diff options
Diffstat (limited to 'indra/llrender/llvertexbuffer.cpp')
| -rw-r--r-- | indra/llrender/llvertexbuffer.cpp | 2660 |
1 files changed, 790 insertions, 1870 deletions
diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp index 4064d35753..de27636c33 100644 --- a/indra/llrender/llvertexbuffer.cpp +++ b/indra/llrender/llvertexbuffer.cpp @@ -62,288 +62,476 @@ U32 wpo2(U32 i) return r; } +struct CompareMappedRegion +{ + bool operator()(const LLVertexBuffer::MappedRegion& lhs, const LLVertexBuffer::MappedRegion& rhs) + { + return lhs.mStart < rhs.mStart; + } +}; -const U32 LL_VBO_BLOCK_SIZE = 2048; -const U32 LL_VBO_POOL_MAX_SEED_SIZE = 256*1024; +#define ENABLE_GL_WORK_QUEUE 0 -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); -} +#if ENABLE_GL_WORK_QUEUE + +#define THREAD_COUNT 1 + +//============================================================================ -U32 vbo_block_index(U32 size) +// High performance WorkQueue for usage in real-time rendering work +class GLWorkQueue { - return vbo_block_size(size)/LL_VBO_BLOCK_SIZE; -} +public: + using Work = std::function<void()>; -const U32 LL_VBO_POOL_SEED_COUNT = vbo_block_index(LL_VBO_POOL_MAX_SEED_SIZE); + GLWorkQueue(); + void post(const Work& value); -//============================================================================ + size_t size(); -//static -LLVBOPool LLVertexBuffer::sStreamVBOPool(GL_STREAM_DRAW, GL_ARRAY_BUFFER); -LLVBOPool LLVertexBuffer::sDynamicVBOPool(GL_DYNAMIC_DRAW, GL_ARRAY_BUFFER); -LLVBOPool LLVertexBuffer::sDynamicCopyVBOPool(GL_DYNAMIC_COPY, GL_ARRAY_BUFFER); -LLVBOPool LLVertexBuffer::sStreamIBOPool(GL_STREAM_DRAW, GL_ELEMENT_ARRAY_BUFFER); -LLVBOPool LLVertexBuffer::sDynamicIBOPool(GL_DYNAMIC_DRAW, GL_ELEMENT_ARRAY_BUFFER); - -U32 LLVBOPool::sBytesPooled = 0; -U32 LLVBOPool::sIndexBytesPooled = 0; -U32 LLVBOPool::sNameIdx = 0; -U32 LLVBOPool::sNamePool[1024]; - -std::list<U32> LLVertexBuffer::sAvailableVAOName; -U32 LLVertexBuffer::sCurVAOName = 1; - -U32 LLVertexBuffer::sAllocatedIndexBytes = 0; -U32 LLVertexBuffer::sIndexCount = 0; - -U32 LLVertexBuffer::sBindCount = 0; -U32 LLVertexBuffer::sSetCount = 0; -S32 LLVertexBuffer::sCount = 0; -S32 LLVertexBuffer::sGLCount = 0; -S32 LLVertexBuffer::sMappedCount = 0; -bool LLVertexBuffer::sDisableVBOMapping = false; -bool LLVertexBuffer::sEnableVBOs = true; -U32 LLVertexBuffer::sGLRenderBuffer = 0; -U32 LLVertexBuffer::sGLRenderArray = 0; -U32 LLVertexBuffer::sGLRenderIndices = 0; -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; + bool done(); + + // Get the next element from the queue + Work pop(); + + void runOne(); + + bool runPending(); + + void runUntilClose(); + + void close(); + + bool isClosed(); + void syncGL(); -U32 LLVBOPool::genBuffer() +private: + std::mutex mMutex; + std::condition_variable mCondition; + std::queue<Work> mQueue; + bool mClosed = false; +}; + +GLWorkQueue::GLWorkQueue() { - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX - if (sNameIdx == 0) - { - glGenBuffers(1024, sNamePool); - sNameIdx = 1024; - } +} + +void GLWorkQueue::syncGL() +{ + /*if (mSync) + { + std::lock_guard<std::mutex> lock(mMutex); + glWaitSync(mSync, 0, GL_TIMEOUT_IGNORED); + mSync = 0; + }*/ +} - return sNamePool[--sNameIdx]; +size_t GLWorkQueue::size() +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; + std::lock_guard<std::mutex> lock(mMutex); + return mQueue.size(); } -void LLVBOPool::deleteBuffer(U32 name) +bool GLWorkQueue::done() { - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX - if (gGLManager.mInited) - { - LLVertexBuffer::unbind(); + return size() == 0 && isClosed(); +} - glBindBuffer(mType, name); - glBufferData(mType, 0, NULL, mUsage); - glBindBuffer(mType, 0); +void GLWorkQueue::post(const GLWorkQueue::Work& value) +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; + { + std::lock_guard<std::mutex> lock(mMutex); + mQueue.push(std::move(value)); + } - glDeleteBuffers(1, &name); - } + mCondition.notify_one(); } +// Get the next element from the queue +GLWorkQueue::Work GLWorkQueue::pop() +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; + // Lock the mutex + { + std::unique_lock<std::mutex> lock(mMutex); + + // Wait for a new element to become available or for the queue to close + { + mCondition.wait(lock, [=] { return !mQueue.empty() || mClosed; }); + } + } + + Work ret; + + { + std::lock_guard<std::mutex> lock(mMutex); + + // Get the next element from the queue + if (mQueue.size() > 0) + { + ret = mQueue.front(); + mQueue.pop(); + } + else + { + ret = []() {}; + } + } + + return ret; +} -LLVBOPool::LLVBOPool(U32 vboUsage, U32 vboType) -: mUsage(vboUsage), mType(vboType) +void GLWorkQueue::runOne() { - mMissCount.resize(LL_VBO_POOL_SEED_COUNT); - std::fill(mMissCount.begin(), mMissCount.end(), 0); + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; + Work w = pop(); + w(); + //mSync = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); } -U8* LLVBOPool::allocate(U32& name, U32 size, bool for_seed) +void GLWorkQueue::runUntilClose() { - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX - llassert(vbo_block_size(size) == size); - - U8* ret = NULL; + while (!isClosed()) + { + runOne(); + } +} - U32 i = vbo_block_index(size); +void GLWorkQueue::close() +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; + { + std::lock_guard<std::mutex> lock(mMutex); + mClosed = true; + } - if (mFreeList.size() <= i) - { - mFreeList.resize(i+1); - } + mCondition.notify_all(); +} - if (mFreeList[i].empty() || for_seed) - { - //make a new buffer - name = genBuffer(); - - glBindBuffer(mType, name); - - if (!for_seed && i < LL_VBO_POOL_SEED_COUNT) - { //record this miss - mMissCount[i]++; - } +bool GLWorkQueue::isClosed() +{ + LL_PROFILE_ZONE_SCOPED_CATEGORY_THREAD; + std::lock_guard<std::mutex> lock(mMutex); + return mClosed; +} - if (mType == GL_ARRAY_BUFFER) - { - LLVertexBuffer::sAllocatedBytes += size; - } - else - { - LLVertexBuffer::sAllocatedIndexBytes += size; - } +#include "llwindow.h" - if (LLVertexBuffer::sDisableVBOMapping || mUsage != GL_DYNAMIC_DRAW) - { - glBufferData(mType, size, 0, mUsage); - if (mUsage != GL_DYNAMIC_COPY) - { //data will be provided by application - ret = (U8*) ll_aligned_malloc<64>(size); - if (!ret) - { - LL_ERRS() << "Failed to allocate "<< size << " bytes for LLVBOPool buffer " << name <<"." << LL_NEWLINE - << "Free list size: " << mFreeList.size() // this happens if we are out of memory so a solution might be to clear some from freelist - << " Allocated Bytes: " << LLVertexBuffer::sAllocatedBytes - << " Allocated Index Bytes: " << LLVertexBuffer::sAllocatedIndexBytes - << " Pooled Bytes: " << sBytesPooled - << " Pooled Index Bytes: " << sIndexBytesPooled - << LL_ENDL; - } - } - } - else - { //always use a true hint of static draw when allocating non-client-backed buffers - glBufferData(mType, size, 0, GL_STATIC_DRAW); - } +class LLGLWorkerThread : public LLThread +{ +public: + LLGLWorkerThread(const std::string& name, GLWorkQueue* queue, LLWindow* window) + : LLThread(name) + { + mWindow = window; + mContext = mWindow->createSharedContext(); + mQueue = queue; + } - glBindBuffer(mType, 0); + void run() override + { + mWindow->makeContextCurrent(mContext); + gGL.init(false); + mQueue->runUntilClose(); + gGL.shutdown(); + mWindow->destroySharedContext(mContext); + } - if (for_seed) - { //put into pool for future use - llassert(mFreeList.size() > i); + GLWorkQueue* mQueue; + LLWindow* mWindow; + void* mContext = nullptr; +}; - Record rec; - rec.mGLName = name; - rec.mClientData = ret; - - if (mType == GL_ARRAY_BUFFER) - { - sBytesPooled += size; - } - else - { - sIndexBytesPooled += size; - } - mFreeList[i].push_back(rec); - } - } - else - { - name = mFreeList[i].front().mGLName; - ret = mFreeList[i].front().mClientData; - if (mType == GL_ARRAY_BUFFER) - { - sBytesPooled -= size; - } - else - { - sIndexBytesPooled -= size; - } +static LLGLWorkerThread* sVBOThread[THREAD_COUNT]; +static GLWorkQueue* sQueue = nullptr; - mFreeList[i].pop_front(); - } +#endif - return ret; -} +//============================================================================ +// Pool of reusable VertexBuffer state -void LLVBOPool::release(U32 name, U8* buffer, U32 size) +// batch calls to glGenBuffers +static GLuint gen_buffer() { - llassert(vbo_block_size(size) == size); + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; - deleteBuffer(name); - ll_aligned_free_fallback((U8*) buffer); + GLuint ret = 0; + constexpr U32 pool_size = 4096; - if (mType == GL_ARRAY_BUFFER) - { - LLVertexBuffer::sAllocatedBytes -= size; - } - else - { - LLVertexBuffer::sAllocatedIndexBytes -= size; - } + thread_local static GLuint sNamePool[pool_size]; + thread_local static U32 sIndex = 0; + + if (sIndex == 0) + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("gen buffer"); + sIndex = pool_size; + if (!gGLManager.mIsAMD) + { + glGenBuffers(pool_size, sNamePool); + } + else + { // work around for AMD driver bug + for (U32 i = 0; i < pool_size; ++i) + { + glGenBuffers(1, sNamePool + i); + } + } + } + + ret = sNamePool[--sIndex]; + return ret; } -void LLVBOPool::seedPool() +#define ANALYZE_VBO_POOL 0 + +class LLVBOPool { - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX - U32 dummy_name = 0; +public: + typedef std::chrono::steady_clock::time_point Time; - if (mFreeList.size() < LL_VBO_POOL_SEED_COUNT) - { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("VBOPool Resize"); - mFreeList.resize(LL_VBO_POOL_SEED_COUNT); - } + struct Entry + { + U8* mData; + GLuint mGLName; + Time mAge; + }; - 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); - } - } - } -} + ~LLVBOPool() + { + clear(); + } + typedef std::unordered_map<U32, std::list<Entry>> Pool; + Pool