src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwLJ_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse2_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     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     krf              = _mm_set1_ps(fr->ic->k_rf);
 109     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 110     crf              = _mm_set1_ps(fr->ic->c_rf);
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 118     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 119     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 120     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     for(iidx=0;iidx<4*DIM;iidx++)
 133     {
 134         scratch[iidx] = 0.0;
 135     }
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 153                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 154
 155         fix0             = _mm_setzero_ps();
 156         fiy0             = _mm_setzero_ps();
 157         fiz0             = _mm_setzero_ps();
 158         fix1             = _mm_setzero_ps();
 159         fiy1             = _mm_setzero_ps();
 160         fiz1             = _mm_setzero_ps();
 161         fix2             = _mm_setzero_ps();
 162         fiy2             = _mm_setzero_ps();
 163         fiz2             = _mm_setzero_ps();
 164         fix3             = _mm_setzero_ps();
 165         fiy3             = _mm_setzero_ps();
 166         fiz3             = _mm_setzero_ps();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm_setzero_ps();
 170         vvdwsum          = _mm_setzero_ps();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               x+j_coord_offsetC,x+j_coord_offsetD,
 189                                               &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm_sub_ps(ix0,jx0);
 193             dy00             = _mm_sub_ps(iy0,jy0);
 194             dz00             = _mm_sub_ps(iz0,jz0);
 195             dx10             = _mm_sub_ps(ix1,jx0);
 196             dy10             = _mm_sub_ps(iy1,jy0);
 197             dz10             = _mm_sub_ps(iz1,jz0);
 198             dx20             = _mm_sub_ps(ix2,jx0);
 199             dy20             = _mm_sub_ps(iy2,jy0);
 200             dz20             = _mm_sub_ps(iz2,jz0);
 201             dx30             = _mm_sub_ps(ix3,jx0);
 202             dy30             = _mm_sub_ps(iy3,jy0);
 203             dz30             = _mm_sub_ps(iz3,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 207             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 208             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 209             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 210
 211             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 212             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 213             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 214
 215             rinvsq00         = gmx_mm_inv_ps(rsq00);
 216             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 217             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 218             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 222                                                               charge+jnrC+0,charge+jnrD+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225             vdwjidx0C        = 2*vdwtype[jnrC+0];
 226             vdwjidx0D        = 2*vdwtype[jnrD+0];
 227
 228             fjx0             = _mm_setzero_ps();
 229             fjy0             = _mm_setzero_ps();
 230             fjz0             = _mm_setzero_ps();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,
 239                                          vdwparam+vdwioffset0+vdwjidx0C,
 240                                          vdwparam+vdwioffset0+vdwjidx0D,
 241                                          &c6_00,&c12_00);
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 247             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 248             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 249             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 253
 254             fscal            = fvdw;
 255
 256             /* Calculate temporary vectorial force */
 257             tx               = _mm_mul_ps(fscal,dx00);
 258             ty               = _mm_mul_ps(fscal,dy00);
 259             tz               = _mm_mul_ps(fscal,dz00);
 260
 261             /* Update vectorial force */
 262             fix0             = _mm_add_ps(fix0,tx);
 263             fiy0             = _mm_add_ps(fiy0,ty);
 264             fiz0             = _mm_add_ps(fiz0,tz);
 265
 266             fjx0             = _mm_add_ps(fjx0,tx);
 267             fjy0             = _mm_add_ps(fjy0,ty);
 268             fjz0             = _mm_add_ps(fjz0,tz);
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq10             = _mm_mul_ps(iq1,jq0);
 276
 277             /* REACTION-FIELD ELECTROSTATICS */
 278             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 279             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_ps(velecsum,velec);
