diff options
Diffstat (limited to 'indra/llmath')
49 files changed, 10804 insertions, 6039 deletions
diff --git a/indra/llmath/CMakeLists.txt b/indra/llmath/CMakeLists.txt index e93fe90650..9dadad7dd3 100644 --- a/indra/llmath/CMakeLists.txt +++ b/indra/llmath/CMakeLists.txt @@ -15,13 +15,16 @@ set(llmath_SOURCE_FILES llcamera.cpp llcoordframe.cpp llline.cpp + llmatrix3a.cpp llmodularmath.cpp llperlin.cpp llquaternion.cpp llrect.cpp llsphere.cpp + llvector4a.cpp llvolume.cpp llvolumemgr.cpp + llvolumeoctree.cpp llsdutil_math.cpp m3math.cpp m4math.cpp @@ -49,21 +52,32 @@ set(llmath_HEADER_FILES llinterp.h llline.h llmath.h + llmatrix3a.h + llmatrix3a.inl llmodularmath.h lloctree.h llperlin.h llplane.h llquantize.h llquaternion.h + llquaternion2.h + llquaternion2.inl llrect.h + llsimdmath.h + llsimdtypes.h + llsimdtypes.inl llsphere.h lltreenode.h + llvector4a.h + llvector4a.inl + llvector4logical.h llv4math.h llv4matrix3.h llv4matrix4.h llv4vector3.h llvolume.h llvolumemgr.h + llvolumeoctree.h llsdutil_math.h m3math.h m4math.h diff --git a/indra/llmath/llcamera.cpp b/indra/llmath/llcamera.cpp index bad4d00fd6..2ffc56644f 100644 --- a/indra/llmath/llcamera.cpp +++ b/indra/llmath/llcamera.cpp @@ -45,7 +45,6 @@ LLCamera::LLCamera() : calculateFrustumPlanes(); } - LLCamera::LLCamera(F32 vertical_fov_rads, F32 aspect_ratio, S32 view_height_in_pixels, F32 near_plane, F32 far_plane) : LLCoordFrame(), mViewHeightInPixels(view_height_in_pixels), @@ -61,6 +60,10 @@ LLCamera::LLCamera(F32 vertical_fov_rads, F32 aspect_ratio, S32 view_height_in_p setView(vertical_fov_rads); } +LLCamera::~LLCamera() +{ + +} // ---------------- LLCamera::getFoo() member functions ---------------- @@ -82,11 +85,11 @@ F32 LLCamera::getMaxView() const // ---------------- LLCamera::setFoo() member functions ---------------- -void LLCamera::setUserClipPlane(LLPlane plane) +void LLCamera::setUserClipPlane(LLPlane& plane) { mPlaneCount = 7; - mAgentPlanes[6].p = plane; - mAgentPlanes[6].mask = calcPlaneMask(plane); + mAgentPlanes[6] = plane; + mPlaneMask[6] = plane.calcPlaneMask(); } void LLCamera::disableUserClipPlane() @@ -158,166 +161,91 @@ size_t LLCamera::readFrustumFromBuffer(const char *buffer) // ---------------- test methods ---------------- -S32 LLCamera::AABBInFrustum(const LLVector3 ¢er, const LLVector3& radius) +S32 LLCamera::AABBInFrustum(const LLVector4a ¢er, const LLVector4a& radius) { - static const LLVector3 scaler[] = { - LLVector3(-1,-1,-1), - LLVector3( 1,-1,-1), - LLVector3(-1, 1,-1), - LLVector3( 1, 1,-1), - LLVector3(-1,-1, 1), - LLVector3( 1,-1, 1), - LLVector3(-1, 1, 1), - LLVector3( 1, 1, 1) + static const LLVector4a scaler[] = { + LLVector4a(-1,-1,-1), + LLVector4a( 1,-1,-1), + LLVector4a(-1, 1,-1), + LLVector4a( 1, 1,-1), + LLVector4a(-1,-1, 1), + LLVector4a( 1,-1, 1), + LLVector4a(-1, 1, 1), + LLVector4a( 1, 1, 1) }; U8 mask = 0; - S32 result = 2; - - /*if (mFrustumCornerDist > 0.f && radius.magVecSquared() > mFrustumCornerDist * mFrustumCornerDist) - { //box is larger than frustum, check frustum quads against box planes - - static const LLVector3 dir[] = - { - LLVector3(1, 0, 0), - LLVector3(-1, 0, 0), - LLVector3(0, 1, 0), - LLVector3(0, -1, 0), - LLVector3(0, 0, 1), - LLVector3(0, 0, -1) - }; - - U32 quads[] = - { - 0, 1, 2, 3, - 0, 1, 5, 4, - 2, 3, 7, 6, - 3, 0, 7, 4, - 1, 2, 6, 4, - 4, 5, 6, 7 - }; - - result = 0; - - BOOL total_inside = TRUE; - for (U32 i = 0; i < 6; i++) - { - LLVector3 p = center + radius.scaledVec(dir[i]); - F32 d = -p*dir[i]; - - for (U32 j = 0; j < 6; j++) - { //for each quad - F32 dist = mAgentFrustum[quads[j*4+0]]*dir[i] + d; - if (dist > 0) - { //at least one frustum point is outside the AABB - total_inside = FALSE; - for (U32 k = 1; k < 4; k++) - { //for each other point on quad - if ( mAgentFrustum[quads[j*4+k]]*dir[i]+d <= 0.f) - { //quad is straddling some plane of AABB - return 1; - } - } - } - else - { - for (U32 k = 1; k < 4; k++) - { - if (mAgentFrustum[quads[j*4+k]]*dir[i]+d > 0.f) - { - return 1; - } - } - } - } - } - - if (total_inside) - { - result = 1; - } - } - else*/ + bool result = false; + LLVector4a rscale, maxp, minp; + LLSimdScalar d; + for (U32 i = 0; i < mPlaneCount; i++) { - for (U32 i = 0; i < mPlaneCount; i++) + mask = mPlaneMask[i]; + if (mask != 0xff) { - mask = mAgentPlanes[i].mask; - if (mask == 0xff) - { - continue; - } - LLPlane p = mAgentPlanes[i].p; - LLVector3 n = LLVector3(p); - float d = p.mV[3]; - LLVector3 rscale = radius.scaledVec(scaler[mask]); - - LLVector3 minp = center - rscale; - LLVector3 maxp = center + rscale; - - if (n * minp > -d) + const LLPlane& p(mAgentPlanes[i]); + p.getAt<3>(d); + rscale.setMul(radius, scaler[mask]); + minp.setSub(center, rscale); + d = -d; + if (p.dot3(minp).getF32() > d) { return 0; } - - if (n * maxp > -d) + + if(!result) { - result = 1; + maxp.setAdd(center, rscale); + result = (p.dot3(maxp).getF32() > d); } } } - - return result; + return result?1:2; } -S32 LLCamera::AABBInFrustumNoFarClip(const LLVector3 ¢er, const LLVector3& radius) + +S32 LLCamera::AABBInFrustumNoFarClip(const LLVector4a& center, const LLVector4a& radius) { - static const LLVector3 scaler[] = { - LLVector3(-1,-1,-1), - LLVector3( 1,-1,-1), - LLVector3(-1, 1,-1), - LLVector3( 1, 1,-1), - LLVector3(-1,-1, 1), - LLVector3( 1,-1, 1), - LLVector3(-1, 1, 1), - LLVector3( 1, 1, 1) + static const LLVector4a scaler[] = { + LLVector4a(-1,-1,-1), + LLVector4a( 1,-1,-1), + LLVector4a(-1, 1,-1), + LLVector4a( 1, 1,-1), + LLVector4a(-1,-1, 1), + LLVector4a( 1,-1, 1), + LLVector4a(-1, 1, 1), + LLVector4a( 1, 1, 1) }; U8 mask = 0; - S32 result = 2; - + bool result = false; + LLVector4a rscale, maxp, minp; + LLSimdScalar d; for (U32 i = 0; i < mPlaneCount; i++) { - if (i == 5) - { - continue; - } - - mask = mAgentPlanes[i].mask; - if (mask == 0xff) - { - continue; - } - LLPlane p = mAgentPlanes[i].p; - LLVector3 n = LLVector3(p); - float d = p.mV[3]; - LLVector3 rscale = radius.scaledVec(scaler[mask]); - - LLVector3 minp = center - rscale; - LLVector3 maxp = center + rscale; - - if (n * minp > -d) + mask = mPlaneMask[i]; + if ((i != 5) && (mask != 0xff)) { - return 0; - } - - if (n * maxp > -d) - { - result = 1; + const LLPlane& p(mAgentPlanes[i]); + p.getAt<3>(d); + rscale.setMul(radius, scaler[mask]); + minp.setSub(center, rscale); + d = -d; + if (p.dot3(minp).getF32() > d) + { + return 0; + } + + if(!result) + { + maxp.setAdd(center, rscale); + result = (p.dot3(maxp).getF32() > d); + } } } - return result; + return result?1:2; } int LLCamera::sphereInFrustumQuick(const LLVector3 &sphere_center, const F32 radius) @@ -438,28 +366,22 @@ int LLCamera::sphereInFrustumOld(const LLVector3 &sphere_center, const F32 radiu int LLCamera::sphereInFrustum(const LLVector3 &sphere_center, const F32 radius) const { // Returns 1 if sphere is in frustum, 0 if not. - int res = 2; + bool res = false; for (int i = 0; i < 6; i++) { - if (mAgentPlanes[i].mask == 0xff) + if (mPlaneMask[i] != 0xff) { - continue; - } - - float d = mAgentPlanes[i].p.dist(sphere_center); + float d = mAgentPlanes[i].dist(sphere_center); - if (d > radius) - { - return 0; - } - - if (d > -radius) - { - res = 1; + if (d > radius) + { + return 0; + } + res = res || (d > -radius); } } - return res; + return res?1:2; } @@ -611,25 +533,6 @@ LLPlane planeFromPoints(LLVector3 p1, LLVector3 p2, LLVector3 p3) return LLPlane(p1, n); } -U8 LLCamera::calcPlaneMask(const LLPlane& plane) -{ - U8 mask = 0; - - if (plane.mV[0] >= 0) - { - mask |= 1; - } - if (plane.mV[1] >= 0) - { - mask |= 2; - } - if (plane.mV[2] >= 0) - { - mask |= 4; - } - - return mask; -} void LLCamera::ignoreAgentFrustumPlane(S32 idx) { @@ -638,12 +541,13 @@ void LLCamera::ignoreAgentFrustumPlane(S32 idx) return; } - mAgentPlanes[idx].mask = 0xff; - mAgentPlanes[idx].p.clearVec(); + mPlaneMask[idx] = 0xff; + mAgentPlanes[idx].clear(); } void LLCamera::calcAgentFrustumPlanes(LLVector3* frust) { + for (int i = 0; i < 8; i++) { mAgentFrustum[i] = frust[i]; @@ -656,27 +560,27 @@ void LLCamera::calcAgentFrustumPlanes(LLVector3* frust) //order of planes is important, keep most likely to fail in the front of the list //near - frust[0], frust[1], frust[2] - mAgentPlanes[2].p = planeFromPoints(frust[0], frust[1], frust[2]); + mAgentPlanes[2] = planeFromPoints(frust[0], frust[1], frust[2]); //far - mAgentPlanes[5].p = planeFromPoints(frust[5], frust[4], frust[6]); + mAgentPlanes[5] = planeFromPoints(frust[5], frust[4], frust[6]); //left - mAgentPlanes[0].p = planeFromPoints(frust[4], frust[0], frust[7]); + mAgentPlanes[0] = planeFromPoints(frust[4], frust[0], frust[7]); //right - mAgentPlanes[1].p = planeFromPoints(frust[1], frust[5], frust[6]); + mAgentPlanes[1] = planeFromPoints(frust[1], frust[5], frust[6]); //top - mAgentPlanes[4].p = planeFromPoints(frust[3], frust[2], frust[6]); + mAgentPlanes[4] = planeFromPoints(frust[3], frust[2], frust[6]); //bottom - mAgentPlanes[3].p = planeFromPoints(frust[1], frust[0], frust[4]); + mAgentPlanes[3] = planeFromPoints(frust[1], frust[0], frust[4]); //cache plane octant facing mask for use in AABBInFrustum for (U32 i = 0; i < mPlaneCount; i++) { - mAgentPlanes[i].mask = calcPlaneMask(mAgentPlanes[i].p); + mPlaneMask[i] = mAgentPlanes[i].calcPlaneMask(); } } @@ -729,9 +633,10 @@ void LLCamera::calculateWorldFrustumPlanes() F32 d; LLVector3 center = mOrigin - mXAxis*mNearPlane; mWorldPlanePos = center; + LLVector3 pnorm; for (int p=0; p<4; p++) { - LLVector3 pnorm = LLVector3(mLocalPlanes[p]); + mLocalPlanes[p].getVector3(pnorm); LLVector3 norm = rotateToAbsolute(pnorm); norm.normVec(); d = -(center * norm); @@ -741,13 +646,15 @@ void LLCamera::calculateWorldFrustumPlanes() LLVector3 zaxis(0, 0, 1.0f); F32 yaw = getYaw(); { - LLVector3 tnorm = LLVector3(mLocalPlanes[PLANE_LEFT]); + LLVector3 tnorm; + mLocalPlanes[PLANE_LEFT].getVector3(tnorm); tnorm.rotVec(yaw, zaxis); d = -(mOrigin * tnorm); mHorizPlanes[HORIZ_PLANE_LEFT] = LLPlane(tnorm, d); } { - LLVector3 tnorm = LLVector3(mLocalPlanes[PLANE_RIGHT]); + LLVector3 tnorm; + mLocalPlanes[PLANE_RIGHT].getVector3(tnorm); tnorm.rotVec(yaw, zaxis); d = -(mOrigin * tnorm); mHorizPlanes[HORIZ_PLANE_RIGHT] = LLPlane(tnorm, d); diff --git a/indra/llmath/llcamera.h b/indra/llmath/llcamera.h index 922d6f9fac..82d80f1057 100644 --- a/indra/llmath/llcamera.h +++ b/indra/llmath/llcamera.h @@ -31,6 +31,7 @@ #include "llmath.h" #include "llcoordframe.h" #include "llplane.h" +#include "llvector4a.h" const F32 DEFAULT_FIELD_OF_VIEW = 60.f * DEG_TO_RAD; const F32 DEFAULT_ASPECT_RATIO = 640.f / 480.f; @@ -73,6 +74,12 @@ class LLCamera : public LLCoordFrame { public: + + LLCamera(const LLCamera& rhs) + { + *this = rhs; + } + enum { PLANE_LEFT = 0, PLANE_RIGHT = 1, @@ -110,6 +117,9 @@ public: }; private: + LLPlane mAgentPlanes[7]; //frustum planes in agent space a la gluUnproject (I'm a bastard, I know) - DaveP + U8 mPlaneMask[8]; // 8 for alignment + F32 mView; // angle between top and bottom frustum planes in radians. F32 mAspect; // width/height S32 mViewHeightInPixels; // for ViewHeightInPixels() only @@ -123,30 +133,22 @@ private: LLPlane mWorldPlanes[PLANE_NUM]; LLPlane mHorizPlanes[HORIZ_PLANE_NUM]; - struct frustum_plane - { - frustum_plane() : mask(0) {} - LLPlane p; - U8 mask; - }; - frustum_plane mAgentPlanes[7]; //frustum planes in agent space a la gluUnproject (I'm a bastard, I know) - DaveP - U32 mPlaneCount; //defaults to 6, if setUserClipPlane is called, uses user supplied clip plane in LLVector3 mWorldPlanePos; // Position of World Planes (may be offset from camera) public: LLVector3 mAgentFrustum[8]; //8 corners of 6-plane frustum F32 mFrustumCornerDist; //distance to corner of frustum against far clip plane - LLPlane getAgentPlane(U32 idx) { return mAgentPlanes[idx].p; } + LLPlane& getAgentPlane(U32 idx) { return mAgentPlanes[idx]; } public: LLCamera(); LLCamera(F32 vertical_fov_rads, F32 aspect_ratio, S32 view_height_in_pixels, F32 near_plane, F32 far_plane); - virtual ~LLCamera(){} // no-op virtual destructor + virtual ~LLCamera(); + - void setUserClipPlane(LLPlane plane); + void setUserClipPlane(LLPlane& plane); void disableUserClipPlane(); - U8 calcPlaneMask(const LLPlane& plane); virtual void setView(F32 vertical_fov_rads); void setViewHeightInPixels(S32 height); void setAspect(F32 new_aspect); @@ -193,8 +195,8 @@ public: S32 sphereInFrustum(const LLVector3 ¢er, const F32 radius) const; S32 pointInFrustum(const LLVector3 &point) const { return sphereInFrustum(point, 0.0f); } S32 sphereInFrustumFull(const LLVector3 ¢er, const F32 radius) const { return sphereInFrustum(center, radius); } - S32 AABBInFrustum(const LLVector3 ¢er, const LLVector3& radius); - S32 AABBInFrustumNoFarClip(const LLVector3 ¢er, const LLVector3& radius); + S32 AABBInFrustum(const LLVector4a& center, const LLVector4a& radius); + S32 AABBInFrustumNoFarClip(const LLVector4a& center, const LLVector4a& radius); //does a quick 'n dirty sphere-sphere check S32 sphereInFrustumQuick(const LLVector3 &sphere_center, const F32 radius); diff --git a/indra/llmath/llmath.h b/indra/llmath/llmath.h index 798f1154d0..25b29ac1a8 100644 --- a/indra/llmath/llmath.h +++ b/indra/llmath/llmath.h @@ -29,7 +29,6 @@ #include <cmath> #include <cstdlib> -#include <complex> #include "lldefs.h" //#include "llstl.h" // *TODO: Remove when LLString is gone //#include "llstring.h" // *TODO: Remove when LLString is gone @@ -55,32 +54,11 @@ #endif // Single Precision Floating Point Routines -#ifndef sqrtf -#define sqrtf(x) ((F32)sqrt((F64)(x))) -#endif -#ifndef fsqrtf -#define fsqrtf(x) sqrtf(x) -#endif - -#ifndef cosf -#define cosf(x) ((F32)cos((F64)(x))) -#endif -#ifndef sinf -#define sinf(x) ((F32)sin((F64)(x))) -#endif -#ifndef tanf +// (There used to be more defined here, but they appeared to be redundant and +// were breaking some other includes. Removed by Falcon, reviewed by Andrew, 11/25/09) +/*#ifndef tanf #define tanf(x) ((F32)tan((F64)(x))) -#endif -#ifndef acosf -#define acosf(x) ((F32)acos((F64)(x))) -#endif - -#ifndef powf -#define powf(x,y) ((F32)pow((F64)(x),(F64)(y))) -#endif -#ifndef expf -#define expf(x) ((F32)exp((F64)(x))) -#endif +#endif*/ const F32 GRAVITY = -9.8f; @@ -200,7 +178,7 @@ inline S32 llfloor( F32 f ) } return result; #else - return (S32)floorf(f); + return (S32)floor(f); #endif } @@ -378,11 +356,14 @@ inline F32 snap_to_sig_figs(F32 foo, S32 sig_figs) bar *= 10.f; } - foo = (F32)llround(foo * bar); + //F32 new_foo = (F32)llround(foo * bar); + // the llround() implementation sucks. Don't us it. + + F32 sign = (foo > 0.f) ? 1.f : -1.f; + F32 new_foo = F32( S64(foo * bar + sign * 0.5f)); + new_foo /= bar; - // shift back - foo /= bar; - return foo; + return new_foo; } inline F32 lerp(F32 a, F32 b, F32 u) @@ -516,4 +497,7 @@ inline F32 llgaussian(F32 x, F32 o) return 1.f/(F_SQRT_TWO_PI*o)*powf(F_E, -(x*x)/(2*o*o)); } +// Include simd math header +#include "llsimdmath.h" + #endif diff --git a/indra/llmath/llmatrix3a.cpp b/indra/llmath/llmatrix3a.cpp new file mode 100644 index 0000000000..ab077abcb0 --- /dev/null +++ b/indra/llmath/llmatrix3a.cpp @@ -0,0 +1,134 @@ +/** + * @file llvector4a.cpp + * @brief SIMD vector implementation + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#include "llmath.h" + +static LL_ALIGN_16(const F32 M_IDENT_3A[12]) = + { 1.f, 0.f, 0.f, 0.f, // Column 1 + 0.f, 1.f, 0.f, 0.f, // Column 2 + 0.f, 0.f, 1.f, 0.f }; // Column 3 + +extern const LLMatrix3a LL_M3A_IDENTITY = *reinterpret_cast<const LLMatrix3a*> (M_IDENT_3A); + +void LLMatrix3a::setMul( const LLMatrix3a& lhs, const LLMatrix3a& rhs ) +{ + const LLVector4a col0 = lhs.getColumn(0); + const LLVector4a col1 = lhs.getColumn(1); + const LLVector4a col2 = lhs.getColumn(2); + + for ( int i = 0; i < 3; i++ ) + { + LLVector4a xxxx = _mm_load_ss( rhs.mColumns[i].getF32ptr() ); + xxxx.splat<0>( xxxx ); + xxxx.mul( col0 ); + + { + LLVector4a yyyy = _mm_load_ss( rhs.mColumns[i].getF32ptr() + 1 ); + yyyy.splat<0>( yyyy ); + yyyy.mul( col1 ); + xxxx.add( yyyy ); + } + + { + LLVector4a zzzz = _mm_load_ss( rhs.mColumns[i].getF32ptr() + 2 ); + zzzz.splat<0>( zzzz ); + zzzz.mul( col2 ); + xxxx.add( zzzz ); + } + + xxxx.store4a( mColumns[i].getF32ptr() ); + } + +} + +/*static */void LLMatrix3a::batchTransform( const LLMatrix3a& xform, const LLVector4a* src, int numVectors, LLVector4a* dst ) +{ + const LLVector4a col0 = xform.getColumn(0); + const LLVector4a col1 = xform.getColumn(1); + const LLVector4a col2 = xform.getColumn(2); + const LLVector4a* maxAddr = src + numVectors; + + if ( numVectors & 0x1 ) + { + LLVector4a xxxx = _mm_load_ss( (const F32*)src ); + LLVector4a yyyy = _mm_load_ss( (const F32*)src + 1 ); + LLVector4a zzzz = _mm_load_ss( (const F32*)src + 2 ); + xxxx.splat<0>( xxxx ); + yyyy.splat<0>( yyyy ); + zzzz.splat<0>( zzzz ); + xxxx.mul( col0 ); + yyyy.mul( col1 ); + zzzz.mul( col2 ); + xxxx.add( yyyy ); + xxxx.add( zzzz ); + xxxx.store4a( (F32*)dst ); + src++; + dst++; + } + + + numVectors >>= 1; + while ( src < maxAddr ) + { + _mm_prefetch( (const char*)(src + 32 ), _MM_HINT_NTA ); + _mm_prefetch( (const char*)(dst + 32), _MM_HINT_NTA ); + LLVector4a xxxx = _mm_load_ss( (const F32*)src ); + LLVector4a xxxx1= _mm_load_ss( (const F32*)(src + 1) ); + + xxxx.splat<0>( xxxx ); + xxxx1.splat<0>( xxxx1 ); + xxxx.mul( col0 ); + xxxx1.mul( col0 ); + + { + LLVector4a yyyy = _mm_load_ss( (const F32*)src + 1 ); + LLVector4a yyyy1 = _mm_load_ss( (const F32*)(src + 1) + 1); + yyyy.splat<0>( yyyy ); + yyyy1.splat<0>( yyyy1 ); + yyyy.mul( col1 ); + yyyy1.mul( col1 ); + xxxx.add( yyyy ); + xxxx1.add( yyyy1 ); + } + + { + LLVector4a zzzz = _mm_load_ss( (const F32*)(src) + 2 ); + LLVector4a zzzz1 = _mm_load_ss( (const F32*)(++src) + 2 ); + zzzz.splat<0>( zzzz ); + zzzz1.splat<0>( zzzz1 ); + zzzz.mul( col2 ); + zzzz1.mul( col2 ); + xxxx.add( zzzz ); + xxxx1.add( zzzz1 ); + } + + xxxx.store4a(dst->getF32ptr()); + src++; + dst++; + + xxxx1.store4a((F32*)dst++); + } +} diff --git a/indra/llmath/llmatrix3a.h b/indra/llmath/llmatrix3a.h new file mode 100644 index 0000000000..adb7e3389d --- /dev/null +++ b/indra/llmath/llmatrix3a.h @@ -0,0 +1,128 @@ +/** + * @file llmatrix3a.h + * @brief LLMatrix3a class header file - memory aligned and vectorized 3x3 matrix + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_LLMATRIX3A_H +#define LL_LLMATRIX3A_H + +///////////////////////////// +// LLMatrix3a, LLRotation +///////////////////////////// +// This class stores a 3x3 (technically 4x3) matrix in column-major order +///////////////////////////// +///////////////////////////// +// These classes are intentionally minimal right now. If you need additional +// functionality, please contact someone with SSE experience (e.g., Falcon or +// Huseby). +///////////////////////////// + +// LLMatrix3a is the base class for LLRotation, which should be used instead any time you're dealing with a +// rotation matrix. +class LLMatrix3a +{ +public: + + // Utility function for quickly transforming an array of LLVector4a's + // For transforming a single LLVector4a, see LLVector4a::setRotated + static void batchTransform( const LLMatrix3a& xform, const LLVector4a* src, int numVectors, LLVector4a* dst ); + + // Utility function to obtain the identity matrix + static inline const LLMatrix3a& getIdentity(); + + ////////////////////////// + // Ctors + ////////////////////////// + + // Ctor + LLMatrix3a() {} + + // Ctor for setting by columns + inline LLMatrix3a( const LLVector4a& c0, const LLVector4a& c1, const LLVector4a& c2 ); + + ////////////////////////// + // Get/Set + ////////////////////////// + + // Loads from an LLMatrix3 + inline void loadu(const LLMatrix3& src); + + // Set rows + inline void setRows(const LLVector4a& r0, const LLVector4a& r1, const LLVector4a& r2); + + // Set columns + inline void setColumns(const LLVector4a& c0, const LLVector4a& c1, const LLVector4a& c2); + + // Get the read-only access to a specified column. Valid columns are 0-2, but the + // function is unchecked. You've been warned. + inline const LLVector4a& getColumn(const U32 column) const; + + ///////////////////////// + // Matrix modification + ///////////////////////// + + // Set this matrix to the product of lhs and rhs ( this = lhs * rhs ) + void setMul( const LLMatrix3a& lhs, const LLMatrix3a& rhs ); + + // Set this matrix to the transpose of src + inline void setTranspose(const LLMatrix3a& src); + + // Set this matrix to a*w + b*(1-w) + inline void setLerp(const LLMatrix3a& a, const LLMatrix3a& b, F32 w); + + ///////////////////////// + // Matrix inspection + ///////////////////////// + + // Sets all 4 elements in 'dest' to the determinant of this matrix. + // If you will be using the determinant in subsequent ops with LLVector4a, use this version + inline void getDeterminant( LLVector4a& dest ) const; + + // Returns the determinant as an LLSimdScalar. Use this if you will be using the determinant + // primary for scalar operations. + inline LLSimdScalar getDeterminant() const; + + // Returns nonzero if rows 0-2 and colums 0-2 contain no NaN or INF values. Row 3 is ignored + inline LLBool32 isFinite() const; + + // Returns true if this matrix is equal to 'rhs' up to 'tolerance' + inline bool isApproximatelyEqual( const LLMatrix3a& rhs, F32 tolerance = F_APPROXIMATELY_ZERO ) const; + +protected: + + LLVector4a mColumns[3]; + +}; + +class LLRotation : public LLMatrix3a +{ +public: + + LLRotation() {} + + // Returns true if this rotation is orthonormal with det ~= 1 + inline bool isOkRotation() const; +}; + +#endif diff --git a/indra/llmath/llmatrix3a.inl b/indra/llmath/llmatrix3a.inl new file mode 100644 index 0000000000..37819fea3c --- /dev/null +++ b/indra/llmath/llmatrix3a.inl @@ -0,0 +1,119 @@ +/** + * @file llmatrix3a.inl + * @brief LLMatrix3a inline definitions + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#include "llmatrix3a.h" +#include "m3math.h" + +inline LLMatrix3a::LLMatrix3a( const LLVector4a& c0, const LLVector4a& c1, const LLVector4a& c2 ) +{ + setColumns( c0, c1, c2 ); +} + +inline void LLMatrix3a::loadu(const LLMatrix3& src) +{ + mColumns[0].load3(src.mMatrix[0]); + mColumns[1].load3(src.mMatrix[1]); + mColumns[2].load3(src.mMatrix[2]); +} + +inline void LLMatrix3a::setRows(const LLVector4a& r0, const LLVector4a& r1, const LLVector4a& r2) +{ + mColumns[0] = r0; + mColumns[1] = r1; + mColumns[2] = r2; + setTranspose( *this ); +} + +inline void LLMatrix3a::setColumns(const LLVector4a& c0, const LLVector4a& c1, const LLVector4a& c2) +{ + mColumns[0] = c0; + mColumns[1] = c1; + mColumns[2] = c2; +} + +inline void LLMatrix3a::setTranspose(const LLMatrix3a& src) +{ + const LLQuad srcCol0 = src.mColumns[0]; + const LLQuad srcCol1 = src.mColumns[1]; + const LLQuad unpacklo = _mm_unpacklo_ps( srcCol0, srcCol1 ); + mColumns[0] = _mm_movelh_ps( unpacklo, src.mColumns[2] ); + mColumns[1] = _mm_shuffle_ps( _mm_movehl_ps( srcCol0, unpacklo ), src.mColumns[2], _MM_SHUFFLE(0, 1, 1, 0) ); + mColumns[2] = _mm_shuffle_ps( _mm_unpackhi_ps( srcCol0, srcCol1 ), src.mColumns[2], _MM_SHUFFLE(0, 2, 1, 0) ); +} + +inline const LLVector4a& LLMatrix3a::getColumn(const U32 column) const +{ + llassert( column < 3 ); + return mColumns[column]; +} + +inline void LLMatrix3a::setLerp(const LLMatrix3a& a, const LLMatrix3a& b, F32 w) +{ + mColumns[0].setLerp( a.mColumns[0], b.mColumns[0], w ); + mColumns[1].setLerp( a.mColumns[1], b.mColumns[1], w ); + mColumns[2].setLerp( a.mColumns[2], b.mColumns[2], w ); +} + +inline LLBool32 LLMatrix3a::isFinite() const +{ + return mColumns[0].isFinite3() && mColumns[1].isFinite3() && mColumns[2].isFinite3(); +} + +inline void LLMatrix3a::getDeterminant( LLVector4a& dest ) const +{ + LLVector4a col1xcol2; col1xcol2.setCross3( mColumns[1], mColumns[2] ); + dest.setAllDot3( col1xcol2, mColumns[0] ); +} + +inline LLSimdScalar LLMatrix3a::getDeterminant() const +{ + LLVector4a col1xcol2; col1xcol2.setCross3( mColumns[1], mColumns[2] ); + return col1xcol2.dot3( mColumns[0] ); +} + +inline bool LLMatrix3a::isApproximatelyEqual( const LLMatrix3a& rhs, F32 tolerance /*= F_APPROXIMATELY_ZERO*/ ) const +{ + return rhs.getColumn(0).equals3(mColumns[0], tolerance) + && rhs.getColumn(1).equals3(mColumns[1], tolerance) + && rhs.getColumn(2).equals3(mColumns[2], tolerance); +} + +inline const LLMatrix3a& LLMatrix3a::getIdentity() +{ + extern const LLMatrix3a LL_M3A_IDENTITY; + return LL_M3A_IDENTITY; +} + +inline bool LLRotation::isOkRotation() const +{ + LLMatrix3a transpose; transpose.setTranspose( *this ); + LLMatrix3a product; product.setMul( *this, transpose ); + + LLSimdScalar detMinusOne = getDeterminant() - 1.f; + + return product.isApproximatelyEqual( LLMatrix3a::getIdentity() ) && (detMinusOne.getAbs() < F_APPROXIMATELY_ZERO); +} + diff --git a/indra/llmath/llmatrix4a.h b/indra/llmath/llmatrix4a.h new file mode 100644 index 0000000000..27cf5b79f6 --- /dev/null +++ b/indra/llmath/llmatrix4a.h @@ -0,0 +1,143 @@ +/** + * @file llmatrix4a.h + * @brief LLMatrix4a class header file - memory aligned and vectorized 4x4 matrix + * + * $LicenseInfo:firstyear=2007&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_LLMATRIX4A_H +#define LL_LLMATRIX4A_H + +#include "llvector4a.h" +#include "m4math.h" +#include "m3math.h" + +class LLMatrix4a +{ +public: + LLVector4a mMatrix[4]; + + inline void clear() + { + mMatrix[0].clear(); + mMatrix[1].clear(); + mMatrix[2].clear(); + mMatrix[3].clear(); + } + + inline void loadu(const LLMatrix4& src) + { + mMatrix[0] = _mm_loadu_ps(src.mMatrix[0]); + mMatrix[1] = _mm_loadu_ps(src.mMatrix[1]); + mMatrix[2] = _mm_loadu_ps(src.mMatrix[2]); + mMatrix[3] = _mm_loadu_ps(src.mMatrix[3]); + + } + + inline void loadu(const LLMatrix3& src) + { + mMatrix[0].load3(src.mMatrix[0]); + mMatrix[1].load3(src.mMatrix[1]); + mMatrix[2].load3(src.mMatrix[2]); + mMatrix[3].set(0,0,0,1.f); + } + + inline void add(const LLMatrix4a& rhs) + { + mMatrix[0].add(rhs.mMatrix[0]); + mMatrix[1].add(rhs.mMatrix[1]); + mMatrix[2].add(rhs.mMatrix[2]); + mMatrix[3].add(rhs.mMatrix[3]); + } + + inline void setRows(const LLVector4a& r0, const LLVector4a& r1, const LLVector4a& r2) + { + mMatrix[0] = r0; + mMatrix[1] = r1; + mMatrix[2] = r2; + } + + inline void setMul(const LLMatrix4a& m, const F32 s) + { + mMatrix[0].setMul(m.mMatrix[0], s); + mMatrix[1].setMul(m.mMatrix[1], s); + mMatrix[2].setMul(m.mMatrix[2], s); + mMatrix[3].setMul(m.mMatrix[3], s); + } + + inline void setLerp(const LLMatrix4a& a, const LLMatrix4a& b, F32 w) + { + LLVector4a d0,d1,d2,d3; + d0.setSub(b.mMatrix[0], a.mMatrix[0]); + d1.setSub(b.mMatrix[1], a.mMatrix[1]); + d2.setSub(b.mMatrix[2], a.mMatrix[2]); + d3.setSub(b.mMatrix[3], a.mMatrix[3]); + + // this = a + d*w + + d0.mul(w); + d1.mul(w); + d2.mul(w); + d3.mul(w); + + mMatrix[0].setAdd(a.mMatrix[0],d0); + mMatrix[1].setAdd(a.mMatrix[1],d1); + mMatrix[2].setAdd(a.mMatrix[2],d2); + mMatrix[3].setAdd(a.mMatrix[3],d3); + } + + inline void rotate(const LLVector4a& v, LLVector4a& res) + { + res = _mm_shuffle_ps(v, v, _MM_SHUFFLE(0, 0, 0, 0)); + res.mul(mMatrix[0]); + + LLVector4a y; + y = _mm_shuffle_ps(v, v, _MM_SHUFFLE(1, 1, 1, 1)); + y.mul(mMatrix[1]); + + LLVector4a z; + z = _mm_shuffle_ps(v, v, _MM_SHUFFLE(2, 2, 2, 2)); + z.mul(mMatrix[2]); + + res.add(y); + res.add(z); + } + + inline void affineTransform(const LLVector4a& v, LLVector4a& res) + { + LLVector4a x,y,z; + + x = _mm_shuffle_ps(v, v, _MM_SHUFFLE(0, 0, 0, 0)); + y = _mm_shuffle_ps(v, v, _MM_SHUFFLE(1, 1, 1, 1)); + z = _mm_shuffle_ps(v, v, _MM_SHUFFLE(2, 2, 2, 2)); + + x.mul(mMatrix[0]); + y.mul(mMatrix[1]); + z.mul(mMatrix[2]); + + x.add(y); + z.add(mMatrix[3]); + res.setAdd(x,z); + } +}; + +#endif diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h index 90d4d742c9..fdfc24f8b7 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -29,14 +29,11 @@ #include "lltreenode.h" #include "v3math.h" +#include "llvector4a.h" #include <vector> #include <set> -#if LL_RELEASE_WITH_DEBUG_INFO || LL_DEBUG -#define OCT_ERRS LL_ERRS("OctreeErrors") -#else #define OCT_ERRS LL_WARNS("OctreeErrors") -#endif #define LL_OCTREE_PARANOIA_CHECK 0 #if LL_DARWIN @@ -67,6 +64,13 @@ public: }; template <class T> +class LLOctreeTravelerDepthFirst : public LLOctreeTraveler<T> +{ +public: + virtual void traverse(const LLOctreeNode<T>* node); +}; + +template <class T> class LLOctreeNode : public LLTreeNode<T> { public: @@ -81,23 +85,30 @@ public: typedef LLOctreeNode<T> oct_node; typedef LLOctreeListener<T> oct_listener; - static const U8 OCTANT_POSITIVE_X = 0x01; - static const U8 OCTANT_POSITIVE_Y = 0x02; - static const U8 OCTANT_POSITIVE_Z = 0x04; - - LLOctreeNode( LLVector3d center, - LLVector3d size, + /*void* operator new(size_t size) + { + return ll_aligned_malloc_16(size); + } + + void operator delete(void* ptr) + { + ll_aligned_free_16(ptr); + }*/ + + LLOctreeNode( const LLVector4a& center, + const LLVector4a& size, BaseType* parent, U8 octant = 255) : mParent((oct_node*)parent), - mCenter(center), - mSize(size), mOctant(octant) { + mCenter = center; + mSize = size; + updateMinMax(); if ((mOctant == 255) && mParent) { - mOctant = ((oct_node*) mParent)->getOctant(mCenter.mdV); + mOctant = ((oct_node*) mParent)->getOctant(mCenter); } clearChildren(); @@ -114,40 +125,24 @@ public: } inline const BaseType* getParent() const { return mParent; } - inline void setParent(BaseType* parent) { mParent = (oct_node*) parent; } - inline const LLVector3d& getCenter() const { return mCenter; } - inline const LLVector3d& getSize() const { return mSize; } - inline void setCenter(LLVector3d center) { mCenter = center; } - inline void setSize(LLVector3d size) { mSize = size; } - inline oct_node* getNodeAt(T* data) { return getNodeAt(data->getPositionGroup(), data->getBinRadius()); } - inline U8 getOctant() const { return mOctant; } - inline void setOctant(U8 octant) { mOctant = octant; } + inline void setParent(BaseType* parent) { mParent = (oct_node*) parent; } + inline const LLVector4a& getCenter() const { return mCenter; } + inline const LLVector4a& getSize() const { return mSize; } + inline void setCenter(const LLVector4a& center) { mCenter = center; } + inline void setSize(const LLVector4a& size) { mSize = size; } + inline oct_node* getNodeAt(T* data) { return getNodeAt(data->getPositionGroup(), data->getBinRadius()); } + inline U8 getOctant() const { return mOctant; } inline const oct_node* getOctParent() const { return (const oct_node*) getParent(); } inline oct_node* getOctParent() { return (oct_node*) getParent(); } - U8 getOctant(const F64 pos[]) const //get the octant pos is in + U8 getOctant(const LLVector4a& pos) const //get the octant pos is in { - U8 ret = 0; - - if (pos[0] > mCenter.mdV[0]) - { - ret |= OCTANT_POSITIVE_X; - } - if (pos[1] > mCenter.mdV[1]) - { - ret |= OCTANT_POSITIVE_Y; - } - if (pos[2] > mCenter.mdV[2]) - { - ret |= OCTANT_POSITIVE_Z; - } - - return ret; + return (U8) (pos.greaterThan(mCenter).getGatheredBits() & 0x7); } - inline bool isInside(const LLVector3d& pos, const F64& rad) const + inline bool isInside(const LLVector4a& pos, const F32& rad) const { - return rad <= mSize.mdV[0]*2.0 && isInside(pos); + return rad <= mSize[0]*2.f && isInside(pos); } inline bool isInside(T* data) const @@ -155,29 +150,27 @@ public: return isInside(data->getPositionGroup(), data->getBinRadius()); } - bool isInside(const LLVector3d& pos) const + bool isInside(const LLVector4a& pos) const { - const F64& x = pos.mdV[0]; - const F64& y = pos.mdV[1]; - const F64& z = pos.mdV[2]; - - if (x > mMax.mdV[0] || x <= mMin.mdV[0] || - y > mMax.mdV[1] || y <= mMin.mdV[1] || - z > mMax.mdV[2] || z <= mMin.mdV[2]) + S32 gt = pos.greaterThan(mMax).getGatheredBits() & 0x7; + if (gt) { return false; } - + + S32 lt = pos.lessEqual(mMin).getGatheredBits() & 0x7; + if (lt) + { + return false; + } + return true; } void updateMinMax() { - for (U32 i = 0; i < 3; i++) - { - mMax.mdV[i] = mCenter.mdV[i] + mSize.mdV[i]; - mMin.mdV[i] = mCenter.mdV[i] - mSize.mdV[i]; - } + mMax.setAdd(mCenter, mSize); + mMin.setSub(mCenter, mSize); } inline oct_listener* getOctListener(U32 index) @@ -190,34 +183,34 @@ public: return contains(xform->getBinRadius()); } - bool contains(F64 radius) + bool contains(F32 radius) { if (mParent == NULL) { //root node contains nothing return false; } - F64 size = mSize.mdV[0]; - F64 p_size = size * 2.0; + F32 size = mSize[0]; + F32 p_size = size * 2.f; - return (radius <= 0.001 && size <= 0.001) || + return (radius <= 0.001f && size <= 0.001f) || (radius <= p_size && radius > size); } - static void pushCenter(LLVector3d ¢er, const LLVector3d &size, const T* data) + static void pushCenter(LLVector4a ¢er, const LLVector4a &size, const T* data) { - const LLVector3d& pos = data->getPositionGroup(); - for (U32 i = 0; i < 3; i++) - { - if (pos.mdV[i] > center.mdV[i]) - { - center.mdV[i] += size.mdV[i]; - } - else - { - center.mdV[i] -= size.mdV[i]; - } - } + const LLVector4a& pos = data->getPositionGroup(); + + LLVector4Logical gt = pos.greaterThan(center); + + LLVector4a up; + up = _mm_and_ps(size, gt); + + LLVector4a down; + down = _mm_andnot_ps(gt, size); + + center.add(up); + center.sub(down); } void accept(oct_traveler* visitor) { visitor->visit(this); } @@ -236,32 +229,49 @@ public: void accept(tree_traveler* visitor) const { visitor->visit(this); } void accept(oct_traveler* visitor) const { visitor->visit(this); } - oct_node* getNodeAt(const LLVector3d& pos, const F64& rad) + void validateChildMap() + { + for (U32 i = 0; i < 8; i++) + { + U8 idx = mChildMap[i]; + if (idx != 255) + { + LLOctreeNode<T>* child = mChild[idx]; + + if (child->getOctant() != i) + { + llerrs << "Invalid child map, bad octant data." << llendl; + } + + if (getOctant(child->getCenter()) != child->getOctant()) + { + llerrs << "Invalid child octant compared to position data." << llendl; + } + } + } + } + + + oct_node* getNodeAt(const LLVector4a& pos, const F32& rad) { LLOctreeNode<T>* node = this; if (node->isInside(pos, rad)) { //do a quick search by octant - U8 octant = node->getOctant(pos.mdV); - BOOL keep_going = TRUE; - + U8 octant = node->getOctant(pos); + //traverse the tree until we find a node that has no node //at the appropriate octant or is smaller than the object. //by definition, that node is the smallest node that contains // the data - while (keep_going && node->getSize().mdV[0] >= rad) + U8 next_node = node->mChildMap[octant]; + + while (next_node != 255 && node->getSize()[0] >= rad) { - keep_going = FALSE; - for (U32 i = 0; i < node->getChildCount() && !keep_going; i++) - { - if (node->getChild(i)->getOctant() == octant) - { - node = node->getChild(i); - octant = node->getOctant(pos.mdV); - keep_going = TRUE; - } - } + node = node->getChild(next_node); + octant = node->getOctant(pos); + next_node = node->mChildMap[octant]; } } else if (!node->contains(rad) && node->getParent()) @@ -276,7 +286,7 @@ public: { if (data == NULL) { - //OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << llendl; + OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE BRANCH !!!" << llendl; return false; } LLOctreeNode<T>* parent = getOctParent(); @@ -284,10 +294,8 @@ public: //is it here? if (isInside(data->getPositionGroup())) { - if (getElementCount() < LL_OCTREE_MAX_CAPACITY && - (contains(data->getBinRadius()) || - (data->getBinRadius() > getSize().mdV[0] && - parent && parent->getElementCount() >= LL_OCTREE_MAX_CAPACITY))) + if ((getElementCount() < LL_OCTREE_MAX_CAPACITY && contains(data->getBinRadius()) || + (data->getBinRadius() > getSize()[0] && parent && parent->getElementCount() >= LL_OCTREE_MAX_CAPACITY))) { //it belongs here #if LL_OCTREE_PARANOIA_CHECK //if this is a redundant insertion, error out (should never happen) @@ -317,16 +325,21 @@ public: } //it's here, but no kids are in the right place, make a new kid - LLVector3d center(getCenter()); - LLVector3d size(getSize()*0.5); + LLVector4a center = getCenter(); + LLVector4a size = getSize(); + size.mul(0.5f); //push center in direction of data LLOctreeNode<T>::pushCenter(center, size, data); // handle case where floating point number gets too small - if( llabs(center.mdV[0] - getCenter().mdV[0]) < F_APPROXIMATELY_ZERO && - llabs(center.mdV[1] - getCenter().mdV[1]) < F_APPROXIMATELY_ZERO && - llabs(center.mdV[2] - getCenter().mdV[2]) < F_APPROXIMATELY_ZERO) + LLVector4a val; + val.setSub(center, getCenter()); + val.setAbs(val); + + S32 lt = val.lessThan(LLVector4a::getEpsilon()).getGatheredBits() & 0x7; + + if( lt == 0x7 ) { mData.insert(data); BaseType::insert(data); @@ -344,7 +357,7 @@ public: //make sure no existing node matches this position for (U32 i = 0; i < getChildCount(); i++) { - if (mChild[i]->getCenter() == center) + if (mChild[i]->getCenter().equals3(center)) { OCT_ERRS << "Octree detected duplicate child center and gave up." << llendl; return false; @@ -362,7 +375,7 @@ public: else { //it's not in here, give it to the root - //OCT_ERRS << "Octree insertion failed, starting over from root!" << llendl; + OCT_ERRS << "Octree insertion failed, starting over from root!" << llendl; oct_node* node = this; @@ -436,6 +449,9 @@ public: void clearChildren() { mChild.clear(); + + U32* foo = (U32*) mChildMap; + foo[0] = foo[1] = 0xFFFFFFFF; } void validate() @@ -469,13 +485,19 @@ public: void addChild(oct_node* child, BOOL silent = FALSE) { #if LL_OCTREE_PARANOIA_CHECK + + if (child->getSize().equals3(getSize())) + { + OCT_ERRS << "Child size is same as parent size!" << llendl; + } + for (U32 i = 0; i < getChildCount(); i++) { - if(mChild[i]->getSize() != child->getSize()) + if(!mChild[i]->getSize().equals3(child->getSize())) { OCT_ERRS <<"Invalid octree child size." << llendl; } - if (mChild[i]->getCenter() == child->getCenter()) + if (mChild[i]->getCenter().equals3(child->getCenter())) { OCT_ERRS <<"Duplicate octree child position." << llendl; } @@ -487,6 +509,8 @@ public: } #endif + mChildMap[child->getOctant()] = (U8) mChild.size(); + mChild.push_back(child); child->setParent(this); @@ -500,7 +524,7 @@ public: } } - void removeChild(U8 index, BOOL destroy = FALSE) + void removeChild(S32 index, BOOL destroy = FALSE) { for (U32 i = 0; i < this->getListenerCount(); i++) { @@ -508,6 +532,8 @@ public: listener->handleChildRemoval(this, getChild(index)); } + + if (destroy) { mChild[index]->destroy(); @@ -515,6 +541,15 @@ public: } mChild.erase(mChild.begin() + index); + //rebuild child map + U32* foo = (U32*) mChildMap; + foo[0] = foo[1] = 0xFFFFFFFF; + + for (U32 i = 0; i < mChild.size(); ++i) + { + mChildMap[mChild[i]->getOctant()] = i; + } + checkAlive(); } @@ -541,19 +576,32 @@ public: } } - //OCT_ERRS << "Octree failed to delete requested child." << llendl; + OCT_ERRS << "Octree failed to delete requested child." << llendl; } protected: - child_list mChild; - element_list mData; + typedef enum + { + CENTER = 0, + SIZE = 1, + MAX = 2, + MIN = 3 + } eDName; + + LLVector4a mCenter; + LLVector4a mSize; + LLVector4a mMax; + LLVector4a mMin; + oct_node* mParent; - LLVector3d mCenter; - LLVector3d mSize; - LLVector3d mMax; - LLVector3d mMin; U8 mOctant; -}; + + child_list mChild; + U8 mChildMap[8]; + + element_list mData; + +}; //just like a regular node, except it might expand on insert and compress on balance template <class T> @@ -563,9 +611,9 @@ public: typedef LLOctreeNode<T> BaseType; typedef LLOctreeNode<T> oct_node; - LLOctreeRoot( LLVector3d center, - LLVector3d size, - BaseType* parent) + LLOctreeRoot(const LLVector4a& center, + const LLVector4a& size, + BaseType* parent) : BaseType(center, size, parent) { } @@ -596,6 +644,8 @@ public: //destroy child child->clearChildren(); delete child; + + return false; } return true; @@ -606,28 +656,33 @@ public: { if (data == NULL) { - //OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE ROOT !!!" << llendl; + OCT_ERRS << "!!! INVALID ELEMENT ADDED TO OCTREE ROOT !!!" << llendl; return false; } if (data->getBinRadius() > 4096.0) { - //OCT_ERRS << "!!! ELEMENT EXCEEDS MAXIMUM SIZE IN OCTREE ROOT !!!" << llendl; + OCT_ERRS << "!!! ELEMENT EXCEEDS MAXIMUM SIZE IN OCTREE ROOT !!!" << llendl; return false; } - const F64 MAX_MAG = 1024.0*1024.0; + LLVector4a MAX_MAG; + MAX_MAG.splat(1024.f*1024.f); - const LLVector3d& v = data->getPositionGroup(); - if (!(fabs(v.mdV[0]-this->mCenter.mdV[0]) < MAX_MAG && - fabs(v.mdV[1]-this->mCenter.mdV[1]) < MAX_MAG && - fabs(v.mdV[2]-this->mCenter.mdV[2]) < MAX_MAG)) + const LLVector4a& v = data->getPositionGroup(); + + LLVector4a val; + val.setSub(v, BaseType::mCenter); + val.setAbs(val); + S32 lt = val.lessThan(MAX_MAG).getGatheredBits() & 0x7; + + if (lt != 0x7) { - //OCT_ERRS << "!!! ELEMENT EXCEEDS RANGE OF SPATIAL PARTITION !!!" << llendl; + OCT_ERRS << "!!! ELEMENT EXCEEDS RANGE OF SPATIAL PARTITION !!!" << llendl; return false; } - if (this->getSize().mdV[0] > data->getBinRadius() && isInside(data->getPositionGroup())) + if (this->getSize()[0] > data->getBinRadius() && isInside(data->getPositionGroup())) { //we got it, just act like a branch oct_node* node = getNodeAt(data); @@ -643,31 +698,34 @@ public: else if (this->getChildCount() == 0) { //first object being added, just wrap it up - while (!(this->getSize().mdV[0] > data->getBinRadius() && isInside(data->getPositionGroup()))) + while (!(this->getSize()[0] > data->getBinRadius() && isInside(data->getPositionGroup()))) { - LLVector3d center, size; + LLVector4a center, size; center = this->getCenter(); size = this->getSize(); LLOctreeNode<T>::pushCenter(center, size, data); this->setCenter(center); - this->setSize(size*2); + size.mul(2.f); + this->setSize(size); this->updateMinMax(); } LLOctreeNode<T>::insert(data); } else { - while (!(this->getSize().mdV[0] > data->getBinRadius() && isInside(data->getPositionGroup()))) + while (!