src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwLJ_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_ElecRF_VdwLJ_GeomW4P1_VF_sse4_1_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_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     __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             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,
 235                                          vdwparam+vdwioffset0+vdwjidx0C,
 236                                          vdwparam+vdwioffset0+vdwjidx0D,
 237                                          &c6_00,&c12_00);
 238
 239             /* LENNARD-JONES DISPERSION/REPULSION */
 240
 241             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 242             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 243             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 244             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 245             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 249
 250             fscal            = fvdw;
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_ps(fscal,dx00);
 254             ty               = _mm_mul_ps(fscal,dy00);
 255             tz               = _mm_mul_ps(fscal,dz00);
 256
 257             /* Update vectorial force */
 258             fix0             = _mm_add_ps(fix0,tx);
 259             fiy0             = _mm_add_ps(fiy0,ty);
 260             fiz0             = _mm_add_ps(fiz0,tz);
 261
 262             fjptrA             = f+j_coord_offsetA;
 263             fjptrB             = f+j_coord_offsetB;
 264             fjptrC             = f+j_coord_offsetC;
 265             fjptrD             = f+j_coord_offsetD;
 266             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq10             = _mm_mul_ps(iq1,jq0);
 274
 275             /* REACTION-FIELD ELECTROSTATICS */
 276             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 277             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 278
 279             /* Update potential sum for this i atom from the interaction with this j atom. */
 280             velecsum         = _mm_add_ps(velecsum,velec);
 281
 282             fscal            = felec;
 283
 284             /* Calculate temporary vectorial force */
 285             tx               = _mm_mul_ps(fscal,dx10);
 286             ty               = _mm_mul_ps(fscal,dy10);
 287             tz               = _mm_mul_ps(fscal,dz10);
 288
 289             /* Update vectorial force */
 290             fix1             = _mm_add_ps(fix1,tx);
 291             fiy1             = _mm_add_ps(fiy1,ty);
 292             fiz1             = _mm_add_ps(fiz1,tz);
 293
 294             fjptrA             = f+j_coord_offsetA;
 295             fjptrB             = f+j_coord_offsetB;
 296             fjptrC             = f+j_coord_offsetC;
 297             fjptrD             = f+j_coord_offsetD;
 298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,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             fjptrA             = f+j_coord_offsetA;
 327             fjptrB             = f+j_coord_offsetB;
 328             fjptrC             = f+j_coord_offsetC;
 329             fjptrD             = f+j_coord_offsetD;
 330             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq30             = _mm_mul_ps(iq3,jq0);
 338
 339             /* REACTION-FIELD ELECTROSTATICS */
 340             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 341             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_ps(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_ps(fscal,dx30);
 350             ty               = _mm_mul_ps(fscal,dy30);
 351             tz               = _mm_mul_ps(fscal,dz30);
 352
 353             /* Update vectorial force */
 354             fix3             = _mm_add_ps(fix3,tx);
 355             fiy3             = _mm_add_ps(fiy3,ty);
 356             fiz3             = _mm_add_ps(fiz3,tz);
 357
 358             fjptrA             = f+j_coord_offsetA;
 359             fjptrB             = f+j_coord_offsetB;
 360             fjptrC             = f+j_coord_offsetC;
 361             fjptrD             = f+j_coord_offsetD;
 362             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 363
 364             /* Inner loop uses 128 flops */
 365         }
 366
 367         if(jidx<j_index_end)
 368         {
 369
 370             /* Get j neighbor index, and coordinate index */
 371             jnrlistA         = jjnr[jidx];
 372             jnrlistB         = jjnr[jidx+1];
 373             jnrlistC         = jjnr[jidx+2];
 374             jnrlistD         = jjnr[jidx+3];
 375             /* Sign of each element will be negative for non-real atoms.
