src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJEw_GeomW3W3_sse4_1_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 sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3W3_VF_sse4_1_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Water3
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3W3_VF_sse4_1_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 SSE, 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           tx,ty,tz,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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  92     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  93     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  94     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  97     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  98     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  99     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 100     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 101     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 102     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 103     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 104     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 105     real             *charge;
 106     int              nvdwtype;
 107     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 108     int              *vdwtype;
 109     real             *vdwparam;
 110     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 111     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 112     __m128           c6grid_00;
 113     __m128           c6grid_01;
 114     __m128           c6grid_02;
 115     __m128           c6grid_10;
 116     __m128           c6grid_11;
 117     __m128           c6grid_12;
 118     __m128           c6grid_20;
 119     __m128           c6grid_21;
 120     __m128           c6grid_22;
 121     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 122     real             *vdwgridparam;
 123     __m128           one_half  = _mm_set1_ps(0.5);
 124     __m128           minus_one = _mm_set1_ps(-1.0);
 125     __m128i          ewitab;
 126     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 127     real             *ewtab;
 128     __m128           dummy_mask,cutoff_mask;
 129     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 130     __m128           one     = _mm_set1_ps(1.0);
 131     __m128           two     = _mm_set1_ps(2.0);
 132     x                = xx[0];
 133     f                = ff[0];
 134
 135     nri              = nlist->nri;
 136     iinr             = nlist->iinr;
 137     jindex           = nlist->jindex;
 138     jjnr             = nlist->jjnr;
 139     shiftidx         = nlist->shift;
 140     gid              = nlist->gid;
 141     shiftvec         = fr->shift_vec[0];
 142     fshift           = fr->fshift[0];
 143     facel            = _mm_set1_ps(fr->epsfac);
 144     charge           = mdatoms->chargeA;
 145     nvdwtype         = fr->ntype;
 146     vdwparam         = fr->nbfp;
 147     vdwtype          = mdatoms->typeA;
 148     vdwgridparam     = fr->ljpme_c6grid;
 149     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 150     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 151     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 152
 153     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 154     ewtab            = fr->ic->tabq_coul_FDV0;
 155     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 156     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 157
 158     /* Setup water-specific parameters */
 159     inr              = nlist->iinr[0];
 160     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 161     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 162     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 163     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 164
 165     jq0              = _mm_set1_ps(charge[inr+0]);
 166     jq1              = _mm_set1_ps(charge[inr+1]);
 167     jq2              = _mm_set1_ps(charge[inr+2]);
 168     vdwjidx0A        = 2*vdwtype[inr+0];
 169     qq00             = _mm_mul_ps(iq0,jq0);
 170     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 171     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 172     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
 173     qq01             = _mm_mul_ps(iq0,jq1);
 174     qq02             = _mm_mul_ps(iq0,jq2);
 175     qq10             = _mm_mul_ps(iq1,jq0);
 176     qq11             = _mm_mul_ps(iq1,jq1);
 177     qq12             = _mm_mul_ps(iq1,jq2);
 178     qq20             = _mm_mul_ps(iq2,jq0);
 179     qq21             = _mm_mul_ps(iq2,jq1);
 180     qq22             = _mm_mul_ps(iq2,jq2);
 181
 182     /* Avoid stupid compiler warnings */
 183     jnrA = jnrB = jnrC = jnrD = 0;
 184     j_coord_offsetA = 0;
 185     j_coord_offsetB = 0;
 186     j_coord_offsetC = 0;
 187     j_coord_offsetD = 0;
 188
 189     outeriter        = 0;
 190     inneriter        = 0;
 191
 192     for(iidx=0;iidx<4*DIM;iidx++)
 193     {
 194         scratch[iidx] = 0.0;
 195     }
 196
 197     /* Start outer loop over neighborlists */
 198     for(iidx=0; iidx<nri; iidx++)
 199     {
 200         /* Load shift vector for this list */
 201         i_shift_offset   = DIM*shiftidx[iidx];
 202
 203         /* Load limits for loop over neighbors */
 204         j_index_start    = jindex[iidx];
 205         j_index_end      = jindex[iidx+1];
 206
 207         /* Get outer coordinate index */
 208         inr              = iinr[iidx];
 209         i_coord_offset   = DIM*inr;
 210
 211         /* Load i particle coords and add shift vector */
 212         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 213                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 214
 215         fix0             = _mm_setzero_ps();
 216         fiy0             = _mm_setzero_ps();
 217         fiz0             = _mm_setzero_ps();
 218         fix1             = _mm_setzero_ps();
 219         fiy1             = _mm_setzero_ps();
 220         fiz1             = _mm_setzero_ps();
 221         fix2             = _mm_setzero_ps();
 222         fiy2             = _mm_setzero_ps();
 223         fiz2             = _mm_setzero_ps();
 224
 225         /* Reset potential sums */
 226         velecsum         = _mm_setzero_ps();
 227         vvdwsum          = _mm_setzero_ps();
 228
 229         /* Start inner kernel loop */
 230         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 231         {
 232
 233             /* Get j neighbor index, and coordinate index */
 234             jnrA             = jjnr[jidx];
 235             jnrB             = jjnr[jidx+1];
 236             jnrC             = jjnr[jidx+2];
 237             jnrD             = jjnr[jidx+3];
 238             j_coord_offsetA  = DIM*jnrA;
 239             j_coord_offsetB  = DIM*jnrB;
 240             j_coord_offsetC  = DIM*jnrC;
 241             j_coord_offsetD  = DIM*jnrD;
 242
 243             /* load j atom coordinates */
 244             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 245                                               x+j_coord_offsetC,x+j_coord_offsetD,
 246                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 247
 248             /* Calculate displacement vector */
 249             dx00             = _mm_sub_ps(ix0,jx0);
 250             dy00             = _mm_sub_ps(iy0,jy0);
 251             dz00             = _mm_sub_ps(iz0,jz0);
 252             dx01             = _mm_sub_ps(ix0,jx1);
 253             dy01             = _mm_sub_ps(iy0,jy1);
 254             dz01             = _mm_sub_ps(iz0,jz1);
 255             dx02             = _mm_sub_ps(ix0,jx2);
 256             dy02             = _mm_sub_ps(iy0,jy2);
 257             dz02             = _mm_sub_ps(iz0,jz2);
 258             dx10             = _mm_sub_ps(ix1,jx0);
 259             dy10             = _mm_sub_ps(iy1,jy0);
 260             dz10             = _mm_sub_ps(iz1,jz0);
 261             dx11             = _mm_sub_ps(ix1,jx1);
 262             dy11             = _mm_sub_ps(iy1,jy1);
 263             dz11             = _mm_sub_ps(iz1,jz1);
 264             dx12             = _mm_sub_ps(ix1,jx2);
 265             dy12             = _mm_sub_ps(iy1,jy2);
 266             dz12             = _mm_sub_ps(iz1,jz2);
 267             dx20             = _mm_sub_ps(ix2,jx0);
 268             dy20             = _mm_sub_ps(iy2,jy0);
 269             dz20             = _mm_sub_ps(iz2,jz0);
 270             dx21             = _mm_sub_ps(ix2,jx1);
 271             dy21             = _mm_sub_ps(iy2,jy1);
 272             dz21             = _mm_sub_ps(iz2,jz1);
 273             dx22             = _mm_sub_ps(ix2,jx2);
 274             dy22             = _mm_sub_ps(iy2,jy2);
 275             dz22             = _mm_sub_ps(iz2,jz2);
 276
 277             /* Calculate squared distance and things based on it */
 278             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 279             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 280             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 281             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 282             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 283             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 284             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 285             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 286             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 287
 288             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 289             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 290             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 291             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 292             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 293             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 294             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 295             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 296             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 297
 298             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 299             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 300             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 301             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 302             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 303             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 304             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 305             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 306             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 307
 308             fjx0             = _mm_setzero_ps();
 309             fjy0             = _mm_setzero_ps();
 310             fjz0             = _mm_setzero_ps();
 311             fjx1             = _mm_setzero_ps();
 312             fjy1             = _mm_setzero_ps();
 313             fjz1             = _mm_setzero_ps();
 314             fjx2             = _mm_setzero_ps();
 315             fjy2             = _mm_setzero_ps();
 316             fjz2             = _mm_setzero_ps();
 317
 318             /**************************
 319              * CALCULATE INTERACTIONS *
 320              **************************/
 321
 322             r00              = _mm_mul_ps(rsq00,rinv00);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm_mul_ps(r00,ewtabscale);
 328             ewitab           = _mm_cvttps_epi32(ewrt);
 329             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 333             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 334             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 335             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 336             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 337             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 338             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 339             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 340
 341             /* Analytical LJ-PME */
 342             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 343             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 344             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 345             exponent         = gmx_simd_exp_r(ewcljrsq);
 346             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 347             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 348             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 349             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 350             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 351             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 352             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 353             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm_add_ps(velecsum,velec);
 357             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 358
 359             fscal            = _mm_add_ps(felec,fvdw);
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm_mul_ps(fscal,dx00);
 363             ty               = _mm_mul_ps(fscal,dy00);
 364             tz               = _mm_mul_ps(fscal,dz00);
 365
 366             /* Update vectorial force */
 367             fix0             = _mm_add_ps(fix0,tx);
 368             fiy0             = _mm_add_ps(fiy0,ty);
 369             fiz0             = _mm_add_ps(fiz0,tz);
 370
 371             fjx0             = _mm_add_ps(fjx0,tx);
 372             fjy0             = _mm_add_ps(fjy0,ty);
 373             fjz0             = _mm_add_ps(fjz0,tz);
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             r01              = _mm_mul_ps(rsq01,rinv01);
 380
 381             /* EWALD ELECTROSTATICS */
 382
 383             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 384             ewrt             = _mm_mul_ps(r01,ewtabscale);
 385             ewitab           = _mm_cvttps_epi32(ewrt);
 386             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 387             ewitab           = _mm_slli_epi32(ewitab,2);
 388             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 389             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 390             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 391             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 392             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 393             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 394             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 395             velec            = _mm_mul_ps(qq01,_mm_sub_ps(rinv01,velec));
 396             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
 397
