void 
nb_kernel310_ppc_altivec  (int *             p_nri,
                       int               iinr[],
                       int               jindex[],
                       int               jjnr[],
                       int               shift[],
                       float             shiftvec[],
                       float             fshift[],
                       int               gid[],
                       float             pos[],
                       float             faction[],
                       float             charge[],
                       float *           p_facel,
                       float *           p_krf,
                       float *           p_crf,
                       float             Vc[],
                       int               type[],
                       int *             p_ntype,
                       float             vdwparam[],
                       float             Vvdw[],
                       float *           p_tabscale,
                       float             VFtab[],
                       float             invsqrta[],
                       float             dvda[],
                       float *           p_gbtabscale,
                       float             GBtab[],
                       int *             p_nthreads,
                       int *             count,
                       void *            mtx,
                       int *             outeriter,
                       int *             inneriter,
					   float *           work)
{
	vector float ix,iy,iz,shvec;
	vector float vfacel,tsc,fs,fs2,nul;
	vector float dx,dy,dz;
	vector float Vvdwtot,vctot,qq,iq,c6,c12,VVc,FFc;
	vector float fix,fiy,fiz;
	vector float tmp1,tmp2,tmp3,tmp4;
	vector float rinv,r,rinvsq,rsq,rinvsix,Vvdw6,Vvdw12;

	int n,k,ii,is3,ii3,ntiA,nj0,nj1;
	int jnra,jnrb,jnrc,jnrd;
	int j3a,j3b,j3c,j3d;
	int nri, ntype, nouter, ninner;
	int tja,tjb,tjc,tjd;
#ifdef GMX_THREADS
	int nn0, nn1;
#endif

    nouter   = 0;
    ninner   = 0;
    nri      = *p_nri;
    ntype    = *p_ntype;
	nul=vec_zero();
	vfacel=load_float_and_splat(p_facel);
	tsc=load_float_and_splat(p_tabscale);

#ifdef GMX_THREADS
    nthreads = *p_nthreads;
	do {
		gmx_thread_mutex_lock((gmx_thread_mutex_t *)mtx);
		nn0              = *count;
		nn1              = nn0+(nri-nn0)/(2*nthreads)+3;
		*count           = nn1;
		gmx_thread_mutex_unlock((gmx_thread_mutex_t *)mtx);
		if(nn1>nri) nn1=nri;
		for(n=nn0; (n<nn1); n++) {
#if 0
		} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without gmx_threads */
		for(n=0;n<nri;n++) {
#endif  
			is3        = 3*shift[n];
			shvec      = load_xyz(shiftvec+is3);
			ii         = iinr[n];
			ii3        = 3*ii;
			ix         = load_xyz(pos+ii3);
			Vvdwtot     = nul;
			vctot      = nul;
			fix        = nul;
			fiy        = nul;
			fiz        = nul;
			ix         = vec_add(ix,shvec);
			nj0        = jindex[n];
			nj1        = jindex[n+1];
			splat_xyz_to_vectors(ix,&ix,&iy,&iz);
			ntiA       = 2*ntype*type[ii];
			iq        = vec_madd(load_float_and_splat(charge+ii),vfacel,nul);

			for(k=nj0; k<(nj1-3); k+=4) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				jnrc            = jjnr[k+2];
				jnrd            = jjnr[k+3];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				j3c             = 3*jnrc;
				j3d             = 3*jnrd;
				transpose_4_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),
								 load_xyz(pos+j3c),
								 load_xyz(pos+j3d),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				rinv            = do_invsqrt(rsq);
				rinvsq          = vec_madd(rinv,rinv,nul);
				r               = vec_madd(rinv,rsq,nul);
				rinvsix         = vec_madd(rinvsq,rinvsq,nul);
				rinvsix         = vec_madd(rinvsix,rinvsq,nul);
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				tjc             = ntiA+2*type[jnrc];
				tjd             = ntiA+2*type[jnrd];
				qq = vec_madd(load_4_float(charge+jnra,charge+jnrb,
										   charge+jnrc,charge+jnrd),iq,nul);
				load_4_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,vdwparam+tjd,&c6,&c12);
				do_4_ctable_coul(VFtab,vec_madd(r,tsc,nul),&VVc,&FFc);
				fs2             = vec_madd(qq,FFc,nul);   /* fijC */
				vctot           = vec_madd(qq,VVc,vctot);
				Vvdw6            = vec_madd(c6,rinvsix,nul);
				Vvdw12           = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),
										   nul);
				fs              = vec_madd(vec_twelve(),Vvdw12,nul);
				fs              = vec_nmsub(vec_six(),Vvdw6,fs);
				fs              = vec_madd(fs,rinv,nul);
				Vvdwtot          = vec_add(Vvdwtot,Vvdw12);
				fs              = vec_nmsub(fs2,tsc,fs);
				fs              = vec_madd(fs,rinv,nul);
				Vvdwtot          = vec_sub(Vvdwtot,Vvdw6);
				fix             = vec_madd(fs,dx,fix); /* +=fx */
				fiy             = vec_madd(fs,dy,fiy); /* +=fy */
				fiz             = vec_madd(fs,dz,fiz); /* +=fz */
				dx              = vec_nmsub(dx,fs,nul); /* -fx */
				dy              = vec_nmsub(dy,fs,nul); /* -fy */
				dz              = vec_nmsub(dz,fs,nul); /* -fz */
				transpose_3_to_4(dx,dy,dz,&tmp1,&tmp2,&tmp3,&tmp4);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);
				add_xyz_to_mem(faction+j3c,tmp3);
				add_xyz_to_mem(faction+j3d,tmp4);
			}
			if(k<(nj1-1)) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				transpose_2_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				zero_highest_2_elements_in_vector(&rsq);
				rinv            = do_invsqrt(rsq);
				zero_highest_2_elements_in_vector(&rinv);
				rinvsq          = vec_madd(rinv,rinv,nul);
				r               = vec_madd(rinv,rsq,nul);
				rinvsix         = vec_madd(rinvsq,rinvsq,nul);
				rinvsix         = vec_madd(rinvsix,rinvsq,nul);
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				qq = vec_madd(load_2_float(charge+jnra,charge+jnrb),iq,nul);
				load_2_pair(vdwparam+tja,vdwparam+tjb,&c6,&c12);
				do_2_ctable_coul(VFtab,vec_madd(r,tsc,nul),&VVc,&FFc);
				fs2             = vec_madd(qq,FFc,nul);   /* fijC */
				vctot           = vec_madd(qq,VVc,vctot);
				Vvdw6            = vec_madd(c6,rinvsix,nul);
				Vvdw12           = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),
										   nul);
				fs              = vec_madd(vec_twelve(),Vvdw12,nul);
				fs              = vec_nmsub(vec_six(),Vvdw6,fs);
				Vvdwtot          = vec_add(Vvdwtot,Vvdw12);
				fs              = vec_madd(fs,rinv,nul);
				fs              = vec_nmsub(fs2,tsc,fs);
				fs              = vec_madd(fs,rinv,nul);
				Vvdwtot          = vec_sub(Vvdwtot,Vvdw6);
				fix             = vec_madd(fs,dx,fix); /* +=fx */
				fiy             = vec_madd(fs,dy,fiy); /* +=fy */
				fiz             = vec_madd(fs,dz,fiz); /* +=fz */
				dx              = vec_nmsub(dx,fs,nul); /* -fx */
				dy              = vec_nmsub(dy,fs,nul); /* -fy */
				dz              = vec_nmsub(dz,fs,nul); /* -fz */
				transpose_3_to_2(dx,dy,dz,&tmp1,&tmp2);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);
				k              += 2;
			}
			if((nj1-nj0) & 0x1) {
				jnra            = jjnr[k];
				j3a             = 3*jnra;
				transpose_1_to_3(load_xyz(pos+j3a),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				zero_highest_3_elements_in_vector(&rsq);
				rinv            = do_invsqrt(rsq);
				zero_highest_3_elements_in_vector(&rinv);
				rinvsq          = vec_madd(rinv,rinv,nul);
				r               = vec_madd(rinv,rsq,nul);
				rinvsix         = vec_madd(rinvsq,rinvsq,nul);
				rinvsix         = vec_madd(rinvsix,rinvsq,nul);
				tja             = ntiA+2*type[jnra];
				qq = vec_madd(load_1_float(charge+jnra),iq,nul);
				load_1_pair(vdwparam+tja,&c6,&c12);
				do_1_ctable_coul(VFtab,vec_madd(r,tsc,nul),&VVc,&FFc);
				fs2             = vec_madd(qq,FFc,nul);   /* fijC */
				vctot           = vec_madd(qq,VVc,vctot);
				Vvdw6            = vec_madd(c6,rinvsix,nul);
				Vvdw12           = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),
										   nul);
				fs              = vec_madd(vec_twelve(),Vvdw12,nul);
				fs              = vec_nmsub(vec_six(),Vvdw6,fs);
				fs              = vec_madd(fs,rinv,nul);
				Vvdwtot          = vec_add(Vvdwtot,Vvdw12);
				fs              = vec_nmsub(fs2,tsc,fs);
				fs              = vec_madd(fs,rinv,nul);
				Vvdwtot          = vec_sub(Vvdwtot,Vvdw6);
				fix             = vec_madd(fs,dx,fix); /* +=fx */
				fiy             = vec_madd(fs,dy,fiy); /* +=fy */
				fiz             = vec_madd(fs,dz,fiz); /* +=fz */
				dx              = vec_nmsub(dx,fs,nul); /* -fx */
				dy              = vec_nmsub(dy,fs,nul); /* -fy */
				dz              = vec_nmsub(dz,fs,nul); /* -fz */
				transpose_3_to_1(dx,dy,dz,&tmp1);
				add_xyz_to_mem(faction+j3a,tmp1);
			}
			/* update outer data */
			transpose_3_to_4(fix,fiy,fiz,&tmp1,&tmp2,&tmp3,&tmp4);
			tmp1 = vec_add(tmp1,tmp3);
			tmp2 = vec_add(tmp2,tmp4);
			tmp1 = vec_add(tmp1,tmp2);

