src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJEw_GeomW4P1_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwioffset3;
  92     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  93     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  94     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  95     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  96     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  97     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  98     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  99     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 100     real             *charge;
 101     int              nvdwtype;
 102     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 103     int              *vdwtype;
 104     real             *vdwparam;
 105     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 106     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 107     __m128           c6grid_00;
 108     __m128           c6grid_10;
 109     __m128           c6grid_20;
 110     __m128           c6grid_30;
 111     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 112     real             *vdwgridparam;
 113     __m128           one_half  = _mm_set1_ps(0.5);
 114     __m128           minus_one = _mm_set1_ps(-1.0);
 115     __m128i          ewitab;
 116     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 117     real             *ewtab;
 118     __m128           dummy_mask,cutoff_mask;
 119     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 120     __m128           one     = _mm_set1_ps(1.0);
 121     __m128           two     = _mm_set1_ps(2.0);
 122     x                = xx[0];
 123     f                = ff[0];
 124
 125     nri              = nlist->nri;
 126     iinr             = nlist->iinr;
 127     jindex           = nlist->jindex;
 128     jjnr             = nlist->jjnr;
 129     shiftidx         = nlist->shift;
 130     gid              = nlist->gid;
 131     shiftvec         = fr->shift_vec[0];
 132     fshift           = fr->fshift[0];
 133     facel            = _mm_set1_ps(fr->epsfac);
 134     charge           = mdatoms->chargeA;
 135     nvdwtype         = fr->ntype;
 136     vdwparam         = fr->nbfp;
 137     vdwtype          = mdatoms->typeA;
 138     vdwgridparam     = fr->ljpme_c6grid;
 139     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 140     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 141     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 142
 143     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 144     ewtab            = fr->ic->tabq_coul_FDV0;
 145     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 146     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 147
 148     /* Setup water-specific parameters */
 149     inr              = nlist->iinr[0];
 150     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 151     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 152     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 153     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 154
 155     /* Avoid stupid compiler warnings */
 156     jnrA = jnrB = jnrC = jnrD = 0;
 157     j_coord_offsetA = 0;
 158     j_coord_offsetB = 0;
 159     j_coord_offsetC = 0;
 160     j_coord_offsetD = 0;
 161
 162     outeriter        = 0;
 163     inneriter        = 0;
 164
 165     for(iidx=0;iidx<4*DIM;iidx++)
 166     {
 167         scratch[iidx] = 0.0;
 168     }
 169
 170     /* Start outer loop over neighborlists */
 171     for(iidx=0; iidx<nri; iidx++)
 172     {
 173         /* Load shift vector for this list */
 174         i_shift_offset   = DIM*shiftidx[iidx];
 175
 176         /* Load limits for loop over neighbors */
 177         j_index_start    = jindex[iidx];
 178         j_index_end      = jindex[iidx+1];
 179
 180         /* Get outer coordinate index */
 181         inr              = iinr[iidx];
 182         i_coord_offset   = DIM*inr;
 183
 184         /* Load i particle coords and add shift vector */
 185         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 186                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 187
 188         fix0             = _mm_setzero_ps();
 189         fiy0             = _mm_setzero_ps();
 190         fiz0             = _mm_setzero_ps();
 191         fix1             = _mm_setzero_ps();
 192         fiy1             = _mm_setzero_ps();
 193         fiz1             = _mm_setzero_ps();
 194         fix2             = _mm_setzero_ps();
 195         fiy2             = _mm_setzero_ps();
 196         fiz2             = _mm_setzero_ps();
 197         fix3             = _mm_setzero_ps();
 198         fiy3             = _mm_setzero_ps();
 199         fiz3             = _mm_setzero_ps();
 200
 201         /* Reset potential sums */
 202         velecsum         = _mm_setzero_ps();
 203         vvdwsum          = _mm_setzero_ps();
 204
 205         /* Start inner kernel loop */
 206         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 207         {
 208
 209             /* Get j neighbor index, and coordinate index */
 210             jnrA             = jjnr[jidx];
 211             jnrB             = jjnr[jidx+1];
 212             jnrC             = jjnr[jidx+2];
 213             jnrD             = jjnr[jidx+3];
 214             j_coord_offsetA  = DIM*jnrA;
 215             j_coord_offsetB  = DIM*jnrB;
 216             j_coord_offsetC  = DIM*jnrC;
 217             j_coord_offsetD  = DIM*jnrD;
 218
 219             /* load j atom coordinates */
 220             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 221                                               x+j_coord_offsetC,x+j_coord_offsetD,
 222                                               &jx0,&jy0,&jz0);
 223
 224             /* Calculate displacement vector */
 225             dx00             = _mm_sub_ps(ix0,jx0);
 226             dy00             = _mm_sub_ps(iy0,jy0);
 227             dz00             = _mm_sub_ps(iz0,jz0);
 228             dx10             = _mm_sub_ps(ix1,jx0);
 229             dy10             = _mm_sub_ps(iy1,jy0);
 230             dz10             = _mm_sub_ps(iz1,jz0);
 231             dx20             = _mm_sub_ps(ix2,jx0);
 232             dy20             = _mm_sub_ps(iy2,jy0);
 233             dz20             = _mm_sub_ps(iz2,jz0);
 234             dx30             = _mm_sub_ps(ix3,jx0);
 235             dy30             = _mm_sub_ps(iy3,jy0);
 236             dz30             = _mm_sub_ps(iz3,jz0);
 237
 238             /* Calculate squared distance and things based on it */
 239             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 240             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 241             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 242             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 243
 244             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 245             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 246             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 247             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 248
 249             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 250             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 251             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 252             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 253
 254             /* Load parameters for j particles */
 255             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 256                                                               charge+jnrC+0,charge+jnrD+0);
 257             vdwjidx0A        = 2*vdwtype[jnrA+0];
 258             vdwjidx0B        = 2*vdwtype[jnrB+0];
 259             vdwjidx0C        = 2*vdwtype[jnrC+0];
 260             vdwjidx0D        = 2*vdwtype[jnrD+0];
 261
 262             fjx0             = _mm_setzero_ps();
 263             fjy0             = _mm_setzero_ps();
 264             fjz0             = _mm_setzero_ps();
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             r00              = _mm_mul_ps(rsq00,rinv00);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 274                                          vdwparam+vdwioffset0+vdwjidx0B,
 275                                          vdwparam+vdwioffset0+vdwjidx0C,
 276                                          vdwparam+vdwioffset0+vdwjidx0D,
 277                                          &c6_00,&c12_00);
 278