 283
 284             fscal            = felec;
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_ps(fscal,dx10);
 288             ty               = _mm_mul_ps(fscal,dy10);
 289             tz               = _mm_mul_ps(fscal,dz10);
 290
 291             /* Update vectorial force */
 292             fix1             = _mm_add_ps(fix1,tx);
 293             fiy1             = _mm_add_ps(fiy1,ty);
 294             fiz1             = _mm_add_ps(fiz1,tz);
 295
 296             fjx0             = _mm_add_ps(fjx0,tx);
 297             fjy0             = _mm_add_ps(fjy0,ty);
 298             fjz0             = _mm_add_ps(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq20             = _mm_mul_ps(iq2,jq0);
 306
 307             /* REACTION-FIELD ELECTROSTATICS */
 308             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 309             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_ps(fscal,dx20);
 318             ty               = _mm_mul_ps(fscal,dy20);
 319             tz               = _mm_mul_ps(fscal,dz20);
 320
 321             /* Update vectorial force */
 322             fix2             = _mm_add_ps(fix2,tx);
 323             fiy2             = _mm_add_ps(fiy2,ty);
 324             fiz2             = _mm_add_ps(fiz2,tz);
 325
 326             fjx0             = _mm_add_ps(fjx0,tx);
 327             fjy0             = _mm_add_ps(fjy0,ty);
 328             fjz0             = _mm_add_ps(fjz0,tz);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq30             = _mm_mul_ps(iq3,jq0);
 336
 337             /* REACTION-FIELD ELECTROSTATICS */
 338             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 339             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx30);
 348             ty               = _mm_mul_ps(fscal,dy30);
 349             tz               = _mm_mul_ps(fscal,dz30);
 350
 351             /* Update vectorial force */
 352             fix3             = _mm_add_ps(fix3,tx);
 353             fiy3             = _mm_add_ps(fiy3,ty);
 354             fiz3             = _mm_add_ps(fiz3,tz);
 355
 356             fjx0             = _mm_add_ps(fjx0,tx);
 357             fjy0             = _mm_add_ps(fjy0,ty);
 358             fjz0             = _mm_add_ps(fjz0,tz);
 359
 360             fjptrA             = f+j_coord_offsetA;
 361             fjptrB             = f+j_coord_offsetB;
 362             fjptrC             = f+j_coord_offsetC;
 363             fjptrD             = f+j_coord_offsetD;
 364
 365             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 366
 367             /* Inner loop uses 128 flops */
 368         }
 369
 370         if(jidx<j_index_end)
 371         {
 372
 373             /* Get j neighbor index, and coordinate index */
 374             jnrlistA         = jjnr[jidx];
 375             jnrlistB         = jjnr[jidx+1];
 376             jnrlistC         = jjnr[jidx+2];
 377             jnrlistD         = jjnr[jidx+3];
 378             /* Sign of each element will be negative for non-real atoms.
 379              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 380              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 381              */
 382             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 383             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 384             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 385             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 386             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 387             j_coord_offsetA  = DIM*jnrA;
 388             j_coord_offsetB  = DIM*jnrB;
 389             j_coord_offsetC  = DIM*jnrC;
 390             j_coord_offsetD  = DIM*jnrD;
 391
 392             /* load j atom coordinates */
 393             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 394                                               x+j_coord_offsetC,x+j_coord_offsetD,
 395                                               &jx0,&jy0,&jz0);
 396
 397             /* Calculate displacement vector */
 398             dx00             = _mm_sub_ps(ix0,jx0);
 399             dy00             = _mm_sub_ps(iy0,jy0);
 400             dz00             = _mm_sub_ps(iz0,jz0);
 401             dx10             = _mm_sub_ps(ix1,jx0);
 402             dy10             = _mm_sub_ps(iy1,jy0);
 403             dz10             = _mm_sub_ps(iz1,jz0);
 404             dx20             = _mm_sub_ps(ix2,jx0);
 405             dy20             = _mm_sub_ps(iy2,jy0);
 406             dz20             = _mm_sub_ps(iz2,jz0);
 407             dx30             = _mm_sub_ps(ix3,jx0);
 408             dy30             = _mm_sub_ps(iy3,jy0);
 409             dz30             = _mm_sub_ps(iz3,jz0);
 410
 411             /* Calculate squared distance and things based on it */
 412             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 413             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 414             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 415             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 416
