SH-2794 WIP -- work on getting glMapBuffer and friends to behave

5 files changed, 199 insertions, 83 deletions

diff --git a/indra/llrender/llvertexbuffer.cpp b/indra/llrender/llvertexbuffer.cpp
index 20a450fbfb..f8c2a55820 100644
--- a/indra/llrender/llvertexbuffer.cpp
+++ b/indra/llrender/llvertexbuffer.cpp
@@ -148,7 +148,7 @@ U32 wpo2(U32 i)
 	return r;
 }
 
-U8* LLVBOPool::allocate(U32& name, U32 size)
+volatile U8* LLVBOPool::allocate(U32& name, U32 size)
 {
 	llassert(nhpo2(size) == size);
 
@@ -159,7 +159,7 @@ U8* LLVBOPool::allocate(U32& name, U32 size)
 		mFreeList.resize(i+1);
 	}
 
-	U8* ret = NULL;
+	volatile U8* ret = NULL;
 
 	if (mFreeList[i].empty())
 	{
@@ -188,7 +188,7 @@ U8* LLVBOPool::allocate(U32& name, U32 size)
 	return ret;
 }
 
-void LLVBOPool::release(U32 name, U8* buffer, U32 size)
+void LLVBOPool::release(U32 name, volatile U8* buffer, U32 size)
 {
 	llassert(nhpo2(size) == size);
 
@@ -221,7 +221,7 @@ void LLVBOPool::cleanup()
 
 			if (r.mClientData)
 			{
-				ll_aligned_free_16(r.mClientData);
+				ll_aligned_free_16((void*) r.mClientData);
 			}
 
 			l.pop_front();
@@ -1042,7 +1042,7 @@ void LLVertexBuffer::destroyGLBuffer()
 		}
 		else
 		{
-			FREE_MEM(sPrivatePoolp, mMappedData) ;
+			FREE_MEM(sPrivatePoolp, (void*) mMappedData) ;
 			mMappedData = NULL;
 			mEmpty = TRUE;
 		}
@@ -1063,7 +1063,7 @@ void LLVertexBuffer::destroyGLIndices()
 		}
 		else
 		{
-			FREE_MEM(sPrivatePoolp, mMappedIndexData) ;
+			FREE_MEM(sPrivatePoolp, (void*) mMappedIndexData) ;
 			mMappedIndexData = NULL;
 			mEmpty = TRUE;
 		}
@@ -1282,8 +1282,11 @@ bool expand_region(LLVertexBuffer::MappedRegion& region, S32 index, S32 count)
 	return true;
 }
 
+static LLFastTimer::DeclareTimer FTM_VBO_MAP_BUFFER_RANGE("VBO Map Range");
+static LLFastTimer::DeclareTimer FTM_VBO_MAP_BUFFER("VBO Map");
+
 // Map for data access
-U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range)
+volatile U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range)
 {
 	bindGLBuffer(true);
 	LLMemType mt2(LLMemType::MTYPE_VERTEX_MAP_BUFFER);
@@ -1346,13 +1349,14 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 			}
 			else
 			{
-				U8* src = NULL;
+				volatile U8* src = NULL;
 				waitFence();
 				if (gGLManager.mHasMapBufferRange)
 				{
 					if (map_range)
 					{
 #ifdef GL_ARB_map_buffer_range
+						LLFastTimer t(FTM_VBO_MAP_BUFFER_RANGE);
 						S32 offset = mOffsets[type] + sTypeSize[type]*index;
 						S32 length = (sTypeSize[type]*count+0xF) & ~0xF;
 						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, offset, length, 
@@ -1376,6 +1380,7 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 							}
 						}
 
+						LLFastTimer t(FTM_VBO_MAP_BUFFER);
 						src = (U8*) glMapBufferRange(GL_ARRAY_BUFFER_ARB, 0, mSize, 
 							GL_MAP_WRITE_BIT | 
 							GL_MAP_FLUSH_EXPLICIT_BIT);
@@ -1403,7 +1408,7 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 
 				llassert(src != NULL);
 
