src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJEw_GeomW4P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LJEwald
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  91     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128           c6grid_00;
 105     __m128           c6grid_10;
 106     __m128           c6grid_20;
 107     __m128           c6grid_30;
 108     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 109     real             *vdwgridparam;
 110     __m128           one_half = _mm_set1_ps(0.5);
 111     __m128           minus_one = _mm_set1_ps(-1.0);
 112     __m128i          ewitab;
 113     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     real             *ewtab;
 115     __m128           dummy_mask,cutoff_mask;
 116     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 117     __m128           one     = _mm_set1_ps(1.0);
 118     __m128           two     = _mm_set1_ps(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_ps(fr->ic->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135     vdwgridparam     = fr->ljpme_c6grid;
 136     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 137     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 138     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 139
 140     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 148     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 149     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = jnrC = jnrD = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156     j_coord_offsetC = 0;
 157     j_coord_offsetD = 0;
 158
 159     outeriter        = 0;
 160     inneriter        = 0;
 161
 162     for(iidx=0;iidx<4*DIM;iidx++)
 163     {
 164         scratch[iidx] = 0.0;
 165     }
 166
 167     /* Start outer loop over neighborlists */
 168     for(iidx=0; iidx<nri; iidx++)
 169     {
 170         /* Load shift vector for this list */
 171         i_shift_offset   = DIM*shiftidx[iidx];
 172
 173         /* Load limits for loop over neighbors */
 174         j_index_start    = jindex[iidx];
 175         j_index_end      = jindex[iidx+1];
 176
 177         /* Get outer coordinate index */
 178         inr              = iinr[iidx];
 179         i_coord_offset   = DIM*inr;
 180
 181         /* Load i particle coords and add shift vector */
 182         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 183                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 184
 185         fix0             = _mm_setzero_ps();
 186         fiy0             = _mm_setzero_ps();
 187         fiz0             = _mm_setzero_ps();
 188         fix1             = _mm_setzero_ps();
 189         fiy1             = _mm_setzero_ps();
 190         fiz1             = _mm_setzero_ps();
 191         fix2             = _mm_setzero_ps();
 192         fiy2             = _mm_setzero_ps();
 193         fiz2             = _mm_setzero_ps();
 194         fix3             = _mm_setzero_ps();
 195         fiy3             = _mm_setzero_ps();
 196         fiz3             = _mm_setzero_ps();
 197
 198         /* Reset potential sums */
 199         velecsum         = _mm_setzero_ps();
 200         vvdwsum          = _mm_setzero_ps();
 201
 202         /* Start inner kernel loop */
 203         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 204         {
 205
 206             /* Get j neighbor index, and coordinate index */
 207             jnrA             = jjnr[jidx];
 208             jnrB             = jjnr[jidx+1];
 209             jnrC             = jjnr[jidx+2];
 210             jnrD             = jjnr[jidx+3];
 211             j_coord_offsetA  = DIM*jnrA;
 212             j_coord_offsetB  = DIM*jnrB;
 213             j_coord_offsetC  = DIM*jnrC;
 214             j_coord_offsetD  = DIM*jnrD;
 215
 216             /* load j atom coordinates */
 217             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 218                                               x+j_coord_offsetC,x+j_coord_offsetD,
 219                                               &jx0,&jy0,&jz0);
 220
 221             /* Calculate displacement vector */
 222             dx00             = _mm_sub_ps(ix0,jx0);
 223             dy00             = _mm_sub_ps(iy0,jy0);
 224             dz00             = _mm_sub_ps(iz0,jz0);
 225             dx10             = _mm_sub_ps(ix1,jx0);
 226             dy10             = _mm_sub_ps(iy1,jy0);
 227             dz10             = _mm_sub_ps(iz1,jz0);
 228             dx20             = _mm_sub_ps(ix2,jx0);
 229             dy20             = _mm_sub_ps(iy2,jy0);
 230             dz20             = _mm_sub_ps(iz2,jz0);
 231             dx30             = _mm_sub_ps(ix3,jx0);
 232             dy30             = _mm_sub_ps(iy3,jy0);
 233             dz30             = _mm_sub_ps(iz3,jz0);
 234
 235             /* Calculate squared distance and things based on it */
 236             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 237             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 238             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 239             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 240
 241             rinv00           = sse2_invsqrt_f(rsq00);
 242             rinv10           = sse2_invsqrt_f(rsq10);
 243             rinv20           = sse2_invsqrt_f(rsq20);
 244             rinv30           = sse2_invsqrt_f(rsq30);
 245
 246             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 247             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 248             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 249             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 250
 251             /* Load parameters for j particles */
 252             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 253                                                               charge+jnrC+0,charge+jnrD+0);
 254             vdwjidx0A        = 2*vdwtype[jnrA+0];
 255             vdwjidx0B        = 2*vdwtype[jnrB+0];
 256             vdwjidx0C        = 2*vdwtype[jnrC+0];
 257             vdwjidx0D        = 2*vdwtype[jnrD+0];
 258
 259             fjx0             = _mm_setzero_ps();
 260             fjy0             = _mm_setzero_ps();
 261             fjz0             = _mm_setzero_ps();
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r00              = _mm_mul_ps(rsq00,rinv00);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 271                                          vdwparam+vdwioffset0+vdwjidx0B,
 272                                          vdwparam+vdwioffset0+vdwjidx0C,
 273                                          vdwparam+vdwioffset0+vdwjidx0D,
 274                                          &c6_00,&c12_00);
 275             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 276                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 277                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 278                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 279
 280             /* Analytical LJ-PME */
 281             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 282             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 283             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 284             exponent         = sse2_exp_f(ewcljrsq);
