src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_avx_128_fma_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_128_fma_single.h"
  48 #include "kernelutil_x86_avx_128_fma_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_VF_avx_128_fma_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water4-Water4
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_VF_avx_128_fma_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwioffset3;
  90     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  94     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  95     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  96     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  97     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
  98     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
  99     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 100     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 101     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 102     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
 103     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 104     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 105     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
 106     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
 107     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
 108     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
 109     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 110     real             *charge;
 111     int              nvdwtype;
 112     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 113     int              *vdwtype;
 114     real             *vdwparam;
 115     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 116     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 117     __m128           c6grid_00;
 118     __m128           c6grid_11;
 119     __m128           c6grid_12;
 120     __m128           c6grid_13;
 121     __m128           c6grid_21;
 122     __m128           c6grid_22;
 123     __m128           c6grid_23;
 124     __m128           c6grid_31;
 125     __m128           c6grid_32;
 126     __m128           c6grid_33;
 127     real             *vdwgridparam;
 128     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 129     __m128           one_half = _mm_set1_ps(0.5);
 130     __m128           minus_one = _mm_set1_ps(-1.0);
 131     __m128i          ewitab;
 132     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 133     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 134     real             *ewtab;
 135     __m128           dummy_mask,cutoff_mask;
 136     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 137     __m128           one     = _mm_set1_ps(1.0);
 138     __m128           two     = _mm_set1_ps(2.0);
 139     x                = xx[0];
 140     f                = ff[0];
 141
 142     nri              = nlist->nri;
 143     iinr             = nlist->iinr;
 144     jindex           = nlist->jindex;
 145     jjnr             = nlist->jjnr;
 146     shiftidx         = nlist->shift;
 147     gid              = nlist->gid;
 148     shiftvec         = fr->shift_vec[0];
 149     fshift           = fr->fshift[0];
 150     facel            = _mm_set1_ps(fr->epsfac);
 151     charge           = mdatoms->chargeA;
 152     nvdwtype         = fr->ntype;
 153     vdwparam         = fr->nbfp;
 154     vdwtype          = mdatoms->typeA;
 155     vdwgridparam     = fr->ljpme_c6grid;
 156     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 157     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 158     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 159
 160     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 161     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
 162     beta2            = _mm_mul_ps(beta,beta);
 163     beta3            = _mm_mul_ps(beta,beta2);
 164     ewtab            = fr->ic->tabq_coul_FDV0;
 165     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 166     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 167
 168     /* Setup water-specific parameters */
 169     inr              = nlist->iinr[0];
 170     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 171     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 172     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 173     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 174
 175     jq1              = _mm_set1_ps(charge[inr+1]);
 176     jq2              = _mm_set1_ps(charge[inr+2]);
 177     jq3              = _mm_set1_ps(charge[inr+3]);
 178     vdwjidx0A        = 2*vdwtype[inr+0];
 179     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 180     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 181     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
 182     qq11             = _mm_mul_ps(iq1,jq1);
 183     qq12             = _mm_mul_ps(iq1,jq2);
 184     qq13             = _mm_mul_ps(iq1,jq3);
 185     qq21             = _mm_mul_ps(iq2,jq1);
 186     qq22             = _mm_mul_ps(iq2,jq2);
 187     qq23             = _mm_mul_ps(iq2,jq3);
 188     qq31             = _mm_mul_ps(iq3,jq1);
 189     qq32             = _mm_mul_ps(iq3,jq2);
 190     qq33             = _mm_mul_ps(iq3,jq3);
 191
 192     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 193     rcutoff_scalar   = fr->rcoulomb;
 194     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 195     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 196
 197     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 198     rvdw             = _mm_set1_ps(fr->rvdw);
 199
 200     /* Avoid stupid compiler warnings */
 201     jnrA = jnrB = jnrC = jnrD = 0;
 202     j_coord_offsetA = 0;
 203     j_coord_offsetB = 0;
 204     j_coord_offsetC = 0;
 205     j_coord_offsetD = 0;
 206
 207     outeriter        = 0;
 208     inneriter        = 0;
 209
 210     for(iidx=0;iidx<4*DIM;iidx++)
 211     {
 212         scratch[iidx] = 0.0;
 213     }
 214
 215     /* Start outer loop over neighborlists */
 216     for(iidx=0; iidx<nri; iidx++)
 217     {
 218         /* Load shift vector for this list */
 219         i_shift_offset   = DIM*shiftidx[iidx];
 220
 221         /* Load limits for loop over neighbors */
 222         j_index_start    = jindex[iidx];
 223         j_index_end      = jindex[iidx+1];
 224
 225         /* Get outer coordinate index */
 226         inr              = iinr[iidx];
 227         i_coord_offset   = DIM*inr;
 228
 229         /* Load i particle coords and add shift vector */
 230         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 231                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 232
 233         fix0             = _mm_setzero_ps();
 234         fiy0             = _mm_setzero_ps();
 235         fiz0             = _mm_setzero_ps();
 236         fix1             = _mm_setzero_ps();
 237         fiy1             = _mm_setzero_ps();
 238         fiz1             = _mm_setzero_ps();
 239         fix2             = _mm_setzero_ps();
 240         fiy2             = _mm_setzero_ps();
 241         fiz2             = _mm_setzero_ps();
 242         fix3             = _mm_setzero_ps();
 243         fiy3             = _mm_setzero_ps();
 244         fiz3             = _mm_setzero_ps();
 245
 246         /* Reset potential sums */
 247         velecsum         = _mm_setzero_ps();
 248         vvdwsum          = _mm_setzero_ps();
 249
 250         /* Start inner kernel loop */
 251         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 252         {
 253
 254             /* Get j neighbor index, and coordinate index */
 255             jnrA             = jjnr[jidx];
 256             jnrB             = jjnr[jidx+1];
 257             jnrC             = jjnr[jidx+2];
 258             jnrD             = jjnr[jidx+3];
 259             j_coord_offsetA  = DIM*jnrA;
 260             j_coord_offsetB  = DIM*jnrB;
 261             j_coord_offsetC  = DIM*jnrC;
 262             j_coord_offsetD  = DIM*jnrD;
 263
 264             /* load j atom coordinates */
 265             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 266                                               x+j_coord_offsetC,x+j_coord_offsetD,
 267                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 268                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 269
 270             /* Calculate displacement vector */
 271             dx00             = _mm_sub_ps(ix0,jx0);
 272             dy00             = _mm_sub_ps(iy0,jy0);
 273             dz00             = _mm_sub_ps(iz0,jz0);
 274             dx11             = _mm_sub_ps(ix1,jx1);
 275             dy11             = _mm_sub_ps(iy1,jy1);
 276             dz11             = _mm_sub_ps(iz1,jz1);
 277             dx12             = _mm_sub_ps(ix1,jx2);
 278             dy12             = _mm_sub_ps(iy1,jy2);
 279             dz12             = _mm_sub_ps(iz1,jz2);
 280             dx13             = _mm_sub_ps(ix1,jx3);
 281             dy13             = _mm_sub_ps(iy1,jy3);
 282             dz13             = _mm_sub_ps(iz1,jz3);
 283             dx21             = _mm_sub_ps(ix2,jx1);
 284             dy21             = _mm_sub_ps(iy2,jy1);
 285             dz21             = _mm_sub_ps(iz2,jz1);
 286             dx22             = _mm_sub_ps(ix2,jx2);
 287             dy22             = _mm_sub_ps(iy2,jy2);
 288             dz22             = _mm_sub_ps(iz2,jz2);
 289             dx23             = _mm_sub_ps(ix2,jx3);
 290             dy23             = _mm_sub_ps(iy2,jy3);
 291             dz23             = _mm_sub_ps(iz2,jz3);
 292             dx31             = _mm_sub_ps(ix3,jx1);
 293             dy31             = _mm_sub_ps(iy3,jy1);
 294             dz31             = _mm_sub_ps(iz3,jz1);
 295             dx32             = _mm_sub_ps(ix3,jx2);
 296             dy32             = _mm_sub_ps(iy3,jy2);
 297             dz32             = _mm_sub_ps(iz3,jz2);
 298             dx33             = _mm_sub_ps(ix3,jx3);
 299             dy33             = _mm_sub_ps(iy3,jy3);
 300             dz33             = _mm_sub_ps(iz3,jz3);
 301
 302             /* Calculate squared distance and things based on it */
 303             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 304             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 305             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 306             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 307             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 308             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 309             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 310             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 311             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 312             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 313
 314             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 315             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 316             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 317             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 318             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 319             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 320             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 321             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 322             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 323             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 324
 325             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 326             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 327             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 328             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 329             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 330             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 331             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 332             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 333             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 334             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 335
 336             fjx0             = _mm_setzero_ps();
 337             fjy0             = _mm_setzero_ps();
 338             fjz0             = _mm_setzero_ps();
 339             fjx1             = _mm_setzero_ps();
 340             fjy1             = _mm_setzero_ps();
 341             fjz1             = _mm_setzero_ps();
 342             fjx2             = _mm_setzero_ps();
 343             fjy2             = _mm_setzero_ps();
 344             fjz2             = _mm_setzero_ps();
 345             fjx3             = _mm_setzero_ps();
 346             fjy3             = _mm_setzero_ps();
 347             fjz3             = _mm_setzero_ps();
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             if (gmx_mm_any_lt(rsq00,rcutoff2))
 354             {
 355
 356             r00              = _mm_mul_ps(rsq00,rinv00);
 357
 358             /* Analytical LJ-PME */
 359             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 360             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 361             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 362             exponent         = gmx_simd_exp_r(ewcljrsq);
 363             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 364             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 365             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 366             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 367             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 368             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 369                                           _mm_mul_ps(_mm_sub_ps(vvdw6,_mm_macc_ps(c6grid_00,sh_lj_ewald,_mm_mul_ps(c6_00,sh_vdw_invrcut6))),one_sixth));
 370             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 371             fvdw             = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
 372
 373             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 377             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 378
 379             fscal            = fvdw;
 380
 381             fscal            = _mm_and_ps(fscal,cutoff_mask);
 382
 383              /* Update vectorial force */
 384             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 385             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 386             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 387
 388             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 389             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 390             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 391
 392             }
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm_any_lt(rsq11,rcutoff2))
 399             {
 400
 401             r11              = _mm_mul_ps(rsq11,rinv11);
 402
 403             /* EWALD ELECTROSTATICS */
 404
 405             /* Analytical PME correction */
 406             zeta2            = _mm_mul_ps(beta2,rsq11);
 407             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 408             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 409             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 410             felec            = _mm_mul_ps(qq11,felec);
 411             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 412             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv11,sh_ewald));
 413             velec            = _mm_mul_ps(qq11,velec);
 414
 415             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 416