 376              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 377              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 378              */
 379             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 380             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 381             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 382             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 383             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 384             j_coord_offsetA  = DIM*jnrA;
 385             j_coord_offsetB  = DIM*jnrB;
 386             j_coord_offsetC  = DIM*jnrC;
 387             j_coord_offsetD  = DIM*jnrD;
 388
 389             /* load j atom coordinates */
 390             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 391                                               x+j_coord_offsetC,x+j_coord_offsetD,
 392                                               &jx0,&jy0,&jz0);
 393
 394             /* Calculate displacement vector */
 395             dx00             = _mm_sub_ps(ix0,jx0);
 396             dy00             = _mm_sub_ps(iy0,jy0);
 397             dz00             = _mm_sub_ps(iz0,jz0);
 398             dx10             = _mm_sub_ps(ix1,jx0);
 399             dy10             = _mm_sub_ps(iy1,jy0);
 400             dz10             = _mm_sub_ps(iz1,jz0);
 401             dx20             = _mm_sub_ps(ix2,jx0);
 402             dy20             = _mm_sub_ps(iy2,jy0);
 403             dz20             = _mm_sub_ps(iz2,jz0);
 404             dx30             = _mm_sub_ps(ix3,jx0);
 405             dy30             = _mm_sub_ps(iy3,jy0);
 406             dz30             = _mm_sub_ps(iz3,jz0);
 407
 408             /* Calculate squared distance and things based on it */
 409             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 410             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 411             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 412             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 413
 414             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 415             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 416             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 417
 418             rinvsq00         = gmx_mm_inv_ps(rsq00);
 419             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 420             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 421             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 422
 423             /* Load parameters for j particles */
 424             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 425                                                               charge+jnrC+0,charge+jnrD+0);
 426             vdwjidx0A        = 2*vdwtype[jnrA+0];
 427             vdwjidx0B        = 2*vdwtype[jnrB+0];
 428             vdwjidx0C        = 2*vdwtype[jnrC+0];
 429             vdwjidx0D        = 2*vdwtype[jnrD+0];
 430
 431             /**************************
 432              * CALCULATE INTERACTIONS *
 433              **************************/
 434
 435             /* Compute parameters for interactions between i and j atoms */
 436             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 437                                          vdwparam+vdwioffset0+vdwjidx0B,
 438                                          vdwparam+vdwioffset0+vdwjidx0C,
 439                                          vdwparam+vdwioffset0+vdwjidx0D,
 440                                          &c6_00,&c12_00);
 441
 442             /* LENNARD-JONES DISPERSION/REPULSION */
 443
 444             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 445             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 446             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 447             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 448             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 449
 450             /* Update potential sum for this i atom from the interaction with this j atom. */
 451             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 452             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 453
 454             fscal            = fvdw;
 455
 456             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 457
 458             /* Calculate temporary vectorial force */
 459             tx               = _mm_mul_ps(fscal,dx00);
 460             ty               = _mm_mul_ps(fscal,dy00);
 461             tz               = _mm_mul_ps(fscal,dz00);
 462
 463             /* Update vectorial force */
 464             fix0             = _mm_add_ps(fix0,tx);
 465             fiy0             = _mm_add_ps(fiy0,ty);
 466             fiz0             = _mm_add_ps(fiz0,tz);
 467
 468             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 469             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 470             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 471             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 472             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             /* Compute parameters for interactions between i and j atoms */
 479             qq10             = _mm_mul_ps(iq1,jq0);
 480
 481             /* REACTION-FIELD ELECTROSTATICS */
 482             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 483             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 484
 485             /* Update potential sum for this i atom from the interaction with this j atom. */
 486             velec            = _mm_andnot_ps(dummy_mask,velec);
 487             velecsum         = _mm_add_ps(velecsum,velec);