 417             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 418             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 419             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 420
 421             rinvsq00         = gmx_mm_inv_ps(rsq00);
 422             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 423             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 424             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 425
 426             /* Load parameters for j particles */
 427             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 428                                                               charge+jnrC+0,charge+jnrD+0);
 429             vdwjidx0A        = 2*vdwtype[jnrA+0];
 430             vdwjidx0B        = 2*vdwtype[jnrB+0];
 431             vdwjidx0C        = 2*vdwtype[jnrC+0];
 432             vdwjidx0D        = 2*vdwtype[jnrD+0];
 433
 434             fjx0             = _mm_setzero_ps();
 435             fjy0             = _mm_setzero_ps();
 436             fjz0             = _mm_setzero_ps();
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             /* Compute parameters for interactions between i and j atoms */
 443             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 444                                          vdwparam+vdwioffset0+vdwjidx0B,
 445                                          vdwparam+vdwioffset0+vdwjidx0C,
 446                                          vdwparam+vdwioffset0+vdwjidx0D,
 447                                          &c6_00,&c12_00);
 448
 449             /* LENNARD-JONES DISPERSION/REPULSION */
 450
 451             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 452             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 453             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 454             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 455             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 459             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 460
 461             fscal            = fvdw;
 462
 463             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 464
 465             /* Calculate temporary vectorial force */
 466             tx               = _mm_mul_ps(fscal,dx00);
 467             ty               = _mm_mul_ps(fscal,dy00);
 468             tz               = _mm_mul_ps(fscal,dz00);
 469
 470             /* Update vectorial force */
 471             fix0             = _mm_add_ps(fix0,tx);
 472             fiy0             = _mm_add_ps(fiy0,ty);
 473             fiz0             = _mm_add_ps(fiz0,tz);
 474
 475             fjx0             = _mm_add_ps(fjx0,tx);
 476             fjy0             = _mm_add_ps(fjy0,ty);
 477             fjz0             = _mm_add_ps(fjz0,tz);
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq10             = _mm_mul_ps(iq1,jq0);
 485
 486             /* REACTION-FIELD ELECTROSTATICS */
 487             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 488             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 489
 490             /* Update potential sum for this i atom from the interaction with this j atom. */
 491             velec            = _mm_andnot_ps(dummy_mask,velec);
 492             velecsum         = _mm_add_ps(velecsum,velec);
 493
 494             fscal            = felec;
 495
 496             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 497
 498             /* Calculate temporary vectorial force */
 499             tx               = _mm_mul_ps(fscal,dx10);
 500             ty               = _mm_mul_ps(fscal,dy10);
 501             tz               = _mm_mul_ps(fscal,dz10);
 502
 503             /* Update vectorial force */
 504             fix1             = _mm_add_ps(fix1,tx);
 505             fiy1             = _mm_add_ps(fiy1,ty);
 506             fiz1             = _mm_add_ps(fiz1,tz);
 507
 508             fjx0             = _mm_add_ps(fjx0,tx);
 509             fjy0             = _mm_add_ps(fjy0,ty);
 510             fjz0             = _mm_add_ps(fjz0,tz);
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             qq20             = _mm_mul_ps(iq2,jq0);
 518
 519             /* REACTION-FIELD ELECTROSTATICS */
 520             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 521             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             velec            = _mm_andnot_ps(dummy_mask,velec);
 525             velecsum         = _mm_add_ps(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_ps(fscal,dx20);
 533             ty               = _mm_mul_ps(fscal,dy20);
 534             tz               = _mm_mul_ps(fscal,dz20);
 535
 536             /* Update vectorial force */
 537             fix2             = _mm_add_ps(fix2,tx);
 538             fiy2             = _mm_add_ps(fiy2,ty);
 539             fiz2             = _mm_add_ps(fiz2,tz);