 417             /* Update potential sum for this i atom from the interaction with this j atom. */
 418             velec            = _mm_and_ps(velec,cutoff_mask);
 419             velecsum         = _mm_add_ps(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             fscal            = _mm_and_ps(fscal,cutoff_mask);
 424
 425              /* Update vectorial force */
 426             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 427             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 428             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 429
 430             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 431             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 432             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 433
 434             }
 435
 436             /**************************
 437              * CALCULATE INTERACTIONS *
 438              **************************/
 439
 440             if (gmx_mm_any_lt(rsq12,rcutoff2))
 441             {
 442
 443             r12              = _mm_mul_ps(rsq12,rinv12);
 444
 445             /* EWALD ELECTROSTATICS */
 446
 447             /* Analytical PME correction */
 448             zeta2            = _mm_mul_ps(beta2,rsq12);
 449             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
 450             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 451             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 452             felec            = _mm_mul_ps(qq12,felec);
 453             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 454             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv12,sh_ewald));
 455             velec            = _mm_mul_ps(qq12,velec);
 456
 457             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velec            = _mm_and_ps(velec,cutoff_mask);
 461             velecsum         = _mm_add_ps(velecsum,velec);
 462
 463             fscal            = felec;
 464
 465             fscal            = _mm_and_ps(fscal,cutoff_mask);
 466
 467              /* Update vectorial force */
 468             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 469             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 470             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 471
 472             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 473             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 474             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 475
 476             }
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm_any_lt(rsq13,rcutoff2))
 483             {
 484
 485             r13              = _mm_mul_ps(rsq13,rinv13);
 486
 487             /* EWALD ELECTROSTATICS */
 488
 489             /* Analytical PME correction */
 490             zeta2            = _mm_mul_ps(beta2,rsq13);
 491             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
 492             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 493             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 494             felec            = _mm_mul_ps(qq13,felec);
 495             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 496             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv13,sh_ewald));
 497             velec            = _mm_mul_ps(qq13,velec);
 498
 499             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
 500
 501             /* Update potential sum for this i atom from the interaction with this j atom. */
 502             velec            = _mm_and_ps(velec,cutoff_mask);
 503             velecsum         = _mm_add_ps(velecsum,velec);
 504
 505             fscal            = felec;
 506
 507             fscal            = _mm_and_ps(fscal,cutoff_mask);
 508
 509              /* Update vectorial force */
 510             fix1             = _mm_macc_ps(dx13,fscal,fix1);
 511             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
 512             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
 513
 514             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
 515             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
 516             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
 517
 518             }
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             if (gmx_mm_any_lt(rsq21,rcutoff2))
 525             {
 526
 527             r21              = _mm_mul_ps(rsq21,rinv21);
 528
 529             /* EWALD ELECTROSTATICS */
 530
 531             /* Analytical PME correction */
 532             zeta2            = _mm_mul_ps(beta2,rsq21);
 533             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
 534             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 535             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 536             felec            = _mm_mul_ps(qq21,felec);
 537             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 538             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv21,sh_ewald));
 539             velec            = _mm_mul_ps(qq21,velec);
 540
 541             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_and_ps(velec,cutoff_mask);
 545             velecsum         = _mm_add_ps(velecsum,velec);
 546
 547             fscal            = felec;
 548
 549             fscal            = _mm_and_ps(fscal,cutoff_mask);
 550
 551              /* Update vectorial force */
 552             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 553             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 554             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 555
 556             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 557             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 558             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 559
 560             }
 561
 562             /**************************
 563              * CALCULATE INTERACTIONS *
 564              **************************/
 565
 566             if (gmx_mm_any_lt(rsq22,rcutoff2))
 567             {
 568
 569             r22              = _mm_mul_ps(rsq22,rinv22);
 570
 571             /* EWALD ELECTROSTATICS */
 572
 573             /* Analytical PME correction */
 574             zeta2            = _mm_mul_ps(beta2,rsq22);
 575             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
 576             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 577             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 578             felec            = _mm_mul_ps(qq22,felec);
 579             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 580             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv22,sh_ewald));
 581             velec            = _mm_mul_ps(qq22,velec);
 582
 583             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velec            = _mm_and_ps(velec,cutoff_mask);
 587             velecsum         = _mm_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_and_ps(fscal,cutoff_mask);
 592
 593              /* Update vectorial force */
 594             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 595             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 596             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 597
 598             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 599             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 600             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 601
 602             }
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_mm_any_lt(rsq23,rcutoff2))
 609             {
 610
 611             r23              = _mm_mul_ps(rsq23,rinv23);
 612
 613             /* EWALD ELECTROSTATICS */
 614
 615             /* Analytical PME correction */
 616             zeta2            = _mm_mul_ps(beta2,rsq23);
 617             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
 618             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 619             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 620             felec            = _mm_mul_ps(qq23,felec);
 621             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 622             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv23,sh_ewald));
 623             velec            = _mm_mul_ps(qq23,velec);
 624
 625             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
 626
 627             /* Update potential sum for this i atom from the interaction with this j atom. */
 628             velec            = _mm_and_ps(velec,cutoff_mask);
 629             velecsum         = _mm_add_ps(velecsum,velec);
 630
 631             fscal            = felec;
 632
 633             fscal            = _mm_and_ps(fscal,cutoff_mask);
 634
 635              /* Update vectorial force */
 636             fix2             = _mm_macc_ps(dx23,fscal,fix2);
 637             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
 638             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
 639
 640             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
 641             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
 642             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
 643
 644             }
 645
 646             /**************************
 647              * CALCULATE INTERACTIONS *
 648              **************************/
 649
 650             if (gmx_mm_any_lt(rsq31,rcutoff2))
 651             {
 652
 653             r31              = _mm_mul_ps(rsq31,rinv31);
 654
 655             /* EWALD ELECTROSTATICS */
 656
 657             /* Analytical PME correction */
 658             zeta2            = _mm_mul_ps(beta2,rsq31);
 659             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
 660             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 661             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 662             felec            = _mm_mul_ps(qq31,felec);
 663             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 664             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv31,sh_ewald));
 665             velec            = _mm_mul_ps(qq31,velec);
 666
 667             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
 668
 669             /* Update potential sum for this i atom from the interaction with this j atom. */
 670             velec            = _mm_and_ps(velec,cutoff_mask);
 671             velecsum         = _mm_add_ps(velecsum,velec);
 672
 673             fscal            = felec;
 674
 675             fscal            = _mm_and_ps(fscal,cutoff_mask);
 676
 677              /* Update vectorial force */
 678             fix3             = _mm_macc_ps(dx31,fscal,fix3);
 679             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
 680             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
 681
 682             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
 683             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
 684             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
 685
 686             }
 687
 688             /**************************
 689              * CALCULATE INTERACTIONS *
 690              **************************/
 691
 692             if (gmx_mm_any_lt(rsq32,rcutoff2))
 693             {
 694
 695             r32              = _mm_mul_ps(rsq32,rinv32);
 696
 697             /* EWALD ELECTROSTATICS */
 698
 699             /* Analytical PME correction */
 700             zeta2            = _mm_mul_ps(beta2,rsq32);
 701             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
 702             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 703             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 704             felec            = _mm_mul_ps(qq32,felec);
 705             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 706             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv32,sh_ewald));
 707             velec            = _mm_mul_ps(qq32,velec);
 708
 709             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
 710
 711             /* Update potential sum for this i atom from the interaction with this j atom. */
 712             velec            = _mm_and_ps(velec,cutoff_mask);
 713             velecsum         = _mm_add_ps(velecsum,velec);
 714
 715             fscal            = felec;
 716
 717             fscal            = _mm_and_ps(fscal,cutoff_mask);
 718
 719              /* Update vectorial force */
 720             fix3             = _mm_macc_ps(dx32,fscal,fix3);
 721             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
 722             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
 723
 724             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
 725             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
 726             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
 727
 728             }
 729
 730             /**************************
 731              * CALCULATE INTERACTIONS *
 732              **************************/
 733
 734             if (gmx_mm_any_lt(rsq33,rcutoff2))
 735             {
 736
 737             r33              = _mm_mul_ps(rsq33,rinv33);
 738
 739             /* EWALD ELECTROSTATICS */
 740
 741             /* Analytical PME correction */
 742             zeta2            = _mm_mul_ps(beta2,rsq33);
 743             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
 744             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 745             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 746             felec            = _mm_mul_ps(qq33,felec);
 747             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 748             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv33,sh_ewald));
 749             velec            = _mm_mul_ps(qq33,velec);
 750
 751             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
 752
 753             /* Update potential sum for this i atom from the interaction with this j atom. */
 754             velec            = _mm_and_ps(velec,cutoff_mask);
 755             velecsum         = _mm_add_ps(velecsum,velec);
 756
 757             fscal            = felec;
 758
 759             fscal            = _mm_and_ps(fscal,cutoff_mask);
 760
 761              /* Update vectorial force */
 762             fix3             = _mm_macc_ps(dx33,fscal,fix3);
 763             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
 764             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
 765
 766             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
 767             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
 768             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
 769
 770             }
 771
 772             fjptrA             = f+j_coord_offsetA;
 773             fjptrB             = f+j_coord_offsetB;
 774             fjptrC             = f+j_coord_offsetC;
 775             fjptrD             = f+j_coord_offsetD;
 776
 777             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 778                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
 779                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
 780
 781             /* Inner loop uses 359 flops */
 782         }
 783
 784         if(jidx<j_index_end)
 785         {
 786
 787             /* Get j neighbor index, and coordinate index */
 788             jnrlistA         = jjnr[jidx];
 789             jnrlistB         = jjnr[jidx+1];
 790             jnrlistC         = jjnr[jidx+2];
 791             jnrlistD         = jjnr[jidx+3];
 792             /* Sign of each element will be negative for non-real atoms.
