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, 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/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_sse2_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_sse2_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             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 279                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 280                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 281                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 282
 283             /* Analytical LJ-PME */
 284             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 285             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 286             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 287             exponent         = gmx_simd_exp_r(ewcljrsq);
 288             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 289             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 290             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 291             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 292             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 293             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 294             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 295             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);
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 299
 300             fscal            = fvdw;
 301
 302             /* Calculate temporary vectorial force */
 303             tx               = _mm_mul_ps(fscal,dx00);
 304             ty               = _mm_mul_ps(fscal,dy00);
 305             tz               = _mm_mul_ps(fscal,dz00);
 306
 307             /* Update vectorial force */
 308             fix0             = _mm_add_ps(fix0,tx);
 309             fiy0             = _mm_add_ps(fiy0,ty);
 310             fiz0             = _mm_add_ps(fiz0,tz);
 311
 312             fjx0             = _mm_add_ps(fjx0,tx);
 313             fjy0             = _mm_add_ps(fjy0,ty);
 314             fjz0             = _mm_add_ps(fjz0,tz);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r10              = _mm_mul_ps(rsq10,rinv10);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_ps(iq1,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 328             ewrt             = _mm_mul_ps(r10,ewtabscale);
 329             ewitab           = _mm_cvttps_epi32(ewrt);
 330             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 334             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 335             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 336             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 337             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 338             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 339             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 340             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             /* Calculate temporary vectorial force */
 348             tx               = _mm_mul_ps(fscal,dx10);
 349             ty               = _mm_mul_ps(fscal,dy10);
 350             tz               = _mm_mul_ps(fscal,dz10);
 351
 352             /* Update vectorial force */
 353             fix1             = _mm_add_ps(fix1,tx);
 354             fiy1             = _mm_add_ps(fiy1,ty);
 355             fiz1             = _mm_add_ps(fiz1,tz);
 356
 357             fjx0             = _mm_add_ps(fjx0,tx);
 358             fjy0             = _mm_add_ps(fjy0,ty);
 359             fjz0             = _mm_add_ps(fjz0,tz);
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             r20              = _mm_mul_ps(rsq20,rinv20);
 366
 367             /* Compute parameters for interactions between i and j atoms */
 368             qq20             = _mm_mul_ps(iq2,jq0);
 369
 370             /* EWALD ELECTROSTATICS */
 371
 372             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 373             ewrt             = _mm_mul_ps(r20,ewtabscale);
 374             ewitab           = _mm_cvttps_epi32(ewrt);
 375             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 376             ewitab           = _mm_slli_epi32(ewitab,2);
 377             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 378             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 379             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 380             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 381             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 382             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 383             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 384             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 385             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velecsum         = _mm_add_ps(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm_mul_ps(fscal,dx20);
 394             ty               = _mm_mul_ps(fscal,dy20);
 395             tz               = _mm_mul_ps(fscal,dz20);
 396
 397             /* Update vectorial force */
 398             fix2             = _mm_add_ps(fix2,tx);
 399             fiy2             = _mm_add_ps(fiy2,ty);
 400             fiz2             = _mm_add_ps(fiz2,tz);
 401
 402             fjx0             = _mm_add_ps(fjx0,tx);
 403             fjy0             = _mm_add_ps(fjy0,ty);
 404             fjz0             = _mm_add_ps(fjz0,tz);
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             r30              = _mm_mul_ps(rsq30,rinv30);
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq30             = _mm_mul_ps(iq3,jq0);
 414
 415             /* EWALD ELECTROSTATICS */
 416
 417             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 418             ewrt             = _mm_mul_ps(r30,ewtabscale);
 419             ewitab           = _mm_cvttps_epi32(ewrt);
 420             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 421             ewitab           = _mm_slli_epi32(ewitab,2);
 422             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 423             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 424             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 425             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 426             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 427             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 428             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 429             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 430             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 431
 432             /* Update potential sum for this i atom from the interaction with this j atom. */
 433             velecsum         = _mm_add_ps(velecsum,velec);
 434
 435             fscal            = felec;
 436
 437             /* Calculate temporary vectorial force */
 438             tx               = _mm_mul_ps(fscal,dx30);
 439             ty               = _mm_mul_ps(fscal,dy30);
 440             tz               = _mm_mul_ps(fscal,dz30);
 441
 442             /* Update vectorial force */
 443             fix3             = _mm_add_ps(fix3,tx);
 444             fiy3             = _mm_add_ps(fiy3,ty);
 445             fiz3             = _mm_add_ps(fiz3,tz);
 446
 447             fjx0             = _mm_add_ps(fjx0,tx);
 448             fjy0             = _mm_add_ps(fjy0,ty);
 449             fjz0             = _mm_add_ps(fjz0,tz);
 450
 451             fjptrA             = f+j_coord_offsetA;
 452             fjptrB             = f+j_coord_offsetB;
 453             fjptrC             = f+j_coord_offsetC;
 454             fjptrD             = f+j_coord_offsetD;
 455
 456             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 457
 458             /* Inner loop uses 174 flops */
 459         }
 460
 461         if(jidx<j_index_end)
 462         {
 463
 464             /* Get j neighbor index, and coordinate index */
 465             jnrlistA         = jjnr[jidx];
 466             jnrlistB         = jjnr[jidx+1];
 467             jnrlistC         = jjnr[jidx+2];
 468             jnrlistD         = jjnr[jidx+3];
 469             /* Sign of each element will be negative for non-real atoms.