 488
 489             fscal            = felec;
 490
 491             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 492
 493             /* Calculate temporary vectorial force */
 494             tx               = _mm_mul_ps(fscal,dx10);
 495             ty               = _mm_mul_ps(fscal,dy10);
 496             tz               = _mm_mul_ps(fscal,dz10);
 497
 498             /* Update vectorial force */
 499             fix1             = _mm_add_ps(fix1,tx);
 500             fiy1             = _mm_add_ps(fiy1,ty);
 501             fiz1             = _mm_add_ps(fiz1,tz);
 502
 503             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 504             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 505             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 506             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 507             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             /* Compute parameters for interactions between i and j atoms */
 514             qq20             = _mm_mul_ps(iq2,jq0);
 515
 516             /* REACTION-FIELD ELECTROSTATICS */
 517             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 518             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm_andnot_ps(dummy_mask,velec);
 522             velecsum         = _mm_add_ps(velecsum,velec);
 523
 524             fscal            = felec;
 525
 526             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 527
 528             /* Calculate temporary vectorial force */
 529             tx               = _mm_mul_ps(fscal,dx20);
 530             ty               = _mm_mul_ps(fscal,dy20);
 531             tz               = _mm_mul_ps(fscal,dz20);
 532
 533             /* Update vectorial force */
 534             fix2             = _mm_add_ps(fix2,tx);
 535             fiy2             = _mm_add_ps(fiy2,ty);
 536             fiz2             = _mm_add_ps(fiz2,tz);
 537
 538             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 539             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 540             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 541             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 542             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq30             = _mm_mul_ps(iq3,jq0);
 550
 551             /* REACTION-FIELD ELECTROSTATICS */
 552             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 553             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 554
 555             /* Update potential sum for this i atom from the interaction with this j atom. */
 556             velec            = _mm_andnot_ps(dummy_mask,velec);
 557             velecsum         = _mm_add_ps(velecsum,velec);
 558
 559             fscal            = felec;
 560
 561             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 562
 563             /* Calculate temporary vectorial force */
 564             tx               = _mm_mul_ps(fscal,dx30);
 565             ty               = _mm_mul_ps(fscal,dy30);
 566             tz               = _mm_mul_ps(fscal,dz30);
 567
 568             /* Update vectorial force */
 569             fix3             = _mm_add_ps(fix3,tx);
 570             fiy3             = _mm_add_ps(fiy3,ty);
 571             fiz3             = _mm_add_ps(fiz3,tz);
 572
 573             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 574             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 575             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 576             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 577             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 578
 579             /* Inner loop uses 128 flops */
 580         }
 581
 582         /* End of innermost loop */
 583
 584         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 585                                               f+i_coord_offset,fshift+i_shift_offset);
 586
 587         ggid                        = gid[iidx];
 588         /* Update potential energies */
 589         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 590         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 591
 592         /* Increment number of inner iterations */
 593         inneriter                  += j_index_end - j_index_start;
 594
 595         /* Outer loop uses 26 flops */
 596     }
 597
 598     /* Increment number of outer iterations */
 599     outeriter        += nri;
 600
 601     /* Update outer/inner flops */
 602
 603     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*128);
 604 }
 605 /*
 606  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse4_1_single
 607  * Electrostatics interaction: ReactionField
 608  * VdW interaction:            LennardJones
 609  * Geometry:                   Water4-Particle
 610  * Calculate force/pot:        Force
 611  */
 612 void
 613 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse4_1_single
 614                     (t_nblist * gmx_restrict                nlist,
 615                      rvec * gmx_restrict                    xx,
 616                      rvec * gmx_restrict                    ff,
 617                      t_forcerec * gmx_restrict              fr,
 618                      t_mdatoms * gmx_restrict               mdatoms,
 619                      nb_kernel_data_t * gmx_restrict        kernel_data,
 620                      t_nrnb * gmx_restrict                  nrnb)
 621 {
 622     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 623      * just 0 for non-waters.
 624      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 625      * jnr indices corresponding to data put in the four positions in the SIMD register.
