src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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_ElecRFCut_VdwLJSh_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128           dummy_mask,cutoff_mask;
  86     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  87     __m128           one     = _mm_set1_ps(1.0);
  88     __m128           two     = _mm_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     krf              = _mm_set1_ps(fr->ic->k_rf);
 103     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 104     crf              = _mm_set1_ps(fr->ic->c_rf);
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     /* Setup water-specific parameters */
 110     inr              = nlist->iinr[0];
 111     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 112     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 113     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 114     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 115
 116     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 117     rcutoff_scalar   = fr->rcoulomb;
 118     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 119     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 120
 121     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 122     rvdw             = _mm_set1_ps(fr->rvdw);
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139         shX              = shiftvec[i_shift_offset+XX];
 140         shY              = shiftvec[i_shift_offset+YY];
 141         shZ              = shiftvec[i_shift_offset+ZZ];
 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         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 153         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 154         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 155         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 156         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 157         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 158         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 159         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 160         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165         fix1             = _mm_setzero_ps();
 166         fiy1             = _mm_setzero_ps();
 167         fiz1             = _mm_setzero_ps();
 168         fix2             = _mm_setzero_ps();
 169         fiy2             = _mm_setzero_ps();
 170         fiz2             = _mm_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174         vvdwsum          = _mm_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185
 186             j_coord_offsetA  = DIM*jnrA;
 187             j_coord_offsetB  = DIM*jnrB;
 188             j_coord_offsetC  = DIM*jnrC;
 189             j_coord_offsetD  = DIM*jnrD;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               x+j_coord_offsetC,x+j_coord_offsetD,
 194                                               &jx0,&jy0,&jz0);
 195
 196             /* Calculate displacement vector */
 197             dx00             = _mm_sub_ps(ix0,jx0);
 198             dy00             = _mm_sub_ps(iy0,jy0);
 199             dz00             = _mm_sub_ps(iz0,jz0);
 200             dx10             = _mm_sub_ps(ix1,jx0);
 201             dy10             = _mm_sub_ps(iy1,jy0);
 202             dz10             = _mm_sub_ps(iz1,jz0);
 203             dx20             = _mm_sub_ps(ix2,jx0);
 204             dy20             = _mm_sub_ps(iy2,jy0);
 205             dz20             = _mm_sub_ps(iz2,jz0);
 206
 207             /* Calculate squared distance and things based on it */
 208             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 209             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 211
 212             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 213             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 214             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 215
 216             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 217             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 218             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 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             if (gmx_mm_any_lt(rsq00,rcutoff2))
 233             {
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq00             = _mm_mul_ps(iq0,jq0);
 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             /* REACTION-FIELD ELECTROSTATICS */
 244             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 245             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 246
 247             /* LENNARD-JONES DISPERSION/REPULSION */
 248
 249             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 250             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 251             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 252             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 253                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 254             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 255
 256             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             velec            = _mm_and_ps(velec,cutoff_mask);
 260             velecsum         = _mm_add_ps(velecsum,velec);
 261             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 262             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 263
 264             fscal            = _mm_add_ps(felec,fvdw);
 265
 266             fscal            = _mm_and_ps(fscal,cutoff_mask);