 793              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 794              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 795              */
 796             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 797             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 798             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 799             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 800             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 801             j_coord_offsetA  = DIM*jnrA;
 802             j_coord_offsetB  = DIM*jnrB;
 803             j_coord_offsetC  = DIM*jnrC;
 804             j_coord_offsetD  = DIM*jnrD;
 805
 806             /* load j atom coordinates */
 807             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 808                                               x+j_coord_offsetC,x+j_coord_offsetD,
 809                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
 810                                               &jy2,&jz2,&jx3,&jy3,&jz3);
 811
 812             /* Calculate displacement vector */
 813             dx00             = _mm_sub_ps(ix0,jx0);
 814             dy00             = _mm_sub_ps(iy0,jy0);
 815             dz00             = _mm_sub_ps(iz0,jz0);
 816             dx11             = _mm_sub_ps(ix1,jx1);
 817             dy11             = _mm_sub_ps(iy1,jy1);
 818             dz11             = _mm_sub_ps(iz1,jz1);
 819             dx12             = _mm_sub_ps(ix1,jx2);
 820             dy12             = _mm_sub_ps(iy1,jy2);
 821             dz12             = _mm_sub_ps(iz1,jz2);
 822             dx13             = _mm_sub_ps(ix1,jx3);
 823             dy13             = _mm_sub_ps(iy1,jy3);
 824             dz13             = _mm_sub_ps(iz1,jz3);
 825             dx21             = _mm_sub_ps(ix2,jx1);
 826             dy21             = _mm_sub_ps(iy2,jy1);
 827             dz21             = _mm_sub_ps(iz2,jz1);
 828             dx22             = _mm_sub_ps(ix2,jx2);
 829             dy22             = _mm_sub_ps(iy2,jy2);
 830             dz22             = _mm_sub_ps(iz2,jz2);
 831             dx23             = _mm_sub_ps(ix2,jx3);
 832             dy23             = _mm_sub_ps(iy2,jy3);
 833             dz23             = _mm_sub_ps(iz2,jz3);
 834             dx31             = _mm_sub_ps(ix3,jx1);
 835             dy31             = _mm_sub_ps(iy3,jy1);
 836             dz31             = _mm_sub_ps(iz3,jz1);
 837             dx32             = _mm_sub_ps(ix3,jx2);
 838             dy32             = _mm_sub_ps(iy3,jy2);
 839             dz32             = _mm_sub_ps(iz3,jz2);
 840             dx33             = _mm_sub_ps(ix3,jx3);
 841             dy33             = _mm_sub_ps(iy3,jy3);
 842             dz33             = _mm_sub_ps(iz3,jz3);
 843
 844             /* Calculate squared distance and things based on it */
 845             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 846             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 847             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 848             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
 849             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 850             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 851             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
 852             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
 853             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
 854             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
 855
 856             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 857             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 858             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 859             rinv13           = gmx_mm_invsqrt_ps(rsq13);
 860             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 861             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 862             rinv23           = gmx_mm_invsqrt_ps(rsq23);
 863             rinv31           = gmx_mm_invsqrt_ps(rsq31);
 864             rinv32           = gmx_mm_invsqrt_ps(rsq32);
 865             rinv33           = gmx_mm_invsqrt_ps(rsq33);
 866
 867             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 868             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 869             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 870             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
 871             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 872             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 873             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
 874             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
 875             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
 876             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
 877
 878             fjx0             = _mm_setzero_ps();
 879             fjy0             = _mm_setzero_ps();
 880             fjz0             = _mm_setzero_ps();
 881             fjx1             = _mm_setzero_ps();
 882             fjy1             = _mm_setzero_ps();
 883             fjz1             = _mm_setzero_ps();
 884             fjx2             = _mm_setzero_ps();
 885             fjy2             = _mm_setzero_ps();
 886             fjz2             = _mm_setzero_ps();
 887             fjx3             = _mm_setzero_ps();
 888             fjy3             = _mm_setzero_ps();
 889             fjz3             = _mm_setzero_ps();
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             if (gmx_mm_any_lt(rsq00,rcutoff2))
 896             {
 897
 898             r00              = _mm_mul_ps(rsq00,rinv00);
 899             r00              = _mm_andnot_ps(dummy_mask,r00);
 900
 901             /* Analytical LJ-PME */
 902             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 903             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 904             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 905             exponent         = gmx_simd_exp_r(ewcljrsq);
 906             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 907             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
 908             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 909             vvdw6            = _mm_mul_ps(_mm_macc_ps(-c6grid_00,_mm_sub_ps(one,poly),c6_00),rinvsix);
 910             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 911             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 912                                           _mm_mul_ps(_mm_sub_ps(vvdw6,_mm_macc_ps(c6grid_00,sh_lj_ewald,_mm_mul_ps(c6_00,sh_vdw_invrcut6))),one_sixth));
 913             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 914             fvdw             = _mm_mul_ps(_mm_add_ps(vvdw12,_mm_msub_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6),vvdw6)),rinvsq00);
 915
 916             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 917
 918             /* Update potential sum for this i atom from the interaction with this j atom. */
 919             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 920             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 921             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 922
 923             fscal            = fvdw;
 924
 925             fscal            = _mm_and_ps(fscal,cutoff_mask);
 926
 927             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 928
 929              /* Update vectorial force */
 930             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 931             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 932             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 933
 934             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 935             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 936             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 937
 938             }
 939
 940             /**************************
 941              * CALCULATE INTERACTIONS *
 942              **************************/
 943
 944             if (gmx_mm_any_lt(rsq11,rcutoff2))
 945             {
 946
 947             r11              = _mm_mul_ps(rsq11,rinv11);
 948             r11              = _mm_andnot_ps(dummy_mask,r11);
 949
 950             /* EWALD ELECTROSTATICS */
 951
 952             /* Analytical PME correction */
 953             zeta2            = _mm_mul_ps(beta2,rsq11);
 954             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
 955             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 956             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 957             felec            = _mm_mul_ps(qq11,felec);
 958             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 959             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv11,sh_ewald));
 960             velec            = _mm_mul_ps(qq11,velec);
 961
 962             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
 963
 964             /* Update potential sum for this i atom from the interaction with this j atom. */
 965             velec            = _mm_and_ps(velec,cutoff_mask);
 966             velec            = _mm_andnot_ps(dummy_mask,velec);
 967             velecsum         = _mm_add_ps(velecsum,velec);
 968
 969             fscal            = felec;
 970
 971             fscal            = _mm_and_ps(fscal,cutoff_mask);
 972
 973             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 974
 975              /* Update vectorial force */
 976             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 977             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 978             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 979
 980             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 981             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 982             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 983
 984             }
 985
 986             /**************************
 987              * CALCULATE INTERACTIONS *
 988              **************************/
 989
 990             if (gmx_mm_any_lt(rsq12,rcutoff2))
 991             {
 992
 993             r12              = _mm_mul_ps(rsq12,rinv12);
 994             r12              = _mm_andnot_ps(dummy_mask,r12);
 995
 996             /* EWALD ELECTROSTATICS */
 997
 998             /* Analytical PME correction */
 999             zeta2            = _mm_mul_ps(beta2,rsq12);
1000             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1001             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1002             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1003             felec            = _mm_mul_ps(qq12,felec);
1004             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1005             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv12,sh_ewald));
1006             velec            = _mm_mul_ps(qq12,velec);
1007
1008             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1009
1010             /* Update potential sum for this i atom from the interaction with this j atom. */
1011             velec            = _mm_and_ps(velec,cutoff_mask);
1012             velec            = _mm_andnot_ps(dummy_mask,velec);
1013             velecsum         = _mm_add_ps(velecsum,velec);
1014
1015             fscal            = felec;
1016
1017             fscal            = _mm_and_ps(fscal,cutoff_mask);
1018
1019             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1020
1021              /* Update vectorial force */
1022             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1023             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1024             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1025
1026             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1027             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1028             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1029
1030             }
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             if (gmx_mm_any_lt(rsq13,rcutoff2))
1037             {
1038
1039             r13              = _mm_mul_ps(rsq13,rinv13);
1040             r13              = _mm_andnot_ps(dummy_mask,r13);
1041
1042             /* EWALD ELECTROSTATICS */
1043
1044             /* Analytical PME correction */
1045             zeta2            = _mm_mul_ps(beta2,rsq13);
1046             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
1047             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1048             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1049             felec            = _mm_mul_ps(qq13,felec);
1050             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1051             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv13,sh_ewald));
1052             velec            = _mm_mul_ps(qq13,velec);
1053
1054             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1055
1056             /* Update potential sum for this i atom from the interaction with this j atom. */
1057             velec            = _mm_and_ps(velec,cutoff_mask);
1058             velec            = _mm_andnot_ps(dummy_mask,velec);
1059             velecsum         = _mm_add_ps(velecsum,velec);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm_and_ps(fscal,cutoff_mask);
1064
1065             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1066
1067              /* Update vectorial force */
1068             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1069             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1070             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1071
1072             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1073             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1074             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1075
1076             }
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             if (gmx_mm_any_lt(rsq21,rcutoff2))
1083             {
1084
1085             r21              = _mm_mul_ps(rsq21,rinv21);
1086             r21              = _mm_andnot_ps(dummy_mask,r21);
1087
1088             /* EWALD ELECTROSTATICS */
1089
1090             /* Analytical PME correction */
