From 3e80fa3dbc943de9b784fedc202ba38cf238f46d Mon Sep 17 00:00:00 2001 From: David Parks Date: Mon, 2 Nov 2009 19:55:37 +0000 Subject: Sync up with render-pipeline-7 ignore-dead-branch --- indra/llmath/llvolume.cpp | 350 +++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 344 insertions(+), 6 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index b8ef92f9a9..afa82ed399 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -46,6 +46,9 @@ #include "lldarray.h" #include "llvolume.h" #include "llstl.h" +#include "llsdserialize.h" +#include "zlib/zlib.h" + #define DEBUG_SILHOUETTE_BINORMALS 0 #define DEBUG_SILHOUETTE_NORMALS 0 // TomY: Use this to display normals using the silhouette @@ -1688,7 +1691,7 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mGenerateSingleFace = generate_single_face; generate(); - if (mParams.getSculptID().isNull()) + if (mParams.getSculptID().isNull() && params.getSculptType() == LL_SCULPT_TYPE_NONE) { createVolumeFaces(); } @@ -1839,6 +1842,295 @@ BOOL LLVolume::generate() return FALSE; } +bool LLVolumeFace::VertexData::operator<(const LLVolumeFace::VertexData& rhs)const +{ + const U8* l = (const U8*) this; + const U8* r = (const U8*) &rhs; + + for (U32 i = 0; i < sizeof(VertexData); ++i) + { + if (l[i] != r[i]) + { + return r[i] < l[i]; + } + } + + return false; +} + +bool LLVolumeFace::VertexData::operator==(const LLVolumeFace::VertexData& rhs)const +{ + const U8* l = (const U8*) this; + const U8* r = (const U8*) &rhs; + + for (U32 i = 0; i < sizeof(VertexData); ++i) + { + if (l[i] != r[i]) + { + return false; + } + } + + return true; +} + + +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 << "not a valid mesh asset!" << llendl; + return FALSE; + } + } + + std::string nm[] = + { + "impostor", + "low_lod", + "medium_lod", + "high_lod" + }; + + S32 lod = llclamp((S32) mDetail, 0, 3); + + while (lod < 4 && header[nm[lod]]["offset"].asInteger() == -1) + { + ++lod; + } + + if (lod >= 4) + { + llwarns << "Couldn't load model for given lod" << llendl; + return FALSE; + } + + is.seekg(header[nm[lod]]["offset"].asInteger(), std::ios_base::cur); + + + U8* result = NULL; + U32 cur_size = 0; + + { + //input stream is now pointing at a zlib compressed block of LLSD + //decompress block + z_stream strm; + + const U32 CHUNK = 65536; + + S32 size = header[nm[lod]]["size"].asInteger(); + U8 *in = new U8[size]; + is.read((char*) in, size); + + U8 out[CHUNK]; + + strm.zalloc = Z_NULL; + strm.zfree = Z_NULL; + strm.opaque = Z_NULL; + strm.avail_in = size; + strm.next_in = in; + + S32 ret = inflateInit(&strm); + + if (ret != Z_OK) + { + llerrs << "WTF?" << llendl; + } + + do + { + strm.avail_out = CHUNK; + strm.next_out = out; + ret = inflate(&strm, Z_NO_FLUSH); + if (ret == Z_STREAM_ERROR) + { + inflateEnd(&strm); + free(result); + delete [] in; + return FALSE; + } + + switch (ret) + { + case Z_NEED_DICT: + ret = Z_DATA_ERROR; + case Z_DATA_ERROR: + case Z_MEM_ERROR: + inflateEnd(&strm); + free(result); + delete [] in; + return FALSE; + break; + } + + U32 have = CHUNK-strm.avail_out; + + result = (U8*) realloc(result, cur_size + have); + memcpy(result+cur_size, out, have); + cur_size += have; + + } while (strm.avail_out == 0); + + inflateEnd(&strm); + delete [] in; + + if (ret != Z_STREAM_END) + { + free(result); + return FALSE; + } + } + + //result now points to the decompressed LLSD block + + LLSD mdl; + + { + std::string res_str((char*) result, cur_size); + std::istringstream istr(res_str); + + if (!LLSDSerialize::deserialize(mdl, istr, cur_size)) + { + llwarns << "not a valid mesh asset!" << llendl; + return FALSE; + } + } + + + free(result); + + + { + U32 face_count = mdl.size(); + + 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]; + + face.mHasBinormals = FALSE; + + //copy out indices + face.mIndices.resize(idx.size()/2); + if (idx.empty()) + { //why is there an empty index list? + 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.mVertices.resize(num_verts); + + LLVector3 min_pos; + LLVector3 max_pos; + LLVector2 min_tc; + LLVector2 max_tc; + + min_pos.setValue(mdl[i]["PositionDomain"]["Min"]); + max_pos.setValue(mdl[i]["PositionDomain"]["Max"]); + min_tc.setValue(mdl[i]["TexCoord0Domain"]["Min"]); + max_tc.setValue(mdl[i]["TexCoord0Domain"]["Max"]); + + F32 scale = llclamp((F32) mdl[i]["Scale"].asReal(), 1.f, 10.f); + + LLVector3 pos_range = max_pos - min_pos; + LLVector2 tc_range = max_tc - min_tc; + + LLVector3& min = face.mExtents[0]; + LLVector3& max = face.mExtents[1]; + + min = max = LLVector3(0,0,0); + + for (U32 j = 0; j < num_verts; ++j) + { + U16* v = (U16*) &(pos[j*3*2]); + + face.mVertices[j].mPosition.setVec( + (F32) v[0] / 65535.f * pos_range.mV[0] + min_pos.mV[0], + (F32) v[1] / 65535.f * pos_range.mV[1] + min_pos.mV[1], + (F32) v[2] / 65535.f * pos_range.mV[2] + min_pos.mV[2]); + + face.mVertices[j].mPosition *= scale; + + if (j == 0) + { + min = max = face.mVertices[j].mPosition; + } + else + { + update_min_max(min,max,face.mVertices[j].mPosition); + } + + U16* n = (U16*) &(norm[j*3*2]); + + face.mVertices[j].mNormal.setVec( + (F32) n[0] / 65535.f * 2.f - 1.f, + (F32) n[1] / 65535.f * 2.f - 1.f, + (F32) n[2] / 65535.f * 2.f - 1.f); + + U16* t = (U16*) &(tc[j*2*2]); + + face.mVertices[j].mTexCoord.setVec( + (F32) t[0] / 65535.f * tc_range.mV[0] + min_tc.mV[0], + (F32) t[1] / 65535.f * tc_range.mV[1] + min_tc.mV[1]); + } + + } + } + + mSculptLevel = 0; // success! + return TRUE; +} + +void LLVolume::copyVolumeFaces(LLVolume* volume) +{ + mVolumeFaces = volume->mVolumeFaces; + mSculptLevel = 0; +} + +S32 const LL_SCULPT_MESH_MAX_FACES = 8; + +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() { @@ -1864,6 +2156,11 @@ void LLVolume::createVolumeFaces() 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; @@ -1907,6 +2204,10 @@ void LLVolume::createVolumeFaces() if (face.mFlat && vf.mNumS > 2) { //flat inner faces have to copy vert normals vf.mNumS = vf.mNumS*2; + if (vf.mNumS < 0) + { + llerrs << "Volume face corruption detected." << llendl; + } } } else @@ -2309,7 +2610,6 @@ bool LLVolumeParams::operator<(const LLVolumeParams ¶ms) const return mSculptID < params.mSculptID; } - return mSculptType < params.mSculptType; @@ -3379,22 +3679,29 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, S32 face_mask) { LLMemType m1(LLMemType::MTYPE_VOLUME); - + vertices.clear(); normals.clear(); segments.clear(); + if (mParams.getSculptType() == LL_SCULPT_TYPE_MESH) + { + return; + } + S32 cur_index = 0; //for each face for (face_list_t::iterator iter = mVolumeFaces.begin(); iter != mVolumeFaces.end(); ++iter) { - const LLVolumeFace& face = *iter; + LLVolumeFace& face = *iter; - if (!(face_mask & (0x1 << cur_index++))) + if (!(face_mask & (0x1 << cur_index++)) || + face.mIndices.empty() || face.mEdge.empty()) { continue; } + if (face.mTypeMask & (LLVolumeFace::CAP_MASK)) { } @@ -3594,6 +3901,8 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, F32 closest_t = 2.f; // must be larger than 1 + end_face = llmin(end_face, getNumVolumeFaces()-1); + for (S32 i = start_face; i <= end_face; i++) { const LLVolumeFace &face = getVolumeFace((U32)i); @@ -4103,11 +4412,28 @@ BOOL LLVolumeParams::exportLegacyStream(std::ostream& output_stream) const return TRUE; } +LLSD LLVolumeParams::sculptAsLLSD() const +{ + LLSD sd = LLSD(); + sd["id"] = getSculptID(); + sd["type"] = getSculptType(); + + return sd; +} + +bool LLVolumeParams::sculptFromLLSD(LLSD& sd) +{ + setSculptID(sd["id"].asUUID(), (U8)sd["type"].asInteger()); + return true; +} + LLSD LLVolumeParams::asLLSD() const { LLSD sd = LLSD(); sd["path"] = mPathParams; sd["profile"] = mProfileParams; + sd["sculpt"] = sculptAsLLSD(); + return sd; } @@ -4115,6 +4441,8 @@ bool LLVolumeParams::fromLLSD(LLSD& sd) { mPathParams.fromLLSD(sd["path"]); mProfileParams.fromLLSD(sd["profile"]); + sculptFromLLSD(sd["sculpt"]); + return true; } @@ -4157,6 +4485,12 @@ const F32 MIN_CONCAVE_PATH_WEDGE = 0.111111f; // 1/9 unity // for collison purposes BOOL LLVolumeParams::isConvex() const { + if (!getSculptID().isNull()) + { + // can't determine, be safe and say no: + return FALSE; + } + F32 path_length = mPathParams.getEnd() - mPathParams.getBegin(); F32 hollow = mProfileParams.getHollow(); @@ -5011,7 +5345,11 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) if (!partial_build) { mIndices.resize(num_indices); - mEdge.resize(num_indices); + + if (volume->getParams().getSculptType() != LL_SCULPT_TYPE_MESH) + { + mEdge.resize(num_indices); + } } else { -- cgit v1.2.3 From 88292104d9a2332e6169f2add8f0b590bb22dbff Mon Sep 17 00:00:00 2001 From: David Parks Date: Wed, 4 Nov 2009 14:19:05 +0000 Subject: Fix for crash when loading some meshes. Added button to auto-fill LODs. --- indra/llmath/llvolume.cpp | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index afa82ed399..ddd1b4b3db 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1690,9 +1690,11 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mGenerateSingleFace = generate_single_face; - generate(); + mLODScaleBias.setVec(1,1,1); + if (mParams.getSculptID().isNull() && params.getSculptType() == LL_SCULPT_TYPE_NONE) { + generate(); createVolumeFaces(); } } -- cgit v1.2.3 From 10069e0e13e3214ba9320fdce915440b2e12f938 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 5 Nov 2009 19:58:10 -0600 Subject: Fix for prims all being 0 lod. Fix for dangling prim references. --- indra/llmath/llvolume.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index ddd1b4b3db..33a8d33ce1 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1692,9 +1692,10 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mLODScaleBias.setVec(1,1,1); + generate(); + if (mParams.getSculptID().isNull() && params.getSculptType() == LL_SCULPT_TYPE_NONE) { - generate(); createVolumeFaces(); } } -- cgit v1.2.3 From 4e420a36c67e611cd7d85652b43d9cd65315e563 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Sat, 7 Nov 2009 08:22:39 -0600 Subject: Fix for missing LOD spam. --- indra/llmath/llvolume.cpp | 21 +++++++++++++++------ 1 file changed, 15 insertions(+), 6 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 33a8d33ce1..c8ef911cc1 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1676,7 +1676,8 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mFaceMask = 0x0; mDetail = detail; mSculptLevel = -2; - + mLODScaleBias.setVec(1,1,1); + // set defaults if (mParams.getPathParams().getCurveType() == LL_PCODE_PATH_FLEXIBLE) { @@ -1690,8 +1691,6 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mGenerateSingleFace = generate_single_face; - mLODScaleBias.setVec(1,1,1); - generate(); if (mParams.getSculptID().isNull() && params.getSculptType() == LL_SCULPT_TYPE_NONE) @@ -1899,7 +1898,7 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) { if (!LLSDSerialize::deserialize(header, is, 1024*1024*1024)) { - llwarns << "not a valid mesh asset!" << llendl; + llwarns << "Mesh header parse error. Not a valid mesh asset!" << llendl; return FALSE; } } @@ -1921,8 +1920,18 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) if (lod >= 4) { - llwarns << "Couldn't load model for given lod" << llendl; - return FALSE; + lod = llclamp((S32) mDetail, 0, 3); + + while (lod >= 0 && header[nm[lod]]["offset"].asInteger() == -1) + { + --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); -- cgit v1.2.3 From c02702f3871979cb7745b49aa502ac3c71f77681 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 13 Nov 2009 17:01:56 -0600 Subject: CTS-7 Add hard edge threshold capability to normal generation. --- indra/llmath/llvolume.cpp | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 6286d1bcea..f252b2a232 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1876,6 +1876,18 @@ bool LLVolumeFace::VertexData::operator==(const LLVolumeFace::VertexData& rhs)co return true; } +bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs, F32 angle_cutoff) const +{ + bool retval = false; + if (rhs.mPosition == mPosition && rhs.mTexCoord == mTexCoord) + { + F32 cur_angle = rhs.mNormal*mNormal; + + retval = cur_angle > angle_cutoff; + } + + return retval; +} BOOL LLVolume::createVolumeFacesFromFile(const std::string& file_name) { -- cgit v1.2.3 From 81bfdcbfae4f203e60f00794966383b01475995b Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 18 Nov 2009 18:10:48 -0600 Subject: Tetrahedron displays in place of unloaded mesh. Still has some LOD issues. --- indra/llmath/llvolume.cpp | 96 ++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 94 insertions(+), 2 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index f252b2a232..84da1b3c62 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1925,7 +1925,9 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) S32 lod = llclamp((S32) mDetail, 0, 3); - while (lod < 4 && header[nm[lod]]["offset"].asInteger() == -1) + while (lod < 4 && + (header[nm[lod]]["offset"].asInteger() == -1 || + header[nm[lod]]["size"].asInteger() == 0 )) { ++lod; } @@ -1934,7 +1936,9 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) { lod = llclamp((S32) mDetail, 0, 3); - while (lod >= 0 && header[nm[lod]]["offset"].asInteger() == -1) + while (lod >= 0 && + (header[nm[lod]]["offset"].asInteger() == -1 || + header[nm[lod]]["size"].asInteger() == 0) ) { --lod; } @@ -2135,6 +2139,94 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) return TRUE; } +void tetrahedron_set_normal(LLVolumeFace::VertexData* cv) +{ + LLVector3 nrm = (cv[1].mPosition-cv[0].mPosition)%(cv[2].mPosition-cv[0].mPosition); + + nrm.normVec(); + + cv[0].mNormal = nrm; + cv[1].mNormal = nrm; + cv[2].mNormal = nrm; +} + +void LLVolume::makeTetrahedron() +{ + mVolumeFaces.clear(); + + LLVolumeFace face; + + F32 x = 0.5f; + LLVector3 p[] = + { //unit tetrahedron corners + LLVector3(x,x,x), + LLVector3(-x,-x,x), + LLVector3(-x,x,-x), + LLVector3(x,-x,-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].mPosition = p[1]; + cv[1].mPosition = p[0]; + cv[2].mPosition = p[2]; + + tetrahedron_set_normal(cv); + + face.mVertices.push_back(cv[0]); + face.mVertices.push_back(cv[1]); + face.mVertices.push_back(cv[2]); + + //side 2 + cv[0].mPosition = p[3]; + cv[1].mPosition = p[0]; + cv[2].mPosition = p[1]; + + tetrahedron_set_normal(cv); + + face.mVertices.push_back(cv[0]); + face.mVertices.push_back(cv[1]); + face.mVertices.push_back(cv[2]); + + //side 3 + cv[0].mPosition = p[3]; + cv[1].mPosition = p[1]; + cv[2].mPosition = p[2]; + + tetrahedron_set_normal(cv); + + face.mVertices.push_back(cv[0]); + face.mVertices.push_back(cv[1]); + face.mVertices.push_back(cv[2]); + + //side 4 + cv[0].mPosition = p[2]; + cv[1].mPosition = p[0]; + cv[2].mPosition = p[3]; + + tetrahedron_set_normal(cv); + + face.mVertices.push_back(cv[0]); + face.mVertices.push_back(cv[1]); + face.mVertices.push_back(cv[2]); + + //set index buffer + for (U32 i = 0; i < 12; i++) + { + face.mIndices.push_back(i); + } + + mVolumeFaces.push_back(face); + mSculptLevel = 0; +} + void LLVolume::copyVolumeFaces(LLVolume* volume) { mVolumeFaces = volume->mVolumeFaces; -- cgit v1.2.3 From 62233f22469cdc66042fc7bbbbd367dbb7212fde Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 24 Nov 2009 07:38:04 -0600 Subject: Fix for copying of tetrahedrons in place of mesh LODs. Fix for bad tetrahedron bounding box. Bad fix for simultaneous loading of multiple LODs. --- indra/llmath/llvolume.cpp | 31 +++++++++++++++++++++++++++++-- 1 file changed, 29 insertions(+), 2 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 84da1b3c62..515b1061f9 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1676,6 +1676,7 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mFaceMask = 0x0; mDetail = detail; mSculptLevel = -2; + mIsTetrahedron = FALSE; mLODScaleBias.setVec(1,1,1); // set defaults @@ -1905,7 +1906,7 @@ BOOL LLVolume::createVolumeFacesFromFile(const std::string& file_name) BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) { mSculptLevel = -1; // default is an error occured - + LLSD header; { if (!LLSDSerialize::deserialize(header, is, 1024*1024*1024)) @@ -2048,6 +2049,11 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) { U32 face_count = mdl.size(); + if (face_count == 0) + { + llerrs << "WTF?" << llendl; + } + mVolumeFaces.resize(face_count); for (U32 i = 0; i < face_count; ++i) @@ -2063,8 +2069,9 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) //copy out indices face.mIndices.resize(idx.size()/2); - if (idx.empty()) + if (idx.empty() || face.mIndices.size() < 3) { //why is there an empty index list? + llerrs <<"WTF?" << llendl; continue; } @@ -2150,6 +2157,11 @@ void tetrahedron_set_normal(LLVolumeFace::VertexData* cv) cv[2].mNormal = nrm; } +BOOL LLVolume::isTetrahedron() +{ + return mIsTetrahedron; +} + void LLVolume::makeTetrahedron() { mVolumeFaces.clear(); @@ -2165,6 +2177,9 @@ void LLVolume::makeTetrahedron() LLVector3(x,-x,-x) }; + face.mExtents[0].setVec(-x,-x,-x); + face.mExtents[1].setVec(x,x,x); + LLVolumeFace::VertexData cv[3]; //set texture coordinates @@ -2225,12 +2240,19 @@ void LLVolume::makeTetrahedron() mVolumeFaces.push_back(face); mSculptLevel = 0; + mIsTetrahedron = TRUE; } void LLVolume::copyVolumeFaces(LLVolume* volume) { + if (volume->isTetrahedron()) + { + llerrs << "WTF?" << llendl; + } + mVolumeFaces = volume->mVolumeFaces; mSculptLevel = 0; + mIsTetrahedron = FALSE; } S32 const LL_SCULPT_MESH_MAX_FACES = 8; @@ -2615,6 +2637,11 @@ void LLVolume::sculpt(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, LLMemType m1(LLMemType::MTYPE_VOLUME); U8 sculpt_type = mParams.getSculptType(); + if (sculpt_type & LL_SCULPT_TYPE_MASK == LL_SCULPT_TYPE_MESH) + { + llerrs << "WTF?" << llendl; + } + BOOL data_is_empty = FALSE; if (sculpt_width == 0 || sculpt_height == 0 || sculpt_components < 3 || sculpt_data == NULL) -- cgit v1.2.3 From 6d66910c6e2fbb25bf8b5c7b90e795f350342104 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 25 Nov 2009 11:35:41 -0600 Subject: Fix for spam on invalid mesh asset. Fix for index buffer overflow spam and crash in llvertexbuffer. --- indra/llmath/llvolume.cpp | 5 ----- 1 file changed, 5 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 515b1061f9..3e547aec6f 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -2245,11 +2245,6 @@ void LLVolume::makeTetrahedron() void LLVolume::copyVolumeFaces(LLVolume* volume) { - if (volume->isTetrahedron()) - { - llerrs << "WTF?" << llendl; - } - mVolumeFaces = volume->mVolumeFaces; mSculptLevel = 0; mIsTetrahedron = FALSE; -- cgit v1.2.3 From 062a2dd309ca5521d4045eb721496476f43d24dc Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 30 Nov 2009 15:32:10 -0600 Subject: Remove zero area triangles from meshes post-import. --- indra/llmath/llvolume.cpp | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 3e547aec6f..1d36da7f52 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1692,10 +1692,9 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mGenerateSingleFace = generate_single_face; - generate(); - if (mParams.getSculptID().isNull() && params.getSculptType() == LL_SCULPT_TYPE_NONE) { + generate(); createVolumeFaces(); } } -- cgit v1.2.3 From f039fa98efedc91965338ef53624279f99914205 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 30 Nov 2009 17:02:38 -0600 Subject: Fix for silly crash due to LLPrimitive having 0 texture entries. --- indra/llmath/llvolume.