 279             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 280                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 281                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 282                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 283
 284             /* Analytical LJ-PME */
 285             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 286             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 287             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 288             exponent         = gmx_simd_exp_r(ewcljrsq);
 289             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 290             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 291             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 292             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 293             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 294             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 295             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 296             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);
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = fvdw;
 302
 303             /* Calculate temporary vectorial force */
 304             tx               = _mm_mul_ps(fscal,dx00);
 305             ty               = _mm_mul_ps(fscal,dy00);
 306             tz               = _mm_mul_ps(fscal,dz00);
 307
 308             /* Update vectorial force */
 309             fix0             = _mm_add_ps(fix0,tx);
 310             fiy0             = _mm_add_ps(fiy0,ty);
 311             fiz0             = _mm_add_ps(fiz0,tz);
 312
 313             fjx0             = _mm_add_ps(fjx0,tx);
 314             fjy0             = _mm_add_ps(fjy0,ty);
 315             fjz0             = _mm_add_ps(fjz0,tz);
 316
 317             /**************************
 318              * CALCULATE INTERACTIONS *
 319              **************************/
 320
 321             r10              = _mm_mul_ps(rsq10,rinv10);
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_ps(iq1,jq0);
 325
 326             /* EWALD ELECTROSTATICS */
 327
 328             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 329             ewrt             = _mm_mul_ps(r10,ewtabscale);
 330             ewitab           = _mm_cvttps_epi32(ewrt);
 331             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 332             ewitab           = _mm_slli_epi32(ewitab,2);
 333             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 334             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 335             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 336             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 337             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 338             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 339             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 340             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 341             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_ps(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_ps(fscal,dx10);
 350             ty               = _mm_mul_ps(fscal,dy10);
 351             tz               = _mm_mul_ps(fscal,dz10);
 352
 353             /* Update vectorial force */
 354             fix1             = _mm_add_ps(fix1,tx);
 355             fiy1             = _mm_add_ps(fiy1,ty);
 356             fiz1             = _mm_add_ps(fiz1,tz);
 357
 358             fjx0             = _mm_add_ps(fjx0,tx);
 359             fjy0             = _mm_add_ps(fjy0,ty);
 360             fjz0             = _mm_add_ps(fjz0,tz);
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             r20              = _mm_mul_ps(rsq20,rinv20);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq20             = _mm_mul_ps(iq2,jq0);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 374             ewrt             = _mm_mul_ps(r20,ewtabscale);
 375             ewitab           = _mm_cvttps_epi32(ewrt);
 376             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 377             ewitab           = _mm_slli_epi32(ewitab,2);
 378             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 379             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 380             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 381             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 382             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 383             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 384             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 385             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 386             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 387
 388             /* Update potential sum for this i atom from the interaction with this j atom. */
 389             velecsum         = _mm_add_ps(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             /* Calculate temporary vectorial force */
 394             tx               = _mm_mul_ps(fscal,dx20);
 395             ty               = _mm_mul_ps(fscal,dy20);
 396             tz               = _mm_mul_ps(fscal,dz20);
 397
 398             /* Update vectorial force */
 399             fix2             = _mm_add_ps(fix2,tx);
 400             fiy2             = _mm_add_ps(fiy2,ty);
 401             fiz2             = _mm_add_ps(fiz2,tz);
 402
 403             fjx0             = _mm_add_ps(fjx0,tx);
 404             fjy0             = _mm_add_ps(fjy0,ty);
 405             fjz0             = _mm_add_ps(fjz0,tz);
 406
 407             /**************************
 408              * CALCULATE INTERACTIONS *
 409              **************************/
 410
 411             r30              = _mm_mul_ps(rsq30,rinv30);
 412
 413             /* Compute parameters for interactions between i and j atoms */
 414             qq30             = _mm_mul_ps(iq3,jq0);
 415
 416             /* EWALD ELECTROSTATICS */
 417
 418             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 419             ewrt             = _mm_mul_ps(r30,ewtabscale);
 420             ewitab           = _mm_cvttps_epi32(ewrt);
 421             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 422             ewitab           = _mm_slli_epi32(ewitab,2);
 423             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 424             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 425             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 426             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 427             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 428             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 429             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 430             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 431             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 432
 433             /* Update potential sum for this i atom from the interaction with this j atom. */
 434             velecsum         = _mm_add_ps(velecsum,velec);
 435
 436             fscal            = felec;
 437
 438             /* Calculate temporary vectorial force */
 439             tx               = _mm_mul_ps(fscal,dx30);
 440             ty               = _mm_mul_ps(fscal,dy30);
 441             tz               = _mm_mul_ps(fscal,dz30);
 442
 443             /* Update vectorial force */
 444             fix3             = _mm_add_ps(fix3,tx);
 445             fiy3             = _mm_add_ps(fiy3,ty);
 446             fiz3             = _mm_add_ps(fiz3,tz);
 447
 448             fjx0             = _mm_add_ps(fjx0,tx);
 449             fjy0             = _mm_add_ps(fjy0,ty);
 450             fjz0             = _mm_add_ps(fjz0,tz);
 451
 452             fjptrA             = f+j_coord_offsetA;
 453             fjptrB             = f+j_coord_offsetB;
 454             fjptrC             = f+j_coord_offsetC;
 455             fjptrD             = f+j_coord_offsetD;
 456
 457             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 458
 459             /* Inner loop uses 174 flops */
 460         }
 461
 462         if(jidx<j_index_end)
 463         {
 464
 465             /* Get j neighbor index, and coordinate index */
 466             jnrlistA         = jjnr[jidx];
 467             jnrlistB         = jjnr[jidx+1];
 468             jnrlistC         = jjnr[jidx+2];
 469             jnrlistD         = jjnr[jidx+3];
 470             /* Sign of each element will be negative for non-real atoms.