1091             zeta2            = _mm_mul_ps(beta2,rsq21);
1092             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1093             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1094             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1095             felec            = _mm_mul_ps(qq21,felec);
1096             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1097             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv21,sh_ewald));
1098             velec            = _mm_mul_ps(qq21,velec);
1099
1100             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1101
1102             /* Update potential sum for this i atom from the interaction with this j atom. */
1103             velec            = _mm_and_ps(velec,cutoff_mask);
1104             velec            = _mm_andnot_ps(dummy_mask,velec);
1105             velecsum         = _mm_add_ps(velecsum,velec);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_and_ps(fscal,cutoff_mask);
1110
1111             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1112
1113              /* Update vectorial force */
1114             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1115             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1116             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1117
1118             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1119             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1120             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1121
1122             }
1123
1124             /**************************
1125              * CALCULATE INTERACTIONS *
1126              **************************/
1127
1128             if (gmx_mm_any_lt(rsq22,rcutoff2))
1129             {
1130
1131             r22              = _mm_mul_ps(rsq22,rinv22);
1132             r22              = _mm_andnot_ps(dummy_mask,r22);
1133
1134             /* EWALD ELECTROSTATICS */
1135
1136             /* Analytical PME correction */
1137             zeta2            = _mm_mul_ps(beta2,rsq22);
1138             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1139             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1140             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1141             felec            = _mm_mul_ps(qq22,felec);
1142             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1143             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv22,sh_ewald));
1144             velec            = _mm_mul_ps(qq22,velec);
1145
1146             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1147
1148             /* Update potential sum for this i atom from the interaction with this j atom. */
1149             velec            = _mm_and_ps(velec,cutoff_mask);
1150             velec            = _mm_andnot_ps(dummy_mask,velec);
1151             velecsum         = _mm_add_ps(velecsum,velec);
1152
1153             fscal            = felec;
1154
1155             fscal            = _mm_and_ps(fscal,cutoff_mask);
1156
1157             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1158
1159              /* Update vectorial force */
1160             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1161             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1162             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1163
1164             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1165             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1166             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1167
1168             }
1169
1170             /**************************
1171              * CALCULATE INTERACTIONS *
1172              **************************/
1173
1174             if (gmx_mm_any_lt(rsq23,rcutoff2))
1175             {
1176
1177             r23              = _mm_mul_ps(rsq23,rinv23);
1178             r23              = _mm_andnot_ps(dummy_mask,r23);
1179
1180             /* EWALD ELECTROSTATICS */
1181
1182             /* Analytical PME correction */
1183             zeta2            = _mm_mul_ps(beta2,rsq23);
1184             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
1185             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1186             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1187             felec            = _mm_mul_ps(qq23,felec);
1188             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1189             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv23,sh_ewald));
1190             velec            = _mm_mul_ps(qq23,velec);
1191
1192             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1193
1194             /* Update potential sum for this i atom from the interaction with this j atom. */
1195             velec            = _mm_and_ps(velec,cutoff_mask);
1196             velec            = _mm_andnot_ps(dummy_mask,velec);
1197             velecsum         = _mm_add_ps(velecsum,velec);
1198
1199             fscal            = felec;
1200
1201             fscal            = _mm_and_ps(fscal,cutoff_mask);
1202
1203             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1204
1205              /* Update vectorial force */
1206             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1207             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1208             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1209
1210             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1211             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1212             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1213
1214             }
1215
1216             /**************************
1217              * CALCULATE INTERACTIONS *
1218              **************************/
1219
1220             if (gmx_mm_any_lt(rsq31,rcutoff2))
1221             {
1222
1223             r31              = _mm_mul_ps(rsq31,rinv31);
1224             r31              = _mm_andnot_ps(dummy_mask,r31);
1225
1226             /* EWALD ELECTROSTATICS */
1227
1228             /* Analytical PME correction */
1229             zeta2            = _mm_mul_ps(beta2,rsq31);
1230             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
1231             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1232             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1233             felec            = _mm_mul_ps(qq31,felec);
1234             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1235             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv31,sh_ewald));
1236             velec            = _mm_mul_ps(qq31,velec);
1237
1238             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1239
1240             /* Update potential sum for this i atom from the interaction with this j atom. */
1241             velec            = _mm_and_ps(velec,cutoff_mask);
1242             velec            = _mm_andnot_ps(dummy_mask,velec);
1243             velecsum         = _mm_add_ps(velecsum,velec);
1244
1245             fscal            = felec;
1246
1247             fscal            = _mm_and_ps(fscal,cutoff_mask);
1248
1249             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1250
1251              /* Update vectorial force */
1252             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1253             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1254             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1255
1256             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1257             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1258             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1259
1260             }
1261
1262             /**************************
1263              * CALCULATE INTERACTIONS *
1264              **************************/
1265
1266             if (gmx_mm_any_lt(rsq32,rcutoff2))
1267             {
1268
1269             r32              = _mm_mul_ps(rsq32,rinv32);
1270             r32              = _mm_andnot_ps(dummy_mask,r32);
1271
1272             /* EWALD ELECTROSTATICS */
1273
1274             /* Analytical PME correction */
1275             zeta2            = _mm_mul_ps(beta2,rsq32);
1276             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
1277             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1278             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1279             felec            = _mm_mul_ps(qq32,felec);
1280             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1281             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv32,sh_ewald));
1282             velec            = _mm_mul_ps(qq32,velec);
1283
1284             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1285
1286             /* Update potential sum for this i atom from the interaction with this j atom. */
1287             velec            = _mm_and_ps(velec,cutoff_mask);
1288             velec            = _mm_andnot_ps(dummy_mask,velec);
1289             velecsum         = _mm_add_ps(velecsum,velec);
1290
1291             fscal            = felec;
1292
1293             fscal            = _mm_and_ps(fscal,cutoff_mask);
1294
1295             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1296
1297              /* Update vectorial force */
1298             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1299             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1300             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1301
1302             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1303             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
1304             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
1305
1306             }
1307
1308             /**************************
1309              * CALCULATE INTERACTIONS *
1310              **************************/
1311
1312             if (gmx_mm_any_lt(rsq33,rcutoff2))
1313             {
1314
1315             r33              = _mm_mul_ps(rsq33,rinv33);
1316             r33              = _mm_andnot_ps(dummy_mask,r33);
1317
1318             /* EWALD ELECTROSTATICS */
1319
1320             /* Analytical PME correction */
1321             zeta2            = _mm_mul_ps(beta2,rsq33);
1322             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
1323             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1324             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1325             felec            = _mm_mul_ps(qq33,felec);
1326             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
1327             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv33,sh_ewald));
1328             velec            = _mm_mul_ps(qq33,velec);
1329
1330             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
1331
1332             /* Update potential sum for this i atom from the interaction with this j atom. */
1333             velec            = _mm_and_ps(velec,cutoff_mask);
1334             velec            = _mm_andnot_ps(dummy_mask,velec);
1335             velecsum         = _mm_add_ps(velecsum,velec);
1336
1337             fscal            = felec;
1338
1339             fscal            = _mm_and_ps(fscal,cutoff_mask);
1340
1341             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1342
1343              /* Update vectorial force */
1344             fix3             = _mm_macc_ps(dx33,fscal,fix3);
1345             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
1346             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
1347
1348             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
1349             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
1350             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
1351
1352             }
1353
1354             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1355             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1356             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1357             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1358
1359             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1360                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
1361                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
1362
1363             /* Inner loop uses 369 flops */
1364         }
1365
1366         /* End of innermost loop */
1367
1368         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1369                                               f+i_coord_offset,fshift+i_shift_offset);
1370
1371         ggid                        = gid[iidx];
1372         /* Update potential energies */
1373         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1374         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1375
1376         /* Increment number of inner iterations */
1377         inneriter                  += j_index_end - j_index_start;
1378
1379         /* Outer loop uses 26 flops */
1380     }
1381
1382     /* Increment number of outer iterations */
1383     outeriter        += nri;
1384
1385     /* Update outer/inner flops */
1386
1387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_VF,outeriter*26 + inneriter*369);
1388 }
1389 /*
1390  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_avx_128_fma_single
1391  * Electrostatics interaction: Ewald
1392  * VdW interaction:            LJEwald
1393  * Geometry:                   Water4-Water4
1394  * Calculate force/pot:        Force
1395  */
1396 void
1397 nb_kernel_ElecEwSh_VdwLJEwSh_GeomW4W4_F_avx_128_fma_single
1398                     (t_nblist                    * gmx_restrict       nlist,
1399                      rvec                        * gmx_restrict          xx,
1400                      rvec                        * gmx_restrict          ff,
1401                      t_forcerec                  * gmx_restrict          fr,
1402                      t_mdatoms                   * gmx_restrict     mdatoms,
1403                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1404                      t_nrnb                      * gmx_restrict        nrnb)
1405 {
1406     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1407      * just 0 for non-waters.
1408      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
1409      * jnr indices corresponding to data put in the four positions in the SIMD register.
