src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_sse4_1_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128i          ewitab;
  96     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 121
 122     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 130     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 131     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix0             = _mm_setzero_ps();
 168         fiy0             = _mm_setzero_ps();
 169         fiz0             = _mm_setzero_ps();
 170         fix1             = _mm_setzero_ps();
 171         fiy1             = _mm_setzero_ps();
 172         fiz1             = _mm_setzero_ps();
 173         fix2             = _mm_setzero_ps();
 174         fiy2             = _mm_setzero_ps();
 175         fiz2             = _mm_setzero_ps();
 176         fix3             = _mm_setzero_ps();
 177         fiy3             = _mm_setzero_ps();
 178         fiz3             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213             dx30             = _mm_sub_ps(ix3,jx0);
 214             dy30             = _mm_sub_ps(iy3,jy0);
 215             dz30             = _mm_sub_ps(iz3,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 222
 223             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 227
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 234                                                               charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             fjx0             = _mm_setzero_ps();
 241             fjy0             = _mm_setzero_ps();
 242             fjz0             = _mm_setzero_ps();
 243
 244             /**************************
 245              * CALCULATE INTERACTIONS *
 246              **************************/
 247
 248             r00              = _mm_mul_ps(rsq00,rinv00);
 249
 250             /* Compute parameters for interactions between i and j atoms */
 251             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,
 253                                          vdwparam+vdwioffset0+vdwjidx0C,
 254                                          vdwparam+vdwioffset0+vdwjidx0D,
 255                                          &c6_00,&c12_00);
 256
 257             /* Calculate table index by multiplying r with table scale and truncate to integer */
 258             rt               = _mm_mul_ps(r00,vftabscale);
 259             vfitab           = _mm_cvttps_epi32(rt);
 260             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 261             vfitab           = _mm_slli_epi32(vfitab,3);
 262
 263             /* CUBIC SPLINE TABLE DISPERSION */
 264             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 265             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 266             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 267             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 268             _MM_TRANSPOSE4_PS(Y,F,G,H);
 269             Heps             = _mm_mul_ps(vfeps,H);
 270             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 271             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 272             vvdw6            = _mm_mul_ps(c6_00,VV);
 273             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 274             fvdw6            = _mm_mul_ps(c6_00,FF);
 275
 276             /* CUBIC SPLINE TABLE REPULSION */
 277             vfitab           = _mm_add_epi32(vfitab,ifour);
 278             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 279             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 280             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 281             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 282             _MM_TRANSPOSE4_PS(Y,F,G,H);
 283             Heps             = _mm_mul_ps(vfeps,H);
 284             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 285             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 286             vvdw12           = _mm_mul_ps(c12_00,VV);
 287             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 288             fvdw12           = _mm_mul_ps(c12_00,FF);
 289             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 290             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 291
 292             /* Update potential sum for this i atom from the interaction with this j atom. */
 293             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 294
 295             fscal            = fvdw;
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_ps(fscal,dx00);
 299             ty               = _mm_mul_ps(fscal,dy00);
 300             tz               = _mm_mul_ps(fscal,dz00);
 301
 302             /* Update vectorial force */
 303             fix0             = _mm_add_ps(fix0,tx);
 304             fiy0             = _mm_add_ps(fiy0,ty);
 305             fiz0             = _mm_add_ps(fiz0,tz);
 306
 307             fjx0             = _mm_add_ps(fjx0,tx);
 308             fjy0             = _mm_add_ps(fjy0,ty);
 309             fjz0             = _mm_add_ps(fjz0,tz);
 310
 311             /**************************
 312              * CALCULATE INTERACTIONS *
 313              **************************/
 314
 315             r10              = _mm_mul_ps(rsq10,rinv10);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq10             = _mm_mul_ps(iq1,jq0);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 323             ewrt             = _mm_mul_ps(r10,ewtabscale);
 324             ewitab           = _mm_cvttps_epi32(ewrt);
 325             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 326             ewitab           = _mm_slli_epi32(ewitab,2);
 327             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 328             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 329             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 330             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 331             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 332             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 333             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 334             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 335             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             /* Calculate temporary vectorial force */
 343             tx               = _mm_mul_ps(fscal,dx10);
 344             ty               = _mm_mul_ps(fscal,dy10);
 345             tz               = _mm_mul_ps(fscal,dz10);
 346
 347             /* Update vectorial force */
 348             fix1             = _mm_add_ps(fix1,tx);
 349             fiy1             = _mm_add_ps(fiy1,ty);
 350             fiz1             = _mm_add_ps(fiz1,tz);
 351
 352             fjx0             = _mm_add_ps(fjx0,tx);
 353             fjy0             = _mm_add_ps(fjy0,ty);
 354             fjz0             = _mm_add_ps(fjz0,tz);
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             r20              = _mm_mul_ps(rsq20,rinv20);
