src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwLJ_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwLJ_GeomW4P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              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              vdwioffset3;
  73     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  75     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128           dummy_mask,cutoff_mask;
  89     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  90     __m128           one     = _mm_set1_ps(1.0);
  91     __m128           two     = _mm_set1_ps(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm_set1_ps(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     krf              = _mm_set1_ps(fr->ic->k_rf);
 106     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 107     crf              = _mm_set1_ps(fr->ic->c_rf);
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 115     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 116     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 117     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     /* Start outer loop over neighborlists */
 130     for(iidx=0; iidx<nri; iidx++)
 131     {
 132         /* Load shift vector for this list */
 133         i_shift_offset   = DIM*shiftidx[iidx];
 134         shX              = shiftvec[i_shift_offset+XX];
 135         shY              = shiftvec[i_shift_offset+YY];
 136         shZ              = shiftvec[i_shift_offset+ZZ];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 148         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 149         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 150         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 151         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 152         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 153         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 154         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 155         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 156         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 157         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 158         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 159
 160         fix0             = _mm_setzero_ps();
 161         fiy0             = _mm_setzero_ps();
 162         fiz0             = _mm_setzero_ps();
 163         fix1             = _mm_setzero_ps();
 164         fiy1             = _mm_setzero_ps();
 165         fiz1             = _mm_setzero_ps();
 166         fix2             = _mm_setzero_ps();
 167         fiy2             = _mm_setzero_ps();
 168         fiz2             = _mm_setzero_ps();
 169         fix3             = _mm_setzero_ps();
 170         fiy3             = _mm_setzero_ps();
 171         fiz3             = _mm_setzero_ps();
 172
 173         /* Reset potential sums */
 174         velecsum         = _mm_setzero_ps();
 175         vvdwsum          = _mm_setzero_ps();
 176
 177         /* Start inner kernel loop */
 178         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 179         {
 180
 181             /* Get j neighbor index, and coordinate index */
 182             jnrA             = jjnr[jidx];
 183             jnrB             = jjnr[jidx+1];
 184             jnrC             = jjnr[jidx+2];
 185             jnrD             = jjnr[jidx+3];
 186
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_ps(ix0,jx0);
 199             dy00             = _mm_sub_ps(iy0,jy0);
 200             dz00             = _mm_sub_ps(iz0,jz0);
 201             dx10             = _mm_sub_ps(ix1,jx0);
 202             dy10             = _mm_sub_ps(iy1,jy0);
 203             dz10             = _mm_sub_ps(iz1,jz0);
 204             dx20             = _mm_sub_ps(ix2,jx0);
 205             dy20             = _mm_sub_ps(iy2,jy0);
 206             dz20             = _mm_sub_ps(iz2,jz0);
 207             dx30             = _mm_sub_ps(ix3,jx0);
 208             dy30             = _mm_sub_ps(iy3,jy0);
 209             dz30             = _mm_sub_ps(iz3,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 215             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 216
 217             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 218             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 219             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 220
 221             rinvsq00         = gmx_mm_inv_ps(rsq00);
 222             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 223             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 224             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 228                                                               charge+jnrC+0,charge+jnrD+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231             vdwjidx0C        = 2*vdwtype[jnrC+0];
 232             vdwjidx0D        = 2*vdwtype[jnrD+0];
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 240                                          vdwparam+vdwioffset0+vdwjidx0B,
 241                                          vdwparam+vdwioffset0+vdwjidx0C,
 242                                          vdwparam+vdwioffset0+vdwjidx0D,
 243                                          &c6_00,&c12_00);
 244
 245             /* LENNARD-JONES DISPERSION/REPULSION */
 246
 247             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 248             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 249             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 250             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 251             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 252
 253             /* Update potential sum for this i atom from the interaction with this j atom. */
