C++ LLVector4a::normalize3fast方法代码示例

BOOL LLFace::calcPixelArea(F32& cos_angle_to_view_dir, F32& radius)
{
	//VECTORIZE THIS
	//get area of circle around face
	LLVector4a center;
	center.load3(getPositionAgent().mV);
	LLVector4a size;
	size.setSub(mExtents[1], mExtents[0]);
	size.mul(0.5f);

	LLViewerCamera* camera = LLViewerCamera::getInstance();

	F32 size_squared = size.dot3(size).getF32();
	LLVector4a lookAt;
	LLVector4a t;
	t.load3(camera->getOrigin().mV);
	lookAt.setSub(center, t);
	F32 dist = lookAt.getLength3().getF32();
	dist = llmax(dist-size.getLength3().getF32(), 0.f);
	lookAt.normalize3fast() ;	

	//get area of circle around node
	F32 app_angle = atanf((F32) sqrt(size_squared) / dist);
	radius = app_angle*LLDrawable::sCurPixelAngle;
	mPixelArea = radius*radius * 3.14159f;
	LLVector4a x_axis;
	x_axis.load3(camera->getXAxis().mV);
	cos_angle_to_view_dir = lookAt.dot3(x_axis).getF32();

	if(dist < mBoundingSphereRadius) //camera is very close
	{
		cos_angle_to_view_dir = 1.0f;
		mImportanceToCamera = 1.0f;
	}
	else
	{
		mImportanceToCamera = LLFace::calcImportanceToCamera(cos_angle_to_view_dir, dist);
	}

	return true;
}

void LLVOPartGroup::getGeometry(const LLViewerPart& part,
								LLStrider<LLVector4a>& verticesp)
{
	if (part.mFlags & LLPartData::LL_PART_RIBBON_MASK)
	{
		LLVector4a axis, pos, paxis, ppos;
		F32 scale, pscale;

		pos.load3(part.mPosAgent.mV);
		axis.load3(part.mAxis.mV);
		scale = part.mScale.mV[0];
		
		if (part.mParent)
		{
			ppos.load3(part.mParent->mPosAgent.mV);
			paxis.load3(part.mParent->mAxis.mV);
			pscale = part.mParent->mScale.mV[0];
		}
		else
		{ //use source object as position
			
			if (part.mPartSourcep->mSourceObjectp.notNull())
			{
				LLVector3 v = LLVector3(0,0,1);
				v *= part.mPartSourcep->mSourceObjectp->getRenderRotation();
				paxis.load3(v.mV);
				ppos.load3(part.mPartSourcep->mPosAgent.mV);
				pscale = part.mStartScale.mV[0];
			}
			else
			{ //no source object, no parent, nothing to draw
				ppos = pos;
				pscale = scale;
				paxis = axis;
			}
		}

		LLVector4a p0, p1, p2, p3;

		scale *= 0.5f;
		pscale *= 0.5f;

		axis.mul(scale);
		paxis.mul(pscale);

		p0.setAdd(pos, axis);
		p1.setSub(pos,axis);
		p2.setAdd(ppos, paxis);
		p3.setSub(ppos, paxis);

		(*verticesp++) = p2;
		(*verticesp++) = p3;
		(*verticesp++) = p0;
		(*verticesp++) = p1;
	}
	else
	{
		LLVector4a part_pos_agent;
		part_pos_agent.load3(part.mPosAgent.mV);
		LLVector4a camera_agent;
	camera_agent.load3(getCameraPosition().mV); 
	LLVector4a at;
	at.setSub(part_pos_agent, camera_agent);
	LLVector4a up(0, 0, 1);
	LLVector4a right;

	right.setCross3(at, up);
	right.normalize3fast();
	up.setCross3(right, at);
	up.normalize3fast();

	if (part.mFlags & LLPartData::LL_PART_FOLLOW_VELOCITY_MASK)
	{
		LLVector4a normvel;
		normvel.load3(part.mVelocity.mV);
		normvel.normalize3fast();
		LLVector2 up_fracs;
		up_fracs.mV[0] = normvel.dot3(right).getF32();
		up_fracs.mV[1] = normvel.dot3(up).getF32();
		up_fracs.normalize();
		LLVector4a new_up;
		LLVector4a new_right;

		//new_up = up_fracs.mV[0] * right + up_fracs.mV[1]*up;
		LLVector4a t = right;
		t.mul(up_fracs.mV[0]);
		new_up = up;
		new_up.mul(up_fracs.mV[1]);
		new_up.add(t);