 398             /* Update potential sum for this i atom from the interaction with this j atom. */
 399             velecsum         = _mm_add_ps(velecsum,velec);
 400
 401             fscal            = felec;
 402
 403             /* Calculate temporary vectorial force */
 404             tx               = _mm_mul_ps(fscal,dx01);
 405             ty               = _mm_mul_ps(fscal,dy01);
 406             tz               = _mm_mul_ps(fscal,dz01);
 407
 408             /* Update vectorial force */
 409             fix0             = _mm_add_ps(fix0,tx);
 410             fiy0             = _mm_add_ps(fiy0,ty);
 411             fiz0             = _mm_add_ps(fiz0,tz);
 412
 413             fjx1             = _mm_add_ps(fjx1,tx);
 414             fjy1             = _mm_add_ps(fjy1,ty);
 415             fjz1             = _mm_add_ps(fjz1,tz);
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             r02              = _mm_mul_ps(rsq02,rinv02);
 422
 423             /* EWALD ELECTROSTATICS */
 424
 425             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 426             ewrt             = _mm_mul_ps(r02,ewtabscale);
 427             ewitab           = _mm_cvttps_epi32(ewrt);
 428             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 429             ewitab           = _mm_slli_epi32(ewitab,2);
 430             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 431             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 432             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 433             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 434             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 435             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 436             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 437             velec            = _mm_mul_ps(qq02,_mm_sub_ps(rinv02,velec));
 438             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
 439
 440             /* Update potential sum for this i atom from the interaction with this j atom. */
 441             velecsum         = _mm_add_ps(velecsum,velec);
 442
 443             fscal            = felec;
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm_mul_ps(fscal,dx02);
 447             ty               = _mm_mul_ps(fscal,dy02);
 448             tz               = _mm_mul_ps(fscal,dz02);
 449
 450             /* Update vectorial force */
 451             fix0             = _mm_add_ps(fix0,tx);
 452             fiy0             = _mm_add_ps(fiy0,ty);
 453             fiz0             = _mm_add_ps(fiz0,tz);
 454
 455             fjx2             = _mm_add_ps(fjx2,tx);
 456             fjy2             = _mm_add_ps(fjy2,ty);
 457             fjz2             = _mm_add_ps(fjz2,tz);
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             r10              = _mm_mul_ps(rsq10,rinv10);
 464
 465             /* EWALD ELECTROSTATICS */
 466
 467             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 468             ewrt             = _mm_mul_ps(r10,ewtabscale);
 469             ewitab           = _mm_cvttps_epi32(ewrt);
 470             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 471             ewitab           = _mm_slli_epi32(ewitab,2);
 472             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 473             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 474             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 475             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 476             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 477             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 478             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 479             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 480             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velecsum         = _mm_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             /* Calculate temporary vectorial force */
 488             tx               = _mm_mul_ps(fscal,dx10);
 489             ty               = _mm_mul_ps(fscal,dy10);
 490             tz               = _mm_mul_ps(fscal,dz10);
 491
 492             /* Update vectorial force */
 493             fix1             = _mm_add_ps(fix1,tx);
 494             fiy1             = _mm_add_ps(fiy1,ty);
 495             fiz1             = _mm_add_ps(fiz1,tz);
 496
 497             fjx0             = _mm_add_ps(fjx0,tx);
 498             fjy0             = _mm_add_ps(fjy0,ty);
 499             fjz0             = _mm_add_ps(fjz0,tz);
 500
 501             /**************************
 502              * CALCULATE INTERACTIONS *
 503              **************************/
 504
 505             r11              = _mm_mul_ps(rsq11,rinv11);
 506
 507             /* EWALD ELECTROSTATICS */
 508
 509             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 510             ewrt             = _mm_mul_ps(r11,ewtabscale);
 511             ewitab           = _mm_cvttps_epi32(ewrt);
 512             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 513             ewitab           = _mm_slli_epi32(ewitab,2);
 514             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 515             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 516             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 517             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 518             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 519             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 520             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 521             velec            = _mm_mul_ps(qq11,_mm_sub_ps(rinv11,velec));
 522             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velecsum         = _mm_add_ps(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_ps(fscal,dx11);
 531             ty               = _mm_mul_ps(fscal,dy11);
 532             tz               = _mm_mul_ps(fscal,dz11);
 533
 534             /* Update vectorial force */
 535             fix1             = _mm_add_ps(fix1,tx);
 536             fiy1             = _mm_add_ps(fiy1,ty);
 537             fiz1             = _mm_add_ps(fiz1,tz);
 538
 539             fjx1             = _mm_add_ps(fjx1,tx);
 540             fjy1             = _mm_add_ps(fjy1,ty);
 541             fjz1             = _mm_add_ps(fjz1,tz);
 542
 543             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             r12              = _mm_mul_ps(rsq12,rinv12);
 548
 549             /* EWALD ELECTROSTATICS */
 550
 551             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 552             ewrt             = _mm_mul_ps(r12,ewtabscale);
 553             ewitab           = _mm_cvttps_epi32(ewrt);
 554             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 555             ewitab           = _mm_slli_epi32(ewitab,2);
 556             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 557             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 558             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 559             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 560             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 561             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 562             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 563             velec            = _mm_mul_ps(qq12,_mm_sub_ps(rinv12,velec));
 564             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             velecsum         = _mm_add_ps(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             /* Calculate temporary vectorial force */
 572             tx               = _mm_mul_ps(fscal,dx12);
 573             ty               = _mm_mul_ps(fscal,dy12);
 574             tz               = _mm_mul_ps(fscal,dz12);
 575
 576             /* Update vectorial force */
 577             fix1             = _mm_add_ps(fix1,tx);
 578             fiy1             = _mm_add_ps(fiy1,ty);
 579             fiz1             = _mm_add_ps(fiz1,tz);
 580
 581             fjx2             = _mm_add_ps(fjx2,tx);
 582             fjy2             = _mm_add_ps(fjy2,ty);
 583             fjz2             = _mm_add_ps(fjz2,tz);
 584
 585             /**************************
 586              * CALCULATE INTERACTIONS *
 587              **************************/
 588
 589             r20              = _mm_mul_ps(rsq20,rinv20);
 590
 591             /* EWALD ELECTROSTATICS */
 592
 593             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 594             ewrt             = _mm_mul_ps(r20,ewtabscale);
 595             ewitab           = _mm_cvttps_epi32(ewrt);
 596             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 597             ewitab           = _mm_slli_epi32(ewitab,2);
 598             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 599             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 600             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 601             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 602             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 603             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 604             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 605             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 606             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velecsum         = _mm_add_ps(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             /* Calculate temporary vectorial force */
 614             tx               = _mm_mul_ps(fscal,dx20);
 615             ty               = _mm_mul_ps(fscal,dy20);
 616             tz               = _mm_mul_ps(fscal,dz20);
 617
 618             /* Update vectorial force */
 619             fix2             = _mm_add_ps(fix2,tx);
 620             fiy2             = _mm_add_ps(fiy2,ty);
 621             fiz2             = _mm_add_ps(fiz2,tz);
 622
 623             fjx0             = _mm_add_ps(fjx0,tx);
 624             fjy0             = _mm_add_ps(fjy0,ty);
 625             fjz0             = _mm_add_ps(fjz0,tz);
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             r21              = _mm_mul_ps(rsq21,rinv21);
 632
 633             /* EWALD ELECTROSTATICS */
 634
 635             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 636             ewrt             = _mm_mul_ps(r21,ewtabscale);
 637             ewitab           = _mm_cvttps_epi32(ewrt);
 638             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 639             ewitab           = _mm_slli_epi32(ewitab,2);
 640             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 641             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 642             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 643             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 644             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 645             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 646             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 647             velec            = _mm_mul_ps(qq21,_mm_sub_ps(rinv21,velec));
 648             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velecsum         = _mm_add_ps(velecsum,velec);
 652
 653             fscal            = felec;
 654
 655             /* Calculate temporary vectorial force */
 656             tx               = _mm_mul_ps(fscal,dx21);
 657             ty               = _mm_mul_ps(fscal,dy21);
 658             tz               = _mm_mul_ps(fscal,dz21);
 659
 660             /* Update vectorial force */
 661             fix2             = _mm_add_ps(fix2,tx);
 662             fiy2             = _mm_add_ps(fiy2,ty);
 663             fiz2             = _mm_add_ps(fiz2,tz);
 664
 665             fjx1             = _mm_add_ps(fjx1,tx);
 666             fjy1             = _mm_add_ps(fjy1,ty);
 667             fjz1             = _mm_add_ps(fjz1,tz);
 668
 669             /**************************
 670              * CALCULATE INTERACTIONS *
 671              **************************/
 672
 673             r22              = _mm_mul_ps(rsq22,rinv22);
 674
 675             /* EWALD ELECTROSTATICS */
 676
 677             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 678             ewrt             = _mm_mul_ps(r22,ewtabscale);
 679             ewitab           = _mm_cvttps_epi32(ewrt);
 680             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 681             ewitab           = _mm_slli_epi32(ewitab,2);
 682             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 683             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 684             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 685             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 686             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 687             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 688             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 689             velec            = _mm_mul_ps(qq22,_mm_sub_ps(rinv22,velec));
 690             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
 691
 692             /* Update potential sum for this i atom from the interaction with this j atom. */
 693             velecsum         = _mm_add_ps(velecsum,velec);
 694
 695             fscal            = felec;
 696
 697             /* Calculate temporary vectorial force */
 698             tx               = _mm_mul_ps(fscal,dx22);
 699             ty               = _mm_mul_ps(fscal,dy22);
 700             tz               = _mm_mul_ps(fscal,dz22);
 701
 702             /* Update vectorial force */
 703             fix2             = _mm_add_ps(fix2,tx);
 704             fiy2             = _mm_add_ps(fiy2,ty);
 705             fiz2             = _mm_add_ps(fiz2,tz);
 706
 707             fjx2             = _mm_add_ps(fjx2,tx);
 708             fjy2             = _mm_add_ps(fjy2,ty);
 709             fjz2             = _mm_add_ps(fjz2,tz);
 710
 711             fjptrA             = f+j_coord_offsetA;
 712             fjptrB             = f+j_coord_offsetB;
 713             fjptrC             = f+j_coord_offsetC;
 714             fjptrD             = f+j_coord_offsetD;
 715
 716             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 717                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 718
 719             /* Inner loop uses 397 flops */
 720         }
 721
 722         if(jidx<j_index_end)
 723         {
 724
 725             /* Get j neighbor index, and coordinate index */
 726             jnrlistA         = jjnr[jidx];
 727             jnrlistB         = jjnr[jidx+1];
 728             jnrlistC         = jjnr[jidx+2];
 729             jnrlistD         = jjnr[jidx+3];
 730             /* Sign of each element will be negative for non-real atoms.