-				mMappedData = LL_NEXT_ALIGNED_ADDRESS<U8>(src);
+				mMappedData = LL_NEXT_ALIGNED_ADDRESS<volatile U8>(src);
 				mAlignedOffset = mMappedData - src;
 			
 				stop_glerror();
@@ -1458,7 +1463,11 @@ U8* LLVertexBuffer::mapVertexBuffer(S32 type, S32 index, S32 count, bool map_ran
 	}
 }
 
-U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
+
+static LLFastTimer::DeclareTimer FTM_VBO_MAP_INDEX_RANGE("IBO Map Range");
+static LLFastTimer::DeclareTimer FTM_VBO_MAP_INDEX("IBO Map");
+
+volatile U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 {
 	LLMemType mt2(LLMemType::MTYPE_VERTEX_MAP_BUFFER);
 	bindGLIndices(true);
@@ -1530,13 +1539,14 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 			}
 			else
 			{
-				U8* src = NULL;
+				volatile U8* src = NULL;
 				waitFence();
 				if (gGLManager.mHasMapBufferRange)
 				{
 					if (map_range)
 					{
 #ifdef GL_ARB_map_buffer_range
+						LLFastTimer t(FTM_VBO_MAP_INDEX_RANGE);
 						S32 offset = sizeof(U16)*index;
 						S32 length = sizeof(U16)*count;
 						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length, 
@@ -1548,6 +1558,7 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 					else
 					{
 #ifdef GL_ARB_map_buffer_range
+						LLFastTimer t(FTM_VBO_MAP_INDEX);
 						src = (U8*) glMapBufferRange(GL_ELEMENT_ARRAY_BUFFER_ARB, 0, sizeof(U16)*mNumIndices, 
 							GL_MAP_WRITE_BIT | 
 							GL_MAP_FLUSH_EXPLICIT_BIT);
@@ -1569,6 +1580,7 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 				}
 				else
 				{
+					LLFastTimer t(FTM_VBO_MAP_INDEX);
 					map_range = false;
 					src = (U8*) glMapBufferARB(GL_ELEMENT_ARRAY_BUFFER_ARB, GL_WRITE_ONLY_ARB);
 				}
@@ -1619,6 +1631,8 @@ U8* LLVertexBuffer::mapIndexBuffer(S32 index, S32 count, bool map_range)
 	}
 }
 
+static LLFastTimer::DeclareTimer FTM_VBO_UNMAP("VBO Unmap");
+
 void LLVertexBuffer::unmapBuffer()
 {
 	LLMemType mt2(LLMemType::MTYPE_VERTEX_UNMAP_BUFFER);
@@ -1627,6 +1641,8 @@ void LLVertexBuffer::unmapBuffer()
 		return ; //nothing to unmap
 	}
 
+	LLFastTimer t(FTM_VBO_UNMAP);
+
 	bool updated_all = false ;
 
 	if (mMappedData && mVertexLocked)
@@ -1644,7 +1660,7 @@ void LLVertexBuffer::unmapBuffer()
 					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;
-					glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, offset, length, mMappedData+offset);
+					glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, offset, length, (U8*) mMappedData+offset);
 					stop_glerror();
 				}
 
@@ -1653,7 +1669,7 @@ void LLVertexBuffer::unmapBuffer()
 			else
 			{
 				stop_glerror();
-				glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, getSize(), mMappedData);
+				glBufferSubDataARB(GL_ARRAY_BUFFER_ARB, 0, getSize(), (U8*) mMappedData);
 				stop_glerror();
 			}
 		}
@@ -1708,7 +1724,7 @@ void LLVertexBuffer::unmapBuffer()
 					const MappedRegion& region = mMappedIndexRegions[i];
 					S32 offset = region.mIndex >= 0 ? sizeof(U16)*region.mIndex : 0;
 					S32 length = sizeof(U16)*region.mCount;
-					glBufferSubDataARB(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length, mMappedIndexData+offset);
+					glBufferSubDataARB(GL_ELEMENT_ARRAY_BUFFER_ARB, offset, length, (U8*) mMappedIndexData+offset);
 					stop_glerror();
 				}
 
