summaryrefslogtreecommitdiff
path: root/indra/llmath
diff options
context:
space:
mode:
Diffstat (limited to 'indra/llmath')
-rw-r--r--indra/llmath/llcamera.cpp222
-rw-r--r--indra/llmath/llcamera.h9
-rw-r--r--indra/llmath/llvolume.cpp487
-rw-r--r--indra/llmath/llvolume.h18
-rw-r--r--indra/llmath/v3color.h51
-rw-r--r--indra/llmath/v4color.h41
6 files changed, 344 insertions, 484 deletions
diff --git a/indra/llmath/llcamera.cpp b/indra/llmath/llcamera.cpp
index 9034182072..18d704dd0f 100644
--- a/indra/llmath/llcamera.cpp
+++ b/indra/llmath/llcamera.cpp
@@ -311,104 +311,6 @@ int LLCamera::sphereInFrustumQuick(const LLVector3 &sphere_center, const F32 rad
return 0;
}
-// HACK: This version is still around because the version below doesn't work
-// unless the agent planes are initialized.
-// Return 1 if sphere is in frustum, 2 if fully in frustum, otherwise 0.
-// NOTE: 'center' is in absolute frame.
-int LLCamera::sphereInFrustumOld(const LLVector3 &sphere_center, const F32 radius) const
-{
- // Returns 1 if sphere is in frustum, 0 if not.
- // modified so that default view frust is along X with Z vertical
- F32 x, y, z, rightDist, leftDist, topDist, bottomDist;
-
- // Subtract the view position
- //LLVector3 relative_center;
- //relative_center = sphere_center - getOrigin();
- LLVector3 rel_center(sphere_center);
- rel_center -= mOrigin;
-
- bool all_in = TRUE;
-
- // Transform relative_center.x to camera frame
- x = mXAxis * rel_center;
- if (x < MIN_NEAR_PLANE - radius)
- {
- return 0;
- }
- else if (x < MIN_NEAR_PLANE + radius)
- {
- all_in = FALSE;
- }
-
- if (x > mFarPlane + radius)
- {
- return 0;
- }
- else if (x > mFarPlane - radius)
- {
- all_in = FALSE;
- }
-
- // Transform relative_center.y to camera frame
- y = mYAxis * rel_center;
-
- // distance to plane is the dot product of (x, y, 0) * plane_normal
- rightDist = x * mLocalPlanes[PLANE_RIGHT][VX] + y * mLocalPlanes[PLANE_RIGHT][VY];
- if (rightDist < -radius)
- {
- return 0;
- }
- else if (rightDist < radius)
- {
- all_in = FALSE;
- }
-
- leftDist = x * mLocalPlanes[PLANE_LEFT][VX] + y * mLocalPlanes[PLANE_LEFT][VY];
- if (leftDist < -radius)
- {
- return 0;
- }
- else if (leftDist < radius)
- {
- all_in = FALSE;
- }
-
- // Transform relative_center.y to camera frame
- z = mZAxis * rel_center;
-
- topDist = x * mLocalPlanes[PLANE_TOP][VX] + z * mLocalPlanes[PLANE_TOP][VZ];
- if (topDist < -radius)
- {
- return 0;
- }
- else if (topDist < radius)
- {
- all_in = FALSE;
- }
-
- bottomDist = x * mLocalPlanes[PLANE_BOTTOM][VX] + z * mLocalPlanes[PLANE_BOTTOM][VZ];
- if (bottomDist < -radius)
- {
- return 0;
- }
- else if (bottomDist < radius)
- {
- all_in = FALSE;
- }
-
- if (all_in)
- {
- return 2;
- }
-
- return 1;
-}
-
-
-// HACK: This (presumably faster) version only currently works if you set up the
-// frustum planes using GL. At some point we should get those planes through another
-// mechanism, and then we can get rid of the "old" version above.
-
// Return 1 if sphere is in frustum, 2 if fully in frustum, otherwise 0.
// NOTE: 'center' is in absolute frame.
int LLCamera::sphereInFrustum(const LLVector3 &sphere_center, const F32 radius) const
@@ -463,65 +365,6 @@ F32 LLCamera::heightInPixels(const LLVector3 &center, F32 radius ) const
}
}
-// If pos is visible, return the distance from pos to the camera.