 731              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 732              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 733              */
 734             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 735             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 736             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 737             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 738             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 739             j_coord_offsetA  = DIM*jnrA;
 740             j_coord_offsetB  = DIM*jnrB;
 741             j_coord_offsetC  = DIM*jnrC;
 742             j_coord_offsetD  = DIM*jnrD;
 743
 744             /* load j atom coordinates */
 745             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 746                                               x+j_coord_offsetC,x+j_coord_offsetD,
 747                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 748
 749             /* Calculate displacement vector */
 750             dx00             = _mm_sub_ps(ix0,jx0);
 751             dy00             = _mm_sub_ps(iy0,jy0);
 752             dz00             = _mm_sub_ps(iz0,jz0);
 753             dx01             = _mm_sub_ps(ix0,jx1);
 754             dy01             = _mm_sub_ps(iy0,jy1);
 755             dz01             = _mm_sub_ps(iz0,jz1);
 756             dx02             = _mm_sub_ps(ix0,jx2);
 757             dy02             = _mm_sub_ps(iy0,jy2);
 758             dz02             = _mm_sub_ps(iz0,jz2);
 759             dx10             = _mm_sub_ps(ix1,jx0);
 760             dy10             = _mm_sub_ps(iy1,jy0);
 761             dz10             = _mm_sub_ps(iz1,jz0);
 762             dx11             = _mm_sub_ps(ix1,jx1);
 763             dy11             = _mm_sub_ps(iy1,jy1);
 764             dz11             = _mm_sub_ps(iz1,jz1);
 765             dx12             = _mm_sub_ps(ix1,jx2);
 766             dy12             = _mm_sub_ps(iy1,jy2);
 767             dz12             = _mm_sub_ps(iz1,jz2);
 768             dx20             = _mm_sub_ps(ix2,jx0);
 769             dy20             = _mm_sub_ps(iy2,jy0);
 770             dz20             = _mm_sub_ps(iz2,jz0);
 771             dx21             = _mm_sub_ps(ix2,jx1);
 772             dy21             = _mm_sub_ps(iy2,jy1);
 773             dz21             = _mm_sub_ps(iz2,jz1);
 774             dx22             = _mm_sub_ps(ix2,jx2);
 775             dy22             = _mm_sub_ps(iy2,jy2);
 776             dz22             = _mm_sub_ps(iz2,jz2);
 777
 778             /* Calculate squared distance and things based on it */
 779             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 780             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 781             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 782             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 783             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 784             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 785             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 786             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 787             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 788
 789             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 790             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 791             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 792             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 793             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 794             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 795             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 796             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 797             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 798
 799             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 800             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 801             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 802             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 803             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 804             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 805             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 806             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 807             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 808
 809             fjx0             = _mm_setzero_ps();
 810             fjy0             = _mm_setzero_ps();
 811             fjz0             = _mm_setzero_ps();
 812             fjx1             = _mm_setzero_ps();
 813             fjy1             = _mm_setzero_ps();
 814             fjz1             = _mm_setzero_ps();
 815             fjx2             = _mm_setzero_ps();
 816             fjy2             = _mm_setzero_ps();
 817             fjz2             = _mm_setzero_ps();
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             r00              = _mm_mul_ps(rsq00,rinv00);
 824             r00              = _mm_andnot_ps(dummy_mask,r00);
 825
 826             /* EWALD ELECTROSTATICS */
 827
 828             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 829             ewrt             = _mm_mul_ps(r00,ewtabscale);
 830             ewitab           = _mm_cvttps_epi32(ewrt);
 831             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 832             ewitab           = _mm_slli_epi32(ewitab,2);
 833             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 834             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 835             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 836             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 837             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 838             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 839             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 840             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 841             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 842
 843             /* Analytical LJ-PME */
 844             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 845             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 846             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 847             exponent         = gmx_simd_exp_r(ewcljrsq);
 848             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 849             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 850             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 851             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 852             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 853             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 854             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 855             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,_mm_sub_ps(vvdw6,_mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6)))),rinvsq00);
 856
 857             /* Update potential sum for this i atom from the interaction with this j atom. */
 858             velec            = _mm_andnot_ps(dummy_mask,velec);
 859             velecsum         = _mm_add_ps(velecsum,velec);
 860             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 861             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 862
 863             fscal            = _mm_add_ps(felec,fvdw);
 864
 865             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 866
 867             /* Calculate temporary vectorial force */
 868             tx               = _mm_mul_ps(fscal,dx00);
 869             ty               = _mm_mul_ps(fscal,dy00);
 870             tz               = _mm_mul_ps(fscal,dz00);
 871
 872             /* Update vectorial force */
 873             fix0             = _mm_add_ps(fix0,tx);
 874             fiy0             = _mm_add_ps(fiy0,ty);
 875             fiz0             = _mm_add_ps(fiz0,tz);
 876
 877             fjx0             = _mm_add_ps(fjx0,tx);
 878             fjy0             = _mm_add_ps(fjy0,ty);
 879             fjz0             = _mm_add_ps(fjz0,tz);
 880
 881             /**************************
 882              * CALCULATE INTERACTIONS *
 883              **************************/
 884
 885             r01              = _mm_mul_ps(rsq01,rinv01);
 886             r01              = _mm_andnot_ps(dummy_mask,r01);
 887
 888             /* EWALD ELECTROSTATICS */
 889
 890             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 891             ewrt             = _mm_mul_ps(r01,ewtabscale);
 892             ewitab           = _mm_cvttps_epi32(ewrt);
 893             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 894             ewitab           = _mm_slli_epi32(ewitab,2);
 895             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 896             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 897             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 898             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 899             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 900             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 901             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 902             velec            = _mm_mul_ps(qq01,_mm_sub_ps(rinv01,velec));
 903             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
 904
 905             /* Update potential sum for this i atom from the interaction with this j atom. */
 906             velec            = _mm_andnot_ps(dummy_mask,velec);
 907             velecsum         = _mm_add_ps(velecsum,velec);
 908
 909             fscal            = felec;
 910
 911             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_ps(fscal,dx01);
 915             ty               = _mm_mul_ps(fscal,dy01);
 916             tz               = _mm_mul_ps(fscal,dz01);
 917
 918             /* Update vectorial force */
 919             fix0             = _mm_add_ps(fix0,tx);
 920             fiy0             = _mm_add_ps(fiy0,ty);
 921             fiz0             = _mm_add_ps(fiz0,tz);
 922
 923             fjx1             = _mm_add_ps(fjx1,tx);
 924             fjy1             = _mm_add_ps(fjy1,ty);
 925             fjz1             = _mm_add_ps(fjz1,tz);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             r02              = _mm_mul_ps(rsq02,rinv02);
 932             r02              = _mm_andnot_ps(dummy_mask,r02);
 933
 934             /* EWALD ELECTROSTATICS */
 935
 936             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 937             ewrt             = _mm_mul_ps(r02,ewtabscale);
 938             ewitab           = _mm_cvttps_epi32(ewrt);
 939             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 940             ewitab           = _mm_slli_epi32(ewitab,2);
 941             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 942             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 943             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 944             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 945             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 946             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 947             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 948             velec            = _mm_mul_ps(qq02,_mm_sub_ps(rinv02,velec));
 949             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
 950
 951             /* Update potential sum for this i atom from the interaction with this j atom. */
 952             velec            = _mm_andnot_ps(dummy_mask,velec);
 953             velecsum         = _mm_add_ps(velecsum,velec);
 954
 955             fscal            = felec;
 956
 957             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 958
 959             /* Calculate temporary vectorial force */
 960             tx               = _mm_mul_ps(fscal,dx02);
 961             ty               = _mm_mul_ps(fscal,dy02);
 962             tz               = _mm_mul_ps(fscal,dz02);
 963
 964             /* Update vectorial force */
 965             fix0             = _mm_add_ps(fix0,tx);
 966             fiy0             = _mm_add_ps(fiy0,ty);
 967             fiz0             = _mm_add_ps(fiz0,tz);
 968
 969             fjx2             = _mm_add_ps(fjx2,tx);
 970             fjy2             = _mm_add_ps(fjy2,ty);
 971             fjz2             = _mm_add_ps(fjz2,tz);
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             r10              = _mm_mul_ps(rsq10,rinv10);
 978             r10              = _mm_andnot_ps(dummy_mask,r10);
 979
 980             /* EWALD ELECTROSTATICS */
 981
 982             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 983             ewrt             = _mm_mul_ps(r10,ewtabscale);
 984             ewitab           = _mm_cvttps_epi32(ewrt);
 985             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 986             ewitab           = _mm_slli_epi32(ewitab,2);
 987             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 988             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 989             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 990             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 991             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 992             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 993             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 994             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 995             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 996
 997             /* Update potential sum for this i atom from the interaction with this j atom. */
 998             velec            = _mm_andnot_ps(dummy_mask,velec);
 999             velecsum         = _mm_add_ps(velecsum,velec);
1000
1001             fscal            = felec;
1002
1003             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1004
