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_kernel030nf_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,tsc;
vector float dx,dy,dz;
vector float Vvdwtot,c6,c12;
vector float rinv,r,rsq;
vector float VVd,VVr;
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);
#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);
rinv = do_invsqrt(rsq);
r = vec_madd(rinv,rsq,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);
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);
}
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);
tja = ntiA+2*type[jnra];
tjb = ntiA+2*type[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);
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);
tja = ntiA+2*type[jnra];
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);
}
/* 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;
}