1410      */
1411     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1412     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1413     int              jnrA,jnrB,jnrC,jnrD;
1414     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1415     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1416     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1417     real             rcutoff_scalar;
1418     real             *shiftvec,*fshift,*x,*f;
1419     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1420     real             scratch[4*DIM];
1421     __m128           fscal,rcutoff,rcutoff2,jidxall;
1422     int              vdwioffset0;
1423     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1424     int              vdwioffset1;
1425     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1426     int              vdwioffset2;
1427     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1428     int              vdwioffset3;
1429     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
1430     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1431     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1432     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1433     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1434     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1435     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1436     int              vdwjidx3A,vdwjidx3B,vdwjidx3C,vdwjidx3D;
1437     __m128           jx3,jy3,jz3,fjx3,fjy3,fjz3,jq3,isaj3;
1438     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1439     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1440     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1441     __m128           dx13,dy13,dz13,rsq13,rinv13,rinvsq13,r13,qq13,c6_13,c12_13;
1442     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1443     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1444     __m128           dx23,dy23,dz23,rsq23,rinv23,rinvsq23,r23,qq23,c6_23,c12_23;
1445     __m128           dx31,dy31,dz31,rsq31,rinv31,rinvsq31,r31,qq31,c6_31,c12_31;
1446     __m128           dx32,dy32,dz32,rsq32,rinv32,rinvsq32,r32,qq32,c6_32,c12_32;
1447     __m128           dx33,dy33,dz33,rsq33,rinv33,rinvsq33,r33,qq33,c6_33,c12_33;
1448     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1449     real             *charge;
1450     int              nvdwtype;
1451     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1452     int              *vdwtype;
1453     real             *vdwparam;
1454     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1455     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1456     __m128           c6grid_00;
1457     __m128           c6grid_11;
1458     __m128           c6grid_12;
1459     __m128           c6grid_13;
1460     __m128           c6grid_21;
1461     __m128           c6grid_22;
1462     __m128           c6grid_23;
1463     __m128           c6grid_31;
1464     __m128           c6grid_32;
1465     __m128           c6grid_33;
1466     real             *vdwgridparam;
1467     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1468     __m128           one_half = _mm_set1_ps(0.5);
1469     __m128           minus_one = _mm_set1_ps(-1.0);
1470     __m128i          ewitab;
1471     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1472     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
1473     real             *ewtab;
1474     __m128           dummy_mask,cutoff_mask;
1475     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1476     __m128           one     = _mm_set1_ps(1.0);
1477     __m128           two     = _mm_set1_ps(2.0);
1478     x                = xx[0];
1479     f                = ff[0];
1480
1481     nri              = nlist->nri;
1482     iinr             = nlist->iinr;
1483     jindex           = nlist->jindex;
1484     jjnr             = nlist->jjnr;
1485     shiftidx         = nlist->shift;
1486     gid              = nlist->gid;
1487     shiftvec         = fr->shift_vec[0];
1488     fshift           = fr->fshift[0];
1489     facel            = _mm_set1_ps(fr->epsfac);
1490     charge           = mdatoms->chargeA;
1491     nvdwtype         = fr->ntype;
1492     vdwparam         = fr->nbfp;
1493     vdwtype          = mdatoms->typeA;
1494     vdwgridparam     = fr->ljpme_c6grid;
1495     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
1496     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
1497     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
1498
1499     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1500     beta             = _mm_set1_ps(fr->ic->ewaldcoeff_q);
1501     beta2            = _mm_mul_ps(beta,beta);
1502     beta3            = _mm_mul_ps(beta,beta2);
1503     ewtab            = fr->ic->tabq_coul_F;
1504     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1505     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1506
1507     /* Setup water-specific parameters */
1508     inr              = nlist->iinr[0];
1509     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1510     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1511     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
1512     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1513
1514     jq1              = _mm_set1_ps(charge[inr+1]);
1515     jq2              = _mm_set1_ps(charge[inr+2]);
1516     jq3              = _mm_set1_ps(charge[inr+3]);
1517     vdwjidx0A        = 2*vdwtype[inr+0];
1518     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1519     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1520     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
1521     qq11             = _mm_mul_ps(iq1,jq1);
1522     qq12             = _mm_mul_ps(iq1,jq2);
1523     qq13             = _mm_mul_ps(iq1,jq3);
1524     qq21             = _mm_mul_ps(iq2,jq1);
1525     qq22             = _mm_mul_ps(iq2,jq2);
1526     qq23             = _mm_mul_ps(iq2,jq3);
1527     qq31             = _mm_mul_ps(iq3,jq1);
1528     qq32             = _mm_mul_ps(iq3,jq2);
1529     qq33             = _mm_mul_ps(iq3,jq3);
1530
1531     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
1532     rcutoff_scalar   = fr->rcoulomb;
1533     rcutoff          = _mm_set1_ps(rcutoff_scalar);
1534     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
1535
1536     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
1537     rvdw             = _mm_set1_ps(fr->rvdw);
1538
1539     /* Avoid stupid compiler warnings */
1540     jnrA = jnrB = jnrC = jnrD = 0;
1541     j_coord_offsetA = 0;
1542     j_coord_offsetB = 0;
1543     j_coord_offsetC = 0;
1544     j_coord_offsetD = 0;
1545
1546     outeriter        = 0;
1547     inneriter        = 0;
1548
1549     for(iidx=0;iidx<4*DIM;iidx++)
1550     {
1551         scratch[iidx] = 0.0;
1552     }
1553
1554     /* Start outer loop over neighborlists */
1555     for(iidx=0; iidx<nri; iidx++)
1556     {
1557         /* Load shift vector for this list */
1558         i_shift_offset   = DIM*shiftidx[iidx];
1559
1560         /* Load limits for loop over neighbors */
1561         j_index_start    = jindex[iidx];
1562         j_index_end      = jindex[iidx+1];
1563
1564         /* Get outer coordinate index */
1565         inr              = iinr[iidx];
1566         i_coord_offset   = DIM*inr;
1567
1568         /* Load i particle coords and add shift vector */
1569         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1570                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
1571
1572         fix0             = _mm_setzero_ps();
1573         fiy0             = _mm_setzero_ps();
1574         fiz0             = _mm_setzero_ps();
1575         fix1             = _mm_setzero_ps();
1576         fiy1             = _mm_setzero_ps();
1577         fiz1             = _mm_setzero_ps();
1578         fix2             = _mm_setzero_ps();
1579         fiy2             = _mm_setzero_ps();
1580         fiz2             = _mm_setzero_ps();
1581         fix3             = _mm_setzero_ps();
1582         fiy3             = _mm_setzero_ps();
1583         fiz3             = _mm_setzero_ps();
1584
1585         /* Start inner kernel loop */
1586         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1587         {
1588
1589             /* Get j neighbor index, and coordinate index */
1590             jnrA             = jjnr[jidx];
1591             jnrB             = jjnr[jidx+1];
1592             jnrC             = jjnr[jidx+2];
1593             jnrD             = jjnr[jidx+3];
1594             j_coord_offsetA  = DIM*jnrA;
1595             j_coord_offsetB  = DIM*jnrB;
1596             j_coord_offsetC  = DIM*jnrC;
1597             j_coord_offsetD  = DIM*jnrD;
1598
1599             /* load j atom coordinates */
1600             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1601                                               x+j_coord_offsetC,x+j_coord_offsetD,
1602                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
1603                                               &jy2,&jz2,&jx3,&jy3,&jz3);
1604
1605             /* Calculate displacement vector */
1606             dx00             = _mm_sub_ps(ix0,jx0);
1607             dy00             = _mm_sub_ps(iy0,jy0);
1608             dz00             = _mm_sub_ps(iz0,jz0);
1609             dx11             = _mm_sub_ps(ix1,jx1);
1610             dy11             = _mm_sub_ps(iy1,jy1);
1611             dz11             = _mm_sub_ps(iz1,jz1);
1612             dx12             = _mm_sub_ps(ix1,jx2);
1613             dy12             = _mm_sub_ps(iy1,jy2);
1614             dz12             = _mm_sub_ps(iz1,jz2);
1615             dx13             = _mm_sub_ps(ix1,jx3);
1616             dy13             = _mm_sub_ps(iy1,jy3);
1617             dz13             = _mm_sub_ps(iz1,jz3);
1618             dx21             = _mm_sub_ps(ix2,jx1);
1619             dy21             = _mm_sub_ps(iy2,jy1);
1620             dz21             = _mm_sub_ps(iz2,jz1);
1621             dx22             = _mm_sub_ps(ix2,jx2);
1622             dy22             = _mm_sub_ps(iy2,jy2);
1623             dz22             = _mm_sub_ps(iz2,jz2);
1624             dx23             = _mm_sub_ps(ix2,jx3);
1625             dy23             = _mm_sub_ps(iy2,jy3);
1626             dz23             = _mm_sub_ps(iz2,jz3);
1627             dx31             = _mm_sub_ps(ix3,jx1);
1628             dy31             = _mm_sub_ps(iy3,jy1);
1629             dz31             = _mm_sub_ps(iz3,jz1);
1630             dx32             = _mm_sub_ps(ix3,jx2);
1631             dy32             = _mm_sub_ps(iy3,jy2);
1632             dz32             = _mm_sub_ps(iz3,jz2);
1633             dx33             = _mm_sub_ps(ix3,jx3);
1634             dy33             = _mm_sub_ps(iy3,jy3);
1635             dz33             = _mm_sub_ps(iz3,jz3);
1636
1637             /* Calculate squared distance and things based on it */
1638             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1639             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1640             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1641             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
1642             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1643             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1644             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
1645             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
1646             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
1647             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
1648
1649             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1650             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1651             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1652             rinv13           = gmx_mm_invsqrt_ps(rsq13);
1653             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1654             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1655             rinv23           = gmx_mm_invsqrt_ps(rsq23);
1656             rinv31           = gmx_mm_invsqrt_ps(rsq31);
1657             rinv32           = gmx_mm_invsqrt_ps(rsq32);
1658             rinv33           = gmx_mm_invsqrt_ps(rsq33);
1659
1660             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1661             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1662             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1663             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
1664             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1665             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1666             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
1667             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
1668             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
1669             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
1670
1671             fjx0             = _mm_setzero_ps();
1672             fjy0             = _mm_setzero_ps();
1673             fjz0             = _mm_setzero_ps();
1674             fjx1             = _mm_setzero_ps();
1675             fjy1             = _mm_setzero_ps();
1676             fjz1             = _mm_setzero_ps();
1677             fjx2             = _mm_setzero_ps();
1678             fjy2             = _mm_setzero_ps();
1679             fjz2             = _mm_setzero_ps();
1680             fjx3             = _mm_setzero_ps();
1681             fjy3             = _mm_setzero_ps();
1682             fjz3             = _mm_setzero_ps();
1683
1684             /**************************
1685              * CALCULATE INTERACTIONS *
1686              **************************/
1687
1688             if (gmx_mm_any_lt(rsq00,rcutoff2))
1689             {
1690
1691             r00              = _mm_mul_ps(rsq00,rinv00);
1692
1693             /* Analytical LJ-PME */
1694             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1695             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1696             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1697             exponent         = gmx_simd_exp_r(ewcljrsq);
1698             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1699             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
1700             /* f6A = 6 * C6grid * (1 - poly) */
1701             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1702             /* f6B = C6grid * exponent * beta^6 */