			add_xyz_to_mem(faction+ii3,tmp1);
			add_xyz_to_mem(fshift+is3,tmp1);

			add_vector_to_float(Vc+gid[n],vctot);
			add_vector_to_float(Vvdw+gid[n],Vvdwtot);
			ninner += nj1 - nj0;
		}
#ifdef GMX_THREADS
		nouter += nn1 - nn0;
	} while (nn1<nri);
#else
	nouter = nri;
#endif
	*outeriter = nouter;
	*inneriter = ninner;
}
void 
nb_kernel010nf_ppc_altivec(int *             p_nri,
                       int               iinr[],
                       int               jindex[],
                       int               jjnr[],
                       int               shift[],
                       float             shiftvec[],
                       float             fshift[],
                       int               gid[],
                       float             pos[],
                       float             faction[],
                       float             charge[],
                       float *           p_facel,
                       float *           p_krf,
                       float *           p_crf,
                       float             Vc[],
                       int               type[],
                       int *             p_ntype,
                       float             vdwparam[],
                       float             Vvdw[],
                       float *           p_tabscale,
                       float             VFtab[],
                       float             invsqrta[],
                       float             dvda[],
                       float *           p_gbtabscale,
                       float             GBtab[],
                       int *             p_nthreads,
                       int *             count,
                       void *            mtx,
                       int *             outeriter,
                       int *             inneriter,
					   float *           work)
{
	vector float ix,iy,iz,shvec;
	vector float nul;
	vector float dx,dy,dz;
	vector float Vvdwtot,c6,c12;
	vector float rinvsq,rsq,rinvsix;
  
	int n,k,ii,is3,ii3,nj0,nj1;
	int jnra,jnrb,jnrc,jnrd;
	int j3a,j3b,j3c,j3d;
	int nri, ntype, nouter, ninner;
	int ntiA,tja,tjb,tjc,tjd;
#ifdef GMX_THREAD_SHM_FDECOMP
	int nn0, nn1;
#endif
  
    nouter   = 0;
    ninner   = 0;
    nri      = *p_nri;
    ntype    = *p_ntype;
	nul=vec_zero();

#ifdef GMX_THREAD_SHM_FDECOMP
    nthreads = *p_nthreads;
	do {
		tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
		nn0              = *count;
		nn1              = nn0+(nri-nn0)/(2*nthreads)+3;
		*count           = nn1;
		tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
		if(nn1>nri) nn1=nri;
		for(n=nn0; (n<nn1); n++) {
#if 0
		} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without tMPI_Threads */
		for(n=0;n<nri;n++) {
#endif
			is3        = 3*shift[n];
			shvec      = load_xyz(shiftvec+is3);
			ii         = iinr[n];
			ii3        = 3*ii;
			ix         = load_xyz(pos+ii3);
			Vvdwtot     = nul;
			ix         = vec_add(ix,shvec);    
			nj0        = jindex[n];
			nj1        = jindex[n+1];
			splat_xyz_to_vectors(ix,&ix,&iy,&iz);
			ntiA       = 2*ntype*type[ii];
			for(k=nj0; k<(nj1-3); k+=4) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				jnrc            = jjnr[k+2];
				jnrd            = jjnr[k+3];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				j3c             = 3*jnrc;
				j3d             = 3*jnrd;
				transpose_4_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),
								 load_xyz(pos+j3c),
								 load_xyz(pos+j3d),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				rinvsq          = do_recip(rsq);
				rinvsix         = vec_madd(rinvsq,rinvsq,nul);
				rinvsix         = vec_madd(rinvsix,rinvsq,nul);
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				tjc             = ntiA+2*type[jnrc];
				tjd             = ntiA+2*type[jnrd];
				load_4_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,vdwparam+tjd,&c6,&c12);
				Vvdwtot          = vec_nmsub(c6,rinvsix,Vvdwtot);
				Vvdwtot          = vec_madd(c12,
										   vec_madd(rinvsix,rinvsix,nul),
										   Vvdwtot);
			}
			if(k<(nj1-1)) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				transpose_2_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				rinvsq          = do_recip(rsq);
				zero_highest_2_elements_in_vector(&rinvsq);
				rinvsix         = vec_madd(rinvsq,rinvsq,nul);
				rinvsix         = vec_madd(rinvsix,rinvsq,nul);
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				load_2_pair(vdwparam+tja,vdwparam+tjb,&c6,&c12);
				Vvdwtot          = vec_nmsub(c6,rinvsix,Vvdwtot);
				Vvdwtot          = vec_madd(c12,
										   vec_madd(rinvsix,rinvsix,nul),
										   Vvdwtot);
				k              += 2;
			}
			if((nj1-nj0) & 0x1) {
				jnra            = jjnr[k];
				j3a             = 3*jnra;
				transpose_1_to_3(load_xyz(pos+j3a),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				rinvsq          = do_recip(rsq);
				zero_highest_3_elements_in_vector(&rinvsq);
				rinvsix         = vec_madd(rinvsq,rinvsq,nul);
				rinvsix         = vec_madd(rinvsix,rinvsq,nul);
				tja             = ntiA+2*type[jnra];
				load_1_pair(vdwparam+tja,&c6,&c12);
				Vvdwtot          = vec_nmsub(c6,rinvsix,Vvdwtot);
				Vvdwtot          = vec_madd(c12,
										   vec_madd(rinvsix,rinvsix,nul),
										   Vvdwtot);
			}
			/* update outer data */
			add_vector_to_float(Vvdw+gid[n],Vvdwtot);
			ninner += nj1 - nj0;
		}
#ifdef GMX_THREAD_SHM_FDECOMP
		nouter += nn1 - nn0;
	} while (nn1<nri);
#else
	nouter = nri;
#endif
	*outeriter = nouter;
	*inneriter = ninner;
}
void 
nb_kernel133nf_ppc_altivec(int *             p_nri,
                       int               iinr[],
                       int               jindex[],
                       int               jjnr[],
                       int               shift[],
                       float             shiftvec[],
                       float             fshift[],
                       int               gid[],
                       float             pos[],
                       float             faction[],
                       float             charge[],
                       float *           p_facel,
                       float *           p_krf,
                       float *           p_crf,
                       float             Vc[],
                       int               type[],
                       int *             p_ntype,
                       float             vdwparam[],
                       float             Vvdw[],
                       float *           p_tabscale,
                       float             VFtab[],
                       float             invsqrta[],
                       float             dvda[],
                       float *           p_gbtabscale,
                       float             GBtab[],
                       int *             p_nthreads,
                       int *             count,
                       void *            mtx,
                       int *             outeriter,
                       int *             inneriter,
					   float *           work)
{
	vector float iOx,iOy,iOz,iH1x,iH1y,iH1z,iH2x,iH2y,iH2z,iMx,iMy,iMz;
	vector float dOx,dOy,dOz,dH1x,dH1y,dH1z,dH2x,dH2y,dH2z,dMx,dMy,dMz;
	vector float Vvdwtot,c6,c12,VVd,VVr,tsc,r;
	vector float vfacel,nul;
	vector float vctot,qqM,qqH,iqM,iqH,jq;
	vector float rinvO,rinvH1,rinvH2,rinvM,rsqO,rsqH1,rsqH2,rsqM;  