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm_mul_ps(fscal,dx00);
 270             ty               = _mm_mul_ps(fscal,dy00);
 271             tz               = _mm_mul_ps(fscal,dz00);
 272
 273             /* Update vectorial force */
 274             fix0             = _mm_add_ps(fix0,tx);
 275             fiy0             = _mm_add_ps(fiy0,ty);
 276             fiz0             = _mm_add_ps(fiz0,tz);
 277
 278             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 279                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 280                                                    tx,ty,tz);
 281
 282             }
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             if (gmx_mm_any_lt(rsq10,rcutoff2))
 289             {
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq10             = _mm_mul_ps(iq1,jq0);
 293
 294             /* REACTION-FIELD ELECTROSTATICS */
 295             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 296             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 297
 298             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velec            = _mm_and_ps(velec,cutoff_mask);
 302             velecsum         = _mm_add_ps(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306             fscal            = _mm_and_ps(fscal,cutoff_mask);
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm_mul_ps(fscal,dx10);
 310             ty               = _mm_mul_ps(fscal,dy10);
 311             tz               = _mm_mul_ps(fscal,dz10);
 312
 313             /* Update vectorial force */
 314             fix1             = _mm_add_ps(fix1,tx);
 315             fiy1             = _mm_add_ps(fiy1,ty);
 316             fiz1             = _mm_add_ps(fiz1,tz);
 317
 318             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 319                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 320                                                    tx,ty,tz);
 321
 322             }
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             if (gmx_mm_any_lt(rsq20,rcutoff2))
 329             {
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq20             = _mm_mul_ps(iq2,jq0);
 333
 334             /* REACTION-FIELD ELECTROSTATICS */
 335             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 336             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 337
 338             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm_and_ps(velec,cutoff_mask);
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm_and_ps(fscal,cutoff_mask);
 347
 348             /* Calculate temporary vectorial force */
 349             tx               = _mm_mul_ps(fscal,dx20);
 350             ty               = _mm_mul_ps(fscal,dy20);
 351             tz               = _mm_mul_ps(fscal,dz20);
 352
 353             /* Update vectorial force */
 354             fix2             = _mm_add_ps(fix2,tx);
 355             fiy2             = _mm_add_ps(fiy2,ty);
 356             fiz2             = _mm_add_ps(fiz2,tz);
 357
 358             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 359                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 360                                                    tx,ty,tz);
 361
 362             }
 363
 364             /* Inner loop uses 126 flops */
 365         }
 366
 367         if(jidx<j_index_end)
 368         {
 369
 370             /* Get j neighbor index, and coordinate index */
 371             jnrA             = jjnr[jidx];
 372             jnrB             = jjnr[jidx+1];
 373             jnrC             = jjnr[jidx+2];
 374             jnrD             = jjnr[jidx+3];
 375
 376             /* Sign of each element will be negative for non-real atoms.
 377              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 378              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 379              */
 380             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 381             jnrA       = (jnrA>=0) ? jnrA : 0;
 382             jnrB       = (jnrB>=0) ? jnrB : 0;
 383             jnrC       = (jnrC>=0) ? jnrC : 0;
 384             jnrD       = (jnrD>=0) ? jnrD : 0;
 385
 386             j_coord_offsetA  = DIM*jnrA;
 387             j_coord_offsetB  = DIM*jnrB;
 388             j_coord_offsetC  = DIM*jnrC;
 389             j_coord_offsetD  = DIM*jnrD;
 390
 391             /* load j atom coordinates */
 392             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 393                                               x+j_coord_offsetC,x+j_coord_offsetD,
 394                                               &jx0,&jy0,&jz0);
 395
 396             /* Calculate displacement vector */
 397             dx00             = _mm_sub_ps(ix0,jx0);
 398             dy00             = _mm_sub_ps(iy0,jy0);
 399             dz00             = _mm_sub_ps(iz0,jz0);
 400             dx10             = _mm_sub_ps(ix1,jx0);
 401             dy10             = _mm_sub_ps(iy1,jy0);
 402             dz10             = _mm_sub_ps(iz1,jz0);
 403             dx20             = _mm_sub_ps(ix2,jx0);
 404             dy20             = _mm_sub_ps(iy2,jy0);
 405             dz20             = _mm_sub_ps(iz2,jz0);
 406
 407             /* Calculate squared distance and things based on it */
 408             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 409             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 410             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 411
 412             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 413             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 414             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 415