 540
 541             fjx0             = _mm_add_ps(fjx0,tx);
 542             fjy0             = _mm_add_ps(fjy0,ty);
 543             fjz0             = _mm_add_ps(fjz0,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq30             = _mm_mul_ps(iq3,jq0);
 551
 552             /* REACTION-FIELD ELECTROSTATICS */
 553             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 554             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_andnot_ps(dummy_mask,velec);
 558             velecsum         = _mm_add_ps(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_ps(fscal,dx30);
 566             ty               = _mm_mul_ps(fscal,dy30);
 567             tz               = _mm_mul_ps(fscal,dz30);
 568
 569             /* Update vectorial force */
 570             fix3             = _mm_add_ps(fix3,tx);
 571             fiy3             = _mm_add_ps(fiy3,ty);
 572             fiz3             = _mm_add_ps(fiz3,tz);
 573
 574             fjx0             = _mm_add_ps(fjx0,tx);
 575             fjy0             = _mm_add_ps(fjy0,ty);
 576             fjz0             = _mm_add_ps(fjz0,tz);
 577
 578             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 579             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 580             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 581             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 582
 583             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 584
 585             /* Inner loop uses 128 flops */
 586         }
 587
 588         /* End of innermost loop */
 589
 590         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 591                                               f+i_coord_offset,fshift+i_shift_offset);
 592
 593         ggid                        = gid[iidx];
 594         /* Update potential energies */
 595         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 596         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 597
 598         /* Increment number of inner iterations */
 599         inneriter                  += j_index_end - j_index_start;
 600
 601         /* Outer loop uses 26 flops */
 602     }
 603
 604     /* Increment number of outer iterations */
 605     outeriter        += nri;
 606
 607     /* Update outer/inner flops */
 608
 609     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*128);
 610 }
 611 /*
 612  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse2_single
 613  * Electrostatics interaction: ReactionField
 614  * VdW interaction:            LennardJones
 615  * Geometry:                   Water4-Particle
 616  * Calculate force/pot:        Force
 617  */
 618 void
 619 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse2_single
 620                     (t_nblist * gmx_restrict                nlist,
 621                      rvec * gmx_restrict                    xx,
 622                      rvec * gmx_restrict                    ff,
 623                      t_forcerec * gmx_restrict              fr,
 624                      t_mdatoms * gmx_restrict               mdatoms,
 625                      nb_kernel_data_t * gmx_restrict        kernel_data,
 626                      t_nrnb * gmx_restrict                  nrnb)
 627 {
 628     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 629      * just 0 for non-waters.
 630      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 631      * jnr indices corresponding to data put in the four positions in the SIMD register.
 632      */
 633     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 634     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 635     int              jnrA,jnrB,jnrC,jnrD;
 636     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 637     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 638     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 639     real             rcutoff_scalar;
 640     real             *shiftvec,*fshift,*x,*f;
 641     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 642     real             scratch[4*DIM];
 643     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 644     int              vdwioffset0;
 645     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 646     int              vdwioffset1;
 647     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 648     int              vdwioffset2;
 649     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 650     int              vdwioffset3;
 651     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 652     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 653     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 654     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 655     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 656     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 657     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 658     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 659     real             *charge;
 660     int              nvdwtype;