	int n,k,ii,is3,ii3,ntiA,nj0,nj1;
	int jnra,jnrb,jnrc,jnrd;
	int j3a,j3b,j3c,j3d;
	int nri, ntype, nouter, ninner;
	int tja,tjb,tjc,tjd;
#ifdef GMX_THREADS
	int nn0, nn1;
#endif
 
    nouter   = 0;
    ninner   = 0;
    nri      = *p_nri;
    ntype    = *p_ntype;
	nul=vec_zero();
	tsc=load_float_and_splat(p_tabscale);
	vfacel=load_float_and_splat(p_facel);
	ii         = iinr[0];
	iqH        = vec_madd(load_float_and_splat(charge+ii+1),vfacel,nul);
	iqM        = vec_madd(load_float_and_splat(charge+ii+3),vfacel,nul);
	ntiA       = 2*ntype*type[ii];
  
#ifdef GMX_THREADS
    nthreads = *p_nthreads;
	do {
		gmx_thread_mutex_lock((gmx_thread_mutex_t *)mtx);
		nn0              = *count;
		nn1              = nn0+(nri-nn0)/(2*nthreads)+3;
		*count           = nn1;
		gmx_thread_mutex_unlock((gmx_thread_mutex_t *)mtx);
		if(nn1>nri) nn1=nri;
		for(n=nn0; (n<nn1); n++) {
#if 0
		} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without gmx_threads */
		for(n=0;n<nri;n++) {
#endif  
			is3        = 3*shift[n];
			ii         = iinr[n];
			ii3        = 3*ii;
			load_1_4atoms_shift_and_splat(pos+ii3,shiftvec+is3,&iOx,&iOy,&iOz,
										  &iH1x,&iH1y,&iH1z,&iH2x,&iH2y,&iH2z,
										  &iMx,&iMy,&iMz);
			vctot      = nul;
			Vvdwtot     = nul;
			nj0        = jindex[n];
			nj1        = jindex[n+1];
    
			for(k=nj0; k<(nj1-3); k+=4) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				jnrc            = jjnr[k+2];
				jnrd            = jjnr[k+3];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				j3c             = 3*jnrc;
				j3d             = 3*jnrd;
				transpose_4_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),
								 load_xyz(pos+j3c),
								 load_xyz(pos+j3d),&dMx,&dMy,&dMz);
				dOx             = vec_sub(iOx,dMx);
				dOy             = vec_sub(iOy,dMy);
				dOz             = vec_sub(iOz,dMz);
				dH1x            = vec_sub(iH1x,dMx);
				dH1y            = vec_sub(iH1y,dMy);
				dH1z            = vec_sub(iH1z,dMz);
				dH2x            = vec_sub(iH2x,dMx);
				dH2y            = vec_sub(iH2y,dMy);
				dH2z            = vec_sub(iH2z,dMz);
				dMx             = vec_sub(iMx,dMx);
				dMy             = vec_sub(iMy,dMy);
				dMz             = vec_sub(iMz,dMz);

				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqM            = vec_madd(dMx,dMx,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqM            = vec_madd(dMy,dMy,rsqM);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);
				rsqM            = vec_madd(dMz,dMz,rsqM);

				rinvO           = do_invsqrt(rsqO);
				do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
				r               = vec_madd(rsqO,rinvO,nul);
				
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				tjc             = ntiA+2*type[jnrc];
				tjd             = ntiA+2*type[jnrd];
				/* load 4 j charges and multiply by iq */
				jq=load_4_float(charge+jnra,charge+jnrb,
								charge+jnrc,charge+jnrd);
				load_4_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,vdwparam+tjd,&c6,&c12);
				do_vonly_4_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&VVr);
				Vvdwtot         = vec_madd(c6,VVd,Vvdwtot);
				Vvdwtot         = vec_madd(c12,VVr,Vvdwtot);

				qqH             = vec_madd(iqH,jq,nul);
				qqM             = vec_madd(iqM,jq,nul);
				vctot           = vec_madd(qqM,rinvM,vctot);
				vctot           = vec_madd(qqH,rinvH1,vctot);
				vctot           = vec_madd(qqH,rinvH2,vctot);
			} 
			if(k<(nj1-2)) 
            {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				jnrc            = jjnr[k+2];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				j3c             = 3*jnrc;
				transpose_4_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),
								 load_xyz(pos+j3c),nul,&dMx,&dMy,&dMz);
				dOx             = vec_sub(iOx,dMx);
				dOy             = vec_sub(iOy,dMy);
				dOz             = vec_sub(iOz,dMz);
				dH1x            = vec_sub(iH1x,dMx);
				dH1y            = vec_sub(iH1y,dMy);
				dH1z            = vec_sub(iH1z,dMz);
				dH2x            = vec_sub(iH2x,dMx);
				dH2y            = vec_sub(iH2y,dMy);
				dH2z            = vec_sub(iH2z,dMz);
				dMx             = vec_sub(iMx,dMx);
				dMy             = vec_sub(iMy,dMy);
				dMz             = vec_sub(iMz,dMz);

				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqM            = vec_madd(dMx,dMx,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqM            = vec_madd(dMy,dMy,rsqM);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);
				rsqM            = vec_madd(dMz,dMz,rsqM);
				rinvO           = do_invsqrt(rsqO);
				do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
				zero_highest_element_in_vector(&rinvO);
				zero_highest_element_in_vector(&rsqO);				
				zero_highest_element_in_3_vectors(&rinvH1,&rinvH2,&rinvM);

				r               = vec_madd(rsqO,rinvO,nul);
				
				jq=load_3_float(charge+jnra,charge+jnrb,charge+jnrc);
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				tjc             = ntiA+2*type[jnrc];
				/* load 3 j charges and multiply by iq */
				load_3_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,&c6,&c12);
				do_vonly_3_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&VVr);
				Vvdwtot         = vec_madd(c6,VVd,Vvdwtot);
				Vvdwtot         = vec_madd(c12,VVr,Vvdwtot);

				qqH             = vec_madd(iqH,jq,nul);
				qqM             = vec_madd(iqM,jq,nul);
				vctot           = vec_madd(qqM,rinvM,vctot);
				vctot           = vec_madd(qqH,rinvH1,vctot);
				vctot           = vec_madd(qqH,rinvH2,vctot);
			}
            else if(k<(nj1-1))
            {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				transpose_2_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),&dMx,&dMy,&dMz);
				dOx             = vec_sub(iOx,dMx);
				dOy             = vec_sub(iOy,dMy);
				dOz             = vec_sub(iOz,dMz);
				dH1x            = vec_sub(iH1x,dMx);
				dH1y            = vec_sub(iH1y,dMy);
				dH1z            = vec_sub(iH1z,dMz);
				dH2x            = vec_sub(iH2x,dMx);
				dH2y            = vec_sub(iH2y,dMy);
				dH2z            = vec_sub(iH2z,dMz);
				dMx             = vec_sub(iMx,dMx);
				dMy             = vec_sub(iMy,dMy);
				dMz             = vec_sub(iMz,dMz);

				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqM            = vec_madd(dMx,dMx,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqM            = vec_madd(dMy,dMy,rsqM);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);
				rsqM            = vec_madd(dMz,dMz,rsqM);

				rinvO           = do_invsqrt(rsqO);
				do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);

				zero_highest_2_elements_in_vector(&rinvO);
				zero_highest_2_elements_in_vector(&rsqO);				
				zero_highest_2_elements_in_3_vectors(&rinvM,&rinvH1,&rinvH2);

				r               = vec_madd(rsqO,rinvO,nul);
				