@@ -1717,7 +1733,7 @@ void LLVertexBuffer::unmapBuffer()
 			else
 			{
 				stop_glerror();
-				glBufferSubDataARB(GL_ELEMENT_ARRAY_BUFFER_ARB, 0, getIndicesSize(), mMappedIndexData);
+				glBufferSubDataARB(GL_ELEMENT_ARRAY_BUFFER_ARB, 0, getIndicesSize(), (U8*) mMappedIndexData);
 				stop_glerror();
 			}
 		}
@@ -1778,7 +1794,7 @@ template <class T,S32 type> struct VertexBufferStrider
 	{
 		if (type == LLVertexBuffer::TYPE_INDEX)
 		{
-			U8* ptr = vbo.mapIndexBuffer(index, count, map_range);
+			volatile U8* ptr = vbo.mapIndexBuffer(index, count, map_range);
 
 			if (ptr == NULL)
 			{
@@ -1794,7 +1810,7 @@ template <class T,S32 type> struct VertexBufferStrider
 		{
 			S32 stride = LLVertexBuffer::sTypeSize[type];
 
-			U8* ptr = vbo.mapVertexBuffer(type, index, count, map_range);
+			volatile U8* ptr = vbo.mapVertexBuffer(type, index, count, map_range);
 
 			if (ptr == NULL)
 			{
@@ -2109,7 +2125,7 @@ void LLVertexBuffer::setupVertexBuffer(U32 data_mask)
 {
 	LLMemType mt2(LLMemType::MTYPE_VERTEX_SETUP_VERTEX_BUFFER);
 	stop_glerror();
-	U8* base = useVBOs() ? (U8*) mAlignedOffset : mMappedData;
+	volatile U8* base = useVBOs() ? (U8*) mAlignedOffset : mMappedData;
 
 	/*if ((data_mask & mTypeMask) != data_mask)
 	{
diff --git a/indra/llrender/llvertexbuffer.h b/indra/llrender/llvertexbuffer.h
index 3e6f6a959a..5b93a0389f 100644
--- a/indra/llrender/llvertexbuffer.h
+++ b/indra/llrender/llvertexbuffer.h
@@ -60,10 +60,10 @@ public:
 	U32 mType;
 
 	//size MUST be a power of 2
-	U8* allocate(U32& name, U32 size);
+	volatile U8* allocate(U32& name, U32 size);
 	
 	//size MUST be the size provided to allocate that returned the given name
-	void release(U32 name, U8* buffer, U32 size);
+	void release(U32 name, volatile U8* buffer, U32 size);
 	
 	//destroy all records in mFreeList
 	void cleanup();
@@ -72,7 +72,7 @@ public:
 	{
 	public:
 		U32 mGLName;
-		U8* mClientData;
+		volatile U8* mClientData;
 	};
 
 	typedef std::list<Record> record_list_t;
@@ -208,8 +208,8 @@ public:
 	LLVertexBuffer(U32 typemask, S32 usage);
 	
 	// map for data access
-	U8*		mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range);
-	U8*		mapIndexBuffer(S32 index, S32 count, bool map_range);
+	volatile U8*		mapVertexBuffer(S32 type, S32 index, S32 count, bool map_range);
+	volatile U8*		mapIndexBuffer(S32 index, S32 count, bool map_range);
 
 	// set for rendering
 	virtual void	setBuffer(U32 data_mask); 	// calls  setupVertexBuffer() if data_mask is not 0
@@ -244,14 +244,14 @@ public:
 	S32 getNumVerts() const					{ return mNumVerts; }
 	S32 getNumIndices() const				{ return mNumIndices; }
 	