mVBOPool; + Pool mIBOPool; -void LLVBOPool::cleanup() -{ - U32 size = LL_VBO_BLOCK_SIZE; + U32 mTouchCount = 0; - for (U32 i = 0; i < mFreeList.size(); ++i) - { - record_list_t& l = mFreeList[i]; + U64 mDistributed = 0; + U64 mAllocated = 0; + U64 mReserved = 0; + U32 mMisses = 0; + U32 mHits = 0; - while (!l.empty()) - { - Record& r = l.front(); + U64 getVramBytesUsed() + { + return mAllocated + mReserved; + } - deleteBuffer(r.mGLName); - - if (r.mClientData) - { - ll_aligned_free<64>((void*) r.mClientData); - } + // increase the size to some common value (e.g. a power of two) to increase hit rate + void adjustSize(U32& size) + { + // size = nhpo2(size); // (193/303)/580 MB (distributed/allocated)/reserved in VBO Pool. Overhead: 66 percent. Hit rate: 77 percent - l.pop_front(); + //(245/276)/385 MB (distributed/allocated)/reserved in VBO Pool. Overhead: 57 percent. Hit rate: 69 percent + //(187/209)/397 MB (distributed/allocated)/reserved in VBO Pool. Overhead: 112 percent. Hit rate: 76 percent + U32 block_size = llmax(nhpo2(size) / 8, (U32) 16); + size += block_size - (size % block_size); + } - if (mType == GL_ARRAY_BUFFER) - { - sBytesPooled -= size; - LLVertexBuffer::sAllocatedBytes -= size; - } - else - { - sIndexBytesPooled -= size; - LLVertexBuffer::sAllocatedIndexBytes -= size; - } - } + void allocate(GLenum type, U32 size, GLuint& name, U8*& data) + { + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + llassert(type == GL_ARRAY_BUFFER || type == GL_ELEMENT_ARRAY_BUFFER); + llassert(name == 0); // non zero name indicates a gl name that wasn't freed + llassert(data == nullptr); // non null data indicates a buffer that wasn't freed + llassert(size >= 2); // any buffer size smaller than a single index is nonsensical + + mDistributed += size; + adjustSize(size); + mAllocated += size; + + auto& pool = type == GL_ELEMENT_ARRAY_BUFFER ? mIBOPool : mVBOPool; + + Pool::iterator iter = pool.find(size); + if (iter == pool.end()) + { // cache miss, allocate a new buffer + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vbo pool miss"); + LL_PROFILE_GPU_ZONE("vbo alloc"); + + mMisses++; + name = gen_buffer(); + glBindBuffer(type, name); + glBufferData(type, size, nullptr, GL_DYNAMIC_DRAW); + if (type == GL_ELEMENT_ARRAY_BUFFER) + { + LLVertexBuffer::sGLRenderIndices = name; + } + else + { + LLVertexBuffer::sGLRenderBuffer = name; + } + + data = (U8*)ll_aligned_malloc_16(size); + } + else + { + mHits++; + llassert(mReserved >= size); // assert if accounting gets messed up + mReserved -= size; + + std::list<Entry>& entries = iter->second; + Entry& entry = entries.back(); + name = entry.mGLName; + data = entry.mData; + + entries.pop_back(); + if (entries.empty()) + { + pool.erase(iter); + } + } - size += LL_VBO_BLOCK_SIZE; - } + clean(); + } + + void free(GLenum type, U32 size, GLuint name, U8* data) + { + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + llassert(type == GL_ARRAY_BUFFER || type == GL_ELEMENT_ARRAY_BUFFER); + llassert(size >= 2); + llassert(name != 0); + llassert(data != nullptr); + + clean(); + + llassert(mDistributed >= size); + mDistributed -= size; + adjustSize(size); + llassert(mAllocated >= size); + mAllocated -= size; + mReserved += size; + + auto& pool = type == GL_ELEMENT_ARRAY_BUFFER ? mIBOPool : mVBOPool; + + Pool::iterator iter = pool.find(size); + + if (iter == pool.end()) + { + std::list<Entry> newlist; + newlist.push_front({ data, name, std::chrono::steady_clock::now() }); + pool[size] = newlist; + } + else + { + iter->second.push_front({ data, name, std::chrono::steady_clock::now() }); + } + + } + + // clean periodically (clean gets called for every alloc/free) + void clean() + { + mTouchCount++; + if (mTouchCount < 1024) // clean every 1k touches + { + return; + } + mTouchCount = 0; + + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + + std::unordered_map<U32, std::list<Entry>>* pools[] = { &mVBOPool, &mIBOPool }; + + using namespace std::chrono_literals; + + Time cutoff = std::chrono::steady_clock::now() - 5s; + + for (auto* pool : pools) + { + for (Pool::iterator iter = pool->begin(); iter != pool->end(); ) + { + auto& entries = iter->second; + + while (!entries.empty() && entries.back().mAge < cutoff) + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vbo cache timeout"); + auto& entry = entries.back(); + ll_aligned_free_16(entry.mData); + glDeleteBuffers(1, &entry.mGLName); + llassert(mReserved >= iter->first); + mReserved -= iter->first; + entries.pop_back(); + + } + + if (entries.empty()) + { + iter = pool->erase(iter); + } + else + { + ++iter; + } + } + } + +#if 0 + LL_INFOS() << llformat("(%d/%d)/%d MB (distributed/allocated)/total in VBO Pool. Overhead: %d percent. Hit rate: %d percent", + mDistributed / 1000000, + mAllocated / 1000000, + (mAllocated + mReserved) / 1000000, // total bytes + ((mAllocated+mReserved-mDistributed)*100)/llmax(mDistributed, (U64) 1), // overhead percent + (mHits*100)/llmax(mMisses+mHits, (U32)1)) // hit rate percent + << LL_ENDL; +#endif + } + + void clear() + { + for (auto& entries : mIBOPool) + { + for (auto& entry : entries.second) + { + ll_aligned_free_16(entry.mData); + glDeleteBuffers(1, &entry.mGLName); + } + } + + for (auto& entries : mVBOPool) + { + for (auto& entry : entries.second) + { + ll_aligned_free_16(entry.mData); + glDeleteBuffers(1, &entry.mGLName); + } + } - //reset miss counts - std::fill(mMissCount.begin(), mMissCount.end(), 0); + mReserved = 0; + + mIBOPool.clear(); + mVBOPool.clear(); + } + + +}; + +static LLVBOPool* sVBOPool = nullptr; + +//static +U64 LLVertexBuffer::getBytesAllocated() +{ + return sVBOPool ? sVBOPool->getVramBytesUsed() : 0; } +//============================================================================ +// +//static +U32 LLVertexBuffer::sGLRenderBuffer = 0; +U32 LLVertexBuffer::sGLRenderIndices = 0; +U32 LLVertexBuffer::sLastMask = 0; +U32 LLVertexBuffer::sVertexCount = 0; + //NOTE: each component must be AT LEAST 4 bytes in size to avoid a performance penalty on AMD hardware -const S32 LLVertexBuffer::sTypeSize[LLVertexBuffer::TYPE_MAX] = +const U32 LLVertexBuffer::sTypeSize[LLVertexBuffer::TYPE_MAX] = { sizeof(LLVector4), // TYPE_VERTEX, sizeof(LLVector4), // TYPE_NORMAL, @@ -392,79 +580,34 @@ const U32 LLVertexBuffer::sGLMode[LLRender::NUM_MODES] = }; //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() -{ - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX - sStreamVBOPool.seedPool(); - sDynamicVBOPool.seedPool(); - sDynamicCopyVBOPool.seedPool(); - sStreamIBOPool.seedPool(); - sDynamicIBOPool.seedPool(); -} - -//static void LLVertexBuffer::setupClientArrays(U32 data_mask) { - if (sLastMask != data_mask) - { + if (sLastMask != data_mask) + { + for (U32 i = 0; i < TYPE_MAX; ++i) + { + S32 loc = i; - if (gGLManager.mGLSLVersionMajor < 2 && gGLManager.mGLSLVersionMinor < 30) - { - //make sure texture index is disabled - data_mask = data_mask & ~MAP_TEXTURE_INDEX; - } + U32 mask = 1 << i; - for (U32 i = 0; i < TYPE_MAX; ++i) - { - S32 loc = i; - - U32 mask = 1 << i; - - if (sLastMask & (1 << i)) - { //was enabled - if (!(data_mask & mask)) - { //needs to be disabled - glDisableVertexAttribArray(loc); - } - } - else - { //was disabled - if (data_mask & mask) - { //needs to be enabled - glEnableVertexAttribArray(loc); - } - } - } - - sLastMask = data_mask; - } + if (sLastMask & (1 << i)) + { //was enabled + if (!(data_mask & mask)) + { //needs to be disabled + glDisableVertexAttribArray(loc); + } + } + else + { //was disabled + if (data_mask & mask) + { //needs to be enabled + glEnableVertexAttribArray(loc); + } + } + } + } + + sLastMask = data_mask; } //static @@ -481,7 +624,7 @@ void LLVertexBuffer::drawArrays(U32 mode, const std::vector<LLVector3>& pos) } //static -void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVector2* tc, S32 num_indices, const U16* indicesp) +void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVector2* tc, U32 num_indices, const U16* indicesp) { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; llassert(LLGLSLShader::sCurBoundShaderPtr != NULL); @@ -514,13 +657,18 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto U16 idx = indicesp[i]; gGL.vertex3fv(pos[idx].getF32ptr()); } - } +} gGL.end(); gGL.flush(); } -void LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_offset) const +bool LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_offset) const { + if (!gDebugGL) + { + return true; + } + llassert(start < (U32)mNumVerts); llassert(end < (U32)mNumVerts); @@ -538,9 +686,8 @@ void LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_of LL_ERRS() << "Bad index buffer draw range: [" << indices_offset << ", " << indices_offset+count << "]" << LL_ENDL; } - if (gDebugGL && !useVBOs()) { - U16* idx = ((U16*) getIndicesPointer())+indices_offset; + U16* idx = (U16*) mMappedIndexData+indices_offset; for (U32 i = 0; i < count; ++i) { llassert(idx[i] >= start); @@ -552,234 +699,93 @@ void LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_of } } + LLVector4a* v = (LLVector4a*)mMappedData; + + for (U32 i = start; i <= end; ++i) + { + if (!v[i].isFinite3()) + { + LL_ERRS() << "Non-finite vertex position data detected." << LL_ENDL; + } + } + LLGLSLShader* shader = LLGLSLShader::sCurBoundShaderPtr; if (shader && shader->mFeatures.mIndexedTextureChannels > 1) { - LLStrider<LLVector4a> v; - //hack to get non-const reference - LLVertexBuffer* vb = (LLVertexBuffer*) this; - vb->getVertexStrider(v); - + LLVector4a* v = (LLVector4a*) mMappedData; + for (U32 i = start; i < end; i++) { - S32 idx = (S32) (v[i][3]+0.25f); - llassert(idx >= 0); - if (idx < 0 || idx >= shader->mFeatures.mIndexedTextureChannels) + U32 idx = (U32) (v[i][3]+0.25f); + if (idx >= shader->mFeatures.mIndexedTextureChannels) { LL_ERRS() << "Bad texture index found in vertex data stream." << LL_ENDL; } } } } -} - -void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const -{ - validateRange(start, end, count, indices_offset); - mMappable = false; - gGL.syncMatrices(); - - llassert(mNumVerts >= 0); - llassert(LLGLSLShader::sCurBoundShaderPtr != NULL); - - if (mGLArray) - { - if (mGLArray != sGLRenderArray) - { - LL_ERRS() << "Wrong vertex array bound." << LL_ENDL; - } - } - else - { - if (mGLIndices != sGLRenderIndices) - { - LL_ERRS() << "Wrong index buffer bound." << LL_ENDL; - } - - if (mGLBuffer != sGLRenderBuffer) - { - LL_ERRS() << "Wrong vertex buffer bound." << LL_ENDL; - } - } - if (gDebugGL && !mGLArray && useVBOs()) - { - GLint elem = 0; - glGetIntegerv(GL_ELEMENT_ARRAY_BUFFER_BINDING, &elem); - - if (elem != mGLIndices) - { - LL_ERRS() << "Wrong index buffer bound!" << LL_ENDL; - } - } - - if (mode >= LLRender::NUM_MODES) - { - LL_ERRS() << "Invalid draw mode: " << mode << LL_ENDL; - return; - } - - U16* idx = ((U16*) getIndicesPointer())+indices_offset; - - stop_glerror(); - LLGLSLShader::startProfile(); - LL_PROFILER_GPU_ZONEC( "gl.DrawRangeElements", 0xFFFF00 ) - glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, - idx); - LLGLSLShader::stopProfile(count, mode); - stop_glerror(); - - + return true; +} - placeFence(); +#ifdef LL_PROFILER_ENABLE_RENDER_DOC +void LLVertexBuffer::setLabel(const char* label) { + LL_LABEL_OBJECT_GL(GL_BUFFER, mGLBuffer, strlen(label), label); } +#endif -void LLVertexBuffer::drawRangeFast(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const +void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const { - mMappable = false; + llassert(validateRange(start, end, count, indices_offset)); + llassert(mGLBuffer == sGLRenderBuffer); + llassert(mGLIndices == sGLRenderIndices); gGL.syncMatrices(); - - U16* idx = ((U16*)getIndicesPointer()) + indices_offset; - - LL_PROFILER_GPU_ZONEC("gl.DrawRangeElements", 0xFFFF00) - glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, - idx); + glDrawRangeElements(sGLMode[mode], start, end, count, GL_UNSIGNED_SHORT, + (GLvoid*) (indices_offset * sizeof(U16))); } void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const { - llassert(LLGLSLShader::sCurBoundShaderPtr != NULL); - mMappable = false; - gGL.syncMatrices(); - - llassert(mNumIndices >= 0); - if (indices_offset >= (U32) mNumIndices || - indices_offset + count > (U32) mNumIndices) - { - LL_ERRS() << "Bad index buffer draw range: [" << indices_offset << ", " << indices_offset+count << "]" << LL_ENDL; - } - - if (mGLArray) - { - if (mGLArray != sGLRenderArray) - { - LL_ERRS() << "Wrong vertex array bound." << LL_ENDL; - } - } - else - { - if (mGLIndices != sGLRenderIndices) - { - LL_ERRS() << "Wrong index buffer bound." << LL_ENDL; - } - - if (mGLBuffer != sGLRenderBuffer) - { - LL_ERRS() << "Wrong vertex buffer bound." << LL_ENDL; - } - } - - if (mode >= LLRender::NUM_MODES) - { - LL_ERRS() << "Invalid draw mode: " << mode << LL_ENDL; - return; - } - - stop_glerror(); - LLGLSLShader::startProfile(); - LL_PROFILER_GPU_ZONEC( "gl.DrawElements", 0xA0FFA0 ) - glDrawElements(sGLMode[mode], count, GL_UNSIGNED_SHORT, - ((U16*) getIndicesPointer()) + indices_offset); - LLGLSLShader::stopProfile(count, mode); - stop_glerror(); - placeFence(); + drawRange(mode, 0, mNumVerts-1, count, indices_offset); } void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const { - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; - llassert(LLGLSLShader::sCurBoundShaderPtr != NULL); - mMappable = false; + llassert(first + count <= mNumVerts); + llassert(mGLBuffer == sGLRenderBuffer); + llassert(mGLIndices == sGLRenderIndices); + gGL.syncMatrices(); + glDrawArrays(sGLMode[mode], first, count); +} -#ifndef LL_RELEASE_FOR_DOWNLOAD - llassert(mNumVerts >= 0); - if (first >= (U32)mNumVerts || - first + count > (U32)mNumVerts) - { - LL_ERRS() << "Bad vertex buffer draw range: [" << first << ", " << first + count << "]" << LL_ENDL; - } +//static +void LLVertexBuffer::initClass(LLWindow* window) +{ + llassert(sVBOPool == nullptr); + sVBOPool = new LLVBOPool(); - if (mGLArray) - { - if (mGLArray != sGLRenderArray) - { - LL_ERRS() << "Wrong vertex array bound." << LL_ENDL; - } - } - else - { - if (mGLBuffer != sGLRenderBuffer || useVBOs() != sVBOActive) - { - LL_ERRS() << "Wrong vertex buffer bound." << LL_ENDL; - } - } +#if ENABLE_GL_WORK_QUEUE + sQueue = new GLWorkQueue(); - if (mode >= LLRender::NUM_MODES) + for (int i = 0; i < THREAD_COUNT; ++i) { - LL_ERRS() << "Invalid draw mode: " << mode << LL_ENDL; - return; + sVBOThread[i] = new LLGLWorkerThread("VBO Worker", sQueue, window); + sVBOThread[i]->start(); } #endif - - LLGLSLShader::startProfile(); - { - LL_PROFILER_GPU_ZONEC("gl.DrawArrays", 0xFF4040) - glDrawArrays(sGLMode[mode], first, count); - } - LLGLSLShader::stopProfile(count, mode); - - stop_glerror(); - placeFence(); -} - -//static -void LLVertexBuffer::initClass(bool use_vbo, bool no_vbo_mapping) -{ - sEnableVBOs = use_vbo && gGLManager.mHasVertexBufferObject; - sDisableVBOMapping = sEnableVBOs && no_vbo_mapping; } //static void LLVertexBuffer::unbind() { - if (sGLRenderArray) - { -//#if GL_ARB_vertex_array_object - glBindVertexArray(0); -//#endif - sGLRenderArray = 0; - sGLRenderIndices = 0; - sIBOActive = false; - } - - if (sVBOActive) - { - glBindBuffer(GL_ARRAY_BUFFER, 0); - sVBOActive = false; - } - if (sIBOActive) - { - glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0); - sIBOActive = false; - } + glBindBuffer(GL_ARRAY_BUFFER, 0); + glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0); sGLRenderBuffer = 0; sGLRenderIndices = 0; - - setupClientArrays(0); } //static @@ -787,98 +793,41 @@ void LLVertexBuffer::cleanupClass() { unbind(); - sStreamIBOPool.cleanup(); - sDynamicIBOPool.cleanup(); - sStreamVBOPool.cleanup(); - sDynamicVBOPool.cleanup(); - sDynamicCopyVBOPool.cleanup(); -} + delete sVBOPool; + sVBOPool = nullptr; -//---------------------------------------------------------------------------- - -S32 LLVertexBuffer::determineUsage(S32 usage) -{ - S32 ret_usage = usage; +#if ENABLE_GL_WORK_QUEUE + sQueue->close(); + for (int i = 0; i < THREAD_COUNT; ++i) + { + sVBOThread[i]->shutdown(); + delete sVBOThread[i]; + sVBOThread[i] = nullptr; + } - if (!sEnableVBOs) - { - ret_usage = 0; - } - - if (ret_usage == GL_STREAM_DRAW && !sUseStreamDraw) - { - ret_usage = 0; - } - - if (ret_usage == GL_DYNAMIC_DRAW && sPreferStreamDraw) - { - ret_usage = GL_STREAM_DRAW; - } - - if (ret_usage == 0 && LLRender::sGLCoreProfile) - { //MUST use VBOs for all rendering - ret_usage = GL_STREAM_DRAW; - } - - if (ret_usage && ret_usage != GL_STREAM_DRAW) - { //only stream_draw and dynamic_draw are supported when using VBOs, dynamic draw is the default - if (ret_usage != GL_DYNAMIC_COPY) - { - if (sDisableVBOMapping) - { //always use stream draw if VBO mapping is disabled - ret_usage = GL_STREAM_DRAW; - } - else - { - ret_usage = GL_DYNAMIC_DRAW; - } - } - } - - return ret_usage; + delete sQueue; + sQueue = nullptr; +#endif } -LLVertexBuffer::LLVertexBuffer(U32 typemask, S32 usage) -: LLRefCount(), - - mNumVerts(0), - mNumIndices(0), - mAlignedOffset(0), - mAlignedIndexOffset(0), - mSize(0), - mIndicesSize(0), - mTypeMask(typemask), - mUsage(LLVertexBuffer::determineUsage(usage)), - mGLBuffer(0), - mGLIndices(0), - mGLArray(0), - mMappedData(NULL), - mMappedIndexData(NULL), - mMappedDataUsingVBOs(false), - mMappedIndexDataUsingVBOs(false), - mVertexLocked(false), - mIndexLocked(false), - mFinal(false), - mEmpty(true), - mMappable(false), - mFence(NULL) -{ - mMappable = (mUsage == GL_DYNAMIC_DRAW && !sDisableVBOMapping); +//---------------------------------------------------------------------------- +LLVertexBuffer::LLVertexBuffer(U32 typemask) +: LLRefCount(), + mTypeMask(typemask) +{ //zero out offsets for (U32 i = 0; i < TYPE_MAX; i++) { mOffsets[i] = 0; } - - sCount++; } //static -S32 LLVertexBuffer::calcOffsets(const U32& typemask, S32* offsets, S32 num_vertices) +U32 LLVertexBuffer::calcOffsets(const U32& typemask, U32* offsets, U32 num_vertices) { - S32 offset = 0; - for (S32 i=0; i<TYPE_TEXTURE_INDEX; i++) + U32 offset = 0; + for (U32 i=0; i<TYPE_TEXTURE_INDEX; i++) { U32 mask = 1<<i; if (typemask & mask) @@ -894,14 +843,14 @@ S32 LLVertexBuffer::calcOffsets(const U32& typemask, S32* offsets, S32 num_verti offsets[TYPE_TEXTURE_INDEX] = offsets[TYPE_VERTEX] + 12; - return offset+16; + return offset; } //static -S32 LLVertexBuffer::calcVertexSize(const U32& typemask) +U32 LLVertexBuffer::calcVertexSize(const U32& typemask) { - S32 size = 0; - for (S32 i = 0; i < TYPE_TEXTURE_INDEX; i++) + U32 size = 0; + for (U32 i = 0; i < TYPE_TEXTURE_INDEX; i++) { U32 mask = 1<<i; if (typemask & mask) @@ -913,11 +862,6 @@ S32 LLVertexBuffer::calcVertexSize(const U32& typemask) return size; } -S32 LLVertexBuffer::getSize() const -{ - return mSize; -} - // protected, use unref() //virtual LLVertexBuffer::~LLVertexBuffer() @@ -925,27 +869,6 @@ LLVertexBuffer::~LLVertexBuffer() destroyGLBuffer(); destroyGLIndices(); - if (mGLArray) - { -//#if GL_ARB_vertex_array_object - releaseVAOName(mGLArray); -//#endif - } - - sCount--; - - if (mFence) - { - // Sanity check. We have weird crashes in this destructor (on delete). Yet mFence is disabled. - // TODO: mFence was added in scope of SH-2038, but was never enabled, consider removing mFence. - LL_ERRS() << "LLVertexBuffer destruction failed" << LL_ENDL; - delete mFence; - mFence = NULL; - } - - sVertexCount -= mNumVerts; - sIndexCount -= mNumIndices; - if (mMappedData) { LL_ERRS() << "Failed to clear