 471              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 472              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 473              */
 474             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 475             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 476             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 477             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 478             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 479             j_coord_offsetA  = DIM*jnrA;
 480             j_coord_offsetB  = DIM*jnrB;
 481             j_coord_offsetC  = DIM*jnrC;
 482             j_coord_offsetD  = DIM*jnrD;
 483
 484             /* load j atom coordinates */
 485             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 486                                               x+j_coord_offsetC,x+j_coord_offsetD,
 487                                               &jx0,&jy0,&jz0);
 488
 489             /* Calculate displacement vector */
 490             dx00             = _mm_sub_ps(ix0,jx0);
 491             dy00             = _mm_sub_ps(iy0,jy0);
 492             dz00             = _mm_sub_ps(iz0,jz0);
 493             dx10             = _mm_sub_ps(ix1,jx0);
 494             dy10             = _mm_sub_ps(iy1,jy0);
 495             dz10             = _mm_sub_ps(iz1,jz0);
 496             dx20             = _mm_sub_ps(ix2,jx0);
 497             dy20             = _mm_sub_ps(iy2,jy0);
 498             dz20             = _mm_sub_ps(iz2,jz0);
 499             dx30             = _mm_sub_ps(ix3,jx0);
 500             dy30             = _mm_sub_ps(iy3,jy0);
 501             dz30             = _mm_sub_ps(iz3,jz0);
 502
 503             /* Calculate squared distance and things based on it */
 504             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 505             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 506             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 507             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 508
 509             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 510             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 511             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 512             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 513
 514             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 515             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 516             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 517             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 518
 519             /* Load parameters for j particles */
 520             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 521                                                               charge+jnrC+0,charge+jnrD+0);
 522             vdwjidx0A        = 2*vdwtype[jnrA+0];
 523             vdwjidx0B        = 2*vdwtype[jnrB+0];
 524             vdwjidx0C        = 2*vdwtype[jnrC+0];
 525             vdwjidx0D        = 2*vdwtype[jnrD+0];
 526
 527             fjx0             = _mm_setzero_ps();
 528             fjy0             = _mm_setzero_ps();
 529             fjz0             = _mm_setzero_ps();
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             r00              = _mm_mul_ps(rsq00,rinv00);
 536             r00              = _mm_andnot_ps(dummy_mask,r00);
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 540                                          vdwparam+vdwioffset0+vdwjidx0B,
 541                                          vdwparam+vdwioffset0+vdwjidx0C,
 542                                          vdwparam+vdwioffset0+vdwjidx0D,
 543                                          &c6_00,&c12_00);
 544
 545             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 546                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 547                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 548                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 549
 550             /* Analytical LJ-PME */
 551             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 552             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 553             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 554             exponent         = gmx_simd_exp_r(ewcljrsq);
 555             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 556             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 557             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 558             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 559             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 560             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 561             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 562             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);
 563
 564             /* Update potential sum for this i atom from the interaction with this j atom. */
 565             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 566             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 567
 568             fscal            = fvdw;
 569
 570             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 571
 572             /* Calculate temporary vectorial force */
 573             tx               = _mm_mul_ps(fscal,dx00);
 574             ty               = _mm_mul_ps(fscal,dy00);
 575             tz               = _mm_mul_ps(fscal,dz00);
 576
 577             /* Update vectorial force */
 578             fix0             = _mm_add_ps(fix0,tx);
 579             fiy0             = _mm_add_ps(fiy0,ty);
 580             fiz0             = _mm_add_ps(fiz0,tz);
 581
 582             fjx0             = _mm_add_ps(fjx0,tx);
 583             fjy0             = _mm_add_ps(fjy0,ty);
 584             fjz0             = _mm_add_ps(fjz0,tz);
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             r10              = _mm_mul_ps(rsq10,rinv10);
 591             r10              = _mm_andnot_ps(dummy_mask,r10);
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq10             = _mm_mul_ps(iq1,jq0);
 595
 596             /* EWALD ELECTROSTATICS */
 597
 598             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 599             ewrt             = _mm_mul_ps(r10,ewtabscale);
 600             ewitab           = _mm_cvttps_epi32(ewrt);
 601             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 602             ewitab           = _mm_slli_epi32(ewitab,2);
 603             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 604             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 605             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 606             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 607             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 608             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 609             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 610             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 611             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velec            = _mm_andnot_ps(dummy_mask,velec);
 615             velecsum         = _mm_add_ps(velecsum,velec);
 616
 617             fscal            = felec;
 618
 619             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 620
 621             /* Calculate temporary vectorial force */