 285             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 286             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 287             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 288             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 289             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 290             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 291             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 292             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);
 293
 294             /* Update potential sum for this i atom from the interaction with this j atom. */
 295             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 296
 297             fscal            = fvdw;
 298
 299             /* Calculate temporary vectorial force */
 300             tx               = _mm_mul_ps(fscal,dx00);
 301             ty               = _mm_mul_ps(fscal,dy00);
 302             tz               = _mm_mul_ps(fscal,dz00);
 303
 304             /* Update vectorial force */
 305             fix0             = _mm_add_ps(fix0,tx);
 306             fiy0             = _mm_add_ps(fiy0,ty);
 307             fiz0             = _mm_add_ps(fiz0,tz);
 308
 309             fjx0             = _mm_add_ps(fjx0,tx);
 310             fjy0             = _mm_add_ps(fjy0,ty);
 311             fjz0             = _mm_add_ps(fjz0,tz);
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             r10              = _mm_mul_ps(rsq10,rinv10);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq10             = _mm_mul_ps(iq1,jq0);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 325             ewrt             = _mm_mul_ps(r10,ewtabscale);
 326             ewitab           = _mm_cvttps_epi32(ewrt);
 327             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 328             ewitab           = _mm_slli_epi32(ewitab,2);
 329             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 330             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 331             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 332             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 333             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 334             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 335             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 336             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 337             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velecsum         = _mm_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm_mul_ps(fscal,dx10);
 346             ty               = _mm_mul_ps(fscal,dy10);
 347             tz               = _mm_mul_ps(fscal,dz10);
 348
 349             /* Update vectorial force */
 350             fix1             = _mm_add_ps(fix1,tx);
 351             fiy1             = _mm_add_ps(fiy1,ty);
 352             fiz1             = _mm_add_ps(fiz1,tz);
 353
 354             fjx0             = _mm_add_ps(fjx0,tx);
 355             fjy0             = _mm_add_ps(fjy0,ty);
 356             fjz0             = _mm_add_ps(fjz0,tz);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             r20              = _mm_mul_ps(rsq20,rinv20);
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq20             = _mm_mul_ps(iq2,jq0);
 366
 367             /* EWALD ELECTROSTATICS */
 368
 369             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 370             ewrt             = _mm_mul_ps(r20,ewtabscale);
 371             ewitab           = _mm_cvttps_epi32(ewrt);
 372             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 373             ewitab           = _mm_slli_epi32(ewitab,2);
 374             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 375             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 376             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 377             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 378             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 379             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 380             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 381             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 382             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 383
 384             /* Update potential sum for this i atom from the interaction with this j atom. */
 385             velecsum         = _mm_add_ps(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             /* Calculate temporary vectorial force */
 390             tx               = _mm_mul_ps(fscal,dx20);
 391             ty               = _mm_mul_ps(fscal,dy20);
 392             tz               = _mm_mul_ps(fscal,dz20);
 393
 394             /* Update vectorial force */
 395             fix2             = _mm_add_ps(fix2,tx);
 396             fiy2             = _mm_add_ps(fiy2,ty);
 397             fiz2             = _mm_add_ps(fiz2,tz);
 398
 399             fjx0             = _mm_add_ps(fjx0,tx);
 400             fjy0             = _mm_add_ps(fjy0,ty);
 401             fjz0             = _mm_add_ps(fjz0,tz);
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             r30              = _mm_mul_ps(rsq30,rinv30);
 408
 409             /* Compute parameters for interactions between i and j atoms */
 410             qq30             = _mm_mul_ps(iq3,jq0);
 411
 412             /* EWALD ELECTROSTATICS */
 413
 414             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 415             ewrt             = _mm_mul_ps(r30,ewtabscale);
 416             ewitab           = _mm_cvttps_epi32(ewrt);
 417             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 418             ewitab           = _mm_slli_epi32(ewitab,2);
 419             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 420             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 421             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 422             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 423             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 424             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 425             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 426             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 427             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velecsum         = _mm_add_ps(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm_mul_ps(fscal,dx30);
 436             ty               = _mm_mul_ps(fscal,dy30);
 437             tz               = _mm_mul_ps(fscal,dz30);
 438
 439             /* Update vectorial force */
 440             fix3             = _mm_add_ps(fix3,tx);
 441             fiy3             = _mm_add_ps(fiy3,ty);
 442             fiz3             = _mm_add_ps(fiz3,tz);
 443
 444             fjx0             = _mm_add_ps(fjx0,tx);
 445             fjy0             = _mm_add_ps(fjy0,ty);
 446             fjz0             = _mm_add_ps(fjz0,tz);
 447
 448             fjptrA             = f+j_coord_offsetA;
 449             fjptrB             = f+j_coord_offsetB;
 450             fjptrC             = f+j_coord_offsetC;
 451             fjptrD             = f+j_coord_offsetD;
 452
 453             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 454
 455             /* Inner loop uses 174 flops */
 456         }
 457
 458         if(jidx<j_index_end)
 459         {
 460
 461             /* Get j neighbor index, and coordinate index */
 462             jnrlistA         = jjnr[jidx];
 463             jnrlistB         = jjnr[jidx+1];
 464             jnrlistC         = jjnr[jidx+2];
 465             jnrlistD         = jjnr[jidx+3];
 466             /* Sign of each element will be negative for non-real atoms.