(this->getSize()[0] > data->getBinRadius() && isInside(data->getPositionGroup()))) { //the data is outside the root node, we need to grow - LLVector3d center(this->getCenter()); - LLVector3d size(this->getSize()); + LLVector4a center(this->getCenter()); + LLVector4a size(this->getSize()); //expand this node - LLVector3d newcenter(center); + LLVector4a newcenter(center); LLOctreeNode<T>::pushCenter(newcenter, size, data); this->setCenter(newcenter); - this->setSize(size*2); + LLVector4a size2 = size; + size2.mul(2.f); + this->setSize(size2); this->updateMinMax(); //copy our children to a new branch @@ -704,4 +762,15 @@ void LLOctreeTraveler<T>::traverse(const LLOctreeNode<T>* node) traverse(node->getChild(i)); } } + +template <class T> +void LLOctreeTravelerDepthFirst<T>::traverse(const LLOctreeNode<T>* node) +{ + for (U32 i = 0; i < node->getChildCount(); i++) + { + traverse(node->getChild(i)); + } + node->accept(this); +} + #endif diff --git a/indra/llmath/llplane.h b/indra/llmath/llplane.h index 443f3f46b9..a611894721 100644 --- a/indra/llmath/llplane.h +++ b/indra/llmath/llplane.h @@ -36,19 +36,23 @@ // The plane normal = [A, B, C] // The closest approach = D / sqrt(A*A + B*B + C*C) -class LLPlane : public LLVector4 +class LLPlane { public: + + // Constructors LLPlane() {}; // no default constructor LLPlane(const LLVector3 &p0, F32 d) { setVec(p0, d); } LLPlane(const LLVector3 &p0, const LLVector3 &n) { setVec(p0, n); } - void setVec(const LLVector3 &p0, F32 d) { LLVector4::setVec(p0[0], p0[1], p0[2], d); } - void setVec(const LLVector3 &p0, const LLVector3 &n) + inline void setVec(const LLVector3 &p0, F32 d) { mV.set(p0[0], p0[1], p0[2], d); } + + // Set + inline void setVec(const LLVector3 &p0, const LLVector3 &n) { F32 d = -(p0 * n); setVec(n, d); } - void setVec(const LLVector3 &p0, const LLVector3 &p1, const LLVector3 &p2) + inline void setVec(const LLVector3 &p0, const LLVector3 &p1, const LLVector3 &p2) { LLVector3 u, v, w; u = p1 - p0; @@ -58,8 +62,38 @@ public: F32 d = -(w * p0); setVec(w, d); } - LLPlane& operator=(const LLVector4& v2) { LLVector4::setVec(v2[0],v2[1],v2[2],v2[3]); return *this;} + + inline LLPlane& operator=(const LLVector4& v2) { mV.set(v2[0],v2[1],v2[2],v2[3]); return *this;} + + inline LLPlane& operator=(const LLVector4a& v2) { mV.set(v2[0],v2[1],v2[2],v2[3]); return *this;} + + inline void set(const LLPlane& p2) { mV = p2.mV; } + + // F32 dist(const LLVector3 &v2) const { return mV[0]*v2[0] + mV[1]*v2[1] + mV[2]*v2[2] + mV[3]; } + + inline LLSimdScalar dot3(const LLVector4a& b) const { return mV.dot3(b); } + + // Read-only access a single float in this vector. Do not use in proximity to any function call that manipulates + // the data at the whole vector level or you will incur a substantial penalty. Consider using the splat functions instead + inline F32 operator[](const S32 idx) const { return mV[idx]; } + + // preferable when index is known at compile time + template <int N> LL_FORCE_INLINE void getAt(LLSimdScalar& v) const { v = mV.getScalarAt<N>(); } + + // reset the vector to 0, 0, 0, 1 + inline void clear() { mV.set(0, 0, 0, 1); } + + inline void getVector3(LLVector3& vec) const { vec.set(mV[0], mV[1], mV[2]); } + + // Retrieve the mask indicating which of the x, y, or z axis are greater or equal to zero. + inline U8 calcPlaneMask() + { + return mV.greaterEqual(LLVector4a::getZero()).getGatheredBits() & LLVector4Logical::MASK_XYZ; + } + +private: + LLVector4a mV; }; diff --git a/indra/llmath/llquantize.h b/indra/llmath/llquantize.h index 7f56ff3448..1595dbecf8 100644 --- a/indra/llmath/llquantize.h +++ b/indra/llmath/llquantize.h @@ -29,10 +29,16 @@ #define LL_LLQUANTIZE_H const U16 U16MAX = 65535; +LL_ALIGN_16( const F32 F_U16MAX_4A[4] ) = { 65535.f, 65535.f, 65535.f, 65535.f }; + const F32 OOU16MAX = 1.f/(F32)(U16MAX); +LL_ALIGN_16( const F32 F_OOU16MAX_4A[4] ) = { OOU16MAX, OOU16MAX, OOU16MAX, OOU16MAX }; const U8 U8MAX = 255; +LL_ALIGN_16( const F32 F_U8MAX_4A[4] ) = { 255.f, 255.f, 255.f, 255.f }; + const F32 OOU8MAX = 1.f/(F32)(U8MAX); +LL_ALIGN_16( const F32 F_OOU8MAX_4A[4] ) = { OOU8MAX, OOU8MAX, OOU8MAX, OOU8MAX }; const U8 FIRSTVALIDCHAR = 54; const U8 MAXSTRINGVAL = U8MAX - FIRSTVALIDCHAR; //we don't allow newline or null diff --git a/indra/llmath/llquaternion.cpp b/indra/llmath/llquaternion.cpp index a51f11072c..7381d5eb99 100644 --- a/indra/llmath/llquaternion.cpp +++ b/indra/llmath/llquaternion.cpp @@ -26,9 +26,10 @@ #include "linden_common.h" +#include "llmath.h" // for F_PI + #include "llquaternion.h" -#include "llmath.h" // for F_PI //#include "vmath.h" #include "v3math.h" #include "v3dmath.h" diff --git a/indra/llmath/llquaternion.h b/indra/llmath/llquaternion.h index 26da14ae20..ca0dfe206b 100644 --- a/indra/llmath/llquaternion.h +++ b/indra/llmath/llquaternion.h @@ -27,7 +27,11 @@ #ifndef LLQUATERNION_H #define LLQUATERNION_H -#include "llmath.h" +#include <iostream> + +#ifndef LLMATH_H //enforce specific include order to avoid tangling inline dependencies +#error "Please include llmath.h first." +#endif class LLVector4; class LLVector3; diff --git a/indra/llmath/llquaternion2.h b/indra/llmath/llquaternion2.h new file mode 100644 index 0000000000..fd9c0cf3ab --- /dev/null +++ b/indra/llmath/llquaternion2.h @@ -0,0 +1,105 @@ +/** + * @file llquaternion2.h + * @brief LLQuaternion2 class header file - SIMD-enabled quaternion class + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_QUATERNION2_H +#define LL_QUATERNION2_H + +///////////////////////////// +// LLQuaternion2 +///////////////////////////// +// This class stores a quaternion x*i + y*j + z*k + w in <x, y, z, w> order +// (i.e., w in high order element of vector) +///////////////////////////// +///////////////////////////// +// These classes are intentionally minimal right now. If you need additional +// functionality, please contact someone with SSE experience (e.g., Falcon or +// Huseby). +///////////////////////////// +#include "llquaternion.h" + +class LLQuaternion2 +{ +public: + + ////////////////////////// + // Ctors + ////////////////////////// + + // Ctor + LLQuaternion2() {} + + // Ctor from LLQuaternion + explicit LLQuaternion2( const class LLQuaternion& quat ); + + ////////////////////////// + // Get/Set + ////////////////////////// + + // Load from an LLQuaternion + inline void operator=( const LLQuaternion& quat ) + { + mQ.loadua( quat.mQ ); + } + + // Return the internal LLVector4a representation of the quaternion + inline const LLVector4a& getVector4a() const; + inline LLVector4a& getVector4aRw(); + + ///////////////////////// + // Quaternion modification + ///////////////////////// + + // Set this quaternion to the conjugate of src + inline void setConjugate(const LLQuaternion2& src); + + // Renormalizes the quaternion. Assumes it has nonzero length. + inline void normalize(); + + // Quantize this quaternion to 8 bit precision + inline void quantize8(); + + // Quantize this quaternion to 16 bit precision + inline void quantize16(); + + ///////////////////////// + // Quaternion inspection + ///////////////////////// + + // Return true if this quaternion is equal to 'rhs'. + // Note! Quaternions exhibit "double-cover", so any rotation has two equally valid + // quaternion representations and they will NOT compare equal. + inline bool equals(const LLQuaternion2& rhs, F32 tolerance = F_APPROXIMATELY_ZERO ) const; + + // Return true if all components are finite and the quaternion is normalized + inline bool isOkRotation() const; + +protected: + + LLVector4a mQ; + +}; + +#endif diff --git a/indra/llmath/llquaternion2.inl b/indra/llmath/llquaternion2.inl new file mode 100644 index 0000000000..2a6987552d --- /dev/null +++ b/indra/llmath/llquaternion2.inl @@ -0,0 +1,102 @@ +/** + * @file llquaternion2.inl + * @brief LLQuaternion2 inline definitions + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#include "llquaternion2.h" + +static const LLQuad LL_V4A_PLUS_ONE = {1.f, 1.f, 1.f, 1.f}; +static const LLQuad LL_V4A_MINUS_ONE = {-1.f, -1.f, -1.f, -1.f}; + +// Ctor from LLQuaternion +inline LLQuaternion2::LLQuaternion2( const LLQuaternion& quat ) +{ + mQ.set(quat.mQ[VX], quat.mQ[VY], quat.mQ[VZ], quat.mQ[VW]); +} + +////////////////////////// +// Get/Set +////////////////////////// + +// Return the internal LLVector4a representation of the quaternion +inline const LLVector4a& LLQuaternion2::getVector4a() const +{ + return mQ; +} + +inline LLVector4a& LLQuaternion2::getVector4aRw() +{ + return mQ; +} + +///////////////////////// +// Quaternion modification +///////////////////////// + +// Set this quaternion to the conjugate of src +inline void LLQuaternion2::setConjugate(const LLQuaternion2& src) +{ + static LL_ALIGN_16( const U32 F_QUAT_INV_MASK_4A[4] ) = { 0x80000000, 0x80000000, 0x80000000, 0x00000000 }; + mQ = _mm_xor_ps(src.mQ, *reinterpret_cast<const LLQuad*>(&F_QUAT_INV_MASK_4A)); +} + +// Renormalizes the quaternion. Assumes it has nonzero length. +inline void LLQuaternion2::normalize() +{ + mQ.normalize4(); +} + +// Quantize this quaternion to 8 bit precision +inline void LLQuaternion2::quantize8() +{ + mQ.quantize8( LL_V4A_MINUS_ONE, LL_V4A_PLUS_ONE ); + normalize(); +} + +// Quantize this quaternion to 16 bit precision +inline void LLQuaternion2::quantize16() +{ + mQ.quantize16( LL_V4A_MINUS_ONE, LL_V4A_PLUS_ONE ); + normalize(); +} + + +///////////////////////// +// Quaternion inspection +///////////////////////// + +// Return true if this quaternion is equal to 'rhs'. +// Note! Quaternions exhibit "double-cover", so any rotation has two equally valid +// quaternion representations and they will NOT compare equal. +inline bool LLQuaternion2::equals(const LLQuaternion2 &rhs, F32 tolerance/* = F_APPROXIMATELY_ZERO*/) const +{ + return mQ.equals4(rhs.mQ, tolerance); +} + +// Return true if all components are finite and the quaternion is normalized +inline bool LLQuaternion2::isOkRotation() const +{ + return mQ.isFinite4() && mQ.isNormalized4(); +} + diff --git a/indra/llmath/llsimdmath.h b/indra/llmath/llsimdmath.h new file mode 100644 index 0000000000..c7cdf7b32c --- /dev/null +++ b/indra/llmath/llsimdmath.h @@ -0,0 +1,93 @@ +/** + * @file llsimdmath.h + * @brief Common header for SIMD-based math library (llvector4a, llmatrix3a, etc.) + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_SIMD_MATH_H +#define LL_SIMD_MATH_H + +#ifndef LLMATH_H +#error "Please include llmath.h before this file." +#endif + +#if ( ( LL_DARWIN || LL_LINUX ) && !(__SSE2__) ) || ( LL_WINDOWS && ( _M_IX86_FP < 2 ) ) +#error SSE2 not enabled. LLVector4a and related class will not compile. +#endif + +#if !LL_WINDOWS +#include <stdint.h> +#endif + +template <typename T> T* LL_NEXT_ALIGNED_ADDRESS(T* address) +{ + return reinterpret_cast<T*>( + (reinterpret_cast<uintptr_t>(address) + 0xF) & ~0xF); +} + +template <typename T> T* LL_NEXT_ALIGNED_ADDRESS_64(T* address) +{ + return reinterpret_cast<T*>( + (reinterpret_cast<uintptr_t>(address) + 0x3F) & ~0x3F); +} + +#if LL_LINUX || LL_DARWIN + +#define LL_ALIGN_PREFIX(x) +#define LL_ALIGN_POSTFIX(x) __attribute__((aligned(x))) + +#elif LL_WINDOWS + +#define LL_ALIGN_PREFIX(x) __declspec(align(x)) +#define LL_ALIGN_POSTFIX(x) + +#else +#error "LL_ALIGN_PREFIX and LL_ALIGN_POSTFIX undefined" +#endif + +#define LL_ALIGN_16(var) LL_ALIGN_PREFIX(16) var LL_ALIGN_POSTFIX(16) + + + +#include <xmmintrin.h> +#include <emmintrin.h> + +#include "llsimdtypes.h" +#include "llsimdtypes.inl" + +class LLMatrix3a; +class LLRotation; +class LLMatrix3; + +#include "llquaternion.h" + +#include "llvector4logical.h" +#include "llvector4a.h" +#include "llmatrix3a.h" +#include "llquaternion2.h" +#include "llvector4a.inl" +#include "llmatrix3a.inl" +#include "llquaternion2.inl" + + +#endif //LL_SIMD_MATH_H diff --git a/indra/llmath/llsimdtypes.h b/indra/llmath/llsimdtypes.h new file mode 100644 index 0000000000..bd991d0e71 --- /dev/null +++ b/indra/llmath/llsimdtypes.h @@ -0,0 +1,124 @@ +/** + * @file llsimdtypes.h + * @brief Declaration of basic SIMD math related types + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_SIMD_TYPES_H +#define LL_SIMD_TYPES_H + +#ifndef LL_SIMD_MATH_H +#error "Please include llmath.h before this file." +#endif + +typedef __m128 LLQuad; + + +#if LL_WINDOWS +#pragma warning(push) +#pragma warning( disable : 4800 3 ) // Disable warning about casting int to bool for this class. +#if defined(_MSC_VER) && (_MSC_VER < 1500) +// VC++ 2005 is missing these intrinsics +// __forceinline is MSVC specific and attempts to override compiler inlining judgment. This is so +// even in debug builds this call is a NOP. +__forceinline const __m128 _mm_castsi128_ps( const __m128i a ) { return reinterpret_cast<const __m128&>(a); } +__forceinline const __m128i _mm_castps_si128( const __m128 a ) { return reinterpret_cast<const __m128i&>(a); } +#endif // _MSC_VER + +#endif // LL_WINDOWS + +class LLBool32 +{ +public: + inline LLBool32() {} + inline LLBool32(int rhs) : m_bool(rhs) {} + inline LLBool32(unsigned int rhs) : m_bool(rhs) {} + inline LLBool32(bool rhs) { m_bool = static_cast<const int>(rhs); } + inline LLBool32& operator= (bool rhs) { m_bool = (int)rhs; return *this; } + inline bool operator== (bool rhs) const { return static_cast<const bool&>(m_bool) == rhs; } + inline bool operator!= (bool rhs) const { return !operator==(rhs); } + inline operator bool() const { return static_cast<const bool&>(m_bool); } + +private: + int m_bool; +}; + +#if LL_WINDOWS +#pragma warning(pop) +#endif + +class LLSimdScalar +{ +public: + inline LLSimdScalar() {} + inline LLSimdScalar(LLQuad q) + { + mQ = q; + } + + inline LLSimdScalar(F32 f) + { + mQ = _mm_set_ss(f); + } + + static inline const LLSimdScalar& getZero() + { + extern const LLQuad F_ZERO_4A; + return reinterpret_cast<const LLSimdScalar&>(F_ZERO_4A); + } + + inline F32 getF32() const; + + inline LLBool32 isApproximatelyEqual(const LLSimdScalar& rhs, F32 tolerance = F_APPROXIMATELY_ZERO) const; + + inline LLSimdScalar getAbs() const; + + inline void setMax( const LLSimdScalar& a, const LLSimdScalar& b ); + + inline void setMin( const LLSimdScalar& a, const LLSimdScalar& b ); + + inline LLSimdScalar& operator=(F32 rhs); + + inline LLSimdScalar& operator+=(const LLSimdScalar& rhs); + + inline LLSimdScalar& operator-=(const LLSimdScalar& rhs); + + inline LLSimdScalar& operator*=(const LLSimdScalar& rhs); + + inline LLSimdScalar& operator/=(const LLSimdScalar& rhs); + + inline operator LLQuad() const + { + return mQ; + } + + inline const LLQuad& getQuad() const + { + return mQ; + } + +private: + LLQuad mQ; +}; + +#endif //LL_SIMD_TYPES_H diff --git a/indra/llmath/llsimdtypes.inl b/indra/llmath/llsimdtypes.inl new file mode 100644 index 0000000000..712239e425 --- /dev/null +++ b/indra/llmath/llsimdtypes.inl @@ -0,0 +1,157 @@ +/** + * @file llsimdtypes.inl + * @brief Inlined definitions of basic SIMD math related types + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + + + + +////////////////// +// LLSimdScalar +////////////////// + +inline LLSimdScalar operator+(const LLSimdScalar& a, const LLSimdScalar& b) +{ + LLSimdScalar t(a); + t += b; + return t; +} + +inline LLSimdScalar operator-(const LLSimdScalar& a, const LLSimdScalar& b) +{ + LLSimdScalar t(a); + t -= b; + return t; +} + +inline LLSimdScalar operator*(const LLSimdScalar& a, const LLSimdScalar& b) +{ + LLSimdScalar t(a); + t *= b; + return t; +} + +inline LLSimdScalar operator/(const LLSimdScalar& a, const LLSimdScalar& b) +{ + LLSimdScalar t(a); + t /= b; + return t; +} + +inline LLSimdScalar operator-(const LLSimdScalar& a) +{ + static LL_ALIGN_16(const U32 signMask[4]) = {0x80000000, 0x80000000, 0x80000000, 0x80000000 }; + return _mm_xor_ps(*reinterpret_cast<const LLQuad*>(signMask), a); +} + +inline LLBool32 operator==(const LLSimdScalar& a, const LLSimdScalar& b) +{ + return _mm_comieq_ss(a, b); +} + +inline LLBool32 operator!=(const LLSimdScalar& a, const LLSimdScalar& b) +{ + return _mm_comineq_ss(a, b); +} + +inline LLBool32 operator<(const LLSimdScalar& a, const LLSimdScalar& b) +{ + return _mm_comilt_ss(a, b); +} + +inline LLBool32 operator<=(const LLSimdScalar& a, const LLSimdScalar& b) +{ + return _mm_comile_ss(a, b); +} + +inline LLBool32 operator>(const LLSimdScalar& a, const LLSimdScalar& b) +{ + return _mm_comigt_ss(a, b); +} + +inline LLBool32 operator>=(const LLSimdScalar& a, const LLSimdScalar& b) +{ + return _mm_comige_ss(a, b); +} + +inline LLBool32 LLSimdScalar::isApproximatelyEqual(const LLSimdScalar& rhs, F32 tolerance /* = F_APPROXIMATELY_ZERO */) const +{ + const LLSimdScalar tol( tolerance ); + const LLSimdScalar diff = _mm_sub_ss( mQ, rhs.mQ ); + const LLSimdScalar absDiff = diff.getAbs(); + return absDiff <= tol; +} + +inline void LLSimdScalar::setMax( const LLSimdScalar& a, const LLSimdScalar& b ) +{ + mQ = _mm_max_ss( a, b ); +} + +inline void LLSimdScalar::setMin( const LLSimdScalar& a, const LLSimdScalar& b ) +{ + mQ = _mm_min_ss( a, b ); +} + +inline LLSimdScalar& LLSimdScalar::operator=(F32 rhs) +{ + mQ = _mm_set_ss(rhs); + return *this; +} + +inline LLSimdScalar& LLSimdScalar::operator+=(const LLSimdScalar& rhs) +{ + mQ = _mm_add_ss( mQ, rhs ); + return *this; +} + +inline LLSimdScalar& LLSimdScalar::operator-=(const LLSimdScalar& rhs) +{ + mQ = _mm_sub_ss( mQ, rhs ); + return *this; +} + +inline LLSimdScalar& LLSimdScalar::operator*=(const LLSimdScalar& rhs) +{ + mQ = _mm_mul_ss( mQ, rhs ); + return *this; +} + +inline LLSimdScalar& LLSimdScalar::operator/=(const LLSimdScalar& rhs) +{ + mQ = _mm_div_ss( mQ, rhs ); + return *this; +} + +inline LLSimdScalar LLSimdScalar::getAbs() const +{ + static const LL_ALIGN_16(U32 F_ABS_MASK_4A[4]) = { 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF }; + return _mm_and_ps( mQ, *reinterpret_cast<const LLQuad*>(F_ABS_MASK_4A)); +} + +inline F32 LLSimdScalar::getF32() const +{ + F32 ret; + _mm_store_ss(&ret, mQ); + return ret; +} diff --git a/indra/llmath/lltreenode.h b/indra/llmath/lltreenode.h index a462d1659e..c66bc26176 100644 --- a/indra/llmath/lltreenode.h +++ b/indra/llmath/lltreenode.h @@ -28,6 +28,9 @@ #include "stdtypes.h" #include "xform.h" +#include "llpointer.h" +#include "llrefcount.h" + #include <vector> template <class T> class LLTreeNode; diff --git a/indra/llmath/llvector4a.cpp b/indra/llmath/llvector4a.cpp new file mode 100644 index 0000000000..b66b7a7076 --- /dev/null +++ b/indra/llmath/llvector4a.cpp @@ -0,0 +1,222 @@ +/** + * @file llvector4a.cpp + * @brief SIMD vector implementation + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#include "llmath.h" +#include "llquantize.h" + +extern const LLQuad F_ZERO_4A = { 0, 0, 0, 0 }; +extern const LLQuad F_APPROXIMATELY_ZERO_4A = { + F_APPROXIMATELY_ZERO, + F_APPROXIMATELY_ZERO, + F_APPROXIMATELY_ZERO, + F_APPROXIMATELY_ZERO +}; + +extern const LLVector4a LL_V4A_ZERO = reinterpret_cast<const LLVector4a&> ( F_ZERO_4A ); +extern const LLVector4a LL_V4A_EPSILON = reinterpret_cast<const LLVector4a&> ( F_APPROXIMATELY_ZERO_4A ); + +/*static */void LLVector4a::memcpyNonAliased16(F32* __restrict dst, const F32* __restrict src, size_t bytes) +{ + assert(src != NULL); + assert(dst != NULL); + assert(bytes > 0); + assert((bytes % sizeof(F32))== 0); + + F32* end = dst + (bytes / sizeof(F32) ); + + if (bytes > 64) + { + F32* begin_64 = LL_NEXT_ALIGNED_ADDRESS_64(dst); + + //at least 64 (16*4) bytes before the end of the destination, switch to 16 byte copies + F32* end_64 = end-16; + + _mm_prefetch((char*)begin_64, _MM_HINT_NTA); + _mm_prefetch((char*)begin_64 + 64, _MM_HINT_NTA); + _mm_prefetch((char*)begin_64 + 128, _MM_HINT_NTA); + _mm_prefetch((char*)begin_64 + 192, _MM_HINT_NTA); + + while (dst < begin_64) + { + copy4a(dst, src); + dst += 4; + src += 4; + } + + while (dst < end_64) + { + _mm_prefetch((char*)src + 512, _MM_HINT_NTA); + _mm_prefetch((char*)dst + 512, _MM_HINT_NTA); + copy4a(dst, src); + copy4a(dst+4, src+4); + copy4a(dst+8, src+8); + copy4a(dst+12, src+12); + + dst += 16; + src += 16; + } + } + + while (dst < end) + { + copy4a(dst, src); + dst += 4; + src += 4; + } +} + +void LLVector4a::setRotated( const LLRotation& rot, const LLVector4a& vec ) +{ + const LLVector4a col0 = rot.getColumn(0); + const LLVector4a col1 = rot.getColumn(1); + const LLVector4a col2 = rot.getColumn(2); + + LLVector4a result = _mm_load_ss( vec.getF32ptr() ); + result.splat<0>( result ); + result.mul( col0 ); + + { + LLVector4a yyyy = _mm_load_ss( vec.getF32ptr() + 1 ); + yyyy.splat<0>( yyyy ); + yyyy.mul( col1 ); + result.add( yyyy ); + } + + { + LLVector4a zzzz = _mm_load_ss( vec.getF32ptr() + 2 ); + zzzz.splat<0>( zzzz ); + zzzz.mul( col2 ); + result.add( zzzz ); + } + + *this = result; +} + +void LLVector4a::setRotated( const LLQuaternion2& quat, const LLVector4a& vec ) +{ + const LLVector4a& quatVec = quat.getVector4a(); + LLVector4a temp; temp.setCross3(quatVec, vec); + temp.add( temp ); + + const LLVector4a realPart( quatVec.getScalarAt<3>() ); + LLVector4a tempTimesReal; tempTimesReal.setMul( temp, realPart ); + + mQ = vec; + add( tempTimesReal ); + + LLVector4a imagCrossTemp; imagCrossTemp.setCross3( quatVec, temp ); + add(imagCrossTemp); +} + +void LLVector4a::quantize8( const LLVector4a& low, const LLVector4a& high ) +{ + LLVector4a val(mQ); + LLVector4a delta; delta.setSub( high, low ); + + { + val.clamp(low, high); + val.sub(low); + + // 8-bit quantization means we can do with just 12 bits of reciprocal accuracy + const LLVector4a oneOverDelta = _mm_rcp_ps(delta.mQ); +// { +// static LL_ALIGN_16( const F32 F_TWO_4A[4] ) = { 2.f, 2.f, 2.f, 2.f }; +// LLVector4a two; two.load4a( F_TWO_4A ); +// +// // Here we use _mm_rcp_ps plus one round of newton-raphson +// // We wish to find 'x' such that x = 1/delta +// // As a first approximation, we take x0 = _mm_rcp_ps(delta) +// // Then x1 = 2 * x0 - a * x0^2 or x1 = x0 * ( 2 - a * x0 ) +// // See Intel AP-803 http://ompf.org/!/Intel_application_note_AP-803.pdf +// const LLVector4a recipApprox = _mm_rcp_ps(delta.mQ); +// oneOverDelta.setMul( delta, recipApprox ); +// oneOverDelta.setSub( two, oneOverDelta ); +// oneOverDelta.mul( recipApprox ); +// } + + val.mul(oneOverDelta); + val.mul(*reinterpret_cast<const LLVector4a*>(F_U8MAX_4A)); + } + + val = _mm_cvtepi32_ps(_mm_cvtps_epi32( val.mQ )); + + { + val.mul(*reinterpret_cast<const LLVector4a*>(F_OOU8MAX_4A)); + val.mul(delta); + val.add(low); + } + + { + LLVector4a maxError; maxError.setMul(delta, *reinterpret_cast<const LLVector4a*>(F_OOU8MAX_4A)); + LLVector4a absVal; absVal.setAbs( val ); + setSelectWithMask( absVal.lessThan( maxError ), F_ZERO_4A, val ); + } +} + +void LLVector4a::quantize16( const LLVector4a& low, const LLVector4a& high ) +{ + LLVector4a val(mQ); + LLVector4a delta; delta.setSub( high, low ); + + { + val.clamp(low, high); + val.sub(low); + + // 16-bit quantization means we need a round of Newton-Raphson + LLVector4a oneOverDelta; + { + static LL_ALIGN_16( const F32 F_TWO_4A[4] ) = { 2.f, 2.f, 2.f, 2.f }; + LLVector4a two; two.load4a( F_TWO_4A ); + + // Here we use _mm_rcp_ps plus one round of newton-raphson + // We wish to find 'x' such that x = 1/delta + // As a first approximation, we take x0 = _mm_rcp_ps(delta) + // Then x1 = 2 * x0 - a * x0^2 or x1 = x0 * ( 2 - a * x0 ) + // See Intel AP-803 http://ompf.org/!/Intel_application_note_AP-803.pdf + const LLVector4a recipApprox = _mm_rcp_ps(delta.mQ); + oneOverDelta.setMul( delta, recipApprox ); + oneOverDelta.setSub( two, oneOverDelta ); + oneOverDelta.mul( recipApprox ); + } + + val.mul(oneOverDelta); + val.mul(*reinterpret_cast<const LLVector4a*>(F_U16MAX_4A)); + } + + val = _mm_cvtepi32_ps(_mm_cvtps_epi32( val.mQ )); + + { + val.mul(*reinterpret_cast<const LLVector4a*>(F_OOU16MAX_4A)); + val.mul(delta); + val.add(low); + } + + { + LLVector4a maxError; maxError.setMul(delta, *reinterpret_cast<const LLVector4a*>(F_OOU16MAX_4A)); + LLVector4a absVal; absVal.setAbs( val ); + setSelectWithMask( absVal.lessThan( maxError ), F_ZERO_4A, val ); + } +} diff --git a/indra/llmath/llvector4a.h b/indra/llmath/llvector4a.h new file mode 100644 index 0000000000..596082509d --- /dev/null +++ b/indra/llmath/llvector4a.h @@ -0,0 +1,324 @@ +/** + * @file llvector4a.h + * @brief LLVector4a class header file - memory aligned and vectorized 4 component vector + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_LLVECTOR4A_H +#define LL_LLVECTOR4A_H + + +class LLRotation; + +#include <assert.h> +#include "llpreprocessor.h" + +/////////////////////////////////// +// FIRST TIME USERS PLEASE READ +////////////////////////////////// +// This is just the beginning of LLVector4a. There are many more useful functions +// yet to be implemented. For example, setNeg to negate a vector, rotate() to apply +// a matrix rotation, various functions to manipulate only the X, Y, and Z elements +// and many others (including a whole variety of accessors). So if you don't see a +// function here that you need, please contact Falcon or someone else with SSE +// experience (Richard, I think, has some and davep has a little as of the time +// of this writing, July 08, 2010) about getting it implemented before you resort to +// LLVector3/LLVector4. +///////////////////////////////// + +class LLVector4a +{ +public: + + /////////////////////////////////// + // STATIC METHODS + /////////////////////////////////// + + // Call initClass() at startup to avoid 15,000+ cycle penalties from denormalized numbers + static void initClass() + { + _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON); + _MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST); + } + + // Return a vector of all zeros + static inline const LLVector4a& getZero() + { + extern const LLVector4a LL_V4A_ZERO; + return LL_V4A_ZERO; + } + + // Return a vector of all epsilon, where epsilon is a small float suitable for approximate equality checks + static inline const LLVector4a& getEpsilon() + { + extern const LLVector4a LL_V4A_EPSILON; + return LL_V4A_EPSILON; + } + + // Copy 16 bytes from src to dst. Source and destination must be 16-byte aligned + static inline void copy4a(F32* dst, const F32* src) + { + _mm_store_ps(dst, _mm_load_ps(src)); + } + + // Copy words 16-byte blocks from src to dst. Source and destination must not overlap. + static void memcpyNonAliased16(F32* __restrict dst, const F32* __restrict src, size_t bytes); + + //////////////////////////////////// + // CONSTRUCTORS + //////////////////////////////////// + + LLVector4a() + { //DO NOT INITIALIZE -- The overhead is completely unnecessary + } + + LLVector4a(F32 x, F32 y, F32 z, F32 w = 0.f) + { + set(x,y,z,w); + } + + LLVector4a(F32 x) + { + splat(x); + } + + LLVector4a(const LLSimdScalar& x) + { + splat(x); + } + + LLVector4a(LLQuad q) + { + mQ = q; + } + + //////////////////////////////////// + // LOAD/STORE + //////////////////////////////////// + + // Load from 16-byte aligned src array (preferred method of loading) + inline void load4a(const F32* src); + + // Load from unaligned src array (NB: Significantly slower than load4a) + inline void loadua(const F32* src); + + // Load only three floats beginning at address 'src'. Slowest method. + inline void load3(const F32* src); + + // Store to a 16-byte aligned memory address + inline void store4a(F32* dst) const; + + //////////////////////////////////// + // BASIC GET/SET + //////////////////////////////////// + + // Return a "this" as an F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) + inline F32* getF32ptr(); + + // Return a "this" as a const F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) + inline const F32* const getF32ptr() const; + + // Read-only access a single float in this vector. Do not use in proximity to any function call that manipulates + // the data at the whole vector level or you will incur a substantial penalty. Consider using the splat functions instead + inline F32 operator[](const S32 idx) const; + + // Prefer this method for read-only access to a single element. Prefer the templated version if the elem is known at compile time. + inline LLSimdScalar getScalarAt(const S32 idx) const; + + // Prefer this method for read-only access to a single element. Prefer the templated version if the elem is known at compile time. + template <int N> LL_FORCE_INLINE LLSimdScalar getScalarAt() const; + + // Set to an x, y, z and optional w provided + inline void set(F32 x, F32 y, F32 z, F32 w = 0.f); + + // Set to all zeros. This is preferred to using ::getZero() + inline void clear(); + + // Set all elements to 'x' + inline void splat(const F32 x); + + // Set all elements to 'x' + inline void splat(const LLSimdScalar& x); + + // Set all 4 elements to element N of src, with N known at compile time + template <int N> void splat(const LLVector4a& src); + + // Set all 4 elements to element i of v, with i NOT known at compile time + inline void splat(const LLVector4a& v, U32 i); + + // Select bits from sourceIfTrue and sourceIfFalse according to bits in mask + inline void setSelectWithMask( const LLVector4Logical& mask, const LLVector4a& sourceIfTrue, const LLVector4a& sourceIfFalse ); + + //////////////////////////////////// + // ALGEBRAIC + //////////////////////////////////// + + // Set this to the element-wise (a + b) + inline void setAdd(const LLVector4a& a, const LLVector4a& b); + + // Set this to element-wise (a - b) + inline void setSub(const LLVector4a& a, const LLVector4a& b); + + // Set this to element-wise multiply (a * b) + inline void setMul(const LLVector4a& a, const LLVector4a& b); + + // Set this to element-wise quotient (a / b) + inline void setDiv(const LLVector4a& a, const LLVector4a& b); + + // Set this to the element-wise absolute value of src + inline void setAbs(const LLVector4a& src); + + // Add to each component in this vector the corresponding component in rhs + inline void add(const LLVector4a& rhs); + + // Subtract from each component in this vector the corresponding component in rhs + inline void sub(const LLVector4a& rhs); + + // Multiply each component in this vector by the corresponding component in rhs + inline void mul(const LLVector4a& rhs); + + // Divide each component in this vector by the corresponding component in rhs + inline void div(const LLVector4a& rhs); + + // Multiply this vector by x in a scalar fashion + inline void mul(const F32 x); + + // Set this to (a x b) (geometric cross-product) + inline void setCross3(const LLVector4a& a, const LLVector4a& b); + + // Set all elements to the dot product of the x, y, and z elements in a and b + inline void setAllDot3(const LLVector4a& a, const LLVector4a& b); + + // Set all elements to the dot product of the x, y, z, and w elements in a and b + inline void setAllDot4(const LLVector4a& a, const LLVector4a& b); + + // Return the 3D dot product of this vector and b + inline LLSimdScalar dot3(const LLVector4a& b) const; + + // Return the 4D dot product of this vector and b + inline LLSimdScalar dot4(const LLVector4a& b) const; + + // Normalize this vector with respect to the x, y, and z components only. Accurate to 22 bites of precision. W component is destroyed + // Note that this does not consider zero length vectors! + inline void normalize3(); + + // Same as normalize3() but with respect to all 4 components + inline void normalize4(); + + // Same as normalize3(), but returns length as a SIMD scalar + inline LLSimdScalar normalize3withLength(); + + // Normalize this vector with respect to the x, y, and z components only. Accurate only to 10-12 bits of precision. W component is destroyed + // Note that this does not consider zero length vectors! + inline void normalize3fast(); + + // Return true if this vector is normalized with respect to x,y,z up to tolerance + inline LLBool32 isNormalized3( F32 tolerance = 1e-3 ) const; + + // Return true if this vector is normalized with respect to all components up to tolerance + inline LLBool32 isNormalized4( F32 tolerance = 1e-3 ) const; + + // Set all elements to the length of vector 'v' + inline void setAllLength3( const LLVector4a& v ); + + // Get this vector's length + inline LLSimdScalar getLength3() const; + + // Set the components of this vector to the minimum of the corresponding components of lhs and rhs + inline void setMin(const LLVector4a& lhs, const LLVector4a& rhs); + + // Set the components of this vector to the maximum of the corresponding components of lhs and rhs + inline void setMax(const LLVector4a& lhs, const LLVector4a& rhs); + + // Clamps this vector to be within the component-wise range low to high (inclusive) + inline void clamp( const LLVector4a& low, const LLVector4a& high ); + + // Set this to (c * lhs) + rhs * ( 1 - c) + inline void setLerp(const LLVector4a& lhs, const LLVector4a& rhs, F32 c); + + // Return true (nonzero) if x, y, z (and w for Finite4) are all finite floats + inline LLBool32 isFinite3() const; + inline LLBool32 isFinite4() const; + + // Set this vector to 'vec' rotated by the LLRotation or LLQuaternion2 provided + void setRotated( const LLRotation& rot, const LLVector4a& vec ); + void setRotated( const class LLQuaternion2& quat, const LLVector4a& vec ); + + // Set this vector to 'vec' rotated by the INVERSE of the LLRotation or LLQuaternion2 provided + inline void setRotatedInv( const LLRotation& rot, const LLVector4a& vec ); + inline void setRotatedInv( const class LLQuaternion2& quat, const LLVector4a& vec ); + + // Quantize this vector to 8 or 16 bit precision + void quantize8( const LLVector4a& low, const LLVector4a& high ); + void quantize16( const LLVector4a& low, const LLVector4a& high ); + + //////////////////////////////////// + // LOGICAL + //////////////////////////////////// + // The functions in this section will compare the elements in this vector + // to those in rhs and return an LLVector4Logical with all bits set in elements + // where the comparison was true and all bits unset in elements where the comparison + // was false. See llvector4logica.h + //////////////////////////////////// + // WARNING: Other than equals3 and equals4, these functions do NOT account + // for floating point tolerance. You should include the appropriate tolerance + // in the inputs. + //////////////////////////////////// + + inline LLVector4Logical greaterThan(const LLVector4a& rhs) const; + + inline LLVector4Logical lessThan(const LLVector4a& rhs) const; + + inline LLVector4Logical greaterEqual(const LLVector4a& rhs) const; + + inline LLVector4Logical lessEqual(const LLVector4a& rhs) const; + + inline LLVector4Logical equal(const LLVector4a& rhs) const; + + // Returns true if this and rhs are componentwise equal up to the specified absolute tolerance + inline bool equals4(const LLVector4a& rhs, F32 tolerance = F_APPROXIMATELY_ZERO ) const; + + inline bool equals3(const LLVector4a& rhs, F32 tolerance = F_APPROXIMATELY_ZERO ) const; + + //////////////////////////////////// + // OPERATORS + //////////////////////////////////// + + // Do NOT add aditional operators without consulting someone with SSE experience + inline const LLVector4a& operator= ( const LLVector4a& rhs ); + + inline const LLVector4a& operator= ( const LLQuad& rhs ); + + inline operator LLQuad() const; + +private: + LLQuad mQ; +}; + +inline void update_min_max(LLVector4a& min, LLVector4a& max, const LLVector4a& p) +{ + min.setMin(min, p); + max.setMax(max, p); +} + +#endif diff --git a/indra/llmath/llvector4a.inl b/indra/llmath/llvector4a.inl new file mode 100644 index 0000000000..7ad22a5631 --- /dev/null +++ b/indra/llmath/llvector4a.inl @@ -0,0 +1,593 @@ +/** + * @file llvector4a.inl + * @brief LLVector4a inline function implementations + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +//////////////////////////////////// +// LOAD/STORE +//////////////////////////////////// + +// Load from 16-byte aligned src array (preferred method of loading) +inline void LLVector4a::load4a(const F32* src) +{ + mQ = _mm_load_ps(src); +} + +// Load from unaligned src array (NB: Significantly slower than load4a) +inline void LLVector4a::loadua(const F32* src) +{ + mQ = _mm_loadu_ps(src); +} + +// Load only three floats beginning at address 'src'. Slowest method. +inline void LLVector4a::load3(const F32* src) +{ + // mQ = { 0.f, src[2], src[1], src[0] } = { W, Z, Y, X } + // NB: This differs from the convention of { Z, Y, X, W } + mQ = _mm_set_ps(0.f, src[2], src[1], src[0]); +} + +// Store to a 16-byte aligned memory address +inline void LLVector4a::store4a(F32* dst) const +{ + _mm_store_ps(dst, mQ); +} + +//////////////////////////////////// +// BASIC GET/SET +//////////////////////////////////// + +// Return a "this" as an F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) +F32* LLVector4a::getF32ptr() +{ + return (F32*) &mQ; +} + +// Return a "this" as a const F32 pointer. Do not use unless you have a very good reason. (Not sure? Ask Falcon) +const F32* const LLVector4a::getF32ptr() const +{ + return (const F32* const) &mQ; +} + +// Read-only access a single float in this vector. Do not use in proximity to any function call that manipulates +// the data at the whole vector level or you will incur a substantial penalty. Consider using the splat functions instead +inline F32 LLVector4a::operator[](const S32 idx) const +{ + return ((F32*)&mQ)[idx]; +} + +// Prefer this method for read-only access to a single element. Prefer the templated version if the elem is known at compile time. +inline LLSimdScalar LLVector4a::getScalarAt(const S32 idx) const +{ + // Return appropriate LLQuad. It will be cast to LLSimdScalar automatically (should be effectively a nop) + switch (idx) + { + case 0: + return mQ; + case 1: + return _mm_shuffle_ps(mQ, mQ, _MM_SHUFFLE(1, 1, 1, 1)); + case 2: + return _mm_shuffle_ps(mQ, mQ, _MM_SHUFFLE(2, 2, 2, 2)); + case 3: + default: + return _mm_shuffle_ps(mQ, mQ, _MM_SHUFFLE(3, 3, 3, 3)); + } +} + +// Prefer this method for read-only access to a single element. Prefer the templated version if the elem is known at compile time. +template <int N> LL_FORCE_INLINE LLSimdScalar LLVector4a::getScalarAt() const +{ + return _mm_shuffle_ps(mQ, mQ, _MM_SHUFFLE(N, N, N, N)); +} + +template<> LL_FORCE_INLINE LLSimdScalar LLVector4a::getScalarAt<0>() const +{ + return mQ; +} + +// Set to an x, y, z and optional w provided +inline void LLVector4a::set(F32 x, F32 y, F32 z, F32 w) +{ + mQ = _mm_set_ps(w, z, y, x); +} + +// Set to all zeros +inline void LLVector4a::clear() +{ + mQ = LLVector4a::getZero().mQ; +} + +inline void LLVector4a::splat(const F32 x) +{ + mQ = _mm_set1_ps(x); +} + +inline void LLVector4a::splat(const LLSimdScalar& x) +{ + mQ = _mm_shuffle_ps( x.getQuad(), x.getQuad(), _MM_SHUFFLE(0,0,0,0) ); +} + +// Set all 4 elements to element N of src, with N known at compile time +template <int N> void LLVector4a::splat(const LLVector4a& src) +{ + mQ = _mm_shuffle_ps(src.mQ, src.mQ, _MM_SHUFFLE(N, N, N, N) ); +} + +// Set all 4 elements to element i of v, with i NOT known at compile time +inline void LLVector4a::splat(const LLVector4a& v, U32 i) +{ + switch (i) + { + case 0: + mQ = _mm_shuffle_ps(v.mQ, v.mQ, _MM_SHUFFLE(0, 0, 0, 0)); + break; + case 1: + mQ = _mm_shuffle_ps(v.mQ, v.mQ, _MM_SHUFFLE(1, 1, 1, 1)); + break; + case 2: + mQ = _mm_shuffle_ps(v.mQ, v.mQ, _MM_SHUFFLE(2, 2, 2, 2)); + break; + case 3: + mQ = _mm_shuffle_ps(v.mQ, v.mQ, _MM_SHUFFLE(3, 3, 3, 3)); + break; + } +} + +// Select bits from sourceIfTrue and sourceIfFalse according to bits in mask +inline void LLVector4a::setSelectWithMask( const LLVector4Logical& mask, const LLVector4a& sourceIfTrue, const LLVector4a& sourceIfFalse ) +{ + // ((( sourceIfTrue ^ sourceIfFalse ) & mask) ^ sourceIfFalse ) + // E.g., sourceIfFalse = 1010b, sourceIfTrue = 0101b, mask = 1100b + // (sourceIfTrue ^ sourceIfFalse) = 1111b --> & mask = 1100b --> ^ sourceIfFalse = 0110b, + // as expected (01 from sourceIfTrue, 10 from sourceIfFalse) + // Courtesy of Mark++, http://markplusplus.wordpress.com/2007/03/14/fast-sse-select-operation/ + mQ = _mm_xor_ps( sourceIfFalse, _mm_and_ps( mask, _mm_xor_ps( sourceIfTrue, sourceIfFalse ) ) ); +} + +//////////////////////////////////// +// ALGEBRAIC +//////////////////////////////////// + +// Set this to the element-wise (a + b) +inline void LLVector4a::setAdd(const LLVector4a& a, const LLVector4a& b) +{ + mQ = _mm_add_ps(a.mQ, b.mQ); +} + +// Set this to element-wise (a - b) +inline void LLVector4a::setSub(const LLVector4a& a, const LLVector4a& b) +{ + mQ = _mm_sub_ps(a.mQ, b.mQ); +} + +// Set this to element-wise multiply (a * b) +inline void LLVector4a::setMul(const LLVector4a& a, const LLVector4a& b) +{ + mQ = _mm_mul_ps(a.mQ, b.mQ); +} + +// Set this to element-wise quotient (a / b) +inline void LLVector4a::setDiv(const LLVector4a& a, const LLVector4a& b) +{ + mQ = _mm_div_ps( a.mQ, b.mQ ); +} + +// Set this to the element-wise absolute value of src +inline void LLVector4a::setAbs(const LLVector4a& src) +{ + static const LL_ALIGN_16(U32 F_ABS_MASK_4A[4]) = { 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF, 0x7FFFFFFF }; + mQ = _mm_and_ps(src.mQ, *reinterpret_cast<const LLQuad*>(F_ABS_MASK_4A)); +} + +// Add to each component in this vector the corresponding component in rhs +inline void LLVector4a::add(const LLVector4a& rhs) +{ + mQ = _mm_add_ps(mQ, rhs.mQ); +} + +// Subtract from each component in this vector the corresponding component in rhs +inline void LLVector4a::sub(const LLVector4a& rhs) +{ + mQ = _mm_sub_ps(mQ, rhs.mQ); +} + +// Multiply each component in this vector by the corresponding component in rhs +inline void LLVector4a::mul(const LLVector4a& rhs) +{ + mQ = _mm_mul_ps(mQ, rhs.mQ); +} + +// Divide each component in this vector by the corresponding component in rhs +inline void LLVector4a::div(const LLVector4a& rhs) +{ + // TODO: Check accuracy, maybe add divFast + mQ = _mm_div_ps(mQ, rhs.mQ); +} + +// Multiply this vector by x in a scalar fashion +inline void LLVector4a::mul(const F32 x) +{ + LLVector4a t; + t.splat(x); + + mQ = _mm_mul_ps(mQ, t.mQ); +} + +// Set this to (a x b) (geometric cross-product) +inline void LLVector4a::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] } + const LLQuad vector1 = _mm_shuffle_ps( a.mQ, a.mQ, _MM_SHUFFLE( 3, 0, 2, 1 )); + // Set vector2 = { b[W], b[Y], b[X], b[Z] } + const LLQuad vector2 = _mm_shuffle_ps( b.mQ, b.mQ, _MM_SHUFFLE( 3, 1, 0, 2 )); + // mQ = { a[W]*b[W], a[X]*b[Y], a[Z]*b[X], a[Y]*b[Z] } + mQ = _mm_mul_ps( vector1, vector2 ); + // vector3 = { a[W], a[Y], a[X], a[Z] } + const LLQuad vector3 = _mm_shuffle_ps( a.mQ, a.mQ, _MM_SHUFFLE( 3, 1, 0, 2 )); + // vector4 = { b[W], b[X], b[Z], b[Y] } + const LLQuad vector4 = _mm_shuffle_ps( b.mQ, b.mQ, _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] } + mQ = _mm_sub_ps( mQ, _mm_mul_ps( vector3, vector4 )); +} + +/* This function works, but may be slightly slower than the one below on older machines + inline void LLVector4a::setAllDot3(const LLVector4a& a, const LLVector4a& b) + { + // ab = { a[W]*b[W], a[Z]*b[Z], a[Y]*b[Y], a[X]*b[X] } + const LLQuad ab = _mm_mul_ps( a.mQ, b.mQ ); + // yzxw = { a[W]*b[W], a[Z]*b[Z], a[X]*b[X], a[Y]*b[Y] } + const LLQuad wzxy = _mm_shuffle_ps( ab, ab, _MM_SHUFFLE(3, 2, 0, 1 )); + // xPlusY = { 2*a[W]*b[W], 2 * a[Z] * b[Z], a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y] } + const LLQuad xPlusY = _mm_add_ps(ab, wzxy); + // xPlusYSplat = { a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y], a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y] } + const LLQuad xPlusYSplat = _mm_movelh_ps(xPlusY, xPlusY); + // zSplat = { a[Z]*b[Z], a[Z]*b[Z], a[Z]*b[Z], a[Z]*b[Z] } + const LLQuad zSplat = _mm_shuffle_ps( ab, ab, _MM_SHUFFLE( 2, 2, 2, 2 )); + // mQ = { a[Z] * b[Z] + a[Y] * b[Y] + a[X] * b[X], same, same, same } + mQ = _mm_add_ps(zSplat, xPlusYSplat); + }*/ + +// Set all elements to the dot product of the x, y, and z elements in a and b +inline void LLVector4a::setAllDot3(const LLVector4a& a, const LLVector4a& b) +{ + // ab = { a[W]*b[W], a[Z]*b[Z], a[Y]*b[Y], a[X]*b[X] } + const LLQuad ab = _mm_mul_ps( a.mQ, b.mQ ); + // yzxw = { a[W]*b[W], a[Z]*b[Z], a[X]*b[X], a[Y]*b[Y] } + const __m128i wzxy = _mm_shuffle_epi32(_mm_castps_si128(ab), _MM_SHUFFLE(3, 2, 0, 1 )); + // xPlusY = { 2*a[W]*b[W], 2 * a[Z] * b[Z], a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y] } + const LLQuad xPlusY = _mm_add_ps(ab, _mm_castsi128_ps(wzxy)); + // xPlusYSplat = { a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y], a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y] } + const LLQuad xPlusYSplat = _mm_movelh_ps(xPlusY, xPlusY); + // zSplat = { a[Z]*b[Z], a[Z]*b[Z], a[Z]*b[Z], a[Z]*b[Z] } + const __m128i zSplat = _mm_shuffle_epi32(_mm_castps_si128(ab), _MM_SHUFFLE( 2, 2, 2, 2 )); + // mQ = { a[Z] * b[Z] + a[Y] * b[Y] + a[X] * b[X], same, same, same } + mQ = _mm_add_ps(_mm_castsi128_ps(zSplat), xPlusYSplat); +} + +// Set all elements to the dot product of the x, y, z, and w elements in a and b +inline void LLVector4a::setAllDot4(const LLVector4a& a, const LLVector4a& b) +{ + // ab = { a[W]*b[W], a[Z]*b[Z], a[Y]*b[Y], a[X]*b[X] } + const LLQuad ab = _mm_mul_ps( a.mQ, b.mQ ); + // yzxw = { a[W]*b[W], a[Z]*b[Z], a[X]*b[X], a[Y]*b[Y] } + const __m128i zwxy = _mm_shuffle_epi32(_mm_castps_si128(ab), _MM_SHUFFLE(2, 3, 0, 1 )); + // zPlusWandXplusY = { a[W]*b[W] + a[Z]*b[Z], a[Z] * b[Z] + a[W]*b[W], a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y] } + const LLQuad zPlusWandXplusY = _mm_add_ps(ab, _mm_castsi128_ps(zwxy)); + // xPlusYSplat = { a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y], a[Y]*b[Y] + a[X] * b[X], a[X] * b[X] + a[Y] * b[Y] } + const LLQuad xPlusYSplat = _mm_movelh_ps(zPlusWandXplusY, zPlusWandXplusY); + const LLQuad zPlusWSplat = _mm_movehl_ps(zPlusWandXplusY, zPlusWandXplusY); + + // mQ = { a[W]*b[W] + a[Z] * b[Z] + a[Y] * b[Y] + a[X] * b[X], same, same, same } + mQ = _mm_add_ps(xPlusYSplat, zPlusWSplat); +} + +// Return the 3D dot product of this vector and b +inline LLSimdScalar LLVector4a::dot3(const LLVector4a& b) const +{ + const LLQuad ab = _mm_mul_ps( mQ, b.mQ ); + const LLQuad splatY = _mm_castsi128_ps( _mm_shuffle_epi32( _mm_castps_si128(ab), _MM_SHUFFLE(1, 1, 1, 1) ) ); + const LLQuad splatZ = _mm_castsi128_ps( _mm_shuffle_epi32( _mm_castps_si128(ab), _MM_SHUFFLE(2, 2, 2, 2) ) ); + const LLQuad xPlusY = _mm_add_ps( ab, splatY ); + return _mm_add_ps( xPlusY, splatZ ); +} + +// Return the 4D dot product of this vector and b +inline LLSimdScalar LLVector4a::dot4(const LLVector4a& b) const +{ + // ab = { w, z, y, x } + const LLQuad ab = _mm_mul_ps( mQ, b.mQ ); + // upperProdsInLowerElems = { y, x, y, x } + const LLQuad upperProdsInLowerElems = _mm_movehl_ps( ab, ab ); + // sumOfPairs = { w+y, z+x, 2y, 2x } + const LLQuad sumOfPairs = _mm_add_ps( upperProdsInLowerElems, ab ); + // shuffled = { z+x, z+x, z+x, z+x } + const LLQuad shuffled = _mm_castsi128_ps( _mm_shuffle_epi32( _mm_castps_si128( sumOfPairs ), _MM_SHUFFLE(1, 1, 1, 1) ) ); + return _mm_add_ss( sumOfPairs, shuffled ); +} + +// Normalize this vector with respect to the x, y, and z components only. Accurate to 22 bites of precision. W component is destroyed +// Note that this does not consider zero length vectors! +inline void LLVector4a::normalize3() +{ + // lenSqrd = a dot a + LLVector4a lenSqrd; lenSqrd.setAllDot3( *this, *this ); + // rsqrt = approximate reciprocal square (i.e., { ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2 } + const LLQuad rsqrt = _mm_rsqrt_ps(lenSqrd.mQ); + static const LLQuad half = { 0.5f, 0.5f, 0.5f, 0.5f }; + static const LLQuad three = {3.f, 3.f, 3.f, 3.f }; + // Now we do one round of Newton-Raphson approximation to get full accuracy + // According to the Newton-Raphson method, given a first 'w' for the root of f(x) = 1/x^2 - a (i.e., x = 1/sqrt(a)) + // the next better approximation w[i+1] = w - f(w)/f'(w) = w - (1/w^2 - a)/(-2*w^(-3)) + // w[i+1] = w + 0.5 * (1/w^2 - a) * w^3 = w + 0.5 * (w - a*w^3) = 1.5 * w - 0.5 * a * w^3 + // = 0.5 * w * (3 - a*w^2) + // Our first approx is w = rsqrt. We need out = a * w[i+1] (this is the input vector 'a', not the 'a' from the above formula + // which is actually lenSqrd). So out = a * [0.5*rsqrt * (3 - lenSqrd*rsqrt*rsqrt)] + const LLQuad AtimesRsqrt = _mm_mul_ps( lenSqrd.mQ, rsqrt ); + const LLQuad AtimesRsqrtTimesRsqrt = _mm_mul_ps( AtimesRsqrt, rsqrt ); + const LLQuad threeMinusAtimesRsqrtTimesRsqrt = _mm_sub_ps(three, AtimesRsqrtTimesRsqrt ); + const LLQuad nrApprox = _mm_mul_ps(half, _mm_mul_ps(rsqrt, threeMinusAtimesRsqrtTimesRsqrt)); + mQ = _mm_mul_ps( mQ, nrApprox ); +} + +// Normalize this vector with respect to all components. Accurate to 22 bites of precision. +// Note that this does not consider zero length vectors! +inline void LLVector4a::normalize4() +{ + // lenSqrd = a dot a + LLVector4a lenSqrd; lenSqrd.setAllDot4( *this, *this ); + // rsqrt = approximate reciprocal square (i.e., { ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2 } + const LLQuad rsqrt = _mm_rsqrt_ps(lenSqrd.mQ); + static const LLQuad half = { 0.5f, 0.5f, 0.5f, 0.5f }; + static const LLQuad three = {3.f, 3.f, 3.f, 3.f }; + // Now we do one round of Newton-Raphson approximation to get full accuracy + // According to the Newton-Raphson method, given a first 'w' for the root of f(x) = 1/x^2 - a (i.e., x = 1/sqrt(a)) + // the next better approximation w[i+1] = w - f(w)/f'(w) = w - (1/w^2 - a)/(-2*w^(-3)) + // w[i+1] = w + 0.5 * (1/w^2 - a) * w^3 = w + 0.5 * (w - a*w^3) = 1.5 * w - 0.5 * a * w^3 + // = 0.5 * w * (3 - a*w^2) + // Our first approx is w = rsqrt. We need out = a * w[i+1] (this is the input vector 'a', not the 'a' from the above formula + // which is actually lenSqrd). So out = a * [0.5*rsqrt * (3 - lenSqrd*rsqrt*rsqrt)] + const LLQuad AtimesRsqrt = _mm_mul_ps( lenSqrd.mQ, rsqrt ); + const LLQuad AtimesRsqrtTimesRsqrt = _mm_mul_ps( AtimesRsqrt, rsqrt ); + const LLQuad threeMinusAtimesRsqrtTimesRsqrt = _mm_sub_ps(three, AtimesRsqrtTimesRsqrt ); + const LLQuad nrApprox = _mm_mul_ps(half, _mm_mul_ps(rsqrt, threeMinusAtimesRsqrtTimesRsqrt)); + mQ = _mm_mul_ps( mQ, nrApprox ); +} + +// Normalize this vector with respect to the x, y, and z components only. Accurate to 22 bites of precision. W component is destroyed +// Note that this does not consider zero length vectors! +inline LLSimdScalar LLVector4a::normalize3withLength() +{ + // lenSqrd = a dot a + LLVector4a lenSqrd; lenSqrd.setAllDot3( *this, *this ); + // rsqrt = approximate reciprocal square (i.e., { ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2, ~1/len(a)^2 } + const LLQuad rsqrt = _mm_rsqrt_ps(lenSqrd.mQ); + static const LLQuad half = { 0.5f, 0.5f, 0.5f, 0.5f }; + static const LLQuad three = {3.f, 3.f, 3.f, 3.f }; + // Now