 661     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 662     int              *vdwtype;
 663     real             *vdwparam;
 664     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 665     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 666     __m128           dummy_mask,cutoff_mask;
 667     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 668     __m128           one     = _mm_set1_ps(1.0);
 669     __m128           two     = _mm_set1_ps(2.0);
 670     x                = xx[0];
 671     f                = ff[0];
 672
 673     nri              = nlist->nri;
 674     iinr             = nlist->iinr;
 675     jindex           = nlist->jindex;
 676     jjnr             = nlist->jjnr;
 677     shiftidx         = nlist->shift;
 678     gid              = nlist->gid;
 679     shiftvec         = fr->shift_vec[0];
 680     fshift           = fr->fshift[0];
 681     facel            = _mm_set1_ps(fr->epsfac);
 682     charge           = mdatoms->chargeA;
 683     krf              = _mm_set1_ps(fr->ic->k_rf);
 684     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 685     crf              = _mm_set1_ps(fr->ic->c_rf);
 686     nvdwtype         = fr->ntype;
 687     vdwparam         = fr->nbfp;
 688     vdwtype          = mdatoms->typeA;
 689
 690     /* Setup water-specific parameters */
 691     inr              = nlist->iinr[0];
 692     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 693     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 694     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 695     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 696
 697     /* Avoid stupid compiler warnings */
 698     jnrA = jnrB = jnrC = jnrD = 0;
 699     j_coord_offsetA = 0;
 700     j_coord_offsetB = 0;
 701     j_coord_offsetC = 0;
 702     j_coord_offsetD = 0;
 703
 704     outeriter        = 0;
 705     inneriter        = 0;
 706
 707     for(iidx=0;iidx<4*DIM;iidx++)
 708     {
 709         scratch[iidx] = 0.0;
 710     }
 711
 712     /* Start outer loop over neighborlists */
 713     for(iidx=0; iidx<nri; iidx++)
 714     {
 715         /* Load shift vector for this list */
 716         i_shift_offset   = DIM*shiftidx[iidx];
 717
 718         /* Load limits for loop over neighbors */
 719         j_index_start    = jindex[iidx];
 720         j_index_end      = jindex[iidx+1];
 721
 722         /* Get outer coordinate index */
 723         inr              = iinr[iidx];
 724         i_coord_offset   = DIM*inr;
 725
 726         /* Load i particle coords and add shift vector */
 727         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 728                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 729
 730         fix0             = _mm_setzero_ps();
 731         fiy0             = _mm_setzero_ps();
 732         fiz0             = _mm_setzero_ps();
 733         fix1             = _mm_setzero_ps();
 734         fiy1             = _mm_setzero_ps();
 735         fiz1             = _mm_setzero_ps();
 736         fix2             = _mm_setzero_ps();
 737         fiy2             = _mm_setzero_ps();
 738         fiz2             = _mm_setzero_ps();
 739         fix3             = _mm_setzero_ps();
 740         fiy3             = _mm_setzero_ps();
 741         fiz3             = _mm_setzero_ps();
 742
 743         /* Start inner kernel loop */
 744         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 745         {
 746
 747             /* Get j neighbor index, and coordinate index */
 748             jnrA             = jjnr[jidx];
 749             jnrB             = jjnr[jidx+1];
 750             jnrC             = jjnr[jidx+2];
 751             jnrD             = jjnr[jidx+3];
 752             j_coord_offsetA  = DIM*jnrA;
 753             j_coord_offsetB  = DIM*jnrB;
 754             j_coord_offsetC  = DIM*jnrC;
 755             j_coord_offsetD  = DIM*jnrD;
 756
 757             /* load j atom coordinates */
 758             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 759                                               x+j_coord_offsetC,x+j_coord_offsetD,
 760                                               &jx0,&jy0,&jz0);
 761
 762             /* Calculate displacement vector */
 763             dx00             = _mm_sub_ps(ix0,jx0);
 764             dy00             = _mm_sub_ps(iy0,jy0);
 765             dz00             = _mm_sub_ps(iz0,jz0);
 766             dx10             = _mm_sub_ps(ix1,jx0);
 767             dy10             = _mm_sub_ps(iy1,jy0);
 768             dz10             = _mm_sub_ps(iz1,jz0);
 769             dx20             = _mm_sub_ps(ix2,jx0);
 770             dy20             = _mm_sub_ps(iy2,jy0);
 771             dz20             = _mm_sub_ps(iz2,jz0);
 772             dx30             = _mm_sub_ps(ix3,jx0);
 773             dy30             = _mm_sub_ps(iy3,jy0);
 774             dz30             = _mm_sub_ps(iz3,jz0);
 775
 776             /* Calculate squared distance and things based on it */