1703             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1704             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1705             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1706
1707             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1708
1709             fscal            = fvdw;
1710
1711             fscal            = _mm_and_ps(fscal,cutoff_mask);
1712
1713              /* Update vectorial force */
1714             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1715             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1716             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1717
1718             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1719             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1720             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1721
1722             }
1723
1724             /**************************
1725              * CALCULATE INTERACTIONS *
1726              **************************/
1727
1728             if (gmx_mm_any_lt(rsq11,rcutoff2))
1729             {
1730
1731             r11              = _mm_mul_ps(rsq11,rinv11);
1732
1733             /* EWALD ELECTROSTATICS */
1734
1735             /* Analytical PME correction */
1736             zeta2            = _mm_mul_ps(beta2,rsq11);
1737             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
1738             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1739             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1740             felec            = _mm_mul_ps(qq11,felec);
1741
1742             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
1743
1744             fscal            = felec;
1745
1746             fscal            = _mm_and_ps(fscal,cutoff_mask);
1747
1748              /* Update vectorial force */
1749             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1750             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1751             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1752
1753             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1754             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1755             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1756
1757             }
1758
1759             /**************************
1760              * CALCULATE INTERACTIONS *
1761              **************************/
1762
1763             if (gmx_mm_any_lt(rsq12,rcutoff2))
1764             {
1765
1766             r12              = _mm_mul_ps(rsq12,rinv12);
1767
1768             /* EWALD ELECTROSTATICS */
1769
1770             /* Analytical PME correction */
1771             zeta2            = _mm_mul_ps(beta2,rsq12);
1772             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
1773             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1774             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1775             felec            = _mm_mul_ps(qq12,felec);
1776
1777             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
1778
1779             fscal            = felec;
1780
1781             fscal            = _mm_and_ps(fscal,cutoff_mask);
1782
1783              /* Update vectorial force */
1784             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1785             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1786             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1787
1788             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1789             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1790             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1791
1792             }
1793
1794             /**************************
1795              * CALCULATE INTERACTIONS *
1796              **************************/
1797
1798             if (gmx_mm_any_lt(rsq13,rcutoff2))
1799             {
1800
1801             r13              = _mm_mul_ps(rsq13,rinv13);
1802
1803             /* EWALD ELECTROSTATICS */
1804
1805             /* Analytical PME correction */
1806             zeta2            = _mm_mul_ps(beta2,rsq13);
1807             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
1808             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1809             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1810             felec            = _mm_mul_ps(qq13,felec);
1811
1812             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
1813
1814             fscal            = felec;
1815
1816             fscal            = _mm_and_ps(fscal,cutoff_mask);
1817
1818              /* Update vectorial force */
1819             fix1             = _mm_macc_ps(dx13,fscal,fix1);
1820             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
1821             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
1822
1823             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
1824             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
1825             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
1826
1827             }
1828
1829             /**************************
1830              * CALCULATE INTERACTIONS *
1831              **************************/
1832
1833             if (gmx_mm_any_lt(rsq21,rcutoff2))
1834             {
1835
1836             r21              = _mm_mul_ps(rsq21,rinv21);
1837
1838             /* EWALD ELECTROSTATICS */
1839
1840             /* Analytical PME correction */
1841             zeta2            = _mm_mul_ps(beta2,rsq21);
1842             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
1843             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1844             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1845             felec            = _mm_mul_ps(qq21,felec);
1846
1847             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
1848
1849             fscal            = felec;
1850
1851             fscal            = _mm_and_ps(fscal,cutoff_mask);
1852
1853              /* Update vectorial force */
1854             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1855             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1856             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1857
1858             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1859             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1860             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1861
1862             }
1863
1864             /**************************
1865              * CALCULATE INTERACTIONS *
1866              **************************/
1867
1868             if (gmx_mm_any_lt(rsq22,rcutoff2))
1869             {
1870
1871             r22              = _mm_mul_ps(rsq22,rinv22);
1872
1873             /* EWALD ELECTROSTATICS */
1874
1875             /* Analytical PME correction */
1876             zeta2            = _mm_mul_ps(beta2,rsq22);
1877             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
1878             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1879             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1880             felec            = _mm_mul_ps(qq22,felec);
1881
1882             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
1883
1884             fscal            = felec;
1885
1886             fscal            = _mm_and_ps(fscal,cutoff_mask);
1887
1888              /* Update vectorial force */
1889             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1890             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1891             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1892
1893             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1894             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1895             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1896
1897             }
1898
1899             /**************************
1900              * CALCULATE INTERACTIONS *
1901              **************************/
1902
1903             if (gmx_mm_any_lt(rsq23,rcutoff2))
1904             {
1905
1906             r23              = _mm_mul_ps(rsq23,rinv23);
1907
1908             /* EWALD ELECTROSTATICS */
1909
1910             /* Analytical PME correction */
1911             zeta2            = _mm_mul_ps(beta2,rsq23);
1912             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
1913             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1914             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1915             felec            = _mm_mul_ps(qq23,felec);
1916
1917             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
1918
1919             fscal            = felec;
1920
1921             fscal            = _mm_and_ps(fscal,cutoff_mask);
1922
1923              /* Update vectorial force */
1924             fix2             = _mm_macc_ps(dx23,fscal,fix2);
1925             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
1926             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
1927
1928             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
1929             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
1930             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
1931
1932             }
1933
1934             /**************************
1935              * CALCULATE INTERACTIONS *
1936              **************************/
1937
1938             if (gmx_mm_any_lt(rsq31,rcutoff2))
1939             {
1940
1941             r31              = _mm_mul_ps(rsq31,rinv31);
1942
1943             /* EWALD ELECTROSTATICS */
1944
1945             /* Analytical PME correction */
1946             zeta2            = _mm_mul_ps(beta2,rsq31);
1947             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
1948             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1949             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1950             felec            = _mm_mul_ps(qq31,felec);
1951
1952             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
1953
1954             fscal            = felec;
1955
1956             fscal            = _mm_and_ps(fscal,cutoff_mask);
1957
1958              /* Update vectorial force */
1959             fix3             = _mm_macc_ps(dx31,fscal,fix3);
1960             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
1961             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
1962
1963             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
1964             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
1965             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
1966
1967             }
1968
1969             /**************************
1970              * CALCULATE INTERACTIONS *
1971              **************************/
1972
1973             if (gmx_mm_any_lt(rsq32,rcutoff2))
1974             {
1975
1976             r32              = _mm_mul_ps(rsq32,rinv32);
1977
1978             /* EWALD ELECTROSTATICS */
1979
1980             /* Analytical PME correction */
1981             zeta2            = _mm_mul_ps(beta2,rsq32);
1982             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
1983             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1984             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1985             felec            = _mm_mul_ps(qq32,felec);
1986
1987             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
1988
1989             fscal            = felec;
1990
1991             fscal            = _mm_and_ps(fscal,cutoff_mask);
1992
1993              /* Update vectorial force */
1994             fix3             = _mm_macc_ps(dx32,fscal,fix3);
1995             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
1996             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
1997
1998             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
1999             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2000             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2001
2002             }
2003
2004             /**************************
2005              * CALCULATE INTERACTIONS *
2006              **************************/
2007
2008             if (gmx_mm_any_lt(rsq33,rcutoff2))
2009             {
2010
2011             r33              = _mm_mul_ps(rsq33,rinv33);
2012
2013             /* EWALD ELECTROSTATICS */
2014
2015             /* Analytical PME correction */
2016             zeta2            = _mm_mul_ps(beta2,rsq33);
2017             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
2018             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2019             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2020             felec            = _mm_mul_ps(qq33,felec);
2021
2022             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2023
2024             fscal            = felec;
2025
2026             fscal            = _mm_and_ps(fscal,cutoff_mask);
2027
2028              /* Update vectorial force */
2029             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2030             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2031             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2032
2033             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2034             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2035             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2036
2037             }
2038
2039             fjptrA             = f+j_coord_offsetA;
2040             fjptrB             = f+j_coord_offsetB;
2041             fjptrC             = f+j_coord_offsetC;
2042             fjptrD             = f+j_coord_offsetD;
2043
2044             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2045                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2046                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2047
2048             /* Inner loop uses 332 flops */
2049         }
2050
2051         if(jidx<j_index_end)
2052         {
2053
2054             /* Get j neighbor index, and coordinate index */
2055             jnrlistA         = jjnr[jidx];
2056             jnrlistB         = jjnr[jidx+1];
2057             jnrlistC         = jjnr[jidx+2];
2058             jnrlistD         = jjnr[jidx+3];
2059             /* Sign of each element will be negative for non-real atoms.