 416             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 417             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 418             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 419
 420             /* Load parameters for j particles */
 421             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 422                                                               charge+jnrC+0,charge+jnrD+0);
 423             vdwjidx0A        = 2*vdwtype[jnrA+0];
 424             vdwjidx0B        = 2*vdwtype[jnrB+0];
 425             vdwjidx0C        = 2*vdwtype[jnrC+0];
 426             vdwjidx0D        = 2*vdwtype[jnrD+0];
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             if (gmx_mm_any_lt(rsq00,rcutoff2))
 433             {
 434
 435             /* Compute parameters for interactions between i and j atoms */
 436             qq00             = _mm_mul_ps(iq0,jq0);
 437             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 438                                          vdwparam+vdwioffset0+vdwjidx0B,
 439                                          vdwparam+vdwioffset0+vdwjidx0C,
 440                                          vdwparam+vdwioffset0+vdwjidx0D,
 441                                          &c6_00,&c12_00);
 442
 443             /* REACTION-FIELD ELECTROSTATICS */
 444             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 445             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 446
 447             /* LENNARD-JONES DISPERSION/REPULSION */
 448
 449             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 450             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 451             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 452             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 453                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 454             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 455
 456             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             velec            = _mm_and_ps(velec,cutoff_mask);
 460             velec            = _mm_andnot_ps(dummy_mask,velec);
 461             velecsum         = _mm_add_ps(velecsum,velec);
 462             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 463             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 464             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 465
 466             fscal            = _mm_add_ps(felec,fvdw);
 467
 468             fscal            = _mm_and_ps(fscal,cutoff_mask);
 469
 470             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 471
 472             /* Calculate temporary vectorial force */
 473             tx               = _mm_mul_ps(fscal,dx00);
 474             ty               = _mm_mul_ps(fscal,dy00);
 475             tz               = _mm_mul_ps(fscal,dz00);
 476
 477             /* Update vectorial force */
 478             fix0             = _mm_add_ps(fix0,tx);
 479             fiy0             = _mm_add_ps(fiy0,ty);
 480             fiz0             = _mm_add_ps(fiz0,tz);
 481
 482             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 483                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 484                                                    tx,ty,tz);
 485
 486             }
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             if (gmx_mm_any_lt(rsq10,rcutoff2))
 493             {
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq10             = _mm_mul_ps(iq1,jq0);
 497
 498             /* REACTION-FIELD ELECTROSTATICS */
 499             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 500             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 501
 502             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 503
 504             /* Update potential sum for this i atom from the interaction with this j atom. */
 505             velec            = _mm_and_ps(velec,cutoff_mask);
 506             velec            = _mm_andnot_ps(dummy_mask,velec);
 507             velecsum         = _mm_add_ps(velecsum,velec);
 508
 509             fscal            = felec;
 510
 511             fscal            = _mm_and_ps(fscal,cutoff_mask);
 512
 513             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 514
 515             /* Calculate temporary vectorial force */
 516             tx               = _mm_mul_ps(fscal,dx10);
 517             ty               = _mm_mul_ps(fscal,dy10);
 518             tz               = _mm_mul_ps(fscal,dz10);
 519
 520             /* Update vectorial force */
 521             fix1             = _mm_add_ps(fix1,tx);
 522             fiy1             = _mm_add_ps(fiy1,ty);
 523             fiz1             = _mm_add_ps(fiz1,tz);
 524
 525             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 526                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 527                                                    tx,ty,tz);
 528
 529             }
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             if (gmx_mm_any_lt(rsq20,rcutoff2))
 536             {
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq20             = _mm_mul_ps(iq2,jq0);
 540
 541             /* REACTION-FIELD ELECTROSTATICS */
 542             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 543             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 544
 545             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 546