we do one round of Newton-Raphson approximation to get full accuracy + // According to the Newton-Raphson method, given a first 'w' for the root of f(x) = 1/x^2 - a (i.e., x = 1/sqrt(a)) + // the next better approximation w[i+1] = w - f(w)/f'(w) = w - (1/w^2 - a)/(-2*w^(-3)) + // w[i+1] = w + 0.5 * (1/w^2 - a) * w^3 = w + 0.5 * (w - a*w^3) = 1.5 * w - 0.5 * a * w^3 + // = 0.5 * w * (3 - a*w^2) + // Our first approx is w = rsqrt. We need out = a * w[i+1] (this is the input vector 'a', not the 'a' from the above formula + // which is actually lenSqrd). So out = a * [0.5*rsqrt * (3 - lenSqrd*rsqrt*rsqrt)] + const LLQuad AtimesRsqrt = _mm_mul_ps( lenSqrd.mQ, rsqrt ); + const LLQuad AtimesRsqrtTimesRsqrt = _mm_mul_ps( AtimesRsqrt, rsqrt ); + const LLQuad threeMinusAtimesRsqrtTimesRsqrt = _mm_sub_ps(three, AtimesRsqrtTimesRsqrt ); + const LLQuad nrApprox = _mm_mul_ps(half, _mm_mul_ps(rsqrt, threeMinusAtimesRsqrtTimesRsqrt)); + mQ = _mm_mul_ps( mQ, nrApprox ); + return _mm_sqrt_ss(lenSqrd); +} + +// Normalize this vector with respect to the x, y, and z components only. Accurate only to 10-12 bits of precision. W component is destroyed +// Note that this does not consider zero length vectors! +inline void LLVector4a::normalize3fast() +{ + LLVector4a lenSqrd; lenSqrd.setAllDot3( *this, *this ); + const LLQuad approxRsqrt = _mm_rsqrt_ps(lenSqrd.mQ); + mQ = _mm_mul_ps( mQ, approxRsqrt ); +} + +// Return true if this vector is normalized with respect to x,y,z up to tolerance +inline LLBool32 LLVector4a::isNormalized3( F32 tolerance ) const +{ + static LL_ALIGN_16(const U32 ones[4]) = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 }; + LLSimdScalar tol = _mm_load_ss( &tolerance ); + tol = _mm_mul_ss( tol, tol ); + LLVector4a lenSquared; lenSquared.setAllDot3( *this, *this ); + lenSquared.sub( *reinterpret_cast<const LLVector4a*>(ones) ); + lenSquared.setAbs(lenSquared); + return _mm_comile_ss( lenSquared, tol ); +} + +// Return true if this vector is normalized with respect to all components up to tolerance +inline LLBool32 LLVector4a::isNormalized4( F32 tolerance ) const +{ + static LL_ALIGN_16(const U32 ones[4]) = { 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000 }; + LLSimdScalar tol = _mm_load_ss( &tolerance ); + tol = _mm_mul_ss( tol, tol ); + LLVector4a lenSquared; lenSquared.setAllDot4( *this, *this ); + lenSquared.sub( *reinterpret_cast<const LLVector4a*>(ones) ); + lenSquared.setAbs(lenSquared); + return _mm_comile_ss( lenSquared, tol ); +} + +// Set all elements to the length of vector 'v' +inline void LLVector4a::setAllLength3( const LLVector4a& v ) +{ + LLVector4a lenSqrd; + lenSqrd.setAllDot3(v, v); + + mQ = _mm_sqrt_ps(lenSqrd.mQ); +} + +// Get this vector's length +inline LLSimdScalar LLVector4a::getLength3() const +{ + return _mm_sqrt_ss( dot3( (const LLVector4a)mQ ) ); +} + +// Set the components of this vector to the minimum of the corresponding components of lhs and rhs +inline void LLVector4a::setMin(const LLVector4a& lhs, const LLVector4a& rhs) +{ + mQ = _mm_min_ps(lhs.mQ, rhs.mQ); +} + +// Set the components of this vector to the maximum of the corresponding components of lhs and rhs +inline void LLVector4a::setMax(const LLVector4a& lhs, const LLVector4a& rhs) +{ + mQ = _mm_max_ps(lhs.mQ, rhs.mQ); +} + +// Set this to (c * lhs) + rhs * ( 1 - c) +inline void LLVector4a::setLerp(const LLVector4a& lhs, const LLVector4a& rhs, F32 c) +{ + LLVector4a a = lhs; + a.mul(c); + + LLVector4a b = rhs; + b.mul(1.f-c); + + setAdd(a, b); +} + +inline LLBool32 LLVector4a::isFinite3() const +{ + static LL_ALIGN_16(const U32 nanOrInfMask[4]) = { 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 }; + const __m128i nanOrInfMaskV = *reinterpret_cast<const __m128i*> (nanOrInfMask); + const __m128i maskResult = _mm_and_si128( _mm_castps_si128(mQ), nanOrInfMaskV ); + const LLVector4Logical equalityCheck = _mm_castsi128_ps(_mm_cmpeq_epi32( maskResult, nanOrInfMaskV )); + return !equalityCheck.areAnySet( LLVector4Logical::MASK_XYZ ); +} + +inline LLBool32 LLVector4a::isFinite4() const +{ + static LL_ALIGN_16(const U32 nanOrInfMask[4]) = { 0x7f800000, 0x7f800000, 0x7f800000, 0x7f800000 }; + const __m128i nanOrInfMaskV = *reinterpret_cast<const __m128i*> (nanOrInfMask); + const __m128i maskResult = _mm_and_si128( _mm_castps_si128(mQ), nanOrInfMaskV ); + const LLVector4Logical equalityCheck = _mm_castsi128_ps(_mm_cmpeq_epi32( maskResult, nanOrInfMaskV )); + return !equalityCheck.areAnySet( LLVector4Logical::MASK_XYZW ); +} + +inline void LLVector4a::setRotatedInv( const LLRotation& rot, const LLVector4a& vec ) +{ + LLRotation inv; inv.setTranspose( rot ); + setRotated( inv, vec ); +} + +inline void LLVector4a::setRotatedInv( const LLQuaternion2& quat, const LLVector4a& vec ) +{ + LLQuaternion2 invRot; invRot.setConjugate( quat ); + setRotated(invRot, vec); +} + +inline void LLVector4a::clamp( const LLVector4a& low, const LLVector4a& high ) +{ + const LLVector4Logical highMask = greaterThan( high ); + const LLVector4Logical lowMask = lessThan( low ); + + setSelectWithMask( highMask, high, *this ); + setSelectWithMask( lowMask, low, *this ); +} + + +//////////////////////////////////// +// LOGICAL +//////////////////////////////////// +// The functions in this section will compare the elements in this vector +// to those in rhs and return an LLVector4Logical with all bits set in elements +// where the comparison was true and all bits unset in elements where the comparison +// was false. See llvector4logica.h +//////////////////////////////////// +// WARNING: Other than equals3 and equals4, these functions do NOT account +// for floating point tolerance. You should include the appropriate tolerance +// in the inputs. +//////////////////////////////////// + +inline LLVector4Logical LLVector4a::greaterThan(const LLVector4a& rhs) const +{ + return _mm_cmpgt_ps(mQ, rhs.mQ); +} + +inline LLVector4Logical LLVector4a::lessThan(const LLVector4a& rhs) const +{ + return _mm_cmplt_ps(mQ, rhs.mQ); +} + +inline LLVector4Logical LLVector4a::greaterEqual(const LLVector4a& rhs) const +{ + return _mm_cmpge_ps(mQ, rhs.mQ); +} + +inline LLVector4Logical LLVector4a::lessEqual(const LLVector4a& rhs) const +{ + return _mm_cmple_ps(mQ, rhs.mQ); +} + +inline LLVector4Logical LLVector4a::equal(const LLVector4a& rhs) const +{ + return _mm_cmpeq_ps(mQ, rhs.mQ); +} + +// Returns true if this and rhs are componentwise equal up to the specified absolute tolerance +inline bool LLVector4a::equals4(const LLVector4a& rhs, F32 tolerance ) const +{ + LLVector4a diff; diff.setSub( *this, rhs ); + diff.setAbs( diff ); + const LLQuad tol = _mm_set1_ps( tolerance ); + const LLQuad cmp = _mm_cmplt_ps( diff, tol ); + return (_mm_movemask_ps( cmp ) & LLVector4Logical::MASK_XYZW) == LLVector4Logical::MASK_XYZW; +} + +inline bool LLVector4a::equals3(const LLVector4a& rhs, F32 tolerance ) const +{ + LLVector4a diff; diff.setSub( *this, rhs ); + diff.setAbs( diff ); + const LLQuad tol = _mm_set1_ps( tolerance ); + const LLQuad t = _mm_cmplt_ps( diff, tol ); + return (_mm_movemask_ps( t ) & LLVector4Logical::MASK_XYZ) == LLVector4Logical::MASK_XYZ; + +} + +//////////////////////////////////// +// OPERATORS +//////////////////////////////////// + +// Do NOT add aditional operators without consulting someone with SSE experience +inline const LLVector4a& LLVector4a::operator= ( const LLVector4a& rhs ) +{ + mQ = rhs.mQ; + return *this; +} + +inline const LLVector4a& LLVector4a::operator= ( const LLQuad& rhs ) +{ + mQ = rhs; + return *this; +} + +inline LLVector4a::operator LLQuad() const +{ + return mQ; +} diff --git a/indra/llmath/llvector4logical.h b/indra/llmath/llvector4logical.h new file mode 100644 index 0000000000..dd66b09d43 --- /dev/null +++ b/indra/llmath/llvector4logical.h @@ -0,0 +1,124 @@ +/** + * @file llvector4logical.h + * @brief LLVector4Logical class header file - Companion class to LLVector4a for logical and bit-twiddling operations + * + * $LicenseInfo:firstyear=2010&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_VECTOR4LOGICAL_H +#define LL_VECTOR4LOGICAL_H + + +//////////////////////////// +// LLVector4Logical +//////////////////////////// +// This class is incomplete. If you need additional functionality, +// for example setting/unsetting particular elements or performing +// other boolean operations, feel free to implement. If you need +// assistance in determining the most optimal implementation, +// contact someone with SSE experience (Falcon, Richard, Davep, e.g.) +//////////////////////////// + +static LL_ALIGN_16(const U32 S_V4LOGICAL_MASK_TABLE[4*4]) = +{ + 0xFFFFFFFF, 0x00000000, 0x00000000, 0x00000000, + 0x00000000, 0xFFFFFFFF, 0x00000000, 0x00000000, + 0x00000000, 0x00000000, 0xFFFFFFFF, 0x00000000, + 0x00000000, 0x00000000, 0x00000000, 0xFFFFFFFF +}; + +class LLVector4Logical +{ +public: + + enum { + MASK_X = 1, + MASK_Y = 1 << 1, + MASK_Z = 1 << 2, + MASK_W = 1 << 3, + MASK_XYZ = MASK_X | MASK_Y | MASK_Z, + MASK_XYZW = MASK_XYZ | MASK_W + }; + + // Empty default ctor + LLVector4Logical() {} + + LLVector4Logical( const LLQuad& quad ) + { + mQ = quad; + } + + // Create and return a mask consisting of the lowest order bit of each element + inline U32 getGatheredBits() const + { + return _mm_movemask_ps(mQ); + }; + + // Invert this mask + inline LLVector4Logical& invert() + { + static const LL_ALIGN_16(U32 allOnes[4]) = { 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF }; + mQ = _mm_andnot_ps( mQ, *(LLQuad*)(allOnes) ); + return *this; + } + + inline LLBool32 areAllSet( U32 mask ) const + { + return ( getGatheredBits() & mask) == mask; + } + + inline LLBool32 areAllSet() const + { + return areAllSet( MASK_XYZW ); + } + + inline LLBool32 areAnySet( U32 mask ) const + { + return getGatheredBits() & mask; + } + + inline LLBool32 areAnySet() const + { + return areAnySet( MASK_XYZW ); + } + + inline operator LLQuad() const + { + return mQ; + } + + inline void clear() + { + mQ = _mm_setzero_ps(); + } + + template<int N> void setElement() + { + mQ = _mm_or_ps( mQ, *reinterpret_cast<const LLQuad*>(S_V4LOGICAL_MASK_TABLE + 4*N) ); + } + +private: + + LLQuad mQ; +}; + +#endif //LL_VECTOR4ALOGICAL_H diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 71b92962fb..2f662b757b 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1,5548 +1,7303 @@ -/** - * @file llvolume.cpp - * - * $LicenseInfo:firstyear=2002&license=viewerlgpl$ - * Second Life Viewer Source Code - * 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. - * - * 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. - * - * 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 - * - * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA - * $/LicenseInfo$ - */ - -#include "linden_common.h" -#include "llmath.h" - -#include <set> - -#include "llerror.h" -#include "llmemtype.h" - -#include "llvolumemgr.h" -#include "v2math.h" -#include "v3math.h" -#include "v4math.h" -#include "m4math.h" -#include "m3math.h" -#include "lldarray.h" -#include "llvolume.h" -#include "llstl.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; -const F32 HOLLOW_MAX = 0.95f; -const F32 HOLLOW_MAX_SQUARE = 0.7f; - -const F32 TWIST_MIN = -1.f; -const F32 TWIST_MAX = 1.f; - -const F32 RATIO_MIN = 0.f; -const F32 RATIO_MAX = 2.f; // Tom Y: Inverted sense here: 0 = top taper, 2 = bottom taper - -const F32 HOLE_X_MIN= 0.05f; -const F32 HOLE_X_MAX= 1.0f; - -const F32 HOLE_Y_MIN= 0.05f; -const F32 HOLE_Y_MAX= 0.5f; - -const F32 SHEAR_MIN = -0.5f; -const F32 SHEAR_MAX = 0.5f; - -const F32 REV_MIN = 1.f; -const F32 REV_MAX = 4.f; - -const F32 TAPER_MIN = -1.f; -const F32 TAPER_MAX = 1.f; - -const F32 SKEW_MIN = -0.95f; -const F32 SKEW_MAX = 0.95f; - -const F32 SCULPT_MIN_AREA = 0.002f; -const S32 SCULPT_MIN_AREA_DETAIL = 1; - -#define GEN_TRI_STRIP 0 - -BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLVector3& pt3, const LLVector3& norm) -{ - LLVector3 test = (pt2-pt1)%(pt3-pt2); - - //answer - if(test * norm < 0) - { - return FALSE; - } - else - { - return TRUE; - } -} - -BOOL LLLineSegmentBoxIntersect(const LLVector3& start, const LLVector3& end, const LLVector3& center, const LLVector3& size) -{ - float fAWdU[3]; - LLVector3 dir; - LLVector3 diff; - - 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; - } - - 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; - - return true; -} - - -// 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) -{ - F32 u, v, t; - - /* find vectors for two edges sharing vert0 */ - LLVector3 edge1 = vert1 - vert0; - - LLVector3 edge2 = 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; - - if (!two_sided) - { - if (det < F_APPROXIMATELY_ZERO) - { - return FALSE; - } - - /* calculate distance from vert0 to ray origin */ - LLVector3 tvec = orig - vert0; - - /* calculate U parameter and test bounds */ - u = tvec * pvec; - - if (u < 0.f || u > det) - { - return FALSE; - } - - /* 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; - } - - /* 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; - } - - else // two sided - { - if (det > -F_APPROXIMATELY_ZERO && det < F_APPROXIMATELY_ZERO) - { - return FALSE; - } - F32 inv_det = 1.0 / det; - - /* 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; - } - - /* 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; - } - - /* calculate t, ray intersects triangle */ - t = (edge2 * qvec) * inv_det; - } - - if (intersection_a != NULL) - *intersection_a = u; - if (intersection_b != NULL) - *intersection_b = v; - if (intersection_t != NULL) - *intersection_t = t; - - - return TRUE; -} - - -//------------------------------------------------------------------- -// statics -//------------------------------------------------------------------- - - -//---------------------------------------------------- - -LLProfile::Face* LLProfile::addCap(S16 faceID) -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - Face *face = vector_append(mFaces, 1); - - face->mIndex = 0; - face->mCount = mTotal; - face->mScaleU= 1.0f; - face->mCap = TRUE; - face->mFaceID = faceID; - return face; -} - -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; - face->mCount = count; - face->mScaleU= scaleU; - - face->mFlat = flat; - face->mCap = FALSE; - face->mFaceID = faceID; - return face; -} - -// 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; - - F32 begin = params.getBegin(); - F32 end = params.getEnd(); - - t_step = 1.0f / sides; - ang_step = 2.0f*F_PI*t_step*ang_scale; - - // 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 - - if (total_sides < 8) - { - scale = tableScale[total_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; - ang = 2.0f*F_PI*(t*ang_scale + offset); - pt1.setVec(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); - - 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); - mProfile.push_back(new_pt); - } - - // 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. - pt1.setVec(cos(ang)*scale,sin(ang)*scale,t); - - if (mProfile.size() > 0) { - LLVector3 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(pt1); - - t += t_step; - ang += ang_step; - } - - t_fraction = (end - (t - t_step))*sides; - - // 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); - - // 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); - - if (mProfile.size() > 0) { - LLVector3 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(new_pt); - } - - // If we're sliced, the profile is open. - if ((end - begin)*ang_scale < 0.99f) - { - if ((end - begin)*ang_scale > 0.5f) - { - mConcave = TRUE; - } - else - { - mConcave = FALSE; - } - mOpen = TRUE; - if (params.getHollow() <= 0) - { - // put center point if not hollow. - mProfile.push_back(LLVector3(0,0,0)); - } - } - else - { - // The profile isn't open. - mOpen = FALSE; - mConcave = FALSE; - } - - 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. -LLProfile::Face* LLProfile::addHole(const LLProfileParams& params, BOOL flat, F32 sides, F32 offset, F32 box_hollow, F32 ang_scale, S32 split) -{ - // Note that addHole will NOT work for non-"circular" profiles, if we ever decide to use them. - - // Total add has number of vertices on outside. - mTotalOut = mTotal; - - // Why is the "bevel" parameter -1? DJS 04/05/02 - genNGon(params, llfloor(sides),offset,-1, ang_scale, split); - - Face *face = addFace(mTotalOut, mTotal-mTotalOut,0,LL_FACE_INNER_SIDE, flat); - - std::vector<LLVector3> pt; - pt.resize(mTotal) ; - - for (S32 i=mTotalOut;i<mTotal;i++) - { - pt[i] = mProfile[i] * box_hollow; - } - - S32 j=mTotal-1; - for (S32 i=mTotalOut;i<mTotal;i++) - { - mProfile[i] = pt[j--]; - } - - for (S32 i=0;i<(S32)mFaces.size();i++) - { - if (mFaces[i].mCap) - { - mFaces[i].mCount *= 2; - } - } - - return face; -} - - - -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; - } - mDirty = FALSE; - - if (detail < MIN_LOD) - { - llinfos << "Generating profile with LOD < MIN_LOD. CLAMPING" << llendl; - detail = MIN_LOD; - } - - mProfile.clear(); - mFaces.clear(); - - // Generate the face data - S32 i; - F32 begin = params.getBegin(); - F32 end = params.getEnd(); - F32 hollow = params.getHollow(); - - // Quick validation to eliminate some server crashes. - if (begin > end - 0.01f) - { - llwarns << "LLProfile::generate() assertion failed (begin >= end)" << llendl; - return FALSE; - } - - S32 face_num = 0; - - switch (params.getCurveType() & LL_PCODE_PROFILE_MASK) - { - case LL_PCODE_PROFILE_SQUARE: - { - genNGon(params, 4,-0.375, 0, 1, split); - if (path_open) - { - addCap (LL_FACE_PATH_BEGIN); - } - - for (i = llfloor(begin * 4.f); i < llfloor(end * 4.f + .999f); i++) - { - addFace((face_num++) * (split +1), split+2, 1, LL_FACE_OUTER_SIDE_0 << i, TRUE); - } - - for (i = 0; i <(S32) mProfile.size(); i++) - { - // Scale by 4 to generate proper tex coords. - mProfile[i].mV[2] *= 4.f; - } - - if (hollow) - { - switch (params.getCurveType() & LL_PCODE_HOLE_MASK) - { - case LL_PCODE_HOLE_TRIANGLE: - // This offset is not correct, but we can't change it now... DK 11/17/04 - addHole(params, TRUE, 3, -0.375f, hollow, 1.f, split); - break; - case LL_PCODE_HOLE_CIRCLE: - // TODO: Compute actual detail levels for cubes - addHole(params, FALSE, MIN_DETAIL_FACES * detail, -0.375f, hollow, 1.f); - break; - case LL_PCODE_HOLE_SAME: - case LL_PCODE_HOLE_SQUARE: - default: - addHole(params, TRUE, 4, -0.375f, hollow, 1.f, split); - break; - } - } - - if (path_open) { - mFaces[0].mCount = mTotal; - } - } - break; - case LL_PCODE_PROFILE_ISOTRI: - case LL_PCODE_PROFILE_RIGHTTRI: - case LL_PCODE_PROFILE_EQUALTRI: - { - genNGon(params, 3,0, 0, 1, split); - for (i = 0; i <(S32) mProfile.size(); i++) - { - // Scale by 3 to generate proper tex coords. - mProfile[i].mV[2] *= 3.f; - } - - if (path_open) - { - addCap(LL_FACE_PATH_BEGIN); - } - - for (i = llfloor(begin * 3.f); i < llfloor(end * 3.f + .999f); i++) - { - addFace((face_num++) * (split +1), split+2, 1, LL_FACE_OUTER_SIDE_0 << i, TRUE); - } - if (hollow) - { - // Swept triangles need smaller hollowness values, - // because the triangle doesn't fill the bounding box. - F32 triangle_hollow = hollow / 2.f; - - switch (params.getCurveType() & LL_PCODE_HOLE_MASK) - { - case LL_PCODE_HOLE_CIRCLE: - // TODO: Actually generate level of detail for triangles - addHole(params, FALSE, MIN_DETAIL_FACES * detail, 0, triangle_hollow, 1.f); - break; - case LL_PCODE_HOLE_SQUARE: - addHole(params, TRUE, 4, 0, triangle_hollow, 1.f, split); - break; - case LL_PCODE_HOLE_SAME: - case LL_PCODE_HOLE_TRIANGLE: - default: - addHole(params, TRUE, 3, 0, triangle_hollow, 1.f, split); - break; - } - } - } - 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; - - genNGon(params, sides); - - if (path_open) - { - addCap (LL_FACE_PATH_BEGIN); - } - - if (mOpen && !hollow) - { - addFace(0,mTotal-1,0,LL_FACE_OUTER_SIDE_0, FALSE); - } - else - { - addFace(0,mTotal,0,LL_FACE_OUTER_SIDE_0, FALSE); - } - - if (hollow) - { - switch (hole_type) - { - case LL_PCODE_HOLE_SQUARE: - addHole(params, TRUE, 4, 0, hollow, 1.f, split); - break; - case LL_PCODE_HOLE_TRIANGLE: - addHole(params, TRUE, 3, 0, hollow, 1.f, split); - break; - case LL_PCODE_HOLE_CIRCLE: - case LL_PCODE_HOLE_SAME: - default: - addHole(params, FALSE, circle_detail, 0, hollow, 1.f); - break; - } - } - } - 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; - } - } - genNGon(params, llfloor(circle_detail), 0.5f, 0.f, 0.5f); - if (path_open) - { - addCap(LL_FACE_PATH_BEGIN); - } - if (mOpen && !params.getHollow()) - { - addFace(0,mTotal-1,0,LL_FACE_OUTER_SIDE_0, FALSE); - } - else - { - addFace(0,mTotal,0,LL_FACE_OUTER_SIDE_0, FALSE); - } - - if (hollow) - { - switch (hole_type) - { - case LL_PCODE_HOLE_SQUARE: - addHole(params, TRUE, 2, 0.5f, hollow, 0.5f, split); - break; - case LL_PCODE_HOLE_TRIANGLE: - addHole(params, TRUE, 3, 0.5f, hollow, 0.5f, split); - break; - case LL_PCODE_HOLE_CIRCLE: - case LL_PCODE_HOLE_SAME: - default: - addHole(params, FALSE, circle_detail, 0.5f, hollow, 0.5f); - break; - } - } - - // Special case for openness of sphere - if ((params.getEnd() - params.getBegin()) < 1.f) - { - mOpen = TRUE; - } - else if (!hollow) - { - mOpen = FALSE; - mProfile.push_back(mProfile[0]); - mTotal++; - } - } - break; - default: - llerrs << "Unknown profile: getCurveType()=" << params.getCurveType() << llendl; - break; - }; - - if (path_open) - { - addCap(LL_FACE_PATH_END); // bottom - } - - if ( mOpen) // interior edge caps - { - addFace(mTotal-1, 2,0.5,LL_FACE_PROFILE_BEGIN, TRUE); - - if (hollow) - { - addFace(mTotalOut-1, 2,0.5,LL_FACE_PROFILE_END, TRUE); - } - else - { - addFace(mTotal-2, 2,0.5,LL_FACE_PROFILE_END, TRUE); - } - } - - //genNormals(params); - - return TRUE; -} - - - -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 - // changing the sscanf below. - char keyword[256]; /* Flawfinder: ignore */ - char valuestr[256]; /* Flawfinder: ignore */ - keyword[0] = 0; - valuestr[0] = 0; - F32 tempF32; - U32 tempU32; - - while (!feof(fp)) - { - if (fgets(buffer, BUFSIZE, fp) == NULL) - { - buffer[0] = '\0'; - } - - sscanf( /* Flawfinder: ignore */ - buffer, - " %255s %255s", - keyword, valuestr); - if (!strcmp("{", keyword)) - { - continue; - } - if (!strcmp("}",keyword)) - { - break; - } - else if (!strcmp("curve", keyword)) - { - sscanf(valuestr,"%d",&tempU32); - setCurveType((U8) tempU32); - } - else if (!strcmp("begin",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setBegin(tempF32); - } - else if (!strcmp("end",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setEnd(tempF32); - } - else if (!strcmp("hollow",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setHollow(tempF32); - } - else - { - llwarns << "unknown keyword " << keyword << " in profile import" << llendl; - } - } - - return TRUE; -} - - -BOOL LLProfileParams::exportFile(LLFILE *fp) const -{ - fprintf(fp,"\t\tprofile 0\n"); - fprintf(fp,"\t\t{\n"); - fprintf(fp,"\t\t\tcurve\t%d\n", getCurveType()); - fprintf(fp,"\t\t\tbegin\t%g\n", getBegin()); - fprintf(fp,"\t\t\tend\t%g\n", getEnd()); - fprintf(fp,"\t\t\thollow\t%g\n", getHollow()); - fprintf(fp, "\t\t}\n"); - return TRUE; -} - - -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 - // changing the sscanf below. - char keyword[256]; /* Flawfinder: ignore */ - char valuestr[256]; /* Flawfinder: ignore */ - keyword[0] = 0; - valuestr[0] = 0; - F32 tempF32; - U32 tempU32; - - while (input_stream.good()) - { - input_stream.getline(buffer, BUFSIZE); - sscanf( /* Flawfinder: ignore */ - buffer, - " %255s %255s", - keyword, - valuestr); - if (!strcmp("{", keyword)) - { - continue; - } - if (!strcmp("}",keyword)) - { - break; - } - else if (!strcmp("curve", keyword)) - { - sscanf(valuestr,"%d",&tempU32); - setCurveType((U8) tempU32); - } - else if (!strcmp("begin",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setBegin(tempF32); - } - else if (!strcmp("end",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setEnd(tempF32); - } - else if (!strcmp("hollow",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setHollow(tempF32); - } - else - { - llwarns << "unknown keyword " << keyword << " in profile import" << llendl; - } - } - - return TRUE; -} - - -BOOL LLProfileParams::exportLegacyStream(std::ostream& output_stream) const -{ - output_stream <<"\t\tprofile 0\n"; - output_stream <<"\t\t{\n"; - output_stream <<"\t\t\tcurve\t" << (S32) getCurveType() << "\n"; - output_stream <<"\t\t\tbegin\t" << getBegin() << "\n"; - output_stream <<"\t\t\tend\t" << getEnd() << "\n"; - output_stream <<"\t\t\thollow\t" << getHollow() << "\n"; - output_stream << "\t\t}\n"; - return TRUE; -} - -LLSD LLProfileParams::asLLSD() const -{ - LLSD sd; - - sd["curve"] = getCurveType(); - sd["begin"] = getBegin(); - sd["end"] = getEnd(); - sd["hollow"] = getHollow(); - return sd; -} - -bool LLProfileParams::fromLLSD(LLSD& sd) -{ - setCurveType(sd["curve"].asInteger()); - setBegin((F32)sd["begin"].asReal()); - setEnd((F32)sd["end"].asReal()); - setHollow((F32)sd["hollow"].asReal()); - return true; -} - -void LLProfileParams::copyParams(const LLProfileParams ¶ms) -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - setCurveType(params.getCurveType()); - setBegin(params.getBegin()); - setEnd(params.getEnd()); - setHollow(params.getHollow()); -} - - -LLPath::~LLPath() -{ -} - -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. - const F32 tableScale[] = { 1, 1, 1, 0.5f, 0.707107f, 0.53f, 0.525f, 0.5f }; - - F32 revolutions = params.getRevolutions(); - F32 skew = params.getSkew(); - F32 skew_mag = fabs(skew); - F32 hole_x = params.getScaleX() * (1.0f - skew_mag); - F32 hole_y = params.getScaleY(); - - // Calculate taper begin/end for x,y (Negative means taper the beginning) - F32 taper_x_begin = 1.0f; - F32 taper_x_end = 1.0f - params.getTaperX(); - F32 taper_y_begin = 1.0f; - F32 taper_y_end = 1.0f - params.getTaperY(); - - if ( taper_x_end > 1.0f ) - { - // Flip tapering. - taper_x_begin = 2.0f - taper_x_end; - taper_x_end = 1.0f; - } - if ( taper_y_end > 1.0f ) - { - // Flip tapering. - taper_y_begin = 2.0f - taper_y_end; - taper_y_end = 1.0f; - } - - // For spheres, the radius is usually zero. - F32 radius_start = 0.5f; - if (sides < 8) - { - radius_start = tableScale[sides]; - } - - // Scale the radius to take the hole size into account. - radius_start *= 1.0f - hole_y; - - // Now check the radius offset to calculate the start,end radius. (Negative means - // decrease the start radius instead). - F32 radius_end = radius_start; - F32 radius_offset = params.getRadiusOffset(); - if (radius_offset < 0.f) - { - radius_start *= 1.f + radius_offset; - } - else - { - radius_end *= 1.f - radius_offset; - } - - // Is the path NOT a closed loop? - mOpen = ( (params.getEnd()*end_scale - params.getBegin() < 1.0f) || - (skew_mag > 0.001f) || - (fabs(taper_x_end - taper_x_begin) > 0.001f) || - (fabs(taper_y_end - taper_y_begin) > 0.001f) || - (fabs(radius_end - radius_start) > 0.001f) ); - - F32 ang, c, s; - LLQuaternion twist, qang; - PathPt *pt; - LLVector3 path_axis (1.f, 0.f, 0.f); - //LLVector3 twist_axis(0.f, 0.f, 1.f); - F32 twist_begin = params.getTwistBegin() * twist_scale; - F32 twist_end = params.getTwist() * twist_scale; - - // We run through this once before the main loop, to make sure - // the path begins at the correct cut. - F32 step= 1.0f / sides; - F32 t = params.getBegin(); - pt = vector_append(mPath, 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) - + 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->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; - - 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()) - { - pt = vector_append(mPath, 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) - + 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->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; - - t+=step; - } - - // Make one final pass for the end cut. - t = params.getEnd(); - pt = vector_append(mPath, 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) - + 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->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; - - mTotal = mPath.size(); -} - -const LLVector2 LLPathParams::getBeginScale() const -{ - LLVector2 begin_scale(1.f, 1.f); - if (getScaleX() > 1) - { - begin_scale.mV[0] = 2-getScaleX(); - } - if (getScaleY() > 1) - { - begin_scale.mV[1] = 2-getScaleY(); - } - return begin_scale; -} - -const LLVector2 LLPathParams::getEndScale() const -{ - LLVector2 end_scale(1.f, 1.f); - if (getScaleX() < 1) - { - end_scale.mV[0] = getScaleX(); - } - if (getScaleY() < 1) - { - end_scale.mV[1] = getScaleY(); - } - return end_scale; -} - -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; - } - - if (detail < MIN_LOD) - { - llinfos << "Generating path with LOD < MIN! Clamping to 1" << llendl; - detail = MIN_LOD; - } - - mDirty = FALSE; - S32 np = 2; // hardcode for line - - mPath.clear(); - mOpen = TRUE; - - // 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; - if (np < split+2) - { - np = split+2; - } - - mStep = 1.0f / (np-1); - - mPath.resize(np); - - LLVector2 start_scale = params.getBeginScale(); - LLVector2 end_scale = params.getEndScale(); - - 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), - 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); - mPath[i].mTexT = t; - } - } - 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()); - - if (is_sculpted) - sides = sculpt_size; - - genNGon(params, sides); - } - break; - - case LL_PCODE_PATH_CIRCLE2: - { - if (params.getEnd() - params.getBegin() >= 0.99f && - params.getScaleX() >= .99f) - { - mOpen = FALSE; - } - - //genNGon(params, llfloor(MIN_DETAIL_FACES * detail), 4.f, 0.f); - genNGon(params, llfloor(MIN_DETAIL_FACES * detail)); - - F32 t = 0.f; - F32 tStep = 1.0f / mPath.size(); - - F32 toggle = 0.5f; - for (S32 i=0;i<(S32)mPath.size();i++) - { - mPath[i].mPos.mV[0] = toggle; - if (toggle == 0.5f) - toggle = -0.5f; - else - toggle = 0.5f; - t += tStep; - } - } - - break; - - case LL_PCODE_PATH_TEST: - - np = 5; - mStep = 1.0f / (np-1); - - mPath.resize(np); - - for (S32 i=0;i<np;i++) - { - F32 t = (F32)i * mStep; - mPath[i].mPos.setVec(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); - mPath[i].mTexT = t; - mPath[i].mRot.setQuat(F_PI * params.getTwist() * t,1,0,0); - } - - break; - }; - - if (params.getTwist() != params.getTwistBegin()) mOpen = TRUE; - - //if ((int(fabsf(params.getTwist() - params.getTwistBegin())*100))%100 != 0) { - // mOpen = TRUE; - //} - - return TRUE; -} - -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); - 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].mTexT = 0; - } - } - - return TRUE; -} - - -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 - // changing the sscanf below. - char keyword[256]; /* Flawfinder: ignore */ - char valuestr[256]; /* Flawfinder: ignore */ - keyword[0] = 0; - valuestr[0] = 0; - - F32 tempF32; - F32 x, y; - U32 tempU32; - - while (!feof(fp)) - { - if (fgets(buffer, BUFSIZE, fp) == NULL) - { - buffer[0] = '\0'; - } - - sscanf( /* Flawfinder: ignore */ - buffer, - " %255s %255s", - keyword, valuestr); - if (!strcmp("{", keyword)) - { - continue; - } - if (!strcmp("}",keyword)) - { - break; - } - else if (!strcmp("curve", keyword)) - { - sscanf(valuestr,"%d",&tempU32); - setCurveType((U8) tempU32); - } - else if (!strcmp("begin",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setBegin(tempF32); - } - else if (!strcmp("end",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setEnd(tempF32); - } - else if (!strcmp("scale",keyword)) - { - // Legacy for one dimensional scale per path - sscanf(valuestr,"%g",&tempF32); - setScale(tempF32, tempF32); - } - else if (!strcmp("scale_x", keyword)) - { - sscanf(valuestr, "%g", &x); - setScaleX(x); - } - else if (!strcmp("scale_y", keyword)) - { - sscanf(valuestr, "%g", &y); - setScaleY(y); - } - else if (!strcmp("shear_x", keyword)) - { - sscanf(valuestr, "%g", &x); - setShearX(x); - } - else if (!strcmp("shear_y", keyword)) - { - sscanf(valuestr, "%g", &y); - setShearY(y); - } - else if (!strcmp("twist",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setTwist(tempF32); - } - else if (!strcmp("twist_begin", keyword)) - { - sscanf(valuestr, "%g", &y); - setTwistBegin(y); - } - else if (!strcmp("radius_offset", keyword)) - { - sscanf(valuestr, "%g", &y); - setRadiusOffset(y); - } - else if (!strcmp("taper_x", keyword)) - { - sscanf(valuestr, "%g", &y); - setTaperX(y); - } - else if (!strcmp("taper_y", keyword)) - { - sscanf(valuestr, "%g", &y); - setTaperY(y); - } - else if (!strcmp("revolutions", keyword)) - { - sscanf(valuestr, "%g", &y); - setRevolutions(y); - } - else if (!strcmp("skew", keyword)) - { - sscanf(valuestr, "%g", &y); - setSkew(y); - } - else - { - llwarns << "unknown keyword " << " in path import" << llendl; - } - } - return TRUE; -} - - -BOOL LLPathParams::exportFile(LLFILE *fp) const -{ - fprintf(fp, "\t\tpath 0\n"); - fprintf(fp, "\t\t{\n"); - fprintf(fp, "\t\t\tcurve\t%d\n", getCurveType()); - fprintf(fp, "\t\t\tbegin\t%g\n", getBegin()); - fprintf(fp, "\t\t\tend\t%g\n", getEnd()); - fprintf(fp, "\t\t\tscale_x\t%g\n", getScaleX() ); - fprintf(fp, "\t\t\tscale_y\t%g\n", getScaleY() ); - fprintf(fp, "\t\t\tshear_x\t%g\n", getShearX() ); - fprintf(fp, "\t\t\tshear_y\t%g\n", getShearY() ); - fprintf(fp,"\t\t\ttwist\t%g\n", getTwist()); - - fprintf(fp,"\t\t\ttwist_begin\t%g\n", getTwistBegin()); - fprintf(fp,"\t\t\tradius_offset\t%g\n", getRadiusOffset()); - fprintf(fp,"\t\t\ttaper_x\t%g\n", getTaperX()); - fprintf(fp,"\t\t\ttaper_y\t%g\n", getTaperY()); - fprintf(fp,"\t\t\trevolutions\t%g\n", getRevolutions()); - fprintf(fp,"\t\t\tskew\t%g\n", getSkew()); - - fprintf(fp, "\t\t}\n"); - return TRUE; -} - - -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 - // changing the sscanf below. - char keyword[256]; /* Flawfinder: ignore */ - char valuestr[256]; /* Flawfinder: ignore */ - keyword[0] = 0; - valuestr[0] = 0; - - F32 tempF32; - F32 x, y; - U32 tempU32; - - while (input_stream.good()) - { - input_stream.getline(buffer, BUFSIZE); - sscanf( /* Flawfinder: ignore */ - buffer, - " %255s %255s", - keyword, valuestr); - if (!strcmp("{", keyword)) - { - continue; - } - if (!strcmp("}",keyword)) - { - break; - } - else if (!strcmp("curve", keyword)) - { - sscanf(valuestr,"%d",&tempU32); - setCurveType((U8) tempU32); - } - else if (!strcmp("begin",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setBegin(tempF32); - } - else if (!strcmp("end",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setEnd(tempF32); - } - else if (!strcmp("scale",keyword)) - { - // Legacy for one dimensional scale per path - sscanf(valuestr,"%g",&tempF32); - setScale(tempF32, tempF32); - } - else if (!strcmp("scale_x", keyword)) - { - sscanf(valuestr, "%g", &x); - setScaleX(x); - } - else if (!strcmp("scale_y", keyword)) - { - sscanf(valuestr, "%g", &y); - setScaleY(y); - } - else if (!strcmp("shear_x", keyword)) - { - sscanf(valuestr, "%g", &x); - setShearX(x); - } - else if (!strcmp("shear_y", keyword)) - { - sscanf(valuestr, "%g", &y); - setShearY(y); - } - else if (!strcmp("twist",keyword)) - { - sscanf(valuestr,"%g",&tempF32); - setTwist(tempF32); - } - else if (!strcmp("twist_begin", keyword)) - { - sscanf(valuestr, "%g", &y); - setTwistBegin(y); - } - else if (!strcmp("radius_offset", keyword)) - { - sscanf(valuestr, "%g", &y); - setRadiusOffset(y); - } - else if (!strcmp("taper_x", keyword)) - { - sscanf(valuestr, "%g", &y); - setTaperX(y); - } - else if (!strcmp("taper_y", keyword)) - { - sscanf(valuestr, "%g", &y); - setTaperY(y); - } - else if (!strcmp("revolutions", keyword)) - { - sscanf(valuestr, "%g", &y); - setRevolutions(y); - } - else if (!strcmp("skew", keyword)) - { - sscanf(valuestr, "%g", &y); - setSkew(y); - } - else - { - llwarns << "unknown keyword " << " in path import" << llendl; - } - } - return TRUE; -} - - -BOOL LLPathParams::exportLegacyStream(std::ostream& output_stream) const -{ - output_stream << "\t\tpath 0\n"; - output_stream << "\t\t{\n"; - output_stream << "\t\t\tcurve\t" << (S32) getCurveType() << "\n"; - output_stream << "\t\t\tbegin\t" << getBegin() << "\n"; - output_stream << "\t\t\tend\t" << getEnd() << "\n"; - output_stream << "\t\t\tscale_x\t" << getScaleX() << "\n"; - output_stream << "\t\t\tscale_y\t" << getScaleY() << "\n"; - output_stream << "\t\t\tshear_x\t" << getShearX() << "\n"; - output_stream << "\t\t\tshear_y\t" << getShearY() << "\n"; - output_stream <<"\t\t\ttwist\t" << getTwist() << "\n"; - - output_stream <<"\t\t\ttwist_begin\t" << getTwistBegin() << "\n"; - output_stream <<"\t\t\tradius_offset\t" << getRadiusOffset() << "\n"; - output_stream <<"\t\t\ttaper_x\t" << getTaperX() << "\n"; - output_stream <<"\t\t\ttaper_y\t" << getTaperY() << "\n"; - output_stream <<"\t\t\trevolutions\t" << getRevolutions() << "\n"; - output_stream <<"\t\t\tskew\t" << getSkew() << "\n"; - - output_stream << "\t\t}\n"; - return TRUE; -} - -LLSD LLPathParams::asLLSD() const -{ - LLSD sd = LLSD(); - sd["curve"] = getCurveType(); - sd["begin"] = getBegin(); - sd["end"] = getEnd(); - sd["scale_x"] = getScaleX(); - sd["scale_y"] = getScaleY(); - sd["shear_x"] = getShearX(); - sd["shear_y"] = getShearY(); - sd["twist"] = getTwist(); - sd["twist_begin"] = getTwistBegin(); - sd["radius_offset"] = getRadiusOffset(); - sd["taper_x"] = getTaperX(); - sd["taper_y"] = getTaperY(); - sd["revolutions"] = getRevolutions(); - sd["skew"] = getSkew(); - - return sd; -} - -bool LLPathParams::fromLLSD(LLSD& sd) -{ - setCurveType(sd["curve"].asInteger()); - setBegin((F32)sd["begin"].asReal()); - setEnd((F32)sd["end"].asReal()); - setScaleX((F32)sd["scale_x"].asReal()); - setScaleY((F32)sd["scale_y"].asReal()); - setShearX((F32)sd["shear_x"].asReal()); - setShearY((F32)sd["shear_y"].asReal()); - setTwist((F32)sd["twist"].asReal()); - setTwistBegin((F32)sd["twist_begin"].asReal()); - setRadiusOffset((F32)sd["radius_offset"].asReal()); - setTaperX((F32)sd["taper_x"].asReal()); - setTaperY((F32)sd["taper_y"].asReal()); - setRevolutions((F32)sd["revolutions"].asReal()); - setSkew((F32)sd["skew"].asReal()); - return true; -} - -void LLPathParams::copyParams(const LLPathParams ¶ms) -{ - setCurveType(params.getCurveType()); - setBegin(params.getBegin()); - setEnd(params.getEnd()); - setScale(params.getScaleX(), params.getScaleY() ); - setShear(params.getShearX(), params.getShearY() ); - setTwist(params.getTwist()); - setTwistBegin(params.getTwistBegin()); - setRadiusOffset(params.getRadiusOffset()); - setTaper( params.getTaperX(), params.getTaperY() ); - setRevolutions(params.getRevolutions()); - setSkew(params.getSkew()); -} - -S32 profile_delete_lock = 1 ; -LLProfile::~LLProfile() -{ - if(profile_delete_lock) - { - llerrs << "LLProfile should not be deleted here!" << llendl ; - } -} - - -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; - - // set defaults - if (mParams.getPathParams().getCurveType() == LL_PCODE_PATH_FLEXIBLE) - { - mPathp = new LLDynamicPath(); - } - else - { - mPathp = new LLPath(); - } - mProfilep = new LLProfile(); - - mGenerateSingleFace = generate_single_face; - - generate(); - if (mParams.getSculptID().isNull() && params.getSculptType() == LL_SCULPT_TYPE_NONE) - { - createVolumeFaces(); - } -} - -void LLVolume::resizePath(S32 length) -{ - mPathp->resizePath(length); - mVolumeFaces.clear(); -} - -void LLVolume::regen() -{ - generate(); - createVolumeFaces(); -} - -void LLVolume::genBinormals(S32 face) -{ - mVolumeFaces[face].createBinormals(); -} - -LLVolume::~LLVolume() -{ - sNumMeshPoints -= mMesh.size(); - delete mPathp; - - profile_delete_lock = 0 ; - delete mProfilep; - profile_delete_lock = 1 ; - - mPathp = NULL; - mProfilep = NULL; - mVolumeFaces.clear(); -} - -BOOL LLVolume::generate() -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - llassert_always(mProfilep); - - //Added 10.03.05 Dave Parks - // Split is a parameter to LLProfile::generate that tesselates edges on the profile - // to prevent lighting and texture interpolation errors on triangles that are - // stretched due to twisting or scaling on the path. - S32 split = (S32) ((mDetail)*0.66f); - - if (mParams.getPathParams().getCurveType() == LL_PCODE_PATH_LINE && - (mParams.getPathParams().getScale().mV[0] != 1.0f || - mParams.getPathParams().getScale().mV[1] != 1.0f) && - (mParams.getProfileParams().getCurveType() == LL_PCODE_PROFILE_SQUARE || - mParams.getProfileParams().getCurveType() == LL_PCODE_PROFILE_ISOTRI || - mParams.getProfileParams().getCurveType() == LL_PCODE_PROFILE_EQUALTRI || - mParams.getProfileParams().getCurveType() == LL_PCODE_PROFILE_RIGHTTRI)) - { - split = 0; - } - - mLODScaleBias.setVec(0.5f, 0.5f, 0.5f); - - F32 profile_detail = mDetail; - F32 path_detail = mDetail; - - U8 path_type = mParams.getPathParams().getCurveType(); - U8 profile_type = mParams.getProfileParams().getCurveType(); - - if (path_type == LL_PCODE_PATH_LINE && profile_type == LL_PCODE_PROFILE_CIRCLE) - { //cylinders don't care about Z-Axis - mLODScaleBias.setVec(0.6f, 0.6f, 0.0f); - } - else if (path_type == LL_PCODE_PATH_CIRCLE) - { - 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); - - if (regenPath || regenProf ) - { - 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(); - - //generate vertex positions - - // Run along the path. - for (S32 s = 0; s < sizeS; ++s) - { - LLVector2 scale = mPathp->mPath[s].mScale; - LLQuaternion rot = mPathp->mPath[s].mRot; - - // Run along the profile. - for (S32 t = 0; t < sizeT; ++t) - { - 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; - } - } - - for (std::vector<LLProfile::Face>::iterator iter = mProfilep->mFaces.begin(); - iter != mProfilep->mFaces.end(); ++iter) - { - LLFaceID id = iter->mFaceID; - mFaceMask |= id; - } - - return TRUE; - } - return FALSE; -} - - -void LLVolume::createVolumeFaces() -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - if (mGenerateSingleFace) - { - // do nothing - } - else - { - S32 num_faces = getNumFaces(); - BOOL partial_build = TRUE; - if (num_faces != mVolumeFaces.size()) - { - partial_build = FALSE; - mVolumeFaces.resize(num_faces); - } - // Initialize volume faces with parameter data - for (S32 i = 0; i < (S32)mVolumeFaces.size(); i++) - { - LLVolumeFace& vf = mVolumeFaces[i]; - LLProfile::Face& face = mProfilep->mFaces[i]; - vf.mBeginS = face.mIndex; - vf.mNumS = face.mCount; - if (vf.mNumS < 0) - { - llerrs << "Volume face corruption detected." << llendl; - } - - vf.mBeginT = 0; - vf.mNumT= getPath().mPath.size(); - vf.mID = i; - - // Set the type mask bits correctly - if (mParams.getProfileParams().getHollow() > 0) - { - vf.mTypeMask |= LLVolumeFace::HOLLOW_MASK; - } - if (mProfilep->isOpen()) - { - vf.mTypeMask |= LLVolumeFace::OPEN_MASK; - } - if (face.mCap) - { - vf.mTypeMask |= LLVolumeFace::CAP_MASK; - if (face.mFaceID == LL_FACE_PATH_BEGIN) - { - vf.mTypeMask |= LLVolumeFace::TOP_MASK; - } - else - { - llassert(face.mFaceID == LL_FACE_PATH_END); - vf.mTypeMask |= LLVolumeFace::BOTTOM_MASK; - } - } - else if (face.mFaceID & (LL_FACE_PROFILE_BEGIN | LL_FACE_PROFILE_END)) - { - vf.mTypeMask |= LLVolumeFace::FLAT_MASK | LLVolumeFace::END_MASK; - } - else - { - vf.mTypeMask |= LLVolumeFace::SIDE_MASK; - if (face.mFlat) - { - vf.mTypeMask |= LLVolumeFace::FLAT_MASK; - } - if (face.mFaceID & LL_FACE_INNER_SIDE) - { - vf.mTypeMask |= LLVolumeFace::INNER_MASK; - if (face.mFlat && vf.mNumS > 2) - { //flat inner faces have to copy vert normals - vf.mNumS = vf.mNumS*2; - if (vf.mNumS < 0) - { - llerrs << "Volume face corruption detected." << llendl; - } - } - } - else - { - vf.mTypeMask |= LLVolumeFace::OUTER_MASK; - } - } - } - - for (face_list_t::iterator iter = mVolumeFaces.begin(); - iter != mVolumeFaces.end(); ++iter) - { - (*iter).create(this, partial_build); - } - } -} - - -inline LLVector3 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; - - return value; -} - -inline U32 sculpt_xy_to_index(U32 x, U32 y, U16 sculpt_width, U16 sculpt_height, S8 sculpt_components) -{ - U32 index = (x + y * sculpt_width) * sculpt_components; - return index; -} - - -inline U32 sculpt_st_to_index(S32 s, S32 t, S32 size_s, S32 size_t, U16 sculpt_width, U16 sculpt_height, S8 sculpt_components) -{ - U32 x = (U32) ((F32)s/(size_s) * (F32) sculpt_width); - U32 y = (U32) ((F32)t/(size_t) * (F32) sculpt_height); - - return sculpt_xy_to_index(x, y, sculpt_width, sculpt_height, sculpt_components); -} - - -inline LLVector3 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]); - - 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) -{ - 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) -{ - U32 index = sculpt_xy_to_index(x, y, sculpt_width, sculpt_height, sculpt_components); - - return sculpt_index_to_vector(index, sculpt_data); -} - - -F32 LLVolume::sculptGetSurfaceArea() -{ - // test to see if image has enough variation to create non-degenerate geometry - - F32 area = 0; - - S32 sizeS = mPathp->mPath.size(); - S32 sizeT = mProfilep->mProfile.size(); - - for (S32 s = 0; s < sizeS-1; s++) - { - 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; - - // 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; - } - } - - return area; -} - -// create placeholder shape -void LLVolume::sculptGeneratePlaceholder() -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - S32 sizeS = mPathp->mPath.size(); - S32 sizeT = mProfilep->mProfile.size(); - - S32 line = 0; - - // for now, this is a sphere. - for (S32 s = 0; s < sizeS; s++) - { - for (S32 t = 0; t < sizeT; t++) - { - S32 i = t + line; - Point& pt = mMesh[i]; - - - F32 u = (F32)s/(sizeS-1); - F32 v = (F32)t/(sizeT-1); - - 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); - - } - line += sizeT; - } -} - -// create the vertices from the map -void LLVolume::sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, U8 sculpt_type) -{ - U8 sculpt_stitching = sculpt_type & LL_SCULPT_TYPE_MASK; - BOOL sculpt_invert = sculpt_type & LL_SCULPT_FLAG_INVERT; - 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(); - - S32 line = 0; - for (S32 s = 0; s < sizeS; s++) - { - // Run along the profile. - for (S32 t = 0; t < sizeT; t++) - { - S32 i = t + line; - Point& pt = mMesh[i]; - - S32 reversed_t = t; - - if (reverse_horizontal) - { - reversed_t = sizeT - t - 1; - } - - U32 x = (U32) ((F32)reversed_t/(sizeT-1) * (F32) sculpt_width); - U32 y = (U32) ((F32)s/(sizeS-1) * (F32) sculpt_height); - - - if (y == 0) // top row stitching - { - // pinch? - if (sculpt_stitching == LL_SCULPT_TYPE_SPHERE) - { - x = sculpt_width / 2; - } - } - - if (y == sculpt_height) // bottom row stitching - { - // wrap? - if (sculpt_stitching == LL_SCULPT_TYPE_TORUS) - { - y = 0; - } - else - { - y = sculpt_height - 1; - } - - // pinch? - if (sculpt_stitching == LL_SCULPT_TYPE_SPHERE) - { - x = sculpt_width / 2; - } - } - - if (x == sculpt_width) // side stitching - { - // wrap? - if ((sculpt_stitching == LL_SCULPT_TYPE_SPHERE) || - (sculpt_stitching == LL_SCULPT_TYPE_TORUS) || - (sculpt_stitching == LL_SCULPT_TYPE_CYLINDER)) - { - x = 0; - } - - else - { - x = sculpt_width - 1; - } - } - - pt.mPos = sculpt_xy_to_vector(x, y, sculpt_width, sculpt_height, sculpt_components, sculpt_data); - - if (sculpt_mirror) - { - pt.mPos.mV[VX] *= -1.f; - } - } - - line += sizeT; - } -} - - -const S32 SCULPT_REZ_1 = 6; // changed from 4 to 6 - 6 looks round whereas 4 looks square -const S32 SCULPT_REZ_2 = 8; -const S32 SCULPT_REZ_3 = 16; -const S32 SCULPT_REZ_4 = 32; - -S32 sculpt_sides(F32 detail) -{ - - // detail is usually one of: 1, 1.5, 2.5, 4.0. - - if (detail <= 1.0) - { - return SCULPT_REZ_1; - } - if (detail <= 2.0) - { - return SCULPT_REZ_2; - } - if (detail <= 3.0) - { - return SCULPT_REZ_3; - } - else - { - return SCULPT_REZ_4; - } -} - - - -// determine the number of vertices in both s and t direction for this sculpt -void sculpt_calc_mesh_resolution(U16 width, U16 height, U8 type, F32 detail, S32& s, S32& t) -{ - // this code has the following properties: - // 1) the aspect ratio of the mesh is as close as possible to the ratio of the map - // while still using all available verts - // 2) the mesh cannot have more verts than is allowed by LOD - // 3) the mesh cannot have more verts than is allowed by the map - - S32 max_vertices_lod = (S32)pow((double)sculpt_sides(detail), 2.0); - S32 max_vertices_map = width * height / 4; - - S32 vertices; - if (max_vertices_map > 0) - vertices = llmin(max_vertices_lod, max_vertices_map); - else - vertices = max_vertices_lod; - - - F32 ratio; - if ((width == 0) || (height == 0)) - ratio = 1.f; - else - ratio = (F32) width / (F32) height; - - - s = (S32)fsqrtf(((F32)vertices / ratio)); - - s = llmax(s, 4); // no degenerate sizes, please - t = vertices / s; - - t = llmax(t, 4); // no degenerate sizes, please - s = vertices / t; -} - -// 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; - - if (sculpt_width == 0 || sculpt_height == 0 || sculpt_components < 3 || sculpt_data == NULL) - { - sculpt_level = -1; - data_is_empty = TRUE; - } - - S32 requested_sizeS = 0; - S32 requested_sizeT = 0; - - sculpt_calc_mesh_resolution(sculpt_width, sculpt_height, sculpt_type, mDetail, requested_sizeS, requested_sizeT); - - mPathp->generate(mParams.getPathParams(), mDetail, 0, TRUE, requested_sizeS); - mProfilep->generate(mParams.getProfileParams(), mPathp->isOpen(), mDetail, 0, TRUE, requested_sizeT); - - S32 sizeS = mPathp->mPath.size(); // we requested a specific size, now see what we really got - S32 sizeT = mProfilep->mProfile.size(); // we requested a specific size, now see what we really got - - // weird crash bug - DEV-11158 - trying to collect more data: - if ((sizeS == 0) || (sizeT == 0)) - { - llwarns << "sculpt bad mesh size " << sizeS << " " << sizeT << llendl; - } - - sNumMeshPoints -= mMesh.size(); - mMesh.resize(sizeS * sizeT); - sNumMeshPoints += mMesh.size(); - - //generate vertex positions - if (!data_is_empty) - { - sculptGenerateMapVertices(sculpt_width, sculpt_height, sculpt_components, sculpt_data, sculpt_type); - - // don't test lowest LOD to support legacy content DEV-33670 - if (mDetail > SCULPT_MIN_AREA_DETAIL) - { - if (sculptGetSurfaceArea() < SCULPT_MIN_AREA) - { - data_is_empty = TRUE; - } - } - } - - if (data_is_empty) - { - sculptGeneratePlaceholder(); - } - - - - for (S32 i = 0; i < (S32)mProfilep->mFaces.size(); i++) - { - mFaceMask |= mProfilep->mFaces[i].mFaceID; - } - - mSculptLevel = sculpt_level; - - // Delete any existing faces so that they get regenerated - mVolumeFaces.clear(); - - createVolumeFaces(); -} - - - - -BOOL LLVolume::isCap(S32 face) -{ - return mProfilep->mFaces[face].mCap; -} - -BOOL LLVolume::isFlat(S32 face) -{ - return mProfilep->mFaces[face].mFlat; -} - - -bool LLVolumeParams::operator==(const LLVolumeParams ¶ms) const -{ - return ( (getPathParams() == params.getPathParams()) && - (getProfileParams() == params.getProfileParams()) && - (mSculptID == params.mSculptID) && - (mSculptType == params.mSculptType) ); -} - -bool LLVolumeParams::operator!