1005             /* Calculate temporary vectorial force */
1006             tx               = _mm_mul_ps(fscal,dx10);
1007             ty               = _mm_mul_ps(fscal,dy10);
1008             tz               = _mm_mul_ps(fscal,dz10);
1009
1010             /* Update vectorial force */
1011             fix1             = _mm_add_ps(fix1,tx);
1012             fiy1             = _mm_add_ps(fiy1,ty);
1013             fiz1             = _mm_add_ps(fiz1,tz);
1014
1015             fjx0             = _mm_add_ps(fjx0,tx);
1016             fjy0             = _mm_add_ps(fjy0,ty);
1017             fjz0             = _mm_add_ps(fjz0,tz);
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             r11              = _mm_mul_ps(rsq11,rinv11);
1024             r11              = _mm_andnot_ps(dummy_mask,r11);
1025
1026             /* EWALD ELECTROSTATICS */
1027
1028             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1029             ewrt             = _mm_mul_ps(r11,ewtabscale);
1030             ewitab           = _mm_cvttps_epi32(ewrt);
1031             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1032             ewitab           = _mm_slli_epi32(ewitab,2);
1033             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1034             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1035             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1036             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1037             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1038             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1039             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1040             velec            = _mm_mul_ps(qq11,_mm_sub_ps(rinv11,velec));
1041             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1042
1043             /* Update potential sum for this i atom from the interaction with this j atom. */
1044             velec            = _mm_andnot_ps(dummy_mask,velec);
1045             velecsum         = _mm_add_ps(velecsum,velec);
1046
1047             fscal            = felec;
1048
1049             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1050
1051             /* Calculate temporary vectorial force */
1052             tx               = _mm_mul_ps(fscal,dx11);
1053             ty               = _mm_mul_ps(fscal,dy11);
1054             tz               = _mm_mul_ps(fscal,dz11);
1055
1056             /* Update vectorial force */
1057             fix1             = _mm_add_ps(fix1,tx);
1058             fiy1             = _mm_add_ps(fiy1,ty);
1059             fiz1             = _mm_add_ps(fiz1,tz);
1060
1061             fjx1             = _mm_add_ps(fjx1,tx);
1062             fjy1             = _mm_add_ps(fjy1,ty);
1063             fjz1             = _mm_add_ps(fjz1,tz);
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             r12              = _mm_mul_ps(rsq12,rinv12);
1070             r12              = _mm_andnot_ps(dummy_mask,r12);
1071
1072             /* EWALD ELECTROSTATICS */
1073
1074             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1075             ewrt             = _mm_mul_ps(r12,ewtabscale);
1076             ewitab           = _mm_cvttps_epi32(ewrt);
1077             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1078             ewitab           = _mm_slli_epi32(ewitab,2);
1079             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1080             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1081             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1082             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1083             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1084             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1085             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1086             velec            = _mm_mul_ps(qq12,_mm_sub_ps(rinv12,velec));
1087             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1088
1089             /* Update potential sum for this i atom from the interaction with this j atom. */
1090             velec            = _mm_andnot_ps(dummy_mask,velec);
1091             velecsum         = _mm_add_ps(velecsum,velec);
1092
1093             fscal            = felec;
1094
1095             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1096
1097             /* Calculate temporary vectorial force */
1098             tx               = _mm_mul_ps(fscal,dx12);
1099             ty               = _mm_mul_ps(fscal,dy12);
1100             tz               = _mm_mul_ps(fscal,dz12);
1101
1102             /* Update vectorial force */
1103             fix1             = _mm_add_ps(fix1,tx);
1104             fiy1             = _mm_add_ps(fiy1,ty);
1105             fiz1             = _mm_add_ps(fiz1,tz);
1106
1107             fjx2             = _mm_add_ps(fjx2,tx);
1108             fjy2             = _mm_add_ps(fjy2,ty);
1109             fjz2             = _mm_add_ps(fjz2,tz);
1110
1111             /**************************
1112              * CALCULATE INTERACTIONS *
1113              **************************/
1114
1115             r20              = _mm_mul_ps(rsq20,rinv20);
1116             r20              = _mm_andnot_ps(dummy_mask,r20);
1117
1118             /* EWALD ELECTROSTATICS */
1119
1120             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1121             ewrt             = _mm_mul_ps(r20,ewtabscale);
1122             ewitab           = _mm_cvttps_epi32(ewrt);
1123             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1124             ewitab           = _mm_slli_epi32(ewitab,2);
1125             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1126             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1127             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1128             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1129             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1130             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1131             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1132             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1133             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1134
1135             /* Update potential sum for this i atom from the interaction with this j atom. */
1136             velec            = _mm_andnot_ps(dummy_mask,velec);
1137             velecsum         = _mm_add_ps(velecsum,velec);
1138
1139             fscal            = felec;
1140
1141             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1142
1143             /* Calculate temporary vectorial force */
1144             tx               = _mm_mul_ps(fscal,dx20);
1145             ty               = _mm_mul_ps(fscal,dy20);
1146             tz               = _mm_mul_ps(fscal,dz20);
1147
1148             /* Update vectorial force */
1149             fix2             = _mm_add_ps(fix2,tx);
1150             fiy2             = _mm_add_ps(fiy2,ty);
1151             fiz2             = _mm_add_ps(fiz2,tz);
1152
1153             fjx0             = _mm_add_ps(fjx0,tx);
1154             fjy0             = _mm_add_ps(fjy0,ty);
1155             fjz0             = _mm_add_ps(fjz0,tz);
1156
1157             /**************************
1158              * CALCULATE INTERACTIONS *
1159              **************************/
1160
1161             r21              = _mm_mul_ps(rsq21,rinv21);
1162             r21              = _mm_andnot_ps(dummy_mask,r21);
1163
1164             /* EWALD ELECTROSTATICS */
1165
1166             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1167             ewrt             = _mm_mul_ps(r21,ewtabscale);
1168             ewitab           = _mm_cvttps_epi32(ewrt);
1169             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1170             ewitab           = _mm_slli_epi32(ewitab,2);
1171             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1172             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1173             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1174             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1175             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1176             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1177             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1178             velec            = _mm_mul_ps(qq21,_mm_sub_ps(rinv21,velec));
1179             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1180
1181             /* Update potential sum for this i atom from the interaction with this j atom. */
1182             velec            = _mm_andnot_ps(dummy_mask,velec);
1183             velecsum         = _mm_add_ps(velecsum,velec);
1184
1185             fscal            = felec;
1186
1187             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1188
1189             /* Calculate temporary vectorial force */
1190             tx               = _mm_mul_ps(fscal,dx21);
1191             ty               = _mm_mul_ps(fscal,dy21);
1192             tz               = _mm_mul_ps(fscal,dz21);
1193
1194             /* Update vectorial force */
1195             fix2             = _mm_add_ps(fix2,tx);
1196             fiy2             = _mm_add_ps(fiy2,ty);
1197             fiz2             = _mm_add_ps(fiz2,tz);
1198
1199             fjx1             = _mm_add_ps(fjx1,tx);
1200             fjy1             = _mm_add_ps(fjy1,ty);
1201             fjz1             = _mm_add_ps(fjz1,tz);
1202
1203             /**************************
1204              * CALCULATE INTERACTIONS *
1205              **************************/
1206
1207             r22              = _mm_mul_ps(rsq22,rinv22);
1208             r22              = _mm_andnot_ps(dummy_mask,r22);
1209
1210             /* EWALD ELECTROSTATICS */
1211
1212             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1213             ewrt             = _mm_mul_ps(r22,ewtabscale);
1214             ewitab           = _mm_cvttps_epi32(ewrt);
1215             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1216             ewitab           = _mm_slli_epi32(ewitab,2);
1217             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1218             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1219             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1220             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1221             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1222             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1223             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1224             velec            = _mm_mul_ps(qq22,_mm_sub_ps(rinv22,velec));
1225             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1226
1227             /* Update potential sum for this i atom from the interaction with this j atom. */
1228             velec            = _mm_andnot_ps(dummy_mask,velec);
1229             velecsum         = _mm_add_ps(velecsum,velec);
1230
1231             fscal            = felec;
1232
1233             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1234
1235             /* Calculate temporary vectorial force */
1236             tx               = _mm_mul_ps(fscal,dx22);
1237             ty               = _mm_mul_ps(fscal,dy22);
1238             tz               = _mm_mul_ps(fscal,dz22);
1239
1240             /* Update vectorial force */
1241             fix2             = _mm_add_ps(fix2,tx);
1242             fiy2             = _mm_add_ps(fiy2,ty);
1243             fiz2             = _mm_add_ps(fiz2,tz);
1244
1245             fjx2             = _mm_add_ps(fjx2,tx);
1246             fjy2             = _mm_add_ps(fjy2,ty);
1247             fjz2             = _mm_add_ps(fjz2,tz);
1248
1249             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1250             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1251             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1252             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1253
1254             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1255                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1256
1257             /* Inner loop uses 406 flops */
1258         }
1259
1260         /* End of innermost loop */
1261
1262         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1263                                               f+i_coord_offset,fshift+i_shift_offset);
1264
1265         ggid                        = gid[iidx];
1266         /* Update potential energies */
1267         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
1268         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
1269
1270         /* Increment number of inner iterations */
1271         inneriter                  += j_index_end - j_index_start;
1272
1273         /* Outer loop uses 20 flops */
1274     }
1275
1276     /* Increment number of outer iterations */
1277     outeriter        += nri;
1278
1279     /* Update outer/inner flops */
1280
1281     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*406);
1282 }
1283 /*
1284  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3W3_F_sse4_1_single
1285  * Electrostatics interaction: Ewald
1286  * VdW interaction:            LJEwald
1287  * Geometry:                   Water3-Water3
1288  * Calculate force/pot:        Force
1289  */
1290 void
1291 nb_kernel_ElecEw_VdwLJEw_GeomW3W3_F_sse4_1_single
1292                     (t_nblist                    * gmx_restrict       nlist,
1293                      rvec                        * gmx_restrict          xx,
1294                      rvec                        * gmx_restrict          ff,
1295                      t_forcerec                  * gmx_restrict          fr,
1296                      t_mdatoms                   * gmx_restrict     mdatoms,
1297                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
1298                      t_nrnb                      * gmx_restrict        nrnb)
1299 {
1300     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
1301      * just 0 for non-waters.