vertex buffer's vertices" << LL_ENDL; @@ -956,101 +879,37 @@ LLVertexBuffer::~LLVertexBuffer() } }; -void LLVertexBuffer::placeFence() const -{ - /*if (!mFence && useVBOs()) - { - if (gGLManager.mHasSync) - { - mFence = new LLGLSyncFence(); - } - } - - if (mFence) - { - mFence->placeFence(); - }*/ -} - -void LLVertexBuffer::waitFence() const -{ - /*if (mFence) - { - mFence->wait(); - }*/ -} - //---------------------------------------------------------------------------- void LLVertexBuffer::genBuffer(U32 size) { - mSize = vbo_block_size(size); - - if (mUsage == GL_STREAM_DRAW) - { - mMappedData = sStreamVBOPool.allocate(mGLBuffer, mSize); - } - else if (mUsage == GL_DYNAMIC_DRAW) - { - mMappedData = sDynamicVBOPool.allocate(mGLBuffer, mSize); - } - else - { - mMappedData = sDynamicCopyVBOPool.allocate(mGLBuffer, mSize); - } - - - sGLCount++; -} - -void LLVertexBuffer::genIndices(U32 size) -{ - mIndicesSize = vbo_block_size(size); - - if (mUsage == GL_STREAM_DRAW) - { - mMappedIndexData = sStreamIBOPool.allocate(mGLIndices, mIndicesSize); - } - else - { - mMappedIndexData = sDynamicIBOPool.allocate(mGLIndices, mIndicesSize); - } - - sGLCount++; -} + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + llassert(sVBOPool); -void LLVertexBuffer::releaseBuffer() -{ - if (mUsage == GL_STREAM_DRAW) - { - sStreamVBOPool.release(mGLBuffer, mMappedData, mSize); - } - else - { - sDynamicVBOPool.release(mGLBuffer, mMappedData, mSize); - } - - mGLBuffer = 0; - mMappedData = NULL; + if (sVBOPool) + { + llassert(mSize == 0); + llassert(mGLBuffer == 0); + llassert(mMappedData == nullptr); - sGLCount--; + mSize = size; + sVBOPool->allocate(GL_ARRAY_BUFFER, mSize, mGLBuffer, mMappedData); + } } -void LLVertexBuffer::releaseIndices() +void LLVertexBuffer::genIndices(U32 size) { - if (mUsage == GL_STREAM_DRAW) - { - sStreamIBOPool.release(mGLIndices, mMappedIndexData, mIndicesSize); - } - else - { - sDynamicIBOPool.release(mGLIndices, mMappedIndexData, mIndicesSize); - } + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + llassert(sVBOPool); - mGLIndices = 0; - mMappedIndexData = NULL; - - sGLCount--; + if (sVBOPool) + { + llassert(mIndicesSize == 0); + llassert(mGLIndices == 0); + llassert(mMappedIndexData == nullptr); + mIndicesSize = size; + sVBOPool->allocate(GL_ELEMENT_ARRAY_BUFFER, mIndicesSize, mGLIndices, mMappedIndexData); + } } bool LLVertexBuffer::createGLBuffer(U32 size) @@ -1067,22 +926,8 @@ bool LLVertexBuffer::createGLBuffer(U32 size) bool success = true; - mEmpty = true; - - mMappedDataUsingVBOs = useVBOs(); + genBuffer(size); - if (mMappedDataUsingVBOs) - { - genBuffer(size); - } - else - { - static int gl_buffer_idx = 0; - mGLBuffer = ++gl_buffer_idx; - mMappedData = (U8*)ll_aligned_malloc_16(size); - mSize = size; - } - if (!mMappedData) { success = false; @@ -1104,27 +949,8 @@ bool LLVertexBuffer::createGLIndices(U32 size) bool success = true; - mEmpty = true; - - //pad by 16 bytes for aligned copies - size += 16; - - mMappedIndexDataUsingVBOs = useVBOs(); - - if (mMappedIndexDataUsingVBOs) - { - //pad by another 16 bytes for VBO pointer adjustment - size += 16; - genIndices(size); - } - else - { - mMappedIndexData = (U8*)ll_aligned_malloc_16(size); - static int gl_buffer_idx = 0; - mGLIndices = ++gl_buffer_idx; - mIndicesSize = size; - } - + genIndices(size); + if (!mMappedIndexData) { success = false; @@ -1136,43 +962,37 @@ void LLVertexBuffer::destroyGLBuffer() { if (mGLBuffer || mMappedData) { - if (mMappedDataUsingVBOs) - { - releaseBuffer(); - } - else - { - ll_aligned_free_16((void*)mMappedData); - mMappedData = NULL; - mEmpty = true; - } + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + //llassert(sVBOPool); + if (sVBOPool) + { + sVBOPool->free(GL_ARRAY_BUFFER, mSize, mGLBuffer, mMappedData); + } + + mSize = 0; + mGLBuffer = 0; + mMappedData = nullptr; } - - mGLBuffer = 0; - //unbind(); } void LLVertexBuffer::destroyGLIndices() { if (mGLIndices || mMappedIndexData) { - if (mMappedIndexDataUsingVBOs) - { - releaseIndices(); - } - else - { - ll_aligned_free_16((void*)mMappedIndexData); - mMappedIndexData = NULL; - mEmpty = true; - } - } + LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; + //llassert(sVBOPool); + if (sVBOPool) + { + sVBOPool->free(GL_ELEMENT_ARRAY_BUFFER, mIndicesSize, mGLIndices, mMappedIndexData); + } - mGLIndices = 0; - //unbind(); + mIndicesSize = 0; + mGLIndices = 0; + mMappedIndexData = nullptr; + } } -bool LLVertexBuffer::updateNumVerts(S32 nverts) +bool LLVertexBuffer::updateNumVerts(U32 nverts) { llassert(nverts >= 0); @@ -1186,19 +1006,17 @@ bool LLVertexBuffer::updateNumVerts(S32 nverts) U32 needed_size = calcOffsets(mTypeMask, mOffsets, nverts); - if (needed_size > mSize || needed_size <= mSize/2) - { - success &= createGLBuffer(needed_size); - } + if (needed_size != mSize) + { + success &= createGLBuffer(needed_size); + } - sVertexCount -= mNumVerts; + llassert(mSize == needed_size); mNumVerts = nverts; - sVertexCount += mNumVerts; - return success; } -bool LLVertexBuffer::updateNumIndices(S32 nindices) +bool LLVertexBuffer::updateNumIndices(U32 nindices) { llassert(nindices >= 0); @@ -1206,22 +1024,18 @@ bool LLVertexBuffer::updateNumIndices(S32 nindices) U32 needed_size = sizeof(U16) * nindices; - if (needed_size > mIndicesSize || needed_size <= mIndicesSize/2) + if (needed_size != mIndicesSize) { success &= createGLIndices(needed_size); } - sIndexCount -= mNumIndices; + llassert(mIndicesSize == needed_size); mNumIndices = nindices; - sIndexCount += mNumIndices; - return success; } -bool LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create) +bool LLVertexBuffer::allocateBuffer(U32 nverts, U32 nindices) { - stop_glerror(); - if (nverts < 0 || nindices < 0 || nverts > 65536) { @@ -1233,735 +1047,215 @@ bool LLVertexBuffer::allocateBuffer(S32 nverts, S32 nindices, bool create) success &= updateNumVerts(nverts); success &= updateNumIndices(nindices); - if (create && (nverts || nindices)) - { - //actually allocate space for the vertex buffer if using VBO mapping - flush(); //unmap - - if (gGLManager.mHasVertexArrayObject && useVBOs() && sUseVAO) - { -//#if GL_ARB_vertex_array_object - mGLArray = getVAOName(); -//#endif - setupVertexArray(); - } - } - return success; } -void LLVertexBuffer::setupVertexArray() -{ - if (!mGLArray) - { - return; - } - - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; -//#if GL_ARB_vertex_array_object - glBindVertexArray(mGLArray); -//#endif - sGLRenderArray = mGLArray; - - static const U32 attrib_size[] = - { - 3, //TYPE_VERTEX, - 3, //TYPE_NORMAL, - 2, //TYPE_TEXCOORD0, - 2, //TYPE_TEXCOORD1, - 2, //TYPE_TEXCOORD2, - 2, //TYPE_TEXCOORD3, - 4, //TYPE_COLOR, - 4, //TYPE_EMISSIVE, - 4, //TYPE_TANGENT, - 1, //TYPE_WEIGHT, - 4, //TYPE_WEIGHT4, - 4, //TYPE_CLOTHWEIGHT, - 1, //TYPE_TEXTURE_INDEX - }; - - static const U32 attrib_type[] = - { - GL_FLOAT, //TYPE_VERTEX, - GL_FLOAT, //TYPE_NORMAL, - GL_FLOAT, //TYPE_TEXCOORD0, - GL_FLOAT, //TYPE_TEXCOORD1, - GL_FLOAT, //TYPE_TEXCOORD2, - GL_FLOAT, //TYPE_TEXCOORD3, - GL_UNSIGNED_BYTE, //TYPE_COLOR, - GL_UNSIGNED_BYTE, //TYPE_EMISSIVE, - GL_FLOAT, //TYPE_TANGENT, - GL_FLOAT, //TYPE_WEIGHT, - GL_FLOAT, //TYPE_WEIGHT4, - GL_FLOAT, //TYPE_CLOTHWEIGHT, - GL_UNSIGNED_INT, //TYPE_TEXTURE_INDEX - }; - - static const bool attrib_integer[] = - { - false, //TYPE_VERTEX, - false, //TYPE_NORMAL, - false, //TYPE_TEXCOORD0, - false, //TYPE_TEXCOORD1, - false, //TYPE_TEXCOORD2, - false, //TYPE_TEXCOORD3, - false, //TYPE_COLOR, - false, //TYPE_EMISSIVE, - false, //TYPE_TANGENT, - false, //TYPE_WEIGHT, - false, //TYPE_WEIGHT4, - false, //TYPE_CLOTHWEIGHT, - true, //TYPE_TEXTURE_INDEX - }; - - static const U32 attrib_normalized[] = - { - GL_FALSE, //TYPE_VERTEX, - GL_FALSE, //TYPE_NORMAL, - GL_FALSE, //TYPE_TEXCOORD0, - GL_FALSE, //TYPE_TEXCOORD1, - GL_FALSE, //TYPE_TEXCOORD2, - GL_FALSE, //TYPE_TEXCOORD3, - GL_TRUE, //TYPE_COLOR, - GL_TRUE, //TYPE_EMISSIVE, - GL_FALSE, //TYPE_TANGENT, - GL_FALSE, //TYPE_WEIGHT, - GL_FALSE, //TYPE_WEIGHT4, - GL_FALSE, //TYPE_CLOTHWEIGHT, - GL_FALSE, //TYPE_TEXTURE_INDEX - }; - - bindGLBuffer(true); - bindGLIndices(true); - - for (U32 i = 0; i < TYPE_MAX; ++i) - { - if (mTypeMask & (1 << i)) - { - glEnableVertexAttribArray(i); - - if (attrib_integer[i]) - { -#if !LL_DARWIN - //glVertexattribIPointer requires GLSL 1.30 or later - if (gGLManager.mGLSLVersionMajor > 1 || gGLManager.mGLSLVersionMinor >= 30) - { - // nat 2018-10-24: VS 2017 also notices the issue - // described below, and warns even with reinterpret_cast. - // Cast via intptr_t to make it painfully obvious to the - // compiler that we're doing this intentionally. - glVertexAttribIPointer(i, attrib_size[i], attrib_type[i], sTypeSize[i], - reinterpret_cast<const GLvoid*>(intptr_t(mOffsets[i]))); - } -#endif - } - else - { - // nat 2016-12-16: With 64-bit clang compile, the compiler - // produces an error if we simply cast mOffsets[i] -- an S32 - // -- to (GLvoid *), the type of the parameter. It correctly - // points out that there's no way an S32 could fit a real - // pointer value. Ruslan asserts that in this case the last - // param is interpreted as an array data offset within the VBO - // rather than as an actual pointer, so it's okay. - glVertexAttribPointer(i, attrib_size[i], attrib_type[i], - attrib_normalized[i], sTypeSize[i], - reinterpret_cast<GLvoid*>(intptr_t(mOffsets[i]))); - } - } - else - { - glDisableVertexAttribArray(i); - } - } - - //draw a dummy triangle to set index array pointer - //glDrawElements(GL_TRIANGLES, 0, GL_UNSIGNED_SHORT, NULL); - - unbind(); -} - -bool LLVertexBuffer::resizeBuffer(S32 newnverts, S32 newnindices) -{ - llassert(newnverts >= 0); - llassert(newnindices >= 0); - - bool