 470              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 471              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 472              */
 473             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 474             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 475             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 476             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 477             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 478             j_coord_offsetA  = DIM*jnrA;
 479             j_coord_offsetB  = DIM*jnrB;
 480             j_coord_offsetC  = DIM*jnrC;
 481             j_coord_offsetD  = DIM*jnrD;
 482
 483             /* load j atom coordinates */
 484             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 485                                               x+j_coord_offsetC,x+j_coord_offsetD,
 486                                               &jx0,&jy0,&jz0);
 487
 488             /* Calculate displacement vector */
 489             dx00             = _mm_sub_ps(ix0,jx0);
 490             dy00             = _mm_sub_ps(iy0,jy0);
 491             dz00             = _mm_sub_ps(iz0,jz0);
 492             dx10             = _mm_sub_ps(ix1,jx0);
 493             dy10             = _mm_sub_ps(iy1,jy0);
 494             dz10             = _mm_sub_ps(iz1,jz0);
 495             dx20             = _mm_sub_ps(ix2,jx0);
 496             dy20             = _mm_sub_ps(iy2,jy0);
 497             dz20             = _mm_sub_ps(iz2,jz0);
 498             dx30             = _mm_sub_ps(ix3,jx0);
 499             dy30             = _mm_sub_ps(iy3,jy0);
 500             dz30             = _mm_sub_ps(iz3,jz0);
 501
 502             /* Calculate squared distance and things based on it */
 503             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 504             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 505             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 506             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 507
 508             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 509             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 510             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 511             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 512
 513             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 514             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 515             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 516             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 517
 518             /* Load parameters for j particles */
 519             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 520                                                               charge+jnrC+0,charge+jnrD+0);
 521             vdwjidx0A        = 2*vdwtype[jnrA+0];
 522             vdwjidx0B        = 2*vdwtype[jnrB+0];
 523             vdwjidx0C        = 2*vdwtype[jnrC+0];
 524             vdwjidx0D        = 2*vdwtype[jnrD+0];
 525
 526             fjx0             = _mm_setzero_ps();
 527             fjy0             = _mm_setzero_ps();
 528             fjz0             = _mm_setzero_ps();
 529
 530             /**************************
 531              * CALCULATE INTERACTIONS *
 532              **************************/
 533
 534             r00              = _mm_mul_ps(rsq00,rinv00);
 535             r00              = _mm_andnot_ps(dummy_mask,r00);
 536
 537             /* Compute parameters for interactions between i and j atoms */
 538             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 539                                          vdwparam+vdwioffset0+vdwjidx0B,
 540                                          vdwparam+vdwioffset0+vdwjidx0C,
 541                                          vdwparam+vdwioffset0+vdwjidx0D,
 542                                          &c6_00,&c12_00);
 543             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 544                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 545                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 546                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 547
 548             /* Analytical LJ-PME */
 549             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 550             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 551             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 552             exponent         = gmx_simd_exp_r(ewcljrsq);
 553             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 554             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 555             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 556             vvdw6            = _mm_mul_ps(_mm_sub_ps(c6_00,_mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly))),rinvsix);
 557             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 558             vvdw             = _mm_sub_ps(_mm_mul_ps(vvdw12,one_twelfth),_mm_mul_ps(vvdw6,one_sixth));
 559             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 560             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);
 561
 562             /* Update potential sum for this i atom from the interaction with this j atom. */
 563             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 564             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 565
 566             fscal            = fvdw;
 567
 568             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 569
 570             /* Calculate temporary vectorial force */
 571             tx               = _mm_mul_ps(fscal,dx00);
 572             ty               = _mm_mul_ps(fscal,dy00);
 573             tz               = _mm_mul_ps(fscal,dz00);
 574
 575             /* Update vectorial force */
 576             fix0             = _mm_add_ps(fix0,tx);
 577             fiy0             = _mm_add_ps(fiy0,ty);
 578             fiz0             = _mm_add_ps(fiz0,tz);
 579
 580             fjx0             = _mm_add_ps(fjx0,tx);
 581             fjy0             = _mm_add_ps(fjy0,ty);
 582             fjz0             = _mm_add_ps(fjz0,tz);
 583
 584             /**************************
 585              * CALCULATE INTERACTIONS *
 586              **************************/
 587
 588             r10              = _mm_mul_ps(rsq10,rinv10);
 589             r10              = _mm_andnot_ps(dummy_mask,r10);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq10             = _mm_mul_ps(iq1,jq0);
 593
 594             /* EWALD ELECTROSTATICS */
 595
 596             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 597             ewrt             = _mm_mul_ps(r10,ewtabscale);
 598             ewitab           = _mm_cvttps_epi32(ewrt);
 599             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 600             ewitab           = _mm_slli_epi32(ewitab,2);
 601             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 602             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 603             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 604             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 605             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 606             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 607             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 608             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 609             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 610