 467              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 468              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 469              */
 470             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 471             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 472             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 473             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 474             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 475             j_coord_offsetA  = DIM*jnrA;
 476             j_coord_offsetB  = DIM*jnrB;
 477             j_coord_offsetC  = DIM*jnrC;
 478             j_coord_offsetD  = DIM*jnrD;
 479
 480             /* load j atom coordinates */
 481             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 482                                               x+j_coord_offsetC,x+j_coord_offsetD,
 483                                               &jx0,&jy0,&jz0);
 484
 485             /* Calculate displacement vector */
 486             dx00             = _mm_sub_ps(ix0,jx0);
 487             dy00             = _mm_sub_ps(iy0,jy0);
 488             dz00             = _mm_sub_ps(iz0,jz0);
 489             dx10             = _mm_sub_ps(ix1,jx0);
 490             dy10             = _mm_sub_ps(iy1,jy0);
 491             dz10             = _mm_sub_ps(iz1,jz0);
 492             dx20             = _mm_sub_ps(ix2,jx0);
 493             dy20             = _mm_sub_ps(iy2,jy0);
 494             dz20             = _mm_sub_ps(iz2,jz0);
 495             dx30             = _mm_sub_ps(ix3,jx0);
 496             dy30             = _mm_sub_ps(iy3,jy0);
 497             dz30             = _mm_sub_ps(iz3,jz0);
 498
 499             /* Calculate squared distance and things based on it */
 500             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 501             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 502             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 503             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 504
 505             rinv00           = sse2_invsqrt_f(rsq00);
 506             rinv10           = sse2_invsqrt_f(rsq10);
 507             rinv20           = sse2_invsqrt_f(rsq20);
 508             rinv30           = sse2_invsqrt_f(rsq30);
 509
 510             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 511             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 512             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 513             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 514
 515             /* Load parameters for j particles */
 516             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 517                                                               charge+jnrC+0,charge+jnrD+0);
 518             vdwjidx0A        = 2*vdwtype[jnrA+0];
 519             vdwjidx0B        = 2*vdwtype[jnrB+0];
 520             vdwjidx0C        = 2*vdwtype[jnrC+0];
 521             vdwjidx0D        = 2*vdwtype[jnrD+0];
 522
 523             fjx0             = _mm_setzero_ps();
 524             fjy0             = _mm_setzero_ps();
 525             fjz0             = _mm_setzero_ps();
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             r00              = _mm_mul_ps(rsq00,rinv00);
 532             r00              = _mm_andnot_ps(dummy_mask,r00);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 536                                          vdwparam+vdwioffset0+vdwjidx0B,
 537                                          vdwparam+vdwioffset0+vdwjidx0C,
 538                                          vdwparam+vdwioffset0+vdwjidx0D,
 539                                          &c6_00,&c12_00);
 540             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 541                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 542                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 543                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 544
 545             /* Analytical LJ-PME */
 546             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 547             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 548             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 549             exponent         = sse2_exp_f(ewcljrsq);
 550             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 551             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 552             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 553             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 554             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 555             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 556             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 557             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);
 558
 559             /* Update potential sum for this i atom from the interaction with this j atom. */
 560             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 561             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 562
 563             fscal            = fvdw;
 564
 565             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 566
 567             /* Calculate temporary vectorial force */
 568             tx               = _mm_mul_ps(fscal,dx00);
 569             ty               = _mm_mul_ps(fscal,dy00);
 570             tz               = _mm_mul_ps(fscal,dz00);
 571
 572             /* Update vectorial force */
 573             fix0             = _mm_add_ps(fix0,tx);
 574             fiy0             = _mm_add_ps(fiy0,ty);
 575             fiz0             = _mm_add_ps(fiz0,tz);
 576
 577             fjx0             = _mm_add_ps(fjx0,tx);
 578             fjy0             = _mm_add_ps(fjy0,ty);
 579             fjz0             = _mm_add_ps(fjz0,tz);
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             r10              = _mm_mul_ps(rsq10,rinv10);
 586             r10              = _mm_andnot_ps(dummy_mask,r10);
 587
 588             /* Compute parameters for interactions between i and j atoms */
 589             qq10             = _mm_mul_ps(iq1,jq0);
 590
 591             /* EWALD ELECTROSTATICS */
 592
 593             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 594             ewrt             = _mm_mul_ps(r10,ewtabscale);
 595             ewitab           = _mm_cvttps_epi32(ewrt);
 596             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 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(qq10,_mm_sub_ps(rinv10,velec));
 606             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velec            = _mm_andnot_ps(dummy_mask,velec);
 610             velecsum         = _mm_add_ps(velecsum,velec);
 611
 612             fscal            = felec;
 613
 614             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 615
 616             /* Calculate temporary vectorial force */
 617             tx               = _mm_mul_ps(fscal,dx10);
 618             ty               = _mm_mul_ps(fscal,dy10);
 619             tz               = _mm_mul_ps(fscal,dz10);
 620
 621             /* Update vectorial force */
 622             fix1             = _mm_add_ps(fix1,tx);
 623             fiy1             = _mm_add_ps(fiy1,ty);