 547             /* Update potential sum for this i atom from the interaction with this j atom. */
 548             velec            = _mm_and_ps(velec,cutoff_mask);
 549             velec            = _mm_andnot_ps(dummy_mask,velec);
 550             velecsum         = _mm_add_ps(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             fscal            = _mm_and_ps(fscal,cutoff_mask);
 555
 556             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 557
 558             /* Calculate temporary vectorial force */
 559             tx               = _mm_mul_ps(fscal,dx20);
 560             ty               = _mm_mul_ps(fscal,dy20);
 561             tz               = _mm_mul_ps(fscal,dz20);
 562
 563             /* Update vectorial force */
 564             fix2             = _mm_add_ps(fix2,tx);
 565             fiy2             = _mm_add_ps(fiy2,ty);
 566             fiz2             = _mm_add_ps(fiz2,tz);
 567
 568             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 569                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 570                                                    tx,ty,tz);
 571
 572             }
 573
 574             /* Inner loop uses 126 flops */
 575         }
 576
 577         /* End of innermost loop */
 578
 579         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 580                                               f+i_coord_offset,fshift+i_shift_offset);
 581
 582         ggid                        = gid[iidx];
 583         /* Update potential energies */
 584         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 585         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 586
 587         /* Increment number of inner iterations */
 588         inneriter                  += j_index_end - j_index_start;
 589
 590         /* Outer loop uses 29 flops */
 591     }
 592
 593     /* Increment number of outer iterations */
 594     outeriter        += nri;
 595
 596     /* Update outer/inner flops */
 597
 598     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*126);
 599 }
 600 /*
 601  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
 602  * Electrostatics interaction: ReactionField
 603  * VdW interaction:            LennardJones
 604  * Geometry:                   Water3-Particle
 605  * Calculate force/pot:        Force
 606  */
 607 void
 608 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_sse2_single
 609                     (t_nblist * gmx_restrict                nlist,
 610                      rvec * gmx_restrict                    xx,
 611                      rvec * gmx_restrict                    ff,
 612                      t_forcerec * gmx_restrict              fr,
 613                      t_mdatoms * gmx_restrict               mdatoms,
 614                      nb_kernel_data_t * gmx_restrict        kernel_data,
 615                      t_nrnb * gmx_restrict                  nrnb)
 616 {
 617     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 618      * just 0 for non-waters.
 619      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 620      * jnr indices corresponding to data put in the four positions in the SIMD register.
 621      */
 622     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 623     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 624     int              jnrA,jnrB,jnrC,jnrD;
 625     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 626     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 627     real             shX,shY,shZ,rcutoff_scalar;
 628     real             *shiftvec,*fshift,*x,*f;
 629     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 630     int              vdwioffset0;
 631     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 632     int              vdwioffset1;
 633     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 634     int              vdwioffset2;
 635     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 636     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 637     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 638     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 639     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 640     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 641     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 642     real             *charge;
 643     int              nvdwtype;
 644     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 645     int              *vdwtype;
 646     real             *vdwparam;
 647     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 648     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 649     __m128           dummy_mask,cutoff_mask;
 650     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 651     __m128           one     = _mm_set1_ps(1.0);
 652     __m128           two     = _mm_set1_ps(2.0);
 653     x                = xx[0];
 654     f                = ff[0];
 655
 656     nri              = nlist->nri;
 657     iinr             = nlist->iinr;
 658     jindex           = nlist->jindex;
 659     jjnr             = nlist->jjnr;
 660     shiftidx         = nlist->shift;
 661     gid              = nlist->gid;
 662     shiftvec         = fr->shift_vec[0];
 663     fshift           = fr->fshift[0];
 664     facel            = _mm_set1_ps(fr->epsfac);
 665     charge           = mdatoms->chargeA;
 666     krf              = _mm_set1_ps(fr->ic->k_rf);
 667     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 668     crf              = _mm_set1_ps(fr->ic->c_rf);
 669     nvdwtype         = fr->ntype;
 670     vdwparam         = fr->nbfp;
