diff options
-rw-r--r-- | indra/cmake/00-Common.cmake | 2 | ||||
-rw-r--r-- | indra/cmake/GooglePerfTools.cmake | 5 | ||||
-rw-r--r-- | indra/llmath/lloctree.h | 127 | ||||
-rw-r--r-- | indra/llmath/llvolume.cpp | 180 | ||||
-rw-r--r-- | indra/llmath/llvolume.h | 4 | ||||
-rw-r--r-- | indra/llmath/llvolumeoctree.cpp | 61 | ||||
-rw-r--r-- | indra/llmath/llvolumeoctree.h | 6 | ||||
-rw-r--r-- | indra/newview/CMakeLists.txt | 6 | ||||
-rw-r--r-- | indra/newview/lldrawable.cpp | 8 | ||||
-rw-r--r-- | indra/newview/lldrawable.h | 56 | ||||
-rw-r--r-- | indra/newview/lldrawpoolavatar.cpp | 10 | ||||
-rw-r--r-- | indra/newview/llface.cpp | 34 | ||||
-rw-r--r-- | indra/newview/llfloatertools.cpp | 8 | ||||
-rw-r--r-- | indra/newview/llselectmgr.cpp | 50 | ||||
-rw-r--r-- | indra/newview/llspatialpartition.cpp | 109 | ||||
-rw-r--r-- | indra/newview/llviewerobject.cpp | 14 | ||||
-rw-r--r-- | indra/newview/llvovolume.cpp | 314 | ||||
-rw-r--r-- | indra/newview/llvovolume.h | 32 | ||||
-rw-r--r-- | install.xml | 8 |
19 files changed, 809 insertions, 225 deletions
diff --git a/indra/cmake/00-Common.cmake b/indra/cmake/00-Common.cmake index beb4f181b5..8b0864a539 100644 --- a/indra/cmake/00-Common.cmake +++ b/indra/cmake/00-Common.cmake @@ -38,7 +38,7 @@ if (WINDOWS) set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /Od /Zi /MDd /MP" CACHE STRING "C++ compiler debug options" FORCE) set(CMAKE_CXX_FLAGS_RELWITHDEBINFO - "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} /Od /Zi /MD /MP /Gm /INCREMENTAL" + "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} /Od /Zi /MD /MP /Gm" CACHE STRING "C++ compiler release-with-debug options" FORCE) set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} ${LL_CXX_FLAGS} /O2 /Zi /MD /MP" diff --git a/indra/cmake/GooglePerfTools.cmake b/indra/cmake/GooglePerfTools.cmake index 946fc6b375..ee231dcde6 100644 --- a/indra/cmake/GooglePerfTools.cmake +++ b/indra/cmake/GooglePerfTools.cmake @@ -6,7 +6,7 @@ if (STANDALONE) else (STANDALONE) use_prebuilt_binary(google) if (WINDOWS) - use_prebuilt_binary(google-perftools) + use_prebuilt_binary(tcmalloc) set(TCMALLOC_LIBRARIES debug libtcmalloc_minimal-debug optimized libtcmalloc_minimal) @@ -29,8 +29,7 @@ if (GOOGLE_PERFTOOLS_FOUND) endif (GOOGLE_PERFTOOLS_FOUND) if (WINDOWS) - # *TODO -reenable this once we get server usage sorted out - #set(USE_GOOGLE_PERFTOOLS ON) + set(USE_GOOGLE_PERFTOOLS ON) endif (WINDOWS) if (USE_GOOGLE_PERFTOOLS) diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h index 73910ef98d..63adfa85b2 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -95,22 +95,30 @@ public: typedef LLOctreeNode<T> oct_node; typedef LLOctreeListener<T> oct_listener; + /*void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + }*/ + LLOctreeNode( const LLVector4a& center, const LLVector4a& size, BaseType* parent, - S32 octant = -1) + U8 octant = 255) : mParent((oct_node*)parent), mOctant(octant) { - mD = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*4); - - mD[CENTER] = center; - mD[SIZE] = size; + mCenter = center; + mSize = size; updateMinMax(); - if ((mOctant == -1) && mParent) + if ((mOctant == 255) && mParent) { - mOctant = ((oct_node*) mParent)->getOctant(mD[CENTER]); + mOctant = ((oct_node*) mParent)->getOctant(mCenter); } clearChildren(); @@ -124,30 +132,27 @@ public: { delete getChild(i); } - - ll_aligned_free_16(mD); } inline const BaseType* getParent() const { return mParent; } inline void setParent(BaseType* parent) { mParent = (oct_node*) parent; } - inline const LLVector4a& getCenter() const { return mD[CENTER]; } - inline const LLVector4a& getSize() const { return mD[SIZE]; } - inline void setCenter(const LLVector4a& center) { mD[CENTER] = center; } - inline void setSize(const LLVector4a& size) { mD[SIZE] = size; } + inline const LLVector4a& getCenter() const { return mCenter; } + inline const LLVector4a& getSize() const { return mSize; } + inline void setCenter(const LLVector4a& center) { mCenter = center; } + inline void setSize(const LLVector4a& size) { mSize = size; } inline oct_node* getNodeAt(T* data) { return getNodeAt(data->getPositionGroup(), data->getBinRadius()); } - inline S32 getOctant() const { return mOctant; } - inline void setOctant(S32 octant) { mOctant = octant; } + inline U8 getOctant() const { return mOctant; } inline const oct_node* getOctParent() const { return (const oct_node*) getParent(); } inline oct_node* getOctParent() { return (oct_node*) getParent(); } - S32 getOctant(const LLVector4a& pos) const //get the octant pos is in + U8 getOctant(const LLVector4a& pos) const //get the octant pos is in { - return pos.greaterThan(mD[CENTER]).getGatheredBits() & 0x7; + return (U8) (pos.greaterThan(mCenter).getGatheredBits() & 0x7); } inline bool isInside(const LLVector4a& pos, const F32& rad) const { - return rad <= mD[SIZE][0]*2.f && isInside(pos); + return rad <= mSize[0]*2.f && isInside(pos); } inline bool isInside(T* data) const @@ -157,13 +162,13 @@ public: bool isInside(const LLVector4a& pos) const { - S32 gt = pos.greaterThan(mD[MAX]).getGatheredBits() & 0x7; + S32 gt = pos.greaterThan(mMax).getGatheredBits() & 0x7; if (gt) { return false; } - S32 lt = pos.lessEqual(mD[MIN]).getGatheredBits() & 0x7; + S32 lt = pos.lessEqual(mMin).getGatheredBits() & 0x7; if (lt) { return false; @@ -174,8 +179,8 @@ public: void updateMinMax() { - mD[MAX].setAdd(mD[CENTER], mD[SIZE]); - mD[MIN].setSub(mD[CENTER], mD[SIZE]); + mMax.setAdd(mCenter, mSize); + mMin.setSub(mCenter, mSize); } inline oct_listener* getOctListener(U32 index) @@ -195,7 +200,7 @@ public: return false; } - F32 size = mD[SIZE][0]; + F32 size = mSize[0]; F32 p_size = size * 2.f; return (radius <= 0.001f && size <= 0.001f) || @@ -234,6 +239,29 @@ public: void accept(tree_traveler* visitor) const { visitor->visit(this); } void accept(oct_traveler* visitor) const { visitor->visit(this); } + void validateChildMap() + { + for (U32 i = 0; i < 8; i++) + { + U8 idx = mChildMap[i]; + if (idx != 255) + { + LLOctreeNode<T>* child = mChild[idx]; + + if (child->getOctant() != i) + { + llerrs << "Invalid child map, bad octant data." << llendl; + } + + if (getOctant(child->getCenter()) != child->getOctant()) + { + llerrs << "Invalid child octant compared to position data." << llendl; + } + } + } + } + + oct_node* getNodeAt(const LLVector4a& pos, const F32& rad) { LLOctreeNode<T>* node = this; @@ -241,25 +269,19 @@ public: if (node->isInside(pos, rad)) { //do a quick search by octant - S32 octant = node->getOctant(pos); - BOOL keep_going = TRUE; - + U8 octant = node->getOctant(pos); + //traverse the tree until we find a node that has no node //at the appropriate octant or is smaller than the object. //by definition, that node is the smallest node that contains // the data - while (keep_going && node->getSize()[0] >= rad) + U8 next_node = node->mChildMap[octant]; + + while (next_node != 255 && node->getSize()[0] >= rad) { - keep_going = FALSE; - for (U32 i = 0; i < node->getChildCount() && !keep_going; i++) - { - if (node->getChild(i)->getOctant() == octant) - { - node = node->getChild(i); - octant = node->getOctant(pos); - keep_going = TRUE; - } - } + node = node->getChild(next_node); + octant = node->getOctant(pos); + next_node = node->mChildMap[octant]; } } else if (!node->contains(rad) && node->getParent()) @@ -439,6 +461,9 @@ public: void clearChildren() { mChild.clear(); + + U32* foo = (U32*) mChildMap; + foo[0] = foo[1] = 0xFFFFFFFF; } void validate() @@ -496,6 +521,8 @@ public: } #endif + mChildMap[child->getOctant()] = (U8) mChild.size(); + mChild.push_back(child); child->setParent(this); @@ -517,6 +544,8 @@ public: listener->handleChildRemoval(this, getChild(index)); } + + if (destroy) { mChild[index]->destroy(); @@ -524,6 +553,15 @@ public: } mChild.erase(mChild.begin() + index); + //rebuild child map + U32* foo = (U32*) mChildMap; + foo[0] = foo[1] = 0xFFFFFFFF; + + for (U32 i = 0; i < mChild.size(); ++i) + { + mChildMap[mChild[i]->getOctant()] = i; + } + checkAlive(); } @@ -562,15 +600,20 @@ protected: MIN = 3 } eDName; - LLVector4a* mD; + LLVector4a mCenter; + LLVector4a mSize; + LLVector4a mMax; + LLVector4a mMin; oct_node* mParent; - S32 mOctant; + U8 mOctant; child_list mChild; + U8 mChildMap[8]; + element_list mData; -}; +}; //just like a regular node, except it might expand on insert and compress on balance template <class T> @@ -613,6 +656,8 @@ public: //destroy child child->clearChildren(); delete child; + + return false; } return true; @@ -639,7 +684,7 @@ public: const LLVector4a& v = data->getPositionGroup(); LLVector4a val; - val.setSub(v, BaseType::mD[BaseType::CENTER]); + val.setSub(v, BaseType::mCenter); val.setAbs(val); S32 lt = val.lessThan(MAX_MAG).getGatheredBits() & 0x7; diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index ab9f8c4c24..24528a8ce9 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -94,6 +94,8 @@ const F32 SKEW_MAX = 0.95f; const F32 SCULPT_MIN_AREA = 0.002f; const S32 SCULPT_MIN_AREA_DETAIL = 1; +extern BOOL gDebugGL; + BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLVector3& pt3, const LLVector3& norm) { LLVector3 test = (pt2-pt1)%(pt3-pt2); @@ -308,22 +310,26 @@ public: } virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch) - { + { //this is a depth first traversal, so it's safe to assum all children have complete + //bounding data + LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0); LLVector4a& min = node->mExtents[0]; LLVector4a& max = node->mExtents[1]; - if (branch->getElementCount() != 0) - { + if (!branch->getData().empty()) + { //node has data, find AABB that binds data set const LLVolumeTriangle* tri = *(branch->getData().begin()); - + + //initialize min/max to first available vertex min = *(tri->mV[0]); max = *(tri->mV[0]); for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = branch->getData().begin(); iter != branch->getData().end(); ++iter) - { + { //for each triangle in node + //stretch by triangles in node tri = *iter; @@ -335,33 +341,27 @@ public: max.setMax(max, *tri->mV[1]); max.setMax(max, *tri->mV[2]); } - - for (S32 i = 0; i < branch->getChildCount(); ++i) - { //stretch by child extents - LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); - min.setMin(min, child->mExtents[0]); - max.setMax(min, child->mExtents[1]); - } } - else if (branch->getChildCount() != 0) - { + else if (!branch->getChildren().empty()) + { //no data, but child nodes exist LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(0)->getListener(0); + //initialize min/max to extents of first child min = child->mExtents[0]; max = child->mExtents[1]; - - for (S32 i = 1; i < branch->getChildCount(); ++i) - { //stretch by child extents - child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); - min.setMin(min, child->mExtents[0]); - max.setMax(max, child->mExtents[1]); - } } else { llerrs << "WTF? Empty leaf" << llendl; } - + + for (S32 i = 0; i < branch->getChildCount(); ++i) + { //stretch by child extents + LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); + min.setMin(min, child->mExtents[0]); + max.setMax(max, child->mExtents[1]); + } + node->mBounds[0].setAdd(min, max); node->mBounds[0].mul(0.5f); @@ -370,7 +370,6 @@ public: } }; - //------------------------------------------------------------------- // statics //------------------------------------------------------------------- @@ -1991,7 +1990,7 @@ void LLVolumeFace::VertexData::init() { if (!mData) { - mData = (LLVector4a*) ll_aligned_malloc_16(32); + mData = new LLVector4a[2]; } } @@ -2020,7 +2019,7 @@ const LLVolumeFace::VertexData& LLVolumeFace::VertexData::operator=(const LLVolu LLVolumeFace::VertexData::~VertexData() { - ll_aligned_free_16(mData); + delete [] mData; } LLVector4a& LLVolumeFace::VertexData::getPosition() @@ -2258,12 +2257,12 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) U32 cur_influence = 0; LLVector4 wght(0,0,0,0); - while (joint != END_INFLUENCES) + while (joint != END_INFLUENCES && idx < weights.size()) { U16 influence = weights[idx++]; influence |= ((U16) weights[idx++] << 8); - F32 w = llmin((F32) influence / 65535.f, 0.99999f); + F32 w = llclamp((F32) influence / 65535.f, 0.f, 0.99999f); wght.mV[cur_influence++] = (F32) joint + w; if (cur_influence >= 4) @@ -2570,14 +2569,13 @@ void LLVolume::makeTetrahedron() mIsTetrahedron = TRUE; } -void LLVolume::copyVolumeFaces(LLVolume* volume) +void LLVolume::copyVolumeFaces(const LLVolume* volume) { mVolumeFaces = volume->mVolumeFaces; mSculptLevel = 0; mIsTetrahedron = FALSE; } - S32 LLVolume::getNumFaces() const { #if LL_MESH_ENABLED @@ -5232,7 +5230,7 @@ LLVolumeFace::LLVolumeFace() : mWeights(NULL), mOctree(NULL) { - mExtents = (LLVector4a*) ll_aligned_malloc_16(48); + mExtents = new LLVector4a[3]; mCenter = mExtents+2; } @@ -5253,7 +5251,7 @@ LLVolumeFace::LLVolumeFace(const LLVolumeFace& src) mWeights(NULL), mOctree(NULL) { - mExtents = (LLVector4a*) ll_aligned_malloc_16(48); + mExtents = new LLVector4a[3]; mCenter = mExtents+2; *this = src; } @@ -5288,7 +5286,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) if (mNumVertices) { - S32 vert_size = mNumVertices*4*sizeof(F32); + S32 vert_size = mNumVertices*sizeof(LLVector4a); S32 tc_size = (mNumVertices*8+0xF) & ~0xF; LLVector4a::memcpyNonAliased16((F32*) mPositions, (F32*) src.mPositions, vert_size); @@ -5303,7 +5301,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) } else { - ll_aligned_free_16(mBinormals); + delete [] mBinormals; mBinormals = NULL; } @@ -5314,7 +5312,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) } else { - ll_aligned_free_16(mWeights); + delete [] mWeights; mWeights = NULL; } } @@ -5332,7 +5330,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) LLVolumeFace::~LLVolumeFace() { - ll_aligned_free_16(mExtents); + delete [] mExtents; mExtents = NULL; freeData(); @@ -5340,17 +5338,17 @@ LLVolumeFace::~LLVolumeFace() void LLVolumeFace::freeData() { - ll_aligned_free_16(mPositions); + delete [] mPositions; mPositions = NULL; - ll_aligned_free_16(mNormals); + delete [] mNormals; mNormals = NULL; - ll_aligned_free_16(mTexCoords); + delete [] mTexCoords; mTexCoords = NULL; - ll_aligned_free_16(mIndices); + delete [] mIndices; mIndices = NULL; - ll_aligned_free_16(mBinormals); + delete [] mBinormals; mBinormals = NULL; - ll_aligned_free_16(mWeights); + delete [] mWeights; mWeights = NULL; delete mOctree; @@ -5463,56 +5461,73 @@ void LLVolumeFace::optimize(F32 angle_cutoff) } -void LLVolumeFace::createOctree() +void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVector4a& size) { - LLVector4a center; - LLVector4a size; - center.splat(0.f); - size.splat(1.f); + if (mOctree) + { + return; + } mOctree = new LLOctreeRoot<LLVolumeTriangle>(center, size, NULL); new LLVolumeOctreeListener(mOctree); for (U32 i = 0; i < mNumIndices; i+= 3) - { + { //for each triangle LLPointer<LLVolumeTriangle> tri = new LLVolumeTriangle(); const LLVector4a& v0 = mPositions[mIndices[i]]; const LLVector4a& v1 = mPositions[mIndices[i+1]]; const LLVector4a& v2 = mPositions[mIndices[i+2]]; + //store pointers to vertex data tri->mV[0] = &v0; tri->mV[1] = &v1; tri->mV[2] = &v2; + //store indices tri->mIndex[0] = mIndices[i]; tri->mIndex[1] = mIndices[i+1]; tri->mIndex[2] = mIndices[i+2]; + //get minimum point LLVector4a min = v0; min.setMin(min, v1); min.setMin(min, v2); + //get maximum point LLVector4a max = v0; max.setMax(max, v1); max.setMax(max, v2); + //compute center LLVector4a center; center.setAdd(min, max); center.mul(0.5f); *tri->mPositionGroup = center; + //compute "radius" LLVector4a size; size.setSub(max,min); - tri->mRadius = size.getLength3().getF32() * 0.5f; + tri->mRadius = size.getLength3().getF32() * scaler; + //insert mOctree->insert(tri); } + //remove unneeded octree layers + while (!mOctree->balance()) { } + + //calculate AABB for each node LLVolumeOctreeRebound rebound(this); rebound.traverse(mOctree); + + if (gDebugGL) + { + LLVolumeOctreeValidate validate; + validate.traverse(mOctree); + } } @@ -6146,21 +6161,21 @@ void LLVolumeFace::createBinormals() void LLVolumeFace::resizeVertices(S32 num_verts) { - ll_aligned_free_16(mPositions); - ll_aligned_free_16(mNormals); - ll_aligned_free_16(mBinormals); - ll_aligned_free_16(mTexCoords); + delete [] mPositions; + delete [] mNormals; + delete [] mBinormals; + delete [] mTexCoords; mBinormals = NULL; if (num_verts) { - mPositions = (LLVector4a*) ll_aligned_malloc_16(num_verts*16); - mNormals = (LLVector4a*) ll_aligned_malloc_16(num_verts*16); + mPositions = new LLVector4a[num_verts]; + mNormals = new LLVector4a[num_verts]; //pad texture coordinate block end to allow for QWORD reads S32 size = ((num_verts*8) + 0xF) & ~0xF; - mTexCoords = (LLVector2*) ll_aligned_malloc_16(size); + mTexCoords = new LLVector2[size/8]; } else { @@ -6184,20 +6199,20 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con S32 old_size = mNumVertices*16; //positions - LLVector4a* dst = (LLVector4a*) ll_aligned_malloc_16(new_size); + LLVector4a* dst = new LLVector4a[new_verts]; if (mPositions) { LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mPositions, old_size); - ll_aligned_free_16(mPositions); + delete [] mPositions; } mPositions = dst; //normals - dst = (LLVector4a*) ll_aligned_malloc_16(new_size); + dst = new LLVector4a[new_verts]; if (mNormals) { LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mNormals, old_size); - ll_aligned_free_16(mNormals); + delete [] mNormals; } mNormals = dst; @@ -6205,19 +6220,18 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con new_size = ((new_verts*8)+0xF) & ~0xF; old_size = ((mNumVertices*8)+0xF) & ~0xF; - dst = (LLVector4a*) ll_aligned_malloc_16(new_size); { - LLVector2* dst = (LLVector2*) ll_aligned_malloc_16(new_size); + LLVector2* dst = new LLVector2[new_size/8]; if (mTexCoords) { LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mTexCoords, old_size); - ll_aligned_free_16(mTexCoords); + delete [] mTexCoords; } + mTexCoords = dst; } - mTexCoords = (LLVector2*) dst; //just clear binormals - ll_aligned_free_16(mBinormals); + delete [] mBinormals; mBinormals = NULL; mPositions[mNumVertices] = pos; @@ -6229,26 +6243,26 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con void LLVolumeFace::allocateBinormals(S32 num_verts) { - ll_aligned_free_16(mBinormals); - mBinormals = (LLVector4a*) ll_aligned_malloc_16(num_verts*16); + delete [] mBinormals; + mBinormals = new LLVector4a[num_verts]; } void LLVolumeFace::allocateWeights(S32 num_verts) { - ll_aligned_free_16(mWeights); - mWeights = (LLVector4a*) ll_aligned_malloc_16(num_verts*16); + delete [] mWeights; + mWeights = new LLVector4a[num_verts]; } void LLVolumeFace::resizeIndices(S32 num_indices) { - ll_aligned_free_16(mIndices); - + delete [] mIndices; + if (num_indices) { //pad index block end to allow for QWORD reads S32 size = ((num_indices*2) + 0xF) & ~0xF; - mIndices = (U16*) ll_aligned_malloc_16(size); + mIndices = new U16[size/2]; } else { @@ -6266,11 +6280,11 @@ void LLVolumeFace::pushIndex(const U16& idx) S32 old_size = ((mNumIndices*2)+0xF) & ~0xF; if (new_size != old_size) { - U16* dst = (U16*) ll_aligned_malloc_16(new_size); + U16* dst = new U16[new_size/2]; if (mIndices) { LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mIndices, old_size); - ll_aligned_free_16(mIndices); + delete [] mIndices; } mIndices = dst; } @@ -6313,9 +6327,9 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat } //allocate new buffer space - LLVector4a* new_pos = (LLVector4a*) ll_aligned_malloc_16(new_count*16); - LLVector4a* new_norm = (LLVector4a*) ll_aligned_malloc_16(new_count*16); - LLVector2* new_tc = (LLVector2*) ll_aligned_malloc_16((new_count*8+0xF) & ~0xF); + LLVector4a* new_pos = new LLVector4a[new_count]; + LLVector4a* new_norm = new LLVector4a[new_count]; + LLVector2* new_tc = new LLVector2[((new_count*8+0xF) & ~0xF)/8]; if (mNumVertices > 0) @@ -6326,10 +6340,10 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat } //free old buffer space - ll_aligned_free_16(mPositions); - ll_aligned_free_16(mNormals); - ll_aligned_free_16(mTexCoords); - + delete [] mPositions; + delete [] mNormals; + delete [] mTexCoords; + //point to new buffers mPositions = new_pos; mNormals = new_norm; @@ -6379,7 +6393,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat new_count = mNumIndices + face.mNumIndices; //allocate new index buffer - U16* new_indices = (U16*) ll_aligned_malloc_16((new_count*2+0xF) & ~0xF); + U16* new_indices = new U16[((new_count*2+0xF) & ~0xF)/2]; if (mNumIndices > 0) { //copy old index buffer S32 old_size = (mNumIndices*2+0xF) & ~0xF; @@ -6387,8 +6401,8 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat } //free old index buffer - ll_aligned_free_16(mIndices); - + delete [] mIndices; + //point to new index buffer mIndices = new_indices; diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h index af28337f57..32364bd4b8 100644 --- a/indra/llmath/llvolume.h +++ b/indra/llmath/llvolume.h @@ -882,7 +882,7 @@ public: }; void optimize(F32 angle_cutoff = 2.f); - void createOctree(); + void createOctree(F32 scaler = 0.25f, const LLVector4a& center = LLVector4a(0,0,0), const LLVector4a& size = LLVector4a(0.5f,0.5f,0.5f)); enum { @@ -1044,7 +1044,7 @@ public: LLVector3 mLODScaleBias; // vector for biasing LOD based on scale void sculpt(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, S32 sculpt_level); - void copyVolumeFaces(LLVolume* volume); + void copyVolumeFaces(const LLVolume* volume); private: void sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, U8 sculpt_type); diff --git a/indra/llmath/llvolumeoctree.cpp b/indra/llmath/llvolumeoctree.cpp index 194b1faf81..cb6211f63c 100644 --- a/indra/llmath/llvolumeoctree.cpp +++ b/indra/llmath/llvolumeoctree.cpp @@ -126,8 +126,8 @@ void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode<LLVolumeTriangle> const F32* center = vl->mBounds[0].getF32(); const F32* size = vl->mBounds[1].getF32();*/ - //if (LLLineSegmentBoxIntersect(mStart.getF32(), mEnd.getF32(), vl->mBounds[0].getF32(), vl->mBounds[1].getF32())) - if (LLLineSegmentBoxIntersect(mStart, mEnd, vl->mBounds[0], vl->mBounds[1])) + //if (LLLineSegmentBoxIntersect(mStart, mEnd, vl->mBounds[0], vl->mBounds[1])) + if (LLLineSegmentBoxIntersect(mStart.getF32ptr(), mEnd.getF32ptr(), vl->mBounds[0].getF32ptr(), vl->mBounds[1].getF32ptr())) { node->accept(this); for (S32 i = 0; i < node->getChildCount(); ++i) @@ -206,3 +206,60 @@ const F32& LLVolumeTriangle::getBinRadius() const } +//TEST CODE + +void LLVolumeOctreeValidate::visit(const LLOctreeNode<LLVolumeTriangle>* branch) +{ + LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0); + + //make sure bounds matches extents + LLVector4a& min = node->mExtents[0]; + LLVector4a& max = node->mExtents[1]; + + LLVector4a& center = node->mBounds[0]; + LLVector4a& size = node->mBounds[1]; + + LLVector4a test_min, test_max; + test_min.setSub(center, size); + test_max.setAdd(center, size); + + if (!test_min.equals3(min, 0.001f) || + !test_max.equals3(max, 0.001f)) + { + llerrs << "Bad bounding box data found." << llendl; + } + + test_min.sub(LLVector4a(0.001f)); + test_max.add(LLVector4a(0.001f)); + + for (U32 i = 0; i < branch->getChildCount(); ++i) + { + LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); + + //make sure all children fit inside this node + if (child->mExtents[0].lessThan(test_min).areAnySet(LLVector4Logical::MASK_XYZ) || + child->mExtents[1].greaterThan(test_max).areAnySet(LLVector4Logical::MASK_XYZ)) + { + llerrs << "Child protrudes from bounding box." << llendl; + } + } + + //children fit, check data + for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = branch->getData().begin(); + iter != branch->getData().end(); ++iter) + { + const LLVolumeTriangle* tri = *iter; + + //validate triangle + for (U32 i = 0; i < 3; i++) + { + if (tri->mV[i]->greaterThan(test_max).areAnySet(LLVector4Logical::MASK_XYZ) || + tri->mV[i]->lessThan(test_min).areAnySet(LLVector4Logical::MASK_XYZ)) + { + llerrs << "Triangle protrudes from node." << llendl; + } + } + } +} + + diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h index 0031626498..0a02a2c597 100644 --- a/indra/llmath/llvolumeoctree.h +++ b/indra/llmath/llvolumeoctree.h @@ -88,8 +88,6 @@ public: F32* mClosestT; bool mHitFace; - LLOctreeTriangleRayIntersect() { }; - LLOctreeTriangleRayIntersect(const LLVector4a& start, const LLVector4a& dir, const LLVolumeFace* face, F32* closest_t, LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal); @@ -134,5 +132,9 @@ public: virtual const F32& getBinRadius() const; }; +class LLVolumeOctreeValidate : public LLOctreeTraveler<LLVolumeTriangle> +{ + virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch); +}; #endif diff --git a/indra/newview/CMakeLists.txt b/indra/newview/CMakeLists.txt index 73f752b218..0f2d461511 100644 --- a/indra/newview/CMakeLists.txt +++ b/indra/newview/CMakeLists.txt @@ -1422,9 +1422,7 @@ if (WINDOWS) set_target_properties(${VIEWER_BINARY_NAME} PROPERTIES - # *TODO -reenable this once we get server usage sorted out - #LINK_FLAGS "/debug /NODEFAULTLIB:LIBCMT /SUBSYSTEM:WINDOWS /INCLUDE:\"__tcmalloc\"" - LINK_FLAGS "/debug /NODEFAULTLIB:LIBCMT /SUBSYSTEM:WINDOWS" + LINK_FLAGS "/debug /NODEFAULTLIB:LIBCMT /SUBSYSTEM:WINDOWS /INCLUDE:__tcmalloc" LINK_FLAGS_DEBUG "/NODEFAULTLIB:\"LIBCMT;LIBCMTD;MSVCRT\" /INCREMENTAL:NO" LINK_FLAGS_RELEASE ${release_flags} ) @@ -1640,6 +1638,7 @@ if (WINDOWS) endif (WINDOWS) target_link_libraries(${VIEWER_BINARY_NAME} + ${GOOGLE_PERFTOOLS_LIBRARIES} ${LLAUDIO_LIBRARIES} ${LLCHARACTER_LIBRARIES} ${LLIMAGE_LIBRARIES} @@ -1676,7 +1675,6 @@ target_link_libraries(${VIEWER_BINARY_NAME} ${OPENSSL_LIBRARIES} ${CRYPTO_LIBRARIES} ${LLLOGIN_LIBRARIES} - ${GOOGLE_PERFTOOLS_LIBRARIES} ${LLCONVEXDECOMP_LIBRARY} ) diff --git a/indra/newview/lldrawable.cpp b/indra/newview/lldrawable.cpp index 68f52e04bc..efbb62011c 100644 --- a/indra/newview/lldrawable.cpp +++ b/indra/newview/lldrawable.cpp @@ -189,6 +189,11 @@ LLVOVolume* LLDrawable::getVOVolume() const } } +const LLMatrix4& LLDrawable::getRenderMatrix() const +{ + return isRoot() ? getWorldMatrix() : getParent()->getWorldMatrix(); +} + BOOL LLDrawable::isLight() const { LLViewerObject* objectp = mVObjp; @@ -713,8 +718,7 @@ void