diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h index c4fc3d5e6..f49c657c0 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -299,7 +299,7 @@ public: { if (data == NULL) { - //OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << llendl; + OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << llendl; return false; } LLOctreeNode* parent = getOctParent(); @@ -389,7 +389,7 @@ public: else { //it's not in here, give it to the root - //OCT_ERRS << "Octree insertion failed, starting over from root!" << llendl; + OCT_ERRS << "Octree insertion failed, starting over from root!" << llendl; oct_node* node = this; @@ -590,7 +590,7 @@ public: } } - //OCT_ERRS << "Octree failed to delete requested child." << llendl; + OCT_ERRS << "Octree failed to delete requested child." << llendl; } protected: @@ -670,13 +670,13 @@ public: { if (data == NULL) { - //OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE ROOT !!!" << llendl; + OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE ROOT !!!" << llendl; return false; } if (data->getBinRadius() > 4096.0) { - //OCT_ERRS << "!!! ELEMENT EXCEEDS MAXIMUM SIZE IN OCTREE ROOT !!!" << llendl; + OCT_ERRS << "!!! ELEMENT EXCEEDS MAXIMUM SIZE IN OCTREE ROOT !!!" << llendl; return false; } @@ -692,7 +692,7 @@ public: if (lt != 0x7) { - //OCT_ERRS << "!!! ELEMENT EXCEEDS RANGE OF SPATIAL PARTITION !!!" << llendl; + OCT_ERRS << "!!! ELEMENT EXCEEDS RANGE OF SPATIAL PARTITION !!!" << llendl; return false; } diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 5ad15bd9b..46a22749a 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -48,6 +48,7 @@ #include "lloctree.h" #include "lldarray.h" #include "llvolume.h" +#include "llvolumeoctree.h" #include "llstl.h" #include "llsdserialize.h" #include "llvector4a.h" @@ -93,6 +94,19 @@ const F32 SKEW_MAX = 0.95f; const F32 SCULPT_MIN_AREA = 0.002f; const S32 SCULPT_MIN_AREA_DETAIL = 1; +extern BOOL gDebugGL; + +void assert_aligned(void* ptr, uintptr_t alignment) +{ +#if 0 + uintptr_t t = (uintptr_t) ptr; + if (t%alignment != 0) + { + llerrs << "Alignment check failed." << llendl; + } +#endif +} + BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLVector3& pt3, const LLVector3& norm) { LLVector3 test = (pt2-pt1)%(pt3-pt2); @@ -296,7 +310,6 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons } } -/* class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst { public: @@ -353,7 +366,7 @@ public: llerrs << "Empty leaf" << llendl; } - for (S32 i = 0; i < branch->getChildCount(); ++i) + for (U32 i = 0; i < branch->getChildCount(); ++i) { //stretch by child extents LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); min.setMin(min, child->mExtents[0]); @@ -366,7 +379,7 @@ public: node->mBounds[1].setSub(max,min); node->mBounds[1].mul(0.5f); } -};*/ +}; //------------------------------------------------------------------- // statics @@ -2233,6 +2246,152 @@ BOOL LLVolume::generate() return FALSE; } +void LLVolumeFace::VertexData::init() +{ + if (!mData) + { + mData = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*2); + } +} + +LLVolumeFace::VertexData::VertexData() +{ + mData = NULL; + init(); +} + +LLVolumeFace::VertexData::VertexData(const VertexData& rhs) +{ + mData = NULL; + *this = rhs; +} + +const LLVolumeFace::VertexData& LLVolumeFace::VertexData::operator=(const LLVolumeFace::VertexData& rhs) +{ + if (this != &rhs) + { + init(); + LLVector4a::memcpyNonAliased16((F32*) mData, (F32*) rhs.mData, 2*sizeof(LLVector4a)); + mTexCoord = rhs.mTexCoord; + } + return *this; +} + +LLVolumeFace::VertexData::~VertexData() +{ + ll_aligned_free_16(mData); + mData = NULL; +} + +LLVector4a& LLVolumeFace::VertexData::getPosition() +{ + return mData[POSITION]; +} + +LLVector4a& LLVolumeFace::VertexData::getNormal() +{ + return mData[NORMAL]; +} + +const LLVector4a& LLVolumeFace::VertexData::getPosition() const +{ + return mData[POSITION]; +} + +const LLVector4a& LLVolumeFace::VertexData::getNormal() const +{ + return mData[NORMAL]; +} + + +void LLVolumeFace::VertexData::setPosition(const LLVector4a& pos) +{ + mData[POSITION] = pos; +} + +void LLVolumeFace::VertexData::setNormal(const LLVector4a& norm) +{ + mData[NORMAL] = norm; +} + +bool LLVolumeFace::VertexData::operator<(const LLVolumeFace::VertexData& rhs)const +{ + const F32* lp = this->getPosition().getF32ptr(); + const F32* rp = rhs.getPosition().getF32ptr(); + + if (lp[0] != rp[0]) + { + return lp[0] < rp[0]; + } + + if (rp[1] != lp[1]) + { + return lp[1] < rp[1]; + } + + if (rp[2] != lp[2]) + { + return lp[2] < rp[2]; + } + + lp = getNormal().getF32ptr(); + rp = rhs.getNormal().getF32ptr(); + + if (lp[0] != rp[0]) + { + return lp[0] < rp[0]; + } + + if (rp[1] != lp[1]) + { + return lp[1] < rp[1]; + } + + if (rp[2] != lp[2]) + { + return lp[2] < rp[2]; + } + + if (mTexCoord.mV[0] != rhs.mTexCoord.mV[0]) + { + return mTexCoord.mV[0] < rhs.mTexCoord.mV[0]; + } + + return mTexCoord.mV[1] < rhs.mTexCoord.mV[1]; +} + +bool LLVolumeFace::VertexData::operator==(const LLVolumeFace::VertexData& rhs)const +{ + return mData[POSITION].equals3(rhs.getPosition()) && + mData[NORMAL].equals3(rhs.getNormal()) && + mTexCoord == rhs.mTexCoord; +} + +bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs, F32 angle_cutoff) const +{ + bool retval = false; + if (rhs.mData[POSITION].equals3(mData[POSITION]) && rhs.mTexCoord == mTexCoord) + { + if (angle_cutoff > 1.f) + { + retval = (mData[NORMAL].equals3(rhs.mData[NORMAL])); + } + else + { + F32 cur_angle = rhs.mData[NORMAL].dot3(mData[NORMAL]).getF32(); + retval = cur_angle > angle_cutoff; + } + } + + return retval; +} + + +S32 LLVolume::getNumFaces() const +{ + return (S32)mProfilep->mFaces.size(); +} + void LLVolume::createVolumeFaces() { @@ -3809,7 +3968,7 @@ S32 LLVolume::getNumTriangles() const for (S32 i = 0; i < getNumVolumeFaces(); ++i) { - triangle_count += getVolumeFace(i).mIndices.size()/3; + triangle_count += getVolumeFace(i).mNumIndices/3; } return triangle_count; @@ -3850,7 +4009,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, const LLVolumeFace& face = *iter; if (!(face_mask & (0x1 << cur_index++)) || - face.mIndices.empty() || face.mEdge.empty()) + face.mNumIndices == 0 || face.mEdge.empty()) { continue; } @@ -3867,7 +4026,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, #if DEBUG_SILHOUETTE_EDGE_MAP //for each triangle - U32 count = face.mIndices.size(); + U32 count = face.mNumIndices; for (U32 j = 0; j < count/3; j++) { //get vertices S32 v1 = face.mIndices[j*3+0]; @@ -3875,9 +4034,9 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, S32 v3 = face.mIndices[j*3+2]; //get current face center - LLVector3 cCenter = (face.mVertices[v1].mPosition + - face.mVertices[v2].mPosition + - face.mVertices[v3].mPosition) / 3.0f; + LLVector3 cCenter = (face.mVertices[v1].getPosition() + + face.mVertices[v2].getPosition() + + face.mVertices[v3].getPosition()) / 3.0f; //for each edge for (S32 k = 0; k < 3; k++) { @@ -3895,9 +4054,9 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, v3 = face.mIndices[nIndex*3+2]; //get neighbor face center - LLVector3 nCenter = (face.mVertices[v1].mPosition + - face.mVertices[v2].mPosition + - face.mVertices[v3].mPosition) / 3.0f; + LLVector3 nCenter = (face.mVertices[v1].getPosition() + + face.mVertices[v2].getPosition() + + face.mVertices[v3].getPosition()) / 3.0f; //draw line vertices.push_back(cCenter); @@ -3920,15 +4079,15 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, #elif DEBUG_SILHOUETTE_NORMALS //for each vertex - for (U32 j = 0; j < face.mVertices.size(); j++) { - vertices.push_back(face.mVertices[j].mPosition); - vertices.push_back(face.mVertices[j].mPosition + face.mVertices[j].mNormal*0.1f); + for (U32 j = 0; j < face.mNumVertices; j++) { + vertices.push_back(face.mVertices[j].getPosition()); + vertices.push_back(face.mVertices[j].getPosition() + face.mVertices[j].getNormal()*0.1f); normals.push_back(LLVector3(0,0,1)); normals.push_back(LLVector3(0,0,1)); segments.push_back(vertices.size()); #if DEBUG_SILHOUETTE_BINORMALS - vertices.push_back(face.mVertices[j].mPosition); - vertices.push_back(face.mVertices[j].mPosition + face.mVertices[j].mBinormal*0.1f); + vertices.push_back(face.mVertices[j].getPosition()); + vertices.push_back(face.mVertices[j].getPosition() + face.mVertices[j].mBinormal*0.1f); normals.push_back(LLVector3(0,0,1)); normals.push_back(LLVector3(0,0,1)); segments.push_back(vertices.size()); @@ -3946,23 +4105,21 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, //for each triangle std::vector fFacing; - vector_append(fFacing, face.mIndices.size()/3); - for (U32 j = 0; j < face.mIndices.size()/3; j++) + vector_append(fFacing, face.mNumIndices/3); + + LLVector4a* v = (LLVector4a*) face.mPositions; + LLVector4a* n = (LLVector4a*) face.mNormals; + + for (U32 j = 0; j < (U32)face.mNumIndices/3; j++) { //approximate normal S32 v1 = face.mIndices[j*3+0]; S32 v2 = face.mIndices[j*3+1]; S32 v3 = face.mIndices[j*3+2]; - //ew. Face verts arent aligned yet. - LLVector4a va1,va3; - va1.load3(face.mVertices[v1].mPosition.mV); - va3.load3(face.mVertices[v3].mPosition.mV); - LLVector4a c1,c2; - c2.load3(face.mVertices[v2].mPosition.mV); - c1.setSub(va1,c2); - c2.sub(va3); + c1.setSub(v[v1], v[v2]); + c2.setSub(v[v2], v[v3]); LLVector4a norm; @@ -3976,7 +4133,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, { //get view vector LLVector4a view; - view.setSub(obj_cam_vec, va1); + view.setSub(obj_cam_vec, v[v1]); bool away = view.dot3(norm) > 0.0f; if (away) { @@ -3990,7 +4147,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, } //for each triangle - for (U32 j = 0; j < face.mIndices.size()/3; j++) + for (U32 j = 0; j < (U32)face.mNumIndices/3; j++) { if (fFacing[j] == (AWAY | TOWARDS)) { //this is a degenerate triangle @@ -4023,15 +4180,21 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, S32 v1 = face.mIndices[j*3+k]; S32 v2 = face.mIndices[j*3+((k+1)%3)]; - vertices.push_back(face.mVertices[v1].mPosition*mat_in); - LLVector3 norm1 = face.mVertices[v1].mNormal * norm_mat_in; - norm1.normVec(); - normals.push_back(norm1); + LLVector4a t; + mat.affineTransform(v[v1], t); + vertices.push_back(LLVector3(t[0], t[1], t[2])); - vertices.push_back(face.mVertices[v2].mPosition*mat_in); - LLVector3 norm2 = face.mVertices[v2].mNormal * norm_mat_in; - norm2.normVec(); - normals.push_back(norm2); + norm_mat.rotate(n[v1], t); + + t.normalize3fast(); + normals.push_back(LLVector3(t[0], t[1], t[2])); + + mat.affineTransform(v[v2], t); + vertices.push_back(LLVector3(t[0], t[1], t[2])); + + norm_mat.rotate(n[v2], t); + t.normalize3fast(); + normals.push_back(LLVector3(t[0], t[1], t[2])); segments.push_back(vertices.size()); } @@ -4080,9 +4243,11 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en F32 closest_t = 2.f; // must be larger than 1 + end_face = llmin(end_face, getNumVolumeFaces()-1); + for (S32 i = start_face; i <= end_face; i++) { - const LLVolumeFace &face = getVolumeFace((U32)i); + LLVolumeFace &face = mVolumeFaces[i]; LLVector4a box_center; box_center.setAdd(face.mExtents[0], face.mExtents[1]); @@ -4097,63 +4262,19 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en { genBinormals(i); } - - for (U32 tri = 0; tri < face.mIndices.size()/3; tri++) + + if (!face.mOctree) { - S32 index1 = face.mIndices[tri*3+0]; - S32 index2 = face.mIndices[tri*3+1]; - S32 index3 = face.mIndices[tri*3+2]; - - F32 a, b, t; - - LLVector4a pos[3]; - pos[0].load3(face.mVertices[index1].mPosition.mV); - pos[1].load3(face.mVertices[index2].mPosition.mV); - pos[2].load3(face.mVertices[index3].mPosition.mV); - if (LLTriangleRayIntersect(pos[0], - pos[1], - pos[2], - start, dir, a, b, t)) - { - if ((t >= 0.f) && // if hit is after start - (t <= 1.f) && // and before end - (t < closest_t)) // and this hit is closer - { - closest_t = t; - hit_face = i; - - if (intersection != NULL) - { - LLVector4a v; - v.setMul(dir,closest_t); - v.add(start); - intersection->set(v.getF32ptr()); - } + face.createOctree(); + } - if (tex_coord != NULL) + //LLVector4a* p = (LLVector4a*) face.mPositions; + + LLOctreeTriangleRayIntersect intersect(start, dir, &face, &closest_t, intersection, tex_coord, normal, bi_normal); + intersect.traverse(face.mOctree); + if (intersect.mHitFace) { - *tex_coord = ((1.f - a - b) * face.mVertices[index1].mTexCoord + - a * face.mVertices[index2].mTexCoord + - b * face.mVertices[index3].mTexCoord); - - } - - if (normal != NULL) - { - *normal = ((1.f - a - b) * face.mVertices[index1].mNormal + - a * face.mVertices[index2].mNormal + - b * face.mVertices[index3].mNormal); - } - - if (bi_normal != NULL) - { - *bi_normal = ((1.f - a - b) * face.mVertices[index1].mBinormal + - a * face.mVertices[index2].mBinormal + - b * face.mVertices[index3].mBinormal); - } - - } - } + hit_face = i; } } } @@ -4923,11 +5044,18 @@ std::ostream& operator<<(std::ostream &s, const LLVolume *volumep) LLVolumeFace::LLVolumeFace() : mID(0), mTypeMask(0), - mHasBinormals(FALSE), mBeginS(0), mBeginT(0), mNumS(0), - mNumT(0) + mNumT(0), + mNumVertices(0), + mNumIndices(0), + mPositions(NULL), + mNormals(NULL), + mBinormals(NULL), + mTexCoords(NULL), + mIndices(NULL), + mOctree(NULL) { mExtents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*3); mExtents[0].splat(-0.5f); @@ -4938,11 +5066,18 @@ LLVolumeFace::LLVolumeFace() : LLVolumeFace::LLVolumeFace(const LLVolumeFace& src) : mID(0), mTypeMask(0), - mHasBinormals(FALSE), mBeginS(0), mBeginT(0), mNumS(0), - mNumT(0) + mNumT(0), + mNumVertices(0), + mNumIndices(0), + mPositions(NULL), + mNormals(NULL), + mBinormals(NULL), + mTexCoords(NULL), + mIndices(NULL), + mOctree(NULL) { mExtents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*3); mCenter = mExtents+2; @@ -4958,7 +5093,6 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) mID = src.mID; mTypeMask = src.mTypeMask; - mHasBinormals = src.mHasBinormals, mBeginS = src.mBeginS; mBeginT = src.mBeginT; mNumS = src.mNumS; @@ -4968,23 +5102,80 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) mExtents[1] = src.mExtents[1]; *mCenter = *src.mCenter; - + mNumVertices = 0; + mNumIndices = 0; - LLVector4a::memcpyNonAliased16((F32*) mExtents, (F32*) src.mExtents, 3*sizeof(LLVector4a)); + freeData(); - mVertices = src.mVertices; - mIndices = src.mIndices; - mEdge = src.mEdge; + LLVector4a::memcpyNonAliased16((F32*) mExtents, (F32*) src.mExtents, 3*sizeof(LLVector4a)); + + resizeVertices(src.mNumVertices); + resizeIndices(src.mNumIndices); + + if (mNumVertices) + { + S32 vert_size = mNumVertices*sizeof(LLVector4a); + S32 tc_size = (mNumVertices*sizeof(LLVector2)+0xF) & ~0xF; + + LLVector4a::memcpyNonAliased16((F32*) mPositions, (F32*) src.mPositions, vert_size); + LLVector4a::memcpyNonAliased16((F32*) mNormals, (F32*) src.mNormals, vert_size); + LLVector4a::memcpyNonAliased16((F32*) mTexCoords, (F32*) src.mTexCoords, tc_size); + + + if (src.mBinormals) + { + allocateBinormals(src.mNumVertices); + LLVector4a::memcpyNonAliased16((F32*) mBinormals, (F32*) src.mBinormals, vert_size); + } + else + { + ll_aligned_free_16(mBinormals); + mBinormals = NULL; + } + + } + if (mNumIndices) + { + S32 idx_size = (mNumIndices*sizeof(U16)+0xF) & ~0xF; + + LLVector4a::memcpyNonAliased16((F32*) mIndices, (F32*) src.mIndices, idx_size); + } + + //delete return *this; } + LLVolumeFace::~LLVolumeFace() { ll_aligned_free_16(mExtents); mExtents = NULL; + freeData(); } + +void LLVolumeFace::freeData() +{ + ll_aligned_free_16(mPositions); + mPositions = NULL; + ll_aligned_free_16( mNormals); + mNormals = NULL; + ll_aligned_free_16(mTexCoords); + mTexCoords = NULL; + ll_aligned_free_16(mIndices); + mIndices = NULL; + ll_aligned_free_16(mBinormals); + mBinormals = NULL; + + delete mOctree; + mOctree = NULL; +} + BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) { + //tree for this face is no longer valid + delete mOctree; + mOctree = NULL; + BOOL ret = FALSE ; if (mTypeMask & CAP_MASK) { @@ -5005,25 +5196,24 @@ BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) mTexCoordExtents[0].setVec(1.f, 1.f) ; mTexCoordExtents[1].setVec(0.f, 0.f) ; - U32 end = mVertices.size() ; - for(U32 i = 0 ; i < end ; i++) + for(U32 i = 0 ; i < (U32)mNumVertices ; i++) { - if(mTexCoordExtents[0].mV[0] > mVertices[i].mTexCoord.mV[0]) + if(mTexCoordExtents[0].mV[0] > mTexCoords[i].mV[0]) { - mTexCoordExtents[0].mV[0] = mVertices[i].mTexCoord.mV[0] ; + mTexCoordExtents[0].mV[0] = mTexCoords[i].mV[0] ; } - if(mTexCoordExtents[1].mV[0] < mVertices[i].mTexCoord.mV[0]) + if(mTexCoordExtents[1].mV[0] < mTexCoords[i].mV[0]) { - mTexCoordExtents[1].mV[0] = mVertices[i].mTexCoord.mV[0] ; + mTexCoordExtents[1].mV[0] = mTexCoords[i].mV[0] ; } - if(mTexCoordExtents[0].mV[1] > mVertices[i].mTexCoord.mV[1]) + if(mTexCoordExtents[0].mV[1] > mTexCoords[i].mV[1]) { - mTexCoordExtents[0].mV[1] = mVertices[i].mTexCoord.mV[1] ; + mTexCoordExtents[0].mV[1] = mTexCoords[i].mV[1] ; } - if(mTexCoordExtents[1].mV[1] < mVertices[i].mTexCoord.mV[1]) + if(mTexCoordExtents[1].mV[1] < mTexCoords[i].mV[1]) { - mTexCoordExtents[1].mV[1] = mVertices[i].mTexCoord.mV[1] ; + mTexCoordExtents[1].mV[1] = mTexCoords[i].mV[1] ; } } mTexCoordExtents[0].mV[0] = llmax(0.f, mTexCoordExtents[0].mV[0]) ; @@ -5035,6 +5225,603 @@ BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) return ret ; } +void LLVolumeFace::getVertexData(U16 index, LLVolumeFace::VertexData& cv) +{ + cv.setPosition(mPositions[index]); + cv.setNormal(mNormals[index]); + cv.mTexCoord = mTexCoords[index]; +} + +bool LLVolumeFace::VertexMapData::operator==(const LLVolumeFace::VertexData& rhs) const +{ + return getPosition().equals3(rhs.getPosition()) && + mTexCoord == rhs.mTexCoord && + getNormal().equals3(rhs.getNormal()); +} + +bool LLVolumeFace::VertexMapData::ComparePosition::operator()(const LLVector3& a, const LLVector3& b) const +{ + if (a.mV[0] != b.mV[0]) + { + return a.mV[0] < b.mV[0]; + } + + if (a.mV[1] != b.mV[1]) + { + return a.mV[1] < b.mV[1]; + } + + return a.mV[2] < b.mV[2]; +} + +void LLVolumeFace::optimize(F32 angle_cutoff) +{ + LLVolumeFace new_face; + + //map of points to vector of vertices at that point + VertexMapData::PointMap point_map; + + //remove redundant vertices + for (U32 i = 0; i < (U32)mNumIndices; ++i) + { + U16 index = mIndices[i]; + + LLVolumeFace::VertexData cv; + getVertexData(index, cv); + + BOOL found = FALSE; + VertexMapData::PointMap::iterator point_iter = point_map.find(LLVector3(cv.getPosition().getF32ptr())); + if (point_iter != point_map.end()) + { //duplicate point might exist + for (U32 j = 0; j < point_iter->second.size(); ++j) + { + LLVolumeFace::VertexData& tv = (point_iter->second)[j]; + if (tv.compareNormal(cv, angle_cutoff)) + { + found = TRUE; + new_face.pushIndex((point_iter->second)[j].mIndex); + break; + } + } + } + + if (!found) + { + new_face.pushVertex(cv); + U16 index = (U16) new_face.mNumVertices-1; + new_face.pushIndex(index); + + VertexMapData d; + d.setPosition(cv.getPosition()); + d.mTexCoord = cv.mTexCoord; + d.setNormal(cv.getNormal()); + d.mIndex = index; + if (point_iter != point_map.end()) + { + point_iter->second.push_back(d); + } + else + { + point_map[LLVector3(d.getPosition().getF32ptr())].push_back(d); + } + } + } + + swapData(new_face); +} + +class LLVCacheTriangleData; + +class LLVCacheVertexData +{ +public: + S32 mIdx; + S32 mCacheTag; + F32 mScore; + U32 mActiveTriangles; + std::vector mTriangles; + + LLVCacheVertexData() + { + mCacheTag = -1; + mScore = 0.f; + mActiveTriangles = 0; + mIdx = -1; + } +}; + +class LLVCacheTriangleData +{ +public: + bool mActive; + F32 mScore; + LLVCacheVertexData* mVertex[3]; + + LLVCacheTriangleData() + { + mActive = true; + mScore = 0.f; + mVertex[0] = mVertex[1] = mVertex[2] = NULL; + } + + void complete() + { + mActive = false; + for (S32 i = 0; i < 3; ++i) + { + if (mVertex[i]) + { + llassert_always(mVertex[i]->mActiveTriangles > 0); + mVertex[i]->mActiveTriangles--; + } + } + } + + bool operator<(const LLVCacheTriangleData& rhs) const + { //highest score first + return rhs.mScore < mScore; + } +}; + +const F32 FindVertexScore_CacheDecayPower = 1.5f; +const F32 FindVertexScore_LastTriScore = 0.75f; +const F32 FindVertexScore_ValenceBoostScale = 2.0f; +const F32 FindVertexScore_ValenceBoostPower = 0.5f; +const U32 MaxSizeVertexCache = 32; + +F32 find_vertex_score(LLVCacheVertexData& data) +{ + if (data.mActiveTriangles == 0) + { //no triangle references this vertex + return -1.f; + } + + F32 score = 0.f; + + S32 cache_idx = data.mCacheTag; + + if (cache_idx < 0) + { + //not in cache + } + else + { + if (cache_idx < 3) + { //vertex was in the last triangle + score = FindVertexScore_LastTriScore; + } + else + { //more points for being higher in the cache + F32 scaler = 1.f/(MaxSizeVertexCache-3); + score = 1.f-((cache_idx-3)*scaler); + score = powf(score, FindVertexScore_CacheDecayPower); + } + } + + //bonus points for having low valence + F32 valence_boost = powf(data.mActiveTriangles, -FindVertexScore_ValenceBoostPower); + score += FindVertexScore_ValenceBoostScale * valence_boost; + + return score; +} + +class LLVCacheFIFO +{ +public: + LLVCacheVertexData* mCache[MaxSizeVertexCache]; + U32 mMisses; + + LLVCacheFIFO() + { + mMisses = 0; + for (U32 i = 0; i < MaxSizeVertexCache; ++i) + { + mCache[i] = NULL; + } + } + + void addVertex(LLVCacheVertexData* data) + { + if (data->mCacheTag == -1) + { + mMisses++; + + S32 end = MaxSizeVertexCache-1; + + if (mCache[end]) + { + mCache[end]->mCacheTag = -1; + } + + for (S32 i = end; i > 0; --i) + { + mCache[i] = mCache[i-1]; + if (mCache[i]) + { + mCache[i]->mCacheTag = i; + } + } + + mCache[0] = data; + data->mCacheTag = 0; + } + } +}; + +class LLVCacheLRU +{ +public: + LLVCacheVertexData* mCache[MaxSizeVertexCache+3]; + + LLVCacheTriangleData* mBestTriangle; + + U32 mMisses; + + LLVCacheLRU() + { + for (U32 i = 0; i < MaxSizeVertexCache+3; ++i) + { + mCache[i] = NULL; + } + + mBestTriangle = NULL; + mMisses = 0; + } + + void