 611             /* Update potential sum for this i atom from the interaction with this j atom. */
 612             velec            = _mm_andnot_ps(dummy_mask,velec);
 613             velecsum         = _mm_add_ps(velecsum,velec);
 614
 615             fscal            = felec;
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_ps(fscal,dx10);
 621             ty               = _mm_mul_ps(fscal,dy10);
 622             tz               = _mm_mul_ps(fscal,dz10);
 623
 624             /* Update vectorial force */
 625             fix1             = _mm_add_ps(fix1,tx);
 626             fiy1             = _mm_add_ps(fiy1,ty);
 627             fiz1             = _mm_add_ps(fiz1,tz);
 628
 629             fjx0             = _mm_add_ps(fjx0,tx);
 630             fjy0             = _mm_add_ps(fjy0,ty);
 631             fjz0             = _mm_add_ps(fjz0,tz);
 632
 633             /**************************
 634              * CALCULATE INTERACTIONS *
 635              **************************/
 636
 637             r20              = _mm_mul_ps(rsq20,rinv20);
 638             r20              = _mm_andnot_ps(dummy_mask,r20);
 639
 640             /* Compute parameters for interactions between i and j atoms */
 641             qq20             = _mm_mul_ps(iq2,jq0);
 642
 643             /* EWALD ELECTROSTATICS */
 644
 645             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 646             ewrt             = _mm_mul_ps(r20,ewtabscale);
 647             ewitab           = _mm_cvttps_epi32(ewrt);
 648             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 649             ewitab           = _mm_slli_epi32(ewitab,2);
 650             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 651             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 652             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 653             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 654             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 655             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 656             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 657             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 658             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm_andnot_ps(dummy_mask,velec);
 662             velecsum         = _mm_add_ps(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm_mul_ps(fscal,dx20);
 670             ty               = _mm_mul_ps(fscal,dy20);
 671             tz               = _mm_mul_ps(fscal,dz20);
 672
 673             /* Update vectorial force */
 674             fix2             = _mm_add_ps(fix2,tx);
 675             fiy2             = _mm_add_ps(fiy2,ty);
 676             fiz2             = _mm_add_ps(fiz2,tz);
 677
 678             fjx0             = _mm_add_ps(fjx0,tx);
 679             fjy0             = _mm_add_ps(fjy0,ty);
 680             fjz0             = _mm_add_ps(fjz0,tz);
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             r30              = _mm_mul_ps(rsq30,rinv30);
 687             r30              = _mm_andnot_ps(dummy_mask,r30);
 688
 689             /* Compute parameters for interactions between i and j atoms */
 690             qq30             = _mm_mul_ps(iq3,jq0);
 691
 692             /* EWALD ELECTROSTATICS */
 693
 694             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 695             ewrt             = _mm_mul_ps(r30,ewtabscale);
 696             ewitab           = _mm_cvttps_epi32(ewrt);
 697             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 698             ewitab           = _mm_slli_epi32(ewitab,2);
 699             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 700             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 701             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 702             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 703             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 704             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 705             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 706             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 707             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 708
 709             /* Update potential sum for this i atom from the interaction with this j atom. */
 710             velec            = _mm_andnot_ps(dummy_mask,velec);
 711             velecsum         = _mm_add_ps(velecsum,velec);
 712
 713             fscal            = felec;
 714
 715             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm_mul_ps(fscal,dx30);
 719             ty               = _mm_mul_ps(fscal,dy30);
 720             tz               = _mm_mul_ps(fscal,dz30);
 721
 722             /* Update vectorial force */
 723             fix3             = _mm_add_ps(fix3,tx);
 724             fiy3             = _mm_add_ps(fiy3,ty);
 725             fiz3             = _mm_add_ps(fiz3,tz);
 726
 727             fjx0             = _mm_add_ps(fjx0,tx);
 728             fjy0             = _mm_add_ps(fjy0,ty);
 729             fjz0             = _mm_add_ps(fjz0,tz);
 730
 731             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 732             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 733             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 734             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 735
 736             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 737
 738             /* Inner loop uses 178 flops */
 739         }
 740
 741         /* End of innermost loop */
 742
 743         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 744                                               f+i_coord_offset,fshift+i_shift_offset);
 745
 746         ggid                        = gid[iidx];
 747         /* Update potential energies */
 748         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 749         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 750
 751         /* Increment number of inner iterations */
 752         inneriter                  += j_index_end - j_index_start;
 753
 754         /* Outer loop uses 26 flops */
 755     }
 756
 757     /* Increment number of outer iterations */
 758     outeriter        += nri;
 759
 760     /* Update outer/inner flops */
 761
 762     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*178);
 763 }
 764 /*
 765  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse2_single
 766  * Electrostatics interaction: Ewald
 767  * VdW interaction:            LJEwald
 768  * Geometry:                   Water4-Particle
 769  * Calculate force/pot:        Force
 770  */
 771 void
 772 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_sse2_single
 773                     (t_nblist                    * gmx_restrict       nlist,
 774                      rvec                        * gmx_restrict          xx,
 775                      rvec                        * gmx_restrict          ff,
 776                      t_forcerec                  * gmx_restrict          fr,
 777                      t_mdatoms                   * gmx_restrict     mdatoms,
 778                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 779                      t_nrnb                      * gmx_restrict        nrnb)
 780 {
 781     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 782      * just 0 for non-waters.
