diff options
Diffstat (limited to 'indra/llmath/llvolume.cpp')
| -rwxr-xr-x[-rw-r--r--] | indra/llmath/llvolume.cpp | 5048 |
1 files changed, 3202 insertions, 1846 deletions
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 34348230b6..c2198b91a7 100644..100755 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1,41 +1,40 @@ /** + * @file llvolume.cpp * - * $LicenseInfo:firstyear=2002&license=viewergpl$ - * - * Copyright (c) 2002-2009, Linden Research, Inc. - * + * $LicenseInfo:firstyear=2002&license=viewerlgpl$ * Second Life Viewer Source Code - * The source code in this file ("Source Code") is provided by Linden Lab - * to you under the terms of the GNU General Public License, version 2.0 - * ("GPL"), unless you have obtained a separate licensing agreement - * ("Other License"), formally executed by you and Linden Lab. Terms of - * the GPL can be found in doc/GPL-license.txt in this distribution, or - * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2 + * Copyright (C) 2010, Linden Research, Inc. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; + * version 2.1 of the License only. * - * There are special exceptions to the terms and conditions of the GPL as - * it is applied to this Source Code. View the full text of the exception - * in the file doc/FLOSS-exception.txt in this software distribution, or - * online at - * http://secondlifegrid.net/programs/open_source/licensing/flossexception + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. * - * By copying, modifying or distributing this software, you acknowledge - * that you have read and understood your obligations described above, - * and agree to abide by those obligations. + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA * - * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO - * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY, - * COMPLETENESS OR PERFORMANCE. + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA * $/LicenseInfo$ */ #include "linden_common.h" +#include "llmemory.h" #include "llmath.h" #include <set> +#if !LL_WINDOWS +#include <stdint.h> +#endif +#include <cmath> #include "llerror.h" -#include "llmemtype.h" #include "llvolumemgr.h" #include "v2math.h" @@ -43,16 +42,20 @@ #include "v4math.h" #include "m4math.h" #include "m3math.h" -#include "lldarray.h" +#include "llmatrix3a.h" +#include "lloctree.h" #include "llvolume.h" +#include "llvolumeoctree.h" #include "llstl.h" +#include "llsdserialize.h" +#include "llvector4a.h" +#include "llmatrix4a.h" +#include "lltimer.h" #define DEBUG_SILHOUETTE_BINORMALS 0 #define DEBUG_SILHOUETTE_NORMALS 0 // TomY: Use this to display normals using the silhouette #define DEBUG_SILHOUETTE_EDGE_MAP 0 // DaveP: Use this to display edge map using the silhouette -const F32 CUT_MIN = 0.f; -const F32 CUT_MAX = 1.f; const F32 MIN_CUT_DELTA = 0.02f; const F32 HOLLOW_MIN = 0.f; @@ -86,6 +89,8 @@ 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); @@ -103,128 +108,339 @@ BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLV BOOL LLLineSegmentBoxIntersect(const LLVector3& start, const LLVector3& end, const LLVector3& center, const LLVector3& size) { - float fAWdU[3]; - LLVector3 dir; - LLVector3 diff; + return LLLineSegmentBoxIntersect(start.mV, end.mV, center.mV, size.mV); +} + +BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* center, const F32* size) +{ + F32 fAWdU[3]; + F32 dir[3]; + F32 diff[3]; for (U32 i = 0; i < 3; i++) { - dir.mV[i] = 0.5f * (end.mV[i] - start.mV[i]); - diff.mV[i] = (0.5f * (end.mV[i] + start.mV[i])) - center.mV[i]; - fAWdU[i] = fabsf(dir.mV[i]); - if(fabsf(diff.mV[i])>size.mV[i] + fAWdU[i]) return false; + dir[i] = 0.5f * (end[i] - start[i]); + diff[i] = (0.5f * (end[i] + start[i])) - center[i]; + fAWdU[i] = fabsf(dir[i]); + if(fabsf(diff[i])>size[i] + fAWdU[i]) return false; } float f; - f = dir.mV[1] * diff.mV[2] - dir.mV[2] * diff.mV[1]; if(fabsf(f)>size.mV[1]*fAWdU[2] + size.mV[2]*fAWdU[1]) return false; - f = dir.mV[2] * diff.mV[0] - dir.mV[0] * diff.mV[2]; if(fabsf(f)>size.mV[0]*fAWdU[2] + size.mV[2]*fAWdU[0]) return false; - f = dir.mV[0] * diff.mV[1] - dir.mV[1] * diff.mV[0]; if(fabsf(f)>size.mV[0]*fAWdU[1] + size.mV[1]*fAWdU[0]) return false; + f = dir[1] * diff[2] - dir[2] * diff[1]; if(fabsf(f)>size[1]*fAWdU[2] + size[2]*fAWdU[1]) return false; + f = dir[2] * diff[0] - dir[0] * diff[2]; if(fabsf(f)>size[0]*fAWdU[2] + size[2]*fAWdU[0]) return false; + f = dir[0] * diff[1] - dir[1] * diff[0]; if(fabsf(f)>size[0]*fAWdU[1] + size[1]*fAWdU[0]) return false; return true; } +// Finds tangent vec based on three vertices with texture coordinates. +// Fills in dummy values if the triangle has degenerate texture coordinates. +void calc_tangent_from_triangle( + LLVector4a& normal, + LLVector4a& tangent_out, + const LLVector4a& v1, + const LLVector2& w1, + const LLVector4a& v2, + const LLVector2& w2, + const LLVector4a& v3, + const LLVector2& w3) +{ + const F32* v1ptr = v1.getF32ptr(); + const F32* v2ptr = v2.getF32ptr(); + const F32* v3ptr = v3.getF32ptr(); + + float x1 = v2ptr[0] - v1ptr[0]; + float x2 = v3ptr[0] - v1ptr[0]; + float y1 = v2ptr[1] - v1ptr[1]; + float y2 = v3ptr[1] - v1ptr[1]; + float z1 = v2ptr[2] - v1ptr[2]; + float z2 = v3ptr[2] - v1ptr[2]; + + float s1 = w2.mV[0] - w1.mV[0]; + float s2 = w3.mV[0] - w1.mV[0]; + float t1 = w2.mV[1] - w1.mV[1]; + float t2 = w3.mV[1] - w1.mV[1]; + + F32 rd = s1*t2-s2*t1; + + float r = ((rd*rd) > FLT_EPSILON) ? (1.0f / rd) + : ((rd > 0.0f) ? 1024.f : -1024.f); //some made up large ratio for division by zero + + llassert(llfinite(r)); + llassert(!llisnan(r)); + + LLVector4a sdir( + (t2 * x1 - t1 * x2) * r, + (t2 * y1 - t1 * y2) * r, + (t2 * z1 - t1 * z2) * r); + + LLVector4a tdir( + (s1 * x2 - s2 * x1) * r, + (s1 * y2 - s2 * y1) * r, + (s1 * z2 - s2 * z1) * r); + + LLVector4a n = normal; + LLVector4a t = sdir; + + LLVector4a ncrosst; + ncrosst.setCross3(n,t); + + // Gram-Schmidt orthogonalize + n.mul(n.dot3(t).getF32()); + + LLVector4a tsubn; + tsubn.setSub(t,n); + + if (tsubn.dot3(tsubn).getF32() > F_APPROXIMATELY_ZERO) + { + tsubn.normalize3fast_checked(); + + // Calculate handedness + F32 handedness = ncrosst.dot3(tdir).getF32() < 0.f ? -1.f : 1.f; + + tsubn.getF32ptr()[3] = handedness; + + tangent_out = tsubn; + } + else + { + // degenerate, make up a value + // + tangent_out.set(0,0,1,1); + } + +} + // intersect test between triangle vert0, vert1, vert2 and a ray from orig in direction dir. // returns TRUE if intersecting and returns barycentric coordinates in intersection_a, intersection_b, // and returns the intersection point along dir in intersection_t. // Moller-Trumbore algorithm -BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, const LLVector3& vert2, const LLVector3& orig, const LLVector3& dir, - F32* intersection_a, F32* intersection_b, F32* intersection_t, BOOL two_sided) +BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, const LLVector4a& vert2, const LLVector4a& orig, const LLVector4a& dir, + F32& intersection_a, F32& intersection_b, F32& intersection_t) { - F32 u, v, t; /* find vectors for two edges sharing vert0 */ - LLVector3 edge1 = vert1 - vert0; + LLVector4a edge1; + edge1.setSub(vert1, vert0); - LLVector3 edge2 = vert2 - vert0;; + LLVector4a edge2; + edge2.setSub(vert2, vert0); /* begin calculating determinant - also used to calculate U parameter */ - LLVector3 pvec = dir % edge2; - - /* if determinant is near zero, ray lies in plane of triangle */ - F32 det = edge1 * pvec; + LLVector4a pvec; + pvec.setCross3(dir, edge2); - if (!two_sided) + /* if determinant is near zero, ray lies in plane of triangle */ + LLVector4a det; + det.setAllDot3(edge1, pvec); + + if (det.greaterEqual(LLVector4a::getEpsilon()).getGatheredBits() & 0x7) { - if (det < F_APPROXIMATELY_ZERO) - { - return FALSE; - } - /* calculate distance from vert0 to ray origin */ - LLVector3 tvec = orig - vert0; + LLVector4a tvec; + tvec.setSub(orig, vert0); /* calculate U parameter and test bounds */ - u = tvec * pvec; + LLVector4a u; + u.setAllDot3(tvec,pvec); - if (u < 0.f || u > det) + if ((u.greaterEqual(LLVector4a::getZero()).getGatheredBits() & 0x7) && + (u.lessEqual(det).getGatheredBits() & 0x7)) { - return FALSE; + /* prepare to test V parameter */ + LLVector4a qvec; + qvec.setCross3(tvec, edge1); + + /* calculate V parameter and test bounds */ + LLVector4a v; + v.setAllDot3(dir, qvec); + + + //if (!(v < 0.f || u + v > det)) + + LLVector4a sum_uv; + sum_uv.setAdd(u, v); + + S32 v_gequal = v.greaterEqual(LLVector4a::getZero()).getGatheredBits() & 0x7; + S32 sum_lequal = sum_uv.lessEqual(det).getGatheredBits() & 0x7; + + if (v_gequal && sum_lequal) + { + /* calculate t, scale parameters, ray intersects triangle */ + LLVector4a t; + t.setAllDot3(edge2,qvec); + + t.div(det); + u.div(det); + v.div(det); + + intersection_a = u[0]; + intersection_b = v[0]; + intersection_t = t[0]; + return TRUE; + } } - - /* prepare to test V parameter */ - LLVector3 qvec = tvec % edge1; + } - /* calculate V parameter and test bounds */ - v = dir * qvec; - if (v < 0.f || u + v > det) - { - return FALSE; - } + return FALSE; +} - /* calculate t, scale parameters, ray intersects triangle */ - t = edge2 * qvec; - F32 inv_det = 1.0 / det; - t *= inv_det; - u *= inv_det; - v *= inv_det; - } +BOOL LLTriangleRayIntersectTwoSided(const LLVector4a& vert0, const LLVector4a& vert1, const LLVector4a& vert2, const LLVector4a& orig, const LLVector4a& dir, + F32& intersection_a, F32& intersection_b, F32& intersection_t) +{ + F32 u, v, t; - else // two sided - { - if (det > -F_APPROXIMATELY_ZERO && det < F_APPROXIMATELY_ZERO) - { - return FALSE; - } - F32 inv_det = 1.0 / det; + /* find vectors for two edges sharing vert0 */ + LLVector4a edge1; + edge1.setSub(vert1, vert0); + + + LLVector4a edge2; + edge2.setSub(vert2, vert0); - /* calculate distance from vert0 to ray origin */ - LLVector3 tvec = orig - vert0; - - /* calculate U parameter and test bounds */ - u = (tvec * pvec) * inv_det; - if (u < 0.f || u > 1.f) - { - return FALSE; - } + /* begin calculating determinant - also used to calculate U parameter */ + LLVector4a pvec; + pvec.setCross3(dir, edge2); - /* prepare to test V parameter */ - LLVector3 qvec = tvec - edge1; - - /* calculate V parameter and test bounds */ - v = (dir * qvec) * inv_det; - - if (v < 0.f || u + v > 1.f) - { - return FALSE; - } + /* if determinant is near zero, ray lies in plane of triangle */ + F32 det = edge1.dot3(pvec).getF32(); + + + if (det > -F_APPROXIMATELY_ZERO && det < F_APPROXIMATELY_ZERO) + { + return FALSE; + } - /* calculate t, ray intersects triangle */ - t = (edge2 * qvec) * inv_det; + F32 inv_det = 1.f / det; + + /* calculate distance from vert0 to ray origin */ + LLVector4a tvec; + tvec.setSub(orig, vert0); + + /* calculate U parameter and test bounds */ + u = (tvec.dot3(pvec).getF32()) * inv_det; + if (u < 0.f || u > 1.f) + { + return FALSE; } + + /* prepare to test V parameter */ + tvec.sub(edge1); + + /* calculate V parameter and test bounds */ + v = (dir.dot3(tvec).getF32()) * inv_det; - if (intersection_a != NULL) - *intersection_a = u; - if (intersection_b != NULL) - *intersection_b = v; - if (intersection_t != NULL) - *intersection_t = t; + if (v < 0.f || u + v > 1.f) + { + return FALSE; + } + + /* calculate t, ray intersects triangle */ + t = (edge2.dot3(tvec).getF32()) * inv_det; + + intersection_a = u; + intersection_b = v; + intersection_t = t; return TRUE; } +//helper for non-aligned vectors +BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, const LLVector3& vert2, const LLVector3& orig, const LLVector3& dir, + F32& intersection_a, F32& intersection_b, F32& intersection_t, BOOL two_sided) +{ + LLVector4a vert0a, vert1a, vert2a, origa, dira; + vert0a.load3(vert0.mV); + vert1a.load3(vert1.mV); + vert2a.load3(vert2.mV); + origa.load3(orig.mV); + dira.load3(dir.mV); + + if (two_sided) + { + return LLTriangleRayIntersectTwoSided(vert0a, vert1a, vert2a, origa, dira, + intersection_a, intersection_b, intersection_t); + } + else + { + return LLTriangleRayIntersect(vert0a, vert1a, vert2a, origa, dira, + intersection_a, intersection_b, intersection_t); + } +} + +class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst<LLVolumeTriangle> +{ +public: + const LLVolumeFace* mFace; + + LLVolumeOctreeRebound(const LLVolumeFace* face) + { + mFace = face; + } + + 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->isEmpty()) + { //node has data, find AABB that binds data set + const LLVolumeTriangle* tri = *(branch->getDataBegin()); + + //initialize min/max to first available vertex + min = *(tri->mV[0]); + max = *(tri->mV[0]); + + for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = + branch->getDataBegin(); iter != branch->getDataEnd(); ++iter) + { //for each triangle in node + + //stretch by triangles in node + tri = *iter; + + min.setMin(min, *tri->mV[0]); + min.setMin(min, *tri->mV[1]); + min.setMin(min, *tri->mV[2]); + + max.setMax(max, *tri->mV[0]); + max.setMax(max, *tri->mV[1]); + max.setMax(max, *tri->mV[2]); + } + } + else if (!branch->isLeaf()) + { //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]; + } + else + { + LL_ERRS() << "Empty leaf" << LL_ENDL; + } + + 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); + + node->mBounds[1].setSub(max,min); + node->mBounds[1].mul(0.5f); + } +}; //------------------------------------------------------------------- // statics @@ -235,8 +451,6 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons LLProfile::Face* LLProfile::addCap(S16 faceID) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - Face *face = vector_append(mFaces, 1); face->mIndex = 0; @@ -249,8 +463,6 @@ LLProfile::Face* LLProfile::addCap(S16 faceID) LLProfile::Face* LLProfile::addFace(S32 i, S32 count, F32 scaleU, S16 faceID, BOOL flat) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - Face *face = vector_append(mFaces, 1); face->mIndex = i; @@ -263,19 +475,80 @@ LLProfile::Face* LLProfile::addFace(S32 i, S32 count, F32 scaleU, S16 faceID, BO return face; } +//static +S32 LLProfile::getNumNGonPoints(const LLProfileParams& params, S32 sides, F32 offset, F32 bevel, F32 ang_scale, S32 split) +{ // this is basically LLProfile::genNGon stripped down to only the operations that influence the number of points + S32 np = 0; + + // Generate an n-sided "circular" path. + // 0 is (1,0), and we go counter-clockwise along a circular path from there. + F32 t, t_step, t_first, t_fraction; + + F32 begin = params.getBegin(); + F32 end = params.getEnd(); + + t_step = 1.0f / sides; + + t_first = floor(begin * sides) / (F32)sides; + + // pt1 is the first point on the fractional face. + // Starting t and ang values for the first face + t = t_first; + + // Increment to the next point. + // pt2 is the end point on the fractional face + t += t_step; + + t_fraction = (begin - t_first)*sides; + + // Only use if it's not almost exactly on an edge. + if (t_fraction < 0.9999f) + { + np++; + } + + // There's lots of potential here for floating point error to generate unneeded extra points - DJS 04/05/02 + while (t < end) + { + // Iterate through all the integer steps of t. + np++; + + t += t_step; + } + + t_fraction = (end - (t - t_step))*sides; + + // Find the fraction that we need to add to the end point. + t_fraction = (end - (t - t_step))*sides; + if (t_fraction > 0.0001f) + { + np++; + } + + // If we're sliced, the profile is open. + if ((end - begin)*ang_scale < 0.99f) + { + if (params.getHollow() <= 0) + { + // put center point if not hollow. + np++; + } + } + + return np; +} + // What is the bevel parameter used for? - DJS 04/05/02 // Bevel parameter is currently unused but presumedly would support // filleted and chamfered corners void LLProfile::genNGon(const LLProfileParams& params, S32 sides, F32 offset, F32 bevel, F32 ang_scale, S32 split) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - // Generate an n-sided "circular" path. // 0 is (1,0), and we go counter-clockwise along a circular path from there. const F32 tableScale[] = { 1, 1, 1, 0.5f, 0.707107f, 0.53f, 0.525f, 0.5f }; F32 scale = 0.5f; F32 t, t_step, t_first, t_fraction, ang, ang_step; - LLVector3 pt1,pt2; + LLVector4a pt1,pt2; F32 begin = params.getBegin(); F32 end = params.getEnd(); @@ -285,7 +558,7 @@ void LLProfile::genNGon(const LLProfileParams& params, S32 sides, F32 offset, F3 // Scale to have size "match" scale. Compensates to get object to generally fill bounding box. - S32 total_sides = llround(sides / ang_scale); // Total number of sides all around + S32 total_sides = ll_round(sides / ang_scale); // Total number of sides all around if (total_sides < 8) { @@ -298,20 +571,21 @@ void LLProfile::genNGon(const LLProfileParams& params, S32 sides, F32 offset, F3 // Starting t and ang values for the first face t = t_first; ang = 2.0f*F_PI*(t*ang_scale + offset); - pt1.setVec(cos(ang)*scale,sin(ang)*scale, t); + pt1.set(cos(ang)*scale,sin(ang)*scale, t); // Increment to the next point. // pt2 is the end point on the fractional face t += t_step; ang += ang_step; - pt2.setVec(cos(ang)*scale,sin(ang)*scale,t); + pt2.set(cos(ang)*scale,sin(ang)*scale,t); t_fraction = (begin - t_first)*sides; // Only use if it's not almost exactly on an edge. if (t_fraction < 0.9999f) { - LLVector3 new_pt = lerp(pt1, pt2, t_fraction); + LLVector4a new_pt; + new_pt.setLerp(pt1, pt2, t_fraction); mProfile.push_back(new_pt); } @@ -319,12 +593,17 @@ void LLProfile::genNGon(const LLProfileParams& params, S32 sides, F32 offset, F3 