=(const LLVolumeParams ¶ms) const -{ - return ( (getPathParams() != params.getPathParams()) || - (getProfileParams() != params.getProfileParams()) || - (mSculptID != params.mSculptID) || - (mSculptType != params.mSculptType) ); -} - -bool LLVolumeParams::operator<(const LLVolumeParams ¶ms) const -{ - if( getPathParams() != params.getPathParams() ) - { - return getPathParams() < params.getPathParams(); - } - - if (getProfileParams() != params.getProfileParams()) - { - return getProfileParams() < params.getProfileParams(); - } - - if (mSculptID != params.mSculptID) - { - return mSculptID < params.mSculptID; - } - - - return mSculptType < params.mSculptType; - - -} - -void LLVolumeParams::copyParams(const LLVolumeParams ¶ms) -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - mProfileParams.copyParams(params.mProfileParams); - mPathParams.copyParams(params.mPathParams); - mSculptID = params.getSculptID(); - mSculptType = params.getSculptType(); -} - -// Less restricitve approx 0 for volumes -const F32 APPROXIMATELY_ZERO = 0.001f; -bool approx_zero( F32 f, F32 tolerance = APPROXIMATELY_ZERO) -{ - return (f >= -tolerance) && (f <= tolerance); -} - -// return true if in range (or nearly so) -static bool limit_range(F32& v, F32 min, F32 max, F32 tolerance = APPROXIMATELY_ZERO) -{ - F32 min_delta = v - min; - if (min_delta < 0.f) - { - v = min; - if (!approx_zero(min_delta, tolerance)) - return false; - } - F32 max_delta = max - v; - if (max_delta < 0.f) - { - v = max; - if (!approx_zero(max_delta, tolerance)) - return false; - } - return true; -} - -bool LLVolumeParams::setBeginAndEndS(const F32 b, const F32 e) -{ - bool valid = true; - - // First, clamp to valid ranges. - F32 begin = b; - valid &= limit_range(begin, 0.f, 1.f - MIN_CUT_DELTA); - - F32 end = e; - if (end >= .0149f && end < MIN_CUT_DELTA) end = MIN_CUT_DELTA; // eliminate warning for common rounding error - valid &= limit_range(end, MIN_CUT_DELTA, 1.f); - - valid &= limit_range(begin, 0.f, end - MIN_CUT_DELTA, .01f); - - // Now set them. - mProfileParams.setBegin(begin); - mProfileParams.setEnd(end); - - return valid; -} - -bool LLVolumeParams::setBeginAndEndT(const F32 b, const F32 e) -{ - bool valid = true; - - // First, clamp to valid ranges. - F32 begin = b; - valid &= limit_range(begin, 0.f, 1.f - MIN_CUT_DELTA); - - F32 end = e; - valid &= limit_range(end, MIN_CUT_DELTA, 1.f); - - valid &= limit_range(begin, 0.f, end - MIN_CUT_DELTA, .01f); - - // Now set them. - mPathParams.setBegin(begin); - mPathParams.setEnd(end); - - return valid; -} - -bool LLVolumeParams::setHollow(const F32 h) -{ - // Validate the hollow based on path and profile. - U8 profile = mProfileParams.getCurveType() & LL_PCODE_PROFILE_MASK; - U8 hole_type = mProfileParams.getCurveType() & LL_PCODE_HOLE_MASK; - - F32 max_hollow = HOLLOW_MAX; - - // Only square holes have trouble. - if (LL_PCODE_HOLE_SQUARE == hole_type) - { - switch(profile) - { - case LL_PCODE_PROFILE_CIRCLE: - case LL_PCODE_PROFILE_CIRCLE_HALF: - case LL_PCODE_PROFILE_EQUALTRI: - max_hollow = HOLLOW_MAX_SQUARE; - } - } - - F32 hollow = h; - bool valid = limit_range(hollow, HOLLOW_MIN, max_hollow); - mProfileParams.setHollow(hollow); - - return valid; -} - -bool LLVolumeParams::setTwistBegin(const F32 b) -{ - F32 twist_begin = b; - bool valid = limit_range(twist_begin, TWIST_MIN, TWIST_MAX); - mPathParams.setTwistBegin(twist_begin); - return valid; -} - -bool LLVolumeParams::setTwistEnd(const F32 e) -{ - F32 twist_end = e; - bool valid = limit_range(twist_end, TWIST_MIN, TWIST_MAX); - mPathParams.setTwistEnd(twist_end); - return valid; -} - -bool LLVolumeParams::setRatio(const F32 x, const F32 y) -{ - F32 min_x = RATIO_MIN; - F32 max_x = RATIO_MAX; - F32 min_y = RATIO_MIN; - F32 max_y = RATIO_MAX; - // If this is a circular path (and not a sphere) then 'ratio' is actually hole size. - U8 path_type = mPathParams.getCurveType(); - U8 profile_type = mProfileParams.getCurveType() & LL_PCODE_PROFILE_MASK; - if ( LL_PCODE_PATH_CIRCLE == path_type && - LL_PCODE_PROFILE_CIRCLE_HALF != profile_type) - { - // Holes are more restricted... - min_x = HOLE_X_MIN; - max_x = HOLE_X_MAX; - min_y = HOLE_Y_MIN; - max_y = HOLE_Y_MAX; - } - - F32 ratio_x = x; - bool valid = limit_range(ratio_x, min_x, max_x); - F32 ratio_y = y; - valid &= limit_range(ratio_y, min_y, max_y); - - mPathParams.setScale(ratio_x, ratio_y); - - return valid; -} - -bool LLVolumeParams::setShear(const F32 x, const F32 y) -{ - F32 shear_x = x; - bool valid = limit_range(shear_x, SHEAR_MIN, SHEAR_MAX); - F32 shear_y = y; - valid &= limit_range(shear_y, SHEAR_MIN, SHEAR_MAX); - mPathParams.setShear(shear_x, shear_y); - return valid; -} - -bool LLVolumeParams::setTaperX(const F32 v) -{ - F32 taper = v; - bool valid = limit_range(taper, TAPER_MIN, TAPER_MAX); - mPathParams.setTaperX(taper); - return valid; -} - -bool LLVolumeParams::setTaperY(const F32 v) -{ - F32 taper = v; - bool valid = limit_range(taper, TAPER_MIN, TAPER_MAX); - mPathParams.setTaperY(taper); - return valid; -} - -bool LLVolumeParams::setRevolutions(const F32 r) -{ - F32 revolutions = r; - bool valid = limit_range(revolutions, REV_MIN, REV_MAX); - mPathParams.setRevolutions(revolutions); - return valid; -} - -bool LLVolumeParams::setRadiusOffset(const F32 offset) -{ - bool valid = true; - - // If this is a sphere, just set it to 0 and get out. - U8 path_type = mPathParams.getCurveType(); - U8 profile_type = mProfileParams.getCurveType() & LL_PCODE_PROFILE_MASK; - if ( LL_PCODE_PROFILE_CIRCLE_HALF == profile_type || - LL_PCODE_PATH_CIRCLE != path_type ) - { - mPathParams.setRadiusOffset(0.f); - return true; - } - - // Limit radius offset, based on taper and hole size y. - F32 radius_offset = offset; - F32 taper_y = getTaperY(); - F32 radius_mag = fabs(radius_offset); - F32 hole_y_mag = fabs(getRatioY()); - F32 taper_y_mag = fabs(taper_y); - // Check to see if the taper effects us. - if ( (radius_offset > 0.f && taper_y < 0.f) || - (radius_offset < 0.f && taper_y > 0.f) ) - { - // The taper does not help increase the radius offset range. - taper_y_mag = 0.f; - } - F32 max_radius_mag = 1.f - hole_y_mag * (1.f - taper_y_mag) / (1.f - hole_y_mag); - - // Enforce the maximum magnitude. - F32 delta = max_radius_mag - radius_mag; - if (delta < 0.f) - { - // Check radius offset sign. - if (radius_offset < 0.f) - { - radius_offset = -max_radius_mag; - } - else - { - radius_offset = max_radius_mag; - } - valid = approx_zero(delta, .1f); - } - - mPathParams.setRadiusOffset(radius_offset); - return valid; -} - -bool LLVolumeParams::setSkew(const F32 skew_value) -{ - bool valid = true; - - // Check the skew value against the revolutions. - F32 skew = llclamp(skew_value, SKEW_MIN, SKEW_MAX); - F32 skew_mag = fabs(skew); - F32 revolutions = getRevolutions(); - F32 scale_x = getRatioX(); - F32 min_skew_mag = 1.0f - 1.0f / (revolutions * scale_x + 1.0f); - // Discontinuity; A revolution of 1 allows skews below 0.5. - if ( fabs(revolutions - 1.0f) < 0.001) - min_skew_mag = 0.0f; - - // Clip skew. - F32 delta = skew_mag - min_skew_mag; - if (delta < 0.f) - { - // Check skew sign. - if (skew < 0.0f) - { - skew = -min_skew_mag; - } - else - { - skew = min_skew_mag; - } - valid = approx_zero(delta, .01f); - } - - mPathParams.setSkew(skew); - return valid; -} - -bool LLVolumeParams::setSculptID(const LLUUID sculpt_id, U8 sculpt_type) -{ - mSculptID = sculpt_id; - mSculptType = sculpt_type; - return true; -} - -bool LLVolumeParams::setType(U8 profile, U8 path) -{ - bool result = true; - // First, check profile and path for validity. - U8 profile_type = profile & LL_PCODE_PROFILE_MASK; - U8 hole_type = (profile & LL_PCODE_HOLE_MASK) >> 4; - U8 path_type = path >> 4; - - if (profile_type > LL_PCODE_PROFILE_MAX) - { - // 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; - } - 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; - } - - if (path_type < LL_PCODE_PATH_MIN || - path_type > LL_PCODE_PATH_MAX) - { - // Bad path. Make it linear. - result = false; - llwarns << "LLVolumeParams::setType changing bad path (" << path - << ") to be LL_PCODE_PATH_LINE" << llendl; - path = LL_PCODE_PATH_LINE; - } - - mProfileParams.setCurveType(profile); - mPathParams.setCurveType(path); - return result; -} - -// static -bool LLVolumeParams::validate(U8 prof_curve, F32 prof_begin, F32 prof_end, F32 hollow, - U8 path_curve, F32 path_begin, F32 path_end, - F32 scx, F32 scy, F32 shx, F32 shy, - F32 twistend, F32 twistbegin, F32 radiusoffset, - F32 tx, F32 ty, F32 revolutions, F32 skew) -{ - LLVolumeParams test_params; - if (!test_params.setType (prof_curve, path_curve)) - { - return false; - } - if (!test_params.setBeginAndEndS (prof_begin, prof_end)) - { - return false; - } - if (!test_params.setBeginAndEndT (path_begin, path_end)) - { - return false; - } - if (!test_params.setHollow (hollow)) - { - return false; - } - if (!test_params.setTwistBegin (twistbegin)) - { - return false; - } - if (!test_params.setTwistEnd (twistend)) - { - return false; - } - if (!test_params.setRatio (scx, scy)) - { - return false; - } - if (!test_params.setShear (shx, shy)) - { - return false; - } - if (!test_params.setTaper (tx, ty)) - { - return false; - } - if (!test_params.setRevolutions (revolutions)) - { - return false; - } - if (!test_params.setRadiusOffset (radiusoffset)) - { - return false; - } - if (!test_params.setSkew (skew)) - { - return false; - } - 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) - { - // 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; - - 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++] = (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; - } - } - } - } - 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; - } - } - - 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 - { - // 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 - - 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; - } - } - } - -#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) - { - 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); - } -#endif - - num_indices = count; - return index; -} - -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, - S32 face_mask) -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - vertices.clear(); - normals.clear(); - segments.clear(); - - S32 cur_index = 0; - //for each face - for (face_list_t::iterator iter = mVolumeFaces.begin(); - iter != mVolumeFaces.end(); ++iter) - { - const LLVolumeFace& face = *iter; - - if (!(face_mask & (0x1 << cur_index++))) - { - continue; - } - if (face.mTypeMask & (LLVolumeFace::CAP_MASK)) { - - } - else { - - //============================================== - //DEBUG draw edge map instead of silhouette edge - //============================================== - -#if DEBUG_SILHOUETTE_EDGE_MAP - - //for each triangle - U32 count = face.mIndices.size(); - for (U32 j = 0; j < count/3; j++) { - //get vertices - S32 v1 = face.mIndices[j*3+0]; - S32 v2 = face.mIndices[j*3+1]; - 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; - - //for each edge - for (S32 k = 0; k < 3; k++) { - S32 nIndex = face.mEdge[j*3+k]; - if (nIndex <= -1) { - continue; - } - - if (nIndex >= (S32) count/3) { - continue; - } - //get neighbor vertices - v1 = face.mIndices[nIndex*3+0]; - v2 = face.mIndices[nIndex*3+1]; - 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; - - //draw line - vertices.push_back(cCenter); - vertices.push_back(nCenter); - normals.push_back(LLVector3(1,1,1)); - normals.push_back(LLVector3(1,1,1)); - segments.push_back(vertices.size()); - } - } - - continue; - - //============================================== - //DEBUG - //============================================== - - //============================================== - //DEBUG draw normals instead of silhouette edge - //============================================== -#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); - 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); - normals.push_back(LLVector3(0,0,1)); - normals.push_back(LLVector3(0,0,1)); - segments.push_back(vertices.size()); -#endif - } - - continue; -#else - //============================================== - //DEBUG - //============================================== - - static const U8 AWAY = 0x01, - TOWARDS = 0x02; - - //for each triangle - std::vector<U8> fFacing; - vector_append(fFacing, face.mIndices.size()/3); - for (U32 j = 0; j < face.mIndices.size()/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) - { - fFacing[j] = AWAY | TOWARDS; - } - else - { - //get view vector - LLVector3 view = (obj_cam_vec-face.mVertices[v1].mPosition); - bool away = view * norm > 0.0f; - if (away) - { - fFacing[j] = AWAY; - } - else - { - fFacing[j] = TOWARDS; - } - } - } - - //for each triangle - for (U32 j = 0; j < face.mIndices.size()/3; j++) - { - if (fFacing[j] == (AWAY | TOWARDS)) - { //this is a degenerate triangle - //take neighbor facing (degenerate faces get facing of one of their neighbors) - // *FIX IF NEEDED: this does not deal with neighboring degenerate faces - for (S32 k = 0; k < 3; k++) - { - S32 index = face.mEdge[j*3+k]; - if (index != -1) - { - fFacing[j] = fFacing[index]; - break; - } - } - continue; //skip degenerate face - } - - //for each edge - for (S32 k = 0; k < 3; k++) { - S32 index = face.mEdge[j*3+k]; - if (index != -1 && fFacing[index] == (AWAY | TOWARDS)) { - //our neighbor is degenerate, make him face our direction - fFacing[face.mEdge[j*3+k]] = fFacing[j]; - continue; - } - - if (index == -1 || //edge has no neighbor, MUST be a silhouette edge - (fFacing[index] & fFacing[j]) == 0) { //we found a silhouette edge - - 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); - - vertices.push_back(face.mVertices[v2].mPosition*mat); - LLVector3 norm2 = face.mVertices[v2].mNormal * norm_mat; - norm2.normVec(); - normals.push_back(norm2); - - segments.push_back(vertices.size()); - } - } - } -#endif - } - } -} - -S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, - S32 face, - LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal) -{ - S32 hit_face = -1; - - S32 start_face; - S32 end_face; - - if (face == -1) // ALL_SIDES - { - start_face = 0; - end_face = getNumVolumeFaces() - 1; - } - else - { - start_face = face; - end_face = face; - } - - LLVector3 dir = end - start; - - F32 closest_t = 2.f; // must be larger than 1 - - for (S32 i = start_face; i <= end_face; i++) - { - const LLVolumeFace &face = getVolumeFace((U32)i); - - LLVector3 box_center = (face.mExtents[0] + face.mExtents[1]) / 2.f; - LLVector3 box_size = 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 - { - genBinormals(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 (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) - { - *normal = ((1.f - a - b) * face.mVertices[index1].mNormal + - a * face.mVertices[index2].mNormal + - b * face.mVertices[index3].mNormal); - } - - if (bi_normal != NULL) - { - *bi_normal = ((1.f - a - b) * face.mVertices[index1].mBinormal + - a * face.mVertices[index2].mBinormal + - b * face.mVertices[index3].mBinormal); - } - - } - } - } - } - } - - - return hit_face; -} - -class LLVertexIndexPair -{ -public: - LLVertexIndexPair(const LLVector3 &vertex, const S32 index); - - LLVector3 mVertex; - S32 mIndex; -}; - -LLVertexIndexPair::LLVertexIndexPair(const LLVector3 &vertex, const S32 index) -{ - mVertex = vertex; - mIndex = index; -} - -const F32 VERTEX_SLOP = 0.00001f; -const F32 VERTEX_SLOP_SQRD = VERTEX_SLOP * VERTEX_SLOP; - -struct lessVertex -{ - bool operator()(const LLVertexIndexPair *a, const LLVertexIndexPair *b) - { - const F32 slop = VERTEX_SLOP; - - if (a->mVertex.mV[0] + slop < b->mVertex.mV[0]) - { - return TRUE; - } - else if (a->mVertex.mV[0] - slop > b->mVertex.mV[0]) - { - return FALSE; - } - - if (a->mVertex.mV[1] + slop < b->mVertex.mV[1]) - { - return TRUE; - } - else if (a->mVertex.mV[1] - slop > b->mVertex.mV[1]) - { - return FALSE; - } - - if (a->mVertex.mV[2] + slop < b->mVertex.mV[2]) - { - return TRUE; - } - else if (a->mVertex.mV[2] - slop > b->mVertex.mV[2]) - { - return FALSE; - } - - return FALSE; - } -}; - -struct lessTriangle -{ - bool operator()(const S32 *a, const S32 *b) - { - if (*a < *b) - { - return TRUE; - } - else if (*a > *b) - { - return FALSE; - } - - if (*(a+1) < *(b+1)) - { - return TRUE; - } - else if (*(a+1) > *(b+1)) - { - return FALSE; - } - - if (*(a+2) < *(b+2)) - { - return TRUE; - } - else if (*(a+2) > *(b+2)) - { - return FALSE; - } - - return FALSE; - } -}; - -BOOL equalTriangle(const S32 *a, const S32 *b) -{ - if ((*a == *b) && (*(a+1) == *(b+1)) && (*(a+2) == *(b+2))) - { - return TRUE; - } - 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; - const S32 BUFSIZE = 16384; - char buffer[BUFSIZE]; /* Flawfinder: ignore */ - // *NOTE: changing the size or type of this buffer will require - // changing the sscanf below. - char keyword[256]; /* Flawfinder: ignore */ - keyword[0] = 0; - - while (!feof(fp)) - { - if (fgets(buffer, BUFSIZE, fp) == NULL) - { - buffer[0] = '\0'; - } - - sscanf(buffer, " %255s", keyword); /* Flawfinder: ignore */ - if (!strcmp("{", keyword)) - { - continue; - } - if (!strcmp("}",keyword)) - { - break; - } - else if (!strcmp("profile", keyword)) - { - mProfileParams.importFile(fp); - } - else if (!strcmp("path",keyword)) - { - mPathParams.importFile(fp); - } - else - { - llwarns << "unknown keyword " << keyword << " in volume import" << llendl; - } - } - - return TRUE; -} - -BOOL LLVolumeParams::exportFile(LLFILE *fp) const -{ - fprintf(fp,"\tshape 0\n"); - fprintf(fp,"\t{\n"); - mPathParams.exportFile(fp); - mProfileParams.exportFile(fp); - fprintf(fp, "\t}\n"); - return TRUE; -} - - -BOOL LLVolumeParams::importLegacyStream(std::istream& input_stream) -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - //llinfos << "importing volume" << llendl; - const S32 BUFSIZE = 16384; - // *NOTE: changing the size or type of this buffer will require - // changing the sscanf below. - char buffer[BUFSIZE]; /* Flawfinder: ignore */ - char keyword[256]; /* Flawfinder: ignore */ - keyword[0] = 0; - - while (input_stream.good()) - { - input_stream.getline(buffer, BUFSIZE); - sscanf(buffer, " %255s", keyword); - if (!strcmp("{", keyword)) - { - continue; - } - if (!strcmp("}",keyword)) - { - break; - } - else if (!strcmp("profile", keyword)) - { - mProfileParams.importLegacyStream(input_stream); - } - else if (!strcmp("path",keyword)) - { - mPathParams.importLegacyStream(input_stream); - } - else - { - llwarns << "unknown keyword " << keyword << " in volume import" << llendl; - } - } - - return TRUE; -} - -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); - mProfileParams.exportLegacyStream(output_stream); - output_stream << "\t}\n"; - return TRUE; -} - -LLSD LLVolumeParams::asLLSD() const -{ - LLSD sd = LLSD(); - sd["path"] = mPathParams; - sd["profile"] = mProfileParams; - return sd; -} - -bool LLVolumeParams::fromLLSD(LLSD& sd) -{ - mPathParams.fromLLSD(sd["path"]); - mProfileParams.fromLLSD(sd["profile"]); - return true; -} - -void LLVolumeParams::reduceS(F32 begin, F32 end) -{ - begin = llclampf(begin); - end = llclampf(end); - if (begin > end) - { - F32 temp = begin; - begin = end; - end = temp; - } - F32 a = mProfileParams.getBegin(); - F32 b = mProfileParams.getEnd(); - mProfileParams.setBegin(a + begin * (b - a)); - mProfileParams.setEnd(a + end * (b - a)); -} - -void LLVolumeParams::reduceT(F32 begin, F32 end) -{ - begin = llclampf(begin); - end = llclampf(end); - if (begin > end) - { - F32 temp = begin; - begin = end; - end = temp; - } - F32 a = mPathParams.getBegin(); - F32 b = mPathParams.getEnd(); - mPathParams.setBegin(a + begin * (b - a)); - mPathParams.setEnd(a + end * (b - a)); -} - -const F32 MIN_CONCAVE_PROFILE_WEDGE = 0.125f; // 1/8 unity -const F32 MIN_CONCAVE_PATH_WEDGE = 0.111111f; // 1/9 unity - -// returns TRUE if the shape can be approximated with a convex shape -// for collison purposes -BOOL LLVolumeParams::isConvex() const -{ - F32 path_length = mPathParams.getEnd() - mPathParams.getBegin(); - F32 hollow = mProfileParams.getHollow(); - - U8 path_type = mPathParams.getCurveType(); - if ( path_length > MIN_CONCAVE_PATH_WEDGE - && ( mPathParams.getTwist() != mPathParams.getTwistBegin() - || (hollow > 0.f - && LL_PCODE_PATH_LINE != path_type) ) ) - { - // twist along a "not too short" path is concave - return FALSE; - } - - F32 profile_length = mProfileParams.getEnd() - mProfileParams.getBegin(); - BOOL same_hole = hollow == 0.f - || (mProfileParams.getCurveType() & LL_PCODE_HOLE_MASK) == LL_PCODE_HOLE_SAME; - - F32 min_profile_wedge = MIN_CONCAVE_PROFILE_WEDGE; - U8 profile_type = mProfileParams.getCurveType() & LL_PCODE_PROFILE_MASK; - if ( LL_PCODE_PROFILE_CIRCLE_HALF == profile_type ) - { - // it is a sphere and spheres get twice the minimum profile wedge - min_profile_wedge = 2.f * MIN_CONCAVE_PROFILE_WEDGE; - } - - BOOL convex_profile = ( ( profile_length == 1.f - || profile_length <= 0.5f ) - && hollow == 0.f ) // trivially convex - || ( profile_length <= min_profile_wedge - && same_hole ); // effectvely convex (even when hollow) - - if (!convex_profile) - { - // profile is concave - return FALSE; - } - - if ( LL_PCODE_PATH_LINE == path_type ) - { - // straight paths with convex profile - return TRUE; - } - - BOOL concave_path = (path_length < 1.0f) && (path_length > 0.5f); - if (concave_path) - { - return FALSE; - } - - // we're left with spheres, toroids and tubes - if ( LL_PCODE_PROFILE_CIRCLE_HALF == profile_type ) - { - // at this stage all spheres must be convex - return TRUE; - } - - // it's a toroid or tube - if ( path_length <= MIN_CONCAVE_PATH_WEDGE ) - { - // effectively convex - return TRUE; - } - - return FALSE; -} - -// debug -void LLVolumeParams::setCube() -{ - mProfileParams.setCurveType(LL_PCODE_PROFILE_SQUARE); - mProfileParams.setBegin(0.f); - mProfileParams.setEnd(1.f); - mProfileParams.setHollow(0.f); - - mPathParams.setBegin(0.f); - mPathParams.setEnd(1.f); - mPathParams.setScale(1.f, 1.f); - mPathParams.setShear(0.f, 0.f); - mPathParams.setCurveType(LL_PCODE_PATH_LINE); - mPathParams.setTwistBegin(0.f); - mPathParams.setTwistEnd(0.f); - mPathParams.setRadiusOffset(0.f); - mPathParams.setTaper(0.f, 0.f); - mPathParams.setRevolutions(0.f); - mPathParams.setSkew(0.f); -} - -LLFaceID LLVolume::generateFaceMask() -{ - LLFaceID new_mask = 0x0000; - - switch(mParams.getProfileParams().getCurveType() & LL_PCODE_PROFILE_MASK) - { - case LL_PCODE_PROFILE_CIRCLE: - case LL_PCODE_PROFILE_CIRCLE_HALF: - new_mask |= LL_FACE_OUTER_SIDE_0; - break; - case LL_PCODE_PROFILE_SQUARE: - { - for(S32 side = (S32)(mParams.getProfileParams().getBegin() * 4.f); side < llceil(mParams.getProfileParams().getEnd() * 4.f); side++) - { - new_mask |= LL_FACE_OUTER_SIDE_0 << side; - } - } - break; - case LL_PCODE_PROFILE_ISOTRI: - case LL_PCODE_PROFILE_EQUALTRI: - case LL_PCODE_PROFILE_RIGHTTRI: - { - for(S32 side = (S32)(mParams.getProfileParams().getBegin() * 3.f); side < llceil(mParams.getProfileParams().getEnd() * 3.f); side++) - { - new_mask |= LL_FACE_OUTER_SIDE_0 << side; - } - } - break; - default: - llerrs << "Unknown profile!" << llendl; - break; - } - - // handle hollow objects - if (mParams.getProfileParams().getHollow() > 0) - { - new_mask |= LL_FACE_INNER_SIDE; - } - - // handle open profile curves - if (mProfilep->isOpen()) - { - new_mask |= LL_FACE_PROFILE_BEGIN | LL_FACE_PROFILE_END; - } - - // handle open path curves - if (mPathp->isOpen()) - { - new_mask |= LL_FACE_PATH_BEGIN | LL_FACE_PATH_END; - } - - return new_mask; -} - -BOOL LLVolume::isFaceMaskValid(LLFaceID face_mask) -{ - LLFaceID test_mask = 0; - for(S32 i = 0; i < getNumFaces(); i++) - { - test_mask |= mProfilep->mFaces[i].mFaceID; - } - - return test_mask == face_mask; -} - -BOOL LLVolume::isConvex() const -{ - // mParams.isConvex() may return FALSE even though the final - // geometry is actually convex due to LOD approximations. - // TODO -- provide LLPath and LLProfile with isConvex() methods - // that correctly determine convexity. -- Leviathan - return mParams.isConvex(); -} - - -std::ostream& operator<<(std::ostream &s, const LLProfileParams &profile_params) -{ - s << "{type=" << (U32) profile_params.mCurveType; - s << ", begin=" << profile_params.mBegin; - s << ", end=" << profile_params.mEnd; - s << ", hollow=" << profile_params.mHollow; - s << "}"; - return s; -} - - -std::ostream& operator<<(std::ostream &s, const LLPathParams &path_params) -{ - s << "{type=" << (U32) path_params.mCurveType; - s << ", begin=" << path_params.mBegin; - s << ", end=" << path_params.mEnd; - s << ", twist=" << path_params.mTwistEnd; - s << ", scale=" << path_params.mScale; - s << ", shear=" << path_params.mShear; - s << ", twist_begin=" << path_params.mTwistBegin; - s << ", radius_offset=" << path_params.mRadiusOffset; - s << ", taper=" << path_params.mTaper; - s << ", revolutions=" << path_params.mRevolutions; - s << ", skew=" << path_params.mSkew; - s << "}"; - return s; -} - - -std::ostream& operator<<(std::ostream &s, const LLVolumeParams &volume_params) -{ - s << "{profileparams = " << volume_params.mProfileParams; - s << ", pathparams = " << volume_params.mPathParams; - s << "}"; - return s; -} - - -std::ostream& operator<<(std::ostream &s, const LLProfile &profile) -{ - s << " {open=" << (U32) profile.mOpen; - s << ", dirty=" << profile.mDirty; - s << ", totalout=" << profile.mTotalOut; - s << ", total=" << profile.mTotal; - s << "}"; - return s; -} - - -std::ostream& operator<<(std::ostream &s, const LLPath &path) -{ - s << "{open=" << (U32) path.mOpen; - s << ", dirty=" << path.mDirty; - s << ", step=" << path.mStep; - s << ", total=" << path.mTotal; - s << "}"; - return s; -} - -std::ostream& operator<<(std::ostream &s, const LLVolume &volume) -{ - s << "{params = " << volume.getParams(); - s << ", path = " << *volume.mPathp; - s << ", profile = " << *volume.mProfilep; - s << "}"; - return s; -} - - -std::ostream& operator<<(std::ostream &s, const LLVolume *volumep) -{ - s << "{params = " << volumep->getParams(); - s << ", path = " << *(volumep->mPathp); - s << ", profile = " << *(volumep->mProfilep); - s << "}"; - return s; -} - - -BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) -{ - BOOL ret = FALSE ; - if (mTypeMask & CAP_MASK) - { - ret = createCap(volume, partial_build); - } - else if ((mTypeMask & END_MASK) || (mTypeMask & SIDE_MASK)) - { - ret = createSide(volume, partial_build); - } - else - { - llerrs << "Unknown/uninitialized face type!" << llendl; - } - - //update the range of the texture coordinates - if(ret) - { - mTexCoordExtents[0].setVec(1.f, 1.f) ; - mTexCoordExtents[1].setVec(0.f, 0.f) ; - - U32 end = mVertices.size() ; - for(U32 i = 0 ; i < end ; i++) - { - if(mTexCoordExtents[0].mV[0] > mVertices[i].mTexCoord.mV[0]) - { - mTexCoordExtents[0].mV[0] = mVertices[i].mTexCoord.mV[0] ; - } - if(mTexCoordExtents[1].mV[0] < mVertices[i].mTexCoord.mV[0]) - { - mTexCoordExtents[1].mV[0] = mVertices[i].mTexCoord.mV[0] ; - } - - if(mTexCoordExtents[0].mV[1] > mVertices[i].mTexCoord.mV[1]) - { - mTexCoordExtents[0].mV[1] = mVertices[i].mTexCoord.mV[1] ; - } - if(mTexCoordExtents[1].mV[1] < mVertices[i].mTexCoord.mV[1]) - { - mTexCoordExtents[1].mV[1] = mVertices[i].mTexCoord.mV[1] ; - } - } - mTexCoordExtents[0].mV[0] = llmax(0.f, mTexCoordExtents[0].mV[0]) ; - mTexCoordExtents[0].mV[1] = llmax(0.f, mTexCoordExtents[0].mV[1]) ; - mTexCoordExtents[1].mV[0] = llmin(1.f, mTexCoordExtents[1].mV[0]) ; - mTexCoordExtents[1].mV[1] = llmin(1.f, mTexCoordExtents[1].mV[1]) ; - } - - return ret ; -} - -void LerpPlanarVertex(LLVolumeFace::VertexData& v0, - LLVolumeFace::VertexData& v1, - LLVolumeFace::VertexData& v2, - LLVolumeFace::VertexData& vout, - F32 coef01, - F32 coef02) -{ - vout.mPosition = v0.mPosition + ((v1.mPosition-v0.mPosition)*coef01)+((v2.mPosition-v0.mPosition)*coef02); - vout.mTexCoord = v0.mTexCoord + ((v1.mTexCoord-v0.mTexCoord)*coef01)+((v2.mTexCoord-v0.mTexCoord)*coef02); - vout.mNormal = v0.mNormal; - vout.mBinormal = v0.mBinormal; -} - -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; - 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]; - - 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(); - } - - 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); - - if (gx == 0 && gy == 0) - { - min = max = newVert.mPosition; - } - else - { - update_min_max(min,max,newVert.mPosition); - } - } - } - - mCenter = (min + max) * 0.5f; - - if (!partial_build) - { -#if GEN_TRI_STRIP - mTriStrip.clear(); -#endif - S32 idxs[] = {0,1,(grid_size+1)+1,(grid_size+1)+1,(grid_size+1),0}; - for(S32 gx = 0;gx<grid_size;gx++) - { - - for(S32 gy = 0;gy<grid_size;gy++) - { - if (mTypeMask & TOP_MASK) - { - for(S32 i=5;i>=0;i--) - { - mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); - } - -#if GEN_TRI_STRIP - 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)); - } -#endif - } - else - { - for(S32 i=0;i<6;i++) - { - mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); - } - -#if GEN_TRI_STRIP - 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)); - } -#endif - } - } - - } - -#if GEN_TRI_STRIP - if (mTriStrip.size()%2 == 1) - { - mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); - } -#endif - } - - 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)&& - (volume->getParams().getPathParams().getEnd()==1.0f))&& - (volume->getParams().getProfileParams().getCurveType()==LL_PCODE_PROFILE_SQUARE && - volume->getParams().getPathParams().getCurveType()==LL_PCODE_PATH_LINE) - ){ - return createUnCutCubeCap(volume, 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; - - // 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) - { - mIndices.resize(num_indices); - } - - S32 max_s = volume->getProfile().getTotal(); - S32 max_t = volume->getPath().mPath.size(); - - mCenter.clearVec(); - - S32 offset = 0; - if (mTypeMask & TOP_MASK) - { - offset = (max_t-1) * max_s; - } - else - { - offset = mBeginS; - } - - // Figure out the normal, assume all caps are flat faces. - // Cross product to get normals. - - LLVector2 cuv; - LLVector2 min_uv, max_uv; - - LLVector3& min = mExtents[0]; - LLVector3& max = mExtents[1]; - - // Copy the vertices into the array - for (S32 i = 0; i < num_vertices; i++) - { - if (mTypeMask & TOP_MASK) - { - mVertices[i].mTexCoord.mV[0] = profile[i].mV[0]+0.5f; - mVertices[i].mTexCoord.mV[1] = profile[i].mV[1]+0.5f; - } - else - { - // Mirror for underside. - mVertices[i].mTexCoord.mV[0] = profile[i].mV[0]+0.5f; - mVertices[i].mTexCoord.mV[1] = 0.5f - profile[i].mV[1]; - } - - mVertices[i].mPosition = mesh[i + offset].mPos; - - 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); - } - } - - 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(); - - LLVector3 d0; - LLVector3 d1; - LLVector3 normal; - - d0 = mCenter-mVertices[0].mPosition; - d1 = mCenter-mVertices[1].mPosition; - - normal = (mTypeMask & TOP_MASK) ? (d0%d1) : (d1%d0); - normal.normVec(); - - VertexData vd; - vd.mPosition = mCenter; - vd.mNormal = normal; - vd.mBinormal = binormal; - vd.mTexCoord = cuv; - - if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK)) - { - mVertices.push_back(vd); - 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; - - if (partial_build) - { - return TRUE; - } - - if (mTypeMask & HOLLOW_MASK) - { - if (mTypeMask & TOP_MASK) - { - // HOLLOW TOP - // Does it matter if it's open or closed? - djs - - S32 pt1 = 0, pt2 = num_vertices - 1; - S32 i = 0; - while (pt2 - pt1 > 1) - { - // 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]; - - 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) - { - mIndices[i++] = pt1; - mIndices[i++] = pt1 + 1; - mIndices[i++] = pt2; - pt1++; - } - else - { - mIndices[i++] = pt1; - mIndices[i++] = pt2 - 1; - mIndices[i++] = pt2; - pt2--; - } - } - - makeTriStrip(); - } - else - { - // HOLLOW BOTTOM - // Does it matter if it's open or closed? - djs - - llassert(mTypeMask & BOTTOM_MASK); - S32 pt1 = 0, pt2 = num_vertices - 1; - - S32 i = 0; - while (pt2 - pt1 > 1) - { - // 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]; - - 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; - } - } - - // Flipped backfacing from top - if (use_tri1a2) - { - mIndices[i++] = pt1; - mIndices[i++] = pt2; - mIndices[i++] = pt1 + 1; - pt1++; - } - else - { - mIndices[i++] = pt1; - mIndices[i++] = pt2; - mIndices[i++] = pt2 - 1; - pt2--; - } - } - - makeTriStrip(); - } - } - else - { - // Not hollow, generate the triangle fan. - U16 v1 = 2; - U16 v2 = 1; - - if (mTypeMask & TOP_MASK) - { - v1 = 1; - v2 = 2; - } - - for (S32 i = 0; i < (num_vertices - 2); i++) - { - mIndices[3*i] = num_vertices - 1; - mIndices[3*i+v1] = i; - mIndices[3*i+v2] = i + 1; - } - -#if GEN_TRI_STRIP - //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]); - } - } -#endif - } - - return TRUE; -} - -void LLVolumeFace::makeTriStrip() -{ -#if GEN_TRI_STRIP - for (U32 i = 0; i < mIndices.size(); i+=3) - { - U16 i0 = mIndices[i]; - U16 i1 = mIndices[i+1]; - U16 i2 = mIndices[i+2]; - - if ((i/3)%2 == 1) - { - mTriStrip.push_back(i0); - mTriStrip.push_back(i0); - mTriStrip.push_back(i1); - mTriStrip.push_back(i2); - mTriStrip.push_back(i2); - } - else - { - mTriStrip.push_back(i2); - mTriStrip.push_back(i2); - mTriStrip.push_back(i1); - mTriStrip.push_back(i0); - mTriStrip.push_back(i0); - } - } - - if (mTriStrip.size()%2 == 1) - { - mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); - } -#endif -} - -void LLVolumeFace::createBinormals() -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - 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; - } - - //even out quad contributions - if (i % 2 == 0) - { - mVertices[mIndices[i*3+2]].mBinormal += binorm; - } - else - { - mVertices[mIndices[i*3+1]].mBinormal += binorm; - } - } - - //normalize binormals - for (U32 i = 0; i < mVertices.size(); i++) - { - mVertices[i].mBinormal.normVec(); - mVertices[i].mNormal.normVec(); - } - - mHasBinormals = TRUE; - } -} - -BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) -{ - LLMemType m1(LLMemType::MTYPE_VOLUME); - - BOOL flat = mTypeMask & FLAT_MASK; - - U8 sculpt_type = volume->getParams().getSculptType(); - U8 sculpt_stitching = sculpt_type & LL_SCULPT_TYPE_MASK; - BOOL sculpt_invert = sculpt_type & LL_SCULPT_FLAG_INVERT; - BOOL sculpt_mirror = sculpt_type & LL_SCULPT_FLAG_MIRROR; - BOOL sculpt_reverse_horizontal = (sculpt_invert ? !sculpt_mirror : sculpt_mirror); // XOR - - 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; - - S32 max_s = volume->getProfile().getTotal(); - - S32 s, t, i; - F32 ss, tt; - - num_vertices = mNumS*mNumT; - num_indices = (mNumS-1)*(mNumT-1)*6; - - mVertices.resize(num_vertices); - - if (!partial_build) - { - mIndices.resize(num_indices); - mEdge.resize(num_indices); - } - else - { - mHasBinormals = FALSE; - } - - S32 begin_stex = llfloor( profile[mBeginS].mV[2] ); - S32 num_s = ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2) ? mNumS/2 : mNumS; - - S32 cur_vertex = 0; - // Copy the vertices into the array - for (t = mBeginT; t < mBeginT + mNumT; t++) - { - tt = path_data[t].mTexT; - for (s = 0; s < num_s; s++) - { - if (mTypeMask & END_MASK) - { - if (s) - { - ss = 1.f; - } - else - { - ss = 0.f; - } - } - else - { - // Get s value for tex-coord. - if (!flat) - { - ss = profile[mBeginS + s].mV[2]; - } - else - { - ss = profile[mBeginS + s].mV[2] - begin_stex; - } - } - - if (sculpt_reverse_horizontal) - { - ss = 1.f - ss; - } - - // Check to see if this triangle wraps around the array. - if (mBeginS + s >= max_s) - { - // We're wrapping - i = mBeginS + s + max_s*(t-1); - } - else - { - i = mBeginS + s + max_s*t; - } - - 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); - - cur_vertex++; - - if ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2 && s > 0) - { - mVertices[cur_vertex].mPosition = mesh[i].mPos; - mVertices[cur_vertex].mTexCoord = LLVector2(ss,tt); - - mVertices[cur_vertex].mNormal = LLVector3(0,0,0); - mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); - cur_vertex++; - } - } - - if ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2) - { - if (mTypeMask & OPEN_MASK) - { - s = num_s-1; - } - else - { - s = 0; - } - - 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); - - cur_vertex++; - } - } - - - //get bounding box for this side - LLVector3& face_min = mExtents[0]; - LLVector3& face_max = mExtents[1]; - mCenter.clearVec(); - - face_min = face_max = mVertices[0].mPosition; - for (U32 i = 1; i < mVertices.size(); ++i) - { - update_min_max(face_min, face_max, mVertices[i].mPosition); - } - - mCenter = (face_min + face_max) * 0.5f; - - S32 cur_index = 0; - S32 cur_edge = 0; - BOOL flat_face = mTypeMask & FLAT_MASK; - - if (!partial_build) - { -#if GEN_TRI_STRIP - mTriStrip.clear(); -#endif - - // Now we generate the indices. - for (t = 0; t < (mNumT-1); t++) - { -#if GEN_TRI_STRIP - //prepend terminating index to strip - mTriStrip.push_back(mNumS*t); -#endif - - for (s = 0; s < (mNumS-1); s++) - { - mIndices[cur_index++] = s + mNumS*t; //bottom left - mIndices[cur_index++] = s+1 + mNumS*(t+1); //top right - mIndices[cur_index++] = s + mNumS*(t+1); //top left - mIndices[cur_index++] = s + mNumS*t; //bottom left - mIndices[cur_index++] = s+1 + mNumS*t; //bottom right - mIndices[cur_index++] = s+1 + mNumS*(t+1); //top right - -#if GEN_TRI_STRIP - 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)); -#endif - - 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; - } - else if (mNumT <= 3 || volume->getPath().isOpen() == TRUE) { //no neighbor - mEdge[cur_edge++] = -1; - } - else { //wrap on T - mEdge[cur_edge++] = s*2+1; - } - if (s > 0) { //top left/bottom left neighbor face - mEdge[cur_edge++] = (mNumS-1)*2*t+s*2-1; - } - else if (flat_face || volume->getProfile().isOpen() == TRUE) { //no neighbor - mEdge[cur_edge++] = -1; - } - else { //wrap on S - mEdge[cur_edge++] = (mNumS-1)*2*t+(mNumS-2)*2+1; - } - - if (t > 0) { //bottom left/bottom right neighbor face - mEdge[cur_edge++] = (mNumS-1)*2*(t-1)+s*2; - } - else if (mNumT <= 3 || volume->getPath().isOpen() == TRUE) { //no neighbor - mEdge[cur_edge++] = -1; - } - else { //wrap on T - mEdge[cur_edge++] = (mNumS-1)*2*(mNumT-2)+s*2; - } - if (s < mNumS-2) { //bottom right/top right neighbor face - mEdge[cur_edge++] = (mNumS-1)*2*t+(s+1)*2; - } - else if (flat_face || volume->getProfile().isOpen() == TRUE) { //no neighbor - mEdge[cur_edge++] = -1; - } - else { //wrap on S - mEdge[cur_edge++] = (mNumS-1)*2*t; - } - mEdge[cur_edge++] = (mNumS-1)*2*t+s*2; //top right/bottom left neighbor face - } -#if GEN_TRI_STRIP - //append terminating vertex to strip - mTriStrip.push_back(mNumS-1+mNumS*(t+1)); -#endif - } - -#if GEN_TRI_STRIP - if (mTriStrip.size()%2 == 1) - { - mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); - } -#endif - } - - //generate normals - for (U32 i = 0; i < mIndices.size()/3; i++) //for each triangle - { - const U16* idx = &(mIndices[i*3]); - - VertexData* v[] = - { &mVertices[idx[0]], &mVertices[idx[1]], &mVertices[idx[2]] }; - - //calculate triangle normal - LLVector3 norm = (v[0]->mPosition-v[1]->mPosition) % (v[0]->mPosition-v[2]->mPosition); - - v[0]->mNormal += norm; - v[1]->mNormal += norm; - v[2]->mNormal += norm; - - //even out quad contributions - v[i%2+1]->mNormal += norm; - } - - // 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); - 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; - } - } - - if ((volume->getProfile().isOpen() == FALSE) && !(s_bottom_converges)) - { //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; - } - } - - if (volume->getPathType() == LL_PCODE_PATH_CIRCLE && - ((volume->getProfileType() & LL_PCODE_PROFILE_MASK) == LL_PCODE_PROFILE_CIRCLE_HALF)) - { - if (s_bottom_converges) - { //all lower S have same normal - for (S32 i = 0; i < mNumT; i++) - { - mVertices[mNumS*i].mNormal = LLVector3(1,0,0); - } - } - - if (s_top_converges) - { //all upper S have same normal - for (S32 i = 0; i < mNumT; i++) - { - mVertices[mNumS*i+mNumS-1].mNormal = LLVector3(-1,0,0); - } - } - } - } - - else // logic for sculpt volumes - { - BOOL average_poles = FALSE; - BOOL wrap_s = FALSE; - BOOL wrap_t = FALSE; - - if (sculpt_stitching == LL_SCULPT_TYPE_SPHERE) - average_poles = TRUE; - - if ((sculpt_stitching == LL_SCULPT_TYPE_SPHERE) || - (sculpt_stitching == LL_SCULPT_TYPE_TORUS) || - (sculpt_stitching == LL_SCULPT_TYPE_CYLINDER)) - wrap_s = TRUE; - - if (sculpt_stitching == LL_SCULPT_TYPE_TORUS) - wrap_t = TRUE; - - - if (average_poles) - { - // average normals for north pole - - LLVector3 average(0.0, 0.0, 0.0); - for (S32 i = 0; i < mNumS; i++) - { - average += mVertices[i].mNormal; - } - - // set average - for (S32 i = 0; i < mNumS; i++) - { - mVertices[i].mNormal = average; - } - - // average normals for south pole - - average = LLVector3(0.0, 0.0, 0.0); - for (S32 i = 0; i < mNumS; i++) - { - average += mVertices[i + mNumS * (mNumT - 1)].mNormal; - } - - // set average - for (S32 i = 0; i < mNumS; i++) - { - mVertices[i + mNumS * (mNumT - 1)].mNormal = average; - } - - } - - - if (wrap_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; - } - } - - - - 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; - } - - } - - } - - 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); - - 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 ); - } -} +/**
+
+ * @file llvolume.cpp
+ *
+ * $LicenseInfo:firstyear=2002&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * 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.