success = true; - - success &= updateNumVerts(newnverts); - success &= updateNumIndices(newnindices); - - if (useVBOs()) - { - flush(); //unmap - - if (mGLArray) - { //if size changed, offsets changed - setupVertexArray(); - } - } - - return success; -} - -bool LLVertexBuffer::useVBOs() const -{ - //it's generally ineffective to use VBO for things that are streaming on apple - return (mUsage != 0); -} - //---------------------------------------------------------------------------- -bool expand_region(LLVertexBuffer::MappedRegion& region, S32 index, S32 count) +// if no gap between region and given range exists, expand region to cover given range and return true +// otherwise return false +bool expand_region(LLVertexBuffer::MappedRegion& region, U32 start, U32 end) { - S32 end = index+count; - S32 region_end = region.mIndex+region.mCount; - if (end < region.mIndex || - index > region_end) + if (end < region.mStart || + start > region.mEnd) { //gap exists, do not merge return false; } - S32 new_end = llmax(end, region_end); - S32 new_index = llmin(index, region.mIndex); - region.mIndex = new_index; - region.mCount = new_end-new_index; + region.mStart = llmin(region.mStart, start); + region.mEnd = llmax(region.mEnd, end); + return true; } // Map for data access -U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range) +U8* LLVertexBuffer::mapVertexBuffer(LLVertexBuffer::AttributeType type, U32 index, S32 count) { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; - bindGLBuffer(true); - if (mFinal) - { - LL_ERRS() << "LLVertexBuffer::mapVeretxBuffer() called on a finalized buffer." << LL_ENDL; - } - if (!useVBOs() && !mMappedData && !mMappedIndexData) - { - LL_ERRS() << "LLVertexBuffer::mapVertexBuffer() called on unallocated buffer." << LL_ENDL; - } - - if (useVBOs()) - { - if (!mMappable || gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange) - { - if (count == -1) - { - count = mNumVerts-index; - } - - bool mapped = false; - //see if range is already mapped - for (U32 i = 0; i < mMappedVertexRegions.size(); ++i) - { - MappedRegion& region = mMappedVertexRegions[i]; - if (region.mType == type) - { - if (expand_region(region, index, count)) - { - mapped = true; - break; - } - } - } - - if (!mapped) - { - //not already mapped, map new region - MappedRegion region(type, mMappable && map_range ? -1 : index, count); - mMappedVertexRegions.push_back(region); - } - } - - if (mVertexLocked && map_range) - { - LL_ERRS() << "Attempted to map a specific range of a buffer that was already mapped." << LL_ENDL; - } - - if (!mVertexLocked) - { - mVertexLocked = true; - sMappedCount++; - stop_glerror(); - - if(!mMappable) - { - map_range = false; - } - else - { - U8* src = NULL; - waitFence(); - if (gGLManager.mHasMapBufferRange) - { - if (map_range) - { -#if GL_ARB_map_buffer_range || GL_EXT_map_buffer_range - S32 offset = mOffsets[type] + sTypeSize[type]*index; - S32 length = (sTypeSize[type]*count+0xF) & ~0xF; - src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER, offset, length, - GL_MAP_WRITE_BIT | - GL_MAP_FLUSH_EXPLICIT_BIT | - GL_MAP_INVALIDATE_RANGE_BIT); -#endif - } - else - { -#if GL_ARB_map_buffer_range || GL_EXT_map_buffer_range - - if (gDebugGL) - { - GLint size = 0; - glGetBufferParameteriv(GL_ARRAY_BUFFER, GL_BUFFER_SIZE, &size); - - if (size < mSize) - { - LL_ERRS() << "Invalid buffer size." << LL_ENDL; - } - } - - src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER, 0, mSize, - GL_MAP_WRITE_BIT | - GL_MAP_FLUSH_EXPLICIT_BIT); -#endif - } - } - else if (gGLManager.mHasFlushBufferRange) - { - if (map_range) - { -#if !LL_MESA_HEADLESS && GL_APPLE_flush_buffer_range && LL_DARWIN - glBufferParameteriAPPLE(GL_ARRAY_BUFFER_ARB, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE); - glBufferParameteriAPPLE(GL_ARRAY_BUFFER_ARB, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE); -#endif - src = (U8*) glMapBuffer(GL_ARRAY_BUFFER, GL_WRITE_ONLY); - } - else - { - src = (U8*) glMapBuffer(GL_ARRAY_BUFFER, GL_WRITE_ONLY); - } - } - else - { - map_range = false; - src = (U8*) glMapBuffer(GL_ARRAY_BUFFER, GL_WRITE_ONLY); - } - - llassert(src != NULL); + + if (count == -1) + { + count = mNumVerts - index; + } - mMappedData = LL_NEXT_ALIGNED_ADDRESS<U8>(src); - mAlignedOffset = mMappedData - src; - - stop_glerror(); - } - - if (!mMappedData) - { - log_glerror(); + U32 start = mOffsets[type] + sTypeSize[type] * index; + U32 end = start + sTypeSize[type] * count-1; - //check the availability of memory - LLMemory::logMemoryInfo(true); - - if(mMappable) - { - //-------------------- - //print out more debug info before crash - LL_INFOS() << "vertex buffer size: (num verts : num indices) = " << getNumVerts() << " : " << getNumIndices() << LL_ENDL; - GLint size; - glGetBufferParameteriv(GL_ARRAY_BUFFER, GL_BUFFER_SIZE, &size); - LL_INFOS() << "GL_ARRAY_BUFFER_ARB size is " << size << LL_ENDL; - //-------------------- - - GLint buff; - glGetIntegerv(GL_ARRAY_BUFFER_BINDING, &buff); - if ((GLuint)buff != mGLBuffer) - { - LL_ERRS() << "Invalid GL vertex buffer bound: " << buff << LL_ENDL; - } - - - LL_ERRS() << "glMapBuffer returned NULL (no vertex data)" << LL_ENDL; - } - else - { - LL_ERRS() << "memory allocation for vertex data failed." << LL_ENDL; - } - } - } - } - else - { - map_range = false; - } - - if (map_range && gGLManager.mHasMapBufferRange && mMappable) + bool flagged = false; + // flag region as mapped + for (U32 i = 0; i < mMappedVertexRegions.size(); ++i) { - return mMappedData; + MappedRegion& region = mMappedVertexRegions[i]; + if (expand_region(region, start, end)) + { + flagged = true; + break; + } } - else + + if (!flagged) { - return mMappedData+mOffsets[type]+sTypeSize[type]*index; + //didn't expand an existing region, make a new one + mMappedVertexRegions.push_back({ start, end }); } + + return mMappedData+mOffsets[type]+sTypeSize[type]*index; } -U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range) +U8* LLVertexBuffer::mapIndexBuffer(U32 index, S32 count) { LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; - bindGLIndices(true); - if (mFinal) - { - LL_ERRS() << "LLVertexBuffer::mapIndexBuffer() called on a finalized buffer." << LL_ENDL; - } - if (!useVBOs() && !mMappedData && !mMappedIndexData) + + if (count == -1) { - LL_ERRS() << "LLVertexBuffer::mapIndexBuffer() called on unallocated buffer." << LL_ENDL; + count = mNumIndices-index; } - if (useVBOs()) - { - if (!mMappable || gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange) - { - if (count == -1) - { - count = mNumIndices-index; - } - - bool mapped = false; - //see if range is already mapped - for (U32 i = 0; i < mMappedIndexRegions.size(); ++i) - { - MappedRegion& region = mMappedIndexRegions[i]; - if (expand_region(region, index, count)) - { - mapped = true; - break; - } - } - - if (!mapped) - { - //not already mapped, map new region - MappedRegion region(TYPE_INDEX, mMappable && map_range ? -1 : index, count); - mMappedIndexRegions.push_back(region); - } - } - - if (mIndexLocked && map_range) - { - LL_ERRS() << "Attempted to map a specific range of a buffer that was already mapped." << LL_ENDL; - } - - if (!mIndexLocked) - { - mIndexLocked = true; - sMappedCount++; - stop_glerror(); - - if (gDebugGL && useVBOs()) - { - GLint elem = 0; - glGetIntegerv(GL_ELEMENT_ARRAY_BUFFER_BINDING, &elem); - - if (elem != mGLIndices) - { - LL_ERRS() << "Wrong index buffer bound!" << LL_ENDL; - } - } - - if(!mMappable) - { - map_range = false; - } - else - { - U8* src = NULL; - waitFence(); - if (gGLManager.mHasMapBufferRange) - { - if (map_range) - { -#if GL_ARB_map_buffer_range || GL_EXT_map_buffer_range - S32 offset = sizeof(U16)*index; - S32 length = sizeof(U16)*count; - src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER, offset, length, - GL_MAP_WRITE_BIT | - GL_MAP_FLUSH_EXPLICIT_BIT | - GL_MAP_INVALIDATE_RANGE_BIT); -#endif - } - else - { -#if GL_ARB_map_buffer_range || GL_EXT_map_buffer_range - src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(U16)*mNumIndices, - GL_MAP_WRITE_BIT | - GL_MAP_FLUSH_EXPLICIT_BIT); -#endif - } - } - else if (gGLManager.mHasFlushBufferRange) - { - if (map_range) - { -#if !LL_MESA_HEADLESS && GL_APPLE_flush_buffer_range && LL_DARWIN - glBufferParameteriAPPLE(GL_ELEMENT_ARRAY_BUFFER_ARB, GL_BUFFER_SERIALIZED_MODIFY_APPLE, GL_FALSE); - glBufferParameteriAPPLE(GL_ELEMENT_ARRAY_BUFFER_ARB, GL_BUFFER_FLUSHING_UNMAP_APPLE, GL_FALSE); -#endif - src = (U8*) glMapBuffer(GL_ELEMENT_ARRAY_BUFFER, GL_WRITE_ONLY); - } - else - { - src = (U8*) glMapBuffer(GL_ELEMENT_ARRAY_BUFFER, GL_WRITE_ONLY); - } - } - else - { - map_range = false; - src = (U8*) glMapBuffer(GL_ELEMENT_ARRAY_BUFFER, GL_WRITE_ONLY); - } - - llassert(src != NULL); + U32 start = sizeof(U16) * index; + U32 end = start + sizeof(U16) * count-1; + bool flagged = false; + // flag region as mapped + for (U32 i = 0; i < mMappedIndexRegions.size(); ++i) + { + MappedRegion& region = mMappedIndexRegions[i]; + if (expand_region(region, start, end)) + { + flagged = true; + break; + } + } - mMappedIndexData = src; //LL_NEXT_ALIGNED_ADDRESS<U8>(src); - mAlignedIndexOffset = mMappedIndexData - src; - stop_glerror(); - } - } + if (!flagged) + { + //didn't expand an existing region, make a new one + mMappedIndexRegions.push_back({ start, end }); + } - if (!mMappedIndexData) - { - log_glerror(); - LLMemory::logMemoryInfo(true); + return mMappedIndexData + sizeof(U16)*index; +} - if(mMappable) - { - GLint buff; - glGetIntegerv(GL_ELEMENT_ARRAY_BUFFER_BINDING, &buff); - if ((GLuint)buff != mGLIndices) - { - LL_ERRS() << "Invalid GL index buffer bound: " << buff << LL_ENDL; - } +// flush the given byte range +// target -- "targret" parameter