-	U8* getIndicesPointer() const			{ return useVBOs() ? (U8*) mAlignedIndexOffset : mMappedIndexData; }
-	U8* getVerticesPointer() const			{ return useVBOs() ? (U8*) mAlignedOffset : mMappedData; }
+	volatile U8* getIndicesPointer() const			{ return useVBOs() ? (U8*) mAlignedIndexOffset : mMappedIndexData; }
+	volatile U8* getVerticesPointer() const			{ return useVBOs() ? (U8*) mAlignedOffset : mMappedData; }
 	U32 getTypeMask() const					{ return mTypeMask; }
 	bool hasDataType(S32 type) const		{ return ((1 << type) & getTypeMask()); }
 	S32 getSize() const;
 	S32 getIndicesSize() const				{ return mIndicesSize; }
-	U8* getMappedData() const				{ return mMappedData; }
-	U8* getMappedIndices() const			{ return mMappedIndexData; }
+	volatile U8* getMappedData() const				{ return mMappedData; }
+	volatile U8* getMappedIndices() const			{ return mMappedIndexData; }
 	S32 getOffset(S32 type) const			{ return mOffsets[type]; }
 	S32 getUsage() const					{ return mUsage; }
 
@@ -278,8 +278,8 @@ protected:
 	U32		mGLIndices;		// GL IBO handle
 	U32		mGLArray;		// GL VAO handle
 	
-	U8*		mMappedData;	// pointer to currently mapped data (NULL if unmapped)
-	U8*		mMappedIndexData;	// pointer to currently mapped indices (NULL if unmapped)
+	volatile U8* mMappedData;	// pointer to currently mapped data (NULL if unmapped)
+	volatile U8* mMappedIndexData;	// pointer to currently mapped indices (NULL if unmapped)
 	BOOL	mVertexLocked;			// if TRUE, vertex buffer is being or has been written to in client memory
 	BOOL	mIndexLocked;			// if TRUE, index buffer is being or has been written to in client memory
 	BOOL	mFinal;			// if TRUE, buffer can not be mapped again
diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp
index 6dbeae6677..941b8db2cf 100644
--- a/indra/newview/llface.cpp
+++ b/indra/newview/llface.cpp
@@ -693,6 +693,49 @@ static void xform(LLVector2 &tex_coord, F32 cosAng, F32 sinAng, F32 offS, F32 of
 	tex_coord.mV[1] = t;
 }
 
+// Transform the texture coordinates for this face.
+static void xform4a(LLVector4a &tex_coord, const LLVector4a& trans, const LLVector4Logical& mask, const LLVector4a& rot0, const LLVector4a& rot1, const LLVector4a& offset, const LLVector4a& scale) 
+{
+	//tex coord is two coords, <s0, t0, s1, t1>
+	LLVector4a st;
+
+	// Texture transforms are done about the center of the face.
+	st.setAdd(tex_coord, trans);
+	
+	// Handle rotation
+	LLVector4a rot_st;
+		
+	// <s0 * cosAng, s0*-sinAng, s1*cosAng, s1*-sinAng>
+	LLVector4a s0;
+	s0.splat(st, 0);
+	LLVector4a s1;
+	s1.splat(st, 2);
+	LLVector4a ss;
+	ss.setSelectWithMask(mask, s1, s0);
+
+	LLVector4a a; 
+	a.setMul(rot0, ss);
+	
+	// <t0*sinAng, t0*cosAng, t1*sinAng, t1*cosAng>
+	LLVector4a t0;
+	t0.splat(st, 1);
+	LLVector4a t1;
+	t1.splat(st, 3);
+	LLVector4a tt;
+	tt.setSelectWithMask(mask, t1, t0);
+
+	LLVector4a b;
+	b.setMul(rot1, tt);
+		
+	st.setAdd(a,b);
+
+	// Then scale
+	st.mul(scale);
+
+	// Then offset
+	tex_coord.setAdd(st, offset);
+}
+
 