 783      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 784      * jnr indices corresponding to data put in the four positions in the SIMD register.
 785      */
 786     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 787     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 788     int              jnrA,jnrB,jnrC,jnrD;
 789     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 790     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 791     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 792     real             rcutoff_scalar;
 793     real             *shiftvec,*fshift,*x,*f;
 794     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 795     real             scratch[4*DIM];
 796     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 797     int              vdwioffset0;
 798     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 799     int              vdwioffset1;
 800     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 801     int              vdwioffset2;
 802     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 803     int              vdwioffset3;
 804     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 805     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 806     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 807     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 808     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 809     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 810     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 811     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 812     real             *charge;
 813     int              nvdwtype;
 814     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 815     int              *vdwtype;
 816     real             *vdwparam;
 817     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 818     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 819     __m128           c6grid_00;
 820     __m128           c6grid_10;
 821     __m128           c6grid_20;
 822     __m128           c6grid_30;
 823     __m128           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 824     real             *vdwgridparam;
 825     __m128           one_half = _mm_set1_ps(0.5);
 826     __m128           minus_one = _mm_set1_ps(-1.0);
 827     __m128i          ewitab;
 828     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 829     real             *ewtab;
 830     __m128           dummy_mask,cutoff_mask;
 831     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 832     __m128           one     = _mm_set1_ps(1.0);
 833     __m128           two     = _mm_set1_ps(2.0);
 834     x                = xx[0];
 835     f                = ff[0];
 836
 837     nri              = nlist->nri;
 838     iinr             = nlist->iinr;
 839     jindex           = nlist->jindex;
 840     jjnr             = nlist->jjnr;
 841     shiftidx         = nlist->shift;
 842     gid              = nlist->gid;
 843     shiftvec         = fr->shift_vec[0];
 844     fshift           = fr->fshift[0];
 845     facel            = _mm_set1_ps(fr->epsfac);
 846     charge           = mdatoms->chargeA;
 847     nvdwtype         = fr->ntype;
 848     vdwparam         = fr->nbfp;
 849     vdwtype          = mdatoms->typeA;
 850     vdwgridparam     = fr->ljpme_c6grid;
 851     sh_lj_ewald      = _mm_set1_ps(fr->ic->sh_lj_ewald);
 852     ewclj            = _mm_set1_ps(fr->ewaldcoeff_lj);
 853     ewclj2           = _mm_mul_ps(minus_one,_mm_mul_ps(ewclj,ewclj));
 854
 855     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 856     ewtab            = fr->ic->tabq_coul_F;
 857     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 858     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 859
 860     /* Setup water-specific parameters */
 861     inr              = nlist->iinr[0];
 862     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 863     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 864     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 865     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 866
 867     /* Avoid stupid compiler warnings */
 868     jnrA = jnrB = jnrC = jnrD = 0;
 869     j_coord_offsetA = 0;
 870     j_coord_offsetB = 0;
 871     j_coord_offsetC = 0;
 872     j_coord_offsetD = 0;
 873
 874     outeriter        = 0;
 875     inneriter        = 0;
 876
 877     for(iidx=0;iidx<4*DIM;iidx++)
 878     {
 879         scratch[iidx] = 0.0;
 880     }
 881
 882     /* Start outer loop over neighborlists */
 883     for(iidx=0; iidx<nri; iidx++)
 884     {
 885         /* Load shift vector for this list */
 886         i_shift_offset   = DIM*shiftidx[iidx];
 887
 888         /* Load limits for loop over neighbors */
 889         j_index_start    = jindex[iidx];
 890         j_index_end      = jindex[iidx+1];
 891
 892         /* Get outer coordinate index */
 893         inr              = iinr[iidx];
 894         i_coord_offset   = DIM*inr;
 895
 896         /* Load i particle coords and add shift vector */
 897         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 898                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 899
 900         fix0             = _mm_setzero_ps();
 901         fiy0             = _mm_setzero_ps();
 902         fiz0             = _mm_setzero_ps();
 903         fix1             = _mm_setzero_ps();
 904         fiy1             = _mm_setzero_ps();
 905         fiz1             = _mm_setzero_ps();
 906         fix2             = _mm_setzero_ps();
 907         fiy2             = _mm_setzero_ps();
 908         fiz2             = _mm_setzero_ps();