1302      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
1303      * jnr indices corresponding to data put in the four positions in the SIMD register.
1304      */
1305     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
1306     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
1307     int              jnrA,jnrB,jnrC,jnrD;
1308     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
1309     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
1310     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
1311     real             rcutoff_scalar;
1312     real             *shiftvec,*fshift,*x,*f;
1313     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
1314     real             scratch[4*DIM];
1315     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
1316     int              vdwioffset0;
1317     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
1318     int              vdwioffset1;
1319     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
1320     int              vdwioffset2;
1321     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
1322     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
1323     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
1324     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
1325     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
1326     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
1327     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
1328     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
1329     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
1330     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
1331     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
1332     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
1333     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
1334     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
1335     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
1336     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
1337     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
1338     real             *charge;
1339     int              nvdwtype;
1340     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
1341     int              *vdwtype;
1342     real             *vdwparam;
1343     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
1344     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
1345     __m128           c6grid_00;
1346     __m128           c6grid_01;
1347     __m128           c6grid_02;
1348     __m128           c6grid_10;
1349     __m128           c6grid_11;
1350     __m128           c6grid_12;
1351     __m128           c6grid_20;
1352     __m128           c6grid_21;
1353     __m128           c6grid_22;
1354     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
1355     real             *vdwgridparam;
1356     __m128           one_half  = _mm_set1_ps(0.5);
1357     __m128           minus_one = _mm_set1_ps(-1.0);
1358     __m128i          ewitab;
1359     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
1360     real             *ewtab;
1361     __m128           dummy_mask,cutoff_mask;
1362     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
1363     __m128           one     = _mm_set1_ps(1.0);
1364     __m128           two     = _mm_set1_ps(2.0);
1365     x                = xx[0];
1366     f                = ff[0];
1367
1368     nri              = nlist->nri;
1369     iinr             = nlist->iinr;
1370     jindex           = nlist->jindex;
1371     jjnr             = nlist->jjnr;
1372     shiftidx         = nlist->shift;
1373     gid              = nlist->gid;
1374     shiftvec         = fr->shift_vec[0];
1375     fshift           = fr->fshift[0];
1376     facel            = _mm_set1_ps(fr->epsfac);
1377     charge           = mdatoms->chargeA;
1378     nvdwtype         = fr->ntype;
1379     vdwparam         = fr->nbfp;
1380     vdwtype          = mdatoms->typeA;
1381     vdwgridparam     = fr->ljpme_c6grid;
1382     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
1383     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
1384     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
1385
1386     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
1387     ewtab            = fr->ic->tabq_coul_F;
1388     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
1389     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
1390
1391     /* Setup water-specific parameters */
1392     inr              = nlist->iinr[0];
1393     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
1394     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
1395     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
1396     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
1397
1398     jq0              = _mm_set1_ps(charge[inr+0]);
1399     jq1              = _mm_set1_ps(charge[inr+1]);
1400     jq2              = _mm_set1_ps(charge[inr+2]);
1401     vdwjidx0A        = 2*vdwtype[inr+0];
1402     qq00             = _mm_mul_ps(iq0,jq0);
1403     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
1404     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
1405     c6grid_00        = _mm_set1_ps(vdwgridparam[vdwioffset0+vdwjidx0A]);
1406     qq01             = _mm_mul_ps(iq0,jq1);
1407     qq02             = _mm_mul_ps(iq0,jq2);
1408     qq10             = _mm_mul_ps(iq1,jq0);
1409     qq11             = _mm_mul_ps(iq1,jq1);
1410     qq12             = _mm_mul_ps(iq1,jq2);
1411     qq20             = _mm_mul_ps(iq2,jq0);
1412     qq21             = _mm_mul_ps(iq2,jq1);
1413     qq22             = _mm_mul_ps(iq2,jq2);
1414
1415     /* Avoid stupid compiler warnings */
1416     jnrA = jnrB = jnrC = jnrD = 0;
1417     j_coord_offsetA = 0;
1418     j_coord_offsetB = 0;
1419     j_coord_offsetC = 0;
1420     j_coord_offsetD = 0;
1421
1422     outeriter        = 0;
1423     inneriter        = 0;
1424
1425     for(iidx=0;iidx<4*DIM;iidx++)
1426     {
1427         scratch[iidx] = 0.0;
1428     }
1429
1430     /* Start outer loop over neighborlists */
1431     for(iidx=0; iidx<nri; iidx++)
1432     {
1433         /* Load shift vector for this list */
1434         i_shift_offset   = DIM*shiftidx[iidx];
1435
1436         /* Load limits for loop over neighbors */
1437         j_index_start    = jindex[iidx];
1438         j_index_end      = jindex[iidx+1];
1439
1440         /* Get outer coordinate index */
1441         inr              = iinr[iidx];
1442         i_coord_offset   = DIM*inr;
1443
1444         /* Load i particle coords and add shift vector */
1445         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1446                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1447
1448         fix0             = _mm_setzero_ps();
1449         fiy0             = _mm_setzero_ps();
1450         fiz0             = _mm_setzero_ps();
1451         fix1             = _mm_setzero_ps();
1452         fiy1             = _mm_setzero_ps();
1453         fiz1             = _mm_setzero_ps();
1454         fix2             = _mm_setzero_ps();
1455         fiy2             = _mm_setzero_ps();
1456         fiz2             = _mm_setzero_ps();
1457
1458         /* Start inner kernel loop */
1459         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1460         {
1461
1462             /* Get j neighbor index, and coordinate index */
1463             jnrA             = jjnr[jidx];
1464             jnrB             = jjnr[jidx+1];
1465             jnrC             = jjnr[jidx+2];
1466             jnrD             = jjnr[jidx+3];
1467             j_coord_offsetA  = DIM*jnrA;
1468             j_coord_offsetB  = DIM*jnrB;
1469             j_coord_offsetC  = DIM*jnrC;
1470             j_coord_offsetD  = DIM*jnrD;
1471
1472             /* load j atom coordinates */
1473             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1474                                               x+j_coord_offsetC,x+j_coord_offsetD,
1475                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1476
1477             /* Calculate displacement vector */
1478             dx00             = _mm_sub_ps(ix0,jx0);
1479             dy00             = _mm_sub_ps(iy0,jy0);
1480             dz00             = _mm_sub_ps(iz0,jz0);
1481             dx01             = _mm_sub_ps(ix0,jx1);
1482             dy01             = _mm_sub_ps(iy0,jy1);
1483             dz01             = _mm_sub_ps(iz0,jz1);
1484             dx02             = _mm_sub_ps(ix0,jx2);
1485             dy02             = _mm_sub_ps(iy0,jy2);
1486             dz02             = _mm_sub_ps(iz0,jz2);
1487             dx10             = _mm_sub_ps(ix1,jx0);
1488             dy10             = _mm_sub_ps(iy1,jy0);
1489             dz10             = _mm_sub_ps(iz1,jz0);
1490             dx11             = _mm_sub_ps(ix1,jx1);
1491             dy11             = _mm_sub_ps(iy1,jy1);
1492             dz11             = _mm_sub_ps(iz1,jz1);
1493             dx12             = _mm_sub_ps(ix1,jx2);
1494             dy12             = _mm_sub_ps(iy1,jy2);
1495             dz12             = _mm_sub_ps(iz1,jz2);
1496             dx20             = _mm_sub_ps(ix2,jx0);
1497             dy20             = _mm_sub_ps(iy2,jy0);
1498             dz20             = _mm_sub_ps(iz2,jz0);
1499             dx21             = _mm_sub_ps(ix2,jx1);
1500             dy21             = _mm_sub_ps(iy2,jy1);
1501             dz21             = _mm_sub_ps(iz2,jz1);
1502             dx22             = _mm_sub_ps(ix2,jx2);
1503             dy22             = _mm_sub_ps(iy2,jy2);
1504             dz22             = _mm_sub_ps(iz2,jz2);
1505
1506             /* Calculate squared distance and things based on it */
1507             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1508             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1509             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1510             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1511             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1512             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1513             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1514             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1515             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1516
1517             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1518             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1519             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1520             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1521             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1522             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1523             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1524             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1525             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1526
1527             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1528             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1529             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1530             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1531             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1532             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1533             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1534             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1535             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1536
1537             fjx0             = _mm_setzero_ps();
1538             fjy0             = _mm_setzero_ps();
1539             fjz0             = _mm_setzero_ps();
1540             fjx1             = _mm_setzero_ps();
1541             fjy1             = _mm_setzero_ps();
1542             fjz1             = _mm_setzero_ps();
1543             fjx2             = _mm_setzero_ps();
1544             fjy2             = _mm_setzero_ps();
1545             fjz2             = _mm_setzero_ps();
1546
1547             /**************************
1548              * CALCULATE INTERACTIONS *
1549              **************************/
1550
1551             r00              = _mm_mul_ps(rsq00,rinv00);
1552
1553             /* EWALD ELECTROSTATICS */
1554
1555             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1556             ewrt             = _mm_mul_ps(r00,ewtabscale);
1557             ewitab           = _mm_cvttps_epi32(ewrt);
1558             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1559             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1560                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1561                                          &ewtabF,&ewtabFn);
1562             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1563             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1564
1565             /* Analytical LJ-PME */
1566             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1567             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1568             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1569             exponent         = gmx_simd_exp_r(ewcljrsq);
1570             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1571             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1572             /* f6A = 6 * C6grid * (1 - poly) */
1573             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1574             /* f6B = C6grid * exponent * beta^6 */
1575             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1576             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1577             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
1578
1579             fscal            = _mm_add_ps(felec,fvdw);
1580
1581             /* Calculate temporary vectorial force */