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				/* load 2 j charges and multiply by iq */
				jq=load_2_float(charge+jnra,charge+jnrb);
				load_2_pair(vdwparam+tja,vdwparam+tjb,&c6,&c12);
				do_vonly_2_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&VVr);
				Vvdwtot         = vec_madd(c6,VVd,Vvdwtot);
				Vvdwtot         = vec_madd(c12,VVr,Vvdwtot);

				qqH             = vec_madd(iqH,jq,nul);
				qqM             = vec_madd(iqM,jq,nul);
				vctot           = vec_madd(qqM,rinvM,vctot);
				vctot           = vec_madd(qqH,rinvH1,vctot);
				vctot           = vec_madd(qqH,rinvH2,vctot);
			} 
            else if(k<nj1) 
            {
				jnra            = jjnr[k];
				j3a             = 3*jnra;
				transpose_1_to_3(load_xyz(pos+j3a),&dMx,&dMy,&dMz);
				dOx             = vec_sub(iOx,dMx);
				dOy             = vec_sub(iOy,dMy);
				dOz             = vec_sub(iOz,dMz);
				dH1x            = vec_sub(iH1x,dMx);
				dH1y            = vec_sub(iH1y,dMy);
				dH1z            = vec_sub(iH1z,dMz);
				dH2x            = vec_sub(iH2x,dMx);
				dH2y            = vec_sub(iH2y,dMy);
				dH2z            = vec_sub(iH2z,dMz);
				dMx             = vec_sub(iMx,dMx);
				dMy             = vec_sub(iMy,dMy);
				dMz             = vec_sub(iMz,dMz);

				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqM            = vec_madd(dMx,dMx,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqM            = vec_madd(dMy,dMy,rsqM);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);
				rsqM            = vec_madd(dMz,dMz,rsqM);
				rinvO           = do_invsqrt(rsqO);
				do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
				zero_highest_3_elements_in_vector(&rinvO);
				zero_highest_3_elements_in_vector(&rsqO);				
				zero_highest_3_elements_in_3_vectors(&rinvH1,&rinvH2,&rinvM);

				r               = vec_madd(rsqO,rinvO,nul);
				
				jq=load_1_float(charge+jnra);
				tja             = ntiA+2*type[jnra];
				/* load 1 j charge and multiply by iq */
				load_1_pair(vdwparam+tja,&c6,&c12);
				do_vonly_1_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&VVr);
				Vvdwtot         = vec_madd(c6,VVd,Vvdwtot);
				Vvdwtot         = vec_madd(c12,VVr,Vvdwtot);
				qqH             = vec_madd(iqH,jq,nul);
				qqM             = vec_madd(iqM,jq,nul);
				vctot           = vec_madd(qqM,rinvM,vctot);
				vctot           = vec_madd(qqH,rinvH1,vctot);
				vctot           = vec_madd(qqH,rinvH2,vctot);
			}
			/* update outer data */
			add_vector_to_float(Vc+gid[n],vctot);
			add_vector_to_float(Vvdw+gid[n],Vvdwtot);
			ninner += nj1 - nj0;
		}
#ifdef GMX_THREADS
		nouter += nn1 - nn0;
	} while (nn1<nri);
#else
	nouter = nri;
#endif
	*outeriter = nouter;
	*inneriter = ninner;
}
void 
nb_kernel100nf_ppc_altivec(int *             p_nri,
                       int               iinr[],
                       int               jindex[],
                       int               jjnr[],
                       int               shift[],
                       float             shiftvec[],
                       float             fshift[],
                       int               gid[],
                       float             pos[],
                       float             faction[],
                       float             charge[],
                       float *           p_facel,
                       float *           p_krf,
                       float *           p_crf,
                       float             Vc[],
                       int               type[],
                       int *             p_ntype,
                       float             vdwparam[],
                       float             Vvdw[],
                       float *           p_tabscale,
                       float             VFtab[],
                       float             invsqrta[],
                       float             dvda[],
                       float *           p_gbtabscale,
                       float             GBtab[],
                       int *             p_nthreads,
                       int *             count,
                       void *            mtx,
                       int *             outeriter,
                       int *             inneriter,
					   float *           work)
{
	vector float ix,iy,iz,shvec;
	vector float vfacel,nul;
	vector float dx,dy,dz;
	vector float vctot,qq,iq;
	vector float rinv,rsq;

	int n,k,ii,is3,ii3,nj0,nj1;
	int jnra,jnrb,jnrc,jnrd;
	int j3a,j3b,j3c,j3d;
	int nri, ntype, nouter, ninner;
#ifdef GMX_THREADS
	int nn0, nn1;
#endif
  
    nouter   = 0;
    ninner   = 0;
    nri      = *p_nri;
    ntype    = *p_ntype;
	nul=vec_zero();
	vfacel=load_float_and_splat(p_facel);

#ifdef GMX_THREADS
    nthreads = *p_nthreads;
	do {
		gmx_thread_mutex_lock((gmx_thread_mutex_t *)mtx);
		nn0              = *count;
		nn1              = nn0+(nri-nn0)/(2*nthreads)+3;
		*count           = nn1;
		gmx_thread_mutex_unlock((gmx_thread_mutex_t *)mtx);
		if(nn1>nri) nn1=nri;
		for(n=nn0; (n<nn1); n++) {
#if 0
		} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without gmx_threads */
		for(n=0;n<nri;n++) {
#endif  
			is3        = 3*shift[n];
			shvec      = load_xyz(shiftvec+is3);
			ii         = iinr[n];
			ii3        = 3*ii;
			ix         = load_xyz(pos+ii3);
			vctot      = nul;
			ix         = vec_add(ix,shvec);    
			nj0        = jindex[n];
			nj1        = jindex[n+1];
			splat_xyz_to_vectors(ix,&ix,&iy,&iz);
			iq         = vec_madd(load_float_and_splat(charge+ii),vfacel,nul);

			for(k=nj0; k<(nj1-3); k+=4) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				jnrc            = jjnr[k+2];
				jnrd            = jjnr[k+3];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				j3c             = 3*jnrc;
				j3d             = 3*jnrd;
				transpose_4_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),
								 load_xyz(pos+j3c),
								 load_xyz(pos+j3d),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				rinv            = do_invsqrt(rsq);
				/* load 4 j charges and multiply by iq */
				qq = vec_madd(load_4_float(charge+jnra,charge+jnrb,
										   charge+jnrc,charge+jnrd),iq,nul);
				vctot           = vec_madd(qq,rinv,vctot);
			}
			if(k<(nj1-1)) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				transpose_2_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				rinv            = do_invsqrt(rsq);
				zero_highest_2_elements_in_vector(&rinv);
				/* load 2 j charges and multiply by iq */
				qq = vec_madd(load_2_float(charge+jnra,charge+jnrb),iq,nul);
				vctot           = vec_madd(qq,rinv,vctot);
				k              += 2;
			}
			if((nj1-nj0) & 0x1) {
				jnra            = jjnr[k];
				j3a             = 3*jnra;
				transpose_1_to_3(load_xyz(pos+j3a),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				rinv            = do_invsqrt(rsq);
				zero_highest_3_elements_in_vector(&rinv);
				/* load 1 j charge and multiply by iq */
				qq = vec_madd(load_1_float(charge+jnra),iq,nul);
				vctot           = vec_madd(qq,rinv,vctot);
			}
			/* update outer data */
			add_vector_to_float(Vc+gid[n],vctot);
			ninner += nj1 - nj0;
		}
#ifdef GMX_THREADS
		nouter += nn1 - nn0;
	} while (nn1<nri);
#else
	nouter = nri;
#endif
	*outeriter = nouter;
	*inneriter = ninner;
}
void 
nb_kernel231_ppc_altivec  (int *             p_nri,
                       int               iinr[],
                       int               jindex[],
                       int               jjnr[],
                       int               shift[],
                       float             shiftvec[],
                       float             fshift[],
                       int               gid[],
                       float             pos[],
                       float             faction[],
                       float             charge[],
                       float *           p_facel,
                       float *           p_krf,
                       float *           p_crf,
                       float             Vc[],
                       int               type[],
                       int *             p_ntype,
                       float             vdwparam[],
                       float             Vvdw[],
                       float *           p_tabscale,
                       float             VFtab[],
                       float             invsqrta[],
                       float             dvda[],
                       float *           p_gbtabscale,
                       float             GBtab[],
                       int *             p_nthreads,
                       int *             count,
                       void *            mtx,
                       int *             outeriter,
                       int *             inneriter,
					   float *           work)
{
	vector float vkrf,vcrf;
	vector float iOx,iOy,iOz,iH1x,iH1y,iH1z,iH2x,iH2y,iH2z;
	vector float dOx,dOy,dOz,dH1x,dH1y,dH1z,dH2x,dH2y,dH2z;
	vector float vfacel,vcoulO,vcoulH1,vcoulH2,nul;
	vector float Vvdwtot,c6,c12,VVd,VVr,FFd,FFr,tsc,r;
	vector float fsO,fsH1,fsH2,krsqO,krsqH1,krsqH2;
	vector float vctot,qqO,qqH,iqO,iqH,jq;
	vector float fiOx,fiOy,fiOz,fiH1x,fiH1y,fiH1z,fiH2x,fiH2y,fiH2z;
	vector float tmp1,tmp2,tmp3,tmp4;
	vector float rinvO,rinvH1,rinvH2,rinvsqH1,rinvsqH2;
	vector float rsqO,rsqH1,rsqH2;
  