-// Use fudge distance to scale rad against top/bot/left/right planes
-// Otherwise, return -distance
-F32 LLCamera::visibleDistance(const LLVector3 &pos, F32 rad, F32 fudgedist, U32 planemask) const
-{
- if (mFixedDistance > 0)
- {
- return mFixedDistance;
- }
- LLVector3 dvec = pos - mOrigin;
- // Check visibility
- F32 dist = dvec.magVec();
- if (dist > rad)
- {
- F32 dp,tdist;
- dp = dvec * mXAxis;
- if (dp < -rad)
- return -dist;
-
- rad *= fudgedist;
- LLVector3 tvec(pos);
- for (int p=0; p<PLANE_NUM; p++)
- {
- if (!(planemask & (1<<p)))
- continue;
- tdist = -(mWorldPlanes[p].dist(tvec));
- if (tdist > rad)
- return -dist;
- }
- }
- return dist;
-}
-
-// Like visibleDistance, except uses mHorizPlanes[], which are left and right
-// planes perpindicular to (0,0,1) in world space
-F32 LLCamera::visibleHorizDistance(const LLVector3 &pos, F32 rad, F32 fudgedist, U32 planemask) const
-{
- if (mFixedDistance > 0)
- {
- return mFixedDistance;
- }
- LLVector3 dvec = pos - mOrigin;
- // Check visibility
- F32 dist = dvec.magVec();
- if (dist > rad)
- {
- rad *= fudgedist;
- LLVector3 tvec(pos);
- for (int p=0; p<HORIZ_PLANE_NUM; p++)
- {
- if (!(planemask & (1<<p)))
- continue;
- F32 tdist = -(mHorizPlanes[p].dist(tvec));
- if (tdist > rad)
- return -dist;
- }
- }
- return dist;
-}
// ---------------- friends and operators ----------------
@@ -536,18 +379,6 @@ std::ostream& operator<<(std::ostream &s, const LLCamera &C)
s << " Aspect = " << C.getAspect() << "\n";
s << " NearPlane = " << C.mNearPlane << "\n";
s << " FarPlane = " << C.mFarPlane << "\n";
- s << " TopPlane = " << C.mLocalPlanes[LLCamera::PLANE_TOP][VX] << " "
- << C.mLocalPlanes[LLCamera::PLANE_TOP][VY] << " "
- << C.mLocalPlanes[LLCamera::PLANE_TOP][VZ] << "\n";
- s << " BottomPlane = " << C.mLocalPlanes[LLCamera::PLANE_BOTTOM][VX] << " "
- << C.mLocalPlanes[LLCamera::PLANE_BOTTOM][VY] << " "
- << C.mLocalPlanes[LLCamera::PLANE_BOTTOM][VZ] << "\n";
- s << " LeftPlane = " << C.mLocalPlanes[LLCamera::PLANE_LEFT][VX] << " "
- << C.mLocalPlanes[LLCamera::PLANE_LEFT][VY] << " "
- << C.mLocalPlanes[LLCamera::PLANE_LEFT][VZ] << "\n";
- s << " RightPlane = " << C.mLocalPlanes[LLCamera::PLANE_RIGHT][VX] << " "
- << C.mLocalPlanes[LLCamera::PLANE_RIGHT][VY] << " "
- << C.mLocalPlanes[LLCamera::PLANE_RIGHT][VZ] << "\n";
s << "}";
return s;
}
@@ -675,26 +506,6 @@ void LLCamera::calcRegionFrustumPlanes(const LLVector3& shift, F32 far_clip_dist
void LLCamera::calculateFrustumPlanes(F32 left, F32 right, F32 top, F32 bottom)
{
- LLVector3 a, b, c;
-
- // For each plane we need to define 3 points (LLVector3's) in camera view space.
- // The order in which we pass the points to planeFromPoints() matters, because the
- // plane normal has a degeneracy of 2; we want it pointing _into_ the frustum.
-
- a.setVec(0.0f, 0.0f, 0.0f);
- b.setVec(mFarPlane, right, top);
- c.setVec(mFarPlane, right, bottom);
- mLocalPlanes[PLANE_RIGHT].setVec(a, b, c);
-
- c.setVec(mFarPlane, left, top);
- mLocalPlanes[PLANE_TOP].setVec(a, c, b);
-
- b.setVec(mFarPlane, left, bottom);
- mLocalPlanes[PLANE_LEFT].setVec(a, b, c);
-
- c.setVec(mFarPlane, right, bottom);
- mLocalPlanes[PLANE_BOTTOM].setVec( a, c, b);
-
//calculate center and radius squared of frustum in world absolute coordinates
static LLVector3 const X_AXIS(1.f, 0.f, 0.f);
mFrustCenter = X_AXIS*mFarPlane*0.5f;
@@ -718,39 +529,6 @@ void LLCamera::calculateFrustumPlanesFromWindow(F32 x1, F32 y1, F32 x2, F32 y2)
calculateFrustumPlanes(left, right, top, bottom);
}
-void LLCamera::calculateWorldFrustumPlanes()
-{
- F32 d;
- LLVector3 center = mOrigin - mXAxis*mNearPlane;
- mWorldPlanePos = center;
- LLVector3 pnorm;
- for (int p = 0; p < PLANE_NUM; p++)
- {
- mLocalPlanes[p].getVector3(pnorm);
- LLVector3 norm = rotateToAbsolute(pnorm);
- norm.normVec();
- d = -(center * norm);
- mWorldPlanes[p] = LLPlane(norm, d);
- }
- // horizontal planes, perpindicular to (0,0,1);
- LLVector3 zaxis(0, 0, 1.0f);
- F32 yaw = getYaw();
- {
- LLVector3 tnorm;
- mLocalPlanes[PLANE_LEFT].getVector3(tnorm);
- tnorm.rotVec(yaw, zaxis);
- d = -(mOrigin * tnorm);
- mHorizPlanes[HORIZ_PLANE_LEFT] = LLPlane(tnorm, d);
- }
- {
- LLVector3 tnorm;
- mLocalPlanes[PLANE_RIGHT].getVector3(tnorm);
- tnorm.rotVec(yaw, zaxis);
- d = -(mOrigin * tnorm);
- mHorizPlanes[HORIZ_PLANE_RIGHT] = LLPlane(tnorm, d);
- }
-}
-
// NOTE: this is the OpenGL matrix that will transform the default OpenGL view
// (-Z=at, Y=up) to the default view of the LLCamera class (X=at, Z=up):
//
diff --git a/indra/llmath/llcamera.h b/indra/llmath/llcamera.h
index d0afa0e88f..27eaa614c9 100644
--- a/indra/llmath/llcamera.h
+++ b/indra/llmath/llcamera.h
@@ -131,14 +131,10 @@ private:
S32 mViewHeightInPixels; // for ViewHeightInPixels() only
F32 mNearPlane;
F32 mFarPlane;
- LL_ALIGN_16(LLPlane mLocalPlanes[PLANE_NUM]);
F32 mFixedDistance; // Always return this distance, unless < 0
LLVector3 mFrustCenter; // center of frustum and radius squared for ultra-quick exclusion test
F32 mFrustRadiusSquared;
- LL_ALIGN_16(LLPlane mWorldPlanes[PLANE_NUM]);
- LL_ALIGN_16(LLPlane mHorizPlanes[HORIZ_PLANE_NUM]);
-
U32 mPlaneCount; //defaults to 6, if setUserClipPlane is called, uses user supplied clip plane in
LLVector3 mWorldPlanePos; // Position of World Planes (may be offset from camera)
@@ -184,7 +180,6 @@ public:
return atan2f(mXAxis[VZ], xylen);
}
- const LLPlane& getWorldPlane(S32 index) const { return mWorldPlanes[index]; }
const LLVector3& getWorldPlanePos() const { return mWorldPlanePos; }
// Copy mView, mAspect, mNearPlane, and mFarPlane to buffer.