 622             tx               = _mm_mul_ps(fscal,dx10);
 623             ty               = _mm_mul_ps(fscal,dy10);
 624             tz               = _mm_mul_ps(fscal,dz10);
 625
 626             /* Update vectorial force */
 627             fix1             = _mm_add_ps(fix1,tx);
 628             fiy1             = _mm_add_ps(fiy1,ty);
 629             fiz1             = _mm_add_ps(fiz1,tz);
 630
 631             fjx0             = _mm_add_ps(fjx0,tx);
 632             fjy0             = _mm_add_ps(fjy0,ty);
 633             fjz0             = _mm_add_ps(fjz0,tz);
 634
 635             /**************************
 636              * CALCULATE INTERACTIONS *
 637              **************************/
 638
 639             r20              = _mm_mul_ps(rsq20,rinv20);
 640             r20              = _mm_andnot_ps(dummy_mask,r20);
 641
 642             /* Compute parameters for interactions between i and j atoms */
 643             qq20             = _mm_mul_ps(iq2,jq0);
 644
 645             /* EWALD ELECTROSTATICS */
 646
 647             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 648             ewrt             = _mm_mul_ps(r20,ewtabscale);
 649             ewitab           = _mm_cvttps_epi32(ewrt);
 650             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 651             ewitab           = _mm_slli_epi32(ewitab,2);
 652             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 653             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 654             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 655             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 656             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 657             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 658             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 659             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 660             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 661
 662             /* Update potential sum for this i atom from the interaction with this j atom. */
 663             velec            = _mm_andnot_ps(dummy_mask,velec);
 664             velecsum         = _mm_add_ps(velecsum,velec);
 665
 666             fscal            = felec;
 667
 668             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm_mul_ps(fscal,dx20);
 672             ty               = _mm_mul_ps(fscal,dy20);
 673             tz               = _mm_mul_ps(fscal,dz20);
 674
 675             /* Update vectorial force */
 676             fix2             = _mm_add_ps(fix2,tx);
 677             fiy2             = _mm_add_ps(fiy2,ty);
 678             fiz2             = _mm_add_ps(fiz2,tz);
 679
 680             fjx0             = _mm_add_ps(fjx0,tx);
 681             fjy0             = _mm_add_ps(fjy0,ty);
 682             fjz0             = _mm_add_ps(fjz0,tz);
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             r30              = _mm_mul_ps(rsq30,rinv30);
 689             r30              = _mm_andnot_ps(dummy_mask,r30);
 690
 691             /* Compute parameters for interactions between i and j atoms */
 692             qq30             = _mm_mul_ps(iq3,jq0);
 693
 694             /* EWALD ELECTROSTATICS */
 695
 696             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 697             ewrt             = _mm_mul_ps(r30,ewtabscale);
 698             ewitab           = _mm_cvttps_epi32(ewrt);
 699             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 700             ewitab           = _mm_slli_epi32(ewitab,2);
 701             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 702             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 703             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 704             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 705             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 706             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 707             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 708             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 709             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 710
 711             /* Update potential sum for this i atom from the interaction with this j atom. */
 712             velec            = _mm_andnot_ps(dummy_mask,velec);
 713             velecsum         = _mm_add_ps(velecsum,velec);
 714
 715             fscal            = felec;
 716
 717             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 718
 719             /* Calculate temporary vectorial force */
 720             tx               = _mm_mul_ps(fscal,dx30);
 721             ty               = _mm_mul_ps(fscal,dy30);
 722             tz               = _mm_mul_ps(fscal,dz30);
 723
 724             /* Update vectorial force */
 725             fix3             = _mm_add_ps(fix3,tx);
 726             fiy3             = _mm_add_ps(fiy3,ty);
 727             fiz3             = _mm_add_ps(fiz3,tz);
 728
 729             fjx0             = _mm_add_ps(fjx0,tx);
 730             fjy0             = _mm_add_ps(fjy0,ty);
 731             fjz0             = _mm_add_ps(fjz0,tz);
 732
 733             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 734             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 735             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 736             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 737
 738             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 739
 740             /* Inner loop uses 178 flops */
 741         }
 742
 743         /* End of innermost loop */
 744
 745         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 746                                               f+i_coord_offset,fshift+i_shift_offset);
 747
 748         ggid                        = gid[iidx];
 749         /* Update potential energies */
 750         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 751         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 752
 753         /* Increment number of inner iterations */
 754         inneriter                  += j_index_end - j_index_start;
 755
 756         /* Outer loop uses 26 flops */
 757     }
 758
 759     /* Increment number of outer iterations */
 760     outeriter        += nri;
 761
 762     /* Update outer/inner flops */
 763
 764     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*178);
 765 }
 766 /*
 767  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_single
 768  * Electrostatics interaction: Ewald
 769  * VdW interaction:            LJEwald
 770  * Geometry:                   Water4-Particle
 771  * Calculate force/pot:        Force
 772  */
 773 void
 774 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse4_1_single
 775                     (t_nblist                    * gmx_restrict       nlist,
 776                      rvec                        * gmx_restrict          xx,
 777                      rvec                        * gmx_restrict          ff,
 778                      t_forcerec                  * gmx_restrict          fr,
 779                      t_mdatoms                   * gmx_restrict     mdatoms,
 780                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 781                      t_nrnb                      * gmx_restrict        nrnb)
 782 {
 783     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 784      * just 0 for non-waters.