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq20             = _mm_mul_ps(iq2,jq0);
 364
 365             /* EWALD ELECTROSTATICS */
 366
 367             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 368             ewrt             = _mm_mul_ps(r20,ewtabscale);
 369             ewitab           = _mm_cvttps_epi32(ewrt);
 370             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 371             ewitab           = _mm_slli_epi32(ewitab,2);
 372             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 373             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 374             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 375             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 376             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 377             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 378             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 379             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 380             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velecsum         = _mm_add_ps(velecsum,velec);
 384
 385             fscal            = felec;
 386
 387             /* Calculate temporary vectorial force */
 388             tx               = _mm_mul_ps(fscal,dx20);
 389             ty               = _mm_mul_ps(fscal,dy20);
 390             tz               = _mm_mul_ps(fscal,dz20);
 391
 392             /* Update vectorial force */
 393             fix2             = _mm_add_ps(fix2,tx);
 394             fiy2             = _mm_add_ps(fiy2,ty);
 395             fiz2             = _mm_add_ps(fiz2,tz);
 396
 397             fjx0             = _mm_add_ps(fjx0,tx);
 398             fjy0             = _mm_add_ps(fjy0,ty);
 399             fjz0             = _mm_add_ps(fjz0,tz);
 400
 401             /**************************
 402              * CALCULATE INTERACTIONS *
 403              **************************/
 404
 405             r30              = _mm_mul_ps(rsq30,rinv30);
 406
 407             /* Compute parameters for interactions between i and j atoms */
 408             qq30             = _mm_mul_ps(iq3,jq0);
 409
 410             /* EWALD ELECTROSTATICS */
 411
 412             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 413             ewrt             = _mm_mul_ps(r30,ewtabscale);
 414             ewitab           = _mm_cvttps_epi32(ewrt);
 415             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 416             ewitab           = _mm_slli_epi32(ewitab,2);
 417             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 418             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 419             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 420             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 421             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 422             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 423             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 424             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 425             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 426
 427             /* Update potential sum for this i atom from the interaction with this j atom. */
 428             velecsum         = _mm_add_ps(velecsum,velec);
 429
 430             fscal            = felec;
 431
 432             /* Calculate temporary vectorial force */
 433             tx               = _mm_mul_ps(fscal,dx30);
 434             ty               = _mm_mul_ps(fscal,dy30);
 435             tz               = _mm_mul_ps(fscal,dz30);
 436
 437             /* Update vectorial force */
 438             fix3             = _mm_add_ps(fix3,tx);
 439             fiy3             = _mm_add_ps(fiy3,ty);
 440             fiz3             = _mm_add_ps(fiz3,tz);
 441
 442             fjx0             = _mm_add_ps(fjx0,tx);
 443             fjy0             = _mm_add_ps(fjy0,ty);
 444             fjz0             = _mm_add_ps(fjz0,tz);
 445
 446             fjptrA             = f+j_coord_offsetA;
 447             fjptrB             = f+j_coord_offsetB;
 448             fjptrC             = f+j_coord_offsetC;
 449             fjptrD             = f+j_coord_offsetD;
 450
 451             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 452
 453             /* Inner loop uses 179 flops */
 454         }
 455
 456         if(jidx<j_index_end)
 457         {
 458
 459             /* Get j neighbor index, and coordinate index */
 460             jnrlistA         = jjnr[jidx];
 461             jnrlistB         = jjnr[jidx+1];
 462             jnrlistC         = jjnr[jidx+2];
 463             jnrlistD         = jjnr[jidx+3];
 464             /* Sign of each element will be negative for non-real atoms.
 465              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 466              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 467              */
 468             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 469             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 470             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 471             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 472             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 473             j_coord_offsetA  = DIM*jnrA;
 474             j_coord_offsetB  = DIM*jnrB;
 475             j_coord_offsetC  = DIM*jnrC;
 476             j_coord_offsetD  = DIM*jnrD;
 477
 478             /* load j atom coordinates */
 479             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 480                                               x+j_coord_offsetC,x+j_coord_offsetD,
 481                                               &jx0,&jy0,&jz0);
 482
 483             /* Calculate displacement vector */
 484             dx00             = _mm_sub_ps(ix0,jx0);
 485             dy00             = _mm_sub_ps(iy0,jy0);
 486             dz00             = _mm_sub_ps(iz0,jz0);
 487             dx10             = _mm_sub_ps(ix1,jx0);
 488             dy10             = _mm_sub_ps(iy1,jy0);
 489             dz10             = _mm_sub_ps(iz1,jz0);
 490             dx20             = _mm_sub_ps(ix2,jx0);
 491             dy20             = _mm_sub_ps(iy2,jy0);
 492             dz20             = _mm_sub_ps(iz2,jz0);
 493             dx30             = _mm_sub_ps(ix3,jx0);
 494             dy30             = _mm_sub_ps(iy3,jy0);
 495             dz30             = _mm_sub_ps(iz3,jz0);
 496
 497             /* Calculate squared distance and things based on it */
 498             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 499             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 500             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 501             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 502
