Merge pull request #2856 from secondlife/marchcat/c-develop

Develop → Maint C sync

1 files changed, 508 insertions, 138 deletions

diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 7caf20f40b..12ae36f4bb 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -36,6 +36,7 @@
 #include "llshadermgr.h"
 #include "llglslshader.h"
 #include "llmemory.h"
+#include <glm/gtc/type_ptr.hpp>
 
 //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
@@ -288,13 +289,103 @@ static GLuint gen_buffer()
     return ret;
 }
 
+static void delete_buffers(S32 count, GLuint* buffers)
+{
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
+    // wait a few frames before actually deleting the buffers to avoid
+    // synchronization issues with the GPU
+    static std::vector<GLuint> sFreeList[4];
+
+    if (gGLManager.mInited)
+    {
+        U32 idx = LLImageGL::sFrameCount % 4;
+
+        for (S32 i = 0; i < count; ++i)
+        {
+            sFreeList[idx].push_back(buffers[i]);
+        }
+
+        idx = (LLImageGL::sFrameCount + 3) % 4;
+
+        if (!sFreeList[idx].empty())
+        {
+            glDeleteBuffers((GLsizei)sFreeList[idx].size(), sFreeList[idx].data());
+            sFreeList[idx].resize(0);
+        }
+    }
+}
+
+
 #define ANALYZE_VBO_POOL 0
 
+// VBO Pool interface
 class LLVBOPool
 {
+    public:
+    virtual ~LLVBOPool() = default;
+    virtual void allocate(GLenum type, U32 size, GLuint& name, U8*& data) = 0;
+    virtual void free(GLenum type, U32 size, GLuint name, U8* data) = 0;
+    virtual U64 getVramBytesUsed() = 0;
+};
+
+// VBO Pool for Apple GPUs (as in M1/M2 etc, not Intel macs)
+// Effectively disables VBO pooling
+class LLAppleVBOPool final: public LLVBOPool
+{
 public:
-    typedef std::chrono::steady_clock::time_point Time;
+    U64 mAllocated = 0;
+
+    U64 getVramBytesUsed() override
+    {
+        return mAllocated;
+    }
+
+    void allocate(GLenum type, U32 size, GLuint& name, U8*& data) override
+    {
+        LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
+        STOP_GLERROR;
+        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
+
+        mAllocated += size;
 
+        { //allocate a new buffer
+            LL_PROFILE_GPU_ZONE("vbo alloc");
+            // ON OS X, we don't allocate a VBO until the last possible moment
+            // in unmapBuffer
+            data = (U8*) ll_aligned_malloc_16(size);
+            STOP_GLERROR;
+        }
+    }
+
+    void free(GLenum type, U32 size, GLuint name, U8* data) override
+    {
+        LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
+        llassert(type == GL_ARRAY_BUFFER || type == GL_ELEMENT_ARRAY_BUFFER);
+        llassert(size >= 2);
+
+        if (data)
+        {
+            ll_aligned_free_16(data);
+        }
+
+        mAllocated -= size;
+        STOP_GLERROR;
+        if (name)
+        {
+            delete_buffers(1, &name);
+        }
+        STOP_GLERROR;
+    }
+};
+
+// VBO Pool for GPUs that benefit from VBO pooling
+class LLDefaultVBOPool final : public LLVBOPool
+{
+public:
+    typedef std::chrono::steady_clock::time_point Time;
     struct Entry
     {
         U8* mData;
@@ -302,7 +393,7 @@ public:
         Time mAge;
     };
 
-    ~LLVBOPool()
+    ~LLDefaultVBOPool() override
     {
         clear();
     }
@@ -320,7 +411,7 @@ public:
     U32 mMisses = 0;
     U32 mHits = 0;
 
-    U64 getVramBytesUsed()
+    U64 getVramBytesUsed() override
     {
         return mAllocated + mReserved;
     }
@@ -336,7 +427,7 @@ public:
         size += block_size - (size % block_size);
     }
 
-    void allocate(GLenum type, U32 size, GLuint& name, U8*& data)
+    void allocate(GLenum type, U32 size, GLuint& name, U8*& data) override
     {
         LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
         llassert(type == GL_ARRAY_BUFFER || type == GL_ELEMENT_ARRAY_BUFFER);
@@ -392,7 +483,7 @@ public:
         clean();
     }
 