1582             tx               = _mm_mul_ps(fscal,dx00);
1583             ty               = _mm_mul_ps(fscal,dy00);
1584             tz               = _mm_mul_ps(fscal,dz00);
1585
1586             /* Update vectorial force */
1587             fix0             = _mm_add_ps(fix0,tx);
1588             fiy0             = _mm_add_ps(fiy0,ty);
1589             fiz0             = _mm_add_ps(fiz0,tz);
1590
1591             fjx0             = _mm_add_ps(fjx0,tx);
1592             fjy0             = _mm_add_ps(fjy0,ty);
1593             fjz0             = _mm_add_ps(fjz0,tz);
1594
1595             /**************************
1596              * CALCULATE INTERACTIONS *
1597              **************************/
1598
1599             r01              = _mm_mul_ps(rsq01,rinv01);
1600
1601             /* EWALD ELECTROSTATICS */
1602
1603             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1604             ewrt             = _mm_mul_ps(r01,ewtabscale);
1605             ewitab           = _mm_cvttps_epi32(ewrt);
1606             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1607             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1608                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1609                                          &ewtabF,&ewtabFn);
1610             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1611             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
1612
1613             fscal            = felec;
1614
1615             /* Calculate temporary vectorial force */
1616             tx               = _mm_mul_ps(fscal,dx01);
1617             ty               = _mm_mul_ps(fscal,dy01);
1618             tz               = _mm_mul_ps(fscal,dz01);
1619
1620             /* Update vectorial force */
1621             fix0             = _mm_add_ps(fix0,tx);
1622             fiy0             = _mm_add_ps(fiy0,ty);
1623             fiz0             = _mm_add_ps(fiz0,tz);
1624
1625             fjx1             = _mm_add_ps(fjx1,tx);
1626             fjy1             = _mm_add_ps(fjy1,ty);
1627             fjz1             = _mm_add_ps(fjz1,tz);
1628
1629             /**************************
1630              * CALCULATE INTERACTIONS *
1631              **************************/
1632
1633             r02              = _mm_mul_ps(rsq02,rinv02);
1634
1635             /* EWALD ELECTROSTATICS */
1636
1637             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1638             ewrt             = _mm_mul_ps(r02,ewtabscale);
1639             ewitab           = _mm_cvttps_epi32(ewrt);
1640             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1641             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1642                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1643                                          &ewtabF,&ewtabFn);
1644             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1645             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
1646
1647             fscal            = felec;
1648
1649             /* Calculate temporary vectorial force */
1650             tx               = _mm_mul_ps(fscal,dx02);
1651             ty               = _mm_mul_ps(fscal,dy02);
1652             tz               = _mm_mul_ps(fscal,dz02);
1653
1654             /* Update vectorial force */
1655             fix0             = _mm_add_ps(fix0,tx);
1656             fiy0             = _mm_add_ps(fiy0,ty);
1657             fiz0             = _mm_add_ps(fiz0,tz);
1658
1659             fjx2             = _mm_add_ps(fjx2,tx);
1660             fjy2             = _mm_add_ps(fjy2,ty);
1661             fjz2             = _mm_add_ps(fjz2,tz);
1662
1663             /**************************
1664              * CALCULATE INTERACTIONS *
1665              **************************/
1666
1667             r10              = _mm_mul_ps(rsq10,rinv10);
1668
1669             /* EWALD ELECTROSTATICS */
1670
1671             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1672             ewrt             = _mm_mul_ps(r10,ewtabscale);
1673             ewitab           = _mm_cvttps_epi32(ewrt);
1674             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1675             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1676                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1677                                          &ewtabF,&ewtabFn);
1678             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1679             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1680
1681             fscal            = felec;
1682
1683             /* Calculate temporary vectorial force */
1684             tx               = _mm_mul_ps(fscal,dx10);
1685             ty               = _mm_mul_ps(fscal,dy10);
1686             tz               = _mm_mul_ps(fscal,dz10);
1687
1688             /* Update vectorial force */
1689             fix1             = _mm_add_ps(fix1,tx);
1690             fiy1             = _mm_add_ps(fiy1,ty);
1691             fiz1             = _mm_add_ps(fiz1,tz);
1692
1693             fjx0             = _mm_add_ps(fjx0,tx);
1694             fjy0             = _mm_add_ps(fjy0,ty);
1695             fjz0             = _mm_add_ps(fjz0,tz);
1696
1697             /**************************
1698              * CALCULATE INTERACTIONS *
1699              **************************/
1700
1701             r11              = _mm_mul_ps(rsq11,rinv11);
1702
1703             /* EWALD ELECTROSTATICS */
1704
1705             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1706             ewrt             = _mm_mul_ps(r11,ewtabscale);
1707             ewitab           = _mm_cvttps_epi32(ewrt);
1708             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1709             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1710                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1711                                          &ewtabF,&ewtabFn);
1712             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1713             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
1714
1715             fscal            = felec;
1716
1717             /* Calculate temporary vectorial force */
1718             tx               = _mm_mul_ps(fscal,dx11);
1719             ty               = _mm_mul_ps(fscal,dy11);
1720             tz               = _mm_mul_ps(fscal,dz11);
1721
1722             /* Update vectorial force */
1723             fix1             = _mm_add_ps(fix1,tx);
1724             fiy1             = _mm_add_ps(fiy1,ty);
1725             fiz1             = _mm_add_ps(fiz1,tz);
1726
1727             fjx1             = _mm_add_ps(fjx1,tx);
1728             fjy1             = _mm_add_ps(fjy1,ty);
1729             fjz1             = _mm_add_ps(fjz1,tz);
1730
1731             /**************************
1732              * CALCULATE INTERACTIONS *
1733              **************************/
1734
1735             r12              = _mm_mul_ps(rsq12,rinv12);
1736
1737             /* EWALD ELECTROSTATICS */
1738
1739             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1740             ewrt             = _mm_mul_ps(r12,ewtabscale);
1741             ewitab           = _mm_cvttps_epi32(ewrt);
1742             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1743             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1744                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1745                                          &ewtabF,&ewtabFn);
1746             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1747             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
1748
1749             fscal            = felec;
1750
1751             /* Calculate temporary vectorial force */
1752             tx               = _mm_mul_ps(fscal,dx12);
1753             ty               = _mm_mul_ps(fscal,dy12);
1754             tz               = _mm_mul_ps(fscal,dz12);
1755
1756             /* Update vectorial force */
1757             fix1             = _mm_add_ps(fix1,tx);
1758             fiy1             = _mm_add_ps(fiy1,ty);
1759             fiz1             = _mm_add_ps(fiz1,tz);
1760
1761             fjx2             = _mm_add_ps(fjx2,tx);
1762             fjy2             = _mm_add_ps(fjy2,ty);
1763             fjz2             = _mm_add_ps(fjz2,tz);
1764
1765             /**************************
1766              * CALCULATE INTERACTIONS *
1767              **************************/
1768
1769             r20              = _mm_mul_ps(rsq20,rinv20);
1770
1771             /* EWALD ELECTROSTATICS */
1772
1773             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1774             ewrt             = _mm_mul_ps(r20,ewtabscale);
1775             ewitab           = _mm_cvttps_epi32(ewrt);
1776             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1777             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1778                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1779                                          &ewtabF,&ewtabFn);
1780             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1781             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1782
1783             fscal            = felec;
1784
1785             /* Calculate temporary vectorial force */
1786             tx               = _mm_mul_ps(fscal,dx20);
1787             ty               = _mm_mul_ps(fscal,dy20);
1788             tz               = _mm_mul_ps(fscal,dz20);
1789
1790             /* Update vectorial force */
1791             fix2             = _mm_add_ps(fix2,tx);
1792             fiy2             = _mm_add_ps(fiy2,ty);
1793             fiz2             = _mm_add_ps(fiz2,tz);
1794
1795             fjx0             = _mm_add_ps(fjx0,tx);
1796             fjy0             = _mm_add_ps(fjy0,ty);
1797             fjz0             = _mm_add_ps(fjz0,tz);
1798
1799             /**************************
1800              * CALCULATE INTERACTIONS *
1801              **************************/
1802
1803             r21              = _mm_mul_ps(rsq21,rinv21);
1804
1805             /* EWALD ELECTROSTATICS */
1806
1807             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1808             ewrt             = _mm_mul_ps(r21,ewtabscale);
1809             ewitab           = _mm_cvttps_epi32(ewrt);
1810             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1811             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1812                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1813                                          &ewtabF,&ewtabFn);
1814             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1815             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
1816
1817             fscal            = felec;
1818
1819             /* Calculate temporary vectorial force */
1820             tx               = _mm_mul_ps(fscal,dx21);
1821             ty               = _mm_mul_ps(fscal,dy21);
1822             tz               = _mm_mul_ps(fscal,dz21);
1823
1824             /* Update vectorial force */
1825             fix2             = _mm_add_ps(fix2,tx);
1826             fiy2             = _mm_add_ps(fiy2,ty);
1827             fiz2             = _mm_add_ps(fiz2,tz);
1828
1829             fjx1             = _mm_add_ps(fjx1,tx);
1830             fjy1             = _mm_add_ps(fjy1,ty);
1831             fjz1             = _mm_add_ps(fjz1,tz);
1832
1833             /**************************
1834              * CALCULATE INTERACTIONS *
1835              **************************/
1836
1837             r22              = _mm_mul_ps(rsq22,rinv22);
1838
1839             /* EWALD ELECTROSTATICS */
1840
1841             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1842             ewrt             = _mm_mul_ps(r22,ewtabscale);
1843             ewitab           = _mm_cvttps_epi32(ewrt);
1844             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1845             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1846                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1847                                          &ewtabF,&ewtabFn);
1848             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1849             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
1850
1851             fscal            = felec;
1852
1853             /* Calculate temporary vectorial force */
1854             tx               = _mm_mul_ps(fscal,dx22);
1855             ty               = _mm_mul_ps(fscal,dy22);
1856             tz               = _mm_mul_ps(fscal,dz22);
1857
1858             /* Update vectorial force */
1859             fix2             = _mm_add_ps(fix2,tx);
1860             fiy2             = _mm_add_ps(fiy2,ty);
1861             fiz2             = _mm_add_ps(fiz2,tz);
1862
1863             fjx2             = _mm_add_ps(fjx2,tx);
1864             fjy2             = _mm_add_ps(fjy2,ty);
1865             fjz2             = _mm_add_ps(fjz2,tz);
1866
1867             fjptrA             = f+j_coord_offsetA;
1868             fjptrB             = f+j_coord_offsetB;
1869             fjptrC             = f+j_coord_offsetC;
1870             fjptrD             = f+j_coord_offsetD;
1871
1872             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1873                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1874
1875             /* Inner loop uses 347 flops */
1876         }
1877
1878         if(jidx<j_index_end)
1879         {
1880
1881             /* Get j neighbor index, and coordinate index */
1882             jnrlistA         = jjnr[jidx];
1883             jnrlistB         = jjnr[jidx+1];
1884             jnrlistC         = jjnr[jidx+2];
1885             jnrlistD         = jjnr[jidx+3];
1886             /* Sign of each element will be negative for non-real atoms.
