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();
		right.normalize3fast();
	}

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


		//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;
	}
}
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]));

	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;

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

	if (!(part.mFlags & LLPartData::LL_PART_EMISSIVE_MASK))
	{ //not fullbright, needs normal
		*normalsp++   = normal;
		*normalsp++   = normal;
		*normalsp++   = normal;
		*normalsp++   = normal;
	}
}