@@ -200,7 +195,6 @@ public:
// Returns 1 if partly in, 2 if fully in.
// NOTE: 'center' is in absolute frame.
- S32 sphereInFrustumOld(const LLVector3 &center, const F32 radius) const;
S32 sphereInFrustum(const LLVector3 &center, const F32 radius) const;
S32 pointInFrustum(const LLVector3 &point) const { return sphereInFrustum(point, 0.0f); }
S32 sphereInFrustumFull(const LLVector3 &center, const F32 radius) const { return sphereInFrustum(center, radius); }
@@ -217,8 +211,6 @@ public:
F32 heightInPixels(const LLVector3 &center, F32 radius ) const;
// return the distance from pos to camera if visible (-distance if not visible)
- F32 visibleDistance(const LLVector3 &pos, F32 rad, F32 fudgescale = 1.0f, U32 planemask = PLANE_ALL_MASK) const;
- F32 visibleHorizDistance(const LLVector3 &pos, F32 rad, F32 fudgescale = 1.0f, U32 planemask = HORIZ_PLANE_ALL_MASK) const;
void setFixedDistance(F32 distance) { mFixedDistance = distance; }
friend std::ostream& operator<<(std::ostream &s, const LLCamera &C);
@@ -227,7 +219,6 @@ protected:
void calculateFrustumPlanes();
void calculateFrustumPlanes(F32 left, F32 right, F32 top, F32 bottom);
void calculateFrustumPlanesFromWindow(F32 x1, F32 y1, F32 x2, F32 y2);
- void calculateWorldFrustumPlanes();
} LL_ALIGN_POSTFIX(16);
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp
index 40f7b1e9fb..91e463cc32 100644
--- a/indra/llmath/llvolume.cpp
+++ b/indra/llmath/llvolume.cpp
@@ -32,6 +32,7 @@
#include <stdint.h>
#endif
#include <cmath>
+#include <unordered_map>
#include "llerror.h"
@@ -52,6 +53,11 @@
#include "llmeshoptimizer.h"
#include "lltimer.h"
+#include "mikktspace/mikktspace.h"
+#include "mikktspace/mikktspace.c" // insert mikktspace implementation into llvolume object file
+
+#include "meshoptimizer/meshoptimizer.h"
+
#define DEBUG_SILHOUETTE_BINORMALS 0
#define DEBUG_SILHOUETTE_NORMALS 0 // TomY: Use this to display normals using the silhouette
#define DEBUG_SILHOUETTE_EDGE_MAP 0 // DaveP: Use this to display edge map using the silhouette
@@ -2093,7 +2099,9 @@ void LLVolume::regen()
void LLVolume::genTangents(S32 face)
{
- mVolumeFaces[face].createTangents();
+ // generate legacy tangents for the specified face
+ llassert(!isMeshAssetLoaded() || mVolumeFaces[face].mTangents != nullptr); // if this is a complete mesh asset, we should already have tangents
+ mVolumeFaces[face].createTangents();
}
LLVolume::~LLVolume()
@@ -2433,11 +2441,10 @@ bool LLVolume::unpackVolumeFacesInternal(const LLSD& mdl)
LLSD::Binary pos = mdl[i]["Position"];
LLSD::Binary norm = mdl[i]["Normal"];
+ LLSD::Binary tangent = mdl[i]["Tangent"];
LLSD::Binary tc = mdl[i]["TexCoord0"];
LLSD::Binary idx = mdl[i]["TriangleList"];
-
-
//copy out indices
S32 num_indices = idx.size() / 2;
const S32 indices_to_discard = num_indices % 3;
@@ -2492,6 +2499,16 @@ bool LLVolume::unpackVolumeFacesInternal(const LLSD& mdl)
min_tc.setValue(mdl[i]["TexCoord0Domain"]["Min"]);
max_tc.setValue(mdl[i]["TexCoord0Domain"]["Max"]);
+ //unpack normalized scale/translation
+ if (mdl[i].has("NormalizedScale"))
+ {
+ face.mNormalizedScale.setValue(mdl[i]["NormalizedScale"]);
+ }
+ else
+ {
+ face.mNormalizedScale.set(1, 1, 1);
+ }
+
LLVector4a pos_range;
pos_range.setSub(max_pos, min_pos);
LLVector2 tc_range2 = max_tc - min_tc;
@@ -2542,6 +2559,34 @@ bool LLVolume::unpackVolumeFacesInternal(const LLSD& mdl)
}
}
+#if 0 // keep this code for now in case we decide to add support for on-the-wire tangents
+ {
+ if (!tangent.empty())
+ {
+ face.allocateTangents(face.mNumVertices);
+ U16* t = (U16*)&(tangent[0]);
+
+ // NOTE: tangents coming from the asset may not be mikkt space, but they should always be used by the GLTF shaders to
+ // maintain compliance with the GLTF spec
+ LLVector4a* t_out = face.mTangents;
+
+ for (U32 j = 0; j < num_verts; ++j)
+ {
+ t_out->set((F32)t[0], (F32)t[1], (F32)t[2], (F32) t[3]);
+ t_out->div(65535.f);
+ t_out->mul(2.f);
+ t_out->sub(1.f);
+
+ F32* tp = t_out->getF32ptr();
+ tp[3] = tp[3] < 0.f ? -1.f : 1.f;
+
+ t_out++;
+ t += 4;
+ }
+ }
+ }
+#endif
+
{
if (!tc.empty())
{
@@ -2745,7 +2790,7 @@ bool LLVolume::unpackVolumeFacesInternal(const LLSD& mdl)
}
}
- if (!cacheOptimize())
+ if (!cacheOptimize(true))
{
// Out of memory?