-    void free(GLenum type, U32 size, GLuint name, U8* data)
+    void free(GLenum type, U32 size, GLuint name, U8* data) override
     {
         LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
         llassert(type == GL_ARRAY_BUFFER || type == GL_ELEMENT_ARRAY_BUFFER);
@@ -455,7 +546,7 @@ public:
                     LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vbo cache timeout");
                     auto& entry = entries.back();
                     ll_aligned_free_16(entry.mData);
-                    glDeleteBuffers(1, &entry.mGLName);
+                    delete_buffers(1, &entry.mGLName);
                     llassert(mReserved >= iter->first);
                     mReserved -= iter->first;
                     entries.pop_back();
@@ -491,7 +582,7 @@ public:
             for (auto& entry : entries.second)
             {
                 ll_aligned_free_16(entry.mData);
-                glDeleteBuffers(1, &entry.mGLName);
+                delete_buffers(1, &entry.mGLName);
             }
         }
 
@@ -500,7 +591,7 @@ public:
             for (auto& entry : entries.second)
             {
                 ll_aligned_free_16(entry.mData);
-                glDeleteBuffers(1, &entry.mGLName);
+                delete_buffers(1, &entry.mGLName);
             }
         }
 
@@ -509,12 +600,72 @@ public:
         mIBOPool.clear();
         mVBOPool.clear();
     }
-
-
 };
 
 static LLVBOPool* sVBOPool = nullptr;
 
+void LLVertexBufferData::drawWithMatrix()
+{
+    if (!mVB)
+    {
+        llassert(false);
+        // Not supposed to happen, check buffer generation
+        return;
+    }
+
+    if (mTexName)
+    {
+        gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, mTexName);
+    }
+    else
+    {
+        gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE);
+    }
+
+    gGL.matrixMode(LLRender::MM_MODELVIEW);
+    gGL.pushMatrix();
+    gGL.loadMatrix(glm::value_ptr(mModelView));
+    gGL.matrixMode(LLRender::MM_PROJECTION);
+    gGL.pushMatrix();
+    gGL.loadMatrix(glm::value_ptr(mProjection));
+    gGL.matrixMode(LLRender::MM_TEXTURE0);
+    gGL.pushMatrix();
+    gGL.loadMatrix(glm::value_ptr(mTexture0));
+
+    mVB->setBuffer();
+    mVB->drawArrays(mMode, 0, mCount);
+
+    gGL.popMatrix();
+    gGL.matrixMode(LLRender::MM_PROJECTION);
+    gGL.popMatrix();
+    gGL.matrixMode(LLRender::MM_MODELVIEW);
+    gGL.popMatrix();
+}
+
+void LLVertexBufferData::draw()
+{
+    if (!mVB)
+    {
+        llassert(false);
+        // Not supposed to happen, check buffer generation
+        return;
+    }
+
+    if (mTexName)
+    {
+        gGL.getTexUnit(0)->bindManual(LLTexUnit::TT_TEXTURE, mTexName);
+    }
+    else
+    {
+        gGL.getTexUnit(0)->unbind(LLTexUnit::TT_TEXTURE);
+    }
+
+    mVB->setBuffer();
+    mVB->drawArrays(mMode, 0, mCount);
+}
+
+//============================================================================
+
 //static
 U64 LLVertexBuffer::getBytesAllocated()
 {
@@ -545,6 +696,7 @@ const U32 LLVertexBuffer::sTypeSize[LLVertexBuffer::TYPE_MAX] =
     sizeof(F32),       // TYPE_WEIGHT,
     sizeof(LLVector4), // TYPE_WEIGHT4,
     sizeof(LLVector4), // TYPE_CLOTHWEIGHT,
+    sizeof(U64),       // TYPE_JOINT,
     sizeof(LLVector4), // TYPE_TEXTURE_INDEX (actually exists as position.w), no extra data, but stride is 16 bytes
 };
 
@@ -562,6 +714,7 @@ static const std::string vb_type_name[] =
     "TYPE_WEIGHT",
     "TYPE_WEIGHT4",
     "TYPE_CLOTHWEIGHT",
+    "TYPE_JOINT"
     "TYPE_TEXTURE_INDEX",
     "TYPE_MAX",
     "TYPE_INDEX",
@@ -575,7 +728,6 @@ const U32 LLVertexBuffer::sGLMode[LLRender::NUM_MODES] =
     GL_POINTS,
     GL_LINES,
     GL_LINE_STRIP,
-    GL_QUADS,
     GL_LINE_LOOP,
 };
 
@@ -629,6 +781,8 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto
     LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
     llassert(LLGLSLShader::sCurBoundShaderPtr != NULL);
 