	int n,k,ii,is3,ii3,ntiA,nj0,nj1;
	int jnra,jnrb,jnrc,jnrd;
	int j3a,j3b,j3c,j3d;
	int nri, ntype, nouter, ninner;
	int tja,tjb,tjc,tjd;
#ifdef GMX_THREAD_SHM_FDECOMP
	int nn0, nn1;
#endif

    nouter   = 0;
    ninner   = 0;
    nri      = *p_nri;
    ntype    = *p_ntype;
	nul=vec_zero();
	tsc=load_float_and_splat(p_tabscale);
	vfacel=load_float_and_splat(p_facel);
	vkrf=load_float_and_splat(p_krf);
	vcrf=load_float_and_splat(p_crf);
	ii         = iinr[0];
	iqO        = vec_madd(load_float_and_splat(charge+ii),vfacel,nul);
	iqH        = vec_madd(load_float_and_splat(charge+ii+1),vfacel,nul);
	ntiA       = 2*ntype*type[ii];
  
#ifdef GMX_THREAD_SHM_FDECOMP
    nthreads = *p_nthreads;
	do {
		tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
		nn0              = *count;
		nn1              = nn0+(nri-nn0)/(2*nthreads)+3;
		*count           = nn1;
		tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
		if(nn1>nri) nn1=nri;
		for(n=nn0; (n<nn1); n++) {
#if 0
		} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without tMPI_Threads */
		for(n=0;n<nri;n++) {
#endif  
			is3        = 3*shift[n];
			ii         = iinr[n];
			ii3        = 3*ii;
			load_1_3atoms_shift_and_splat(pos+ii3,shiftvec+is3,&iOx,&iOy,&iOz,
										  &iH1x,&iH1y,&iH1z,&iH2x,&iH2y,&iH2z);
			vctot      = nul;
			Vvdwtot     = nul;
			fiOx       = nul;
			fiOy       = nul;
			fiOz       = nul;
			fiH1x      = nul;
			fiH1y      = nul;
			fiH1z      = nul;
			fiH2x      = nul;
			fiH2y      = nul;
			fiH2z      = nul;
			nj0        = jindex[n];
			nj1        = jindex[n+1];
    
			for(k=nj0; k<(nj1-3); k+=4) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				jnrc            = jjnr[k+2];
				jnrd            = jjnr[k+3];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				j3c             = 3*jnrc;
				j3d             = 3*jnrd;
				transpose_4_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),
								 load_xyz(pos+j3c),
								 load_xyz(pos+j3d),&dH2x,&dH2y,&dH2z);
				dOx             = vec_sub(iOx,dH2x);
				dOy             = vec_sub(iOy,dH2y);
				dOz             = vec_sub(iOz,dH2z);
				dH1x            = vec_sub(iH1x,dH2x);
				dH1y            = vec_sub(iH1y,dH2y);
				dH1z            = vec_sub(iH1z,dH2z);
				dH2x            = vec_sub(iH2x,dH2x);
				dH2y            = vec_sub(iH2y,dH2y);
				dH2z            = vec_sub(iH2z,dH2z);
      
				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);
				do_3_invsqrt(rsqO,rsqH1,rsqH2,&rinvO,&rinvH1,&rinvH2);
				rinvsqH1        = vec_madd(rinvH1,rinvH1,nul);
				rinvsqH2        = vec_madd(rinvH2,rinvH2,nul);
				r               = vec_madd(rinvO,rsqO,nul);
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				tjc             = ntiA+2*type[jnrc];
				tjd             = ntiA+2*type[jnrd];
				/* load 4 j charges and multiply by iq */
				jq=load_4_float(charge+jnra,charge+jnrb,
								charge+jnrc,charge+jnrd);
				load_4_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,vdwparam+tjd,&c6,&c12);
				do_4_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&FFd,&VVr,&FFr);
				Vvdwtot         = vec_madd(c6,VVd,Vvdwtot);
				tmp1            = vec_nmsub(c6,FFd,nul);
				Vvdwtot         = vec_madd(c12,VVr,Vvdwtot);
				tmp1            = vec_nmsub(c12,FFr,tmp1);
				tmp1            = vec_madd(tmp1,tsc,nul);

				qqO             = vec_madd(iqO,jq,nul);
				qqH             = vec_madd(iqH,jq,nul);
				krsqO           = vec_madd(vkrf,rsqO,nul);
				krsqH1          = vec_madd(vkrf,rsqH1,nul);
				krsqH2          = vec_madd(vkrf,rsqH2,nul);

				fsO             = vec_nmsub(vec_two(),krsqO,rinvO);
				vcoulO          = vec_add(rinvO,krsqO);
				vcoulH1         = vec_add(rinvH1,krsqH1);
				fsO             = vec_madd(qqO,fsO,nul);          
				vcoulH2         = vec_add(rinvH2,krsqH2);
				vcoulO          = vec_sub(vcoulO,vcrf);
				vcoulH1         = vec_sub(vcoulH1,vcrf);
				vcoulH2         = vec_sub(vcoulH2,vcrf);
				vctot           = vec_madd(qqO,vcoulO,vctot);
				fsH1            = vec_nmsub(vec_two(),krsqH1,rinvH1);
				fsH2            = vec_nmsub(vec_two(),krsqH2,rinvH2);
				vctot           = vec_madd(qqH,vcoulH1,vctot);
				fsO             = vec_madd(fsO,rinvO,tmp1);
				fsO             = vec_madd(fsO,rinvO,nul);
				fsH1            = vec_madd(fsH1,qqH,nul);
				fsH2            = vec_madd(fsH2,qqH,nul);
				vctot           = vec_madd(qqH,vcoulH2,vctot);
				fsH1            = vec_madd(fsH1,rinvsqH1,nul);
				fsH2            = vec_madd(fsH2,rinvsqH2,nul);

				fiOx            = vec_madd(fsO,dOx,fiOx); /* +=fx */
				dOx             = vec_nmsub(fsO,dOx,nul); /* -fx */
				fiOy            = vec_madd(fsO,dOy,fiOy); /* +=fy */
				dOy             = vec_nmsub(fsO,dOy,nul); /* -fy */
				fiOz            = vec_madd(fsO,dOz,fiOz); /* +=fz */
				dOz             = vec_nmsub(fsO,dOz,nul); /* -fz */
				fiH1x           = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
				dOx             = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
				fiH1y           = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
				dOy             = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
				fiH1z           = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
				dOz             = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
				fiH2x           = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
				dOx             = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
				fiH2y           = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
				dOy             = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
				fiH2z           = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
				dOz             = vec_nmsub(fsH2,dH2z,dOz); /* -fz */

				transpose_3_to_4(dOx,dOy,dOz,&tmp1,&tmp2,&tmp3,&tmp4);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);
				add_xyz_to_mem(faction+j3c,tmp3);
				add_xyz_to_mem(faction+j3d,tmp4);
			} 
			if(k<(nj1-2)) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				jnrc            = jjnr[k+2];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				j3c             = 3*jnrc;
				transpose_4_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),
								 load_xyz(pos+j3c),nul,&dH2x,&dH2y,&dH2z);
				dOx             = vec_sub(iOx,dH2x);
				dOy             = vec_sub(iOy,dH2y);
				dOz             = vec_sub(iOz,dH2z);
				dH1x            = vec_sub(iH1x,dH2x);
				dH1y            = vec_sub(iH1y,dH2y);
				dH1z            = vec_sub(iH1z,dH2z);
				dH2x            = vec_sub(iH2x,dH2x);
				dH2y            = vec_sub(iH2y,dH2y);
				dH2z            = vec_sub(iH2z,dH2z);
      