LL_WARNS() << "Failed to optimize!" << LL_ENDL;
@@ -2786,11 +2831,11 @@ void LLVolume::copyVolumeFaces(const LLVolume* volume)
mSculptLevel = 0;
}
-bool LLVolume::cacheOptimize()
+bool LLVolume::cacheOptimize(bool gen_tangents)
{
for (S32 i = 0; i < mVolumeFaces.size(); ++i)
{
- if (!mVolumeFaces[i].cacheOptimize())
+ if (!mVolumeFaces[i].cacheOptimize(gen_tangents))
{
return false;
}
@@ -3306,12 +3351,12 @@ BOOL LLVolume::isFlat(S32 face)
bool LLVolumeParams::isSculpt() const
{
- return mSculptID.notNull();
+ return (mSculptType & LL_SCULPT_TYPE_MASK) != LL_SCULPT_TYPE_NONE;
}
bool LLVolumeParams::isMeshSculpt() const
{
- return isSculpt() && ((mSculptType & LL_SCULPT_TYPE_MASK) == LL_SCULPT_TYPE_MESH);
+ return (mSculptType & LL_SCULPT_TYPE_MASK) == LL_SCULPT_TYPE_MESH;
}
bool LLVolumeParams::operator==(const LLVolumeParams &params) const
@@ -3726,6 +3771,7 @@ bool LLVolumeParams::validate(U8 prof_curve, F32 prof_begin, F32 prof_end, F32 h
void LLVolume::getLoDTriangleCounts(const LLVolumeParams& params, S32* counts)
{ //attempt to approximate the number of triangles that will result from generating a volume LoD set for the
//supplied LLVolumeParams -- inaccurate, but a close enough approximation for determining streaming cost
+ LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME;
F32 detail[] = {1.f, 1.5f, 2.5f, 4.f};
for (S32 i = 0; i < 4; i++)
{
@@ -4073,7 +4119,7 @@ S32 LLVolume::lineSegmentIntersect(const LLVector4a& start, const LLVector4a& en
{
if (tangent_out != NULL) // if the caller wants tangents, we may need to generate them
{
- genTangents(i);
+ genTangents(i);
}
if (isUnique())
@@ -4861,6 +4907,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src)
}
mOptimized = src.mOptimized;
+ mNormalizedScale = src.mNormalizedScale;
//delete
return *this;
@@ -5383,256 +5430,218 @@ public:
}
};
+// data structures for tangent generation
-bool LLVolumeFace::cacheOptimize()
-{ //optimize for vertex cache according to Forsyth method:
- // http://home.comcast.net/~tom_forsyth/papers/fast_vert_cache_opt.html
-
- llassert(!mOptimized);
- mOptimized = TRUE;
+struct MikktData
+{
+ LLVolumeFace* face;
+ std::vector<LLVector3> p;
+ std::vector<LLVector3> n;
+ std::vector<LLVector2> tc;
+ std::vector<LLVector4> w;
+ std::vector<LLVector4> t;
- LLVCacheLRU cache;
-
- if (mNumVertices < 3 || mNumIndices < 3)
- { //nothing to do
- return true;
- }
+ MikktData(LLVolumeFace* f)
+ : face(f)
+ {
+ U32 count = face->mNumIndices;
- //mapping of vertices to triangles and indices
- std::vector<LLVCacheVertexData> vertex_data;
+ p.resize(count);
+ n.resize(count);
+ tc.resize(count);
+ t.resize(count);
- //mapping of triangles do vertices
- std::vector<LLVCacheTriangleData> triangle_data;
+ if (face->mWeights)
+ {
+ w.resize(count);
+ }
- try
- {
- triangle_data.resize(mNumIndices / 3);
- vertex_data.resize(mNumVertices);
- for (U32 i = 0; i < mNumIndices; i++)
- { //populate vertex data and triangle data arrays
- U16 idx = mIndices[i];
- U32 tri_idx = i / 3;
+ LLVector3 inv_scale(1.f / face->mNormalizedScale.mV[0], 1.f / face->mNormalizedScale.mV[1], 1.f / face->mNormalizedScale.mV[2]);
+
- if (idx >= mNumVertices)
+ for (int i = 0; i < face->mNumIndices; ++i)
+ {
+ U32 idx = face->mIndices[i];
+
+ p[i].set(face->mPositions[idx].getF32ptr());
+ p[i].scaleVec(face->mNormalizedScale); //put mesh in original coordinate frame when reconstructing tangents
+ n[i].set(face->mNormals[idx].getF32ptr());
+ n[i].scaleVec(inv_scale);
+ n[i].normalize();
+ tc[i].set(face->mTexCoords[idx]);
+
+ if (idx >= face->mNumVertices)
{
// invalid index
// replace with a valid index to avoid crashes
- idx = mNumVertices - 1;
- mIndices[i] = idx;
+ idx = face->mNumVertices - 1;
+ face->mIndices[i] = idx;
// Needs better logging
LL_DEBUGS_ONCE("LLVOLUME") << "Invalid index, substituting" << LL_ENDL;
}
- vertex_data[idx].mTriangles.push_back(&(triangle_data[tri_idx]));
- vertex_data[idx].mIdx = idx;
- triangle_data[tri_idx].mVertex[i % 3] = &(vertex_data[idx]);
+ if (face->mWeights)
+ {
+ w[i].set(face->mWeights[idx].getF32ptr());
+ }
}
}
- catch (std::bad_alloc&)
- {
- // resize or push_back failed