+    STOP_GLERROR;
+
     gGL.syncMatrices();
 
     U32 mask = LLVertexBuffer::MAP_VERTEX;
@@ -643,7 +797,7 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto
 
     if (tc != nullptr)
     {
-        for (int i = 0; i < num_indices; ++i)
+        for (U32 i = 0; i < num_indices; ++i)
         {
             U16 idx = indicesp[i];
             gGL.texCoord2fv(tc[idx].mV);
@@ -652,7 +806,7 @@ void LLVertexBuffer::drawElements(U32 mode, const LLVector4a* pos, const LLVecto
     }
     else
     {
-        for (int i = 0; i < num_indices; ++i)
+        for (U32 i = 0; i < num_indices; ++i)
         {
             U16 idx = indicesp[i];
             gGL.vertex3fv(pos[idx].getF32ptr());
@@ -669,24 +823,23 @@ bool LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_of
         return true;
     }
 
-    llassert(start < (U32)mNumVerts);
-    llassert(end < (U32)mNumVerts);
+    llassert(start < mNumVerts);
+    llassert(end < mNumVerts);
 
-    if (start >= (U32) mNumVerts ||
-        end >= (U32) mNumVerts)
+    if (start >= mNumVerts ||
+        end >= mNumVerts)
     {
         LL_ERRS() << "Bad vertex buffer draw range: [" << start << ", " << end << "] vs " << mNumVerts << LL_ENDL;
     }
 
-    llassert(mNumIndices >= 0);
-
-    if (indices_offset >= (U32) mNumIndices ||
-        indices_offset + count > (U32) mNumIndices)
+    if (indices_offset >= mNumIndices ||
+        indices_offset + count > mNumIndices)
     {
         LL_ERRS() << "Bad index buffer draw range: [" << indices_offset << ", " << indices_offset+count << "]" << LL_ENDL;
     }
 
     {
+#if 0  // not a reliable test for VBOs that are not backed by a CPU buffer
         U16* idx = (U16*) mMappedIndexData+indices_offset;
         for (U32 i = 0; i < count; ++i)
         {
@@ -718,12 +871,13 @@ bool LLVertexBuffer::validateRange(U32 start, U32 end, U32 count, U32 indices_of
             for (U32 i = start; i < end; i++)
             {
                 U32 idx = (U32) (v[i][3]+0.25f);
-                if (idx >= shader->mFeatures.mIndexedTextureChannels)
+                if (idx >= (U32)shader->mFeatures.mIndexedTextureChannels)
                 {
                     LL_ERRS() << "Bad texture index found in vertex data stream." << LL_ENDL;
                 }
             }
         }
+#endif
     }
 
     return true;
@@ -735,16 +889,37 @@ void LLVertexBuffer::setLabel(const char* label) {
 }
 #endif
 
+void LLVertexBuffer::clone(LLVertexBuffer& target) const
+{
+    target.mTypeMask = mTypeMask;
+    target.mIndicesType = mIndicesType;
+    target.mIndicesStride = mIndicesStride;
+    if (target.getNumVerts() != getNumVerts() ||
+        target.getNumIndices() != getNumIndices())
+    {
+        target.allocateBuffer(getNumVerts(), getNumIndices());
+    }
+}
+
 void LLVertexBuffer::drawRange(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const
 {
     llassert(validateRange(start, end, count, indices_offset));
     llassert(mGLBuffer == sGLRenderBuffer);
     llassert(mGLIndices == sGLRenderIndices);
     gGL.syncMatrices();
+    STOP_GLERROR;
     glDrawRangeElements(sGLMode[mode], start, end, count, mIndicesType,
         (GLvoid*) (indices_offset * (size_t) mIndicesStride));
+    STOP_GLERROR;
+}
+
+void LLVertexBuffer::drawRangeFast(U32 mode, U32 start, U32 end, U32 count, U32 indices_offset) const
+{
+    glDrawRangeElements(sGLMode[mode], start, end, count, mIndicesType,
+        (GLvoid*)(indices_offset * (size_t)mIndicesStride));
 }
 
+
 void LLVertexBuffer::draw(U32 mode, U32 count, U32 indices_offset) const
 {
     drawRange(mode, 0, mNumVerts-1, count, indices_offset);
@@ -758,14 +933,25 @@ void LLVertexBuffer::drawArrays(U32 mode, U32 first, U32 count) const
     llassert(mGLIndices == sGLRenderIndices);
 
     gGL.syncMatrices();
+    STOP_GLERROR;
     glDrawArrays(sGLMode[mode], first, count);
+    STOP_GLERROR;
 }
 