				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);

				zero_highest_element_in_3_vectors(&rsqO,&rsqH1,&rsqH2);
				do_3_invsqrt(rsqO,rsqH1,rsqH2,&rinvO,&rinvH1,&rinvH2);
				zero_highest_element_in_3_vectors(&rinvO,&rinvH1,&rinvH2);

				r               = vec_madd(rinvO,rsqO,nul);
				rinvsqH1        = vec_madd(rinvH1,rinvH1,nul);
				rinvsqH2        = vec_madd(rinvH2,rinvH2,nul);

				jq=load_3_float(charge+jnra,charge+jnrb,charge+jnrc);      
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				tjc             = ntiA+2*type[jnrc];
				load_3_pair(vdwparam+tja,vdwparam+tjb,vdwparam+tjc,&c6,&c12);
				do_3_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&FFd,&VVr,&FFr);
				Vvdwtot         = vec_madd(c6,VVd,Vvdwtot);
				tmp1            = vec_nmsub(c6,FFd,nul);
				Vvdwtot         = vec_madd(c12,VVr,Vvdwtot);
				tmp1            = vec_nmsub(c12,FFr,tmp1);
				tmp1            = vec_madd(tmp1,tsc,nul);

				qqO             = vec_madd(iqO,jq,nul);
				qqH             = vec_madd(iqH,jq,nul);
				krsqO           = vec_madd(vkrf,rsqO,nul);
				krsqH1          = vec_madd(vkrf,rsqH1,nul);
				krsqH2          = vec_madd(vkrf,rsqH2,nul);
				fsO             = vec_nmsub(vec_two(),krsqO,rinvO);
				vcoulO          = vec_add(rinvO,krsqO);
				vcoulH1         = vec_add(rinvH1,krsqH1);
				fsO             = vec_madd(qqO,fsO,nul);          
				vcoulH2         = vec_add(rinvH2,krsqH2);
				vcoulO          = vec_sub(vcoulO,vcrf);
				vcoulH1         = vec_sub(vcoulH1,vcrf);
				vcoulH2         = vec_sub(vcoulH2,vcrf);
				vctot           = vec_madd(qqO,vcoulO,vctot);
				fsH1            = vec_nmsub(vec_two(),krsqH1,rinvH1);
				fsH2            = vec_nmsub(vec_two(),krsqH2,rinvH2);
				vctot           = vec_madd(qqH,vcoulH1,vctot);
				fsO             = vec_madd(fsO,rinvO,tmp1);
				fsO             = vec_madd(fsO,rinvO,nul);
				fsH1            = vec_madd(fsH1,qqH,nul);
				fsH2            = vec_madd(fsH2,qqH,nul);
				vctot           = vec_madd(qqH,vcoulH2,vctot);
				fsH1            = vec_madd(fsH1,rinvsqH1,nul);
				fsH2            = vec_madd(fsH2,rinvsqH2,nul);

				fiOx            = vec_madd(fsO,dOx,fiOx); /* +=fx */
				dOx             = vec_nmsub(fsO,dOx,nul); /* -fx */
				fiOy            = vec_madd(fsO,dOy,fiOy); /* +=fy */
				dOy             = vec_nmsub(fsO,dOy,nul); /* -fy */
				fiOz            = vec_madd(fsO,dOz,fiOz); /* +=fz */
				dOz             = vec_nmsub(fsO,dOz,nul); /* -fz */
				fiH1x           = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
				dOx             = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
				fiH1y           = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
				dOy             = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
				fiH1z           = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
				dOz             = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
				fiH2x           = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
				dOx             = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
				fiH2y           = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
				dOy             = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
				fiH2z           = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
				dOz             = vec_nmsub(fsH2,dH2z,dOz); /* -fz */

				transpose_4_to_3(dOx,dOy,dOz,nul,&tmp1,&tmp2,&tmp3);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);
				add_xyz_to_mem(faction+j3c,tmp3);
			} else if(k<(nj1-1)) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				transpose_2_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),&dH2x,&dH2y,&dH2z);
				dOx             = vec_sub(iOx,dH2x);
				dOy             = vec_sub(iOy,dH2y);
				dOz             = vec_sub(iOz,dH2z);
				dH1x            = vec_sub(iH1x,dH2x);
				dH1y            = vec_sub(iH1y,dH2y);
				dH1z            = vec_sub(iH1z,dH2z);
				dH2x            = vec_sub(iH2x,dH2x);
				dH2y            = vec_sub(iH2y,dH2y);
				dH2z            = vec_sub(iH2z,dH2z);
      
				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);

				zero_highest_2_elements_in_3_vectors(&rsqO,&rsqH1,&rsqH2);
				do_3_invsqrt(rsqO,rsqH1,rsqH2,&rinvO,&rinvH1,&rinvH2);
				zero_highest_2_elements_in_3_vectors(&rinvO,&rinvH1,&rinvH2);

				r               = vec_madd(rinvO,rsqO,nul);
				rinvsqH1        = vec_madd(rinvH1,rinvH1,nul);
				rinvsqH2        = vec_madd(rinvH2,rinvH2,nul);
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				/* load 2 j charges and multiply by iq */
				jq=load_2_float(charge+jnra,charge+jnrb);
				load_2_pair(vdwparam+tja,vdwparam+tjb,&c6,&c12);
				do_2_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&FFd,&VVr,&FFr);
				Vvdwtot         = vec_madd(c6,VVd,Vvdwtot);
				tmp1            = vec_nmsub(c6,FFd,nul);
				Vvdwtot         = vec_madd(c12,VVr,Vvdwtot);
				tmp1            = vec_nmsub(c12,FFr,tmp1);
				tmp1            = vec_madd(tmp1,tsc,nul);

				qqO             = vec_madd(iqO,jq,nul);
				qqH             = vec_madd(iqH,jq,nul);
				krsqO           = vec_madd(vkrf,rsqO,nul);
				krsqH1          = vec_madd(vkrf,rsqH1,nul);
				krsqH2          = vec_madd(vkrf,rsqH2,nul);
				fsO             = vec_nmsub(vec_two(),krsqO,rinvO);
				vcoulO          = vec_add(rinvO,krsqO);
				vcoulH1         = vec_add(rinvH1,krsqH1);
				fsO             = vec_madd(qqO,fsO,nul);          
				vcoulH2         = vec_add(rinvH2,krsqH2);
				vcoulO          = vec_sub(vcoulO,vcrf);
				vcoulH1         = vec_sub(vcoulH1,vcrf);
				vcoulH2         = vec_sub(vcoulH2,vcrf);
				vctot           = vec_madd(qqO,vcoulO,vctot);
				fsH1            = vec_nmsub(vec_two(),krsqH1,rinvH1);
				fsH2            = vec_nmsub(vec_two(),krsqH2,rinvH2);
				vctot           = vec_madd(qqH,vcoulH1,vctot);
				fsO             = vec_madd(fsO,rinvO,tmp1);
				fsO             = vec_madd(fsO,rinvO,nul);
				fsH1            = vec_madd(fsH1,qqH,nul);
				fsH2            = vec_madd(fsH2,qqH,nul);
				vctot           = vec_madd(qqH,vcoulH2,vctot);
				fsH1            = vec_madd(fsH1,rinvsqH1,nul);
				fsH2            = vec_madd(fsH2,rinvsqH2,nul);

				fiOx            = vec_madd(fsO,dOx,fiOx); /* +=fx */
				dOx             = vec_nmsub(fsO,dOx,nul); /* -fx */
				fiOy            = vec_madd(fsO,dOy,fiOy); /* +=fy */
				dOy             = vec_nmsub(fsO,dOy,nul); /* -fy */
				fiOz            = vec_madd(fsO,dOz,fiOz); /* +=fz */
				dOz             = vec_nmsub(fsO,dOz,nul); /* -fz */
				fiH1x           = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
				dOx             = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
				fiH1y           = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
				dOy             = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
				fiH1z           = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
				dOz             = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
				fiH2x           = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
				dOx             = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
				fiH2y           = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
				dOy             = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
				fiH2z           = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
				dOz             = vec_nmsub(fsH2,dH2z,dOz); /* -fz */

				transpose_3_to_2(dOx,dOy,dOz,&tmp1,&tmp2);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);
			} else if(k<nj1) {
				jnra            = jjnr[k];
				j3a             = 3*jnra;
				transpose_1_to_3(load_xyz(pos+j3a),&dH2x,&dH2y,&dH2z);
				dOx             = vec_sub(iOx,dH2x);
				dOy             = vec_sub(iOy,dH2y);
				dOz             = vec_sub(iOz,dH2z);
				dH1x            = vec_sub(iH1x,dH2x);
				dH1y            = vec_sub(iH1y,dH2y);
				dH1z            = vec_sub(iH1z,dH2z);
				dH2x            = vec_sub(iH2x,dH2x);
				dH2y            = vec_sub(iH2y,dH2y);
				dH2z            = vec_sub(iH2z,dH2z);
      