 254             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 255
 256             fscal            = fvdw;
 257
 258             /* Calculate temporary vectorial force */
 259             tx               = _mm_mul_ps(fscal,dx00);
 260             ty               = _mm_mul_ps(fscal,dy00);
 261             tz               = _mm_mul_ps(fscal,dz00);
 262
 263             /* Update vectorial force */
 264             fix0             = _mm_add_ps(fix0,tx);
 265             fiy0             = _mm_add_ps(fiy0,ty);
 266             fiz0             = _mm_add_ps(fiz0,tz);
 267
 268             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 269                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 270                                                    tx,ty,tz);
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq10             = _mm_mul_ps(iq1,jq0);
 278
 279             /* REACTION-FIELD ELECTROSTATICS */
 280             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 281             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm_add_ps(velecsum,velec);
 285
 286             fscal            = felec;
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm_mul_ps(fscal,dx10);
 290             ty               = _mm_mul_ps(fscal,dy10);
 291             tz               = _mm_mul_ps(fscal,dz10);
 292
 293             /* Update vectorial force */
 294             fix1             = _mm_add_ps(fix1,tx);
 295             fiy1             = _mm_add_ps(fiy1,ty);
 296             fiz1             = _mm_add_ps(fiz1,tz);
 297
 298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 299                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 300                                                    tx,ty,tz);
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq20             = _mm_mul_ps(iq2,jq0);
 308
 309             /* REACTION-FIELD ELECTROSTATICS */
 310             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 311             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velecsum         = _mm_add_ps(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             /* Calculate temporary vectorial force */
 319             tx               = _mm_mul_ps(fscal,dx20);
 320             ty               = _mm_mul_ps(fscal,dy20);
 321             tz               = _mm_mul_ps(fscal,dz20);
 322
 323             /* Update vectorial force */
 324             fix2             = _mm_add_ps(fix2,tx);
 325             fiy2             = _mm_add_ps(fiy2,ty);
 326             fiz2             = _mm_add_ps(fiz2,tz);
 327
 328             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 329                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 330                                                    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             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             /* Inner loop uses 128 flops */
 363         }
 364
 365         if(jidx<j_index_end)
 366         {
 367
 368             /* Get j neighbor index, and coordinate index */
 369             jnrA             = jjnr[jidx];
 370             jnrB             = jjnr[jidx+1];
 371             jnrC             = jjnr[jidx+2];
 372             jnrD             = jjnr[jidx+3];
 373
 374             /* Sign of each element will be negative for non-real atoms.
 375              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 376              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 377              */
 378             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 379             jnrA       = (jnrA>=0) ? jnrA : 0;
 380             jnrB       = (jnrB>=0) ? jnrB : 0;
 381             jnrC       = (jnrC>=0) ? jnrC : 0;
 382             jnrD       = (jnrD>=0) ? jnrD : 0;
 383
 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             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 469                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 470                                                    tx,ty,tz);
 471
 472             /**************************
 473              * CALCULATE INTERACTIONS *
 474              **************************/
 475
 476             /* Compute parameters for interactions between i and j atoms */
 477             qq10             = _mm_mul_ps(iq1,jq0);
 478
 479             /* REACTION-FIELD ELECTROSTATICS */
 480             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 481             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 482
 483             /* Update potential sum for this i atom from the interaction with this j atom. */
 484             velec            = _mm_andnot_ps(dummy_mask,velec);
 485             velecsum         = _mm_add_ps(velecsum,velec);
 486
 487             fscal            = felec;
 488
 489             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 490
 491             /* Calculate temporary vectorial force */
 492             tx               = _mm_mul_ps(fscal,dx10);
 493             ty               = _mm_mul_ps(fscal,dy10);
 494             tz               = _mm_mul_ps(fscal,dz10);
 495
 496             /* Update vectorial force */
 497             fix1             = _mm_add_ps(fix1,tx);
 498             fiy1             = _mm_add_ps(fiy1,ty);
 499             fiz1             = _mm_add_ps(fiz1,tz);
 500
 501             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 502                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 503                                                    tx,ty,tz);
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq20             = _mm_mul_ps(iq2,jq0);
 511
 512             /* REACTION-FIELD ELECTROSTATICS */
 513             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 514             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm_andnot_ps(dummy_mask,velec);