while (t < end) { // Iterate through all the integer steps of t. - pt1.setVec(cos(ang)*scale,sin(ang)*scale,t); + pt1.set(cos(ang)*scale,sin(ang)*scale,t); if (mProfile.size() > 0) { - LLVector3 p = mProfile[mProfile.size()-1]; + LLVector4a p = mProfile[mProfile.size()-1]; for (S32 i = 0; i < split && mProfile.size() > 0; i++) { - mProfile.push_back(p+(pt1-p) * 1.0f/(float)(split+1) * (float)(i+1)); + //mProfile.push_back(p+(pt1-p) * 1.0f/(float)(split+1) * (float)(i+1)); + LLVector4a new_pt; + new_pt.setSub(pt1, p); + new_pt.mul(1.0f/(float)(split+1) * (float)(i+1)); + new_pt.add(p); + mProfile.push_back(new_pt); } } mProfile.push_back(pt1); @@ -337,18 +616,25 @@ void LLProfile::genNGon(const LLProfileParams& params, S32 sides, F32 offset, F3 // pt1 is the first point on the fractional face // pt2 is the end point on the fractional face - pt2.setVec(cos(ang)*scale,sin(ang)*scale,t); + pt2.set(cos(ang)*scale,sin(ang)*scale,t); // Find the fraction that we need to add to the end point. t_fraction = (end - (t - t_step))*sides; if (t_fraction > 0.0001f) { - LLVector3 new_pt = lerp(pt1, pt2, t_fraction); + LLVector4a new_pt; + new_pt.setLerp(pt1, pt2, t_fraction); if (mProfile.size() > 0) { - LLVector3 p = mProfile[mProfile.size()-1]; + LLVector4a p = mProfile[mProfile.size()-1]; for (S32 i = 0; i < split && mProfile.size() > 0; i++) { - mProfile.push_back(p+(new_pt-p) * 1.0f/(float)(split+1) * (float)(i+1)); + //mProfile.push_back(p+(new_pt-p) * 1.0f/(float)(split+1) * (float)(i+1)); + + LLVector4a pt1; + pt1.setSub(new_pt, p); + pt1.mul(1.0f/(float)(split+1) * (float)(i+1)); + pt1.add(p); + mProfile.push_back(pt1); } } mProfile.push_back(new_pt); @@ -369,7 +655,7 @@ void LLProfile::genNGon(const LLProfileParams& params, S32 sides, F32 offset, F3 if (params.getHollow() <= 0) { // put center point if not hollow. - mProfile.push_back(LLVector3(0,0,0)); + mProfile.push_back(LLVector4a(0,0,0)); } } else @@ -382,103 +668,6 @@ void LLProfile::genNGon(const LLProfileParams& params, S32 sides, F32 offset, F3 mTotal = mProfile.size(); } -void LLProfile::genNormals(const LLProfileParams& params) -{ - S32 count = mProfile.size(); - - S32 outer_count; - if (mTotalOut) - { - outer_count = mTotalOut; - } - else - { - outer_count = mTotal / 2; - } - - mEdgeNormals.resize(count * 2); - mEdgeCenters.resize(count * 2); - mNormals.resize(count); - - LLVector2 pt0,pt1; - - BOOL hollow = (params.getHollow() > 0); - - S32 i0, i1, i2, i3, i4; - - // Parametrically generate normal - for (i2 = 0; i2 < count; i2++) - { - mNormals[i2].mV[0] = mProfile[i2].mV[0]; - mNormals[i2].mV[1] = mProfile[i2].mV[1]; - if (hollow && (i2 >= outer_count)) - { - mNormals[i2] *= -1.f; - } - if (mNormals[i2].magVec() < 0.001) - { - // Special case for point at center, get adjacent points. - i1 = (i2 - 1) >= 0 ? i2 - 1 : count - 1; - i0 = (i1 - 1) >= 0 ? i1 - 1 : count - 1; - i3 = (i2 + 1) < count ? i2 + 1 : 0; - i4 = (i3 + 1) < count ? i3 + 1 : 0; - - pt0.setVec(mProfile[i1].mV[VX] + mProfile[i1].mV[VX] - mProfile[i0].mV[VX], - mProfile[i1].mV[VY] + mProfile[i1].mV[VY] - mProfile[i0].mV[VY]); - pt1.setVec(mProfile[i3].mV[VX] + mProfile[i3].mV[VX] - mProfile[i4].mV[VX], - mProfile[i3].mV[VY] + mProfile[i3].mV[VY] - mProfile[i4].mV[VY]); - - mNormals[i2] = pt0 + pt1; - mNormals[i2] *= 0.5f; - } - mNormals[i2].normVec(); - } - - S32 num_normal_sets = isConcave() ? 2 : 1; - for (S32 normal_set = 0; normal_set < num_normal_sets; normal_set++) - { - S32 point_num; - for (point_num = 0; point_num < mTotal; point_num++) - { - LLVector3 point_1 = mProfile[point_num]; - point_1.mV[VZ] = 0.f; - - LLVector3 point_2; - - if (isConcave() && normal_set == 0 && point_num == (mTotal - 1) / 2) - { - point_2 = mProfile[mTotal - 1]; - } - else if (isConcave() && normal_set == 1 && point_num == mTotal - 1) - { - point_2 = mProfile[(mTotal - 1) / 2]; - } - else - { - LLVector3 delta_pos; - S32 neighbor_point = (point_num + 1) % mTotal; - while(delta_pos.magVecSquared() < 0.01f * 0.01f) - { - point_2 = mProfile[neighbor_point]; - delta_pos = point_2 - point_1; - neighbor_point = (neighbor_point + 1) % mTotal; - if (neighbor_point == point_num) - { - break; - } - } - } - - point_2.mV[VZ] = 0.f; - LLVector3 face_normal = (point_2 - point_1) % LLVector3::z_axis; - face_normal.normVec(); - mEdgeNormals[normal_set * count + point_num] = face_normal; - mEdgeCenters[normal_set * count + point_num] = lerp(point_1, point_2, 0.5f); - } - } -} - - // Hollow is percent of the original bounding box, not of this particular // profile's geometry. Thus, a swept triangle needs lower hollow values than // a swept square. @@ -494,12 +683,13 @@ LLProfile::Face* LLProfile::addHole(const LLProfileParams& params, BOOL flat, F3 Face *face = addFace(mTotalOut, mTotal-mTotalOut,0,LL_FACE_INNER_SIDE, flat); - std::vector<LLVector3> pt; + static LLAlignedArray<LLVector4a,64> pt; pt.resize(mTotal) ; for (S32 i=mTotalOut;i<mTotal;i++) { - pt[i] = mProfile[i] * box_hollow; + pt[i] = mProfile[i]; + pt[i].mul(box_hollow); } S32 j=mTotal-1; @@ -519,13 +709,120 @@ LLProfile::Face* LLProfile::addHole(const LLProfileParams& params, BOOL flat, F3 return face; } +//static +S32 LLProfile::getNumPoints(const LLProfileParams& params, BOOL path_open,F32 detail, S32 split, + BOOL is_sculpted, S32 sculpt_size) +{ // this is basically LLProfile::generate stripped down to only operations that influence the number of points + if (detail < MIN_LOD) + { + detail = MIN_LOD; + } + + // Generate the face data + F32 hollow = params.getHollow(); + + S32 np = 0; + + switch (params.getCurveType() & LL_PCODE_PROFILE_MASK) + { + case LL_PCODE_PROFILE_SQUARE: + { + np = getNumNGonPoints(params, 4,-0.375, 0, 1, split); + + if (hollow) + { + np *= 2; + } + } + break; + case LL_PCODE_PROFILE_ISOTRI: + case LL_PCODE_PROFILE_RIGHTTRI: + case LL_PCODE_PROFILE_EQUALTRI: + { + np = getNumNGonPoints(params, 3,0, 0, 1, split); + + if (hollow) + { + np *= 2; + } + } + break; + case LL_PCODE_PROFILE_CIRCLE: + { + // If this has a square hollow, we should adjust the + // number of faces a bit so that the geometry lines up. + U8 hole_type=0; + F32 circle_detail = MIN_DETAIL_FACES * detail; + if (hollow) + { + hole_type = params.getCurveType() & LL_PCODE_HOLE_MASK; + if (hole_type == LL_PCODE_HOLE_SQUARE) + { + // Snap to the next multiple of four sides, + // so that corners line up. + circle_detail = llceil(circle_detail / 4.0f) * 4.0f; + } + } + + S32 sides = (S32)circle_detail; + + if (is_sculpted) + sides = sculpt_size; + + np = getNumNGonPoints(params, sides); + + if (hollow) + { + np *= 2; + } + } + break; + case LL_PCODE_PROFILE_CIRCLE_HALF: + { + // If this has a square hollow, we should adjust the + // number of faces a bit so that the geometry lines up. + U8 hole_type=0; + // Number of faces is cut in half because it's only a half-circle. + F32 circle_detail = MIN_DETAIL_FACES * detail * 0.5f; + if (hollow) + { + hole_type = params.getCurveType() & LL_PCODE_HOLE_MASK; + if (hole_type == LL_PCODE_HOLE_SQUARE) + { + // Snap to the next multiple of four sides (div 2), + // so that corners line up. + circle_detail = llceil(circle_detail / 2.0f) * 2.0f; + } + } + np = getNumNGonPoints(params, llfloor(circle_detail), 0.5f, 0.f, 0.5f); + + if (hollow) + { + np *= 2; + } + + // Special case for openness of sphere + if ((params.getEnd() - params.getBegin()) < 1.f) + { + } + else if (!hollow) + { + np++; + } + } + break; + default: + break; + }; + + + return np; +} BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detail, S32 split, BOOL is_sculpted, S32 sculpt_size) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - if ((!mDirty) && (!is_sculpted)) { return FALSE; @@ -534,12 +831,12 @@ BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detai if (detail < MIN_LOD) { - llinfos << "Generating profile with LOD < MIN_LOD. CLAMPING" << llendl; + LL_INFOS() << "Generating profile with LOD < MIN_LOD. CLAMPING" << LL_ENDL; detail = MIN_LOD; } - mProfile.clear(); - mFaces.clear(); + mProfile.resize(0); + mFaces.resize(0); // Generate the face data S32 i; @@ -550,7 +847,7 @@ BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detai // Quick validation to eliminate some server crashes. if (begin > end - 0.01f) { - llwarns << "LLProfile::generate() assertion failed (begin >= end)" << llendl; + LL_WARNS() << "LLProfile::generate() assertion failed (begin >= end)" << LL_ENDL; return FALSE; } @@ -571,10 +868,13 @@ BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detai addFace((face_num++) * (split +1), split+2, 1, LL_FACE_OUTER_SIDE_0 << i, TRUE); } + LLVector4a scale(1,1,4,1); + for (i = 0; i <(S32) mProfile.size(); i++) { // Scale by 4 to generate proper tex coords. - mProfile[i].mV[2] *= 4.f; + mProfile[i].mul(scale); + llassert(mProfile[i].isFinite3()); } if (hollow) @@ -607,10 +907,12 @@ BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detai case LL_PCODE_PROFILE_EQUALTRI: { genNGon(params, 3,0, 0, 1, split); + LLVector4a scale(1,1,3,1); for (i = 0; i <(S32) mProfile.size(); i++) { // Scale by 3 to generate proper tex coords. - mProfile[i].mV[2] *= 3.f; + mProfile[i].mul(scale); + llassert(mProfile[i].isFinite3()); } if (path_open) @@ -766,7 +1068,7 @@ BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detai } break; default: - llerrs << "Unknown profile: getCurveType()=" << params.getCurveType() << llendl; + LL_ERRS() << "Unknown profile: getCurveType()=" << params.getCurveType() << LL_ENDL; break; }; @@ -789,8 +1091,6 @@ BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detai } } - //genNormals(params); - return TRUE; } @@ -798,8 +1098,6 @@ BOOL LLProfile::generate(const LLProfileParams& params, BOOL path_open,F32 detai BOOL LLProfileParams::importFile(LLFILE *fp) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - const S32 BUFSIZE = 16384; char buffer[BUFSIZE]; /* Flawfinder: ignore */ // *NOTE: changing the size or type of these buffers will require @@ -852,7 +1150,7 @@ BOOL LLProfileParams::importFile(LLFILE *fp) } else { - llwarns << "unknown keyword " << keyword << " in profile import" << llendl; + LL_WARNS() << "unknown keyword " << keyword << " in profile import" << LL_ENDL; } } @@ -875,8 +1173,6 @@ BOOL LLProfileParams::exportFile(LLFILE *fp) const BOOL LLProfileParams::importLegacyStream(std::istream& input_stream) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - const S32 BUFSIZE = 16384; char buffer[BUFSIZE]; /* Flawfinder: ignore */ // *NOTE: changing the size or type of these buffers will require @@ -926,7 +1222,7 @@ BOOL LLProfileParams::importLegacyStream(std::istream& input_stream) } else { - llwarns << "unknown keyword " << keyword << " in profile import" << llendl; + LL_WARNS() << "unknown keyword " << keyword << " in profile import" << LL_ENDL; } } @@ -968,7 +1264,6 @@ bool LLProfileParams::fromLLSD(LLSD& sd) void LLProfileParams::copyParams(const LLProfileParams ¶ms) { - LLMemType m1(LLMemType::MTYPE_VOLUME); setCurveType(params.getCurveType()); setBegin(params.getBegin()); setEnd(params.getEnd()); @@ -980,6 +1275,32 @@ LLPath::~LLPath() { } +S32 LLPath::getNumNGonPoints(const LLPathParams& params, S32 sides, F32 startOff, F32 end_scale, F32 twist_scale) +{ //this is basically LLPath::genNGon stripped down to only operations that influence the number of points added + S32 ret = 0; + + F32 step= 1.0f / sides; + F32 t = params.getBegin(); + ret = 1; + + t+=step; + + // Snap to a quantized parameter, so that cut does not + // affect most sample points. + t = ((S32)(t * sides)) / (F32)sides; + + // Run through the non-cut dependent points. + while (t < params.getEnd()) + { + ret++; + t+=step; + } + + ret++; + + return ret; +} + void LLPath::genNGon(const LLPathParams& params, S32 sides, F32 startOff, F32 end_scale, F32 twist_scale) { // Generates a circular path, starting at (1, 0, 0), counterclockwise along the xz plane. @@ -1052,25 +1373,29 @@ void LLPath::genNGon(const LLPathParams& params, S32 sides, F32 startOff, F32 en // the path begins at the correct cut. F32 step= 1.0f / sides; F32 t = params.getBegin(); - pt = vector_append(mPath, 1); + pt = mPath.append(1); ang = 2.0f*F_PI*revolutions * t; s = sin(ang)*lerp(radius_start, radius_end, t); c = cos(ang)*lerp(radius_start, radius_end, t); - pt->mPos.setVec(0 + lerp(0,params.getShear().mV[0],s) + pt->mPos.set(0 + lerp(0,params.getShear().mV[0],s) + lerp(-skew ,skew, t) * 0.5f, c + lerp(0,params.getShear().mV[1],s), s); - pt->mScale.mV[VX] = hole_x * lerp(taper_x_begin, taper_x_end, t); - pt->mScale.mV[VY] = hole_y * lerp(taper_y_begin, taper_y_end, t); + pt->mScale.set(hole_x * lerp(taper_x_begin, taper_x_end, t), + hole_y * lerp(taper_y_begin, taper_y_end, t), + 0,1); pt->mTexT = t; - + // Twist rotates the path along the x,y plane (I think) - DJS 04/05/02 twist.setQuat (lerp(twist_begin,twist_end,t) * 2.f * F_PI - F_PI,0,0,1); // Rotate the point around the circle's center. qang.setQuat (ang,path_axis); - pt->mRot = twist * qang; + + LLMatrix3 rot(twist * qang); + + pt->mRot.loadu(rot); t+=step; @@ -1081,50 +1406,54 @@ void LLPath::genNGon(const LLPathParams& params, S32 sides, F32 startOff, F32 en // Run through the non-cut dependent points. while (t < params.getEnd()) { - pt = vector_append(mPath, 1); + pt = mPath.append(1); ang = 2.0f*F_PI*revolutions * t; c = cos(ang)*lerp(radius_start, radius_end, t); s = sin(ang)*lerp(radius_start, radius_end, t); - pt->mPos.setVec(0 + lerp(0,params.getShear().mV[0],s) + pt->mPos.set(0 + lerp(0,params.getShear().mV[0],s) + lerp(-skew ,skew, t) * 0.5f, c + lerp(0,params.getShear().mV[1],s), s); - pt->mScale.mV[VX] = hole_x * lerp(taper_x_begin, taper_x_end, t); - pt->mScale.mV[VY] = hole_y * lerp(taper_y_begin, taper_y_end, t); + pt->mScale.set(hole_x * lerp(taper_x_begin, taper_x_end, t), + hole_y * lerp(taper_y_begin, taper_y_end, t), + 0,1); pt->mTexT = t; // Twist rotates the path along the x,y plane (I think) - DJS 04/05/02 twist.setQuat (lerp(twist_begin,twist_end,t) * 2.f * F_PI - F_PI,0,0,1); // Rotate the point around the circle's center. qang.setQuat (ang,path_axis); - pt->mRot = twist * qang; + LLMatrix3 tmp(twist*qang); + pt->mRot.loadu(tmp); t+=step; } // Make one final pass for the end cut. t = params.getEnd(); - pt = vector_append(mPath, 1); + pt = mPath.append(1); ang = 2.0f*F_PI*revolutions * t; c = cos(ang)*lerp(radius_start, radius_end, t); s = sin(ang)*lerp(radius_start, radius_end, t); - pt->mPos.setVec(0 + lerp(0,params.getShear().mV[0],s) + pt->mPos.set(0 + lerp(0,params.getShear().mV[0],s) + lerp(-skew ,skew, t) * 0.5f, c + lerp(0,params.getShear().mV[1],s), s); - pt->mScale.mV[VX] = hole_x * lerp(taper_x_begin, taper_x_end, t); - pt->mScale.mV[VY] = hole_y * lerp(taper_y_begin, taper_y_end, t); + pt->mScale.set(hole_x * lerp(taper_x_begin, taper_x_end, t), + hole_y * lerp(taper_y_begin, taper_y_end, t), + 0,1); pt->mTexT = t; - + // Twist rotates the path along the x,y plane (I think) - DJS 04/05/02 twist.setQuat (lerp(twist_begin,twist_end,t) * 2.f * F_PI - F_PI,0,0,1); // Rotate the point around the circle's center. qang.setQuat (ang,path_axis); - pt->mRot = twist * qang; + LLMatrix3 tmp(twist*qang); + pt->mRot.loadu(tmp); mTotal = mPath.size(); } @@ -1157,11 +1486,57 @@ const LLVector2 LLPathParams::getEndScale() const return end_scale; } +S32 LLPath::getNumPoints(const LLPathParams& params, F32 detail) +{ // this is basically LLPath::generate stripped down to only the operations that influence the number of points + if (detail < MIN_LOD) + { + detail = MIN_LOD; + } + + S32 np = 2; // hardcode for line + + // Is this 0xf0 mask really necessary? DK 03/02/05 + + switch (params.getCurveType() & 0xf0) + { + default: + case LL_PCODE_PATH_LINE: + { + // Take the begin/end twist into account for detail. + np = llfloor(fabs(params.getTwistBegin() - params.getTwist()) * 3.5f * (detail-0.5f)) + 2; + } + break; + + case LL_PCODE_PATH_CIRCLE: + { + // Increase the detail as the revolutions and twist increase. + F32 twist_mag = fabs(params.getTwistBegin() - params.getTwist()); + + S32 sides = (S32)llfloor(llfloor((MIN_DETAIL_FACES * detail + twist_mag * 3.5f * (detail-0.5f))) * params.getRevolutions()); + + np = sides; + } + break; + + case LL_PCODE_PATH_CIRCLE2: + { + //genNGon(params, llfloor(MIN_DETAIL_FACES * detail), 4.f, 0.f); + np = getNumNGonPoints(params, llfloor(MIN_DETAIL_FACES * detail)); + } + break; + + case LL_PCODE_PATH_TEST: + + np = 5; + break; + }; + + return np; +} + BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split, BOOL is_sculpted, S32 sculpt_size) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - if ((!mDirty) && (!is_sculpted)) { return FALSE; @@ -1169,14 +1544,14 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split, if (detail < MIN_LOD) { - llinfos << "Generating path with LOD < MIN! Clamping to 1" << llendl; + LL_INFOS() << "Generating path with LOD < MIN! Clamping to 1" << LL_ENDL; detail = MIN_LOD; } mDirty = FALSE; S32 np = 2; // hardcode for line - mPath.clear(); + mPath.resize(0); mOpen = TRUE; // Is this 0xf0 mask really necessary? DK 03/02/05 @@ -1202,12 +1577,16 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split, for (S32 i=0;i<np;i++) { F32 t = lerp(params.getBegin(),params.getEnd(),(F32)i * mStep); - mPath[i].mPos.setVec(lerp(0,params.getShear().mV[0],t), + mPath[i].mPos.set(lerp(0,params.getShear().mV[0],t), lerp(0,params.getShear().mV[1],t), t - 0.5f); - mPath[i].mRot.setQuat(lerp(F_PI * params.getTwistBegin(),F_PI * params.getTwist(),t),0,0,1); - mPath[i].mScale.mV[0] = lerp(start_scale.mV[0],end_scale.mV[0],t); - mPath[i].mScale.mV[1] = lerp(start_scale.mV[1],end_scale.mV[1],t); + LLQuaternion quat; + quat.setQuat(lerp(F_PI * params.getTwistBegin(),F_PI * params.getTwist(),t),0,0,1); + LLMatrix3 tmp(quat); + mPath[i].mRot.loadu(tmp); + mPath[i].mScale.set(lerp(start_scale.mV[0],end_scale.mV[0],t), + lerp(start_scale.mV[1],end_scale.mV[1],t), + 0,1); mPath[i].mTexT = t; } } @@ -1221,7 +1600,7 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split, S32 sides = (S32)llfloor(llfloor((MIN_DETAIL_FACES * detail + twist_mag * 3.5f * (detail-0.5f))) * params.getRevolutions()); if (is_sculpted) - sides = sculpt_size; + sides = llmax(sculpt_size, 1); genNGon(params, sides); } @@ -1244,7 +1623,7 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split, F32 toggle = 0.5f; for (S32 i=0;i<(S32)mPath.size();i++) { - mPath[i].mPos.mV[0] = toggle; + mPath[i].mPos.getF32ptr()[0] = toggle; if (toggle == 0.5f) toggle = -0.5f; else @@ -1265,13 +1644,16 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split, for (S32 i=0;i<np;i++) { F32 t = (F32)i * mStep; - mPath[i].mPos.setVec(0, + mPath[i].mPos.set(0, lerp(0, -sin(F_PI*params.getTwist()*t)*0.5f,t), - lerp(-0.5, cos(F_PI*params.getTwist()*t)*0.5f,t)); - mPath[i].mScale.mV[0] = lerp(1,params.getScale().mV[0],t); - mPath[i].mScale.mV[1] = lerp(1,params.getScale().mV[1],t); + lerp(-0.5f, cos(F_PI*params.getTwist()*t)*0.5f,t)); + mPath[i].mScale.set(lerp(1,params.getScale().mV[0],t), + lerp(1,params.getScale().mV[1],t), 0,1); mPath[i].mTexT = t; - mPath[i].mRot.setQuat(F_PI * params.getTwist() * t,1,0,0); + LLQuaternion quat; + quat.setQuat(F_PI * params.getTwist() * t,1,0,0); + LLMatrix3 tmp(quat); + mPath[i].mRot.loadu(tmp); } break; @@ -1289,19 +1671,21 @@ BOOL LLPath::generate(const LLPathParams& params, F32 detail, S32 split, BOOL LLDynamicPath::generate(const LLPathParams& params, F32 detail, S32 split, BOOL is_sculpted, S32 sculpt_size) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - mOpen = TRUE; // Draw end caps if (getPathLength() == 0) { // Path hasn't been generated yet. // Some algorithms later assume at least TWO path points. resizePath(2); + LLQuaternion quat; + quat.setQuat(0,0,0); + LLMatrix3 tmp(quat); + for (U32 i = 0; i < 2; i++) { - mPath[i].mPos.setVec(0, 0, 0); - mPath[i].mRot.setQuat(0, 0, 0); - mPath[i].mScale.setVec(1, 1); + mPath[i].mPos.set(0, 0, 0); + mPath[i].mRot.loadu(tmp); + mPath[i].mScale.set(1, 1, 0, 1); mPath[i].mTexT = 0; } } @@ -1312,8 +1696,6 @@ BOOL LLDynamicPath::generate(const LLPathParams& params, F32 detail, S32 split, BOOL LLPathParams::importFile(LLFILE *fp) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - const S32 BUFSIZE = 16384; char buffer[BUFSIZE]; /* Flawfinder: ignore */ // *NOTE: changing the size or type of these buffers will require @@ -1424,7 +1806,7 @@ BOOL LLPathParams::importFile(LLFILE *fp) } else { - llwarns << "unknown keyword " << " in path import" << llendl; + LL_WARNS() << "unknown keyword " << " in path import" << LL_ENDL; } } return TRUE; @@ -1458,8 +1840,6 @@ BOOL LLPathParams::exportFile(LLFILE *fp) const BOOL LLPathParams::importLegacyStream(std::istream& input_stream) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - const S32 BUFSIZE = 16384; char buffer[BUFSIZE]; /* Flawfinder: ignore */ // *NOTE: changing the size or type of these buffers will require @@ -1566,7 +1946,7 @@ BOOL LLPathParams::importLegacyStream(std::istream& input_stream) } else { - llwarns << "unknown keyword " << " in path import" << llendl; + LL_WARNS() << "unknown keyword " << " in path import" << LL_ENDL; } } return TRUE; @@ -1657,7 +2037,7 @@ LLProfile::~LLProfile() { if(profile_delete_lock) { - llerrs << "LLProfile should not be deleted here!" << llendl ; + LL_ERRS() << "LLProfile should not be deleted here!" << LL_ENDL ; } } @@ -1667,13 +2047,18 @@ S32 LLVolume::sNumMeshPoints = 0; LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL generate_single_face, const BOOL is_unique) : mParams(params) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - mUnique = is_unique; mFaceMask = 0x0; mDetail = detail; mSculptLevel = -2; - + mSurfaceArea = 1.f; //only calculated for sculpts, defaults to 1 for all other prims + mIsMeshAssetLoaded = FALSE; + mLODScaleBias.setVec(1,1,1); + mHullPoints = NULL; + mHullIndices = NULL; + mNumHullPoints = 0; + mNumHullIndices = 0; + // set defaults if (mParams.getPathParams().getCurveType() == LL_PCODE_PATH_FLEXIBLE) { @@ -1688,7 +2073,8 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mGenerateSingleFace = generate_single_face; generate(); - if (mParams.getSculptID().isNull()) + + if ((mParams.getSculptID().isNull() && mParams.getSculptType() == LL_SCULPT_TYPE_NONE) || mParams.getSculptType() == LL_SCULPT_TYPE_MESH) { createVolumeFaces(); } @@ -1698,6 +2084,7 @@ void LLVolume::resizePath(S32 length) { mPathp->resizePath(length); mVolumeFaces.clear(); + setDirty(); } void LLVolume::regen() @@ -1706,9 +2093,9 @@ void LLVolume::regen() createVolumeFaces(); } -void LLVolume::genBinormals(S32 face) +void LLVolume::genTangents(S32 face) { - mVolumeFaces[face].createBinormals(); + mVolumeFaces[face].createTangents(); } LLVolume::~LLVolume() @@ -1723,11 +2110,16 @@ LLVolume::~LLVolume() mPathp = NULL; mProfilep = NULL; mVolumeFaces.clear(); + + ll_aligned_free_16(mHullPoints); + mHullPoints = NULL; + ll_aligned_free_16(mHullIndices); + mHullIndices = NULL; } BOOL LLVolume::generate() { - LLMemType m1(LLMemType::MTYPE_VOLUME); + LL_CHECK_MEMORY llassert_always(mProfilep); //Added 10.03.05 Dave Parks @@ -1764,20 +2156,6 @@ BOOL LLVolume::generate() mLODScaleBias.setVec(0.6f, 0.6f, 0.6f); } - //******************************************************************** - //debug info, to be removed - if((U32)(mPathp->mPath.size() * mProfilep->mProfile.size()) > (1u << 20)) - { - llinfos << "sizeS: " << mPathp->mPath.size() << " sizeT: " << mProfilep->mProfile.size() << llendl ; - llinfos << "path_detail : " << path_detail << " split: " << split << " profile_detail: " << profile_detail << llendl ; - llinfos << mParams << llendl ; - llinfos << "more info to check if mProfilep is deleted or not." << llendl ; - llinfos << mProfilep->mNormals.size() << " : " << mProfilep->mFaces.size() << " : " << mProfilep->mEdgeNormals.size() << " : " << mProfilep->mEdgeCenters.size() << llendl ; - - llerrs << "LLVolume corrupted!" << llendl ; - } - //******************************************************************** - BOOL regenPath = mPathp->generate(mParams.getPathParams(), path_detail, split); BOOL regenProf = mProfilep->generate(mParams.getProfileParams(), mPathp->isOpen(),profile_detail, split); @@ -1786,21 +2164,6 @@ BOOL LLVolume::generate() S32 sizeS = mPathp->mPath.size(); S32 sizeT = mProfilep->mProfile.size(); - //******************************************************************** - //debug info, to be removed - if((U32)(sizeS * sizeT) > (1u << 20)) - { - llinfos << "regenPath: " << (S32)regenPath << " regenProf: " << (S32)regenProf << llendl ; - llinfos << "sizeS: " << sizeS << " sizeT: " << sizeT << llendl ; - llinfos << "path_detail : " << path_detail << " split: " << split << " profile_detail: " << profile_detail << llendl ; - llinfos << mParams << llendl ; - llinfos << "more info to check if mProfilep is deleted or not." << llendl ; - llinfos << mProfilep->mNormals.size() << " : " << mProfilep->mFaces.size() << " : " << mProfilep->mEdgeNormals.size() << " : " << mProfilep->mEdgeCenters.size() << llendl ; - - llerrs << "LLVolume corrupted!" << llendl ; - } - //******************************************************************** - sNumMeshPoints -= mMesh.size(); mMesh.resize(sizeT * sizeS); sNumMeshPoints += mMesh.size(); @@ -1808,22 +2171,39 @@ BOOL LLVolume::generate() //generate vertex positions // Run along the path. + LLVector4a* dst = mMesh.mArray; + for (S32 s = 0; s < sizeS; ++s) { - LLVector2 scale = mPathp->mPath[s].mScale; - LLQuaternion rot = mPathp->mPath[s].mRot; + F32* scale = mPathp->mPath[s].mScale.getF32ptr(); + + F32 sc [] = + { scale[0], 0, 0, 0, + 0, scale[1], 0, 0, + 0, 0, scale[2], 0, + 0, 0, 0, 1 }; + + LLMatrix4 rot((F32*) mPathp->mPath[s].mRot.mMatrix); + LLMatrix4 scale_mat(sc); + + scale_mat *= rot; + + LLMatrix4a rot_mat; + rot_mat.loadu(scale_mat); + + LLVector4a* profile = mProfilep->mProfile.mArray; + LLVector4a* end_profile = profile+sizeT; + LLVector4a offset = mPathp->mPath[s].mPos; + + LLVector4a tmp; // Run along the profile. - for (S32 t = 0; t < sizeT; ++t) + while (profile < end_profile) { - S32 m = s*sizeT + t; - Point& pt = mMesh[m]; - - pt.mPos.mV[0] = mProfilep->mProfile[t].mV[0] * scale.mV[0]; - pt.mPos.mV[1] = mProfilep->mProfile[t].mV[1] * scale.mV[1]; - pt.mPos.mV[2] = 0.0f; - pt.mPos = pt.mPos * rot; - pt.mPos += mPathp->mPath[s].mPos; + rot_mat.rotate(*profile++, tmp); + dst->setAdd(tmp,offset); + llassert(dst->isFinite3()); + ++dst; } } @@ -1833,17 +2213,510 @@ BOOL LLVolume::generate() LLFaceID id = iter->mFaceID; mFaceMask |= id; } - + LL_CHECK_MEMORY return TRUE; } + + LL_CHECK_MEMORY 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; +} -void LLVolume::createVolumeFaces() +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) { - LLMemType m1(LLMemType::MTYPE_VOLUME); + 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; + + const F32 epsilon = 0.00001f; + + if (rhs.mData[POSITION].equals3(mData[POSITION], epsilon) && + fabs(rhs.mTexCoord[0]-mTexCoord[0]) < epsilon && + fabs(rhs.mTexCoord[1]-mTexCoord[1]) < epsilon) + { + if (angle_cutoff > 1.f) + { + retval = (mData[NORMAL].equals3(rhs.mData[NORMAL], epsilon)); + } + else + { + F32 cur_angle = rhs.mData[NORMAL].dot3(mData[NORMAL]).getF32(); + retval = cur_angle > angle_cutoff; + } + } + + return retval; +} + +bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) +{ + //input stream is now pointing at a zlib compressed block of LLSD + //decompress block + LLSD mdl; + if (!unzip_llsd(mdl, is, size)) + { + LL_DEBUGS("MeshStreaming") << "Failed to unzip LLSD blob for LoD, will probably fetch from sim again." << LL_ENDL; + return false; + } + + { + U32 face_count = mdl.size(); + + if (face_count == 0) + { //no faces unpacked, treat as failed decode + LL_WARNS() << "found no faces!" << LL_ENDL; + return false; + } + + mVolumeFaces.resize(face_count); + + for (U32 i = 0; i < face_count; ++i) + { + LLVolumeFace& face = mVolumeFaces[i]; + + if (mdl[i].has("NoGeometry")) + { //face has no geometry, continue + face.resizeIndices(3); + face.resizeVertices(1); + memset(face.mPositions, 0, sizeof(LLVector4a)); + memset(face.mNormals, 0, sizeof(LLVector4a)); + memset(face.mTexCoords, 0, sizeof(LLVector2)); + memset(face.mIndices, 0, sizeof(U16)*3); + continue; + } + + LLSD::Binary pos = mdl[i]["Position"]; + LLSD::Binary norm = mdl[i]["Normal"]; + LLSD::Binary tc = mdl[i]["TexCoord0"]; + LLSD::Binary idx = mdl[i]["TriangleList"]; + + + + //copy out indices + face.resizeIndices(idx.size()/2); + + if (idx.empty() || face.mNumIndices < 3) + { //why is there an empty index list? + LL_WARNS() <<"Empty face present!" << LL_ENDL; + continue; + } + + U16* indices = (U16*) &(idx[0]); + U32 count = idx.size()/2; + for (U32 j = 0; j < count; ++j) + { + face.mIndices[j] = indices[j]; + } + + //copy out vertices + U32 num_verts = pos.size()/(3*2); + face.resizeVertices(num_verts); + + LLVector3 minp; + LLVector3 maxp; + LLVector2 min_tc; + LLVector2 max_tc; + + minp.setValue(mdl[i]["PositionDomain"]["Min"]); + maxp.setValue(mdl[i]["PositionDomain"]["Max"]); + LLVector4a min_pos, max_pos; + min_pos.load3(minp.mV); + max_pos.load3(maxp.mV); + + min_tc.setValue(mdl[i]["TexCoord0Domain"]["Min"]); + max_tc.setValue(mdl[i]["TexCoord0Domain"]["Max"]); + + LLVector4a pos_range; + pos_range.setSub(max_pos, min_pos); + LLVector2 tc_range2 = max_tc - min_tc; + + LLVector4a tc_range; + tc_range.set(tc_range2[0], tc_range2[1], tc_range2[0], tc_range2[1]); + LLVector4a min_tc4(min_tc[0], min_tc[1], min_tc[0], min_tc[1]); + + LLVector4a* pos_out = face.mPositions; + LLVector4a* norm_out = face.mNormals; + LLVector4a* tc_out = (LLVector4a*) face.mTexCoords; + + { + U16* v = (U16*) &(pos[0]); + for (U32 j = 0; j < num_verts; ++j) + { + pos_out->set((F32) v[0], (F32) v[1], (F32) v[2]); + pos_out->div(65535.f); + pos_out->mul(pos_range); + pos_out->add(min_pos); + pos_out++; + v += 3; + } + + } + + { + if (!norm.empty()) + { + U16* n = (U16*) &(norm[0]); + for (U32 j = 0; j < num_verts; ++j) + { + norm_out->set((F32) n[0], (F32) n[1], (F32) n[2]); + norm_out->div(65535.f); + norm_out->mul(2.f); + norm_out->sub(1.f); + norm_out++; + n += 3; + } + } + else + { + memset(norm_out, 0, sizeof(LLVector4a)*num_verts); + } + } + + { + if (!tc.empty()) + { + U16* t = (U16*) &(tc[0]); + for (U32 j = 0; j < num_verts; j+=2) + { + if (j < num_verts-1) + { + tc_out->set((F32) t[0], (F32) t[1], (F32) t[2], (F32) t[3]); + } + else + { + tc_out->set((F32) t[0], (F32) t[1], 0.f, 0.f); + } + + t += 4; + + tc_out->div(65535.f); + tc_out->mul(tc_range); + tc_out->add(min_tc4); + + tc_out++; + } + } + else + { + memset(tc_out, 0, sizeof(LLVector2)*num_verts); + } + } + + if (mdl[i].has("Weights")) + { + face.allocateWeights(num_verts); + + LLSD::Binary weights = mdl[i]["Weights"]; + + U32 idx = 0; + + U32 cur_vertex = 0; + while (idx < weights.size() && cur_vertex < num_verts) + { + const U8 END_INFLUENCES = 0xFF; + U8 joint = weights[idx++]; + + U32 cur_influence = 0; + LLVector4 wght(0,0,0,0); + while (joint != END_INFLUENCES && idx < weights.size()) + { + U16 influence = weights[idx++]; + influence |= ((U16) weights[idx++] << 8); + + F32 w = llclamp((F32) influence / 65535.f, 0.f, 0.99999f); + wght.mV[cur_influence++] = (F32) joint + w; + + if (cur_influence >= 4) + { + joint = END_INFLUENCES; + } + else + { + joint = weights[idx++]; + } + } + + face.mWeights[cur_vertex].loadua(wght.mV); + + cur_vertex++; + } + + if (cur_vertex != num_verts || idx != weights.size()) + { + LL_WARNS() << "Vertex weight count does not match vertex count!" << LL_ENDL; + } + + } + + // modifier flags? + bool do_mirror = (mParams.getSculptType() & LL_SCULPT_FLAG_MIRROR); + bool do_invert = (mParams.getSculptType() &LL_SCULPT_FLAG_INVERT); + + + // translate to actions: + bool do_reflect_x = false; + bool do_reverse_triangles = false; + bool do_invert_normals = false; + + if (do_mirror) + { + do_reflect_x = true; + do_reverse_triangles = !do_reverse_triangles; + } + + if (do_invert) + { + do_invert_normals = true; + do_reverse_triangles = !do_reverse_triangles; + } + + // now do the work + + if (do_reflect_x) + { + LLVector4a* p = (LLVector4a*) face.mPositions; + LLVector4a* n = (LLVector4a*) face.mNormals; + + for (S32 i = 0; i < face.mNumVertices; i++) + { + p[i].mul(-1.0f); + n[i].mul(-1.0f); + } + } + + if (do_invert_normals) + { + LLVector4a* n = (LLVector4a*) face.mNormals; + + for (S32 i = 0; i < face.mNumVertices; i++) + { + n[i].mul(-1.0f); + } + } + + if (do_reverse_triangles) + { + for (U32 j = 0; j < face.mNumIndices; j += 3) + { + // swap the 2nd and 3rd index + S32 swap = face.mIndices[j+1]; + face.mIndices[j+1] = face.mIndices[j+2]; + face.mIndices[j+2] = swap; + } + } + + //calculate bounding box + LLVector4a& min = face.mExtents[0]; + LLVector4a& max = face.mExtents[1]; + + if (face.mNumVertices < 3) + { //empty face, use a dummy 1cm (at 1m scale) bounding box + min.splat(-0.005f); + max.splat(0.005f); + } + else + { + min = max = face.mPositions[0]; + + for (S32 i = 1; i < face.mNumVertices; ++i) + { + min.setMin(min, face.mPositions[i]); + max.setMax(max, face.mPositions[i]); + } + + if (face.mTexCoords) + { + LLVector2& min_tc = face.mTexCoordExtents[0]; + LLVector2& max_tc = face.mTexCoordExtents[1]; + + min_tc = face.mTexCoords[0]; + max_tc = face.mTexCoords[0]; + + for (U32 j = 1; j < face.mNumVertices; ++j) + { + update_min_max(min_tc, max_tc, face.mTexCoords[j]); + } + } + else + { + face.mTexCoordExtents[0].set(0,0); + face.mTexCoordExtents[1].set(1,1); + } + } + } + } + + mSculptLevel = 0; // success! + + cacheOptimize(); + + return true; +} + + +BOOL LLVolume::isMeshAssetLoaded() +{ + return mIsMeshAssetLoaded; +} + +void LLVolume::setMeshAssetLoaded(BOOL loaded) +{ + mIsMeshAssetLoaded = loaded; +} + +void LLVolume::copyFacesTo(std::vector<LLVolumeFace> &faces) const +{ + faces = mVolumeFaces; +} + +void LLVolume::copyFacesFrom(const std::vector<LLVolumeFace> &faces) +{ + mVolumeFaces = faces; + mSculptLevel = 0; +} + +void LLVolume::copyVolumeFaces(const LLVolume* volume) +{ + mVolumeFaces = volume->mVolumeFaces; + mSculptLevel = 0; +} + +void LLVolume::cacheOptimize() +{ + for (S32 i = 0; i < mVolumeFaces.size(); ++i) + { + mVolumeFaces[i].cacheOptimize(); + } +} + + +S32 LLVolume::getNumFaces() const +{ + return mIsMeshAssetLoaded ? getNumVolumeFaces() : (S32)mProfilep->mFaces.size(); +} + + +void LLVolume::createVolumeFaces() +{ if (mGenerateSingleFace) { // do nothing @@ -1864,6 +2737,11 @@ void LLVolume::createVolumeFaces() LLProfile::Face& face = mProfilep->mFaces[i]; vf.mBeginS = face.mIndex; vf.mNumS = face.mCount; + if (vf.mNumS < 0) + { + LL_ERRS() << "Volume face corruption detected." << LL_ENDL; + } + vf.mBeginT = 0; vf.mNumT= getPath().mPath.size(); vf.mID = i; @@ -1907,6 +2785,10 @@ void LLVolume::createVolumeFaces() if (face.mFlat && vf.mNumS > 2) { //flat inner faces have to copy vert normals vf.mNumS = vf.mNumS*2; + if (vf.mNumS < 0) + { + LL_ERRS() << "Volume face corruption