 //static
 void LLVertexBuffer::initClass(LLWindow* window)
 {
     llassert(sVBOPool == nullptr);
-    sVBOPool = new LLVBOPool();
+    if (gGLManager.mIsApple)
+    {
+        LL_INFOS() << "VBO Pooling Disabled" << LL_ENDL;
+        sVBOPool = new LLAppleVBOPool();
+    }
+    else
+    {
+        LL_INFOS() << "VBO Pooling Enabled" << LL_ENDL;
+        sVBOPool = new LLDefaultVBOPool();
+    }
 
 #if ENABLE_GL_WORK_QUEUE
     sQueue = new GLWorkQueue();
@@ -781,9 +967,10 @@ void LLVertexBuffer::initClass(LLWindow* window)
 //static
 void LLVertexBuffer::unbind()
 {
+    STOP_GLERROR;
     glBindBuffer(GL_ARRAY_BUFFER, 0);
     glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, 0);
-
+    STOP_GLERROR;
     sGLRenderBuffer = 0;
     sGLRenderIndices = 0;
 }
@@ -823,6 +1010,24 @@ LLVertexBuffer::LLVertexBuffer(U32 typemask)
     }
 }
 
+// list of mapped buffers
+// NOTE: must not be LLPointer<LLVertexBuffer> to avoid breaking non-ref-counted LLVertexBuffer instances
+static std::vector<LLVertexBuffer*> sMappedBuffers;
+
+//static
+void LLVertexBuffer::flushBuffers()
+{
+    LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
+    // must only be called from main thread
+    for (auto& buffer : sMappedBuffers)
+    {
+        buffer->_unmapBuffer();
+        buffer->mMapped = false;
+    }
+
+    sMappedBuffers.resize(0);
+}
+
 //static
 U32 LLVertexBuffer::calcOffsets(const U32& typemask, U32* offsets, U32 num_vertices)
 {
@@ -866,6 +1071,12 @@ U32 LLVertexBuffer::calcVertexSize(const U32& typemask)
 //virtual
 LLVertexBuffer::~LLVertexBuffer()
 {
+    if (mMapped)
+    { // is on the mapped buffer list but doesn't need to be flushed
+        mMapped = false;
+        unmapBuffer();
+    }
+
     destroyGLBuffer();
     destroyGLIndices();
 
@@ -998,12 +1209,6 @@ bool LLVertexBuffer::updateNumVerts(U32 nverts)
 
     bool success = true;
 
-    if (nverts > 65536)
-    {
-        LL_WARNS() << "Vertex buffer overflow!" << LL_ENDL;
-        nverts = 65536;
-    }
-
     U32 needed_size = calcOffsets(mTypeMask, mOffsets, nverts);
 
     if (needed_size != mSize)
@@ -1036,8 +1241,7 @@ bool LLVertexBuffer::updateNumIndices(U32 nindices)
 
 bool LLVertexBuffer::allocateBuffer(U32 nverts, U32 nindices)
 {
-    if (nverts < 0 || nindices < 0 ||
-        nverts > 65536)
+    if (nverts < 0 || nindices < 0)
     {
         LL_ERRS() << "Bad vertex buffer allocation: " << nverts << " : " << nindices << LL_ENDL;
     }
@@ -1074,33 +1278,36 @@ bool expand_region(LLVertexBuffer::MappedRegion& region, U32 start, U32 end)
 U8* LLVertexBuffer::mapVertexBuffer(LLVertexBuffer::AttributeType type, U32 index, S32 count)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
+    _mapBuffer();
 
     if (count == -1)
     {
         count = mNumVerts - index;
     }
 
-    U32 start = mOffsets[type] + sTypeSize[type] * index;
-    U32 end = start + sTypeSize[type] * count-1;
-
-    bool flagged = false;
-    // flag region as mapped
-    for (U32 i = 0; i < mMappedVertexRegions.size(); ++i)
+    if (!gGLManager.mIsApple)
     {
-        MappedRegion& region = mMappedVertexRegions[i];
-        if (expand_region(region, start, end))
+        U32 start = mOffsets[type] + sTypeSize[type] * index;
+        U32 end = start + sTypeSize[type] * count-1;
+
+        bool flagged = false;
+        // flag region as mapped
+        for (U32 i = 0; i < mMappedVertexRegions.size(); ++i)
         {
-            flagged = true;
-            break;
+            MappedRegion& region = mMappedVertexRegions[i];
+            if (expand_region(region, start, end))
+            {
+                flagged = true;
+                break;
+            }
         }
-    }
 