for glBufferSubData +// start -- first byte to copy +// end -- last byte to copy (NOT last byte + 1) +// data -- mMappedData or mMappedIndexData +static void flush_vbo(GLenum target, U32 start, U32 end, void* data) +{ + if (end != 0) + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("glBufferSubData"); + LL_PROFILE_ZONE_NUM(start); + LL_PROFILE_ZONE_NUM(end); + LL_PROFILE_ZONE_NUM(end-start); - LL_ERRS() << "glMapBuffer returned NULL (no index data)" << LL_ENDL; - } - else - { - LL_ERRS() << "memory allocation for Index data failed. " << LL_ENDL; - } - } - } - else - { - map_range = false; - } + constexpr U32 block_size = 8192; - if (map_range && gGLManager.mHasMapBufferRange && mMappable) - { - return mMappedIndexData; - } - else - { - return mMappedIndexData + sizeof(U16)*index; - } + for (U32 i = start; i <= end; i += block_size) + { + LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("glBufferSubData block"); + //LL_PROFILE_GPU_ZONE("glBufferSubData"); + U32 tend = llmin(i + block_size, end); + U32 size = tend - i + 1; + glBufferSubData(target, i, size, (U8*) data + (i-start)); + } + } } void LLVertexBuffer::unmapBuffer() { - if (!useVBOs()) - { - return; //nothing to unmap - } + struct SortMappedRegion + { + bool operator()(const MappedRegion& lhs, const MappedRegion& rhs) + { + return lhs.mStart < rhs.mStart; + } + }; - bool updated_all = false; - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; - if (mMappedData && mVertexLocked) + if (!mMappedVertexRegions.empty()) { LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - vertex"); - bindGLBuffer(true); - updated_all = mIndexLocked; //both vertex and index buffers done updating + if (sGLRenderBuffer != mGLBuffer) + { + glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer); + sGLRenderBuffer = mGLBuffer; + } + + U32 start = 0; + U32 end = 0; - if(!mMappable) - { - if (!mMappedVertexRegions.empty()) - { - stop_glerror(); - for (U32 i = 0; i < mMappedVertexRegions.size(); ++i) - { - const MappedRegion& region = mMappedVertexRegions[i]; - S32 offset = region.mIndex >= 0 ? mOffsets[region.mType]+sTypeSize[region.mType]*region.mIndex : 0; - S32 length = sTypeSize[region.mType]*region.mCount; - if (mSize >= length + offset) - { - glBufferSubData(GL_ARRAY_BUFFER, offset, length, (U8*)mMappedData + offset); - } - else - { - GLint size = 0; - glGetBufferParameteriv(GL_ARRAY_BUFFER, GL_BUFFER_SIZE, &size); - LL_WARNS() << "Attempted to map regions to a buffer that is too small, " - << "mapped size: " << mSize - << ", gl buffer size: " << size - << ", length: " << length - << ", offset: " << offset - << LL_ENDL; - } - stop_glerror(); - } + std::sort(mMappedVertexRegions.begin(), mMappedVertexRegions.end(), SortMappedRegion()); - mMappedVertexRegions.clear(); - } - else - { - stop_glerror(); - glBufferSubData(GL_ARRAY_BUFFER, 0, getSize(), (U8*) mMappedData); - stop_glerror(); - } - } - else + for (U32 i = 0; i < mMappedVertexRegions.size(); ++i) { - if (gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange) - { - if (!mMappedVertexRegions.empty()) - { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - flush vertex"); - for (U32 i = 0; i < mMappedVertexRegions.size(); ++i) - { - const MappedRegion& region = mMappedVertexRegions[i]; - S32 offset = region.mIndex >= 0 ? mOffsets[region.mType]+sTypeSize[region.mType]*region.mIndex : 0; - S32 length = sTypeSize[region.mType]*region.mCount; - if (gGLManager.mHasMapBufferRange) - { -#if GL_ARB_map_buffer_range || GL_EXT_map_buffer_range - glFlushMappedBufferRange(GL_ARRAY_BUFFER, offset, length); -#endif - } - else if (gGLManager.mHasFlushBufferRange) - { -#ifndef LL_MESA_HEADLESS - glFlushMappedBufferRange(GL_ARRAY_BUFFER, offset, length); -#endif - } - } - - mMappedVertexRegions.clear(); - } - } - stop_glerror(); - glUnmapBuffer(GL_ARRAY_BUFFER); - stop_glerror(); - - mMappedData = NULL; + const MappedRegion& region = mMappedVertexRegions[i]; + if (region.mStart == end + 1) + { + end = region.mEnd; + } + else + { + flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start); + start = region.mStart; + end = region.mEnd; + } } - mVertexLocked = false; - sMappedCount--; + flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start); + + mMappedVertexRegions.clear(); } - if (mMappedIndexData && mIndexLocked) + if (!mMappedIndexRegions.empty()) { LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - index"); - bindGLIndices(); - if(!mMappable) - { - if (!mMappedIndexRegions.empty()) - { - for (U32 i = 0; i < mMappedIndexRegions.size(); ++i) - { - const MappedRegion& region = mMappedIndexRegions[i]; - S32 offset = region.mIndex >= 0 ? sizeof(U16)*region.mIndex : 0; - S32 length = sizeof(U16)*region.mCount; - if (mIndicesSize >= length + offset) - { - glBufferSubData(GL_ELEMENT_ARRAY_BUFFER, offset, length, (U8*) mMappedIndexData+offset); - } - else - { - GLint size = 0; - glGetBufferParameteriv(GL_ELEMENT_ARRAY_BUFFER, GL_BUFFER_SIZE, &size); - LL_WARNS() << "Attempted to map regions to a buffer that is too small, " - << "mapped size: " << mIndicesSize - << ", gl buffer size: " << size - << ", length: " << length - << ", offset: " << offset - << LL_ENDL; - } - stop_glerror(); - } - mMappedIndexRegions.clear(); - } - else - { - stop_glerror(); - glBufferSubData(GL_ELEMENT_ARRAY_BUFFER, 0, getIndicesSize(), (U8*) mMappedIndexData); - stop_glerror(); - } - } - else - { - if (gGLManager.mHasMapBufferRange || gGLManager.mHasFlushBufferRange) - { - if (!mMappedIndexRegions.empty()) - { - for (U32 i = 0; i < mMappedIndexRegions.size(); ++i) - { - LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - flush index"); - const MappedRegion& region = mMappedIndexRegions[i]; - S32 offset = region.mIndex >= 0 ? sizeof(U16)*region.mIndex : 0; - S32 length = sizeof(U16)*region.mCount; - if (gGLManager.mHasMapBufferRange) - { -#if GL_ARB_map_buffer_range || GL_EXT_map_buffer_range - glFlushMappedBufferRange(GL_ELEMENT_ARRAY_BUFFER, offset, length); -#endif - } - else if (gGLManager.mHasFlushBufferRange) - { -#if GL_APPLE_flush_buffer_range && LL_DARWIN -#ifndef LL_MESA_HEADLESS - glFlushMappedBufferRangeAPPLE(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length); -#endif -#endif - } - stop_glerror(); - } + if (mGLIndices != sGLRenderIndices) + { + glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices); + sGLRenderIndices = mGLIndices; + } + U32 start = 0; + U32 end = 0; - mMappedIndexRegions.clear(); - } - } - - glUnmapBuffer(GL_ELEMENT_ARRAY_BUFFER); + std::sort(mMappedIndexRegions.begin(), mMappedIndexRegions.end(), SortMappedRegion()); - mMappedIndexData = NULL; - } + for (U32 i = 0; i < mMappedIndexRegions.size(); ++i) + { + const MappedRegion& region = mMappedIndexRegions[i]; + if (region.mStart == end + 1) + { + end = region.mEnd; + } + else + { + flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start); + start = region.mStart; + end = region.mEnd; + } + } - mIndexLocked = false; - sMappedCount--; - } + flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start); - if(updated_all) - { - mEmpty = false; + mMappedIndexRegions.clear(); } } //---------------------------------------------------------------------------- -template <class T,S32 type> struct VertexBufferStrider +template <class T,LLVertexBuffer::AttributeType type> struct VertexBufferStrider { typedef LLStrider<T> strider_t; static bool get(LLVertexBuffer& vbo, strider_t& strider, - S32 index, S32 count, bool map_range) + S32 index, S32 count) { if (type == LLVertexBuffer::TYPE_INDEX) { - U8* ptr = vbo.mapIndexBuffer(index, count, map_range); + U8* ptr = vbo.mapIndexBuffer(index, count); if (ptr == NULL) { @@ -1975,9 +1269,9 @@ template <class T,S32 type> struct VertexBufferStrider } else if (vbo.hasDataType(type)) { - S32 stride = LLVertexBuffer::sTypeSize[type]; + U32 stride = LLVertexBuffer::sTypeSize[type]; - U8* ptr = vbo.mapVertexBuffer(type, index, count, map_range); + U8* ptr = vbo.mapVertexBuffer(type, index, count); if (ptr == NULL) { @@ -1997,543 +1291,159 @@ template <class T,S32 type> struct VertexBufferStrider } }; -bool LLVertexBuffer::getVertexStrider(LLStrider<LLVector3>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getVertexStrider(LLStrider<LLVector3>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector3,TYPE_VERTEX>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector3,TYPE_VERTEX>::get(*this, strider, index, count); } -bool LLVertexBuffer::getVertexStrider(LLStrider<LLVector4a>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getVertexStrider(LLStrider<LLVector4a>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector4a,TYPE_VERTEX>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector4a,TYPE_VERTEX>::get(*this, strider, index, count); } -bool LLVertexBuffer::getIndexStrider(LLStrider<U16>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getIndexStrider(LLStrider<U16>& strider, U32 index, S32 count) { - return VertexBufferStrider<U16,TYPE_INDEX>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<U16,TYPE_INDEX>::get(*this, strider, index, count); } -bool LLVertexBuffer::getTexCoord0Strider(LLStrider<LLVector2>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getTexCoord0Strider(LLStrider<LLVector2>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector2,TYPE_TEXCOORD0>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector2,TYPE_TEXCOORD0>::get(*this, strider, index, count); } -bool LLVertexBuffer::getTexCoord1Strider(LLStrider<LLVector2>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getTexCoord1Strider(LLStrider<LLVector2>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector2,TYPE_TEXCOORD1>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector2,TYPE_TEXCOORD1>::get(*this, strider, index, count); } -bool