detected." << LL_ENDL; + } } } else @@ -1925,13 +2807,15 @@ void LLVolume::createVolumeFaces() } -inline LLVector3 sculpt_rgb_to_vector(U8 r, U8 g, U8 b) +inline LLVector4a sculpt_rgb_to_vector(U8 r, U8 g, U8 b) { // maps RGB values to vector values [0..255] -> [-0.5..0.5] - LLVector3 value; - value.mV[VX] = r / 255.f - 0.5f; - value.mV[VY] = g / 255.f - 0.5f; - value.mV[VZ] = b / 255.f - 0.5f; + LLVector4a value; + LLVector4a sub(0.5f, 0.5f, 0.5f); + + value.set(r,g,b); + value.mul(1.f/255.f); + value.sub(sub); return value; } @@ -1952,21 +2836,21 @@ inline U32 sculpt_st_to_index(S32 s, S32 t, S32 size_s, S32 size_t, U16 sculpt_w } -inline LLVector3 sculpt_index_to_vector(U32 index, const U8* sculpt_data) +inline LLVector4a sculpt_index_to_vector(U32 index, const U8* sculpt_data) { - LLVector3 v = sculpt_rgb_to_vector(sculpt_data[index], sculpt_data[index+1], sculpt_data[index+2]); + LLVector4a v = sculpt_rgb_to_vector(sculpt_data[index], sculpt_data[index+1], sculpt_data[index+2]); return v; } -inline LLVector3 sculpt_st_to_vector(S32 s, S32 t, S32 size_s, S32 size_t, U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data) +inline LLVector4a sculpt_st_to_vector(S32 s, S32 t, S32 size_s, S32 size_t, U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data) { U32 index = sculpt_st_to_index(s, t, size_s, size_t, sculpt_width, sculpt_height, sculpt_components); return sculpt_index_to_vector(index, sculpt_data); } -inline LLVector3 sculpt_xy_to_vector(U32 x, U32 y, U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data) +inline LLVector4a sculpt_xy_to_vector(U32 x, U32 y, U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data) { U32 index = sculpt_xy_to_index(x, y, sculpt_width, sculpt_height, sculpt_components); @@ -1988,15 +2872,26 @@ F32 LLVolume::sculptGetSurfaceArea() for (S32 t = 0; t < sizeT-1; t++) { // get four corners of quad - LLVector3 p1 = mMesh[(s )*sizeT + (t )].mPos; - LLVector3 p2 = mMesh[(s+1)*sizeT + (t )].mPos; - LLVector3 p3 = mMesh[(s )*sizeT + (t+1)].mPos; - LLVector3 p4 = mMesh[(s+1)*sizeT + (t+1)].mPos; + LLVector4a& p1 = mMesh[(s )*sizeT + (t )]; + LLVector4a& p2 = mMesh[(s+1)*sizeT + (t )]; + LLVector4a& p3 = mMesh[(s )*sizeT + (t+1)]; + LLVector4a& p4 = mMesh[(s+1)*sizeT + (t+1)]; // compute the area of the quad by taking the length of the cross product of the two triangles - LLVector3 cross1 = (p1 - p2) % (p1 - p3); - LLVector3 cross2 = (p4 - p2) % (p4 - p3); - area += (cross1.magVec() + cross2.magVec()) / 2.0; + LLVector4a v0,v1,v2,v3; + v0.setSub(p1,p2); + v1.setSub(p1,p3); + v2.setSub(p4,p2); + v3.setSub(p4,p3); + + LLVector4a cross1, cross2; + cross1.setCross3(v0,v1); + cross2.setCross3(v2,v3); + + //LLVector3 cross1 = (p1 - p2) % (p1 - p3); + //LLVector3 cross2 = (p4 - p2) % (p4 - p3); + + area += (cross1.getLength3() + cross2.getLength3()).getF32() / 2.f; } } @@ -2006,8 +2901,6 @@ F32 LLVolume::sculptGetSurfaceArea() // create placeholder shape void LLVolume::sculptGeneratePlaceholder() { - LLMemType m1(LLMemType::MTYPE_VOLUME); - S32 sizeS = mPathp->mPath.size(); S32 sizeT = mProfilep->mProfile.size(); @@ -2019,7 +2912,7 @@ void LLVolume::sculptGeneratePlaceholder() for (S32 t = 0; t < sizeT; t++) { S32 i = t + line; - Point& pt = mMesh[i]; + LLVector4a& pt = mMesh[i]; F32 u = (F32)s/(sizeS-1); @@ -2027,9 +2920,13 @@ void LLVolume::sculptGeneratePlaceholder() const F32 RADIUS = (F32) 0.3; - pt.mPos.mV[0] = (F32)(sin(F_PI * v) * cos(2.0 * F_PI * u) * RADIUS); - pt.mPos.mV[1] = (F32)(sin(F_PI * v) * sin(2.0 * F_PI * u) * RADIUS); - pt.mPos.mV[2] = (F32)(cos(F_PI * v) * RADIUS); + F32* p = pt.getF32ptr(); + + p[0] = (F32)(sin(F_PI * v) * cos(2.0 * F_PI * u) * RADIUS); + p[1] = (F32)(sin(F_PI * v) * sin(2.0 * F_PI * u) * RADIUS); + p[2] = (F32)(cos(F_PI * v) * RADIUS); + + llassert(pt.isFinite3()); } line += sizeT; @@ -2044,9 +2941,6 @@ void LLVolume::sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 BOOL sculpt_mirror = sculpt_type & LL_SCULPT_FLAG_MIRROR; BOOL reverse_horizontal = (sculpt_invert ? !sculpt_mirror : sculpt_mirror); // XOR - - LLMemType m1(LLMemType::MTYPE_VOLUME); - S32 sizeS = mPathp->mPath.size(); S32 sizeT = mProfilep->mProfile.size(); @@ -2057,7 +2951,7 @@ void LLVolume::sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 for (S32 t = 0; t < sizeT; t++) { S32 i = t + line; - Point& pt = mMesh[i]; + LLVector4a& pt = mMesh[i]; S32 reversed_t = t; @@ -2114,12 +3008,15 @@ void LLVolume::sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 } } - pt.mPos = sculpt_xy_to_vector(x, y, sculpt_width, sculpt_height, sculpt_components, sculpt_data); + pt = sculpt_xy_to_vector(x, y, sculpt_width, sculpt_height, sculpt_components, sculpt_data); if (sculpt_mirror) { - pt.mPos.mV[VX] *= -1.f; + LLVector4a scale(-1.f,1,1,1); + pt.mul(scale); } + + llassert(pt.isFinite3()); } line += sizeT; @@ -2183,7 +3080,7 @@ void sculpt_calc_mesh_resolution(U16 width, U16 height, U8 type, F32 detail, S32 ratio = (F32) width / (F32) height; - s = (S32)fsqrtf(((F32)vertices / ratio)); + s = (S32)(F32) sqrt(((F32)vertices / ratio)); s = llmax(s, 4); // no degenerate sizes, please t = vertices / s; @@ -2195,7 +3092,6 @@ void sculpt_calc_mesh_resolution(U16 width, U16 height, U8 type, F32 detail, S32 // sculpt replaces generate() for sculpted surfaces void LLVolume::sculpt(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, S32 sculpt_level) { - LLMemType m1(LLMemType::MTYPE_VOLUME); U8 sculpt_type = mParams.getSculptType(); BOOL data_is_empty = FALSE; @@ -2220,7 +3116,7 @@ void LLVolume::sculpt(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, // weird crash bug - DEV-11158 - trying to collect more data: if ((sizeS == 0) || (sizeT == 0)) { - llwarns << "sculpt bad mesh size " << sizeS << " " << sizeT << llendl; + LL_WARNS() << "sculpt bad mesh size " << sizeS << " " << sizeT << LL_ENDL; } sNumMeshPoints -= mMesh.size(); @@ -2235,7 +3131,13 @@ void LLVolume::sculpt(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, // don't test lowest LOD to support legacy content DEV-33670 if (mDetail > SCULPT_MIN_AREA_DETAIL) { - if (sculptGetSurfaceArea() < SCULPT_MIN_AREA) + F32 area = sculptGetSurfaceArea(); + + mSurfaceArea = area; + + const F32 SCULPT_MAX_AREA = 384.f; + + if (area < SCULPT_MIN_AREA || area > SCULPT_MAX_AREA) { data_is_empty = TRUE; } @@ -2276,6 +3178,16 @@ BOOL LLVolume::isFlat(S32 face) } +bool LLVolumeParams::isSculpt() const +{ + return mSculptID.notNull(); +} + +bool LLVolumeParams::isMeshSculpt() const +{ + return isSculpt() && ((mSculptType & LL_SCULPT_TYPE_MASK) == LL_SCULPT_TYPE_MESH); +} + bool LLVolumeParams::operator==(const LLVolumeParams ¶ms) const { return ( (getPathParams() == params.getPathParams()) && @@ -2309,7 +3221,6 @@ bool LLVolumeParams::operator<(const LLVolumeParams ¶ms) const return mSculptID < params.mSculptID; } - return mSculptType < params.mSculptType; @@ -2317,7 +3228,6 @@ bool LLVolumeParams::operator<(const LLVolumeParams ¶ms) const void LLVolumeParams::copyParams(const LLVolumeParams ¶ms) { - LLMemType m1(LLMemType::MTYPE_VOLUME); mProfileParams.copyParams(params.mProfileParams); mPathParams.copyParams(params.mPathParams); mSculptID = params.getSculptID(); @@ -2601,16 +3511,16 @@ bool LLVolumeParams::setType(U8 profile, U8 path) // Bad profile. Make it square. profile = LL_PCODE_PROFILE_SQUARE; result = false; - llwarns << "LLVolumeParams::setType changing bad profile type (" << profile_type - << ") to be LL_PCODE_PROFILE_SQUARE" << llendl; + LL_WARNS() << "LLVolumeParams::setType changing bad profile type (" << profile_type + << ") to be LL_PCODE_PROFILE_SQUARE" << LL_ENDL; } else if (hole_type > LL_PCODE_HOLE_MAX) { // Bad hole. Make it the same. profile = profile_type; result = false; - llwarns << "LLVolumeParams::setType changing bad hole type (" << hole_type - << ") to be LL_PCODE_HOLE_SAME" << llendl; + LL_WARNS() << "LLVolumeParams::setType changing bad hole type (" << hole_type + << ") to be LL_PCODE_HOLE_SAME" << LL_ENDL; } if (path_type < LL_PCODE_PATH_MIN || @@ -2618,8 +3528,8 @@ bool LLVolumeParams::setType(U8 profile, U8 path) { // Bad path. Make it linear. result = false; - llwarns << "LLVolumeParams::setType changing bad path (" << path - << ") to be LL_PCODE_PATH_LINE" << llendl; + LL_WARNS() << "LLVolumeParams::setType changing bad path (" << path + << ") to be LL_PCODE_PATH_LINE" << LL_ENDL; path = LL_PCODE_PATH_LINE; } @@ -2687,714 +3597,87 @@ bool LLVolumeParams::validate(U8 prof_curve, F32 prof_begin, F32 prof_end, F32 h return true; } -S32 *LLVolume::getTriangleIndices(U32 &num_indices) const -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - S32 expected_num_triangle_indices = getNumTriangleIndices(); - if (expected_num_triangle_indices > MAX_VOLUME_TRIANGLE_INDICES) +void LLVolume::getLoDTriangleCounts(const LLVolumeParams& params, S32* counts) +{ //attempt to approximate the number of triangles that will result from generating a volume LoD set for the + //supplied LLVolumeParams -- inaccurate, but a close enough approximation for determining streaming cost + F32 detail[] = {1.f, 1.5f, 2.5f, 4.f}; + for (S32 i = 0; i < 4; i++) { - // we don't allow LLVolumes with this many vertices - llwarns << "Couldn't allocate triangle indices" << llendl; - num_indices = 0; - return NULL; - } - - S32* index = new S32[expected_num_triangle_indices]; - S32 count = 0; - - // Let's do this totally diffently, as we don't care about faces... - // Counter-clockwise triangles are forward facing... - - BOOL open = getProfile().isOpen(); - BOOL hollow = (mParams.getProfileParams().getHollow() > 0); - BOOL path_open = getPath().isOpen(); - S32 size_s, size_s_out, size_t; - S32 s, t, i; - size_s = getProfile().getTotal(); - size_s_out = getProfile().getTotalOut(); - size_t = getPath().mPath.size(); - - // NOTE -- if the construction of the triangles below ever changes - // then getNumTriangleIndices() method may also have to be updated. - - if (open) /* Flawfinder: ignore */ - { - if (hollow) - { - // Open hollow -- much like the closed solid, except we - // we need to stitch up the gap between s=0 and s=size_s-1 - - for (t = 0; t < size_t - 1; t++) - { - // The outer face, first cut, and inner face - for (s = 0; s < size_s - 1; s++) - { - i = s + t*size_s; - index[count++] = i; // x,y - index[count++] = i + 1; // x+1,y - index[count++] = i + size_s; // x,y+1 - - index[count++] = i + size_s; // x,y+1 - index[count++] = i + 1; // x+1,y - index[count++] = i + size_s + 1; // x+1,y+1 - } - - // The other cut face - index[count++] = s + t*size_s; // x,y - index[count++] = 0 + t*size_s; // x+1,y - index[count++] = s + (t+1)*size_s; // x,y+1 - - index[count++] = s + (t+1)*size_s; // x,y+1 - index[count++] = 0 + t*size_s; // x+1,y - index[count++] = 0 + (t+1)*size_s; // x+1,y+1 - } - - // Do the top and bottom caps, if necessary - if (path_open) - { - // Top cap - S32 pt1 = 0; - S32 pt2 = size_s-1; - S32 i = (size_t - 1)*size_s; - - while (pt2 - pt1 > 1) - { - // Use the profile points instead of the mesh, since you want - // the un-transformed profile distances. - LLVector3 p1 = getProfile().mProfile[pt1]; - LLVector3 p2 = getProfile().mProfile[pt2]; - LLVector3 pa = getProfile().mProfile[pt1+1]; - LLVector3 pb = getProfile().mProfile[pt2-1]; - - p1.mV[VZ] = 0.f; - p2.mV[VZ] = 0.f; - pa.mV[VZ] = 0.f; - pb.mV[VZ] = 0.f; - - // Use area of triangle to determine backfacing - F32 area_1a2, area_1ba, area_21b, area_2ab; - area_1a2 = (p1.mV[0]*pa.mV[1] - pa.mV[0]*p1.mV[1]) + - (pa.mV[0]*p2.mV[1] - p2.mV[0]*pa.mV[1]) + - (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]); - - area_1ba = (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*pa.mV[1] - pa.mV[0]*pb.mV[1]) + - (pa.mV[0]*p1.mV[1] - p1.mV[0]*pa.mV[1]); - - area_21b = (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]) + - (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); - - area_2ab = (p2.mV[0]*pa.mV[1] - pa.mV[0]*p2.mV[1]) + - (pa.mV[0]*pb.mV[1] - pb.mV[0]*pa.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); - - BOOL use_tri1a2 = TRUE; - BOOL tri_1a2 = TRUE; - BOOL tri_21b = TRUE; - - if (area_1a2 < 0) - { - tri_1a2 = FALSE; - } - if (area_2ab < 0) - { - // Can't use, because it contains point b - tri_1a2 = FALSE; - } - if (area_21b < 0) - { - tri_21b = FALSE; - } - if (area_1ba < 0) - { - // Can't use, because it contains point b - tri_21b = FALSE; - } - - if (!tri_1a2) - { - use_tri1a2 = FALSE; - } - else if (!tri_21b) - { - use_tri1a2 = TRUE; - } - else - { - LLVector3 d1 = p1 - pa; - LLVector3 d2 = p2 - pb; - - if (d1.magVecSquared() < d2.magVecSquared()) - { - use_tri1a2 = TRUE; - } - else - { - use_tri1a2 = FALSE; - } - } - - if (use_tri1a2) - { - index[count++] = pt1 + i; - index[count++] = pt1 + 1 + i; - index[count++] = pt2 + i; - pt1++; - } - else - { - index[count++] = pt1 + i; - index[count++] = pt2 - 1 + i; - index[count++] = pt2 + i; - pt2--; - } - } - - // Bottom cap - pt1 = 0; - pt2 = size_s-1; - while (pt2 - pt1 > 1) - { - // Use the profile points instead of the mesh, since you want - // the un-transformed profile distances. - LLVector3 p1 = getProfile().mProfile[pt1]; - LLVector3 p2 = getProfile().mProfile[pt2]; - LLVector3 pa = getProfile().mProfile[pt1+1]; - LLVector3 pb = getProfile().mProfile[pt2-1]; - - p1.mV[VZ] = 0.f; - p2.mV[VZ] = 0.f; - pa.mV[VZ] = 0.f; - pb.mV[VZ] = 0.f; - - // Use area of triangle to determine backfacing - F32 area_1a2, area_1ba, area_21b, area_2ab; - area_1a2 = (p1.mV[0]*pa.mV[1] - pa.mV[0]*p1.mV[1]) + - (pa.mV[0]*p2.mV[1] - p2.mV[0]*pa.mV[1]) + - (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]); - - area_1ba = (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*pa.mV[1] - pa.mV[0]*pb.mV[1]) + - (pa.mV[0]*p1.mV[1] - p1.mV[0]*pa.mV[1]); - - area_21b = (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]) + - (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); - - area_2ab = (p2.mV[0]*pa.mV[1] - pa.mV[0]*p2.mV[1]) + - (pa.mV[0]*pb.mV[1] - pb.mV[0]*pa.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); - - BOOL use_tri1a2 = TRUE; - BOOL tri_1a2 = TRUE; - BOOL tri_21b = TRUE; - - if (area_1a2 < 0) - { - tri_1a2 = FALSE; - } - if (area_2ab < 0) - { - // Can't use, because it contains point b - tri_1a2 = FALSE; - } - if (area_21b < 0) - { - tri_21b = FALSE; - } - if (area_1ba < 0) - { - // Can't use, because it contains point b - tri_21b = FALSE; - } - - if (!tri_1a2) - { - use_tri1a2 = FALSE; - } - else if (!tri_21b) - { - use_tri1a2 = TRUE; - } - else - { - LLVector3 d1 = p1 - pa; - LLVector3 d2 = p2 - pb; - - if (d1.magVecSquared() < d2.magVecSquared()) - { - use_tri1a2 = TRUE; - } - else - { - use_tri1a2 = FALSE; - } - } - - if (use_tri1a2) - { - index[count++] = pt1; - index[count++] = pt2; - index[count++] = pt1 + 1; - pt1++; - } - else - { - index[count++] = pt1; - index[count++] = pt2; - index[count++] = pt2 - 1; - pt2--; - } - } - } - } - else - { - // Open solid - - for (t = 0; t < size_t - 1; t++) - { - // Outer face + 1 cut face - for (s = 0; s < size_s - 1; s++) - { - i = s + t*size_s; + S32 count = 0; + S32 path_points = LLPath::getNumPoints(params.getPathParams(), detail[i]); + S32 profile_points = LLProfile::getNumPoints(params.getProfileParams(), false, detail[i]); - index[count++] = i; // x,y - index[count++] = i + 1; // x+1,y - index[count++] = i + size_s; // x,y+1 + count = (profile_points-1)*2*(path_points-1); + count += profile_points*2; - index[count++] = i + size_s; // x,y+1 - index[count++] = i + 1; // x+1,y - index[count++] = i + size_s + 1; // x+1,y+1 - } - - // The other cut face - index[count++] = (size_s - 1) + (t*size_s); // x,y - index[count++] = 0 + t*size_s; // x+1,y - index[count++] = (size_s - 1) + (t+1)*size_s; // x,y+1 - - index[count++] = (size_s - 1) + (t+1)*size_s; // x,y+1 - index[count++] = 0 + (t*size_s); // x+1,y - index[count++] = 0 + (t+1)*size_s; // x+1,y+1 - } - - // Do the top and bottom caps, if necessary - if (path_open) - { - for (s = 0; s < size_s - 2; s++) - { - index[count++] = s+1; - index[count++] = s; - index[count++] = size_s - 1; - } - - // We've got a top cap - S32 offset = (size_t - 1)*size_s; - for (s = 0; s < size_s - 2; s++) - { - // Inverted ordering from bottom cap. - index[count++] = offset + size_s - 1; - index[count++] = offset + s; - index[count++] = offset + s + 1; - } - } - } + counts[i] = count; } - else if (hollow) - { - // Closed hollow - // Outer face - - for (t = 0; t < size_t - 1; t++) - { - for (s = 0; s < size_s_out - 1; s++) - { - i = s + t*size_s; - - index[count++] = i; // x,y - index[count++] = i + 1; // x+1,y - index[count++] = i + size_s; // x,y+1 - - index[count++] = i + size_s; // x,y+1 - index[count++] = i + 1; // x+1,y - index[count++] = i + 1 + size_s; // x+1,y+1 - } - } - - // Inner face - // Invert facing from outer face - for (t = 0; t < size_t - 1; t++) - { - for (s = size_s_out; s < size_s - 1; s++) - { - i = s + t*size_s; - - index[count++] = i; // x,y - index[count++] = i + 1; // x+1,y - index[count++] = i + size_s; // x,y+1 - - index[count++] = i + size_s; // x,y+1 - index[count++] = i + 1; // x+1,y - index[count++] = i + 1 + size_s; // x+1,y+1 - } - } - - // Do the top and bottom caps, if necessary - if (path_open) - { - // Top cap - S32 pt1 = 0; - S32 pt2 = size_s-1; - S32 i = (size_t - 1)*size_s; - - while (pt2 - pt1 > 1) - { - // Use the profile points instead of the mesh, since you want - // the un-transformed profile distances. - LLVector3 p1 = getProfile().mProfile[pt1]; - LLVector3 p2 = getProfile().mProfile[pt2]; - LLVector3 pa = getProfile().mProfile[pt1+1]; - LLVector3 pb = getProfile().mProfile[pt2-1]; - - p1.mV[VZ] = 0.f; - p2.mV[VZ] = 0.f; - pa.mV[VZ] = 0.f; - pb.mV[VZ] = 0.f; - - // Use area of triangle to determine backfacing - F32 area_1a2, area_1ba, area_21b, area_2ab; - area_1a2 = (p1.mV[0]*pa.mV[1] - pa.mV[0]*p1.mV[1]) + - (pa.mV[0]*p2.mV[1] - p2.mV[0]*pa.mV[1]) + - (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]); - - area_1ba = (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*pa.mV[1] - pa.mV[0]*pb.mV[1]) + - (pa.mV[0]*p1.mV[1] - p1.mV[0]*pa.mV[1]); - - area_21b = (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]) + - (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); - - area_2ab = (p2.mV[0]*pa.mV[1] - pa.mV[0]*p2.mV[1]) + - (pa.mV[0]*pb.mV[1] - pb.mV[0]*pa.