatrix[2][2] = final_mat.mMatrix[3][3] = 0.f; diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index ffb0ce4056..2530f1f0d4 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -2454,7 +2454,7 @@ void pushBufferVerts(LLVertexBuffer* buffer, U32 mask) if (buffer) { buffer->setBuffer(mask); - buffer->drawRange(LLRender::TRIANGLES, 0, buffer->getRequestedVerts()-1, buffer->getRequestedIndices(), 0); + buffer->drawRange(LLRender::TRIANGLES, 0, buffer->getNumVerts()-1, buffer->getNumIndices(), 0); } } @@ -2526,7 +2526,7 @@ void renderOctree(LLSpatialGroup* group) //coded by buffer usage and activity gGL.setSceneBlendType(LLRender::BT_ADD_WITH_ALPHA); LLVector4 col; - /*if (group->mBuilt > 0.f) + if (group->mBuilt > 0.f) { group->mBuilt -= 2.f * gFrameIntervalSeconds; if (group->mBufferUsage == GL_STATIC_DRAW_ARB) @@ -2595,7 +2595,7 @@ void renderOctree(LLSpatialGroup* group) gGL.diffuseColor4f(1,1,1,1); } } - else*/ + else { if (group->mBufferUsage == GL_STATIC_DRAW_ARB && !group->getData().empty() && group->mSpatialPartition->mRenderByGroup) @@ -3442,11 +3442,11 @@ void renderPhysicsShapes(LLSpatialGroup* group) buff->setBuffer(LLVertexBuffer::MAP_VERTEX); gGL.diffuseColor3f(0.2f, 0.5f, 0.3f); - buff->draw(LLRender::TRIANGLES, buff->getRequestedIndices(), 0); + buff->draw(LLRender::TRIANGLES, buff->getNumIndices(), 0); gGL.diffuseColor3f(0.2f, 1.f, 0.3f); glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); - buff->draw(LLRender::TRIANGLES, buff->getRequestedIndices(), 0); + buff->draw(LLRender::TRIANGLES, buff->getNumIndices(), 0); } } } diff --git a/indra/newview/llviewerdisplay.cpp b/indra/newview/llviewerdisplay.cpp index 1832416a4b..e0359cc61d 100644 --- a/indra/newview/llviewerdisplay.cpp +++ b/indra/newview/llviewerdisplay.cpp @@ -661,10 +661,6 @@ void display(BOOL rebuild, F32 zoom_factor, int subfield, BOOL for_snapshot) { LLMemType mt_ds(LLMemType::MTYPE_DISPLAY_SWAP); - { - LLFastTimer ftm(FTM_CLIENT_COPY); - LLVertexBuffer::clientCopy(0.016); - } if (gResizeScreenTexture) { diff --git a/indra/newview/llviewerwindow.cpp b/indra/newview/llviewerwindow.cpp index 2aac43d99e..f3e9bc711a 100644 --- a/indra/newview/llviewerwindow.cpp +++ b/indra/newview/llviewerwindow.cpp @@ -532,7 +532,7 @@ public: } - addText(xpos, ypos, llformat("%d MB Vertex Data", LLVertexBuffer::sAllocatedBytes/(1024*1024))); + addText(xpos, ypos, llformat("%d MB Vertex Data (%d MB Pooled)", LLVertexBuffer::sAllocatedBytes/(1024*1024), LLVBOPool::sBytesPooled/(1024*1024))); ypos += y_inc; addText(xpos, ypos, llformat("%d Vertex Buffers", LLVertexBuffer::sGLCount)); diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp index 5687ba5064..6506938766 100644 --- a/indra/newview/llvoavatar.cpp +++ b/indra/newview/llvoavatar.cpp @@ -2117,8 +2117,8 @@ void LLVOAvatar::updateMeshData() } else { - if (buff->getRequestedIndices() == num_indices && - buff->getRequestedVerts() == num_vertices) + if (buff->getNumIndices() == num_indices && + buff->getNumVerts() == num_vertices) { terse_update = true; } diff --git a/indra/newview/llvowlsky.cpp b/indra/newview/llvowlsky.cpp index 14fd0a1eb1..f1c5499d84 100644 --- a/indra/newview/llvowlsky.cpp +++ b/indra/newview/llvowlsky.cpp @@ -516,9 +516,9 @@ void LLVOWLSky::drawDome(void) strips_segment->drawRange( LLRender::TRIANGLE_STRIP, - 0, strips_segment->getRequestedVerts()-1, strips_segment->getRequestedIndices(), + 0, strips_segment->getNumVerts()-1, strips_segment->getNumIndices(), 0); - gPipeline.addTrianglesDrawn(strips_segment->getRequestedIndices(), LLRender::TRIANGLE_STRIP); + gPipeline.addTrianglesDrawn(strips_segment->getNumIndices(), LLRender::TRIANGLE_STRIP); } #else diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index e4125c8dc8..5035e0197d 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -264,15 +264,7 @@ std::string gPoolNames[] = void drawBox(const LLVector3& c, const LLVector3& r); void drawBoxOutline(const LLVector3& pos, const LLVector3& size); - -U32 nhpo2(U32 v) -{ - U32 r = 1; - while (r < v) { - r *= 2; - } - return r; -} +U32 nhpo2(U32 v); glh::matrix4f glh_copy_matrix(F32* src) { @@ -2902,11 +2894,7 @@ void LLPipeline::stateSort(LLCamera& camera, LLCullResult &result) } } } - { - LLFastTimer ftm(FTM_CLIENT_COPY); - LLVertexBuffer::clientCopy(); - } - + postSort(camera); } @@ -6122,13 +6110,7 @@ void LLPipeline::resetVertexBuffers() llwarns << "VBO wipe failed." << llendl; } - if (!LLVertexBuffer::sStreamIBOPool.mNameList.empty() || - !LLVertexBuffer::sStreamVBOPool.mNameList.empty() || - !LLVertexBuffer::sDynamicIBOPool.mNameList.empty() || - !LLVertexBuffer::sDynamicVBOPool.mNameList.empty()) - { - llwarns << "VBO name pool cleanup failed." << llendl; - } + llassert(LLVertexBuffer::sGLCount == 0); LLVertexBuffer::unbind(); @@ -6142,6 +6124,8 @@ void LLPipeline::resetVertexBuffers() sBakeSunlight = gSavedSettings.getBOOL("RenderBakeSunlight"); sNoAlpha = gSavedSettings.getBOOL("RenderNoAlpha"); LLPipeline::sTextureBindTest = gSavedSettings.getBOOL("RenderDebugTextureBind"); + + LLVertexBuffer::initClass(LLVertexBuffer::sEnableVBOs, LLVertexBuffer::sDisableVBOMapping); } void LLPipeline::renderObjects(U32 type, U32 mask, BOOL texture) |