 909         fix3             = _mm_setzero_ps();
 910         fiy3             = _mm_setzero_ps();
 911         fiz3             = _mm_setzero_ps();
 912
 913         /* Start inner kernel loop */
 914         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 915         {
 916
 917             /* Get j neighbor index, and coordinate index */
 918             jnrA             = jjnr[jidx];
 919             jnrB             = jjnr[jidx+1];
 920             jnrC             = jjnr[jidx+2];
 921             jnrD             = jjnr[jidx+3];
 922             j_coord_offsetA  = DIM*jnrA;
 923             j_coord_offsetB  = DIM*jnrB;
 924             j_coord_offsetC  = DIM*jnrC;
 925             j_coord_offsetD  = DIM*jnrD;
 926
 927             /* load j atom coordinates */
 928             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 929                                               x+j_coord_offsetC,x+j_coord_offsetD,
 930                                               &jx0,&jy0,&jz0);
 931
 932             /* Calculate displacement vector */
 933             dx00             = _mm_sub_ps(ix0,jx0);
 934             dy00             = _mm_sub_ps(iy0,jy0);
 935             dz00             = _mm_sub_ps(iz0,jz0);
 936             dx10             = _mm_sub_ps(ix1,jx0);
 937             dy10             = _mm_sub_ps(iy1,jy0);
 938             dz10             = _mm_sub_ps(iz1,jz0);
 939             dx20             = _mm_sub_ps(ix2,jx0);
 940             dy20             = _mm_sub_ps(iy2,jy0);
 941             dz20             = _mm_sub_ps(iz2,jz0);
 942             dx30             = _mm_sub_ps(ix3,jx0);
 943             dy30             = _mm_sub_ps(iy3,jy0);
 944             dz30             = _mm_sub_ps(iz3,jz0);
 945
 946             /* Calculate squared distance and things based on it */
 947             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 948             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 949             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 950             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 951
 952             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 953             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 954             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 955             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 956
 957             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 958             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 959             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 960             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 961
 962             /* Load parameters for j particles */
 963             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 964                                                               charge+jnrC+0,charge+jnrD+0);
 965             vdwjidx0A        = 2*vdwtype[jnrA+0];
 966             vdwjidx0B        = 2*vdwtype[jnrB+0];
 967             vdwjidx0C        = 2*vdwtype[jnrC+0];
 968             vdwjidx0D        = 2*vdwtype[jnrD+0];
 969
 970             fjx0             = _mm_setzero_ps();
 971             fjy0             = _mm_setzero_ps();
 972             fjz0             = _mm_setzero_ps();
 973
 974             /**************************
 975              * CALCULATE INTERACTIONS *
 976              **************************/
 977
 978             r00              = _mm_mul_ps(rsq00,rinv00);
 979
 980             /* Compute parameters for interactions between i and j atoms */
 981             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 982                                          vdwparam+vdwioffset0+vdwjidx0B,
 983                                          vdwparam+vdwioffset0+vdwjidx0C,
 984                                          vdwparam+vdwioffset0+vdwjidx0D,
 985                                          &c6_00,&c12_00);
 986             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
 987                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
 988                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
 989                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
 990
 991             /* Analytical LJ-PME */
 992             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 993             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
 994             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
 995             exponent         = gmx_simd_exp_r(ewcljrsq);
 996             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 997             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
 998             /* f6A = 6 * C6grid * (1 - poly) */
 999             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1000             /* f6B = C6grid * exponent * beta^6 */
1001             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1002             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1003             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);
1004
1005             fscal            = fvdw;
1006
1007             /* Calculate temporary vectorial force */
1008             tx               = _mm_mul_ps(fscal,dx00);
1009             ty               = _mm_mul_ps(fscal,dy00);
1010             tz               = _mm_mul_ps(fscal,dz00);
1011
1012             /* Update vectorial force */
1013             fix0             = _mm_add_ps(fix0,tx);
1014             fiy0             = _mm_add_ps(fiy0,ty);
1015             fiz0             = _mm_add_ps(fiz0,tz);
1016
1017             fjx0             = _mm_add_ps(fjx0,tx);