- LL_WARNS("LLVOLUME") << "Resize for " << mNumVertices << " vertices failed" << LL_ENDL;
- return false;
- }
+};
- /*F32 pre_acmr = 1.f;
- //measure cache misses from before rebuild
- {
- LLVCacheFIFO test_cache;
- for (U32 i = 0; i < mNumIndices; ++i)
- {
- test_cache.addVertex(&vertex_data[mIndices[i]]);
- }
- for (U32 i = 0; i < mNumVertices; i++)
- {
- vertex_data[i].mCacheTag = -1;
- }
+bool LLVolumeFace::cacheOptimize(bool gen_tangents)
+{ //optimize for vertex cache according to Forsyth method:
+ LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME;
+ llassert(!mOptimized);
+ mOptimized = TRUE;
- pre_acmr = (F32) test_cache.mMisses/(mNumIndices/3);
- }*/
+ if (gen_tangents && mNormals && mTexCoords)
+ { // generate mikkt space tangents before cache optimizing since the index buffer may change
+ // a bit of a hack to do this here, but this function gets called exactly once for the lifetime of a mesh
+ // and is executed on a background thread
+ SMikkTSpaceInterface ms;
- for (U32 i = 0; i < mNumVertices; i++)
- { //initialize score values (no cache -- might try a fifo cache here)
- LLVCacheVertexData& data = vertex_data[i];
+ ms.m_getNumFaces = [](const SMikkTSpaceContext* pContext)
+ {
+ MikktData* data = (MikktData*)pContext->m_pUserData;
+ LLVolumeFace* face = data->face;
+ return face->mNumIndices / 3;
+ };
- data.mScore = find_vertex_score(data);
- data.mActiveTriangles = data.mTriangles.size();
+ ms.m_getNumVerticesOfFace = [](const SMikkTSpaceContext* pContext, const int iFace)
+ {
+ return 3;
+ };
- for (U32 j = 0; j < data.mActiveTriangles; ++j)
- {
- data.mTriangles[j]->mScore += data.mScore;
- }
- }
+ ms.m_getPosition = [](const SMikkTSpaceContext* pContext, float fvPosOut[], const int iFace, const int iVert)
+ {
+ MikktData* data = (MikktData*)pContext->m_pUserData;
+ F32* v = data->p[iFace * 3 + iVert].mV;
+ fvPosOut[0] = v[0];
+ fvPosOut[1] = v[1];
+ fvPosOut[2] = v[2];
+ };
+
+ ms.m_getNormal = [](const SMikkTSpaceContext* pContext, float fvNormOut[], const int iFace, const int iVert)
+ {
+ MikktData* data = (MikktData*)pContext->m_pUserData;
+ F32* n = data->n[iFace * 3 + iVert].mV;
+ fvNormOut[0] = n[0];
+ fvNormOut[1] = n[1];
+ fvNormOut[2] = n[2];
+ };
+
+ ms.m_getTexCoord = [](const SMikkTSpaceContext* pContext, float fvTexcOut[], const int iFace, const int iVert)
+ {
+ MikktData* data = (MikktData*)pContext->m_pUserData;
+ F32* tc = data->tc[iFace * 3 + iVert].mV;
+ fvTexcOut[0] = tc[0];
+ fvTexcOut[1] = tc[1];
+ };
- //sort triangle data by score
- std::sort(triangle_data.begin(), triangle_data.end());
+ ms.m_setTSpaceBasic = [](const SMikkTSpaceContext* pContext, const float fvTangent[], const float fSign, const int iFace, const int iVert)
+ {
+ MikktData* data = (MikktData*)pContext->m_pUserData;
+ S32 i = iFace * 3 + iVert;
+
+ data->t[i].set(fvTangent);
+ data->t[i].mV[3] = fSign;
+ };
- std::vector<U16> new_indices;
+ ms.m_setTSpace = nullptr;
- LLVCacheTriangleData* tri;
+ MikktData data(this);
- //prime pump by adding first triangle to cache;
- tri = &(triangle_data[0]);
- cache.addTriangle(tri);
- new_indices.push_back(tri->mVertex[0]->mIdx);
- new_indices.push_back(tri->mVertex[1]->mIdx);
- new_indices.push_back(tri->mVertex[2]->mIdx);
- tri->complete();
+ SMikkTSpaceContext ctx = { &ms, &data };
- U32 breaks = 0;
- for (U32 i = 1; i < mNumIndices/3; ++i)
- {
- cache.updateScores();
- tri = cache.mBestTriangle;
- if (!tri)
- {
- breaks++;
- for (U32 j = 0; j < triangle_data.size(); ++j)
- {
- if (triangle_data[j].mActive)
- {
- tri = &(triangle_data[j]);
- break;
- }
- }
- }
-
- cache.addTriangle(tri);
- new_indices.push_back(tri->mVertex[0]->mIdx);
- new_indices.push_back(tri->mVertex[1]->mIdx);
- new_indices.push_back(tri->mVertex[2]->mIdx);
- tri->complete();
- }
+ genTangSpaceDefault(&ctx);
- for (U32 i = 0; i < mNumIndices; ++i)
- {
- mIndices[i] = new_indices[i];
- }
+ //re-weld
+ meshopt_Stream mos[] =
+ {
+ { &data.p[0], sizeof(LLVector3), sizeof(LLVector3) },
+ { &data.n[0], sizeof(LLVector3), sizeof(LLVector3) },
+ { &data.t[0], sizeof(LLVector4), sizeof(LLVector4) },
+ { &data.tc[0], sizeof(LLVector2), sizeof(LLVector2) },
+ { data.w.empty() ? nullptr : &data.w[0], sizeof(LLVector4), sizeof(LLVector4) }