 503             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 504             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 505             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 506             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 507
 508             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 509             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 510             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 511
 512             /* Load parameters for j particles */
 513             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 514                                                               charge+jnrC+0,charge+jnrD+0);
 515             vdwjidx0A        = 2*vdwtype[jnrA+0];
 516             vdwjidx0B        = 2*vdwtype[jnrB+0];
 517             vdwjidx0C        = 2*vdwtype[jnrC+0];
 518             vdwjidx0D        = 2*vdwtype[jnrD+0];
 519
 520             fjx0             = _mm_setzero_ps();
 521             fjy0             = _mm_setzero_ps();
 522             fjz0             = _mm_setzero_ps();
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             r00              = _mm_mul_ps(rsq00,rinv00);
 529             r00              = _mm_andnot_ps(dummy_mask,r00);
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 533                                          vdwparam+vdwioffset0+vdwjidx0B,
 534                                          vdwparam+vdwioffset0+vdwjidx0C,
 535                                          vdwparam+vdwioffset0+vdwjidx0D,
 536                                          &c6_00,&c12_00);
 537
 538             /* Calculate table index by multiplying r with table scale and truncate to integer */
 539             rt               = _mm_mul_ps(r00,vftabscale);
 540             vfitab           = _mm_cvttps_epi32(rt);
 541             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 542             vfitab           = _mm_slli_epi32(vfitab,3);
 543
 544             /* CUBIC SPLINE TABLE DISPERSION */
 545             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 546             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 547             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 548             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 549             _MM_TRANSPOSE4_PS(Y,F,G,H);
 550             Heps             = _mm_mul_ps(vfeps,H);
 551             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 552             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 553             vvdw6            = _mm_mul_ps(c6_00,VV);
 554             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 555             fvdw6            = _mm_mul_ps(c6_00,FF);
 556
 557             /* CUBIC SPLINE TABLE REPULSION */
 558             vfitab           = _mm_add_epi32(vfitab,ifour);
 559             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 560             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 561             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 562             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 563             _MM_TRANSPOSE4_PS(Y,F,G,H);
 564             Heps             = _mm_mul_ps(vfeps,H);
 565             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 566             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 567             vvdw12           = _mm_mul_ps(c12_00,VV);
 568             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 569             fvdw12           = _mm_mul_ps(c12_00,FF);
 570             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 571             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 572
 573             /* Update potential sum for this i atom from the interaction with this j atom. */
 574             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 575             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 576
 577             fscal            = fvdw;
 578
 579             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 580
 581             /* Calculate temporary vectorial force */
 582             tx               = _mm_mul_ps(fscal,dx00);
 583             ty               = _mm_mul_ps(fscal,dy00);
 584             tz               = _mm_mul_ps(fscal,dz00);
 585
 586             /* Update vectorial force */
 587             fix0             = _mm_add_ps(fix0,tx);
 588             fiy0             = _mm_add_ps(fiy0,ty);
 589             fiz0             = _mm_add_ps(fiz0,tz);
 590
 591             fjx0             = _mm_add_ps(fjx0,tx);
 592             fjy0             = _mm_add_ps(fjy0,ty);
 593             fjz0             = _mm_add_ps(fjz0,tz);
 594
 595             /**************************
 596              * CALCULATE INTERACTIONS *
 597              **************************/
 598
 599             r10              = _mm_mul_ps(rsq10,rinv10);
 600             r10              = _mm_andnot_ps(dummy_mask,r10);
 601
 602             /* Compute parameters for interactions between i and j atoms */
 603             qq10             = _mm_mul_ps(iq1,jq0);
 604
 605             /* EWALD ELECTROSTATICS */
 606
 607             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 608             ewrt             = _mm_mul_ps(r10,ewtabscale);
 609             ewitab           = _mm_cvttps_epi32(ewrt);
 610             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 611             ewitab           = _mm_slli_epi32(ewitab,2);
 612             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 613             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 614             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 615             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 616             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 617             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 618             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 619             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 620             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 621
 622             /* Update potential sum for this i atom from the interaction with this j atom. */
 623             velec            = _mm_andnot_ps(dummy_mask,velec);
 624             velecsum         = _mm_add_ps(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 629
 630             /* Calculate temporary vectorial force */
 631             tx               = _mm_mul_ps(fscal,dx10);
 632             ty               = _mm_mul_ps(fscal,dy10);
 633             tz               = _mm_mul_ps(fscal,dz10);
 634
 635             /* Update vectorial force */
 636             fix1             = _mm_add_ps(fix1,tx);
 637             fiy1             = _mm_add_ps(fiy1,ty);
 638             fiz1             = _mm_add_ps(fiz1,tz);
 639
 640             fjx0             = _mm_add_ps(fjx0,tx);
 641             fjy0             = _mm_add_ps(fjy0,ty);
 642             fjz0             = _mm_add_ps(fjz0,tz);
 643
 644             /**************************
 645              * CALCULATE INTERACTIONS *