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 1d36da7f52..858bd9edea 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1692,9 +1692,10 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge mGenerateSingleFace = generate_single_face; + generate(); + if (mParams.getSculptID().isNull() && params.getSculptType() == LL_SCULPT_TYPE_NONE) { - generate(); createVolumeFaces(); } } -- cgit v1.2.3 From bb2631180a85df343e6d816fc37d881af31d49fb Mon Sep 17 00:00:00 2001 From: "Karl Stiefvater (qarl)" Date: Tue, 1 Dec 2009 17:40:52 -0600 Subject: CTS-4 Only part of an uploaded mesh renders. --- indra/llmath/llvolume.cpp | 1 - 1 file changed, 1 deletion(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 858bd9edea..fb2de92e35 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -2250,7 +2250,6 @@ void LLVolume::copyVolumeFaces(LLVolume* volume) mIsTetrahedron = FALSE; } -S32 const LL_SCULPT_MESH_MAX_FACES = 8; S32 LLVolume::getNumFaces() const { -- cgit v1.2.3 From 695969c77066de5032bdc9caefecf9b32b076b2f Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 11 Dec 2009 14:47:11 -0600 Subject: HTTP Mesh fetch FTW.. still busted --- indra/llmath/llvolume.cpp | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index fb2de92e35..44ff173502 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1952,8 +1952,12 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) } 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) +{ U8* result = NULL; U32 cur_size = 0; @@ -1964,7 +1968,6 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) const U32 CHUNK = 65536; - S32 size = header[nm[lod]]["size"].asInteger(); U8 *in = new U8[size]; is.read((char*) in, size); -- cgit v1.2.3 From 081fa98a47d2b592ada0fbb049ff959ac2cd6294 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 15 Dec 2009 17:43:05 -0600 Subject: HTTP Mesh transfer relatively blocking-free. --- indra/llmath/llvolume.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 44ff173502..844918432d 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -2171,7 +2171,7 @@ void LLVolume::makeTetrahedron() LLVolumeFace face; - F32 x = 0.5f; + F32 x = 0.25f; LLVector3 p[] = { //unit tetrahedron corners LLVector3(x,x,x), -- cgit v1.2.3 From 512a5736dceb1cc6db286e5f5baad867ac7a5935 Mon Sep 17 00:00:00 2001 From: "Karl Stiefvater (qarl)" Date: Wed, 23 Dec 2009 14:40:48 -0600 Subject: LODs for scene upload --- indra/llmath/llvolume.cpp | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 844918432d..de32070da1 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -3807,6 +3807,20 @@ S32 LLVolume::getNumTriangleIndices() const return count; } + +S32 LLVolume::getNumTriangles() const +{ + U32 triangle_count = 0; + + for (S32 i = 0; i < getNumVolumeFaces(); ++i) + { + triangle_count += getVolumeFace(i).mIndices.size()/3; + } + + return triangle_count; +} + + //----------------------------------------------------------------------------- // generateSilhouetteVertices() //----------------------------------------------------------------------------- -- cgit v1.2.3 From 024c0ebe19588f8452bae7ea01756fd7b4b30540 Mon Sep 17 00:00:00 2001 From: "Karl Stiefvater (qarl)" Date: Fri, 29 Jan 2010 14:33:04 -0600 Subject: enable mirror and invert flags for meshes. --- indra/llmath/llvolume.cpp | 59 ++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 56 insertions(+), 3 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index de32070da1..0328c09c9a 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1694,7 +1694,7 @@ LLVolume::LLVolume(const LLVolumeParams ¶ms, const F32 detail, const BOOL ge generate(); - if (mParams.getSculptID().isNull() && params.getSculptType() == LL_SCULPT_TYPE_NONE) + if (mParams.getSculptID().isNull() && mParams.getSculptType() == LL_SCULPT_TYPE_NONE) { createVolumeFaces(); } @@ -2142,6 +2142,59 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) (F32) t[1] / 65535.f * tc_range.mV[1] + min_tc.mV[1]); } + + // 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) + { + for (S32 i = 0; i < face.mVertices.size(); i++) + { + face.mVertices[i].mPosition.mV[VX] *= -1.0f; + face.mVertices[i].mNormal.mV[VX] *= -1.0f; + } + } + + if (do_invert_normals) + { + for (S32 i = 0; i < face.mVertices.size(); i++) + { + face.mVertices[i].mNormal *= -1.0f; + } + } + + if (do_reverse_triangles) + { + for (U32 j = 0; j < face.mIndices.size(); 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; + } + } + } } @@ -3838,7 +3891,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, normals.clear(); segments.clear(); - if (mParams.getSculptType() == LL_SCULPT_TYPE_MESH) + if ((mParams.getSculptType() & LL_SCULPT_TYPE_MASK) == LL_SCULPT_TYPE_MESH) { return; } @@ -5500,7 +5553,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { mIndices.resize(num_indices); - if (volume->getParams().getSculptType() != LL_SCULPT_TYPE_MESH) + if ((volume->getParams().getSculptType() & LL_SCULPT_TYPE_MASK) != LL_SCULPT_TYPE_MESH) { mEdge.resize(num_indices); } -- cgit v1.2.3 From 095a5e84408b47ef3c5610e111aefe51d77633ca Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Sat, 6 Feb 2010 17:33:12 -0600 Subject: Draw prims using triangle strips instead of triangle lists. --- indra/llmath/llvolume.cpp | 201 ++++++++++++++++++++++++++++++++++++---------- 1 file changed, 157 insertions(+), 44 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index df4c618ac1..cd7d7a12e3 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -4520,15 +4520,65 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) if (!partial_build) { - int idxs[] = {0,1,(grid_size+1)+1,(grid_size+1)+1,(grid_size+1),0}; - for(int gx = 0;gx=0;i--)mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); - }else{ - for(int i=0;i<6;i++)mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); + mTriStrip.clear(); + S32 idxs[] = {0,1,(grid_size+1)+1,(grid_size+1)+1,(grid_size+1),0}; + for(S32 gx = 0;gx=0;i--) + { + mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); + } + + if (gy == 0) + { + mTriStrip.push_back((gx+1)*(grid_size+1)); + mTriStrip.push_back((gx+1)*(grid_size+1)); + mTriStrip.push_back(gx*(grid_size+1)); + } + + mTriStrip.push_back(gy+1+(gx+1)*(grid_size+1)); + mTriStrip.push_back(gy+1+gx*(grid_size+1)); + + + if (gy == grid_size-1) + { + mTriStrip.push_back(gy+1+gx*(grid_size+1)); + } + } + else + { + for(S32 i=0;i<6;i++) + { + mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); + } + + if (gy == 0) + { + mTriStrip.push_back(gx*(grid_size+1)); + mTriStrip.push_back(gx*(grid_size+1)); + mTriStrip.push_back((gx+1)*(grid_size+1)); + } + + mTriStrip.push_back(gy+1+gx*(grid_size+1)); + mTriStrip.push_back(gy+1+(gx+1)*(grid_size+1)); + + if (gy == grid_size-1) + { + mTriStrip.push_back(gy+1+(gx+1)*(grid_size+1)); + } } } + + } + + if (mTriStrip.size()%2 == 1) + { + mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); } } @@ -4770,6 +4820,8 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) pt2--; } } + + makeTriStrip(); } else { @@ -4874,67 +4926,108 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) pt2--; } } + + makeTriStrip(); } } else { // Not hollow, generate the triangle fan. + U16 v1 = 2; + U16 v2 = 1; + if (mTypeMask & TOP_MASK) { - if (mTypeMask & OPEN_MASK) - { - // SOLID OPEN TOP - // Generate indices - // This is a tri-fan, so we reuse the same first point for all triangles. - for (S32 i = 0; i < (num_vertices - 2); i++) - { - mIndices[3*i] = num_vertices - 1; - mIndices[3*i+1] = i; - mIndices[3*i+2] = i + 1; - } - } - else - { - // SOLID CLOSED TOP - for (S32 i = 0; i < (num_vertices - 2); i++) - { - //MSMSM fix these caps but only for the un-cut case - mIndices[3*i] = num_vertices - 1; - mIndices[3*i+1] = i; - mIndices[3*i+2] = i + 1; - } - } + 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; + } + + //make tri strip + if (mTypeMask & OPEN_MASK) + { + makeTriStrip(); } else { - if (mTypeMask & OPEN_MASK) + S32 j = num_vertices-2; + if (mTypeMask & TOP_MASK) { - // SOLID OPEN BOTTOM - // Generate indices - // This is a tri-fan, so we reuse the same first point for all triangles. - for (S32 i = 0; i < (num_vertices - 2); i++) + mTriStrip.push_back(0); + for (S32 i = 1; i <= j; ++i) { - mIndices[3*i] = num_vertices - 1; - mIndices[3*i+1] = i + 1; - mIndices[3*i+2] = i; + mTriStrip.push_back(i); + if (i != j) + { + mTriStrip.push_back(j); + } + --j; } } else { - // SOLID CLOSED BOTTOM - for (S32 i = 0; i < (num_vertices - 2); i++) + mTriStrip.push_back(j); + for (S32 i = 1; i <= j; ++i) { - //MSMSM fix these caps but only for the un-cut case - mIndices[3*i] = num_vertices - 1; - mIndices[3*i+1] = i + 1; - mIndices[3*i+2] = 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]); + } } } + return TRUE; } +void LLVolumeFace::makeTriStrip() +{ + 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]); + } +} + void LLVolumeFace::createBinormals() { LLMemType m1(LLMemType::MTYPE_VOLUME); @@ -5135,9 +5228,14 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) if (!partial_build) { + mTriStrip.clear(); + // Now we generate the indices. for (t = 0; t < (mNumT-1); t++) { + //prepend terminating index to strip + mTriStrip.push_back(mNumS*t); + for (s = 0; s < (mNumS-1); s++) { mIndices[cur_index++] = s + mNumS*t; //bottom left @@ -5147,6 +5245,14 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) mIndices[cur_index++] = s+1 + mNumS*t; //bottom right mIndices[cur_index++] = s+1 + mNumS*(t+1); //top right + if (s == 0) + { + mTriStrip.push_back(s+mNumS*t); + mTriStrip.push_back(s+mNumS*(t+1)); + } + mTriStrip.push_back(s+1+mNumS*t); + mTriStrip.push_back(s+1+mNumS*(t+1)); + mEdge[cur_edge++] = (mNumS-1)*2*t+s*2+1; //bottom left/top right neighbor face if (t < mNumT-2) { //top right/top left neighbor face mEdge[cur_edge++] = (mNumS-1)*2*(t+1)+s*2+1; @@ -5187,6 +5293,13 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) } mEdge[cur_edge++] = (mNumS-1)*2*t+s*2; //top right/bottom left neighbor face } + //append terminating vertex to strip + mTriStrip.push_back(mNumS-1+mNumS*(t+1)); + } + + if (mTriStrip.size()%2 == 1) + { + mTriStrip.push_back(mTriStrip[mTriStrip.size()-1]); } } -- cgit v1.2.3 From 42df75bafeab49b408f23d79feb4f2213d2560eb Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 8 Feb 2010 10:14:11 -0600 Subject: Enable FBO multisampling for OSX. Fix bad triangle in prim caps. --- indra/llmath/llvolume.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index cd7d7a12e3..ae5c9bc8cf 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -4960,7 +4960,7 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) if (mTypeMask & TOP_MASK) { mTriStrip.push_back(0); - for (S32 i = 1; i <= j; ++i) + for (S32 i = 0; i <= j; ++i) { mTriStrip.push_back(i); if (i != j) @@ -4973,7 +4973,7 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) else { mTriStrip.push_back(j); - for (S32 i = 1; i <= j; ++i) + for (S32 i = 0; i <= j; ++i) { if (i != j) { -- cgit v1.2.3 From 2cb5b0b66ec9633d4c6563acf5ff9d0f7bc7cbf7 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 9 Feb 2010 12:26:09 -0600 Subject: consolidate button work in progress --- indra/llmath/llvolume.cpp | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index de32070da1..596c5fe231 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -5468,6 +5468,25 @@ void LLVolumeFace::createBinormals() } } +void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& transform, LLMatrix4& norm_transform) +{ + for (U32 i = 0; i < face.mVertices.size(); ++i) + { + VertexData v = face.mVertices[i]; + v.mPosition *= mat; + v.mNormal *= norm_transform; + + + mVertices.push_back(v); + } + + U16 offset = mIndices.size(); + for (U32 i = 0; i < face.mIndices.size(); ++i) + { + mIndices.push_back(face.mIndices[i]+offset); + } +} + BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { LLMemType m1(LLMemType::MTYPE_VOLUME); -- cgit v1.2.3 From ffcbbf4aaabc652c2050ca6147a9388217cfcaa7 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 11 Feb 2010 18:00:00 -0600 Subject: Multi-threaded asset uploading with proper ordering first draft. --- indra/llmath/llvolume.cpp | 20 ++++++++++++++++---- 1 file changed, 16 insertions(+), 4 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index d1716e1407..7e1517fba7 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -5614,19 +5614,31 @@ void LLVolumeFace::createBinormals() } } -void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& transform, LLMatrix4& norm_transform) +void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat, LLMatrix4& norm_mat) { + U16 offset = mVertices.size(); + + for (U32 i = 0; i < face.mVertices.size(); ++i) { VertexData v = face.mVertices[i]; - v.mPosition *= mat; - v.mNormal *= norm_transform; + v.mPosition = v.mPosition*mat; + v.mNormal = v.mNormal * norm_mat; mVertices.push_back(v); + + if (offset == 0 && i == 0) + { + mExtents[0] = mExtents[1] = v.mPosition; + } + else + { + update_min_max(mExtents[0], mExtents[1], v.mPosition); + } } - U16 offset = mIndices.size(); + for (U32 i = 0; i < face.mIndices.size(); ++i) { mIndices.push_back(face.mIndices[i]+offset); -- cgit v1.2.3 From ee8036712847315141c78d37646d629796442d09 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 17 Feb 2010 18:08:00 -0600 Subject: 16-bit limit awareness when consolidating models. --- indra/llmath/llvolume.cpp | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 7c98536e72..33a00b80ca 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -5619,6 +5619,10 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat, LLMatrix { U16 offset = mVertices.size(); + if (face.mVertices.size() + mVertices.size() > 65536) + { + llerrs << "Cannot append face -- 16-bit overflow will occur." << llendl; + } for (U32 i = 0; i < face.mVertices.size(); ++i) { -- cgit v1.2.3 From 066f9de07ecfcf142103f646695e5be63a22a667 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 23 Feb 2010 16:57:06 -0600 Subject: Fix for normals getting squished on consolidation. Replaced some magic numbers with constants. Switched up throttling of mesh upload HTTP posts to prevent overloading one capability at a time. Added some feedback on upload progress via debug text. Made debug text move with side panel (keep debug text from rendering on top of side panel). --- indra/llmath/llvolume.cpp | 1 + 1 file changed, 1 insertion(+) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 33a00b80ca..704308f20f 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -5630,6 +5630,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat, LLMatrix v.mPosition = v.mPosition*mat; v.mNormal = v.mNormal * norm_mat; + v.mNormal.normalize(); mVertices.push_back(v); -- cgit v1.2.3 From 3c78771acee787e087bd2e2391397794d4d98f6d Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 24 Feb 2010 22:02:01 -0600 Subject: Removed scale from model importer. Removed support for scale entry in mesh assets. Fixed MeshMaxConcurrentRequests being ignored. Added mesh download queue debug text. --- indra/llmath/llvolume.cpp | 4 ---- 1 file changed, 4 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 704308f20f..904786079f 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -2098,8 +2098,6 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) min_tc.setValue(mdl[i]["TexCoord0Domain"]["Min"]); max_tc.setValue(mdl[i]["TexCoord0Domain"]["Max"]); - F32 scale = llclamp((F32) mdl[i]["Scale"].asReal(), 1.f, 10.f); - LLVector3 pos_range = max_pos - min_pos; LLVector2 tc_range = max_tc - min_tc; @@ -2117,8 +2115,6 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) (F32) v[1] / 65535.f * pos_range.mV[1] + min_pos.mV[1], (F32) v[2] / 65535.f * pos_range.mV[2] + min_pos.mV[2]); - face.mVertices[j].mPosition *= scale; - if (j == 0) { min = max = face.mVertices[j].mPosition; -- cgit v1.2.3 From d60f5e937f2ed264f3e01eec7e32b9260e3d772f Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 9 Mar 2010 14:28:06 -0600 Subject: Tool tips for model preview. Rename "Impostor" to "Lowest" --- indra/llmath/llvolume.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 904786079f..d85c56046f 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1918,7 +1918,7 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) std::string nm[] = { - "impostor", + "lowest_lod", "low_lod", "medium_lod", "high_lod" -- cgit v1.2.3 From 4c0e2f79219913b57424bfe136b75a6a58fb8639 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 11 Mar 2010 12:02:37 -0600 Subject: "Generate Normals" is less busted now. --- indra/llmath/llvolume.cpp | 67 ++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 64 insertions(+), 3 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index d85c56046f..9ea3912a88 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1882,9 +1882,15 @@ bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs bool retval = false; if (rhs.mPosition == mPosition && rhs.mTexCoord == mTexCoord) { - F32 cur_angle = rhs.mNormal*mNormal; - - retval = cur_angle > angle_cutoff; + if (angle_cutoff > 1.f) + { + retval = (mNormal == rhs.mNormal); + } + else + { + F32 cur_angle = rhs.mNormal*mNormal; + retval = cur_angle > angle_cutoff; + } } return retval; @@ -4953,6 +4959,61 @@ BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) } } +void LLVolumeFace::optimize(F32 angle_cutoff) +{ + LLVolumeFace new_face; + + VertexMapData::PointMap point_map; + + //remove redundant vertices + for (U32 i = 0; i < mIndices.size(); ++i) + { + U16 index = mIndices[i]; + + LLVolumeFace::VertexData cv = mVertices[index]; + + BOOL found = FALSE; + VertexMapData::PointMap::iterator point_iter = point_map.find(cv.mPosition); + 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.mIndices.push_back((point_iter->second)[j].mIndex); + break; + } + } + } + + if (!found) + { + new_face.mVertices.push_back(cv); + U16 index = (U16) new_face.mVertices.size()-1; + new_face.mIndices.push_back(index); + + VertexMapData d; + d.mPosition = cv.mPosition; + d.mTexCoord = cv.mTexCoord; + d.mNormal = cv.mNormal; + d.mIndex = index; + if (point_iter != point_map.end()) + { + point_iter->second.push_back(d); + } + else + { + point_map[d.mPosition].push_back(d); + } + } + } + + mVertices = new_face.mVertices; + mIndices = new_face.mIndices; +} + void LerpPlanarVertex(LLVolumeFace::VertexData& v0, LLVolumeFace::VertexData& v1, LLVolumeFace::VertexData& v2, -- cgit v1.2.3 From 71d11af31083ced30da7b67a2a63e624c93b44a3 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Sat, 13 Mar 2010 17:39:32 -0600 Subject: Mesh cache. Has a bug. --- indra/llmath/llvolume.cpp | 28 ++++++++++++++-------------- 1 file changed, 14 insertions(+), 14 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 9ea3912a88..52a3fb2195 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1962,7 +1962,7 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) return unpackVolumeFaces(is, header[nm[lod]]["size"].asInteger()); } -BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) +bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) { U8* result = NULL; U32 cur_size = 0; @@ -2002,7 +2002,7 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) inflateEnd(&strm); free(result); delete [] in; - return FALSE; + return false; } switch (ret) @@ -2014,7 +2014,7 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) inflateEnd(&strm); free(result); delete [] in; - return FALSE; + return false; break; } @@ -2032,7 +2032,7 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) if (ret != Z_STREAM_END) { free(result); - return FALSE; + return false; } } @@ -2047,7 +2047,7 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) if (!LLSDSerialize::deserialize(mdl, istr, cur_size)) { llwarns << "not a valid mesh asset!" << llendl; - return FALSE; + return false; } } @@ -2074,7 +2074,7 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) LLVolumeFace& face = mVolumeFaces[i]; - face.mHasBinormals = FALSE; + face.mHasBinormals = false; //copy out indices face.mIndices.resize(idx.size()/2); @@ -2146,24 +2146,24 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) // modifier flags? - BOOL do_mirror = (mParams.getSculptType() & LL_SCULPT_FLAG_MIRROR); - BOOL do_invert = (mParams.getSculptType() &LL_SCULPT_FLAG_INVERT); + 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; + bool do_reflect_x = false; + bool do_reverse_triangles = false; + bool do_invert_normals = false; if (do_mirror) { - do_reflect_x = TRUE; + do_reflect_x = true; do_reverse_triangles = !do_reverse_triangles; } if (do_invert) { - do_invert_normals = TRUE; + do_invert_normals = true; do_reverse_triangles = !do_reverse_triangles; } @@ -2201,7 +2201,7 @@ BOOL LLVolume::unpackVolumeFaces(std::istream& is, S32 size) } mSculptLevel = 0; // success! - return TRUE; + return true; } void tetrahedron_set_normal(LLVolumeFace::VertexData* cv) -- cgit v1.2.3 From 807d835c2bfc5d794a74f9690d1fafbe55ff88cc Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 2 Apr 2010 14:43:05 -0500 Subject: First draft of skin weights in .mesh asset --- indra/llmath/llvolume.cpp | 135 ++++++++++++++++------------------------------ 1 file changed, 45 insertions(+), 90 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 52a3fb2195..c563af592f 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -47,8 +47,6 @@ #include "llvolume.h" #include "llstl.h" #include "llsdserialize.h" -#include "zlib/zlib.h" - #define DEBUG_SILHOUETTE_BINORMALS 0 #define DEBUG_SILHOUETTE_NORMALS 0 // TomY: Use this to display normals using the silhouette @@ -1964,97 +1962,15 @@ BOOL LLVolume::createVolumeFacesFromStream(std::istream& is) bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) { - U8* result = NULL; - U32 cur_size = 0; - - { - //input stream is now pointing at a zlib compressed block of LLSD - //decompress block - z_stream strm; - - const U32 CHUNK = 65536; - - U8 *in = new U8[size]; - is.read((char*) in, size); - - U8 out[CHUNK]; - - strm.zalloc = Z_NULL; - strm.zfree = Z_NULL; - strm.opaque = Z_NULL; - strm.avail_in = size; - strm.next_in = in; - - S32 ret = inflateInit(&strm); - - if (ret != Z_OK) - { - llerrs << "WTF?" << llendl; - } - - do - { - strm.avail_out = CHUNK; - strm.next_out = out; - ret = inflate(&strm, Z_NO_FLUSH); - if (ret == Z_STREAM_ERROR) - { - inflateEnd(&strm); - free(result); - delete [] in; - return false; - } - - switch (ret) - { - case Z_NEED_DICT: - ret = Z_DATA_ERROR; - case Z_DATA_ERROR: - case Z_MEM_ERROR: - inflateEnd(&strm); - free(result); - delete [] in; - return false; - break; - } - - U32 have = CHUNK-strm.avail_out; - - result = (U8*) realloc(result, cur_size + have); - memcpy(result+cur_size, out, have); - cur_size += have; - - } while (strm.avail_out == 0); - - inflateEnd(&strm); - delete [] in; - - if (ret != Z_STREAM_END) - { - free(result); - return false; - } - } - - //result now points to the decompressed LLSD block - + //input stream is now pointing at a zlib compressed block of LLSD + //decompress block LLSD mdl; - + if (!unzip_llsd(mdl, is, size)) { - std::string res_str((char*) result, cur_size); - std::istringstream istr(res_str); - - if (!LLSDSerialize::deserialize(mdl, istr, cur_size)) - { - llwarns << "not a valid mesh asset!" << llendl; - return false; - } + llwarns << "not a valid mesh asset!" << llendl; + return false; } - - - free(result); - - + { U32 face_count = mdl.size(); @@ -2094,11 +2010,50 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) U32 num_verts = pos.size()/(3*2); face.mVertices.resize(num_verts); + if (mdl[i].has("Weights")) + { + face.mWeights.resize(num_verts); + LLSD::Binary weights = mdl[i]["Weights"]; + + LLSD::Binary::iterator iter = weights.begin(); + + U32 cur_vertex = 0; + while (iter != weights.end()) + { + const S32 END_INFLUENCES = 0xFF; + U8 joint = *(iter++); + + U32 cur_influence = 0; + while (joint != END_INFLUENCES) + { + U16 influence = *(iter++); + influence = influence << 8; + influence |= *(iter++); + + F32 w = llmin((F32) influence / 65535.f, 0.99999f); + face.mWeights[cur_vertex].mV[cur_influence++] = (F32) joint + w; + + if (cur_influence >= 4) + { + joint = END_INFLUENCES; + } + else + { + joint = *(iter++); + } + } + + cur_vertex++; + iter++; + } + } + LLVector3 min_pos; LLVector3 max_pos; LLVector2 min_tc; LLVector2 max_tc; + min_pos.setValue(mdl[i]["PositionDomain"]["Min"]); max_pos.setValue(mdl[i]["PositionDomain"]["Max"]); min_tc.setValue(mdl[i]["TexCoord0Domain"]["Min"]); -- cgit v1.2.3 From 47ffcdb93d6e2ac1f9d497e43e0213c98d129254 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 6 Apr 2010 16:24:08 -0500 Subject: Rigged attachments (almost works). --- indra/llmath/llvolume.cpp | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index c563af592f..fdd48b9e9e 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -2013,22 +2013,23 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) if (mdl[i].has("Weights")) { face.mWeights.resize(num_verts); + LLSD::Binary weights = mdl[i]["Weights"]; - LLSD::Binary::iterator iter = weights.begin(); + U32 idx = 0; U32 cur_vertex = 0; - while (iter != weights.end()) + while (idx < weights.size() && cur_vertex < num_verts) { - const S32 END_INFLUENCES = 0xFF; - U8 joint = *(iter++); + const U8 END_INFLUENCES = 0xFF; + U8 joint = weights[idx++]; U32 cur_influence = 0; while (joint != END_INFLUENCES) { - U16 influence = *(iter++); + U16 influence = weights[idx++]; influence = influence << 8; - influence |= *(iter++); + influence |= weights[idx++]; F32 w = llmin((F32) influence / 65535.f, 0.99999f); face.mWeights[cur_vertex].mV[cur_influence++] = (F32) joint + w; @@ -2039,13 +2040,18 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) } else { - joint = *(iter++); + joint = weights[idx++]; } } cur_vertex++; - iter++; } + + if (cur_vertex != num_verts || idx != weights.size()) + { + llwarns << "Vertex weight count does not match vertex count!" << llendl; + } + } LLVector3 min_pos; -- cgit v1.2.3 From 07c0389f50ccef13ad2699e149dc4b87de3dbd70 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 27 Apr 2010 01:50:06 -0500 Subject: Proper byte ordering when decoding skin weights. --- indra/llmath/llvolume.cpp | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index fdd48b9e9e..9d2d157c76 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -2028,8 +2028,7 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) while (joint != END_INFLUENCES) { U16 influence = weights[idx++]; - influence = influence << 8; - influence |= weights[idx++]; + influence |= ((U16) weights[idx++] << 8); F32 w = llmin((F32) influence / 65535.f, 0.99999f); face.mWeights[cur_vertex].mV[cur_influence++] = (F32) joint + w; -- cgit v1.2.3 From d71716aa6dde434b6356cfe85e3a8fce376056dd Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 28 Apr 2010 02:53:12 -0500 Subject: Make LLVolume::createSide a little faster. --- indra/llmath/llvolume.cpp | 82 ++++++++++++++++++++++++++--------------------- 1 file changed, 45 insertions(+), 37 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 9d2d157c76..5ffc61ce9c 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -87,6 +87,8 @@ 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); @@ -5079,7 +5081,9 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) 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 2) ? mNumS/2 : mNumS; @@ -5771,15 +5779,6 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) mVertices[cur_vertex].mNormal = LLVector3(0,0,0); mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); - - if (cur_vertex == 0) - { - face_min = face_max = mesh[i].mPos; - } - else - { - update_min_max(face_min, face_max, mesh[i].mPos); - } cur_vertex++; @@ -5813,12 +5812,22 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) mVertices[cur_vertex].mNormal = LLVector3(0,0,0); mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); - update_min_max(face_min,face_max,mesh[i].mPos); - 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; @@ -5827,13 +5836,17 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) 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++) { @@ -5844,6 +5857,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) mIndices[cur_index++] = s+1 + mNumS*t; //bottom right mIndices[cur_index++] = s+1 + mNumS*(t+1); //top right +#if GEN_TRI_STRIP if (s == 0) { mTriStrip.push_back(s+mNumS*t); @@ -5851,6 +5865,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) } 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 @@ -5892,44 +5907,37 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) } 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 S32 i0 = mIndices[i*3+0]; - const S32 i1 = mIndices[i*3+1]; - const S32 i2 = mIndices[i*3+2]; - const VertexData& v0 = mVertices[i0]; - const VertexData& v1 = mVertices[i1]; - const VertexData& v2 = mVertices[i2]; + const U16* idx = &(mIndices[i*3]); + + VertexData* v[] = + { &mVertices[idx[0]], &mVertices[idx[1]], &mVertices[idx[2]] }; //calculate triangle normal - LLVector3 norm = (v0.mPosition-v1.mPosition) % (v0.mPosition-v2.mPosition); + LLVector3 norm = (v[0]->mPosition-v[1]->mPosition) % (v[0]->mPosition-v[2]->mPosition); - for (U32 j = 0; j < 3; j++) - { //add triangle normal to vertices - const S32 idx = mIndices[i*3+j]; - mVertices[idx].mNormal += norm; // * (weight_sum - d[j])/weight_sum; - } + v[0]->mNormal += norm; + v[1]->mNormal += norm; + v[2]->mNormal += norm; //even out quad contributions - if ((i & 1) == 0) - { - mVertices[i2].mNormal += norm; - } - else - { - mVertices[i1].mNormal += norm; - } + v[i%2+1]->mNormal += norm; } // adjust normals based on wrapping and stitching -- cgit v1.2.3 From 8919f4811a7dcaf47dc58159e0ba4ba042183325 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 21 May 2010 23:55:18 -0500 Subject: blah --- indra/llmath/llvolume.cpp | 262 +++++++++++++++++++++++++++++++++------------- 1 file changed, 191 insertions(+), 71 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 5ffc61ce9c..4e342b0b48 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -43,10 +43,12 @@ #include "v4math.h" #include "m4math.h" #include "m3math.h" +#include "llmatrix4a.h" #include "lldarray.h" #include "llvolume.h" #include "llstl.h" #include "llsdserialize.h" +#include "llvector4a.h" #define DEBUG_SILHOUETTE_BINORMALS 0 #define DEBUG_SILHOUETTE_NORMALS 0 // TomY: Use this to display normals using the silhouette @@ -1992,11 +1994,10 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) LLVolumeFace& face = mVolumeFaces[i]; - face.mHasBinormals = false; - //copy out indices - face.mIndices.resize(idx.size()/2); - if (idx.empty() || face.mIndices.size() < 3) + face.resizeIndices(idx.size()/2); + + if (idx.empty() || face.mNumIndices < 3) { //why is there an empty index list? llerrs <<"WTF?" << llendl; continue; @@ -2010,7 +2011,7 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) //copy out vertices U32 num_verts = pos.size()/(3*2); - face.mVertices.resize(num_verts); + face.resizeVertices(num_verts); if (mdl[i].has("Weights")) { @@ -2059,7 +2060,6 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) LLVector3 max_pos; LLVector2 min_tc; LLVector2 max_tc; - min_pos.setValue(mdl[i]["PositionDomain"]["Min"]); max_pos.setValue(mdl[i]["PositionDomain"]["Max"]); @@ -2074,36 +2074,44 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) min = max = LLVector3(0,0,0); + F32* pos_out = face.mPositions; + F32* norm_out = face.mNormals; + F32* tc_out = face.mTexCoords; + for (U32 j = 0; j < num_verts; ++j) { U16* v = (U16*) &(pos[j*3*2]); - face.mVertices[j].mPosition.setVec( - (F32) v[0] / 65535.f * pos_range.mV[0] + min_pos.mV[0], - (F32) v[1] / 65535.f * pos_range.mV[1] + min_pos.mV[1], - (F32) v[2] / 65535.f * pos_range.mV[2] + min_pos.mV[2]); + pos_out[0] = (F32) v[0] / 65535.f * pos_range.mV[0] + min_pos.mV[0]; + pos_out[1] = (F32) v[1] / 65535.f * pos_range.mV[1] + min_pos.mV[1]; + pos_out[2] = (F32) v[2] / 65535.f * pos_range.mV[2] + min_pos.mV[2]; + if (j == 0) { - min = max = face.mVertices[j].mPosition; + min = max = LLVector3(pos_out); } else { - update_min_max(min,max,face.mVertices[j].mPosition); + update_min_max(min,max,pos_out); } + pos_out += 4; + U16* n = (U16*) &(norm[j*3*2]); - face.mVertices[j].mNormal.setVec( - (F32) n[0] / 65535.f * 2.f - 1.f, - (F32) n[1] / 65535.f * 2.f - 1.f, - (F32) n[2] / 65535.f * 2.f - 1.f); + + norm_out[0] = (F32) n[0] / 65535.f * 2.f - 1.f; + norm_out[1] = (F32) n[1] / 65535.f * 2.f - 1.f; + norm_out[2] = (F32) n[2] / 65535.f * 2.f - 1.f; + norm_out += 4; U16* t = (U16*) &(tc[j*2*2]); - face.mVertices[j].mTexCoord.setVec( - (F32) t[0] / 65535.f * tc_range.mV[0] + min_tc.mV[0], - (F32) t[1] / 65535.f * tc_range.mV[1] + min_tc.mV[1]); + tc_out[0] = (F32) t[0] / 65535.f * tc_range.mV[0] + min_tc.mV[0]; + tc_out[1] = (F32) t[1] / 65535.f * tc_range.mV[1] + min_tc.mV[1]; + + tc_out += 8; } @@ -2133,24 +2141,29 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) if (do_reflect_x) { - for (S32 i = 0; i < face.mVertices.size(); i++) + LLVector4a* p = (LLVector4a*) face.mPositions; + LLVector4a* n = (LLVector4a*) face.mNormals; + + for (S32 i = 0; i < face.mNumVertices; i++) { - face.mVertices[i].mPosition.mV[VX] *= -1.0f; - face.mVertices[i].mNormal.mV[VX] *= -1.0f; + p[i].mul(-1.0f); + n[i].mul(-1.0f); } } if (do_invert_normals) { - for (S32 i = 0; i < face.mVertices.size(); i++) + LLVector4a* n = (LLVector4a*) face.mNormals; + + for (S32 i = 0; i < face.mNumVertices; i++) { - face.mVertices[i].mNormal *= -1.0f; + n[i].mul(-1.0f); } } if (do_reverse_triangles) { - for (U32 j = 0; j < face.mIndices.size(); j += 3) + for (U32 j = 0; j < face.mNumIndices; j += 3) { // swap the 2nd and 3rd index S32 swap = face.mIndices[j+1]; @@ -2215,9 +2228,28 @@ void LLVolume::makeTetrahedron() tetrahedron_set_normal(cv); - face.mVertices.push_back(cv[0]); - face.mVertices.push_back(cv[1]); - face.mVertices.push_back(cv[2]); + face.resizeVertices(12); + face.resizeIndices(12); + + LLVector4a* v = (LLVector4a*) face.mPositions; + LLVector4a* n = (LLVector4a*) face.mNormals; + LLVector2* tc = (LLVector2*) face.mTexCoords; + + v[0].load3(cv[0].mPosition.mV); + v[1].load3(cv[1].mPosition.mV); + v[2].load3(cv[2].mPosition.mV); + v += 3; + + n[0].load3(cv[0].mNormal.mV); + n[1].load3(cv[1].mNormal.mV); + n[2].load3(cv[2].mNormal.mV); + n += 3; + + tc[0] = cv[0].mTexCoord; + tc[1] = cv[1].mTexCoord; + tc[2] = cv[2].mTexCoord; + tc += 3; + //side 2 cv[0].mPosition = p[3]; @@ -2226,9 +2258,20 @@ void LLVolume::makeTetrahedron() tetrahedron_set_normal(cv); - face.mVertices.push_back(cv[0]); - face.mVertices.push_back(cv[1]); - face.mVertices.push_back(cv[2]); + v[0].load3(cv[0].mPosition.mV); + v[1].load3(cv[1].mPosition.mV); + v[2].load3(cv[2].mPosition.mV); + v += 3; + + n[0].load3(cv[0].mNormal.mV); + n[1].load3(cv[1].mNormal.mV); + n[2].load3(cv[2].mNormal.mV); + n += 3; + + tc[0] = cv[0].mTexCoord; + tc[1] = cv[1].mTexCoord; + tc[2] = cv[2].mTexCoord; + tc += 3; //side 3 cv[0].mPosition = p[3]; @@ -2237,9 +2280,20 @@ void LLVolume::makeTetrahedron() tetrahedron_set_normal(cv); - face.mVertices.push_back(cv[0]); - face.mVertices.push_back(cv[1]); - face.mVertices.push_back(cv[2]); + v[0].load3(cv[0].mPosition.mV); + v[1].load3(cv[1].mPosition.mV); + v[2].load3(cv[2].mPosition.mV); + v += 3; + + n[0].load3(cv[0].mNormal.mV); + n[1].load3(cv[1].mNormal.mV); + n[2].load3(cv[2].mNormal.mV); + n += 3; + + tc[0] = cv[0].mTexCoord; + tc[1] = cv[1].mTexCoord; + tc[2] = cv[2].mTexCoord; + tc += 3; //side 4 cv[0].mPosition = p[2]; @@ -2248,14 +2302,25 @@ void LLVolume::makeTetrahedron() tetrahedron_set_normal(cv); - face.mVertices.push_back(cv[0]); - face.mVertices.push_back(cv[1]); - face.mVertices.push_back(cv[2]); + v[0].load3(cv[0].mPosition.mV); + v[1].load3(cv[1].mPosition.mV); + v[2].load3(cv[2].mPosition.mV); + v += 3; + + n[0].load3(cv[0].mNormal.mV); + n[1].load3(cv[1].mNormal.mV); + n[2].load3(cv[2].mNormal.mV); + n += 3; + + tc[0] = cv[0].mTexCoord; + tc[1] = cv[1].mTexCoord; + tc[2] = cv[2].mTexCoord; + tc += 3; //set index buffer - for (U32 i = 0; i < 12; i++) + for (U16 i = 0; i < 12; i++) { - face.mIndices.push_back(i); + face.mIndices[i] = i; } mVolumeFaces.push_back(face); @@ -3831,7 +3896,7 @@ S32 LLVolume::getNumTriangles() const for (S32 i = 0; i < getNumVolumeFaces(); ++i) { - triangle_count += getVolumeFace(i).mIndices.size()/3; + triangle_count += getVolumeFace(i).mNumIndices/3; } return triangle_count; @@ -3844,13 +3909,22 @@ S32 LLVolume::getNumTriangles() const void LLVolume::generateSilhouetteVertices(std::vector &vertices, std::vector &normals, std::vector &segments, - const LLVector3& obj_cam_vec, - const LLMatrix4& mat, - const LLMatrix3& norm_mat, + const LLVector3& obj_cam_vec_in, + const LLMatrix4& mat_in, + const LLMatrix3& norm_mat_in, S32 face_mask) { LLMemType m1(LLMemType::MTYPE_VOLUME); + LLMatrix4a mat; + mat.loadu(mat_in); + + LLMatrix4a norm_mat; + norm_mat.loadu(norm_mat_in); + + LLVector4a obj_cam_vec; + obj_cam_vec.load3(obj_cam_vec_in.mV); + vertices.clear(); normals.clear(); segments.clear(); @@ -3868,7 +3942,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, LLVolumeFace& face = *iter; if (!