 bool less_than_max_mag(const LLVector4a& vec)
 {
@@ -1060,6 +1103,16 @@ static LLFastTimer::DeclareTimer FTM_FACE_GEOM_EMISSIVE("Emissive");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_WEIGHTS("Weights");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_BINORMAL("Binormal");
 static LLFastTimer::DeclareTimer FTM_FACE_GEOM_INDEX("Index");
+static LLFastTimer::DeclareTimer FTM_FACE_GEOM_INDEX_TAIL("Tail");
+static LLFastTimer::DeclareTimer FTM_FACE_POSITION_STORE("Pos");
+static LLFastTimer::DeclareTimer FTM_FACE_TEXTURE_INDEX_STORE("TexIdx");
+static LLFastTimer::DeclareTimer FTM_FACE_POSITION_PAD("Pad");
+static LLFastTimer::DeclareTimer FTM_FACE_TEX_DEFAULT("Default");
+static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK("Quick");
+static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_NO_XFORM("No Xform");
+static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_XFORM("Xform");
+
+static LLFastTimer::DeclareTimer FTM_FACE_TEX_QUICK_PLANAR("Quick Planar");
 
 BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 							   const S32 &f,
@@ -1104,16 +1157,12 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 	}
 
 	LLStrider<LLVector3> vert;
-	LLVector4a* vertices = NULL;
 	LLStrider<LLVector2> tex_coords;
 	LLStrider<LLVector2> tex_coords2;
-	LLVector4a* normals = NULL;
 	LLStrider<LLVector3> norm;
 	LLStrider<LLColor4U> colors;
-	LLVector4a* binormals = NULL;
 	LLStrider<LLVector3> binorm;
 	LLStrider<U16> indicesp;
-	LLVector4a* weights = NULL;
 	LLStrider<LLVector4> wght;
 
 	BOOL full_rebuild = force_rebuild || mDrawablep->isState(LLDrawable::REBUILD_VOLUME);
@@ -1202,7 +1251,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		LLFastTimer t(FTM_FACE_GEOM_INDEX);
 		mVertexBuffer->getIndexStrider(indicesp, mIndicesIndex, mIndicesCount, map_range);
 
-		__m128i* dst = (__m128i*) indicesp.get();
+		volatile __m128i* dst = (__m128i*) indicesp.get();
 		__m128i* src = (__m128i*) vf.mIndices;
 		__m128i offset = _mm_set1_epi16(index_offset);
 
@@ -1211,12 +1260,17 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		for (S32 i = 0; i < end; i++)
 		{
 			__m128i res = _mm_add_epi16(src[i], offset);
-			_mm_storeu_si128(dst+i, res);
+			_mm_storeu_si128((__m128i*) dst++, res);
 		}
 
-		for (S32 i = end*8; i < num_indices; ++i)
 		{
-			indicesp[i] = vf.mIndices[i]+index_offset;
+			LLFastTimer t(FTM_FACE_GEOM_INDEX_TAIL);
+			U16* idx = (U16*) dst;
+
+			for (S32 i = end*8; i < num_indices; ++i)
+			{
+				*idx++ = vf.mIndices[i]+index_offset;
+			}
 		}
 