1018             fjy0             = _mm_add_ps(fjy0,ty);
1019             fjz0             = _mm_add_ps(fjz0,tz);
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             r10              = _mm_mul_ps(rsq10,rinv10);
1026
1027             /* Compute parameters for interactions between i and j atoms */
1028             qq10             = _mm_mul_ps(iq1,jq0);
1029
1030             /* EWALD ELECTROSTATICS */
1031
1032             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1033             ewrt             = _mm_mul_ps(r10,ewtabscale);
1034             ewitab           = _mm_cvttps_epi32(ewrt);
1035             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1036             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1037                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1038                                          &ewtabF,&ewtabFn);
1039             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1040             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1041
1042             fscal            = felec;
1043
1044             /* Calculate temporary vectorial force */
1045             tx               = _mm_mul_ps(fscal,dx10);
1046             ty               = _mm_mul_ps(fscal,dy10);
1047             tz               = _mm_mul_ps(fscal,dz10);
1048
1049             /* Update vectorial force */
1050             fix1             = _mm_add_ps(fix1,tx);
1051             fiy1             = _mm_add_ps(fiy1,ty);
1052             fiz1             = _mm_add_ps(fiz1,tz);
1053
1054             fjx0             = _mm_add_ps(fjx0,tx);
1055             fjy0             = _mm_add_ps(fjy0,ty);
1056             fjz0             = _mm_add_ps(fjz0,tz);
1057
1058             /**************************
1059              * CALCULATE INTERACTIONS *
1060              **************************/
1061
1062             r20              = _mm_mul_ps(rsq20,rinv20);
1063
1064             /* Compute parameters for interactions between i and j atoms */
1065             qq20             = _mm_mul_ps(iq2,jq0);
1066
1067             /* EWALD ELECTROSTATICS */
1068
1069             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1070             ewrt             = _mm_mul_ps(r20,ewtabscale);
1071             ewitab           = _mm_cvttps_epi32(ewrt);
1072             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1073             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1074                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1075                                          &ewtabF,&ewtabFn);
1076             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1077             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1078
1079             fscal            = felec;
1080
1081             /* Calculate temporary vectorial force */
1082             tx               = _mm_mul_ps(fscal,dx20);
1083             ty               = _mm_mul_ps(fscal,dy20);
1084             tz               = _mm_mul_ps(fscal,dz20);
1085
1086             /* Update vectorial force */
1087             fix2             = _mm_add_ps(fix2,tx);
1088             fiy2             = _mm_add_ps(fiy2,ty);
1089             fiz2             = _mm_add_ps(fiz2,tz);
1090
1091             fjx0             = _mm_add_ps(fjx0,tx);
1092             fjy0             = _mm_add_ps(fjy0,ty);
1093             fjz0             = _mm_add_ps(fjz0,tz);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r30              = _mm_mul_ps(rsq30,rinv30);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq30             = _mm_mul_ps(iq3,jq0);
1103
1104             /* EWALD ELECTROSTATICS */
1105
1106             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1107             ewrt             = _mm_mul_ps(r30,ewtabscale);
1108             ewitab           = _mm_cvttps_epi32(ewrt);
1109             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1110             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1111                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1112                                          &ewtabF,&ewtabFn);
1113             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1114             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1115
1116             fscal            = felec;
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm_mul_ps(fscal,dx30);
1120             ty               = _mm_mul_ps(fscal,dy30);
1121             tz               = _mm_mul_ps(fscal,dz30);
1122
1123             /* Update vectorial force */
1124             fix3             = _mm_add_ps(fix3,tx);
1125             fiy3             = _mm_add_ps(fiy3,ty);
1126             fiz3             = _mm_add_ps(fiz3,tz);
1127
1128             fjx0             = _mm_add_ps(fjx0,tx);
1129             fjy0             = _mm_add_ps(fjy0,ty);
1130             fjz0             = _mm_add_ps(fjz0,tz);
1131
1132             fjptrA             = f+j_coord_offsetA;
1133             fjptrB             = f+j_coord_offsetB;
1134             fjptrC             = f+j_coord_offsetC;
1135             fjptrD             = f+j_coord_offsetD;
1136
1137             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1138
1139             /* Inner loop uses 154 flops */
1140         }
1141
1142         if(jidx<j_index_end)
1143         {
1144
1145             /* Get j neighbor index, and coordinate index */
1146             jnrlistA         = jjnr[jidx];
1147             jnrlistB         = jjnr[jidx+1];
1148             jnrlistC         = jjnr[jidx+2];
1149             jnrlistD         = jjnr[jidx+3];
1150             /* Sign of each element will be negative for non-real atoms.