		//new_right = up_fracs.mV[1] * right - up_fracs.mV[0]*up;
		t = right;
		t.mul(up_fracs.mV[1]);
		new_right = up;
		new_right.mul(up_fracs.mV[0]);
		t.sub(new_right);

		up = new_up;
		right = t;
		up.normalize3fast();
//.........这里部分代码省略.........

void LLVOClouds::getGeometry(S32 idx,
								LLStrider<LLVector4a>& verticesp,
								LLStrider<LLVector3>& normalsp, 
								LLStrider<LLVector2>& texcoordsp,
								LLStrider<LLColor4U>& colorsp, 
								LLStrider<U16>& indicesp)
{

	if (idx >= mCloudGroupp->getNumPuffs())
	{
		return;
	}

	LLDrawable* drawable = mDrawable;
	LLFace *facep = drawable->getFace(idx);

	if (!facep->hasGeometry())
	{
		return;
	}
	

	const LLCloudPuff &puff = mCloudGroupp->getPuff(idx);

	LLColor4 float_color(LLColor3(gSky.getSunDiffuseColor() + gSky.getSunAmbientColor()),puff.getAlpha());
	LLColor4U color;
	color.setVec(float_color);
	facep->setFaceColor(float_color);
	
	U32 vert_offset = facep->getGeomIndex();
		
	LLVector4a part_pos_agent;
	part_pos_agent.load3(facep->mCenterLocal.mV);	
	LLVector4a at;
	at.load3(LLViewerCamera::getInstance()->getAtAxis().mV);
	LLVector4a up(0, 0, 1);
	LLVector4a right;

	right.setCross3(at, up);
	right.normalize3fast();
	up.setCross3(right, at);
	up.normalize3fast();
	right.mul(0.5f*CLOUD_PUFF_WIDTH);
	up.mul(0.5f*CLOUD_PUFF_HEIGHT);
		

	LLVector3 normal(0.f,0.f,-1.f);

	//HACK -- the verticesp->mV[3] = 0.f here are to set the texture index to 0 (particles don't use texture batching, maybe they should)
	// this works because there is actually a 4th float stored after the vertex position which is used as a texture index
	// also, somebody please VECTORIZE THIS

	LLVector4a ppapu;
	LLVector4a ppamu;
	
	ppapu.setAdd(part_pos_agent, up);
	ppamu.setSub(part_pos_agent, up);
	
	verticesp->setSub(ppapu, right);
	(*verticesp++).getF32ptr()[3] = 0.f;
	verticesp->setSub(ppamu, right);
	(*verticesp++).getF32ptr()[3] = 0.f;
	verticesp->setAdd(ppapu, right);
	(*verticesp++).getF32ptr()[3] = 0.f;
	verticesp->setAdd(ppamu, right);
	(*verticesp++).getF32ptr()[3] = 0.f;

	// *verticesp++ = puff_pos_agent - right + up;
	// *verticesp++ = puff_pos_agent - right - up;
	// *verticesp++ = puff_pos_agent + right + up;
	// *verticesp++ = puff_pos_agent + right - up;

	*colorsp++ = color;
	*colorsp++ = color;
	*colorsp++ = color;
	*colorsp++ = color;

	*texcoordsp++ = LLVector2(0.f, 1.f);
	*texcoordsp++ = LLVector2(0.f, 0.f);
	*texcoordsp++ = LLVector2(1.f, 1.f);
	*texcoordsp++ = LLVector2(1.f, 0.f);

	*normalsp++   = normal;
	*normalsp++   = normal;
	*normalsp++   = normal;
	*normalsp++   = normal;

	*indicesp++ = vert_offset + 0;
	*indicesp++ = vert_offset + 1;
	*indicesp++ = vert_offset + 2;

	*indicesp++ = vert_offset + 1;
	*indicesp++ = vert_offset + 3;
	*indicesp++ = vert_offset + 2;
}

void LLPolyMorphTarget::apply( ESex avatar_sex )
{
	if (!mMorphData || mNumMorphMasksPending > 0)
	{
		return;
	}

	LL_RECORD_BLOCK_TIME(FTM_APPLY_MORPH_TARGET);

	mLastSex = avatar_sex;

	// Check for NaN condition (NaN is detected if a variable doesn't equal itself.
	if (mCurWeight != mCurWeight)
	{
		mCurWeight = 0.0;
	}
	if (mLastWeight != mLastWeight)
	{
		mLastWeight = mCurWeight+.001;
	}