 		if (map_range)
@@ -1373,19 +1427,48 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 
 			if (texgen != LLTextureEntry::TEX_GEN_PLANAR)
 			{
+				LLFastTimer t(FTM_FACE_TEX_QUICK);
 				if (!do_tex_mat)
 				{
 					if (!do_xform)
 					{
+						LLFastTimer t(FTM_FACE_TEX_QUICK_NO_XFORM);
 						LLVector4a::memcpyNonAliased16((F32*) tex_coords.get(), (F32*) vf.mTexCoords, num_vertices*2*sizeof(F32));
 					}
 					else
 					{
-						for (S32 i = 0; i < num_vertices; i++)
+						LLFastTimer t(FTM_FACE_TEX_QUICK_XFORM);
+						F32* dst = (F32*) tex_coords.get();
+						LLVector4a* src = (LLVector4a*) vf.mTexCoords;
+
+						LLVector4a trans;
+						trans.splat(-0.5f);
+
+						LLVector4a rot0;
+						rot0.set(cos_ang, -sin_ang, cos_ang, -sin_ang);
+
+						LLVector4a rot1;
+						rot1.set(sin_ang, cos_ang, sin_ang, cos_ang);
+
+						LLVector4a scale;
+						scale.set(ms, mt, ms, mt);
+
+						LLVector4a offset;
+						offset.set(os+0.5f, ot+0.5f, os+0.5f, ot+0.5f);
+
+						LLVector4Logical mask;
+						mask.clear();
+						mask.setElement<2>();
+						mask.setElement<3>();
+
+						U32 count = num_vertices/2 + num_vertices%2;
+
+						for (S32 i = 0; i < count; i++)
 						{	
-							LLVector2 tc(vf.mTexCoords[i]);
-							xform(tc, cos_ang, sin_ang, os, ot, ms, mt);
-							*tex_coords++ = tc;	
+							LLVector4a res = *src++;
+							xform4a(res, trans, mask, rot0, rot1, offset, scale);
+							res.store4a(dst);
+							dst += 4;
 						}
 					}
 				}
@@ -1407,6 +1490,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 			}
 			else
 			{ //no bump, no atlas, tex gen planar
+				LLFastTimer t(FTM_FACE_TEX_QUICK_PLANAR);
 				if (do_tex_mat)
 				{
 					for (S32 i = 0; i < num_vertices; i++)
@@ -1451,6 +1535,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		}
 		else
 		{ //either bump mapped or in atlas, just do the whole expensive loop
+			LLFastTimer t(FTM_FACE_TEX_DEFAULT);
 			mVertexBuffer->getTexCoord0Strider(tex_coords, mGeomIndex, mGeomCount, map_range);
 
 			std::vector<LLVector2> bump_tc;
@@ -1642,44 +1727,55 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		llassert(num_vertices > 0);
 		
 		mVertexBuffer->getVertexStrider(vert, mGeomIndex, mGeomCount, map_range);
-		vertices = (LLVector4a*) vert.get();
-	
+			
+
 		LLMatrix4a mat_vert;
 		mat_vert.loadu(mat_vert_in);
 
 		LLVector4a* src = vf.mPositions;
-		LLVector4a* dst = vertices;
+		volatile F32* dst = (volatile F32*) vert.get();
 
-		LLVector4a* end = dst+num_vertices;
-		do
-		{	
-			mat_vert.affineTransform(*src++, *dst++);
-		}
-		while(dst < end);
+		volatile F32* end = dst+num_vertices*4;
+		LLVector4a res;
 
-		F32 index = (F32) (mTextureIndex < 255 ? mTextureIndex : 0);
+		LLVector4a texIdx;
 
+		F32 index = (F32) (mTextureIndex < 255 ? mTextureIndex : 0);
 		llassert(index <= LLGLSLShader::sIndexedTextureChannels-1);
-		F32 *index_dst = (F32*) vertices;
-		F32 *index_end = (F32*) end;
 