 777             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 778             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 779             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 780             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 781
 782             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 783             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 784             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 785
 786             rinvsq00         = gmx_mm_inv_ps(rsq00);
 787             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 788             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 789             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 790
 791             /* Load parameters for j particles */
 792             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 793                                                               charge+jnrC+0,charge+jnrD+0);
 794             vdwjidx0A        = 2*vdwtype[jnrA+0];
 795             vdwjidx0B        = 2*vdwtype[jnrB+0];
 796             vdwjidx0C        = 2*vdwtype[jnrC+0];
 797             vdwjidx0D        = 2*vdwtype[jnrD+0];
 798
 799             fjx0             = _mm_setzero_ps();
 800             fjy0             = _mm_setzero_ps();
 801             fjz0             = _mm_setzero_ps();
 802
 803             /**************************
 804              * CALCULATE INTERACTIONS *
 805              **************************/
 806
 807             /* Compute parameters for interactions between i and j atoms */
 808             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 809                                          vdwparam+vdwioffset0+vdwjidx0B,
 810                                          vdwparam+vdwioffset0+vdwjidx0C,
 811                                          vdwparam+vdwioffset0+vdwjidx0D,
 812                                          &c6_00,&c12_00);
 813
 814             /* LENNARD-JONES DISPERSION/REPULSION */
 815
 816             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 817             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 818
 819             fscal            = fvdw;
 820
 821             /* Calculate temporary vectorial force */
 822             tx               = _mm_mul_ps(fscal,dx00);
 823             ty               = _mm_mul_ps(fscal,dy00);
 824             tz               = _mm_mul_ps(fscal,dz00);
 825
 826             /* Update vectorial force */
 827             fix0             = _mm_add_ps(fix0,tx);
 828             fiy0             = _mm_add_ps(fiy0,ty);
 829             fiz0             = _mm_add_ps(fiz0,tz);
 830
 831             fjx0             = _mm_add_ps(fjx0,tx);
 832             fjy0             = _mm_add_ps(fjy0,ty);
 833             fjz0             = _mm_add_ps(fjz0,tz);
 834
 835             /**************************
 836              * CALCULATE INTERACTIONS *
 837              **************************/
 838
 839             /* Compute parameters for interactions between i and j atoms */
 840             qq10             = _mm_mul_ps(iq1,jq0);
 841
 842             /* REACTION-FIELD ELECTROSTATICS */
 843             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 844
 845             fscal            = felec;
 846
 847             /* Calculate temporary vectorial force */
 848             tx               = _mm_mul_ps(fscal,dx10);
 849             ty               = _mm_mul_ps(fscal,dy10);
 850             tz               = _mm_mul_ps(fscal,dz10);
 851
 852             /* Update vectorial force */
 853             fix1             = _mm_add_ps(fix1,tx);
 854             fiy1             = _mm_add_ps(fiy1,ty);
 855             fiz1             = _mm_add_ps(fiz1,tz);
 856
 857             fjx0             = _mm_add_ps(fjx0,tx);
 858             fjy0             = _mm_add_ps(fjy0,ty);
 859             fjz0             = _mm_add_ps(fjz0,tz);
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             /* Compute parameters for interactions between i and j atoms */
 866             qq20             = _mm_mul_ps(iq2,jq0);
 867
 868             /* REACTION-FIELD ELECTROSTATICS */
 869             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 870
 871             fscal            = felec;
 872
 873             /* Calculate temporary vectorial force */
 874             tx               = _mm_mul_ps(fscal,dx20);
 875             ty               = _mm_mul_ps(fscal,dy20);
 876             tz               = _mm_mul_ps(fscal,dz20);
 877
 878             /* Update vectorial force */
 879             fix2             = _mm_add_ps(fix2,tx);
 880             fiy2             = _mm_add_ps(fiy2,ty);
 881             fiz2             = _mm_add_ps(fiz2,tz);
 882
 883             fjx0             = _mm_add_ps(fjx0,tx);
 884             fjy0             = _mm_add_ps(fjy0,ty);
 885             fjz0             = _mm_add_ps(fjz0,tz);
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq30             = _mm_mul_ps(iq3,jq0);
 893
 894             /* REACTION-FIELD ELECTROSTATICS */