 671     vdwtype          = mdatoms->typeA;
 672
 673     /* Setup water-specific parameters */
 674     inr              = nlist->iinr[0];
 675     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 676     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 677     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 678     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 679
 680     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 681     rcutoff_scalar   = fr->rcoulomb;
 682     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 683     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 684
 685     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 686     rvdw             = _mm_set1_ps(fr->rvdw);
 687
 688     /* Avoid stupid compiler warnings */
 689     jnrA = jnrB = jnrC = jnrD = 0;
 690     j_coord_offsetA = 0;
 691     j_coord_offsetB = 0;
 692     j_coord_offsetC = 0;
 693     j_coord_offsetD = 0;
 694
 695     outeriter        = 0;
 696     inneriter        = 0;
 697
 698     /* Start outer loop over neighborlists */
 699     for(iidx=0; iidx<nri; iidx++)
 700     {
 701         /* Load shift vector for this list */
 702         i_shift_offset   = DIM*shiftidx[iidx];
 703         shX              = shiftvec[i_shift_offset+XX];
 704         shY              = shiftvec[i_shift_offset+YY];
 705         shZ              = shiftvec[i_shift_offset+ZZ];
 706
 707         /* Load limits for loop over neighbors */
 708         j_index_start    = jindex[iidx];
 709         j_index_end      = jindex[iidx+1];
 710
 711         /* Get outer coordinate index */
 712         inr              = iinr[iidx];
 713         i_coord_offset   = DIM*inr;
 714
 715         /* Load i particle coords and add shift vector */
 716         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 717         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 718         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 719         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 720         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 721         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 722         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 723         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 724         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 725
 726         fix0             = _mm_setzero_ps();
 727         fiy0             = _mm_setzero_ps();
 728         fiz0             = _mm_setzero_ps();
 729         fix1             = _mm_setzero_ps();
 730         fiy1             = _mm_setzero_ps();
 731         fiz1             = _mm_setzero_ps();
 732         fix2             = _mm_setzero_ps();
 733         fiy2             = _mm_setzero_ps();
 734         fiz2             = _mm_setzero_ps();
 735
 736         /* Start inner kernel loop */
 737         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 738         {
 739
 740             /* Get j neighbor index, and coordinate index */
 741             jnrA             = jjnr[jidx];
 742             jnrB             = jjnr[jidx+1];
 743             jnrC             = jjnr[jidx+2];
 744             jnrD             = jjnr[jidx+3];
 745
 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
 767             /* Calculate squared distance and things based on it */
 768             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 769             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 770             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 771
 772             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 773             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 774             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 775
 776             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 777             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 778             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 779
 780             /* Load parameters for j particles */
 781             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 782                                                               charge+jnrC+0,charge+jnrD+0);
 783             vdwjidx0A        = 2*vdwtype[jnrA+0];
 784             vdwjidx0B        = 2*vdwtype[jnrB+0];
 785             vdwjidx0C        = 2*vdwtype[jnrC+0];
 786             vdwjidx0D        = 2*vdwtype[jnrD+0];
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             if (gmx_mm_any_lt(rsq00,rcutoff2))
 793             {
 794
 795             /* Compute parameters for interactions between i and j atoms */
 796             qq00             = _mm_mul_ps(iq0,jq0);
 797             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 798                                          vdwparam+vdwioffset0+vdwjidx0B,
 799                                          vdwparam+vdwioffset0+vdwjidx0C,
 800                                          vdwparam+vdwioffset0+vdwjidx0D,
 801                                          &c6_00,&c12_00);
 802
 803             /* REACTION-FIELD ELECTROSTATICS */
 804             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 805
 806             /* LENNARD-JONES DISPERSION/REPULSION */
 807
 808             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 809             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 810
 811             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 812
 813             fscal            = _mm_add_ps(felec,fvdw);
 814
 815             fscal            = _mm_and_ps(fscal,cutoff_mask);
 816
 817             /* Calculate temporary vectorial force */