 785      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 786      * jnr indices corresponding to data put in the four positions in the SIMD register.
 787      */
 788     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 789     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 790     int              jnrA,jnrB,jnrC,jnrD;
 791     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 792     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 793     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 794     real             rcutoff_scalar;
 795     real             *shiftvec,*fshift,*x,*f;
 796     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 797     real             scratch[4*DIM];
 798     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 799     int              vdwioffset0;
 800     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 801     int              vdwioffset1;
 802     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 803     int              vdwioffset2;
 804     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 805     int              vdwioffset3;
 806     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 807     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 808     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 809     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 810     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 811     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 812     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 813     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 814     real             *charge;
 815     int              nvdwtype;
 816     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 817     int              *vdwtype;
 818     real             *vdwparam;
 819     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 820     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 821     __m128           c6grid_00;
 822     __m128           c6grid_10;
 823     __m128           c6grid_20;
 824     __m128           c6grid_30;
 825     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 826     real             *vdwgridparam;
 827     __m128           one_half  = _mm_set1_ps(0.5);
 828     __m128           minus_one = _mm_set1_ps(-1.0);
 829     __m128i          ewitab;
 830     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 831     real             *ewtab;
 832     __m128           dummy_mask,cutoff_mask;
 833     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 834     __m128           one     = _mm_set1_ps(1.0);
 835     __m128           two     = _mm_set1_ps(2.0);
 836     x                = xx[0];
 837     f                = ff[0];
 838
 839     nri              = nlist->nri;
 840     iinr             = nlist->iinr;
 841     jindex           = nlist->jindex;
 842     jjnr             = nlist->jjnr;
 843     shiftidx         = nlist->shift;
 844     gid              = nlist->gid;
 845     shiftvec         = fr->shift_vec[0];
 846     fshift           = fr->fshift[0];
 847     facel            = _mm_set1_ps(fr->epsfac);
 848     charge           = mdatoms->chargeA;
 849     nvdwtype         = fr->ntype;
 850     vdwparam         = fr->nbfp;
 851     vdwtype          = mdatoms->typeA;
 852     vdwgridparam     = fr->ljpme_c6grid;
 853     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 854     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 855     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 856
 857     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 858     ewtab            = fr->ic->tabq_coul_F;
 859     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 860     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 861
 862     /* Setup water-specific parameters */
 863     inr              = nlist->iinr[0];
 864     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 865     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 866     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 867     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 868
 869     /* Avoid stupid compiler warnings */
 870     jnrA = jnrB = jnrC = jnrD = 0;
 871     j_coord_offsetA = 0;
 872     j_coord_offsetB = 0;
 873     j_coord_offsetC = 0;
 874     j_coord_offsetD = 0;
 875
 876     outeriter        = 0;
 877     inneriter        = 0;
 878
 879     for(iidx=0;iidx<4*DIM;iidx++)
 880     {
 881         scratch[iidx] = 0.0;
 882     }
 883
 884     /* Start outer loop over neighborlists */
 885     for(iidx=0; iidx<nri; iidx++)
 886     {
 887         /* Load shift vector for this list */
 888         i_shift_offset   = DIM*shiftidx[iidx];
 889
 890         /* Load limits for loop over neighbors */
 891         j_index_start    = jindex[iidx];
 892         j_index_end      = jindex[iidx+1];
 893
 894         /* Get outer coordinate index */
 895         inr              = iinr[iidx];
 896         i_coord_offset   = DIM*inr;
 897
 898         /* Load i particle coords and add shift vector */
 899         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 900                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 901
 902         fix0             = _mm_setzero_ps();
 903         fiy0             = _mm_setzero_ps();
 904         fiz0             = _mm_setzero_ps();
 905         fix1             = _mm_setzero_ps();
 906         fiy1             = _mm_setzero_ps();
 907         fiz1             = _mm_setzero_ps();
 908         fix2             = _mm_setzero_ps();
 909         fiy2             = _mm_setzero_ps();
 910         fiz2             = _mm_setzero_ps();
 911         fix3             = _mm_setzero_ps();
 912         fiy3             = _mm_setzero_ps();
 913         fiz3             = _mm_setzero_ps();
 914
 915         /* Start inner kernel loop */
 916         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 917         {
 918
 919             /* Get j neighbor index, and coordinate index */
 920             jnrA             = jjnr[jidx];
 921             jnrB             = jjnr[jidx+1];
 922             jnrC             = jjnr[jidx+2];
 923             jnrD             = jjnr[jidx+3];
 924             j_coord_offsetA  = DIM*jnrA;
 925             j_coord_offsetB  = DIM*jnrB;
 926             j_coord_offsetC  = DIM*jnrC;
 927             j_coord_offsetD  = DIM*jnrD;
 928
 929             /* load j atom coordinates */
 930             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 931                                               x+j_coord_offsetC,x+j_coord_offsetD,
 932                                               &jx0,&jy0,&jz0);
 933
 934             /* Calculate displacement vector */
 935             dx00             = _mm_sub_ps(ix0,jx0);
 936             dy00             = _mm_sub_ps(iy0,jy0);
 937             dz00             = _mm_sub_ps(iz0,jz0);
 938             dx10             = _mm_sub_ps(ix1,jx0);
 939             dy10             = _mm_sub_ps(iy1,jy0);
 940             dz10             = _mm_sub_ps(iz1,jz0);
 941             dx20             = _mm_sub_ps(ix2,jx0);
 942             dy20             = _mm_sub_ps(iy2,jy0);
 943             dz20             = _mm_sub_ps(iz2,jz0);
 944             dx30             = _mm_sub_ps(ix3,jx0);
 945             dy30             = _mm_sub_ps(iy3,jy0);