 646              **************************/
 647
 648             r20              = _mm_mul_ps(rsq20,rinv20);
 649             r20              = _mm_andnot_ps(dummy_mask,r20);
 650
 651             /* Compute parameters for interactions between i and j atoms */
 652             qq20             = _mm_mul_ps(iq2,jq0);
 653
 654             /* EWALD ELECTROSTATICS */
 655
 656             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 657             ewrt             = _mm_mul_ps(r20,ewtabscale);
 658             ewitab           = _mm_cvttps_epi32(ewrt);
 659             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 660             ewitab           = _mm_slli_epi32(ewitab,2);
 661             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 662             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 663             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 664             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 665             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 666             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 667             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 668             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 669             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 670
 671             /* Update potential sum for this i atom from the interaction with this j atom. */
 672             velec            = _mm_andnot_ps(dummy_mask,velec);
 673             velecsum         = _mm_add_ps(velecsum,velec);
 674
 675             fscal            = felec;
 676
 677             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm_mul_ps(fscal,dx20);
 681             ty               = _mm_mul_ps(fscal,dy20);
 682             tz               = _mm_mul_ps(fscal,dz20);
 683
 684             /* Update vectorial force */
 685             fix2             = _mm_add_ps(fix2,tx);
 686             fiy2             = _mm_add_ps(fiy2,ty);
 687             fiz2             = _mm_add_ps(fiz2,tz);
 688
 689             fjx0             = _mm_add_ps(fjx0,tx);
 690             fjy0             = _mm_add_ps(fjy0,ty);
 691             fjz0             = _mm_add_ps(fjz0,tz);
 692
 693             /**************************
 694              * CALCULATE INTERACTIONS *
 695              **************************/
 696
 697             r30              = _mm_mul_ps(rsq30,rinv30);
 698             r30              = _mm_andnot_ps(dummy_mask,r30);
 699
 700             /* Compute parameters for interactions between i and j atoms */
 701             qq30             = _mm_mul_ps(iq3,jq0);
 702
 703             /* EWALD ELECTROSTATICS */
 704
 705             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 706             ewrt             = _mm_mul_ps(r30,ewtabscale);
 707             ewitab           = _mm_cvttps_epi32(ewrt);
 708             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 709             ewitab           = _mm_slli_epi32(ewitab,2);
 710             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 711             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 712             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 713             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 714             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 715             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 716             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 717             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 718             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 719
 720             /* Update potential sum for this i atom from the interaction with this j atom. */
 721             velec            = _mm_andnot_ps(dummy_mask,velec);
 722             velecsum         = _mm_add_ps(velecsum,velec);
 723
 724             fscal            = felec;
 725
 726             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 727
 728             /* Calculate temporary vectorial force */
 729             tx               = _mm_mul_ps(fscal,dx30);
 730             ty               = _mm_mul_ps(fscal,dy30);
 731             tz               = _mm_mul_ps(fscal,dz30);
 732
 733             /* Update vectorial force */
 734             fix3             = _mm_add_ps(fix3,tx);
 735             fiy3             = _mm_add_ps(fiy3,ty);
 736             fiz3             = _mm_add_ps(fiz3,tz);
 737
 738             fjx0             = _mm_add_ps(fjx0,tx);
 739             fjy0             = _mm_add_ps(fjy0,ty);
 740             fjz0             = _mm_add_ps(fjz0,tz);
 741
 742             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 743             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 744             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 745             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 746
 747             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 748
 749             /* Inner loop uses 183 flops */
 750         }
 751
 752         /* End of innermost loop */
 753
 754         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 755                                               f+i_coord_offset,fshift+i_shift_offset);
 756
 757         ggid                        = gid[iidx];
 758         /* Update potential energies */
 759         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 760         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 761
 762         /* Increment number of inner iterations */
 763         inneriter                  += j_index_end - j_index_start;
 764
 765         /* Outer loop uses 26 flops */
 766     }
 767
 768     /* Increment number of outer iterations */
 769     outeriter        += nri;
 770
 771     /* Update outer/inner flops */
 772
 773     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*183);
 774 }
 775 /*
 776  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse4_1_single
 777  * Electrostatics interaction: Ewald
 778  * VdW interaction:            CubicSplineTable
 779  * Geometry:                   Water4-Particle
 780  * Calculate force/pot:        Force
 781  */
 782 void
 783 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse4_1_single
 784                     (t_nblist * gmx_restrict                nlist,
 785                      rvec * gmx_restrict                    xx,
 786                      rvec * gmx_restrict                    ff,
 787                      t_forcerec * gmx_restrict              fr,
 788                      t_mdatoms * gmx_restrict               mdatoms,
 789                      nb_kernel_data_t * gmx_restrict        kernel_data,
 790                      t_nrnb * gmx_restrict                  nrnb)
 791 {
 792     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 793      * just 0 for non-waters.
 794      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 795      * jnr indices corresponding to data put in the four positions in the SIMD register.