+ *
+ * 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.
+ *
+ * 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
+ *
+ * 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 "llerror.h"
+#include "llmemtype.h"
+
+#include "llvolumemgr.h"
+#include "v2math.h"
+#include "v3math.h"
+#include "v4math.h"
+#include "m4math.h"
+#include "m3math.h"
+#include "llmatrix3a.h"
+#include "lloctree.h"
+#include "lldarray.h"
+#include "llvolume.h"
+#include "llvolumeoctree.h"
+#include "llstl.h"
+#include "llsdserialize.h"
+#include "llvector4a.h"
+#include "llmatrix4a.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;
+const F32 HOLLOW_MAX = 0.95f;
+const F32 HOLLOW_MAX_SQUARE = 0.7f;
+
+const F32 TWIST_MIN = -1.f;
+const F32 TWIST_MAX = 1.f;
+
+const F32 RATIO_MIN = 0.f;
+const F32 RATIO_MAX = 2.f; // Tom Y: Inverted sense here: 0 = top taper, 2 = bottom taper
+
+const F32 HOLE_X_MIN= 0.05f;
+const F32 HOLE_X_MAX= 1.0f;
+
+const F32 HOLE_Y_MIN= 0.05f;
+const F32 HOLE_Y_MAX= 0.5f;
+
+const F32 SHEAR_MIN = -0.5f;
+const F32 SHEAR_MAX = 0.5f;
+
+const F32 REV_MIN = 1.f;
+const F32 REV_MAX = 4.f;
+
+const F32 TAPER_MIN = -1.f;
+const F32 TAPER_MAX = 1.f;
+
+const F32 SKEW_MIN = -0.95f;
+const F32 SKEW_MAX = 0.95f;
+
+const F32 SCULPT_MIN_AREA = 0.002f;
+const S32 SCULPT_MIN_AREA_DETAIL = 1;
+
+extern BOOL gDebugGL;
+
+void assert_aligned(void* ptr, uintptr_t alignment)
+{
+#if 0
+ uintptr_t t = (uintptr_t) ptr;
+ if (t%alignment != 0)
+ {
+ llerrs << "WTF?" << llendl;
+ }
+#endif
+}
+
+BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLVector3& pt3, const LLVector3& norm)
+{
+ LLVector3 test = (pt2-pt1)%(pt3-pt2);
+
+ //answer
+ if(test * norm < 0)
+ {
+ return FALSE;
+ }
+ else
+ {
+ return TRUE;
+ }
+}
+
+BOOL LLLineSegmentBoxIntersect(const LLVector3& start, const LLVector3& end, const LLVector3& center, const LLVector3& size)
+{
+ 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[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[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;
+}
+
+
+
+// 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 LLVector4a& vert0, const LLVector4a& vert1, const LLVector4a& vert2, const LLVector4a& orig, const LLVector4a& dir,
+ F32& intersection_a, F32& intersection_b, F32& intersection_t)
+{
+
+ /* find vectors for two edges sharing vert0 */
+ LLVector4a edge1;
+ edge1.setSub(vert1, vert0);
+
+ LLVector4a edge2;
+ edge2.setSub(vert2, vert0);
+
+ /* begin calculating determinant - also used to calculate U parameter */
+ LLVector4a pvec;
+ pvec.setCross3(dir, edge2);
+
+ /* if determinant is near zero, ray lies in plane of triangle */
+ LLVector4a det;
+ det.setAllDot3(edge1, pvec);
+
+ if (det.greaterEqual(LLVector4a::getEpsilon()).getGatheredBits() & 0x7)
+ {
+ /* calculate distance from vert0 to ray origin */
+ LLVector4a tvec;
+ tvec.setSub(orig, vert0);
+
+ /* calculate U parameter and test bounds */
+ LLVector4a u;
+ u.setAllDot3(tvec,pvec);
+
+ if ((u.greaterEqual(LLVector4a::getZero()).getGatheredBits() & 0x7) &&
+ (u.lessEqual(det).getGatheredBits() & 0x7))
+ {
+ /* 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;
+ }
+ }
+ }
+
+ return FALSE;
+}
+
+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;
+
+ /* find vectors for two edges sharing vert0 */
+ LLVector4a edge1;
+ edge1.setSub(vert1, vert0);
+
+
+ LLVector4a edge2;
+ edge2.setSub(vert2, vert0);
+
+ /* begin calculating determinant - also used to calculate U parameter */
+ LLVector4a pvec;
+ pvec.setCross3(dir, edge2);
+
+ /* 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;
+ }
+
+ 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 (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->getData().empty())
+ { //node has data, find AABB that binds data set
+ const LLVolumeTriangle* tri = *(branch->getData().begin());
+
+ //initialize min/max to first available vertex
+ min = *(tri->mV[0]);
+ max = *(tri->mV[0]);
+
+ for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter =
+ branch->getData().begin(); iter != branch->getData().end(); ++iter)
+ { //for each triangle in node
+
+ //stretch by triangles in node
+ tri = *iter;
+
+ 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->getChildren().empty())
+ { //no data, but child nodes exist
+ LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(0)->getListener(0);
+
+ //initialize min/max to extents of first child
+ min = child->mExtents[0];
+ max = child->mExtents[1];
+ }
+ else
+ {
+ llerrs << "WTF? Empty leaf" << llendl;
+ }
+
+ for (S32 i = 0; i < branch->getChildCount(); ++i)
+ { //stretch by child extents
+ LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0);
+ min.setMin(min, child->mExtents[0]);
+ max.setMax(max, child->mExtents[1]);
+ }
+
+ node->mBounds[0].setAdd(min, max);
+ node->mBounds[0].mul(0.5f);
+
+ node->mBounds[1].setSub(max,min);
+ node->mBounds[1].mul(0.5f);
+ }
+};
+
+//-------------------------------------------------------------------
+// statics
+//-------------------------------------------------------------------
+
+
+//----------------------------------------------------
+
+LLProfile::Face* LLProfile::addCap(S16 faceID)
+{
+ LLMemType m1(LLMemType::MTYPE_VOLUME);
+
+ Face *face = vector_append(mFaces, 1);
+
+ face->mIndex = 0;
+ face->mCount = mTotal;
+ face->mScaleU= 1.0f;
+ face->mCap = TRUE;
+ face->mFaceID = faceID;
+ return face;
+}
+
+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;
+ face->mCount = count;
+ face->mScaleU= scaleU;
+
+ face->mFlat = flat;
+ face->mCap = FALSE;
+ face->mFaceID = faceID;
+ return face;
+}
+
+// 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;
+
+ F32 begin = params.getBegin();
+ F32 end = params.getEnd();
+
+ t_step = 1.0f / sides;
+ ang_step = 2.0f*F_PI*t_step*ang_scale;
+
+ // 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
+
+ if (total_sides < 8)
+ {
+ scale = tableScale[total_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;
+ ang = 2.0f*F_PI*(t*ang_scale + offset);
+ pt1.setVec(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);
+
+ 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);
+ mProfile.push_back(new_pt);
+ }
+
+ // 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.
+ pt1.setVec(cos(ang)*scale,sin(ang)*scale,t);
+
+ if (mProfile.size() > 0) {
+ LLVector3 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(pt1);
+
+ t += t_step;
+ ang += ang_step;
+ }
+
+ t_fraction = (end - (t - t_step))*sides;
+
+ // 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);
+
+ // 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);
+
+ if (mProfile.size() > 0) {
+ LLVector3 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(new_pt);
+ }
+
+ // If we're sliced, the profile is open.
+ if ((end - begin)*ang_scale < 0.99f)
+ {
+ if ((end - begin)*ang_scale > 0.5f)
+ {
+ mConcave = TRUE;
+ }
+ else
+ {
+ mConcave = FALSE;
+ }
+ mOpen = TRUE;
+ if (params.getHollow() <= 0)
+ {
+ // put center point if not hollow.
+ mProfile.push_back(LLVector3(0,0,0));
+ }
+ }
+ else
+ {
+ // The profile isn't open.
+ mOpen = FALSE;
+ mConcave = FALSE;
+ }
+
+ 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.
+LLProfile::Face* LLProfile::addHole(const LLProfileParams& params, BOOL flat, F32 sides, F32 offset, F32 box_hollow, F32 ang_scale, S32 split)
+{
+ // Note that addHole will NOT work for non-"circular" profiles, if we ever decide to use them.
+
+ // Total add has number of vertices on outside.
+ mTotalOut = mTotal;
+
+ // Why is the "bevel" parameter -1? DJS 04/05/02
+ genNGon(params, llfloor(sides),offset,-1, ang_scale, split);
+
+ Face *face = addFace(mTotalOut, mTotal-mTotalOut,0,LL_FACE_INNER_SIDE, flat);
+
+ std::vector<LLVector3> pt;
+ pt.resize(mTotal) ;
+
+ for (S32 i=mTotalOut;i<mTotal;i++)
+ {
+ pt[i] = mProfile[i] * box_hollow;
+ }
+
+ S32 j=mTotal-1;
+ for (S32 i=mTotalOut;i<mTotal;i++)
+ {
+ mProfile[i] = pt[j--];
+ }
+
+ for (S32 i=0;i<(S32)mFaces.size();i++)
+ {
+ if (mFaces[i].mCap)
+ {
+ mFaces[i].mCount *= 2;
+ }
+ }
+
+ return face;
+}
+
+
+
+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;
+ }
+ mDirty = FALSE;
+
+ if (detail < MIN_LOD)
+ {
+ llinfos << "Generating profile with LOD < MIN_LOD. CLAMPING" << llendl;
+ detail = MIN_LOD;
+ }
+
+ mProfile.clear();
+ mFaces.clear();
+
+ // Generate the face data
+ S32 i;
+ F32 begin = params.getBegin();
+ F32 end = params.getEnd();
+ F32 hollow = params.getHollow();
+
+ // Quick validation to eliminate some server crashes.
+ if (begin > end - 0.01f)
+ {
+ llwarns << "LLProfile::generate() assertion failed (begin >= end)" << llendl;
+ return FALSE;
+ }
+
+ S32 face_num = 0;
+
+ switch (params.getCurveType() & LL_PCODE_PROFILE_MASK)
+ {
+ case LL_PCODE_PROFILE_SQUARE:
+ {
+ genNGon(params, 4,-0.375, 0, 1, split);
+ if (path_open)
+ {
+ addCap (LL_FACE_PATH_BEGIN);
+ }
+
+ for (i = llfloor(begin * 4.f); i < llfloor(end * 4.f + .999f); i++)
+ {
+ addFace((face_num++) * (split +1), split+2, 1, LL_FACE_OUTER_SIDE_0 << i, TRUE);
+ }
+
+ for (i = 0; i <(S32) mProfile.size(); i++)
+ {
+ // Scale by 4 to generate proper tex coords.
+ mProfile[i].mV[2] *= 4.f;
+ }
+
+ if (hollow)
+ {
+ switch (params.getCurveType() & LL_PCODE_HOLE_MASK)
+ {
+ case LL_PCODE_HOLE_TRIANGLE:
+ // This offset is not correct, but we can't change it now... DK 11/17/04
+ addHole(params, TRUE, 3, -0.375f, hollow, 1.f, split);
+ break;
+ case LL_PCODE_HOLE_CIRCLE:
+ // TODO: Compute actual detail levels for cubes
+ addHole(params, FALSE, MIN_DETAIL_FACES * detail, -0.375f, hollow, 1.f);
+ break;
+ case LL_PCODE_HOLE_SAME:
+ case LL_PCODE_HOLE_SQUARE:
+ default:
+ addHole(params, TRUE, 4, -0.375f, hollow, 1.f, split);
+ break;
+ }
+ }
+
+ if (path_open) {
+ mFaces[0].mCount = mTotal;
+ }
+ }
+ break;
+ case LL_PCODE_PROFILE_ISOTRI:
+ case LL_PCODE_PROFILE_RIGHTTRI:
+ case LL_PCODE_PROFILE_EQUALTRI:
+ {
+ genNGon(params, 3,0, 0, 1, split);
+ for (i = 0; i <(S32) mProfile.size(); i++)
+ {
+ // Scale by 3 to generate proper tex coords.
+ mProfile[i].mV[2] *= 3.f;
+ }
+
+ if (path_open)
+ {
+ addCap(LL_FACE_PATH_BEGIN);
+ }
+
+ for (i = llfloor(begin * 3.f); i < llfloor(end * 3.f + .999f); i++)
+ {
+ addFace((face_num++) * (split +1), split+2, 1, LL_FACE_OUTER_SIDE_0 << i, TRUE);
+ }
+ if (hollow)
+ {
+ // Swept triangles need smaller hollowness values,
+ // because the triangle doesn't fill the bounding box.
+ F32 triangle_hollow = hollow / 2.f;
+
+ switch (params.getCurveType() & LL_PCODE_HOLE_MASK)
+ {
+ case LL_PCODE_HOLE_CIRCLE:
+ // TODO: Actually generate level of detail for triangles
+ addHole(params, FALSE, MIN_DETAIL_FACES * detail, 0, triangle_hollow, 1.f);
+ break;
+ case LL_PCODE_HOLE_SQUARE:
+ addHole(params, TRUE, 4, 0, triangle_hollow, 1.f, split);
+ break;
+ case LL_PCODE_HOLE_SAME:
+ case LL_PCODE_HOLE_TRIANGLE:
+ default:
+ addHole(params, TRUE, 3, 0, triangle_hollow, 1.f, split);
+ break;
+ }
+ }
+ }
+ 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;
+
+ genNGon(params, sides);
+
+ if (path_open)
+ {
+ addCap (LL_FACE_PATH_BEGIN);
+ }
+
+ if (mOpen && !hollow)
+ {
+ addFace(0,mTotal-1,0,LL_FACE_OUTER_SIDE_0, FALSE);
+ }
+ else
+ {
+ addFace(0,mTotal,0,LL_FACE_OUTER_SIDE_0, FALSE);
+ }
+
+ if (hollow)
+ {
+ switch (hole_type)
+ {
+ case LL_PCODE_HOLE_SQUARE:
+ addHole(params, TRUE, 4, 0, hollow, 1.f, split);
+ break;
+ case LL_PCODE_HOLE_TRIANGLE:
+ addHole(params, TRUE, 3, 0, hollow, 1.f, split);
+ break;
+ case LL_PCODE_HOLE_CIRCLE:
+ case LL_PCODE_HOLE_SAME:
+ default:
+ addHole(params, FALSE, circle_detail, 0, hollow, 1.f);
+ break;
+ }
+ }
+ }
+ 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;
+ }
+ }
+ genNGon(params, llfloor(circle_detail), 0.5f, 0.f, 0.5f);
+ if (path_open)
+ {
+ addCap(LL_FACE_PATH_BEGIN);
+ }
+ if (mOpen && !params.getHollow())
+ {
+ addFace(0,mTotal-1,0,LL_FACE_OUTER_SIDE_0, FALSE);
+ }
+ else
+ {
+ addFace(0,mTotal,0,LL_FACE_OUTER_SIDE_0, FALSE);
+ }
+
+ if (hollow)
+ {
+ switch (hole_type)
+ {
+ case LL_PCODE_HOLE_SQUARE:
+ addHole(params, TRUE, 2, 0.5f, hollow, 0.5f, split);
+ break;
+ case LL_PCODE_HOLE_TRIANGLE:
+ addHole(params, TRUE, 3, 0.5f, hollow, 0.5f, split);
+ break;
+ case LL_PCODE_HOLE_CIRCLE:
+ case LL_PCODE_HOLE_SAME:
+ default:
+ addHole(params, FALSE, circle_detail, 0.5f, hollow, 0.5f);
+ break;
+ }
+ }
+
+ // Special case for openness of sphere
+ if ((params.getEnd() - params.getBegin()) < 1.f)
+ {
+ mOpen = TRUE;
+ }
+ else if (!hollow)
+ {
+ mOpen = FALSE;
+ mProfile.push_back(mProfile[0]);
+ mTotal++;
+ }
+ }
+ break;
+ default:
+ llerrs << "Unknown profile: getCurveType()=" << params.getCurveType() << llendl;
+ break;
+ };
+
+ if (path_open)
+ {
+ addCap(LL_FACE_PATH_END); // bottom
+ }
+
+ if ( mOpen) // interior edge caps
+ {
+ addFace(mTotal-1, 2,0.5,LL_FACE_PROFILE_BEGIN, TRUE);
+
+ if (hollow)
+ {
+ addFace(mTotalOut-1, 2,0.5,LL_FACE_PROFILE_END, TRUE);
+ }
+ else
+ {
+ addFace(mTotal-2, 2,0.5,LL_FACE_PROFILE_END, TRUE);
+ }
+ }
+
+ //genNormals(params);
+
+ return TRUE;
+}
+
+
+
+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
+ // changing the sscanf below.
+ char keyword[256]; /* Flawfinder: ignore */
+ char valuestr[256]; /* Flawfinder: ignore */
+ keyword[0] = 0;
+ valuestr[0] = 0;
+ F32 tempF32;
+ U32 tempU32;
+
+ while (!feof(fp))
+ {
+ if (fgets(buffer, BUFSIZE, fp) == NULL)
+ {
+ buffer[0] = '\0';
+ }
+
+ sscanf( /* Flawfinder: ignore */
+ buffer,
+ " %255s %255s",
+ keyword, valuestr);
+ if (!strcmp("{", keyword))
+ {
+ continue;
+ }
+ if (!strcmp("}",keyword))
+ {
+ break;
+ }
+ else if (!strcmp("curve", keyword))
+ {
+ sscanf(valuestr,"%d",&tempU32);
+ setCurveType((U8) tempU32);
+ }
+ else if (!strcmp("begin",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setBegin(tempF32);
+ }
+ else if (!strcmp("end",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setEnd(tempF32);
+ }
+ else if (!strcmp("hollow",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setHollow(tempF32);
+ }
+ else
+ {
+ llwarns << "unknown keyword " << keyword << " in profile import" << llendl;
+ }
+ }
+
+ return TRUE;
+}
+
+
+BOOL LLProfileParams::exportFile(LLFILE *fp) const
+{
+ fprintf(fp,"\t\tprofile 0\n");
+ fprintf(fp,"\t\t{\n");
+ fprintf(fp,"\t\t\tcurve\t%d\n", getCurveType());
+ fprintf(fp,"\t\t\tbegin\t%g\n", getBegin());
+ fprintf(fp,"\t\t\tend\t%g\n", getEnd());
+ fprintf(fp,"\t\t\thollow\t%g\n", getHollow());
+ fprintf(fp, "\t\t}\n");
+ return TRUE;
+}
+
+
+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
+ // changing the sscanf below.
+ char keyword[256]; /* Flawfinder: ignore */
+ char valuestr[256]; /* Flawfinder: ignore */
+ keyword[0] = 0;
+ valuestr[0] = 0;
+ F32 tempF32;
+ U32 tempU32;
+
+ while (input_stream.good())
+ {
+ input_stream.getline(buffer, BUFSIZE);
+ sscanf( /* Flawfinder: ignore */
+ buffer,
+ " %255s %255s",
+ keyword,
+ valuestr);
+ if (!strcmp("{", keyword))
+ {
+ continue;
+ }
+ if (!strcmp("}",keyword))
+ {
+ break;
+ }
+ else if (!strcmp("curve", keyword))
+ {
+ sscanf(valuestr,"%d",&tempU32);
+ setCurveType((U8) tempU32);
+ }
+ else if (!strcmp("begin",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setBegin(tempF32);
+ }
+ else if (!strcmp("end",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setEnd(tempF32);
+ }
+ else if (!strcmp("hollow",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setHollow(tempF32);
+ }
+ else
+ {
+ llwarns << "unknown keyword " << keyword << " in profile import" << llendl;
+ }
+ }
+
+ return TRUE;
+}
+
+
+BOOL LLProfileParams::exportLegacyStream(std::ostream& output_stream) const
+{
+ output_stream <<"\t\tprofile 0\n";
+ output_stream <<"\t\t{\n";
+ output_stream <<"\t\t\tcurve\t" << (S32) getCurveType() << "\n";
+ output_stream <<"\t\t\tbegin\t" << getBegin() << "\n";
+ output_stream <<"\t\t\tend\t" << getEnd() << "\n";
+ output_stream <<"\t\t\thollow\t" << getHollow() << "\n";
+ output_stream << "\t\t}\n";
+ return TRUE;
+}
+
+LLSD LLProfileParams::asLLSD() const
+{
+ LLSD sd;
+
+ sd["curve"] = getCurveType();
+ sd["begin"] = getBegin();
+ sd["end"] = getEnd();
+ sd["hollow"] = getHollow();
+ return sd;
+}
+
+bool LLProfileParams::fromLLSD(LLSD& sd)
+{
+ setCurveType(sd["curve"].asInteger());
+ setBegin((F32)sd["begin"].asReal());
+ setEnd((F32)sd["end"].asReal());
+ setHollow((F32)sd["hollow"].asReal());
+ return true;
+}
+
+void LLProfileParams::copyParams(const LLProfileParams ¶ms)
+{
+ LLMemType m1(LLMemType::MTYPE_VOLUME);
+ setCurveType(params.getCurveType());
+ setBegin(params.getBegin());
+ setEnd(params.getEnd());
+ setHollow(params.getHollow());
+}
+
+
+LLPath::~LLPath()
+{
+}
+
+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.
+ const F32 tableScale[] = { 1, 1, 1, 0.5f, 0.707107f, 0.53f, 0.525f, 0.5f };
+
+ F32 revolutions = params.getRevolutions();
+ F32 skew = params.getSkew();
+ F32 skew_mag = fabs(skew);
+ F32 hole_x = params.getScaleX() * (1.0f - skew_mag);
+ F32 hole_y = params.getScaleY();
+
+ // Calculate taper begin/end for x,y (Negative means taper the beginning)
+ F32 taper_x_begin = 1.0f;
+ F32 taper_x_end = 1.0f - params.getTaperX();
+ F32 taper_y_begin = 1.0f;
+ F32 taper_y_end = 1.0f - params.getTaperY();
+
+ if ( taper_x_end > 1.0f )
+ {
+ // Flip tapering.
+ taper_x_begin = 2.0f - taper_x_end;
+ taper_x_end = 1.0f;
+ }
+ if ( taper_y_end > 1.0f )
+ {
+ // Flip tapering.
+ taper_y_begin = 2.0f - taper_y_end;
+ taper_y_end = 1.0f;
+ }
+
+ // For spheres, the radius is usually zero.
+ F32 radius_start = 0.5f;
+ if (sides < 8)
+ {
+ radius_start = tableScale[sides];
+ }
+
+ // Scale the radius to take the hole size into account.
+ radius_start *= 1.0f - hole_y;
+
+ // Now check the radius offset to calculate the start,end radius. (Negative means
+ // decrease the start radius instead).
+ F32 radius_end = radius_start;
+ F32 radius_offset = params.getRadiusOffset();
+ if (radius_offset < 0.f)
+ {
+ radius_start *= 1.f + radius_offset;
+ }
+ else
+ {
+ radius_end *= 1.f - radius_offset;
+ }
+
+ // Is the path NOT a closed loop?
+ mOpen = ( (params.getEnd()*end_scale - params.getBegin() < 1.0f) ||
+ (skew_mag > 0.001f) ||
+ (fabs(taper_x_end - taper_x_begin) > 0.001f) ||
+ (fabs(taper_y_end - taper_y_begin) > 0.001f) ||
+ (fabs(radius_end - radius_start) > 0.001f) );
+
+ F32 ang, c, s;
+ LLQuaternion twist, qang;
+ PathPt *pt;
+ LLVector3 path_axis (1.f, 0.f, 0.f);
+ //LLVector3 twist_axis(0.f, 0.f, 1.f);
+ F32 twist_begin = params.getTwistBegin() * twist_scale;
+ F32 twist_end = params.getTwist() * twist_scale;
+
+ // We run through this once before the main loop, to make sure
+ // the path begins at the correct cut.
+ F32 step= 1.0f / sides;
+ F32 t = params.getBegin();
+ pt = vector_append(mPath, 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)
+ + 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->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;
+
+ 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())
+ {
+ pt = vector_append(mPath, 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)
+ + 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->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;
+
+ t+=step;
+ }
+
+ // Make one final pass for the end cut.
+ t = params.getEnd();
+ pt = vector_append(mPath, 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)
+ + 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->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;
+
+ mTotal = mPath.size();
+}
+
+const LLVector2 LLPathParams::getBeginScale() const
+{
+ LLVector2 begin_scale(1.f, 1.f);
+ if (getScaleX() > 1)
+ {
+ begin_scale.mV[0] = 2-getScaleX();
+ }
+ if (getScaleY() > 1)
+ {
+ begin_scale.mV[1] = 2-getScaleY();
+ }
+ return begin_scale;
+}
+
+const LLVector2 LLPathParams::getEndScale() const
+{
+ LLVector2 end_scale(1.f, 1.f);
+ if (getScaleX() < 1)
+ {
+ end_scale.mV[0] = getScaleX();
+ }
+ if (getScaleY() < 1)
+ {
+ end_scale.mV[1] = getScaleY();
+ }
+ return end_scale;
+}
+
+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;
+ }
+
+ if (detail < MIN_LOD)
+ {
+ llinfos << "Generating path with LOD < MIN! Clamping to 1" << llendl;
+ detail = MIN_LOD;
+ }
+
+ mDirty = FALSE;
+ S32 np = 2; // hardcode for line
+
+ mPath.clear();
+ mOpen = TRUE;
+
+ // 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;
+ if (np < split+2)
+ {
+ np = split+2;
+ }
+
+ mStep = 1.0f / (np-1);
+
+ mPath.resize(np);
+
+ LLVector2 start_scale = params.getBeginScale();
+ LLVector2 end_scale = params.getEndScale();
+
+ 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),
+ 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);
+ mPath[i].mTexT = t;
+ }
+ }
+ 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());
+
+ if (is_sculpted)
+ sides = sculpt_size;
+
+ genNGon(params, sides);
+ }
+ break;
+
+ case LL_PCODE_PATH_CIRCLE2:
+ {
+ if (params.getEnd() - params.getBegin() >= 0.99f &&
+ params.getScaleX() >= .99f)
+ {
+ mOpen = FALSE;
+ }
+
+ //genNGon(params, llfloor(MIN_DETAIL_FACES * detail), 4.f, 0.f);
+ genNGon(params, llfloor(MIN_DETAIL_FACES * detail));
+
+ F32 t = 0.f;
+ F32 tStep = 1.0f / mPath.size();
+
+ F32 toggle = 0.5f;
+ for (S32 i=0;i<(S32)mPath.size();i++)
+ {
+ mPath[i].mPos.mV[0] = toggle;
+ if (toggle == 0.5f)
+ toggle = -0.5f;
+ else
+ toggle = 0.5f;
+ t += tStep;
+ }
+ }
+
+ break;
+
+ case LL_PCODE_PATH_TEST:
+
+ np = 5;
+ mStep = 1.0f / (np-1);
+
+ mPath.resize(np);
+
+ for (S32 i=0;i<np;i++)
+ {
+ F32 t = (F32)i * mStep;
+ mPath[i].mPos.setVec(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);
+ mPath[i].mTexT = t;
+ mPath[i].mRot.setQuat(F_PI * params.getTwist() * t,1,0,0);
+ }
+
+ break;
+ };
+
+ if (params.getTwist() != params.getTwistBegin()) mOpen = TRUE;
+
+ //if ((int(fabsf(params.getTwist() - params.getTwistBegin())*100))%100 != 0) {
+ // mOpen = TRUE;
+ //}
+
+ return TRUE;
+}
+
+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);
+ 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].mTexT = 0;
+ }
+ }
+
+ return TRUE;
+}
+
+
+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
+ // changing the sscanf below.
+ char keyword[256]; /* Flawfinder: ignore */
+ char valuestr[256]; /* Flawfinder: ignore */
+ keyword[0] = 0;
+ valuestr[0] = 0;
+
+ F32 tempF32;
+ F32 x, y;
+ U32 tempU32;
+
+ while (!feof(fp))
+ {
+ if (fgets(buffer, BUFSIZE, fp) == NULL)
+ {
+ buffer[0] = '\0';
+ }
+
+ sscanf( /* Flawfinder: ignore */
+ buffer,
+ " %255s %255s",
+ keyword, valuestr);
+ if (!strcmp("{", keyword))
+ {
+ continue;
+ }
+ if (!strcmp("}",keyword))
+ {
+ break;
+ }
+ else if (!strcmp("curve", keyword))
+ {
+ sscanf(valuestr,"%d",&tempU32);
+ setCurveType((U8) tempU32);
+ }
+ else if (!strcmp("begin",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setBegin(tempF32);
+ }
+ else if (!strcmp("end",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setEnd(tempF32);
+ }
+ else if (!strcmp("scale",keyword))
+ {
+ // Legacy for one dimensional scale per path
+ sscanf(valuestr,"%g",&tempF32);
+ setScale(tempF32, tempF32);
+ }
+ else if (!strcmp("scale_x", keyword))
+ {
+ sscanf(valuestr, "%g", &x);
+ setScaleX(x);
+ }
+ else if (!strcmp("scale_y", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setScaleY(y);
+ }
+ else if (!strcmp("shear_x", keyword))
+ {
+ sscanf(valuestr, "%g", &x);
+ setShearX(x);
+ }
+ else if (!strcmp("shear_y", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setShearY(y);
+ }
+ else if (!strcmp("twist",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setTwist(tempF32);
+ }
+ else if (!strcmp("twist_begin", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setTwistBegin(y);
+ }
+ else if (!strcmp("radius_offset", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setRadiusOffset(y);
+ }
+ else if (!strcmp("taper_x", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setTaperX(y);
+ }
+ else if (!strcmp("taper_y", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setTaperY(y);
+ }
+ else if (!strcmp("revolutions", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setRevolutions(y);
+ }
+ else if (!strcmp("skew", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setSkew(y);
+ }
+ else
+ {
+ llwarns << "unknown keyword " << " in path import" << llendl;
+ }
+ }
+ return TRUE;
+}
+
+
+BOOL LLPathParams::exportFile(LLFILE *fp) const
+{
+ fprintf(fp, "\t\tpath 0\n");
+ fprintf(fp, "\t\t{\n");
+ fprintf(fp, "\t\t\tcurve\t%d\n", getCurveType());
+ fprintf(fp, "\t\t\tbegin\t%g\n", getBegin());
+ fprintf(fp, "\t\t\tend\t%g\n", getEnd());
+ fprintf(fp, "\t\t\tscale_x\t%g\n", getScaleX() );
+ fprintf(fp, "\t\t\tscale_y\t%g\n", getScaleY() );
+ fprintf(fp, "\t\t\tshear_x\t%g\n", getShearX() );
+ fprintf(fp, "\t\t\tshear_y\t%g\n", getShearY() );
+ fprintf(fp,"\t\t\ttwist\t%g\n", getTwist());
+
+ fprintf(fp,"\t\t\ttwist_begin\t%g\n", getTwistBegin());
+ fprintf(fp,"\t\t\tradius_offset\t%g\n", getRadiusOffset());
+ fprintf(fp,"\t\t\ttaper_x\t%g\n", getTaperX());
+ fprintf(fp,"\t\t\ttaper_y\t%g\n", getTaperY());
+ fprintf(fp,"\t\t\trevolutions\t%g\n", getRevolutions());
+ fprintf(fp,"\t\t\tskew\t%g\n", getSkew());
+
+ fprintf(fp, "\t\t}\n");
+ return TRUE;
+}
+
+
+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
+ // changing the sscanf below.
+ char keyword[256]; /* Flawfinder: ignore */
+ char valuestr[256]; /* Flawfinder: ignore */
+ keyword[0] = 0;
+ valuestr[0] = 0;
+
+ F32 tempF32;
+ F32 x, y;
+ U32 tempU32;
+
+ while (input_stream.good())
+ {
+ input_stream.getline(buffer, BUFSIZE);
+ sscanf( /* Flawfinder: ignore */
+ buffer,
+ " %255s %255s",
+ keyword, valuestr);
+ if (!strcmp("{", keyword))
+ {
+ continue;
+ }
+ if (!strcmp("}",keyword))
+ {
+ break;
+ }
+ else if (!strcmp("curve", keyword))
+ {
+ sscanf(valuestr,"%d",&tempU32);
+ setCurveType((U8) tempU32);
+ }
+ else if (!strcmp("begin",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setBegin(tempF32);
+ }
+ else if (!strcmp("end",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setEnd(tempF32);
+ }
+ else if (!strcmp("scale",keyword))
+ {
+ // Legacy for one dimensional scale per path
+ sscanf(valuestr,"%g",&tempF32);
+ setScale(tempF32, tempF32);
+ }
+ else if (!strcmp("scale_x", keyword))
+ {
+ sscanf(valuestr, "%g", &x);
+ setScaleX(x);
+ }
+ else if (!strcmp("scale_y", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setScaleY(y);
+ }
+ else if (!strcmp("shear_x", keyword))
+ {
+ sscanf(valuestr, "%g", &x);
+ setShearX(x);
+ }
+ else if (!strcmp("shear_y", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setShearY(y);
+ }
+ else if (!strcmp("twist",keyword))
+ {
+ sscanf(valuestr,"%g",&tempF32);
+ setTwist(tempF32);
+ }
+ else if (!strcmp("twist_begin", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setTwistBegin(y);
+ }
+ else if (!strcmp("radius_offset", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setRadiusOffset(y);
+ }
+ else if (!strcmp("taper_x", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setTaperX(y);
+ }
+ else if (!strcmp("taper_y", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setTaperY(y);
+ }
+ else if (!strcmp("revolutions", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setRevolutions(y);
+ }
+ else if (!strcmp("skew", keyword))
+ {
+ sscanf(valuestr, "%g", &y);
+ setSkew(y);
+ }
+ else
+ {
+ llwarns << "unknown keyword " << " in path import" << llendl;
+ }
+ }
+ return TRUE;
+}
+
+
+BOOL LLPathParams::exportLegacyStream(std::ostream& output_stream) const
+{
+ output_stream << "\t\tpath 0\n";
+ output_stream << "\t\t{\n";
+ output_stream << "\t\t\tcurve\t" << (S32) getCurveType() << "\n";
+ output_stream << "\t\t\tbegin\t" << getBegin() << "\n";
+ output_stream << "\t\t\tend\t" << getEnd() << "\n";
+ output_stream << "\t\t\tscale_x\t" << getScaleX() << "\n";
+ output_stream << "\t\t\tscale_y\t" << getScaleY() << "\n";
+ output_stream << "\t\t\tshear_x\t" << getShearX() << "\n";
+ output_stream << "\t\t\tshear_y\t" << getShearY() << "\n";
+ output_stream <<"\t\t\ttwist\t" << getTwist() << "\n";
+
+ output_stream <<"\t\t\ttwist_begin\t" << getTwistBegin() << "\n";
+ output_stream <<"\t\t\tradius_offset\t" << getRadiusOffset() << "\n";
+ output_stream <<"\t\t\ttaper_x\t" << getTaperX() << "\n";
+ output_stream <<"\t\t\ttaper_y\t" << getTaperY() << "\n";
+ output_stream <<"\t\t\trevolutions\t" << getRevolutions() << "\n";
+ output_stream <<"\t\t\tskew\t" << getSkew() << "\n";
+
+ output_stream << "\t\t}\n";
+ return TRUE;
+}
+
+LLSD LLPathParams::asLLSD() const
+{
+ LLSD sd = LLSD();
+ sd["curve"] = getCurveType();
+ sd["begin"] = getBegin();
+ sd["end"] = getEnd();
+ sd["scale_x"] = getScaleX();
+ sd["scale_y"] = getScaleY();
+ sd["shear_x"] = getShearX();
+ sd["shear_y"] = getShearY();
+ sd["twist"] = getTwist();
+ sd["twist_begin"] = getTwistBegin();
+ sd["radius_offset"] = getRadiusOffset();
+ sd["taper_x"] = getTaperX();
+ sd["taper_y"] = getTaperY();
+ sd["revolutions"] = getRevolutions();
+ sd["skew"] = getSkew();
+
+ return sd;
+}
+
+bool LLPathParams::fromLLSD(LLSD& sd)
+{
+ setCurveType(sd["curve"].asInteger());
+ setBegin((F32)sd["begin"].asReal());
+ setEnd((F32)sd["end"].asReal());
+ setScaleX((F32)sd["scale_x"].asReal());
+ setScaleY((F32)sd["scale_y"].asReal());
+ setShearX((F32)sd["shear_x"].asReal());
+ setShearY((F32)sd["shear_y"].asReal());
+ setTwist((F32)sd["twist"].asReal());
+ setTwistBegin((F32)sd["twist_begin"].asReal());
+ setRadiusOffset((F32)sd["radius_offset"].asReal());
+ setTaperX((F32)sd["taper_x"].asReal());
+ setTaperY((F32)sd["taper_y"].asReal());
+ setRevolutions((F32)sd["revolutions"].asReal());
+ setSkew((F32)sd["skew"].asReal());
+ return true;
+}
+
+void LLPathParams::copyParams(const LLPathParams ¶ms)
+{
+ setCurveType(params.getCurveType());
+ setBegin(params.getBegin());
+ setEnd(params.getEnd());
+ setScale(params.getScaleX(), params.getScaleY() );
+ setShear(params.getShearX(), params.getShearY() );
+ setTwist(params.getTwist());
+ setTwistBegin(params.getTwistBegin());
+ setRadiusOffset(params.getRadiusOffset());
+ setTaper( params.getTaperX(), params.getTaperY() );
+ setRevolutions(params.getRevolutions());
+ setSkew(params.getSkew());
+}
+
+S32 profile_delete_lock = 1 ;
+LLProfile::~LLProfile()
+{
+ if(profile_delete_lock)
+ {
+ llerrs << "LLProfile should not be deleted here!" << llendl ;
+ }
+}
+
+
+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;
+ mIsTetrahedron = FALSE;
+ mLODScaleBias.setVec(1,1,1);
+ mHullPoints = NULL;
+ mHullIndices = NULL;
+ mNumHullPoints = 0;
+ mNumHullIndices = 0;
+
+ // set defaults
+ if (mParams.getPathParams().getCurveType() == LL_PCODE_PATH_FLEXIBLE)
+ {
+ mPathp = new LLDynamicPath();
+ }
+ else
+ {
+ mPathp = new LLPath();
+ }
+ mProfilep = new LLProfile();
+
+ mGenerateSingleFace = generate_single_face;
+
+ generate();
+
+ if (mParams.getSculptID().isNull() && mParams.getSculptType() == LL_SCULPT_TYPE_NONE)
+ {
+ createVolumeFaces();
+ }
+}
+
+void LLVolume::resizePath(S32 length)
+{
+ mPathp->resizePath(length);
+ mVolumeFaces.clear();
+}
+
+void LLVolume::regen()
+{
+ generate();
+ createVolumeFaces();
+}
+
+void LLVolume::genBinormals(S32 face)
+{
+ mVolumeFaces[face].createBinormals();
+}
+
+LLVolume::~LLVolume()
+{
+ sNumMeshPoints -= mMesh.size();
+ delete mPathp;
+
+ profile_delete_lock = 0 ;
+ delete mProfilep;
+ profile_delete_lock = 1 ;
+
+ mPathp = NULL;
+ mProfilep = NULL;
+ mVolumeFaces.clear();
+
+ free(mHullPoints);
+ mHullPoints = NULL;
+ free(mHullIndices);
+ mHullIndices = NULL;
+}
+
+BOOL LLVolume::generate()
+{
+ LLMemType m1(LLMemType::MTYPE_VOLUME);
+ llassert_always(mProfilep);
+
+ //Added 10.03.05 Dave Parks
+ // Split is a parameter to LLProfile::generate that tesselates edges on the profile
+ // to prevent lighting and texture interpolation errors on triangles that are
+ // stretched due to twisting or scaling on the path.
+ S32 split = (S32) ((mDetail)*0.66f);
+
+ if (mParams.getPathParams().getCurveType() == LL_PCODE_PATH_LINE &&
+ (mParams.getPathParams().getScale().mV[0] != 1.0f ||
+ mParams.getPathParams().getScale().mV[1] != 1.0f) &&
+ (mParams.getProfileParams().getCurveType() == LL_PCODE_PROFILE_SQUARE ||
+ mParams.getProfileParams().getCurveType() == LL_PCODE_PROFILE_ISOTRI ||
+ mParams.getProfileParams().getCurveType() == LL_PCODE_PROFILE_EQUALTRI ||
+ mParams.getProfileParams().getCurveType() == LL_PCODE_PROFILE_RIGHTTRI))
+ {
+ split = 0;
+ }
+
+ mLODScaleBias.setVec(0.5f, 0.5f, 0.5f);
+
+ F32 profile_detail = mDetail;
+ F32 path_detail = mDetail;
+
+ U8 path_type = mParams.getPathParams().getCurveType();
+ U8 profile_type = mParams.getProfileParams().getCurveType();
+
+ if (path_type == LL_PCODE_PATH_LINE && profile_type == LL_PCODE_PROFILE_CIRCLE)
+ { //cylinders don't care about Z-Axis
+ mLODScaleBias.setVec(0.6f, 0.6f, 0.0f);
+ }
+ else if (path_type == LL_PCODE_PATH_CIRCLE)
+ {
+ 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);
+
+ if (regenPath || regenProf )
+ {
+ 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();
+
+ //generate vertex positions
+
+ // Run along the path.
+ for (S32 s = 0; s < sizeS; ++s)
+ {
+ LLVector2 scale = mPathp->mPath[s].mScale;
+ LLQuaternion rot = mPathp->mPath[s].mRot;
+
+ // Run along the profile.
+ for (S32 t = 0; t < sizeT; ++t)
+ {
+ 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;
+ }
+ }
+
+ for (std::vector<LLProfile::Face>::iterator iter = mProfilep->mFaces.begin();
+ iter != mProfilep->mFaces.end(); ++iter)
+ {
+ LLFaceID id = iter->mFaceID;
+ mFaceMask |= id;
+ }
+
+ return TRUE;
+ }
+ return FALSE;
+}
+
+void LLVolumeFace::VertexData::init()
+{
+ if (!mData)
+ {
+ mData = (LLVector4a*) malloc(sizeof(LLVector4a)*2);
+ }
+}
+
+LLVolumeFace::VertexData::VertexData()
+{
+ mData = NULL;
+ init();
+}
+
+LLVolumeFace::VertexData::VertexData(const VertexData& rhs)
+{
+ mData = NULL;
+ *this = rhs;
+}
+
+const LLVolumeFace::VertexData& LLVolumeFace::VertexData::operator=(const LLVolumeFace::VertexData& rhs)
+{
+ if (this != &rhs)
+ {
+ init();
+ LLVector4a::memcpyNonAliased16((F32*) mData, (F32*) rhs.mData, 2*sizeof(LLVector4a));
+ mTexCoord = rhs.mTexCoord;
+ }
+ return *this;
+}
+
+LLVolumeFace::VertexData::~VertexData()
+{
+ free(mData);
+ mData = NULL;
+}
+
+LLVector4a& LLVolumeFace::VertexData::getPosition()
+{
+ return mData[POSITION];
+}
+
+LLVector4a& LLVolumeFace::VertexData::getNormal()
+{
+ return mData[NORMAL];
+}
+
+const LLVector4a& LLVolumeFace::VertexData::getPosition() const
+{
+ return mData[POSITION];
+}
+
+const LLVector4a& LLVolumeFace::VertexData::getNormal() const
+{
+ return mData[NORMAL];
+}
+
+
+void LLVolumeFace::VertexData::setPosition(const LLVector4a& pos)
+{
+ mData[POSITION] = pos;
+}
+
+void LLVolumeFace::VertexData::setNormal(const LLVector4a& norm)
+{
+ mData[NORMAL] = norm;
+}
+
+bool LLVolumeFace::VertexData::operator<(const LLVolumeFace::VertexData& rhs)const
+{
+ const F32* lp = this->getPosition().getF32ptr();
+ const F32* rp = rhs.getPosition().getF32ptr();
+
+ if (lp[0] != rp[0])
+ {
+ return lp[0] < rp[0];
+ }
+
+ if (rp[1] != lp[1])
+ {
+ return lp[1] < rp[1];
+ }
+
+ if (rp[2] != lp[2])
+ {
+ return lp[2] < rp[2];
+ }
+
+ lp = getNormal().getF32ptr();
+ rp = rhs.getNormal().getF32ptr();
+
+ if (lp[0] != rp[0])
+ {
+ return lp[0] < rp[0];
+ }
+
+ if (rp[1] != lp[1])
+ {
+ return lp[1] < rp[1];
+ }
+
+ if (rp[2] != lp[2])
+ {
+ return lp[2] < rp[2];
+ }
+
+ if (mTexCoord.mV[0] != rhs.mTexCoord.mV[0])
+ {
+ return mTexCoord.mV[0] < rhs.mTexCoord.mV[0];
+ }
+
+ return mTexCoord.mV[1] < rhs.mTexCoord.mV[1];
+}
+
+bool LLVolumeFace::VertexData::operator==(const LLVolumeFace::VertexData& rhs)const
+{
+ return mData[POSITION].equals3(rhs.getPosition()) &&
+ mData[NORMAL].equals3(rhs.getNormal()) &&
+ mTexCoord == rhs.mTexCoord;
+}
+
+bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs, F32 angle_cutoff) const
+{
+ bool retval = false;
+ if (rhs.mData[POSITION].equals3(mData[POSITION]) && rhs.mTexCoord == mTexCoord)
+ {
+ if (angle_cutoff > 1.f)
+ {
+ retval = (mData[NORMAL].equals3(rhs.mData[NORMAL]));
+ }
+ else
+ {
+ F32 cur_angle = rhs.mData[NORMAL].dot3(mData[NORMAL]).getF32();
+ retval = cur_angle > angle_cutoff;
+ }
+ }
+
+ return retval;
+}
+
+BOOL LLVolume::createVolumeFacesFromFile(const std::string& file_name)
+{
+ std::ifstream is;
+
+ is.open(file_name.c_str(), std::ifstream::in | std::ifstream::binary);
+
+ BOOL success = createVolumeFacesFromStream(is);
+
+ is.close();
+
+ return success;
+}
+
+BOOL LLVolume::createVolumeFacesFromStream(std::istream& is)
+{
+ mSculptLevel = -1; // default is an error occured
+
+ LLSD header;
+ {
+ if (!LLSDSerialize::deserialize(header, is, 1024*1024*1024))
+ {
+ llwarns << "Mesh header parse error. Not a valid mesh asset!" << llendl;
+ return FALSE;
+ }
+ }
+
+ std::string nm[] =
+ {
+ "lowest_lod",
+ "low_lod",
+ "medium_lod",
+ "high_lod"
+ };
+
+ S32 lod = llclamp((S32) mDetail, 0, 3);
+
+ while (lod < 4 &&
+ (header[nm[lod]]["offset"].asInteger() == -1 ||
+ header[nm[lod]]["size"].asInteger() == 0 ))
+ {
+ ++lod;
+ }
+
+ if (lod >= 4)
+ {
+ lod = llclamp((S32) mDetail, 0, 3);
+
+ while (lod >= 0 &&
+ (header[nm[lod]]["offset"].asInteger() == -1 ||
+ header[nm[lod]]["size"].asInteger() == 0) )
+ {
+ --lod;
+ }
+
+ if (lod < 0)
+ {
+ llwarns << "Mesh header missing LOD offsets. Not a valid mesh asset!" << llendl;
+ return FALSE;
+ }
+ }
+
+ is.seekg(header[nm[lod]]["offset"].asInteger(), std::ios_base::cur);
+
+ return unpackVolumeFaces(is, header[nm[lod]]["size"].asInteger());
+}
+
+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))
+ {
+ llwarns << "not a valid mesh asset!" << llendl;
+ return false;
+ }
+
+ {
+ U32 face_count = mdl.size();
+
+ if (face_count == 0)
+ {
+ llerrs << "WTF?" << llendl;
+ }
+
+ mVolumeFaces.resize(face_count);
+
+ for (U32 i = 0; i < face_count; ++i)
+ {
+ LLSD::Binary pos = mdl[i]["Position"];
+ LLSD::Binary norm = mdl[i]["Normal"];
+ LLSD::Binary tc = mdl[i]["TexCoord0"];
+ LLSD::Binary idx = mdl[i]["TriangleList"];
+
+ LLVolumeFace& face = mVolumeFaces[i];
+
+ //copy out indices
+ face.resizeIndices(idx.size()/2);
+
+ if (idx.empty() || face.mNumIndices < 3)
+ { //why is there an empty index list?