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); - - BOOL use_tri1a2 = TRUE; - BOOL tri_1a2 = TRUE; - BOOL tri_21b = TRUE; - - if (area_1a2 < 0) - { - tri_1a2 = FALSE; - } - if (area_2ab < 0) - { - // Can't use, because it contains point b - tri_1a2 = FALSE; - } - if (area_21b < 0) - { - tri_21b = FALSE; - } - if (area_1ba < 0) - { - // Can't use, because it contains point b - tri_21b = FALSE; - } - - if (!tri_1a2) - { - use_tri1a2 = FALSE; - } - else if (!tri_21b) - { - use_tri1a2 = TRUE; - } - else - { - LLVector3 d1 = p1 - pa; - LLVector3 d2 = p2 - pb; - - if (d1.magVecSquared() < d2.magVecSquared()) - { - use_tri1a2 = TRUE; - } - else - { - use_tri1a2 = FALSE; - } - } - - if (use_tri1a2) - { - index[count++] = pt1 + i; - index[count++] = pt1 + 1 + i; - index[count++] = pt2 + i; - pt1++; - } - else - { - index[count++] = pt1 + i; - index[count++] = pt2 - 1 + i; - index[count++] = pt2 + i; - pt2--; - } - } - - // Bottom cap - pt1 = 0; - pt2 = size_s-1; - while (pt2 - pt1 > 1) - { - // Use the profile points instead of the mesh, since you want - // the un-transformed profile distances. - LLVector3 p1 = getProfile().mProfile[pt1]; - LLVector3 p2 = getProfile().mProfile[pt2]; - LLVector3 pa = getProfile().mProfile[pt1+1]; - LLVector3 pb = getProfile().mProfile[pt2-1]; - - p1.mV[VZ] = 0.f; - p2.mV[VZ] = 0.f; - pa.mV[VZ] = 0.f; - pb.mV[VZ] = 0.f; - - // Use area of triangle to determine backfacing - F32 area_1a2, area_1ba, area_21b, area_2ab; - area_1a2 = (p1.mV[0]*pa.mV[1] - pa.mV[0]*p1.mV[1]) + - (pa.mV[0]*p2.mV[1] - p2.mV[0]*pa.mV[1]) + - (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]); - - area_1ba = (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*pa.mV[1] - pa.mV[0]*pb.mV[1]) + - (pa.mV[0]*p1.mV[1] - p1.mV[0]*pa.mV[1]); - - area_21b = (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]) + - (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); - - area_2ab = (p2.mV[0]*pa.mV[1] - pa.mV[0]*p2.mV[1]) + - (pa.mV[0]*pb.mV[1] - pb.mV[0]*pa.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); - - BOOL use_tri1a2 = TRUE; - BOOL tri_1a2 = TRUE; - BOOL tri_21b = TRUE; - - if (area_1a2 < 0) - { - tri_1a2 = FALSE; - } - if (area_2ab < 0) - { - // Can't use, because it contains point b - tri_1a2 = FALSE; - } - if (area_21b < 0) - { - tri_21b = FALSE; - } - if (area_1ba < 0) - { - // Can't use, because it contains point b - tri_21b = FALSE; - } +} - if (!tri_1a2) - { - use_tri1a2 = FALSE; - } - else if (!tri_21b) - { - use_tri1a2 = TRUE; - } - else - { - LLVector3 d1 = p1 - pa; - LLVector3 d2 = p2 - pb; - if (d1.magVecSquared() < d2.magVecSquared()) - { - use_tri1a2 = TRUE; - } - else - { - use_tri1a2 = FALSE; - } - } +S32 LLVolume::getNumTriangles(S32* vcount) const +{ + U32 triangle_count = 0; + U32 vertex_count = 0; - if (use_tri1a2) - { - index[count++] = pt1; - index[count++] = pt2; - index[count++] = pt1 + 1; - pt1++; - } - else - { - index[count++] = pt1; - index[count++] = pt2; - index[count++] = pt2 - 1; - pt2--; - } - } - } - } - else + for (S32 i = 0; i < getNumVolumeFaces(); ++i) { - // Closed solid. Easy case. - for (t = 0; t < size_t - 1; t++) - { - for (s = 0; s < size_s - 1; s++) - { - // Should wrap properly, but for now... - i = s + t*size_s; - - index[count++] = i; // x,y - index[count++] = i + 1; // x+1,y - index[count++] = i + size_s; // x,y+1 + const LLVolumeFace& face = getVolumeFace(i); + triangle_count += face.mNumIndices/3; - index[count++] = i + size_s; // x,y+1 - index[count++] = i + 1; // x+1,y - index[count++] = i + size_s + 1; // x+1,y+1 - } - } - - // Do the top and bottom caps, if necessary - if (path_open) - { - // bottom cap - for (s = 1; s < size_s - 2; s++) - { - index[count++] = s+1; - index[count++] = s; - index[count++] = 0; - } - - // top cap - S32 offset = (size_t - 1)*size_s; - for (s = 1; s < size_s - 2; s++) - { - // Inverted ordering from bottom cap. - index[count++] = offset; - index[count++] = offset + s; - index[count++] = offset + s + 1; - } - } + vertex_count += face.mNumVertices; } -#ifdef LL_DEBUG - // assert that we computed the correct number of indices - if (count != expected_num_triangle_indices ) - { - llerrs << "bad index count prediciton:" - << " expected=" << expected_num_triangle_indices - << " actual=" << count << llendl; - } -#endif -#if 0 - // verify that each index does not point beyond the size of the mesh - S32 num_vertices = mMesh.size(); - for (i = 0; i < count; i+=3) + if (vcount) { - llinfos << index[i] << ":" << index[i+1] << ":" << index[i+2] << llendl; - llassert(index[i] < num_vertices); - llassert(index[i+1] < num_vertices); - llassert(index[i+2] < num_vertices); + *vcount = vertex_count; } -#endif - - num_indices = count; - return index; + + return triangle_count; } -S32 LLVolume::getNumTriangleIndices() const -{ - BOOL profile_open = getProfile().isOpen(); - BOOL hollow = (mParams.getProfileParams().getHollow() > 0); - BOOL path_open = getPath().isOpen(); - - S32 size_s, size_s_out, size_t; - size_s = getProfile().getTotal(); - size_s_out = getProfile().getTotalOut(); - size_t = getPath().mPath.size(); - - S32 count = 0; - if (profile_open) /* Flawfinder: ignore */ - { - if (hollow) - { - // Open hollow -- much like the closed solid, except we - // we need to stitch up the gap between s=0 and s=size_s-1 - count = (size_t - 1) * (((size_s -1) * 6) + 6); - } - else - { - count = (size_t - 1) * (((size_s -1) * 6) + 6); - } - } - else if (hollow) - { - // Closed hollow - // Outer face - count = (size_t - 1) * (size_s_out - 1) * 6; - - // Inner face - count += (size_t - 1) * ((size_s - 1) - size_s_out) * 6; - } - else - { - // Closed solid. Easy case. - count = (size_t - 1) * (size_s - 1) * 6; - } - - if (path_open) - { - S32 cap_triangle_count = size_s - 3; - if ( profile_open - || hollow ) - { - cap_triangle_count = size_s - 2; - } - if ( cap_triangle_count > 0 ) - { - // top and bottom caps - count += cap_triangle_count * 2 * 3; - } - } - return count; -} //----------------------------------------------------------------------------- // generateSilhouetteVertices() //----------------------------------------------------------------------------- void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices, std::vector<LLVector3> &normals, - std::vector<S32> &segments, - const LLVector3& obj_cam_vec, - const LLMatrix4& mat, - const LLMatrix3& norm_mat, + const LLVector3& obj_cam_vec_in, + const LLMatrix4& mat_in, + const LLMatrix3& norm_mat_in, S32 face_mask) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - + LLMatrix4a mat; + mat.loadu(mat_in); + + LLMatrix4a norm_mat; + norm_mat.loadu(norm_mat_in); + + LLVector4a obj_cam_vec; + obj_cam_vec.load3(obj_cam_vec_in.mV); + vertices.clear(); normals.clear(); - segments.clear(); + if ((mParams.getSculptType() & LL_SCULPT_TYPE_MASK) == LL_SCULPT_TYPE_MESH) + { + return; + } + S32 cur_index = 0; //for each face for (face_list_t::iterator iter = mVolumeFaces.begin(); iter != mVolumeFaces.end(); ++iter) { - const LLVolumeFace& face = *iter; + LLVolumeFace& face = *iter; - if (!(face_mask & (0x1 << cur_index++))) + if (!(face_mask & (0x1 << cur_index++)) || + face.mNumIndices == 0 || face.mEdge.empty()) { continue; } + if (face.mTypeMask & (LLVolumeFace::CAP_MASK)) { } @@ -3407,7 +3690,7 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &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]; @@ -3415,9 +3698,9 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &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++) { @@ -3435,9 +3718,9 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &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); @@ -3460,15 +3743,15 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &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].mTangent*0.1f); normals.push_back(LLVector3(0,0,1)); normals.push_back(LLVector3(0,0,1)); segments.push_back(vertices.size()); @@ -3486,26 +3769,36 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices, //for each triangle std::vector<U8> 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 < 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]; - LLVector3 norm = (face.mVertices[v1].mPosition - face.mVertices[v2].mPosition) % - (face.mVertices[v2].mPosition - face.mVertices[v3].mPosition); - - if (norm.magVecSquared() < 0.00000001f) + LLVector4a c1,c2; + c1.setSub(v[v1], v[v2]); + c2.setSub(v[v2], v[v3]); + + LLVector4a norm; + + norm.setCross3(c1, c2); + + if (norm.dot3(norm) < 0.00000001f) { fFacing[j] = AWAY | TOWARDS; } else { //get view vector - LLVector3 view = (obj_cam_vec-face.mVertices[v1].mPosition); - bool away = view * norm > 0.0f; + LLVector4a view; + view.setSub(obj_cam_vec, v[v1]); + bool away = view.dot3(norm) > 0.0f; if (away) { fFacing[j] = AWAY; @@ -3518,7 +3811,7 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices, } //for each triangle - for (U32 j = 0; j < face.mIndices.size()/3; j++) + for (U32 j = 0; j < face.mNumIndices/3; j++) { if (fFacing[j] == (AWAY | TOWARDS)) { //this is a degenerate triangle @@ -3551,17 +3844,21 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &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); - LLVector3 norm1 = face.mVertices[v1].mNormal * norm_mat; - norm1.normVec(); - normals.push_back(norm1); + LLVector4a t; + mat.affineTransform(v[v1], t); + vertices.push_back(LLVector3(t[0], t[1], t[2])); + + norm_mat.rotate(n[v1], t); - vertices.push_back(face.mVertices[v2].mPosition*mat); - LLVector3 norm2 = face.mVertices[v2].mNormal * norm_mat; - norm2.normVec(); - normals.push_back(norm2); + t.normalize3fast(); + normals.push_back(LLVector3(t[0], t[1], t[2])); - segments.push_back(vertices.size()); + 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])); } } } @@ -3570,9 +3867,9 @@ void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices, } } -S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, +S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, S32 face, - LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal) + LLVector4a* intersection,LLVector2* tex_coord, LLVector4a* normal, LLVector4a* tangent_out) { S32 hit_face = -1; @@ -3590,74 +3887,132 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, end_face = face; } - LLVector3 dir = end - start; + LLVector4a dir; + dir.setSub(end, start); 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]; - LLVector3 box_center = (face.mExtents[0] + face.mExtents[1]) / 2.f; - LLVector3 box_size = face.mExtents[1] - face.mExtents[0]; + LLVector4a box_center; + box_center.setAdd(face.mExtents[0], face.mExtents[1]); + box_center.mul(0.5f); + + LLVector4a box_size; + box_size.setSub(face.mExtents[1], face.mExtents[0]); if (LLLineSegmentBoxIntersect(start, end, box_center, box_size)) { - if (bi_normal != NULL) // if the caller wants binormals, we may need to generate them + if (tangent_out != NULL) // if the caller wants tangents, we may need to generate them { - genBinormals(i); + genTangents(i); } - - for (U32 tri = 0; tri < face.mIndices.size()/3; tri++) - { - 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; - - if (LLTriangleRayIntersect(face.mVertices[index1].mPosition, - face.mVertices[index2].mPosition, - face.mVertices[index3].mPosition, - start, dir, &a, &b, &t, FALSE)) - { - 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 (isUnique()) + { //don't bother with an octree for flexi volumes + U32 tri_count = face.mNumIndices/3; - if (intersection != NULL) - { - *intersection = start + dir * closest_t; - } - - if (tex_coord != NULL) - { - *tex_coord = ((1.f - a - b) * face.mVertices[index1].mTexCoord + - a * face.mVertices[index2].mTexCoord + - b * face.mVertices[index3].mTexCoord); - - } - - if (normal != NULL) + for (U32 j = 0; j < tri_count; ++j) { - *normal = ((1.f - a - b) * face.mVertices[index1].mNormal + - a * face.mVertices[index2].mNormal + - b * face.mVertices[index3].mNormal); - } + U16 idx0 = face.mIndices[j*3+0]; + U16 idx1 = face.mIndices[j*3+1]; + U16 idx2 = face.mIndices[j*3+2]; - if (bi_normal != NULL) + const LLVector4a& v0 = face.mPositions[idx0]; + const LLVector4a& v1 = face.mPositions[idx1]; + const LLVector4a& v2 = face.mPositions[idx2]; + + F32 a,b,t; + + if (LLTriangleRayIntersect(v0, v1, v2, + start, dir, a, b, t)) { - *bi_normal = ((1.f - a - b) * face.mVertices[index1].mBinormal + - a * face.mVertices[index2].mBinormal + - b * face.mVertices[index3].mBinormal); + 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 intersect = dir; + intersect.mul(closest_t); + intersect.add(start); + *intersection = intersect; + } + + + if (tex_coord != NULL) + { + LLVector2* tc = (LLVector2*) face.mTexCoords; + *tex_coord = ((1.f - a - b) * tc[idx0] + + a * tc[idx1] + + b * tc[idx2]); + + } + + if (normal!= NULL) + { + LLVector4a* norm = face.mNormals; + + LLVector4a n1,n2,n3; + n1 = norm[idx0]; + n1.mul(1.f-a-b); + + n2 = norm[idx1]; + n2.mul(a); + + n3 = norm[idx2]; + n3.mul(b); + + n1.add(n2); + n1.add(n3); + + *normal = n1; + } + + if (tangent_out != NULL) + { + LLVector4a* tangents = face.mTangents; + + LLVector4a t1,t2,t3; + t1 = tangents[idx0]; + t1.mul(1.f-a-b); + + t2 = tangents[idx1]; + t2.mul(a); + + t3 = tangents[idx2]; + t3.mul(b); + + t1.add(t2); + t1.add(t3); + + *tangent_out = t1; + } } - } } } + else + { + if (!face.mOctree) + { + face.createOctree(); + } + + LLOctreeTriangleRayIntersect intersect(start, dir, &face, &closest_t, intersection, tex_coord, normal, tangent_out); + intersect.traverse(face.mOctree); + if (intersect.mHitFace) + { + hit_face = i; + } + } } } @@ -3681,7 +4036,6 @@ LLVertexIndexPair::LLVertexIndexPair(const LLVector3 &vertex, const S32 index) } const F32 VERTEX_SLOP = 0.00001f; -const F32 VERTEX_SLOP_SQRD = VERTEX_SLOP * VERTEX_SLOP; struct lessVertex { @@ -3764,242 +4118,9 @@ BOOL equalTriangle(const S32 *a, const S32 *b) return FALSE; } -BOOL LLVolume::cleanupTriangleData( const S32 num_input_vertices, - const std::vector<Point>& input_vertices, - const S32 num_input_triangles, - S32 *input_triangles, - S32 &num_output_vertices, - LLVector3 **output_vertices, - S32 &num_output_triangles, - S32 **output_triangles) -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - /* Testing: avoid any cleanup - static BOOL skip_cleanup = TRUE; - if ( skip_cleanup ) - { - num_output_vertices = num_input_vertices; - num_output_triangles = num_input_triangles; - - *output_vertices = new LLVector3[num_input_vertices]; - for (S32 index = 0; index < num_input_vertices; index++) - { - (*output_vertices)[index] = input_vertices[index].mPos; - } - - *output_triangles = new S32[num_input_triangles*3]; - memcpy(*output_triangles, input_triangles, 3*num_input_triangles*sizeof(S32)); // Flawfinder: ignore - return TRUE; - } - */ - - // Here's how we do this: - // Create a structure which contains the original vertex index and the - // LLVector3 data. - // "Sort" the data by the vectors - // Create an array the size of the old vertex list, with a mapping of - // old indices to new indices. - // Go through triangles, shift so the lowest index is first - // Sort triangles by first index - // Remove duplicate triangles - // Allocate and pack new triangle data. - - //LLTimer cleanupTimer; - //llinfos << "In vertices: " << num_input_vertices << llendl; - //llinfos << "In triangles: " << num_input_triangles << llendl; - - S32 i; - typedef std::multiset<LLVertexIndexPair*, lessVertex> vertex_set_t; - vertex_set_t vertex_list; - - LLVertexIndexPair *pairp = NULL; - for (i = 0; i < num_input_vertices; i++) - { - LLVertexIndexPair *new_pairp = new LLVertexIndexPair(input_vertices[i].mPos, i); - vertex_list.insert(new_pairp); - } - - // Generate the vertex mapping and the list of vertices without - // duplicates. This will crash if there are no vertices. - llassert(num_input_vertices > 0); // check for no vertices! - S32 *vertex_mapping = new S32[num_input_vertices]; - LLVector3 *new_vertices = new LLVector3[num_input_vertices]; - LLVertexIndexPair *prev_pairp = NULL; - - S32 new_num_vertices; - - new_num_vertices = 0; - for (vertex_set_t::iterator iter = vertex_list.begin(), - end = vertex_list.end(); - iter != end; iter++) - { - pairp = *iter; - if (!prev_pairp || ((pairp->mVertex - prev_pairp->mVertex).magVecSquared() >= VERTEX_SLOP_SQRD)) - { - new_vertices[new_num_vertices] = pairp->mVertex; - //llinfos << "Added vertex " << new_num_vertices << " : " << pairp->mVertex << llendl; - new_num_vertices++; - // Update the previous - prev_pairp = pairp; - } - else - { - //llinfos << "Removed duplicate vertex " << pairp->mVertex << ", distance magVecSquared() is " << (pairp->mVertex - prev_pairp->mVertex).magVecSquared() << llendl; - } - vertex_mapping[pairp->mIndex] = new_num_vertices - 1; - } - - // Iterate through triangles and remove degenerates, re-ordering vertices - // along the way. - S32 *new_triangles = new S32[num_input_triangles * 3]; - S32 new_num_triangles = 0; - - for (i = 0; i < num_input_triangles; i++) - { - S32 v1 = i*3; - S32 v2 = v1 + 1; - S32 v3 = v1 + 2; - - //llinfos << "Checking triangle " << input_triangles[v1] << ":" << input_triangles[v2] << ":" << input_triangles[v3] << llendl; - input_triangles[v1] = vertex_mapping[input_triangles[v1]]; - input_triangles[v2] = vertex_mapping[input_triangles[v2]]; - input_triangles[v3] = vertex_mapping[input_triangles[v3]]; - - if ((input_triangles[v1] == input_triangles[v2]) - || (input_triangles[v1] == input_triangles[v3]) - || (input_triangles[v2] == input_triangles[v3])) - { - //llinfos << "Removing degenerate triangle " << input_triangles[v1] << ":" << input_triangles[v2] << ":" << input_triangles[v3] << llendl; - // Degenerate triangle, skip - continue; - } - - if (input_triangles[v1] < input_triangles[v2]) - { - if (input_triangles[v1] < input_triangles[v3]) - { - // (0 < 1) && (0 < 2) - new_triangles[new_num_triangles*3] = input_triangles[v1]; - new_triangles[new_num_triangles*3+1] = input_triangles[v2]; - new_triangles[new_num_triangles*3+2] = input_triangles[v3]; - } - else - { - // (0 < 1) && (2 < 0) - new_triangles[new_num_triangles*3] = input_triangles[v3]; - new_triangles[new_num_triangles*3+1] = input_triangles[v1]; - new_triangles[new_num_triangles*3+2] = input_triangles[v2]; - } - } - else if (input_triangles[v2] < input_triangles[v3]) - { - // (1 < 0) && (1 < 2) - new_triangles[new_num_triangles*3] = input_triangles[v2]; - new_triangles[new_num_triangles*3+1] = input_triangles[v3]; - new_triangles[new_num_triangles*3+2] = input_triangles[v1]; - } - else - { - // (1 < 0) && (2 < 1) - new_triangles[new_num_triangles*3] = input_triangles[v3]; - new_triangles[new_num_triangles*3+1] = input_triangles[v1]; - new_triangles[new_num_triangles*3+2] = input_triangles[v2]; - } - new_num_triangles++; - } - - if (new_num_triangles == 0) - { - llwarns << "Created volume object with 0 faces." << llendl; - delete[] new_triangles; - delete[] vertex_mapping; - delete[] new_vertices; - return FALSE; - } - - typedef std::set<S32*, lessTriangle> triangle_set_t; - triangle_set_t triangle_list; - - for (i = 0; i < new_num_triangles; i++) - { - triangle_list.insert(&new_triangles[i*3]); - } - - // Sort through the triangle list, and delete duplicates - - S32 *prevp = NULL; - S32 *curp = NULL; - - S32 *sorted_tris = new S32[new_num_triangles*3]; - S32 cur_tri = 0; - for (triangle_set_t::iterator iter = triangle_list.begin(), - end = triangle_list.end(); - iter != end; iter++) - { - curp = *iter; - if (!prevp || !equalTriangle(prevp, curp)) - { - //llinfos << "Added triangle " << *curp << ":" << *(curp+1) << ":" << *(curp+2) << llendl; - sorted_tris[cur_tri*3] = *curp; - sorted_tris[cur_tri*3+1] = *(curp+1); - sorted_tris[cur_tri*3+2] = *(curp+2); - cur_tri++; - prevp = curp; - } - else - { - //llinfos << "Skipped triangle " << *curp << ":" << *(curp+1) << ":" << *(curp+2) << llendl; - } - } - - *output_vertices = new LLVector3[new_num_vertices]; - num_output_vertices = new_num_vertices; - for (i = 0; i < new_num_vertices; i++) - { - (*output_vertices)[i] = new_vertices[i]; - } - - *output_triangles = new S32[cur_tri*3]; - num_output_triangles = cur_tri; - memcpy(*output_triangles, sorted_tris, 3*cur_tri*sizeof(S32)); /* Flawfinder: ignore */ - - /* - llinfos << "Out vertices: " << num_output_vertices << llendl; - llinfos << "Out triangles: " << num_output_triangles << llendl; - for (i = 0; i < num_output_vertices; i++) - { - llinfos << i << ":" << (*output_vertices)[i] << llendl; - } - for (i = 0; i < num_output_triangles; i++) - { - llinfos << i << ":" << (*output_triangles)[i*3] << ":" << (*output_triangles)[i*3+1] << ":" << (*output_triangles)[i*3+2] << llendl; - } - */ - - //llinfos << "Out vertices: " << num_output_vertices << llendl; - //llinfos << "Out triangles: " << num_output_triangles << llendl; - delete[] vertex_mapping; - vertex_mapping = NULL; - delete[] new_vertices; - new_vertices = NULL; - delete[] new_triangles; - new_triangles = NULL; - delete[] sorted_tris; - sorted_tris = NULL; - triangle_list.clear(); - std::for_each(vertex_list.begin(), vertex_list.end(), DeletePointer()); - vertex_list.clear(); - - return TRUE; -} - - BOOL LLVolumeParams::importFile(LLFILE *fp) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - - //llinfos << "importing volume" << llendl; + //LL_INFOS() << "importing volume" << LL_ENDL; const S32 BUFSIZE = 16384; char buffer[BUFSIZE]; /* Flawfinder: ignore */ // *NOTE: changing the size or type of this buffer will require @@ -4033,7 +4154,7 @@ BOOL LLVolumeParams::importFile(LLFILE *fp) } else { - llwarns << "unknown keyword " << keyword << " in volume import" << llendl; + LL_WARNS() << "unknown keyword " << keyword << " in volume import" << LL_ENDL; } } @@ -4053,9 +4174,7 @@ BOOL LLVolumeParams::exportFile(LLFILE *fp) const BOOL LLVolumeParams::importLegacyStream(std::istream& input_stream) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - - //llinfos << "importing volume" << llendl; + //LL_INFOS() << "importing volume" << LL_ENDL; const S32 BUFSIZE = 16384; // *NOTE: changing the size or type of this buffer will require // changing the sscanf below. @@ -4085,7 +4204,7 @@ BOOL LLVolumeParams::importLegacyStream(std::istream& input_stream) } else { - llwarns << "unknown keyword " << keyword << " in volume import" << llendl; + LL_WARNS() << "unknown keyword " << keyword << " in volume import" << LL_ENDL; } } @@ -4094,8 +4213,6 @@ BOOL LLVolumeParams::importLegacyStream(std::istream& input_stream) BOOL LLVolumeParams::exportLegacyStream(std::ostream& output_stream) const { - LLMemType m1(LLMemType::MTYPE_VOLUME); - output_stream <<"\tshape 0\n"; output_stream <<"\t{\n"; mPathParams.exportLegacyStream(output_stream); @@ -4104,11 +4221,28 @@ BOOL LLVolumeParams::exportLegacyStream(std::ostream& output_stream) const return TRUE; } +LLSD LLVolumeParams::sculptAsLLSD() const +{ + LLSD sd = LLSD(); + sd["id"] = getSculptID(); + sd["type"] = getSculptType(); + + return sd; +} + +bool LLVolumeParams::sculptFromLLSD(LLSD& sd) +{ + setSculptID(sd["id"].asUUID(), (U8)sd["type"].asInteger()); + return true; +} + LLSD LLVolumeParams::asLLSD() const { LLSD sd = LLSD(); sd["path"] = mPathParams; sd["profile"] = mProfileParams; + sd["sculpt"] = sculptAsLLSD(); + return sd; } @@ -4116,6 +4250,8 @@ bool LLVolumeParams::fromLLSD(LLSD& sd) { mPathParams.fromLLSD(sd["path"]); mProfileParams.fromLLSD(sd["profile"]); + sculptFromLLSD(sd["sculpt"]); + return true; } @@ -4158,6 +4294,12 @@ const F32 MIN_CONCAVE_PATH_WEDGE = 0.111111f; // 1/9 unity // for collison purposes BOOL LLVolumeParams::isConvex() const { + if (!getSculptID().isNull()) + { + // can't determine, be safe and say no: + return FALSE; + } + F32 path_length = mPathParams.getEnd() - mPathParams.getBegin(); F32 hollow = mProfileParams.getHollow(); @@ -4274,7 +4416,7 @@ LLFaceID LLVolume::generateFaceMask() } break; default: - llerrs << "Unknown profile!" << llendl; + LL_ERRS() << "Unknown profile!" << LL_ENDL; break; } @@ -4398,24 +4540,865 @@ std::ostream& operator<<(std::ostream &s, const LLVolume *volumep) return s; } +LLVolumeFace::LLVolumeFace() : + mID(0), + mTypeMask(0), + mBeginS(0), + mBeginT(0), + mNumS(0), + mNumT(0), + mNumVertices(0), + mNumAllocatedVertices(0), + mNumIndices(0), + mPositions(NULL), + mNormals(NULL), + mTangents(NULL), + mTexCoords(NULL), + mIndices(NULL), + mWeights(NULL), + mOctree(NULL), + mOptimized(FALSE) +{ + mExtents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*3); + mExtents[0].splat(-0.5f); + mExtents[1].splat(0.5f); + mCenter = mExtents+2; +} + +LLVolumeFace::LLVolumeFace(const LLVolumeFace& src) +: mID(0), + mTypeMask(0), + mBeginS(0), + mBeginT(0), + mNumS(0), + mNumT(0), + mNumVertices(0), + mNumAllocatedVertices(0), + mNumIndices(0), + mPositions(NULL), + mNormals(NULL), + mTangents(NULL), + mTexCoords(NULL), + mIndices(NULL), + mWeights(NULL), + mOctree(NULL) +{ + mExtents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*3); + mCenter = mExtents+2; + *this = src; +} + +LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) +{ + if (&src == this) + { //self assignment, do nothing + return *this; + } + + mID = src.mID; + mTypeMask = src.mTypeMask; + mBeginS = src.mBeginS; + mBeginT = src.mBeginT; + mNumS = src.mNumS; + mNumT = src.mNumT; + + mExtents[0] = src.mExtents[0]; + mExtents[1] = src.mExtents[1]; + *mCenter = *src.mCenter; + + mNumVertices = 0; + mNumIndices = 0; + + freeData(); + + 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); + + if (src.mNormals) + { + LLVector4a::memcpyNonAliased16((F32*) mNormals, (F32*) src.mNormals, vert_size); + } + + if(src.mTexCoords) + { + LLVector4a::memcpyNonAliased16((F32*) mTexCoords, (F32*) src.mTexCoords, tc_size); + } + + if (src.mTangents) + { + allocateTangents(src.mNumVertices); + LLVector4a::memcpyNonAliased16((F32*) mTangents, (F32*) src.mTangents, vert_size); + } + else + { + ll_aligned_free_16(mTangents); + mTangents = NULL; + } + + if (src.mWeights) + { + allocateWeights(src.mNumVertices); + LLVector4a::memcpyNonAliased16((F32*) mWeights, (F32*) src.mWeights, vert_size); + } + else + { + ll_aligned_free_16(mWeights); + mWeights = NULL; + } + } + + if (mNumIndices) + { + S32 idx_size = (mNumIndices*sizeof(U16)+0xF) & ~0xF; + + LLVector4a::memcpyNonAliased16((F32*) mIndices, (F32*) src.mIndices, idx_size); + } + + mOptimized = src.mOptimized; + + //delete + return *this; +} + +LLVolumeFace::~LLVolumeFace() +{ + ll_aligned_free_16(mExtents); + mExtents = NULL; + + freeData(); +} + +void LLVolumeFace::freeData() +{ + ll_aligned_free<64>(mPositions); + mPositions = NULL; + + //normals and texture coordinates are part of the same buffer as mPositions, do not free them separately + mNormals = NULL; + mTexCoords = NULL; + + ll_aligned_free_16(mIndices); + mIndices = NULL; + ll_aligned_free_16(mTangents); + mTangents = NULL; + ll_aligned_free_16(mWeights); + mWeights = 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; + + LL_CHECK_MEMORY + BOOL ret = FALSE ; if (mTypeMask & CAP_MASK) { - return createCap(volume, partial_build); + ret = createCap(volume, partial_build); + LL_CHECK_MEMORY } else if ((mTypeMask & END_MASK) || (mTypeMask & SIDE_MASK)) { - return createSide(volume, partial_build); + ret = createSide(volume, partial_build); + LL_CHECK_MEMORY } else { - llerrs << "Unknown/uninitialized face type!" << llendl; - return FALSE; + LL_ERRS() << "Unknown/uninitialized face type!" << LL_ENDL; + } + + return ret ; +} + +void LLVolumeFace::getVertexData(U16 index, LLVolumeFace::VertexData& cv) +{ + cv.setPosition(mPositions[index]); + if (mNormals) + { + cv.setNormal(mNormals[index]); + } + else + { + cv.getNormal().clear(); + } + + if (mTexCoords) + { + cv.mTexCoord = mTexCoords[index]; + } + else + { + cv.mTexCoord.clear(); + } +} + +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 + std::map<U64, std::vector<VertexMapData> > point_map; + + LLVector4a range; + range.setSub(mExtents[1],mExtents[0]); + + //remove redundant vertices + for (U32 i = 0; i < mNumIndices; ++i) + { + U16 index = mIndices[i]; + + LLVolumeFace::VertexData cv; + getVertexData(index, cv); + + BOOL found = FALSE; + + LLVector4a pos; + pos.setSub(mPositions[index], mExtents[0]); + pos.div(range); + + U64 pos64 = 0; + + pos64 = (U16) (pos[0]*65535); + pos64 = pos64 | (((U64) (pos[1]*65535)) << 16); + pos64 = pos64 | (((U64) (pos[2]*65535)) << 32); + + std::map<U64, std::vector<VertexMapData> >::iterator point_iter = point_map.find(pos64); + + 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[pos64].push_back(d); + } + } + } + + + if (angle_cutoff > 1.f && !mNormals) + { + // Now alloc'd with positions + //ll_aligned_free_16(new_face.mNormals); + new_face.mNormals = NULL; + } + + if (!mTexCoords) + { + // Now alloc'd with positions + //ll_aligned_free_16(new_face.mTexCoords); + new_face.mTexCoords = NULL; + } + + // Only swap data if we've actually optimized the mesh + // + if (new_face.mNumVertices <= mNumVertices) + { + llassert(new_face.mNumIndices == mNumIndices); + swapData(new_face); + } + +} + +class LLVCacheTriangleData; + +class LLVCacheVertexData +{ +public: + S32 mIdx; + S32 mCacheTag; + F64 mScore; + U32 mActiveTriangles; + std::vector<LLVCacheTriangleData*> mTriangles; + + LLVCacheVertexData() + { + mCacheTag = -1; + mScore = 0.0; + mActiveTriangles = 0; + mIdx = -1; + } +}; + +class LLVCacheTriangleData +{ +public: + bool mActive; + F64 mScore; + LLVCacheVertexData* mVertex[3]; + + LLVCacheTriangleData() + { + mActive = true; + mScore = 0.0; + mVertex[0] = mVertex[1] = mVertex[2] = NULL; + } + + void complete() + { + mActive = false; + for (S32 i = 0; i < 3; ++i) + { + if (mVertex[i]) + { + llassert(mVertex[i]->mActiveTriangles > 0); + mVertex[i]->mActiveTriangles--; + } + } + } + + bool operator<(const LLVCacheTriangleData& rhs) const + { //highest score first + return rhs.mScore < mScore; + } +}; + +const F64 FindVertexScore_CacheDecayPower = 1.5; +const F64 FindVertexScore_LastTriScore = 0.75; +const F64 FindVertexScore_ValenceBoostScale = 2.0; +const F64 FindVertexScore_ValenceBoostPower = 0.5; +const U32 MaxSizeVertexCache = 32; +const F64 FindVertexScore_Scaler = 1.0/(MaxSizeVertexCache-3); + +F64 find_vertex_score(LLVCacheVertexData& data) +{ + F64 score = -1.0; + + score = 0.0; + + 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 + score = 1.0-((cache_idx-3)*FindVertexScore_Scaler); + score = pow(score, FindVertexScore_CacheDecayPower); + } + } + + //bonus points for having low valence + F64 valence_boost = pow((F64)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() + { + LLVCacheVertexData** data_iter = mCache+MaxSizeVertexCache; + LLVCacheVertexData** end_data = mCache+MaxSizeVertexCache+3; + + while(data_iter != end_data) + { + LLVCacheVertexData* data = *data_iter++; + //trailing 3 vertices aren't actually in the cache for scoring purposes + if (data) + { + data->mCacheTag = -1; + } + } + + data_iter = mCache; + end_data = mCache+MaxSizeVertexCache; + + while (data_iter != end_data) + { //update scores of vertices in cache + LLVCacheVertexData* data = *data_iter++; + if (data) + { + data->mScore = find_vertex_score(*data); + } + } + + mBestTriangle = NULL; + //update triangle scores + data_iter = mCache; + end_data = mCache+MaxSizeVertexCache+3; + + while (data_iter != end_data) + { + LLVCacheVertexData* data = *data_iter++; + if (data) + { + for (std::vector<LLVCacheTriangleData*>::iterator iter = data->mTriangles.begin(), end_iter = data->mTriangles.end(); iter != end_iter; ++iter) + { + LLVCacheTriangleData* tri = *iter; + 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 + data_iter = mCache+MaxSizeVertexCache; + end_data = mCache+MaxSizeVertexCache+3; + while (data_iter != end_data) + { + LLVCacheVertexData* data = *data_iter; + if (data) + { + llassert(data->mCacheTag == -1); + *data_iter = NULL; + } + ++data_iter; + } + } +}; + + +void LLVolumeFace::cacheOptimize() +{ //optimize for vertex cache according to Forsyth method: + // http://home.comcast.net/~tom_forsyth/papers/fast_vert_cache_opt.html + + llassert(!mOptimized); + mOptimized = TRUE; + + LLVCacheLRU cache; + + if (mNumVertices < 3) + { //nothing to do + return; + } + + //mapping of vertices to triangles and indices + std::vector<LLVCacheVertexData> vertex_data; + + //mapping of triangles do vertices + std::vector<LLVCacheTriangleData> triangle_data; + + triangle_data.resize(mNumIndices/3); + vertex_data.resize(mNumVertices); + + for (U32 i = 0; i < 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 < mNumVertices; i++) + { //initialize score values (no cache -- might try a fifo cache here) + LLVCacheVertexData& data = vertex_data[i]; + + data.mScore = find_vertex_score(data); + data.mActiveTriangles = data.mTriangles.size(); + + for (U32 j = 0; j < data.mActiveTriangles; ++j) + { + data.mTriangles[j]->mScore += data.mScore; + } + } + + //sort triangle data by score + std::sort(triangle_data.begin(), triangle_data.end()); + + std::vector<U16> 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 < 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 < 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; + S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF; + LLVector4a* pos = (LLVector4a*) ll_aligned_malloc<64>(sizeof(LLVector4a)*2*num_verts+size); + LLVector4a* norm = pos + num_verts; + LLVector2* tc = (LLVector2*) (norm + num_verts); + + LLVector4a* wght = NULL; + if (mWeights) + { + wght = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); + } + + LLVector4a* binorm = NULL; + if (mTangents) + { + binorm = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); + } + + //allocate mapping of old indices to new indices + std::vector<S32> new_idx; + new_idx.resize(mNumVertices, -1); + + S32 cur_idx = 0; + for (U32 i = 0; i < 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 (mWeights) + { + wght[cur_idx] = mWeights[idx]; + } + if (mTangents) + { + binorm[cur_idx] = mTangents[idx]; + } + + cur_idx++; + } + } + + for (U32 i = 0; i < mNumIndices; ++i) + { + mIndices[i] = new_idx[mIndices[i]]; + } + + ll_aligned_free<64>(mPositions); + // DO NOT free mNormals and mTexCoords as they are part of mPositions buffer + ll_aligned_free_16(mWeights); + ll_aligned_free_16(mTangents); + + mPositions = pos; + mNormals = norm; + mTexCoords = tc; + mWeights = wght; + mTangents = binorm; + + //std::string result = llformat("ACMR pre/post: %.3f/%.3f -- %d triangles %d breaks", pre_acmr, post_acmr, mNumIndices/3, breaks); + //LL_INFOS() << result << LL_ENDL; + +} + +void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVector4a& size) +{ + 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() * 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.mTangents, mTangents); + 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, @@ -4423,105 +5406,130 @@ 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) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - - const std::vector<LLVolume::Point>& mesh = volume->getMesh(); - const std::vector<LLVector3>& profile = volume->getProfile().mProfile; + LL_CHECK_MEMORY + + const LLAlignedArray<LLVector4a,64>& mesh = volume->getMesh(); + const LLAlignedArray<LLVector4a,64>& profile = volume->getProfile().mProfile; S32 max_s = volume->getProfile().getTotal(); S32 max_t = volume->getPath().mPath.size(); // S32 i; - S32 num_vertices = 0, num_indices = 0; S32 grid_size = (profile.size()-1)/4; - S32 quad_count = (grid_size * grid_size); - num_vertices = (grid_size+1)*(grid_size+1); - num_indices = quad_count * 4; - - LLVector3& min = mExtents[0]; - LLVector3& max = mExtents[1]; + LLVector4a& min = mExtents[0]; + LLVector4a& max = mExtents[1]; S32 offset = 0; if (mTypeMask & TOP_MASK) + { offset = (max_t-1) * max_s; + } else + { offset = mBeginS; + } - VertexData corners[4]; - VertexData baseVert; - for(int t = 0; t < 4; t++){ - corners[t].mPosition = mesh[offset + (grid_size*t)].mPos; - 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.normVec(); - if(!