-    if (!flagged)
-    {
-        //didn't expand an existing region, make a new one
-        mMappedVertexRegions.push_back({ start, end });
+        if (!flagged)
+        {
+            //didn't expand an existing region, make a new one
+            mMappedVertexRegions.push_back({ start, end });
+        }
     }
-
     return mMappedData+mOffsets[type]+sTypeSize[type]*index;
 }
 
@@ -1108,31 +1315,35 @@ U8* LLVertexBuffer::mapVertexBuffer(LLVertexBuffer::AttributeType type, U32 inde
 U8* LLVertexBuffer::mapIndexBuffer(U32 index, S32 count)
 {
     LL_PROFILE_ZONE_SCOPED_CATEGORY_VERTEX;
+    _mapBuffer();
 
     if (count == -1)
     {
         count = mNumIndices-index;
     }
 
-    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)
+    if (!gGLManager.mIsApple)
     {
-        MappedRegion& region = mMappedIndexRegions[i];
-        if (expand_region(region, start, end))
+        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)
         {
-            flagged = true;
-            break;
+            MappedRegion& region = mMappedIndexRegions[i];
+            if (expand_region(region, start, end))
+            {
+                flagged = true;
+                break;
+            }
         }
-    }
 
-    if (!flagged)
-    {
-        //didn't expand an existing region, make a new one
-        mMappedIndexRegions.push_back({ start, end });
+        if (!flagged)
+        {
+            //didn't expand an existing region, make a new one
+            mMappedIndexRegions.push_back({ start, end });
+        }
     }
 
     return mMappedIndexData + sizeof(U16)*index;
@@ -1142,31 +1353,68 @@ U8* LLVertexBuffer::mapIndexBuffer(U32 index, S32 count)
 //  target -- "target" 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)
+//  data -- data to be flushed
+//  dst -- mMappedData or mMappedIndexData
+void LLVertexBuffer::flush_vbo(GLenum target, U32 start, U32 end, void* data, U8* dst)
 {
-    if (end != 0)
+    if (gGLManager.mIsApple)
     {
-        LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("glBufferSubData");
-        LL_PROFILE_ZONE_NUM(start);
-        LL_PROFILE_ZONE_NUM(end);
-        LL_PROFILE_ZONE_NUM(end-start);
-
-        constexpr U32 block_size = 8192;
+        // on OS X, flush_vbo doesn't actually write to the GL buffer, so be sure to call
+        // _mapBuffer to tag the buffer for flushing to GL
+        _mapBuffer();
+        LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("vb memcpy");
+        STOP_GLERROR;
+        // copy into mapped buffer
+        memcpy(dst+start, data, end-start+1);
+    }
+    else
+    {
+        llassert(target == GL_ARRAY_BUFFER ? sGLRenderBuffer == mGLBuffer : sGLRenderIndices == mGLIndices);
 
-        for (U32 i = start; i <= end; i += block_size)
+        // skip mapped data and stream to GPU via glBufferSubData
+        if (end != 0)
         {
-            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));
+            LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("glBufferSubData");
+            LL_PROFILE_ZONE_NUM(start);
+            LL_PROFILE_ZONE_NUM(end);
+            LL_PROFILE_ZONE_NUM(end-start);
+
+            constexpr U32 block_size = 65536;
+
+            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()
 {
+    flushBuffers();
+}
+
+void LLVertexBuffer::_mapBuffer()
+{
+    if (!mMapped)
+    {
+        mMapped = true;
+        sMappedBuffers.push_back(this);
+    }
+}
+
+void LLVertexBuffer::_unmapBuffer()
+{
+    STOP_GLERROR;
+    if (!mMapped)
+    {
+        return;
+    }
+
     struct SortMappedRegion
     {
         bool operator()(const MappedRegion& lhs, const MappedRegion& rhs)
@@ -1175,72 +1423,115 @@ void LLVertexBuffer::unmapBuffer()
         }
     };
 