-		index_dst += 3;
-		index_end += 3;
-		do
+		LLVector4Logical mask;
+		mask.clear();
+		mask.setElement<3>();
+		
+		texIdx.set(0,0,0,index);
+
 		{
-			*index_dst = index;
-			index_dst += 4;
+			LLFastTimer t(FTM_FACE_POSITION_STORE);
+			LLVector4a tmp;
+
+			do
+			{	
+				mat_vert.affineTransform(*src++, res);
+				tmp.setSelectWithMask(mask, texIdx, res);
+				tmp.store4a((F32*) dst);
+				dst += 4;
+			}
+			while(dst < end);
 		}
-		while (index_dst < index_end);
-		
-		S32 aligned_pad_vertices = mGeomCount - num_vertices;
-		LLVector4a* last_vec = end - 1;
-		while (aligned_pad_vertices > 0)
+
 		{
-			--aligned_pad_vertices;
-			*dst++ = *last_vec;
+			LLFastTimer t(FTM_FACE_POSITION_PAD);
+			S32 aligned_pad_vertices = mGeomCount - num_vertices;
+			res.set(res[0], res[1], res[2], 0.f);
+
+			while (aligned_pad_vertices > 0)
+			{
+				--aligned_pad_vertices;
+				res.store4a((F32*) dst);
+				dst += 4;
+			}
 		}
-		
+
 		if (map_range)
 		{
 			mVertexBuffer->flush();
@@ -1690,14 +1786,15 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 	{
 		LLFastTimer t(FTM_FACE_GEOM_NORMAL);
 		mVertexBuffer->getNormalStrider(norm, mGeomIndex, mGeomCount, map_range);
-		normals = (LLVector4a*) norm.get();
+		F32* normals = (F32*) norm.get();
 	
 		for (S32 i = 0; i < num_vertices; i++)
 		{	
 			LLVector4a normal;
 			mat_normal.rotate(vf.mNormals[i], normal);
 			normal.normalize3fast();
-			normals[i] = normal;
+			normal.store4a(normals);
+			normals += 4;
 		}
 
 		if (map_range)
@@ -1710,14 +1807,15 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 	{
 		LLFastTimer t(FTM_FACE_GEOM_BINORMAL);
 		mVertexBuffer->getBinormalStrider(binorm, mGeomIndex, mGeomCount, map_range);
-		binormals = (LLVector4a*) binorm.get();
+		F32* binormals = (F32*) binorm.get();
 		
 		for (S32 i = 0; i < num_vertices; i++)
 		{	
 			LLVector4a binormal;
 			mat_normal.rotate(vf.mBinormals[i], binormal);
 			binormal.normalize3fast();
-			binormals[i] = binormal;
+			binormal.store4a(binormals);
+			binormals += 4;
 		}
 
 		if (map_range)
@@ -1730,8 +1828,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 	{
 		LLFastTimer t(FTM_FACE_GEOM_WEIGHTS);
 		mVertexBuffer->getWeight4Strider(wght, mGeomIndex, mGeomCount, map_range);
-		weights = (LLVector4a*) wght.get();
-		LLVector4a::memcpyNonAliased16((F32*) weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32));
+		F32* weights = (F32*) wght.get();
+		LLVector4a::memcpyNonAliased16(weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32));
 		if (map_range)
 		{
 			mVertexBuffer->flush();
@@ -1750,7 +1848,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		
 		src.loadua((F32*) vec);
 
-		LLVector4a* dst = (LLVector4a*) colors.get();
+		F32* dst = (F32*) colors.get();
 		S32 num_vecs = num_vertices/4;
 		if (num_vertices%4 > 0)
 		{
@@ -1759,7 +1857,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 
 		for (S32 i = 0; i < num_vecs; i++)
 		{	
-			dst[i] = src;
+			src.store4a(dst);
+			dst += 4;
 		}
 
 		if (map_range)
@@ -1789,7 +1888,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 		
 		src.loadua((F32*) vec);
 