 626      */
 627     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 628     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 629     int              jnrA,jnrB,jnrC,jnrD;
 630     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 631     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 632     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 633     real             rcutoff_scalar;
 634     real             *shiftvec,*fshift,*x,*f;
 635     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 636     real             scratch[4*DIM];
 637     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 638     int              vdwioffset0;
 639     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 640     int              vdwioffset1;
 641     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 642     int              vdwioffset2;
 643     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 644     int              vdwioffset3;
 645     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 646     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 647     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 648     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 649     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 650     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 651     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 652     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 653     real             *charge;
 654     int              nvdwtype;
 655     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 656     int              *vdwtype;
 657     real             *vdwparam;
 658     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 659     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 660     __m128           dummy_mask,cutoff_mask;
 661     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 662     __m128           one     = _mm_set1_ps(1.0);
 663     __m128           two     = _mm_set1_ps(2.0);
 664     x                = xx[0];
 665     f                = ff[0];
 666
 667     nri              = nlist->nri;
 668     iinr             = nlist->iinr;
 669     jindex           = nlist->jindex;
 670     jjnr             = nlist->jjnr;
 671     shiftidx         = nlist->shift;
 672     gid              = nlist->gid;
 673     shiftvec         = fr->shift_vec[0];
 674     fshift           = fr->fshift[0];
 675     facel            = _mm_set1_ps(fr->epsfac);
 676     charge           = mdatoms->chargeA;
 677     krf              = _mm_set1_ps(fr->ic->k_rf);
 678     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 679     crf              = _mm_set1_ps(fr->ic->c_rf);
 680     nvdwtype         = fr->ntype;
 681     vdwparam         = fr->nbfp;
 682     vdwtype          = mdatoms->typeA;
 683
 684     /* Setup water-specific parameters */
 685     inr              = nlist->iinr[0];
 686     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 687     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 688     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 689     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 690
 691     /* Avoid stupid compiler warnings */
 692     jnrA = jnrB = jnrC = jnrD = 0;
 693     j_coord_offsetA = 0;
 694     j_coord_offsetB = 0;
 695     j_coord_offsetC = 0;
 696     j_coord_offsetD = 0;
 697
 698     outeriter        = 0;
 699     inneriter        = 0;
 700
 701     for(iidx=0;iidx<4*DIM;iidx++)
 702     {
 703         scratch[iidx] = 0.0;
 704     }
 705
 706     /* Start outer loop over neighborlists */
 707     for(iidx=0; iidx<nri; iidx++)
 708     {
 709         /* Load shift vector for this list */
 710         i_shift_offset   = DIM*shiftidx[iidx];
 711
 712         /* Load limits for loop over neighbors */
 713         j_index_start    = jindex[iidx];
 714         j_index_end      = jindex[iidx+1];
 715
 716         /* Get outer coordinate index */
 717         inr              = iinr[iidx];
 718         i_coord_offset   = DIM*inr;
 719
 720         /* Load i particle coords and add shift vector */
 721         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 722                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 723
 724         fix0             = _mm_setzero_ps();
 725         fiy0             = _mm_setzero_ps();
 726         fiz0             = _mm_setzero_ps();
 727         fix1             = _mm_setzero_ps();
 728         fiy1             = _mm_setzero_ps();
 729         fiz1             = _mm_setzero_ps();
 730         fix2             = _mm_setzero_ps();
 731         fiy2             = _mm_setzero_ps();
 732         fiz2             = _mm_setzero_ps();
 733         fix3             = _mm_setzero_ps();
 734         fiy3             = _mm_setzero_ps();
 735         fiz3             = _mm_setzero_ps();
 736
 737         /* Start inner kernel loop */
 738         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 739         {
 740
 741             /* Get j neighbor index, and coordinate index */
 742             jnrA             = jjnr[jidx];
 743             jnrB             = jjnr[jidx+1];
 744             jnrC             = jjnr[jidx+2];
 745             jnrD             = jjnr[jidx+3];
 746             j_coord_offsetA  = DIM*jnrA;
 747             j_coord_offsetB  = DIM*jnrB;
 748             j_coord_offsetC  = DIM*jnrC;
 749             j_coord_offsetD  = DIM*jnrD;
 750
 751             /* load j atom coordinates */
 752             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 753                                               x+j_coord_offsetC,x+j_coord_offsetD,