1151              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1152              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1153              */
1154             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1155             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1156             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1157             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1158             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1159             j_coord_offsetA  = DIM*jnrA;
1160             j_coord_offsetB  = DIM*jnrB;
1161             j_coord_offsetC  = DIM*jnrC;
1162             j_coord_offsetD  = DIM*jnrD;
1163
1164             /* load j atom coordinates */
1165             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1166                                               x+j_coord_offsetC,x+j_coord_offsetD,
1167                                               &jx0,&jy0,&jz0);
1168
1169             /* Calculate displacement vector */
1170             dx00             = _mm_sub_ps(ix0,jx0);
1171             dy00             = _mm_sub_ps(iy0,jy0);
1172             dz00             = _mm_sub_ps(iz0,jz0);
1173             dx10             = _mm_sub_ps(ix1,jx0);
1174             dy10             = _mm_sub_ps(iy1,jy0);
1175             dz10             = _mm_sub_ps(iz1,jz0);
1176             dx20             = _mm_sub_ps(ix2,jx0);
1177             dy20             = _mm_sub_ps(iy2,jy0);
1178             dz20             = _mm_sub_ps(iz2,jz0);
1179             dx30             = _mm_sub_ps(ix3,jx0);
1180             dy30             = _mm_sub_ps(iy3,jy0);
1181             dz30             = _mm_sub_ps(iz3,jz0);
1182
1183             /* Calculate squared distance and things based on it */
1184             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1185             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1186             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1187             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1188
1189             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1190             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1191             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1192             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1193
1194             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1195             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1196             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1197             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1198
1199             /* Load parameters for j particles */
1200             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1201                                                               charge+jnrC+0,charge+jnrD+0);
1202             vdwjidx0A        = 2*vdwtype[jnrA+0];
1203             vdwjidx0B        = 2*vdwtype[jnrB+0];
1204             vdwjidx0C        = 2*vdwtype[jnrC+0];
1205             vdwjidx0D        = 2*vdwtype[jnrD+0];
1206
1207             fjx0             = _mm_setzero_ps();
1208             fjy0             = _mm_setzero_ps();
1209             fjz0             = _mm_setzero_ps();
1210
1211             /**************************
1212              * CALCULATE INTERACTIONS *
1213              **************************/
1214
1215             r00              = _mm_mul_ps(rsq00,rinv00);
1216             r00              = _mm_andnot_ps(dummy_mask,r00);
1217
1218             /* Compute parameters for interactions between i and j atoms */
1219             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1220                                          vdwparam+vdwioffset0+vdwjidx0B,
1221                                          vdwparam+vdwioffset0+vdwjidx0C,
1222                                          vdwparam+vdwioffset0+vdwjidx0D,
1223                                          &c6_00,&c12_00);
1224             c6grid_00       = gmx_mm_load_4real_swizzle_ps(vdwgridparam+vdwioffset0+vdwjidx0A,
1225                                                                vdwgridparam+vdwioffset0+vdwjidx0B,
1226                                                                vdwgridparam+vdwioffset0+vdwjidx0C,
1227                                                                vdwgridparam+vdwioffset0+vdwjidx0D);
1228
1229             /* Analytical LJ-PME */
1230             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1231             ewcljrsq         = _mm_mul_ps(ewclj2,rsq00);
1232             ewclj6           = _mm_mul_ps(ewclj2,_mm_mul_ps(ewclj2,ewclj2));
1233             exponent         = gmx_simd_exp_r(ewcljrsq);
1234             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1235             poly             = _mm_mul_ps(exponent,_mm_add_ps(_mm_sub_ps(one,ewcljrsq),_mm_mul_ps(_mm_mul_ps(ewcljrsq,ewcljrsq),one_half)));
1236             /* f6A = 6 * C6grid * (1 - poly) */
1237             f6A              = _mm_mul_ps(c6grid_00,_mm_sub_ps(one,poly));
1238             /* f6B = C6grid * exponent * beta^6 */
1239             f6B              = _mm_mul_ps(_mm_mul_ps(c6grid_00,one_sixth),_mm_mul_ps(exponent,ewclj6));
1240             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1241             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);
1242
1243             fscal            = fvdw;
1244
1245             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1246
1247             /* Calculate temporary vectorial force */
1248             tx               = _mm_mul_ps(fscal,dx00);
1249             ty               = _mm_mul_ps(fscal,dy00);
1250             tz               = _mm_mul_ps(fscal,dz00);
1251
1252             /* Update vectorial force */
1253             fix0             = _mm_add_ps(fix0,tx);
1254             fiy0             = _mm_add_ps(fiy0,ty);
1255             fiz0             = _mm_add_ps(fiz0,tz);
1256
1257             fjx0             = _mm_add_ps(fjx0,tx);
1258             fjy0             = _mm_add_ps(fjy0,ty);
1259             fjz0             = _mm_add_ps(fjz0,tz);
1260
1261             /**************************
1262              * CALCULATE INTERACTIONS *
1263              **************************/
1264
1265             r10              = _mm_mul_ps(rsq10,rinv10);