 818             tx               = _mm_mul_ps(fscal,dx00);
 819             ty               = _mm_mul_ps(fscal,dy00);
 820             tz               = _mm_mul_ps(fscal,dz00);
 821
 822             /* Update vectorial force */
 823             fix0             = _mm_add_ps(fix0,tx);
 824             fiy0             = _mm_add_ps(fiy0,ty);
 825             fiz0             = _mm_add_ps(fiz0,tz);
 826
 827             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 828                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 829                                                    tx,ty,tz);
 830
 831             }
 832
 833             /**************************
 834              * CALCULATE INTERACTIONS *
 835              **************************/
 836
 837             if (gmx_mm_any_lt(rsq10,rcutoff2))
 838             {
 839
 840             /* Compute parameters for interactions between i and j atoms */
 841             qq10             = _mm_mul_ps(iq1,jq0);
 842
 843             /* REACTION-FIELD ELECTROSTATICS */
 844             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 845
 846             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 847
 848             fscal            = felec;
 849
 850             fscal            = _mm_and_ps(fscal,cutoff_mask);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm_mul_ps(fscal,dx10);
 854             ty               = _mm_mul_ps(fscal,dy10);
 855             tz               = _mm_mul_ps(fscal,dz10);
 856
 857             /* Update vectorial force */
 858             fix1             = _mm_add_ps(fix1,tx);
 859             fiy1             = _mm_add_ps(fiy1,ty);
 860             fiz1             = _mm_add_ps(fiz1,tz);
 861
 862             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 863                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 864                                                    tx,ty,tz);
 865
 866             }
 867
 868             /**************************
 869              * CALCULATE INTERACTIONS *
 870              **************************/
 871
 872             if (gmx_mm_any_lt(rsq20,rcutoff2))
 873             {
 874
 875             /* Compute parameters for interactions between i and j atoms */
 876             qq20             = _mm_mul_ps(iq2,jq0);
 877
 878             /* REACTION-FIELD ELECTROSTATICS */
 879             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 880
 881             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 882
 883             fscal            = felec;
 884
 885             fscal            = _mm_and_ps(fscal,cutoff_mask);
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm_mul_ps(fscal,dx20);
 889             ty               = _mm_mul_ps(fscal,dy20);
 890             tz               = _mm_mul_ps(fscal,dz20);
 891
 892             /* Update vectorial force */
 893             fix2             = _mm_add_ps(fix2,tx);
 894             fiy2             = _mm_add_ps(fiy2,ty);
 895             fiz2             = _mm_add_ps(fiz2,tz);
 896
 897             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 898                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 899                                                    tx,ty,tz);
 900
 901             }
 902
 903             /* Inner loop uses 97 flops */
 904         }
 905
 906         if(jidx<j_index_end)
 907         {
 908
 909             /* Get j neighbor index, and coordinate index */
 910             jnrA             = jjnr[jidx];
 911             jnrB             = jjnr[jidx+1];
 912             jnrC             = jjnr[jidx+2];
 913             jnrD             = jjnr[jidx+3];
 914
 915             /* Sign of each element will be negative for non-real atoms.
 916              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 917              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 918              */
 919             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 920             jnrA       = (jnrA>=0) ? jnrA : 0;
 921             jnrB       = (jnrB>=0) ? jnrB : 0;
 922             jnrC       = (jnrC>=0) ? jnrC : 0;
 923             jnrD       = (jnrD>=0) ? jnrD : 0;
 924
 925             j_coord_offsetA  = DIM*jnrA;
 926             j_coord_offsetB  = DIM*jnrB;
 927             j_coord_offsetC  = DIM*jnrC;
 928             j_coord_offsetD  = DIM*jnrD;
 929
 930             /* load j atom coordinates */
 931             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 932                                               x+j_coord_offsetC,x+j_coord_offsetD,
 933                                               &jx0,&jy0,&jz0);
 934
 935             /* Calculate displacement vector */
 936             dx00             = _mm_sub_ps(ix0,jx0);
 937             dy00             = _mm_sub_ps(iy0,jy0);
 938             dz00             = _mm_sub_ps(iz0,jz0);
 939             dx10             = _mm_sub_ps(ix1,jx0);
 940             dy10             = _mm_sub_ps(iy1,jy0);
 941             dz10             = _mm_sub_ps(iz1,jz0);
 942             dx20             = _mm_sub_ps(ix2,jx0);
 943             dy20             = _mm_sub_ps(iy2,jy0);
 944             dz20             = _mm_sub_ps(iz2,jz0);
 945
 946             /* Calculate squared distance and things based on it */
 947             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 948             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 949             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 950