				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);

				zero_highest_3_elements_in_3_vectors(&rsqO,&rsqH1,&rsqH2);
				do_3_invsqrt(rsqO,rsqH1,rsqH2,&rinvO,&rinvH1,&rinvH2);
				zero_highest_3_elements_in_3_vectors(&rinvO,&rinvH1,&rinvH2);

				r               = vec_madd(rinvO,rsqO,nul);
				rinvsqH1        = vec_madd(rinvH1,rinvH1,nul);
				rinvsqH2        = vec_madd(rinvH2,rinvH2,nul);
				tja             = ntiA+2*type[jnra];
				/* load 1 j charges and multiply by iq */
				jq=load_1_float(charge+jnra);
				load_1_pair(vdwparam+tja,&c6,&c12);
				do_1_ljtable_lj(VFtab,vec_madd(r,tsc,nul),&VVd,&FFd,&VVr,&FFr);
				Vvdwtot         = vec_madd(c6,VVd,Vvdwtot);
				tmp1            = vec_nmsub(c6,FFd,nul);
				Vvdwtot         = vec_madd(c12,VVr,Vvdwtot);
				tmp1            = vec_nmsub(c12,FFr,tmp1);
				tmp1            = vec_madd(tmp1,tsc,nul);

				qqO             = vec_madd(iqO,jq,nul);
				qqH             = vec_madd(iqH,jq,nul);
				krsqO           = vec_madd(vkrf,rsqO,nul);
				krsqH1          = vec_madd(vkrf,rsqH1,nul);
				krsqH2          = vec_madd(vkrf,rsqH2,nul);
				fsO             = vec_nmsub(vec_two(),krsqO,rinvO);
				vcoulO          = vec_add(rinvO,krsqO);
				vcoulH1         = vec_add(rinvH1,krsqH1);
				fsO             = vec_madd(qqO,fsO,nul);          
				vcoulH2         = vec_add(rinvH2,krsqH2);
				vcoulO          = vec_sub(vcoulO,vcrf);
				vcoulH1         = vec_sub(vcoulH1,vcrf);
				vcoulH2         = vec_sub(vcoulH2,vcrf);
				vctot           = vec_madd(qqO,vcoulO,vctot);
				fsH1            = vec_nmsub(vec_two(),krsqH1,rinvH1);
				fsH2            = vec_nmsub(vec_two(),krsqH2,rinvH2);
				vctot           = vec_madd(qqH,vcoulH1,vctot);
				fsO             = vec_madd(fsO,rinvO,tmp1);
				fsO             = vec_madd(fsO,rinvO,nul);
				fsH1            = vec_madd(fsH1,qqH,nul);
				fsH2            = vec_madd(fsH2,qqH,nul);
				vctot           = vec_madd(qqH,vcoulH2,vctot);
				fsH1            = vec_madd(fsH1,rinvsqH1,nul);
				fsH2            = vec_madd(fsH2,rinvsqH2,nul);

				fiOx            = vec_madd(fsO,dOx,fiOx); /* +=fx */
				dOx             = vec_nmsub(fsO,dOx,nul); /* -fx */
				fiOy            = vec_madd(fsO,dOy,fiOy); /* +=fy */
				dOy             = vec_nmsub(fsO,dOy,nul); /* -fy */
				fiOz            = vec_madd(fsO,dOz,fiOz); /* +=fz */
				dOz             = vec_nmsub(fsO,dOz,nul); /* -fz */
				fiH1x           = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
				dOx             = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
				fiH1y           = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
				dOy             = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
				fiH1z           = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
				dOz             = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
				fiH2x           = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
				dOx             = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
				fiH2y           = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
				dOy             = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
				fiH2z           = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
				dOz             = vec_nmsub(fsH2,dH2z,dOz); /* -fz */

				transpose_3_to_1(dOx,dOy,dOz,&tmp1);
				add_xyz_to_mem(faction+j3a,tmp1);
			}
			/* update outer data */
			update_i_3atoms_forces(faction+ii3,fshift+is3,
								   fiOx,fiOy,fiOz,fiH1x,fiH1y,fiH1z,
								   fiH2x,fiH2y,fiH2z);

			add_vector_to_float(Vc+gid[n],vctot);
			add_vector_to_float(Vvdw+gid[n],Vvdwtot);
			ninner += nj1 - nj0;
		}
#ifdef GMX_THREAD_SHM_FDECOMP
		nouter += nn1 - nn0;
	} while (nn1<nri);
#else
	nouter = nri;
#endif
	*outeriter = nouter;
	*inneriter = ninner;
}
void 
nb_kernel400_ppc_altivec  (int *             p_nri,
                       int               iinr[],
                       int               jindex[],
                       int               jjnr[],
                       int               shift[],
                       float             shiftvec[],
                       float             fshift[],
                       int               gid[],
                       float             pos[],
                       float             faction[],
                       float             charge[],
                       float *           p_facel,
                       float *           p_krf,
                       float *           p_crf,
                       float             Vc[],
                       int               type[],
                       int *             p_ntype,
                       float             vdwparam[],
                       float             Vvdw[],
                       float *           p_tabscale,
                       float             VFtab[],
                       float             invsqrta[],
                       float             dvda[],
                       float *           p_gbtabscale,
                       float             GBtab[],
                       int *             p_nthreads,
                       int *             count,
                       void *            mtx,
                       int *             outeriter,
                       int *             inneriter,
					   float *           work)
{
	vector float ix,iy,iz,shvec;
	vector float vfacel,fs,nul;
	vector float dx,dy,dz;
	vector float vctot,qq,iq;
	vector float fix,fiy,fiz;
	vector float tmp1,tmp2,tmp3,tmp4;
	vector float rinv,r,rsq,VVc,FFc;
	vector float isai,isaj,isaprod,gbtsc,dvdasum,dvdaj,dvdatmp,gbscale,half;

	int n,k,ii,is3,ii3,nj0,nj1;
	int jnra,jnrb,jnrc,jnrd;
	int j3a,j3b,j3c,j3d;
	int nri, ntype, nouter, ninner;
#ifdef GMX_THREADS
	int nn0, nn1;
#endif

    nouter   = 0;
    ninner   = 0;
    nri      = *p_nri;
    ntype    = *p_ntype;
	nul=vec_zero();
	half=vec_half();
	vfacel=load_float_and_splat(p_facel);
	gbtsc=load_float_and_splat(p_gbtabscale);