(face_mask & (0x1 << cur_index++)) || - face.mIndices.empty() || face.mEdge.empty()) + face.mNumIndices == 0 || face.mEdge.empty()) { continue; } @@ -3885,7 +3959,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, #if DEBUG_SILHOUETTE_EDGE_MAP //for each triangle - U32 count = face.mIndices.size(); + U32 count = face.mNumIndices; for (U32 j = 0; j < count/3; j++) { //get vertices S32 v1 = face.mIndices[j*3+0]; @@ -3938,7 +4012,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, #elif DEBUG_SILHOUETTE_NORMALS //for each vertex - for (U32 j = 0; j < face.mVertices.size(); j++) { + for (U32 j = 0; j < face.mNumVertices; 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)); @@ -3964,26 +4038,36 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, //for each triangle std::vector fFacing; - vector_append(fFacing, face.mIndices.size()/3); - for (U32 j = 0; j < face.mIndices.size()/3; j++) + vector_append(fFacing, face.mNumIndices/3); + + LLVector4a* v = (LLVector4a*) face.mPositions; + LLVector4a* n = (LLVector4a*) face.mNormals; + + for (U32 j = 0; j < face.mNumIndices/3; j++) { //approximate normal S32 v1 = face.mIndices[j*3+0]; S32 v2 = face.mIndices[j*3+1]; S32 v3 = face.mIndices[j*3+2]; - LLVector3 norm = (face.mVertices[v1].mPosition - face.mVertices[v2].mPosition) % - (face.mVertices[v2].mPosition - face.mVertices[v3].mPosition); - - if (norm.magVecSquared() < 0.00000001f) + LLVector4a c1,c2; + c1.setSub(v[v1], v[v2]); + c2.setSub(v[v2], v[v3]); + + LLVector4a norm; + + norm.setCross3(c1, c2); + + if (norm.dot3(norm) < 0.00000001f) { fFacing[j] = AWAY | TOWARDS; } else { //get view vector - LLVector3 view = (obj_cam_vec-face.mVertices[v1].mPosition); - bool away = view * norm > 0.0f; + LLVector4a view; + view.setSub(obj_cam_vec, v[v1]); + bool away = view.dot3(norm) > 0.0f; if (away) { fFacing[j] = AWAY; @@ -3996,7 +4080,7 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, } //for each triangle - for (U32 j = 0; j < face.mIndices.size()/3; j++) + for (U32 j = 0; j < face.mNumIndices/3; j++) { if (fFacing[j] == (AWAY | TOWARDS)) { //this is a degenerate triangle @@ -4029,9 +4113,14 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, S32 v1 = face.mIndices[j*3+k]; S32 v2 = face.mIndices[j*3+((k+1)%3)]; - vertices.push_back(face.mVertices[v1].mPosition*mat); - LLVector3 norm1 = face.mVertices[v1].mNormal * norm_mat; - norm1.normVec(); + 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(); + LLVector3 norm1 = LLVector3(t[0], t[1], t[2]); normals.push_back(norm1); vertices.push_back(face.mVertices[v2].mPosition*mat); @@ -4088,7 +4177,7 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, genBinormals(i); } - for (U32 tri = 0; tri < face.mIndices.size()/3; tri++) + for (U32 tri = 0; tri < face.mNumIndices/3; tri++) { S32 index1 = face.mIndices[tri*3+0]; S32 index2 = face.mIndices[tri*3+1]; @@ -4928,7 +5017,7 @@ void LLVolumeFace::optimize(F32 angle_cutoff) VertexMapData::PointMap point_map; //remove redundant vertices - for (U32 i = 0; i < mIndices.size(); ++i) + for (U32 i = 0; i < mNumIndices; ++i) { U16 index = mIndices[i]; @@ -4953,7 +5042,7 @@ void LLVolumeFace::optimize(F32 angle_cutoff) if (!found) { new_face.mVertices.push_back(cv); - U16 index = (U16) new_face.mVertices.size()-1; + U16 index = (U16) new_face.mNumVertices-1; new_face.mIndices.push_back(index); VertexMapData d; @@ -5053,7 +5142,7 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) mVertices.clear(); } - S32 vtop = mVertices.size(); + S32 vtop = mNumVertices; for(int gx = 0;gx 65536) + if (face.mNumVertices + mNumVertices > 65536) { llerrs << "Cannot append face -- 16-bit overflow will occur." << llendl; } - for (U32 i = 0; i < face.mVertices.size(); ++i) + for (U32 i = 0; i < face.mNumVertices; ++i) { VertexData v = face.mVertices[i]; v.mPosition = v.mPosition*mat; @@ -5676,7 +5796,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat, LLMatrix } - for (U32 i = 0; i < face.mIndices.size(); ++i) + for (U32 i = 0; i < face.mNumIndices; ++i) { mIndices.push_back(face.mIndices[i]+offset); } @@ -5823,7 +5943,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) mCenter.clearVec(); face_min = face_max = mVertices[0].mPosition; - for (U32 i = 1; i < mVertices.size(); ++i) + for (U32 i = 1; i < mNumVertices; ++i) { update_min_max(face_min, face_max, mVertices[i].mPosition); } @@ -5922,7 +6042,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) } //generate normals - for (U32 i = 0; i < mIndices.size()/3; i++) //for each triangle + for (U32 i = 0; i < mNumIndices/3; i++) //for each triangle { const U16* idx = &(mIndices[i*3]); -- cgit v1.2.3 From 4d57cb3c0975ff0bcea0d6fb3498f2d90962ff16 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Sat, 22 May 2010 04:35:02 -0500 Subject: Vectorize/memory align buffers in llvolumeface WIP --- indra/llmath/llvolume.cpp | 720 +++++++++++++++++++++++----------------------- 1 file changed, 365 insertions(+), 355 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 4e342b0b48..01fe2be371 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1,4 +1,5 @@ /** + * @file llvolume.cpp * * $LicenseInfo:firstyear=2002&license=viewergpl$ @@ -89,8 +90,6 @@ 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); @@ -134,21 +133,25 @@ BOOL LLLineSegmentBoxIntersect(const LLVector3& start, const LLVector3& end, con // 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, +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 two_sided) { F32 u, v, t; /* find vectors for two edges sharing vert0 */ - LLVector3 edge1 = vert1 - vert0; + LLVector4a edge1; + edge1.setSub(vert1, vert0); - LLVector3 edge2 = vert2 - vert0;; + + LLVector4a edge2; + edge2.setSub(vert2, vert0); /* begin calculating determinant - also used to calculate U parameter */ - LLVector3 pvec = dir % edge2; - + LLVector4a pvec; + pvec.setCross3(dir, edge2); + /* if determinant is near zero, ray lies in plane of triangle */ - F32 det = edge1 * pvec; + F32 det = edge1.dot3(pvec); if (!two_sided) { @@ -158,10 +161,11 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons } /* calculate distance from vert0 to ray origin */ - LLVector3 tvec = orig - vert0; + LLVector4a tvec; + tvec.setSub(orig, vert0); /* calculate U parameter and test bounds */ - u = tvec * pvec; + u = tvec.dot3(pvec); if (u < 0.f || u > det) { @@ -169,17 +173,18 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons } /* prepare to test V parameter */ - LLVector3 qvec = tvec % edge1; + LLVector4a qvec; + qvec.setCross3(tvec, edge1); /* calculate V parameter and test bounds */ - v = dir * qvec; + v = dir.dot3(qvec); if (v < 0.f || u + v > det) { return FALSE; } /* calculate t, scale parameters, ray intersects triangle */ - t = edge2 * qvec; + t = edge2.dot3(qvec); F32 inv_det = 1.0 / det; t *= inv_det; u *= inv_det; @@ -195,20 +200,22 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons F32 inv_det = 1.0 / det; /* calculate distance from vert0 to ray origin */ - LLVector3 tvec = orig - vert0; + LLVector4a tvec; + tvec.setSub(orig, vert0); /* calculate U parameter and test bounds */ - u = (tvec * pvec) * inv_det; + u = (tvec.dot3(pvec)) * inv_det; if (u < 0.f || u > 1.f) { return FALSE; } /* prepare to test V parameter */ - LLVector3 qvec = tvec - edge1; + LLVector4a qvec; + qvec.setSub(tvec, edge1); /* calculate V parameter and test bounds */ - v = (dir * qvec) * inv_det; + v = (dir.dot3(qvec)) * inv_det; if (v < 0.f || u + v > 1.f) { @@ -216,7 +223,7 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons } /* calculate t, ray intersects triangle */ - t = (edge2 * qvec) * inv_det; + t = (edge2.dot3(qvec)) * inv_det; } if (intersection_a != NULL) @@ -4120,13 +4127,14 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, norm_mat.rotate(n[v1], t); t.normalize3Fast(); - LLVector3 norm1 = LLVector3(t[0], t[1], t[2]); - normals.push_back(norm1); + normals.push_back(LLVector3(t[0], t[1], t[2])); - vertices.push_back(face.mVertices[v2].mPosition*mat); - LLVector3 norm2 = face.mVertices[v2].mNormal * norm_mat; - norm2.normVec(); - normals.push_back(norm2); + 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()); } @@ -4177,6 +4185,10 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, genBinormals(i); } + LLVector4a starta, dira; + + LLVector4a* p = (LLVector4a*) face.mPositions; + for (U32 tri = 0; tri < face.mNumIndices/3; tri++) { S32 index1 = face.mIndices[tri*3+0]; @@ -4185,15 +4197,15 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, 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 (LLTriangleRayIntersect(p[index1], + p[index2], + p[index3], + starta, dira, &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; @@ -4201,27 +4213,35 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, { *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); + { + LLVector2* tc = (LLVector2*) face.mTexCoords; + *tex_coord = ((1.f - a - b) * tc[index1] + + a * tc[index2] + + b * tc[index3]); } if (normal != NULL) - { - *normal = ((1.f - a - b) * face.mVertices[index1].mNormal + - a * face.mVertices[index2].mNormal + - b * face.mVertices[index3].mNormal); + { + LLVector4* norm = (LLVector4*) face.mNormals; + + *normal = ((1.f - a - b) * LLVector3(norm[index1]) + + a * LLVector3(norm[index2]) + + b * LLVector3(norm[index3])); } if (bi_normal != NULL) - { - *bi_normal = ((1.f - a - b) * face.mVertices[index1].mBinormal + - a * face.mVertices[index2].mBinormal + - b * face.mVertices[index3].mBinormal); + { + LLVector4* binormal = (LLVector4*) face.mBinormals; + if (binormal) + { + *bi_normal = ((1.f - a - b) * LLVector3(binormal[index1]) + + a * LLVector3(binormal[index2]) + + b * LLVector3(binormal[index3])); + } } } @@ -4992,6 +5012,14 @@ std::ostream& operator<<(std::ostream &s, const LLVolume *volumep) return s; } +LLVolumeFace::~LLVolumeFace() +{ + _mm_free(mPositions); + _mm_free(mNormals); + _mm_free(mTexCoords); + _mm_free(mIndices); + _mm_free(mBinormals); +} BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) { @@ -5012,6 +5040,7 @@ BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) void LLVolumeFace::optimize(F32 angle_cutoff) { +#if 0 //disabling until a vectorized version is available LLVolumeFace new_face; VertexMapData::PointMap point_map; @@ -5063,6 +5092,7 @@ void LLVolumeFace::optimize(F32 angle_cutoff) mVertices = new_face.mVertices; mIndices = new_face.mIndices; +#endif } void LerpPlanarVertex(LLVolumeFace::VertexData& v0, @@ -5127,20 +5157,22 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) corners[1].mTexCoord=corners[2].mTexCoord; corners[2].mTexCoord=swap; } - baseVert.mBinormal = calc_binormal_from_triangle( + + LLVector4a binormal; + + calc_binormal_from_triangle( binormal, 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 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; S32 vtop = mNumVertices; for(int gx = 0;gx=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 + *out++ = (vtop+(gy*(grid_size+1))+gx+idxs[i]); + } } else { for(S32 i=0;i<6;i++) { - mIndices.push_back(vtop+(gy*(grid_size+1))+gx+idxs[i]); - } - -#if 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)); + *out++ = (vtop+(gy*(grid_size+1))+gx+idxs[i]); } - - 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; @@ -5267,11 +5262,25 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) num_vertices = profile.size(); num_indices = (profile.size() - 2)*3; - mVertices.resize(num_vertices); + if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK)) + { + resizeVertices(num_vertices+1); + allocateBinormals(num_vertices+1); - if (!partial_build) + if (!partial_build) + { + resizeIndices(num_indices+3); + } + } + else { - mIndices.resize(num_indices); + resizeVertices(num_vertices); + allocateBinormals(num_vertices); + + if (!partial_build) + { + resizeIndices(num_indices); + } } S32 max_s = volume->getProfile().getTotal(); @@ -5298,79 +5307,87 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) LLVector3& min = mExtents[0]; LLVector3& max = mExtents[1]; + LLVector2* tc = (LLVector2*) mTexCoords; + LLVector4a* pos = (LLVector4a*) mPositions; + LLVector4a* norm = (LLVector4a*) mNormals; + LLVector4a* binorm = (LLVector4a*) mBinormals; + // Copy the vertices into the array for (S32 i = 0; i < num_vertices; i++) { if (mTypeMask & TOP_MASK) { - mVertices[i].mTexCoord.mV[0] = profile[i].mV[0]+0.5f; - mVertices[i].mTexCoord.mV[1] = profile[i].mV[1]+0.5f; + tc[i].mV[0] = profile[i].mV[0]+0.5f; + tc[i].mV[1] = profile[i].mV[1]+0.5f; } else { // Mirror for underside. - mVertices[i].mTexCoord.mV[0] = profile[i].mV[0]+0.5f; - mVertices[i].mTexCoord.mV[1] = 0.5f - profile[i].mV[1]; + tc[i].mV[0] = profile[i].mV[0]+0.5f; + tc[i].mV[1] = 0.5f - profile[i].mV[1]; } - mVertices[i].mPosition = mesh[i + offset].mPos; + pos[i].load3(mesh[i + offset].mPos.mV); if (i == 0) { - min = max = mVertices[i].mPosition; - min_uv = max_uv = mVertices[i].mTexCoord; + min = max = mesh[i+offset].mPos; + min_uv = max_uv = tc[i]; } else { - update_min_max(min,max, mVertices[i].mPosition); - update_min_max(min_uv, max_uv, mVertices[i].mTexCoord); + update_min_max(min,max, mesh[i+offset].mPos); + update_min_max(min_uv, max_uv, tc[i]); } } mCenter = (min+max)*0.5f; cuv = (min_uv + max_uv)*0.5f; - LLVector3 binormal = calc_binormal_from_triangle( + LLVector4a binormal; + calc_binormal_from_triangle(binormal, mCenter, cuv, - mVertices[0].mPosition, mVertices[0].mTexCoord, - mVertices[1].mPosition, mVertices[1].mTexCoord); - binormal.normVec(); + mesh[0+offset].mPos, tc[0], + mesh[1+offset].mPos, tc[1]); + binormal.normalize3Fast(); + + LLVector4a normal; + LLVector4a d0, d1; + LLVector4a center; + + center.load3(mCenter.mV); - LLVector3 d0; - LLVector3 d1; - LLVector3 normal; + d0.setSub(center, pos[0]); + d1.setSub(center, pos[1]); - d0 = mCenter-mVertices[0].mPosition; - d1 = mCenter-mVertices[1].mPosition; + if (mTypeMask & TOP_MASK) + { + normal.setCross3(d0, d1); + } + else + { + normal.setCross3(d1, d0); + } - normal = (mTypeMask & TOP_MASK) ? (d0%d1) : (d1%d0); - normal.normVec(); + normal.normalize3Fast(); 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); + pos[num_vertices].load4a((F32*) ¢er.mQ); + tc[num_vertices] = cuv; num_vertices++; - if (!partial_build) - { - vector_append(mIndices, 3); - } } - for (S32 i = 0; i < num_vertices; i++) { - mVertices[i].mBinormal = binormal; - mVertices[i].mNormal = normal; + binorm[i].load4a((F32*) &binormal.mQ); + norm[i].load4a((F32*) &normal.mQ); } - mHasBinormals = TRUE; - if (partial_build) { return TRUE; @@ -5478,8 +5495,6 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) pt2--; } } - - makeTriStrip(); } else { @@ -5584,8 +5599,6 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) pt2--; } } - - makeTriStrip(); } } else @@ -5607,131 +5620,63 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) 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 < mNumIndices; 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) + if (!mBinormals) { + allocateBinormals(mNumVertices); + //generate binormals + LLStrider pos; + pos = (LLVector3*) mPositions; + pos.setStride(16); + + LLVector2* tc = (LLVector2*) mTexCoords; + LLVector4a* binorm = (LLVector4a*) mBinormals; + for (U32 i = 0; i < mNumIndices/3; i++) { //for each triangle - const VertexData& v0 = mVertices[mIndices[i*3+0]]; - const VertexData& v1 = mVertices[mIndices[i*3+1]]; - const VertexData& v2 = mVertices[mIndices[i*3+2]]; + const U16& i0 = mIndices[i*3+0]; + const U16& i1 = mIndices[i*3+1]; + const U16& i2 = mIndices[i*3+2]; //calculate binormal - LLVector3 binorm = calc_binormal_from_triangle(v0.mPosition, v0.mTexCoord, - v1.mPosition, v1.mTexCoord, - v2.mPosition, v2.mTexCoord); + LLVector4a binormal; + calc_binormal_from_triangle(binormal, + pos[i0], tc[i0], + pos[i1], tc[i1], + pos[i2], tc[i2]); - for (U32 j = 0; j < 3; j++) - { //add triangle normal to vertices - mVertices[mIndices[i*3+j]].mBinormal += binorm; // * (weight_sum - d[j])/weight_sum; - } + + //add triangle normal to vertices + binorm[i0].add(binormal); + binorm[i1].add(binormal); + binorm[i2].add(binormal); //even out quad contributions if (i % 2 == 0) { - mVertices[mIndices[i*3+2]].mBinormal += binorm; + binorm[i2].add(binormal); } else { - mVertices[mIndices[i*3+1]].mBinormal += binorm; + binorm[i1].add(binormal); } } //normalize binormals for (U32 i = 0; i < mNumVertices; i++) { - mVertices[i].mBinormal.normVec(); - mVertices[i].mNormal.normVec(); + binorm[i].normalize3Fast(); } - - mHasBinormals = TRUE; } } @@ -5754,6 +5699,13 @@ void LLVolumeFace::resizeVertices(S32 num_verts) mNumVertices = num_verts; } +void LLVolumeFace::allocateBinormals(S32 num_verts) +{ + _mm_free(mBinormals); + mBinormals = (F32*) _mm_malloc(num_verts*16, 16); +} + + void LLVolumeFace::resizeIndices(S32 num_indices) { _mm_free(mIndices); @@ -5761,44 +5713,107 @@ void LLVolumeFace::resizeIndices(S32 num_indices) //pad index block end to allow for QWORD reads S32 size = ((num_indices*2) + 0xF) & ~0xF; - mIndices = (U16*) _mm_malloc(size); + mIndices = (U16*) _mm_malloc(size,16); mNumIndices = num_indices; } -void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat, LLMatrix4& norm_mat) +void LLVolumeFace::fillFromLegacyData(std::vector& v, std::vector& idx) +{ + resizeVertices(v.size()); + resizeIndices(idx.size()); + + for (U32 i = 0; i < v.size(); ++i) + { + for (U32 j = 0; j < 3; ++j) + { + mPositions[i*4+j] = v[i].mPosition[j]; + mNormals[i*4+j] = v[i].mNormal[j]; + } + + mTexCoords[i*2+0] = v[i].mTexCoord.mV[0]; + mTexCoords[i*2+1] = v[i].mTexCoord.mV[1]; + } + + 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; - if (face.mNumVertices + mNumVertices > 65536) + S32 new_count = face.mNumVertices + mNumVertices; + + if (new_count > 65536) { llerrs << "Cannot append face -- 16-bit overflow will occur." << llendl; } + + F32* new_pos = (F32*) _mm_malloc(new_count*16, 16); + F32* new_norm = (F32*) _mm_malloc(new_count*16, 16); + F32* new_tc = (F32*) _mm_malloc((new_count*8+0xF) & ~0xF, 16); + + LLVector4a::memcpyNonAliased16(new_pos, mPositions, new_count*4); + LLVector4a::memcpyNonAliased16(new_norm, mNormals, new_count*4); + LLVector4a::memcpyNonAliased16(new_tc, mTexCoords, new_count*2); + + _mm_free(mPositions); + _mm_free(mNormals); + _mm_free(mTexCoords); + + mPositions = new_pos; + mNormals = new_norm; + mTexCoords = new_tc; + + mNumVertices = new_count; + + LLVector4a* dst_pos = (LLVector4a*) mPositions+offset; + LLVector2* dst_tc = (LLVector2*) mTexCoords+offset; + LLVector4a* dst_norm = (LLVector4a*) mNormals+offset; + + LLVector4a* src_pos = (LLVector4a*) face.mPositions; + LLVector2* src_tc = (LLVector2*) face.mTexCoords; + LLVector4a* src_norm = (LLVector4a*) face.mNormals; + + LLMatrix4a mat, norm_mat; + mat.loadu(mat_in); + norm_mat.loadu(norm_mat_in); + for (U32 i = 0; i < face.mNumVertices; ++i) { - VertexData v = face.mVertices[i]; - v.mPosition = v.mPosition*mat; - v.mNormal = v.mNormal * norm_mat; + mat.affineTransform(src_pos[i], dst_pos[i]); + norm_mat.rotate(src_norm[i], dst_norm[i]); + dst_norm[i].normalize3Fast(); - v.mNormal.normalize(); - - mVertices.push_back(v); + dst_tc[i] = src_tc[i]; if (offset == 0 && i == 0) { - mExtents[0] = mExtents[1] = v.mPosition; + mExtents[0] = mExtents[1] = LLVector3((F32*) &(dst_pos[i].mQ)); } else { - update_min_max(mExtents[0], mExtents[1], v.mPosition); + update_min_max(mExtents[0], mExtents[1], (F32*) &(dst_pos[i].mQ)); } } - + + new_count = mNumIndices + face.mNumIndices; + U16* new_indices = (U16*) _mm_malloc((new_count*2+0xF) & ~0xF, 16); + LLVector4a::memcpyNonAliased16((F32*) new_indices, (F32*) mIndices, new_count/2); + _mm_free(mIndices); + mIndices = new_indices; + mNumIndices = new_count; + + U16* dst_idx = mIndices+offset; + for (U32 i = 0; i < face.mNumIndices; ++i) { - mIndices.push_back(face.mIndices[i]+offset); + dst_idx[i] = face.mIndices[i]+offset; } } @@ -5828,21 +5843,20 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) num_vertices = mNumS*mNumT; num_indices = (mNumS-1)*(mNumT-1)*6; - mVertices.resize(num_vertices); - if (!partial_build) { - mIndices.resize(num_indices); + resizeVertices(num_vertices); + resizeIndices(num_indices); if ((volume->getParams().getSculptType() & LL_SCULPT_TYPE_MASK) != LL_SCULPT_TYPE_MESH) { mEdge.resize(num_indices); } } - else - { - mHasBinormals = FALSE; - } + + 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; @@ -5894,21 +5908,21 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) i = mBeginS + s + max_s*t; } - mVertices[cur_vertex].mPosition = mesh[i].mPos; - mVertices[cur_vertex].mTexCoord = LLVector2(ss,tt); + pos[cur_vertex].load3(mesh[i].mPos.mV); + tc[cur_vertex] = LLVector2(ss,tt); - mVertices[cur_vertex].mNormal = LLVector3(0,0,0); - mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); + norm[cur_vertex].clear(); cur_vertex++; if ((mTypeMask & INNER_MASK) && (mTypeMask & FLAT_MASK) && mNumS > 2 && s > 0) { - mVertices[cur_vertex].mPosition = mesh[i].mPos; - mVertices[cur_vertex].mTexCoord = LLVector2(ss,tt); + + pos[cur_vertex].load3(mesh[i].mPos.mV); + tc[cur_vertex] = LLVector2(ss,tt); - mVertices[cur_vertex].mNormal = LLVector3(0,0,0); - mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); + norm[cur_vertex].clear(); + cur_vertex++; } } @@ -5926,12 +5940,10 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) i = mBeginS + s + max_s*t; ss = profile[mBeginS + s].mV[2] - begin_stex; - mVertices[cur_vertex].mPosition = mesh[i].mPos; - mVertices[cur_vertex].mTexCoord = LLVector2(ss,tt); - - mVertices[cur_vertex].mNormal = LLVector3(0,0,0); - mVertices[cur_vertex].mBinormal = LLVector3(0,0,0); - + pos[cur_vertex].load3(mesh[i].mPos.mV); + tc[cur_vertex] = LLVector2(ss,tt); + norm[cur_vertex].clear(); + cur_vertex++; } } @@ -5942,10 +5954,11 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) LLVector3& face_max = mExtents[1]; mCenter.clearVec(); - face_min = face_max = mVertices[0].mPosition; + face_min = face_max = LLVector3((F32*) &(pos[i].mQ)); + for (U32 i = 1; i < mNumVertices; ++i) { - update_min_max(face_min, face_max, mVertices[i].mPosition); + update_min_max(face_min, face_max, (F32*) &(pos[i].mQ)); } mCenter = (face_min + face_max) * 0.5f; @@ -5956,18 +5969,9 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) 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 @@ -5977,16 +5981,6 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) mIndices[cur_index++] = s+1 + mNumS*t; //bottom right mIndices[cur_index++] = s+1 + mNumS*(t+1); //top right -#if 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; @@ -6027,52 +6021,55 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) } 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 < mNumIndices/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; + 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 - v[i%2+1]->mNormal += norm; + n[i%2+1]->add(c); } // adjust normals based on wrapping and stitching - BOOL s_bottom_converges = ((mVertices[0].mPosition - mVertices[mNumS*(mNumT-2)].mPosition).magVecSquared() < 0.000001f); - BOOL s_top_converges = ((mVertices[mNumS-1].mPosition - mVertices[mNumS*(mNumT-2)+mNumS-1].mPosition).magVecSquared() < 0.000001f); + LLVector4a top; + top.setSub(pos[0], pos[mNumS*(mNumT-2)]); + BOOL s_bottom_converges = (top.dot3(top) < 0.000001f); + + top.setSub(pos[mNumS-1], pos[mNumS*(mNumT-2)+mNumS-1]); + BOOL s_top_converges = (top.dot3(top) < 0.000001f); + if (sculpt_stitching == LL_SCULPT_TYPE_NONE) // logic for non-sculpt volumes { if (volume->getPath().isOpen() == FALSE) { //wrap normals on T for (S32 i = 0; i < mNumS; i++) { - LLVector3 norm = mVertices[i].mNormal + mVertices[mNumS*(mNumT-1)+i].mNormal; - mVertices[i].mNormal = norm; - mVertices[mNumS*(mNumT-1)+i].mNormal = norm; + LLVector4a n; + n.setAdd(norm[i], norm[mNumS*(mNumT-1)+i]); + norm[i] = n; + norm[mNumS*(mNumT-1)+i] = n; } } @@ -6080,9 +6077,10 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { //wrap normals on S for (S32 i = 0; i < mNumT; i++) { - LLVector3 norm = mVertices[mNumS*i].mNormal + mVertices[mNumS*i+mNumS-1].mNormal; - mVertices[mNumS * i].mNormal = norm; - mVertices[mNumS * i+mNumS-1].mNormal = norm; + LLVector4a n; + n.setAdd(norm[mNumS*i], norm[mNumS*i+mNumS-1]); + norm[mNumS * i] = n; + norm[mNumS * i+mNumS-1] = n; } } @@ -6093,7 +6091,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { //all lower S have same normal for (S32 i = 0; i < mNumT; i++) { - mVertices[mNumS*i].mNormal = LLVector3(1,0,0); + norm[mNumS*i].set(1,0,0); } } @@ -6101,7 +6099,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { //all upper S have same normal for (S32 i = 0; i < mNumT; i++) { - mVertices[mNumS*i+mNumS-1].mNormal = LLVector3(-1,0,0); + norm[mNumS*i+mNumS-1].set(-1,0,0); } } } @@ -6129,30 +6127,33 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { // average normals for north pole - LLVector3 average(0.0, 0.0, 0.0); + LLVector4a average; + average.clear(); + for (S32 i = 0; i < mNumS; i++) { - average += mVertices[i].mNormal; + average.add(norm[i]); } // set average for (S32 i = 0; i < mNumS; i++) { - mVertices[i].mNormal = average; + norm[i] = average; } // average normals for south pole - average = LLVector3(0.0, 0.0, 0.0); + average.clear(); + for (S32 i = 0; i < mNumS; i++) { - average += mVertices[i + mNumS * (mNumT - 1)].mNormal; + average.add(norm[i + mNumS * (mNumT - 1)]); } // set average for (S32 i = 0; i < mNumS; i++) { - mVertices[i + mNumS * (mNumT - 1)].mNormal = average; + norm[i + mNumS * (mNumT - 1)] = average; } } @@ -6162,23 +6163,22 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) { for (S32 i = 0; i < mNumT; i++) { - LLVector3 norm = mVertices[mNumS*i].mNormal + mVertices[mNumS*i+mNumS-1].mNormal; - mVertices[mNumS * i].mNormal = norm; - mVertices[mNumS * i+mNumS-1].mNormal = norm; + LLVector4a n; + n.setAdd(norm[mNumS*i], norm[mNumS*i+mNumS-1]); + norm[mNumS * i] = n; + norm[mNumS * i+mNumS-1] = n; } } - - if (wrap_t) { for (S32 i = 0; i < mNumS; i++) { - LLVector3 norm = mVertices[i].mNormal + mVertices[mNumS*(mNumT-1)+i].mNormal; - mVertices[i].mNormal = norm; - mVertices[mNumS*(mNumT-1)+i].mNormal = norm; + LLVector4a n; + n.setAdd(norm[i], norm[mNumS*(mNumT-1)+i]); + norm[i] = n; + norm[mNumS*(mNumT-1)+i] = n; } - } } @@ -6188,7 +6188,8 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) // Finds binormal based on three vertices with texture coordinates. // Fills in dummy values if the triangle has degenerate texture coordinates. -LLVector3 calc_binormal_from_triangle( +void calc_binormal_from_triangle(LLVector4a& binormal, + const LLVector3& pos0, const LLVector2& tex0, const LLVector3& pos1, @@ -6196,33 +6197,42 @@ LLVector3 calc_binormal_from_triangle( 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] ); + LLVector4a rx0; rx0.set( pos0.mV[VX], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a rx1; rx1.set( pos1.mV[VX], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a rx2; rx2.set( 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] ); + LLVector4a ry0; ry0.set( pos0.mV[VY], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a ry1; ry1.set( pos1.mV[VY], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a ry2; ry2.set( 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] ); + LLVector4a rz0; rz0.set( pos0.mV[VZ], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a rz1; rz1.set( pos1.mV[VZ], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a rz2; rz2.set( 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); + 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.mV[VX] && r1.mV[VX] && r2.mV[VX] ) + if( r0[VX] && r1[VX] && r2[VX] ) { - LLVector3 binormal( - -r0.mV[VZ] / r0.mV[VX], - -r1.mV[VZ] / r1.mV[VX], - -r2.mV[VZ] / r2.mV[VX]); + binormal.set( + -r0[VZ] / r0[VX], + -r1[VZ] / r1[VX], + -r2[VZ] / r2[VX]); // binormal.normVec(); - return binormal; } else { - return LLVector3( 0, 1 , 0 ); + binormal.set( 0, 1 , 0 ); } } -- cgit v1.2.3 From d7cab99ba74b214c557d9b5e02a7800b6a25c109 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Sat, 22 May 2010 12:37:53 -0500 Subject: Fix for a couple dumb mistakes. --- indra/llmath/llvolume.cpp | 38 +++++++++++++++++++++++++++----------- 1 file changed, 27 insertions(+), 11 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 01fe2be371..88969af4bd 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -5206,7 +5206,7 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) if (!partial_build) { - resizeIndices(grid_size*6); + resizeIndices(grid_size*grid_size*6); U16* out = mIndices; @@ -5689,12 +5689,21 @@ void LLVolumeFace::resizeVertices(S32 num_verts) mBinormals = NULL; - mPositions = (F32*) _mm_malloc(num_verts*16, 16); - mNormals = (F32*) _mm_malloc(num_verts*16, 16); + if (num_verts) + { + mPositions = (F32*) _mm_malloc(num_verts*16, 16); + mNormals = (F32*) _mm_malloc(num_verts*16, 16); - //pad texture coordinate block end to allow for QWORD reads - S32 size = ((num_verts*8) + 0xF) & ~0xF; - mTexCoords = (F32*) _mm_malloc(size, 16); + //pad texture coordinate block end to allow for QWORD reads + S32 size = ((num_verts*8) + 0xF) & ~0xF; + mTexCoords = (F32*) _mm_malloc(size, 16); + } + else + { + mPositions = NULL; + mNormals = NULL; + mTexCoords = NULL; + } mNumVertices = num_verts; } @@ -5710,10 +5719,17 @@ void LLVolumeFace::resizeIndices(S32 num_indices) { _mm_free(mIndices); - //pad index block end to allow for QWORD reads - S32 size = ((num_indices*2) + 0xF) & ~0xF; - - mIndices = (U16*) _mm_malloc(size,16); + if (num_indices) + { + //pad index block end to allow for QWORD reads + S32 size = ((num_indices*2) + 0xF) & ~0xF; + + mIndices = (U16*) _mm_malloc(size,16); + } + else + { + mIndices = NULL; + } mNumIndices = num_indices; } @@ -5954,7 +5970,7 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) LLVector3& face_max = mExtents[1]; mCenter.clearVec(); - face_min = face_max = LLVector3((F32*) &(pos[i].mQ)); + face_min = face_max = LLVector3((F32*) &(pos[0].mQ)); for (U32 i = 1; i < mNumVertices; ++i) { -- cgit v1.2.3 From 8c32e3bf29337e330a313d0e4865ebd03ad9ca50 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 24 May 2010 14:03:10 -0500 Subject: Fix for bad indexes on cube faces. Extra validation on vertex buffers. --- indra/llmath/llvolume.cpp | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 88969af4bd..31544016db 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -4173,7 +4173,7 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, for (S32 i = start_face; i <= end_face; i++) { - const LLVolumeFace &face = getVolumeFace((U32)i); + LLVolumeFace &face = mVolumeFaces[i]; LLVector3 box_center = (face.mExtents[0] + face.mExtents[1]) / 2.f; LLVector3 box_size = face.mExtents[1] - face.mExtents[0]; @@ -4235,6 +4235,10 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, if (bi_normal != NULL) { + if (!face.mBinormals) + { + face.createBinormals(); + } LLVector4* binormal = (LLVector4*) face.mBinormals; if (binormal) { @@ -5174,7 +5178,6 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) LLVector4a* binorm = (LLVector4a*) mBinormals; LLVector2* tc = (LLVector2*) mTexCoords; - S32 vtop = mNumVertices; for(int gx = 0;gx=0;i--) { - *out++ = (vtop+(gy*(grid_size+1))+gx+idxs[i]); + *out++ = ((gy*(grid_size+1))+gx+idxs[i]); } } else { for(S32 i=0;i<6;i++) { - *out++ = (vtop+(gy*(grid_size+1))+gx+idxs[i]); + *out++ = ((gy*(grid_size+1))+gx+idxs[i]); } } } -- cgit v1.2.3 From a2eb86b00927439afcf27219e38e58eba421294f Mon Sep 17 00:00:00 2001 From: "Matthew Breindel (Falcon)" Date: Mon, 24 May 2010 13:37:59 -0700 Subject: Ack. Fixed a bunch of stupid type mistakes in llvector4a. --- indra/llmath/llvolume.cpp | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 88969af4bd..d7d36d901d 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -4126,14 +4126,14 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, norm_mat.rotate(n[v1], t); - t.normalize3Fast(); + 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(); + t.normalize3fast(); normals.push_back(LLVector3(t[0], t[1], t[2])); segments.push_back(vertices.size()); @@ -5349,7 +5349,7 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) mCenter, cuv, mesh[0+offset].mPos, tc[0], mesh[1+offset].mPos, tc[1]); - binormal.normalize3Fast(); + binormal.normalize3fast(); LLVector4a normal; LLVector4a d0, d1; @@ -5369,7 +5369,7 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) normal.setCross3(d1, d0); } - normal.normalize3Fast(); + normal.normalize3fast(); VertexData vd; vd.mPosition = mCenter; @@ -5675,7 +5675,7 @@ void LLVolumeFace::createBinormals() //normalize binormals for (U32 i = 0; i < mNumVertices; i++) { - binorm[i].normalize3Fast(); + binorm[i].normalize3fast(); } } } @@ -5803,7 +5803,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat { mat.affineTransform(src_pos[i], dst_pos[i]); norm_mat.rotate(src_norm[i], dst_norm[i]); - dst_norm[i].normalize3Fast(); + dst_norm[i].normalize3fast(); dst_tc[i] = src_tc[i]; -- cgit v1.2.3 From c0b654dd4bee466a2ccbf050e532fb4a05acc549 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Mon, 24 May 2010 17:33:41 -0500 Subject: Fix for bad feeding of vectorized raycast. --- indra/llmath/llvolume.cpp | 12 ++++-------- 1 file changed, 4 insertions(+), 8 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 2ff1463b7c..f05e6eb9d9 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -4187,6 +4187,9 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, LLVector4a starta, dira; + starta.load3(start.mV); + dira.load3(dir.mV); + LLVector4a* p = (LLVector4a*) face.mPositions; for (U32 tri = 0; tri < face.mNumIndices/3; tri++) @@ -4235,17 +4238,10 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, if (bi_normal != NULL) { - if (!face.mBinormals) - { - face.createBinormals(); - } LLVector4* binormal = (LLVector4*) face.mBinormals; - if (binormal) - { - *bi_normal = ((1.f - a - b) * LLVector3(binormal[index1]) + + *bi_normal = ((1.f - a - b) * LLVector3(binormal[index1]) + a * LLVector3(binormal[index2]) + b * LLVector3(binormal[index3])); - } } } -- cgit v1.2.3 From e6fe3b1f1aa888e4594c89154ef895b3cf5498e9 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Tue, 25 May 2010 03:55:01 -0500 Subject: Better vectorization of various things. Turn off debug gl by default. --- indra/llmath/llvolume.cpp | 383 ++++++++++++++++++++++++++++++---------------- 1 file changed, 247 insertions(+), 136 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index f05e6eb9d9..d8fbc081fa 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -237,6 +237,21 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co 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); + + return LLTriangleRayIntersect(vert0a, vert1a, vert2a, origa, dira, + intersection_a, intersection_b, intersection_t, two_sided); +} + //------------------------------------------------------------------- // statics @@ -1889,15 +1904,15 @@ bool LLVolumeFace::VertexData::operator==(const LLVolumeFace::VertexData& rhs)co bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs, F32 angle_cutoff) const { bool retval = false; - if (rhs.mPosition == mPosition && rhs.mTexCoord == mTexCoord) + if (rhs.mData[POSITION].equal3(mData[POSITION]) && rhs.mTexCoord == mTexCoord) { if (angle_cutoff > 1.f) { - retval = (mNormal == rhs.mNormal); + retval = (mData[NORMAL].equal3(rhs.mData[NORMAL])); } else { - F32 cur_angle = rhs.mNormal*mNormal; + F32 cur_angle = rhs.mData[NORMAL].dot3(mData[NORMAL]); retval = cur_angle > angle_cutoff; } } @@ -2081,9 +2096,9 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) min = max = LLVector3(0,0,0); - F32* pos_out = face.mPositions; - F32* norm_out = face.mNormals; - F32* tc_out = face.mTexCoords; + F32* pos_out = (F32*) face.mPositions; + F32* norm_out = (F32*) face.mNormals; + F32* tc_out = (F32*) face.mTexCoords; for (U32 j = 0; j < num_verts; ++j) { @@ -2188,13 +2203,15 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) void tetrahedron_set_normal(LLVolumeFace::VertexData* cv) { - LLVector3 nrm = (cv[1].mPosition-cv[0].mPosition)%(cv[2].mPosition-cv[0].mPosition); - - nrm.normVec(); - - cv[0].mNormal = nrm; - cv[1].mNormal = nrm; - cv[2].mNormal = nrm; + 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() @@ -2209,12 +2226,12 @@ void LLVolume::makeTetrahedron() LLVolumeFace face; F32 x = 0.25f; - LLVector3 p[] = + LLVector4a p[] = { //unit tetrahedron corners - LLVector3(x,x,x), - LLVector3(-x,-x,x), - LLVector3(-x,x,-x), - LLVector3(x,-x,-x) + LLVector4a(x,x,x), + LLVector4a(-x,-x,x), + LLVector4a(-x,x,-x), + LLVector4a(x,-x,-x) }; face.mExtents[0].setVec(-x,-x,-x); @@ -2229,9 +2246,9 @@ void LLVolume::makeTetrahedron() //side 1 - cv[0].mPosition = p[1]; - cv[1].mPosition = p[0]; - cv[2].mPosition = p[2]; + cv[0].setPosition(p[1]); + cv[1].setPosition(p[0]); + cv[2].setPosition(p[2]); tetrahedron_set_normal(cv); @@ -2242,14 +2259,14 @@ void LLVolume::makeTetrahedron() LLVector4a* n = (LLVector4a*) face.mNormals; LLVector2* tc = (LLVector2*) face.mTexCoords; - v[0].load3(cv[0].mPosition.mV); - v[1].load3(cv[1].mPosition.mV); - v[2].load3(cv[2].mPosition.mV); + v[0] = cv[0].getPosition(); + v[1] = cv[1].getPosition(); + v[2] = cv[2].getPosition(); v += 3; - n[0].load3(cv[0].mNormal.mV); - n[1].load3(cv[1].mNormal.mV); - n[2].load3(cv[2].mNormal.mV); + n[0] = cv[0].getNormal(); + n[1] = cv[1].getNormal(); + n[2] = cv[2].getNormal(); n += 3; tc[0] = cv[0].mTexCoord; @@ -2259,20 +2276,20 @@ void LLVolume::makeTetrahedron() //side 2 - cv[0].mPosition = p[3]; - cv[1].mPosition = p[0]; - cv[2].mPosition = p[1]; + cv[0].setPosition(p[3]); + cv[1].setPosition(p[0]); + cv[2].setPosition(p[1]); tetrahedron_set_normal(cv); - v[0].load3(cv[0].mPosition.mV); - v[1].load3(cv[1].mPosition.mV); - v[2].load3(cv[2].mPosition.mV); + v[0] = cv[0].getPosition(); + v[1] = cv[1].getPosition(); + v[2] = cv[2].getPosition(); v += 3; - n[0].load3(cv[0].mNormal.mV); - n[1].load3(cv[1].mNormal.mV); - n[2].load3(cv[2].mNormal.mV); + n[0] = cv[0].getNormal(); + n[1] = cv[1].getNormal(); + n[2] = cv[2].getNormal(); n += 3; tc[0] = cv[0].mTexCoord; @@ -2281,20 +2298,20 @@ void LLVolume::makeTetrahedron() tc += 3; //side 3 - cv[0].mPosition = p[3]; - cv[1].mPosition = p[1]; - cv[2].mPosition = p[2]; + cv[0].setPosition(p[3]); + cv[1].setPosition(p[1]); + cv[2].setPosition(p[2]); tetrahedron_set_normal(cv); - v[0].load3(cv[0].mPosition.mV); - v[1].load3(cv[1].mPosition.mV); - v[2].load3(cv[2].mPosition.mV); + v[0] = cv[0].getPosition(); + v[1] = cv[1].getPosition(); + v[2] = cv[2].getPosition(); v += 3; - n[0].load3(cv[0].mNormal.mV); - n[1].load3(cv[1].mNormal.mV); - n[2].load3(cv[2].mNormal.mV); + n[0] = cv[0].getNormal(); + n[1] = cv[1].getNormal(); + n[2] = cv[2].getNormal(); n += 3; tc[0] = cv[0].mTexCoord; @@ -2303,20 +2320,20 @@ void LLVolume::makeTetrahedron() tc += 3; //side 4 - cv[0].mPosition = p[2]; - cv[1].mPosition = p[0]; - cv[2].mPosition = p[3]; + cv[0].setPosition(p[2]); + cv[1].setPosition(p[0]); + cv[2].setPosition(p[3]); tetrahedron_set_normal(cv); - v[0].load3(cv[0].mPosition.mV); - v[1].load3(cv[1].mPosition.mV); - v[2].load3(cv[2].mPosition.mV); + v[0] = cv[0].getPosition(); + v[1] = cv[1].getPosition(); + v[2] = cv[2].getPosition(); v += 3; - n[0].load3(cv[0].mNormal.mV); - n[1].load3(cv[1].mNormal.mV); - n[2].load3(cv[2].mNormal.mV); + n[0] = cv[0].getNormal(); + n[1] = cv[1].getNormal(); + n[2] = cv[2].getNormal(); n += 3; tc[0] = cv[0].mTexCoord; @@ -3974,9 +3991,9 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, S32 v3 = face.mIndices[j*3+2]; //get current face center - LLVector3 cCenter = (face.mVertices[v1].mPosition + - face.mVertices[v2].mPosition + - face.mVertices[v3].mPosition) / 