 624             fiz1             = _mm_add_ps(fiz1,tz);
 625
 626             fjx0             = _mm_add_ps(fjx0,tx);
 627             fjy0             = _mm_add_ps(fjy0,ty);
 628             fjz0             = _mm_add_ps(fjz0,tz);
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             r20              = _mm_mul_ps(rsq20,rinv20);
 635             r20              = _mm_andnot_ps(dummy_mask,r20);
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             qq20             = _mm_mul_ps(iq2,jq0);
 639
 640             /* EWALD ELECTROSTATICS */
 641
 642             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 643             ewrt             = _mm_mul_ps(r20,ewtabscale);
 644             ewitab           = _mm_cvttps_epi32(ewrt);
 645             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 646             ewitab           = _mm_slli_epi32(ewitab,2);
 647             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 648             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 649             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 650             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 651             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 652             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 653             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 654             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 655             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 656
 657             /* Update potential sum for this i atom from the interaction with this j atom. */
 658             velec            = _mm_andnot_ps(dummy_mask,velec);
 659             velecsum         = _mm_add_ps(velecsum,velec);
 660
 661             fscal            = felec;
 662
 663             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 664
 665             /* Calculate temporary vectorial force */
 666             tx               = _mm_mul_ps(fscal,dx20);
 667             ty               = _mm_mul_ps(fscal,dy20);
 668             tz               = _mm_mul_ps(fscal,dz20);
 669
 670             /* Update vectorial force */
 671             fix2             = _mm_add_ps(fix2,tx);
 672             fiy2             = _mm_add_ps(fiy2,ty);
 673             fiz2             = _mm_add_ps(fiz2,tz);
 674
 675             fjx0             = _mm_add_ps(fjx0,tx);
 676             fjy0             = _mm_add_ps(fjy0,ty);
 677             fjz0             = _mm_add_ps(fjz0,tz);
 678
 679             /**************************
 680              * CALCULATE INTERACTIONS *
 681              **************************/
 682
 683             r30              = _mm_mul_ps(rsq30,rinv30);
 684             r30              = _mm_andnot_ps(dummy_mask,r30);
 685
 686             /* Compute parameters for interactions between i and j atoms */
 687             qq30             = _mm_mul_ps(iq3,jq0);
 688
 689             /* EWALD ELECTROSTATICS */
 690
 691             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 692             ewrt             = _mm_mul_ps(r30,ewtabscale);
 693             ewitab           = _mm_cvttps_epi32(ewrt);
 694             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 695             ewitab           = _mm_slli_epi32(ewitab,2);
 696             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 697             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 698             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 699             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 700             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 701             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 702             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 703             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 704             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 705
 706             /* Update potential sum for this i atom from the interaction with this j atom. */
 707             velec            = _mm_andnot_ps(dummy_mask,velec);
 708             velecsum         = _mm_add_ps(velecsum,velec);
 709
 710             fscal            = felec;
 711
 712             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 713
 714             /* Calculate temporary vectorial force */
 715             tx               = _mm_mul_ps(fscal,dx30);
 716             ty               = _mm_mul_ps(fscal,dy30);
 717             tz               = _mm_mul_ps(fscal,dz30);
 718
 719             /* Update vectorial force */
 720             fix3             = _mm_add_ps(fix3,tx);
 721             fiy3             = _mm_add_ps(fiy3,ty);
 722             fiz3             = _mm_add_ps(fiz3,tz);
 723
 724             fjx0             = _mm_add_ps(fjx0,tx);
 725             fjy0             = _mm_add_ps(fjy0,ty);
 726             fjz0             = _mm_add_ps(fjz0,tz);
 727
 728             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 729             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 730             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 731             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 732
 733             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 734
 735             /* Inner loop uses 178 flops */
 736         }
 737
 738         /* End of innermost loop */
 739
 740         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 741                                               f+i_coord_offset,fshift+i_shift_offset);
 742
 743         ggid                        = gid[iidx];
 744         /* Update potential energies */
 745         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 746         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 747
 748         /* Increment number of inner iterations */
 749         inneriter                  += j_index_end - j_index_start;
 750
 751         /* Outer loop uses 26 flops */
 752     }
 753
 754     /* Increment number of outer iterations */
 755     outeriter        += nri;
 756
 757     /* Update outer/inner flops */
 758
 759     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*178);
 760 }
 761 /*
 762  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse2_single
 763  * Electrostatics interaction: Ewald
 764  * VdW interaction:            LJEwald
 765  * Geometry:                   Water4-Particle
 766  * Calculate force/pot:        Force
 767  */
 768 void
 769 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse2_single
 770                     (t_nblist                    * gmx_restrict       nlist,
 771                      rvec                        * gmx_restrict          xx,
 772                      rvec                        * gmx_restrict          ff,
 773                      struct t_forcerec           * gmx_restrict          fr,
 774                      t_mdatoms                   * gmx_restrict     mdatoms,
 775                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 776                      t_nrnb                      * gmx_restrict        nrnb)
 777 {
 778     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 779      * just 0 for non-waters.
 780      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 781      * jnr indices corresponding to data put in the four positions in the SIMD register.