LLVertexBuffer::getTexCoord2Strider(LLStrider<LLVector2>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getTexCoord2Strider(LLStrider<LLVector2>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector2,TYPE_TEXCOORD2>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector2,TYPE_TEXCOORD2>::get(*this, strider, index, count); } -bool LLVertexBuffer::getNormalStrider(LLStrider<LLVector3>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getNormalStrider(LLStrider<LLVector3>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector3,TYPE_NORMAL>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector3,TYPE_NORMAL>::get(*this, strider, index, count); } -bool LLVertexBuffer::getTangentStrider(LLStrider<LLVector3>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getTangentStrider(LLStrider<LLVector3>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector3,TYPE_TANGENT>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector3,TYPE_TANGENT>::get(*this, strider, index, count); } -bool LLVertexBuffer::getTangentStrider(LLStrider<LLVector4a>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getTangentStrider(LLStrider<LLVector4a>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector4a,TYPE_TANGENT>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector4a,TYPE_TANGENT>::get(*this, strider, index, count); } -bool LLVertexBuffer::getColorStrider(LLStrider<LLColor4U>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getColorStrider(LLStrider<LLColor4U>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLColor4U,TYPE_COLOR>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLColor4U,TYPE_COLOR>::get(*this, strider, index, count); } -bool LLVertexBuffer::getEmissiveStrider(LLStrider<LLColor4U>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getEmissiveStrider(LLStrider<LLColor4U>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLColor4U,TYPE_EMISSIVE>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLColor4U,TYPE_EMISSIVE>::get(*this, strider, index, count); } -bool LLVertexBuffer::getWeightStrider(LLStrider<F32>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getWeightStrider(LLStrider<F32>& strider, U32 index, S32 count) { - return VertexBufferStrider<F32,TYPE_WEIGHT>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<F32,TYPE_WEIGHT>::get(*this, strider, index, count); } -bool LLVertexBuffer::getWeight4Strider(LLStrider<LLVector4>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getWeight4Strider(LLStrider<LLVector4>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector4,TYPE_WEIGHT4>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector4,TYPE_WEIGHT4>::get(*this, strider, index, count); } -bool LLVertexBuffer::getClothWeightStrider(LLStrider<LLVector4>& strider, S32 index, S32 count, bool map_range) +bool LLVertexBuffer::getClothWeightStrider(LLStrider<LLVector4>& strider, U32 index, S32 count) { - return VertexBufferStrider<LLVector4,TYPE_CLOTHWEIGHT>::get(*this, strider, index, count, map_range); + return VertexBufferStrider<LLVector4,TYPE_CLOTHWEIGHT>::get(*this, strider, index, count); } //---------------------------------------------------------------------------- -bool LLVertexBuffer::bindGLArray() -{ - if (mGLArray && sGLRenderArray != mGLArray) - { - { - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; -//#if GL_ARB_vertex_array_object - glBindVertexArray(mGLArray); -//#endif - sGLRenderArray = mGLArray; - } - - //really shouldn't be necessary, but some drivers don't properly restore the - //state of GL_ELEMENT_ARRAY_BUFFER_BINDING - bindGLIndices(); - - return true; - } - - return false; -} -bool LLVertexBuffer::bindGLBuffer(bool force_bind) +// Set for rendering +void LLVertexBuffer::setBuffer() { - bindGLArray(); + // no data may be pending + llassert(mMappedVertexRegions.empty()); + llassert(mMappedIndexRegions.empty()); - bool ret = false; + // a shader must be bound + llassert(LLGLSLShader::sCurBoundShaderPtr); - if (useVBOs() && (force_bind || (mGLBuffer && (mGLBuffer != sGLRenderBuffer || !sVBOActive)))) - { - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; - glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer); - sGLRenderBuffer = mGLBuffer; - sBindCount++; - sVBOActive = true; + U32 data_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; - llassert(!mGLArray || sGLRenderArray == mGLArray); - - ret = true; - } + // this Vertex Buffer must provide all necessary attributes for currently bound shader + llassert_msg((data_mask & mTypeMask) == data_mask, + "Attribute mask mismatch! mTypeMask should be a superset of data_mask. data_mask: 0x" + << std::hex << data_mask << " mTypeMask: 0x" << mTypeMask << " Missing: 0x" << (data_mask & ~mTypeMask) << std::dec); - return ret; -} - -bool LLVertexBuffer::bindGLBufferFast() -{ - if (mGLBuffer != sGLRenderBuffer || !sVBOActive) + if (sGLRenderBuffer != mGLBuffer) { glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer); sGLRenderBuffer = mGLBuffer; - sBindCount++; - sVBOActive = true; - return true; + setupVertexBuffer(); } - - return false; -} - -bool LLVertexBuffer::bindGLIndices(bool force_bind) -{ - LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX; - bindGLArray(); - - bool ret = false; - if (useVBOs() && (force_bind || (mGLIndices && (mGLIndices != sGLRenderIndices || !sIBOActive)))) - { - /*if (sMapped) - { - LL_ERRS() << "VBO bound while another VBO mapped!" << LL_ENDL; - }*/ - glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices); - sGLRenderIndices = mGLIndices; - stop_glerror(); - sBindCount++; - sIBOActive = true; - ret = true; - } - - return ret; -} - -bool LLVertexBuffer::bindGLIndicesFast() -{ - if (mGLIndices != sGLRenderIndices || !sIBOActive) + else if (sLastMask != data_mask) { - glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices); - sGLRenderIndices = mGLIndices; - sBindCount++; - sIBOActive = true; - - return true; + setupVertexBuffer(); + sLastMask = data_mask; } - - return false; -} - -void LLVertexBuffer::flush() -{ - if (useVBOs()) - { - unmapBuffer(); - } -} - -// 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) -{ - flush(); - - //set up pointers if the data mask is different ... - bool setup = (sLastMask != data_mask); - - if (gDebugGL && data_mask != 0) - { //make sure data requirements are fulfilled - LLGLSLShader* shader = LLGLSLShader::sCurBoundShaderPtr; - if (shader) - { - U32 required_mask = 0; - for (U32 i = 0; i < LLVertexBuffer::TYPE_TEXTURE_INDEX; ++i) - { - if (shader->getAttribLocation(i) > -1) - { - U32 required = 1 << i; - if ((data_mask & required) == 0) - { - LL_WARNS() << "Missing attribute: " << LLShaderMgr::instance()->mReservedAttribs[i] << LL_ENDL; - } - - required_mask |= required; - } - } - - if ((data_mask & required_mask) != required_mask) - { - - U32 unsatisfied_mask = (required_mask & ~data_mask); - - for (U32 i = 0; i < TYPE_MAX; i++) - { - U32 unsatisfied_flag = unsatisfied_mask & (1 << i); - switch (unsatisfied_flag) - { - case 0: break; - case MAP_VERTEX: LL_INFOS() << "Missing vert pos" << LL_ENDL; break; - case MAP_NORMAL: LL_INFOS() << "Missing normals" << LL_ENDL; break; - case MAP_TEXCOORD0: LL_INFOS() << "Missing TC 0" << LL_ENDL; break; - case MAP_TEXCOORD1: LL_INFOS() << "Missing TC 1" << LL_ENDL; break; - case MAP_TEXCOORD2: LL_INFOS() << "Missing TC 2" << LL_ENDL; break; - case MAP_TEXCOORD3: LL_INFOS() << "Missing TC 3" << LL_ENDL; break; - case MAP_COLOR: LL_INFOS() << "Missing vert color" << LL_ENDL; break; - case MAP_EMISSIVE: LL_INFOS() << "Missing emissive" << LL_ENDL; break; - case MAP_TANGENT: LL_INFOS() << "Missing tangent" << LL_ENDL; break; - case MAP_WEIGHT: LL_INFOS() << "Missing weight" << LL_ENDL; break; - case MAP_WEIGHT4: LL_INFOS() << "Missing weightx4" << LL_ENDL; break; - case MAP_CLOTHWEIGHT: LL_INFOS() << "Missing clothweight" << LL_ENDL; break; - case MAP_TEXTURE_INDEX: LL_INFOS() << "Missing tex index" << LL_ENDL; break; - default: LL_INFOS() << "Missing who effin knows: " << unsatisfied_flag << LL_ENDL; - } - } - - // TYPE_INDEX is beyond TYPE_MAX, so check for it individually - if (unsatisfied_mask & (1 << TYPE_INDEX)) - { - LL_INFOS() << "Missing indices" << LL_ENDL; - } - - LL_ERRS() << "Shader consumption mismatches data provision." << LL_ENDL; - } - } - } - - if (useVBOs()) - { - if (mGLArray) - { - bindGLArray(); - setup = false; //do NOT perform pointer setup if using VAO - } - else - { - const bool bindBuffer = bindGLBuffer(); - const bool bindIndices = bindGLIndices(); - - setup = setup || bindBuffer || bindIndices; - } - - if (gDebugGL && !mGLArray) - { - GLint buff; - glGetIntegerv(GL_ARRAY_BUFFER_BINDING, &buff); - if ((GLuint)buff != mGLBuffer) - { - if (gDebugSession) - { - gFailLog << "Invalid GL vertex buffer bound: " << buff << std::endl; - } - else - { - LL_ERRS() << "Invalid GL vertex buffer bound: " << buff << LL_ENDL; - } - } - - if (mGLIndices) - { - glGetIntegerv(GL_ELEMENT_ARRAY_BUFFER_BINDING, &buff); - if ((GLuint)buff != mGLIndices) - { - if (gDebugSession) - { - gFailLog << "Invalid GL index buffer bound: " << buff << std::endl; - } - else - { - LL_ERRS() << "Invalid GL index buffer bound: " << buff << LL_ENDL; - } - } - } - } - - - } - else - { - if (sGLRenderArray) - { -//#if GL_ARB_vertex_array_object - glBindVertexArray(0); -//#endif - sGLRenderArray = 0; - sGLRenderIndices = 0; - sIBOActive = false; - } - - if (mGLBuffer) - { - if (sVBOActive) - { - glBindBuffer(GL_ARRAY_BUFFER, 0); - sBindCount++; - sVBOActive = false; - setup = true; // ... or a VBO is deactivated - } - if (sGLRenderBuffer != mGLBuffer) - { - sGLRenderBuffer = mGLBuffer; - setup = true; // ... or a client memory