 518             velecsum         = _mm_add_ps(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm_mul_ps(fscal,dx20);
 526             ty               = _mm_mul_ps(fscal,dy20);
 527             tz               = _mm_mul_ps(fscal,dz20);
 528
 529             /* Update vectorial force */
 530             fix2             = _mm_add_ps(fix2,tx);
 531             fiy2             = _mm_add_ps(fiy2,ty);
 532             fiz2             = _mm_add_ps(fiz2,tz);
 533
 534             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 535                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 536                                                    tx,ty,tz);
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq30             = _mm_mul_ps(iq3,jq0);
 544
 545             /* REACTION-FIELD ELECTROSTATICS */
 546             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 547             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm_andnot_ps(dummy_mask,velec);
 551             velecsum         = _mm_add_ps(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm_mul_ps(fscal,dx30);
 559             ty               = _mm_mul_ps(fscal,dy30);
 560             tz               = _mm_mul_ps(fscal,dz30);
 561
 562             /* Update vectorial force */
 563             fix3             = _mm_add_ps(fix3,tx);
 564             fiy3             = _mm_add_ps(fiy3,ty);
 565             fiz3             = _mm_add_ps(fiz3,tz);
 566
 567             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 568                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 569                                                    tx,ty,tz);
 570
 571             /* Inner loop uses 128 flops */
 572         }
 573
 574         /* End of innermost loop */
 575
 576         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 577                                               f+i_coord_offset,fshift+i_shift_offset);
 578
 579         ggid                        = gid[iidx];
 580         /* Update potential energies */
 581         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 582         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 583
 584         /* Increment number of inner iterations */
 585         inneriter                  += j_index_end - j_index_start;
 586
 587         /* Outer loop uses 38 flops */
 588     }
 589
 590     /* Increment number of outer iterations */
 591     outeriter        += nri;
 592
 593     /* Update outer/inner flops */
 594
 595     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*128);
 596 }
 597 /*
 598  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse2_single
 599  * Electrostatics interaction: ReactionField
 600  * VdW interaction:            LennardJones
 601  * Geometry:                   Water4-Particle
 602  * Calculate force/pot:        Force
 603  */
 604 void
 605 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_sse2_single
 606                     (t_nblist * gmx_restrict                nlist,
 607                      rvec * gmx_restrict                    xx,
 608                      rvec * gmx_restrict                    ff,
 609                      t_forcerec * gmx_restrict              fr,
 610                      t_mdatoms * gmx_restrict               mdatoms,
 611                      nb_kernel_data_t * gmx_restrict        kernel_data,
 612                      t_nrnb * gmx_restrict                  nrnb)
 613 {
 614     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 615      * just 0 for non-waters.
 616      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 617      * jnr indices corresponding to data put in the four positions in the SIMD register.
 618      */
 619     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 620     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 621     int              jnrA,jnrB,jnrC,jnrD;
 622     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 623     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 624     real             shX,shY,shZ,rcutoff_scalar;
 625     real             *shiftvec,*fshift,*x,*f;
 626     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 627     int              vdwioffset0;
 628     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 629     int              vdwioffset1;
 630     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 631     int              vdwioffset2;
 632     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 633     int              vdwioffset3;
 634     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 635     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 636     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 637     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 638     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 639     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 640     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 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     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 676     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 677     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 678     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 679
 680     /* Avoid stupid compiler warnings */
 681     jnrA = jnrB = jnrC = jnrD = 0;
 682     j_coord_offsetA = 0;
 683     j_coord_offsetB = 0;
 684     j_coord_offsetC = 0;
 685     j_coord_offsetD = 0;
 686
 687     outeriter        = 0;
 688     inneriter        = 0;
 689
 690     /* Start outer loop over neighborlists */
 691     for(iidx=0; iidx<nri; iidx++)
 692     {
 693         /* Load shift vector for this list */
 694         i_shift_offset   = DIM*shiftidx[iidx];
 695         shX              = shiftvec[i_shift_offset+XX];
 696         shY              = shiftvec[i_shift_offset+YY];