 946             dz30             = _mm_sub_ps(iz3,jz0);
 947
 948             /* Calculate squared distance and things based on it */
 949             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 950             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 951             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 952             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 953
 954             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 955             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 956             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 957             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 958
 959             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 960             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 961             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 962             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 963
 964             /* Load parameters for j particles */
 965             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 966                                                               charge+jnrC+0,charge+jnrD+0);
 967             vdwjidx0A        = 2*vdwtype[jnrA+0];
 968             vdwjidx0B        = 2*vdwtype[jnrB+0];
 969             vdwjidx0C        = 2*vdwtype[jnrC+0];
 970             vdwjidx0D        = 2*vdwtype[jnrD+0];
 971
 972             fjx0             = _mm_setzero_ps();
 973             fjy0             = _mm_setzero_ps();
 974             fjz0             = _mm_setzero_ps();
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             r00              = _mm_mul_ps(rsq00,rinv00);
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 984                                          vdwparam+vdwioffset0+vdwjidx0B,
 985                                          vdwparam+vdwioffset0+vdwjidx0C,
 986                                          vdwparam+vdwioffset0+vdwjidx0D,
 987                                          &c6_00,&c12_00);
 988
 989             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 990                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 991                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 992                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 993
 994             /* Analytical LJ-PME */
 995             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 996             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 997             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 998             exponent         = gmx_simd_exp_r(ewcljrsq);
 999             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1000             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1001             /* f6A = 6 * C6grid * (1 - poly) */
1002             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1003             /* f6B = C6grid * exponent * beta^6 */
1004             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1005             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1006             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);
1007
1008             fscal            = fvdw;
1009
1010             /* Calculate temporary vectorial force */
1011             tx               = _mm_mul_ps(fscal,dx00);
1012             ty               = _mm_mul_ps(fscal,dy00);
1013             tz               = _mm_mul_ps(fscal,dz00);
1014
1015             /* Update vectorial force */
1016             fix0             = _mm_add_ps(fix0,tx);
1017             fiy0             = _mm_add_ps(fiy0,ty);
1018             fiz0             = _mm_add_ps(fiz0,tz);
1019
1020             fjx0             = _mm_add_ps(fjx0,tx);
1021             fjy0             = _mm_add_ps(fjy0,ty);
1022             fjz0             = _mm_add_ps(fjz0,tz);
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             r10              = _mm_mul_ps(rsq10,rinv10);
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             qq10             = _mm_mul_ps(iq1,jq0);
1032
1033             /* EWALD ELECTROSTATICS */
1034
1035             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1036             ewrt             = _mm_mul_ps(r10,ewtabscale);
1037             ewitab           = _mm_cvttps_epi32(ewrt);
1038             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1039             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1040                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1041                                          &ewtabF,&ewtabFn);
1042             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1043             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1044
1045             fscal            = felec;
1046
1047             /* Calculate temporary vectorial force */
1048             tx               = _mm_mul_ps(fscal,dx10);
1049             ty               = _mm_mul_ps(fscal,dy10);
1050             tz               = _mm_mul_ps(fscal,dz10);
1051
1052             /* Update vectorial force */
1053             fix1             = _mm_add_ps(fix1,tx);
1054             fiy1             = _mm_add_ps(fiy1,ty);
1055             fiz1             = _mm_add_ps(fiz1,tz);
1056
1057             fjx0             = _mm_add_ps(fjx0,tx);
1058             fjy0             = _mm_add_ps(fjy0,ty);
1059             fjz0             = _mm_add_ps(fjz0,tz);
1060
1061             /**************************
1062              * CALCULATE INTERACTIONS *
1063              **************************/
1064
1065             r20              = _mm_mul_ps(rsq20,rinv20);
1066
1067             /* Compute parameters for interactions between i and j atoms */
1068             qq20             = _mm_mul_ps(iq2,jq0);
1069
1070             /* EWALD ELECTROSTATICS */
1071
1072             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1073             ewrt             = _mm_mul_ps(r20,ewtabscale);
1074             ewitab           = _mm_cvttps_epi32(ewrt);
1075             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1076             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1077                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1078                                          &ewtabF,&ewtabFn);
1079             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1080             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1081
1082             fscal            = felec;
1083
1084             /* Calculate temporary vectorial force */
1085             tx               = _mm_mul_ps(fscal,dx20);
1086             ty               = _mm_mul_ps(fscal,dy20);
1087             tz               = _mm_mul_ps(fscal,dz20);
1088
1089             /* Update vectorial force */
1090             fix2             = _mm_add_ps(fix2,tx);
1091             fiy2             = _mm_add_ps(fiy2,ty);
1092             fiz2             = _mm_add_ps(fiz2,tz);
1093
1094             fjx0             = _mm_add_ps(fjx0,tx);
1095             fjy0             = _mm_add_ps(fjy0,ty);
1096             fjz0             = _mm_add_ps(fjz0,tz);
1097
1098             /**************************
1099              * CALCULATE INTERACTIONS *
1100              **************************/
1101
1102             r30              = _mm_mul_ps(rsq30,rinv30);
1103