1887              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1888              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1889              */
1890             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1891             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1892             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1893             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1894             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1895             j_coord_offsetA  = DIM*jnrA;
1896             j_coord_offsetB  = DIM*jnrB;
1897             j_coord_offsetC  = DIM*jnrC;
1898             j_coord_offsetD  = DIM*jnrD;
1899
1900             /* load j atom coordinates */
1901             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1902                                               x+j_coord_offsetC,x+j_coord_offsetD,
1903                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1904
1905             /* Calculate displacement vector */
1906             dx00             = _mm_sub_ps(ix0,jx0);
1907             dy00             = _mm_sub_ps(iy0,jy0);
1908             dz00             = _mm_sub_ps(iz0,jz0);
1909             dx01             = _mm_sub_ps(ix0,jx1);
1910             dy01             = _mm_sub_ps(iy0,jy1);
1911             dz01             = _mm_sub_ps(iz0,jz1);
1912             dx02             = _mm_sub_ps(ix0,jx2);
1913             dy02             = _mm_sub_ps(iy0,jy2);
1914             dz02             = _mm_sub_ps(iz0,jz2);
1915             dx10             = _mm_sub_ps(ix1,jx0);
1916             dy10             = _mm_sub_ps(iy1,jy0);
1917             dz10             = _mm_sub_ps(iz1,jz0);
1918             dx11             = _mm_sub_ps(ix1,jx1);
1919             dy11             = _mm_sub_ps(iy1,jy1);
1920             dz11             = _mm_sub_ps(iz1,jz1);
1921             dx12             = _mm_sub_ps(ix1,jx2);
1922             dy12             = _mm_sub_ps(iy1,jy2);
1923             dz12             = _mm_sub_ps(iz1,jz2);
1924             dx20             = _mm_sub_ps(ix2,jx0);
1925             dy20             = _mm_sub_ps(iy2,jy0);
1926             dz20             = _mm_sub_ps(iz2,jz0);
1927             dx21             = _mm_sub_ps(ix2,jx1);
1928             dy21             = _mm_sub_ps(iy2,jy1);
1929             dz21             = _mm_sub_ps(iz2,jz1);
1930             dx22             = _mm_sub_ps(ix2,jx2);
1931             dy22             = _mm_sub_ps(iy2,jy2);
1932             dz22             = _mm_sub_ps(iz2,jz2);
1933
1934             /* Calculate squared distance and things based on it */
1935             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1936             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1937             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1938             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1939             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1940             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1941             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1942             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1943             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1944
1945             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1946             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1947             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1948             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1949             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1950             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1951             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1952             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1953             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1954
1955             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1956             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1957             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1958             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1959             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1960             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1961             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1962             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1963             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1964
1965             fjx0             = _mm_setzero_ps();
1966             fjy0             = _mm_setzero_ps();
1967             fjz0             = _mm_setzero_ps();
1968             fjx1             = _mm_setzero_ps();
1969             fjy1             = _mm_setzero_ps();
1970             fjz1             = _mm_setzero_ps();
1971             fjx2             = _mm_setzero_ps();
1972             fjy2             = _mm_setzero_ps();
1973             fjz2             = _mm_setzero_ps();
1974
1975             /**************************
1976              * CALCULATE INTERACTIONS *
1977              **************************/
1978
1979             r00              = _mm_mul_ps(rsq00,rinv00);
1980             r00              = _mm_andnot_ps(dummy_mask,r00);
1981
1982             /* EWALD ELECTROSTATICS */
1983
1984             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1985             ewrt             = _mm_mul_ps(r00,ewtabscale);
1986             ewitab           = _mm_cvttps_epi32(ewrt);
1987             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1988             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1989                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1990                                          &ewtabF,&ewtabFn);
1991             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1992             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1993
1994             /* Analytical LJ-PME */
1995             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1996             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1997             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1998             exponent         = gmx_simd_exp_r(ewcljrsq);
1999             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
2000             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
2001             /* f6A = 6 * C6grid * (1 - poly) */
2002             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
2003             /* f6B = C6grid * exponent * beta^6 */
2004             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
2005             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
2006             fvdw              = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),_mm_sub_ps(c6_00,f6A)),rinvsix),f6B),rinvsq00);
2007
2008             fscal            = _mm_add_ps(felec,fvdw);
2009
2010             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2011
2012             /* Calculate temporary vectorial force */
2013             tx               = _mm_mul_ps(fscal,dx00);
2014             ty               = _mm_mul_ps(fscal,dy00);
2015             tz               = _mm_mul_ps(fscal,dz00);
2016
2017             /* Update vectorial force */
2018             fix0             = _mm_add_ps(fix0,tx);
2019             fiy0             = _mm_add_ps(fiy0,ty);
2020             fiz0             = _mm_add_ps(fiz0,tz);
2021
2022             fjx0             = _mm_add_ps(fjx0,tx);
2023             fjy0             = _mm_add_ps(fjy0,ty);
2024             fjz0             = _mm_add_ps(fjz0,tz);
2025
2026             /**************************
2027              * CALCULATE INTERACTIONS *
2028              **************************/
2029
2030             r01              = _mm_mul_ps(rsq01,rinv01);
2031             r01              = _mm_andnot_ps(dummy_mask,r01);
2032
2033             /* EWALD ELECTROSTATICS */
2034
2035             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2036             ewrt             = _mm_mul_ps(r01,ewtabscale);
2037             ewitab           = _mm_cvttps_epi32(ewrt);
2038             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2039             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2040                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2041                                          &ewtabF,&ewtabFn);
2042             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2043             felec            = _mm_mul_ps(_mm_mul_ps(qq01,rinv01),_mm_sub_ps(rinvsq01,felec));
2044
2045             fscal            = felec;
2046
2047             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2048
2049             /* Calculate temporary vectorial force */
2050             tx               = _mm_mul_ps(fscal,dx01);
2051             ty               = _mm_mul_ps(fscal,dy01);
2052             tz               = _mm_mul_ps(fscal,dz01);
2053
2054             /* Update vectorial force */
2055             fix0             = _mm_add_ps(fix0,tx);
2056             fiy0             = _mm_add_ps(fiy0,ty);
2057             fiz0             = _mm_add_ps(fiz0,tz);
2058
2059             fjx1             = _mm_add_ps(fjx1,tx);
2060             fjy1             = _mm_add_ps(fjy1,ty);
2061             fjz1             = _mm_add_ps(fjz1,tz);
2062
2063             /**************************
2064              * CALCULATE INTERACTIONS *
2065              **************************/
2066
2067             r02              = _mm_mul_ps(rsq02,rinv02);
2068             r02              = _mm_andnot_ps(dummy_mask,r02);
2069
2070             /* EWALD ELECTROSTATICS */
2071
2072             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2073             ewrt             = _mm_mul_ps(r02,ewtabscale);
2074             ewitab           = _mm_cvttps_epi32(ewrt);
2075             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2076             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2077                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2078                                          &ewtabF,&ewtabFn);
2079             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2080             felec            = _mm_mul_ps(_mm_mul_ps(qq02,rinv02),_mm_sub_ps(rinvsq02,felec));
2081
2082             fscal            = felec;
2083
2084             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2085
2086             /* Calculate temporary vectorial force */
2087             tx               = _mm_mul_ps(fscal,dx02);
2088             ty               = _mm_mul_ps(fscal,dy02);
2089             tz               = _mm_mul_ps(fscal,dz02);
2090
2091             /* Update vectorial force */
2092             fix0             = _mm_add_ps(fix0,tx);
2093             fiy0             = _mm_add_ps(fiy0,ty);
2094             fiz0             = _mm_add_ps(fiz0,tz);
2095
2096             fjx2             = _mm_add_ps(fjx2,tx);
2097             fjy2             = _mm_add_ps(fjy2,ty);
2098             fjz2             = _mm_add_ps(fjz2,tz);
2099
2100             /**************************
2101              * CALCULATE INTERACTIONS *
2102              **************************/
2103