 754                                               &jx0,&jy0,&jz0);
 755
 756             /* Calculate displacement vector */
 757             dx00             = _mm_sub_ps(ix0,jx0);
 758             dy00             = _mm_sub_ps(iy0,jy0);
 759             dz00             = _mm_sub_ps(iz0,jz0);
 760             dx10             = _mm_sub_ps(ix1,jx0);
 761             dy10             = _mm_sub_ps(iy1,jy0);
 762             dz10             = _mm_sub_ps(iz1,jz0);
 763             dx20             = _mm_sub_ps(ix2,jx0);
 764             dy20             = _mm_sub_ps(iy2,jy0);
 765             dz20             = _mm_sub_ps(iz2,jz0);
 766             dx30             = _mm_sub_ps(ix3,jx0);
 767             dy30             = _mm_sub_ps(iy3,jy0);
 768             dz30             = _mm_sub_ps(iz3,jz0);
 769
 770             /* Calculate squared distance and things based on it */
 771             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 772             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 773             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 774             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 775
 776             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 777             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 778             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 779
 780             rinvsq00         = gmx_mm_inv_ps(rsq00);
 781             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 782             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 783             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 784
 785             /* Load parameters for j particles */
 786             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 787                                                               charge+jnrC+0,charge+jnrD+0);
 788             vdwjidx0A        = 2*vdwtype[jnrA+0];
 789             vdwjidx0B        = 2*vdwtype[jnrB+0];
 790             vdwjidx0C        = 2*vdwtype[jnrC+0];
 791             vdwjidx0D        = 2*vdwtype[jnrD+0];
 792
 793             /**************************
 794              * CALCULATE INTERACTIONS *
 795              **************************/
 796
 797             /* Compute parameters for interactions between i and j atoms */
 798             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 799                                          vdwparam+vdwioffset0+vdwjidx0B,
 800                                          vdwparam+vdwioffset0+vdwjidx0C,
 801                                          vdwparam+vdwioffset0+vdwjidx0D,
 802                                          &c6_00,&c12_00);
 803
 804             /* LENNARD-JONES DISPERSION/REPULSION */
 805
 806             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 807             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 808
 809             fscal            = fvdw;
 810
 811             /* Calculate temporary vectorial force */
 812             tx               = _mm_mul_ps(fscal,dx00);
 813             ty               = _mm_mul_ps(fscal,dy00);
 814             tz               = _mm_mul_ps(fscal,dz00);
 815
 816             /* Update vectorial force */
 817             fix0             = _mm_add_ps(fix0,tx);
 818             fiy0             = _mm_add_ps(fiy0,ty);
 819             fiz0             = _mm_add_ps(fiz0,tz);
 820
 821             fjptrA             = f+j_coord_offsetA;
 822             fjptrB             = f+j_coord_offsetB;
 823             fjptrC             = f+j_coord_offsetC;
 824             fjptrD             = f+j_coord_offsetD;
 825             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             /* Compute parameters for interactions between i and j atoms */
 832             qq10             = _mm_mul_ps(iq1,jq0);
 833
 834             /* REACTION-FIELD ELECTROSTATICS */
 835             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 836
 837             fscal            = felec;
 838
 839             /* Calculate temporary vectorial force */
 840             tx               = _mm_mul_ps(fscal,dx10);
 841             ty               = _mm_mul_ps(fscal,dy10);
 842             tz               = _mm_mul_ps(fscal,dz10);
 843
 844             /* Update vectorial force */
 845             fix1             = _mm_add_ps(fix1,tx);
 846             fiy1             = _mm_add_ps(fiy1,ty);
 847             fiz1             = _mm_add_ps(fiz1,tz);
 848
 849             fjptrA             = f+j_coord_offsetA;
 850             fjptrB             = f+j_coord_offsetB;
 851             fjptrC             = f+j_coord_offsetC;
 852             fjptrD             = f+j_coord_offsetD;
 853             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 854
 855             /**************************
 856              * CALCULATE INTERACTIONS *
 857              **************************/
 858
 859             /* Compute parameters for interactions between i and j atoms */
 860             qq20             = _mm_mul_ps(iq2,jq0);
 861
 862             /* REACTION-FIELD ELECTROSTATICS */
 863             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 864
 865             fscal            = felec;
 866
 867             /* Calculate temporary vectorial force */
 868             tx               = _mm_mul_ps(fscal,dx20);
 869             ty               = _mm_mul_ps(fscal,dy20);
 870             tz               = _mm_mul_ps(fscal,dz20);
 871