addVertex(LLVCacheVertexData* data) + { + S32 end = MaxSizeVertexCache+2; + if (data->mCacheTag != -1) + { //just moving a vertex to the front of the cache + end = data->mCacheTag; + } + else + { + mMisses++; + if (mCache[end]) + { //adding a new vertex, vertex at end of cache falls off + mCache[end]->mCacheTag = -1; + } + } + + for (S32 i = end; i > 0; --i) + { //adjust cache pointers and tags + mCache[i] = mCache[i-1]; + + if (mCache[i]) + { + mCache[i]->mCacheTag = i; + } + } + + mCache[0] = data; + mCache[0]->mCacheTag = 0; + } + + void addTriangle(LLVCacheTriangleData* data) + { + addVertex(data->mVertex[0]); + addVertex(data->mVertex[1]); + addVertex(data->mVertex[2]); + } + + void updateScores() + { + for (U32 i = MaxSizeVertexCache; i < MaxSizeVertexCache+3; ++i) + { //trailing 3 vertices aren't actually in the cache for scoring purposes + if (mCache[i]) + { + mCache[i]->mCacheTag = -1; + } + } + + for (U32 i = 0; i < MaxSizeVertexCache; ++i) + { //update scores of vertices in cache + if (mCache[i]) + { + mCache[i]->mScore = find_vertex_score(*(mCache[i])); + llassert_always(mCache[i]->mCacheTag == i); + } + } + + mBestTriangle = NULL; + //update triangle scores + for (U32 i = 0; i < MaxSizeVertexCache+3; ++i) + { + if (mCache[i]) + { + for (U32 j = 0; j < mCache[i]->mTriangles.size(); ++j) + { + LLVCacheTriangleData* tri = mCache[i]->mTriangles[j]; + if (tri->mActive) + { + tri->mScore = tri->mVertex[0]->mScore; + tri->mScore += tri->mVertex[1]->mScore; + tri->mScore += tri->mVertex[2]->mScore; + + if (!mBestTriangle || mBestTriangle->mScore < tri->mScore) + { + mBestTriangle = tri; + } + } + } + } + } + + //knock trailing 3 vertices off the cache + for (U32 i = MaxSizeVertexCache; i < MaxSizeVertexCache+3; ++i) + { + if (mCache[i]) + { + llassert_always(mCache[i]->mCacheTag == -1); + mCache[i] = NULL; + } + } + } +}; + + +void LLVolumeFace::cacheOptimize() +{ //optimize for vertex cache according to Forsyth method: + // http://home.comcast.net/~tom_forsyth/papers/fast_vert_cache_opt.html + + LLVCacheLRU cache; + + if (mNumVertices < 3) + { //nothing to do + return; + } + + //mapping of vertices to triangles and indices + std::vector vertex_data; + + //mapping of triangles do vertices + std::vector triangle_data; + + triangle_data.resize(mNumIndices/3); + vertex_data.resize(mNumVertices); + + for (U32 i = 0; i < (U32)mNumIndices; i++) + { //populate vertex data and triangle data arrays + U16 idx = mIndices[i]; + U32 tri_idx = i/3; + + vertex_data[idx].mTriangles.push_back(&(triangle_data[tri_idx])); + vertex_data[idx].mIdx = idx; + triangle_data[tri_idx].mVertex[i%3] = &(vertex_data[idx]); + } + + /*F32 pre_acmr = 1.f; + //measure cache misses from before rebuild + { + LLVCacheFIFO test_cache; + for (U32 i = 0; i < mNumIndices; ++i) + { + test_cache.addVertex(&vertex_data[mIndices[i]]); + } + + for (U32 i = 0; i < mNumVertices; i++) + { + vertex_data[i].mCacheTag = -1; + } + + pre_acmr = (F32) test_cache.mMisses/(mNumIndices/3); + }*/ + + for (U32 i = 0; i < (U32)mNumVertices; i++) + { //initialize score values (no cache -- might try a fifo cache here) + vertex_data[i].mScore = find_vertex_score(vertex_data[i]); + vertex_data[i].mActiveTriangles = vertex_data[i].mTriangles.size(); + + for (U32 j = 0; j < vertex_data[i].mTriangles.size(); ++j) + { + vertex_data[i].mTriangles[j]->mScore += vertex_data[i].mScore; + } + } + + //sort triangle data by score + std::sort(triangle_data.begin(), triangle_data.end()); + + std::vector new_indices; + + LLVCacheTriangleData* tri; + + //prime pump by adding first triangle to cache; + tri = &(triangle_data[0]); + cache.addTriangle(tri); + new_indices.push_back(tri->mVertex[0]->mIdx); + new_indices.push_back(tri->mVertex[1]->mIdx); + new_indices.push_back(tri->mVertex[2]->mIdx); + tri->complete(); + + U32 breaks = 0; + for (U32 i = 1; i < (U32)mNumIndices/3; ++i) + { + cache.updateScores(); + tri = cache.mBestTriangle; + if (!tri) + { + breaks++; + for (U32 j = 0; j < triangle_data.size(); ++j) + { + if (triangle_data[j].mActive) + { + tri = &(triangle_data[j]); + break; + } + } + } + + cache.addTriangle(tri); + new_indices.push_back(tri->mVertex[0]->mIdx); + new_indices.push_back(tri->mVertex[1]->mIdx); + new_indices.push_back(tri->mVertex[2]->mIdx); + tri->complete(); + } + + for (U32 i = 0; i < (U32)mNumIndices; ++i) + { + mIndices[i] = new_indices[i]; + } + + /*F32 post_acmr = 1.f; + //measure cache misses from after rebuild + { + LLVCacheFIFO test_cache; + for (U32 i = 0; i < mNumVertices; i++) + { + vertex_data[i].mCacheTag = -1; + } + + for (U32 i = 0; i < mNumIndices; ++i) + { + test_cache.addVertex(&vertex_data[mIndices[i]]); + } + + post_acmr = (F32) test_cache.mMisses/(mNumIndices/3); + }*/ + + //optimize for pre-TnL cache + + //allocate space for new buffer + S32 num_verts = mNumVertices; + LLVector4a* pos = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); + LLVector4a* norm = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); + S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF; + LLVector2* tc = (LLVector2*) ll_aligned_malloc_16(size); + + + LLVector4a* binorm = NULL; + if (mBinormals) + { + binorm = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); + } + + //allocate mapping of old indices to new indices + std::vector new_idx; + new_idx.resize(mNumVertices, -1); + + S32 cur_idx = 0; + for (U32 i = 0; i < (U32)mNumIndices; ++i) + { + U16 idx = mIndices[i]; + if (new_idx[idx] == -1) + { //this vertex hasn't been added yet + new_idx[idx] = cur_idx; + + //copy vertex data + pos[cur_idx] = mPositions[idx]; + norm[cur_idx] = mNormals[idx]; + tc[cur_idx] = mTexCoords[idx]; + if (mBinormals) + { + binorm[cur_idx] = mBinormals[idx]; + } + + cur_idx++; + } + } + + for (U32 i = 0; i < (U32)mNumIndices; ++i) + { + mIndices[i] = new_idx[mIndices[i]]; + } + + ll_aligned_free_16(mPositions); + ll_aligned_free_16(mNormals); + ll_aligned_free_16(mTexCoords); + ll_aligned_free_16(mBinormals); + + mPositions = pos; + mNormals = norm; + mTexCoords = tc; + mBinormals = binorm; + + //std::string result = llformat("ACMR pre/post: %.3f/%.3f -- %d triangles %d breaks", pre_acmr, post_acmr, mNumIndices/3, breaks); + //llinfos << result << llendl; + +} + +void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVector4a& size) +{ + if (mOctree) + { + return; + } + + mOctree = new LLOctreeRoot(center, size, NULL); + new LLVolumeOctreeListener(mOctree); + + for (U32 i = 0; i < (U32)mNumIndices; i+= 3) + { //for each triangle + LLPointer 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() * 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); + } +} + + +void LLVolumeFace::swapData(LLVolumeFace& rhs) +{ + llswap(rhs.mPositions, mPositions); + llswap(rhs.mNormals, mNormals); + llswap(rhs.mBinormals, mBinormals); + llswap(rhs.mTexCoords, mTexCoords); + llswap(rhs.mIndices,mIndices); + llswap(rhs.mNumVertices, mNumVertices); + llswap(rhs.mNumIndices, mNumIndices); +} void LerpPlanarVertex(LLVolumeFace::VertexData& v0, LLVolumeFace::VertexData& v1, LLVolumeFace::VertexData& v2, @@ -5042,10 +5829,21 @@ void LerpPlanarVertex(LLVolumeFace::VertexData& v0, F32 coef01, F32 coef02) { - vout.mPosition = v0.mPosition + ((v1.mPosition-v0.mPosition)*coef01)+((v2.mPosition-v0.mPosition)*coef02); + + LLVector4a lhs; + lhs.setSub(v1.getPosition(), v0.getPosition()); + lhs.mul(coef01); + LLVector4a rhs; + rhs.setSub(v2.getPosition(), v0.getPosition()); + rhs.mul(coef02); + + rhs.add(lhs); + rhs.add(v0.getPosition()); + + vout.setPosition(rhs); + vout.mTexCoord = v0.mTexCoord + ((v1.mTexCoord-v0.mTexCoord)*coef01)+((v2.mTexCoord-v0.mTexCoord)*coef02); - vout.mNormal = v0.mNormal; - vout.mBinormal = v0.mBinormal; + vout.setNormal(v0.getNormal()); } BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) @@ -5078,26 +5876,28 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) offset = mBeginS; } - S32 vtop; { VertexData corners[4]; VertexData baseVert; - for(S32 t = 0; t < 4; t++){ - corners[t].mPosition = mesh[offset + (grid_size*t)].mPos; + for(S32 t = 0; t < 4; t++) + { + corners[t].getPosition().load3( mesh[offset + (grid_size*t)].mPos.mV); corners[t].mTexCoord.mV[0] = profile[grid_size*t].mV[0]+0.5f; corners[t].mTexCoord.mV[1] = 0.5f - profile[grid_size*t].mV[1]; } { - baseVert.mNormal = - ((corners[1].mPosition-corners[0].mPosition) % - (corners[2].mPosition-corners[1].mPosition)); - baseVert.mNormal.normalize(); + LLVector4a lhs; + lhs.setSub(corners[1].getPosition(), corners[0].getPosition()); + LLVector4a rhs; + rhs.setSub(corners[2].getPosition(), corners[1].getPosition()); + baseVert.getNormal().setCross3(lhs, rhs); + baseVert.getNormal().normalize3fast(); } if(!(mTypeMask & TOP_MASK)) { - baseVert.mNormal *= -.1f; + baseVert.getNormal().mul(-1.0f); } else { @@ -5110,22 +5910,24 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) corners[1].mTexCoord=corners[2].mTexCoord; corners[2].mTexCoord=swap; } - baseVert.mBinormal = calc_binormal_from_triangle( - corners[0].mPosition, corners[0].mTexCoord, - corners[1].mPosition, corners[1].mTexCoord, - corners[2].mPosition, corners[2].mTexCoord); - for(int t = 0; t < 4; t++){ - corners[t].mBinormal = baseVert.mBinormal; - corners[t].mNormal = baseVert.mNormal; - } - mHasBinormals = TRUE; - if (partial_build) - { - mVertices.clear(); - } + LLVector4a binormal; - vtop = mVertices.size(); + calc_binormal_from_triangle( binormal, + corners[0].getPosition(), corners[0].mTexCoord, + corners[1].getPosition(), corners[1].mTexCoord, + corners[2].getPosition(), corners[2].mTexCoord); + + binormal.normalize3fast(); + + S32 size = (grid_size+1)*(grid_size+1); + resizeVertices(size); + allocateBinormals(size); + + LLVector4a* pos = (LLVector4a*) mPositions; + LLVector4a* norm = (LLVector4a*) mNormals; + LLVector4a* binorm = (LLVector4a*) mBinormals; + LLVector2* tc = (LLVector2*) mTexCoords; for(int gx = 0;gx=0;i--) { - mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); - } + *out++ = ((gy*(grid_size+1))+gx+idxs[i]); + } } else { for(S32 i=0;i<6;i++) { - mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); + *out++ = ((gy*(grid_size+1))+gx+idxs[i]); } } } @@ -5212,11 +6021,25 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) num_vertices = profile.size(); num_indices = (profile.size() - 2)*3; - mVertices.resize(num_vertices); - - if (!partial_build) + if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK)) { - mIndices.resize(num_indices); + resizeVertices(num_vertices+1); + allocateBinormals(num_vertices+1); + + if (!partial_build) + { + resizeIndices(num_indices+3); + } + } + else + { + resizeVertices(num_vertices); + allocateBinormals(num_vertices); + + if (!partial_build) + { + resizeIndices(num_indices); + } } S32 max_s = volume->getProfile().getTotal(); @@ -5243,35 +6066,38 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) LLVector4a& min = mExtents[0]; LLVector4a& max = mExtents[1]; + LLVector2* tc = (LLVector2*) mTexCoords; + LLVector4a* pos = (LLVector4a*) mPositions; + LLVector4a* norm = (LLVector4a*) mNormals; + LLVector4a* binorm = (LLVector4a*) mBinormals; + // Copy the vertices into the array for (S32 i = 0; i < num_vertices; i++) { if (mTypeMask & TOP_MASK) { - mVertices[i].mTexCoord.mV[0] = profile[i].mV[0]+0.5f; - mVertices[i].mTexCoord.mV[1] = profile[i].mV[1]+0.5f; + tc[i].mV[0] = profile[i].mV[0]+0.5f; + tc[i].mV[1] = profile[i].mV[1]+0.5f; } else { // Mirror for underside. - mVertices[i].mTexCoord.mV[0] = profile[i].mV[0]+0.5f; - mVertices[i].mTexCoord.mV[1] = 0.5f - profile[i].mV[1]; + tc[i].mV[0] = profile[i].mV[0]+0.5f; + tc[i].mV[1] = 0.5f - profile[i].mV[1]; } - mVertices[i].mPosition = mesh[i + offset].mPos; + pos[i].load3(mesh[i + offset].mPos.mV); if (i == 0) { - max.load3(mVertices[i].mPosition.mV); + max = pos[i]; min = max; - min_uv = max_uv = mVertices[i].mTexCoord; + min_uv = max_uv = tc[i]; } else { - LLVector4a pos; - pos.load3(mVertices[i].mPosition.mV); - update_min_max(min,max, pos); - update_min_max(min_uv, max_uv, mVertices[i].mTexCoord); + update_min_max(min,max,pos[i]); + update_min_max(min_uv, max_uv, tc[i]); } } @@ -5280,15 +6106,13 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) cuv = (min_uv + max_uv)*0.5f; - LLVector3 binormal = calc_binormal_from_triangle( - LLVector3(mCenter->getF32ptr()), cuv, - mVertices[0].mPosition, mVertices[0].mTexCoord, - mVertices[1].mPosition, mVertices[1].mTexCoord); - binormal.normVec(); + LLVector4a binormal; + calc_binormal_from_triangle(binormal, + *mCenter, cuv, + pos[0], tc[0], + pos[1], tc[1]); + binormal.normalize3fast(); - LLVector4a pos[2]; - pos[0].load3(mVertices[0].mPosition.mV); - pos[1].load3(mVertices[1].mPosition.mV); LLVector4a normal; LLVector4a d0, d1; @@ -5308,30 +6132,22 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) normal.normalize3fast(); VertexData vd; - vd.mPosition.set(mCenter->getF32ptr()); - vd.mNormal.set(normal.getF32ptr()); - vd.mBinormal = binormal; + vd.setPosition(*mCenter); vd.mTexCoord = cuv; if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK)) { - mVertices.push_back(vd); + pos[num_vertices] = *mCenter; + tc[num_vertices] = cuv; num_vertices++; - if (!partial_build) - { - vector_append(mIndices, 3); - } } - for (S32 i = 0; i < num_vertices; i++) { - mVertices[i].mBinormal = binormal; - mVertices[i].mNormal.set(normal.getF32ptr()); + binorm[i].load4a(binormal.getF32ptr()); + norm[i].load4a(normal.getF32ptr()); } - mHasBinormals = TRUE; - if (partial_build) { return TRUE; @@ -5563,7 +6379,10 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) mIndices[3*i+v1] = i; mIndices[3*i+v2] = i + 1; } + + } + return TRUE; } @@ -5571,44 +6390,261 @@ void LLVolumeFace::createBinormals() { LLMemType m1(LLMemType::MTYPE_VOLUME); - if (!mHasBinormals) + if (!mBinormals) { + allocateBinormals(mNumVertices); + //generate binormals - for (U32 i = 0; i < mIndices.size()/3; i++) + LLVector4a* pos = mPositions; + LLVector2* tc = (LLVector2*) mTexCoords; + LLVector4a* binorm = (LLVector4a*) mBinormals; + + LLVector4a* end = mBinormals+mNumVertices; + while (binorm < end) + { + (*binorm++).clear(); + } + + binorm = mBinormals; + + for (U32 i = 0; i < (U32)mNumIndices/3; i++) { //for each triangle - const VertexData& v0 = mVertices[mIndices[i*3+0]]; - const VertexData& v1 = mVertices[mIndices[i*3+1]]; - const VertexData& v2 = mVertices[mIndices[i*3+2]]; + const U16& i0 = mIndices[i*3+0]; + const U16& i1 = mIndices[i*3+1]; + const U16& i2 = mIndices[i*3+2]; //calculate binormal - LLVector3 binorm = calc_binormal_from_triangle(v0.mPosition, v0.mTexCoord, - v1.mPosition, v1.mTexCoord, - v2.mPosition, v2.mTexCoord); + LLVector4a binormal; + calc_binormal_from_triangle(binormal, + pos[i0], tc[i0], + pos[i1], tc[i1], + pos[i2], tc[i2]); - for (U32 j = 0; j < 3; j++) - { //add triangle normal to vertices - mVertices[mIndices[i*3+j]].mBinormal += binorm; // * (weight_sum - d[j])/weight_sum; - } + + //add triangle normal to vertices + binorm[i0].add(binormal); + binorm[i1].add(binormal); + binorm[i2].add(binormal); //even out quad contributions if (i % 2 == 0) { - mVertices[mIndices[i*3+2]].mBinormal += binorm; + binorm[i2].add(binormal); } else { - mVertices[mIndices[i*3+1]].mBinormal += binorm; + binorm[i1].add(binormal); } } //normalize binormals - for (U32 i = 0; i < mVertices.size(); i++) + for (U32 i = 0; i < (U32)mNumVertices; i++) { - mVertices[i].mBinormal.normVec(); - mVertices[i].mNormal.normVec(); + binorm[i].normalize3fast(); + //bump map/planar projection code requires normals to be normalized + mNormals[i].normalize3fast(); } + } +} - mHasBinormals = TRUE; +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); + + mBinormals = NULL; + + if (num_verts) + { + mPositions = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); + assert_aligned(mPositions, 16); + mNormals = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); + assert_aligned(mNormals, 16); + + //pad texture coordinate block end to allow for QWORD reads + S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF; + mTexCoords = (LLVector2*) ll_aligned_malloc_16(size); + assert_aligned(mTexCoords, 16); + } + else + { + mPositions = NULL; + mNormals = NULL; + mTexCoords = NULL; + } + + mNumVertices = num_verts; +} + +void LLVolumeFace::pushVertex(const LLVolumeFace::VertexData& cv) +{ + pushVertex(cv.getPosition(), cv.getNormal(), cv.mTexCoord); +} + +void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, const LLVector2& tc) +{ + S32 new_verts = mNumVertices+1; + S32 new_size = new_verts*16; +// S32 old_size = mNumVertices*16; + + //positions + mPositions = (LLVector4a*) realloc(mPositions, new_size); + + //normals + mNormals = (LLVector4a*) realloc(mNormals, new_size); + + //tex coords + new_size = ((new_verts*8)+0xF) & ~0xF; + mTexCoords = (LLVector2*) realloc(mTexCoords, new_size); + + + //just clear binormals + ll_aligned_free_16(mBinormals); + mBinormals = NULL; + + mPositions[mNumVertices] = pos; + mNormals[mNumVertices] = norm; + mTexCoords[mNumVertices] = tc; + + mNumVertices++; +} + +void LLVolumeFace::allocateBinormals(S32 num_verts) +{ + ll_aligned_free_16(mBinormals); + mBinormals = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); +} + +void LLVolumeFace::resizeIndices(S32 num_indices) +{ + ll_aligned_free_16(mIndices); + + if (num_indices) + { + //pad index block end to allow for QWORD reads + S32 size = ((num_indices*sizeof(U16)) + 0xF) & ~0xF; + + mIndices = (U16*) ll_aligned_malloc_16(size); + } + else + { + mIndices = NULL; + } + + mNumIndices = num_indices; +} + +void LLVolumeFace::pushIndex(const U16& idx) +{ + S32 new_count = mNumIndices + 1; + S32 new_size = ((new_count*2)+0xF) & ~0xF; + + S32 old_size = ((mNumIndices*2)+0xF) & ~0xF; + if (new_size != old_size) + { + mIndices = (U16*) realloc(mIndices, new_size); + } + + mIndices[mNumIndices++] = idx; +} + +void LLVolumeFace::fillFromLegacyData(std::vector& v, std::vector& idx) +{ + resizeVertices(v.size()); + resizeIndices(idx.size()); + + for (U32 i = 0; i < v.size(); ++i) + { + mPositions[i] = v[i].getPosition(); + mNormals[i] = v[i].getNormal(); + mTexCoords[i] = v[i].mTexCoord; + } + + for (U32 i = 0; i < idx.size(); ++i) + { + mIndices[i] = idx[i]; + } +} + +void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMatrix4& norm_mat_in) +{ + U16 offset = mNumVertices; + + S32 new_count = face.mNumVertices + mNumVertices; + + if (new_count > 65536) + { + llerrs << "Cannot append face -- 16-bit overflow will occur." << llendl; + } + + if (face.mNumVertices == 0) + { + llerrs << "Cannot append empty face." << llendl; + } + + //allocate new buffer space + mPositions = (LLVector4a*) realloc(mPositions, new_count*sizeof(LLVector4a)); + assert_aligned(mPositions, 16); + mNormals = (LLVector4a*) realloc(mNormals, new_count*sizeof(LLVector4a)); + assert_aligned(mNormals, 16); + mTexCoords = (LLVector2*) realloc(mTexCoords, (new_count*sizeof(LLVector2)+0xF) & ~0xF); + assert_aligned(mTexCoords, 16); + + mNumVertices = new_count; + + //get destination address of appended face + LLVector4a* dst_pos = mPositions+offset; + LLVector2* dst_tc = mTexCoords+offset; + LLVector4a* dst_norm = mNormals+offset; + + //get source addresses of appended face + const LLVector4a* src_pos = face.mPositions; + const LLVector2* src_tc = face.mTexCoords; + const LLVector4a* src_norm = face.mNormals; + + //load aligned matrices + LLMatrix4a mat, norm_mat; + mat.loadu(mat_in); + norm_mat.loadu(norm_mat_in); + + for (U32 i = 0; i < (U32)face.mNumVertices; ++i) + { + //transform appended face position and store + mat.affineTransform(src_pos[i], dst_pos[i]); + + //transform appended face normal and store + norm_mat.rotate(src_norm[i], dst_norm[i]); + dst_norm[i].normalize3fast(); + + //copy appended face texture coordinate + dst_tc[i] = src_tc[i]; + + if (offset == 0 && i == 0) + { //initialize bounding box + mExtents[0] = mExtents[1] = dst_pos[i]; + } + else + { + //stretch bounding box + update_min_max(mExtents[0], mExtents[1], dst_pos[i]); + } + } + + + new_count = mNumIndices + face.mNumIndices; + + //allocate new index buffer + mIndices = (U16*) realloc(mIndices, (new_count*sizeof(U16)+0xF) & ~0xF); + + //get destination address into new index buffer + U16* dst_idx = mIndices+mNumIndices; + mNumIndices = new_count; + + for (U32 i = 0; i < (U32)face.mNumIndices; ++i) + { //copy indices, offsetting by old vertex count + dst_idx[i] = face.mIndices[i]+offset; } } @@ -5638,18 +6674,20 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) num_vertices = mNumS*mNumT; num_indices = (mNumS-1)*(mNumT-1)*6; - mVertices.resize(num_vertices); - if (!partial_build) { - mIndices.resize(num_indices); - mEdge.resize(num_indices); - } - else - { - mHasBinormals = FALSE; + resizeVertices(num_vertices); + resizeIndices(num_indices); + + //if ((volume->getParams().getSculptType() & LL_SCULPT_TYPE_MASK) != LL_SCULPT_TYPE_MESH) + { + mEdge.resize(num_indices); + } } + LLVector4a* pos = (LLVector4a*) mPositions; + LLVector4a* norm = (LLVector4a*) mNormals; + LLVector2* tc = (LLVector2*) mTexCoords; S32 begin_stex = llfloor( profile[mBeginS].mV[2] ); S32 num_s = ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2) ? mNumS/2 : mNumS; @@ -5700,21 +6738,20 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) i = mBeginS + s + max_s*t; } - mVertices[cur_vertex].mPosition = mesh[i].mPos; - mVertices[cur_vertex].mTexCoord = LLVector2(ss,tt); + pos[cur_vertex].load3(mesh[i].mPos.mV); + tc[cur_vertex] = LLVector2(ss,tt); - mVertices[cur_vertex].mNormal = LLVector3(0,0,0); - mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); - + norm[cur_vertex].clear(); cur_vertex++; if ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2 && s > 0) { - mVertices[cur_vertex].mPosition = mesh[i].mPos; - mVertices[cur_vertex].mTexCoord = LLVector2(ss,tt); + + pos[cur_vertex].load3(mesh[i].mPos.mV); + tc[cur_vertex] = LLVector2(ss,tt); - mVertices[cur_vertex].mNormal = LLVector3(0,0,0); - mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); + norm[cur_vertex].clear(); + cur_vertex++; } } @@ -5732,12 +6769,10 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) i = mBeginS + s + max_s*t; ss = profile[mBeginS + s].mV[2] - begin_stex; - mVertices[cur_vertex].mPosition = mesh[i].mPos; - mVertices[cur_vertex].mTexCoord = LLVector2(ss,tt); - - mVertices[cur_vertex].mNormal = LLVector3(0,0,0); - mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); - + pos[cur_vertex].load3(mesh[i].mPos.mV); + tc[cur_vertex] = LLVector2(ss,tt); + norm[cur_vertex].clear(); + cur_vertex++; } } @@ -5748,13 +6783,11 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) LLVector4a& face_max = mExtents[1]; mCenter->clear(); - face_max.load3(mVertices[0].mPosition.mV); - face_min = face_max; - for (U32 i = 1; i < mVertices.size(); ++i) + face_min = face_max = pos[0]; + + for (U32 i = 1; i < (U32)mNumVertices; ++i) { - LLVector4a pos; - pos.load3(mVertices[i].mPosition.mV); - update_min_max(face_min, face_max, pos); + update_min_max(face_min, face_max, pos[i]); } mCenter->setAdd(face_min, face_max); @@ -5821,38 +6854,58 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) } } + //clear normals + for (U32 i = 0; i < (U32)mNumVertices; i++) + { + mNormals[i].clear(); + } + //generate normals - for (U32 i = 0; i < mIndices.size()/3; i++) //for each triangle + for (U32 i = 0; i < (U32)mNumIndices/3; i++) //for each triangle { const U16* idx = &(mIndices[i*3]); - - VertexData* v[] = - { &mVertices[idx[0]], &mVertices[idx[1]], &mVertices[idx[2]] }; - + + + LLVector4a* v[] = + { pos+idx[0], pos+idx[1], pos+idx[2] }; + + LLVector4a* n[] = + { norm+idx[0], norm+idx[1], norm+idx[2] }; + //calculate triangle normal - LLVector3 norm = (v[0]->mPosition-v[1]->mPosition) % (v[0]->mPosition-v[2]->mPosition); - - v[0]->mNormal += norm; - v[1]->mNormal += norm; - v[2]->mNormal += norm; + LLVector4a a, b, c; + + a.setSub(*v[0], *v[1]); + b.setSub(*v[0], *v[2]); + c.setCross3(a,b); + n[0]->add(c); + n[1]->add(c); + n[2]->add(c); + //even out quad contributions - v[i%2+1]->mNormal += norm; + n[i%2+1]->add(c); } // adjust normals based on wrapping and stitching - BOOL s_bottom_converges = ((mVertices[0].mPosition - mVertices[mNumS*(mNumT-2)].mPosition).magVecSquared() < 0.000001f); - BOOL s_top_converges = ((mVertices[mNumS-1].mPosition - mVertices[mNumS*(mNumT-2)+mNumS-1].mPosition).magVecSquared() < 0.000001f); + LLVector4a top; + top.setSub(pos[0], pos[mNumS*(mNumT-2)]); + BOOL s_bottom_converges = (top.dot3(top) < 0.000001f); + + top.setSub(pos[mNumS-1], pos[mNumS*(mNumT-2)+mNumS-1]); + BOOL s_top_converges = (top.dot3(top) < 0.000001f); + if (sculpt_stitching == LL_SCULPT_TYPE_NONE) // logic for non-sculpt volumes { if (volume->getPath().isOpen() == FALSE) { //wrap normals on T for (S32 i = 0; i < mNumS; i++) { - LLVector3 norm = mVertices[i].mNormal + mVertices[mNumS*(mNumT-1)+i].mNormal; - mVertices[i].mNormal = norm; - mVertices[mNumS*(mNumT-1)+i].mNormal = norm; + LLVector4a n; + n.setAdd(norm[i], norm[mNumS*(mNumT-1)+i]); + norm[i] = n; + norm[mNumS*(mNumT-1)+i] = n; } } @@ -5860,9 +6913,10 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { //wrap normals on S for (S32 i = 0; i < mNumT; i++) { - LLVector3 norm = mVertices[mNumS*i].mNormal + mVertices[mNumS*i+mNumS-1].mNormal; - mVertices[mNumS * i].mNormal = norm; - mVertices[mNumS * i+mNumS-1].mNormal = norm; + LLVector4a n; + n.setAdd(norm[mNumS*i], norm[mNumS*i+mNumS-1]); + norm[mNumS * i] = n; + norm[mNumS * i+mNumS-1] = n; } } @@ -5873,7 +6927,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { //all lower S have same normal for (S32 i = 0; i < mNumT; i++) { - mVertices[mNumS*i].mNormal = LLVector3(1,0,0); + norm[mNumS*i].set(1,0,0); } } @@ -5881,12 +6935,11 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { //all upper S have same normal for (S32 i = 0; i < mNumT; i++) { - mVertices[mNumS*i+mNumS-1].mNormal = LLVector3(-1,0,0); + norm[mNumS*i+mNumS-1].set(-1,0,0); } } } } - else // logic for sculpt volumes { BOOL average_poles = FALSE; @@ -5909,30 +6962,33 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { // average normals for north pole - LLVector3 average(0.0, 0.0, 0.0); + LLVector4a average; + average.clear(); + for (S32 i = 0; i < mNumS; i++) { - average += mVertices[i].mNormal; + average.add(norm[i]); } // set average for (S32 i = 0; i < mNumS; i++) { - mVertices[i].mNormal = average; + norm[i] = average; } // average normals for south pole - average = LLVector3(0.0, 0.0, 0.0); + average.clear(); + for (S32 i = 0; i < mNumS; i++) { - average += mVertices[i + mNumS * (mNumT - 1)].mNormal; + average.add(norm[i + mNumS * (mNumT - 1)]); } // set average for (S32 i = 0; i < mNumS; i++) { - mVertices[i + mNumS * (mNumT - 1)].mNormal = average; + norm[i + mNumS * (mNumT - 1)] = average; } } @@ -5942,23 +6998,22 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { for (S32 i = 0; i < mNumT; i++) { - LLVector3 norm = mVertices[mNumS*i].mNormal + mVertices[mNumS*i+mNumS-1].mNormal; - mVertices[mNumS * i].mNormal = norm; - mVertices[mNumS * i+mNumS-1].mNormal = norm; + LLVector4a n; + n.setAdd(norm[mNumS*i], norm[mNumS*i+mNumS-1]); + norm[mNumS * i] = n; + norm[mNumS * i+mNumS-1] = n; } } - - if (wrap_t) { for (S32 i = 0; i < mNumS; i++) { - LLVector3 norm = mVertices[i].mNormal + mVertices[mNumS*(mNumT-1)+i].mNormal; - mVertices[i].mNormal = norm; - mVertices[mNumS*(mNumT-1)+i].mNormal = norm; + LLVector4a n; + n.setAdd(norm[i], norm[mNumS*(mNumT-1)+i]); + norm[i] = n; + norm[mNumS*(mNumT-1)+i] = n; } - } } @@ -5968,41 +7023,51 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) // Finds binormal based on three vertices with texture coordinates. // Fills in dummy values if the triangle has degenerate texture coordinates. -LLVector3 calc_binormal_from_triangle( - const LLVector3& pos0, +void calc_binormal_from_triangle(LLVector4a& binormal, + + const LLVector4a& pos0, const LLVector2& tex0, - const LLVector3& pos1, + const LLVector4a& pos1, const LLVector2& tex1, - const LLVector3& pos2, + const LLVector4a& pos2, const LLVector2& tex2) { - LLVector3 rx0( pos0.mV[VX], tex0.mV[VX], tex0.mV[VY] ); - LLVector3 rx1( pos1.mV[VX], tex1.mV[VX], tex1.mV[VY] ); - LLVector3 rx2( pos2.mV[VX], tex2.mV[VX], tex2.mV[VY] ); + LLVector4a rx0( pos0[VX], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a rx1( pos1[VX], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a rx2( pos2[VX], tex2.mV[VX], tex2.mV[VY] ); - LLVector3 ry0( pos0.mV[VY], tex0.mV[VX], tex0.mV[VY] ); - LLVector3 ry1( pos1.mV[VY], tex1.mV[VX], tex1.mV[VY] ); - LLVector3 ry2( pos2.mV[VY], tex2.mV[VX], tex2.mV[VY] ); + LLVector4a ry0( pos0[VY], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a ry1( pos1[VY], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a ry2( pos2[VY], tex2.mV[VX], tex2.mV[VY] ); - LLVector3 rz0( pos0.mV[VZ], tex0.mV[VX], tex0.mV[VY] ); - LLVector3 rz1( pos1.mV[VZ], tex1.mV[VX], tex1.mV[VY] ); - LLVector3 rz2( pos2.mV[VZ], tex2.mV[VX], tex2.mV[VY] ); + LLVector4a rz0( pos0[VZ], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a rz1( pos1[VZ], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a rz2( pos2[VZ], tex2.mV[VX], tex2.mV[VY] ); - LLVector3 r0 = (rx0 - rx1) % (rx0 - rx2); - LLVector3 r1 = (ry0 - ry1) % (ry0 - ry2); - LLVector3 r2 = (rz0 - rz1) % (rz0 - rz2); + LLVector4a lhs, rhs; - if( r0.mV[VX] && r1.mV[VX] && r2.mV[VX] ) + LLVector4a r0; + lhs.setSub(rx0, rx1); rhs.setSub(rx0, rx2); + r0.setCross3(lhs, rhs); + + LLVector4a r1; + lhs.setSub(ry0, ry1); rhs.setSub(ry0, ry2); + r1.setCross3(lhs, rhs); + + LLVector4a r2; + lhs.setSub(rz0, rz1); rhs.setSub(rz0, rz2); + r2.setCross3(lhs, rhs); + + if( r0[VX] && r1[VX] && r2[VX] ) { - LLVector3 binormal( - -r0.mV[VZ] / r0.mV[VX], - -r1.mV[VZ] / r1.mV[VX], - -r2.mV[VZ] / r2.mV[VX]); + binormal.set( + -r0[VZ] / r0[VX], + -r1[VZ] / r1[VX], + -r2[VZ] / r2[VX]); // binormal.normVec(); - return binormal; } else { - return LLVector3( 0, 1 , 0 ); + binormal.set( 0, 1 , 0 ); } } diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h index 03498881d..2f0dcb616 100644 --- a/indra/llmath/llvolume.h +++ b/indra/llmath/llvolume.h @@ -40,8 +40,13 @@ class LLPathParams; class LLVolumeParams; class LLProfile; class LLPath; + +template class LLOctreeNode; + +class LLVector4a; class LLVolumeFace; class LLVolume; +class LLVolumeTriangle; #include "lldarray.h" #include "lluuid.h" @@ -800,20 +805,84 @@ class LLVolumeFace public: class VertexData { + enum + { + POSITION = 0, + NORMAL = 1 + }; + + private: + void init(); public: - LLVector3 mPosition; - LLVector3 mNormal; - LLVector3 mBinormal; + VertexData(); + VertexData(const VertexData& rhs); + const VertexData& operator=(const VertexData& rhs); + + ~VertexData(); + LLVector4a& getPosition(); + LLVector4a& getNormal(); + const LLVector4a& getPosition() const; + const LLVector4a& getNormal() const; + void setPosition(const LLVector4a& pos); + void setNormal(const LLVector4a& norm); + + LLVector2 mTexCoord; + + bool operator<(const VertexData& rhs) const; + bool operator==(const VertexData& rhs) const; + bool compareNormal(const VertexData& rhs, F32 angle_cutoff) const; + + private: + LLVector4a* mData; }; + LLVolumeFace(); LLVolumeFace(const LLVolumeFace& src); LLVolumeFace& operator=(const LLVolumeFace& rhs); ~LLVolumeFace(); +private: + void freeData(); +public: + BOOL create(LLVolume* volume, BOOL partial_build = FALSE); void createBinormals(); + + void appendFace(const LLVolumeFace& face, LLMatrix4& transform, LLMatrix4& normal_tranform); + void resizeVertices(S32 num_verts); + void allocateBinormals(S32 num_verts); + void resizeIndices(S32 num_indices); + void fillFromLegacyData(std::vector& v, std::vector& idx); + + void pushVertex(const VertexData& cv); + void pushVertex(const LLVector4a& pos, const LLVector4a& norm, const LLVector2& tc); + void pushIndex(const U16& idx); + + void swapData(LLVolumeFace& rhs); + + void getVertexData(U16 indx, LLVolumeFace::VertexData& cv); + + class VertexMapData : public LLVolumeFace::VertexData + { + public: + U16 mIndex; + + bool operator==(const LLVolumeFace::VertexData& rhs) const; + + struct ComparePosition + { + bool operator()(const LLVector3& a, const LLVector3& b) const; + }; + + typedef std::map, VertexMapData::ComparePosition > PointMap; + }; + + void optimize(F32 angle_cutoff = 2.f); + void cacheOptimize(); + + void createOctree(F32 scaler = 0.25f, const LLVector4a& center = LLVector4a(0,0,0), const LLVector4a& size = LLVector4a(0.5f,0.5f,0.5f)); enum { @@ -833,7 +902,6 @@ public: public: S32 mID; U32 mTypeMask; - BOOL mHasBinormals; // Only used for INNER/OUTER faces S32 mBeginS; @@ -845,9 +913,17 @@ public: LLVector4a* mCenter; LLVector2 mTexCoordExtents[2]; //minimum and maximum of texture coordinates of the face. - std::vector mVertices; - std::vector mIndices; + S32 mNumVertices; + S32 mNumIndices; + + LLVector4a* mPositions; + LLVector4a* mNormals; + LLVector4a* mBinormals; + LLVector2* mTexCoords; + U16* mIndices; + std::vector