(mTypeMask & TOP_MASK)){ - baseVert.mNormal *= -1.0f; - }else{ - //Swap the UVs on the U(X) axis for top face - LLVector2 swap; - swap = corners[0].mTexCoord; - corners[0].mTexCoord=corners[3].mTexCoord; - corners[3].mTexCoord=swap; - swap = corners[1].mTexCoord; - 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(); - } + VertexData corners[4]; + VertexData baseVert; + for(S32 t = 0; t < 4; t++) + { + corners[t].getPosition().load4a(mesh[offset + (grid_size*t)].getF32ptr()); + corners[t].mTexCoord.mV[0] = profile[grid_size*t][0]+0.5f; + corners[t].mTexCoord.mV[1] = 0.5f - profile[grid_size*t][1]; + } - S32 vtop = mVertices.size(); - for(int gx = 0;gx<grid_size+1;gx++){ - for(int gy = 0;gy<grid_size+1;gy++){ - VertexData newVert; - LerpPlanarVertex( - corners[0], - corners[1], - corners[3], - newVert, - (F32)gx/(F32)grid_size, - (F32)gy/(F32)grid_size); - mVertices.push_back(newVert); + { + 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 (gx == 0 && gy == 0) - { - min = max = newVert.mPosition; - } - else + if(!(mTypeMask & TOP_MASK)) + { + baseVert.getNormal().mul(-1.0f); + } + else + { + //Swap the UVs on the U(X) axis for top face + LLVector2 swap; + swap = corners[0].mTexCoord; + corners[0].mTexCoord=corners[3].mTexCoord; + corners[3].mTexCoord=swap; + swap = corners[1].mTexCoord; + corners[1].mTexCoord=corners[2].mTexCoord; + corners[2].mTexCoord=swap; + } + + S32 size = (grid_size+1)*(grid_size+1); + resizeVertices(size); + + LLVector4a* pos = (LLVector4a*) mPositions; + LLVector4a* norm = (LLVector4a*) mNormals; + LLVector2* tc = (LLVector2*) mTexCoords; + + for(int gx = 0;gx<grid_size+1;gx++) + { + for(int gy = 0;gy<grid_size+1;gy++) { - update_min_max(min,max,newVert.mPosition); + VertexData newVert; + LerpPlanarVertex( + corners[0], + corners[1], + corners[3], + newVert, + (F32)gx/(F32)grid_size, + (F32)gy/(F32)grid_size); + + *pos++ = newVert.getPosition(); + *norm++ = baseVert.getNormal(); + *tc++ = newVert.mTexCoord; + + if (gx == 0 && gy == 0) + { + min = newVert.getPosition(); + max = min; + } + else + { + min.setMin(min, newVert.getPosition()); + max.setMax(max, newVert.getPosition()); + } } } - } - mCenter = (min + max) * 0.5f; + mCenter->setAdd(min, max); + mCenter->mul(0.5f); + } if (!partial_build) { - mTriStrip.clear(); + resizeIndices(grid_size*grid_size*6); + + U16* out = mIndices; + S32 idxs[] = {0,1,(grid_size+1)+1,(grid_size+1)+1,(grid_size+1),0}; for(S32 gx = 0;gx<grid_size;gx++) { @@ -4532,65 +5540,27 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) { for(S32 i=5;i>=0;i--) { - mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); - } - - if (gy == 0) - { - mTriStrip.push_back((gx+1)*(grid_size+1)); - mTriStrip.push_back((gx+1)*(grid_size+1)); - mTriStrip.push_back(gx*(grid_size+1)); - } - - mTriStrip.push_back(gy+1+(gx+1)*(grid_size+1)); - mTriStrip.push_back(gy+1+gx*(grid_size+1)); - - - if (gy == grid_size-1) - { - mTriStrip.push_back(gy+1+gx*(grid_size+1)); - } + *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]); - } - - if (gy == 0) - { - mTriStrip.push_back(gx*(grid_size+1)); - mTriStrip.push_back(gx*(grid_size+1)); - mTriStrip.push_back((gx+1)*(grid_size+1)); - } - - mTriStrip.push_back(gy+1+gx*(grid_size+1)); - mTriStrip.push_back(gy+1+(gx+1)*(grid_size+1)); - - if (gy == grid_size-1) - { - mTriStrip.push_back(gy+1+(gx+1)*(grid_size+1)); + *out++ = ((gy*(grid_size+1))+gx+idxs[i]); } } - } - - } - - if (mTriStrip.size()%2 == 1) - { - mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); + } } } + LL_CHECK_MEMORY return TRUE; } BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK) && ((volume->getParams().getPathParams().getBegin()==0.0f)&& @@ -4603,24 +5573,37 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) S32 num_vertices = 0, num_indices = 0; - const std::vector<LLVolume::Point>& mesh = volume->getMesh(); - const std::vector<LLVector3>& profile = volume->getProfile().mProfile; + const LLAlignedArray<LLVector4a,64>& mesh = volume->getMesh(); + const LLAlignedArray<LLVector4a,64>& profile = volume->getProfile().mProfile; // All types of caps have the same number of vertices and indices num_vertices = profile.size(); num_indices = (profile.size() - 2)*3; - mVertices.resize(num_vertices); - - if (!partial_build) + if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK)) + { + resizeVertices(num_vertices+1); + + if (!partial_build) + { + resizeIndices(num_indices+3); + } + } + else { - mIndices.resize(num_indices); + resizeVertices(num_vertices); + if (!partial_build) + { + resizeIndices(num_indices); + } } + LL_CHECK_MEMORY; + S32 max_s = volume->getProfile().getTotal(); S32 max_t = volume->getPath().mPath.size(); - mCenter.clearVec(); + mCenter->clear(); S32 offset = 0; if (mTypeMask & TOP_MASK) @@ -4638,82 +5621,99 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) LLVector2 cuv; LLVector2 min_uv, max_uv; - LLVector3& min = mExtents[0]; - LLVector3& max = mExtents[1]; + LLVector4a& min = mExtents[0]; + LLVector4a& max = mExtents[1]; + LLVector2* tc = (LLVector2*) mTexCoords; + LLVector4a* pos = (LLVector4a*) mPositions; + LLVector4a* norm = (LLVector4a*) mNormals; + // Copy the vertices into the array - for (S32 i = 0; i < num_vertices; i++) + + const LLVector4a* src = mesh.mArray+offset; + const LLVector4a* end = src+num_vertices; + + min = *src; + max = min; + + + const LLVector4a* p = profile.mArray; + + if (mTypeMask & TOP_MASK) { - if (mTypeMask & TOP_MASK) + min_uv.set((*p)[0]+0.5f, + (*p)[1]+0.5f); + + max_uv = min_uv; + + while(src < end) { - mVertices[i].mTexCoord.mV[0] = profile[i].mV[0]+0.5f; - mVertices[i].mTexCoord.mV[1] = profile[i].mV[1]+0.5f; + tc->mV[0] = (*p)[0]+0.5f; + tc->mV[1] = (*p)[1]+0.5f; + + llassert(src->isFinite3()); + update_min_max(min,max,*src); + update_min_max(min_uv, max_uv, *tc); + + *pos = *src; + + llassert(pos->isFinite3()); + + ++p; + ++tc; + ++src; + ++pos; + } } else { + + min_uv.set((*p)[0]+0.5f, + 0.5f - (*p)[1]); + max_uv = min_uv; + + while(src < end) + { // 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->mV[0] = (*p)[0]+0.5f; + tc->mV[1] = 0.5f - (*p)[1]; + + llassert(src->isFinite3()); + update_min_max(min,max,*src); + update_min_max(min_uv, max_uv, *tc); - mVertices[i].mPosition = mesh[i + offset].mPos; + *pos = *src; - if (i == 0) - { - min = max = mVertices[i].mPosition; - min_uv = max_uv = mVertices[i].mTexCoord; - } - else - { - update_min_max(min,max, mVertices[i].mPosition); - update_min_max(min_uv, max_uv, mVertices[i].mTexCoord); + llassert(pos->isFinite3()); + + ++p; + ++tc; + ++src; + ++pos; } } - mCenter = (min+max)*0.5f; - cuv = (min_uv + max_uv)*0.5f; - - LLVector3 binormal = calc_binormal_from_triangle( - mCenter, cuv, - mVertices[0].mPosition, mVertices[0].mTexCoord, - mVertices[1].mPosition, mVertices[1].mTexCoord); - binormal.normVec(); + LL_CHECK_MEMORY - LLVector3 d0; - LLVector3 d1; - LLVector3 normal; + mCenter->setAdd(min, max); + mCenter->mul(0.5f); - d0 = mCenter-mVertices[0].mPosition; - d1 = mCenter-mVertices[1].mPosition; + cuv = (min_uv + max_uv)*0.5f; - normal = (mTypeMask & TOP_MASK) ? (d0%d1) : (d1%d0); - normal.normVec(); VertexData vd; - vd.mPosition = mCenter; - vd.mNormal = normal; - vd.mBinormal = binormal; + vd.setPosition(*mCenter); vd.mTexCoord = cuv; if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK)) { - mVertices.push_back(vd); + *pos++ = *mCenter; + *tc++ = 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 = normal; - } - - mHasBinormals = TRUE; - + LL_CHECK_MEMORY + if (partial_build) { return TRUE; @@ -4732,33 +5732,38 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) { // Use the profile points instead of the mesh, since you want // the un-transformed profile distances. - LLVector3 p1 = profile[pt1]; - LLVector3 p2 = profile[pt2]; - LLVector3 pa = profile[pt1+1]; - LLVector3 pb = profile[pt2-1]; + const LLVector4a& p1 = profile[pt1]; + const LLVector4a& p2 = profile[pt2]; + const LLVector4a& pa = profile[pt1+1]; + const LLVector4a& pb = profile[pt2-1]; - p1.mV[VZ] = 0.f; - p2.mV[VZ] = 0.f; - pa.mV[VZ] = 0.f; - pb.mV[VZ] = 0.f; + const F32* p1V = p1.getF32ptr(); + const F32* p2V = p2.getF32ptr(); + const F32* paV = pa.getF32ptr(); + const F32* pbV = pb.getF32ptr(); + + //p1.mV[VZ] = 0.f; + //p2.mV[VZ] = 0.f; + //pa.mV[VZ] = 0.f; + //pb.mV[VZ] = 0.f; // Use area of triangle to determine backfacing F32 area_1a2, area_1ba, area_21b, area_2ab; - area_1a2 = (p1.mV[0]*pa.mV[1] - pa.mV[0]*p1.mV[1]) + - (pa.mV[0]*p2.mV[1] - p2.mV[0]*pa.mV[1]) + - (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]); + area_1a2 = (p1V[0]*paV[1] - paV[0]*p1V[1]) + + (paV[0]*p2V[1] - p2V[0]*paV[1]) + + (p2V[0]*p1V[1] - p1V[0]*p2V[1]); - area_1ba = (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*pa.mV[1] - pa.mV[0]*pb.mV[1]) + - (pa.mV[0]*p1.mV[1] - p1.mV[0]*pa.mV[1]); + area_1ba = (p1V[0]*pbV[1] - pbV[0]*p1V[1]) + + (pbV[0]*paV[1] - paV[0]*pbV[1]) + + (paV[0]*p1V[1] - p1V[0]*paV[1]); - area_21b = (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]) + - (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); + area_21b = (p2V[0]*p1V[1] - p1V[0]*p2V[1]) + + (p1V[0]*pbV[1] - pbV[0]*p1V[1]) + + (pbV[0]*p2V[1] - p2V[0]*pbV[1]); - area_2ab = (p2.mV[0]*pa.mV[1] - pa.mV[0]*p2.mV[1]) + - (pa.mV[0]*pb.mV[1] - pb.mV[0]*pa.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); + area_2ab = (p2V[0]*paV[1] - paV[0]*p2V[1]) + + (paV[0]*pbV[1] - pbV[0]*paV[1]) + + (pbV[0]*p2V[1] - p2V[0]*pbV[1]); BOOL use_tri1a2 = TRUE; BOOL tri_1a2 = TRUE; @@ -4793,10 +5798,13 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) } else { - LLVector3 d1 = p1 - pa; - LLVector3 d2 = p2 - pb; + LLVector4a d1; + d1.setSub(p1, pa); + + LLVector4a d2; + d2.setSub(p2, pb); - if (d1.magVecSquared() < d2.magVecSquared()) + if (d1.dot3(d1) < d2.dot3(d2)) { use_tri1a2 = TRUE; } @@ -4821,8 +5829,6 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) pt2--; } } - - makeTriStrip(); } else { @@ -4837,33 +5843,33 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) { // Use the profile points instead of the mesh, since you want // the un-transformed profile distances. - LLVector3 p1 = profile[pt1]; - LLVector3 p2 = profile[pt2]; - LLVector3 pa = profile[pt1+1]; - LLVector3 pb = profile[pt2-1]; + const LLVector4a& p1 = profile[pt1]; + const LLVector4a& p2 = profile[pt2]; + const LLVector4a& pa = profile[pt1+1]; + const LLVector4a& pb = profile[pt2-1]; - p1.mV[VZ] = 0.f; - p2.mV[VZ] = 0.f; - pa.mV[VZ] = 0.f; - pb.mV[VZ] = 0.f; + const F32* p1V = p1.getF32ptr(); + const F32* p2V = p2.getF32ptr(); + const F32* paV = pa.getF32ptr(); + const F32* pbV = pb.getF32ptr(); // Use area of triangle to determine backfacing F32 area_1a2, area_1ba, area_21b, area_2ab; - area_1a2 = (p1.mV[0]*pa.mV[1] - pa.mV[0]*p1.mV[1]) + - (pa.mV[0]*p2.mV[1] - p2.mV[0]*pa.mV[1]) + - (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]); + area_1a2 = (p1V[0]*paV[1] - paV[0]*p1V[1]) + + (paV[0]*p2V[1] - p2V[0]*paV[1]) + + (p2V[0]*p1V[1] - p1V[0]*p2V[1]); - area_1ba = (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*pa.mV[1] - pa.mV[0]*pb.mV[1]) + - (pa.mV[0]*p1.mV[1] - p1.mV[0]*pa.mV[1]); + area_1ba = (p1V[0]*pbV[1] - pbV[0]*p1V[1]) + + (pbV[0]*paV[1] - paV[0]*pbV[1]) + + (paV[0]*p1V[1] - p1V[0]*paV[1]); - area_21b = (p2.mV[0]*p1.mV[1] - p1.mV[0]*p2.mV[1]) + - (p1.mV[0]*pb.mV[1] - pb.mV[0]*p1.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); + area_21b = (p2V[0]*p1V[1] - p1V[0]*p2V[1]) + + (p1V[0]*pbV[1] - pbV[0]*p1V[1]) + + (pbV[0]*p2V[1] - p2V[0]*pbV[1]); - area_2ab = (p2.mV[0]*pa.mV[1] - pa.mV[0]*p2.mV[1]) + - (pa.mV[0]*pb.mV[1] - pb.mV[0]*pa.mV[1]) + - (pb.mV[0]*p2.mV[1] - p2.mV[0]*pb.mV[1]); + area_2ab = (p2V[0]*paV[1] - paV[0]*p2V[1]) + + (paV[0]*pbV[1] - pbV[0]*paV[1]) + + (pbV[0]*p2V[1] - p2V[0]*pbV[1]); BOOL use_tri1a2 = TRUE; BOOL tri_1a2 = TRUE; @@ -4898,10 +5904,12 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) } else { - LLVector3 d1 = p1 - pa; - LLVector3 d2 = p2 - pb; + LLVector4a d1; + d1.setSub(p1,pa); + LLVector4a d2; + d2.setSub(p2,pb); - if (d1.magVecSquared() < d2.magVecSquared()) + if (d1.dot3(d1) < d2.dot3(d2)) { use_tri1a2 = TRUE; } @@ -4927,8 +5935,6 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) pt2--; } } - - makeTriStrip(); } } else @@ -4950,134 +5956,320 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) mIndices[3*i+v2] = i + 1; } - //make tri strip - if (mTypeMask & OPEN_MASK) - { - makeTriStrip(); - } - else - { - S32 j = num_vertices-2; - if (mTypeMask & TOP_MASK) - { - mTriStrip.push_back(0); - for (S32 i = 0; i <= j; ++i) - { - mTriStrip.push_back(i); - if (i != j) - { - mTriStrip.push_back(j); - } - --j; - } - } - else - { - mTriStrip.push_back(j); - for (S32 i = 0; i <= j; ++i) - { - if (i != j) - { - mTriStrip.push_back(j); - } - mTriStrip.push_back(i); - --j; - } - } - - mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); - if (mTriStrip.size()%2 == 1) - { - mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); - } - } } + + LLVector4a d0,d1; + LL_CHECK_MEMORY + + d0.setSub(mPositions[mIndices[1]], mPositions[mIndices[0]]); + d1.setSub(mPositions[mIndices[2]], mPositions[mIndices[0]]); + + LLVector4a normal; + normal.setCross3(d0,d1); + + if (normal.dot3(normal).getF32() > F_APPROXIMATELY_ZERO) + { + normal.normalize3fast(); + } + else + { //degenerate, make up a value + if(normal.getF32ptr()[2] >= 0) + normal.set(0.f,0.f,1.f); + else + normal.set(0.f,0.f,-1.f); + } + + llassert(llfinite(normal.getF32ptr()[0])); + llassert(llfinite(normal.getF32ptr()[1])); + llassert(llfinite(normal.getF32ptr()[2])); + + llassert(!llisnan(normal.getF32ptr()[0])); + llassert(!llisnan(normal.getF32ptr()[1])); + llassert(!llisnan(normal.getF32ptr()[2])); + + for (S32 i = 0; i < num_vertices; i++) + { + norm[i].load4a(normal.getF32ptr()); + } + return TRUE; } -void LLVolumeFace::makeTriStrip() +void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVector4a *normal, + const LLVector2 *texcoord, U32 triangleCount, const U16* index_array, LLVector4a *tangent); + +void LLVolumeFace::createTangents() { - for (U32 i = 0; i < mIndices.size(); i+=3) + if (!mTangents) { - U16 i0 = mIndices[i]; - U16 i1 = mIndices[i+1]; - U16 i2 = mIndices[i+2]; + allocateTangents(mNumVertices); + + //generate tangents + //LLVector4a* pos = mPositions; + //LLVector2* tc = (LLVector2*) mTexCoords; + LLVector4a* binorm = (LLVector4a*) mTangents; - if ((i/3)%2 == 1) + LLVector4a* end = mTangents+mNumVertices; + while (binorm < end) { - mTriStrip.push_back(i0); - mTriStrip.push_back(i0); - mTriStrip.push_back(i1); - mTriStrip.push_back(i2); - mTriStrip.push_back(i2); + (*binorm++).clear(); } - else + + binorm = mTangents; + + CalculateTangentArray(mNumVertices, mPositions, mNormals, mTexCoords, mNumIndices/3, mIndices, mTangents); + + //normalize tangents + for (U32 i = 0; i < mNumVertices; i++) { - mTriStrip.push_back(i2); - mTriStrip.push_back(i2); - mTriStrip.push_back(i1); - mTriStrip.push_back(i0); - mTriStrip.push_back(i0); + //binorm[i].normalize3fast(); + //bump map/planar projection code requires normals to be normalized + mNormals[i].normalize3fast(); } } +} - if (mTriStrip.size()%2 == 1) +void LLVolumeFace::resizeVertices(S32 num_verts) +{ + ll_aligned_free<64>(mPositions); + //DO NOT free mNormals and mTexCoords as they are part of mPositions buffer + ll_aligned_free_16(mTangents); + + mTangents = NULL; + + if (num_verts) + { + //pad texture coordinate block end to allow for QWORD reads + S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF; + + mPositions = (LLVector4a*) ll_aligned_malloc<64>(sizeof(LLVector4a)*2*num_verts+size); + mNormals = mPositions+num_verts; + mTexCoords = (LLVector2*) (mNormals+num_verts); + + ll_assert_aligned(mPositions, 