 782      */
 783     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 784     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 785     int              jnrA,jnrB,jnrC,jnrD;
 786     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 787     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 788     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 789     real             rcutoff_scalar;
 790     real             *shiftvec,*fshift,*x,*f;
 791     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 792     real             scratch[4*DIM];
 793     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 794     int              vdwioffset0;
 795     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 796     int              vdwioffset1;
 797     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 798     int              vdwioffset2;
 799     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 800     int              vdwioffset3;
 801     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 802     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 803     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 804     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 805     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 806     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 807     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 808     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 809     real             *charge;
 810     int              nvdwtype;
 811     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 812     int              *vdwtype;
 813     real             *vdwparam;
 814     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 815     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 816     __m128           c6grid_00;
 817     __m128           c6grid_10;
 818     __m128           c6grid_20;
 819     __m128           c6grid_30;
 820     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 821     real             *vdwgridparam;
 822     __m128           one_half = _mm_set1_ps(0.5);
 823     __m128           minus_one = _mm_set1_ps(-1.0);
 824     __m128i          ewitab;
 825     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 826     real             *ewtab;
 827     __m128           dummy_mask,cutoff_mask;
 828     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 829     __m128           one     = _mm_set1_ps(1.0);
 830     __m128           two     = _mm_set1_ps(2.0);
 831     x                = xx[0];
 832     f                = ff[0];
 833
 834     nri              = nlist->nri;
 835     iinr             = nlist->iinr;
 836     jindex           = nlist->jindex;
 837     jjnr             = nlist->jjnr;
 838     shiftidx         = nlist->shift;
 839     gid              = nlist->gid;
 840     shiftvec         = fr->shift_vec[0];
 841     fshift           = fr->fshift[0];
 842     facel            = _mm_set1_ps(fr->ic->epsfac);
 843     charge           = mdatoms->chargeA;
 844     nvdwtype         = fr->ntype;
 845     vdwparam         = fr->nbfp;
 846     vdwtype          = mdatoms->typeA;
 847     vdwgridparam     = fr->ljpme_c6grid;
 848     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 849     ewclj            = _mm_set1_ps(fr->ic->ewaldcoeff_lj);
 850     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 851
 852     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 853     ewtab            = fr->ic->tabq_coul_F;
 854     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 855     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 856
 857     /* Setup water-specific parameters */
 858     inr              = nlist->iinr[0];
 859     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 860     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 861     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 862     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 863
 864     /* Avoid stupid compiler warnings */
 865     jnrA = jnrB = jnrC = jnrD = 0;
 866     j_coord_offsetA = 0;
 867     j_coord_offsetB = 0;
 868     j_coord_offsetC = 0;
 869     j_coord_offsetD = 0;
 870
 871     outeriter        = 0;
 872     inneriter        = 0;
 873
 874     for(iidx=0;iidx<4*DIM;iidx++)
 875     {
 876         scratch[iidx] = 0.0;
 877     }
 878
 879     /* Start outer loop over neighborlists */
 880     for(iidx=0; iidx<nri; iidx++)
 881     {
 882         /* Load shift vector for this list */
 883         i_shift_offset   = DIM*shiftidx[iidx];
 884
 885         /* Load limits for loop over neighbors */
 886         j_index_start    = jindex[iidx];
 887         j_index_end      = jindex[iidx+1];
 888
 889         /* Get outer coordinate index */
 890         inr              = iinr[iidx];
 891         i_coord_offset   = DIM*inr;
 892
 893         /* Load i particle coords and add shift vector */
 894         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 895                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 896
 897         fix0             = _mm_setzero_ps();
 898         fiy0             = _mm_setzero_ps();
 899         fiz0             = _mm_setzero_ps();
 900         fix1             = _mm_setzero_ps();
 901         fiy1             = _mm_setzero_ps();
 902         fiz1             = _mm_setzero_ps();
 903         fix2             = _mm_setzero_ps();
 904         fiy2             = _mm_setzero_ps();
 905         fiz2             = _mm_setzero_ps();
 906         fix3             = _mm_setzero_ps();
 907         fiy3             = _mm_setzero_ps();
 908         fiz3             = _mm_setzero_ps();
 909
 910         /* Start inner kernel loop */
 911         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 912         {
 913
 914             /* Get j neighbor index, and coordinate index */
 915             jnrA             = jjnr[jidx];
 916             jnrB             = jjnr[jidx+1];
 917             jnrC             = jjnr[jidx+2];
 918             jnrD             = jjnr[jidx+3];
 919             j_coord_offsetA  = DIM*jnrA;
 920             j_coord_offsetB  = DIM*jnrB;
 921             j_coord_offsetC  = DIM*jnrC;
 922             j_coord_offsetD  = DIM*jnrD;
 923
 924             /* load j atom coordinates */
 925             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 926                                               x+j_coord_offsetC,x+j_coord_offsetD,
 927                                               &jx0,&jy0,&jz0);
 928
 929             /* Calculate displacement vector */
 930             dx00             = _mm_sub_ps(ix0,jx0);
 931             dy00             = _mm_sub_ps(iy0,jy0);
 932             dz00             = _mm_sub_ps(iz0,jz0);
 933             dx10             = _mm_sub_ps(ix1,jx0);
 934             dy10             = _mm_sub_ps(iy1,jy0);
 935             dz10             = _mm_sub_ps(iz1,jz0);
 936             dx20             = _mm_sub_ps(ix2,jx0);
 937             dy20             = _mm_sub_ps(iy2,jy0);
 938             dz20             = _mm_sub_ps(iz2,jz0);
 939             dx30             = _mm_sub_ps(ix3,jx0);
 940             dy30             = _mm_sub_ps(iy3,jy0);
 941             dz30             = _mm_sub_ps(iz3,jz0);
 942