 895             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 896
 897             fscal            = felec;
 898
 899             /* Calculate temporary vectorial force */
 900             tx               = _mm_mul_ps(fscal,dx30);
 901             ty               = _mm_mul_ps(fscal,dy30);
 902             tz               = _mm_mul_ps(fscal,dz30);
 903
 904             /* Update vectorial force */
 905             fix3             = _mm_add_ps(fix3,tx);
 906             fiy3             = _mm_add_ps(fiy3,ty);
 907             fiz3             = _mm_add_ps(fiz3,tz);
 908
 909             fjx0             = _mm_add_ps(fjx0,tx);
 910             fjy0             = _mm_add_ps(fjy0,ty);
 911             fjz0             = _mm_add_ps(fjz0,tz);
 912
 913             fjptrA             = f+j_coord_offsetA;
 914             fjptrB             = f+j_coord_offsetB;
 915             fjptrC             = f+j_coord_offsetC;
 916             fjptrD             = f+j_coord_offsetD;
 917
 918             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 919
 920             /* Inner loop uses 108 flops */
 921         }
 922
 923         if(jidx<j_index_end)
 924         {
 925
 926             /* Get j neighbor index, and coordinate index */
 927             jnrlistA         = jjnr[jidx];
 928             jnrlistB         = jjnr[jidx+1];
 929             jnrlistC         = jjnr[jidx+2];
 930             jnrlistD         = jjnr[jidx+3];
 931             /* Sign of each element will be negative for non-real atoms.
 932              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 933              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 934              */
 935             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 936             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 937             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 938             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 939             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 940             j_coord_offsetA  = DIM*jnrA;
 941             j_coord_offsetB  = DIM*jnrB;
 942             j_coord_offsetC  = DIM*jnrC;
 943             j_coord_offsetD  = DIM*jnrD;
 944
 945             /* load j atom coordinates */
 946             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 947                                               x+j_coord_offsetC,x+j_coord_offsetD,
 948                                               &jx0,&jy0,&jz0);
 949
 950             /* Calculate displacement vector */
 951             dx00             = _mm_sub_ps(ix0,jx0);
 952             dy00             = _mm_sub_ps(iy0,jy0);
 953             dz00             = _mm_sub_ps(iz0,jz0);
 954             dx10             = _mm_sub_ps(ix1,jx0);
 955             dy10             = _mm_sub_ps(iy1,jy0);
 956             dz10             = _mm_sub_ps(iz1,jz0);
 957             dx20             = _mm_sub_ps(ix2,jx0);
 958             dy20             = _mm_sub_ps(iy2,jy0);
 959             dz20             = _mm_sub_ps(iz2,jz0);
 960             dx30             = _mm_sub_ps(ix3,jx0);
 961             dy30             = _mm_sub_ps(iy3,jy0);
 962             dz30             = _mm_sub_ps(iz3,jz0);
 963
 964             /* Calculate squared distance and things based on it */
 965             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 966             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 967             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 968             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 969
 970             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 971             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 972             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 973
 974             rinvsq00         = gmx_mm_inv_ps(rsq00);
 975             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 976             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 977             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 978
 979             /* Load parameters for j particles */
 980             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 981                                                               charge+jnrC+0,charge+jnrD+0);
 982             vdwjidx0A        = 2*vdwtype[jnrA+0];
 983             vdwjidx0B        = 2*vdwtype[jnrB+0];
 984             vdwjidx0C        = 2*vdwtype[jnrC+0];
 985             vdwjidx0D        = 2*vdwtype[jnrD+0];
 986
 987             fjx0             = _mm_setzero_ps();
 988             fjy0             = _mm_setzero_ps();
 989             fjz0             = _mm_setzero_ps();
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             /* Compute parameters for interactions between i and j atoms */
 996             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 997                                          vdwparam+vdwioffset0+vdwjidx0B,
 998                                          vdwparam+vdwioffset0+vdwjidx0C,