2060              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
2061              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
2062              */
2063             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
2064             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
2065             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
2066             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
2067             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
2068             j_coord_offsetA  = DIM*jnrA;
2069             j_coord_offsetB  = DIM*jnrB;
2070             j_coord_offsetC  = DIM*jnrC;
2071             j_coord_offsetD  = DIM*jnrD;
2072
2073             /* load j atom coordinates */
2074             gmx_mm_load_4rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
2075                                               x+j_coord_offsetC,x+j_coord_offsetD,
2076                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,
2077                                               &jy2,&jz2,&jx3,&jy3,&jz3);
2078
2079             /* Calculate displacement vector */
2080             dx00             = _mm_sub_ps(ix0,jx0);
2081             dy00             = _mm_sub_ps(iy0,jy0);
2082             dz00             = _mm_sub_ps(iz0,jz0);
2083             dx11             = _mm_sub_ps(ix1,jx1);
2084             dy11             = _mm_sub_ps(iy1,jy1);
2085             dz11             = _mm_sub_ps(iz1,jz1);
2086             dx12             = _mm_sub_ps(ix1,jx2);
2087             dy12             = _mm_sub_ps(iy1,jy2);
2088             dz12             = _mm_sub_ps(iz1,jz2);
2089             dx13             = _mm_sub_ps(ix1,jx3);
2090             dy13             = _mm_sub_ps(iy1,jy3);
2091             dz13             = _mm_sub_ps(iz1,jz3);
2092             dx21             = _mm_sub_ps(ix2,jx1);
2093             dy21             = _mm_sub_ps(iy2,jy1);
2094             dz21             = _mm_sub_ps(iz2,jz1);
2095             dx22             = _mm_sub_ps(ix2,jx2);
2096             dy22             = _mm_sub_ps(iy2,jy2);
2097             dz22             = _mm_sub_ps(iz2,jz2);
2098             dx23             = _mm_sub_ps(ix2,jx3);
2099             dy23             = _mm_sub_ps(iy2,jy3);
2100             dz23             = _mm_sub_ps(iz2,jz3);
2101             dx31             = _mm_sub_ps(ix3,jx1);
2102             dy31             = _mm_sub_ps(iy3,jy1);
2103             dz31             = _mm_sub_ps(iz3,jz1);
2104             dx32             = _mm_sub_ps(ix3,jx2);
2105             dy32             = _mm_sub_ps(iy3,jy2);
2106             dz32             = _mm_sub_ps(iz3,jz2);
2107             dx33             = _mm_sub_ps(ix3,jx3);
2108             dy33             = _mm_sub_ps(iy3,jy3);
2109             dz33             = _mm_sub_ps(iz3,jz3);
2110
2111             /* Calculate squared distance and things based on it */
2112             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
2113             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
2114             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
2115             rsq13            = gmx_mm_calc_rsq_ps(dx13,dy13,dz13);
2116             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
2117             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
2118             rsq23            = gmx_mm_calc_rsq_ps(dx23,dy23,dz23);
2119             rsq31            = gmx_mm_calc_rsq_ps(dx31,dy31,dz31);
2120             rsq32            = gmx_mm_calc_rsq_ps(dx32,dy32,dz32);
2121             rsq33            = gmx_mm_calc_rsq_ps(dx33,dy33,dz33);
2122
2123             rinv00           = gmx_mm_invsqrt_ps(rsq00);
2124             rinv11           = gmx_mm_invsqrt_ps(rsq11);
2125             rinv12           = gmx_mm_invsqrt_ps(rsq12);
2126             rinv13           = gmx_mm_invsqrt_ps(rsq13);
2127             rinv21           = gmx_mm_invsqrt_ps(rsq21);
2128             rinv22           = gmx_mm_invsqrt_ps(rsq22);
2129             rinv23           = gmx_mm_invsqrt_ps(rsq23);
2130             rinv31           = gmx_mm_invsqrt_ps(rsq31);
2131             rinv32           = gmx_mm_invsqrt_ps(rsq32);
2132             rinv33           = gmx_mm_invsqrt_ps(rsq33);
2133
2134             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
2135             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
2136             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
2137             rinvsq13         = _mm_mul_ps(rinv13,rinv13);
2138             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
2139             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
2140             rinvsq23         = _mm_mul_ps(rinv23,rinv23);
2141             rinvsq31         = _mm_mul_ps(rinv31,rinv31);
2142             rinvsq32         = _mm_mul_ps(rinv32,rinv32);
2143             rinvsq33         = _mm_mul_ps(rinv33,rinv33);
2144
2145             fjx0             = _mm_setzero_ps();
2146             fjy0             = _mm_setzero_ps();
2147             fjz0             = _mm_setzero_ps();
2148             fjx1             = _mm_setzero_ps();
2149             fjy1             = _mm_setzero_ps();
2150             fjz1             = _mm_setzero_ps();
2151             fjx2             = _mm_setzero_ps();
2152             fjy2             = _mm_setzero_ps();
2153             fjz2             = _mm_setzero_ps();
2154             fjx3             = _mm_setzero_ps();
2155             fjy3             = _mm_setzero_ps();
2156             fjz3             = _mm_setzero_ps();
2157
2158             /**************************
2159              * CALCULATE INTERACTIONS *
2160              **************************/
2161
2162             if (gmx_mm_any_lt(rsq00,rcutoff2))
2163             {
2164
2165             r00              = _mm_mul_ps(rsq00,rinv00);
2166             r00              = _mm_andnot_ps(dummy_mask,r00);
2167
2168             /* Analytical LJ-PME */
2169             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
2170             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
2171             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
2172             exponent         = gmx_simd_exp_r(ewcljrsq);
2173             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2174             poly             = _mm_mul_ps(exponent,_mm_macc_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half,_mm_sub_ps(one,ewcljrsq)));
2175             /* f6A = 6 * C6grid * (1 - poly) */
2176             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
2177             /* f6B = C6grid * exponent * beta^6 */
2178             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
2179             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2180             fvdw              = _mm_mul_ps(_mm_macc_ps(_mm_msub_ps(c12_00,rinvsix,_mm_sub_ps(c6_00,f6A)),rinvsix,f6B),rinvsq00);
2181
2182             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
2183
2184             fscal            = fvdw;
2185
2186             fscal            = _mm_and_ps(fscal,cutoff_mask);
2187
2188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2189
2190              /* Update vectorial force */
2191             fix0             = _mm_macc_ps(dx00,fscal,fix0);
2192             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
2193             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
2194
2195             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
2196             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
2197             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
2198
2199             }
2200
2201             /**************************
2202              * CALCULATE INTERACTIONS *
2203              **************************/
2204
2205             if (gmx_mm_any_lt(rsq11,rcutoff2))
2206             {
2207
2208             r11              = _mm_mul_ps(rsq11,rinv11);
2209             r11              = _mm_andnot_ps(dummy_mask,r11);
2210
2211             /* EWALD ELECTROSTATICS */
2212
2213             /* Analytical PME correction */
2214             zeta2            = _mm_mul_ps(beta2,rsq11);
2215             rinv3            = _mm_mul_ps(rinvsq11,rinv11);
2216             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2217             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2218             felec            = _mm_mul_ps(qq11,felec);
2219
2220             cutoff_mask      = _mm_cmplt_ps(rsq11,rcutoff2);
2221
2222             fscal            = felec;
2223
2224             fscal            = _mm_and_ps(fscal,cutoff_mask);
2225
2226             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2227
2228              /* Update vectorial force */
2229             fix1             = _mm_macc_ps(dx11,fscal,fix1);
2230             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
2231             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
2232
2233             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
2234             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
2235             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
2236
2237             }
2238
2239             /**************************
2240              * CALCULATE INTERACTIONS *
2241              **************************/
2242
2243             if (gmx_mm_any_lt(rsq12,rcutoff2))
2244             {
2245
2246             r12              = _mm_mul_ps(rsq12,rinv12);
2247             r12              = _mm_andnot_ps(dummy_mask,r12);
2248
2249             /* EWALD ELECTROSTATICS */
2250
2251             /* Analytical PME correction */
2252             zeta2            = _mm_mul_ps(beta2,rsq12);
2253             rinv3            = _mm_mul_ps(rinvsq12,rinv12);
2254             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2255             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2256             felec            = _mm_mul_ps(qq12,felec);
2257
2258             cutoff_mask      = _mm_cmplt_ps(rsq12,rcutoff2);
2259
2260             fscal            = felec;
2261
2262             fscal            = _mm_and_ps(fscal,cutoff_mask);
2263
2264             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2265
2266              /* Update vectorial force */
2267             fix1             = _mm_macc_ps(dx12,fscal,fix1);
2268             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
2269             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
2270
2271             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
2272             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
2273             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
2274
2275             }
2276
2277             /**************************
2278              * CALCULATE INTERACTIONS *
2279              **************************/
2280
2281             if (gmx_mm_any_lt(rsq13,rcutoff2))
2282             {
2283
2284             r13              = _mm_mul_ps(rsq13,rinv13);
2285             r13              = _mm_andnot_ps(dummy_mask,r13);
2286
2287             /* EWALD ELECTROSTATICS */
2288
2289             /* Analytical PME correction */
2290             zeta2            = _mm_mul_ps(beta2,rsq13);
2291             rinv3            = _mm_mul_ps(rinvsq13,rinv13);