64); + } + else { - mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); + mPositions = NULL; + mNormals = NULL; + mTexCoords = NULL; } + + mNumVertices = num_verts; + mNumAllocatedVertices = num_verts; +} + +void LLVolumeFace::pushVertex(const LLVolumeFace::VertexData& cv) +{ + pushVertex(cv.getPosition(), cv.getNormal(), cv.mTexCoord); } -void LLVolumeFace::createBinormals() +void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, const LLVector2& tc) { - LLMemType m1(LLMemType::MTYPE_VOLUME); + S32 new_verts = mNumVertices+1; + + if (new_verts > mNumAllocatedVertices) + { + //double buffer size on expansion + new_verts *= 2; + + S32 new_tc_size = ((new_verts*8)+0xF) & ~0xF; + S32 old_tc_size = ((mNumVertices*8)+0xF) & ~0xF; + + S32 old_vsize = mNumVertices*16; + + S32 new_size = new_verts*16*2+new_tc_size; + + LLVector4a* old_buf = mPositions; + + mPositions = (LLVector4a*) ll_aligned_malloc<64>(new_size); + mNormals = mPositions+new_verts; + mTexCoords = (LLVector2*) (mNormals+new_verts); + + //positions + LLVector4a::memcpyNonAliased16((F32*) mPositions, (F32*) old_buf, old_vsize); - if (!mHasBinormals) - { - //generate binormals - for (U32 i = 0; i < mIndices.size()/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]]; - - //calculate binormal - LLVector3 binorm = calc_binormal_from_triangle(v0.mPosition, v0.mTexCoord, - v1.mPosition, v1.mTexCoord, - v2.mPosition, v2.mTexCoord); - - 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; - } + //normals + LLVector4a::memcpyNonAliased16((F32*) mNormals, (F32*) (old_buf+mNumVertices), old_vsize); - //even out quad contributions - if (i % 2 == 0) - { - mVertices[mIndices[i*3+2]].mBinormal += binorm; - } - else - { - mVertices[mIndices[i*3+1]].mBinormal += binorm; - } - } + //tex coords + LLVector4a::memcpyNonAliased16((F32*) mTexCoords, (F32*) (old_buf+mNumVertices*2), old_tc_size); + + //just clear tangents + ll_aligned_free_16(mTangents); + mTangents = NULL; + ll_aligned_free<64>(old_buf); + + mNumAllocatedVertices = new_verts; + + } + + mPositions[mNumVertices] = pos; + mNormals[mNumVertices] = norm; + mTexCoords[mNumVertices] = tc; + + mNumVertices++; +} + +void LLVolumeFace::allocateTangents(S32 num_verts) +{ + ll_aligned_free_16(mTangents); + mTangents = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts); +} + +void LLVolumeFace::allocateWeights(S32 num_verts) +{ + ll_aligned_free_16(mWeights); + mWeights = (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*) ll_aligned_realloc_16(mIndices, new_size, old_size); + ll_assert_aligned(mIndices,16); + } + + mIndices[mNumIndices++] = idx; +} + +void LLVolumeFace::fillFromLegacyData(std::vector<LLVolumeFace::VertexData>& v, std::vector<U16>& 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) + { + LL_ERRS() << "Cannot append face -- 16-bit overflow will occur." << LL_ENDL; + } + + if (face.mNumVertices == 0) + { + LL_ERRS() << "Cannot append empty face." << LL_ENDL; + } + + U32 old_vsize = mNumVertices*16; + U32 new_vsize = new_count * 16; + U32 old_tcsize = (mNumVertices*sizeof(LLVector2)+0xF) & ~0xF; + U32 new_tcsize = (new_count*sizeof(LLVector2)+0xF) & ~0xF; + U32 new_size = new_vsize * 2 + new_tcsize; + + //allocate new buffer space + LLVector4a* old_buf = mPositions; + mPositions = (LLVector4a*) ll_aligned_malloc<64>(new_size); + mNormals = mPositions + new_count; + mTexCoords = (LLVector2*) (mNormals+new_count); + + mNumAllocatedVertices = new_count; + + LLVector4a::memcpyNonAliased16((F32*) mPositions, (F32*) old_buf, old_vsize); + LLVector4a::memcpyNonAliased16((F32*) mNormals, (F32*) (old_buf+mNumVertices), old_vsize); + LLVector4a::memcpyNonAliased16((F32*) mTexCoords, (F32*) (old_buf+mNumVertices*2), old_tcsize); + + 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 < 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]; - //normalize binormals - for (U32 i = 0; i < mVertices.size(); i++) + if (offset == 0 && i == 0) + { //initialize bounding box + mExtents[0] = mExtents[1] = dst_pos[i]; + } + else { - mVertices[i].mBinormal.normVec(); - mVertices[i].mNormal.normVec(); + //stretch bounding box + update_min_max(mExtents[0], mExtents[1], dst_pos[i]); } + } + - mHasBinormals = TRUE; + new_count = mNumIndices + face.mNumIndices; + + //allocate new index buffer + mIndices = (U16*) ll_aligned_realloc_16(mIndices, (new_count*sizeof(U16)+0xF) & ~0xF, (mNumIndices*sizeof(U16)+0xF) & ~0xF); + + //get destination address into new index buffer + U16* dst_idx = mIndices+mNumIndices; + mNumIndices = new_count; + + for (U32 i = 0; i < face.mNumIndices; ++i) + { //copy indices, offsetting by old vertex count + dst_idx[i] = face.mIndices[i]+offset; } } BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { - LLMemType m1(LLMemType::MTYPE_VOLUME); - + LL_CHECK_MEMORY BOOL flat = mTypeMask & FLAT_MASK; U8 sculpt_type = volume->getParams().getSculptType(); @@ -5088,9 +6280,9 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) S32 num_vertices, num_indices; - const std::vector<LLVolume::Point>& mesh = volume->getMesh(); - const std::vector<LLVector3>& profile = volume->getProfile().mProfile; - const std::vector<LLPath::PathPt>& path_data = volume->getPath().mPath; + const LLAlignedArray<LLVector4a,64>& mesh = volume->getMesh(); + const LLAlignedArray<LLVector4a,64>& profile = volume->getProfile().mProfile; + const LLAlignedArray<LLPath::PathPt,64>& path_data = volume->getPath().mPath; S32 max_s = volume->getProfile().getTotal(); @@ -5100,30 +6292,32 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) num_vertices = mNumS*mNumT; num_indices = (mNumS-1)*(mNumT-1)*6; - mVertices.resize(num_vertices); + partial_build = (num_vertices > mNumVertices || num_indices > mNumIndices) ? FALSE : partial_build; if (!partial_build) { - mIndices.resize(num_indices); - mEdge.resize(num_indices); - } - else - { - mHasBinormals = FALSE; - } + resizeVertices(num_vertices); + resizeIndices(num_indices); + if (!volume->isMeshAssetLoaded()) + { + mEdge.resize(num_indices); + } + } - LLVector3& face_min = mExtents[0]; - LLVector3& face_max = mExtents[1]; - - mCenter.clearVec(); + LL_CHECK_MEMORY - S32 begin_stex = llfloor( profile[mBeginS].mV[2] ); + LLVector4a* pos = (LLVector4a*) mPositions; + LLVector2* tc = (LLVector2*) mTexCoords; + F32 begin_stex = floorf(profile[mBeginS][2]); S32 num_s = ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2) ? mNumS/2 : mNumS; S32 cur_vertex = 0; + S32 end_t = mBeginT+mNumT; + bool test = (mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2; + // Copy the vertices into the array - for (t = mBeginT; t < mBeginT + mNumT; t++) + for (t = mBeginT; t < end_t; t++) { tt = path_data[t].mTexT; for (s = 0; s < num_s; s++) @@ -5144,11 +6338,11 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) // Get s value for tex-coord. if (!flat) { - ss = profile[mBeginS + s].mV[2]; + ss = profile[mBeginS + s][2]; } else { - ss = profile[mBeginS + s].mV[2] - begin_stex; + ss = profile[mBeginS + s][2] - begin_stex; } } @@ -5168,30 +6362,15 @@ 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); + mesh[i].store4a((F32*)(pos+cur_vertex)); + tc[cur_vertex].set(ss,tt); - mVertices[cur_vertex].mNormal = LLVector3(0,0,0); - mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); - - if (cur_vertex == 0) - { - face_min = face_max = mesh[i].mPos; - } - else - { - update_min_max(face_min, face_max, mesh[i].mPos); - } - cur_vertex++; - if ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2 && s > 0) + if (test && s > 0) { - 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); + mesh[i].store4a((F32*)(pos+cur_vertex)); + tc[cur_vertex].set(ss,tt); cur_vertex++; } } @@ -5208,20 +6387,65 @@ 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); - - update_min_max(face_min,face_max,mesh[i].mPos); + ss = profile[mBeginS + s][2] - begin_stex; + mesh[i].store4a((F32*)(pos+cur_vertex)); + tc[cur_vertex].set(ss,tt); + cur_vertex++; } - } + } + LL_CHECK_MEMORY + + mCenter->clear(); + + LLVector4a* cur_pos = pos; + LLVector4a* end_pos = pos + mNumVertices; + + //get bounding box for this side + LLVector4a face_min; + LLVector4a face_max; - mCenter = (face_min + face_max) * 0.5f; + face_min = face_max = *cur_pos++; + + while (cur_pos < end_pos) + { + update_min_max(face_min, face_max, *cur_pos++); + } + + mExtents[0] = face_min; + mExtents[1] = face_max; + + U32 tc_count = mNumVertices; + if (tc_count%2 == 1) + { //odd number of texture coordinates, duplicate last entry to padded end of array + tc_count++; + mTexCoords[mNumVertices] = mTexCoords[mNumVertices-1]; + } + + LLVector4a* cur_tc = (LLVector4a*) mTexCoords; + LLVector4a* end_tc = (LLVector4a*) (mTexCoords+tc_count); + + LLVector4a tc_min; + LLVector4a tc_max; + + tc_min = tc_max = *cur_tc++; + + while (cur_tc < end_tc) + { + update_min_max(tc_min, tc_max, *cur_tc++); + } + + F32* minp = tc_min.getF32ptr(); + F32* maxp = tc_max.getF32ptr(); + + mTexCoordExtents[0].mV[0] = llmin(minp[0], minp[2]); + mTexCoordExtents[0].mV[1] = llmin(minp[1], minp[3]); + mTexCoordExtents[1].mV[0] = llmax(maxp[0], maxp[2]); + mTexCoordExtents[1].mV[1] = llmax(maxp[1], maxp[3]); + + mCenter->setAdd(face_min, face_max); + mCenter->mul(0.5f); S32 cur_index = 0; S32 cur_edge = 0; @@ -5229,14 +6453,9 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) if (!partial_build) { - mTriStrip.clear(); - // Now we generate the indices. for (t = 0; t < (mNumT-1); t++) { - //prepend terminating index to strip - mTriStrip.push_back(mNumS*t); - for (s = 0; s < (mNumS-1); s++) { mIndices[cur_index++] = s + mNumS*t; //bottom left @@ -5246,14 +6465,6 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) mIndices[cur_index++] = s+1 + mNumS*t; //bottom right mIndices[cur_index++] = s+1 + mNumS*(t+1); //top right - if (s == 0) - { - mTriStrip.push_back(s+mNumS*t); - mTriStrip.push_back(s+mNumS*(t+1)); - } - mTriStrip.push_back(s+1+mNumS*t); - mTriStrip.push_back(s+1+mNumS*(t+1)); - mEdge[cur_edge++] = (mNumS-1)*2*t+s*2+1; //bottom left/top right neighbor face if (t < mNumT-2) { //top right/top left neighbor face mEdge[cur_edge++] = (mNumS-1)*2*(t+1)+s*2+1; @@ -5294,59 +6505,141 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) } mEdge[cur_edge++] = (mNumS-1)*2*t+s*2; //top right/bottom left neighbor face } - //append terminating vertex to strip - mTriStrip.push_back(mNumS-1+mNumS*(t+1)); } + } - if (mTriStrip.size()%2 == 1) - { - mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); - } + LL_CHECK_MEMORY + + //clear normals + F32* dst = (F32*) mNormals; + F32* end = (F32*) (mNormals+mNumVertices); + LLVector4a zero = LLVector4a::getZero(); + + while (dst < end) + { + zero.store4a(dst); + dst += 4; } + LL_CHECK_MEMORY + //generate normals - for (U32 i = 0; i < mIndices.size()/3; i++) //for each triangle - { - const S32 i0 = mIndices[i*3+0]; - const S32 i1 = mIndices[i*3+1]; - const S32 i2 = mIndices[i*3+2]; - const VertexData& v0 = mVertices[i0]; - const VertexData& v1 = mVertices[i1]; - const VertexData& v2 = mVertices[i2]; - + U32 count = mNumIndices/3; + + LLVector4a* norm = mNormals; + + static LLAlignedArray<LLVector4a, 64> triangle_normals; + triangle_normals.resize(count); + LLVector4a* output = triangle_normals.mArray; + LLVector4a* end_output = output+count; + + U16* idx = mIndices; + + while (output < end_output) + { + LLVector4a b,v1,v2; + b.load4a((F32*) (pos+idx[0])); + v1.load4a((F32*) (pos+idx[1])); + v2.load4a((F32*) (pos+idx[2])); + //calculate triangle normal - LLVector3 norm = (v0.mPosition-v1.mPosition) % (v0.mPosition-v2.mPosition); + LLVector4a a; + + a.setSub(b, v1); + b.sub(v2); - for (U32 j = 0; j < 3; j++) - { //add triangle normal to vertices - const S32 idx = mIndices[i*3+j]; - mVertices[idx].mNormal += norm; // * (weight_sum - d[j])/weight_sum; - } + + LLQuad& vector1 = *((LLQuad*) &v1); + LLQuad& vector2 = *((LLQuad*) &v2); + + LLQuad& amQ = *((LLQuad*) &a); + LLQuad& bmQ = *((LLQuad*) &b); + + //v1.setCross3(t,v0); + //setCross3(const LLVector4a& a, const LLVector4a& b) + // Vectors are stored in memory in w, z, y, x order from high to low + // Set vector1 = { a[W], a[X], a[Z], a[Y] } + vector1 = _mm_shuffle_ps( amQ, amQ, _MM_SHUFFLE( 3, 0, 2, 1 )); + // Set vector2 = { b[W], b[Y], b[X], b[Z] } + vector2 = _mm_shuffle_ps( bmQ, bmQ, _MM_SHUFFLE( 3, 1, 0, 2 )); + // mQ = { a[W]*b[W], a[X]*b[Y], a[Z]*b[X], a[Y]*b[Z] } + vector2 = _mm_mul_ps( vector1, vector2 ); + // vector3 = { a[W], a[Y], a[X], a[Z] } + amQ = _mm_shuffle_ps( amQ, amQ, _MM_SHUFFLE( 3, 1, 0, 2 )); + // vector4 = { b[W], b[X], b[Z], b[Y] } + bmQ = _mm_shuffle_ps( bmQ, bmQ, _MM_SHUFFLE( 3, 0, 2, 1 )); + // mQ = { 0, a[X]*b[Y] - a[Y]*b[X], a[Z]*b[X] - a[X]*b[Z], a[Y]*b[Z] - a[Z]*b[Y] } + vector1 = _mm_sub_ps( vector2, _mm_mul_ps( amQ, bmQ )); + + llassert(v1.isFinite3()); + + v1.store4a((F32*) output); + + + output++; + idx += 3; + } + + idx = mIndices; + + LLVector4a* src = triangle_normals.mArray; + + for (U32 i = 0; i < count; i++) //for each triangle + { + LLVector4a c; + c.load4a((F32*) (src++)); + + LLVector4a* n0p = norm+idx[0]; + LLVector4a* n1p = norm+idx[1]; + LLVector4a* n2p = norm+idx[2]; + + idx += 3; + + LLVector4a n0,n1,n2; + n0.load4a((F32*) n0p); + n1.load4a((F32*) n1p); + n2.load4a((F32*) n2p); + + n0.add(c); + n1.add(c); + n2.add(c); + + llassert(c.isFinite3()); //even out quad contributions - if ((i & 1) == 0) + switch (i%2+1) { - mVertices[i2].mNormal += norm; - } - else - { - mVertices[i1].mNormal += norm; - } + case 0: n0.add(c); break; + case 1: n1.add(c); break; + case 2: n2.add(c); break; + }; + + n0.store4a((F32*) n0p); + n1.store4a((F32*) n1p); + n2.store4a((F32*) n2p); } + LL_CHECK_MEMORY + // 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; } } @@ -5354,9 +6647,10 @@ 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; } } @@ -5367,7 +6661,7 @@ 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); } } @@ -5375,12 +6669,11 @@ 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; @@ -5403,30 +6696,33 @@ 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; } } @@ -5436,67 +6732,127 @@ 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; } - } } + LL_CHECK_MEMORY + return TRUE; } -// 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, - const LLVector2& tex0, - const LLVector3& pos1, - const LLVector2& tex1, - const LLVector3& 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] ); - - 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] ); - - 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] ); - - LLVector3 r0 = (rx0 - rx1) % (rx0 - rx2); - LLVector3 r1 = (ry0 - ry1) % (ry0 - ry2); - LLVector3 r2 = (rz0 - rz1) % (rz0 - rz2); +//adapted from Lengyel, Eric. "Computing Tangent Space Basis Vectors for an Arbitrary Mesh". Terathon Software 3D Graphics Library, 2001. http://www.terathon.com/code/tangent.html +void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVector4a *normal, + const LLVector2 *texcoord, U32 triangleCount, const U16* index_array, LLVector4a *tangent) +{ + //LLVector4a *tan1 = new LLVector4a[vertexCount * 2]; + LLVector4a* tan1 = (LLVector4a*) ll_aligned_malloc_16(vertexCount*2*sizeof(LLVector4a)); + + LLVector4a* tan2 = tan1 + vertexCount; + + memset(tan1, 0, vertexCount*2*sizeof(LLVector4a)); + + for (U32 a = 0; a < triangleCount; a++) + { + U32 i1 = *index_array++; + U32 i2 = *index_array++; + U32 i3 = *index_array++; + + const LLVector4a& v1 = vertex[i1]; + const LLVector4a& v2 = vertex[i2]; + const LLVector4a& v3 = vertex[i3]; + + const LLVector2& w1 = texcoord[i1]; + const LLVector2& w2 = texcoord[i2]; + const LLVector2& w3 = texcoord[i3]; + + const F32* v1ptr = v1.getF32ptr(); + const F32* v2ptr = v2.getF32ptr(); + const F32* v3ptr = v3.getF32ptr(); + + float x1 = v2ptr[0] - v1ptr[0]; + float x2 = v3ptr[0] - v1ptr[0]; + float y1 = v2ptr[1] - v1ptr[1]; + float y2 = v3ptr[1] - v1ptr[1]; + float z1 = v2ptr[2] - v1ptr[2]; + float z2 = v3ptr[2] - v1ptr[2]; + + float s1 = w2.mV[0] - w1.mV[0]; + float s2 = w3.mV[0] - w1.mV[0]; + float t1 = w2.mV[1] - w1.mV[1]; + float t2 = w3.mV[1] - w1.mV[1]; + + F32 rd = s1*t2-s2*t1; + + float r = ((rd*rd) > FLT_EPSILON) ? (1.0f / rd) + : ((rd > 0.0f) ? 1024.f : -1024.f); //some made up large ratio for division by zero + + llassert(llfinite(r)); + llassert(!llisnan(r)); + + LLVector4a sdir((t2 * x1 - t1 * x2) * r, (t2 * y1 - t1 * y2) * r, + (t2 * z1 - t1 * z2) * r); + LLVector4a tdir((s1 * x2 - s2 * x1) * r, (s1 * y2 - s2 * y1) * r, + (s1 * z2 - s2 * z1) * r); + + tan1[i1].add(sdir); + tan1[i2].add(sdir); + tan1[i3].add(sdir); + + tan2[i1].add(tdir); + tan2[i2].add(tdir); + tan2[i3].add(tdir); + } + + for (U32 a = 0; a < vertexCount; a++) + { + LLVector4a n = normal[a]; + + const LLVector4a& t = tan1[a]; + + LLVector4a ncrosst; + ncrosst.setCross3(n,t); + + // Gram-Schmidt orthogonalize + n.mul(n.dot3(t).getF32()); + + LLVector4a tsubn; + tsubn.setSub(t,n); + + if (tsubn.dot3(tsubn).getF32() > F_APPROXIMATELY_ZERO) + { + tsubn.normalize3fast(); + + // Calculate handedness + F32 handedness = ncrosst.dot3(tan2[a]).getF32() < 0.f ? -1.f : 1.f; + + tsubn.getF32ptr()[3] = handedness; - if( r0.mV[VX] && r1.mV[VX] && r2.mV[VX] ) - { - LLVector3 binormal( - -r0.mV[VZ] / r0.mV[VX], - -r1.mV[VZ] / r1.mV[VX], - -r2.mV[VZ] / r2.mV[VX]); - // binormal.normVec(); - return binormal; - } - else - { - return LLVector3( 0, 1 , 0 ); - } + tangent[a] = tsubn; + } + else + { //degenerate, make up a value + tangent[a].set(0,0,1,1); + } + } + + ll_aligned_free_16(tan1); } + + |