-		LLVector4a* dst = (LLVector4a*) emissive.get();
+		F32* dst = (F32*) emissive.get();
 		S32 num_vecs = num_vertices/4;
 		if (num_vertices%4 > 0)
 		{
@@ -1798,7 +1897,8 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume,
 
 		for (S32 i = 0; i < num_vecs; i++)
 		{	
-			dst[i] = src;
+			src.store4a(dst);
+			dst += 4;
 		}
 
 		if (map_range)
diff --git a/indra/newview/llfloatermodelpreview.cpp b/indra/newview/llfloatermodelpreview.cpp
index 64bdcccd9f..9122e5a8f5 100755
--- a/indra/newview/llfloatermodelpreview.cpp
+++ b/indra/newview/llfloatermodelpreview.cpp
@@ -3969,7 +3969,7 @@ void LLModelPreview::genLODs(S32 which_lod, U32 decimation, bool enforce_tri_lim
 				U32 num_indices = mVertexBuffer[5][mdl][i]->getNumIndices();
 				if (num_indices > 2)
 				{
-					glodInsertElements(mObject[mdl], i, GL_TRIANGLES, num_indices, GL_UNSIGNED_SHORT, mVertexBuffer[5][mdl][i]->getIndicesPointer(), 0, 0.f);
+					glodInsertElements(mObject[mdl], i, GL_TRIANGLES, num_indices, GL_UNSIGNED_SHORT, (U8*) mVertexBuffer[5][mdl][i]->getIndicesPointer(), 0, 0.f);
 				}
 				tri_count += num_indices/3;
 				stop_gloderror();
@@ -4083,14 +4083,14 @@ void LLModelPreview::genLODs(S32 which_lod, U32 decimation, bool enforce_tri_lim
 				{
 					buff->allocateBuffer(sizes[i*2+1], sizes[i*2], true);
 					buff->setBuffer(type_mask);
-					glodFillElements(mObject[base], names[i], GL_UNSIGNED_SHORT, buff->getIndicesPointer());
+					glodFillElements(mObject[base], names[i], GL_UNSIGNED_SHORT, (U8*) buff->getIndicesPointer());
 					stop_gloderror();
 				}
 				else
 				{ //this face was eliminated, create a dummy triangle (one vertex, 3 indices, all 0)
 					buff->allocateBuffer(1, 3, true);
-					memset(buff->getMappedData(), 0, buff->getSize());
-					memset(buff->getIndicesPointer(), 0, buff->getIndicesSize());
+					memset((U8*) buff->getMappedData(), 0, buff->getSize());
+					memset((U8*) buff->getIndicesPointer(), 0, buff->getIndicesSize());
 				}
 
 				buff->validateRange(0, buff->getNumVerts()-1, buff->getNumIndices(), 0);
@@ -4880,8 +4880,8 @@ void LLModelPreview::addEmptyFace( LLModel* pTarget )
 	LLPointer<LLVertexBuffer> buff = new LLVertexBuffer(type_mask, 0);
 	
 	buff->allocateBuffer(1, 3, true);
-	memset( buff->getMappedData(), 0, buff->getSize() );
-	memset( buff->getIndicesPointer(), 0, buff->getIndicesSize() );
+	memset( (U8*) buff->getMappedData(), 0, buff->getSize() );
+	memset( (U8*) buff->getIndicesPointer(), 0, buff->getIndicesSize() );
 		
 	buff->validateRange( 0, buff->getNumVerts()-1, buff->getNumIndices(), 0 );
 		
diff --git a/indra/newview/llvosurfacepatch.cpp b/indra/newview/llvosurfacepatch.cpp
index c3a2e6a712..bf6158eeaf 100644
--- a/indra/newview/llvosurfacepatch.cpp
+++ b/indra/newview/llvosurfacepatch.cpp
@@ -65,7 +65,7 @@ public:
 			return;
 		}
 
-		U8* base = useVBOs() ? (U8*) mAlignedOffset : mMappedData;
+		volatile U8* base = useVBOs() ? (U8*) mAlignedOffset : mMappedData;
 
 		//assume tex coords 2 and 3 are present
 		U32 type_mask = mTypeMask | MAP_TEXCOORD2 | MAP_TEXCOORD3;