#ifdef GMX_THREADS
    nthreads = *p_nthreads;
	do {
		gmx_thread_mutex_lock((gmx_thread_mutex_t *)mtx);
		nn0              = *count;
		nn1              = nn0+(nri-nn0)/(2*nthreads)+3;
		*count           = nn1;
		gmx_thread_mutex_unlock((gmx_thread_mutex_t *)mtx);
		if(nn1>nri) nn1=nri;
		for(n=nn0; (n<nn1); n++) {
#if 0
		} /* maintain correct indentation even with conditional left braces */
#endif
#else /* without gmx_threads */
		for(n=0;n<nri;n++) {
#endif  
			is3        = 3*shift[n];
			shvec      = load_xyz(shiftvec+is3);
			ii         = iinr[n];
			ii3        = 3*ii;
			ix         = load_xyz(pos+ii3);
			vctot      = nul;
			dvdasum    = nul;
			fix        = nul;
			fiy        = nul;
			fiz        = nul;
			ix         = vec_add(ix,shvec);    
			nj0        = jindex[n];
			nj1        = jindex[n+1];
			splat_xyz_to_vectors(ix,&ix,&iy,&iz);
			iq         = vec_madd(load_float_and_splat(charge+ii),vfacel,nul);
			isai       = load_float_and_splat(invsqrta+ii);

			for(k=nj0; k<(nj1-3); k+=4) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				jnrc            = jjnr[k+2];
				jnrd            = jjnr[k+3];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				j3c             = 3*jnrc;
				j3d             = 3*jnrd;
				transpose_4_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),
								 load_xyz(pos+j3c),
								 load_xyz(pos+j3d),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				rinv            = do_invsqrt(rsq);
				r               = vec_madd(rinv,rsq,nul);
				/* load 1/sqrt(a2) and multiply with 1/sqrt(a1) */
				isaj    = load_4_float(invsqrta+jnra,invsqrta+jnrb,
									   invsqrta+jnrc,invsqrta+jnrd);
				isaprod = vec_madd(isai,isaj,nul);
				/* load 4 j charges and multiply by iq and 1/sqrt(a1*a2) */
				qq = vec_madd(load_4_float(charge+jnra,charge+jnrb,
										   charge+jnrc,charge+jnrd),iq,nul);
				qq = vec_madd(isaprod,qq,nul);
				gbscale = vec_madd(isaprod,gbtsc,nul);
				do_4_ctable_coul(GBtab,vec_madd(r,gbscale,nul),&VVc,&FFc);
				dvdaj           = load_4_float(dvda+jnra,dvda+jnrb,
											   dvda+jnrc,dvda+jnrd);
				fs              = vec_madd(qq,FFc,nul);
				fs              = vec_madd(fs,gbscale,nul);
				vctot           = vec_madd(qq,VVc,vctot);
				dvdatmp         = vec_madd(fs,r,nul);
				dvdatmp         = vec_madd(qq,VVc,dvdatmp);
				dvdasum         = vec_sub(dvdasum,dvdatmp);
				dvdaj		      = vec_sub(dvdaj,dvdatmp);
				store_4_float(dvdaj,dvda+jnra,dvda+jnrb,dvda+jnrc,dvda+jnrd);
				fs              = vec_nmsub(fs,rinv,nul);
				fix             = vec_madd(fs,dx,fix); /* +=fx */
				fiy             = vec_madd(fs,dy,fiy); /* +=fy */
				fiz             = vec_madd(fs,dz,fiz); /* +=fz */
				dx              = vec_nmsub(dx,fs,nul); /* -fx */
				dy              = vec_nmsub(dy,fs,nul); /* -fy */
				dz              = vec_nmsub(dz,fs,nul); /* -fz */
				transpose_3_to_4(dx,dy,dz,&tmp1,&tmp2,&tmp3,&tmp4);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);
				add_xyz_to_mem(faction+j3c,tmp3);
				add_xyz_to_mem(faction+j3d,tmp4);
			}
			if(k<(nj1-1)) {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				transpose_2_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				zero_highest_2_elements_in_vector(&rsq);
				rinv            = do_invsqrt(rsq);
				zero_highest_2_elements_in_vector(&rinv);
				r               = vec_madd(rinv,rsq,nul);
				/* load 1/sqrt(a2) and multiply with 1/sqrt(a1) */
				isaj    = load_2_float(invsqrta+jnra,invsqrta+jnrb);
				isaprod = vec_madd(isai,isaj,nul);
				/* load 2 j charges and multiply by iq and 1/sqrt(a1*a2) */
				qq = vec_madd(load_2_float(charge+jnra,charge+jnrb),iq,nul);
				qq = vec_madd(isaprod,qq,nul);
				gbscale = vec_madd(isaprod,gbtsc,nul);
				do_2_ctable_coul(GBtab,vec_madd(r,gbscale,nul),&VVc,&FFc);
				dvdaj           = load_2_float(dvda+jnra,dvda+jnrb);
				fs              = vec_madd(qq,FFc,nul);
				fs              = vec_madd(fs,gbscale,nul);
				vctot           = vec_madd(qq,VVc,vctot);
				dvdatmp         = vec_madd(fs,r,nul);
				dvdatmp         = vec_madd(qq,VVc,dvdatmp);
				dvdasum         = vec_sub(dvdasum,dvdatmp);
				dvdaj           = vec_sub(dvdaj,dvdatmp);
				store_2_float(dvdaj,dvda+jnra,dvda+jnrb);
				fs              = vec_nmsub(fs,rinv,nul);
				fix             = vec_madd(fs,dx,fix); /* +=fx */
				fiy             = vec_madd(fs,dy,fiy); /* +=fy */
				fiz             = vec_madd(fs,dz,fiz); /* +=fz */
				dx              = vec_nmsub(dx,fs,nul); /* -fx */
				dy              = vec_nmsub(dy,fs,nul); /* -fy */
				dz              = vec_nmsub(dz,fs,nul); /* -fz */
				transpose_3_to_2(dx,dy,dz,&tmp1,&tmp2);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);     
				k              += 2;
			}
			if((nj1-nj0) & 0x1) {
				jnra            = jjnr[k];
				j3a             = 3*jnra;
				transpose_1_to_3(load_xyz(pos+j3a),&dx,&dy,&dz);
				dx              = vec_sub(ix,dx);
				dy              = vec_sub(iy,dy);
				dz              = vec_sub(iz,dz);
				rsq             = vec_madd(dx,dx,nul);
				rsq             = vec_madd(dy,dy,rsq);
				rsq             = vec_madd(dz,dz,rsq);
				zero_highest_3_elements_in_vector(&rsq);
				rinv            = do_invsqrt(rsq);
				zero_highest_3_elements_in_vector(&rinv);
				r               = vec_madd(rinv,rsq,nul);
				/* load 1/sqrt(a2) and multiply with 1/sqrt(a1) */
				isaj    = load_1_float(invsqrta+jnra);
				isaprod = vec_madd(isai,isaj,nul);
				/* load 1 j charge and multiply by iq and 1/sqrt(a1*a2) */
				qq = vec_madd(load_1_float(charge+jnra),iq,nul);
				qq = vec_madd(isaprod,qq,nul);
				gbscale = vec_madd(isaprod,gbtsc,nul);
				do_1_ctable_coul(GBtab,vec_madd(r,gbscale,nul),&VVc,&FFc);
				dvdaj           = load_1_float(dvda+jnra);
				fs              = vec_madd(qq,FFc,nul);
				fs              = vec_madd(fs,gbscale,nul);
				vctot           = vec_madd(qq,VVc,vctot);
				dvdatmp         = vec_madd(fs,r,nul);
				dvdatmp         = vec_madd(qq,VVc,dvdatmp);
				dvdasum         = vec_sub(dvdasum,dvdatmp);
				dvdaj           = vec_sub(dvdaj,dvdatmp);
				store_1_float(dvdaj,dvda+jnra);
				fs              = vec_nmsub(fs,rinv,nul);
				fix             = vec_madd(fs,dx,fix); /* +=fx */
				fiy             = vec_madd(fs,dy,fiy); /* +=fy */
				fiz             = vec_madd(fs,dz,fiz); /* +=fz */
				dx              = vec_nmsub(dx,fs,nul); /* -fx */
				dy              = vec_nmsub(dy,fs,nul); /* -fy */
				dz              = vec_nmsub(dz,fs,nul); /* -fz */
				transpose_3_to_1(dx,dy,dz,&tmp1);
				add_xyz_to_mem(faction+j3a,tmp1);
			}
			/* update outer data */
			transpose_3_to_4(fix,fiy,fiz,&tmp1,&tmp2,&tmp3,&tmp4);
			tmp1 = vec_add(tmp1,tmp3);
			tmp2 = vec_add(tmp2,tmp4);
			tmp1 = vec_add(tmp1,tmp2);    
			add_xyz_to_mem(faction+ii3,tmp1);
			add_xyz_to_mem(fshift+is3,tmp1);

			add_vector_to_float(Vc+gid[n],vctot);

			add_vector_to_float(dvda+ii,dvdasum);
			ninner += nj1 - nj0;
		}
#ifdef GMX_THREADS
		nouter += nn1 - nn0;
	} while (nn1<nri);
#else
	nouter = nri;
#endif
	*outeriter = nouter;
	*inneriter = ninner;
}