pointer changed - } - } - if (mGLIndices) - { - if (sIBOActive) - { - glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0); - sBindCount++; - sIBOActive = false; - } - - sGLRenderIndices = mGLIndices; - } - } - - if (!mGLArray) - { - setupClientArrays(data_mask); - } - - if (mGLBuffer) - { - if (data_mask && setup) - { - setupVertexBuffer(data_mask); // subclass specific setup (virtual function) - sSetCount++; - } - } -} - -void LLVertexBuffer::setBufferFast(U32 data_mask) -{ - if (useVBOs()) - { - //set up pointers if the data mask is different ... - bool setup = (sLastMask != data_mask); - - const bool bindBuffer = bindGLBufferFast(); - const bool bindIndices = bindGLIndicesFast(); - - setup = setup || bindBuffer || bindIndices; - - setupClientArrays(data_mask); - - if (data_mask && setup) - { - setupVertexBufferFast(data_mask); - sSetCount++; - } - } - else + + if (mGLIndices != sGLRenderIndices) { - //fallback to slow path when not using VBOs - setBuffer(data_mask); + glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices); + sGLRenderIndices = mGLIndices; } } // virtual (default) -void LLVertexBuffer::setupVertexBuffer(U32 data_mask) +void LLVertexBuffer::setupVertexBuffer() { - stop_glerror(); - U8* base = useVBOs() ? (U8*) mAlignedOffset : mMappedData; + U8* base = nullptr; - if (gDebugGL && ((data_mask & mTypeMask) != data_mask)) - { - for (U32 i = 0; i < LLVertexBuffer::TYPE_MAX; ++i) - { - U32 mask = 1 << i; - if (mask & data_mask && !(mask & mTypeMask)) - { //bit set in data_mask, but not set in mTypeMask - LL_WARNS() << "Missing required component " << vb_type_name[i] << LL_ENDL; - } - } - LL_ERRS() << "LLVertexBuffer::setupVertexBuffer missing required components for supplied data mask." << LL_ENDL; - } - - if (data_mask & MAP_NORMAL) - { - S32 loc = TYPE_NORMAL; - void* ptr = (void*)(base + mOffsets[TYPE_NORMAL]); - glVertexAttribPointer(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_NORMAL], ptr); - } - if (data_mask & MAP_TEXCOORD3) - { - S32 loc = TYPE_TEXCOORD3; - void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD3]); - glVertexAttribPointer(loc,2,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD3], ptr); - } - if (data_mask & MAP_TEXCOORD2) - { - S32 loc = TYPE_TEXCOORD2; - void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD2]); - glVertexAttribPointer(loc,2,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD2], ptr); - } - if (data_mask & MAP_TEXCOORD1) - { - S32 loc = TYPE_TEXCOORD1; - void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD1]); - glVertexAttribPointer(loc,2,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD1], ptr); - } - if (data_mask & MAP_TANGENT) - { - S32 loc = TYPE_TANGENT; - void* ptr = (void*)(base + mOffsets[TYPE_TANGENT]); - glVertexAttribPointer(loc, 4,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TANGENT], ptr); - } - if (data_mask & MAP_TEXCOORD0) - { - S32 loc = TYPE_TEXCOORD0; - void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD0]); - glVertexAttribPointer(loc,2,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD0], ptr); - } - if (data_mask & MAP_COLOR) - { - S32 loc = TYPE_COLOR; - //bind emissive instead of color pointer if emissive is present - void* ptr = (data_mask & MAP_EMISSIVE) ? (void*)(base + mOffsets[TYPE_EMISSIVE]) : (void*)(base + mOffsets[TYPE_COLOR]); - glVertexAttribPointer(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_COLOR], ptr); - } - if (data_mask & MAP_EMISSIVE) - { - S32 loc = TYPE_EMISSIVE; - void* ptr = (void*)(base + mOffsets[TYPE_EMISSIVE]); - glVertexAttribPointer(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_EMISSIVE], ptr); - - if (!(data_mask & MAP_COLOR)) - { //map emissive to color channel when color is not also being bound to avoid unnecessary shader swaps - loc = TYPE_COLOR; - glVertexAttribPointer(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_EMISSIVE], ptr); - } - } - if (data_mask & MAP_WEIGHT) - { - S32 loc = TYPE_WEIGHT; - void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT]); - glVertexAttribPointer(loc, 1, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT], ptr); - } - if (data_mask & MAP_WEIGHT4) - { - S32 loc = TYPE_WEIGHT4; - void* ptr = (void*)(base+mOffsets[TYPE_WEIGHT4]); - glVertexAttribPointer(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT4], ptr); - } - if (data_mask & MAP_CLOTHWEIGHT) - { - S32 loc = TYPE_CLOTHWEIGHT; - void* ptr = (void*)(base + mOffsets[TYPE_CLOTHWEIGHT]); - glVertexAttribPointer(loc, 4, GL_FLOAT, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_CLOTHWEIGHT], ptr); - } - if (data_mask & MAP_TEXTURE_INDEX && - (gGLManager.mGLSLVersionMajor >= 2 || gGLManager.mGLSLVersionMinor >= 30)) //indexed texture rendering requires GLSL 1.30 or later - { -#if !LL_DARWIN - S32 loc = TYPE_TEXTURE_INDEX; - void *ptr = (void*) (base + mOffsets[TYPE_VERTEX] + 12); - glVertexAttribIPointer(loc, 1, GL_UNSIGNED_INT, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); -#endif - } - if (data_mask & MAP_VERTEX) - { - S32 loc = TYPE_VERTEX; - void* ptr = (void*)(base + mOffsets[TYPE_VERTEX]); - glVertexAttribPointer(loc, 3,GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); - } - - llglassertok(); -} - -void LLVertexBuffer::setupVertexBufferFast(U32 data_mask) -{ - U8* base = (U8*)mAlignedOffset; + U32 data_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask; if (data_mask & MAP_NORMAL) { - S32 loc = TYPE_NORMAL; + AttributeType loc = TYPE_NORMAL; void* ptr = (void*)(base + mOffsets[TYPE_NORMAL]); glVertexAttribPointer(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_NORMAL], ptr); } if (data_mask & MAP_TEXCOORD3) { - S32 loc = TYPE_TEXCOORD3; + AttributeType loc = TYPE_TEXCOORD3; void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD3]); glVertexAttribPointer(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD3], ptr); } if (data_mask & MAP_TEXCOORD2) { - S32 loc = TYPE_TEXCOORD2; + AttributeType loc = TYPE_TEXCOORD2; void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD2]); glVertexAttribPointer(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD2], ptr); } if (data_mask & MAP_TEXCOORD1) { - S32 loc = TYPE_TEXCOORD1; + AttributeType loc = TYPE_TEXCOORD1; void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD1]); glVertexAttribPointer(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD1], ptr); } if (data_mask & MAP_TANGENT) { - S32 loc = TYPE_TANGENT; + AttributeType loc = TYPE_TANGENT; void* ptr = (void*)(base + mOffsets[TYPE_TANGENT]); glVertexAttribPointer(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TANGENT], ptr); } if (data_mask & MAP_TEXCOORD0) { - S32 loc = TYPE_TEXCOORD0; + AttributeType loc = TYPE_TEXCOORD0; void* ptr = (void*)(base + mOffsets[TYPE_TEXCOORD0]); glVertexAttribPointer(loc, 2, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_TEXCOORD0], ptr); } if (data_mask & MAP_COLOR) { - S32 loc = TYPE_COLOR; + AttributeType loc = TYPE_COLOR; //bind emissive instead of color pointer if emissive is present void* ptr = (data_mask & MAP_EMISSIVE) ? (void*)(base + mOffsets[TYPE_EMISSIVE]) : (void*)(base + mOffsets[TYPE_COLOR]); glVertexAttribPointer(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_COLOR], ptr); } if (data_mask & MAP_EMISSIVE) { - S32 loc = TYPE_EMISSIVE; + AttributeType loc = TYPE_EMISSIVE; void* ptr = (void*)(base + mOffsets[TYPE_EMISSIVE]); glVertexAttribPointer(loc, 4, GL_UNSIGNED_BYTE, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_EMISSIVE], ptr); @@ -2545,42 +1455,52 @@ void LLVertexBuffer::setupVertexBufferFast(U32 data_mask) } if (data_mask & MAP_WEIGHT) { - S32 loc = TYPE_WEIGHT; + AttributeType loc = TYPE_WEIGHT; void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT]); glVertexAttribPointer(loc, 1, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT], ptr); } if (data_mask & MAP_WEIGHT4) { - S32 loc = TYPE_WEIGHT4; + AttributeType loc = TYPE_WEIGHT4; void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT4]); glVertexAttribPointer(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT4], ptr); } if (data_mask & MAP_CLOTHWEIGHT) { - S32 loc = TYPE_CLOTHWEIGHT; + AttributeType loc = TYPE_CLOTHWEIGHT; void* ptr = (void*)(base + mOffsets[TYPE_CLOTHWEIGHT]); glVertexAttribPointer(loc, 4, GL_FLOAT, GL_TRUE, LLVertexBuffer::sTypeSize[TYPE_CLOTHWEIGHT], ptr); } if (data_mask & MAP_TEXTURE_INDEX) { -#if !LL_DARWIN - S32 loc = TYPE_TEXTURE_INDEX; + AttributeType loc = TYPE_TEXTURE_INDEX; void* ptr = (void*)(base + mOffsets[TYPE_VERTEX] + 12); glVertexAttribIPointer(loc, 1, GL_UNSIGNED_INT, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); -#endif } if (data_mask & MAP_VERTEX) { - S32 loc = TYPE_VERTEX; + AttributeType loc = TYPE_VERTEX; void* ptr = (void*)(base + mOffsets[TYPE_VERTEX]); glVertexAttribPointer(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr); } } -LLVertexBuffer::MappedRegion::MappedRegion(S32 type, S32 index, S32 count) -: mType(type), mIndex(index), mCount(count) -{ - mEnd = mIndex+mCount; -} +void LLVertexBuffer::setPositionData(const LLVector4a* data) +{ + llassert(sGLRenderBuffer == mGLBuffer); + flush_vbo(GL_ARRAY_BUFFER, 0, sizeof(LLVector4a) * getNumVerts()-1, (U8*) data); +} + +void LLVertexBuffer::setTexCoordData(const LLVector2* data) +{ + llassert(sGLRenderBuffer == mGLBuffer); + flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD0], mOffsets[TYPE_TEXCOORD0] + sTypeSize[TYPE_TEXCOORD0] * getNumVerts() - 1, (U8*)data); +} + +void LLVertexBuffer::setColorData(const LLColor4U* data) +{ + llassert(sGLRenderBuffer == mGLBuffer); + flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_COLOR], mOffsets[TYPE_COLOR] + sTypeSize[TYPE_COLOR] * getNumVerts() - 1, (U8*) data); +} |