	// perform differential update of morph
	F32 delta_weight = ( getSex() & avatar_sex ) ? (mCurWeight - mLastWeight) : (getDefaultWeight() - mLastWeight);
	// store last weight
	mLastWeight += delta_weight;

	if (delta_weight != 0.f)
	{
		llassert(!mMesh->isLOD());
		LLVector4a *coords = mMesh->getWritableCoords();

		LLVector4a *scaled_normals = mMesh->getScaledNormals();
		LLVector4a *normals = mMesh->getWritableNormals();

		LLVector4a *scaled_binormals = mMesh->getScaledBinormals();
		LLVector4a *binormals = mMesh->getWritableBinormals();

		LLVector4a *clothing_weights = mMesh->getWritableClothingWeights();
		LLVector2 *tex_coords = mMesh->getWritableTexCoords();

		F32 *maskWeightArray = (mVertMask) ? mVertMask->getMorphMaskWeights() : NULL;

		for(U32 vert_index_morph = 0; vert_index_morph < mMorphData->mNumIndices; vert_index_morph++)
		{
			S32 vert_index_mesh = mMorphData->mVertexIndices[vert_index_morph];

			F32 maskWeight = 1.f;
			if (maskWeightArray)
			{
				maskWeight = maskWeightArray[vert_index_morph];
			}


			LLVector4a pos = mMorphData->mCoords[vert_index_morph];
			pos.mul(delta_weight*maskWeight);
			coords[vert_index_mesh].add(pos);

			if (getInfo()->mIsClothingMorph && clothing_weights)
			{
				LLVector4a clothing_offset = mMorphData->mCoords[vert_index_morph];
				clothing_offset.mul(delta_weight * maskWeight);
				LLVector4a* clothing_weight = &clothing_weights[vert_index_mesh];
				clothing_weight->add(clothing_offset);
				clothing_weight->getF32ptr()[VW] = maskWeight;
			}

			// calculate new normals based on half angles
			LLVector4a norm = mMorphData->mNormals[vert_index_morph];
			norm.mul(delta_weight*maskWeight*NORMAL_SOFTEN_FACTOR);
			scaled_normals[vert_index_mesh].add(norm);
			norm = scaled_normals[vert_index_mesh];

			// guard against degenerate input data before we create NaNs below!
			//
			norm.normalize3fast();
			normals[vert_index_mesh] = norm;

			// calculate new binormals
			LLVector4a binorm = mMorphData->mBinormals[vert_index_morph];

			// guard against degenerate input data before we create NaNs below!
			//
			if (!binorm.isFinite3() || (binorm.dot3(binorm).getF32() <= F_APPROXIMATELY_ZERO))
			{
				binorm.set(1,0,0,1);
			}

			binorm.mul(delta_weight*maskWeight*NORMAL_SOFTEN_FACTOR);
			scaled_binormals[vert_index_mesh].add(binorm);
			LLVector4a tangent;
			tangent.setCross3(scaled_binormals[vert_index_mesh], norm);
			LLVector4a& normalized_binormal = binormals[vert_index_mesh];

			normalized_binormal.setCross3(norm, tangent); 
			normalized_binormal.normalize3fast();
			
			tex_coords[vert_index_mesh] += mMorphData->mTexCoords[vert_index_morph] * delta_weight * maskWeight;
		}

		// now apply volume changes
//.........这里部分代码省略.........

void LLVOPartGroup::getGeometry(S32 idx,
								LLStrider<LLVector4a>& verticesp,
								LLStrider<LLVector3>& normalsp, 
								LLStrider<LLVector2>& texcoordsp,
								LLStrider<LLColor4U>& colorsp, 
								LLStrider<U16>& indicesp)
{
	if (idx >= (S32) mViewerPartGroupp->mParticles.size())
	{
		return;
	}

	const LLViewerPart &part = *((LLViewerPart*) (mViewerPartGroupp->mParticles[idx]));

	U32 vert_offset = mDrawable->getFace(idx)->getGeomIndex();

	
	LLVector4a part_pos_agent;
	part_pos_agent.load3(part.mPosAgent.mV);
	LLVector4a camera_agent;
	camera_agent.load3(getCameraPosition().mV); 
	LLVector4a at;
	at.setSub(part_pos_agent, camera_agent);
	LLVector4a up(0, 0, 1);
	LLVector4a right;

	right.setCross3(at, up);
	right.normalize3fast();
	up.setCross3(right, at);
	up.normalize3fast();

	if (part.mFlags & LLPartData::LL_PART_FOLLOW_VELOCITY_MASK)
	{
		LLVector4a normvel;
		normvel.load3(part.mVelocity.mV);
		normvel.normalize3fast();
		LLVector2 up_fracs;
		up_fracs.mV[0] = normvel.dot3(right).getF32();
		up_fracs.mV[1] = normvel.dot3(up).getF32();
		up_fracs.normalize();
		LLVector4a new_up;
		LLVector4a new_right;