LLDrawable::updateDistance(LLCamera& camera, bool force_update) LLVOVolume* volume = getVOVolume(); if (volume) { - volume->updateRelativeXform(); - pos = volume->getRelativeXform().getTranslation(); + pos.set(getPositionGroup().getF32ptr()); if (isStatic()) { pos += volume->getRegion()->getOriginAgent(); diff --git a/indra/newview/lldrawable.h b/indra/newview/lldrawable.h index fc90bc57c4..a17e2172aa 100644 --- a/indra/newview/lldrawable.h +++ b/indra/newview/lldrawable.h @@ -101,7 +101,7 @@ public: LLVOVolume* getVOVolume() const; // cast mVObjp tp LLVOVolume if OK const LLMatrix4& getWorldMatrix() const { return mXform.getWorldMatrix(); } - const LLMatrix4& getRenderMatrix() const { return isRoot() ? getWorldMatrix() : getParent()->getWorldMatrix(); } + const LLMatrix4& getRenderMatrix() const; void setPosition(LLVector3 v) const { } const LLVector3& getPosition() const { return mXform.getPosition(); } const LLVector3& getWorldPosition() const { return mXform.getPositionW(); } @@ -250,35 +250,37 @@ public: typedef enum e_drawable_flags { - IN_REBUILD_Q1 = 0x00000002, - IN_REBUILD_Q2 = 0x00000004, - IN_LIGHT_Q = 0x00000008, - EARLY_MOVE = 0x00000010, - MOVE_UNDAMPED = 0x00000020, - ON_MOVE_LIST = 0x00000040, - USE_BACKLIGHT = 0x00000080, - UV = 0x00000100, - UNLIT = 0x00000200, - LIGHT = 0x00000400, - LIGHTING_BUILT = 0x00000800, - REBUILD_VOLUME = 0x00001000, //volume changed LOD or parameters, or vertex buffer changed - REBUILD_TCOORD = 0x00002000, //texture coordinates changed - REBUILD_COLOR = 0x00004000, //color changed - REBUILD_POSITION= 0x00010000, //vertex positions/normals changed + IN_REBUILD_Q1 = 0x00000001, + IN_REBUILD_Q2 = 0x00000002, + IN_LIGHT_Q = 0x00000004, + EARLY_MOVE = 0x00000008, + MOVE_UNDAMPED = 0x00000010, + ON_MOVE_LIST = 0x00000020, + USE_BACKLIGHT = 0x00000040, + UV = 0x00000080, + UNLIT = 0x00000100, + LIGHT = 0x00000200, + LIGHTING_BUILT = 0x00000400, + REBUILD_VOLUME = 0x00000800, //volume changed LOD or parameters, or vertex buffer changed + REBUILD_TCOORD = 0x00001000, //texture coordinates changed + REBUILD_COLOR = 0x00002000, //color changed + REBUILD_POSITION= 0x00004000, //vertex positions/normals changed REBUILD_GEOMETRY= REBUILD_POSITION|REBUILD_TCOORD|REBUILD_COLOR, REBUILD_MATERIAL= REBUILD_TCOORD|REBUILD_COLOR, REBUILD_ALL = REBUILD_GEOMETRY|REBUILD_VOLUME, - ON_SHIFT_LIST = 0x00100000, - BLOCKER = 0x00400000, - ACTIVE = 0x00800000, - DEAD = 0x01000000, - INVISIBLE = 0x02000000, // stay invisible until flag is cleared - NEARBY_LIGHT = 0x04000000, // In gPipeline.mNearbyLightSet - BUILT = 0x08000000, - FORCE_INVISIBLE = 0x10000000, // stay invis until CLEAR_INVISIBLE is set (set of orphaned) - CLEAR_INVISIBLE = 0x20000000, // clear FORCE_INVISIBLE next draw frame - REBUILD_SHADOW = 0x40000000, - HAS_ALPHA = 0x80000000, + REBUILD_RIGGED = 0x00008000, + ON_SHIFT_LIST = 0x00010000, + BLOCKER = 0x00020000, + ACTIVE = 0x00040000, + DEAD = 0x00080000, + INVISIBLE = 0x00100000, // stay invisible until flag is cleared + NEARBY_LIGHT = 0x00200000, // In gPipeline.mNearbyLightSet + BUILT = 0x00400000, + FORCE_INVISIBLE = 0x00800000, // stay invis until CLEAR_INVISIBLE is set (set of orphaned) + CLEAR_INVISIBLE = 0x01000000, // clear FORCE_INVISIBLE next draw frame + REBUILD_SHADOW = 0x02000000, + HAS_ALPHA = 0x04000000, + RIGGED = 0x08000000, } EDrawableFlags; LLXformMatrix mXform; diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index ccc060f3fa..5cf6082f12 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -1325,6 +1325,12 @@ void LLDrawPoolAvatar::renderAvatars(LLVOAvatar* single_avatar, S32 pass) #if LL_MESH_ENABLED void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, LLFace* face, const LLMeshSkinInfo* skin, LLVolume* volume, const LLVolumeFace& vol_face) { + LLVector4a* weight = vol_face.mWeights; + if (!weight) + { + return; + } + LLVertexBuffer* buffer = face->mVertexBuffer; U32 data_mask = 0; @@ -1403,8 +1409,6 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, LLFace* } } - LLVector4a* weight = vol_face.mWeights; - LLMatrix4a bind_shape_matrix; bind_shape_matrix.loadu(skin->mBindShapeMatrix); @@ -1422,7 +1426,7 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, LLFace* { F32 w = weight[j][k]; - idx[k] = (S32) floorf(w); + idx[k] = llclamp((S32) floorf(w), 0, 63); wght[k] = w - floorf(w); scale += wght[k]; } diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 7c24eb77cd..3fa60e9f1e 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -556,8 +556,36 @@ void LLFace::renderSelected(LLViewerTexture *imagep, const LLColor4& color) } glColor4fv(color.mV); - mVertexBuffer->setBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0); - mVertexBuffer->draw(LLRender::TRIANGLES, mIndicesCount, mIndicesIndex); + + if (mDrawablep->isState(LLDrawable::RIGGED)) + { + LLVOVolume* volume = mDrawablep->getVOVolume(); + if (volume) + { + LLRiggedVolume* rigged = volume->getRiggedVolume(); + if (rigged) + { + LLGLEnable offset(GL_POLYGON_OFFSET_FILL); + glPolygonOffset(-1.f, -1.f); + glMultMatrixf((F32*) volume->getRelativeXform().mMatrix); + const LLVolumeFace& vol_face = rigged->getVolumeFace(getTEOffset()); + LLVertexBuffer::unbind(); + glVertexPointer(3, GL_FLOAT, 16, vol_face.mPositions); + if (vol_face.mTexCoords) + { + glEnableClientState(GL_TEXTURE_COORD_ARRAY); + glTexCoordPointer(2, GL_FLOAT, 8, vol_face.mTexCoords); + } + glDrawElements(GL_TRIANGLES, vol_face.mNumIndices, GL_UNSIGNED_SHORT, vol_face.mIndices); + glDisableClientState(GL_TEXTURE_COORD_ARRAY); + } + } + } + else + { + mVertexBuffer->draw(LLRender::TRIANGLES, mIndicesCount, mIndicesIndex); + mVertexBuffer->setBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_TEXCOORD0); + } gGL.popMatrix(); } @@ -720,7 +748,7 @@ BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f, LLMemType mt1(LLMemType::MTYPE_DRAWABLE); //get bounding box - if (mDrawablep->isState(LLDrawable::REBUILD_VOLUME | LLDrawable::REBUILD_POSITION)) + if (mDrawablep->isState(LLDrawable::REBUILD_VOLUME | LLDrawable::REBUILD_POSITION | LLDrawable::REBUILD_RIGGED)) { //VECTORIZE THIS LLMatrix4a mat_vert; diff --git a/indra/newview/llfloatertools.cpp b/indra/newview/llfloatertools.cpp index fa5d9b0892..653f838779 100644 --- a/indra/newview/llfloatertools.cpp +++ b/indra/newview/llfloatertools.cpp @@ -1016,10 +1016,12 @@ S32 LLFloaterTools::calcRenderCost() LLSelectNode *select_node = *selection_iter; if (select_node) { - LLVOVolume *viewer_volume = (LLVOVolume*)select_node->getObject(); - if (viewer_volume) + LLViewerObject *vobj = select_node->getObject(); + if (vobj->getVolume()) { - cost += viewer_volume->getRenderCost(textures); + LLVOVolume* volume = (LLVOVolume*) vobj; + + cost += volume->getRenderCost(textures); for (LLVOVolume::texture_cost_t::iterator iter = textures.begin(); iter != textures.end(); ++iter) { // add the cost of each individual texture in the linkset diff --git a/indra/newview/llselectmgr.cpp b/indra/newview/llselectmgr.cpp index ff2929695d..21f7e780e2 100644 --- a/indra/newview/llselectmgr.cpp +++ b/indra/newview/llselectmgr.cpp @@ -5395,6 +5395,42 @@ BOOL LLSelectNode::allowOperationOnNode(PermissionBit op, U64 group_proxy_power) return (mPermissions->allowOperationBy(op, proxy_agent_id, group_id)); } + +//helper function for pushing relevant vertices from drawable to GL +void pushWireframe(LLDrawable* drawable) +{ + if (drawable->isState(LLDrawable::RIGGED)) + { //render straight from rigged volume if this is a rigged attachment + LLVOVolume* vobj = drawable->getVOVolume(); + if (vobj) + { + vobj->updateRiggedVolume(); + LLRiggedVolume* rigged_volume = vobj->getRiggedVolume(); + if (rigged_volume) + { + LLVertexBuffer::unbind(); + gGL.pushMatrix(); + glMultMatrixf((F32*) vobj->getRelativeXform().mMatrix); + for (S32 i = 0; i < rigged_volume->getNumVolumeFaces(); ++i) + { + const LLVolumeFace& face = rigged_volume->getVolumeFace(i); + glVertexPointer(3, GL_FLOAT, 16, face.mPositions); + glDrawElements(GL_TRIANGLES, face.mNumIndices, GL_UNSIGNED_SHORT, face.mIndices); + } + gGL.popMatrix(); + } + } + } + else + { + for (S32 i = 0; i < drawable->getNumFaces(); ++i) + { + LLFace* face = drawable->getFace(i); + pushVerts(face, LLVertexBuffer::MAP_VERTEX); + } + } +} + void LLSelectNode::renderOneWireframe(const LLColor4& color) { LLViewerObject* objectp = getObject(); @@ -5442,11 +5478,7 @@ void LLSelectNode::renderOneWireframe(const LLColor4& color) gGL.setAlphaRejectSettings(LLRender::CF_DEFAULT); { glColor4f(color.mV[VRED], color.mV[VGREEN], color.mV[VBLUE], 0.4f); - for (S32 i = 0; i < drawable->getNumFaces(); ++i) - { - LLFace* face = drawable->getFace(i); - pushVerts(face, LLVertexBuffer::MAP_VERTEX); - } + pushWireframe(drawable); } } @@ -5455,13 +5487,9 @@ void LLSelectNode::renderOneWireframe(const LLColor4& color) glColor4f(color.mV[VRED]*2, color.mV[VGREEN]*2, color.mV[VBLUE]*2, LLSelectMgr::sHighlightAlpha*2); LLGLEnable offset(GL_POLYGON_OFFSET_LINE); - glPolygonOffset(3.f, 2.f); + glPolygonOffset(3.f, 3.f); glLineWidth(3.f); - for (S32 i = 0; i < drawable->getNumFaces(); ++i) - { - LLFace* face = drawable->getFace(i); - pushVerts(face, LLVertexBuffer::MAP_VERTEX); - } + pushWireframe(drawable); glLineWidth(1.f); glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); gGL.popMatrix(); diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index 425fa42339..0f9f35dc57 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -3035,11 +3035,11 @@ class LLRenderOctreeRaycast : public LLOctreeTriangleRayIntersect { public: - LLRenderOctreeRaycast(const LLVector3& start, const LLVector3& end) + + LLRenderOctreeRaycast(const LLVector4a& start, const LLVector4a& dir, F32* closest_t) + : LLOctreeTriangleRayIntersect(start, dir, NULL, closest_t, NULL, NULL, NULL, NULL) { - mStart.load3(start.mV); - mEnd.load3(end.mV); - mDir.setSub(mEnd, mStart); + } void visit(const LLOctreeNode<LLVolumeTriangle>* branch) @@ -3047,9 +3047,22 @@ public: LLVolumeOctreeListener* vl = (LLVolumeOctreeListener*) branch->getListener(0); LLVector3 center, size; - center.set(vl->mBounds[0].getF32ptr()); - size.set(vl->mBounds[1].getF32ptr()); + + if (branch->getData().empty()) + { + gGL.color3f(1.f,0.2f,0.f); + center.set(branch->getCenter().getF32ptr()); + size.set(branch->getSize().getF32ptr()); + } + else + { + gGL.color3f(0.75f, 1.f, 0.f); + center.set(vl->mBounds[0].getF32ptr()); + size.set(vl->mBounds[1].getF32ptr()); + } + drawBoxOutline(center, size); + for (U32 i = 0; i < 2; i++) { LLGLDepthTest depth(GL_TRUE, GL_FALSE, i == 1 ? GL_LEQUAL : GL_GREATER); @@ -3061,9 +3074,14 @@ public: else { gGL.color4f(0,0.5f,0.5f, 0.25f); + drawBoxOutline(center, size); + } + + if (i == 1) + { + gGL.flush(); + glLineWidth(3.f); } - - drawBoxOutline(center, size); gGL.begin(LLRender::TRIANGLES); for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = branch->getData().begin(); @@ -3077,6 +3095,12 @@ public: gGL.vertex3fv(tri->mV[2]->getF32ptr()); } gGL.end(); + + if (i == 1) + { + gGL.flush(); + glLineWidth(1.f); + } } } }; @@ -3096,26 +3120,63 @@ void renderRaycast(LLDrawable* drawablep) LLVOVolume* vobj = drawablep->getVOVolume(); LLVolume* volume = vobj->getVolume(); - for (S32 i = 0; i < volume->getNumVolumeFaces(); ++i) + + bool transform = true; + if (drawablep->isState(LLDrawable::RIGGED)) { - const LLVolumeFace& face = volume->getVolumeFace(i); - if (!face.mOctree) + volume = vobj->getRiggedVolume(); + transform = false; + } + + if (volume) + { + for (S32 i = 0; i < volume->getNumVolumeFaces(); ++i) { - ((LLVolumeFace*) &face)->createOctree(); - } + const LLVolumeFace& face = volume->getVolumeFace(i); + if (!face.mOctree) + { + ((LLVolumeFace*) &face)->createOctree(); + } - gGL.pushMatrix(); - glMultMatrixf((F32*) vobj->getRelativeXform().mMatrix); - LLVector3 start, end; - start = vobj->agentPositionToVolume(gDebugRaycastStart); - end = vobj->agentPositionToVolume(gDebugRaycastEnd); + gGL.pushMatrix(); + glMultMatrixf((F32*) vobj->getRelativeXform().mMatrix); - LLRenderOctreeRaycast render(start, end); - gGL.flush(); - glPolygonMode(GL_FRONT_AND_BACK, GL_LINE); - render.traverse(face.mOctree); - gGL.popMatrix(); - glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); + LLVector3 start, end; + if (transform) + { + start = vobj->agentPositionToVolume(gDebugRaycastStart); + end = vobj->agentPositionToVolume(gDebugRaycastEnd); + } + else + { + start = gDebugRaycastStart; + end = gDebugRaycastEnd; + } + + LLVector4a starta, enda; + starta.load3(start.mV); + enda.load3(end.mV); + LLVector4a dir; + dir.setSub(enda, starta); + + F32 t = 1.f; + + LLRenderOctreeRaycast render(starta, dir, &t); + gGL.flush(); + glPolygonMode(GL_FRONT_AND_BACK, GL_LINE); + + { + //render face positions + LLVertexBuffer::unbind(); + glColor4f(0,1,1,0.5f); + glVertexPointer(3, GL_FLOAT, sizeof(LLVector4a), face.mPositions); + glDrawElements(GL_TRIANGLES, face.mNumIndices, GL_UNSIGNED_SHORT, face.mIndices); + } + + render.traverse(face.mOctree); + gGL.popMatrix(); + glPolygonMode(GL_FRONT_AND_BACK, GL_FILL); + } } } else if (drawablep->isAvatar()) diff --git a/indra/newview/llviewerobject.cpp