-    if (!mMappedVertexRegions.empty())
+    if (gGLManager.mIsApple)
     {
-        LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - vertex");
-        if (sGLRenderBuffer != mGLBuffer)
+        STOP_GLERROR;
+        if (mMappedData)
         {
+            if (mGLBuffer)
+            {
+                delete_buffers(1, &mGLBuffer);
+            }
+            mGLBuffer = gen_buffer();
             glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer);
             sGLRenderBuffer = mGLBuffer;
+            glBufferData(GL_ARRAY_BUFFER, mSize, mMappedData, GL_STATIC_DRAW);
         }
+        else if (mGLBuffer != sGLRenderBuffer)
+        {
+            glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer);
+            sGLRenderBuffer = mGLBuffer;
+        }
+        STOP_GLERROR;
 
-        U32 start = 0;
-        U32 end = 0;
-
-        std::sort(mMappedVertexRegions.begin(), mMappedVertexRegions.end(), SortMappedRegion());
-
-        for (U32 i = 0; i < mMappedVertexRegions.size(); ++i)
+        if (mMappedIndexData)
         {
-            const MappedRegion& region = mMappedVertexRegions[i];
-            if (region.mStart == end + 1)
-            {
-                end = region.mEnd;
-            }
-            else
+            if (mGLIndices)
             {
-                flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start);
-                start = region.mStart;
-                end = region.mEnd;
+                delete_buffers(1, &mGLIndices);
             }
-        }
-
-        flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start);
-
-        mMappedVertexRegions.clear();
-    }
 
-    if (!mMappedIndexRegions.empty())
-    {
-        LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - index");
+            mGLIndices = gen_buffer();
+            glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices);
+            sGLRenderIndices = mGLIndices;
 
-        if (mGLIndices != sGLRenderIndices)
+            glBufferData(GL_ELEMENT_ARRAY_BUFFER, mIndicesSize, mMappedIndexData, GL_STATIC_DRAW);
+        }
+        else if (mGLIndices != sGLRenderIndices)
         {
             glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices);
             sGLRenderIndices = mGLIndices;
         }
-        U32 start = 0;
-        U32 end = 0;
-
-        std::sort(mMappedIndexRegions.begin(), mMappedIndexRegions.end(), SortMappedRegion());
-
-        for (U32 i = 0; i < mMappedIndexRegions.size(); ++i)
+        STOP_GLERROR;
+    }
+    else
+    {
+        if (!mMappedVertexRegions.empty())
         {
-            const MappedRegion& region = mMappedIndexRegions[i];
-            if (region.mStart == end + 1)
+            LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - vertex");
+
+            if (sGLRenderBuffer != mGLBuffer)
             {
-                end = region.mEnd;
+                glBindBuffer(GL_ARRAY_BUFFER, mGLBuffer);
+                sGLRenderBuffer = mGLBuffer;
             }
-            else
+
+            U32 start = 0;
+            U32 end = 0;
+
+            std::sort(mMappedVertexRegions.begin(), mMappedVertexRegions.end(), SortMappedRegion());
+
+            for (U32 i = 0; i < mMappedVertexRegions.size(); ++i)
             {
-                flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start);
-                start = region.mStart;
-                end = region.mEnd;
+                const MappedRegion& region = mMappedVertexRegions[i];
+                if (region.mStart == end + 1)
+                {
+                    end = region.mEnd;
+                }
+                else
+                {
+                    flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start, mMappedData);
+                    start = region.mStart;
+                    end = region.mEnd;
+                }
             }
+
+            flush_vbo(GL_ARRAY_BUFFER, start, end, (U8*)mMappedData + start, mMappedData);
+            mMappedVertexRegions.clear();
         }
 
-        flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start);
+        if (!mMappedIndexRegions.empty())
+        {
+            LL_PROFILE_ZONE_NAMED_CATEGORY_VERTEX("unmapBuffer - index");
+
+            if (mGLIndices != sGLRenderIndices)
+            {
+                glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices);
+                sGLRenderIndices = mGLIndices;
+            }
+            U32 start = 0;
+            U32 end = 0;
+
+            std::sort(mMappedIndexRegions.begin(), mMappedIndexRegions.end(), SortMappedRegion());
+
+            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, mMappedIndexData);
+                    start = region.mStart;
+                    end = region.mEnd;
+                }
+            }
 
-        mMappedIndexRegions.clear();
+            flush_vbo(GL_ELEMENT_ARRAY_BUFFER, start, end, (U8*)mMappedIndexData + start, mMappedIndexData);
+            mMappedIndexRegions.clear();
+        }
     }
 }
 