+ };
- /*F32 post_acmr = 1.f;
- //measure cache misses from after rebuild
- {
- LLVCacheFIFO test_cache;
- for (U32 i = 0; i < mNumVertices; i++)
- {
- vertex_data[i].mCacheTag = -1;
- }
+ std::vector<U32> remap;
+ remap.resize(data.p.size());
- for (U32 i = 0; i < mNumIndices; ++i)
- {
- test_cache.addVertex(&vertex_data[mIndices[i]]);
- }
-
- post_acmr = (F32) test_cache.mMisses/(mNumIndices/3);
- }*/
+ U32 stream_count = data.w.empty() ? 4 : 5;
- //optimize for pre-TnL cache
-
- //allocate space for new buffer
- S32 num_verts = mNumVertices;
- S32 size = ((num_verts*sizeof(LLVector2)) + 0xF) & ~0xF;
- LLVector4a* pos = (LLVector4a*) ll_aligned_malloc<64>(sizeof(LLVector4a)*2*num_verts+size);
- if (pos == NULL)
- {
- LL_WARNS("LLVOLUME") << "Allocation of positions vector[" << sizeof(LLVector4a) * 2 * num_verts + size << "] failed. " << LL_ENDL;
- return false;
- }
- LLVector4a* norm = pos + num_verts;
- LLVector2* tc = (LLVector2*) (norm + num_verts);
+ U32 vert_count = meshopt_generateVertexRemapMulti(&remap[0], nullptr, data.p.size(), data.p.size(), mos, stream_count);
- LLVector4a* wght = NULL;
- if (mWeights)
- {
- wght = (LLVector4a*)ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts);
- if (wght == NULL)
- {
- ll_aligned_free<64>(pos);
- LL_WARNS("LLVOLUME") << "Allocation of weights[" << sizeof(LLVector4a) * num_verts << "] failed" << LL_ENDL;
- return false;
- }
- }
+ if (vert_count < 65535)
+ {
+ std::vector<U32> indices;
+ indices.resize(mNumIndices);
- LLVector4a* binorm = NULL;
- if (mTangents)
- {
- binorm = (LLVector4a*) ll_aligned_malloc_16(sizeof(LLVector4a)*num_verts);
- if (binorm == NULL)
- {
- ll_aligned_free<64>(pos);
- ll_aligned_free_16(wght);
- LL_WARNS("LLVOLUME") << "Allocation of binormals[" << sizeof(LLVector4a)*num_verts << "] failed" << LL_ENDL;
- return false;
- }
- }
+ //copy results back into volume
+ resizeVertices(vert_count);
- //allocate mapping of old indices to new indices
- std::vector<S32> new_idx;
+ if (!data.w.empty())
+ {
+ allocateWeights(vert_count);
+ }
- try
- {
- new_idx.resize(mNumVertices, -1);
- }
- catch (std::bad_alloc&)
- {
- ll_aligned_free<64>(pos);
- ll_aligned_free_16(wght);
- ll_aligned_free_16(binorm);
- LL_WARNS("LLVOLUME") << "Resize failed: " << mNumVertices << LL_ENDL;
- return false;
- }
+ allocateTangents(mNumVertices);
- S32 cur_idx = 0;
- for (U32 i = 0; i < mNumIndices; ++i)
- {
- U16 idx = mIndices[i];
- if (new_idx[idx] == -1)
- { //this vertex hasn't been added yet
- new_idx[idx] = cur_idx;
+ for (int i = 0; i < mNumIndices; ++i)
+ {
+ U32 src_idx = i;
+ U32 dst_idx = remap[i];
+ mIndices[i] = dst_idx;
- //copy vertex data
- pos[cur_idx] = mPositions[idx];
- norm[cur_idx] = mNormals[idx];
- tc[cur_idx] = mTexCoords[idx];
- if (mWeights)
- {
- wght[cur_idx] = mWeights[idx];
- }
- if (mTangents)
- {
- binorm[cur_idx] = mTangents[idx];
- }
+ mPositions[dst_idx].load3(data.p[src_idx].mV);
+ mNormals[dst_idx].load3(data.n[src_idx].mV);
+ mTexCoords[dst_idx] = data.tc[src_idx];
- cur_idx++;
- }
- }
+ mTangents[dst_idx].loadua(data.t[src_idx].mV);
- for (U32 i = 0; i < mNumIndices; ++i)
- {
- mIndices[i] = new_idx[mIndices[i]];
- }
-
- ll_aligned_free<64>(mPositions);
- // DO NOT free mNormals and mTexCoords as they are part of mPositions buffer
- ll_aligned_free_16(mWeights);
- ll_aligned_free_16(mTangents);
-#if USE_SEPARATE_JOINT_INDICES_AND_WEIGHTS
- ll_aligned_free_16(mJointIndices);
- ll_aligned_free_16(mJustWeights);
- mJustWeights = NULL;
- mJointIndices = NULL; // filled in later as necessary by skinning code for acceleration
-#endif
+ if (mWeights)
+ {
+ mWeights[dst_idx].loadua(data.w[src_idx].mV);
+ }
+ }
+ }
+ else
+ {
+ // blew past the max vertex size limit, use legacy tangent generation which never adds verts
+ createTangents();
+ }
- mPositions = pos;
- mNormals = norm;
- mTexCoords = tc;
- mWeights = wght;
- mTangents = binorm;
+ // put back in normalized coordinate frame
+ LLVector4a inv_scale(1.f/mNormalizedScale.mV[0], 1.f / mNormalizedScale.mV[1], 1.f / mNormalizedScale.mV[2]);
+ LLVector4a scale;
+ scale.load3(mNormalizedScale.mV);
+ scale.getF32ptr()[3] = 1.f;
+