b/indra/newview/llviewerobject.cpp index fd2212f25d..4c1a1958e8 100644 --- a/indra/newview/llviewerobject.cpp +++ b/indra/newview/llviewerobject.cpp @@ -3279,6 +3279,15 @@ const LLVector3 LLViewerObject::getPositionEdit() const const LLVector3 LLViewerObject::getRenderPosition() const { + if (mDrawable.notNull() && mDrawable->isState(LLDrawable::RIGGED)) + { + LLVOAvatar* avatar = getAvatar(); + if (avatar) + { + return avatar->getPositionAgent(); + } + } + if (mDrawable.isNull() || mDrawable->getGeneration() < 0) { return getPositionAgent(); @@ -3297,6 +3306,11 @@ const LLVector3 LLViewerObject::getPivotPositionAgent() const const LLQuaternion LLViewerObject::getRenderRotation() const { LLQuaternion ret; + if (mDrawable.notNull() && mDrawable->isState(LLDrawable::RIGGED)) + { + return ret; + } + if (mDrawable.isNull() || mDrawable->isStatic()) { ret = getRotationEdit(); diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index 5448e7dcd9..f09ce5b363 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -41,9 +41,11 @@ #include "llviewercontrol.h" #include "lldir.h" #include "llflexibleobject.h" +#include "llfloatertools.h" #include "llmaterialtable.h" #include "llprimitive.h" #include "llvolume.h" +#include "llvolumeoctree.h" #include "llvolumemgr.h" #include "llvolumemessage.h" #include "material_codes.h" @@ -70,6 +72,7 @@ #include "llselectmgr.h" #include "pipeline.h" #include "llsdutil.h" +#include "llmatrix4a.h" #include "llmediaentry.h" #include "llmediadataclient.h" #include "llmeshrepository.h" @@ -1355,7 +1358,14 @@ BOOL LLVOVolume::genBBoxes(BOOL force_global) LLVector4a min,max; - BOOL rebuild = mDrawable->isState(LLDrawable::REBUILD_VOLUME | LLDrawable::REBUILD_POSITION); + BOOL rebuild = mDrawable->isState(LLDrawable::REBUILD_VOLUME | LLDrawable::REBUILD_POSITION | LLDrawable::REBUILD_RIGGED); + + bool rigged = false; + LLVolume* volume = mRiggedVolume; + if (!volume) + { + volume = getVolume(); + } for (S32 i = 0; i < getVolume()->getNumVolumeFaces(); i++) { @@ -1364,7 +1374,7 @@ BOOL LLVOVolume::genBBoxes(BOOL force_global) { continue; } - res &= face->genVolumeBBoxes(*getVolume(), i, + res &= face->genVolumeBBoxes(*volume, i, mRelativeXform, mRelativeXformInvTrans, (mVolumeImpl && mVolumeImpl->isVolumeGlobal()) || force_global); @@ -1415,7 +1425,21 @@ void LLVOVolume::updateRelativeXform() LLDrawable* drawable = mDrawable; - if (drawable->isActive()) + if (drawable->isState(LLDrawable::RIGGED) && mRiggedVolume.notNull()) + { //rigged volume (which is in agent space) is used for generating bounding boxes etc + //inverse of render matrix should go to partition space + mRelativeXform = getRenderMatrix(); + + F32* dst = (F32*) mRelativeXformInvTrans.mMatrix; + F32* src = (F32*) mRelativeXform.mMatrix; + dst[0] = src[0]; dst[1] = src[1]; dst[2] = src[2]; + dst[3] = src[4]; dst[4] = src[5]; dst[5] = src[6]; + dst[6] = src[8]; dst[7] = src[9]; dst[8] = src[10]; + + mRelativeXform.invert(); + mRelativeXformInvTrans.transpose(); + } + else if (drawable->isActive()) { // setup relative transforms LLQuaternion delta_rot; @@ -1497,11 +1521,22 @@ void LLVOVolume::updateRelativeXform() static LLFastTimer::DeclareTimer FTM_GEN_FLEX("Generate Flexies"); static LLFastTimer::DeclareTimer FTM_UPDATE_PRIMITIVES("Update Primitives"); +static LLFastTimer::DeclareTimer FTM_UPDATE_RIGGED_VOLUME("Update Rigged"); BOOL LLVOVolume::updateGeometry(LLDrawable *drawable) { LLFastTimer t(FTM_UPDATE_PRIMITIVES); + if (mDrawable->isState(LLDrawable::REBUILD_RIGGED)) + { + { + LLFastTimer t(FTM_UPDATE_RIGGED_VOLUME); + updateRiggedVolume(); + } + genBBoxes(FALSE); + mDrawable->clearState(LLDrawable::REBUILD_RIGGED); + } + if (mVolumeImpl != NULL) { BOOL res; @@ -1597,7 +1632,7 @@ BOOL LLVOVolume::updateGeometry(LLDrawable *drawable) { LLPipeline::sCompiles++; } - + mVolumeChanged = FALSE; mLODChanged = FALSE; mSculptChanged = FALSE; @@ -2724,6 +2759,11 @@ BOOL LLVOVolume::isVolumeGlobal() const { return mVolumeImpl->isVolumeGlobal() ? TRUE : FALSE; } + else if (mRiggedVolume.notNull()) + { + return TRUE; + } + return FALSE; } @@ -3390,12 +3430,37 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e BOOL ret = FALSE; LLVolume* volume = getVolume(); + + bool transform = true; + + if (mDrawable->isState(LLDrawable::RIGGED)) + { + if (LLFloater::isVisible(gFloaterTools) && getAvatar()->isSelf()) + { + gPipeline.markRebuild(mDrawable, LLDrawable::REBUILD_RIGGED, TRUE); + volume = mRiggedVolume; + transform = false; + } + else + { //cannot pick rigged attachments on other avatars or when not in build mode + return FALSE; + } + } + if (volume) { LLVector3 v_start, v_end, v_dir; - v_start = agentPositionToVolume(start); - v_end = agentPositionToVolume(end); + if (transform) + { + v_start = agentPositionToVolume(start); + v_end = agentPositionToVolume(end); + } + else + { + v_start = start; + v_end = end; + } LLVector3 p; LLVector3 n; @@ -3455,18 +3520,40 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e if (intersection != NULL) { - *intersection = volumePositionToAgent(p); // must map back to agent space + if (transform) + { + *intersection = volumePositionToAgent(p); // must map back to agent space + } + else + { + *intersection = p; + } } if (normal != NULL) { - *normal = volumeDirectionToAgent(n); + if (transform) + { + *normal = volumeDirectionToAgent(n); + } + else + { + *normal = n; + } + (*normal).normVec(); } if (bi_normal != NULL) { - *bi_normal = volumeDirectionToAgent(bn); + if (transform) + { + *bi_normal = volumeDirectionToAgent(bn); + } + else + { + *bi_normal = bn; + } (*bi_normal).normVec(); } @@ -3484,6 +3571,201 @@ BOOL LLVOVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& e return ret; } +bool LLVOVolume::treatAsRigged() +{ + return LLFloater::isVisible(gFloaterTools) && + isAttachment() && + getAvatar() && + getAvatar()->isSelf() && + mDrawable.notNull() && + mDrawable->isState(LLDrawable::RIGGED); +} + +LLRiggedVolume* LLVOVolume::getRiggedVolume() +{ + return mRiggedVolume; +} + +void LLVOVolume::clearRiggedVolume() +{ + if (mRiggedVolume.notNull()) + { + mRiggedVolume = NULL; + updateRelativeXform(); + } +} + +void LLVOVolume::updateRiggedVolume() +{ + //Update mRiggedVolume to match current animation frame of avatar. + //Also update position/size in octree. + + if (!treatAsRigged()) + { + clearRiggedVolume(); + + return; + } + + LLVolume* volume = getVolume(); + + const LLMeshSkinInfo* skin = gMeshRepo.getSkinInfo(volume->getParams().getSculptID()); + + if (!skin) + { + clearRiggedVolume(); + return; + } + + LLVOAvatar* avatar = getAvatar(); + + if (!avatar) + { + clearRiggedVolume(); + return; + } + + if (!mRiggedVolume) + { + LLVolumeParams p; + mRiggedVolume = new LLRiggedVolume(p); + updateRelativeXform(); + } + + mRiggedVolume->update(skin, avatar, volume); + +} + +static LLFastTimer::DeclareTimer FTM_SKIN_RIGGED("Skin"); +static LLFastTimer::DeclareTimer FTM_RIGGED_OCTREE("Octree"); + +void LLRiggedVolume::update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, const LLVolume* volume) +{ + bool copy = false; + if (volume->getNumVolumeFaces() != getNumVolumeFaces()) + { + copy = true; + } + + for (S32 i = 0; i < volume->getNumVolumeFaces() && !copy; ++i) + { + const LLVolumeFace& src_face = volume->getVolumeFace(i); + const LLVolumeFace& dst_face = getVolumeFace(i); + + if (src_face.mNumIndices != dst_face.mNumIndices || + src_face.mNumVertices != dst_face.mNumVertices) + { + copy = true; + } + } + + if (copy) + { + copyVolumeFaces(volume); + } + + //build matrix palette + LLMatrix4a mp[64]; + LLMatrix4* mat = (LLMatrix4*) mp; + + for (U32 j = 0; j < skin->mJointNames.size(); ++j) + { + LLJoint* joint = avatar->getJoint(skin->mJointNames[j]); + if (joint) + { + mat[j] = skin->mInvBindMatrix[j]; + mat[j] *= joint->getWorldMatrix(); + } + } + + for (S32 i = 0; i < volume->getNumVolumeFaces(); ++i) + { + const LLVolumeFace& vol_face = volume->getVolumeFace(i); + + LLVolumeFace& dst_face = mVolumeFaces[i]; + + LLVector4a* weight = vol_face.mWeights; + + LLMatrix4a bind_shape_matrix; + bind_shape_matrix.loadu(skin->mBindShapeMatrix); + + LLVector4a* pos = dst_face.mPositions; + + { + LLFastTimer t(FTM_SKIN_RIGGED); + + for (U32 j = 0; j < dst_face.mNumVertices; ++j) + { + LLMatrix4a final_mat; + final_mat.clear(); + + S32 idx[4]; + + LLVector4 wght; + + F32 scale = 0.f; + for (U32 k = 0; k < 4; k++) + { + F32 w = weight[j][k]; + + idx[k] = (S32) floorf(w); + wght[k] = w - floorf(w); + scale += wght[k]; + } + + wght *= 1.f/scale; + + for (U32 k = 0; k < 4; k++) + { + F32 w = wght[k]; + + LLMatrix4a src; + src.setMul(mp[idx[k]], w); + + final_mat.add(src); + } + + + LLVector4a& v = vol_face.mPositions[j]; + LLVector4a t; + LLVector4a dst; + bind_shape_matrix.affineTransform(v, t); + final_mat.affineTransform(t, dst); + pos[j] = dst; + } + + //update bounding box + LLVector4a& min = dst_face.mExtents[0]; + LLVector4a& max = dst_face.mExtents[1]; + + min = pos[0]; + max = pos[1]; + + for (U32 j = 1; j < dst_face.mNumVertices; ++j) + { + min.setMin(min, pos[j]); + max.setMax(max, pos[j]); + } + + dst_face.mCenter->setAdd(dst_face.mExtents[0], dst_face.mExtents[1]); + dst_face.mCenter->mul(0.5f); + + } + + { + LLFastTimer t(FTM_RIGGED_OCTREE); + delete dst_face.mOctree; + dst_face.mOctree = NULL; + + LLVector4a size; + size.setSub(dst_face.mExtents[1], dst_face.mExtents[0]); + size.splat(size.getLength3().getF32()*0.5f); + + dst_face.createOctree(1.f); + } + } +} + U32 LLVOVolume::getPartitionType() const { if (isHUDAttachment()) @@ -3732,6 +4014,8 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) bool bake_sunlight = LLPipeline::sBakeSunlight && drawablep->isStatic(); + bool is_rigged = false; + //for each face for (S32 i = 0; i < drawablep->getNumFaces(); i++) { @@ -3747,8 +4031,9 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) facep->mVertexBuffer = NULL; facep->mLastVertexBuffer = NULL; } - + facep->setState(LLFace::RIGGED); + is_rigged = true; //get drawpool of avatar with rigged face LLDrawPoolAvatar* pool = get_avatar_drawpool(vobj); @@ -3958,6 +4243,15 @@ void LLVolumeGeometryManager::rebuildGeom(LLSpatialGroup* group) facep->mLastVertexBuffer = NULL; } } + + if (is_rigged) + { + drawablep->setState(LLDrawable::RIGGED); + } + else + { + drawablep->clearState(LLDrawable::RIGGED); + } } group->mBufferUsage = useage; diff --git a/indra/newview/llvovolume.h b/indra/newview/llvovolume.h index eeb98726c9..f058710a27 100644 --- a/indra/newview/llvovolume.h +++ b/indra/newview/llvovolume.h @@ -46,6 +46,8 @@ class LLDrawPool; class LLSelectNode; class LLObjectMediaDataClient; class LLObjectMediaNavigateClient; +class LLVOAvatar; +class LLMeshSkinInfo; typedef std::vector<viewer_media_t> media_list_t; @@ -54,6 +56,18 @@ enum LLVolumeInterfaceType INTERFACE_FLEXIBLE = 1, }; + +class LLRiggedVolume : public LLVolume +{ +public: + LLRiggedVolume(const LLVolumeParams& params) + : LLVolume(params, 0.f) + { + } + + void update(const LLMeshSkinInfo* skin, LLVOAvatar* avatar, const LLVolume* src_volume); +}; + // Base class for implementations of the volume - Primitive, Flexible Object, etc. class LLVolumeInterface { @@ -289,6 +303,21 @@ public: void removeMDCImpl() { --mMDCImplCount; } S32 getMDCImplCount() { return mMDCImplCount; } + + //rigged volume update (for raycasting) + void updateRiggedVolume(); + LLRiggedVolume* getRiggedVolume(); + + //returns true if volume should be treated as a rigged volume + // - Build tools are open + // - object is an attachment + // - object is attached to self + // - object is rendered as rigged + bool treatAsRigged(); + + //clear out rigged volume and revert back to non-rigged state for picking/LOD/distance updates + void clearRiggedVolume(); + protected: S32 computeLODDetail(F32 distance, F32 radius); BOOL calcLOD(); @@ -322,6 +351,9 @@ private: S32 mLastFetchedMediaVersion; // as fetched from the server, starts as -1 S32 mIndexInTex; S32 mMDCImplCount; + + LLPointer<LLRiggedVolume> mRiggedVolume; + // statics public: static F32 sLODSlopDistanceFactor;// Changing this to zero, effectively disables the LOD transition slop diff --git a/install.xml b/install.xml index 18d061e81f..e896956c1b 100644 --- a/install.xml +++ b/install.xml @@ -630,12 +630,12 @@ </map> </map> </map> - <key>google-perftools</key> + <key>tcmalloc</key> <map> <key>copyright</key> <string>Copyright (c) 2005, Google Inc.</string> <key>description</key> - <string>Heap performance and validity checking tools from google. Includes TCMalloc, heap-checker, heap-profiler and cpu-profiler.</string> + <string>High performance memory allocator.</string> <key>license</key> <string>bsd</string> <key>packages</key> @@ -643,9 +643,9 @@ <key>windows</key> <map> <key>md5sum</key> - <string>32dba32ddd460a08e082898ebba6315c</string> + <string>8b11a7d3caee8a4ba70da41d6a5ee131</string> <key>url</key> - <uri>http://s3.amazonaws.com/viewer-source-downloads/install_pkgs/google-perftools-1.0-windows-20090406.tar.bz2</uri> + <uri>http://s3.amazonaws.com/viewer-source-downloads/install_pkgs/tcmalloc-1.6-windows-20100916b.tar.bz2</uri> </map> </map> </map> |