 697         shZ              = shiftvec[i_shift_offset+ZZ];
 698
 699         /* Load limits for loop over neighbors */
 700         j_index_start    = jindex[iidx];
 701         j_index_end      = jindex[iidx+1];
 702
 703         /* Get outer coordinate index */
 704         inr              = iinr[iidx];
 705         i_coord_offset   = DIM*inr;
 706
 707         /* Load i particle coords and add shift vector */
 708         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 709         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 710         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 711         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 712         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 713         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 714         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 715         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 716         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 717         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 718         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 719         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 720
 721         fix0             = _mm_setzero_ps();
 722         fiy0             = _mm_setzero_ps();
 723         fiz0             = _mm_setzero_ps();
 724         fix1             = _mm_setzero_ps();
 725         fiy1             = _mm_setzero_ps();
 726         fiz1             = _mm_setzero_ps();
 727         fix2             = _mm_setzero_ps();
 728         fiy2             = _mm_setzero_ps();
 729         fiz2             = _mm_setzero_ps();
 730         fix3             = _mm_setzero_ps();
 731         fiy3             = _mm_setzero_ps();
 732         fiz3             = _mm_setzero_ps();
 733
 734         /* Start inner kernel loop */
 735         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 736         {
 737
 738             /* Get j neighbor index, and coordinate index */
 739             jnrA             = jjnr[jidx];
 740             jnrB             = jjnr[jidx+1];
 741             jnrC             = jjnr[jidx+2];
 742             jnrD             = jjnr[jidx+3];
 743
 744             j_coord_offsetA  = DIM*jnrA;
 745             j_coord_offsetB  = DIM*jnrB;
 746             j_coord_offsetC  = DIM*jnrC;
 747             j_coord_offsetD  = DIM*jnrD;
 748
 749             /* load j atom coordinates */
 750             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 751                                               x+j_coord_offsetC,x+j_coord_offsetD,
 752                                               &jx0,&jy0,&jz0);
 753
 754             /* Calculate displacement vector */
 755             dx00             = _mm_sub_ps(ix0,jx0);
 756             dy00             = _mm_sub_ps(iy0,jy0);
 757             dz00             = _mm_sub_ps(iz0,jz0);
 758             dx10             = _mm_sub_ps(ix1,jx0);
 759             dy10             = _mm_sub_ps(iy1,jy0);
 760             dz10             = _mm_sub_ps(iz1,jz0);
 761             dx20             = _mm_sub_ps(ix2,jx0);
 762             dy20             = _mm_sub_ps(iy2,jy0);
 763             dz20             = _mm_sub_ps(iz2,jz0);
 764             dx30             = _mm_sub_ps(ix3,jx0);
 765             dy30             = _mm_sub_ps(iy3,jy0);
 766             dz30             = _mm_sub_ps(iz3,jz0);
 767
 768             /* Calculate squared distance and things based on it */
 769             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 770             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 771             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 772             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 773
 774             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 775             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 776             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 777
 778             rinvsq00         = gmx_mm_inv_ps(rsq00);
 779             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 780             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 781             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 782
 783             /* Load parameters for j particles */
 784             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 785                                                               charge+jnrC+0,charge+jnrD+0);
 786             vdwjidx0A        = 2*vdwtype[jnrA+0];
 787             vdwjidx0B        = 2*vdwtype[jnrB+0];
 788             vdwjidx0C        = 2*vdwtype[jnrC+0];
 789             vdwjidx0D        = 2*vdwtype[jnrD+0];
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             /* Compute parameters for interactions between i and j atoms */
 796             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 797                                          vdwparam+vdwioffset0+vdwjidx0B,
 798                                          vdwparam+vdwioffset0+vdwjidx0C,
 799                                          vdwparam+vdwioffset0+vdwjidx0D,
 800                                          &c6_00,&c12_00);
 801
 802             /* LENNARD-JONES DISPERSION/REPULSION */
 803
 804             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 805             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 806
 807             fscal            = fvdw;
 808
 809             /* Calculate temporary vectorial force */
 810             tx               = _mm_mul_ps(fscal,dx00);
 811             ty               = _mm_mul_ps(fscal,dy00);
 812             tz               = _mm_mul_ps(fscal,dz00);
 813
 814             /* Update vectorial force */
 815             fix0             = _mm_add_ps(fix0,tx);
 816             fiy0             = _mm_add_ps(fiy0,ty);