 943             /* Calculate squared distance and things based on it */
 944             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 945             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 946             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 947             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 948
 949             rinv00           = sse2_invsqrt_f(rsq00);
 950             rinv10           = sse2_invsqrt_f(rsq10);
 951             rinv20           = sse2_invsqrt_f(rsq20);
 952             rinv30           = sse2_invsqrt_f(rsq30);
 953
 954             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 955             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 956             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 957             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 958
 959             /* Load parameters for j particles */
 960             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 961                                                               charge+jnrC+0,charge+jnrD+0);
 962             vdwjidx0A        = 2*vdwtype[jnrA+0];
 963             vdwjidx0B        = 2*vdwtype[jnrB+0];
 964             vdwjidx0C        = 2*vdwtype[jnrC+0];
 965             vdwjidx0D        = 2*vdwtype[jnrD+0];
 966
 967             fjx0             = _mm_setzero_ps();
 968             fjy0             = _mm_setzero_ps();
 969             fjz0             = _mm_setzero_ps();
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             r00              = _mm_mul_ps(rsq00,rinv00);
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 979                                          vdwparam+vdwioffset0+vdwjidx0B,
 980                                          vdwparam+vdwioffset0+vdwjidx0C,
 981                                          vdwparam+vdwioffset0+vdwjidx0D,
 982                                          &c6_00,&c12_00);
 983             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 984                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 985                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 986                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 987
 988             /* Analytical LJ-PME */
 989             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 990             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 991             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 992             exponent         = sse2_exp_f(ewcljrsq);
 993             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 994             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 995             /* f6A = 6 * C6grid * (1 - poly) */
 996             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
 997             /* f6B = C6grid * exponent * beta^6 */
 998             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
 999             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1000             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);
1001
1002             fscal            = fvdw;
1003
1004             /* Calculate temporary vectorial force */
1005             tx               = _mm_mul_ps(fscal,dx00);
1006             ty               = _mm_mul_ps(fscal,dy00);
1007             tz               = _mm_mul_ps(fscal,dz00);
1008
1009             /* Update vectorial force */
1010             fix0             = _mm_add_ps(fix0,tx);
1011             fiy0             = _mm_add_ps(fiy0,ty);
1012             fiz0             = _mm_add_ps(fiz0,tz);
1013
1014             fjx0             = _mm_add_ps(fjx0,tx);
1015             fjy0             = _mm_add_ps(fjy0,ty);
1016             fjz0             = _mm_add_ps(fjz0,tz);
1017
1018             /**************************
1019              * CALCULATE INTERACTIONS *
1020              **************************/
1021
1022             r10              = _mm_mul_ps(rsq10,rinv10);
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq10             = _mm_mul_ps(iq1,jq0);
1026
1027             /* EWALD ELECTROSTATICS */
1028
1029             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1030             ewrt             = _mm_mul_ps(r10,ewtabscale);
1031             ewitab           = _mm_cvttps_epi32(ewrt);
1032             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1033             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1034                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1035                                          &ewtabF,&ewtabFn);
1036             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1037             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1038
1039             fscal            = felec;
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm_mul_ps(fscal,dx10);
1043             ty               = _mm_mul_ps(fscal,dy10);
1044             tz               = _mm_mul_ps(fscal,dz10);
1045
1046             /* Update vectorial force */
1047             fix1             = _mm_add_ps(fix1,tx);
1048             fiy1             = _mm_add_ps(fiy1,ty);
1049             fiz1             = _mm_add_ps(fiz1,tz);
1050
1051             fjx0             = _mm_add_ps(fjx0,tx);
1052             fjy0             = _mm_add_ps(fjy0,ty);
1053             fjz0             = _mm_add_ps(fjz0,tz);
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             r20              = _mm_mul_ps(rsq20,rinv20);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq20             = _mm_mul_ps(iq2,jq0);
1063
1064             /* EWALD ELECTROSTATICS */
1065
1066             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1067             ewrt             = _mm_mul_ps(r20,ewtabscale);
1068             ewitab           = _mm_cvttps_epi32(ewrt);
1069             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1070             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1071                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1072                                          &ewtabF,&ewtabFn);
1073             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1074             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1075
1076             fscal            = felec;
1077
1078             /* Calculate temporary vectorial force */
1079             tx               = _mm_mul_ps(fscal,dx20);
1080             ty               = _mm_mul_ps(fscal,dy20);
1081             tz               = _mm_mul_ps(fscal,dz20);
1082
1083             /* Update vectorial force */
1084             fix2             = _mm_add_ps(fix2,tx);
1085             fiy2             = _mm_add_ps(fiy2,ty);
1086             fiz2             = _mm_add_ps(fiz2,tz);
1087
1088             fjx0             = _mm_add_ps(fjx0,tx);
1089             fjy0             = _mm_add_ps(fjy0,ty);
1090             fjz0             = _mm_add_ps(fjz0,tz);
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             r30              = _mm_mul_ps(rsq30,rinv30);
1097
1098             /* Compute parameters for interactions between i and j atoms */
1099             qq30             = _mm_mul_ps(iq3,jq0);
1100
1101             /* EWALD ELECTROSTATICS */
1102