mEdge; + LLOctreeNode* mOctree; private: BOOL createUnCutCubeCap(LLVolume* volume, BOOL partial_build = FALSE); @@ -882,7 +958,7 @@ public: U8 getProfileType() const { return mParams.getProfileParams().getCurveType(); } U8 getPathType() const { return mParams.getPathParams().getCurveType(); } - S32 getNumFaces() const { return (S32)mProfilep->mFaces.size(); } + S32 getNumFaces() const; S32 getNumVolumeFaces() const { return mVolumeFaces.size(); } F32 getDetail() const { return mDetail; } const LLVolumeParams& getParams() const { return mParams; } @@ -965,6 +1041,9 @@ 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(const LLVolume* volume); + void cacheOptimize(); + private: void sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, U8 sculpt_type); F32 sculptGetSurfaceArea(); @@ -993,12 +1072,13 @@ protected: std::ostream& operator<<(std::ostream &s, const LLVolumeParams &volume_params); -LLVector3 calc_binormal_from_triangle( - const LLVector3& pos0, +void calc_binormal_from_triangle( + LLVector4a& binormal, + const LLVector4a& pos0, const LLVector2& tex0, - const LLVector3& pos1, + const LLVector4a& pos1, const LLVector2& tex1, - const LLVector3& pos2, + const LLVector4a& pos2, const LLVector2& tex2); BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* center, const F32* size); diff --git a/indra/llmath/llvolumeoctree.cpp b/indra/llmath/llvolumeoctree.cpp index 0ac6b1350..a420c8e51 100644 --- a/indra/llmath/llvolumeoctree.cpp +++ b/indra/llmath/llvolumeoctree.cpp @@ -74,8 +74,6 @@ BOOL LLLineSegmentBoxIntersect(const LLVector4a& start, const LLVector4a& end, c return (grt & 0x7) ? false : true; } - -#if 0 //MESH LLVolumeOctreeListener::LLVolumeOctreeListener(LLOctreeNode* node) { node->addListener(this); @@ -122,7 +120,7 @@ void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode 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) + for (U32 i = 0; i < node->getChildCount(); ++i) { traverse(node->getChild(i)); } @@ -253,5 +251,5 @@ void LLVolumeOctreeValidate::visit(const LLOctreeNode* branch) } } } -#endif + diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h index 31da2acbf..688d91dc4 100644 --- a/indra/llmath/llvolumeoctree.h +++ b/indra/llmath/llvolumeoctree.h @@ -34,7 +34,6 @@ #include "llvolume.h" #include "llvector4a.h" -#if 0 //MESH class LLVolumeTriangle : public LLRefCount { public: @@ -131,5 +130,5 @@ class LLVolumeOctreeValidate : public LLOctreeTraveler { virtual void visit(const LLOctreeNode* branch); }; -#endif //0 + #endif diff --git a/indra/llui/lllineeditor.h b/indra/llui/lllineeditor.h index 2272c450b..8c8229050 100644 --- a/indra/llui/lllineeditor.h +++ b/indra/llui/lllineeditor.h @@ -431,7 +431,8 @@ public: void setText(const LLStringExplicit &new_text) { mSearchEdit->setText(new_text); } - void setSearchCallback(void (*search_callback)(const std::string& search_string, void* user_data), void* data) { mSearchCallback = search_callback; mCallbackUserData = data; } + typedef boost::function search_callback_t; + void setSearchCallback(search_callback_t cb,void *user_data) { mSearchCallback = boost::bind(cb,_1,user_data); } // LLUICtrl interface virtual void setValue(const LLSD& value ); @@ -446,7 +447,8 @@ private: LLLineEditor* mSearchEdit; class LLButton* mClearSearchButton; - void (*mSearchCallback)(const std::string& search_string, void* user_data); + + search_callback_t mSearchCallback; }; diff --git a/indra/newview/floatersculptpreview.cpp b/indra/newview/floatersculptpreview.cpp index da8d33331..c6b85561f 100644 --- a/indra/newview/floatersculptpreview.cpp +++ b/indra/newview/floatersculptpreview.cpp @@ -730,8 +730,8 @@ void LLPreviewSculpted::setPreviewTarget(LLImageRaw* imagep, F32 distance) } const LLVolumeFace &vf = mVolume->getVolumeFace(0); - U32 num_indices = vf.mIndices.size(); - U32 num_vertices = vf.mVertices.size(); + U32 num_indices = vf.mNumIndices; + U32 num_vertices = vf.mNumVertices; mVertexBuffer = new LLVertexBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_NORMAL, 0); mVertexBuffer->allocateBuffer(num_vertices, num_indices, TRUE); @@ -747,8 +747,8 @@ void LLPreviewSculpted::setPreviewTarget(LLImageRaw* imagep, F32 distance) // build vertices and normals for (U32 i = 0; (S32)i < num_vertices; i++) { - *(vertex_strider++) = vf.mVertices[i].mPosition; - LLVector3 normal = vf.mVertices[i].mNormal; + (vertex_strider++)->set(vf.mPositions[i].getF32ptr()); + LLVector3 normal(vf.mNormals[i].getF32ptr()); normal.normalize(); *(normal_strider++) = normal; } @@ -812,7 +812,7 @@ BOOL LLPreviewSculpted::render() LLViewerCamera::getInstance()->setPerspective(FALSE, mOrigin.mX, mOrigin.mY, mFullWidth, mFullHeight, FALSE); const LLVolumeFace &vf = mVolume->getVolumeFace(0); - U32 num_indices = vf.mIndices.size(); + U32 num_indices = vf.mNumIndices; mVertexBuffer->setBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_NORMAL); diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 918637a23..f977b41c4 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -924,10 +924,8 @@ void LLFace::getPlanarProjectedParams(LLQuaternion* face_rot, LLVector3* face_po { const LLMatrix4& vol_mat = getWorldMatrix(); const LLVolumeFace& vf = getViewerObject()->getVolume()->getVolumeFace(mTEOffset); - LLVector4a normal4a; - normal4a.load3(vf.mVertices[0].mNormal.mV); - LLVector4a binormal4a; - binormal4a.load3(vf.mVertices[0].mBinormal.mV); + const LLVector4a& normal4a = vf.mNormals[0]; + const LLVector4a& binormal4a = vf.mBinormals[0]; LLVector2 projected_binormal; planarProjection(projected_binormal, normal4a, *vf.mCenter, binormal4a); projected_binormal -= LLVector2(0.5f, 0.5f); // this normally happens in xform() @@ -1041,14 +1039,14 @@ bool LLFace::canRenderAsMask() BOOL LLFace::getGeometryVolume(const LLVolume& volume, const S32 &f, - const LLMatrix4& mat_vert, const LLMatrix3& mat_normal, + const LLMatrix4& mat_vert_in, const LLMatrix3& mat_norm_in, const U16 &index_offset, bool force_rebuild) { llassert(verify()); const LLVolumeFace &vf = volume.getVolumeFace(f); - S32 num_vertices = (S32)vf.mVertices.size(); - S32 num_indices = (S32)vf.mIndices.size(); + S32 num_vertices = (S32)vf.mNumVertices; + S32 num_indices = (S32) vf.mNumIndices; if (mVertexBuffer.notNull()) { @@ -1081,7 +1079,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, LLStrider binormals; LLStrider indicesp; - BOOL full_rebuild = mDrawablep->isState(LLDrawable::REBUILD_VOLUME); + BOOL full_rebuild = force_rebuild || mDrawablep->isState(LLDrawable::REBUILD_VOLUME); BOOL global_volume = mDrawablep->getVOVolume()->isVolumeGlobal(); LLVector3 scale; @@ -1094,11 +1092,12 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, scale = mVObjp->getScale(); } - BOOL rebuild_pos = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_POSITION); - BOOL rebuild_color = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_COLOR); - BOOL rebuild_tcoord = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_TCOORD); - BOOL rebuild_normal = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL); - BOOL rebuild_binormal = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_BINORMAL); + bool rebuild_pos = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_POSITION); + bool rebuild_color = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_COLOR); + bool rebuild_tcoord = full_rebuild || mDrawablep->isState(LLDrawable::REBUILD_TCOORD); + bool rebuild_normal = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL); + bool rebuild_binormal = rebuild_pos && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_BINORMAL); + const LLTextureEntry *tep = mVObjp->getTE(f); if (!tep) rebuild_color = FALSE; // can't get color when tep is NULL @@ -1130,7 +1129,6 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, mVertexBuffer->getColorStrider(colors, mGeomIndex); } - F32 r = 0, os = 0, ot = 0, ms = 0, mt = 0, cos_ang = 0, sin_ang = 0; BOOL is_static = mDrawablep->isStatic(); BOOL is_global = is_static; @@ -1146,59 +1144,6 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, clearState(GLOBAL); } - LLVector2 tmin, tmax; - - - - if (rebuild_tcoord) - { - if (tep) - { - r = tep->getRotation(); - os = tep->mOffsetS; - ot = tep->mOffsetT; - ms = tep->mScaleS; - mt = tep->mScaleT; - cos_ang = cos(r); - sin_ang = sin(r); - } - else - { - cos_ang = 1.0f; - sin_ang = 0.0f; - os = 0.0f; - ot = 0.0f; - ms = 1.0f; - mt = 1.0f; - } - } - - U8 tex_mode = 0; - - if (isState(TEXTURE_ANIM)) - { - LLVOVolume* vobj = (LLVOVolume*) (LLViewerObject*) mVObjp; - tex_mode = vobj->mTexAnimMode; - - if (!tex_mode) - { - clearState(TEXTURE_ANIM); - } - else - { - os = ot = 0.f; - r = 0.f; - cos_ang = 1.f; - sin_ang = 0.f; - ms = mt = 1.f; - } - - if (getVirtualSize() >= MIN_TEX_ANIM_SIZE) - { //don't override texture transform during tc bake - tex_mode = 0; - } - } - LLColor4U color = (tep ? LLColor4U(tep->getColor()) : LLColor4U::white); if (rebuild_color) // FALSE if tep == NULL @@ -1226,167 +1171,361 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, mVertexBuffer->getIndexStrider(indicesp, mIndicesIndex); for (U32 i = 0; i < (U32) num_indices; i++) { - *indicesp++ = vf.mIndices[i] + index_offset; + indicesp[i] = vf.mIndices[i] + index_offset; } } + LLMatrix4a mat_normal; + mat_normal.loadu(mat_norm_in); - //bump setup - LLVector3 binormal_dir( -sin_ang, cos_ang, 0 ); - LLVector3 bump_s_primary_light_ray; - LLVector3 bump_t_primary_light_ray; + //if it's not fullbright and has no normals, bake sunlight based on face normal + //bool bake_sunlight = !getTextureEntry()->getFullbright() && + // !mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_NORMAL); - LLQuaternion bump_quat; - if (mDrawablep->isActive()) + F32 r = 0, os = 0, ot = 0, ms = 0, mt = 0, cos_ang = 0, sin_ang = 0; + + if (rebuild_tcoord) { - bump_quat = LLQuaternion(mDrawablep->getRenderMatrix()); - } - - if (bump_code) - { - mVObjp->getVolume()->genBinormals(f); - F32 offset_multiple; - switch( bump_code ) + bool do_xform; + + if (tep) { - case BE_NO_BUMP: - offset_multiple = 0.f; - break; - case BE_BRIGHTNESS: - case BE_DARKNESS: - if( mTexture.notNull() && mTexture->hasGLTexture()) + r = tep->getRotation(); + os = tep->mOffsetS; + ot = tep->mOffsetT; + ms = tep->mScaleS; + mt = tep->mScaleT; + cos_ang = cos(r); + sin_ang = sin(r); + + if (cos_ang != 1.f || + sin_ang != 0.f || + os != 0.f || + ot != 0.f || + ms != 1.f || + mt != 1.f) { - // Offset by approximately one texel - S32 cur_discard = mTexture->getDiscardLevel(); - S32 max_size = llmax( mTexture->getWidth(), mTexture->getHeight() ); - max_size <<= cur_discard; - const F32 ARTIFICIAL_OFFSET = 2.f; - offset_multiple = ARTIFICIAL_OFFSET / (F32)max_size; + do_xform = true; } else { - offset_multiple = 1.f/256; - } - break; - - default: // Standard bumpmap textures. Assumed to be 256x256 - offset_multiple = 1.f / 256; - break; - } - - F32 s_scale = 1.f; - F32 t_scale = 1.f; - if( tep ) - { - tep->getScale( &s_scale, &t_scale ); - } - // Use the nudged south when coming from above sun angle, such - // that emboss mapping always shows up on the upward faces of cubes when - // it's noon (since a lot of builders build with the sun forced to noon). - LLVector3 sun_ray = gSky.mVOSkyp->mBumpSunDir; - LLVector3 moon_ray = gSky.getMoonDirection(); - LLVector3& primary_light_ray = (sun_ray.mV[VZ] > 0) ? sun_ray : moon_ray; - - bump_s_primary_light_ray = offset_multiple * s_scale * primary_light_ray; - bump_t_primary_light_ray = offset_multiple * t_scale * primary_light_ray; - } - - U8 texgen = getTextureEntry()->getTexGen(); - if (rebuild_tcoord && texgen != LLTextureEntry::TEX_GEN_DEFAULT) - { //planar texgen needs binormals - mVObjp->getVolume()->genBinormals(f); - } - - LLVector4a scalea; - scalea.load3(scale.mV); - - for (S32 i = 0; i < num_vertices; i++) - { - if (rebuild_tcoord) - { - LLVector2 tc = vf.mVertices[i].mTexCoord; - - if (texgen != LLTextureEntry::TEX_GEN_DEFAULT) - { - LLVector4a vec; - vec.load3(vf.mVertices[i].mPosition.mV); - vec.mul(scalea); - LLVector4a norm; - norm.load3(vf.mVertices[i].mNormal.mV); - - switch (texgen) - { - case LLTextureEntry::TEX_GEN_PLANAR: - planarProjection(tc, norm, *(vf.mCenter), vec); - break; - case LLTextureEntry::TEX_GEN_SPHERICAL: - sphericalProjection(tc, norm, *(vf.mCenter), vec); - break; - case LLTextureEntry::TEX_GEN_CYLINDRICAL: - cylindricalProjection(tc, norm, *(vf.mCenter), vec); - break; - default: - break; - } - } - - if (tex_mode && mTextureMatrix) - { - LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f); - tmp = tmp * *mTextureMatrix; - tc.mV[0] = tmp.mV[0]; - tc.mV[1] = tmp.mV[1]; - } - else - { - xform(tc, cos_ang, sin_ang, os, ot, ms, mt); - } - - *tex_coords++ = tc; - - if (bump_code && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TEXCOORD1)) - { - LLVector3 tangent = vf.mVertices[i].mBinormal % vf.mVertices[i].mNormal; - - LLMatrix3 tangent_to_object; - tangent_to_object.setRows(tangent, vf.mVertices[i].mBinormal, vf.mVertices[i].mNormal); - LLVector3 binormal = binormal_dir * tangent_to_object; - binormal = binormal * mat_normal; - - if (mDrawablep->isActive()) - { - binormal *= bump_quat; - } - - binormal.normVec(); - tc += LLVector2( bump_s_primary_light_ray * tangent, bump_t_primary_light_ray * binormal ); - - *tex_coords2++ = tc; + do_xform = false; } } + else + { + do_xform = false; + } + + //bump setup + LLVector4a binormal_dir( -sin_ang, cos_ang, 0.f ); + LLVector4a bump_s_primary_light_ray(0.f, 0.f, 0.f); + LLVector4a bump_t_primary_light_ray(0.f, 0.f, 0.f); + + LLQuaternion bump_quat; + if (mDrawablep->isActive()) + { + bump_quat = LLQuaternion(mDrawablep->getRenderMatrix()); + } + + if (bump_code) + { + mVObjp->getVolume()->genBinormals(f); + F32 offset_multiple; + switch( bump_code ) + { + case BE_NO_BUMP: + offset_multiple = 0.f; + break; + case BE_BRIGHTNESS: + case BE_DARKNESS: + if( mTexture.notNull() && mTexture->hasGLTexture()) + { + // Offset by approximately one texel + S32 cur_discard = mTexture->getDiscardLevel(); + S32 max_size = llmax( mTexture->getWidth(), mTexture->getHeight() ); + max_size <<= cur_discard; + const F32 ARTIFICIAL_OFFSET = 2.f; + offset_multiple = ARTIFICIAL_OFFSET / (F32)max_size; + } + else + { + offset_multiple = 1.f/256; + } + break; + + default: // Standard bumpmap textures. Assumed to be 256x256 + offset_multiple = 1.f / 256; + break; + } + + F32 s_scale = 1.f; + F32 t_scale = 1.f; + if( tep ) + { + tep->getScale( &s_scale, &t_scale ); + } + // Use the nudged south when coming from above sun angle, such + // that emboss mapping always shows up on the upward faces of cubes when + // it's noon (since a lot of builders build with the sun forced to noon). + LLVector3 sun_ray = gSky.mVOSkyp->mBumpSunDir; + LLVector3 moon_ray = gSky.getMoonDirection(); + LLVector3& primary_light_ray = (sun_ray.mV[VZ] > 0) ? sun_ray : moon_ray; + + bump_s_primary_light_ray.load3((offset_multiple * s_scale * primary_light_ray).mV); + bump_t_primary_light_ray.load3((offset_multiple * t_scale * primary_light_ray).mV); + } + + U8 texgen = getTextureEntry()->getTexGen(); + if (rebuild_tcoord && texgen != LLTextureEntry::TEX_GEN_DEFAULT) + { //planar texgen needs binormals + mVObjp->getVolume()->genBinormals(f); + } + + U8 tex_mode = 0; + + if (isState(TEXTURE_ANIM)) + { + LLVOVolume* vobj = (LLVOVolume*) (LLViewerObject*) mVObjp; + tex_mode = vobj->mTexAnimMode; + + if (!tex_mode) + { + clearState(TEXTURE_ANIM); + } + else + { + os = ot = 0.f; + r = 0.f; + cos_ang = 1.f; + sin_ang = 0.f; + ms = mt = 1.f; + + do_xform = false; + } + + if (getVirtualSize() >= MIN_TEX_ANIM_SIZE) + { //don't override texture transform during tc bake + tex_mode = 0; + } + } + + LLVector4a scalea; + scalea.load3(scale.mV); + + bool do_bump = bump_code && mVertexBuffer->hasDataType(LLVertexBuffer::TYPE_TEXCOORD1); + bool do_tex_mat = tex_mode && mTextureMatrix; + + if (!do_bump) + { //not in atlas or not bump mapped, might be able to do a cheap update + if (texgen != LLTextureEntry::TEX_GEN_PLANAR) + { + if (!do_tex_mat) + { + /*if (!do_xform) + { + LLVector4a::memcpyNonAliased16((F32*) tex_coords.get(), (F32*) vf.mTexCoords, num_vertices*2*sizeof(F32)); + } + else*/ + { + for (S32 i = 0; i < num_vertices; i++) + { + LLVector2 tc(vf.mTexCoords[i]); + if(do_xform) + xform(tc, cos_ang, sin_ang, os, ot, ms, mt); + *tex_coords++ = tc; + } + } + } + else + { //do tex mat, no texgen, no atlas, no bump + for (S32 i = 0; i < num_vertices; i++) + { + LLVector2 tc(vf.mTexCoords[i]); + //LLVector4a& norm = vf.mNormals[i]; + //LLVector4a& center = *(vf.mCenter); + + LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f); + tmp = tmp * *mTextureMatrix; + tc.mV[0] = tmp.mV[0]; + tc.mV[1] = tmp.mV[1]; + *tex_coords++ = tc; + } + } + } + else + { //no bump, no atlas, tex gen planar + if (do_tex_mat) + { + for (S32 i = 0; i < num_vertices; i++) + { + LLVector2 tc(vf.mTexCoords[i]); + LLVector4a& norm = vf.mNormals[i]; + LLVector4a& center = *(vf.mCenter); + LLVector4a vec = vf.mPositions[i]; + vec.mul(scalea); + planarProjection(tc, norm, center, vec); + + LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f); + tmp = tmp * *mTextureMatrix; + tc.mV[0] = tmp.mV[0]; + tc.mV[1] = tmp.mV[1]; + + *tex_coords++ = tc; + } + } + else + { + for (S32 i = 0; i < num_vertices; i++) + { + LLVector2 tc(vf.mTexCoords[i]); + LLVector4a& norm = vf.mNormals[i]; + LLVector4a& center = *(vf.mCenter); + LLVector4a vec = vf.mPositions[i]; + vec.mul(scalea); + planarProjection(tc, norm, center, vec); + + xform(tc, cos_ang, sin_ang, os, ot, ms, mt); + + *tex_coords++ = tc; + } + } + } + } + else + { //either bump mapped or in atlas, just do the whole expensive loop + + std::vector bump_tc; + + for (S32 i = 0; i < num_vertices; i++) + { + LLVector2 tc(vf.mTexCoords[i]); - if (rebuild_pos) - { - *vertices++ = vf.mVertices[i].mPosition * mat_vert; - } + LLVector4a& norm = vf.mNormals[i]; + + LLVector4a& center = *(vf.mCenter); + + if (texgen != LLTextureEntry::TEX_GEN_DEFAULT) + { + LLVector4a vec = vf.mPositions[i]; + + vec.mul(scalea); + + switch (texgen) + { + case LLTextureEntry::TEX_GEN_PLANAR: + planarProjection(tc, norm, center, vec); + break; + case LLTextureEntry::TEX_GEN_SPHERICAL: + sphericalProjection(tc, norm, center, vec); + break; + case LLTextureEntry::TEX_GEN_CYLINDRICAL: + cylindricalProjection(tc, norm, center, vec); + break; + default: + break; + } + } + + if (tex_mode && mTextureMatrix) + { + LLVector3 tmp(tc.mV[0], tc.mV[1], 0.f); + tmp = tmp * *mTextureMatrix; + tc.mV[0] = tmp.mV[0]; + tc.mV[1] = tmp.mV[1]; + } + else + { + xform(tc, cos_ang, sin_ang, os, ot, ms, mt); + } + + + *tex_coords++ = tc; + if (do_bump) + { + bump_tc.push_back(tc); + } + } + + + if (do_bump) + { - if (rebuild_normal) + for (S32 i = 0; i < num_vertices; i++) + { + LLVector4a tangent; + tangent.setCross3(vf.mBinormals[i], vf.mNormals[i]); + + LLMatrix4a tangent_to_object; + tangent_to_object.setRows(tangent, vf.mBinormals[i], vf.mNormals[i]); + LLVector4a t; + tangent_to_object.rotate(binormal_dir, t); + LLVector4a binormal; + mat_normal.rotate(t, binormal); + + //VECTORIZE THIS + if (mDrawablep->isActive()) + { + LLVector3 t; + t.set(binormal.getF32ptr()); + t *= bump_quat; + binormal.load3(t.mV); + } + + binormal.normalize3fast(); + LLVector2 tc = bump_tc[i]; + tc += LLVector2( bump_s_primary_light_ray.dot3(tangent).getF32(), bump_t_primary_light_ray.dot3(binormal).getF32() ); + + *tex_coords2++ = tc; + } + } + } + } + + if (rebuild_pos) + { + LLMatrix4a mat_vert; + mat_vert.loadu(mat_vert_in); + + LLVector4a* src = vf.mPositions; + LLVector4a position; + for (S32 i = 0; i < num_vertices; i++) { - LLVector3 normal = vf.mVertices[i].mNormal * mat_normal; - normal.normVec(); + mat_vert.affineTransform(src[i], position); + vertices[i].set(position.getF32ptr()); + } - *normals++ = normal; - } + + } - if (rebuild_binormal) - { - LLVector3 binormal = vf.mVertices[i].mBinormal * mat_normal; - binormal.normVec(); - *binormals++ = binormal; + if (rebuild_normal) + { + for (S32 i = 0; i < num_vertices; i++) + { + LLVector4a normal; + mat_normal.rotate(vf.mNormals[i], normal); + normal.normalize3fast(); + normals[i].set(normal.getF32ptr()); } + } - if (rebuild_color) + if (rebuild_binormal) + { + for (S32 i = 0; i < num_vertices; i++) + { + LLVector4a binormal; + mat_normal.rotate(vf.mBinormals[i], binormal); + binormal.normalize3fast(); + binormals[i].set(binormal.getF32ptr()); + } + } + + + if (rebuild_color) + { + for (S32 i = 0; i < num_vertices; i++) { - *colors++ = color; + colors[i] = color; } } @@ -1395,7 +1534,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, mTexExtents[0].setVec(0,0); mTexExtents[1].setVec(1,1); xform(mTexExtents[0], cos_ang, sin_ang, os, ot, ms, mt); - xform(mTexExtents[1], cos_ang, sin_ang, os, ot, ms, mt); + xform(mTexExtents[1], cos_ang, sin_ang, os, ot, ms, mt); F32 es = vf.mTexCoordExtents[1].mV[0] - vf.mTexCoordExtents[0].mV[0] ; F32 et = vf.mTexCoordExtents[1].mV[1] - vf.mTexCoordExtents[0].mV[1] ; diff --git a/indra/newview/llfloaterimagepreview.cpp b/indra/newview/llfloaterimagepreview.cpp index 27cd5055d..2fc241cc4 100644 --- a/indra/newview/llfloaterimagepreview.cpp +++ b/indra/newview/llfloaterimagepreview.cpp @@ -348,24 +348,7 @@ bool LLFloaterImagePreview::loadImage(const std::string& src_filename) { std::string exten = gDirUtilp->getExtension(src_filename); - U32 codec = IMG_CODEC_INVALID; - std::string temp_str; - if( exten == "bmp") - { - codec = IMG_CODEC_BMP; - } - else if( exten == "tga") - { - codec = IMG_CODEC_TGA; - } - else if( exten == "jpg" || exten == "jpeg") - { - codec = IMG_CODEC_JPEG; - } - else if( exten == "png" ) - { - codec = IMG_CODEC_PNG; - } + U32 codec = LLImageBase::getCodecFromExtension(exten); LLPointer raw_image = new LLImageRaw; @@ -854,8 +837,8 @@ void LLImagePreviewSculpted::setPreviewTarget(LLImageRaw* imagep, F32 distance) } const LLVolumeFace &vf = mVolume->getVolumeFace(0); - U32 