2104             r10              = _mm_mul_ps(rsq10,rinv10);
2105             r10              = _mm_andnot_ps(dummy_mask,r10);
2106
2107             /* EWALD ELECTROSTATICS */
2108
2109             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2110             ewrt             = _mm_mul_ps(r10,ewtabscale);
2111             ewitab           = _mm_cvttps_epi32(ewrt);
2112             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2113             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2114                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2115                                          &ewtabF,&ewtabFn);
2116             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2117             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
2118
2119             fscal            = felec;
2120
2121             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2122
2123             /* Calculate temporary vectorial force */
2124             tx               = _mm_mul_ps(fscal,dx10);
2125             ty               = _mm_mul_ps(fscal,dy10);
2126             tz               = _mm_mul_ps(fscal,dz10);
2127
2128             /* Update vectorial force */
2129             fix1             = _mm_add_ps(fix1,tx);
2130             fiy1             = _mm_add_ps(fiy1,ty);
2131             fiz1             = _mm_add_ps(fiz1,tz);
2132
2133             fjx0             = _mm_add_ps(fjx0,tx);
2134             fjy0             = _mm_add_ps(fjy0,ty);
2135             fjz0             = _mm_add_ps(fjz0,tz);
2136
2137             /**************************
2138              * CALCULATE INTERACTIONS *
2139              **************************/
2140
2141             r11              = _mm_mul_ps(rsq11,rinv11);
2142             r11              = _mm_andnot_ps(dummy_mask,r11);
2143
2144             /* EWALD ELECTROSTATICS */
2145
2146             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2147             ewrt             = _mm_mul_ps(r11,ewtabscale);
2148             ewitab           = _mm_cvttps_epi32(ewrt);
2149             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2150             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2151                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2152                                          &ewtabF,&ewtabFn);
2153             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2154             felec            = _mm_mul_ps(_mm_mul_ps(qq11,rinv11),_mm_sub_ps(rinvsq11,felec));
2155
2156             fscal            = felec;
2157
2158             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2159
2160             /* Calculate temporary vectorial force */
2161             tx               = _mm_mul_ps(fscal,dx11);
2162             ty               = _mm_mul_ps(fscal,dy11);
2163             tz               = _mm_mul_ps(fscal,dz11);
2164
2165             /* Update vectorial force */
2166             fix1             = _mm_add_ps(fix1,tx);
2167             fiy1             = _mm_add_ps(fiy1,ty);
2168             fiz1             = _mm_add_ps(fiz1,tz);
2169
2170             fjx1             = _mm_add_ps(fjx1,tx);
2171             fjy1             = _mm_add_ps(fjy1,ty);
2172             fjz1             = _mm_add_ps(fjz1,tz);
2173
2174             /**************************
2175              * CALCULATE INTERACTIONS *
2176              **************************/
2177
2178             r12              = _mm_mul_ps(rsq12,rinv12);
2179             r12              = _mm_andnot_ps(dummy_mask,r12);
2180
2181             /* EWALD ELECTROSTATICS */
2182
2183             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2184             ewrt             = _mm_mul_ps(r12,ewtabscale);
2185             ewitab           = _mm_cvttps_epi32(ewrt);
2186             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2187             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2188                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2189                                          &ewtabF,&ewtabFn);
2190             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2191             felec            = _mm_mul_ps(_mm_mul_ps(qq12,rinv12),_mm_sub_ps(rinvsq12,felec));
2192
2193             fscal            = felec;
2194
2195             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2196
2197             /* Calculate temporary vectorial force */
2198             tx               = _mm_mul_ps(fscal,dx12);
2199             ty               = _mm_mul_ps(fscal,dy12);
2200             tz               = _mm_mul_ps(fscal,dz12);
2201
2202             /* Update vectorial force */
2203             fix1             = _mm_add_ps(fix1,tx);
2204             fiy1             = _mm_add_ps(fiy1,ty);
2205             fiz1             = _mm_add_ps(fiz1,tz);
2206
2207             fjx2             = _mm_add_ps(fjx2,tx);
2208             fjy2             = _mm_add_ps(fjy2,ty);
2209             fjz2             = _mm_add_ps(fjz2,tz);
2210
2211             /**************************
2212              * CALCULATE INTERACTIONS *
2213              **************************/
2214
2215             r20              = _mm_mul_ps(rsq20,rinv20);
2216             r20              = _mm_andnot_ps(dummy_mask,r20);
2217
2218             /* EWALD ELECTROSTATICS */
2219
2220             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2221             ewrt             = _mm_mul_ps(r20,ewtabscale);
2222             ewitab           = _mm_cvttps_epi32(ewrt);
2223             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2224             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2225                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2226                                          &ewtabF,&ewtabFn);
2227             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2228             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
2229
2230             fscal            = felec;
2231
2232             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2233
2234             /* Calculate temporary vectorial force */
2235             tx               = _mm_mul_ps(fscal,dx20);
2236             ty               = _mm_mul_ps(fscal,dy20);
2237             tz               = _mm_mul_ps(fscal,dz20);
2238
2239             /* Update vectorial force */
2240             fix2             = _mm_add_ps(fix2,tx);
2241             fiy2             = _mm_add_ps(fiy2,ty);
2242             fiz2             = _mm_add_ps(fiz2,tz);
2243
2244             fjx0             = _mm_add_ps(fjx0,tx);
2245             fjy0             = _mm_add_ps(fjy0,ty);
2246             fjz0             = _mm_add_ps(fjz0,tz);
2247
2248             /**************************
2249              * CALCULATE INTERACTIONS *
2250              **************************/
2251
2252             r21              = _mm_mul_ps(rsq21,rinv21);
2253             r21              = _mm_andnot_ps(dummy_mask,r21);
2254
2255             /* EWALD ELECTROSTATICS */
2256
2257             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2258             ewrt             = _mm_mul_ps(r21,ewtabscale);
2259             ewitab           = _mm_cvttps_epi32(ewrt);
2260             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2261             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2262                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2263                                          &ewtabF,&ewtabFn);
2264             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2265             felec            = _mm_mul_ps(_mm_mul_ps(qq21,rinv21),_mm_sub_ps(rinvsq21,felec));
2266
2267             fscal            = felec;
2268
2269             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2270
2271             /* Calculate temporary vectorial force */
2272             tx               = _mm_mul_ps(fscal,dx21);
2273             ty               = _mm_mul_ps(fscal,dy21);
2274             tz               = _mm_mul_ps(fscal,dz21);
2275
2276             /* Update vectorial force */
2277             fix2             = _mm_add_ps(fix2,tx);
2278             fiy2             = _mm_add_ps(fiy2,ty);
2279             fiz2             = _mm_add_ps(fiz2,tz);
2280
2281             fjx1             = _mm_add_ps(fjx1,tx);
2282             fjy1             = _mm_add_ps(fjy1,ty);
2283             fjz1             = _mm_add_ps(fjz1,tz);
2284
2285             /**************************
2286              * CALCULATE INTERACTIONS *
2287              **************************/
2288
2289             r22              = _mm_mul_ps(rsq22,rinv22);
2290             r22              = _mm_andnot_ps(dummy_mask,r22);
2291
2292             /* EWALD ELECTROSTATICS */
2293
2294             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
2295             ewrt             = _mm_mul_ps(r22,ewtabscale);
2296             ewitab           = _mm_cvttps_epi32(ewrt);
2297             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
2298             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
2299                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
2300                                          &ewtabF,&ewtabFn);
2301             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
2302             felec            = _mm_mul_ps(_mm_mul_ps(qq22,rinv22),_mm_sub_ps(rinvsq22,felec));
2303
2304             fscal            = felec;
2305
2306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
2307
2308             /* Calculate temporary vectorial force */
2309             tx               = _mm_mul_ps(fscal,dx22);
2310             ty               = _mm_mul_ps(fscal,dy22);
2311             tz               = _mm_mul_ps(fscal,dz22);
2312
2313             /* Update vectorial force */
2314             fix2             = _mm_add_ps(fix2,tx);
2315             fiy2             = _mm_add_ps(fiy2,ty);
2316             fiz2             = _mm_add_ps(fiz2,tz);
2317
2318             fjx2             = _mm_add_ps(fjx2,tx);
2319             fjy2             = _mm_add_ps(fjy2,ty);
2320             fjz2             = _mm_add_ps(fjz2,tz);
2321
2322             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
2323             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
2324             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
2325             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
2326
2327             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
2328                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
2329
2330             /* Inner loop uses 356 flops */
2331         }
2332
2333         /* End of innermost loop */
2334
2335         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
2336                                               f+i_coord_offset,fshift+i_shift_offset);
2337
2338         /* Increment number of inner iterations */
2339         inneriter                  += j_index_end - j_index_start;
2340
2341         /* Outer loop uses 18 flops */
2342     }
2343
2344     /* Increment number of outer iterations */
2345     outeriter        += nri;
2346
2347     /* Update outer/inner flops */
2348
2349     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*356);
2350 }