+ for (int i = 0; i < mNumVertices; ++i)
+ {
+ mPositions[i].mul(inv_scale);
+ mNormals[i].mul(scale);
+ mNormals[i].normalize3();
+ F32 w = mTangents[i].getF32ptr()[3];
+ mTangents[i].mul(scale);
+ mTangents[i].normalize3();
+ mTangents[i].getF32ptr()[3] = w;
+ }
+ }
- //std::string result = llformat("ACMR pre/post: %.3f/%.3f -- %d triangles %d breaks", pre_acmr, post_acmr, mNumIndices/3, breaks);
- //LL_INFOS() << result << LL_ENDL;
+ // cache optimize index buffer
+
+ // meshopt needs scratch space, do some pointer shuffling to avoid an extra index buffer copy
+ U16* src_indices = mIndices;
+ mIndices = nullptr;
+ resizeIndices(mNumIndices);
+
+ meshopt_optimizeVertexCache<U16>(mIndices, src_indices, mNumIndices, mNumVertices);
+
+ ll_aligned_free_16(src_indices);
return true;
}
@@ -6442,35 +6451,31 @@ void CalculateTangentArray(U32 vertexCount, const LLVector4a *vertex, const LLVe
void LLVolumeFace::createTangents()
{
- LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME
+ LL_PROFILE_ZONE_SCOPED_CATEGORY_VOLUME;
- if (!mTangents)
- {
- allocateTangents(mNumVertices);
-
- //generate tangents
- //LLVector4a* pos = mPositions;
- //LLVector2* tc = (LLVector2*) mTexCoords;
- LLVector4a* binorm = (LLVector4a*) mTangents;
+ if (!mTangents)
+ {
+ allocateTangents(mNumVertices);
+
+ //generate tangents
+ LLVector4a* ptr = (LLVector4a*)mTangents;
- LLVector4a* end = mTangents+mNumVertices;
- while (binorm < end)
- {
- (*binorm++).clear();
- }
+ LLVector4a* end = mTangents + mNumVertices;
+ while (ptr < end)
+ {
+ (*ptr++).clear();
+ }
- binorm = mTangents;
+ CalculateTangentArray(mNumVertices, mPositions, mNormals, mTexCoords, mNumIndices / 3, mIndices, mTangents);
- CalculateTangentArray(mNumVertices, mPositions, mNormals, mTexCoords, mNumIndices/3, mIndices, mTangents);
+ //normalize normals
+ for (U32 i = 0; i < mNumVertices; i++)
+ {
+ //bump map/planar projection code requires normals to be normalized
+ mNormals[i].normalize3fast();
+ }
+ }
- //normalize tangents
- for (U32 i = 0; i < mNumVertices; i++)
- {
- //binorm[i].normalize3fast();
- //bump map/planar projection code requires normals to be normalized
- mNormals[i].normalize3fast();
- }
- }
}
void LLVolumeFace::resizeVertices(S32 num_verts)
diff --git a/indra/llmath/llvolume.h b/indra/llmath/llvolume.h
index 1509241623..ad6a669531 100644
--- a/indra/llmath/llvolume.h
+++ b/indra/llmath/llvolume.h
@@ -908,7 +908,7 @@ public:
void remap();
void optimize(F32 angle_cutoff = 2.f);
- bool cacheOptimize();
+ bool cacheOptimize(bool gen_tangents = false);
void createOctree(F32 scaler = 0.25f, const LLVector4a& center = LLVector4a(0,0,0), const LLVector4a& size = LLVector4a(0.5f,0.5f,0.5f));
void destroyOctree();
@@ -960,10 +960,6 @@ public:
// indexes for mPositions/mNormals/mTexCoords
U16* mIndices;
- // vertex buffer filled in by LLFace to cache this volume face geometry in vram
- // (declared as a LLPointer to LLRefCount to avoid dependency on LLVertexBuffer)
- mutable LLPointer<LLRefCount> mVertexBuffer;
-
std::vector<S32> mEdge;
//list of skin weights for rigged volumes
@@ -985,6 +981,11 @@ public:
//whether or not face has been cache optimized
BOOL mOptimized;
+ // if this is a mesh asset, scale and translation that were applied
+ // when encoding the source mesh into a unit cube
+ // used for regenerating tangents
+ LLVector3 mNormalizedScale = LLVector3(1,1,1);
+
private:
LLOctreeNode<LLVolumeTriangle, LLVolumeTriangle*>* mOctree;
LLVolumeTriangle* mOctreeTriangles;
@@ -1033,7 +1034,7 @@ public:
void setDirty() { mPathp->setDirty(); mProfilep->setDirty(); }
void regen();
- void genTangents(S32 face);
+ void genTangents(S32 face);
BOOL isConvex() const;
BOOL isCap(S32 face);
@@ -1087,7 +1088,10 @@ public:
void copyVolumeFaces(const LLVolume* volume);
void copyFacesTo(std::vector<LLVolumeFace> &faces) const;
void copyFacesFrom(const std::vector<LLVolumeFace> &faces);
- bool cacheOptimize();
+
+ // use meshoptimizer to optimize index buffer for vertex shader cache
+ // gen_tangents - if true, generate MikkTSpace tangents if needed before optimizing index buffer
+ bool cacheOptimize(bool gen_tangents = false);
private:
void sculptGenerateMapVertices(U16 sculpt_width, U16 sculpt_height, S8 sculpt_components, const U8* sculpt_data, U8 sculpt_type);
diff --git a/indra/llmath/v3color.h b/indra/llmath/v3color.h
index 43a632408c..d925f56e97 100644
--- a/indra/llmath/v3color.h
+++ b/indra/llmath/v3color.h
@@ -33,6 +33,7 @@ class LLVector4;
#include "llerror.h"
#include "llmath.h"
#include "llsd.h"
+#include "v3math.h" // needed for linearColor3v implemtation below
#include <string.h>
// LLColor3 = |r g b|
@@ -87,6 +88,16 @@ public:
const LLColor3& set(F32 x, F32 y, F32 z); // Sets LLColor3 to (x, y, z)
const LLColor3& set(const LLColor3 &vec); // Sets LLColor3 to vec
const LLColor3& set(const F32 *vec); // Sets LLColor3 to vec
+
+ // set from a vector of unknown type and size
+ // may leave some data unmodified
+ template<typename T>
+ const LLColor3& set(const std::vector<T>& v);
+
+ // write to a vector of unknown type and size
+ // maye leave some data unmodified
+ template<typename T>
+ void write(std::vector<T>& v) const;
F32 magVec() const; // deprecated
F32 magVecSquared() const; // deprecated
@@ -484,13 +495,45 @@ inline const LLColor3 srgbColor3(const LLColor3 &a) {
return srgbColor;
}
-inline const LLColor3 linearColor3(const LLColor3 &a) {
+inline const LLColor3 linearColor3p(const F32* v) {
LLColor3 linearColor;
- linearColor.mV[0] = sRGBtoLinear(a.mV[0]);
- linearColor.mV[1] = sRGBtoLinear(a.mV[1]);
- linearColor.mV[2] = sRGBtoLinear(a.mV[2]);
+ linearColor.mV[0] = sRGBtoLinear(v[0]);
+ linearColor.mV[1] = sRGBtoLinear(v[1]);
+ linearColor.mV[2] = sRGBtoLinear(v[2]);
return linearColor;
}
+template<class T>
+inline const LLColor3 linearColor3(const T& a) {
+ return linearColor3p(a.mV);
+}
+
+template<class T>
+inline const LLVector3 linearColor3v(const T& a) {
+ return LLVector3(linearColor3p(a.mV).mV);
+}
+
+template<typename T>
+const LLColor3& LLColor3::set(const std::vector<T>& v)
+{
+ for (S32 i = 0; i < llmin((S32)v.size(), 3); ++i)
+ {
+ mV[i] = v[i];
+ }
+
+ return *this;
+}
+
+// write to a vector of unknown type and size
+// maye leave some data unmodified
+template<typename T>
+void LLColor3::write(std::vector<T>& v) const
+{
+ for (int i = 0; i < llmin((S32)v.size(), 3); ++i)
+ {
+ v[i] = mV[i];
+ }
+}
+
#endif
diff --git a/indra/llmath/v4color.h b/indra/llmath/v4color.h
index 175edf1471..daa61594fb 100644
--- a/indra/llmath/v4color.h
+++ b/indra/llmath/v4color.h
@@ -88,8 +88,18 @@ class LLColor4
const LLColor4& set(const LLColor3 &vec); // Sets LLColor4 to LLColor3 vec (no change in alpha)
const LLColor4& set(const LLColor3 &vec, F32 a); // Sets LLColor4 to LLColor3 vec, with alpha specified
const LLColor4& set(const F32 *vec); // Sets LLColor4 to vec
- const LLColor4& set(const LLColor4U& color4u); // Sets LLColor4 to color4u, rescaled.
+ const LLColor4& set(const F64 *vec); // Sets LLColor4 to (double)vec
+ const LLColor4& set(const LLColor4U& color4u); // Sets LLColor4 to color4u, rescaled.
+ // set from a vector of unknown type and size
+ // may leave some data unmodified
+ template<typename T>
+ const LLColor4& set(const std::vector<T>& v);
+
+ // write to a vector of unknown type and size
+ // maye leave some data unmodified
+ template<typename T>
+ void write(std::vector<T>& v) const;
const LLColor4& setAlpha(F32 a);
@@ -334,6 +344,15 @@ inline const LLColor4& LLColor4::set(const F32 *vec)
return (*this);
}
+inline const LLColor4& LLColor4::set(const F64 *vec)
+{
+ mV[VX] = static_cast<F32>(vec[VX]);
+ mV[VY] = static_cast<F32>(vec[VY]);
+ mV[VZ] = static_cast<F32>(vec[VZ]);
+ mV[VW] = static_cast<F32>(vec[VW]);
+ return (*this);
+}
+
// deprecated
inline const LLColor4& LLColor4::setVec(F32 x, F32 y, F32 z)
{
@@ -680,5 +699,25 @@ inline const LLColor4 linearColor4(const LLColor4 &a)
return linearColor;
}
+template<typename T>
+const LLColor4& LLColor4::set(const std::vector<T>& v)
+{
+ for (S32 i = 0; i < llmin((S32)v.size(), 4); ++i)
+ {
+ mV[i] = v[i];
+ }
+
+ return *this;
+}
+
+template<typename T>
+void LLColor4::write(std::vector<T>& v) const
+{
+ for (int i = 0; i < llmin((S32)v.size(), 4); ++i)
+ {
+ v[i] = mV[i];
+ }
+}
+
#endif
generated by cgit v1.2.3 (git 2.25.1) at 2024-11-10 05:25:36 +0000