num_indices = vf.mIndices.size(); - U32 num_vertices = vf.mVertices.size(); + U32 num_indices = vf.mNumIndices; + U32 num_vertices = vf.mNumVertices; mVertexBuffer = new LLVertexBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_NORMAL, 0); mVertexBuffer->allocateBuffer(num_vertices, num_indices, TRUE); @@ -869,12 +852,12 @@ void LLImagePreviewSculpted::setPreviewTarget(LLImageRaw* imagep, F32 distance) mVertexBuffer->getIndexStrider(index_strider); // build vertices and normals - for (U32 i = 0; (S32)i < num_vertices; i++) + for (U32 i = 0; (U32)i < num_vertices; i++) { - *(vertex_strider++) = vf.mVertices[i].mPosition; - LLVector3 normal = vf.mVertices[i].mNormal; - normal.normalize(); - *(normal_strider++) = normal; + (vertex_strider++)->set(vf.mPositions[i].getF32ptr()); + LLVector4a normal(vf.mNormals[i]); + normal.normalize3fast(); + (normal_strider++)->set(normal.getF32ptr()); } // build indices @@ -936,7 +919,7 @@ BOOL LLImagePreviewSculpted::render() LLViewerCamera::getInstance()->setPerspective(FALSE, mOrigin.mX, mOrigin.mY, mFullWidth, mFullHeight, FALSE); const LLVolumeFace &vf = mVolume->getVolumeFace(0); - U32 num_indices = vf.mIndices.size(); + U32 num_indices = (U32)vf.mNumIndices; mVertexBuffer->setBuffer(LLVertexBuffer::MAP_VERTEX | LLVertexBuffer::MAP_NORMAL); @@ -948,8 +931,7 @@ BOOL LLImagePreviewSculpted::render() gGL.color3f(BRIGHTNESS, BRIGHTNESS, BRIGHTNESS); mVertexBuffer->draw(LLRender::TRIANGLES, num_indices, 0); - gGL.popMatrix(); - + gGL.popMatrix(); return TRUE; } diff --git a/indra/newview/llpolymesh.cpp b/indra/newview/llpolymesh.cpp index 3a82e7304..7c433cb43 100644 --- a/indra/newview/llpolymesh.cpp +++ b/indra/newview/llpolymesh.cpp @@ -1426,20 +1426,24 @@ BOOL LLPolyMesh::loadOBJ(LLFILE *fp) S32 f1 = faces[f][1]; S32 f2 = faces[f][2]; - LLVector3& v0 = coords[f0]; - LLVector3& v1 = coords[f1]; - LLVector3& v2 = coords[f2]; - + LLVector4a v0; + LLVector4a v1; + LLVector4a v2; + v0.load3(coords[f0].mV);; + v1.load3(coords[f1].mV);; + v2.load3(coords[f2].mV);; LLVector2& t0 = tex[f0]; LLVector2& t1 = tex[f1]; LLVector2& t2 = tex[f2]; - LLVector3 binorm = calc_binormal_from_triangle(v0, t0, v1, t1, v2, t2); - binorm.normVec(); + LLVector4a binorm; + calc_binormal_from_triangle(binorm, v0, t0, v1, t1, v2, t2); + binorm.normalize3fast(); - binormals[f0] += binorm; - binormals[f1] += binorm; - binormals[f2] += binorm; + LLVector3 binormdelta(binorm.getF32ptr()); + binormals[f0] += binormdelta; + binormals[f1] += binormdelta; + binormals[f2] += binormdelta; } for ( v=0; vgetNumVolumeFaces(); ++i) { const LLVolumeFace& face = volume->getVolumeFace(i); - glVertexPointer(3, GL_FLOAT, 16, face.mVertices[i].mPosition.mV); - //This probably doesn't work. - glDrawElements(GL_TRIANGLES, face.mIndices.size(), GL_UNSIGNED_SHORT, (GLvoid*)face.mIndices[i]); + glVertexPointer(3, GL_FLOAT, 16, face.mPositions); + glDrawElements(GL_TRIANGLES, face.mNumIndices, GL_UNSIGNED_SHORT, face.mIndices); } } diff --git a/indra/newview/llspatialpartition.h b/indra/newview/llspatialpartition.h index fd2075092..bc11da84d 100644 --- a/indra/newview/llspatialpartition.h +++ b/indra/newview/llspatialpartition.h @@ -71,6 +71,18 @@ protected: ~LLDrawInfo(); public: + + LLDrawInfo(const LLDrawInfo& rhs) + { + *this = rhs; + } + + const LLDrawInfo& operator=(const LLDrawInfo& rhs) + { + llerrs << "Illegal operation!" << llendl; + return *this; + } + LLDrawInfo(U16 start, U16 end, U32 count, U32 offset, LLViewerTexture* image, LLVertexBuffer* buffer, BOOL fullbright = FALSE, U8 bump = 0, BOOL particle = FALSE, F32 part_size = 0); @@ -164,6 +176,18 @@ class LLSpatialGroup : public LLOctreeListener friend class LLSpatialPartition; friend class LLOctreeStateCheck; public: + + LLSpatialGroup(const LLSpatialGroup& rhs) + { + *this = rhs; + } + + const LLSpatialGroup& operator=(const LLSpatialGroup& rhs) + { + llerrs << "Illegal operation!" << llendl; + return *this; + } + static std::set sPendingQueries; //pending occlusion queries static U32 sNodeCount; static BOOL sNoDelete; //deletion of spatial groups and draw info not allowed if TRUE @@ -173,7 +197,7 @@ public: typedef std::vector > drawmap_elem_t; typedef std::map draw_map_t; typedef std::vector > buffer_list_t; - typedef std::map, buffer_list_t> buffer_texture_map_t; + typedef std::map buffer_texture_map_t; typedef std::map buffer_map_t; typedef LLOctreeListener BaseType; diff --git a/indra/newview/llviewercamera.cpp b/indra/newview/llviewercamera.cpp index efdf9e1fe..f08404069 100644 --- a/indra/newview/llviewercamera.cpp +++ b/indra/newview/llviewercamera.cpp @@ -773,10 +773,9 @@ BOOL LLViewerCamera::areVertsVisible(LLViewerObject* volumep, BOOL all_verts) { const LLVolumeFace& face = volume->getVolumeFace(i); - for (U32 v = 0; v < face.mVertices.size(); v++) + for (U32 v = 0; v < (U32)face.mNumVertices; v++) { - LLVector4a src_vec; - src_vec.load3(face.mVertices[v].mPosition.mV); + const LLVector4a& src_vec = face.mPositions[v]; LLVector4a vec; mata.affineTransform(src_vec, vec); diff --git a/indra/newview/llvotextbubble.cpp b/indra/newview/llvotextbubble.cpp index fcbf93367..5b46ec681 100644 --- a/indra/newview/llvotextbubble.cpp +++ b/indra/newview/llvotextbubble.cpp @@ -216,7 +216,7 @@ void LLVOTextBubble::updateFaceSize(S32 idx) else { const LLVolumeFace& vol_face = getVolume()->getVolumeFace(idx); - face->setSize(vol_face.mVertices.size(), vol_face.mIndices.size()); + face->setSize(vol_face.mNumVertices, vol_face.mNumIndices); } } @@ -234,19 +234,27 @@ void LLVOTextBubble::getGeometry(S32 idx, const LLVolumeFace& face = getVolume()->getVolumeFace(idx); - LLVector3 pos = getPositionAgent(); + LLVector4a pos; + pos.load3(getPositionAgent().mV); + + LLVector4a scale; + scale.load3(getScale().mV); + LLColor4U color = LLColor4U(getTE(idx)->getColor()); U32 offset = mDrawable->getFace(idx)->getGeomIndex(); - for (U32 i = 0; i < face.mVertices.size(); i++) + for (U32 i = 0; i < (U32)face.mNumVertices; i++) { - *verticesp++ = face.mVertices[i].mPosition.scaledVec(getScale()) + pos; - *normalsp++ = face.mVertices[i].mNormal; - *texcoordsp++ = face.mVertices[i].mTexCoord; + LLVector4a vertpos; + vertpos.setMul(face.mPositions[i],scale); + vertpos.add(pos); + (verticesp++)->set(vertpos.getF32ptr()); + (normalsp++)->set(face.mNormals[i].getF32ptr()); + *texcoordsp++ = face.mTexCoords[i]; *colorsp++ = color; } - for (U32 i = 0; i < face.mIndices.size(); i++) + for (U32 i = 0; i < (U32)face.mNumIndices; i++) { *indicesp++ = face.mIndices[i] + offset; } diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index 55e898888..421be412f 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -1342,7 +1342,9 @@ void LLVOVolume::updateFaceSize(S32 idx) else { const LLVolumeFace& vol_face = getVolume()->getVolumeFace(idx); - facep->setSize(vol_face.mVertices.size(), vol_face.mIndices.size(),true); + facep->setSize(vol_face.mNumVertices, vol_face.mNumIndices, + true); // <--- volume faces should be padded for 16-byte alignment + } } @@ -2033,6 +2035,49 @@ const LLMatrix4 LLVOVolume::getRenderMatrix() const return mDrawable->getWorldMatrix(); } + +U32 LLVOVolume::getTriangleCount() +{ + U32 count = 0; + LLVolume* volume = getVolume(); + if (volume) + { + count = volume->getNumTriangles(); + } + + return count; +} + +U32 LLVOVolume::getHighLODTriangleCount() +{ + U32 ret = 0; + + LLVolume* volume = getVolume(); + + if (!isSculpted()) + { + LLVolume* ref = LLPrimitive::getVolumeManager()->refVolume(volume->getParams(), 3); + ret = ref->getNumTriangles(); + LLPrimitive::getVolumeManager()->unrefVolume(ref); + } + /*else if (isMesh()) + { + LLVolume* ref = LLPrimitive::getVolumeManager()->refVolume(volume->getParams(), 3); + if (ref->isTetrahedron() || ref->getNumVolumeFaces() == 0) + { + gMeshRepo.loadMesh(this, volume->getParams(), LLModel::LOD_HIGH); + } + ret = ref->getNumTriangles(); + LLPrimitive::getVolumeManager()->unrefVolume(ref); + }*/ + else + { //default sculpts have a constant number of triangles + ret = 31*2*31; //31 rows of 31 columns of quads for a 32x32 vertex patch + } + + return ret; +} + //static void LLVOVolume::preUpdateGeom() { @@ -2070,7 +2115,7 @@ void LLVOVolume::setSelected(BOOL sel) } } -void LLVOVolume::updateSpatialExtents(LLVector4a& min, LLVector4a& max) +void LLVOVolume::updateSpatialExtents(LLVector4a& newMin, LLVector4a& newMax) { } @@ -2805,6 +2850,32 @@ void LLVolumeGeometryManager::rebuildMesh(LLSpatialGroup* group) } } +struct CompareBatchBreakerModified +{ + bool operator()(const LLFace* const& lhs, const LLFace* const& rhs) + { + const LLTextureEntry* lte = lhs->getTextureEntry(); + const LLTextureEntry* rte = rhs->getTextureEntry(); + + if (lte->getBumpmap() != rte->getBumpmap()) + { + return lte->getBumpmap() < rte->getBumpmap(); + } + else if (lte->getFullbright() != rte->getFullbright()) + { + return lte->getFullbright() < rte->getFullbright(); + } + else if (lte->getGlow() != rte->getGlow()) + { + return lte->getGlow() < rte->getGlow(); + } + else + { + return lhs->getTexture() < rhs->getTexture(); + } + + } +}; void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std::vector& faces, BOOL distance_sort) { //calculate maximum number of vertices to store in a single buffer @@ -2815,7 +2886,7 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std:: if (!distance_sort) { //sort faces by things that break batches - std::sort(faces.begin(), faces.end(), LLFace::CompareBatchBreaker()); + std::sort(faces.begin(), faces.end(), CompareBatchBreakerModified()); } else { @@ -2859,7 +2930,7 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std:: U32 index_count = facep->getIndicesCount(); U32 geom_count = facep->getGeomCount(); - //sum up vertices needed for this texture + //sum up vertices needed for this render batch std::vector::iterator i = face_iter; ++i; @@ -2880,7 +2951,7 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std:: //create/delete/resize vertex buffer if needed LLVertexBuffer* buffer = NULL; - LLSpatialGroup::buffer_texture_map_t::iterator found_iter = group->mBufferMap[mask].find(tex); + LLSpatialGroup::buffer_texture_map_t::iterator found_iter = group->mBufferMap[mask].find(*face_iter); if (found_iter != group->mBufferMap[mask].end()) { @@ -2911,7 +2982,7 @@ void LLVolumeGeometryManager::genDrawInfo(LLSpatialGroup* group, U32 mask, std:: } } - buffer_map[mask][tex].push_back(buffer); + buffer_map[mask][*face_iter].push_back(buffer); //add face geometry diff --git a/indra/newview/llvovolume.h b/indra/newview/llvovolume.h index 420dda150..22bed531f 100644 --- a/indra/newview/llvovolume.h +++ b/indra/newview/llvovolume.h @@ -118,6 +118,8 @@ public: /*virtual*/ const LLMatrix4 getRenderMatrix() const; + /*virtual*/ U32 getTriangleCount(); + /*virtual*/ U32 getHighLODTriangleCount(); /*virtual*/ BOOL lineSegmentIntersect(const LLVector3& start, const LLVector3& end, S32 face = -1, // which face to check, -1 = ALL_SIDES BOOL pick_transparent = FALSE,