1104             /* Compute parameters for interactions between i and j atoms */
1105             qq30             = _mm_mul_ps(iq3,jq0);
1106
1107             /* EWALD ELECTROSTATICS */
1108
1109             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1110             ewrt             = _mm_mul_ps(r30,ewtabscale);
1111             ewitab           = _mm_cvttps_epi32(ewrt);
1112             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1113             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1114                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1115                                          &ewtabF,&ewtabFn);
1116             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1117             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1118
1119             fscal            = felec;
1120
1121             /* Calculate temporary vectorial force */
1122             tx               = _mm_mul_ps(fscal,dx30);
1123             ty               = _mm_mul_ps(fscal,dy30);
1124             tz               = _mm_mul_ps(fscal,dz30);
1125
1126             /* Update vectorial force */
1127             fix3             = _mm_add_ps(fix3,tx);
1128             fiy3             = _mm_add_ps(fiy3,ty);
1129             fiz3             = _mm_add_ps(fiz3,tz);
1130
1131             fjx0             = _mm_add_ps(fjx0,tx);
1132             fjy0             = _mm_add_ps(fjy0,ty);
1133             fjz0             = _mm_add_ps(fjz0,tz);
1134
1135             fjptrA             = f+j_coord_offsetA;
1136             fjptrB             = f+j_coord_offsetB;
1137             fjptrC             = f+j_coord_offsetC;
1138             fjptrD             = f+j_coord_offsetD;
1139
1140             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1141
1142             /* Inner loop uses 154 flops */
1143         }
1144
1145         if(jidx<j_index_end)
1146         {
1147
1148             /* Get j neighbor index, and coordinate index */
1149             jnrlistA         = jjnr[jidx];
1150             jnrlistB         = jjnr[jidx+1];
1151             jnrlistC         = jjnr[jidx+2];
1152             jnrlistD         = jjnr[jidx+3];
1153             /* Sign of each element will be negative for non-real atoms.
1154              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1155              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1156              */
1157             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1158             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1159             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1160             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1161             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1162             j_coord_offsetA  = DIM*jnrA;
1163             j_coord_offsetB  = DIM*jnrB;
1164             j_coord_offsetC  = DIM*jnrC;
1165             j_coord_offsetD  = DIM*jnrD;
1166
1167             /* load j atom coordinates */
1168             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1169                                               x+j_coord_offsetC,x+j_coord_offsetD,
1170                                               &jx0,&jy0,&jz0);
1171
1172             /* Calculate displacement vector */
1173             dx00             = _mm_sub_ps(ix0,jx0);
1174             dy00             = _mm_sub_ps(iy0,jy0);
1175             dz00             = _mm_sub_ps(iz0,jz0);
1176             dx10             = _mm_sub_ps(ix1,jx0);
1177             dy10             = _mm_sub_ps(iy1,jy0);
1178             dz10             = _mm_sub_ps(iz1,jz0);
1179             dx20             = _mm_sub_ps(ix2,jx0);
1180             dy20             = _mm_sub_ps(iy2,jy0);
1181             dz20             = _mm_sub_ps(iz2,jz0);
1182             dx30             = _mm_sub_ps(ix3,jx0);
1183             dy30             = _mm_sub_ps(iy3,jy0);
1184             dz30             = _mm_sub_ps(iz3,jz0);
1185
1186             /* Calculate squared distance and things based on it */
1187             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1188             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1189             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1190             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1191
1192             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1193             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1194             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1195             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1196
1197             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1198             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1199             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1200             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1201
1202             /* Load parameters for j particles */
1203             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1204                                                               charge+jnrC+0,charge+jnrD+0);
1205             vdwjidx0A        = 2*vdwtype[jnrA+0];
1206             vdwjidx0B        = 2*vdwtype[jnrB+0];
1207             vdwjidx0C        = 2*vdwtype[jnrC+0];
1208             vdwjidx0D        = 2*vdwtype[jnrD+0];
1209
1210             fjx0             = _mm_setzero_ps();
1211             fjy0             = _mm_setzero_ps();
1212             fjz0             = _mm_setzero_ps();
1213
1214             /**************************
1215              * CALCULATE INTERACTIONS *
1216              **************************/
1217
1218             r00              = _mm_mul_ps(rsq00,rinv00);
1219             r00              = _mm_andnot_ps(dummy_mask,r00);
1220
1221             /* Compute parameters for interactions between i and j atoms */
1222             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1223                                          vdwparam+vdwioffset0+vdwjidx0B,
1224                                          vdwparam+vdwioffset0+vdwjidx0C,
1225                                          vdwparam+vdwioffset0+vdwjidx0D,
1226                                          &c6_00,&c12_00);
1227
1228             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1229                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1230                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1231                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1232
1233             /* Analytical LJ-PME */
1234             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1235             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1236             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1237             exponent         = gmx_simd_exp_r(ewcljrsq);
1238             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1239             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1240             /* f6A = 6 * C6grid * (1 - poly) */
1241             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1242             /* f6B = C6grid * exponent * beta^6 */
1243             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1244             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1245             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);