 817             fiz0             = _mm_add_ps(fiz0,tz);
 818
 819             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 820                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 821                                                    tx,ty,tz);
 822
 823             /**************************
 824              * CALCULATE INTERACTIONS *
 825              **************************/
 826
 827             /* Compute parameters for interactions between i and j atoms */
 828             qq10             = _mm_mul_ps(iq1,jq0);
 829
 830             /* REACTION-FIELD ELECTROSTATICS */
 831             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 832
 833             fscal            = felec;
 834
 835             /* Calculate temporary vectorial force */
 836             tx               = _mm_mul_ps(fscal,dx10);
 837             ty               = _mm_mul_ps(fscal,dy10);
 838             tz               = _mm_mul_ps(fscal,dz10);
 839
 840             /* Update vectorial force */
 841             fix1             = _mm_add_ps(fix1,tx);
 842             fiy1             = _mm_add_ps(fiy1,ty);
 843             fiz1             = _mm_add_ps(fiz1,tz);
 844
 845             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 846                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 847                                                    tx,ty,tz);
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq20             = _mm_mul_ps(iq2,jq0);
 855
 856             /* REACTION-FIELD ELECTROSTATICS */
 857             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 858
 859             fscal            = felec;
 860
 861             /* Calculate temporary vectorial force */
 862             tx               = _mm_mul_ps(fscal,dx20);
 863             ty               = _mm_mul_ps(fscal,dy20);
 864             tz               = _mm_mul_ps(fscal,dz20);
 865
 866             /* Update vectorial force */
 867             fix2             = _mm_add_ps(fix2,tx);
 868             fiy2             = _mm_add_ps(fiy2,ty);
 869             fiz2             = _mm_add_ps(fiz2,tz);
 870
 871             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 872                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 873                                                    tx,ty,tz);
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             /* Compute parameters for interactions between i and j atoms */
 880             qq30             = _mm_mul_ps(iq3,jq0);
 881
 882             /* REACTION-FIELD ELECTROSTATICS */
 883             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 884
 885             fscal            = felec;
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm_mul_ps(fscal,dx30);
 889             ty               = _mm_mul_ps(fscal,dy30);
 890             tz               = _mm_mul_ps(fscal,dz30);
 891
 892             /* Update vectorial force */
 893             fix3             = _mm_add_ps(fix3,tx);
 894             fiy3             = _mm_add_ps(fiy3,ty);
 895             fiz3             = _mm_add_ps(fiz3,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             /* Inner loop uses 108 flops */
 902         }
 903
 904         if(jidx<j_index_end)
 905         {
 906
 907             /* Get j neighbor index, and coordinate index */
 908             jnrA             = jjnr[jidx];
 909             jnrB             = jjnr[jidx+1];
 910             jnrC             = jjnr[jidx+2];
 911             jnrD             = jjnr[jidx+3];
 912
 913             /* Sign of each element will be negative for non-real atoms.
 914              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 915              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 916              */
 917             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 918             jnrA       = (jnrA>=0) ? jnrA : 0;
 919             jnrB       = (jnrB>=0) ? jnrB : 0;
 920             jnrC       = (jnrC>=0) ? jnrC : 0;
 921             jnrD       = (jnrD>=0) ? jnrD : 0;
 922
 923             j_coord_offsetA  = DIM*jnrA;
 924             j_coord_offsetB  = DIM*jnrB;
 925             j_coord_offsetC  = DIM*jnrC;
 926             j_coord_offsetD  = DIM*jnrD;
 927
 928             /* load j atom coordinates */
 929             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 930                                               x+j_coord_offsetC,x+j_coord_offsetD,
 931                                               &jx0,&jy0,&jz0);
 932
 933             /* Calculate displacement vector */
 934             dx00             = _mm_sub_ps(ix0,jx0);
 935             dy00             = _mm_sub_ps(iy0,jy0);
 936             dz00             = _mm_sub_ps(iz0,jz0);
 937             dx10             = _mm_sub_ps(ix1,jx0);
 938             dy10             = _mm_sub_ps(iy1,jy0);
 939             dz10             = _mm_sub_ps(iz1,jz0);
 940             dx20             = _mm_sub_ps(ix2,jx0);
 941             dy20             = _mm_sub_ps(iy2,jy0);
 942             dz20             = _mm_sub_ps(iz2,jz0);
 943             dx30             = _mm_sub_ps(ix3,jx0);
 944             dy30             = _mm_sub_ps(iy3,jy0);
 945             dz30             = _mm_sub_ps(iz3,jz0);
 946
 947             /* Calculate squared distance and things based on it */
 948             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 949             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 950             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 951             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 952
 953             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 954             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 955             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 956