 999                                          vdwparam+vdwioffset0+vdwjidx0D,
1000                                          &c6_00,&c12_00);
1001
1002             /* LENNARD-JONES DISPERSION/REPULSION */
1003
1004             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1005             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1006
1007             fscal            = fvdw;
1008
1009             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1010
1011             /* Calculate temporary vectorial force */
1012             tx               = _mm_mul_ps(fscal,dx00);
1013             ty               = _mm_mul_ps(fscal,dy00);
1014             tz               = _mm_mul_ps(fscal,dz00);
1015
1016             /* Update vectorial force */
1017             fix0             = _mm_add_ps(fix0,tx);
1018             fiy0             = _mm_add_ps(fiy0,ty);
1019             fiz0             = _mm_add_ps(fiz0,tz);
1020
1021             fjx0             = _mm_add_ps(fjx0,tx);
1022             fjy0             = _mm_add_ps(fjy0,ty);
1023             fjz0             = _mm_add_ps(fjz0,tz);
1024
1025             /**************************
1026              * CALCULATE INTERACTIONS *
1027              **************************/
1028
1029             /* Compute parameters for interactions between i and j atoms */
1030             qq10             = _mm_mul_ps(iq1,jq0);
1031
1032             /* REACTION-FIELD ELECTROSTATICS */
1033             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1034
1035             fscal            = felec;
1036
1037             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm_mul_ps(fscal,dx10);
1041             ty               = _mm_mul_ps(fscal,dy10);
1042             tz               = _mm_mul_ps(fscal,dz10);
1043
1044             /* Update vectorial force */
1045             fix1             = _mm_add_ps(fix1,tx);
1046             fiy1             = _mm_add_ps(fiy1,ty);
1047             fiz1             = _mm_add_ps(fiz1,tz);
1048
1049             fjx0             = _mm_add_ps(fjx0,tx);
1050             fjy0             = _mm_add_ps(fjy0,ty);
1051             fjz0             = _mm_add_ps(fjz0,tz);
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             /* Compute parameters for interactions between i and j atoms */
1058             qq20             = _mm_mul_ps(iq2,jq0);
1059
1060             /* REACTION-FIELD ELECTROSTATICS */
1061             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1062
1063             fscal            = felec;
1064
1065             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm_mul_ps(fscal,dx20);
1069             ty               = _mm_mul_ps(fscal,dy20);
1070             tz               = _mm_mul_ps(fscal,dz20);
1071
1072             /* Update vectorial force */
1073             fix2             = _mm_add_ps(fix2,tx);
1074             fiy2             = _mm_add_ps(fiy2,ty);
1075             fiz2             = _mm_add_ps(fiz2,tz);
1076
1077             fjx0             = _mm_add_ps(fjx0,tx);
1078             fjy0             = _mm_add_ps(fjy0,ty);
1079             fjz0             = _mm_add_ps(fjz0,tz);
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq30             = _mm_mul_ps(iq3,jq0);
1087
1088             /* REACTION-FIELD ELECTROSTATICS */
1089             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1090
1091             fscal            = felec;
1092
1093             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm_mul_ps(fscal,dx30);
1097             ty               = _mm_mul_ps(fscal,dy30);
1098             tz               = _mm_mul_ps(fscal,dz30);
1099
1100             /* Update vectorial force */
1101             fix3             = _mm_add_ps(fix3,tx);
1102             fiy3             = _mm_add_ps(fiy3,ty);
1103             fiz3             = _mm_add_ps(fiz3,tz);
1104
1105             fjx0             = _mm_add_ps(fjx0,tx);
1106             fjy0             = _mm_add_ps(fjy0,ty);
1107             fjz0             = _mm_add_ps(fjz0,tz);
1108
1109             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1110             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1111             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1112             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1113
1114             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1115
1116             /* Inner loop uses 108 flops */
1117         }
1118
1119         /* End of innermost loop */
1120
1121         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1122                                               f+i_coord_offset,fshift+i_shift_offset);
1123
1124         /* Increment number of inner iterations */
1125         inneriter                  += j_index_end - j_index_start;
1126
1127         /* Outer loop uses 24 flops */
1128     }
1129
1130     /* Increment number of outer iterations */
1131     outeriter        += nri;
1132
1133     /* Update outer/inner flops */
1134
1135     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1136 }