1103             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1104             ewrt             = _mm_mul_ps(r30,ewtabscale);
1105             ewitab           = _mm_cvttps_epi32(ewrt);
1106             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1107             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1108                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1109                                          &ewtabF,&ewtabFn);
1110             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1111             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1112
1113             fscal            = felec;
1114
1115             /* Calculate temporary vectorial force */
1116             tx               = _mm_mul_ps(fscal,dx30);
1117             ty               = _mm_mul_ps(fscal,dy30);
1118             tz               = _mm_mul_ps(fscal,dz30);
1119
1120             /* Update vectorial force */
1121             fix3             = _mm_add_ps(fix3,tx);
1122             fiy3             = _mm_add_ps(fiy3,ty);
1123             fiz3             = _mm_add_ps(fiz3,tz);
1124
1125             fjx0             = _mm_add_ps(fjx0,tx);
1126             fjy0             = _mm_add_ps(fjy0,ty);
1127             fjz0             = _mm_add_ps(fjz0,tz);
1128
1129             fjptrA             = f+j_coord_offsetA;
1130             fjptrB             = f+j_coord_offsetB;
1131             fjptrC             = f+j_coord_offsetC;
1132             fjptrD             = f+j_coord_offsetD;
1133
1134             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1135
1136             /* Inner loop uses 154 flops */
1137         }
1138
1139         if(jidx<j_index_end)
1140         {
1141
1142             /* Get j neighbor index, and coordinate index */
1143             jnrlistA         = jjnr[jidx];
1144             jnrlistB         = jjnr[jidx+1];
1145             jnrlistC         = jjnr[jidx+2];
1146             jnrlistD         = jjnr[jidx+3];
1147             /* Sign of each element will be negative for non-real atoms.
1148              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1149              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1150              */
1151             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1152             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1153             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1154             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1155             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1156             j_coord_offsetA  = DIM*jnrA;
1157             j_coord_offsetB  = DIM*jnrB;
1158             j_coord_offsetC  = DIM*jnrC;
1159             j_coord_offsetD  = DIM*jnrD;
1160
1161             /* load j atom coordinates */
1162             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1163                                               x+j_coord_offsetC,x+j_coord_offsetD,
1164                                               &jx0,&jy0,&jz0);
1165
1166             /* Calculate displacement vector */
1167             dx00             = _mm_sub_ps(ix0,jx0);
1168             dy00             = _mm_sub_ps(iy0,jy0);
1169             dz00             = _mm_sub_ps(iz0,jz0);
1170             dx10             = _mm_sub_ps(ix1,jx0);
1171             dy10             = _mm_sub_ps(iy1,jy0);
1172             dz10             = _mm_sub_ps(iz1,jz0);
1173             dx20             = _mm_sub_ps(ix2,jx0);
1174             dy20             = _mm_sub_ps(iy2,jy0);
1175             dz20             = _mm_sub_ps(iz2,jz0);
1176             dx30             = _mm_sub_ps(ix3,jx0);
1177             dy30             = _mm_sub_ps(iy3,jy0);
1178             dz30             = _mm_sub_ps(iz3,jz0);
1179
1180             /* Calculate squared distance and things based on it */
1181             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1182             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1183             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1184             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1185
1186             rinv00           = sse2_invsqrt_f(rsq00);
1187             rinv10           = sse2_invsqrt_f(rsq10);
1188             rinv20           = sse2_invsqrt_f(rsq20);
1189             rinv30           = sse2_invsqrt_f(rsq30);
1190
1191             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1192             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1193             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1194             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1195
1196             /* Load parameters for j particles */
1197             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1198                                                               charge+jnrC+0,charge+jnrD+0);
1199             vdwjidx0A        = 2*vdwtype[jnrA+0];
1200             vdwjidx0B        = 2*vdwtype[jnrB+0];
1201             vdwjidx0C        = 2*vdwtype[jnrC+0];
1202             vdwjidx0D        = 2*vdwtype[jnrD+0];
1203
1204             fjx0             = _mm_setzero_ps();
1205             fjy0             = _mm_setzero_ps();
1206             fjz0             = _mm_setzero_ps();
1207
1208             /**************************
1209              * CALCULATE INTERACTIONS *
1210              **************************/
1211
1212             r00              = _mm_mul_ps(rsq00,rinv00);
1213             r00              = _mm_andnot_ps(dummy_mask,r00);
1214
1215             /* Compute parameters for interactions between i and j atoms */
1216             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1217                                          vdwparam+vdwioffset0+vdwjidx0B,
1218                                          vdwparam+vdwioffset0+vdwjidx0C,
1219                                          vdwparam+vdwioffset0+vdwjidx0D,
1220                                          &c6_00,&c12_00);
1221             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1222                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1223                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1224                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1225
1226             /* Analytical LJ-PME */
1227             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1228             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1229             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1230             exponent         = sse2_exp_f(ewcljrsq);
1231             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1232             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1233             /* f6A = 6 * C6grid * (1 - poly) */
1234             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1235             /* f6B = C6grid * exponent * beta^6 */
1236             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1237             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1238             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);
1239
1240             fscal            = fvdw;
1241
1242             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1243