3.0f; + LLVector3 cCenter = (face.mVertices[v1].getPosition() + + face.mVertices[v2].getPosition() + + face.mVertices[v3].getPosition()) / 3.0f; //for each edge for (S32 k = 0; k < 3; k++) { @@ -3994,9 +4011,9 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, v3 = face.mIndices[nIndex*3+2]; //get neighbor face center - LLVector3 nCenter = (face.mVertices[v1].mPosition + - face.mVertices[v2].mPosition + - face.mVertices[v3].mPosition) / 3.0f; + LLVector3 nCenter = (face.mVertices[v1].getPosition() + + face.mVertices[v2].getPosition() + + face.mVertices[v3].getPosition()) / 3.0f; //draw line vertices.push_back(cCenter); @@ -4020,14 +4037,14 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, //for each vertex for (U32 j = 0; j < face.mNumVertices; j++) { - vertices.push_back(face.mVertices[j].mPosition); - vertices.push_back(face.mVertices[j].mPosition + face.mVertices[j].mNormal*0.1f); + vertices.push_back(face.mVertices[j].getPosition()); + vertices.push_back(face.mVertices[j].getPosition() + face.mVertices[j].getNormal()*0.1f); normals.push_back(LLVector3(0,0,1)); normals.push_back(LLVector3(0,0,1)); segments.push_back(vertices.size()); #if DEBUG_SILHOUETTE_BINORMALS - vertices.push_back(face.mVertices[j].mPosition); - vertices.push_back(face.mVertices[j].mPosition + face.mVertices[j].mBinormal*0.1f); + vertices.push_back(face.mVertices[j].getPosition()); + vertices.push_back(face.mVertices[j].getPosition() + face.mVertices[j].mBinormal*0.1f); normals.push_back(LLVector3(0,0,1)); normals.push_back(LLVector3(0,0,1)); segments.push_back(vertices.size()); @@ -5038,9 +5055,15 @@ BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) } } +void LLVolumeFace::getVertexData(U16 index, LLVolumeFace::VertexData& cv) +{ + cv.setPosition(mPositions[index]); + cv.setNormal(mNormals[index]); + cv.mTexCoord = mTexCoords[index]; +} + void LLVolumeFace::optimize(F32 angle_cutoff) { -#if 0 //disabling until a vectorized version is available LLVolumeFace new_face; VertexMapData::PointMap point_map; @@ -5050,10 +5073,11 @@ void LLVolumeFace::optimize(F32 angle_cutoff) { U16 index = mIndices[i]; - LLVolumeFace::VertexData cv = mVertices[index]; - + LLVolumeFace::VertexData cv; + getVertexData(index, cv); + BOOL found = FALSE; - VertexMapData::PointMap::iterator point_iter = point_map.find(cv.mPosition); + VertexMapData::PointMap::iterator point_iter = point_map.find(cv.getPosition()); if (point_iter != point_map.end()) { //duplicate point might exist for (U32 j = 0; j < point_iter->second.size(); ++j) @@ -5062,7 +5086,7 @@ void LLVolumeFace::optimize(F32 angle_cutoff) if (tv.compareNormal(cv, angle_cutoff)) { found = TRUE; - new_face.mIndices.push_back((point_iter->second)[j].mIndex); + new_face.pushIndex((point_iter->second)[j].mIndex); break; } } @@ -5070,14 +5094,14 @@ void LLVolumeFace::optimize(F32 angle_cutoff) if (!found) { - new_face.mVertices.push_back(cv); + new_face.pushVertex(cv); U16 index = (U16) new_face.mNumVertices-1; - new_face.mIndices.push_back(index); + new_face.pushIndex(index); VertexMapData d; - d.mPosition = cv.mPosition; + d.setPosition(cv.getPosition()); d.mTexCoord = cv.mTexCoord; - d.mNormal = cv.mNormal; + d.setNormal(cv.getNormal()); d.mIndex = index; if (point_iter != point_map.end()) { @@ -5085,14 +5109,23 @@ void LLVolumeFace::optimize(F32 angle_cutoff) } else { - point_map[d.mPosition].push_back(d); + point_map[d.getPosition()].push_back(d); } } } - mVertices = new_face.mVertices; - mIndices = new_face.mIndices; -#endif + swapData(new_face); +} + +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, @@ -5102,10 +5135,21 @@ void LerpPlanarVertex(LLVolumeFace::VertexData& v0, F32 coef01, F32 coef02) { - vout.mPosition = v0.mPosition + ((v1.mPosition-v0.mPosition)*coef01)+((v2.mPosition-v0.mPosition)*coef02); + + LLVector4a lhs; + lhs.setSub(v1.getPosition(), v0.getPosition()); + lhs.mul(coef01); + LLVector4a rhs; + rhs.setSub(v2.getPosition(), v0.getPosition()); + rhs.mul(coef02); + + rhs.add(lhs); + rhs.add(v0.getPosition()); + + vout.setPosition(rhs); + vout.mTexCoord = v0.mTexCoord + ((v1.mTexCoord-v0.mTexCoord)*coef01)+((v2.mTexCoord-v0.mTexCoord)*coef02); - vout.mNormal = v0.mNormal; - vout.mBinormal = v0.mBinormal; + vout.setNormal(v0.getNormal()); } BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) @@ -5137,16 +5181,22 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) VertexData corners[4]; VertexData baseVert; for(int t = 0; t < 4; t++){ - corners[t].mPosition = mesh[offset + (grid_size*t)].mPos; + corners[t].getPosition().load3( mesh[offset + (grid_size*t)].mPos.mV); corners[t].mTexCoord.mV[0] = profile[grid_size*t].mV[0]+0.5f; corners[t].mTexCoord.mV[1] = 0.5f - profile[grid_size*t].mV[1]; } - baseVert.mNormal = - ((corners[1].mPosition-corners[0].mPosition) % - (corners[2].mPosition-corners[1].mPosition)); - baseVert.mNormal.normVec(); + + { + LLVector4a lhs; + lhs.setSub(corners[1].getPosition(), corners[0].getPosition()); + LLVector4a rhs; + rhs.setSub(corners[2].getPosition(), corners[1].getPosition()); + baseVert.getNormal().setCross3(lhs, rhs); + baseVert.getNormal().normalize3fast(); + } + if(!(mTypeMask & TOP_MASK)){ - baseVert.mNormal *= -1.0f; + baseVert.getNormal().mul(-1.0f); }else{ //Swap the UVs on the U(X) axis for top face LLVector2 swap; @@ -5161,9 +5211,9 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) LLVector4a binormal; calc_binormal_from_triangle( binormal, - corners[0].mPosition, corners[0].mTexCoord, - corners[1].mPosition, corners[1].mTexCoord, - corners[2].mPosition, corners[2].mTexCoord); + corners[0].getPosition(), corners[0].mTexCoord, + corners[1].getPosition(), corners[1].mTexCoord, + corners[2].getPosition(), corners[2].mTexCoord); S32 size = (grid_size+1)*(grid_size+1); resizeVertices(size); @@ -5185,18 +5235,18 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) (F32)gx/(F32)grid_size, (F32)gy/(F32)grid_size); - (*pos++).load3(newVert.mPosition.mV); - (*norm++).load3(baseVert.mNormal.mV); - (*tc++) = newVert.mTexCoord; - (*binorm++).load4a((F32*) &binormal.mQ); + *pos++ = newVert.getPosition(); + *norm++ = baseVert.getNormal(); + *tc++ = newVert.mTexCoord; + *binorm++ = binormal; if (gx == 0 && gy == 0) { - min = max = newVert.mPosition; + min = max = LLVector3(newVert.getPosition().getF32()); } else { - update_min_max(min,max,newVert.mPosition); + update_min_max(min,max,newVert.getPosition().getF32()); } } } @@ -5343,18 +5393,19 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) mCenter = (min+max)*0.5f; cuv = (min_uv + max_uv)*0.5f; + LLVector4a center; + center.load3(mCenter.mV); + LLVector4a binormal; calc_binormal_from_triangle(binormal, - mCenter, cuv, - mesh[0+offset].mPos, tc[0], - mesh[1+offset].mPos, tc[1]); + center, cuv, + pos[0], tc[0], + pos[1], tc[1]); binormal.normalize3fast(); LLVector4a normal; LLVector4a d0, d1; - LLVector4a center; - - center.load3(mCenter.mV); + d0.setSub(center, pos[0]); d1.setSub(center, pos[1]); @@ -5371,7 +5422,7 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) normal.normalize3fast(); VertexData vd; - vd.mPosition = mCenter; + vd.getPosition().load3(mCenter.mV); vd.mTexCoord = cuv; if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK)) @@ -5634,10 +5685,7 @@ void LLVolumeFace::createBinormals() allocateBinormals(mNumVertices); //generate binormals - LLStrider pos; - pos = (LLVector3*) mPositions; - pos.setStride(16); - + LLVector4a* pos = mPositions; LLVector2* tc = (LLVector2*) mTexCoords; LLVector4a* binorm = (LLVector4a*) mBinormals; @@ -5690,12 +5738,12 @@ void LLVolumeFace::resizeVertices(S32 num_verts) if (num_verts) { - mPositions = (F32*) _mm_malloc(num_verts*16, 16); - mNormals = (F32*) _mm_malloc(num_verts*16, 16); + mPositions = (LLVector4a*) _mm_malloc(num_verts*16, 16); + mNormals = (LLVector4a*) _mm_malloc(num_verts*16, 16); //pad texture coordinate block end to allow for QWORD reads S32 size = ((num_verts*8) + 0xF) & ~0xF; - mTexCoords = (F32*) _mm_malloc(size, 16); + mTexCoords = (LLVector2*) _mm_malloc(size, 16); } else { @@ -5707,10 +5755,61 @@ void LLVolumeFace::resizeVertices(S32 num_verts) 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; + + //positions + LLVector4a* dst = (LLVector4a*) _mm_malloc(new_size, 16); + if (mPositions) + { + LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mPositions, new_size/4); + _mm_free(mPositions); + } + mPositions = dst; + + //normals + dst = (LLVector4a*) _mm_malloc(new_size, 16); + if (mNormals) + { + LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mNormals, new_size/4); + _mm_free(mNormals); + } + mNormals = dst; + + //tex coords + new_size = ((new_verts*8)+0xF) & ~0xF; + + { + LLVector2* dst = (LLVector2*) _mm_malloc(new_size, 16); + if (mTexCoords) + { + LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mTexCoords, new_size/4); + _mm_free(mTexCoords); + } + } + + //just clear binormals + _mm_free(mBinormals); + mBinormals = NULL; + + mPositions[mNumVertices] = pos; + mNormals[mNumVertices] = norm; + mTexCoords[mNumVertices] = tc; + + mNumVertices++; +} + void LLVolumeFace::allocateBinormals(S32 num_verts) { _mm_free(mBinormals); - mBinormals = (F32*) _mm_malloc(num_verts*16, 16); + mBinormals = (LLVector4a*) _mm_malloc(num_verts*16, 16); } @@ -5733,6 +5832,23 @@ void LLVolumeFace::resizeIndices(S32 num_indices) 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+0xF) & ~0xF; + if (new_size != old_size) + { + U16* dst = (U16*) _mm_malloc(new_size, 16); + LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mIndices, new_size/4); + _mm_free(mIndices); + mIndices = dst; + } + + mIndices[mNumIndices++] = idx; +} + void LLVolumeFace::fillFromLegacyData(std::vector& v, std::vector& idx) { resizeVertices(v.size()); @@ -5740,14 +5856,9 @@ void LLVolumeFace::fillFromLegacyData(std::vector& v, for (U32 i = 0; i < v.size(); ++i) { - for (U32 j = 0; j < 3; ++j) - { - mPositions[i*4+j] = v[i].mPosition[j]; - mNormals[i*4+j] = v[i].mNormal[j]; - } - - mTexCoords[i*2+0] = v[i].mTexCoord.mV[0]; - mTexCoords[i*2+1] = v[i].mTexCoord.mV[1]; + mPositions[i] = v[i].getPosition(); + mNormals[i] = v[i].getNormal(); + mTexCoords[i] = v[i].mTexCoord; } for (U32 i = 0; i < idx.size(); ++i) @@ -5768,13 +5879,13 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat } - F32* new_pos = (F32*) _mm_malloc(new_count*16, 16); - F32* new_norm = (F32*) _mm_malloc(new_count*16, 16); - F32* new_tc = (F32*) _mm_malloc((new_count*8+0xF) & ~0xF, 16); + LLVector4a* new_pos = (LLVector4a*) _mm_malloc(new_count*16, 16); + LLVector4a* new_norm = (LLVector4a*) _mm_malloc(new_count*16, 16); + LLVector2* new_tc = (LLVector2*) _mm_malloc((new_count*8+0xF) & ~0xF, 16); - LLVector4a::memcpyNonAliased16(new_pos, mPositions, new_count*4); - LLVector4a::memcpyNonAliased16(new_norm, mNormals, new_count*4); - LLVector4a::memcpyNonAliased16(new_tc, mTexCoords, new_count*2); + LLVector4a::memcpyNonAliased16((F32*) new_pos, (F32*) mPositions, new_count*4); + LLVector4a::memcpyNonAliased16((F32*) new_norm, (F32*) mNormals, new_count*4); + LLVector4a::memcpyNonAliased16((F32*) new_tc, (F32*) mTexCoords, new_count*2); _mm_free(mPositions); _mm_free(mNormals); @@ -6205,24 +6316,24 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) // Fills in dummy values if the triangle has degenerate texture coordinates. void calc_binormal_from_triangle(LLVector4a& binormal, - const LLVector3& pos0, + const LLVector4a& pos0, const LLVector2& tex0, - const LLVector3& pos1, + const LLVector4a& pos1, const LLVector2& tex1, - const LLVector3& pos2, + const LLVector4a& pos2, const LLVector2& tex2) { - LLVector4a rx0; rx0.set( pos0.mV[VX], tex0.mV[VX], tex0.mV[VY] ); - LLVector4a rx1; rx1.set( pos1.mV[VX], tex1.mV[VX], tex1.mV[VY] ); - LLVector4a rx2; rx2.set( pos2.mV[VX], tex2.mV[VX], tex2.mV[VY] ); + LLVector4a rx0( pos0[VX], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a rx1( pos1[VX], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a rx2( pos2[VX], tex2.mV[VX], tex2.mV[VY] ); - LLVector4a ry0; ry0.set( pos0.mV[VY], tex0.mV[VX], tex0.mV[VY] ); - LLVector4a ry1; ry1.set( pos1.mV[VY], tex1.mV[VX], tex1.mV[VY] ); - LLVector4a ry2; ry2.set( pos2.mV[VY], tex2.mV[VX], tex2.mV[VY] ); + LLVector4a ry0( pos0[VY], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a ry1( pos1[VY], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a ry2( pos2[VY], tex2.mV[VX], tex2.mV[VY] ); - LLVector4a rz0; rz0.set( pos0.mV[VZ], tex0.mV[VX], tex0.mV[VY] ); - LLVector4a rz1; rz1.set( pos1.mV[VZ], tex1.mV[VX], tex1.mV[VY] ); - LLVector4a rz2; rz2.set( pos2.mV[VZ], tex2.mV[VX], tex2.mV[VY] ); + LLVector4a rz0( pos0[VZ], tex0.mV[VX], tex0.mV[VY] ); + LLVector4a rz1( pos1[VZ], tex1.mV[VX], tex1.mV[VY] ); + LLVector4a rz2( pos2[VZ], tex2.mV[VX], tex2.mV[VY] ); LLVector4a lhs, rhs; -- cgit v1.2.3 From c98b1b3fd9341834978aff0e841714e206d28c0a Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Wed, 26 May 2010 03:29:19 -0500 Subject: Fully aligned llvolume --- indra/llmath/llvolume.cpp | 353 +++++++++++++++++++++++++++++++++++----------- 1 file changed, 274 insertions(+), 79 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index d8fbc081fa..9b6e2488e6 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -107,22 +107,27 @@ BOOL check_same_clock_dir( const LLVector3& pt1, const LLVector3& pt2, const LLV BOOL LLLineSegmentBoxIntersect(const LLVector3& start, const LLVector3& end, const LLVector3& center, const LLVector3& size) { - float fAWdU[3]; - LLVector3 dir; - LLVector3 diff; + return LLLineSegmentBoxIntersect(start.mV, end.mV, center.mV, size.mV); +} + +BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* center, const F32* size) +{ + F32 fAWdU[3]; + F32 dir[3]; + F32 diff[3]; for (U32 i = 0; i < 3; i++) { - dir.mV[i] = 0.5f * (end.mV[i] - start.mV[i]); - diff.mV[i] = (0.5f * (end.mV[i] + start.mV[i])) - center.mV[i]; - fAWdU[i] = fabsf(dir.mV[i]); - if(fabsf(diff.mV[i])>size.mV[i] + fAWdU[i]) return false; + dir[i] = 0.5f * (end[i] - start[i]); + diff[i] = (0.5f * (end[i] + start[i])) - center[i]; + fAWdU[i] = fabsf(dir[i]); + if(fabsf(diff[i])>size[i] + fAWdU[i]) return false; } float f; - f = dir.mV[1] * diff.mV[2] - dir.mV[2] * diff.mV[1]; if(fabsf(f)>size.mV[1]*fAWdU[2] + size.mV[2]*fAWdU[1]) return false; - f = dir.mV[2] * diff.mV[0] - dir.mV[0] * diff.mV[2]; if(fabsf(f)>size.mV[0]*fAWdU[2] + size.mV[2]*fAWdU[0]) return false; - f = dir.mV[0] * diff.mV[1] - dir.mV[1] * diff.mV[0]; if(fabsf(f)>size.mV[0]*fAWdU[1] + size.mV[1]*fAWdU[0]) return false; + f = dir[1] * diff[2] - dir[2] * diff[1]; if(fabsf(f)>size[1]*fAWdU[2] + size[2]*fAWdU[1]) return false; + f = dir[2] * diff[0] - dir[0] * diff[2]; if(fabsf(f)>size[0]*fAWdU[2] + size[2]*fAWdU[0]) return false; + f = dir[0] * diff[1] - dir[1] * diff[0]; if(fabsf(f)>size[0]*fAWdU[1] + size[1]*fAWdU[0]) return false; return true; } @@ -1869,6 +1874,59 @@ BOOL LLVolume::generate() return FALSE; } +void LLVolumeFace::VertexData::init() +{ + mData = (LLVector4a*) _mm_malloc(32, 16); +} + +LLVolumeFace::VertexData::VertexData() +{ + init(); +} + +LLVolumeFace::VertexData::VertexData(const VertexData& rhs) +{ + init(); + LLVector4a::memcpyNonAliased16((F32*) mData, (F32*) rhs.mData, 8); + mTexCoord = rhs.mTexCoord; +} + +LLVolumeFace::VertexData::~VertexData() +{ + _mm_free(mData); +} + +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 U8* l = (const U8*) this; @@ -2037,7 +2095,7 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) if (mdl[i].has("Weights")) { - face.mWeights.resize(num_verts); + face.allocateWeights(num_verts); LLSD::Binary weights = mdl[i]["Weights"]; @@ -2050,13 +2108,15 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) U8 joint = weights[idx++]; U32 cur_influence = 0; + LLVector4 wght(0,0,0,0); + while (joint != END_INFLUENCES) { U16 influence = weights[idx++]; influence |= ((U16) weights[idx++] << 8); F32 w = llmin((F32) influence / 65535.f, 0.99999f); - face.mWeights[cur_vertex].mV[cur_influence++] = (F32) joint + w; + wght.mV[cur_influence++] = (F32) joint + w; if (cur_influence >= 4) { @@ -2068,6 +2128,8 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) } } + face.mWeights[cur_vertex].loadua(wght.mV); + cur_vertex++; } @@ -2078,62 +2140,70 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) } - LLVector3 min_pos; - LLVector3 max_pos; + LLVector3 minp; + LLVector3 maxp; LLVector2 min_tc; LLVector2 max_tc; - min_pos.setValue(mdl[i]["PositionDomain"]["Min"]); - max_pos.setValue(mdl[i]["PositionDomain"]["Max"]); + 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"]); - LLVector3 pos_range = max_pos - min_pos; + LLVector4a pos_range; + pos_range.setSub(max_pos, min_pos); LLVector2 tc_range = max_tc - min_tc; - LLVector3& min = face.mExtents[0]; - LLVector3& max = face.mExtents[1]; - - min = max = LLVector3(0,0,0); + LLVector4a& min = face.mExtents[0]; + LLVector4a& max = face.mExtents[1]; - F32* pos_out = (F32*) face.mPositions; - F32* norm_out = (F32*) face.mNormals; - F32* tc_out = (F32*) face.mTexCoords; + min.clear(); + max.clear(); + + 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[0] = (F32) v[0] / 65535.f * pos_range.mV[0] + min_pos.mV[0]; - pos_out[1] = (F32) v[1] / 65535.f * pos_range.mV[1] + min_pos.mV[1]; - pos_out[2] = (F32) v[2] / 65535.f * pos_range.mV[2] + min_pos.mV[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); if (j == 0) { - min = max = LLVector3(pos_out); + min = *pos_out; + max = min; } else { - update_min_max(min,max,pos_out); + min.setMin(*pos_out); + max.setMax(*pos_out); } - pos_out += 4; + pos_out++; U16* n = (U16*) &(norm[j*3*2]); - - norm_out[0] = (F32) n[0] / 65535.f * 2.f - 1.f; - norm_out[1] = (F32) n[1] / 65535.f * 2.f - 1.f; - norm_out[2] = (F32) n[2] / 65535.f * 2.f - 1.f; - norm_out += 4; + 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[0] = (F32) t[0] / 65535.f * tc_range.mV[0] + min_tc.mV[0]; - tc_out[1] = (F32) t[1] / 65535.f * tc_range.mV[1] + min_tc.mV[1]; + 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 += 8; + tc_out++; } @@ -2234,8 +2304,8 @@ void LLVolume::makeTetrahedron() LLVector4a(x,-x,-x) }; - face.mExtents[0].setVec(-x,-x,-x); - face.mExtents[1].setVec(x,x,x); + face.mExtents[0].splat(-x); + face.mExtents[1].splat(x); LLVolumeFace::VertexData cv[3]; @@ -4165,6 +4235,18 @@ void LLVolume::generateSilhouetteVertices(std::vector &vertices, 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; @@ -4182,7 +4264,8 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, end_face = face; } - LLVector3 dir = end - start; + LLVector4a dir; + dir.setSub(end, start); F32 closest_t = 2.f; // must be larger than 1 @@ -4192,21 +4275,20 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, { LLVolumeFace &face = mVolumeFaces[i]; - LLVector3 box_center = (face.mExtents[0] + face.mExtents[1]) / 2.f; - LLVector3 box_size = face.mExtents[1] - face.mExtents[0]; + LLVector4a box_center; + box_center.setAdd(face.mExtents[0], face.mExtents[1]); + box_center.mul(0.5f); + + LLVector4a box_size; + box_size.setSub(face.mExtents[1], face.mExtents[0]); - if (LLLineSegmentBoxIntersect(start, end, box_center, box_size)) + if (LLLineSegmentBoxIntersect(start.getF32(), end.getF32(), box_center.getF32(), box_size.getF32())) { if (bi_normal != NULL) // if the caller wants binormals, we may need to generate them { genBinormals(i); } - LLVector4a starta, dira; - - starta.load3(start.mV); - dira.load3(dir.mV); - LLVector4a* p = (LLVector4a*) face.mPositions; for (U32 tri = 0; tri < face.mNumIndices/3; tri++) @@ -4220,7 +4302,7 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, if (LLTriangleRayIntersect(p[index1], p[index2], p[index3], - starta, dira, &a, &b, &t, FALSE)) + start, dir, &a, &b, &t, FALSE)) { if ((t >= 0.f) && // if hit is after start (t <= 1.f) && // and before end @@ -4231,7 +4313,10 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, if (intersection != NULL) { - *intersection = start + dir * closest_t; + LLVector4a intersect = dir; + intersect.mul(closest_t); + intersect.add(start); + intersection->set(intersect.getF32()); } @@ -5029,6 +5114,107 @@ std::ostream& operator<<(std::ostream &s, const LLVolume *volumep) return s; } +LLVolumeFace::LLVolumeFace() : + mID(0), + mTypeMask(0), + mBeginS(0), + mBeginT(0), + mNumS(0), + mNumT(0), + mNumVertices(0), + mNumIndices(0), + mPositions(NULL), + mNormals(NULL), + mBinormals(NULL), + mTexCoords(NULL), + mIndices(NULL), + mWeights(NULL) +{ + mExtents = (LLVector4a*) _mm_malloc(48, 16); + mCenter = mExtents+2; +} + +LLVolumeFace::LLVolumeFace(const LLVolumeFace& src) +{ + mExtents = (LLVector4a*) _mm_malloc(48, 16); + 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; + + mNumVertices = 0; + mNumIndices = 0; + mPositions = NULL; + mNormals = NULL; + mBinormals = NULL; + mTexCoords = NULL; + mWeights = NULL; + mIndices = NULL; + + LLVector4a::memcpyNonAliased16((F32*) mExtents, (F32*) src.mExtents, 12); + + resizeVertices(src.mNumVertices); + resizeIndices(src.mNumIndices); + + if (mNumVertices) + { + S32 vert_size = mNumVertices*4; + S32 tc_size = (mNumVertices*8+0xF) & ~0xF; + tc_size /= 4; + + 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, vert_size); + + if (src.mBinormals) + { + allocateBinormals(src.mNumVertices); + LLVector4a::memcpyNonAliased16((F32*) mBinormals, (F32*) src.mBinormals, vert_size); + } + else + { + _mm_free(mBinormals); + mBinormals = NULL; + } + + if (src.mWeights) + { + allocateWeights(src.mNumVertices); + LLVector4a::memcpyNonAliased16((F32*) mWeights, (F32*) src.mWeights, vert_size); + } + else + { + _mm_free(mWeights); + mWeights = NULL; + } + } + + if (mNumIndices) + { + S32 idx_size = (mNumIndices*2+0xF) & ~0xF; + idx_size /= 4; + + LLVector4a::memcpyNonAliased16((F32*) mIndices, (F32*) src.mIndices, idx_size); + } + + + //delete + return *this; +} + LLVolumeFace::~LLVolumeFace() { _mm_free(mPositions); @@ -5036,6 +5222,8 @@ LLVolumeFace::~LLVolumeFace() _mm_free(mTexCoords); _mm_free(mIndices); _mm_free(mBinormals); + _mm_free(mWeights); + _mm_free(mExtents); } BOOL LLVolumeFace::create(LLVolume* volume, BOOL partial_build) @@ -5169,8 +5357,8 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) num_vertices = (grid_size+1)*(grid_size+1); num_indices = quad_count * 4; - LLVector3& min = mExtents[0]; - LLVector3& max = mExtents[1]; + LLVector4a& min = mExtents[0]; + LLVector4a& max = mExtents[1]; S32 offset = 0; if (mTypeMask & TOP_MASK) @@ -5242,16 +5430,18 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) if (gx == 0 && gy == 0) { - min = max = LLVector3(newVert.getPosition().getF32()); + min = max = newVert.getPosition(); } else { - update_min_max(min,max,newVert.getPosition().getF32()); + min.setMin(newVert.getPosition()); + max.setMax(newVert.getPosition()); } } } - mCenter = (min + max) * 0.5f; + mCenter->setAdd(min, max); + mCenter->mul(0.5f); if (!partial_build) { @@ -5335,7 +5525,7 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) S32 max_s = volume->getProfile().getTotal(); S32 max_t = volume->getPath().mPath.size(); - mCenter.clearVec(); + mCenter->clear(); S32 offset = 0; if (mTypeMask & TOP_MASK) @@ -5353,8 +5543,8 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) LLVector2 cuv; LLVector2 min_uv, max_uv; - LLVector3& min = mExtents[0]; - LLVector3& max = mExtents[1]; + LLVector4a& min = mExtents[0]; + LLVector4a& max = mExtents[1]; LLVector2* tc = (LLVector2*) mTexCoords; LLVector4a* pos = (LLVector4a*) mPositions; @@ -5380,25 +5570,24 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) if (i == 0) { - min = max = mesh[i+offset].mPos; + min = max = pos[i]; min_uv = max_uv = tc[i]; } else { - update_min_max(min,max, mesh[i+offset].mPos); + update_min_max(min,max,pos[i]); update_min_max(min_uv, max_uv, tc[i]); } } - mCenter = (min+max)*0.5f; - cuv = (min_uv + max_uv)*0.5f; + mCenter->setAdd(min, max); + mCenter->mul(0.5f); - LLVector4a center; - center.load3(mCenter.mV); + cuv = (min_uv + max_uv)*0.5f; LLVector4a binormal; calc_binormal_from_triangle(binormal, - center, cuv, + *mCenter, cuv, pos[0], tc[0], pos[1], tc[1]); binormal.normalize3fast(); @@ -5407,8 +5596,8 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) LLVector4a d0, d1; - d0.setSub(center, pos[0]); - d1.setSub(center, pos[1]); + d0.setSub(*mCenter, pos[0]); + d1.setSub(*mCenter, pos[1]); if (mTypeMask & TOP_MASK) { @@ -5422,12 +5611,12 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) normal.normalize3fast(); VertexData vd; - vd.getPosition().load3(mCenter.mV); + vd.setPosition(*mCenter); vd.mTexCoord = cuv; if (!(mTypeMask & HOLLOW_MASK) && !(mTypeMask & OPEN_MASK)) { - pos[num_vertices].load4a((F32*) ¢er.mQ); + pos[num_vertices] = *mCenter; tc[num_vertices] = cuv; num_vertices++; } @@ -5812,6 +6001,11 @@ void LLVolumeFace::allocateBinormals(S32 num_verts) mBinormals = (LLVector4a*) _mm_malloc(num_verts*16, 16); } +void LLVolumeFace::allocateWeights(S32 num_verts) +{ + _mm_free(mWeights); + mWeights = (LLVector4a*) _mm_malloc(num_verts*16, 16); +} void LLVolumeFace::resizeIndices(S32 num_indices) { @@ -5919,11 +6113,11 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat if (offset == 0 && i == 0) { - mExtents[0] = mExtents[1] = LLVector3((F32*) &(dst_pos[i].mQ)); + mExtents[0] = mExtents[1] = dst_pos[i]; } else { - update_min_max(mExtents[0], mExtents[1], (F32*) &(dst_pos[i].mQ)); + update_min_max(mExtents[0], mExtents[1], dst_pos[i]); } } @@ -6076,18 +6270,19 @@ BOOL LLVolumeFace::createSide(LLVolume* volume, BOOL partial_build) //get bounding box for this side - LLVector3& face_min = mExtents[0]; - LLVector3& face_max = mExtents[1]; - mCenter.clearVec(); + LLVector4a& face_min = mExtents[0]; + LLVector4a& face_max = mExtents[1]; + mCenter->clear(); - face_min = face_max = LLVector3((F32*) &(pos[0].mQ)); + face_min = face_max = pos[0]; for (U32 i = 1; i < mNumVertices; ++i) { - update_min_max(face_min, face_max, (F32*) &(pos[i].mQ)); + update_min_max(face_min, face_max, pos[i]); } - mCenter = (face_min + face_max) * 0.5f; + mCenter->setAdd(face_min, face_max); + mCenter->mul(0.5f); S32 cur_index = 0; S32 cur_edge = 0; -- cgit v1.2.3 From 0e7f4dc5cef8a97cb1dd08aa2f79538ced267888 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Sat, 29 May 2010 05:37:38 -0500 Subject: Octree per LLVolumeFace WIP --- indra/llmath/llvolume.cpp | 625 ++++++++++++++++++++++++++++++++++++---------- 1 file changed, 491 insertions(+), 134 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 9b6e2488e6..d261811aa2 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -45,6 +45,7 @@ #include "m4math.h" #include "m3math.h" #include "llmatrix4a.h" +#include "lloctree.h" #include "lldarray.h" #include "llvolume.h" #include "llstl.h" @@ -132,6 +133,51 @@ BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* cent return true; } +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); + + S32 grt = lhs.greaterThan4(rhs).getComparisonMask(); + + if (grt & 0x7) + { + return false; + } + + LLVector4a f; + f.setCross3(dir, diff); + f.setAbs(f); + + LLVector4a v0; v0.mQ = _mm_shuffle_ps(size.mQ, size.mQ, _MM_SHUFFLE(3,1,0,0)); + LLVector4a v1; v1.mQ = _mm_shuffle_ps(fAWdU.mQ, fAWdU.mQ, _MM_SHUFFLE(3,2,2,1)); + lhs.setMul(v0, v1); + + v0.mQ = _mm_shuffle_ps(size.mQ, size.mQ, _MM_SHUFFLE(3,2,2,1)); + v1.mQ = _mm_shuffle_ps(fAWdU.mQ, fAWdU.mQ, _MM_SHUFFLE(3,1,0,0)); + rhs.setMul(v0, v1); + rhs.add(lhs); + + grt = f.greaterThan4(rhs).getComparisonMask(); + + return (grt & 0x7) ? false : 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, @@ -139,15 +185,13 @@ BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* cent // 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, BOOL two_sided) + 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); @@ -156,87 +200,116 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co pvec.setCross3(dir, edge2); /* if determinant is near zero, ray lies in plane of triangle */ - F32 det = edge1.dot3(pvec); - - if (!two_sided) + LLVector4a det; + det.setAllDot3(edge1, pvec); + + if (det.greaterEqual4(LLVector4a::getApproximatelyZero()).getComparisonMask()) { - if (det < F_APPROXIMATELY_ZERO) - { - return FALSE; - } - /* calculate distance from vert0 to ray origin */ LLVector4a tvec; tvec.setSub(orig, vert0); /* calculate U parameter and test bounds */ - u = tvec.dot3(pvec); + LLVector4a u; + u.setAllDot3(tvec,pvec); - if (u < 0.f || u > det) + if (u.greaterEqual4(LLVector4a::getZero()).getComparisonMask() && + u.lessEqual4(det).getComparisonMask()) { - return FALSE; + /* prepare to test V parameter */ + LLVector4a qvec; + qvec.setCross3(tvec, edge1); + + /* calculate V parameter and test bounds */ + LLVector4a v; + v.setAllDot3(dir, qvec); + + + //if (!(v < 0.f || u + v > det)) + + LLVector4a sum_uv; + sum_uv.setAdd(u, v); + + S32 v_gequal = v.greaterEqual4(LLVector4a::getZero()).getComparisonMask(); + S32 sum_lequal = sum_uv.lessEqual4(det).getComparisonMask(); + + if (v_gequal && sum_lequal) + { + /* calculate t, scale parameters, ray intersects triangle */ + LLVector4a t; + t.setAllDot3(edge2,qvec); + + t.div(det); + u.div(det); + v.div(det); + + intersection_a = u[0]; + intersection_b = v[0]; + intersection_t = t[0]; + return TRUE; + } } - - /* prepare to test V parameter */ - LLVector4a qvec; - qvec.setCross3(tvec, edge1); + } - /* calculate V parameter and test bounds */ - v = dir.dot3(qvec); - if (v < 0.f || u + v > det) - { - return FALSE; - } + return FALSE; +} - /* calculate t, scale parameters, ray intersects triangle */ - t = edge2.dot3(qvec); - F32 inv_det = 1.0 / det; - t *= inv_det; - u *= inv_det; - v *= inv_det; - } +BOOL LLTriangleRayIntersectTwoSided(const LLVector4a& vert0, const LLVector4a& vert1, const LLVector4a& vert2, const LLVector4a& orig, const LLVector4a& dir, + F32& intersection_a, F32& intersection_b, F32& intersection_t) +{ + F32 u, v, t; - else // two sided - { - if (det > -F_APPROXIMATELY_ZERO && det < F_APPROXIMATELY_ZERO) - { - return FALSE; - } - F32 inv_det = 1.0 / det; + /* find vectors for two edges sharing vert0 */ + LLVector4a edge1; + edge1.setSub(vert1, vert0); + + + LLVector4a edge2; + edge2.setSub(vert2, vert0); - /* calculate distance from vert0 to ray origin */ - LLVector4a tvec; - tvec.setSub(orig, vert0); - - /* calculate U parameter and test bounds */ - u = (tvec.dot3(pvec)) * inv_det; - if (u < 0.f || u > 1.f) - { - return FALSE; - } + /* begin calculating determinant - also used to calculate U parameter */ + LLVector4a pvec; + pvec.setCross3(dir, edge2); - /* prepare to test V parameter */ - LLVector4a qvec; - qvec.setSub(tvec, edge1); - - /* calculate V parameter and test bounds */ - v = (dir.dot3(qvec)) * inv_det; - - if (v < 0.f || u + v > 1.f) - { - return FALSE; - } + /* if determinant is near zero, ray lies in plane of triangle */ + F32 det = edge1.dot3(pvec); + + + 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)) * inv_det; + if (u < 0.f || u > 1.f) + { + return FALSE; + } - /* calculate t, ray intersects triangle */ - t = (edge2.dot3(qvec)) * inv_det; + /* prepare to test V parameter */ + tvec.sub(edge1); + + /* calculate V parameter and test bounds */ + v = (dir.dot3(tvec)) * inv_det; + + if (v < 0.f || u + v > 1.f) + { + return FALSE; } + + /* calculate t, ray intersects triangle */ + t = (edge2.dot3(tvec)) * inv_det; - if (intersection_a != NULL) - *intersection_a = u; - if (intersection_b != NULL) - *intersection_b = v; - if (intersection_t != NULL) - *intersection_t = t; + intersection_a = u; + intersection_b = v; + intersection_t = t; return TRUE; @@ -244,7 +317,7 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co //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) + F32& intersection_a, F32& intersection_b, F32& intersection_t, BOOL two_sided) { LLVector4a vert0a, vert1a, vert2a, origa, dira; vert0a.load3(vert0.mV); @@ -253,11 +326,130 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons origa.load3(orig.mV); dira.load3(dir.mV); - return LLTriangleRayIntersect(vert0a, vert1a, vert2a, origa, dira, - intersection_a, intersection_b, intersection_t, two_sided); + 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 LLVolumeOctreeListener : public LLOctreeListener +{ +public: + + LLVolumeOctreeListener(LLOctreeNode* node) + { + node->addListener(this); + + mBounds = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*4, 16); + mExtents = mBounds+2; + } + + ~LLVolumeOctreeListener() + { + _mm_free(mBounds); + } + + //LISTENER FUNCTIONS + virtual void handleChildAddition(const LLOctreeNode* parent, + LLOctreeNode* child) + { + new LLVolumeOctreeListener(child); + } + + virtual void handleStateChange(const LLTreeNode* node) { } + virtual void handleChildRemoval(const LLOctreeNode* parent, + const LLOctreeNode* child) { } + virtual void handleInsertion(const LLTreeNode* node, LLVolumeFace::Triangle* tri) { } + virtual void handleRemoval(const LLTreeNode* node, LLVolumeFace::Triangle* tri) { } + virtual void handleDestruction(const LLTreeNode* node) { } + + +public: + LLVector4a* mBounds; // bounding box (center, size) of this node and all its children (tight fit to objects) + LLVector4a* mExtents; // extents (min, max) of this node and all its children +}; + +class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst +{ +public: + const LLVolumeFace* mFace; + + LLVolumeOctreeRebound(const LLVolumeFace* face) + { + mFace = face; + } + + virtual void visit(const LLOctreeNode* branch) + { + LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0); + + LLVector4a& min = node->mExtents[0]; + LLVector4a& max = node->mExtents[1]; + + if (branch->getElementCount() != 0) + { + const LLVolumeFace::Triangle* tri = *(branch->getData().begin()); + + min = *(tri->mV[0]); + max = *(tri->mV[0]); + + for (LLOctreeNode::const_element_iter iter = + branch->getData().begin(); iter != branch->getData().end(); ++iter) + { + //stretch by triangles in node + tri = *iter; + + min.setMin(*tri->mV[0]); + min.setMin(*tri->mV[1]); + min.setMin(*tri->mV[2]); + + max.setMax(*tri->mV[0]); + max.setMax(*tri->mV[1]); + max.setMax(*tri->mV[2]); + } + + for (S32 i = 0; i < branch->getChildCount(); ++i) + { //stretch by child extents + LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); + min.setMin(child->mExtents[0]); + max.setMax(child->mExtents[1]); + } + } + else if (branch->getChildCount() != 0) + { + LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(0)->getListener(0); + + min = child->mExtents[0]; + max = child->mExtents[1]; + + for (S32 i = 1; i < branch->getChildCount(); ++i) + { //stretch by child extents + child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); + min.setMin(child->mExtents[0]); + max.setMax(child->mExtents[1]); + } + } + else + { + llerrs << "WTF? Empty leaf" << llendl; + } + + 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 //------------------------------------------------------------------- @@ -4244,6 +4436,114 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, } +class LLOctreeTriangleRayIntersect : public LLOctreeTraveler +{ +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) + : 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 traverse(const LLOctreeNode* 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])) + { + node->accept(this); + for (S32 i = 0; i < node->getChildCount(); ++i) + { + traverse(node->getChild(i)); + } + } + } + + void visit(const LLOctreeNode* node) + { + for (LLOctreeNode::const_element_iter iter = + node->getData().begin(); iter != node->getData().end(); ++iter) + { + const LLVolumeFace::Triangle* 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.getF32()); + } + + + 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]])); + } + } + } + } + } +}; + S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, S32 face, LLVector3* intersection,LLVector2* tex_coord, LLVector3* normal, LLVector3* bi_normal) @@ -4288,66 +4588,19 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en { genBinormals(i); } - - LLVector4a* p = (LLVector4a*) face.mPositions; - for (U32 tri = 0; tri < face.mNumIndices/3; tri++) + if (!face.mOctree) { - S32 index1 = face.mIndices[tri*3+0]; - S32 index2 = face.mIndices[tri*3+1]; - S32 index3 = face.mIndices[tri*3+2]; - - F32 a, b, t; + face.createOctree(); + } - if (LLTriangleRayIntersect(p[index1], - p[index2], - p[index3], - 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) - { - LLVector4a intersect = dir; - intersect.mul(closest_t); - intersect.add(start); - intersection->set(intersect.getF32()); - } - - - if (tex_coord != NULL) - { - LLVector2* tc = (LLVector2*) face.mTexCoords; - *tex_coord = ((1.f - a - b) * tc[index1] + - a * tc[index2] + - b * tc[index3]); - - } - - if (normal != NULL) - { - LLVector4* norm = (LLVector4*) face.mNormals; - - *normal = ((1.f - a - b) * LLVector3(norm[index1]) + - a * LLVector3(norm[index2]) + - b * LLVector3(norm[index3])); - } - - if (bi_normal != NULL) - { - LLVector4* binormal = (LLVector4*) face.mBinormals; - *bi_normal = ((1.f - a - b) * LLVector3(binormal[index1]) + - a * LLVector3(binormal[index2]) + - b * LLVector3(binormal[index3])); - } + 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; } } } @@ -5128,13 +5381,29 @@ LLVolumeFace::LLVolumeFace() : mBinormals(NULL), mTexCoords(NULL), mIndices(NULL), - mWeights(NULL) + mWeights(NULL), + mOctree(NULL) { mExtents = (LLVector4a*) _mm_malloc(48, 16); 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*) _mm_malloc(48, 16); mCenter = mExtents+2; @@ -5157,13 +5426,9 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) mNumVertices = 0; mNumIndices = 0; - mPositions = NULL; - mNormals = NULL; - mBinormals = NULL; - mTexCoords = NULL; - mWeights = NULL; - mIndices = NULL; + freeData(); + LLVector4a::memcpyNonAliased16((F32*) mExtents, (F32*) src.mExtents, 12); resizeVertices(src.mNumVertices); @@ -5179,6 +5444,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) LLVector4a::memcpyNonAliased16((F32*) mNormals, (F32*) src.mNormals, vert_size); LLVector4a::memcpyNonAliased16((F32*) mTexCoords, (F32*) src.mTexCoords, vert_size); + if (src.mBinormals) { allocateBinormals(src.mNumVertices); @@ -5216,18 +5482,38 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) } LLVolumeFace::~LLVolumeFace() +{ + _mm_free(mExtents); + mExtents = NULL; + + freeData(); +} + +void LLVolumeFace::freeData() { _mm_free(mPositions); + mPositions = NULL; _mm_free(mNormals); + mNormals = NULL; _mm_free(mTexCoords); + mTexCoords = NULL; _mm_free(mIndices); + mIndices = NULL; _mm_free(mBinormals); + mBinormals = NULL; _mm_free(mWeights); - _mm_free(mExtents); + 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; + if (mTypeMask & CAP_MASK) { return createCap(volume, partial_build); @@ -5250,6 +5536,18 @@ void LLVolumeFace::getVertexData(U16 index, LLVolumeFace::VertexData& cv) cv.mTexCoord = mTexCoords[index]; } +bool LLVolumeFace::VertexMapData::operator==(const LLVolumeFace::VertexData& rhs) const +{ + return getPosition().equal3(rhs.getPosition()) && + mTexCoord == rhs.mTexCoord && + getNormal().equal3(rhs.getNormal()); +} + +bool LLVolumeFace::VertexMapData::ComparePosition::operator()(const LLVector4a& a, const LLVector4a& b) const +{ + return a.less3(b); +} + void LLVolumeFace::optimize(F32 angle_cutoff) { LLVolumeFace new_face; @@ -5305,6 +5603,65 @@ void LLVolumeFace::optimize(F32 angle_cutoff) swapData(new_face); } + +void LLVolumeFace::createOctree() +{ + mOctree = new LLOctreeRoot(LLVector3d(0,0,0), LLVector3d(1,1,1), NULL); + new LLVolumeOctreeListener(mOctree); + + for (U32 i = 0; i < mNumIndices; i+= 3) + { + Triangle* tri = new Triangle(); + + const LLVector4a& v0 = mPositions[mIndices[i]]; + const LLVector4a& v1 = mPositions[mIndices[i+1]]; + const LLVector4a& v2 = mPositions[mIndices[i+2]]; + + tri->mV[0] = &v0; + tri->mV[1] = &v1; + tri->mV[2] = &v2; + + tri->mIndex[0] = mIndices[i]; + tri->mIndex[1] = mIndices[i+1]; + tri->mIndex[2] = mIndices[i+2]; + + LLVector4a min = v0; + min.setMin(v1); + min.setMin(v2); + + LLVector4a max = v0; + max.setMax(v1); + max.setMax(v2); + + LLVector4a center; + center.setAdd(min, max); + center.mul(0.5f); + + + tri->mPositionGroup.setVec(center[0], center[1], center[2]); + + LLVector4a size; + size.setSub(max,min); + + tri->mRadius = size.length3() * 0.5f; + + mOctree->insert(tri); + } + + LLVolumeOctreeRebound rebound(this); + rebound.traverse(mOctree); +} + +const LLVector3d& LLVolumeFace::Triangle::getPositionGroup() const +{ + return mPositionGroup; +} + +const F64& LLVolumeFace::Triangle::getBinRadius() const +{ + return mRadius; +} + void LLVolumeFace::swapData(LLVolumeFace& rhs) { llswap(rhs.mPositions, mPositions); -- cgit v1.2.3 From 9a869d630162292864e01fdd1707efc609fbd6b4 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Sat, 29 May 2010 19:55:13 -0500 Subject: Octree triven raycast works, time to profile. --- indra/llmath/llvolume.cpp | 202 ++-------------------------------------------- 1 file changed, 7 insertions(+), 195 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index d261811aa2..c4172de651 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -48,6 +48,7 @@ #include "lloctree.h" #include "lldarray.h" #include "llvolume.h" +#include "llvolumeoctree.h" #include "llstl.h" #include "llsdserialize.h" #include "llvector4a.h" @@ -133,50 +134,6 @@ BOOL LLLineSegmentBoxIntersect(const F32* start, const F32* end, const F32* cent return true; } -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); - - S32 grt = lhs.greaterThan4(rhs).getComparisonMask(); - - if (grt & 0x7) - { - return false; - } - - LLVector4a f; - f.setCross3(dir, diff); - f.setAbs(f); - - LLVector4a v0; v0.mQ = _mm_shuffle_ps(size.mQ, size.mQ, _MM_SHUFFLE(3,1,0,0)); - LLVector4a v1; v1.mQ = _mm_shuffle_ps(fAWdU.mQ, fAWdU.mQ, _MM_SHUFFLE(3,2,2,1)); - lhs.setMul(v0, v1); - - v0.mQ = _mm_shuffle_ps(size.mQ, size.mQ, _MM_SHUFFLE(3,2,2,1)); - v1.mQ = _mm_shuffle_ps(fAWdU.mQ, fAWdU.mQ, _MM_SHUFFLE(3,1,0,0)); - rhs.setMul(v0, v1); - rhs.add(lhs); - - grt = f.greaterThan4(rhs).getComparisonMask(); - - return (grt & 0x7) ? false : true; -} // intersect test between triangle vert0, vert1, vert2 and a ray from orig in direction dir. @@ -203,7 +160,7 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co LLVector4a det; det.setAllDot3(edge1, pvec); - if (det.greaterEqual4(LLVector4a::getApproximatelyZero()).getComparisonMask()) + if (det.greaterEqual4(LLVector4a::getApproximatelyZero()).getComparisonMask() & 0x7) { /* calculate distance from vert0 to ray origin */ LLVector4a tvec; @@ -213,8 +170,8 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co LLVector4a u; u.setAllDot3(tvec,pvec); - if (u.greaterEqual4(LLVector4a::getZero()).getComparisonMask() && - u.lessEqual4(det).getComparisonMask()) + if ((u.greaterEqual4(LLVector4a::getZero()).getComparisonMask() & 0x7) && + (u.lessEqual4(det).getComparisonMask() & 0x7)) { /* prepare to test V parameter */ LLVector4a qvec; @@ -230,10 +187,10 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co LLVector4a sum_uv; sum_uv.setAdd(u, v); - S32 v_gequal = v.greaterEqual4(LLVector4a::getZero()).getComparisonMask(); - S32 sum_lequal = sum_uv.lessEqual4(det).getComparisonMask(); + S32 v_gequal = v.greaterEqual4(LLVector4a::getZero()).getComparisonMask() & 0x7; + S32 sum_lequal = sum_uv.lessEqual4(det).getComparisonMask() & 0x7; - if (v_gequal && sum_lequal) + if (v_gequal && sum_lequal) { /* calculate t, scale parameters, ray intersects triangle */ LLVector4a t; @@ -338,44 +295,6 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons } } - -class LLVolumeOctreeListener : public LLOctreeListener -{ -public: - - LLVolumeOctreeListener(LLOctreeNode* node) - { - node->addListener(this); - - mBounds = (LLVector4a*) _mm_malloc(sizeof(LLVector4a)*4, 16); - mExtents = mBounds+2; - } - - ~LLVolumeOctreeListener() - { - _mm_free(mBounds); - } - - //LISTENER FUNCTIONS - virtual void handleChildAddition(const LLOctreeNode* parent, - LLOctreeNode* child) - { - new LLVolumeOctreeListener(child); - } - - virtual void handleStateChange(const LLTreeNode* node) { } - virtual void handleChildRemoval(const LLOctreeNode* parent, - const LLOctreeNode* child) { } - virtual void handleInsertion(const LLTreeNode* node, LLVolumeFace::Triangle* tri) { } - virtual void handleRemoval(const LLTreeNode* node, LLVolumeFace::Triangle* tri) { } - virtual void handleDestruction(const LLTreeNode* node) { } - - -public: - LLVector4a* mBounds; // bounding box (center, size) of this node and all its children (tight fit to objects) - LLVector4a* mExtents; // extents (min, max) of this node and all its children -}; - class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst { public: @@ -4436,113 +4355,6 @@ S32 LLVolume::lineSegmentIntersect(const LLVector3& start, const LLVector3& end, } -class LLOctreeTriangleRayIntersect : public LLOctreeTraveler -{ -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) - : 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 traverse(const LLOctreeNode* 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])) - { - node->accept(this); - for (S32 i = 0; i < node->getChildCount(); ++i) - { - traverse(node->getChild(i)); - } - } - } - - void visit(const LLOctreeNode* node) - { - for (LLOctreeNode::const_element_iter iter = - node->getData().begin(); iter != node->getData().end(); ++iter) - { - const LLVolumeFace::Triangle* 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.getF32()); - } - - - 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]])); - } - } - } - } - } -}; S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& end, S32 face, -- cgit v1.2.3 From 26ba00b5554d20ee958693ced87b36fa7f6e3d99 Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Thu, 3 Jun 2010 12:52:28 -0500 Subject: Vectorized octree and much of llspatialpartition and lldrawable. Octree driven raycast. --- indra/llmath/llvolume.cpp | 31 +++++++++++++------------------ 1 file changed, 13 insertions(+), 18 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index c4172de651..72833c019f 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -295,7 +295,7 @@ BOOL LLTriangleRayIntersect(const LLVector3& vert0, const LLVector3& vert1, cons } } -class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst +class LLVolumeOctreeRebound : public LLOctreeTravelerDepthFirst { public: const LLVolumeFace* mFace; @@ -305,7 +305,7 @@ public: mFace = face; } - virtual void visit(const LLOctreeNode* branch) + virtual void visit(const LLOctreeNode* branch) { LLVolumeOctreeListener* node = (LLVolumeOctreeListener*) branch->getListener(0); @@ -314,12 +314,12 @@ public: if (branch->getElementCount() != 0) { - const LLVolumeFace::Triangle* tri = *(branch->getData().begin()); + const LLVolumeTriangle* tri = *(branch->getData().begin()); min = *(tri->mV[0]); max = *(tri->mV[0]); - for (LLOctreeNode::const_element_iter iter = + for (LLOctreeNode::const_element_iter iter = branch->getData().begin(); iter != branch->getData().end(); ++iter) { //stretch by triangles in node @@ -4394,7 +4394,7 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en LLVector4a box_size; box_size.setSub(face.mExtents[1], face.mExtents[0]); - if (LLLineSegmentBoxIntersect(start.getF32(), end.getF32(), box_center.getF32(), box_size.getF32())) + if (LLLineSegmentBoxIntersect(start, end, box_center, box_size)) { if (bi_normal != NULL) // if the caller wants binormals, we may need to generate them { @@ -5418,12 +5418,17 @@ void LLVolumeFace::optimize(F32 angle_cutoff) void LLVolumeFace::createOctree() { - mOctree = new LLOctreeRoot(LLVector3d(0,0,0), LLVector3d(1,1,1), NULL); + LLVector4a center; + LLVector4a size; + center.splat(0.f); + size.splat(1.f); + + mOctree = new LLOctreeRoot(center, size, NULL); new LLVolumeOctreeListener(mOctree); for (U32 i = 0; i < mNumIndices; i+= 3) { - Triangle* tri = new Triangle(); + LLPointer tri = new LLVolumeTriangle(); const LLVector4a& v0 = mPositions[mIndices[i]]; const LLVector4a& v1 = mPositions[mIndices[i+1]]; @@ -5449,8 +5454,7 @@ void LLVolumeFace::createOctree() center.setAdd(min, max); center.mul(0.5f); - - tri->mPositionGroup.setVec(center[0], center[1], center[2]); + *tri->mPositionGroup = center; LLVector4a size; size.setSub(max,min); @@ -5464,15 +5468,6 @@ void LLVolumeFace::createOctree() rebound.traverse(mOctree); } -const LLVector3d& LLVolumeFace::Triangle::getPositionGroup() const -{ - return mPositionGroup; -} - -const F64& LLVolumeFace::Triangle::getBinRadius() const -{ - return mRadius; -} void LLVolumeFace::swapData(LLVolumeFace& rhs) { -- cgit v1.2.3 From 087b7499082c7f0ae867990a102bc8f90a83471d Mon Sep 17 00:00:00 2001 From: Tofu Linden Date: Fri, 4 Jun 2010 08:46:00 +0100 Subject: fix scoping issues for gcc --- indra/llmath/llvolume.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 72833c019f..ef1ab57036 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -4406,7 +4406,7 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en face.createOctree(); } - LLVector4a* p = (LLVector4a*) face.mPositions; + //LLVector4a* p = (LLVector4a*) face.mPositions; LLOctreeTriangleRayIntersect intersect(start, dir, &face, &closest_t, intersection, tex_coord, normal, bi_normal); intersect.traverse(face.mOctree); -- cgit v1.2.3 From dc2f50642bf6c785263d91ca53ec337f3898b990 Mon Sep 17 00:00:00 2001 From: Tofu Linden Date: Fri, 4 Jun 2010 08:54:03 +0100 Subject: lots of _mm_malloc and _mm_free -> ll_aligned_malloc_16 and ll_aligned_free_16 more to come. --- indra/llmath/llvolume.cpp | 86 +++++++++++++++++++++++------------------------ 1 file changed, 43 insertions(+), 43 deletions(-) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index ef1ab57036..a8684759f3 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -1987,7 +1987,7 @@ BOOL LLVolume::generate() void LLVolumeFace::VertexData::init() { - mData = (LLVector4a*) _mm_malloc(32, 16); + mData = (LLVector4a*) ll_aligned_malloc_16(32); } LLVolumeFace::VertexData::VertexData() @@ -2004,7 +2004,7 @@ LLVolumeFace::VertexData::VertexData(const VertexData& rhs) LLVolumeFace::VertexData::~VertexData() { - _mm_free(mData); + ll_aligned_free_16(mData); } LLVector4a& LLVolumeFace::VertexData::getPosition() @@ -5196,7 +5196,7 @@ LLVolumeFace::LLVolumeFace() : mWeights(NULL), mOctree(NULL) { - mExtents = (LLVector4a*) _mm_malloc(48, 16); + mExtents = (LLVector4a*) ll_aligned_malloc_16(48); mCenter = mExtents+2; } @@ -5217,7 +5217,7 @@ LLVolumeFace::LLVolumeFace(const LLVolumeFace& src) mWeights(NULL), mOctree(NULL) { - mExtents = (LLVector4a*) _mm_malloc(48, 16); + mExtents = (LLVector4a*) ll_aligned_malloc_16(48); mCenter = mExtents+2; *this = src; } @@ -5264,7 +5264,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) } else { - _mm_free(mBinormals); + ll_aligned_free_16(mBinormals); mBinormals = NULL; } @@ -5275,7 +5275,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) } else { - _mm_free(mWeights); + ll_aligned_free_16(mWeights); mWeights = NULL; } } @@ -5295,7 +5295,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) LLVolumeFace::~LLVolumeFace() { - _mm_free(mExtents); + ll_aligned_free_16(mExtents); mExtents = NULL; freeData(); @@ -5303,17 +5303,17 @@ LLVolumeFace::~LLVolumeFace() void LLVolumeFace::freeData() { - _mm_free(mPositions); + ll_aligned_free_16(mPositions); mPositions = NULL; - _mm_free(mNormals); + ll_aligned_free_16(mNormals); mNormals = NULL; - _mm_free(mTexCoords); + ll_aligned_free_16(mTexCoords); mTexCoords = NULL; - _mm_free(mIndices); + ll_aligned_free_16(mIndices); mIndices = NULL; - _mm_free(mBinormals); + ll_aligned_free_16(mBinormals); mBinormals = NULL; - _mm_free(mWeights); + ll_aligned_free_16(mWeights); mWeights = NULL; delete mOctree; @@ -6082,21 +6082,21 @@ void LLVolumeFace::createBinormals() void LLVolumeFace::resizeVertices(S32 num_verts) { - _mm_free(mPositions); - _mm_free(mNormals); - _mm_free(mBinormals); - _mm_free(mTexCoords); + ll_aligned_free_16(mPositions); + ll_aligned_free_16(mNormals); + ll_aligned_free_16(mBinormals); + ll_aligned_free_16(mTexCoords); mBinormals = NULL; if (num_verts) { - mPositions = (LLVector4a*) _mm_malloc(num_verts*16, 16); - mNormals = (LLVector4a*) _mm_malloc(num_verts*16, 16); + mPositions = (LLVector4a*) ll_aligned_malloc_16(num_verts*16); + mNormals = (LLVector4a*) ll_aligned_malloc_16(num_verts*16); //pad texture coordinate block end to allow for QWORD reads S32 size = ((num_verts*8) + 0xF) & ~0xF; - mTexCoords = (LLVector2*) _mm_malloc(size, 16); + mTexCoords = (LLVector2*) ll_aligned_malloc_16(size); } else { @@ -6119,20 +6119,20 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con S32 new_size = new_verts*16; //positions - LLVector4a* dst = (LLVector4a*) _mm_malloc(new_size, 16); + LLVector4a* dst = (LLVector4a*) ll_aligned_malloc_16(new_size); if (mPositions) { LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mPositions, new_size/4); - _mm_free(mPositions); + ll_aligned_free_16(mPositions); } mPositions = dst; //normals - dst = (LLVector4a*) _mm_malloc(new_size, 16); + dst = (LLVector4a*) ll_aligned_malloc_16(new_size); if (mNormals) { LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mNormals, new_size/4); - _mm_free(mNormals); + ll_aligned_free_16(mNormals); } mNormals = dst; @@ -6140,16 +6140,16 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con new_size = ((new_verts*8)+0xF) & ~0xF; { - LLVector2* dst = (LLVector2*) _mm_malloc(new_size, 16); + LLVector2* dst = (LLVector2*) ll_aligned_malloc_16(new_size); if (mTexCoords) { LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mTexCoords, new_size/4); - _mm_free(mTexCoords); + ll_aligned_free_16(mTexCoords); } } //just clear binormals - _mm_free(mBinormals); + ll_aligned_free_16(mBinormals); mBinormals = NULL; mPositions[mNumVertices] = pos; @@ -6161,26 +6161,26 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con void LLVolumeFace::allocateBinormals(S32 num_verts) { - _mm_free(mBinormals); - mBinormals = (LLVector4a*) _mm_malloc(num_verts*16, 16); + ll_aligned_free_16(mBinormals); + mBinormals = (LLVector4a*) ll_aligned_malloc_16(num_verts*16); } void LLVolumeFace::allocateWeights(S32 num_verts) { - _mm_free(mWeights); - mWeights = (LLVector4a*) _mm_malloc(num_verts*16, 16); + ll_aligned_free_16(mWeights); + mWeights = (LLVector4a*) ll_aligned_malloc_16(num_verts*16); } void LLVolumeFace::resizeIndices(S32 num_indices) { - _mm_free(mIndices); + ll_aligned_free_16(mIndices); if (num_indices) { //pad index block end to allow for QWORD reads S32 size = ((num_indices*2) + 0xF) & ~0xF; - mIndices = (U16*) _mm_malloc(size,16); + mIndices = (U16*) ll_aligned_malloc_16(size); } else { @@ -6198,9 +6198,9 @@ void LLVolumeFace::pushIndex(const U16& idx) S32 old_size = (mNumIndices+0xF) & ~0xF; if (new_size != old_size) { - U16* dst = (U16*) _mm_malloc(new_size, 16); + U16* dst = (U16*) ll_aligned_malloc_16(new_size); LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mIndices, new_size/4); - _mm_free(mIndices); + ll_aligned_free_16(mIndices); mIndices = dst; } @@ -6237,17 +6237,17 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat } - LLVector4a* new_pos = (LLVector4a*) _mm_malloc(new_count*16, 16); - LLVector4a* new_norm = (LLVector4a*) _mm_malloc(new_count*16, 16); - LLVector2* new_tc = (LLVector2*) _mm_malloc((new_count*8+0xF) & ~0xF, 16); + LLVector4a* new_pos = (LLVector4a*) ll_aligned_malloc_16(new_count*16); + LLVector4a* new_norm = (LLVector4a*) ll_aligned_malloc_16(new_count*16); + LLVector2* new_tc = (LLVector2*) ll_aligned_malloc_16((new_count*8+0xF) & ~0xF); LLVector4a::memcpyNonAliased16((F32*) new_pos, (F32*) mPositions, new_count*4); LLVector4a::memcpyNonAliased16((F32*) new_norm, (F32*) mNormals, new_count*4); LLVector4a::memcpyNonAliased16((F32*) new_tc, (F32*) mTexCoords, new_count*2); - _mm_free(mPositions); - _mm_free(mNormals); - _mm_free(mTexCoords); + ll_aligned_free_16(mPositions); + ll_aligned_free_16(mNormals); + ll_aligned_free_16(mTexCoords); mPositions = new_pos; mNormals = new_norm; @@ -6287,9 +6287,9 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat new_count = mNumIndices + face.mNumIndices; - U16* new_indices = (U16*) _mm_malloc((new_count*2+0xF) & ~0xF, 16); + U16* new_indices = (U16*) ll_aligned_malloc_16((new_count*2+0xF) & ~0xF); LLVector4a::memcpyNonAliased16((F32*) new_indices, (F32*) mIndices, new_count/2); - _mm_free(mIndices); + ll_aligned_free_16(mIndices); mIndices = new_indices; mNumIndices = new_count; -- cgit v1.2.3 From 6ca40c7afbd794b161904b63b88a95273ae80c9c Mon Sep 17 00:00:00 2001 From: Tofu Linden Date: Fri, 4 Jun 2010 09:10:00 +0100 Subject: more needed #includes --- indra/llmath/llvolume.cpp | 1 + 1 file changed, 1 insertion(+) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index a8684759f3..ce7e20e9b4 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -31,6 +31,7 @@ */ #include "linden_common.h" +#include "llmemory.h" #include "llmath.h" #include -- cgit v1.2.3 From a8f0e47fd5deee1e45b4126ee43955a7bc68bb5d Mon Sep 17 00:00:00 2001 From: Dave Parks Date: Fri, 4 Jun 2010 12:07:55 -0500 Subject: Normal debug display and fix for bad bump mapping and planar texture coordinates. --- indra/llmath/llvolume.cpp | 2 ++ 1 file changed, 2 insertions(+) (limited to 'indra/llmath/llvolume.cpp') diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index 72833c019f..8cb9475994 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -6076,6 +6076,8 @@ void LLVolumeFace::createBinormals() for (U32 i = 0; i < mNumVertices; i++) { binorm[i].normalize3fast(); + //bump map/planar projection code requires normals to be normalized + mNormals[i].normalize3fast(); } } } -- cgit v1.2.3