+ llerrs <<"WTF?" << llendl;
+ continue;
+ }
+
+ U16* indices = (U16*) &(idx[0]);
+ for (U32 j = 0; j < idx.size()/2; ++j)
+ {
+ face.mIndices[j] = indices[j];
+ }
+
+ //copy out vertices
+ U32 num_verts = pos.size()/(3*2);
+ face.resizeVertices(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())
+ {
+ llwarns << "Vertex weight count does not match vertex count!" << llendl;
+ }
+
+ }
+
+ 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_range = max_tc - min_tc;
+
+ LLVector4a* pos_out = face.mPositions;
+ LLVector4a* norm_out = face.mNormals;
+ LLVector2* tc_out = face.mTexCoords;
+
+ for (U32 j = 0; j < num_verts; ++j)
+ {
+ U16* v = (U16*) &(pos[j*3*2]);
+
+ 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++;
+
+ U16* n = (U16*) &(norm[j*3*2]);
+
+ 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++;
+
+ U16* t = (U16*) &(tc[j*2*2]);
+
+ tc_out->mV[0] = (F32) t[0] / 65535.f * tc_range.mV[0] + min_tc.mV[0];
+ tc_out->mV[1] = (F32) t[1] / 65535.f * tc_range.mV[1] + min_tc.mV[1];
+
+ tc_out++;
+ }
+
+
+ // 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];
+
+ min.clear();
+ max.clear();
+ 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]);
+ }
+ }
+ }
+
+ mSculptLevel = 0; // success!
+
+ cacheOptimize();
+
+ return true;
+}
+
+void tetrahedron_set_normal(LLVolumeFace::VertexData* cv)
+{
+ LLVector4a v0;
+ v0.setSub(cv[1].getPosition(), cv[0].getNormal());
+ LLVector4a v1;
+ v1.setSub(cv[2].getNormal(), cv[0].getPosition());
+
+ cv[0].getNormal().setCross3(v0,v1);
+ cv[0].getNormal().normalize3fast();
+ cv[1].setNormal(cv[0].getNormal());
+ cv[2].setNormal(cv[1].getNormal());
+}
+
+BOOL LLVolume::isTetrahedron()
+{
+ return mIsTetrahedron;
+}
+
+void LLVolume::makeTetrahedron()
+{
+ mVolumeFaces.clear();
+
+ LLVolumeFace face;
+
+ F32 x = 0.25f;
+ LLVector4a p[] =
+ { //unit tetrahedron corners
+ LLVector4a(x,x,x),
+ LLVector4a(-x,-x,x),
+ LLVector4a(-x,x,-x),
+ LLVector4a(x,-x,-x)
+ };
+
+ face.mExtents[0].splat(-x);
+ face.mExtents[1].splat(x);
+
+ LLVolumeFace::VertexData cv[3];
+
+ //set texture coordinates
+ cv[0].mTexCoord = LLVector2(0,0);
+ cv[1].mTexCoord = LLVector2(1,0);
+ cv[2].mTexCoord = LLVector2(0.5f, 0.5f*F_SQRT3);
+
+
+ //side 1
+ cv[0].setPosition(p[1]);
+ cv[1].setPosition(p[0]);
+ cv[2].setPosition(p[2]);
+
+ tetrahedron_set_normal(cv);
+
+ face.resizeVertices(12);
+ face.resizeIndices(12);
+
+ LLVector4a* v = (LLVector4a*) face.mPositions;
+ LLVector4a* n = (LLVector4a*) face.mNormals;
+ LLVector2* tc = (LLVector2*) face.mTexCoords;
+
+ v[0] = cv[0].getPosition();
+ v[1] = cv[1].getPosition();
+ v[2] = cv[2].getPosition();
+ v += 3;
+
+ n[0] = cv[0].getNormal();
+ n[1] = cv[1].getNormal();
+ n[2] = cv[2].getNormal();
+ n += 3;
+
+ tc[0] = cv[0].mTexCoord;
+ tc[1] = cv[1].mTexCoord;
+ tc[2] = cv[2].mTexCoord;
+ tc += 3;
+
+
+ //side 2
+ cv[0].setPosition(p[3]);
+ cv[1].setPosition(p[0]);
+ cv[2].setPosition(p[1]);
+
+ tetrahedron_set_normal(cv);
+
+ v[0] = cv[0].getPosition();
+ v[1] = cv[1].getPosition();
+ v[2] = cv[2].getPosition();
+ v += 3;
+
+ n[0] = cv[0].getNormal();
+ n[1] = cv[1].getNormal();
+ n[2] = cv[2].getNormal();
+ n += 3;
+
+ tc[0] = cv[0].mTexCoord;
+ tc[1] = cv[1].mTexCoord;
+ tc[2] = cv[2].mTexCoord;
+ tc += 3;
+
+ //side 3
+ cv[0].setPosition(p[3]);
+ cv[1].setPosition(p[1]);
+ cv[2].setPosition(p[2]);
+
+ tetrahedron_set_normal(cv);
+
+ v[0] = cv[0].getPosition();
+ v[1] = cv[1].getPosition();
+ v[2] = cv[2].getPosition();
+ v += 3;
+
+ n[0] = cv[0].getNormal();
+ n[1] = cv[1].getNormal();
+ n[2] = cv[2].getNormal();
+ n += 3;
+
+ tc[0] = cv[0].mTexCoord;
+ tc[1] = cv[1].mTexCoord;
+ tc[2] = cv[2].mTexCoord;
+ tc += 3;
+
+ //side 4
+ cv[0].setPosition(p[2]);
+ cv[1].setPosition(p[0]);
+ cv[2].setPosition(p[3]);
+
+ tetrahedron_set_normal(cv);
+
+ v[0] = cv[0].getPosition();
+ v[1] = cv[1].getPosition();
+ v[2] = cv[2].getPosition();
+ v += 3;
+
+ n[0] = cv[0].getNormal();
+ n[1] = cv[1].getNormal();
+ n[2] = cv[2].getNormal();
+ n += 3;
+
+ tc[0] = cv[0].mTexCoord;
+ tc[1] = cv[1].mTexCoord;
+ tc[2] = cv[2].mTexCoord;
+ tc += 3;
+
+ //set index buffer
+ for (U16 i = 0; i < 12; i++)
+ {
+ face.mIndices[i] = i;
+ }
+
+ mVolumeFaces.push_back(face);
+ mSculptLevel = 0;
+ mIsTetrahedron = TRUE;
+}
+
+void LLVolume::copyVolumeFaces(const LLVolume* volume)
+{
+ mVolumeFaces = volume->mVolumeFaces;
+ mSculptLevel = 0;
+ mIsTetrahedron = FALSE;
+}
+
+void LLVolume::cacheOptimize()
+{
+ for (S32 i = 0; i < mVolumeFaces.size(); ++i)
+ {
+ mVolumeFaces[i].cacheOptimize();
+ }
+}
+
+
+S32 LLVolume::getNumFaces() const
+{
+ U8 sculpt_type = (mParams.getSculptType() & LL_SCULPT_TYPE_MASK);
+
+ if (sculpt_type == LL_SCULPT_TYPE_MESH)
+ {
+ return LL_SCULPT_MESH_MAX_FACES;
+ }
+
+ return (S32)mProfilep->mFaces.size();
+}
+
+
+void LLVolume::createVolumeFaces()
+{
+ LLMemType m1(LLMemType::MTYPE_VOLUME);
+
+ if (mGenerateSingleFace)
+ {
+ // do nothing
+ }
+ else
+ {
+ S32 num_faces = getNumFaces();
+ BOOL partial_build = TRUE;
+ if (num_faces != mVolumeFaces.size())
+ {
+ partial_build = FALSE;
+ mVolumeFaces.resize(num_faces);
+ }
+ // Initialize volume faces with parameter data
+ for (S32 i = 0; i < (S32)mVolumeFaces.size(); i++)
+ {
+ LLVolumeFace& vf = mVolumeFaces[i];
+ LLProfile::Face& face = mProfilep->mFaces[i];
+ vf.mBeginS = face.mIndex;
+ vf.mNumS = face.mCount;
+ if (vf.mNumS < 0)
+ {
+ llerrs << "Volume face corruption detected." << llendl;
+ }
+
+ vf.mBeginT = 0;
+ vf.mNumT= getPath().mPath.size();
+ vf.mID = i;
+
+ // Set the type mask bits correctly
+ if (mParams.getProfileParams().getHollow() > 0)
+ {
+ vf.mTypeMask |= LLVolumeFace::HOLLOW_MASK;
+ }
+ if (mProfilep->isOpen())
+ {
+ vf.mTypeMask |= LLVolumeFace::OPEN_MASK;
+ }
+ if (face.mCap)
+ {
+ vf.mTypeMask |= LLVolumeFace::CAP_MASK;
+ if (face.mFaceID == LL_FACE_PATH_BEGIN)
+ {
+ vf.mTypeMask |= LLVolumeFace::TOP_MASK;
+ }
+ else
+ {
+ llassert(face.mFaceID == LL_FACE_PATH_END);
+ vf.mTypeMask |= LLVolumeFace::BOTTOM_MASK;
+ }
+ }
+ else if (face.mFaceID & (LL_FACE_PROFILE_BEGIN | LL_FACE_PROFILE_END))
+ {
+ vf.mTypeMask |= LLVolumeFace::FLAT_MASK | LLVolumeFace::END_MASK;
+ }
+ else
+ {
+ vf.mTypeMask |= LLVolumeFace::SIDE_MASK;
+ if (face.mFlat)
+ {
+ vf.mTypeMask |= LLVolumeFace::FLAT_MASK;
+ }
+ if (face.mFaceID & LL_FACE_INNER_SIDE)
+ {
+ vf.mTypeMask |= LLVolumeFace::INNER_MASK;
+ if (face.mFlat && vf.mNumS > 2)
+ { //flat inner faces have to copy vert normals
+ vf.mNumS = vf.mNumS*2;
+ if (vf.mNumS < 0)
+ {
+ llerrs << "Volume face corruption detected." << llendl;
+ }
+ }
+ }
+ else
+ {
+ vf.mTypeMask |= LLVolumeFace::OUTER_MASK;
+ }
+ }
+ }
+
+ for (face_list_t::iterator iter = mVolumeFaces.begin();
+ iter != mVolumeFaces.end(); ++iter)
+ {
+ (*iter).create(this, partial_build);
+ }
+ }
+}
+
+
+inline LLVector3 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;
+
+ return value;
+}
+
+inline U32 sculpt_xy_to_index(U32 x, U32 y, U16 sculpt_width, U16 sculpt_height, S8 sculpt_components)
+{
+ U32 index = (x + y * sculpt_width) * sculpt_components;
+ return index;
+}
+
+
+inline U32 sculpt_st_to_index(S32 s, S32 t, S32 size_s, S32 size_t, U16 sculpt_width, U16 sculpt_height, S8 sculpt_components)
+{
+ U32 x = (U32) ((F32)s/(size_s) * (F32) sculpt_width);
+ U32 y = (U32) ((F32)t/(size_t) * (F32) sculpt_height);
+
+ return sculpt_xy_to_index(x, y, sculpt_width, sculpt_height, sculpt_components);
+}
+
+
+inline LLVector3 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]);
+
+ 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)
+{
+ 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)
+{
+ U32 index = sculpt_xy_to_index(x, y, sculpt_width, sculpt_height, sculpt_components);
+
+ return sculpt_index_to_vector(index, sculpt_data);
+}
+
+
+F32 LLVolume::sculptGetSurfaceArea()
+{
+ // test to see if image has enough variation to create non-degenerate geometry
+
+ F32 area = 0;
+
+ S32 sizeS = mPathp->mPath.size();
+ S32 sizeT = mProfilep->mProfile.size();
+
+ for (S32 s = 0; s < sizeS-1; s++)
+ {
+ 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;
+
+ // 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;
+ }
+ }
+
+ return area;
+}
+
+// create placeholder shape
+void LLVolume::sculptGeneratePlaceholder()
+{
+ LLMemType m1(LLMemType::MTYPE_VOLUME);
+
+ S32 sizeS = mPathp->mPath.size();
+ S32 sizeT = mProfilep->mProfile.size();
+
+ S32 line = 0;
+
+ // for now, this is a sphere.
+ for (S32 s = 0; s < sizeS; s++)
+ {
+ for (S32 t = 0; t < sizeT; t++)
+ {
+ S32 i = t + line;
+ Point& pt = mMesh[i];
+
+
+ F32 u = (F32)s/(sizeS-1);
+ F32 v = (F32)t/(sizeT-1);
+
+ 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);
+
+ }
+ line += sizeT;
+ }
+}
+
+// create the vertices from the map
+void LLVolume::sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, U8 sculpt_type)
+{
+ U8 sculpt_stitching = sculpt_type & LL_SCULPT_TYPE_MASK;
+ BOOL sculpt_invert = sculpt_type & LL_SCULPT_FLAG_INVERT;
+ 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();
+
+ S32 line = 0;
+ for (S32 s = 0; s < sizeS; s++)
+ {
+ // Run along the profile.
+ for (S32 t = 0; t < sizeT; t++)
+ {
+ S32 i = t + line;
+ Point& pt = mMesh[i];
+
+ S32 reversed_t = t;
+
+ if (reverse_horizontal)
+ {
+ reversed_t = sizeT - t - 1;
+ }
+
+ U32 x = (U32) ((F32)reversed_t/(sizeT-1) * (F32) sculpt_width);
+ U32 y = (U32) ((F32)s/(sizeS-1) * (F32) sculpt_height);
+
+
+ if (y == 0) // top row stitching
+ {
+ // pinch?
+ if (sculpt_stitching == LL_SCULPT_TYPE_SPHERE)
+ {
+ x = sculpt_width / 2;
+ }
+ }
+
+ if (y == sculpt_height) // bottom row stitching
+ {
+ // wrap?
+ if (sculpt_stitching == LL_SCULPT_TYPE_TORUS)
+ {
+ y = 0;
+ }
+ else
+ {
+ y = sculpt_height - 1;
+ }
+
+ // pinch?
+ if (sculpt_stitching == LL_SCULPT_TYPE_SPHERE)
+ {
+ x = sculpt_width / 2;
+ }
+ }
+
+ if (x == sculpt_width) // side stitching
+ {
+ // wrap?
+ if ((sculpt_stitching == LL_SCULPT_TYPE_SPHERE) ||
+ (sculpt_stitching == LL_SCULPT_TYPE_TORUS) ||
+ (sculpt_stitching == LL_SCULPT_TYPE_CYLINDER))
+ {
+ x = 0;
+ }
+
+ else
+ {
+ x = sculpt_width - 1;
+ }
+ }
+
+ pt.mPos = sculpt_xy_to_vector(x, y, sculpt_width, sculpt_height, sculpt_components, sculpt_data);
+
+ if (sculpt_mirror)
+ {
+ pt.mPos.mV[VX] *= -1.f;
+ }
+ }
+
+ line += sizeT;
+ }
+}
+
+
+const S32 SCULPT_REZ_1 = 6; // changed from 4 to 6 - 6 looks round whereas 4 looks square
+const S32 SCULPT_REZ_2 = 8;
+const S32 SCULPT_REZ_3 = 16;
+const S32 SCULPT_REZ_4 = 32;
+
+S32 sculpt_sides(F32 detail)
+{
+
+ // detail is usually one of: 1, 1.5, 2.5, 4.0.
+
+ if (detail <= 1.0)
+ {
+ return SCULPT_REZ_1;
+ }
+ if (detail <= 2.0)
+ {
+ return SCULPT_REZ_2;
+ }
+ if (detail <= 3.0)
+ {
+ return SCULPT_REZ_3;
+ }
+ else
+ {
+ return SCULPT_REZ_4;
+ }
+}
+
+
+
+// determine the number of vertices in both s and t direction for this sculpt
+void sculpt_calc_mesh_resolution(U16 width, U16 height, U8 type, F32 detail, S32& s, S32& t)
+{
+ // this code has the following properties:
+ // 1) the aspect ratio of the mesh is as close as possible to the ratio of the map
+ // while still using all available verts
+ // 2) the mesh cannot have more verts than is allowed by LOD
+ // 3) the mesh cannot have more verts than is allowed by the map
+
+ S32 max_vertices_lod = (S32)pow((double)sculpt_sides(detail), 2.0);
+ S32 max_vertices_map = width * height / 4;
+
+ S32 vertices;
+ if (max_vertices_map > 0)
+ vertices = llmin(max_vertices_lod, max_vertices_map);
+ else
+ vertices = max_vertices_lod;
+
+
+ F32 ratio;
+ if ((width == 0) || (height == 0))
+ ratio = 1.f;
+ else
+ ratio = (F32) width / (F32) height;
+
+
+ s = (S32)(F32) sqrt(((F32)vertices / ratio));
+
+ s = llmax(s, 4); // no degenerate sizes, please
+ t = vertices / s;
+
+ t = llmax(t, 4); // no degenerate sizes, please
+ s = vertices / t;
+}
+
+// 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;
+
+ if (sculpt_width == 0 || sculpt_height == 0 || sculpt_components < 3 || sculpt_data == NULL)
+ {
+ sculpt_level = -1;
+ data_is_empty = TRUE;
+ }
+
+ S32 requested_sizeS = 0;
+ S32 requested_sizeT = 0;
+
+ sculpt_calc_mesh_resolution(sculpt_width, sculpt_height, sculpt_type, mDetail, requested_sizeS, requested_sizeT);
+
+ mPathp->generate(mParams.getPathParams(), mDetail, 0, TRUE, requested_sizeS);
+ mProfilep->generate(mParams.getProfileParams(), mPathp->isOpen(), mDetail, 0, TRUE, requested_sizeT);
+
+ S32 sizeS = mPathp->mPath.size(); // we requested a specific size, now see what we really got
+ S32 sizeT = mProfilep->mProfile.size(); // we requested a specific size, now see what we really got
+
+ // weird crash bug - DEV-11158 - trying to collect more data:
+ if ((sizeS == 0) || (sizeT == 0))
+ {
+ llwarns << "sculpt bad mesh size " << sizeS << " " << sizeT << llendl;
+ }
+
+ sNumMeshPoints -= mMesh.size();
+ mMesh.resize(sizeS * sizeT);
+ sNumMeshPoints += mMesh.size();
+
+ //generate vertex positions
+ if (!data_is_empty)
+ {
+ sculptGenerateMapVertices(sculpt_width, sculpt_height, sculpt_components, sculpt_data, sculpt_type);
+
+ // don't test lowest LOD to support legacy content DEV-33670
+ if (mDetail > SCULPT_MIN_AREA_DETAIL)
+ {
+ if (sculptGetSurfaceArea() < SCULPT_MIN_AREA)
+ {
+ data_is_empty = TRUE;
+ }
+ }
+ }
+
+ if (data_is_empty)
+ {
+ sculptGeneratePlaceholder();
+ }
+
+
+
+ for (S32 i = 0; i < (S32)mProfilep->mFaces.size(); i++)
+ {
+ mFaceMask |= mProfilep->mFaces[i].mFaceID;
+ }
+
+ mSculptLevel = sculpt_level;
+
+ // Delete any existing faces so that they get regenerated
+ mVolumeFaces.clear();
+
+ createVolumeFaces();
+}
+
+
+
+
+BOOL LLVolume::isCap(S32 face)
+{
+ return mProfilep->mFaces[face].mCap;
+}
+
+BOOL LLVolume::isFlat(S32 face)
+{
+ return mProfilep->mFaces[face].mFlat;
+}
+
+
+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()) &&
+ (getProfileParams() == params.getProfileParams()) &&
+ (mSculptID == params.mSculptID) &&
+ (mSculptType == params.mSculptType) );
+}
+
+bool LLVolumeParams::operator!=(const LLVolumeParams ¶ms) const
+{
+ return ( (getPathParams() != params.getPathParams()) ||
+ (getProfileParams() != params.getProfileParams()) ||
+ (mSculptID != params.mSculptID) ||
+ (mSculptType != params.mSculptType) );
+}
+
+bool LLVolumeParams::operator<(const LLVolumeParams ¶ms) const
+{
+ if( getPathParams() != params.getPathParams() )
+ {
+ return getPathParams() < params.getPathParams();
+ }
+
+ if (getProfileParams() != params.getProfileParams())
+ {
+ return getProfileParams() < params.getProfileParams();
+ }
+
+ if (mSculptID != params.mSculptID)
+ {
+ return mSculptID < params.mSculptID;
+ }
+
+ return mSculptType < params.mSculptType;
+
+
+}
+
+void LLVolumeParams::copyParams(const LLVolumeParams ¶ms)
+{
+ LLMemType m1(LLMemType::MTYPE_VOLUME);
+ mProfileParams.copyParams(params.mProfileParams);
+ mPathParams.copyParams(params.mPathParams);
+ mSculptID = params.getSculptID();
+ mSculptType = params.getSculptType();
+}
+
+// Less restricitve approx 0 for volumes
+const F32 APPROXIMATELY_ZERO = 0.001f;
+bool approx_zero( F32 f, F32 tolerance = APPROXIMATELY_ZERO)
+{
+ return (f >= -tolerance) && (f <= tolerance);
+}
+
+// return true if in range (or nearly so)
+static bool limit_range(F32& v, F32 min, F32 max, F32 tolerance = APPROXIMATELY_ZERO)
+{
+ F32 min_delta = v - min;
+ if (min_delta < 0.f)
+ {
+ v = min;
+ if (!approx_zero(min_delta, tolerance))
+ return false;
+ }
+ F32 max_delta = max - v;
+ if (max_delta < 0.f)
+ {
+ v = max;
+ if (!approx_zero(max_delta, tolerance))
+ return false;
+ }
+ return true;
+}
+
+bool LLVolumeParams::setBeginAndEndS(const F32 b, const F32 e)
+{
+ bool valid = true;
+
+ // First, clamp to valid ranges.
+ F32 begin = b;
+ valid &= limit_range(begin, 0.f, 1.f - MIN_CUT_DELTA);
+
+ F32 end = e;
+ if (end >= .0149f && end < MIN_CUT_DELTA) end = MIN_CUT_DELTA; // eliminate warning for common rounding error
+ valid &= limit_range(end, MIN_CUT_DELTA, 1.f);
+
+ valid &= limit_range(begin, 0.f, end - MIN_CUT_DELTA, .01f);
+
+ // Now set them.
+ mProfileParams.setBegin(begin);
+ mProfileParams.setEnd(end);
+
+ return valid;
+}
+
+bool LLVolumeParams::setBeginAndEndT(const F32 b, const F32 e)
+{
+ bool valid = true;
+
+ // First, clamp to valid ranges.
+ F32 begin = b;
+ valid &= limit_range(begin, 0.f, 1.f - MIN_CUT_DELTA);
+
+ F32 end = e;
+ valid &= limit_range(end, MIN_CUT_DELTA, 1.f);
+
+ valid &= limit_range(begin, 0.f, end - MIN_CUT_DELTA, .01f);
+
+ // Now set them.
+ mPathParams.setBegin(begin);
+ mPathParams.setEnd(end);
+
+ return valid;
+}
+
+bool LLVolumeParams::setHollow(const F32 h)
+{
+ // Validate the hollow based on path and profile.
+ U8 profile = mProfileParams.getCurveType() & LL_PCODE_PROFILE_MASK;
+ U8 hole_type = mProfileParams.getCurveType() & LL_PCODE_HOLE_MASK;
+
+ F32 max_hollow = HOLLOW_MAX;
+
+ // Only square holes have trouble.
+ if (LL_PCODE_HOLE_SQUARE == hole_type)
+ {
+ switch(profile)
+ {
+ case LL_PCODE_PROFILE_CIRCLE:
+ case LL_PCODE_PROFILE_CIRCLE_HALF:
+ case LL_PCODE_PROFILE_EQUALTRI:
+ max_hollow = HOLLOW_MAX_SQUARE;
+ }
+ }
+
+ F32 hollow = h;
+ bool valid = limit_range(hollow, HOLLOW_MIN, max_hollow);
+ mProfileParams.setHollow(hollow);
+
+ return valid;
+}
+
+bool LLVolumeParams::setTwistBegin(const F32 b)
+{
+ F32 twist_begin = b;
+ bool valid = limit_range(twist_begin, TWIST_MIN, TWIST_MAX);
+ mPathParams.setTwistBegin(twist_begin);
+ return valid;
+}
+
+bool LLVolumeParams::setTwistEnd(const F32 e)
+{
+ F32 twist_end = e;
+ bool valid = limit_range(twist_end, TWIST_MIN, TWIST_MAX);
+ mPathParams.setTwistEnd(twist_end);
+ return valid;
+}
+
+bool LLVolumeParams::setRatio(const F32 x, const F32 y)
+{
+ F32 min_x = RATIO_MIN;
+ F32 max_x = RATIO_MAX;
+ F32 min_y = RATIO_MIN;
+ F32 max_y = RATIO_MAX;
+ // If this is a circular path (and not a sphere) then 'ratio' is actually hole size.
+ U8 path_type = mPathParams.getCurveType();
+ U8 profile_type = mProfileParams.getCurveType() & LL_PCODE_PROFILE_MASK;
+ if ( LL_PCODE_PATH_CIRCLE == path_type &&
+ LL_PCODE_PROFILE_CIRCLE_HALF != profile_type)
+ {
+ // Holes are more restricted...
+ min_x = HOLE_X_MIN;
+ max_x = HOLE_X_MAX;
+ min_y = HOLE_Y_MIN;
+ max_y = HOLE_Y_MAX;
+ }
+
+ F32 ratio_x = x;
+ bool valid = limit_range(ratio_x, min_x, max_x);
+ F32 ratio_y = y;
+ valid &= limit_range(ratio_y, min_y, max_y);
+
+ mPathParams.setScale(ratio_x, ratio_y);
+
+ return valid;
+}
+
+bool LLVolumeParams::setShear(const F32 x, const F32 y)
+{
+ F32 shear_x = x;
+ bool valid = limit_range(shear_x, SHEAR_MIN, SHEAR_MAX);
+ F32 shear_y = y;
+ valid &= limit_range(shear_y, SHEAR_MIN, SHEAR_MAX);
+ mPathParams.setShear(shear_x, shear_y);
+ return valid;
+}
+
+bool LLVolumeParams::setTaperX(const F32 v)
+{
+ F32 taper = v;
+ bool valid = limit_range(taper, TAPER_MIN, TAPER_MAX);
+ mPathParams.setTaperX(taper);
+ return valid;
+}
+
+bool LLVolumeParams::setTaperY(const F32 v)
+{
+ F32 taper = v;
+ bool valid = limit_range(taper, TAPER_MIN, TAPER_MAX);
+ mPathParams.setTaperY(taper);
+ return valid;
+}
+
+bool LLVolumeParams::setRevolutions(const F32 r)
+{
+ F32 revolutions = r;
+ bool valid = limit_range(revolutions, REV_MIN, REV_MAX);
+ mPathParams.setRevolutions(revolutions);
+ return valid;
+}
+
+bool LLVolumeParams::setRadiusOffset(const F32 offset)
+{
+ bool valid = true;
+
+ // If this is a sphere, just set it to 0 and get out.
+ U8 path_type = mPathParams.getCurveType();
+ U8 profile_type = mProfileParams.getCurveType() & LL_PCODE_PROFILE_MASK;
+ if ( LL_PCODE_PROFILE_CIRCLE_HALF == profile_type ||
+ LL_PCODE_PATH_CIRCLE != path_type )
+ {
+ mPathParams.setRadiusOffset(0.f);
+ return true;
+ }
+
+ // Limit radius offset, based on taper and hole size y.
+ F32 radius_offset = offset;
+ F32 taper_y = getTaperY();
+ F32 radius_mag = fabs(radius_offset);
+ F32 hole_y_mag = fabs(getRatioY());
+ F32 taper_y_mag = fabs(taper_y);
+ // Check to see if the taper effects us.
+ if ( (radius_offset > 0.f && taper_y < 0.f) ||
+ (radius_offset < 0.f && taper_y > 0.f) )
+ {
+ // The taper does not help increase the radius offset range.
+ taper_y_mag = 0.f;
+ }
+ F32 max_radius_mag = 1.f - hole_y_mag * (1.f - taper_y_mag) / (1.f - hole_y_mag);
+
+ // Enforce the maximum magnitude.
+ F32 delta = max_radius_mag - radius_mag;
+ if (delta < 0.f)
+ {
+ // Check radius offset sign.
+ if (radius_offset < 0.f)
+ {
+ radius_offset = -max_radius_mag;
+ }
+ else
+ {
+ radius_offset = max_radius_mag;
+ }
+ valid = approx_zero(delta, .1f);
+ }
+
+ mPathParams.setRadiusOffset(radius_offset);
+ return valid;
+}
+
+bool LLVolumeParams::setSkew(const F32 skew_value)
+{
+ bool valid = true;
+
+ // Check the skew value against the revolutions.
+ F32 skew = llclamp(skew_value, SKEW_MIN, SKEW_MAX);
+ F32 skew_mag = fabs(skew);
+ F32 revolutions = getRevolutions();
+ F32 scale_x = getRatioX();
+ F32 min_skew_mag = 1.0f - 1.0f / (revolutions * scale_x + 1.0f);
+ // Discontinuity; A revolution of 1 allows skews below 0.5.
+ if ( fabs(revolutions - 1.0f) < 0.001)
+ min_skew_mag = 0.0f;
+
+ // Clip skew.
+ F32 delta = skew_mag - min_skew_mag;
+ if (delta < 0.f)
+ {
+ // Check skew sign.
+ if (skew < 0.0f)
+ {
+ skew = -min_skew_mag;
+ }
+ else
+ {
+ skew = min_skew_mag;
+ }
+ valid = approx_zero(delta, .01f);
+ }
+
+ mPathParams.setSkew(skew);
+ return valid;
+}
+
+bool LLVolumeParams::setSculptID(const LLUUID sculpt_id, U8 sculpt_type)
+{
+ mSculptID = sculpt_id;
+ mSculptType = sculpt_type;
+ return true;
+}
+
+bool LLVolumeParams::setType(U8 profile, U8 path)
+{
+ bool result = true;
+ // First, check profile and path for validity.
+ U8 profile_type = profile & LL_PCODE_PROFILE_MASK;
+ U8 hole_type = (profile & LL_PCODE_HOLE_MASK) >> 4;
+ U8 path_type = path >> 4;
+
+ if (profile_type > LL_PCODE_PROFILE_MAX)
+ {
+ // 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;
+ }
+ 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;
+ }
+
+ if (path_type < LL_PCODE_PATH_MIN ||
+ path_type > LL_PCODE_PATH_MAX)
+ {
+ // Bad path. Make it linear.
+ result = false;
+ llwarns << "LLVolumeParams::setType changing bad path (" << path
+ << ") to be LL_PCODE_PATH_LINE" << llendl;
+ path = LL_PCODE_PATH_LINE;
+ }
+
+ mProfileParams.setCurveType(profile);
+ mPathParams.setCurveType(path);
+ return result;
+}
+
+// static
+bool LLVolumeParams::validate(U8 prof_curve, F32 prof_begin, F32 prof_end, F32 hollow,
+ U8 path_curve, F32 path_begin, F32 path_end,
+ F32 scx, F32 scy, F32 shx, F32 shy,
+ F32 twistend, F32 twistbegin, F32 radiusoffset,
+ F32 tx, F32 ty, F32 revolutions, F32 skew)
+{
+ LLVolumeParams test_params;
+ if (!test_params.setType (prof_curve, path_curve))
+ {
+ return false;
+ }
+ if (!test_params.setBeginAndEndS (prof_begin, prof_end))
+ {
+ return false;
+ }
+ if (!test_params.setBeginAndEndT (path_begin, path_end))
+ {
+ return false;
+ }
+ if (!test_params.setHollow (hollow))
+ {
+ return false;
+ }
+ if (!test_params.setTwistBegin (twistbegin))
+ {
+ return false;
+ }
+ if (!test_params.setTwistEnd (twistend))
+ {
+ return false;
+ }
+ if (!test_params.setRatio (scx, scy))
+ {
+ return false;
+ }
+ if (!test_params.setShear (shx, shy))
+ {
+ return false;
+ }
+ if (!test_params.setTaper (tx, ty))
+ {
+ return false;
+ }
+ if (!test_params.setRevolutions (revolutions))
+ {
+ return false;
+ }
+ if (!test_params.setRadiusOffset (radiusoffset))
+ {
+ return false;
+ }
+ if (!test_params.setSkew (skew))
+ {
+ return false;
+ }
+ 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)
+ {
+ // 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;
+
+ 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++] = (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;
+ }
+ }
+ }
+ }
+ 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;
+ }
+ }
+
+ 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
+ {
+ // 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
+
+ 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;
+ }
+ }
+ }
+
+#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)
+ {
+ 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);
+ }
+#endif
+
+ num_indices = count;
+ return index;
+}
+
+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;
+}
+
+
+S32 LLVolume::getNumTriangles() const
+{
+ U32 triangle_count = 0;
+
+ for (S32 i = 0; i < getNumVolumeFaces(); ++i)
+ {
+ triangle_count += getVolumeFace(i).mNumIndices/3;
+ }
+
+ return triangle_count;
+}
+
+
+//-----------------------------------------------------------------------------
+// generateSilhouetteVertices()
+//-----------------------------------------------------------------------------
+void LLVolume::generateSilhouetteVertices(std::vector<LLVector3> &vertices,
+ std::vector<LLVector3> &normals,
+ std::vector<S32> &segments,
+ 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)
+ {
+ LLVolumeFace& face = *iter;
+
+ if (!(face_mask & (0x1 << cur_index++)) ||
+ face.mNumIndices == 0 || face.mEdge.empty())
+ {
+ continue;
+ }
+
+ if (face.mTypeMask & (LLVolumeFace::CAP_MASK)) {
+
+ }
+ else {
+
+ //==============================================
+ //DEBUG draw edge map instead of silhouette edge
+ //==============================================
+
+#if DEBUG_SILHOUETTE_EDGE_MAP
+
+ //for each triangle
+ U32 count = face.mNumIndices;
+ for (U32 j = 0; j < count/3; j++) {
+ //get vertices
+ S32 v1 = face.mIndices[j*3+0];
+ S32 v2 = face.mIndices[j*3+1];
+ S32 v3 = face.mIndices[j*3+2];
+
+ //get current face center
+ 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++) {
+ S32 nIndex = face.mEdge[j*3+k];
+ if (nIndex <= -1) {
+ continue;
+ }
+
+ if (nIndex >= (S32) count/3) {
+ continue;
+ }
+ //get neighbor vertices
+ v1 = face.mIndices[nIndex*3+0];
+ v2 = face.mIndices[nIndex*3+1];
+ v3 = face.mIndices[nIndex*3+2];
+
+ //get neighbor face center
+ LLVector3 nCenter = (face.mVertices[v1].getPosition() +
+ face.mVertices[v2].getPosition() +
+ face.mVertices[v3].getPosition()) / 3.0f;
+
+ //draw line
+ vertices.push_back(cCenter);
+ vertices.push_back(nCenter);
+ normals.push_back(LLVector3(1,1,1));
+ normals.push_back(LLVector3(1,1,1));
+ segments.push_back(vertices.size());
+ }
+ }
+
+ continue;
+
+ //==============================================
+ //DEBUG
+ //==============================================
+
+ //==============================================
+ //DEBUG draw normals instead of silhouette edge
+ //==============================================
+#elif DEBUG_SILHOUETTE_NORMALS
+
+ //for each vertex
+ 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].getPosition());
+ vertices.push_back(face.mVertices[j].getPosition() + face.mVertices[j].mBinormal*0.1f);
+ normals.push_back(LLVector3(0,0,1));
+ normals.push_back(LLVector3(0,0,1));
+ segments.push_back(vertices.size());
+#endif
+ }
+
+ continue;
+#else
+ //==============================================
+ //DEBUG
+ //==============================================
+
+ static const U8 AWAY = 0x01,
+ TOWARDS = 0x02;
+
+ //for each triangle
+ std::vector<U8> fFacing;
+ 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];
+
+ 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
+ LLVector4a view;
+ view.setSub(obj_cam_vec, v[v1]);
+ bool away = view.dot3(norm) > 0.0f;
+ if (away)
+ {
+ fFacing[j] = AWAY;
+ }
+ else
+ {
+ fFacing[j] = TOWARDS;
+ }
+ }
+ }
+
+ //for each triangle
+ for (U32 j = 0; j < face.mNumIndices/3; j++)
+ {
+ if (fFacing[j] == (AWAY | TOWARDS))
+ { //this is a degenerate triangle
+ //take neighbor facing (degenerate faces get facing of one of their neighbors)
+ // *FIX IF NEEDED: this does not deal with neighboring degenerate faces
+ for (S32 k = 0; k < 3; k++)
+ {
+ S32 index = face.mEdge[j*3+k];
+ if (index != -1)
+ {
+ fFacing[j] = fFacing[index];
+ break;
+ }
+ }
+ continue; //skip degenerate face
+ }
+
+ //for each edge
+ for (S32 k = 0; k < 3; k++) {
+ S32 index = face.mEdge[j*3+k];
+ if (index != -1 && fFacing[index] == (AWAY | TOWARDS)) {
+ //our neighbor is degenerate, make him face our direction
+ fFacing[face.mEdge[j*3+k]] = fFacing[j];
+ continue;
+ }
+
+ if (index == -1 || //edge has no neighbor, MUST be a silhouette edge
+ (fFacing[index] & fFacing[j]) == 0) { //we found a silhouette edge
+
+ S32 v1 = face.mIndices[j*3+k];
+ S32 v2 = face.mIndices[j*3+((k+1)%3)];
+
+ LLVector4a t;
+ mat.affineTransform(v[v1], t);
+ vertices.push_back(LLVector3(t[0], t[1], t[2]));
+
+ norm_mat.rotate(n[v1], t);
+
+ t.normalize3fast();
+ normals.push_back(LLVector3(t[0], t[1], t[2]));
+
+ mat.affineTransform(v[v2], t);
+ vertices.push_back(LLVector3(t[0], t[1], t[2]));
+
+ norm_mat.rotate(n[v2], t);
+ t.normalize3fast();
+ normals.push_back(LLVector3(t[0], t[1], t[2]));
+
+ segments.push_back(vertices.size());
+ }
+ }
+ }
+#endif
+ }
+ }
+}
+
+S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end,
+ S32 face,
+ LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal)
+{
+ LLVector4a starta, enda;
+ starta.load3(start.mV);
+ enda.load3(end.mV);
+
+ return lineSegmentIntersect(starta, enda, face, intersection, tex_coord, normal, bi_normal);
+
+}
+
+
+S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end,
+ S32 face,
+ LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal)
+{
+ S32 hit_face = -1;
+
+ S32 start_face;
+ S32 end_face;
+
+ if (face == -1) // ALL_SIDES
+ {
+ start_face = 0;
+ end_face = getNumVolumeFaces() - 1;
+ }
+ else
+ {
+ start_face = face;
+ end_face = face;
+ }
+
+ 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++)
+ {
+ LLVolumeFace &face = mVolumeFaces[i];
+
+ 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
+ {
+ genBinormals(i);
+ }
+
+ if (!face.mOctree)
+ {
+ face.createOctree();
+ }
+
+ //LLVector4a* p = (LLVector4a*) face.mPositions;
+
+ LLOctreeTriangleRayIntersect intersect(start, dir, &face, &closest_t, intersection, tex_coord, normal, bi_normal);
+ intersect.traverse(face.mOctree);
+ if (intersect.mHitFace)
+ {
+ hit_face = i;
+ }
+ }
+ }
+
+
+ return hit_face;
+}
+
+class LLVertexIndexPair
+{
+public:
+ LLVertexIndexPair(const LLVector3 &vertex, const S32 index);
+
+ LLVector3 mVertex;
+ S32 mIndex;
+};
+
+LLVertexIndexPair::LLVertexIndexPair(const LLVector3 &vertex, const S32 index)
+{
+ mVertex = vertex;
+ mIndex = index;
+}
+
+const F32 VERTEX_SLOP = 0.00001f;
+const F32 VERTEX_SLOP_SQRD = VERTEX_SLOP * VERTEX_SLOP;
+
+struct lessVertex
+{
+ bool operator()(const LLVertexIndexPair *a, const LLVertexIndexPair *b)
+ {
+ const F32 slop = VERTEX_SLOP;
+
+ if (a->mVertex.mV[0] + slop < b->mVertex.mV[0])
+ {
+ return TRUE;
+ }
+ else if (a->mVertex.mV[0] - slop > b->mVertex.mV[0])
+ {
+ return FALSE;
+ }
+
+ if (a->mVertex.mV[1] + slop < b->mVertex.mV[1])
+ {
+ return TRUE;
+ }
+ else if (a->mVertex.mV[1] - slop > b->mVertex.mV[1])
+ {
+ return FALSE;
+ }
+
+ if (a->mVertex.mV[2] + slop < b->mVertex.mV[2])
+ {
+ return TRUE;
+ }
+ else if (a->mVertex.mV[2] - slop > b->mVertex.mV[2])
+ {
+ return FALSE;
+ }
+
+ return FALSE;
+ }
+};
+
+struct lessTriangle
+{
+ bool operator()(const S32 *a, const S32 *b)
+ {
+ if (*a < *b)
+ {
+ return TRUE;
+ }
+ else if (*a > *b)
+ {
+ return FALSE;
+ }
+
+ if (*(a+1) < *(b+1))
+ {
+ return TRUE;
+ }
+ else if (*(a+1) > *(b+1))
+ {
+ return FALSE;
+ }
+
+ if (*(a+2) < *(b+2))
+ {
+ return TRUE;
+ }
+ else if (*(a+2) > *(b+2))
+ {
+ return FALSE;
+ }
+
+ return FALSE;
+ }
+};
+
+BOOL equalTriangle(const S32 *a, const S32 *b)
+{
+ if ((*a == *b) && (*(a+1) == *(b+1)) && (*(a+2) == *(b+2)))
+ {
+ return TRUE;
+ }
+ 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;
+ const S32 BUFSIZE = 16384;
+ char buffer[BUFSIZE]; /* Flawfinder: ignore */
+ // *NOTE: changing the size or type of this buffer will require
+ // changing the sscanf below.
+ char keyword[256]; /* Flawfinder: ignore */
+ keyword[0] = 0;
+
+ while (!feof(fp))
+ {
+ if (fgets(buffer, BUFSIZE, fp) == NULL)
+ {
+ buffer[0] = '\0';
+ }
+
+ sscanf(buffer, " %255s", keyword); /* Flawfinder: ignore */
+ if (!strcmp("{", keyword))
+ {
+ continue;
+ }
+ if (!strcmp("}",keyword))
+ {
+ break;
+ }
+ else if (!strcmp("profile", keyword))
+ {
+ mProfileParams.importFile(fp);
+ }
+ else if (!strcmp("path",keyword))
+ {
+ mPathParams.importFile(fp);
+ }
+ else
+ {
+ llwarns << "unknown keyword " << keyword << " in volume import" << llendl;
+ }
+ }
+
+ return TRUE;
+}
+
+BOOL LLVolumeParams::exportFile(LLFILE *fp) const
+{
+ fprintf(fp,"\tshape 0\n");
+ fprintf(fp,"\t{\n");
+ mPathParams.exportFile(fp);
+ mProfileParams.exportFile(fp);
+ fprintf(fp, "\t}\n");
+ return TRUE;
+}
+
+
+BOOL LLVolumeParams::importLegacyStream(std::istream& input_stream)
+{
+ LLMemType m1(LLMemType::MTYPE_VOLUME);
+
+ //llinfos << "importing volume" << llendl;
+ const S32 BUFSIZE = 16384;
+ // *NOTE: changing the size or type of this buffer will require
+ // changing the sscanf below.
+ char buffer[BUFSIZE]; /* Flawfinder: ignore */
+ char keyword[256]; /* Flawfinder: ignore */
+ keyword[0] = 0;
+
+ while (input_stream.good())
+ {
+ input_stream.getline(buffer, BUFSIZE);
+ sscanf(buffer, " %255s", keyword);
+ if (!strcmp("{", keyword))
+ {
+ continue;
+ }
+ if (!strcmp("}",keyword))
+ {
+ break;
+ }
+ else if (!strcmp("profile", keyword))
+ {
+ mProfileParams.importLegacyStream(input_stream);
+ }
+ else if (!strcmp("path",keyword))
+ {
+ mPathParams.importLegacyStream(input_stream);
+ }
+ else
+ {
+ llwarns << "unknown keyword " << keyword << " in volume import" << llendl;
+ }
+ }
+
+ return TRUE;
+}
+
+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);
+ mProfileParams.exportLegacyStream(output_stream);
+ output_stream << "\t}\n";
+ 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;
+}
+
+bool LLVolumeParams::fromLLSD(LLSD& sd)
+{
+ mPathParams.fromLLSD(sd["path"]);
+ mProfileParams.fromLLSD(sd["profile"]);
+ sculptFromLLSD(sd["sculpt"]);
+
+ return true;
+}
+
+void LLVolumeParams::reduceS(F32 begin, F32 end)
+{
+ begin = llclampf(begin);
+ end = llclampf(end);
+ if (begin > end)
+ {
+ F32 temp = begin;
+ begin = end;
+ end = temp;
+ }
+ F32 a = mProfileParams.getBegin();
+ F32 b = mProfileParams.getEnd();
+ mProfileParams.setBegin(a + begin * (b - a));
+ mProfileParams.setEnd(a + end * (b - a));
+}
+
+void LLVolumeParams::reduceT(F32 begin, F32 end)
+{
+ begin = llclampf(begin);
+ end = llclampf(end);
+ if (begin > end)
+ {
+ F32 temp = begin;
+ begin = end;
+ end = temp;
+ }
+ F32 a = mPathParams.getBegin();
+ F32 b = mPathParams.getEnd();
+ mPathParams.setBegin(a + begin * (b - a));
+ mPathParams.setEnd(a + end * (b - a));
+}
+
+const F32 MIN_CONCAVE_PROFILE_WEDGE = 0.125f; // 1/8 unity
+const F32 MIN_CONCAVE_PATH_WEDGE = 0.111111f; // 1/9 unity
+
+// returns TRUE if the shape can be approximated with a convex shape
+// 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();
+
+ U8 path_type = mPathParams.getCurveType();
+ if ( path_length > MIN_CONCAVE_PATH_WEDGE
+ && ( mPathParams.getTwist() != mPathParams.getTwistBegin()
+ || (hollow > 0.f
+ && LL_PCODE_PATH_LINE != path_type) ) )
+ {
+ // twist along a "not too short" path is concave
+ return FALSE;
+ }
+
+ F32 profile_length = mProfileParams.getEnd() - mProfileParams.getBegin();
+ BOOL same_hole = hollow == 0.f
+ || (mProfileParams.getCurveType() & LL_PCODE_HOLE_MASK) == LL_PCODE_HOLE_SAME;
+
+ F32 min_profile_wedge = MIN_CONCAVE_PROFILE_WEDGE;
+ U8 profile_type = mProfileParams.getCurveType() & LL_PCODE_PROFILE_MASK;
+ if ( LL_PCODE_PROFILE_CIRCLE_HALF == profile_type )
+ {
+ // it is a sphere and spheres get twice the minimum profile wedge
+ min_profile_wedge = 2.f * MIN_CONCAVE_PROFILE_WEDGE;
+ }
+
+ BOOL convex_profile = ( ( profile_length == 1.f
+ || profile_length <= 0.5f )
+ && hollow == 0.f ) // trivially convex
+ || ( profile_length <= min_profile_wedge
+ && same_hole ); // effectvely convex (even when hollow)
+
+ if (!convex_profile)
+ {
+ // profile is concave
+ return FALSE;
+ }
+
+ if ( LL_PCODE_PATH_LINE == path_type )
+ {
+ // straight paths with convex profile
+ return TRUE;
+ }
+
+ BOOL concave_path = (path_length < 1.0f) && (path_length > 0.5f);
+ if (concave_path)
+ {
+ return FALSE;
+ }
+
+ // we're left with spheres, toroids and tubes
+ if ( LL_PCODE_PROFILE_CIRCLE_HALF == profile_type )
+ {
+ // at this stage all spheres must be convex
+ return TRUE;
+ }
+
+ // it's a toroid or tube
+ if ( path_length <= MIN_CONCAVE_PATH_WEDGE )
+ {
+ // effectively convex
+ return TRUE;
+ }
+
+ return FALSE;
+}
+
+// debug
+void LLVolumeParams::setCube()
+{
+ mProfileParams.setCurveType(LL_PCODE_PROFILE_SQUARE);
+ mProfileParams.setBegin(0.f);
+ mProfileParams.setEnd(1.f);
+ mProfileParams.setHollow(0.f);
+
+ mPathParams.setBegin(0.f);
+ mPathParams.setEnd(1.f);
+ mPathParams.setScale(1.f, 1.f);
+ mPathParams.setShear(0.f, 0.f);
+ mPathParams.setCurveType(LL_PCODE_PATH_LINE);
+ mPathParams.setTwistBegin(0.f);
+ mPathParams.setTwistEnd(0.f);
+ mPathParams.setRadiusOffset(0.f);
+ mPathParams.setTaper(0.f, 0.f);
+ mPathParams.setRevolutions(0.f);
+ mPathParams.setSkew(0.f);
+}
+
+LLFaceID LLVolume::generateFaceMask()
+{
+ LLFaceID new_mask = 0x0000;
+
+ switch(mParams.getProfileParams().getCurveType() & LL_PCODE_PROFILE_MASK)
+ {
+ case LL_PCODE_PROFILE_CIRCLE:
+ case LL_PCODE_PROFILE_CIRCLE_HALF:
+ new_mask |= LL_FACE_OUTER_SIDE_0;
+ break;
+ case LL_PCODE_PROFILE_SQUARE:
+ {
+ for(S32 side = (S32)(mParams.getProfileParams().getBegin() * 4.f); side < llceil(mParams.getProfileParams().getEnd() * 4.f); side++)
+ {
+ new_mask |= LL_FACE_OUTER_SIDE_0 << side;
+ }
+ }
+ break;
+ case LL_PCODE_PROFILE_ISOTRI:
+ case LL_PCODE_PROFILE_EQUALTRI:
+ case LL_PCODE_PROFILE_RIGHTTRI:
+ {
+ for(S32 side = (S32)(mParams.getProfileParams().getBegin() * 3.f); side < llceil(mParams.getProfileParams().getEnd() * 3.f); side++)
+ {
+ new_mask |= LL_FACE_OUTER_SIDE_0 << side;
+ }
+ }
+ break;
+ default:
+ llerrs << "Unknown profile!" << llendl;
+ break;
+ }
+
+ // handle hollow objects
+ if (mParams.getProfileParams().getHollow() > 0)
+ {
+ new_mask |= LL_FACE_INNER_SIDE;
+ }
+
+ // handle open profile curves
+ if (mProfilep->isOpen())
+ {
+ new_mask |= LL_FACE_PROFILE_BEGIN | LL_FACE_PROFILE_END;
+ }
+
+ // handle open path curves
+ if (mPathp->isOpen())
+ {
+ new_mask |= LL_FACE_PATH_BEGIN | LL_FACE_PATH_END;
+ }
+
+ return new_mask;
+}
+
+BOOL LLVolume::isFaceMaskValid(LLFaceID face_mask)
+{
+ LLFaceID test_mask = 0;
+ for(S32 i = 0; i < getNumFaces(); i++)
+ {
+ test_mask |= mProfilep->mFaces[i].mFaceID;
+ }
+
+ return test_mask == face_mask;
+}
+
+BOOL LLVolume::isConvex() const
+{
+ // mParams.isConvex() may return FALSE even though the final
+ // geometry is actually convex due to LOD approximations.