 796      */
 797     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 798     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 799     int              jnrA,jnrB,jnrC,jnrD;
 800     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 801     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 802     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 803     real             rcutoff_scalar;
 804     real             *shiftvec,*fshift,*x,*f;
 805     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 806     real             scratch[4*DIM];
 807     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 808     int              vdwioffset0;
 809     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 810     int              vdwioffset1;
 811     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 812     int              vdwioffset2;
 813     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 814     int              vdwioffset3;
 815     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 816     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 817     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 818     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 819     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 820     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 821     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 822     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 823     real             *charge;
 824     int              nvdwtype;
 825     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 826     int              *vdwtype;
 827     real             *vdwparam;
 828     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 829     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 830     __m128i          vfitab;
 831     __m128i          ifour       = _mm_set1_epi32(4);
 832     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 833     real             *vftab;
 834     __m128i          ewitab;
 835     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 836     real             *ewtab;
 837     __m128           dummy_mask,cutoff_mask;
 838     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 839     __m128           one     = _mm_set1_ps(1.0);
 840     __m128           two     = _mm_set1_ps(2.0);
 841     x                = xx[0];
 842     f                = ff[0];
 843
 844     nri              = nlist->nri;
 845     iinr             = nlist->iinr;
 846     jindex           = nlist->jindex;
 847     jjnr             = nlist->jjnr;
 848     shiftidx         = nlist->shift;
 849     gid              = nlist->gid;
 850     shiftvec         = fr->shift_vec[0];
 851     fshift           = fr->fshift[0];
 852     facel            = _mm_set1_ps(fr->epsfac);
 853     charge           = mdatoms->chargeA;
 854     nvdwtype         = fr->ntype;
 855     vdwparam         = fr->nbfp;
 856     vdwtype          = mdatoms->typeA;
 857
 858     vftab            = kernel_data->table_vdw->data;
 859     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 860
 861     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 862     ewtab            = fr->ic->tabq_coul_F;
 863     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 864     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 865
 866     /* Setup water-specific parameters */
 867     inr              = nlist->iinr[0];
 868     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 869     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 870     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 871     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 872
 873     /* Avoid stupid compiler warnings */
 874     jnrA = jnrB = jnrC = jnrD = 0;
 875     j_coord_offsetA = 0;
 876     j_coord_offsetB = 0;
 877     j_coord_offsetC = 0;
 878     j_coord_offsetD = 0;
 879
 880     outeriter        = 0;
 881     inneriter        = 0;
 882
 883     for(iidx=0;iidx<4*DIM;iidx++)
 884     {
 885         scratch[iidx] = 0.0;
 886     }
 887
 888     /* Start outer loop over neighborlists */
 889     for(iidx=0; iidx<nri; iidx++)
 890     {
 891         /* Load shift vector for this list */
 892         i_shift_offset   = DIM*shiftidx[iidx];
 893
 894         /* Load limits for loop over neighbors */
 895         j_index_start    = jindex[iidx];
 896         j_index_end      = jindex[iidx+1];
 897
 898         /* Get outer coordinate index */
 899         inr              = iinr[iidx];
 900         i_coord_offset   = DIM*inr;
 901
 902         /* Load i particle coords and add shift vector */
 903         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 904                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 905
 906         fix0             = _mm_setzero_ps();
 907         fiy0             = _mm_setzero_ps();
 908         fiz0             = _mm_setzero_ps();
 909         fix1             = _mm_setzero_ps();
 910         fiy1             = _mm_setzero_ps();
 911         fiz1             = _mm_setzero_ps();
 912         fix2             = _mm_setzero_ps();
 913         fiy2             = _mm_setzero_ps();
 914         fiz2             = _mm_setzero_ps();
 915         fix3             = _mm_setzero_ps();
 916         fiy3             = _mm_setzero_ps();
 917         fiz3             = _mm_setzero_ps();
 918
 919         /* Start inner kernel loop */
 920         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 921         {
 922
 923             /* Get j neighbor index, and coordinate index */
 924             jnrA             = jjnr[jidx];
 925             jnrB             = jjnr[jidx+1];
 926             jnrC             = jjnr[jidx+2];
 927             jnrD             = jjnr[jidx+3];
 928             j_coord_offsetA  = DIM*jnrA;
 929             j_coord_offsetB  = DIM*jnrB;
 930             j_coord_offsetC  = DIM*jnrC;
 931             j_coord_offsetD  = DIM*jnrD;
 932
 933             /* load j atom coordinates */
 934             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 935                                               x+j_coord_offsetC,x+j_coord_offsetD,
 936                                               &jx0,&jy0,&jz0);
 937
 938             /* Calculate displacement vector */
 939             dx00             = _mm_sub_ps(ix0,jx0);
 940             dy00             = _mm_sub_ps(iy0,jy0);
 941             dz00             = _mm_sub_ps(iz0,jz0);
 942             dx10             = _mm_sub_ps(ix1,jx0);
 943             dy10             = _mm_sub_ps(iy1,jy0);
 944             dz10             = _mm_sub_ps(iz1,jz0);
 945             dx20             = _mm_sub_ps(ix2,jx0);
 946             dy20             = _mm_sub_ps(iy2,jy0);
 947             dz20             = _mm_sub_ps(iz2,jz0);
 948             dx30             = _mm_sub_ps(ix3,jx0);
 949             dy30             = _mm_sub_ps(iy3,jy0);
 950             dz30             = _mm_sub_ps(iz3,jz0);
 951
 952             /* Calculate squared distance and things based on it */
 953             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 954             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 955             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 956             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 957
 958             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 959             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 960             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 961             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 962
 963             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 964             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 965             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 966
 967             /* Load parameters for j particles */
 968             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 969                                                               charge+jnrC+0,charge+jnrD+0);
 970             vdwjidx0A        = 2*vdwtype[jnrA+0];
 971             vdwjidx0B        = 2*vdwtype[jnrB+0];
 972             vdwjidx0C        = 2*vdwtype[jnrC+0];
 973             vdwjidx0D        = 2*vdwtype[jnrD+0];
 974
 975             fjx0             = _mm_setzero_ps();
 976             fjy0             = _mm_setzero_ps();
 977             fjz0             = _mm_setzero_ps();