 872             /* Update vectorial force */
 873             fix2             = _mm_add_ps(fix2,tx);
 874             fiy2             = _mm_add_ps(fiy2,ty);
 875             fiz2             = _mm_add_ps(fiz2,tz);
 876
 877             fjptrA             = f+j_coord_offsetA;
 878             fjptrB             = f+j_coord_offsetB;
 879             fjptrC             = f+j_coord_offsetC;
 880             fjptrD             = f+j_coord_offsetD;
 881             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 882
 883             /**************************
 884              * CALCULATE INTERACTIONS *
 885              **************************/
 886
 887             /* Compute parameters for interactions between i and j atoms */
 888             qq30             = _mm_mul_ps(iq3,jq0);
 889
 890             /* REACTION-FIELD ELECTROSTATICS */
 891             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 892
 893             fscal            = felec;
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm_mul_ps(fscal,dx30);
 897             ty               = _mm_mul_ps(fscal,dy30);
 898             tz               = _mm_mul_ps(fscal,dz30);
 899
 900             /* Update vectorial force */
 901             fix3             = _mm_add_ps(fix3,tx);
 902             fiy3             = _mm_add_ps(fiy3,ty);
 903             fiz3             = _mm_add_ps(fiz3,tz);
 904
 905             fjptrA             = f+j_coord_offsetA;
 906             fjptrB             = f+j_coord_offsetB;
 907             fjptrC             = f+j_coord_offsetC;
 908             fjptrD             = f+j_coord_offsetD;
 909             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 910
 911             /* Inner loop uses 108 flops */
 912         }
 913
 914         if(jidx<j_index_end)
 915         {
 916
 917             /* Get j neighbor index, and coordinate index */
 918             jnrlistA         = jjnr[jidx];
 919             jnrlistB         = jjnr[jidx+1];
 920             jnrlistC         = jjnr[jidx+2];
 921             jnrlistD         = jjnr[jidx+3];
 922             /* Sign of each element will be negative for non-real atoms.
 923              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 924              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 925              */
 926             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 927             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 928             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 929             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 930             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 931             j_coord_offsetA  = DIM*jnrA;
 932             j_coord_offsetB  = DIM*jnrB;
 933             j_coord_offsetC  = DIM*jnrC;
 934             j_coord_offsetD  = DIM*jnrD;
 935
 936             /* load j atom coordinates */
 937             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 938                                               x+j_coord_offsetC,x+j_coord_offsetD,
 939                                               &jx0,&jy0,&jz0);
 940
 941             /* Calculate displacement vector */
 942             dx00             = _mm_sub_ps(ix0,jx0);
 943             dy00             = _mm_sub_ps(iy0,jy0);
 944             dz00             = _mm_sub_ps(iz0,jz0);
 945             dx10             = _mm_sub_ps(ix1,jx0);
 946             dy10             = _mm_sub_ps(iy1,jy0);
 947             dz10             = _mm_sub_ps(iz1,jz0);
 948             dx20             = _mm_sub_ps(ix2,jx0);
 949             dy20             = _mm_sub_ps(iy2,jy0);
 950             dz20             = _mm_sub_ps(iz2,jz0);
 951             dx30             = _mm_sub_ps(ix3,jx0);
 952             dy30             = _mm_sub_ps(iy3,jy0);
 953             dz30             = _mm_sub_ps(iz3,jz0);
 954
 955             /* Calculate squared distance and things based on it */
 956             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 957             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 958             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 959             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 960
 961             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 962             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 963             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 964
 965             rinvsq00         = gmx_mm_inv_ps(rsq00);
 966             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 967             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 968             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 969
 970             /* Load parameters for j particles */
 971             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 972                                                               charge+jnrC+0,charge+jnrD+0);
 973             vdwjidx0A        = 2*vdwtype[jnrA+0];
 974             vdwjidx0B        = 2*vdwtype[jnrB+0];
 975             vdwjidx0C        = 2*vdwtype[jnrC+0];
 976             vdwjidx0D        = 2*vdwtype[jnrD+0];
 977
 978             /**************************
 979              * CALCULATE INTERACTIONS *
 980              **************************/
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 984                                          vdwparam+vdwioffset0+vdwjidx0B,
 985                                          vdwparam+vdwioffset0+vdwjidx0C,
 986                                          vdwparam+vdwioffset0+vdwjidx0D,