+ // TODO -- provide LLPath and LLProfile with isConvex() methods
+ // that correctly determine convexity. -- Leviathan
+ return mParams.isConvex();
+}
+
+
+std::ostream& operator<<(std::ostream &s, const LLProfileParams &profile_params)
+{
+ s << "{type=" << (U32) profile_params.mCurveType;
+ s << ", begin=" << profile_params.mBegin;
+ s << ", end=" << profile_params.mEnd;
+ s << ", hollow=" << profile_params.mHollow;
+ s << "}";
+ return s;
+}
+
+
+std::ostream& operator<<(std::ostream &s, const LLPathParams &path_params)
+{
+ s << "{type=" << (U32) path_params.mCurveType;
+ s << ", begin=" << path_params.mBegin;
+ s << ", end=" << path_params.mEnd;
+ s << ", twist=" << path_params.mTwistEnd;
+ s << ", scale=" << path_params.mScale;
+ s << ", shear=" << path_params.mShear;
+ s << ", twist_begin=" << path_params.mTwistBegin;
+ s << ", radius_offset=" << path_params.mRadiusOffset;
+ s << ", taper=" << path_params.mTaper;
+ s << ", revolutions=" << path_params.mRevolutions;
+ s << ", skew=" << path_params.mSkew;
+ s << "}";
+ return s;
+}
+
+
+std::ostream& operator<<(std::ostream &s, const LLVolumeParams &volume_params)
+{
+ s << "{profileparams = " << volume_params.mProfileParams;
+ s << ", pathparams = " << volume_params.mPathParams;
+ s << "}";
+ return s;
+}
+
+
+std::ostream& operator<<(std::ostream &s, const LLProfile &profile)
+{
+ s << " {open=" << (U32) profile.mOpen;
+ s << ", dirty=" << profile.mDirty;
+ s << ", totalout=" << profile.mTotalOut;
+ s << ", total=" << profile.mTotal;
+ s << "}";
+ return s;
+}
+
+
+std::ostream& operator<<(std::ostream &s, const LLPath &path)
+{
+ s << "{open=" << (U32) path.mOpen;
+ s << ", dirty=" << path.mDirty;
+ s << ", step=" << path.mStep;
+ s << ", total=" << path.mTotal;
+ s << "}";
+ return s;
+}
+
+std::ostream& operator<<(std::ostream &s, const LLVolume &volume)
+{
+ s << "{params = " << volume.getParams();
+ s << ", path = " << *volume.mPathp;
+ s << ", profile = " << *volume.mProfilep;
+ s << "}";
+ return s;
+}
+
+
+std::ostream& operator<<(std::ostream &s, const LLVolume *volumep)
+{
+ s << "{params = " << volumep->getParams();
+ s << ", path = " << *(volumep->mPathp);
+ s << ", profile = " << *(volumep->mProfilep);
+ s << "}";
+ return s;
+}
+
+LLVolumeFace::LLVolumeFace() :
+ mID(0),
+ mTypeMask(0),
+ mBeginS(0),
+ mBeginT(0),
+ mNumS(0),
+ mNumT(0),
+ mNumVertices(0),
+ mNumIndices(0),
+ mPositions(NULL),
+ mNormals(NULL),
+ mBinormals(NULL),
+ mTexCoords(NULL),
+ mIndices(NULL),
+ mWeights(NULL),
+ mOctree(NULL)
+{
+ mExtents = (LLVector4a*) malloc(sizeof(LLVector4a)*3);
+ mCenter = mExtents+2;
+}
+
+LLVolumeFace::LLVolumeFace(const LLVolumeFace& src)
+: mID(0),
+ mTypeMask(0),
+ mBeginS(0),
+ mBeginT(0),
+ mNumS(0),
+ mNumT(0),
+ mNumVertices(0),
+ mNumIndices(0),
+ mPositions(NULL),
+ mNormals(NULL),
+ mBinormals(NULL),
+ mTexCoords(NULL),
+ mIndices(NULL),
+ mWeights(NULL),
+ mOctree(NULL)
+{
+ mExtents = (LLVector4a*) malloc(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();
+
+ LLVector4a::memcpyNonAliased16((F32*) mExtents, (F32*) src.mExtents, 3*sizeof(LLVector4a));
+
+ resizeVertices(src.mNumVertices);
+ resizeIndices(src.mNumIndices);
+
+ if (mNumVertices)
+ {
+ S32 vert_size = mNumVertices*sizeof(LLVector4a);
+ S32 tc_size = (mNumVertices*sizeof(LLVector2)+0xF) & ~0xF;
+
+ LLVector4a::memcpyNonAliased16((F32*) mPositions, (F32*) src.mPositions, vert_size);
+ LLVector4a::memcpyNonAliased16((F32*) mNormals, (F32*) src.mNormals, vert_size);
+ LLVector4a::memcpyNonAliased16((F32*) mTexCoords, (F32*) src.mTexCoords, tc_size);
+
+
+ if (src.mBinormals)
+ {
+ allocateBinormals(src.mNumVertices);
+ LLVector4a::memcpyNonAliased16((F32*) mBinormals, (F32*) src.mBinormals, vert_size);
+ }
+ else
+ {
+ free(mBinormals);
+ mBinormals = NULL;
+ }
+
+ if (src.mWeights)
+ {
+ allocateWeights(src.mNumVertices);
+ LLVector4a::memcpyNonAliased16((F32*) mWeights, (F32*) src.mWeights, vert_size);
+ }
+ else
+ {
+ free(mWeights);
+ mWeights = NULL;
+ }
+ }
+
+ if (mNumIndices)
+ {
+ S32 idx_size = (mNumIndices*sizeof(U16)+0xF) & ~0xF;
+
+ LLVector4a::memcpyNonAliased16((F32*) mIndices, (F32*) src.mIndices, idx_size);
+ }
+
+ //delete
+ return *this;
+}
+
+LLVolumeFace::~LLVolumeFace()
+{
+ free(mExtents);
+ mExtents = NULL;
+
+ freeData();
+}
+
+void LLVolumeFace::freeData()
+{
+ free(mPositions);
+ mPositions = NULL;
+ free( mNormals);
+ mNormals = NULL;
+ free(mTexCoords);
+ mTexCoords = NULL;
+ free(mIndices);
+ mIndices = NULL;
+ free(mBinormals);
+ mBinormals = NULL;
+ free(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;
+
+ BOOL ret = FALSE ;
+ if (mTypeMask & CAP_MASK)
+ {
+ ret = createCap(volume, partial_build);
+ }
+ else if ((mTypeMask & END_MASK) || (mTypeMask & SIDE_MASK))
+ {
+ ret = createSide(volume, partial_build);
+ }
+ else
+ {
+ llerrs << "Unknown/uninitialized face type!" << llendl;
+ }
+
+ //update the range of the texture coordinates
+ if(ret)
+ {
+ mTexCoordExtents[0].setVec(1.f, 1.f) ;
+ mTexCoordExtents[1].setVec(0.f, 0.f) ;
+
+ for(U32 i = 0 ; i < mNumVertices ; i++)
+ {
+ if(mTexCoordExtents[0].mV[0] > mTexCoords[i].mV[0])
+ {
+ mTexCoordExtents[0].mV[0] = mTexCoords[i].mV[0] ;
+ }
+ if(mTexCoordExtents[1].mV[0] < mTexCoords[i].mV[0])
+ {
+ mTexCoordExtents[1].mV[0] = mTexCoords[i].mV[0] ;
+ }
+
+ if(mTexCoordExtents[0].mV[1] > mTexCoords[i].mV[1])
+ {
+ mTexCoordExtents[0].mV[1] = mTexCoords[i].mV[1] ;
+ }
+ if(mTexCoordExtents[1].mV[1] < mTexCoords[i].mV[1])
+ {
+ mTexCoordExtents[1].mV[1] = mTexCoords[i].mV[1] ;
+ }
+ }
+ mTexCoordExtents[0].mV[0] = llmax(0.f, mTexCoordExtents[0].mV[0]) ;
+ mTexCoordExtents[0].mV[1] = llmax(0.f, mTexCoordExtents[0].mV[1]) ;
+ mTexCoordExtents[1].mV[0] = llmin(1.f, mTexCoordExtents[1].mV[0]) ;
+ mTexCoordExtents[1].mV[1] = llmin(1.f, mTexCoordExtents[1].mV[1]) ;
+ }
+
+ return ret ;
+}
+
+void LLVolumeFace::getVertexData(U16 index, LLVolumeFace::VertexData& cv)
+{
+ cv.setPosition(mPositions[index]);
+ cv.setNormal(mNormals[index]);
+ cv.mTexCoord = mTexCoords[index];
+}
+
+bool LLVolumeFace::VertexMapData::operator==(const LLVolumeFace::VertexData& rhs) const
+{
+ return getPosition().equals3(rhs.getPosition()) &&
+ mTexCoord == rhs.mTexCoord &&
+ getNormal().equals3(rhs.getNormal());
+}
+
+bool LLVolumeFace::VertexMapData::ComparePosition::operator()(const LLVector3& a, const LLVector3& b) const
+{
+ if (a.mV[0] != b.mV[0])
+ {
+ return a.mV[0] < b.mV[0];
+ }
+
+ if (a.mV[1] != b.mV[1])
+ {
+ return a.mV[1] < b.mV[1];
+ }
+
+ return a.mV[2] < b.mV[2];
+}
+
+void LLVolumeFace::optimize(F32 angle_cutoff)
+{
+ LLVolumeFace new_face;
+
+ //map of points to vector of vertices at that point
+ VertexMapData::PointMap point_map;
+
+ //remove redundant vertices
+ for (U32 i = 0; i < mNumIndices; ++i)
+ {
+ U16 index = mIndices[i];
+
+ LLVolumeFace::VertexData cv;
+ getVertexData(index, cv);
+
+ BOOL found = FALSE;
+ VertexMapData::PointMap::iterator point_iter = point_map.find(LLVector3(cv.getPosition().getF32ptr()));
+ if (point_iter != point_map.end())
+ { //duplicate point might exist
+ for (U32 j = 0; j < point_iter->second.size(); ++j)
+ {
+ LLVolumeFace::VertexData& tv = (point_iter->second)[j];
+ if (tv.compareNormal(cv, angle_cutoff))
+ {
+ found = TRUE;
+ new_face.pushIndex((point_iter->second)[j].mIndex);
+ break;
+ }
+ }
+ }
+
+ if (!found)
+ {
+ new_face.pushVertex(cv);
+ U16 index = (U16) new_face.mNumVertices-1;
+ new_face.pushIndex(index);
+
+ VertexMapData d;
+ d.setPosition(cv.getPosition());
+ d.mTexCoord = cv.mTexCoord;
+ d.setNormal(cv.getNormal());
+ d.mIndex = index;
+ if (point_iter != point_map.end())
+ {
+ point_iter->second.push_back(d);
+ }
+ else
+ {
+ point_map[LLVector3(d.getPosition().getF32ptr())].push_back(d);
+ }
+ }
+ }
+
+ swapData(new_face);
+}
+
+class LLVCacheTriangleData;
+
+class LLVCacheVertexData
+{
+public:
+ S32 mIdx;
+ S32 mCacheTag;
+ F32 mScore;
+ U32 mActiveTriangles;
+ std::vector<LLVCacheTriangleData*> mTriangles;
+
+ LLVCacheVertexData()
+ {
+ mCacheTag = -1;
+ mScore = 0.f;
+ mActiveTriangles = 0;
+ mIdx = -1;
+ }
+};
+
+class LLVCacheTriangleData
+{
+public:
+ bool mActive;
+ F32 mScore;
+ LLVCacheVertexData* mVertex[3];
+
+ LLVCacheTriangleData()
+ {
+ mActive = true;
+ mScore = 0.f;
+ mVertex[0] = mVertex[1] = mVertex[2] = NULL;
+ }
+
+ void complete()
+ {
+ mActive = false;
+ for (S32 i = 0; i < 3; ++i)
+ {
+ if (mVertex[i])
+ {
+ llassert_always(mVertex[i]->mActiveTriangles > 0);
+ mVertex[i]->mActiveTriangles--;
+ }
+ }
+ }
+
+ bool operator<(const LLVCacheTriangleData& rhs) const
+ { //highest score first
+ return rhs.mScore < mScore;
+ }
+};
+
+const F32 FindVertexScore_CacheDecayPower = 1.5f;
+const F32 FindVertexScore_LastTriScore = 0.75f;
+const F32 FindVertexScore_ValenceBoostScale = 2.0f;
+const F32 FindVertexScore_ValenceBoostPower = 0.5f;
+const U32 MaxSizeVertexCache = 32;
+
+F32 find_vertex_score(LLVCacheVertexData& data)
+{
+ if (data.mActiveTriangles == 0)
+ { //no triangle references this vertex
+ return -1.f;
+ }
+
+ F32 score = 0.f;
+
+ S32 cache_idx = data.mCacheTag;
+
+ if (cache_idx < 0)
+ {
+ //not in cache
+ }
+ else
+ {
+ if (cache_idx < 3)
+ { //vertex was in the last triangle
+ score = FindVertexScore_LastTriScore;
+ }
+ else
+ { //more points for being higher in the cache
+ F32 scaler = 1.f/(MaxSizeVertexCache-3);
+ score = 1.f-((cache_idx-3)*scaler);
+ score = powf(score, FindVertexScore_CacheDecayPower);
+ }
+ }
+
+ //bonus points for having low valence
+ F32 valence_boost = powf(data.mActiveTriangles, -FindVertexScore_ValenceBoostPower);
+ score += FindVertexScore_ValenceBoostScale * valence_boost;
+
+ return score;
+}
+
+class LLVCacheFIFO
+{
+public:
+ LLVCacheVertexData* mCache[MaxSizeVertexCache];
+ U32 mMisses;
+
+ LLVCacheFIFO()
+ {
+ mMisses = 0;
+ for (U32 i = 0; i < MaxSizeVertexCache; ++i)
+ {
+ mCache[i] = NULL;
+ }
+ }
+
+ void addVertex(LLVCacheVertexData* data)
+ {
+ if (data->mCacheTag == -1)
+ {
+ mMisses++;
+
+ S32 end = MaxSizeVertexCache-1;
+
+ if (mCache[end])
+ {
+ mCache[end]->mCacheTag = -1;
+ }
+
+ for (S32 i = end; i > 0; --i)
+ {
+ mCache[i] = mCache[i-1];
+ if (mCache[i])
+ {
+ mCache[i]->mCacheTag = i;
+ }
+ }
+
+ mCache[0] = data;
+ data->mCacheTag = 0;
+ }
+ }
+};
+
+class LLVCacheLRU
+{
+public:
+ LLVCacheVertexData* mCache[MaxSizeVertexCache+3];
+
+ LLVCacheTriangleData* mBestTriangle;
+
+ U32 mMisses;
+
+ LLVCacheLRU()
+ {
+ for (U32 i = 0; i < MaxSizeVertexCache+3; ++i)
+ {
+ mCache[i] = NULL;
+ }
+
+ mBestTriangle = NULL;
+ mMisses = 0;
+ }
+
+ void addVertex(LLVCacheVertexData* data)
+ {
+ S32 end = MaxSizeVertexCache+2;
+ if (data->mCacheTag != -1)
+ { //just moving a vertex to the front of the cache
+ end = data->mCacheTag;
+ }
+ else
+ {
+ mMisses++;
+ if (mCache[end])
+ { //adding a new vertex, vertex at end of cache falls off
+ mCache[end]->mCacheTag = -1;
+ }
+ }
+
+ for (S32 i = end; i > 0; --i)
+ { //adjust cache pointers and tags
+ mCache[i] = mCache[i-1];
+
+ if (mCache[i])
+ {
+ mCache[i]->mCacheTag = i;
+ }
+ }
+
+ mCache[0] = data;
+ mCache[0]->mCacheTag = 0;
+ }
+
+ void addTriangle(LLVCacheTriangleData* data)
+ {
+ addVertex(data->mVertex[0]);
+ addVertex(data->mVertex[1]);
+ addVertex(data->mVertex[2]);
+ }
+
+ void updateScores()
+ {
+ for (U32 i = MaxSizeVertexCache; i < MaxSizeVertexCache+3; ++i)
+ { //trailing 3 vertices aren't actually in the cache for scoring purposes
+ if (mCache[i])
+ {
+ mCache[i]->mCacheTag = -1;
+ }
+ }
+
+ for (U32 i = 0; i < MaxSizeVertexCache; ++i)
+ { //update scores of vertices in cache
+ if (mCache[i])
+ {
+ mCache[i]->mScore = find_vertex_score(*(mCache[i]));
+ llassert_always(mCache[i]->mCacheTag == i);
+ }
+ }
+
+ mBestTriangle = NULL;
+ //update triangle scores
+ for (U32 i = 0; i < MaxSizeVertexCache+3; ++i)
+ {
+ if (mCache[i])
+ {
+ for (U32 j = 0; j < mCache[i]->mTriangles.size(); ++j)
+ {
+ LLVCacheTriangleData* tri = mCache[i]->mTriangles[j];
+ if (tri->mActive)
+ {
+ tri->mScore = tri->mVertex[0]->mScore;
+ tri->mScore += tri->mVertex[1]->mScore;
+ tri->mScore += tri->mVertex[2]->mScore;
+
+ if (!mBestTriangle || mBestTriangle->mScore < tri->mScore)
+ {
+ mBestTriangle = tri;
+ }
+ }
+ }
+ }
+ }
+
+ //knock trailing 3 vertices off the cache
+ for (U32 i = MaxSizeVertexCache; i < MaxSizeVertexCache+3; ++i)
+ {
+ if (mCache[i])
+ {
+ llassert_always(mCache[i]->mCacheTag == -1);
+ mCache[i] = NULL;
+ }
+ }
+ }
+};
+
+
+void LLVolumeFace::cacheOptimize()
+{ //optimize for vertex cache according to Forsyth method:
+ // http://home.comcast.net/~tom_forsyth/papers/fast_vert_cache_opt.html
+
+ LLVCacheLRU cache;
+
+ //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)
+ vertex_data[i].mScore = find_vertex_score(vertex_data[i]);
+ vertex_data[i].mActiveTriangles = vertex_data[i].mTriangles.size();
+
+ for (U32 j = 0; j < vertex_data[i].mTriangles.size(); ++j)
+ {
+ vertex_data[i].mTriangles[j]->mScore += vertex_data[i].mScore;
+ }
+ }
+
+ //sort triangle data by score
+ std::sort(triangle_data.begin(), triangle_data.end());
+
+ std::vector<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;
+ LLVector4a* pos = (LLVector4a*) malloc(sizeof(LLVector4a)*num_verts);
+ LLVector4a* norm = (LLVector4a*) malloc(sizeof(LLVector4a)*num_verts);
+ S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF;
+ LLVector2* tc = (LLVector2*) malloc(size);
+
+ LLVector4a* wght = NULL;
+ if (mWeights)
+ {
+ wght = (LLVector4a*) malloc(sizeof(LLVector4a)*num_verts);
+ }
+
+ LLVector4a* binorm = NULL;
+ if (mBinormals)
+ {
+ binorm = (LLVector4a*) malloc(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 (mBinormals)
+ {
+ binorm[cur_idx] = mBinormals[idx];
+ }
+
+ cur_idx++;
+ }
+ }
+
+ for (U32 i = 0; i < mNumIndices; ++i)
+ {
+ mIndices[i] = new_idx[mIndices[i]];
+ }
+
+ free(mPositions);
+ free(mNormals);
+ free(mTexCoords);
+ free(mWeights);
+ free(mBinormals);
+
+ mPositions = pos;
+ mNormals = norm;
+ mTexCoords = tc;
+ mWeights = wght;
+ mBinormals = binorm;
+
+ //std::string result = llformat("ACMR pre/post: %.3f/%.3f -- %d triangles %d breaks", pre_acmr, post_acmr, mNumIndices/3, breaks);
+ //llinfos << result << llendl;
+
+}
+
+void LLVolumeFace::createOctree(F32 scaler, const LLVector4a& center, const LLVector4a& size)
+{
+ if (mOctree)
+ {
+ return;
+ }
+
+ mOctree = new LLOctreeRoot<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.mBinormals, mBinormals);
+ llswap(rhs.mTexCoords, mTexCoords);
+ llswap(rhs.mIndices,mIndices);
+ llswap(rhs.mNumVertices, mNumVertices);
+ llswap(rhs.mNumIndices, mNumIndices);
+}
+
+void LerpPlanarVertex(LLVolumeFace::VertexData& v0,
+ LLVolumeFace::VertexData& v1,
+ LLVolumeFace::VertexData& v2,
+ LLVolumeFace::VertexData& vout,
+ F32 coef01,
+ F32 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.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;
+ 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;
+
+ 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(S32 t = 0; t < 4; t++)
+ {
+ corners[t].getPosition().load3( mesh[offset + (grid_size*t)].mPos.mV);
+ corners[t].mTexCoord.mV[0] = profile[grid_size*t].mV[0]+0.5f;
+ corners[t].mTexCoord.mV[1] = 0.5f - profile[grid_size*t].mV[1];
+ }
+
+ {
+ LLVector4a lhs;
+ lhs.setSub(corners[1].getPosition(), corners[0].getPosition());
+ LLVector4a rhs;
+ rhs.setSub(corners[2].getPosition(), corners[1].getPosition());
+ baseVert.getNormal().setCross3(lhs, rhs);
+ baseVert.getNormal().normalize3fast();
+ }
+
+ if(!(mTypeMask & TOP_MASK))
+ {
+ baseVert.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;
+ }
+
+ LLVector4a binormal;
+
+ calc_binormal_from_triangle( binormal,
+ corners[0].getPosition(), corners[0].mTexCoord,
+ corners[1].getPosition(), corners[1].mTexCoord,
+ corners[2].getPosition(), corners[2].mTexCoord);
+
+ binormal.normalize3fast();
+
+ S32 size = (grid_size+1)*(grid_size+1);
+ resizeVertices(size);
+ allocateBinormals(size);
+
+ LLVector4a* pos = (LLVector4a*) mPositions;
+ LLVector4a* norm = (LLVector4a*) mNormals;
+ LLVector4a* binorm = (LLVector4a*) mBinormals;
+ LLVector2* tc = (LLVector2*) mTexCoords;
+
+ for(int gx = 0;gx<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);
+
+ *pos++ = newVert.getPosition();
+ *norm++ = baseVert.getNormal();
+ *tc++ = newVert.mTexCoord;
+ *binorm++ = binormal;
+
+ if (gx == 0 && gy == 0)
+ {
+ min = newVert.getPosition();
+ max = min;
+ }
+ else
+ {
+ min.setMin(min, newVert.getPosition());
+ max.setMax(max, newVert.getPosition());
+ }
+ }
+ }
+
+ mCenter->setAdd(min, max);
+ mCenter->mul(0.5f);
+ }
+
+ if (!partial_build)
+ {
+ 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++)
+ {
+
+ for(S32 gy = 0;gy<grid_size;gy++)
+ {
+ if (mTypeMask & TOP_MASK)
+ {
+ for(S32 i=5;i>=0;i--)
+ {
+ *out++ = ((gy*(grid_size+1))+gx+idxs[i]);
+ }
+ }
+ else
+ {
+ for(S32 i=0;i<6;i++)
+ {
+ *out++ = ((gy*(grid_size+1))+gx+idxs[i]);
+ }
+ }
+ }
+ }
+ }
+
+ 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)&&
+ (volume->getParams().getPathParams().getEnd()==1.0f))&&
+ (volume->getParams().getProfileParams().getCurveType()==LL_PCODE_PROFILE_SQUARE &&
+ volume->getParams().getPathParams().getCurveType()==LL_PCODE_PATH_LINE)
+ ){
+ return createUnCutCubeCap(volume, 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;
+
+ // All types of caps have the same number of vertices and indices
+ num_vertices = profile.size();
+ num_indices = (profile.size() - 2)*3;
+
+ if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK))
+ {
+ resizeVertices(num_vertices+1);
+ allocateBinormals(num_vertices+1);
+
+ if (!partial_build)
+ {
+ resizeIndices(num_indices+3);
+ }
+ }
+ else
+ {
+ resizeVertices(num_vertices);
+ allocateBinormals(num_vertices);
+
+ if (!partial_build)
+ {
+ resizeIndices(num_indices);
+ }
+ }
+
+ S32 max_s = volume->getProfile().getTotal();
+ S32 max_t = volume->getPath().mPath.size();
+
+ mCenter->clear();
+
+ S32 offset = 0;
+ if (mTypeMask & TOP_MASK)
+ {
+ offset = (max_t-1) * max_s;
+ }
+ else
+ {
+ offset = mBeginS;
+ }
+
+ // Figure out the normal, assume all caps are flat faces.
+ // Cross product to get normals.
+
+ LLVector2 cuv;
+ LLVector2 min_uv, max_uv;
+
+ LLVector4a& min = mExtents[0];
+ LLVector4a& max = mExtents[1];
+
+ LLVector2* tc = (LLVector2*) mTexCoords;
+ LLVector4a* pos = (LLVector4a*) mPositions;
+ LLVector4a* norm = (LLVector4a*) mNormals;
+ LLVector4a* binorm = (LLVector4a*) mBinormals;
+
+ // Copy the vertices into the array
+ for (S32 i = 0; i < num_vertices; i++)
+ {
+ if (mTypeMask & TOP_MASK)
+ {
+ tc[i].mV[0] = profile[i].mV[0]+0.5f;
+ tc[i].mV[1] = profile[i].mV[1]+0.5f;
+ }
+ else
+ {
+ // Mirror for underside.
+ tc[i].mV[0] = profile[i].mV[0]+0.5f;
+ tc[i].mV[1] = 0.5f - profile[i].mV[1];
+ }
+
+ pos[i].load3(mesh[i + offset].mPos.mV);
+
+ if (i == 0)
+ {
+ max = pos[i];
+ min = max;
+ min_uv = max_uv = tc[i];
+ }
+ else
+ {
+ update_min_max(min,max,pos[i]);
+ update_min_max(min_uv, max_uv, tc[i]);
+ }
+ }
+
+ mCenter->setAdd(min, max);
+ mCenter->mul(0.5f);
+
+ cuv = (min_uv + max_uv)*0.5f;
+
+ LLVector4a binormal;
+ calc_binormal_from_triangle(binormal,
+ *mCenter, cuv,
+ pos[0], tc[0],
+ pos[1], tc[1]);
+ binormal.normalize3fast();
+
+ LLVector4a normal;
+ LLVector4a d0, d1;
+
+
+ d0.setSub(*mCenter, pos[0]);
+ d1.setSub(*mCenter, pos[1]);
+
+ if (mTypeMask & TOP_MASK)
+ {
+ normal.setCross3(d0, d1);
+ }
+ else
+ {
+ normal.setCross3(d1, d0);
+ }
+
+ normal.normalize3fast();
+
+ VertexData vd;
+ vd.setPosition(*mCenter);
+ vd.mTexCoord = cuv;
+
+ if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK))
+ {
+ pos[num_vertices] = *mCenter;
+ tc[num_vertices] = cuv;
+ num_vertices++;
+ }
+
+ for (S32 i = 0; i < num_vertices; i++)
+ {
+ binorm[i].load4a(binormal.getF32ptr());
+ norm[i].load4a(normal.getF32ptr());
+ }
+
+ if (partial_build)
+ {
+ return TRUE;
+ }
+
+ if (mTypeMask & HOLLOW_MASK)
+ {
+ if (mTypeMask & TOP_MASK)
+ {
+ // HOLLOW TOP
+ // Does it matter if it's open or closed? - djs
+
+ S32 pt1 = 0, pt2 = num_vertices - 1;
+ S32 i = 0;
+ while (pt2 - pt1 > 1)
+ {
+ // 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];
+
+ 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)
+ {
+ mIndices[i++] = pt1;
+ mIndices[i++] = pt1 + 1;
+ mIndices[i++] = pt2;
+ pt1++;
+ }
+ else
+ {
+ mIndices[i++] = pt1;
+ mIndices[i++] = pt2 - 1;
+ mIndices[i++] = pt2;
+ pt2--;
+ }
+ }
+ }
+ else
+ {
+ // HOLLOW BOTTOM
+ // Does it matter if it's open or closed? - djs
+
+ llassert(mTypeMask & BOTTOM_MASK);
+ S32 pt1 = 0, pt2 = num_vertices - 1;
+
+ S32 i = 0;
+ while (pt2 - pt1 > 1)
+ {
+ // 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];
+
+ 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;
+ }
+ }
+
+ // Flipped backfacing from top
+ if (use_tri1a2)
+ {
+ mIndices[i++] = pt1;
+ mIndices[i++] = pt2;
+ mIndices[i++] = pt1 + 1;
+ pt1++;
+ }
+ else
+ {
+ mIndices[i++] = pt1;
+ mIndices[i++] = pt2;
+ mIndices[i++] = pt2 - 1;
+ pt2--;
+ }
+ }
+ }
+ }
+ else
+ {
+ // Not hollow, generate the triangle fan.
+ U16 v1 = 2;
+ U16 v2 = 1;
+
+ if (mTypeMask & TOP_MASK)
+ {
+ v1 = 1;
+ v2 = 2;
+ }
+
+ for (S32 i = 0; i < (num_vertices - 2); i++)
+ {
+ mIndices[3*i] = num_vertices - 1;
+ mIndices[3*i+v1] = i;
+ mIndices[3*i+v2] = i + 1;
+ }
+
+
+ }
+
+ return TRUE;
+}
+
+void LLVolumeFace::createBinormals()
+{
+ LLMemType m1(LLMemType::MTYPE_VOLUME);
+
+ if (!mBinormals)
+ {
+ allocateBinormals(mNumVertices);
+
+ //generate binormals
+ LLVector4a* pos = mPositions;
+ LLVector2* tc = (LLVector2*) mTexCoords;
+ LLVector4a* binorm = (LLVector4a*) mBinormals;
+
+ LLVector4a* end = mBinormals+mNumVertices;
+ while (binorm < end)
+ {
+ (*binorm++).clear();
+ }
+
+ binorm = mBinormals;
+
+ for (U32 i = 0; i < mNumIndices/3; i++)
+ { //for each triangle
+ const U16& i0 = mIndices[i*3+0];
+ const U16& i1 = mIndices[i*3+1];
+ const U16& i2 = mIndices[i*3+2];
+
+ //calculate binormal
+ LLVector4a binormal;
+ calc_binormal_from_triangle(binormal,
+ pos[i0], tc[i0],
+ pos[i1], tc[i1],
+ pos[i2], tc[i2]);
+
+
+ //add triangle normal to vertices
+ binorm[i0].add(binormal);
+ binorm[i1].add(binormal);
+ binorm[i2].add(binormal);
+
+ //even out quad contributions
+ if (i % 2 == 0)
+ {
+ binorm[i2].add(binormal);
+ }
+ else
+ {
+ binorm[i1].add(binormal);
+ }
+ }
+
+ //normalize binormals
+ for (U32 i = 0; i < mNumVertices; i++)
+ {
+ binorm[i].normalize3fast();
+ //bump map/planar projection code requires normals to be normalized
+ mNormals[i].normalize3fast();
+ }
+ }
+}
+
+void LLVolumeFace::resizeVertices(S32 num_verts)
+{
+ free(mPositions);
+ free(mNormals);
+ free(mBinormals);
+ free(mTexCoords);
+
+ mBinormals = NULL;
+
+ if (num_verts)
+ {
+ mPositions = (LLVector4a*) malloc(sizeof(LLVector4a)*num_verts);
+ assert_aligned(mPositions, 16);
+ mNormals = (LLVector4a*) malloc(sizeof(LLVector4a)*num_verts);
+ assert_aligned(mNormals, 16);
+
+ //pad texture coordinate block end to allow for QWORD reads
+ S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF;
+ mTexCoords = (LLVector2*) malloc(size);
+ assert_aligned(mTexCoords, 16);
+ }
+ else
+ {
+ mPositions = NULL;
+ mNormals = NULL;
+ mTexCoords = NULL;
+ }
+
+ mNumVertices = num_verts;
+}
+
+void LLVolumeFace::pushVertex(const LLVolumeFace::VertexData& cv)
+{
+ pushVertex(cv.getPosition(), cv.getNormal(), cv.mTexCoord);
+}
+
+void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, const LLVector2& tc)
+{
+ S32 new_verts = mNumVertices+1;
+ S32 new_size = new_verts*16;
+// S32 old_size = mNumVertices*16;
+
+ //positions
+ mPositions = (LLVector4a*) realloc(mPositions, new_size);
+
+ //normals
+ mNormals = (LLVector4a*) realloc(mNormals, new_size);
+
+ //tex coords
+ new_size = ((new_verts*8)+0xF) & ~0xF;
+ mTexCoords = (LLVector2*) realloc(mTexCoords, new_size);
+
+
+ //just clear binormals
+ free(mBinormals);
+ mBinormals = NULL;
+
+ mPositions[mNumVertices] = pos;
+ mNormals[mNumVertices] = norm;
+ mTexCoords[mNumVertices] = tc;
+
+ mNumVertices++;
+}
+
+void LLVolumeFace::allocateBinormals(S32 num_verts)
+{
+ free(mBinormals);
+ mBinormals = (LLVector4a*) malloc(sizeof(LLVector4a)*num_verts);
+}
+
+void LLVolumeFace::allocateWeights(S32 num_verts)
+{
+ free(mWeights);
+ mWeights = (LLVector4a*) malloc(sizeof(LLVector4a)*num_verts);
+}
+
+void LLVolumeFace::resizeIndices(S32 num_indices)
+{
+ free(mIndices);
+
+ if (num_indices)
+ {
+ //pad index block end to allow for QWORD reads
+ S32 size = ((num_indices*sizeof(U16)) + 0xF) & ~0xF;
+
+ mIndices = (U16*) malloc(size);
+ }
+ else
+ {
+ mIndices = NULL;
+ }
+
+ mNumIndices = num_indices;
+}
+
+void LLVolumeFace::pushIndex(const U16& idx)
+{
+ S32 new_count = mNumIndices + 1;
+ S32 new_size = ((new_count*2)+0xF) & ~0xF;
+
+ S32 old_size = ((mNumIndices*2)+0xF) & ~0xF;
+ if (new_size != old_size)
+ {
+ mIndices = (U16*) realloc(mIndices, new_size);
+ }
+
+ mIndices[mNumIndices++] = idx;
+}
+
+void LLVolumeFace::fillFromLegacyData(std::vector<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)
+ {
+ llerrs << "Cannot append face -- 16-bit overflow will occur." << llendl;
+ }
+
+ if (face.mNumVertices == 0)
+ {
+ llerrs << "Cannot append empty face." << llendl;
+ }
+
+ //allocate new buffer space
+ mPositions = (LLVector4a*) realloc(mPositions, new_count*sizeof(LLVector4a));
+ assert_aligned(mPositions, 16);
+ mNormals = (LLVector4a*) realloc(mNormals, new_count*sizeof(LLVector4a));
+ assert_aligned(mNormals, 16);
+ mTexCoords = (LLVector2*) realloc(mTexCoords, (new_count*sizeof(LLVector2)+0xF) & ~0xF);
+ assert_aligned(mTexCoords, 16);
+
+ mNumVertices = new_count;
+
+ //get destination address of appended face
+ LLVector4a* dst_pos = mPositions+offset;
+ LLVector2* dst_tc = mTexCoords+offset;
+ LLVector4a* dst_norm = mNormals+offset;
+
+ //get source addresses of appended face
+ const LLVector4a* src_pos = face.mPositions;
+ const LLVector2* src_tc = face.mTexCoords;
+ const LLVector4a* src_norm = face.mNormals;
+
+ //load aligned matrices
+ LLMatrix4a mat, norm_mat;
+ mat.loadu(mat_in);
+ norm_mat.loadu(norm_mat_in);
+
+ for (U32 i = 0; i < face.mNumVertices; ++i)
+ {
+ //transform appended face position and store
+ mat.affineTransform(src_pos[i], dst_pos[i]);
+
+ //transform appended face normal and store
+ norm_mat.rotate(src_norm[i], dst_norm[i]);
+ dst_norm[i].normalize3fast();
+
+ //copy appended face texture coordinate
+ dst_tc[i] = src_tc[i];
+
+ if (offset == 0 && i == 0)
+ { //initialize bounding box
+ mExtents[0] = mExtents[1] = dst_pos[i];
+ }
+ else
+ {
+ //stretch bounding box
+ update_min_max(mExtents[0], mExtents[1], dst_pos[i]);
+ }
+ }
+
+
+ new_count = mNumIndices + face.mNumIndices;
+
+ //allocate new index buffer
+ mIndices = (U16*) realloc(mIndices, (new_count*sizeof(U16)+0xF) & ~0xF);
+
+ //get destination address into new index buffer
+ U16* dst_idx = mIndices+mNumIndices;
+ mNumIndices = new_count;
+
+ for (U32 i = 0; i < 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);
+
+ BOOL flat = mTypeMask & FLAT_MASK;
+
+ U8 sculpt_type = volume->getParams().getSculptType();
+ U8 sculpt_stitching = sculpt_type & LL_SCULPT_TYPE_MASK;
+ BOOL sculpt_invert = sculpt_type & LL_SCULPT_FLAG_INVERT;
+ BOOL sculpt_mirror = sculpt_type & LL_SCULPT_FLAG_MIRROR;
+ BOOL sculpt_reverse_horizontal = (sculpt_invert ? !sculpt_mirror : sculpt_mirror); // XOR
+
+ 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;
+
+ S32 max_s = volume->getProfile().getTotal();
+
+ S32 s, t, i;
+ F32 ss, tt;
+
+ num_vertices = mNumS*mNumT;
+ num_indices = (mNumS-1)*(mNumT-1)*6;
+
+ if (!partial_build)
+ {
+ resizeVertices(num_vertices);
+ resizeIndices(num_indices);
+
+ if ((volume->getParams().getSculptType() & LL_SCULPT_TYPE_MASK) != LL_SCULPT_TYPE_MESH)
+ {
+ mEdge.resize(num_indices);
+ }
+ }
+
+ LLVector4a* pos = (LLVector4a*) mPositions;
+ LLVector4a* norm = (LLVector4a*) mNormals;
+ LLVector2* tc = (LLVector2*) mTexCoords;
+ S32 begin_stex = llfloor( profile[mBeginS].mV[2] );
+ S32 num_s = ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2) ? mNumS/2 : mNumS;
+
+ S32 cur_vertex = 0;
+ // Copy the vertices into the array
+ for (t = mBeginT; t < mBeginT + mNumT; t++)
+ {
+ tt = path_data[t].mTexT;
+ for (s = 0; s < num_s; s++)
+ {
+ if (mTypeMask & END_MASK)
+ {
+ if (s)
+ {
+ ss = 1.f;
+ }
+ else
+ {
+ ss = 0.f;
+ }
+ }
+ else
+ {
+ // Get s value for tex-coord.
+ if (!flat)
+ {
+ ss = profile[mBeginS + s].mV[2];
+ }
+ else
+ {
+ ss = profile[mBeginS + s].mV[2] - begin_stex;
+ }
+ }
+
+ if (sculpt_reverse_horizontal)
+ {
+ ss = 1.f - ss;
+ }
+
+ // Check to see if this triangle wraps around the array.
+ if (mBeginS + s >= max_s)
+ {
+ // We're wrapping
+ i = mBeginS + s + max_s*(t-1);
+ }
+ else
+ {
+ i = mBeginS + s + max_s*t;
+ }
+
+ pos[cur_vertex].load3(mesh[i].mPos.mV);
+ tc[cur_vertex] = LLVector2(ss,tt);
+
+ norm[cur_vertex].clear();
+ cur_vertex++;
+
+ if ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2 && s > 0)
+ {
+
+ pos[cur_vertex].load3(mesh[i].mPos.mV);
+ tc[cur_vertex] = LLVector2(ss,tt);
+
+ norm[cur_vertex].clear();
+
+ cur_vertex++;
+ }
+ }
+
+ if ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2)
+ {
+ if (mTypeMask & OPEN_MASK)
+ {
+ s = num_s-1;
+ }
+ else
+ {
+ s = 0;
+ }
+
+ i = mBeginS + s + max_s*t;
+ ss = profile[mBeginS + s].mV[2] - begin_stex;
+ pos[cur_vertex].load3(mesh[i].mPos.mV);
+ tc[cur_vertex] = LLVector2(ss,tt);
+ norm[cur_vertex].clear();
+
+ cur_vertex++;
+ }
+ }
+
+
+ //get bounding box for this side
+ LLVector4a& face_min = mExtents[0];
+ LLVector4a& face_max = mExtents[1];
+ mCenter->clear();
+
+ face_min = face_max = pos[0];
+
+ for (U32 i = 1; i < mNumVertices; ++i)
+ {
+ update_min_max(face_min, face_max, pos[i]);
+ }
+
+ mCenter->setAdd(face_min, face_max);
+ mCenter->mul(0.5f);
+
+ S32 cur_index = 0;
+ S32 cur_edge = 0;
+ BOOL flat_face = mTypeMask & FLAT_MASK;
+
+ if (!partial_build)
+ {
+ // Now we generate the indices.
+ for (t = 0; t < (mNumT-1); t++)
+ {
+ for (s = 0; s < (mNumS-1); s++)
+ {
+ mIndices[cur_index++] = s + mNumS*t; //bottom left
+ mIndices[cur_index++] = s+1 + mNumS*(t+1); //top right
+ mIndices[cur_index++] = s + mNumS*(t+1); //top left
+ mIndices[cur_index++] = s + mNumS*t; //bottom left
+ mIndices[cur_index++] = s+1 + mNumS*t; //bottom right
+ mIndices[cur_index++] = s+1 + mNumS*(t+1); //top right
+
+ 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;
+ }
+ else if (mNumT <= 3 || volume->getPath().isOpen() == TRUE) { //no neighbor
+ mEdge[cur_edge++] = -1;
+ }
+ else { //wrap on T
+ mEdge[cur_edge++] = s*2+1;
+ }
+ if (s > 0) { //top left/bottom left neighbor face
+ mEdge[cur_edge++] = (mNumS-1)*2*t+s*2-1;
+ }
+ else if (flat_face || volume->getProfile().isOpen() == TRUE) { //no neighbor
+ mEdge[cur_edge++] = -1;
+ }
+ else { //wrap on S
+ mEdge[cur_edge++] = (mNumS-1)*2*t+(mNumS-2)*2+1;
+ }
+
+ if (t > 0) { //bottom left/bottom right neighbor face
+ mEdge[cur_edge++] = (mNumS-1)*2*(t-1)+s*2;
+ }
+ else if (mNumT <= 3 || volume->getPath().isOpen() == TRUE) { //no neighbor
+ mEdge[cur_edge++] = -1;
+ }
+ else { //wrap on T
+ mEdge[cur_edge++] = (mNumS-1)*2*(mNumT-2)+s*2;
+ }
+ if (s < mNumS-2) { //bottom right/top right neighbor face
+ mEdge[cur_edge++] = (mNumS-1)*2*t+(s+1)*2;
+ }
+ else if (flat_face || volume->getProfile().isOpen() == TRUE) { //no neighbor
+ mEdge[cur_edge++] = -1;
+ }
+ else { //wrap on S
+ mEdge[cur_edge++] = (mNumS-1)*2*t;
+ }
+ mEdge[cur_edge++] = (mNumS-1)*2*t+s*2; //top right/bottom left neighbor face
+ }
+ }
+ }
+
+ //clear normals
+ for (U32 i = 0; i < mNumVertices; i++)
+ {
+ mNormals[i].clear();
+ }
+
+ //generate normals
+ for (U32 i = 0; i < mNumIndices/3; i++) //for each triangle
+ {
+ const U16* idx = &(mIndices[i*3]);
+
+
+ LLVector4a* v[] =
+ { pos+idx[0], pos+idx[1], pos+idx[2] };
+
+ LLVector4a* n[] =
+ { norm+idx[0], norm+idx[1], norm+idx[2] };
+
+ //calculate triangle normal
+ LLVector4a a, b, c;
+
+ a.setSub(*v[0], *v[1]);
+ b.setSub(*v[0], *v[2]);
+ c.setCross3(a,b);
+
+ n[0]->add(c);
+ n[1]->add(c);
+ n[2]->add(c);
+
+ //even out quad contributions
+ n[i%2+1]->add(c);
+ }
+
+ // adjust normals based on wrapping and stitching
+
+ 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++)
+ {
+ LLVector4a n;
+ n.setAdd(norm[i], norm[mNumS*(mNumT-1)+i]);
+ norm[i] = n;
+ norm[mNumS*(mNumT-1)+i] = n;
+ }
+ }
+
+ if ((volume->getProfile().isOpen() == FALSE) && !(s_bottom_converges))
+ { //wrap normals on S
+ for (S32 i = 0; i < mNumT; i++)
+ {
+ LLVector4a n;
+ n.setAdd(norm[mNumS*i], norm[mNumS*i+mNumS-1]);
+ norm[mNumS * i] = n;
+ norm[mNumS * i+mNumS-1] = n;
+ }
+ }
+
+ if (volume->getPathType() == LL_PCODE_PATH_CIRCLE &&
+ ((volume->getProfileType() & LL_PCODE_PROFILE_MASK) == LL_PCODE_PROFILE_CIRCLE_HALF))
+ {
+ if (s_bottom_converges)
+ { //all lower S have same normal
+ for (S32 i = 0; i < mNumT; i++)
+ {
+ norm[mNumS*i].set(1,0,0);
+ }
+ }
+
+ if (s_top_converges)
+ { //all upper S have same normal
+ for (S32 i = 0; i < mNumT; i++)
+ {
+ norm[mNumS*i+mNumS-1].set(-1,0,0);
+ }
+ }
+ }
+ }
+ else // logic for sculpt volumes
+ {
+ BOOL average_poles = FALSE;
+ BOOL wrap_s = FALSE;
+ BOOL wrap_t = FALSE;
+
+ if (sculpt_stitching == LL_SCULPT_TYPE_SPHERE)
+ average_poles = TRUE;
+
+ if ((sculpt_stitching == LL_SCULPT_TYPE_SPHERE) ||
+ (sculpt_stitching == LL_SCULPT_TYPE_TORUS) ||
+ (sculpt_stitching == LL_SCULPT_TYPE_CYLINDER))
+ wrap_s = TRUE;
+
+ if (sculpt_stitching == LL_SCULPT_TYPE_TORUS)
+ wrap_t = TRUE;
+
+
+ if (average_poles)
+ {
+ // average normals for north pole
+
+ LLVector4a average;
+ average.clear();
+
+ for (S32 i = 0; i < mNumS; i++)
+ {
+ average.add(norm[i]);
+ }
+
+ // set average
+ for (S32 i = 0; i < mNumS; i++)
+ {
+ norm[i] = average;
+ }
+
+ // average normals for south pole
+
+ average.clear();
+
+ for (S32 i = 0; i < mNumS; i++)
+ {
+ average.add(norm[i + mNumS * (mNumT - 1)]);
+ }
+
+ // set average
+ for (S32 i = 0; i < mNumS; i++)
+ {
+ norm[i + mNumS * (mNumT - 1)] = average;
+ }
+
+ }
+
+
+ if (wrap_s)
+ {
+ for (S32 i = 0; i < mNumT; i++)
+ {
+ 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++)
+ {
+ LLVector4a n;
+ n.setAdd(norm[i], norm[mNumS*(mNumT-1)+i]);
+ norm[i] = n;
+ norm[mNumS*(mNumT-1)+i] = n;
+ }
+ }
+
+ }
+
+ return TRUE;
+}
+
+// Finds binormal based on three vertices with texture coordinates.
+// Fills in dummy values if the triangle has degenerate texture coordinates.