1266             r10              = _mm_andnot_ps(dummy_mask,r10);
1267
1268             /* Compute parameters for interactions between i and j atoms */
1269             qq10             = _mm_mul_ps(iq1,jq0);
1270
1271             /* EWALD ELECTROSTATICS */
1272
1273             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1274             ewrt             = _mm_mul_ps(r10,ewtabscale);
1275             ewitab           = _mm_cvttps_epi32(ewrt);
1276             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1277             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1278                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1279                                          &ewtabF,&ewtabFn);
1280             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1281             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1282
1283             fscal            = felec;
1284
1285             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1286
1287             /* Calculate temporary vectorial force */
1288             tx               = _mm_mul_ps(fscal,dx10);
1289             ty               = _mm_mul_ps(fscal,dy10);
1290             tz               = _mm_mul_ps(fscal,dz10);
1291
1292             /* Update vectorial force */
1293             fix1             = _mm_add_ps(fix1,tx);
1294             fiy1             = _mm_add_ps(fiy1,ty);
1295             fiz1             = _mm_add_ps(fiz1,tz);
1296
1297             fjx0             = _mm_add_ps(fjx0,tx);
1298             fjy0             = _mm_add_ps(fjy0,ty);
1299             fjz0             = _mm_add_ps(fjz0,tz);
1300
1301             /**************************
1302              * CALCULATE INTERACTIONS *
1303              **************************/
1304
1305             r20              = _mm_mul_ps(rsq20,rinv20);
1306             r20              = _mm_andnot_ps(dummy_mask,r20);
1307
1308             /* Compute parameters for interactions between i and j atoms */
1309             qq20             = _mm_mul_ps(iq2,jq0);
1310
1311             /* EWALD ELECTROSTATICS */
1312
1313             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1314             ewrt             = _mm_mul_ps(r20,ewtabscale);
1315             ewitab           = _mm_cvttps_epi32(ewrt);
1316             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1317             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1318                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1319                                          &ewtabF,&ewtabFn);
1320             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1321             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1322
1323             fscal            = felec;
1324
1325             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1326
1327             /* Calculate temporary vectorial force */
1328             tx               = _mm_mul_ps(fscal,dx20);
1329             ty               = _mm_mul_ps(fscal,dy20);
1330             tz               = _mm_mul_ps(fscal,dz20);
1331
1332             /* Update vectorial force */
1333             fix2             = _mm_add_ps(fix2,tx);
1334             fiy2             = _mm_add_ps(fiy2,ty);
1335             fiz2             = _mm_add_ps(fiz2,tz);
1336
1337             fjx0             = _mm_add_ps(fjx0,tx);
1338             fjy0             = _mm_add_ps(fjy0,ty);
1339             fjz0             = _mm_add_ps(fjz0,tz);
1340
1341             /**************************
1342              * CALCULATE INTERACTIONS *
1343              **************************/
1344
1345             r30              = _mm_mul_ps(rsq30,rinv30);
1346             r30              = _mm_andnot_ps(dummy_mask,r30);
1347
1348             /* Compute parameters for interactions between i and j atoms */
1349             qq30             = _mm_mul_ps(iq3,jq0);
1350
1351             /* EWALD ELECTROSTATICS */
1352
1353             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1354             ewrt             = _mm_mul_ps(r30,ewtabscale);
1355             ewitab           = _mm_cvttps_epi32(ewrt);
1356             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1357             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1358                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1359                                          &ewtabF,&ewtabFn);
1360             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1361             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1362
1363             fscal            = felec;
1364
1365             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1366
1367             /* Calculate temporary vectorial force */
1368             tx               = _mm_mul_ps(fscal,dx30);
1369             ty               = _mm_mul_ps(fscal,dy30);
1370             tz               = _mm_mul_ps(fscal,dz30);
1371
1372             /* Update vectorial force */
1373             fix3             = _mm_add_ps(fix3,tx);
1374             fiy3             = _mm_add_ps(fiy3,ty);
1375             fiz3             = _mm_add_ps(fiz3,tz);
1376
1377             fjx0             = _mm_add_ps(fjx0,tx);
1378             fjy0             = _mm_add_ps(fjy0,ty);
1379             fjz0             = _mm_add_ps(fjz0,tz);
1380
1381             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1382             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1383             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1384             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1385
1386             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1387
1388             /* Inner loop uses 158 flops */
1389         }
1390
1391         /* End of innermost loop */
1392
1393         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1394                                               f+i_coord_offset,fshift+i_shift_offset);
1395
1396         /* Increment number of inner iterations */
1397         inneriter                  += j_index_end - j_index_start;
1398
1399         /* Outer loop uses 24 flops */
1400     }
1401
1402     /* Increment number of outer iterations */
1403     outeriter        += nri;
1404
1405     /* Update outer/inner flops */
1406
1407     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*158);
1408 }