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             r00              = _mm_mul_ps(rsq00,rinv00);
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 987                                          vdwparam+vdwioffset0+vdwjidx0B,
 988                                          vdwparam+vdwioffset0+vdwjidx0C,
 989                                          vdwparam+vdwioffset0+vdwjidx0D,
 990                                          &c6_00,&c12_00);
 991
 992             /* Calculate table index by multiplying r with table scale and truncate to integer */
 993             rt               = _mm_mul_ps(r00,vftabscale);
 994             vfitab           = _mm_cvttps_epi32(rt);
 995             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 996             vfitab           = _mm_slli_epi32(vfitab,3);
 997
 998             /* CUBIC SPLINE TABLE DISPERSION */
 999             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1000             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1001             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1002             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1003             _MM_TRANSPOSE4_PS(Y,F,G,H);
1004             Heps             = _mm_mul_ps(vfeps,H);
1005             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1006             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1007             fvdw6            = _mm_mul_ps(c6_00,FF);
1008
1009             /* CUBIC SPLINE TABLE REPULSION */
1010             vfitab           = _mm_add_epi32(vfitab,ifour);
1011             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1012             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1013             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1014             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1015             _MM_TRANSPOSE4_PS(Y,F,G,H);
1016             Heps             = _mm_mul_ps(vfeps,H);
1017             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1018             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1019             fvdw12           = _mm_mul_ps(c12_00,FF);
1020             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1021
1022             fscal            = fvdw;
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm_mul_ps(fscal,dx00);
1026             ty               = _mm_mul_ps(fscal,dy00);
1027             tz               = _mm_mul_ps(fscal,dz00);
1028
1029             /* Update vectorial force */
1030             fix0             = _mm_add_ps(fix0,tx);
1031             fiy0             = _mm_add_ps(fiy0,ty);
1032             fiz0             = _mm_add_ps(fiz0,tz);
1033
1034             fjx0             = _mm_add_ps(fjx0,tx);
1035             fjy0             = _mm_add_ps(fjy0,ty);
1036             fjz0             = _mm_add_ps(fjz0,tz);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             r10              = _mm_mul_ps(rsq10,rinv10);
1043
1044             /* Compute parameters for interactions between i and j atoms */
1045             qq10             = _mm_mul_ps(iq1,jq0);
1046
1047             /* EWALD ELECTROSTATICS */
1048
1049             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1050             ewrt             = _mm_mul_ps(r10,ewtabscale);
1051             ewitab           = _mm_cvttps_epi32(ewrt);
1052             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1053             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1054                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1055                                          &ewtabF,&ewtabFn);
1056             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1057             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1058
1059             fscal            = felec;
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm_mul_ps(fscal,dx10);
1063             ty               = _mm_mul_ps(fscal,dy10);
1064             tz               = _mm_mul_ps(fscal,dz10);
1065
1066             /* Update vectorial force */
1067             fix1             = _mm_add_ps(fix1,tx);
1068             fiy1             = _mm_add_ps(fiy1,ty);
1069             fiz1             = _mm_add_ps(fiz1,tz);
1070
1071             fjx0             = _mm_add_ps(fjx0,tx);
1072             fjy0             = _mm_add_ps(fjy0,ty);
1073             fjz0             = _mm_add_ps(fjz0,tz);
1074
1075             /**************************
1076              * CALCULATE INTERACTIONS *
1077              **************************/
1078
1079             r20              = _mm_mul_ps(rsq20,rinv20);
1080
1081             /* Compute parameters for interactions between i and j atoms */
1082             qq20             = _mm_mul_ps(iq2,jq0);
1083
1084             /* EWALD ELECTROSTATICS */
1085
1086             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1087             ewrt             = _mm_mul_ps(r20,ewtabscale);
1088             ewitab           = _mm_cvttps_epi32(ewrt);
1089             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1090             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1091                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1092                                          &ewtabF,&ewtabFn);
1093             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1094             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1095
1096             fscal            = felec;
1097
1098             /* Calculate temporary vectorial force */
1099             tx               = _mm_mul_ps(fscal,dx20);
1100             ty               = _mm_mul_ps(fscal,dy20);
1101             tz               = _mm_mul_ps(fscal,dz20);
1102
1103             /* Update vectorial force */
1104             fix2             = _mm_add_ps(fix2,tx);
1105             fiy2             = _mm_add_ps(fiy2,ty);
1106             fiz2             = _mm_add_ps(fiz2,tz);
1107
1108             fjx0             = _mm_add_ps(fjx0,tx);
1109             fjy0             = _mm_add_ps(fjy0,ty);
1110             fjz0             = _mm_add_ps(fjz0,tz);
1111
1112             /**************************
1113              * CALCULATE INTERACTIONS *
1114              **************************/
1115
1116             r30              = _mm_mul_ps(rsq30,rinv30);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq30             = _mm_mul_ps(iq3,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1124             ewrt             = _mm_mul_ps(r30,ewtabscale);
1125             ewitab           = _mm_cvttps_epi32(ewrt);
1126             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1127             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1128                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1129                                          &ewtabF,&ewtabFn);
1130             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1131             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1132
1133             fscal            = felec;
1134
1135             /* Calculate temporary vectorial force */
1136             tx               = _mm_mul_ps(fscal,dx30);
1137             ty               = _mm_mul_ps(fscal,dy30);
1138             tz               = _mm_mul_ps(fscal,dz30);
1139
1140             /* Update vectorial force */
1141             fix3             = _mm_add_ps(fix3,tx);
1142             fiy3             = _mm_add_ps(fiy3,ty);
1143             fiz3             = _mm_add_ps(fiz3,tz);
1144
1145             fjx0             = _mm_add_ps(fjx0,tx);
1146             fjy0             = _mm_add_ps(fjy0,ty);
1147             fjz0             = _mm_add_ps(fjz0,tz);
1148
1149             fjptrA             = f+j_coord_offsetA;
1150             fjptrB             = f+j_coord_offsetB;
1151             fjptrC             = f+j_coord_offsetC;
1152             fjptrD             = f+j_coord_offsetD;
1153
1154             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1155
1156             /* Inner loop uses 156 flops */
1157         }
1158
1159         if(jidx<j_index_end)
1160         {
1161
1162             /* Get j neighbor index, and coordinate index */
1163             jnrlistA         = jjnr[jidx];
1164             jnrlistB         = jjnr[jidx+1];
1165             jnrlistC         = jjnr[jidx+2];
1166             jnrlistD         = jjnr[jidx+3];
1167             /* Sign of each element will be negative for non-real atoms.