@@ -1362,6 +1653,14 @@ bool LLVertexBuffer::getClothWeightStrider(LLStrider<LLVector4>& strider, U32 in
 // Set for rendering
 void LLVertexBuffer::setBuffer()
 {
+    STOP_GLERROR;
+
+    if (mMapped)
+    {
+        LL_WARNS_ONCE() << "Missing call to unmapBuffer or flushBuffers" << LL_ENDL;
+        _unmapBuffer();
+    }
+
     // no data may be pending
     llassert(mMappedVertexRegions.empty());
     llassert(mMappedIndexRegions.empty());
@@ -1394,12 +1693,15 @@ void LLVertexBuffer::setBuffer()
         glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, mGLIndices);
         sGLRenderIndices = mGLIndices;
     }
+
+    STOP_GLERROR;
 }
 
 
 // virtual (default)
 void LLVertexBuffer::setupVertexBuffer()
 {
+    STOP_GLERROR;
     U8* base = nullptr;
 
     U32 data_mask = LLGLSLShader::sCurBoundShaderPtr->mAttributeMask;
@@ -1471,6 +1773,12 @@ void LLVertexBuffer::setupVertexBuffer()
         void* ptr = (void*)(base + mOffsets[TYPE_WEIGHT4]);
         glVertexAttribPointer(loc, 4, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_WEIGHT4], ptr);
     }
+    if (data_mask & MAP_JOINT)
+    {
+        AttributeType loc = TYPE_JOINT;
+        void* ptr = (void*)(base + mOffsets[TYPE_JOINT]);
+        glVertexAttribIPointer(loc, 4, GL_UNSIGNED_SHORT, LLVertexBuffer::sTypeSize[TYPE_JOINT], ptr);
+    }
     if (data_mask & MAP_CLOTHWEIGHT)
     {
         AttributeType loc = TYPE_CLOTHWEIGHT;
@@ -1489,59 +1797,121 @@ void LLVertexBuffer::setupVertexBuffer()
         void* ptr = (void*)(base + mOffsets[TYPE_VERTEX]);
         glVertexAttribPointer(loc, 3, GL_FLOAT, GL_FALSE, LLVertexBuffer::sTypeSize[TYPE_VERTEX], ptr);
     }
+    STOP_GLERROR;
 }
 
 void LLVertexBuffer::setPositionData(const LLVector4a* data)
 {
-    llassert(sGLRenderBuffer == mGLBuffer);
-    flush_vbo(GL_ARRAY_BUFFER, 0, sizeof(LLVector4a) * getNumVerts()-1, (U8*) data);
+    flush_vbo(GL_ARRAY_BUFFER, 0, sizeof(LLVector4a) * getNumVerts()-1, (U8*) data, mMappedData);
+}
+
+void LLVertexBuffer::setTexCoord0Data(const LLVector2* data)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD0], mOffsets[TYPE_TEXCOORD0] + sTypeSize[TYPE_TEXCOORD0] * getNumVerts() - 1, (U8*)data, mMappedData);
 }
 
-void LLVertexBuffer::setTexCoordData(const LLVector2* data)
+void LLVertexBuffer::setTexCoord1Data(const LLVector2* data)
 {
-    llassert(sGLRenderBuffer == mGLBuffer);
-    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD0], mOffsets[TYPE_TEXCOORD0] + sTypeSize[TYPE_TEXCOORD0] * getNumVerts() - 1, (U8*)data);
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD1], mOffsets[TYPE_TEXCOORD1] + sTypeSize[TYPE_TEXCOORD1] * getNumVerts() - 1, (U8*)data, mMappedData);
 }
 
 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);
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_COLOR], mOffsets[TYPE_COLOR] + sTypeSize[TYPE_COLOR] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setNormalData(const LLVector4a* data)
 {
-    llassert(sGLRenderBuffer == mGLBuffer);
-    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_NORMAL], mOffsets[TYPE_NORMAL] + sTypeSize[TYPE_NORMAL] * getNumVerts() - 1, (U8*) data);
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_NORMAL], mOffsets[TYPE_NORMAL] + sTypeSize[TYPE_NORMAL] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setTangentData(const LLVector4a* data)
 {
-    llassert(sGLRenderBuffer == mGLBuffer);
-    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TANGENT], mOffsets[TYPE_TANGENT] + sTypeSize[TYPE_TANGENT] * getNumVerts() - 1, (U8*) data);
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TANGENT], mOffsets[TYPE_TANGENT] + sTypeSize[TYPE_TANGENT] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setWeight4Data(const LLVector4a* data)
 {
-    llassert(sGLRenderBuffer == mGLBuffer);
-    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_WEIGHT4], mOffsets[TYPE_WEIGHT4] + sTypeSize[TYPE_WEIGHT4] * getNumVerts() - 1, (U8*) data);
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_WEIGHT4], mOffsets[TYPE_WEIGHT4] + sTypeSize[TYPE_WEIGHT4] * getNumVerts() - 1, (U8*) data, mMappedData);
+}
+
+void LLVertexBuffer::setJointData(const U64* data)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_JOINT], mOffsets[TYPE_JOINT] + sTypeSize[TYPE_JOINT] * getNumVerts() - 1, (U8*) data, mMappedData);
 }
 