+void calc_binormal_from_triangle(LLVector4a& binormal,
+
+ const LLVector4a& pos0,
+ const LLVector2& tex0,
+ const LLVector4a& pos1,
+ const LLVector2& tex1,
+ const LLVector4a& pos2,
+ const LLVector2& tex2)
+{
+ LLVector4a rx0( pos0[VX], tex0.mV[VX], tex0.mV[VY] );
+ LLVector4a rx1( pos1[VX], tex1.mV[VX], tex1.mV[VY] );
+ LLVector4a rx2( pos2[VX], tex2.mV[VX], tex2.mV[VY] );
+
+ LLVector4a ry0( pos0[VY], tex0.mV[VX], tex0.mV[VY] );
+ LLVector4a ry1( pos1[VY], tex1.mV[VX], tex1.mV[VY] );
+ LLVector4a ry2( pos2[VY], tex2.mV[VX], tex2.mV[VY] );
+
+ LLVector4a rz0( pos0[VZ], tex0.mV[VX], tex0.mV[VY] );
+ LLVector4a rz1( pos1[VZ], tex1.mV[VX], tex1.mV[VY] );
+ LLVector4a rz2( pos2[VZ], tex2.mV[VX], tex2.mV[VY] );
+
+ LLVector4a lhs, rhs;
+
+ LLVector4a r0;
+ lhs.setSub(rx0, rx1); rhs.setSub(rx0, rx2);
+ r0.setCross3(lhs, rhs);
+
+ LLVector4a r1;
+ lhs.setSub(ry0, ry1); rhs.setSub(ry0, ry2);
+ r1.setCross3(lhs, rhs);
+
+ LLVector4a r2;
+ lhs.setSub(rz0, rz1); rhs.setSub(rz0, rz2);
+ r2.setCross3(lhs, rhs);
+
+ if( r0[VX] && r1[VX] && r2[VX] )
+ {
+ binormal.set(
+ -r0[VZ] / r0[VX],
+ -r1[VZ] / r1[VX],
+ -r2[VZ] / r2[VX]);
+ // binormal.normVec();
+ }
+ else
+ {
+ binormal.set( 0, 1 , 0 );
+ }
+}
diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h index 28b9895ff3..5bd65c2bf2 100644 --- a/indra/llmath/llvolume.h +++ b/indra/llmath/llvolume.h @@ -34,8 +34,13 @@ class LLPathParams; class LLVolumeParams; class LLProfile; class LLPath; + +template <class T> class LLOctreeNode; + +class LLVector4a; class LLVolumeFace; class LLVolume; +class LLVolumeTriangle; #include "lldarray.h" #include "lluuid.h" @@ -43,6 +48,8 @@ class LLVolume; //#include "vmath.h" #include "v2math.h" #include "v3math.h" +#include "v3dmath.h" +#include "v4math.h" #include "llquaternion.h" #include "llstrider.h" #include "v4coloru.h" @@ -177,12 +184,14 @@ const U8 LL_SCULPT_TYPE_SPHERE = 1; const U8 LL_SCULPT_TYPE_TORUS = 2; const U8 LL_SCULPT_TYPE_PLANE = 3; const U8 LL_SCULPT_TYPE_CYLINDER = 4; - -const U8 LL_SCULPT_TYPE_MASK = LL_SCULPT_TYPE_SPHERE | LL_SCULPT_TYPE_TORUS | LL_SCULPT_TYPE_PLANE | LL_SCULPT_TYPE_CYLINDER; +const U8 LL_SCULPT_TYPE_MESH = 5; +const U8 LL_SCULPT_TYPE_MASK = LL_SCULPT_TYPE_SPHERE | LL_SCULPT_TYPE_TORUS | LL_SCULPT_TYPE_PLANE | + LL_SCULPT_TYPE_CYLINDER | LL_SCULPT_TYPE_MESH; const U8 LL_SCULPT_FLAG_INVERT = 64; const U8 LL_SCULPT_FLAG_MIRROR = 128; +const S32 LL_SCULPT_MESH_MAX_FACES = 8; class LLProfileParams { @@ -569,6 +578,9 @@ public: BOOL importLegacyStream(std::istream& input_stream); BOOL exportLegacyStream(std::ostream& output_stream) const; + LLSD sculptAsLLSD() const; + bool sculptFromLLSD(LLSD& sd); + LLSD asLLSD() const; operator LLSD() const { return asLLSD(); } bool fromLLSD(LLSD& sd); @@ -628,7 +640,8 @@ public: const F32& getSkew() const { return mPathParams.getSkew(); } const LLUUID& getSculptID() const { return mSculptID; } const U8& getSculptType() const { return mSculptType; } - + bool isSculpt() const; + bool isMeshSculpt() const; BOOL isConvex() const; // 'begin' and 'end' should be in range [0, 1] (they will be clamped) @@ -779,30 +792,88 @@ public: class LLVolumeFace { public: - LLVolumeFace() : - mID(0), - mTypeMask(0), - mHasBinormals(FALSE), - mBeginS(0), - mBeginT(0), - mNumS(0), - mNumT(0) + class VertexData { - } + enum + { + POSITION = 0, + NORMAL = 1 + }; + + private: + void init(); + public: + VertexData(); + VertexData(const VertexData& rhs); + const VertexData& operator=(const VertexData& rhs); + + ~VertexData(); + LLVector4a& getPosition(); + LLVector4a& getNormal(); + const LLVector4a& getPosition() const; + const LLVector4a& getNormal() const; + void setPosition(const LLVector4a& pos); + void setNormal(const LLVector4a& norm); + + + LLVector2 mTexCoord; + + bool operator<(const VertexData& rhs) const; + bool operator==(const VertexData& rhs) const; + bool compareNormal(const VertexData& rhs, F32 angle_cutoff) const; + + private: + LLVector4a* mData; + }; + + LLVolumeFace(); + LLVolumeFace(const LLVolumeFace& src); + LLVolumeFace& operator=(const LLVolumeFace& rhs); + + ~LLVolumeFace(); +private: + void freeData(); +public: BOOL create(LLVolume* volume, BOOL partial_build = FALSE); void createBinormals(); - void makeTriStrip(); - class VertexData + void appendFace(const LLVolumeFace& face, LLMatrix4& transform, LLMatrix4& normal_tranform); + + void resizeVertices(S32 num_verts); + void allocateBinormals(S32 num_verts); + void allocateWeights(S32 num_verts); + void resizeIndices(S32 num_indices); + void fillFromLegacyData(std::vector<LLVolumeFace::VertexData>& v, std::vector<U16>& idx); + + void pushVertex(const VertexData& cv); + void pushVertex(const LLVector4a& pos, const LLVector4a& norm, const LLVector2& tc); + void pushIndex(const U16& idx); + + void swapData(LLVolumeFace& rhs); + + void getVertexData(U16 indx, LLVolumeFace::VertexData& cv); + + class VertexMapData : public LLVolumeFace::VertexData { public: - LLVector3 mPosition; - LLVector3 mNormal; - LLVector3 mBinormal; - LLVector2 mTexCoord; + U16 mIndex; + + bool operator==(const LLVolumeFace::VertexData& rhs) const; + + struct ComparePosition + { + bool operator()(const LLVector3& a, const LLVector3& b) const; + }; + + typedef std::map<LLVector3, std::vector<VertexMapData>, VertexMapData::ComparePosition > PointMap; }; + void optimize(F32 angle_cutoff = 2.f); + void cacheOptimize(); + + void createOctree(F32 scaler = 0.25f, const LLVector4a& center = LLVector4a(0,0,0), const LLVector4a& size = LLVector4a(0.5f,0.5f,0.5f)); + enum { SINGLE_MASK = 0x0001, @@ -821,23 +892,35 @@ public: public: S32 mID; U32 mTypeMask; - LLVector3 mCenter; - BOOL mHasBinormals; - + // Only used for INNER/OUTER faces S32 mBeginS; S32 mBeginT; S32 mNumS; S32 mNumT; - LLVector3 mExtents[2]; //minimum and maximum point of face - LLVector2 mTexCoordExtents[2]; //minimum and maximum of texture coordinates of the face. + LLVector4a* mExtents; //minimum and maximum point of face + LLVector4a* mCenter; + LLVector2 mTexCoordExtents[2]; //minimum and maximum of texture coordinates of the face. + + S32 mNumVertices; + S32 mNumIndices; + + LLVector4a* mPositions; + LLVector4a* mNormals; + LLVector4a* mBinormals; + LLVector2* mTexCoords; + U16* mIndices; - std::vector<VertexData> mVertices; - std::vector<U16> mIndices; - std::vector<U16> mTriStrip; std::vector<S32> mEdge; + //list of skin weights for rigged volumes + // format is mWeights[vertex_index].mV[influence] = <joint_index>.<weight> + // mWeights.size() should be empty or match mVertices.size() + LLVector4a* mWeights; + + LLOctreeNode<LLVolumeTriangle>* mOctree; + private: BOOL createUnCutCubeCap(LLVolume* volume, BOOL partial_build = FALSE); BOOL createCap(LLVolume* volume, BOOL partial_build = FALSE); @@ -848,8 +931,7 @@ class LLVolume : public LLRefCount { friend class LLVolumeLODGroup; -private: - LLVolume(const LLVolume&); // Don't implement +protected: ~LLVolume(); // use unref public: @@ -871,7 +953,7 @@ public: U8 getProfileType() const { return mParams.getProfileParams().getCurveType(); } U8 getPathType() const { return mParams.getPathParams().getCurveType(); } - S32 getNumFaces() const { return (S32)mProfilep->mFaces.size(); } + S32 getNumFaces() const; S32 getNumVolumeFaces() const { return mVolumeFaces.size(); } F32 getDetail() const { return mDetail; } const LLVolumeParams& getParams() const { return mParams; } @@ -893,12 +975,15 @@ public: BOOL isUnique() const { return mUnique; } S32 getSculptLevel() const { return mSculptLevel; } - + void setSculptLevel(S32 level) { mSculptLevel = level; } + S32 *getTriangleIndices(U32 &num_indices) const; // returns number of triangle indeces required for path/profile mesh S32 getNumTriangleIndices() const; + S32 getNumTriangles() const; + void generateSilhouetteVertices(std::vector<LLVector3> &vertices, std::vector<LLVector3> &normals, std::vector<S32> &segments, @@ -917,6 +1002,13 @@ public: LLVector3* normal = NULL, // return the surface normal at the intersection point LLVector3* bi_normal = NULL // return the surface bi-normal at the intersection point ); + + S32 lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, + S32 face = 1, + LLVector3* intersection = NULL, + LLVector2* tex_coord = NULL, + LLVector3* normal = NULL, + LLVector3* bi_normal = NULL); // The following cleans up vertices and triangles, // getting rid of degenerate triangles and duplicate vertices, @@ -938,11 +1030,14 @@ public: friend std::ostream& operator<<(std::ostream &s, const LLVolume *volumep); // HACK to bypass Windoze confusion over // conversion if *(LLVolume*) to LLVolume& const LLVolumeFace &getVolumeFace(const S32 f) const {return mVolumeFaces[f];} // DO NOT DELETE VOLUME WHILE USING THIS REFERENCE, OR HOLD A POINTER TO THIS VOLUMEFACE - + U32 mFaceMask; // bit array of which faces exist in this volume LLVector3 mLODScaleBias; // vector for biasing LOD based on scale void sculpt(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, S32 sculpt_level); + void copyVolumeFaces(const LLVolume* volume); + void cacheOptimize(); + private: void sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, U8 sculpt_type); F32 sculptGetSurfaceArea(); @@ -953,35 +1048,58 @@ private: protected: BOOL generate(); void createVolumeFaces(); +public: + virtual BOOL createVolumeFacesFromFile(const std::string& file_name); + virtual BOOL createVolumeFacesFromStream(std::istream& is); + virtual bool unpackVolumeFaces(std::istream& is, S32 size); + + virtual void makeTetrahedron(); + virtual BOOL isTetrahedron(); protected: BOOL mUnique; F32 mDetail; S32 mSculptLevel; + BOOL mIsTetrahedron; LLVolumeParams mParams; LLPath *mPathp; LLProfile *mProfilep; std::vector<Point> mMesh; - + BOOL mGenerateSingleFace; typedef std::vector<LLVolumeFace> face_list_t; face_list_t mVolumeFaces; + +public: + LLVector4a* mHullPoints; + U16* mHullIndices; + S32 mNumHullPoints; + S32 mNumHullIndices; }; std::ostream& operator<<(std::ostream &s, const LLVolumeParams &volume_params); -LLVector3 calc_binormal_from_triangle( - const LLVector3& pos0, +void calc_binormal_from_triangle( + LLVector4a& binormal, + const LLVector4a& pos0, const LLVector2& tex0, - const LLVector3& pos1, + const LLVector4a& pos1, const LLVector2& tex1, - const LLVector3& pos2, + const LLVector4a& pos2, const LLVector2& tex2); +BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* center, const F32* size); BOOL LLLineSegmentBoxIntersect(const LLVector3& start, const LLVector3& end, const LLVector3& center, const LLVector3& size); +BOOL LLLineSegmentBoxIntersect(const LLVector4a& start, const LLVector4a& end, const LLVector4a& center, const LLVector4a& size); + 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); + 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); +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); diff --git a/indra/llmath/llvolumemgr.cpp b/indra/llmath/llvolumemgr.cpp index 88c195936c..c60b750088 100644 --- a/indra/llmath/llvolumemgr.cpp +++ b/indra/llmath/llvolumemgr.cpp @@ -314,7 +314,7 @@ BOOL LLVolumeLODGroup::derefLOD(LLVolume *volumep) { llassert_always(mLODRefs[i] > 0); mLODRefs[i]--; -#if 1 // SJB: Possible opt: keep other lods around +#if 0 // SJB: Possible opt: keep other lods around if (!mLODRefs[i]) { mVolumeLODs[i] = NULL; @@ -369,6 +369,19 @@ F32 LLVolumeLODGroup::getVolumeScaleFromDetail(const S32 detail) return mDetailScales[detail]; } +S32 LLVolumeLODGroup::getVolumeDetailFromScale(const F32 detail) +{ + for (S32 i = 1; i < 4; i++) + { + if (mDetailScales[i] > detail) + { + return i-1; + } + } + + return 3; +} + F32 LLVolumeLODGroup::dump() { F32 usage = 0.f; diff --git a/indra/llmath/llvolumemgr.h b/indra/llmath/llvolumemgr.h index 5257da2693..c75906f675 100644 --- a/indra/llmath/llvolumemgr.h +++ b/indra/llmath/llvolumemgr.h @@ -53,6 +53,7 @@ public: static S32 getDetailFromTan(const F32 tan_angle); static void getDetailProximity(const F32 tan_angle, F32 &to_lower, F32& to_higher); static F32 getVolumeScaleFromDetail(const S32 detail); + static S32 getVolumeDetailFromScale(F32 scale); LLVolume* refLOD(const S32 detail); BOOL derefLOD(LLVolume *volumep); diff --git a/indra/llmath/llvolumeoctree.cpp b/indra/llmath/llvolumeoctree.cpp new file mode 100644 index 0000000000..b5a935c2b5 --- /dev/null +++ b/indra/llmath/llvolumeoctree.cpp @@ -0,0 +1,256 @@ +/** + + * @file llvolumeoctree.cpp + * + * $LicenseInfo:firstyear=2002&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#include "llvolumeoctree.h" +#include "llvector4a.h" + +BOOL LLLineSegmentBoxIntersect(const LLVector4a& start, const LLVector4a& end, const LLVector4a& center, const LLVector4a& size) +{ + LLVector4a fAWdU; + LLVector4a dir; + LLVector4a diff; + + dir.setSub(end, start); + dir.mul(0.5f); + + diff.setAdd(end,start); + diff.mul(0.5f); + diff.sub(center); + fAWdU.setAbs(dir); + + LLVector4a rhs; + rhs.setAdd(size, fAWdU); + + LLVector4a lhs; + lhs.setAbs(diff); + + U32 grt = lhs.greaterThan(rhs).getGatheredBits(); + + if (grt & 0x7) + { + return false; + } + + LLVector4a f; + f.setCross3(dir, diff); + f.setAbs(f); + + LLVector4a v0, v1; + + v0 = _mm_shuffle_ps(size, size,_MM_SHUFFLE(3,0,0,1)); + v1 = _mm_shuffle_ps(fAWdU, fAWdU, _MM_SHUFFLE(3,1,2,2)); + lhs.setMul(v0, v1); + + v0 = _mm_shuffle_ps(size, size, _MM_SHUFFLE(3,1,2,2)); + v1 = _mm_shuffle_ps(fAWdU, fAWdU, _MM_SHUFFLE(3,0,0,1)); + rhs.setMul(v0, v1); + rhs.add(lhs); + + grt = f.greaterThan(rhs).getGatheredBits(); + + return (grt & 0x7) ? false : true; +} + + +LLVolumeOctreeListener::LLVolumeOctreeListener(LLOctreeNode<LLVolumeTriangle>* node) +{ + node->addListener(this); +} + +LLVolumeOctreeListener::~LLVolumeOctreeListener() +{ + +} + +void LLVolumeOctreeListener::handleChildAddition(const LLOctreeNode<LLVolumeTriangle>* parent, + LLOctreeNode<LLVolumeTriangle>* child) +{ + new LLVolumeOctreeListener(child); +} + + +LLOctreeTriangleRayIntersect::LLOctreeTriangleRayIntersect(const LLVector4a& start, const LLVector4a& dir, + const LLVolumeFace* face, F32* closest_t, + LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal) + : mFace(face), + mStart(start), + mDir(dir), + mIntersection(intersection), + mTexCoord(tex_coord), + mNormal(normal), + mBinormal(bi_normal), + mClosestT(closest_t), + mHitFace(false) +{ + mEnd.setAdd(mStart, mDir); +} + +void LLOctreeTriangleRayIntersect::traverse(const LLOctreeNode<LLVolumeTriangle>* node) +{ + LLVolumeOctreeListener* vl = (LLVolumeOctreeListener*) node->getListener(0); + + /*const F32* start = mStart.getF32(); + const F32* end = mEnd.getF32(); + const F32* center = vl->mBounds[0].getF32(); + const F32* size = vl->mBounds[1].getF32();*/ + + //if (LLLineSegmentBoxIntersect(mStart, mEnd, vl->mBounds[0], vl->mBounds[1])) + if (LLLineSegmentBoxIntersect(mStart.getF32ptr(), mEnd.getF32ptr(), vl->mBounds[0].getF32ptr(), vl->mBounds[1].getF32ptr())) + { + node->accept(this); + for (S32 i = 0; i < node->getChildCount(); ++i) + { + traverse(node->getChild(i)); + } + } +} + +void LLOctreeTriangleRayIntersect::visit(const LLOctreeNode<LLVolumeTriangle>* node) +{ + for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = + node->getData().begin(); iter != node->getData().end(); ++iter) + { + const LLVolumeTriangle* tri = *iter; + + F32 a, b, t; + + if (LLTriangleRayIntersect(*tri->mV[0], *tri->mV[1], *tri->mV[2], + mStart, mDir, a, b, t)) + { + if ((t >= 0.f) && // if hit is after start + (t <= 1.f) && // and before end + (t < *mClosestT)) // and this hit is closer + { + *mClosestT = t; + mHitFace = true; + + if (mIntersection != NULL) + { + LLVector4a intersect = mDir; + intersect.mul(*mClosestT); + intersect.add(mStart); + mIntersection->set(intersect.getF32ptr()); + } + + + if (mTexCoord != NULL) + { + LLVector2* tc = (LLVector2*) mFace->mTexCoords; + *mTexCoord = ((1.f - a - b) * tc[tri->mIndex[0]] + + a * tc[tri->mIndex[1]] + + b * tc[tri->mIndex[2]]); + + } + + if (mNormal != NULL) + { + LLVector4* norm = (LLVector4*) mFace->mNormals; + + *mNormal = ((1.f - a - b) * LLVector3(norm[tri->mIndex[0]]) + + a * LLVector3(norm[tri->mIndex[1]]) + + b * LLVector3(norm[tri->mIndex[2]])); + } + + if (mBinormal != NULL) + { + LLVector4* binormal = (LLVector4*) mFace->mBinormals; + *mBinormal = ((1.f - a - b) * LLVector3(binormal[tri->mIndex[0]]) + + a * LLVector3(binormal[tri->mIndex[1]]) + + b * LLVector3(binormal[tri->mIndex[2]])); + } + } + } + } +} + +const LLVector4a& LLVolumeTriangle::getPositionGroup() const +{ + return mPositionGroup; +} + +const F32& LLVolumeTriangle::getBinRadius() const +{ + return mRadius; +} + + +//TEST CODE + +void LLVolumeOctreeValidate::visit(const LLOctreeNode<LLVolumeTriangle>* branch) +{ + LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0); + + //make sure bounds matches extents + LLVector4a& min = node->mExtents[0]; + LLVector4a& max = node->mExtents[1]; + + LLVector4a& center = node->mBounds[0]; + LLVector4a& size = node->mBounds[1]; + + LLVector4a test_min, test_max; + test_min.setSub(center, size); + test_max.setAdd(center, size); + + if (!test_min.equals3(min, 0.001f) || + !test_max.equals3(max, 0.001f)) + { + llerrs << "Bad bounding box data found." << llendl; + } + + test_min.sub(LLVector4a(0.001f)); + test_max.add(LLVector4a(0.001f)); + + for (U32 i = 0; i < branch->getChildCount(); ++i) + { + LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); + + //make sure all children fit inside this node + if (child->mExtents[0].lessThan(test_min).areAnySet(LLVector4Logical::MASK_XYZ) || + child->mExtents[1].greaterThan(test_max).areAnySet(LLVector4Logical::MASK_XYZ)) + { + llerrs << "Child protrudes from bounding box." << llendl; + } + } + + //children fit, check data + for (LLOctreeNode<LLVolumeTriangle>::const_element_iter iter = branch->getData().begin(); + iter != branch->getData().end(); ++iter) + { + const LLVolumeTriangle* tri = *iter; + + //validate triangle + for (U32 i = 0; i < 3; i++) + { + if (tri->mV[i]->greaterThan(test_max).areAnySet(LLVector4Logical::MASK_XYZ) || + tri->mV[i]->lessThan(test_min).areAnySet(LLVector4Logical::MASK_XYZ)) + { + llerrs << "Triangle protrudes from node." << llendl; + } + } + } +} + + diff --git a/indra/llmath/llvolumeoctree.h b/indra/llmath/llvolumeoctree.h new file mode 100644 index 0000000000..688d91dc40 --- /dev/null +++ b/indra/llmath/llvolumeoctree.h @@ -0,0 +1,134 @@ +/** + * @file llvolumeoctree.h + * @brief LLVolume octree classes. + * + * $LicenseInfo:firstyear=2002&license=viewerlgpl$ + * Second Life Viewer Source Code + * 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. + * + * 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. + * + * 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 + * + * Linden Research, Inc., 945 Battery Street, San Francisco, CA 94111 USA + * $/LicenseInfo$ + */ + +#ifndef LL_LLVOLUME_OCTREE_H +#define LL_LLVOLUME_OCTREE_H + +#include "linden_common.h" +#include "llmemory.h" + +#include "lloctree.h" +#include "llvolume.h" +#include "llvector4a.h" + +class LLVolumeTriangle : public LLRefCount +{ +public: + LLVolumeTriangle() + { + + } + + LLVolumeTriangle(const LLVolumeTriangle& rhs) + { + *this = rhs; + } + + const LLVolumeTriangle& operator=(const LLVolumeTriangle& rhs) + { + llerrs << "Illegal operation!" << llendl; + return *this; + } + + ~LLVolumeTriangle() + { + + } + + LLVector4a mPositionGroup; + + const LLVector4a* mV[3]; + U16 mIndex[3]; + + F32 mRadius; + + virtual const LLVector4a& getPositionGroup() const; + virtual const F32& getBinRadius() const; +}; + +class LLVolumeOctreeListener : public LLOctreeListener<LLVolumeTriangle> +{ +public: + + LLVolumeOctreeListener(LLOctreeNode<LLVolumeTriangle>* node); + ~LLVolumeOctreeListener(); + + LLVolumeOctreeListener(const LLVolumeOctreeListener& rhs) + { + *this = rhs; + } + + const LLVolumeOctreeListener& operator=(const LLVolumeOctreeListener& rhs) + { + llerrs << "Illegal operation!" << llendl; + return *this; + } + + //LISTENER FUNCTIONS + virtual void handleChildAddition(const LLOctreeNode<LLVolumeTriangle>* parent, + LLOctreeNode<LLVolumeTriangle>* child); + virtual void handleStateChange(const LLTreeNode<LLVolumeTriangle>* node) { } + virtual void handleChildRemoval(const LLOctreeNode<LLVolumeTriangle>* parent, + const LLOctreeNode<LLVolumeTriangle>* child) { } + virtual void handleInsertion(const LLTreeNode<LLVolumeTriangle>* node, LLVolumeTriangle* tri) { } + virtual void handleRemoval(const LLTreeNode<LLVolumeTriangle>* node, LLVolumeTriangle* tri) { } + virtual void handleDestruction(const LLTreeNode<LLVolumeTriangle>* node) { } + + +public: + LLVector4a mBounds[2]; // bounding box (center, size) of this node and all its children (tight fit to objects) + LLVector4a mExtents[2]; // extents (min, max) of this node and all its children +}; + +class LLOctreeTriangleRayIntersect : public LLOctreeTraveler<LLVolumeTriangle> +{ +public: + const LLVolumeFace* mFace; + LLVector4a mStart; + LLVector4a mDir; + LLVector4a mEnd; + LLVector3* mIntersection; + LLVector2* mTexCoord; + LLVector3* mNormal; + LLVector3* mBinormal; + F32* mClosestT; + bool mHitFace; + + LLOctreeTriangleRayIntersect(const LLVector4a& start, const LLVector4a& dir, + const LLVolumeFace* face, F32* closest_t, + LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal); + + void traverse(const LLOctreeNode<LLVolumeTriangle>* node); + + virtual void visit(const LLOctreeNode<LLVolumeTriangle>* node); +}; + +class LLVolumeOctreeValidate : public LLOctreeTraveler<LLVolumeTriangle> +{ + virtual void visit(const LLOctreeNode<LLVolumeTriangle>* branch); +}; + +#endif diff --git a/indra/llmath/m4math.cpp b/indra/llmath/m4math.cpp index 946b1553fe..8741400e52 100644 --- a/indra/llmath/m4math.cpp +++ b/indra/llmath/m4math.cpp @@ -215,8 +215,33 @@ const LLMatrix4& LLMatrix4::transpose() F32 LLMatrix4::determinant() const { - llerrs << "Not implemented!" << llendl; - return 0.f; + F32 value = + mMatrix[0][3] * mMatrix[1][2] * mMatrix[2][1] * mMatrix[3][0] - + mMatrix[0][2] * mMatrix[1][3] * mMatrix[2][1] * mMatrix[3][0] - + mMatrix[0][3] * mMatrix[1][1] * mMatrix[2][2] * mMatrix[3][0] + + mMatrix[0][1] * mMatrix[1][3] * mMatrix[2][2] * mMatrix[3][0] + + mMatrix[0][2] * mMatrix[1][1] * mMatrix[2][3] * mMatrix[3][0] - + mMatrix[0][1] * mMatrix[1][2] * mMatrix[2][3] * mMatrix[3][0] - + mMatrix[0][3] * mMatrix[1][2] * mMatrix[2][0] * mMatrix[3][1] + + mMatrix[0][2] * mMatrix[1][3] * mMatrix[2][0] * mMatrix[3][1] + + mMatrix[0][3] * mMatrix[1][0] * mMatrix[2][2] * mMatrix[3][1] - + mMatrix[0][0] * mMatrix[1][3] * mMatrix[2][2] * mMatrix[3][1] - + mMatrix[0][2] * mMatrix[1][0] * mMatrix[2][3] * mMatrix[3][1] + + mMatrix[0][0] * mMatrix[1][2] * mMatrix[2][3] * mMatrix[3][1] + + mMatrix[0][3] * mMatrix[1][1] * mMatrix[2][0] * mMatrix[3][2] - + mMatrix[0][1] * mMatrix[1][3] * mMatrix[2][0] * mMatrix[3][2] - + mMatrix[0][3] * mMatrix[1][0] * mMatrix[2][1] * mMatrix[3][2] + + mMatrix[0][0] * mMatrix[1][3] * mMatrix[2][1] * mMatrix[3][2] + + mMatrix[0][1] * mMatrix[1][0] * mMatrix[2][3] * mMatrix[3][2] - + mMatrix[0][0] * mMatrix[1][1] * mMatrix[2][3] * mMatrix[3][2] - + mMatrix[0][2] * mMatrix[1][1] * mMatrix[2][0] * mMatrix[3][3] + + mMatrix[0][1] * mMatrix[1][2] * mMatrix[2][0] * mMatrix[3][3] + + mMatrix[0][2] * mMatrix[1][0] * mMatrix[2][1] * mMatrix[3][3] - + mMatrix[0][0] * mMatrix[1][2] * mMatrix[2][1] * mMatrix[3][3] - + mMatrix[0][1] * mMatrix[1][0] * mMatrix[2][2] * mMatrix[3][3] + + mMatrix[0][0] * mMatrix[1][1] * mMatrix[2][2] * mMatrix[3][3]; + + return value; } // Only works for pure orthonormal, homogeneous transform matrices. @@ -422,6 +447,17 @@ const LLMatrix4& LLMatrix4::initRotTrans(const LLQuaternion &q, const LLVector return (*this); } +const LLMatrix4& LLMatrix4::initScale(const LLVector3 &scale) +{ + setIdentity(); + + mMatrix[VX][VX] = scale.mV[VX]; + mMatrix[VY][VY] = scale.mV[VY]; + mMatrix[VZ][VZ] = scale.mV[VZ]; + + return (*this); +} + const LLMatrix4& LLMatrix4::initAll(const LLVector3 &scale, const LLQuaternion &q, const LLVector3 &pos) { F32 sx, sy, sz; @@ -642,37 +678,6 @@ const LLMatrix4& LLMatrix4::initMatrix(const LLMatrix3 &mat, const LLVector4 & // LLMatrix4 Operators - -/* Not implemented to help enforce code consistency with the syntax of - row-major notation. This is a Good Thing. -LLVector4 operator*(const LLMatrix4 &a, const LLVector4 &b) -{ - // Operate "to the right" on column-vector b - LLVector4 vec; - vec.mV[VX] = a.mMatrix[VX][VX] * b.mV[VX] + - a.mMatrix[VY][VX] * b.mV[VY] + - a.mMatrix[VZ][VX] * b.mV[VZ] + - a.mMatrix[VW][VX] * b.mV[VW]; - - vec.mV[VY] = a.mMatrix[VX][VY] * b.mV[VX] + - a.mMatrix[VY][VY] * b.mV[VY] + - a.mMatrix[VZ][VY] * b.mV[VZ] + - a.mMatrix[VW][VY] * b.mV[VW]; - - vec.mV[VZ] = a.mMatrix[VX][VZ] * b.mV[VX] + - a.mMatrix[VY][VZ] * b.mV[VY] + - a.mMatrix[VZ][VZ] * b.mV[VZ] + - a.mMatrix[VW][VZ] * b.mV[VW]; - - vec.mV[VW] = a.mMatrix[VX][VW] * b.mV[VX] + - a.mMatrix[VY][VW] * b.mV[VY] + - a.mMatrix[VZ][VW] * b.mV[VZ] + - a.mMatrix[VW][VW] * b.mV[VW]; - return vec; -} -*/ - - LLVector4 operator*(const LLVector4 &a, const LLMatrix4 &b) { // Operate "to the left" on row-vector a @@ -768,6 +773,23 @@ bool operator!=(const LLMatrix4 &a, const LLMatrix4 &b) return FALSE; } +bool operator<(const LLMatrix4& a, const LLMatrix4 &b) +{ + U32 i, j; + for (i = 0; i < NUM_VALUES_IN_MAT4; i++) + { + for (j = 0; j < NUM_VALUES_IN_MAT4; j++) + { + if (a.mMatrix[i][j] != b.mMatrix[i][j]) + { + return a.mMatrix[i][j] < b.mMatrix[i][j]; + } + } + } + + return false; +} + const LLMatrix4& operator*=(LLMatrix4 &a, F32 k) { U32 i, j; diff --git a/indra/llmath/m4math.h b/indra/llmath/m4math.h index 6ec9958491..10c88464e5 100644 --- a/indra/llmath/m4math.h +++ b/indra/llmath/m4math.h @@ -153,6 +153,7 @@ public: const LLMatrix4& initRotTrans(const F32 roll, const F32 pitch, const F32 yaw, const LLVector4 &pos); // Rotation from Euler + translation const LLMatrix4& initRotTrans(const LLQuaternion &q, const LLVector4 &pos); // Set with Quaternion and position + const LLMatrix4& initScale(const LLVector3 &scale); // Set all const LLMatrix4& initAll(const LLVector3 &scale, const LLQuaternion &q, const LLVector3 &pos); @@ -219,10 +220,7 @@ public: // Operators // -// Not implemented to enforce code that agrees with symbolic syntax -// friend LLVector4 operator*(const LLMatrix4 &a, const LLVector4 &b); // Apply rotation a to vector b - -// friend inline LLMatrix4 operator*(const LLMatrix4 &a, const LLMatrix4 &b); // Return a * b + // friend inline LLMatrix4 operator*(const LLMatrix4 &a, const LLMatrix4 &b); // Return a * b friend LLVector4 operator*(const LLVector4 &a, const LLMatrix4 &b); // Return transform of vector a by matrix b friend const LLVector3 operator*(const LLVector3 &a, const LLMatrix4 &b); // Return full transform of a by matrix b friend LLVector4 rotate_vector(const LLVector4 &a, const LLMatrix4 &b); // Rotates a but does not translate @@ -230,6 +228,7 @@ public: friend bool operator==(const LLMatrix4 &a, const LLMatrix4 &b); // Return a == b friend bool operator!=(const LLMatrix4 &a, const LLMatrix4 &b); // Return a != b + friend bool operator<(const LLMatrix4 &a, const LLMatrix4& b); // Return a < b friend const LLMatrix4& operator+=(LLMatrix4 &a, const LLMatrix4 &b); // Return a + b friend const LLMatrix4& operator-=(LLMatrix4 &a, const LLMatrix4 &b); // Return a - b diff --git a/indra/llmath/tests/llquaternion_test.cpp b/indra/llmath/tests/llquaternion_test.cpp index 9e79b299ff..e69010b2d6 100644 --- a/indra/llmath/tests/llquaternion_test.cpp +++ b/indra/llmath/tests/llquaternion_test.cpp @@ -29,12 +29,12 @@ #include "linden_common.h" #include "../test/lltut.h" -#include "../llquaternion.h" #include "../v4math.h" #include "../v3math.h" #include "../v3dmath.h" #include "../m4math.h" #include "../m3math.h" +#include "../llquaternion.h" namespace tut { diff --git a/indra/llmath/tests/v2math_test.cpp b/indra/llmath/tests/v2math_test.cpp index 9747996b25..4d6a2eca93 100644 --- a/indra/llmath/tests/v2math_test.cpp +++ b/indra/llmath/tests/v2math_test.cpp @@ -85,7 +85,7 @@ namespace tut F32 x = 2.2345f, y = 3.5678f ; LLVector2 vec2(x,y); ensure("magVecSquared:Fail ", is_approx_equal(vec2.magVecSquared(), (x*x + y*y))); - ensure("magVec:Fail ", is_approx_equal(vec2.magVec(), fsqrtf(x*x + y*y))); + ensure("magVec:Fail ", is_approx_equal(vec2.magVec(), (F32) sqrt(x*x + y*y))); } template<> template<> @@ -407,7 +407,7 @@ namespace tut ensure_equals("dist_vec_squared values are not equal",val2, val1); val1 = dist_vec(vec2, vec3); - val2 = fsqrtf((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2)); + val2 = (F32) sqrt((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2)); ensure_equals("dist_vec values are not equal",val2, val1); } @@ -431,7 +431,7 @@ namespace tut LLVector2 vec2(x1, y1); F32 vecMag = vec2.normVec(); - F32 mag = fsqrtf(x1*x1 + y1*y1); + F32 mag = (F32) sqrt(x1*x1 + y1*y1); F32 oomag = 1.f / mag; val1 = x1 * oomag; diff --git a/indra/llmath/tests/v3color_test.cpp b/indra/llmath/tests/v3color_test.cpp index 2c00f00ab3..29d1c483ab 100644 --- a/indra/llmath/tests/v3color_test.cpp +++ b/indra/llmath/tests/v3color_test.cpp @@ -93,7 +93,7 @@ namespace tut F32 r = 2.3436212f, g = 1231.f, b = 4.7849321232f; LLColor3 llcolor3(r,g,b); ensure("magVecSquared:Fail ", is_approx_equal(llcolor3.magVecSquared(), (r*r + g*g + b*b))); - ensure("magVec:Fail ", is_approx_equal(llcolor3.magVec(), fsqrtf(r*r + g*g + b*b))); + ensure("magVec:Fail ", is_approx_equal(llcolor3.magVec(), (F32) sqrt(r*r + g*g + b*b))); } template<> template<> @@ -103,7 +103,7 @@ namespace tut F32 val1, val2,val3; LLColor3 llcolor3(r,g,b); F32 vecMag = llcolor3.normVec(); - F32 mag = fsqrtf(r*r + g*g + b*b); + F32 mag = (F32) sqrt(r*r + g*g + b*b); F32 oomag = 1.f / mag; val1 = r * oomag; val2 = g * oomag; @@ -286,7 +286,7 @@ namespace tut F32 r1 =1.f, g1 = 2.f,b1 = 1.2f, r2 = -2.3f, g2 = 1.11f, b2 = 1234.234f; LLColor3 llcolor3(r1,g1,b1),llcolor3a(r2,g2,b2); F32 val = distVec(llcolor3,llcolor3a); - ensure("distVec failed ", is_approx_equal(fsqrtf((r1-r2)*(r1-r2) + (g1-g2)*(g1-g2) + (b1-b2)*(b1-b2)) ,val)); + ensure("distVec failed ", is_approx_equal((F32) sqrt((r1-r2)*(r1-r2) + (g1-g2)*(g1-g2) + (b1-b2)*(b1-b2)) ,val)); F32 val1 = distVec_squared(llcolor3,llcolor3a); ensure("distVec_squared failed ", is_approx_equal(((r1-r2)*(r1-r2) + (g1-g2)*(g1-g2) + (b1-b2)*(b1-b2)) ,val1)); diff --git a/indra/llmath/tests/v3dmath_test.cpp b/indra/llmath/tests/v3dmath_test.cpp index b67346f4e5..20b26faa12 100644 --- a/indra/llmath/tests/v3dmath_test.cpp +++ b/indra/llmath/tests/v3dmath_test.cpp @@ -30,11 +30,11 @@ #include "llsd.h" #include "../test/lltut.h" -#include "../llquaternion.h" #include "../m3math.h" #include "../v4math.h" #include "../v3dmath.h" #include "../v3dmath.h" +#include "../llquaternion.h" namespace tut { @@ -403,7 +403,7 @@ namespace tut LLVector3d vec3D(x,y,z); F64 res = (x*x + y*y + z*z) - vec3D.magVecSquared(); ensure("1:magVecSquared:Fail ", ((-F_APPROXIMATELY_ZERO <= res)&& (res <=F_APPROXIMATELY_ZERO))); - res = fsqrtf(x*x + y*y + z*z) - vec3D.magVec(); + res = (F32) sqrt(x*x + y*y + z*z) - vec3D.magVec(); ensure("2:magVec: Fail ", ((-F_APPROXIMATELY_ZERO <= res)&& (res <=F_APPROXIMATELY_ZERO))); } diff --git a/indra/llmath/tests/v3math_test.cpp b/indra/llmath/tests/v3math_test.cpp index e4732bf861..df7a77002f 100644 --- a/indra/llmath/tests/v3math_test.cpp +++ b/indra/llmath/tests/v3math_test.cpp @@ -30,12 +30,12 @@ #include "../test/lltut.h" #include "llsd.h" -#include "../llquaternion.h" -#include "../llquantize.h" #include "../v3dmath.h" #include "../m3math.h" #include "../v4math.h" #include "../v3math.h" +#include "../llquaternion.h" +#include "../llquantize.h" namespace tut @@ -149,7 +149,7 @@ namespace tut F32 x = 2.32f, y = 1.212f, z = -.12f; LLVector3 vec3(x,y,z); ensure("1:magVecSquared:Fail ", is_approx_equal(vec3.magVecSquared(), (x*x + y*y + z*z))); - ensure("2:magVec:Fail ", is_approx_equal(vec3.magVec(), fsqrtf(x*x + y*y + z*z))); + ensure("2:magVec:Fail ", is_approx_equal(vec3.magVec(), (F32) sqrt(x*x + y*y + z*z))); } template<> template<> @@ -509,7 +509,7 @@ namespace tut F32 val1,val2; LLVector3 vec3(x1,y1,z1),vec3a(x2,y2,z2); val1 = dist_vec(vec3,vec3a); - val2 = fsqrtf((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); + val2 = (F32) sqrt((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); ensure_equals("1:dist_vec: Fail ",val2, val1); val1 = dist_vec_squared(vec3,vec3a); val2 =((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); diff --git a/indra/llmath/tests/v4color_test.cpp b/indra/llmath/tests/v4color_test.cpp index fbd43625d1..d7eec3c87f 100644 --- a/indra/llmath/tests/v4color_test.cpp +++ b/indra/llmath/tests/v4color_test.cpp @@ -155,7 +155,7 @@ namespace tut F32 r = 0x20, g = 0xFFFF, b = 0xFF; LLColor4 llcolor4(r,g,b); ensure("magVecSquared:Fail ", is_approx_equal(llcolor4.magVecSquared(), (r*r + g*g + b*b))); - ensure("magVec:Fail ", is_approx_equal(llcolor4.magVec(), fsqrtf(r*r + g*g + b*b))); + ensure("magVec:Fail ", is_approx_equal(llcolor4.magVec(), (F32) sqrt(r*r + g*g + b*b))); } template<> template<> @@ -164,7 +164,7 @@ namespace tut F32 r = 0x20, g = 0xFFFF, b = 0xFF; LLColor4 llcolor4(r,g,b); F32 vecMag = llcolor4.normVec(); - F32 mag = fsqrtf(r*r + g*g + b*b); + F32 mag = (F32) sqrt(r*r + g*g + b*b); F32 oomag = 1.f / mag; F32 val1 = r * oomag, val2 = g * oomag, val3 = b * oomag; ensure("1:normVec failed ", (is_approx_equal(val1, llcolor4.mV[0]) && is_approx_equal(val2, llcolor4.mV[1]) && is_approx_equal(val3, llcolor4.mV[2]) && is_approx_equal(vecMag, mag))); diff --git a/indra/llmath/tests/v4coloru_test.cpp b/indra/llmath/tests/v4coloru_test.cpp index 6d84ba41ef..128f6f3564 100644 --- a/indra/llmath/tests/v4coloru_test.cpp +++ b/indra/llmath/tests/v4coloru_test.cpp @@ -135,7 +135,7 @@ namespace tut U8 r = 0x12, g = 0xFF, b = 0xAF; LLColor4U llcolor4u(r,g,b); ensure("magVecSquared:Fail ", is_approx_equal(llcolor4u.magVecSquared(), (F32)(r*r + g*g + b*b))); - ensure("magVec:Fail ", is_approx_equal(llcolor4u.magVec(), fsqrtf(r*r + g*g + b*b))); + ensure("magVec:Fail ", is_approx_equal(llcolor4u.magVec(), (F32) sqrt((F32) (r*r + g*g + b*b)))); } template<> template<> diff --git a/indra/llmath/tests/v4math_test.cpp b/indra/llmath/tests/v4math_test.cpp index b1f934e555..191ac864df 100644 --- a/indra/llmath/tests/v4math_test.cpp +++ b/indra/llmath/tests/v4math_test.cpp @@ -30,9 +30,9 @@ #include "../test/lltut.h" #include "llsd.h" -#include "../llquaternion.h" #include "../m4math.h" #include "../v4math.h" +#include "../llquaternion.h" namespace tut { @@ -96,7 +96,7 @@ namespace tut { F32 x = 10.f, y = -2.3f, z = -.023f; LLVector4 vec4(x,y,z); - ensure("magVec:Fail ", is_approx_equal(vec4.magVec(), fsqrtf(x*x + y*y + z*z))); + ensure("magVec:Fail ", is_approx_equal(vec4.magVec(), (F32) sqrt(x*x + y*y + z*z))); ensure("magVecSquared:Fail ", is_approx_equal(vec4.magVecSquared(), (x*x + y*y + z*z))); } @@ -337,7 +337,7 @@ namespace tut F32 val1,val2; LLVector4 vec4(x1,y1,z1),vec4a(x2,y2,z2); val1 = dist_vec(vec4,vec4a); - val2 = fsqrtf((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); + val2 = (F32) sqrt((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); ensure_equals("dist_vec: Fail ",val2, val1); val1 = dist_vec_squared(vec4,vec4a); val2 =((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); diff --git a/indra/llmath/v2math.cpp b/indra/llmath/v2math.cpp index 0180049b5d..a0cd642853 100644 --- a/indra/llmath/v2math.cpp +++ b/indra/llmath/v2math.cpp @@ -86,7 +86,7 @@ F32 dist_vec(const LLVector2 &a, const LLVector2 &b) { F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; - return fsqrtf( x*x + y*y ); + return (F32) sqrt( x*x + y*y ); } F32 dist_vec_squared(const LLVector2 &a, const LLVector2 &b) @@ -109,3 +109,18 @@ LLVector2 lerp(const LLVector2 &a, const LLVector2 &b, F32 u) a.mV[VX] + (b.mV[VX] - a.mV[VX]) * u, a.mV[VY] + (b.mV[VY] - a.mV[VY]) * u ); } + +LLSD LLVector2::getValue() const +{ + LLSD ret; + ret[0] = mV[0]; + ret[1] = mV[1]; + return ret; +} + +void LLVector2::setValue(LLSD& sd) +{ + mV[0] = (F32) sd[0].asReal(); + mV[1] = (F32) sd[1].asReal(); +} + diff --git a/indra/llmath/v2math.h b/indra/llmath/v2math.h index f50a5e6633..8d5db96f5e 100644 --- a/indra/llmath/v2math.h +++ b/indra/llmath/v2math.h @@ -60,6 +60,9 @@ class LLVector2 void set(const LLVector2 &vec); // Sets LLVector2 to vec void set(const F32 *vec); // Sets LLVector2 to vec + LLSD getValue() const; + void setValue(LLSD& sd); + void setVec(F32 x, F32 y); // deprecated void setVec(const LLVector2 &vec); // deprecated void setVec(const F32 *vec); // deprecated @@ -216,7 +219,7 @@ inline void LLVector2::setVec(const F32 *vec) inline F32 LLVector2::length(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1]); } inline F32 LLVector2::lengthSquared(void) const @@ -226,7 +229,7 @@ inline F32 LLVector2::lengthSquared(void) const inline F32 LLVector2::normalize(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1]); F32 oomag; if (mag > FP_MAG_THRESHOLD) @@ -253,7 +256,7 @@ inline bool LLVector2::isFinite() const // deprecated inline F32 LLVector2::magVec(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1]); } // deprecated @@ -265,7 +268,7 @@ inline F32 LLVector2::magVecSquared(void) const // deprecated inline F32 LLVector2::normVec(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1]); F32 oomag; if (mag > FP_MAG_THRESHOLD) diff --git a/indra/llmath/v3color.h b/indra/llmath/v3color.h index 327e452bf7..56cb2ae73e 100644 --- a/indra/llmath/v3color.h +++ b/indra/llmath/v3color.h @@ -278,7 +278,7 @@ inline F32 LLColor3::brightness(void) const inline F32 LLColor3::length(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); } inline F32 LLColor3::lengthSquared(void) const @@ -288,7 +288,7 @@ inline F32 LLColor3::lengthSquared(void) const inline F32 LLColor3::normalize(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); F32 oomag; if (mag) @@ -304,7 +304,7 @@ inline F32 LLColor3::normalize(void) // deprecated inline F32 LLColor3::magVec(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); } // deprecated @@ -316,7 +316,7 @@ inline F32 LLColor3::magVecSquared(void) const // deprecated inline F32 LLColor3::normVec(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); F32 oomag; if (mag) @@ -438,7 +438,7 @@ inline F32 distVec(const LLColor3 &a, const LLColor3 &b) F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; F32 z = a.mV[2] - b.mV[2]; - return fsqrtf( x*x + y*y + z*z ); + return (F32) sqrt( x*x + y*y + z*z ); } inline F32 distVec_squared(const LLColor3 &a, const LLColor3 &b) diff --git a/indra/llmath/v3dmath.h b/indra/llmath/v3dmath.h index 664c986ad0..578dcdc8ea 100644 --- a/indra/llmath/v3dmath.h +++ b/indra/llmath/v3dmath.h @@ -234,7 +234,7 @@ inline const LLVector3d& LLVector3d::setVec(const F64 *vec) inline F64 LLVector3d::normVec(void) { - F64 mag = fsqrtf(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); + F64 mag = (F32) sqrt(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); F64 oomag; if (mag > FP_MAG_THRESHOLD) @@ -256,7 +256,7 @@ inline F64 LLVector3d::normVec(void) inline F64 LLVector3d::normalize(void) { - F64 mag = fsqrtf(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); + F64 mag = (F32) sqrt(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); F64 oomag; if (mag > FP_MAG_THRESHOLD) @@ -280,7 +280,7 @@ inline F64 LLVector3d::normalize(void) inline F64 LLVector3d::magVec(void) const { - return fsqrtf(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); + return (F32) sqrt(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); } inline F64 LLVector3d::magVecSquared(void) const @@ -290,7 +290,7 @@ inline F64 LLVector3d::magVecSquared(void) const inline F64 LLVector3d::length(void) const { - return fsqrtf(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); + return (F32) sqrt(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); } inline F64 LLVector3d::lengthSquared(void) const @@ -400,7 +400,7 @@ inline F64 dist_vec(const LLVector3d &a, const LLVector3d &b) F64 x = a.mdV[0] - b.mdV[0]; F64 y = a.mdV[1] - b.mdV[1]; F64 z = a.mdV[2] - b.mdV[2]; - return fsqrtf( x*x + y*y + z*z ); + return (F32) sqrt( x*x + y*y + z*z ); } inline F64 dist_vec_squared(const LLVector3d &a, const LLVector3d &b) diff --git a/indra/llmath/v3math.cpp b/indra/llmath/v3math.cpp index fd08df02d8..e88b34f8d6 100644 --- a/indra/llmath/v3math.cpp +++ b/indra/llmath/v3math.cpp @@ -191,6 +191,28 @@ const LLVector3& LLVector3::rotVec(const LLQuaternion &q) return *this; } +const LLVector3& LLVector3::transVec(const LLMatrix4& mat) +{ + setVec( + mV[VX] * mat.mMatrix[VX][VX] + + mV[VY] * mat.mMatrix[VX][VY] + + mV[VZ] * mat.mMatrix[VX][VZ] + + mat.mMatrix[VX][VW], + + mV[VX] * mat.mMatrix[VY][VX] + + mV[VY] * mat.mMatrix[VY][VY] + + mV[VZ] * mat.mMatrix[VY][VZ] + + mat.mMatrix[VY][VW], + + mV[VX] * mat.mMatrix[VZ][VX] + + mV[VY] * mat.mMatrix[VZ][VY] + + mV[VZ] * mat.mMatrix[VZ][VZ] + + mat.mMatrix[VZ][VW]); + + return *this; +} + + const LLVector3& LLVector3::rotVec(F32 angle, const LLVector3 &vec) { if ( !vec.isExactlyZero() && angle ) diff --git a/indra/llmath/v3math.h b/indra/llmath/v3math.h index dbd38c1c3f..b6425087b3 100644 --- a/indra/llmath/v3math.h +++ b/indra/llmath/v3math.h @@ -34,6 +34,7 @@ class LLVector2; class LLVector4; class LLMatrix3; +class LLMatrix4; class LLVector3d; class LLQuaternion; @@ -109,6 +110,7 @@ class LLVector3 const LLVector3& rotVec(F32 angle, F32 x, F32 y, F32 z); // Rotates about x,y,z by angle radians const LLVector3& rotVec(const LLMatrix3 &mat); // Rotates by LLMatrix4 mat const LLVector3& rotVec(const LLQuaternion &q); // Rotates by LLQuaternion q + const LLVector3& transVec(const LLMatrix4& mat); // Transforms by LLMatrix4 mat (mat * v) const LLVector3& scaleVec(const LLVector3& vec); // scales per component by vec LLVector3 scaledVec(const LLVector3& vec) const; // get a copy of this vector scaled by vec @@ -274,7 +276,7 @@ inline void LLVector3::setVec(const F32 *vec) inline F32 LLVector3::normalize(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); F32 oomag; if (mag > FP_MAG_THRESHOLD) @@ -297,7 +299,7 @@ inline F32 LLVector3::normalize(void) // deprecated inline F32 LLVector3::normVec(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); F32 oomag; if (mag > FP_MAG_THRESHOLD) @@ -321,7 +323,7 @@ inline F32 LLVector3::normVec(void) inline F32 LLVector3::length(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); } inline F32 LLVector3::lengthSquared(void) const @@ -331,7 +333,7 @@ inline F32 LLVector3::lengthSquared(void) const inline F32 LLVector3::magVec(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); } inline F32 LLVector3::magVecSquared(void) const @@ -465,7 +467,7 @@ inline F32 dist_vec(const LLVector3 &a, const LLVector3 &b) F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; F32 z = a.mV[2] - b.mV[2]; - return fsqrtf( x*x + y*y + z*z ); + return (F32) sqrt( x*x + y*y + z*z ); } inline F32 dist_vec_squared(const LLVector3 &a, const LLVector3 &b) @@ -523,6 +525,21 @@ inline void update_min_max(LLVector3& min, LLVector3& max, const LLVector3& pos) } } +inline void update_min_max(LLVector3& min, LLVector3& max, const F32* pos) +{ + for (U32 i = 0; i < 3; i++) + { + if (min.mV[i] > pos[i]) + { + min.mV[i] = pos[i]; + } + if (max.mV[i] < pos[i]) + { + max.mV[i] = pos[i]; + } + } +} + inline F32 angle_between(const LLVector3& a, const LLVector3& b) { LLVector3 an = a; diff --git a/indra/llmath/v4color.h b/indra/llmath/v4color.h index 60d24e2e11..b047f86e6e 100644 --- a/indra/llmath/v4color.h +++ b/indra/llmath/v4color.h @@ -108,6 +108,7 @@ class LLColor4 const LLColor4& operator=(const LLColor3 &a); // Assigns vec3 to vec4 and returns vec4 + bool operator<(const LLColor4& rhs) const; friend std::ostream& operator<<(std::ostream& s, const LLColor4 &a); // Print a friend LLColor4 operator+(const LLColor4 &a, const LLColor4 &b); // Return vector a + b friend LLColor4 operator-(const LLColor4 &a, const LLColor4 &b); // Return vector a minus b @@ -385,7 +386,7 @@ inline const LLColor4& LLColor4::setAlpha(F32 a) inline F32 LLColor4::length(void) const { - return fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + return (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); } inline F32 LLColor4::lengthSquared(void) const @@ -395,7 +396,7 @@ inline F32 LLColor4::lengthSquared(void) const inline F32 LLColor4::normalize(void) { - F32 mag = fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + F32 mag = (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); F32 oomag; if (mag) @@ -411,7 +412,7 @@ inline F32 LLColor4::normalize(void) // deprecated inline F32 LLColor4::magVec(void) const { - return fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + return (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); } // deprecated @@ -423,7 +424,7 @@ inline F32 LLColor4::magVecSquared(void) const // deprecated inline F32 LLColor4::normVec(void) { - F32 mag = fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + F32 mag = (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); F32 oomag; if (mag) @@ -589,6 +590,23 @@ inline LLColor4 lerp(const LLColor4 &a, const LLColor4 &b, F32 u) a.mV[VW] + (b.mV[VW] - a.mV[VW]) * u); } +inline bool LLColor4::operator<(const LLColor4& rhs) const +{ + if (mV[0] != rhs.mV[0]) + { + return mV[0] < rhs.mV[0]; + } + if (mV[1] != rhs.mV[1]) + { + return mV[1] < rhs.mV[1]; + } + if (mV[2] != rhs.mV[2]) + { + return mV[2] < rhs.mV[2]; + } + + return mV[3] < rhs.mV[3]; +} void LLColor4::clamp() { diff --git a/indra/llmath/v4coloru.h b/indra/llmath/v4coloru.h index 7471aebe02..12da7e2dd7 100644 --- a/indra/llmath/v4coloru.h +++ b/indra/llmath/v4coloru.h @@ -294,7 +294,7 @@ inline const LLColor4U& LLColor4U::setAlpha(U8 a) inline F32 LLColor4U::length(void) const { - return fsqrtf( ((F32)mV[VX]) * mV[VX] + ((F32)mV[VY]) * mV[VY] + ((F32)mV[VZ]) * mV[VZ] ); + return (F32) sqrt( ((F32)mV[VX]) * mV[VX] + ((F32)mV[VY]) * mV[VY] + ((F32)mV[VZ]) * mV[VZ] ); } inline F32 LLColor4U::lengthSquared(void) const @@ -305,7 +305,7 @@ inline F32 LLColor4U::lengthSquared(void) const // deprecated inline F32 LLColor4U::magVec(void) const { - return fsqrtf( ((F32)mV[VX]) * mV[VX] + ((F32)mV[VY]) * mV[VY] + ((F32)mV[VZ]) * mV[VZ] ); + return (F32) sqrt( ((F32)mV[VX]) * mV[VX] + ((F32)mV[VY]) * mV[VY] + ((F32)mV[VZ]) * mV[VZ] ); } // deprecated diff --git a/indra/llmath/v4math.h b/indra/llmath/v4math.h index e7028626f9..623c8b2003 100644 --- a/indra/llmath/v4math.h +++ b/indra/llmath/v4math.h @@ -315,7 +315,7 @@ inline void LLVector4::setVec(const F32 *vec) inline F32 LLVector4::length(void) const { - return fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + return (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); } inline F32 LLVector4::lengthSquared(void) const @@ -325,7 +325,7 @@ inline F32 LLVector4::lengthSquared(void) const inline F32 LLVector4::magVec(void) const { - return fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + return (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); } inline F32 LLVector4::magVecSquared(void) const @@ -457,7 +457,7 @@ inline LLVector4 lerp(const LLVector4 &a, const LLVector4 &b, F32 u) inline F32 LLVector4::normalize(void) { - F32 mag = fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + F32 mag = (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); F32 oomag; if (mag > FP_MAG_THRESHOLD) @@ -480,7 +480,7 @@ inline F32 LLVector4::normalize(void) // deprecated inline F32 LLVector4::normVec(void) { - F32 mag = fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + F32 mag = (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); F32 oomag; if (mag > FP_MAG_THRESHOLD) diff --git a/indra/llmath/xform.h b/indra/llmath/xform.h index 5159c1cbfe..1b50749b3e 100644 --- a/indra/llmath/xform.h +++ b/indra/llmath/xform.h @@ -32,11 +32,11 @@ const F32 MAX_OBJECT_Z = 4096.f; // should match REGION_HEIGHT_METERS, Pre-havok4: 768.f const F32 MIN_OBJECT_Z = -256.f; -const F32 DEFAULT_MAX_PRIM_SCALE = 10.f; +const F32 DEFAULT_MAX_PRIM_SCALE = 64.f; +const F32 DEFAULT_MAX_PRIM_SCALE_NO_MESH = 10.f; const F32 MIN_PRIM_SCALE = 0.01f; const F32 MAX_PRIM_SCALE = 65536.f; // something very high but not near FLT_MAX - class LLXform { protected: |