 951             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 952             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 953             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 954
 955             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 956             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 957             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 958
 959             /* Load parameters for j particles */
 960             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 961                                                               charge+jnrC+0,charge+jnrD+0);
 962             vdwjidx0A        = 2*vdwtype[jnrA+0];
 963             vdwjidx0B        = 2*vdwtype[jnrB+0];
 964             vdwjidx0C        = 2*vdwtype[jnrC+0];
 965             vdwjidx0D        = 2*vdwtype[jnrD+0];
 966
 967             /**************************
 968              * CALCULATE INTERACTIONS *
 969              **************************/
 970
 971             if (gmx_mm_any_lt(rsq00,rcutoff2))
 972             {
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             qq00             = _mm_mul_ps(iq0,jq0);
 976             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 977                                          vdwparam+vdwioffset0+vdwjidx0B,
 978                                          vdwparam+vdwioffset0+vdwjidx0C,
 979                                          vdwparam+vdwioffset0+vdwjidx0D,
 980                                          &c6_00,&c12_00);
 981
 982             /* REACTION-FIELD ELECTROSTATICS */
 983             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 984
 985             /* LENNARD-JONES DISPERSION/REPULSION */
 986
 987             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 988             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 989
 990             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 991
 992             fscal            = _mm_add_ps(felec,fvdw);
 993
 994             fscal            = _mm_and_ps(fscal,cutoff_mask);
 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             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1009                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1010                                                    tx,ty,tz);
1011
1012             }
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             if (gmx_mm_any_lt(rsq10,rcutoff2))
1019             {
1020
1021             /* Compute parameters for interactions between i and j atoms */
1022             qq10             = _mm_mul_ps(iq1,jq0);
1023
1024             /* REACTION-FIELD ELECTROSTATICS */
1025             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1026
1027             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1028
1029             fscal            = felec;
1030
1031             fscal            = _mm_and_ps(fscal,cutoff_mask);
1032
1033             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1034
1035             /* Calculate temporary vectorial force */
1036             tx               = _mm_mul_ps(fscal,dx10);
1037             ty               = _mm_mul_ps(fscal,dy10);
1038             tz               = _mm_mul_ps(fscal,dz10);
1039
1040             /* Update vectorial force */
1041             fix1             = _mm_add_ps(fix1,tx);
1042             fiy1             = _mm_add_ps(fiy1,ty);
1043             fiz1             = _mm_add_ps(fiz1,tz);
1044
1045             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1046                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1047                                                    tx,ty,tz);
1048
1049             }
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             if (gmx_mm_any_lt(rsq20,rcutoff2))
1056             {
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq20             = _mm_mul_ps(iq2,jq0);
1060
1061             /* REACTION-FIELD ELECTROSTATICS */
1062             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1063
1064             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1065
1066             fscal            = felec;
1067
1068             fscal            = _mm_and_ps(fscal,cutoff_mask);
1069
1070             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1071
1072             /* Calculate temporary vectorial force */
1073             tx               = _mm_mul_ps(fscal,dx20);
1074             ty               = _mm_mul_ps(fscal,dy20);
1075             tz               = _mm_mul_ps(fscal,dz20);
1076
1077             /* Update vectorial force */
1078             fix2             = _mm_add_ps(fix2,tx);
1079             fiy2             = _mm_add_ps(fiy2,ty);
1080             fiz2             = _mm_add_ps(fiz2,tz);
1081
1082             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1083                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1084                                                    tx,ty,tz);
1085
1086             }
1087
1088             /* Inner loop uses 97 flops */
1089         }
1090
1091         /* End of innermost loop */
1092
1093         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1094                                               f+i_coord_offset,fshift+i_shift_offset);
1095
1096         /* Increment number of inner iterations */
1097         inneriter                  += j_index_end - j_index_start;
1098
1099         /* Outer loop uses 27 flops */
1100     }
1101
1102     /* Increment number of outer iterations */
1103     outeriter        += nri;
1104
1105     /* Update outer/inner flops */
1106
1107     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*97);
1108 }