1168              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1169              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1170              */
1171             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1172             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1173             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1174             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1175             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1176             j_coord_offsetA  = DIM*jnrA;
1177             j_coord_offsetB  = DIM*jnrB;
1178             j_coord_offsetC  = DIM*jnrC;
1179             j_coord_offsetD  = DIM*jnrD;
1180
1181             /* load j atom coordinates */
1182             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1183                                               x+j_coord_offsetC,x+j_coord_offsetD,
1184                                               &jx0,&jy0,&jz0);
1185
1186             /* Calculate displacement vector */
1187             dx00             = _mm_sub_ps(ix0,jx0);
1188             dy00             = _mm_sub_ps(iy0,jy0);
1189             dz00             = _mm_sub_ps(iz0,jz0);
1190             dx10             = _mm_sub_ps(ix1,jx0);
1191             dy10             = _mm_sub_ps(iy1,jy0);
1192             dz10             = _mm_sub_ps(iz1,jz0);
1193             dx20             = _mm_sub_ps(ix2,jx0);
1194             dy20             = _mm_sub_ps(iy2,jy0);
1195             dz20             = _mm_sub_ps(iz2,jz0);
1196             dx30             = _mm_sub_ps(ix3,jx0);
1197             dy30             = _mm_sub_ps(iy3,jy0);
1198             dz30             = _mm_sub_ps(iz3,jz0);
1199
1200             /* Calculate squared distance and things based on it */
1201             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1202             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1203             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1204             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1205
1206             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1207             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1208             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1209             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1210
1211             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1212             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1213             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1214
1215             /* Load parameters for j particles */
1216             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1217                                                               charge+jnrC+0,charge+jnrD+0);
1218             vdwjidx0A        = 2*vdwtype[jnrA+0];
1219             vdwjidx0B        = 2*vdwtype[jnrB+0];
1220             vdwjidx0C        = 2*vdwtype[jnrC+0];
1221             vdwjidx0D        = 2*vdwtype[jnrD+0];
1222
1223             fjx0             = _mm_setzero_ps();
1224             fjy0             = _mm_setzero_ps();
1225             fjz0             = _mm_setzero_ps();
1226
1227             /**************************
1228              * CALCULATE INTERACTIONS *
1229              **************************/
1230
1231             r00              = _mm_mul_ps(rsq00,rinv00);
1232             r00              = _mm_andnot_ps(dummy_mask,r00);
1233
1234             /* Compute parameters for interactions between i and j atoms */
1235             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1236                                          vdwparam+vdwioffset0+vdwjidx0B,
1237                                          vdwparam+vdwioffset0+vdwjidx0C,
1238                                          vdwparam+vdwioffset0+vdwjidx0D,
1239                                          &c6_00,&c12_00);
1240
1241             /* Calculate table index by multiplying r with table scale and truncate to integer */
1242             rt               = _mm_mul_ps(r00,vftabscale);
1243             vfitab           = _mm_cvttps_epi32(rt);
1244             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1245             vfitab           = _mm_slli_epi32(vfitab,3);
1246
1247             /* CUBIC SPLINE TABLE DISPERSION */
1248             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1249             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1250             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1251             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1252             _MM_TRANSPOSE4_PS(Y,F,G,H);
1253             Heps             = _mm_mul_ps(vfeps,H);
1254             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1255             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1256             fvdw6            = _mm_mul_ps(c6_00,FF);
1257
1258             /* CUBIC SPLINE TABLE REPULSION */
1259             vfitab           = _mm_add_epi32(vfitab,ifour);
1260             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1261             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1262             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1263             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1264             _MM_TRANSPOSE4_PS(Y,F,G,H);
1265             Heps             = _mm_mul_ps(vfeps,H);
1266             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1267             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1268             fvdw12           = _mm_mul_ps(c12_00,FF);
1269             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1270
1271             fscal            = fvdw;
1272
1273             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1274
1275             /* Calculate temporary vectorial force */
1276             tx               = _mm_mul_ps(fscal,dx00);
1277             ty               = _mm_mul_ps(fscal,dy00);
1278             tz               = _mm_mul_ps(fscal,dz00);
1279
1280             /* Update vectorial force */
1281             fix0             = _mm_add_ps(fix0,tx);
1282             fiy0             = _mm_add_ps(fiy0,ty);
1283             fiz0             = _mm_add_ps(fiz0,tz);
1284
1285             fjx0             = _mm_add_ps(fjx0,tx);
1286             fjy0             = _mm_add_ps(fjy0,ty);
1287             fjz0             = _mm_add_ps(fjz0,tz);
1288
1289             /**************************