 957             rinvsq00         = gmx_mm_inv_ps(rsq00);
 958             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 959             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 960             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 961
 962             /* Load parameters for j particles */
 963             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 964                                                               charge+jnrC+0,charge+jnrD+0);
 965             vdwjidx0A        = 2*vdwtype[jnrA+0];
 966             vdwjidx0B        = 2*vdwtype[jnrB+0];
 967             vdwjidx0C        = 2*vdwtype[jnrC+0];
 968             vdwjidx0D        = 2*vdwtype[jnrD+0];
 969
 970             /**************************
 971              * CALCULATE INTERACTIONS *
 972              **************************/
 973
 974             /* Compute parameters for interactions between i and j atoms */
 975             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 976                                          vdwparam+vdwioffset0+vdwjidx0B,
 977                                          vdwparam+vdwioffset0+vdwjidx0C,
 978                                          vdwparam+vdwioffset0+vdwjidx0D,
 979                                          &c6_00,&c12_00);
 980
 981             /* LENNARD-JONES DISPERSION/REPULSION */
 982
 983             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 984             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 985
 986             fscal            = fvdw;
 987
 988             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm_mul_ps(fscal,dx00);
 992             ty               = _mm_mul_ps(fscal,dy00);
 993             tz               = _mm_mul_ps(fscal,dz00);
 994
 995             /* Update vectorial force */
 996             fix0             = _mm_add_ps(fix0,tx);
 997             fiy0             = _mm_add_ps(fiy0,ty);
 998             fiz0             = _mm_add_ps(fiz0,tz);
 999
1000             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1001                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1002                                                    tx,ty,tz);
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             /* Compute parameters for interactions between i and j atoms */
1009             qq10             = _mm_mul_ps(iq1,jq0);
1010
1011             /* REACTION-FIELD ELECTROSTATICS */
1012             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1013
1014             fscal            = felec;
1015
1016             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm_mul_ps(fscal,dx10);
1020             ty               = _mm_mul_ps(fscal,dy10);
1021             tz               = _mm_mul_ps(fscal,dz10);
1022
1023             /* Update vectorial force */
1024             fix1             = _mm_add_ps(fix1,tx);
1025             fiy1             = _mm_add_ps(fiy1,ty);
1026             fiz1             = _mm_add_ps(fiz1,tz);
1027
1028             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1029                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1030                                                    tx,ty,tz);
1031
1032             /**************************
1033              * CALCULATE INTERACTIONS *
1034              **************************/
1035
1036             /* Compute parameters for interactions between i and j atoms */
1037             qq20             = _mm_mul_ps(iq2,jq0);
1038
1039             /* REACTION-FIELD ELECTROSTATICS */
1040             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1041
1042             fscal            = felec;
1043
1044             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1045
1046             /* Calculate temporary vectorial force */
1047             tx               = _mm_mul_ps(fscal,dx20);
1048             ty               = _mm_mul_ps(fscal,dy20);
1049             tz               = _mm_mul_ps(fscal,dz20);
1050
1051             /* Update vectorial force */
1052             fix2             = _mm_add_ps(fix2,tx);
1053             fiy2             = _mm_add_ps(fiy2,ty);
1054             fiz2             = _mm_add_ps(fiz2,tz);
1055
1056             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1057                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1058                                                    tx,ty,tz);
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             /* Compute parameters for interactions between i and j atoms */
1065             qq30             = _mm_mul_ps(iq3,jq0);
1066
1067             /* REACTION-FIELD ELECTROSTATICS */
1068             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1069
1070             fscal            = felec;
1071
1072             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1073
1074             /* Calculate temporary vectorial force */
1075             tx               = _mm_mul_ps(fscal,dx30);
1076             ty               = _mm_mul_ps(fscal,dy30);
1077             tz               = _mm_mul_ps(fscal,dz30);
1078
1079             /* Update vectorial force */
1080             fix3             = _mm_add_ps(fix3,tx);
1081             fiy3             = _mm_add_ps(fiy3,ty);
1082             fiz3             = _mm_add_ps(fiz3,tz);
1083
1084             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1085                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1086                                                    tx,ty,tz);
1087
1088             /* Inner loop uses 108 flops */
1089         }
1090
1091         /* End of innermost loop */
1092
1093         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
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 36 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_W4_F,outeriter*36 + inneriter*108);
1108 }