1244             /* Calculate temporary vectorial force */
1245             tx               = _mm_mul_ps(fscal,dx00);
1246             ty               = _mm_mul_ps(fscal,dy00);
1247             tz               = _mm_mul_ps(fscal,dz00);
1248
1249             /* Update vectorial force */
1250             fix0             = _mm_add_ps(fix0,tx);
1251             fiy0             = _mm_add_ps(fiy0,ty);
1252             fiz0             = _mm_add_ps(fiz0,tz);
1253
1254             fjx0             = _mm_add_ps(fjx0,tx);
1255             fjy0             = _mm_add_ps(fjy0,ty);
1256             fjz0             = _mm_add_ps(fjz0,tz);
1257
1258             /**************************
1259              * CALCULATE INTERACTIONS *
1260              **************************/
1261
1262             r10              = _mm_mul_ps(rsq10,rinv10);
1263             r10              = _mm_andnot_ps(dummy_mask,r10);
1264
1265             /* Compute parameters for interactions between i and j atoms */
1266             qq10             = _mm_mul_ps(iq1,jq0);
1267
1268             /* EWALD ELECTROSTATICS */
1269
1270             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1271             ewrt             = _mm_mul_ps(r10,ewtabscale);
1272             ewitab           = _mm_cvttps_epi32(ewrt);
1273             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1274             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1275                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1276                                          &ewtabF,&ewtabFn);
1277             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1278             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1279
1280             fscal            = felec;
1281
1282             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1283
1284             /* Calculate temporary vectorial force */
1285             tx               = _mm_mul_ps(fscal,dx10);
1286             ty               = _mm_mul_ps(fscal,dy10);
1287             tz               = _mm_mul_ps(fscal,dz10);
1288
1289             /* Update vectorial force */
1290             fix1             = _mm_add_ps(fix1,tx);
1291             fiy1             = _mm_add_ps(fiy1,ty);
1292             fiz1             = _mm_add_ps(fiz1,tz);
1293
1294             fjx0             = _mm_add_ps(fjx0,tx);
1295             fjy0             = _mm_add_ps(fjy0,ty);
1296             fjz0             = _mm_add_ps(fjz0,tz);
1297
1298             /**************************
1299              * CALCULATE INTERACTIONS *
1300              **************************/
1301
1302             r20              = _mm_mul_ps(rsq20,rinv20);
1303             r20              = _mm_andnot_ps(dummy_mask,r20);
1304
1305             /* Compute parameters for interactions between i and j atoms */
1306             qq20             = _mm_mul_ps(iq2,jq0);
1307
1308             /* EWALD ELECTROSTATICS */
1309
1310             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1311             ewrt             = _mm_mul_ps(r20,ewtabscale);
1312             ewitab           = _mm_cvttps_epi32(ewrt);
1313             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1314             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1315                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1316                                          &ewtabF,&ewtabFn);
1317             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1318             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1319
1320             fscal            = felec;
1321
1322             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1323
1324             /* Calculate temporary vectorial force */
1325             tx               = _mm_mul_ps(fscal,dx20);
1326             ty               = _mm_mul_ps(fscal,dy20);
1327             tz               = _mm_mul_ps(fscal,dz20);
1328
1329             /* Update vectorial force */
1330             fix2             = _mm_add_ps(fix2,tx);
1331             fiy2             = _mm_add_ps(fiy2,ty);
1332             fiz2             = _mm_add_ps(fiz2,tz);
1333
1334             fjx0             = _mm_add_ps(fjx0,tx);
1335             fjy0             = _mm_add_ps(fjy0,ty);
1336             fjz0             = _mm_add_ps(fjz0,tz);
1337
1338             /**************************
1339              * CALCULATE INTERACTIONS *
1340              **************************/
1341
1342             r30              = _mm_mul_ps(rsq30,rinv30);
1343             r30              = _mm_andnot_ps(dummy_mask,r30);
1344
1345             /* Compute parameters for interactions between i and j atoms */
1346             qq30             = _mm_mul_ps(iq3,jq0);
1347
1348             /* EWALD ELECTROSTATICS */
1349
1350             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1351             ewrt             = _mm_mul_ps(r30,ewtabscale);
1352             ewitab           = _mm_cvttps_epi32(ewrt);
1353             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1354             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1355                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1356                                          &ewtabF,&ewtabFn);
1357             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1358             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1359
1360             fscal            = felec;
1361
1362             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1363
1364             /* Calculate temporary vectorial force */
1365             tx               = _mm_mul_ps(fscal,dx30);
1366             ty               = _mm_mul_ps(fscal,dy30);
1367             tz               = _mm_mul_ps(fscal,dz30);
1368
1369             /* Update vectorial force */
1370             fix3             = _mm_add_ps(fix3,tx);
1371             fiy3             = _mm_add_ps(fiy3,ty);
1372             fiz3             = _mm_add_ps(fiz3,tz);
1373
1374             fjx0             = _mm_add_ps(fjx0,tx);
1375             fjy0             = _mm_add_ps(fjy0,ty);
1376             fjz0             = _mm_add_ps(fjz0,tz);
1377
1378             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1379             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1380             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1381             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1382
1383             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1384
1385             /* Inner loop uses 158 flops */
1386         }
1387
1388         /* End of innermost loop */
1389
1390         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1391                                               f+i_coord_offset,fshift+i_shift_offset);
1392
1393         /* Increment number of inner iterations */
1394         inneriter                  += j_index_end - j_index_start;
1395
1396         /* Outer loop uses 24 flops */
1397     }
1398
1399     /* Increment number of outer iterations */
1400     outeriter        += nri;
1401
1402     /* Update outer/inner flops */
1403
1404     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*158);
1405 }