1290              * CALCULATE INTERACTIONS *
1291              **************************/
1292
1293             r10              = _mm_mul_ps(rsq10,rinv10);
1294             r10              = _mm_andnot_ps(dummy_mask,r10);
1295
1296             /* Compute parameters for interactions between i and j atoms */
1297             qq10             = _mm_mul_ps(iq1,jq0);
1298
1299             /* EWALD ELECTROSTATICS */
1300
1301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1302             ewrt             = _mm_mul_ps(r10,ewtabscale);
1303             ewitab           = _mm_cvttps_epi32(ewrt);
1304             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1305             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1306                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1307                                          &ewtabF,&ewtabFn);
1308             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1309             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1310
1311             fscal            = felec;
1312
1313             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1314
1315             /* Calculate temporary vectorial force */
1316             tx               = _mm_mul_ps(fscal,dx10);
1317             ty               = _mm_mul_ps(fscal,dy10);
1318             tz               = _mm_mul_ps(fscal,dz10);
1319
1320             /* Update vectorial force */
1321             fix1             = _mm_add_ps(fix1,tx);
1322             fiy1             = _mm_add_ps(fiy1,ty);
1323             fiz1             = _mm_add_ps(fiz1,tz);
1324
1325             fjx0             = _mm_add_ps(fjx0,tx);
1326             fjy0             = _mm_add_ps(fjy0,ty);
1327             fjz0             = _mm_add_ps(fjz0,tz);
1328
1329             /**************************
1330              * CALCULATE INTERACTIONS *
1331              **************************/
1332
1333             r20              = _mm_mul_ps(rsq20,rinv20);
1334             r20              = _mm_andnot_ps(dummy_mask,r20);
1335
1336             /* Compute parameters for interactions between i and j atoms */
1337             qq20             = _mm_mul_ps(iq2,jq0);
1338
1339             /* EWALD ELECTROSTATICS */
1340
1341             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1342             ewrt             = _mm_mul_ps(r20,ewtabscale);
1343             ewitab           = _mm_cvttps_epi32(ewrt);
1344             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1345             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1346                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1347                                          &ewtabF,&ewtabFn);
1348             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1349             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1350
1351             fscal            = felec;
1352
1353             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1354
1355             /* Calculate temporary vectorial force */
1356             tx               = _mm_mul_ps(fscal,dx20);
1357             ty               = _mm_mul_ps(fscal,dy20);
1358             tz               = _mm_mul_ps(fscal,dz20);
1359
1360             /* Update vectorial force */
1361             fix2             = _mm_add_ps(fix2,tx);
1362             fiy2             = _mm_add_ps(fiy2,ty);
1363             fiz2             = _mm_add_ps(fiz2,tz);
1364
1365             fjx0             = _mm_add_ps(fjx0,tx);
1366             fjy0             = _mm_add_ps(fjy0,ty);
1367             fjz0             = _mm_add_ps(fjz0,tz);
1368
1369             /**************************
1370              * CALCULATE INTERACTIONS *
1371              **************************/
1372
1373             r30              = _mm_mul_ps(rsq30,rinv30);
1374             r30              = _mm_andnot_ps(dummy_mask,r30);
1375
1376             /* Compute parameters for interactions between i and j atoms */
1377             qq30             = _mm_mul_ps(iq3,jq0);
1378
1379             /* EWALD ELECTROSTATICS */
1380
1381             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1382             ewrt             = _mm_mul_ps(r30,ewtabscale);
1383             ewitab           = _mm_cvttps_epi32(ewrt);
1384             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1385             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1386                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1387                                          &ewtabF,&ewtabFn);
1388             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1389             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1390
1391             fscal            = felec;
1392
1393             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1394
1395             /* Calculate temporary vectorial force */
1396             tx               = _mm_mul_ps(fscal,dx30);
1397             ty               = _mm_mul_ps(fscal,dy30);
1398             tz               = _mm_mul_ps(fscal,dz30);
1399
1400             /* Update vectorial force */
1401             fix3             = _mm_add_ps(fix3,tx);
1402             fiy3             = _mm_add_ps(fiy3,ty);
1403             fiz3             = _mm_add_ps(fiz3,tz);
1404
1405             fjx0             = _mm_add_ps(fjx0,tx);
1406             fjy0             = _mm_add_ps(fjy0,ty);
1407             fjz0             = _mm_add_ps(fjz0,tz);
1408
1409             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1410             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1411             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1412             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1413
1414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1415
1416             /* Inner loop uses 160 flops */
1417         }
1418
1419         /* End of innermost loop */
1420
1421         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1422                                               f+i_coord_offset,fshift+i_shift_offset);
1423
1424         /* Increment number of inner iterations */
1425         inneriter                  += j_index_end - j_index_start;
1426
1427         /* Outer loop uses 24 flops */
1428     }
1429
1430     /* Increment number of outer iterations */
1431     outeriter        += nri;
1432
1433     /* Update outer/inner flops */
1434
1435     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*160);
1436 }