		//new_up = up_fracs.mV[0] * right + up_fracs.mV[1]*up;
		LLVector4a t = right;
		t.mul(up_fracs.mV[0]);
		new_up = up;
		new_up.mul(up_fracs.mV[1]);
		new_up.add(t);

		//new_right = up_fracs.mV[1] * right - up_fracs.mV[0]*up;
		t = right;
		t.mul(up_fracs.mV[1]);
		new_right = up;
		new_right.mul(up_fracs.mV[0]);
		t.sub(new_right);

		up = new_up;
		right = t;
		up.normalize3fast();
		right.normalize3fast();
	}

	right.mul(0.5f*part.mScale.mV[0]);
	up.mul(0.5f*part.mScale.mV[1]);


	LLVector3 normal = -LLViewerCamera::getInstance()->getXAxis();

	//HACK -- the verticesp->mV[3] = 0.f here are to set the texture index to 0 (particles don't use texture batching, maybe they should)
	// this works because there is actually a 4th float stored after the vertex position which is used as a texture index
	// also, somebody please VECTORIZE THIS

	LLVector4a ppapu;
	LLVector4a ppamu;

	ppapu.setAdd(part_pos_agent, up);
	ppamu.setSub(part_pos_agent, up);

	verticesp->setSub(ppapu, right);
	(*verticesp++).getF32ptr()[3] = 0.f;
	verticesp->setSub(ppamu, right);
	(*verticesp++).getF32ptr()[3] = 0.f;
	verticesp->setAdd(ppapu, right);
	(*verticesp++).getF32ptr()[3] = 0.f;
	verticesp->setAdd(ppamu, right);
	(*verticesp++).getF32ptr()[3] = 0.f;

	//*verticesp++ = part_pos_agent + up - right;
	//*verticesp++ = part_pos_agent - up - right;
	//*verticesp++ = part_pos_agent + up + right;
	//*verticesp++ = part_pos_agent - up + right;

	*colorsp++ = part.mColor;
	*colorsp++ = part.mColor;
	*colorsp++ = part.mColor;
	*colorsp++ = part.mColor;

	*texcoordsp++ = LLVector2(0.f, 1.f);
	*texcoordsp++ = LLVector2(0.f, 0.f);
//.........这里部分代码省略.........

//.........这里部分代码省略.........
				{
					bump_tc.push_back(tc);
				}
			}

			//mVertexBuffer->setBuffer(0);


			if (do_bump)
			{
				mVertexBuffer->getTexCoord1Strider(tex_coords2, mGeomIndex);
		
				for (S32 i = 0; i < num_vertices; i++)
				{
					LLVector4a tangent;
					tangent.setCross3(vf.mBinormals[i], vf.mNormals[i]);

					LLMatrix4a tangent_to_object;
					tangent_to_object.setRows(tangent, vf.mBinormals[i], vf.mNormals[i]);
					LLVector4a t;
					tangent_to_object.rotate(binormal_dir, t);
					LLVector4a binormal;
					mat_normal.rotate(t, binormal);
						
					//VECTORIZE THIS
					if (mDrawablep->isActive())
					{
						LLVector3 t;
						t.set(binormal.getF32ptr());
						t *= bump_quat;
						binormal.load3(t.mV);
					}

					binormal.normalize3fast();
					LLVector2 tc = bump_tc[i];
					tc += LLVector2( bump_s_primary_light_ray.dot3(tangent).getF32(), bump_t_primary_light_ray.dot3(binormal).getF32() );
					
					*tex_coords2++ = tc;
				}

				//mVertexBuffer->setBuffer(0);
			}
		}
	}

	if (rebuild_pos)
	{
		llassert(num_vertices > 0);
		mVertexBuffer->getVertexStrider(vertices, mGeomIndex);
		LLMatrix4a mat_vert;
		mat_vert.loadu(mat_vert_in);
		
		LLVector4a* src = vf.mPositions;
		LLVector4a position;
		for (S32 i = 0; i < num_vertices; i++)
		{
			mat_vert.affineTransform(src[i], position);
			vertices[i].set(position.getF32ptr());
		}
			
			
		//mVertexBuffer->setBuffer(0);
	}
		
	if (rebuild_normal)
	{