 987                                          &c6_00,&c12_00);
 988
 989             /* LENNARD-JONES DISPERSION/REPULSION */
 990
 991             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 992             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 993
 994             fscal            = fvdw;
 995
 996             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 997
 998             /* Calculate temporary vectorial force */
 999             tx               = _mm_mul_ps(fscal,dx00);
1000             ty               = _mm_mul_ps(fscal,dy00);
1001             tz               = _mm_mul_ps(fscal,dz00);
1002
1003             /* Update vectorial force */
1004             fix0             = _mm_add_ps(fix0,tx);
1005             fiy0             = _mm_add_ps(fiy0,ty);
1006             fiz0             = _mm_add_ps(fiz0,tz);
1007
1008             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1009             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1010             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1011             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1012             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq10             = _mm_mul_ps(iq1,jq0);
1020
1021             /* REACTION-FIELD ELECTROSTATICS */
1022             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1023
1024             fscal            = felec;
1025
1026             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1027
1028             /* Calculate temporary vectorial force */
1029             tx               = _mm_mul_ps(fscal,dx10);
1030             ty               = _mm_mul_ps(fscal,dy10);
1031             tz               = _mm_mul_ps(fscal,dz10);
1032
1033             /* Update vectorial force */
1034             fix1             = _mm_add_ps(fix1,tx);
1035             fiy1             = _mm_add_ps(fiy1,ty);
1036             fiz1             = _mm_add_ps(fiz1,tz);
1037
1038             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1039             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1040             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1041             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1042             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             /* Compute parameters for interactions between i and j atoms */
1049             qq20             = _mm_mul_ps(iq2,jq0);
1050
1051             /* REACTION-FIELD ELECTROSTATICS */
1052             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1053
1054             fscal            = felec;
1055
1056             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1057
1058             /* Calculate temporary vectorial force */
1059             tx               = _mm_mul_ps(fscal,dx20);
1060             ty               = _mm_mul_ps(fscal,dy20);
1061             tz               = _mm_mul_ps(fscal,dz20);
1062
1063             /* Update vectorial force */
1064             fix2             = _mm_add_ps(fix2,tx);
1065             fiy2             = _mm_add_ps(fiy2,ty);
1066             fiz2             = _mm_add_ps(fiz2,tz);
1067
1068             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1069             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1070             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1071             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1072             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1073
1074             /**************************
1075              * CALCULATE INTERACTIONS *
1076              **************************/
1077
1078             /* Compute parameters for interactions between i and j atoms */
1079             qq30             = _mm_mul_ps(iq3,jq0);
1080
1081             /* REACTION-FIELD ELECTROSTATICS */
1082             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1083
1084             fscal            = felec;
1085
1086             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1087
1088             /* Calculate temporary vectorial force */
1089             tx               = _mm_mul_ps(fscal,dx30);
1090             ty               = _mm_mul_ps(fscal,dy30);
1091             tz               = _mm_mul_ps(fscal,dz30);
1092
1093             /* Update vectorial force */
1094             fix3             = _mm_add_ps(fix3,tx);
1095             fiy3             = _mm_add_ps(fiy3,ty);
1096             fiz3             = _mm_add_ps(fiz3,tz);
1097
1098             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1099             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1100             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1101             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1102             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1103
1104             /* Inner loop uses 108 flops */
1105         }
1106
1107         /* End of innermost loop */
1108
1109         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1110                                               f+i_coord_offset,fshift+i_shift_offset);
1111
1112         /* Increment number of inner iterations */
1113         inneriter                  += j_index_end - j_index_start;
1114
1115         /* Outer loop uses 24 flops */
1116     }
1117
1118     /* Increment number of outer iterations */
1119     outeriter        += nri;
1120
1121     /* Update outer/inner flops */
1122
1123     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1124 }