 void LLVertexBuffer::setIndexData(const U16* data)
 {
-    llassert(sGLRenderIndices == mGLIndices);
-    flush_vbo(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(U16) * getNumIndices() - 1, (U8*) data);
+    flush_vbo(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(U16) * getNumIndices() - 1, (U8*) data, mMappedIndexData);
 }
 
 void LLVertexBuffer::setIndexData(const U32* data)
 {
-    llassert(sGLRenderIndices == mGLIndices);
     if (mIndicesType != GL_UNSIGNED_INT)
     { // HACK -- vertex buffers are initialized as 16-bit indices, but can be switched to 32-bit indices
         mIndicesType = GL_UNSIGNED_INT;
         mIndicesStride = 4;
         mNumIndices /= 2;
     }
-    flush_vbo(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(U32) * getNumIndices() - 1, (U8*)data);
+    flush_vbo(GL_ELEMENT_ARRAY_BUFFER, 0, sizeof(U32) * getNumIndices() - 1, (U8*)data, mMappedIndexData);
+}
+
+void LLVertexBuffer::setPositionData(const LLVector4a* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, offset * sizeof(LLVector4a), (offset + count) * sizeof(LLVector4a) - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setNormalData(const LLVector4a* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_NORMAL] + offset * sTypeSize[TYPE_NORMAL], mOffsets[TYPE_NORMAL] + (offset + count) * sTypeSize[TYPE_NORMAL] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setTexCoord0Data(const LLVector2* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD0] + offset * sTypeSize[TYPE_TEXCOORD0], mOffsets[TYPE_TEXCOORD0] + (offset + count) * sTypeSize[TYPE_TEXCOORD0] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setTexCoord1Data(const LLVector2* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TEXCOORD1] + offset * sTypeSize[TYPE_TEXCOORD1], mOffsets[TYPE_TEXCOORD1] + (offset + count) * sTypeSize[TYPE_TEXCOORD1] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setColorData(const LLColor4U* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_COLOR] + offset * sTypeSize[TYPE_COLOR], mOffsets[TYPE_COLOR] + (offset + count) * sTypeSize[TYPE_COLOR] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setTangentData(const LLVector4a* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_TANGENT] + offset * sTypeSize[TYPE_TANGENT], mOffsets[TYPE_TANGENT] + (offset + count) * sTypeSize[TYPE_TANGENT] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setWeight4Data(const LLVector4a* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_WEIGHT4] + offset * sTypeSize[TYPE_WEIGHT4], mOffsets[TYPE_WEIGHT4] + (offset + count) * sTypeSize[TYPE_WEIGHT4] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setJointData(const U64* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ARRAY_BUFFER, mOffsets[TYPE_JOINT] + offset * sTypeSize[TYPE_JOINT], mOffsets[TYPE_JOINT] + (offset + count) * sTypeSize[TYPE_JOINT] - 1, (U8*)data, mMappedData);
+}
+
+void LLVertexBuffer::setIndexData(const U16* data, U32 offset, U32 count)
+{
+    flush_vbo(GL_ELEMENT_ARRAY_BUFFER, offset * sizeof(U16), (offset + count) * sizeof(U16) - 1, (U8*)data, mMappedIndexData);
+}
+
+void LLVertexBuffer::setIndexData(const U32* data, U32 offset, U32 count)
+{
+    if (mIndicesType != GL_UNSIGNED_INT)
+    { // HACK -- vertex buffers are initialized as 16-bit indices, but can be switched to 32-bit indices
+        mIndicesType = GL_UNSIGNED_INT;
+        mIndicesStride = 4;
+        mNumIndices /= 2;
+    }
+    flush_vbo(GL_ELEMENT_ARRAY_BUFFER, offset * sizeof(U32), (offset + count) * sizeof(U32) - 1, (U8*)data, mMappedIndexData);
 }
 
+
+
+