1246
1247             fscal            = fvdw;
1248
1249             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1250
1251             /* Calculate temporary vectorial force */
1252             tx               = _mm_mul_ps(fscal,dx00);
1253             ty               = _mm_mul_ps(fscal,dy00);
1254             tz               = _mm_mul_ps(fscal,dz00);
1255
1256             /* Update vectorial force */
1257             fix0             = _mm_add_ps(fix0,tx);
1258             fiy0             = _mm_add_ps(fiy0,ty);
1259             fiz0             = _mm_add_ps(fiz0,tz);
1260
1261             fjx0             = _mm_add_ps(fjx0,tx);
1262             fjy0             = _mm_add_ps(fjy0,ty);
1263             fjz0             = _mm_add_ps(fjz0,tz);
1264
1265             /**************************
1266              * CALCULATE INTERACTIONS *
1267              **************************/
1268
1269             r10              = _mm_mul_ps(rsq10,rinv10);
1270             r10              = _mm_andnot_ps(dummy_mask,r10);
1271
1272             /* Compute parameters for interactions between i and j atoms */
1273             qq10             = _mm_mul_ps(iq1,jq0);
1274
1275             /* EWALD ELECTROSTATICS */
1276
1277             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1278             ewrt             = _mm_mul_ps(r10,ewtabscale);
1279             ewitab           = _mm_cvttps_epi32(ewrt);
1280             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1281             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1282                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1283                                          &ewtabF,&ewtabFn);
1284             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1285             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1286
1287             fscal            = felec;
1288
1289             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1290
1291             /* Calculate temporary vectorial force */
1292             tx               = _mm_mul_ps(fscal,dx10);
1293             ty               = _mm_mul_ps(fscal,dy10);
1294             tz               = _mm_mul_ps(fscal,dz10);
1295
1296             /* Update vectorial force */
1297             fix1             = _mm_add_ps(fix1,tx);
1298             fiy1             = _mm_add_ps(fiy1,ty);
1299             fiz1             = _mm_add_ps(fiz1,tz);
1300
1301             fjx0             = _mm_add_ps(fjx0,tx);
1302             fjy0             = _mm_add_ps(fjy0,ty);
1303             fjz0             = _mm_add_ps(fjz0,tz);
1304
1305             /**************************
1306              * CALCULATE INTERACTIONS *
1307              **************************/
1308
1309             r20              = _mm_mul_ps(rsq20,rinv20);
1310             r20              = _mm_andnot_ps(dummy_mask,r20);
1311
1312             /* Compute parameters for interactions between i and j atoms */
1313             qq20             = _mm_mul_ps(iq2,jq0);
1314
1315             /* EWALD ELECTROSTATICS */
1316
1317             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1318             ewrt             = _mm_mul_ps(r20,ewtabscale);
1319             ewitab           = _mm_cvttps_epi32(ewrt);
1320             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1321             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1322                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1323                                          &ewtabF,&ewtabFn);
1324             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1325             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1326
1327             fscal            = felec;
1328
1329             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1330
1331             /* Calculate temporary vectorial force */
1332             tx               = _mm_mul_ps(fscal,dx20);
1333             ty               = _mm_mul_ps(fscal,dy20);
1334             tz               = _mm_mul_ps(fscal,dz20);
1335
1336             /* Update vectorial force */
1337             fix2             = _mm_add_ps(fix2,tx);
1338             fiy2             = _mm_add_ps(fiy2,ty);
1339             fiz2             = _mm_add_ps(fiz2,tz);
1340
1341             fjx0             = _mm_add_ps(fjx0,tx);
1342             fjy0             = _mm_add_ps(fjy0,ty);
1343             fjz0             = _mm_add_ps(fjz0,tz);
1344
1345             /**************************
1346              * CALCULATE INTERACTIONS *
1347              **************************/
1348
1349             r30              = _mm_mul_ps(rsq30,rinv30);
1350             r30              = _mm_andnot_ps(dummy_mask,r30);
1351
1352             /* Compute parameters for interactions between i and j atoms */
1353             qq30             = _mm_mul_ps(iq3,jq0);
1354
1355             /* EWALD ELECTROSTATICS */
1356
1357             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1358             ewrt             = _mm_mul_ps(r30,ewtabscale);
1359             ewitab           = _mm_cvttps_epi32(ewrt);
1360             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1361             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1362                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1363                                          &ewtabF,&ewtabFn);
1364             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1365             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1366
1367             fscal            = felec;
1368
1369             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1370
1371             /* Calculate temporary vectorial force */
1372             tx               = _mm_mul_ps(fscal,dx30);
1373             ty               = _mm_mul_ps(fscal,dy30);
1374             tz               = _mm_mul_ps(fscal,dz30);
1375
1376             /* Update vectorial force */
1377             fix3             = _mm_add_ps(fix3,tx);
1378             fiy3             = _mm_add_ps(fiy3,ty);
1379             fiz3             = _mm_add_ps(fiz3,tz);
1380
1381             fjx0             = _mm_add_ps(fjx0,tx);
1382             fjy0             = _mm_add_ps(fjy0,ty);
1383             fjz0             = _mm_add_ps(fjz0,tz);
1384
1385             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1386             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1387             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1388             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1389
1390             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1391
1392             /* Inner loop uses 158 flops */
1393         }
1394
1395         /* End of innermost loop */
1396
1397         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1398                                               f+i_coord_offset,fshift+i_shift_offset);
1399
1400         /* Increment number of inner iterations */
1401         inneriter                  += j_index_end - j_index_start;
1402
1403         /* Outer loop uses 24 flops */
1404     }
1405
1406     /* Increment number of outer iterations */
1407     outeriter        += nri;
1408
1409     /* Update outer/inner flops */
1410
1411     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*158);
1412 }