2292             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2293             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2294             felec            = _mm_mul_ps(qq13,felec);
2295
2296             cutoff_mask      = _mm_cmplt_ps(rsq13,rcutoff2);
2297
2298             fscal            = felec;
2299
2300             fscal            = _mm_and_ps(fscal,cutoff_mask);
2301
2302             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2303
2304              /* Update vectorial force */
2305             fix1             = _mm_macc_ps(dx13,fscal,fix1);
2306             fiy1             = _mm_macc_ps(dy13,fscal,fiy1);
2307             fiz1             = _mm_macc_ps(dz13,fscal,fiz1);
2308
2309             fjx3             = _mm_macc_ps(dx13,fscal,fjx3);
2310             fjy3             = _mm_macc_ps(dy13,fscal,fjy3);
2311             fjz3             = _mm_macc_ps(dz13,fscal,fjz3);
2312
2313             }
2314
2315             /**************************
2316              * CALCULATE INTERACTIONS *
2317              **************************/
2318
2319             if (gmx_mm_any_lt(rsq21,rcutoff2))
2320             {
2321
2322             r21              = _mm_mul_ps(rsq21,rinv21);
2323             r21              = _mm_andnot_ps(dummy_mask,r21);
2324
2325             /* EWALD ELECTROSTATICS */
2326
2327             /* Analytical PME correction */
2328             zeta2            = _mm_mul_ps(beta2,rsq21);
2329             rinv3            = _mm_mul_ps(rinvsq21,rinv21);
2330             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2331             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2332             felec            = _mm_mul_ps(qq21,felec);
2333
2334             cutoff_mask      = _mm_cmplt_ps(rsq21,rcutoff2);
2335
2336             fscal            = felec;
2337
2338             fscal            = _mm_and_ps(fscal,cutoff_mask);
2339
2340             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2341
2342              /* Update vectorial force */
2343             fix2             = _mm_macc_ps(dx21,fscal,fix2);
2344             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
2345             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
2346
2347             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
2348             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
2349             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
2350
2351             }
2352
2353             /**************************
2354              * CALCULATE INTERACTIONS *
2355              **************************/
2356
2357             if (gmx_mm_any_lt(rsq22,rcutoff2))
2358             {
2359
2360             r22              = _mm_mul_ps(rsq22,rinv22);
2361             r22              = _mm_andnot_ps(dummy_mask,r22);
2362
2363             /* EWALD ELECTROSTATICS */
2364
2365             /* Analytical PME correction */
2366             zeta2            = _mm_mul_ps(beta2,rsq22);
2367             rinv3            = _mm_mul_ps(rinvsq22,rinv22);
2368             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2369             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2370             felec            = _mm_mul_ps(qq22,felec);
2371
2372             cutoff_mask      = _mm_cmplt_ps(rsq22,rcutoff2);
2373
2374             fscal            = felec;
2375
2376             fscal            = _mm_and_ps(fscal,cutoff_mask);
2377
2378             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2379
2380              /* Update vectorial force */
2381             fix2             = _mm_macc_ps(dx22,fscal,fix2);
2382             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
2383             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
2384
2385             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
2386             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
2387             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
2388
2389             }
2390
2391             /**************************
2392              * CALCULATE INTERACTIONS *
2393              **************************/
2394
2395             if (gmx_mm_any_lt(rsq23,rcutoff2))
2396             {
2397
2398             r23              = _mm_mul_ps(rsq23,rinv23);
2399             r23              = _mm_andnot_ps(dummy_mask,r23);
2400
2401             /* EWALD ELECTROSTATICS */
2402
2403             /* Analytical PME correction */
2404             zeta2            = _mm_mul_ps(beta2,rsq23);
2405             rinv3            = _mm_mul_ps(rinvsq23,rinv23);
2406             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2407             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2408             felec            = _mm_mul_ps(qq23,felec);
2409
2410             cutoff_mask      = _mm_cmplt_ps(rsq23,rcutoff2);
2411
2412             fscal            = felec;
2413
2414             fscal            = _mm_and_ps(fscal,cutoff_mask);
2415
2416             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2417
2418              /* Update vectorial force */
2419             fix2             = _mm_macc_ps(dx23,fscal,fix2);
2420             fiy2             = _mm_macc_ps(dy23,fscal,fiy2);
2421             fiz2             = _mm_macc_ps(dz23,fscal,fiz2);
2422
2423             fjx3             = _mm_macc_ps(dx23,fscal,fjx3);
2424             fjy3             = _mm_macc_ps(dy23,fscal,fjy3);
2425             fjz3             = _mm_macc_ps(dz23,fscal,fjz3);
2426
2427             }
2428
2429             /**************************
2430              * CALCULATE INTERACTIONS *
2431              **************************/
2432
2433             if (gmx_mm_any_lt(rsq31,rcutoff2))
2434             {
2435
2436             r31              = _mm_mul_ps(rsq31,rinv31);
2437             r31              = _mm_andnot_ps(dummy_mask,r31);
2438
2439             /* EWALD ELECTROSTATICS */
2440
2441             /* Analytical PME correction */
2442             zeta2            = _mm_mul_ps(beta2,rsq31);
2443             rinv3            = _mm_mul_ps(rinvsq31,rinv31);
2444             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2445             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2446             felec            = _mm_mul_ps(qq31,felec);
2447
2448             cutoff_mask      = _mm_cmplt_ps(rsq31,rcutoff2);
2449
2450             fscal            = felec;
2451
2452             fscal            = _mm_and_ps(fscal,cutoff_mask);
2453
2454             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2455
2456              /* Update vectorial force */
2457             fix3             = _mm_macc_ps(dx31,fscal,fix3);
2458             fiy3             = _mm_macc_ps(dy31,fscal,fiy3);
2459             fiz3             = _mm_macc_ps(dz31,fscal,fiz3);
2460
2461             fjx1             = _mm_macc_ps(dx31,fscal,fjx1);
2462             fjy1             = _mm_macc_ps(dy31,fscal,fjy1);
2463             fjz1             = _mm_macc_ps(dz31,fscal,fjz1);
2464
2465             }
2466
2467             /**************************
2468              * CALCULATE INTERACTIONS *
2469              **************************/
2470
2471             if (gmx_mm_any_lt(rsq32,rcutoff2))
2472             {
2473
2474             r32              = _mm_mul_ps(rsq32,rinv32);
2475             r32              = _mm_andnot_ps(dummy_mask,r32);
2476
2477             /* EWALD ELECTROSTATICS */
2478
2479             /* Analytical PME correction */
2480             zeta2            = _mm_mul_ps(beta2,rsq32);
2481             rinv3            = _mm_mul_ps(rinvsq32,rinv32);
2482             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2483             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2484             felec            = _mm_mul_ps(qq32,felec);
2485
2486             cutoff_mask      = _mm_cmplt_ps(rsq32,rcutoff2);
2487
2488             fscal            = felec;
2489
2490             fscal            = _mm_and_ps(fscal,cutoff_mask);
2491
2492             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2493
2494              /* Update vectorial force */
2495             fix3             = _mm_macc_ps(dx32,fscal,fix3);
2496             fiy3             = _mm_macc_ps(dy32,fscal,fiy3);
2497             fiz3             = _mm_macc_ps(dz32,fscal,fiz3);
2498
2499             fjx2             = _mm_macc_ps(dx32,fscal,fjx2);
2500             fjy2             = _mm_macc_ps(dy32,fscal,fjy2);
2501             fjz2             = _mm_macc_ps(dz32,fscal,fjz2);
2502
2503             }
2504
2505             /**************************
2506              * CALCULATE INTERACTIONS *
2507              **************************/
2508
2509             if (gmx_mm_any_lt(rsq33,rcutoff2))
2510             {
2511
2512             r33              = _mm_mul_ps(rsq33,rinv33);
2513             r33              = _mm_andnot_ps(dummy_mask,r33);
2514
2515             /* EWALD ELECTROSTATICS */
2516
2517             /* Analytical PME correction */
2518             zeta2            = _mm_mul_ps(beta2,rsq33);
2519             rinv3            = _mm_mul_ps(rinvsq33,rinv33);
2520             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
2521             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
2522             felec            = _mm_mul_ps(qq33,felec);
2523
2524             cutoff_mask      = _mm_cmplt_ps(rsq33,rcutoff2);
2525
2526             fscal            = felec;
2527
2528             fscal            = _mm_and_ps(fscal,cutoff_mask);
2529
2530             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2531
2532              /* Update vectorial force */
2533             fix3             = _mm_macc_ps(dx33,fscal,fix3);
2534             fiy3             = _mm_macc_ps(dy33,fscal,fiy3);
2535             fiz3             = _mm_macc_ps(dz33,fscal,fiz3);
2536
2537             fjx3             = _mm_macc_ps(dx33,fscal,fjx3);
2538             fjy3             = _mm_macc_ps(dy33,fscal,fjy3);
2539             fjz3             = _mm_macc_ps(dz33,fscal,fjz3);
2540
2541             }
2542
2543             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2544             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2545             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2546             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2547
2548             gmx_mm_decrement_4rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2549                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,
2550                                                    fjx2,fjy2,fjz2,fjx3,fjy3,fjz3);
2551
2552             /* Inner loop uses 342 flops */
2553         }
2554
2555         /* End of innermost loop */
2556
2557         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
2558                                               f+i_coord_offset,fshift+i_shift_offset);
2559
2560         /* Increment number of inner iterations */
2561         inneriter                  += j_index_end - j_index_start;
2562
2563         /* Outer loop uses 24 flops */
2564     }
2565
2566     /* Increment number of outer iterations */
2567     outeriter        += nri;
2568
2569     /* Update outer/inner flops */
2570
2571     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4W4_F,outeriter*24 + inneriter*342);
2572 }